Tensor SECOND EDITION

Analysis
Theory and Applications to Geometry and
Mechanics of Continua

I. S. Sokolnikoff
 RICHARD C. FREY
1003 SUNSET
CINCINNAT! 5, OHIO

about the book...

The second edition of this popular
text continues to provide a clear and
explicit introduction to the applica-
tions of mathematics. Because of the
- importance of linear transformations
in motivating the development of
tensor theory, the author begins the
book with a discussion of linear
transformations and matrices. He
then proceeds with a self-contained
presentation of algebra and calculus
of tensors.
Dr. Sokolnikoff has expanded cer-
tain sections on the uses of calculus
of variations in geometry in a gen-
eral discussion of those geometrical
topics that are important in the study
of analytical dynamics and mechan-
ics of continuous media. The essen-
tial concepts of analytical mechanics
are presented in a somewhat ex-
panded but still concise manner.
Relativistic mechanics is introduced
and illuminated by a number of illus-
trative examples. The concluding
chapter, devoted to mechanics of
continua, has been rewritten to pre-
sent the essentials of the non-linear
theory of mechanics of deformable
media from a unified point of view.
The Second Edition of TENSOR
ANALYSIS provides a careful and
broad introduction to the develop-
ment of tensor theory and its appli-
cations to geometry, mechanics,
relativity and mechanics of continu-
ous media.
N.U.—5829
 | TBienago C, Sy
Yo) ‘ ; ~~

° betI a

F A i
 APPLIED MATHEMATICS SERIES

Edited by

I. S. SOKOLNIKOFF

TENSOR ANALYSIS
THEORY AND APPLICATIONS TO GEOMETRY
AND MECHANICS OF CONTINUA
 APPLIED MATHEMATICS SERIES

The Applied Mathematics Series is devoted to books

dealing with mathematical theories underlying physical
and biological sciences, and with advanced mathematical
techniques needed for solving problems of these sciences.
TENSOR ANALYSIS
THEORY AND APPLICATIONS TO GEOMETRY

AND MECHANICS OF CONTINUA

Second Edition

I. S. SOKOLNIKOFF
PROFESSOR OF MATHEMATICS
UNIVERSITY OF CALIFORNIA
LOS ANGELES

JOHN WILEY & SONS, INC.

NEW YORK - LONDON - SYDNEY
 CopyYRIGHT,1951 © 1964
BY
JOHN WILEY & Sons, INC.

All Rights Reserved.

This book or any part thereof must not

be reproduced in any form without the
written permission of the publisher.

SECOND PRINTING, DECEMBER, 1965

Library of Congress Catalog Card Number: 64-13223 ,

PRINTED IN THE UNITED STATES OF AMERICA
 PREFACE TO THE SECOND EDITION

In preparing the Second Edition of this book I have been guided by

suggestions kindly made to me by users of the First Edition. There
appeared to be no compelling reasons for making major changes in the
introductory chapter concerned with linear transformations and matrices,
or in the second chapter, devoted to algebra and calculus of tensors.
In Chapter 3 some sections concerned with the uses of calculus of
variations in geometry have been expanded, some new illustrative material
introduced, and two new sections, on parallel surfaces and the Gauss-
Bonnet theorem, have been added. Chapters 2 and 3 in the present
edition contain adequate material for an introductory course on metric
differential geometry at the beginning graduate level or, for that matter,
at the upper-division undergraduate level.
Chapter 4, dealing with analytical mechanics, has been expanded. It
contains a distillation of the essentials of classical analytical mechanics
and potential theory, which, together with Chapter 5 on relativistic me-
chanics, should be, but often is not, a part of the equipment of every
student of mathematics. A number of illustrative examples that further
illuminate the theory have been introduced, and the discussion of non-
holonomic dynamical systems, of Hamilton’s canonical equations, and
of potential theory has been made more detailed.
The concluding chapter, devoted to mechanics of continua, was entirely
rewritten. It presents from a unified point of view and, it is hoped, with
sufficient clarity, the essentials of the nonlinear theory of mechanics of
deformable media. This chapter provides a common basis for a careful
development of the mathematical theories of elasticity, plasticity, hydro-
dynamics, and gas dynamics.
‘ I. S. SOKOLNIKOFF
Pacific Palisades, California
January 1964
 PREFACE TO THE FIRST EDITION

This book is an outgrowth of a course of lectures I gave over a period of

years at the University of Wisconsin, Brown University, and the Univer-
sity of California. My audience consisted, for the most part, of graduate
students interested in applications of mathematics, and this fact shaped
both the content and the character of exposition.
Because of the importance of linear transformations in motivating the
development of tensor theory, the first chapter in this book is given to a
discussion of linear transformations and matrices, in which stress is placed
on the geometry and physics of the situation. Although a large part of
the subject matter treated in this chapter is normally covered in courses
on matrix algebra, only a few of my listeners have had the sort of appre-
ciation of matrix transformations that an applied mathematician should
have.
The second chapter is concerned with algebra and calculus of tensors.
The treatment in it is self-contained and is not made to depend on some
special field of mathematics as a vehicle for the development of tensor
analysis. This is a departure from the customary practice of making
geometry or relativity a medium for the unfolding of tensor analysis.
Although this latter practice has a great deal to commend it because it
provides a simple means for motivating the study of tensors, it often
leaves an erroneous impression that the formulation of tensor analysis
depends somehow on geometry or relativity.
The remaining four chapters in this volume deal with the applica-
tions of tensor calculus to geometry, analytical mechanics, relativistic
mechanics, and mechanics of deformable media. Thus, Chapter 3 con-
tains a selection of those geometrical topics that are important in the
study of analytical dynamics and in such portions of elasticity and plas-
ticity as deal with the deformation of plates and shells. This chapter
provides a substantial introduction to the subject of metric differential
geometry. In Chapter 4, the essential concepts of analytical mechanics
are presented adequately and concisely. An introduction to relativistic
mechanics is contained in Chapter 5. The treatment there was inten-
have
tionally made very brief because some excellent books on relativity
appeared recently and there seems little point in duplicatin g their contents.
Vil
Vili PREFACE TO THE FIRST EDITION
The final chapter of the book is concerned with a formulation of the
essential ideas of nonlinear mechanics of continuous media in the most
general tensor form. The classical linearized equations of elasticity and
fluid mechanics appear as special cases of the general treatment.
Perhaps the best evidence of the remarkable effectiveness of the tensor
apparatus in the study of Nature is in the fact that it was possible to
include, between the covers of one small volume, a large amount of
material that is of interest to mathematicians, physicists, and engineers.
A survey of applied mathematics as broad as that in this book must
inevitably reflect contributions of so many scholars that it is futile to
attempt to assign proper credit for original ideas or methods of attack.
However, in the treatment of geometry, the influence of T. Levi-Civita
and A. J. McConnell, whose books (especially McConnell’s Applications
of the Absolute Differential Calculus) I used in my classes for many
years as required reading, is clearly discernible. Specific acknowledg-
ments to these and other authors are made in the appropriate places in
the text. However, my greatest debt is to my listeners, who have made
the job of writing this book seem both enjoyable and worth while.
It is a particular pleasure to single out among my listeners Mr. William
R. Seugling, Research Assistant at the University of California at Los
Angeles, who gave unstintingly of his time in following this book through
press.
I. S. SOKOLNIKOFF
Los Angeles, California
November 1951
 CONTENTS ;

1 LINEAR VECTOR SPACES. MATRICES

. Coordinate Systems
The Geometric Concept of a Vector
Linear Vector Spaces. Dimensionality of Space
. N-Dimensional Spaces
Linear Vector Spaces of n Dimensions
. Complex Linear Vector Spaces
. Summation Convention. Review of Determinants
. Linear Transformations and Matrices
. Linear Transformations in Euclidean 3-space
_ . Orthogonal Transformation in E,
_ Linear Transformations in n-Dimensional Euclidean Spaces
—VN. Reduction of Matrices to the Diagonal Form
133 Real Symmetric Matrices and Quadratic Forms
. Illustrations of Reduction of Quadratic Forms
. Classification and Properties of Quadratic Forms
. Simultaneous Reduction of Two Quadratic Forms to a Sum of Squares
. Unitary Transformations and Hermitean Matrices

2 TENSOR THEORY

. Scope of Tensor Analysis. Invariance

. Transformation of Coordinates
. Properties of Admissible Transformations of Coordinates
. Transformation by Invariance
. Transformation by Covariance and Contravariance
_ The Tensor Concept. Contravariant and Covariant Tensors
. Tensor Character of Covariant and Contravariant Laws
. Algebra of Tensors
. Quotient Laws
. Symmetric and Skew-Symmetric Tensors
. Relative Tensors 5
. The Metric Tensor
. The Fundamental and Associated Tensors
. Christoffel’s Symbols
. Transformation of Christoffel’s Symbols
. Covariant Differentiation of Tensors
. Formulas for Covariant Differentiation
. Ricci’s Theorem
. Riemann-Christoffel Tensor
xii CONTENTS
122. Fluid Mechanics. Equations of Continuity 344
123. Ideal Fluids. Euler’s Equations 346
124. Viscous Fluids. Navier’s Equations 349
125. Remarks on Turbulent Flows and Dissipative Media 352
Bibliography 353
Index 355
 1 ;

LINEAR VECTOR SPACES. MATRICES

1. Coordinate Systems

In order to locate a geometrical configuration a reference frame is

needed. Among the simplest reference frames used in mathematics are
the cartesian coordinate systems. Although the construction of such
coordinate systems is familiar to the reader from courses in analytic
geometry, we review it here in order to set in relief certain basic notions
that underlie the concept of coordinates covering the space of our physical
intuition. This review will pave the ground for some far-reaching generali-
zations of the concept of physical space, which we formulate in Sec. 4.
The cardinal idea responsible for the invention of coordinate systems
by Descartes is the identification of the set of points composing a straight
line with the totality of real numbers. It consists of the assumption that
to each real number there corresponds a unique point on a straight line,
and conversely."
We choose a straight line X and a point O on it (Fig. 1). This point
O, which we call the origin, divides the line into two half-rays. We

Q A P xe
———_ OO OCS

Fig. 1.
between the set of
1 Although the idea of one-to-one reciprocal correspondence
and the totality of real numbers had it roots in the Eudoxus
points composing a line
B.c., the invention of
theory of incommensurables, dating back to the fourth century
nth century. It should
coordinate systems did not come until the first part of the seventee
linear sets of points and
be also noted that a rigorous analysis of the relation between
last century, chiefly through
real numbers was made only during the closing years of the
of rigor depends entirely on conven-
the efforts of Dedekind and Cantor. The concept sophistication
ng tastes indicati ve of the degree of mathema tical
tions dictated by prevaili made rigorous
concepts are usually
in a given chronological period. Fruitful intuitive of definable
agreeme nts as to which ideas fall into a category
by (a) making explicit
into mathematical theories new modes
concepts and which do not, and (6) introducing
hopes) are free of contradi ction.
of reasoning which (one
1
M LINEAR VECTOR SPACES. MATRICES [CHaP. 1

Fig. 2.

designate one of these as the positive and the other as the negative half-ray.
On the positive half-ray we choose a point A and call the length of the
line segment OA the unit length. We next coordinate points on X with a
set of real numbers in the following way: If P is any point on the positive
half-ray, we define a number x associated with P by the formula

where OP and OA are lengths of the line segments OP and OA. The
number 2 is the coordinate of P. The coordinate x of the point Q on
the negative half-ray is defined by the ratio

We also assume that each real number x corresponds to one and only
one point on X. This association of the set of points on X with the set
of real numbers constitutes a coordinate system of the one-dimensional
space consisting of points on X.
The coordination of the set of points lying in the plane with sets of
real numbers is accomplished by taking two straight lines X, and X,
intersecting at a single point O-(Fig. 2). On each line a coordinate system
is constructed as above, but the units on each line need not be equal.
A pair of such lines with unit points 4 and B marked on them form the
coordinate axes X,, X,. With each point P in the plane of coordinate
axes we associate an ordered pair of real numbers (x,, x.) determined as
follows. The line through P drawn parallel to the X-axis intersects the
X-axis in a point M, with coordinate a,, and the line through P parallel
to the X,-axis cuts X, in a point M, with coordinate x. The ordered
pair of numbers (x,, x.) are the coordinates of P in the plane, and the
SEG is THE, GEOMETRIC CONCEPT (OF Ax VECTOR 3
one-to-one correspondence of ordered pairs of numbers with the set of
points in the plane X,X, is the coordinate system of the two-dimensional
space consisting of points in the plane.
The extension of this representation to points in a three-dimensional
space is obvious. We take three noncoplanar lines Xj, X, X3 intersecting
at the common point O. On each of these lines we establish coordinate
systems, and we associate with each point P an ordered triplet of numbers
(x1, Xa, %3) determined by the intersection with the axes of three planes
drawn through P parallel to the coordinate planes X,X>, XX, and X,X3.
The coordinate systems just described are called oblique cartesian
systems. Their construction makes use of the notions of length and
parallelism of ordinary Euclidean geometry, and the essential feature of
it is the concept of one-to-one correspondence of points with ordered
sets of numbers. In the event the coordinate axes Xj, X2, X, intersect
at right angles, the coordinate system is said to be orthogonal cartesian,
or rectangular cartesian. In applications, orthogonal coordinate systems
are generally used because the expression for the length d of the line
segment AB joining a pair of points with coordinates A(q,, a2, a3) and
B(b,, by, bs) has the simple form

(1) d = V(b, — a4)? + (by — a)? + (3 — as)”.

This is the familiar formula of Pythagoras. If the coordinate system is
oblique, the formula for the distance d is somewhat more complicated,
We will learn in Sec. 9 that one can pass from an orthogonal system of
coordinates to an oblique system by making a linear transformation of
coordinates. From this fact and from the structure of formula 1.1, it
would follow that the length of the line segment joining the points with
oblique coordinates (x, #2, 73) and (Yr, Yo Ys) 18
3

(1.2) d= Bilt — x)(y; — %;),

where the g,,’s are constants that depend on the coefficients in the above-
mentioned linear transformation of coordinates. We will be concerned
in the sequel with a detailed study of quadratic forms appearing under
of
the radical in formula 1.2 and with their bearing on metric properties
space.

2. The Geometric Concept of a Vector

ate
In the preceding section we recalled the construction of coordin
formula of
systems in the familiar three-dimensional space where the
points. Spaces
Pythagoras is used to measure distances between pairs of
4 LINEAR VECTOR SPACES. MATRICES [CuaP. 1

B b C

Fig. 3.

where it is possible to construct a coordinate system such that the length

of a line segment is given by the formula of Pythagoras are called Euclidean
spaces. In these spaces the notion of displacement is fundamental. Thus,
if a point A is moved to a new position B, the displacement from A to B
—>
can be visualized as directed line segment AB (Fig. 3). If B is displaced
to a new position C, the resultant displacement can be achieved by moving
the point A to the position C. These operations can be denoted sym-
bolically by the equation
—> — —
AB BO= TAG,

In the elementary treatment of vector analysis, directed line segments

are termed vectors, and they are usually denoted by a single letter printed
in boldface type. Thus the foregoing formula can be written

(251) atb= Cc,

—_ — —>
wheresAB =a, BC =b, AC=c
The rule for the composition of vectors indicated in Fig. 3 was first
formulated by Stevinus in 1586 in connection with the experimental
study of laws governing the composition of forces. It is known as the
parallelogram law of addition. The fact that many entities occurring in
physics can be represented by directed line segments, whose law of
composition is symbolized by formula 2.1, is responsible for the usefulness
of vector analysis in applications. We have here an instance of geometriza-
tion of physics which had no less important influence on the evolution
of this subject than the arithmetization of geometry had on the growth
of mathematical analysis.
From the idea of a vector as displacement determined by a pair of
points, we are led to conclude that two vectors are equal if the line seg-
ments representing them are equal in length and their directions parallel.
We shall denote the length of the vector a by the symbol |a|. We will
SEGa2|) [THE GEOMETRIC ‘CONCEPT OF A’ VECTOR 5
assume that the concept of length is independent of the chosen reference
frame, so that the length |a| can be calculated (by Pythagorean formula)
from the coordinates of the initial and terminal points of a.
The negative of the vector a (written —a) is the vector whose length
is the same as that of a but whose direction is opposite. We define the
vector zero (written 0) corresponding to a zero displacement by the
formula
a+ (—a) = 0.

From the geometrical properties of directed line segments we deduce that

(I) at+b=b+a.
(II) (a+b) +ce=a+(b+ 0c).
(II]) If a and b are vectors, there exists a unique vector x such that
a=b+x.

We next define the operation of multiplication of vectors by real numbers.

If « is a real number the symbol «a = aa is a vector whose length is
|x| |a| and whose direction is the same as that of a if « > 0, opposite to
aif
a < 0. If « = 0, then «a = 0.
From this definition and from properties of real numbers we conclude
that
(IV) (a, + a )a = aa + Ha
(V) a(a + b) = aa + ab
(VI) a4 (% a) = (%,%2)a, l-a=a,

for any real numbers «, and 4».

We introduce next the definition of scalar product of two vectors, which
will provide us with a new notation for the length of a vector.
DEFINITION. The scalar product of two vectors a and b, written a-b,
is a real number |a| |b| cos (a, b), where cos (a, b) is the cosine of the angle
between a and b.
Stated in the language of geometry, a - b is equal to the product of the
of
projection of a on b multiplied by the length of b. Thus the length
square root of a-a. We also note
the vector a is given by the positive
that a and b are orthogonal if, and only if, a- b = 0.
can easily
From this definition and the properties of real numbers we
deduce the following theorems.
(VID) a-a=|al?> 0, unless a = 0.
(VIID) a-b=b-a.
(IX) a-(b+c)=a-b+arc.
(X) a(a+b) = (axa-b), where « is a real number.
 VECTOR SPACES. MATRICES [CHaP. 1
6 LINEAR

3. Linear Vector Spaces. Dimensionality of Space

dence of a set of vectors
We formulate next the definition of Jinear depen
impor tant conne ction with the concept
Aja Bot Wenn ay: which will have an
of dimensionality of space.
+++ > An is called linearly
Linear Dependence. A set of n vectors Ay, Ay,
Og) ++ +5 ns Not all of which are zero,
dependent if there exist NUMDErS 4,
such that
048, + &ag to + 4,8, = 0.
linearly independent.
If no such numbers exist, the vectors are said to be
ely, directed
Consider two vectors a and b which are like, or opposit
(Fig. 4). Then there exists a number k ~ 0 such that
(3.1) b='ka.

If we set k = —a/B, we can write this equation as

aa + fb = 0,
since
and hence two collinear (or parallel) vectors are linearly dependent
neither « nor f is zero. We will say that the totality of vectors ka for
an arbitrary real k and a # 0 forms a one-dimensiona l real linear vector
space. The reason for this terminology is clear since every point on the
line can be represented by some position vector ka.
If a and b are two noncollinear vectors, represented by directed line
segments with common origin O (Fig. 5), any vector ¢ lying mm the plane
of a and b can be represented in the form

(3.2) = ma + nb.
Formula 3.2 follows at once from the rule for addition of vectors and
from the definition of multiplication of vectors by scalars. Equation 3.2
can be rewritten in symmetric form to read
aa + fb + ye = 0,
which is the condition for linear dependence of the set of three vectors,
since not all constants in this formula vanish. The formula ma + nb,
where a and b are two linearly independent vectors and m and n are
arbitrary real numbers, defines a two-dimensional real linear vector space.
We see that in a two-dimensional linear vector space a set of three vectors
is always linearly dependent.

O a b

Fig. 4.
SEC. 3] DIMENSIONALITY OF SPACE f|

Fig. 5. Fig. 6.

If we start with three noncoplanar vectors a,b,¢ issuing from the

common origin O (Fig. 6), we can clearly represent every vector d in
the form

(3.3) d = ma + nb +
pe,

from which it follows that among four vectors a,b,c,d there always
exists a nontrivial relation of the form

aa + Pb + ye + dd = 0.

Formula 3.3, for an arbitrary choice of real numbers m, n, p, defines a

three-dimensional real linear vector space. The terminal points of position
vectors d sweep out a three-dimensional space of points if m, n, and p are
allowed to range over the entire set of real numbers. Ina three-dimensional
linear vector space every set of four vectors is linearly dependent. We
will make use of the connection of the number of linearly independent
vectors with the dimensionality of space to formulate the concept of
dimensionality of a linear vector space of n dimensions.
The vectors a, b, and ¢ in (3.3) are called base or coordinate vectors,
and the numbers m, n, and p are the measure numbers or components of
the vector d. Once a set of base vectors is specified, every vector is
determined uniquely by a triplet of measure numbers.
A set of three mutually orthogonal vectors in a three-dimensional
space is obviously linearly independent, and if we choose as our coordinate
vectors three mutually orthogonal vectors aj, a:, a3, each of length 1,
the resulting set of base vectors is said to be orthonormal.
We can visualize a set of orthonormal vectors directed along the axes
of a suitable rectangular cartesian reference frame; in this case every
vector x has the representation

X = Ta, + Lay + TAs,

 8 LINEAR VECTOR SPACES. MATRICES [CHAP.1

where (2,, 2, #3) are called the physical components of x, and the terminal
points of the base vectors a,, (i = 1, 2, 3), have the coordinates

sare (15050):
asin 0.4180):
as (0, 07 1%,

We conclude this section by noting the rules for the addition and
multiplication of vectors when the latter are referred to an orthonormal
system of base vectors a,, (i = 1, 2,3). If we have two vectors x and y
whose components are (a, 2, %3) and (Yj, Ys, Ys), respectively, then the
vector x + y has the components (7, + 41, 22 + Yo, 73 + Ys). If « is a
real number, the components of the vector «x are (a, «22, x3). From
the distributive law of scalar multiplication of vectors it follows at once
that the product of
X = Xa, + Xa, + Xa
and
y Yay + Yoas + Y3a3
is
X°Y = XWYy + TeYo + Xz,

since a,-a; = 0,,, where 6,,; = 1 if i=/, and 6,,;= 0 when i #/j. This
follows from the assumed orthonormal nature of the base vectors a;.
The foregoing formula leads at once to the familiar expression for the
length |x| of the vector x referred to an orthogonal cartesian reference
system. Thus
Xe
x = v7 ey?
4 aw?
= [xis
so that
[x] = Via? + 2? + a5”.
Clearly |x|-> 0, unless z°=="a5 =", = 0:

4. N-Dimensional Spaces

In a variety of circumstances one encounters a correspondence of sets

of objects with ordered sets of numbers where the number of independent
entities exceeds three. For instance, in dealing with the states of gas
determined by the pressure (p), the volume (v), the temperature (7), and
the time (ft), one may wish to coordinate these entities with ordered sets
of four real numbers (a, x2, 73, 74). Here a diagrammatic representation
of the states of gas by points in the three-dimensional physical space is
SEc. 4] N-DIMENSIONAL SPACES 9
clearly impossible. However, the essential idea in the concept of coordi-
nate systems is not a pictorial representation but the one-to-one reciprocal
correspondence of objects with sets of numbers. The notion of distance
between pairs of arbitrary points is, likewise, irrelevant. Indeed, the
idea of distance becomes devoid of geometric sense even in the familiar
representation of the states of gas [the pressure (p) and the volume (v)]
by points in the cartesian pu-plane. It is manifestly absurd to speak
of the distance between two states characterized by ordered pairs of
numbers (p, v).
The utility of analytic treatment of physical problems is so great that
we are naturally led to form the concept of spaces of higher dimensions
by utilizing the idea of one-to-one correspondence between the sets of
numbers and objects. The “objects” here might be of quite diverse sorts.
In certain situations they might be pressures, volumes, and temperatures;
in others they might be the amounts of electrical charge and the complex
potentials produced by the motion of such charge, and so on.
We define® a space (or manifold) of N dimensions as any set of objects
that can be placed in a one-to-one correspondence with the totality of
ordered sets of N (real or complex) numbers 1, %,...,%N such that
|jz, — A,| < k,, (i=. 12. es N),
numbers.
where A,,..., Ay are constants and the ky, ks, ..., Ky are real
The inequalities in this definition specify the range of variation of the
numbers z;. If the numbers 2; are real, the N-dimensional space is real,
and we can write the inequalities in the form
a, SX < A, by < % < be,..-, tity sh.
for the
Some of the equality signs may be omitted, and we may have
range of variables ~,, for example, 0 < 4 < ©.
we use
We denote the space of N dimensions by the symbol Vy, and
the term “‘points” to mean “object s.”
ordered sets
Any particular one-to-one association of the points with the
and the numbers
of numbers (x1, %,+++ ay) is called a coordinate system,
OFe termed the coordin ates of points in the coordinate
aghast
system. :
t of distance
There is no implication in these definitions that the concep
meanin g. If a rule is specifi ed for the
between pairs of points has any
n points, the space Viy is called metric.
measurement of the distance betwee
assume that our spaces are metriz ed.
For the time being we will not
Differential Forms, p. 13. In speaking
2 Compare O. Veblen, Invariants of Quadratic cal corre-
in mind one-to-one recipro
of one-to-one correspondence we always ‘have
spondence.
 10 LINEAR VECTOR SPACES. MATRICES [CHAP.1

A set of equations of the form

X; = (41, Yor synods (i = 1, phe eran Bs
(4.1)

in which the functions x; are single-valued and are such that, in the region
under consideration, they yield N single-valued solutions
Ys = Yl%1, Te, --- > xy),

defines a transformation of coordinates.

We will defer a discussion of the general functional transformation
(4.1) to Chapter 2. In the remainder of this chapter we are concerned
with a detailed study of an important case of linear (or affine) trans-
formations of coordinates
N
Y=) ayes; (c= pee eaN),
j=1

and with the bearing of such transformations on linear vector spaces.

5. Linear Vector Spaces of n Dimensions

A sketch of the rudiments of vector analysis in Sec. 2, based on the
concept of directed displacement, contained a set of ten theorems em-
bodied in the formulas identified by the Roman numerals. These theorems
can be taken as a point of departure in the generalization of the concept
of a vector in the n-dimensional space since the idea of directed displace-
ment and length become devoid of familiar sense whenever n exceeds
three. Accordingly, we will postulate that there exist points in the
n-dimensional space and that

A. Every two points in the real n-dimensional space determine an entity

which we call a vector. We denote this entity by the symbol a.
B. Every two vectors a and b have a sum a + b which obeys laws I, II,
and III stated in Sec. 2.
It follows from the third of these laws that the operation of subtraction
of vectors is unique and that there exists a vector 0 such thata +0=a
for every vector a.
C. For every real number « and vector a there exists a vector «a = ax
obeying laws IV, V, and VI of Sec. 2.

We retain the definition of linear dependence of the set of n vectors,

with respect to the field of real numbers «,,..., «,, and take as our
axiom of dimensionality the assumption that
D. There exist n linearly independent vectors in the n-dimensional space,
but every set of n + 1 vectors is linearly dependent.
Sec. 5} LINEAR VECTOR SPACES OF n DIMENSIONS 11
This axiom implies that every vector x can be represented in the form

(5.1) X = aa, + wa, +°-'+ a@,a,,

where a,, a),...,4, is any set of m linearly independent vectors. We

say that the totality of vectors determined by formula 5.1, where the «,;
are arbitrary real numbers, constitutes a real linear vector space of n
dimensions.
To give a meaning to the concepts of length and orthogonality of
vectors we need a postulate
E. With every two vectors a and b we can associate a number a-b,
called their scalar product, which obeys laws VII, VII, 1X, and X of
DEC. 2.
At this stage we are not concerned with the nature of the formula used
to calculate the number a- b. Suffice to say that the properties embodied
in the laws of scalar multiplication lead to a definite rule for computing
a-b once a coordinate system is introduced for the specification of
coordinates of points determining the vectors.
A vector satisfying postulates A through E is said to be defined in the
n-dimensional Euclidean space E,,.
We shall use the language of Euclidean geometry and will mean by the
length of the vector a the positive square root of the scalar product of
the vector a by itself. Thus the length |a| = Va-a. If ja| = 1 the
vector a is called a unit vector. Two vectors a and b are said to be
orthogonal whenever a+b = 0.
We proceed to demonstrate that every set of m linearly independent
vectors in E, (m <n) can be orthogonalized. This means that from
a given set of m linearly independent vectors X;, X:,.--,Xm We can
construct a set of vectors a;,...,,, Such that a; - a; = 0 whenever i # j.
Moreover, it is possible to choose the vectors a, so that they are unit
vectors.
We assume that the set of vectors {x,}, (@=1,...,m) is
Proof.
linearly independent. Thus the equation

CyXy + CoX&o ie ae Sie CmXm re 0

(5.2)

follows
can be satisfied only by choosing ¢y = ¢, =*"* = Cm = 0. It
that x, ¥ 0, for, if it were zero, the numbers

=¢, =90
¢=1@=G=-°

y dependent,
would satisfy (5.2) and hence the vectors would be linearl
produc t of x, by
which is contrary to our hypothesis. Denote by a, the
12 LINEAR VECTOR SPACES. MATRICES [CuaP.1

the reciprocal of its length so that

Clearly a, - a, = 1, so that a, is a unit vector.

The set of vectors
Ay, Xo, +--+ 5 Xp

is obviously a linearly independent set. Consider next the vector

as =X, — (Xp °a,)ay.

/

The product of this vector by a, vanishes since

Xp * a, — (X2) + a,)a, + a, = 0.

Thus a,’ is orthogonal to a, and a,’/|a,’| = a, is a unit vector.

The set of vectors
PEACE Rail) sip oe

is linearly independent, and we can define the vector aj’ by the formula

As = X3 — (X3+ a)a; — (Xs + ag)ay,

which is orthogonal to both a, and a,. The vector a; = a,'/|a,'| is a unit

vector, and the set .
DRAgwasiK esew ke

is a linearly independent set of vectors.

A repetition of this procedure will yield a set of m linearly independent
unit vectors

(5.3) 930 Ao ae

each of which is expressed in terms of the x; The set of orthogonal

unit vectors (5.3) is called an orthonormal set.
Ifm = n, the set of orthonormal vectors aj, a3, ..., a, is called complete
because every vector x in E,, can be represented in the form

(5.4) X = a8, + aa, +°-*+ a, a,.

By analogy with the three-dimensional case, a complete set of ortho-

normal vectors can be taken as a set of coordinate vectors oriented along
the axes of the n-dimensional orthogonal cartesian reference frame. The
Sec. 5] LINEAR VECTOR SPACES OF n DIMENSIONS 13
terminal points of these vectors then have the coordinates
DOs. aia
i abie reir @
O91 , 0

O50: 0: a
The constants a, %,...,«, in (5.4) are called the components of the
vector x. Multiplying (5.4) scalarly by aj, a,...,a n in turn, and
remembering that? a, +a; = 0;;, we obtain
a,°X = a, a, +X = %, oe a
Thus the vector x can be represented in the form
(5,5) X = (a,-x)a, + (a,-x)a. +°+* + (a, + x)a,.
If we introduce the notation a; +x = 2,, equation 5.5 assumes the form

eS 1 Se tes Sp OO Sea

Using the distributive property of scalar multiplication, we get

(5.6) Rema? aye a,
so that
Ix] =< bape
Var Fae
This is the formula of Pythagoras in E,,.
If y = y,8 + YoMe + °°* + nan, then
KY = yyy + Hayy to + UY ye
This formula has the same structure as the expression for the scalar
product of two vectors in ordinary three-dimensional space of Euclidean
geometry.
We note that in an orthogonal cartesian reference frame a vector x is
uniquely determined by an n-tuple of numbers (21, %,...,%,). This
property is taken by some authors as the definition of a vector in E,,.
For the sum of two vectors x and y, with components
Ke PW asiete ela)>
Y: (Yas Yous + +9 Yn»
8 The symbol 6;;, the Kronecker delta, means
61, it =J;
=0, ifi+;.
14 LINEAR VECTOR SPACES. MATRICES [CHapP. |

we have the formula

x+y: @ + py Ve + Ye ---> 2m + Yn)

and for the product of x by the scalar «,

OKs (OMe OH, tsrepeky arn).

The formula
XY = LY + Leo +--+ 2UYn

serves to define metric properties of vectors in E£,,.

The passage from an orthonormal set of vectors a; to any other set of
base vectors is accomplished by subjecting the elements of the n-tuple to
a suitable linear transformation. In essence the approach to vectors by
way of n-tuples of numbers reduces the study of vectors to the study of
algebraic properties of linear transformations. In this book we prefer
to stress the geometric concepts that underlie the idea of a vector and not
submerge them in a purely algebraic formalism.

6. Complex Linear Vector Spaces

The considerations of Sec. 5 can be easily extended to the field of

complex numbers.
In a complex n-dimensional linear vector space the vector x is determined
by the ordered set of m complex numbers (2, #2,...,2,), the elements
«, of which are the components of x.
We define the sum x + y of two vectors

xX: Gee tes

y: (Y, a
by the rule
PG (2, + Yy, Fe + Yo... + 5.%y + ¥,),

and the product «x by

ON 2 (Oa, Maren :,. ore).

It is customary to define the scalar product x+y by the formula

(6.1) xy = > y,,

where a bar over x; denotes the conjugate of the complex number z,.
SEC. 6] COMPLEX LINEAR VECTOR SPACES 15

We note that

(6.2) {=x = > YX;

i=1
so that

(6.3) xe y = y-xX,

since the conjugate of the sum is the sum of the conjugates and the
conjugate of the product is equal to the product of conjugates.
Formula 6.1 is adopted for the calculation of the scalar product in
order to ensure that

be a real number. It clearly specializes to (5.6) when the numbers 2;

are real.
The vectors x, y are said to be orthogonal if x» y = 0. As regards the
notion of linear independence, we retain the definition given in Sec. 3 with
the understanding that the coefficients «; now belong to the field of com-
plex numbers.
Problems

1. If we start with the definition of a vector x as an n-tuple of n real or complex

numbers (2, %,...,%,), and use for the definition of sum and product the
formulas
xe ¥ Grit Vy,0 02> at Ua)
Rx (hess eh ea)s
n

2 OC >)LYis
i=1
then
(xp yy Ze Xe ZiT y 2,
x-(y +z) =x-y + x-Z,
(kx) -y =K(x-y),
x + (ky) = k(x-y).
2. Prove that, ifa™, a, ..., a‘ is a set of n linearly independent vectors in
l to each
a complex n-dimensional vector space, then the only vector x orthogona
of the vectors a‘) is the zero vector.
3. Prove that a set of mutually orthogonal nonzero vectors is always linearly
independent. — :
4. Let the set of vectors a® in E,: (a, af,... , ae) im 1,25. 5 A, de
linearly dependent, and suppose that r of them, a, a®),.. .,a', r <n, are
linearly independent. Show that every vector x that is orthogon al to this set of r
linearly independent vectors forming the subset of E,, is also orthogonal to the
remaining n — r vectors in the given set.
16 LINEAR VECTOR SPACES. MATRICES [CuaP.1

7. Summation Convention. Review of Determinants

It is clear from the developments in preceding sections that the linear

forms and matrices associated with them enter prominently in the study
of vectors in the n-dimensional manifolds. Since such forms wil] occur
frequently throughout the remainder of this chapter, it is desirable to
introduce a compact abridged notation and to rewrite with its aid certain
familiar results from the theory of determinants. .
From now on we shall adhere to the following summation convention.
If in some expression a certain index occurs twice, we shall mean that this
expression is summed with respect to that index for all admissible values
of the index. Thus the linear form > a;,#, has the index / occurring in it
4 .

i=1
twice; we will omit the summation symbol & and write a,;7; to mean
Ay%, + AX. + ast, + a,x, Of course, the range of admissible values of
the index, | to 4 in this case, must be specified. If the symbol 7 has the
range of values | to 3 andj ranges from | to 4, the expression

(7A) Aj ;X 5, (i a L 2. 3),

o™
5,
I 1, 2, 3, 4),
represents three linear forms

AyX + AyX%y + AyyX3 + Ay4Xq,

(7.2) Ag1X + AgyX%y + AggX3 + AgyXq,
A31X1 + Ag%y + Agg%3 + AgyXy.

In expression 7.1 the index i is the identifying index. It denotes one of the
forms in (7.2), depending on the chosen value of i. The index j, however,
since it occurs twice, is the summation index. The summation (or dummy)
index can be changed at will. Thus (7.1) can be written in the form a,,2,
if k has the same range of values as 7. The summation index is analogous
to a variable of integration in a definite integral, which also can be
changed at will.
Unless a statement to the contrary is made, we will assume that the
summation and the identifying indices have the ranges of values from
1 ton. Thus a,x, will represent a linear form

Q,Xy + Ak, ++°* + a,%,.

Although in the last term of this expression the letter n occurs twice,
it does not represent the sum, since 7 here has a fixed value. In order to
avoid ambiguity, or when we want to suspend the summation convention,
SEC. 7] REVIEW OF DETERMINANTS 17
we may enclose the index in parentheses. Thus we can write the linear
form as
AX TF Agky + + * + A(y)X(n)-
n n

The quadratic form > > a,;,2, will be written a,,v,7;. An expression
t=1j=1
a,;«y; represents a bilinear form containing n? terms, whereas 4;,a,,
represents n® sums of the type

Ay Ay, F AzeQoy +" * 1 Asn Ang,

since each of the identifying, or free, indices i and k can have values
from | to n. We will not trouble to enclose the indices in parentheses
when the context makes it clear (as in the above expression) that such
indices have fixed values. If, however, we wish to discuss a particular
term in this sum we will write @,,;)@,,),.
Frequently, it is convenient to identify the different symbols by using
superscripts rather than subscripts. For instance, we may write the
sequence of terms x’, x, ... , #", where the superscripts are not the powers
of the variable but the identifying indices. The typical term in this
sequence is x’, (i= 1,2,...,m). A linear form in the x’, with the co-
efficients a;, will be written as a,x‘. A bilinear form, with the coefficients
a’), in the variables x, and y; will be written as a’’a,y;.
A determinant
4, U2 Qn
Gx, Ag9 Aen

ani ane Sai ann

whose elements are a,; will be written, as is customary, |q,;|. If the

elements of this determinant are denoted by a,’, where the superscript /
indicates the row and the subscript j the column in which this element
appears, we will write the determinant as |a;*|. Thus
1
ay" a, ay
2 2
[a;|= a, 42 ay,
Ga) oie ie) xe gels. Pyros) usice ©

For the multiplication of two determinants |a;*| and |b,*| we have the
familiar rule: by
la,'|-[5;"| = lel,

where c,’ = a,‘b;*. If we deal with determinants |a,,| and |b,;|, then the
and
element c,, in the ith row and the jth column of the product of |a,;|
|b;;| isc, = AinDy 3:
18 LINEAR VECTOR SPACES. MATRICES [CuaP. 1

The cofactor of the element a,’ in |a;'| is denoted by 4,’. If we write

the Kronecker delta as 6;', where

éi=1,
I
ifi=y,
=0, ifi¥),
then for the expansion of |a,‘| in terms of cofactors we have the following
formulas:

(7.3) GA oo
(7.4) a;'A* — ao;*,

where a = |a;‘|. These formulas include the familiar simple Laplace

developments of |a;‘|. The first of these then represents the expansion
in terms of the elements of the ith row; the second, in terms of the elements
of the jth column of |a;‘|.
If the elements of the determinant a are denoted by a;;, we shall write
the cofactor of a;; as A,;. Simple Laplace developments corresponding
to (7.3) and (7.4) assume the forms

A.Aiy; = and A: Ain) = 4.

We can derive Cramer’s rule for the solution of the system of 7 linear
equations

(7.5) ao =o. lif sly seen

in n unknowns x‘, where |a,'| # 0, as follows: Multiply both sides of

equations in (7.5) by A,*, and sum with respect to i. This yields

By (7.4) this reduces to

adfa! = btAS.
or
ax* = biAS:

Thus
tak

Frequently, the cofactor of the element a;; in |a,;| is denoted by 4%,

so that the Laplace developments (7.3) and (7.4) assume the forms

ai;A = a,
A;;)A™) = a,
Sec. 8] LINEAR TRANSFORMATIONS AND MATRICES 19
To gain familiarity with this notation, the reader is advised to derive
Cramer’s rule when the system of linear equations is written in the form
a,;x? =-b;. He will also prove that, if a;‘b,? = 6,', then |a,'| = 1/[b,'|.
We will return to the subject of determinants in Sec. 41, where a
different notation permits us to eliminate references to rows and columns
of the determinant and enables us to write it in terms of its elements,
without reference to cofactors.
Problems
1. Write out in full the following expressions:

(a) 6;'a?. (b) 6 ,j0'x?. (c) aijbjx = Six. (d) A;54H.

a ,
of;
(e) meee (f)0;*. (g) a = yi b!, (h) aijayv'y™.

é ay* dy*
© 26 —3253,)° (j) aiqjye. (k) 5;;0™.

The symbols 6;', 6;;, and 6” all denote the Kronecker deltas.
2. Verify that (7.6) is the solution of (7.5).

8. Linear Transformations and Matrices

A set of n relations of the form

(8.1) L, = Ay; ‘Cbg ia eda ey a)

where the q;,’s are constants, is called a linear homogeneous transformation

of the set of variables x, into a set x,. We shall suppose that the trans-
formation 8.1 is nonsingular, so that the set of n linear equations 8.1
can be solved for the x, in terms of the z,’. This implies that the deter-
minant |a,,| of the coefficients of x,’s is different from zero.
The solution of (8.1) for the «’s yields
Aji ,
x;
a
where 4,, is the cofactor of the element @;; in |a,;| = a.
The set of equations 8.1 can be interpreted in two essentially different
ways:
vector x:
(a) The quantities x; may be regarded as components of a
of another vector
(a, %,.--,5%,), and the numbers a,/ as components
to a referenc e
x’: (ay, X',...,%,), Where both x and x’ are referred
think of equatio ns
frame with the system of base factors a;; 1n this case we
vector x’.
8.1 as representing a transformation of the vector x into another
20 LINEAR VECTOR SPACES. MATRICES [CHaP.|

(b) The two sets of numbers (a1, %,..., ,) and (2,’, Bike Sse ean
be regarded as components of the same vector x when x is referred to two
different sets of cartesian reference frames determined by the base vectors
41, Mg, 4.58, and a,’ a, 400.9185 in thisvevent equations 61 give a
transformation of coordinate axes.

Before proceeding to a specific discussion of these two interpretations

of the set of equations 8.1, it is necessary to review the operations with
matrices.
An array of mn numbers, arranged in m rows and n columns, is called
an m Xn matrix. We denote the matrix formed from the elements a;;
(or a;’) by

1 1 1
44, 12 Ain ay a, a,
2
(a,;) = Ao, Ag9 Gan or (aj)= a, 4a: Ane

m™m mm ™m
ami Ame Amn ay as a,

We shall also write the symbol A for the matrix (a,;). We shall say that the
matrix A = (a;,;) is equal to the matrix B = (6,,) if, and only if, a,; = 5,;
for each i and j. That is, if 4 = B the elements in the corresponding
rows and columns of the matrices must be equal.
By the sum 4 + B of two matrices A = (a,;) and B = (b,,;) of the same
type, that is, containing the same number of rows and columns, we mean
the matrix
A+ B= (a, + b,).
If we have an m X n matrix A and ann X p matrix B, we can define
the product of matrices A and B, written AB, by the formula

(8.3) AB = (4,;b;,).

Thus the product AB is an m x p matrix; we can multiply two matrices

only if the number of columns in the first factor is equal to the number
of rows in the second.
For the most part we shall deal with square matrices, that is, matrices
containing an equal number of rows and columns.
A matrix all of whose elements are zero is called a zero matrix. It is
denoted by the symbol O.
We note two peculiaritiesofmatrix multiplication. From the definition
8.3 it follows that, if A and B are two n X n matrices, then AB is not
necessarily equal to BA.
Sec. 8] LINEAR TRANSFORMATIONS AND MATRICES 21
For example, if
Oe —1 0
A= and B= :
LU Oa |
then s
Oe 0 —-1l
AB = - whereas BA = :
—1 0 1 0

Thus the product of matrices, in general, is not commutative. However,

if we have two matrices of order n, which contain zero elements everywhere
except possibly along the diagonal, then they are commutative, and obey
the simple law of multiplication.

Pee .23 20 ree 1) ah eae Aiba 0 5 OMe 0

Cae ees) Oe aa 0 0 / Wee or 0

‘ah Pacer = Oe Oe ier 0 0 cua) helos

Such matrices are called diagonal matrices. The diagonal matrices will
be found to be of considerable importance in what follows.
A particular diagonal matrix

L4,0 0
ase | 0
[=

OO 1

is called the identity matrix. We note that, if A is any matrix, then

AI =1A=A.
when
We also observe that the product of two matrices may vanish
neither of the matrices is a zero matrix.

le bad 0.00 00 0

0 0} and B= 0 0 O|, then AB=|0 O O}.

Thus, if 4=|0
0 0 0 1980, 40 0 0 0
22 LINEAR VECTOR SPACES. MATRICES [CHaP. |
However, the determinant |AB| of the product of two square matrices
is precisely equal to the product of the determinants |A| and |B| of the
matrices A and B. This follows at once from the observation that the
law of formation of the element in the ith row and kth column of
the product of two determinants is identical with the corresponding rule
for the product of two matrices. We call ann x n matrix whose determi-
nant is zero a singular matrix.
Finally, we define the multiplication of the matrix A = (a,,;) by the
number k, written kA, as the matrix each of whose elements is multiplied
by k. Thus kA = (ka,,).
As an exercise the reader will verify the following theorems, which
follow directly from the definitions given previously.
(1) A+B=B+A.
(II) (AB) | C= A + (8 Cy
(II) (A + B)C = AC + BC.
(IV) C(A + B)=CA+4 CB.
The notation just developed permits us to write the system of equations
8.1 in the form of a vector equation
(8.4) x’ = Ax,
where A = (a;;) and where we agree to interpret x either as a column
matrix
vy
* 2 9:Dag) eee
or a square matrix Ps ey

z, 0 0 0
a n.

The inverse transformation 8.2 can be written

(8.5) x = A-ly’,
where

An An Ana
|A| |A| |A|
Aj. Ags Ans

)8. A ph [Al ee1A

| ae
|A| |>
Ary Agn Aix
Sec. 8] LINEAR TRANSFORMATIONS AND MATRICES 23

and the 4,,’s are the cofactors of the elements a,,; in the determinant |A].
The matrix A-! is called the inverse of the matrix A, and it is defined
for any nonsingular matrix A. From definition 8.6 it follows that the
matrices A and A~! are related by the formulas ‘

AAt=1, AtA=],

where J is the identity matrix. For, AA! = (ayA,/|A|) and a,.4j,. =

6,;|A|. The identity matrix / corresponds to an identity transformation
a,/ = x,; this transformation when written in the matrix form (8.4)
appears as x’ = /x, or
x =x

We call the matrix

ay, Ag Ant

= ay2 Ag9 an2 5

A

Qi, Aan Ann

44, Arp Ain

A= |421 4o2 Gen |,

Ani ane Ann

the transpose of A.
and multi-
Using the definition of transpose and the laws of addition
plication of matrices it is easy to show that

(V) CeO Bieta ta

(VI) (KAY = kA’.
(VII) (AB) ="B A> (Note order.)

If A is nonsingular, then the matric equations

AX=I and XA=I

y verified by multi-
have unique solutions X¥ = A™, as can be immediatel
plying them by 4~? on both sides and noting that
A1A = AAT =I.
24 LINEAR VECTOR SPACES. MATRICES [CHaP. 1

If we take 4-14 = AA and form the transpose, we get

AAD) =A) An

Multiplying by (A’)-? on the left, we obtain

(A')-1A'(A74)’ — CAA a:

(A) = (4’) (A474)

=(A)o
Thus
(fy = (4.
We can also readily show that
(AB) = BA: (Note order.)

If we have two successive linear transformations

L, = A,X; and x," = b,,%;,, Gj =1;-.+.7),

/ . .

the direct transformation from the variables x; to the variables 2,” is

£," = Db,Ankp, (2,J,Klee os

this is called the product transformation. Writing these transformations
in matrix notation yields
xo— Ax and x" = Bx,
so that
x” = BAx.

Since the product BA, in general, is not equal to AB, we see that the order
in which the transformations are performed is not immaterial.
It should be observed that the matrix A in the equation 2’ = Az can
be interpreted as an operator which converts a vector x into another
vector x’. Because of the properties
A(kx) = kAx
and
A(x + y) = Ax + Ay,
where k is any scalar, A is frequently called a linear vector operator or
linear vector function. It can be viewed as an apparatus for the manu-
facture of a new vector from a given vector. We shall expound these
points in greater detail by considering a number of examples of the uses
of matrices in several situations familiar from analytic geometry and
elementary vector analysis.
Sec. 9] TRANSFORMATIONS IN EUCLIDEAN 3-SPACE 25

9, Linear Transformations in Euclidean 3-Space

Let us refer our Euclidean 3-space (E;) to a system of coordinates with

base vectors a‘), a), a‘), linearly independent but not necessarily orthog-
onal. Then any vector x can be represented in the form
(9.1) x = a,a"), (j = 1, 2, 3),
where the x; are appropriate real measure numbers. If we introduce
a real linear transformation
(9.2) a =a,e, with |a,| #0, j= 1,2, 3),
Or

(9.3) x, = Ax,

we can interpret the resulting vector x’ as a deformed vector produced

by the deformation of space which is characterized by the operator A.
In general, the length of the vector x’ will be different from that of x,
and its orientation relative to our fixed reference frame will differ from
the orientation of the vector x.
Obviously there are infinitely many reference frames that may be
imbedded in our space, and in each frame the vector x is characterized
uniquely by a triplet of numbers. Let us inquire: What is the form of
the transformation giving the same deformation of space as that charac-
terized by the matrix A, when the vector x is referred to a new frame of
old
reference in which the base vectors a, a), a‘) are related to the
base vectors a), a), a) by the formulas

(9.4) =)
al ba?
denote
We shall suppose that the matrix (6,,) = B is nonsingular and
of x relative to the new system by (&,, &, &3), so that
the components
(9.5) x = Gal”.
a‘) in
If we insert in (9.5) the expressions 9.4 for the base vectors
terms of a‘), we obtain

(9.6) x = &,b,a”.
tion between
A comparison of this equation with (9.1) yields the connec
the components &; and 2,, namely,
a, = bi56;.
(9.7)
of base vectors
We note that the matrix B in the transformation 9.4
9.7 of components
a) differs from the matrix B’ in the transformation
26 LINEAR VECTOR SPACES. MATRICES [CuapP.|

of the vector x in that the rows and columns in these matrices are inter-
changed. Thus the matrix B’ is the transpose of the matrix B. We write
(9.7) in the form
(9.8) ena bk
The solution of (9.8) for & is given by
(9.9) EB = (B’)"x.
To simplify writing we denote (B’)! by C, so that (9.9) becomes
(9.10) E = Cx,
where
(9.11) C= (By.
Formula 9.10 permits us to calculate the components of the vector x
when it is referred to a new system of base vectors a"), determined by
(9.4). Consequently the components &,’, 4’, 5’ of x’, relative to the
reference frame with base vectors a‘), are given by
(9.12) E’ = Cx’,
and the question of the expression (in the new frame) for the deformation
of space characterized by (9.3) amounts to finding the relation connecting
the components &,, &, &; with §&,’, &’, €;’. The substitution from (9.3)
in (9.12) gives
E = CAx:
and, since by (9.10)
x= CF,
we get the desired relation
(9.13) &’ = CACHE,
The transformation determined by the matrix S = CAC~? is called similar
to the transformation produced by A because formulas 9.13 and 9.3
characterize the same deformation of space relative to two different
reference frames.
If we recall the definition (9.11), we can write (9.13) in the form
(9.14) B= (BY) 1ABE,
which brings into explicit evidence the matrices A and B characterizing,
respectively, the deformation of space and the transformation of base
vectors. We note that the determinants of all similar transformations
are equal. An important special case of the transformation 9.2, corre-
sponding to the rotation of the vector x to a new position, is discussed
in the next section.
Sec. 10] ORTHOGONAL TRANSFORMATION IN £; os|

10. Orthogonal Transformation in EF,

Let us suppose that the base vectors a, a), a) in Sec. 9 are orthogonal
unit vectors, so that the measure numbers «; in (9.1) are the, physical
components of x. Then the square of the length of the vector x is given
by the formula
|x|? = a ,, (i =e 1275):

Let us inquire about restrictions that must be imposed on the matrix

A in (9.3) if the length of x is to be unchanged by the transformation 9.2.
This restriction demands that
Ge te at Pt Pe
(10.1) ts

Substituting in (10.1) from (9.2), we obtain

— UX, (i,tf k = iF 2 3)
(4;;%;)(AixX)

or

(10.2) jpXjy, = OX Upp

Aj

since
Oj4XjXy = UX, = UX;
equations
Equating the coefficients of like products in (10.2), we obtain six
Ay? + Gy;” + ag = 1,
Ayo? + doe” + Ago” = 1,
Ays? + eg” + 433” = |,
y9Ay3 + A223 1 432433 = 0,
43411 + 23421 + 433431 = 0,
AyAya + Ao1429 + 431432 = 0,
or

(10.3) ijn = On:

sis that the length of x
These equations are consequences of the hypothe value
the
remains invariant. The determinant of the matrix in (10.3) has

(10.4) |a,;4i| = 1.
unchanged when its rows
Since the value of the determinant |a,;| is
see from the rule for multiplication of
and columns are interchanged, we
determinants (Sec. 7) that
\a;;4%| = |a,;| ° |a;;| = |A|?.
28 LINEAR VECTOR SPACES. MATRICES [CHAP.1

Thus (10.4) yields the result that the square of the determinant |a,,| in
(9.2) has the value 1 whenever the length of the vector is unchanged by
the transformation. We conclude that |A| = +1. The case when
|A| = +1 corresponds to the transformation of rotation of space relative
to fixed axes. The circumstance when |A| = —1 corresponds to the
transformation of reflection (say, 7,’ = —2,, %2' = —%,%3) = —3) or a
reflection followed by a rotation.
A linear transformation

(10.5) Ly = Azz;
in which a;,,a,;, = 6,, is called an orthogonal transformation. It is called
the transformation of rotation when |a;;| = +1. If we denote by A’
the transpose of A in (10.5), we can write the orthogonality conditions
(10.3) in the form
AA =I.
Multiplying this equation on the right by 4, we get
(10.6) A’ = A,
so that in an orthogonal transformation the inverse matrix A~ is equal to
the transpose A’ of A when the base vectors are orthonormal.
It follows that, if we write equations 10.5 in the form
XestAy:
then
xX = Aix.
and by virtue of (10.6)
x = A’x’
or
(10.7) XL, = a;;X;'.

11. Linear Transformations in n-Dimensional Euclidean Spaces

Our discussion of linear transformations in Euclidean 3-space can be

immediately extended to n-dimensional manifolds E,, referred to a coor-
dinate system such that the length of the vector x is determined from
formula 5.6.
We introduce n-orthonormal vectors,

ag di, 0, 0, Vo 0),

a): (0,1,0,...,0),
Sec. 11] LINEAR TRANSFORMATIONS Dose)
and represent any vector x: (#, 2% ,...,2,) in the form (cf. equation 9.1)
(11.1) x = za, (i at ot rt)
A linear transformation of components, corresponding to equation
9.2, is

(1122) L. = G03, (of ed ae 7h

We can write it in matrix notation as

(11:3) Ss be

where A = (4;;).
We suppose that |A| ¥ 0, and denote the solution of (11.3) by
x = Aly’,
where
An = (A;,) ;

|A|
The A,,’s denote the cofactors of the elements a,; in |A|.
Just as was done in the three-dimensional case, we can show that the
product of transformations x’ = Ax and x” = Bx’ is x” = BAx. We
can still use the suggestive language of geometry and speak of the set of
equations 11.3 as representing the deformation of space E,, and consider
that the transformation of the form

(11.4) x’ = CACx
by the
represents the same deformation of space as that characterized
similar.
matrix A in (11.3). The matrices A and CA C- are still termed
rmation
By analogy with the three-dimensional case, a real linear transfo
, x,) invaria nt is
that leaves the length of every real vector x: (%,...
10 it is obvious that the
called orthogonal. From computations of Sec.
(11.2) satisfy the relation s
coefficients a,, in an orthogonal transformation
(11:5) jin = Ojn9
transformation is related
and that the matrix A = (a,,) of an orthogonal
condition (11.5) is both
to its inverse by the formula A’ = A, The
be orthogonal. Since
necessary and sufficient for a transformation to
ormation is equal to
the transpose of the matrix of an orthogonal transf
its inverse, we deduce that 4;,4,; = 9jx-
ions (11.5) is called
Any matrix satisfying the orthogonality condit
such a matrix has the
orthogonal. The square of the determinant of
value 1.
30 LINEAR VECTOR SPACES. MATRICES [CuaP.1

As in the three-dimensional case we introduce a matrix B = (5,;)

defining a transformation of the base vectors a‘” into a new set of base
vectors a” in accordance with the formula

(11.6) ote ha), Gefen 2);

then C = (8).
If the vectors a are orthonormal and the matrix B orthogonal, the
new set of vectors a” will obviously be orthonormal. Whenever |5;;| = 1,
we shall speak of (11.6) as representing a rotation of base vectors in E,,.
We now raise the question: Is it possible to find a matrix C such that
the matrix CAC has the diagonal form

0 0
an Bidecd be 0 |2

0 0 os
This means that relative to a suitable reference frame the deformatioa
of space, characterized by (11.2), assumes the form

(Liz) oa = A161, E,! ae Ao& 2, ae | Sa a Aen

the &,”s being the components of x’ and the €,’s of x in the new coordinate
system.
In the language of transformations in £3, equations 11.7 state that for
a suitably chosen reference frame the linear deformation of space is
equivalent to simple extensions or contractions along the coordinate axes.
Clearly the possibility of such reduction depends on the nature of
coefficients a,,; in (11.2).
A detailed discussion of the problem of reduction of matrices to various
canonical forms is involved. In the following sections we treat only
those cases that occur most frequently in applications, referring the
reader for an exhaustive treatment to standard treatises on higher algebra.

12. Reduction of Matrices to the Diagonal Form

We return to the problem posed in Sec. 11, concerning the possibility

of finding a nonsingular matrix C such that an arbitrary matrix A can
be reduced to the diagonal form A by means of the similitude trans-
formation CAC. From the point of view of linear transformation of
SEc. 12] REDUCTION OF MATRICES 31

space, this problem is equivalent to determining the base system a,

(i= 1,...,n), relative to which the transformation

XL, = a,x;
assumes the form [see (11.7)]

Ey =A4, Ey’ = Agee, = 9 E, = AEn

We write C-! = S, and seek a solution of the matric equation

(12.1) ipo, Pelee tae

or
(12.2) AS = SA,
where A = (a,;) and
A, 0 0
ares ‘ay, 5 0

0 O A

The matric equation 12.2 is equivalent to the system of linear equations

(12.3) a,Sjn = Sade (no sum on k), (iy ek eee, )

where
Siy Siz Sin

See Sai Son Son

Shuster kaion vw O38

obtain a system of n
If in (12.3) we set i = 1,2,...,m, and fix k, we
appearing in the kth
equations containing the elements (Sips Sons ++ > Sng)
can be viewed as com-
column of S. The elements (5), 52x; - - . 5 Snp)
ination of the matrix S is
ponents of the vector s‘), so that the determ
(k = 1,...,7”), whose com-
equivalent to finding a set of n vectors s®,
ponents satisfy equations 12.3. Accordingly we write equation 12.3 in
the form

(12.4) As®) = s),, (no sum on k),

and note that (12.3) is equivalent to

(12.5) (a;; — OiAdSix = 9 (k not summed).

32 LINEAR VECTOR SPACES. MATRICES [CHAP.1

If this system of linear homogeneous equations is to have a nontrivial

solution for the s,,, then /, must be a root of the determinantal equation

ee 6,54 = 0,

which, when written out in full, is

ay,—A ay ey Ay

a Api fh ee” a>

(12.6) 21 " |=0,
ani ane ann mar us

This nth order algebraic equation in A has n roots, which are known as
the characteristic values* of the matrix A. If these n roots are distinct,
we can readily show that the system of equations 12.4 yields a set of n
linearly independent vectors s“), and hence a nonsingular matrix S, as
required by (12.1), exists. If the roots are not distinct, it may not be
possible to determine the desired matrix S.
‘We consider the case when the roots are distinct, and denote them
by A,, 4,,...,4,. If we set A, for A, in (12.5) we obtain a system of n
homogeneous equations. This system will have a nontrivial solution
$1, So1,+++55S_,- Setting A, = A, in (12.5) we get the system yielding a
solution S19, 5g, ..., Sng. This gives the second column of S. Proceeding
in this fashion we can determine the remaining columns and hence the
matrix S, which satisfies the equation 12.2. To show that the trans-
formation 12.1 is possible, we must demonstrate that the vectors s\) so
calculated are linearly independent, so that S possesses an inverse S~1.
We shall prove this by supposing that the matrix S is singular and reaching
a contradiction.
If |S| = 0, the vectors s“) appearing in the columns of S are linearly
dependent, and hence there exists a set of constants c,, not all zero, such
that
cs) + cs@ + -+-+c,5™ = 0.

In this expression some c’s may be zero. We may suppose, without

loss of generality, that the first r c’s do not vanish, so that we have the
relation
(12.7) cs + cs +---+ cs = 0, ne
in which none of the c’s (or s‘’s) vanishes.

* They are also called eigenvalues and the corresponding vectors s‘*) are termed
characteristic vectors or eigenvectors.
Sec. 12] REDUCTION OF MATRICES 33
From (12.4) we deduce the relations

w,Ce BI Me A( As) = As), = s\1,2,

A[A(As™)] = 8,3, oe (Ayashi s

If we multiply (12.7) by A successively r — 1 times and take account

of the chain of relations just written, we obtain a system of equations

es? + cs +-:> 4+ ¢s% = 0,

CSA, + c8@A, +°°- + ¢8MA, = 0,

3
+ cys@A5-b= + +++ + ¢8MA-* = 0.
1
csMAL-}
. —1

Since none of the c’s or ss vanishes, this system can be satisfied

only if
1 1 1

A= tei te Ca

a re aoe 3 He?

The determinant A, however, is a Vandermondian,°® and its value is

known to be

A= (A, ma AAs suAg) ioe * (A, — A,)

(As ae As) ae (A, ee. Ag)

Se) wl esis! 8 10.8. (6) 0

Cent Sah ees

the
This is never zero if the /’s are distinct. Thus the assumption that
be
matrix S is singular is incorrect, and hence the matrix can always
of the
reduced to the diagonal form whenever the characteristic values
matrix A are distinct.
, the reduction
If the roots of the equation |A — A/| = 0 are not distinct
impossible.
ofA to the diagonal form by the transformation S-1AS may be
are discussed
In this event there are other canonical representations which
however , the
in books on higher algebra.° In several special cases,
Algebra, Ginn and Co., Boston
5 See L. J. Paige and J. D. Swift, Elements of Linear
(1961).
John Wiley and Sons, New York,
6 See F. D. Murnaghan, Applied Mathematics,
A Survey of Modern Algebra, The Macmillan Co.,
(1948); G. Birkhoff and S. MacLane,
New York (1941).
34 LINEAR VECTOR SPACES. MATRICES [CHAP.1

reduction of the matrix A to the diagonal form, even when the charac-
teristic equation |A — A/| = 0 has multiple roots, can be achieved. We
turn to the consideration of these cases in the following sections.

13. Real Symmetric Matrices and Quadratic Forms

Let us assume that the matrix A = (q;,;) in a linear transformation

(13:8) Te — 49° 792 Ui ja Wlge: ss

is real and symmetric, so that a;; = a;; (or A’ = A) for all values of 7
and j. We will show that the matrix A can be reduced to the diagonal
form by the transformation S-14S. Moreover, S can be an orthogonal
matrix.
Linear transformations with real symmetric matrices occur commonly
in the study of deformations taking place in elastic media. Real symmetric
matrices also enter prominently in the study of real quadratic forms
(13.2) OCs, Sane 3) = wea (Gj = A. 25.7)
which arise in many problems in dynamics and geometry. We can assume
without loss of generality that the coefficients a,; in (13.2) are symmetric,
since (13.2) can always be written

Q(x, Vo,--+ +5 x,,) = Bae U;X j,

in which the coefficients are obviously symmetric. In déaling with

quadratic forms we shall always suppose that they have been symmetrized.
It will follow from the developments in this section that the problems
of reduction of the set of linear forms (13.1) to the form

gy = Aig, E,’ —— Agés, Ss) Pas = Pa ae

and of the quadratic form (13.2) to the form

(13.3) QO=16°+ 462 +°-:4+4,6,?
are mathematically identical.
We note first several properties of quadratic forms (13.2). If we intro-
duce a linear transformation
(13.4) %, = Sbp or x = SE,
the form Q in (13.2) becomes

OQ = 4,i(S.F)(552F,)
= S755
1F461
SEc. 13] REAL SYMMETRIC MATRICES 35
We denote the coefficients of &,&, by c;,,, so that

0 = cubs
RISkS
where

(13.5) Cer = VjSix551- :

Since a;; = a;,, and i andj in (13.5) are the summation indices, an inter-
change of k and / does not alter the value of (13.5). Thus c,, = cj, and
hence the matrix C = (c,;) is symmetric. We thus have the result that
the symmetry of quadratic form (13.2) is not destroyed by subjecting the
variables x, to a linear transformation.
Let us write (13.5) in the form
Cyy = Six(4ij551)5
and observe that a;,;s;, is an element in the ith row and /th column of
the matrix

Thus
(13.6), Cer = Sibir
can be regarded as the element in the Ath row and the /th column of the
matrix S’B, and
(i237) C= S’AS.
We have established a
TueorEM. If the variables x, in the quadratic form Q = 4,;x,x;, with
a matrix A, are subjected to a linear transformation x, = 8;;§;, with a
matrix S, the resulting quadratic form has the matrix S'AS.
We note, as a corollary of this theorem, that the determinant of the
resulting quadratic form has the value |A| |S|?.
If the transformation 13.4 is orthogonal, then S’ = S~! and we can
write (13.7) as
C = S-1AS.
It follows from this result that the determination of an orthogonal trans-
formation which reduces the form 13.2 to the sum of the squares 13.3
reduces to the solution of the matrix equation
(13.8) SOAS = A.
This is precisely the problem we considered in Sec. 12. It follows from
the discussion of that section that the system of homogeneous equations
(13.9) AijSin = Sirdw (no sum on k),

obtained from
AS = SA
 VECTOR SPACES. MATRICES [CHaP. |
36 LINEAR
solution for the vectors slay
(see equations 12.3) will have a nontrivial
4’s in (13.9) satisfy the equation
(Sit Sea ree) IG and only if, the
la,; — ;;4| = 9, or
(13.10) |A — Al| = 0.
ion 13.10, in general,
If the matrix A is arbitrary, the characteristic equat
discussed
has complex roots; and if these roots are distinct, the methods
ly independent vectors
in Sec. 12 permit us to calculate a set of n linear
er, the matrix S
s composing the matrix S. In the present case, howev
roots of the charac-
has to be orthogonal and hence real. Now, if the
from equation
teristic equation 13.10 are real, then it follows at once
be taken to be real
13.9 that the solutions s™: (Sy, Sex + - > Snx) Can
since the a,,’s are real. We prove a
of the
TuHeoreM. Jf the matrix A is real and symmetric, then the roots
characteristic equation |A — AI| = 0 are all real.
The system of equations 13.9 can be written compactly as

(get) As™ =s], (no sum on k).

We can regard As“) as a vector with components

AiyS1y, + Ai2Sox + °° * 1 VinSnx (i= 1,2,..., 7).

or
Let A, be a root of (13.10), real or complex, and s“) a vector, real
13.11. We multiply (13.11) scalarly by
complex, satisfying the system
s® and get :

(13.12) si*)- Ag = |g]? A,.

Now, the left-hand member in this product (recall definition 6.1)

s™ - As = A,,F4Sixs (no sum on k)

is real if a,, = a,;,. To prove this, note that the conjugate of 4,;5;.5;, is
equal to the original expression,
A SxSix = GSuSin = ViSaSin-

Since the left-hand member of (13.12) is real, and |s“)|? is real, it follows
that J, is real. This completes the proof of the theorem.
We prove next that, if 4; and 4; are two distinct roots of (13.10), then
the vectors s‘ and s“), corresponding to these roots, are orthogonal.
Since s‘ and s‘ satisfy (13.11), we have the identities

As = s2,;, (no sum),

Sec. 13] REAL SYMMETRIC MATRICES 37
where all the vectors involved are real. If we multiply the first of these
scalarly by s”) on the right and the second by s™) on the left and subtract,
we get
As e s) — sg ° As”) = (A; = A,)s . si), ns

and the left-hand member vanishes since s‘” - As‘) = As‘ - s) because
of symmetry of 4. This establishes the orthogonality of s‘) and s"),
whenever the roots 4; and A; are unequal. Since equation 13.11 is homo-
geneous, we can multiply it by a suitable constant making the length of
s) equal to 1. We shall suppose that this has been done.
We recall that a set of orthogonal vectors is necessarily linearly inde-
pendent. Hence, if all roots of |A — AJ| = 0 are distinct, the vectors s®)
will be orthonormal, and, accordingly, the matrix S, accomplishing the
transformation S-1AS = A, will be orthogonal.
It remains to consider the case of reduction of real quadratic forms
13.2 to the diagonal form 13.3 when the equation

(13.10) |A — Al| =0
has multiple roots. The demonstration that the reduction is possible in
this case hinges on one important property of all similar matrices, namely:
the characteristic roots of all similar matrices are equal. The proof of this
is easy. We replace A in the left-hand member of (13.10) by some similar
matrix S-!AS and obtain the polynomial in 4,
ISAS — Al| = |S(A — ADS!
== |S-4|-|A — Al |S|
See
and
It follows that the characteristic equations associated with S'AS
A are identical, and hence their roots are equal.
A, be
Now let us suppose that (13.10) has multiple roots. Let A =
of (13.11) s™:
some root of (13.10), and let us determine the solution
= 1.
(eRe SD corresponding to A = Ay, which is such that s™ -s@)
We can adjoin
This can be done whether A, is a multiple root or not.
a complete
to the vector s a set of n — | orthonormal vectors forming
vectors can be
system of vectors in our n-dimensional manifold. These
of orthono rmal
used as a basis for our space instead of the original set
and we can pass from the reference frame
base vectors a™,...,a'”),
transformation.
determined by the a‘”s to the new frame by an orthogonal
referred to the new
Hence the matrix of the quadratic form 13.2, when
the form A, = S,AS,, where S, is orthogonal.
frame, will assume
38 LINEAR VECTOR SPACES. MATRICES [CHaP.1
Moreover,

(13.13) |A, — Al| = 0

has the same characteristic roots as (13.10). The equation [cf. (13.11)]

(13.14) As = sh
for A = A, has the solutions s™: (1,0,0,...,0), since we chose it to
be a unit vector, and s® is one of the base vectors of the new reference
frame. If we insert this solution in (13.14), we get an identity

1 hy
0 0

A, 7 = °

0 0

from which it follows that the matrix A, has the following elements:

(315) ay AP = ae Sa = 0.
The original matrix A is symmetric, and, since orthogonal transforma-
tions do not destroy the symmetry, the matrix A, is also symmetric.”
Thus
A,’ = A,,

and we can write instead of (13.15)

(li) Fa ee ‘aS iN pes ie eee ee AD cl 1)
ay =A, a = ayy = as = ats = = at = ef = 0,
so that
Asli 0

Ay faire
0 af) +++ ale
gem.
(08 14: 2

Ci ee ee
0 ane ann

Thus the quadratic form 13.2, when referred to our new frame, has the
structure
O71, 8 PAREEL ey = Dae n):
We succeeded in separating one square and reduced the problem to
a consideration of the form a{},é; in n — 1 variables. We can apply
"For Ay’ = (S,1AS,)’ = Sy'A(S,>)’ = S,1AS,, since S-1 = S’ for orthogonal
matrices.
 REAL SYMMETRIC MATRICES 39
Sec. 13]
consider
similar reasoning to the (n — 1) x (m — 1) matrix 4, = (a\?) and
onal subspace
the form aWPE.é, Gj = 2,3,.-. _n), in the n — | dimensi
s™. In E,_4, we
E,_; of E,, determined by the base vectors other than
can calculate a unit vector s®) satisfying the equation
A,s = SA, -
by an orthog-
corresponding to A = ,, and construct a new base system
will yield a matrix
onal transformation in which s® is a base vector. This

(2) (2)
As — 0 Ass agp

(2) (2)
0 ang ann

and hence Q of the form

O = A,F? + Ag? + aD €;, (isp 3, ne 5s
original quadratic form
The continuation of this process will reduce the
13.2 to the form
Q= Ae? + AE Sale gine Sie& een
by an orthogonal trans-
Since each successive reduction is performed
formations is equivalent to
formation, the product of orthogonal trans
resulting diagonal matrix A,
a single orthogonal transformation S. The
Ay (0) aed 0)

0 he Sees 0

SHAS = A= he ; ids

0) @) Aviles Ae

equal to the multiplicity of the roots

contains the number of like roots 4 to A, the characteristic
S-1AS is similar
in |A — AI| = 0. Since the matrix with those of
AED ae Bee of |A — Al| =0, are identical
|A — Al| = 0.
characteristic vectors s“ associated
The directions determined by the ix A.
d the principal directions of the matr
with the matrix A are calle
Problem
symmetric
vector and Q = 4;;%;%; is a real
If x: (4, 22, --->%n) is a unit eme valu es of Q are the
matrix A, then the extr
quadratic form with nonsingular mize Q subj ect to the con-
it. Hint : Maxi
characteristic values of A. Prove s (a;; — 6,jA) u; =
deduce the system of equation
straining condition 7,7; = 1 and
er.
0, where A is the Lagrange multipli
40 LINEAR VECTOR SPACES. MATRICES [CHAP. 1

14. Illustrations of Reduction of Quadratic Forms

We shall interpret the results of Sec. 13 in the language of analytic

geometry and give two examples providing concrete illustrations of
reduction of quadratic forms to the canonical form by means of orthogonal
transformations.
If we suppose that the dimensionality of space n = 3, and set 4,;7,x,;
equal to a constant c, then the equation
(14.1) iio
ig 4"9 (i,
7 = 1,2, 3)

represents a quadratic surface Q referred to a reference frame with base

vectors a’. An orthogonal transformation S-14S = A, leading to the
quadratic form
(14.2) Aé12 + AS? + Asés” = c,
can be interpreted as a transformation of coordinate axes yielding a
frame with base vectors directed along the principal axes of the quadratic.
Let the quadratic exemplifying (14.1) be

OQ = 22,7 + 2x? — 15x,? + 82,7, — 12xx, — 12x73 = €.

In order to determine the coefficients 4; in (14.2) for this particular

case, we symmetrize Q and obtain
O = 22,7? + 4a,2, — 62,25
+ 4xy2, + 2x," — 62,25
— 6x3x, — 624%, — 152”,
from which the characteristic equation |A — AZ| = 0 can be written down
at once. We have
2—A 4 —6
|A—AI|=| 4 2—A —6 sx):
—6 —6 —-15—A4
Expanding this determinant leads to a cubic
Po Vit 144i te
which has the roots
A,= —2, A,= —18, A, = 9,
Thus, relative to a new reference frame, Q assumes the form

—2§,? — 18€,? + 9&,? = ¢,

representing an hyperboloid.
Sec. 14] REDUCTIONS OF QUADRATIC FORMS 4]

For the determination of the new base vectors s‘, we have the system
of equations 13.9,
QijSin = Sihys (no sum on k),
or
(a;; — 9:;A,)Sjn = 9.
Writing these out, we have
(2 — Ay)Sixz + 482% — 953, = 0,
(14.3) 454, + (2 — Ay)Sox — 653, = 9;
65}, a OSs, aa (15 4+ Ax) Sax = 0.
—

Substituting A; = —2 gives three equations, two of which are identical.

The linearly independent equations are
4s, + 4501 a 6531 — 0,

— 65,1 a 651 a 1355) == 0.

Solving these yields the components ors.

$41 = ¢, oa =e Sy, = 0,

where c is arbitrary. We determine the constant ¢ so that the length of

s® is unity. Thus
2 2 i |,
Sy + Soy" 1 S31
are
and hence c = 1/,/2 and our normalized components

S.
These determine the first column of the matrix
(14.3) leads to three homogeneous
The substitution of A, = —18 in
equations
+ 4555 ae 6532 — 0,
208i

20559 Se 653 = 0,
4545 +

6515 ae 65Sa0 + 330 = 0,

—

be
the solution of which is readily found to
res wey =
iC, Soo = 4¢, Sg. = C.
Sig =
42 LINEAR VECTOR SPACES. MATRICES [CHAP.1

The elements entering in the third column of S are determined from

the system 14.3 by setting 24; = 9. This yields the equations
— 7545 + 4555 = 6535 — ‘).

4513 — 783 — 6533 = 0,

— 6513 == 6593 a 24535 = 0,

which are satisfied by

Sig Fc S93 5 Gs S33 = —3C.

Normalizing to unity, we obtain s) in the form
S131
=
es
ae
Seg = 3> S33 =
oa, mss1

Accordingly, the orthogonal transformation yielding the desired canonical

form is
1 1
& = = % — = % + 0° 4,
az 2
1 1 -
&, = —= 2, + —=2, + —= 25,
3.42 3,/2 a2
§5 = 3a, + §r_ — 323.

To illustrate reduction in the event the characteristic equation has

multiple roots, we take
QO = 3x? + 2x,? + 32,7 + 22,25 = €.
Here the characteristic equation of the matrix of Q is
3—A 0 1
sols (enlistNie? tea’ 0 | =/? — 82? + 201 — 16 = 0,
I 0 3-A
whose roots are A, = A, = 2,4; = 4. Hence the quadric surface is an
ellipsoid of revolution whose equation can be taken in the form
2(E,? + &") + 46,2 =.
The equations for the determination of the new base vectors are
(3 — A)sy, + O52, + 53, = 0,
Os, + (2 = A)se, + O55, = 0,
Six + OSq, + (3 — A)sg, = 0.
Setting 2, = 2 yields only one equation

Su + 53, = 0
Sec. 15] REAL QUADRATIC FORMS 43

for the determination of s™, so that the normalized solution can be taken
as

Sy = 0, tegen

The second characteristic root A, = 2 gives the equation

(14.4) Syo + Sg = 0,

and since s) must be normal to s, we have the orthogonality condition

541512 + SeiS22 + 531532 = 9,

or

(14.5) ia “ are = 0.

Equations 14.4 and 14.5 state that 51, = 0, Son = 1, Sgg = 0.

Finally, for the determination of the third base vector we have the
system of equations
—S13 + 533 = 9,
— 2595 = 0,

Sig — S33 = 0,

obtained by setting A= 4. The normalized solution of this system is

S has the form
S33 = 1//2, Sog = 0, S33 = 1//2. Hence the matrix

ig eS1 | pS1
oe cule
the variables x; and &;
from which the equations of connection between
can be written down at once.

atic Forms
15. Classification and Properties of Real Quadr
rties of real quadratic forms
In this section we summarize several prope
O = 4,4 ;%;, (G,j=1,..-; n),
(15.1)
applications.
which are of considerable importance in
44 LINEAR VECTOR SPACES. MATRICES [CHAP.|

We have shown that the real-quadratic form Q can be reduced by an |

orthogonal transformation
(15:2) E, = 5104,

to the canonical form

(15.3) OD = AEP + AGP +00 + AED

The problem of reduction of (15.1) to the form 15.3 is equivalent to
the search of an orthogonal matrix S = (s,,) satisfying the matric equation
(15.4) S14S=A, (or S‘'AS=A),
where the elements along the diagonal in the A matrix are the roots of
the determinantal equation
(15.5) |A — All| = 0,
and A is a real symmetric matrix.
Since the determinant of S does not vanish, it is clear from (15.4) that
the rank of A is equal to the rank of A. If the characteristic equation
15.5 has n nonvanishing roots, then the number of terms actually appearing
in (15.3) is n. If, however, equation 15.5 has r < nm nonvanishing roots,
then the reduced form 15.3 will have the appearance

(15.6) 0) = Ages" a ne Se 99% Se Ac

and we shall say that the rank of (15.1) is r. The number of positive 2’s
appearing in (15.6) is called the index of Q. If we have a form (15.6)
with p positive and r — p negative /’s, we can introduce a real trans-
formation &; = (UNDE for terms with positive ’s and €; = (1/V —A,)é;’
for terms with negative /’s so that it assumes the form

(15.7) OQ = &" + &’? +-++4 6% — eh — ee ws Ee

Thus every real quadratic form Q can be reduced by a real linear trans-
formation &;' = c,;x; to the canonical form 15.7. The matrix (c;,), of
course, is not necessarily orthogonal.
The form 15.7 provides a means for the classification of quadratic
forms.
We consider the following cases.
1. The index p in (15.7) is equal to n, so that equation 15.5 has n
Positive roots. In this case we say that the form 15.1 is positive definite.
2. If the index p = 0, so that all roots of (15.5) are negative and the
rank of Q isn, the form 15.1 is negative definite.
Sec. 16] SIMULTANEOUS REDUCTION 45

3. If the index p is equal to the rank r and r <n, then the form is
said to be positive. On the other hand, if the index is zero and the rank
r <n, the form Q is negative.
4. The forms whose canonical representation 15.3 contains both
positive and negative /’s are called indefinite.
We observe that positive and negative definite forms never vanish for
real nonzero values of the variables x; They vanish if, and only if, all
a,s vanish. In contradistinction, the positive and negative forms may
vanish for nonzero values of the arguments 2,. To see this, note that,
if r <n, then
QO = AE? + Ak? + °° + Ag,
We can make (15.1) vanish by choosing the 2; in (15.2) so that
fos, = os, = 0.
of r
The nonvanishing values of x, will surely exist, since the system
homogeneous equations,
$,;%; = 9, fp ot Se ee dP
er r < 7.
in n unknowns 2,, has nontrivial solutions whenev
that in a positive
It follows at once from (15.4), and from the fact
inant |a,,| of
definite form the A,’s in A are all positive, that the determ
is necessa rily positiv e. The convers e of this,
the positive definite form
by noting that |A| = |Al,
clearly, is not true. This can be readily seen
as well as definite forms.
and the positive value of |A| admits indefinite

Forms to a Sum of Squares

16. Simultaneous Reduction of Two Quadratic
investigating the possi-
We conclude our study of quadratic forms by
quadratic forms to the sum
bility of simultaneous reduction of two real
This problem arises, among
of squares by a single linear transformation.
nical systems about the
other places, in a study of oscillations of mecha
state of equilibrium.
Consider two real quadratic forms
QO, = AyiX%; and Oo = djjX,X5,
(16.1)
ive definite. Let it be required
each of rank n, one of which, say Q,, 1s posit
sarily orthogonal, such that
to find a linear transformation, not neces
both forms reduce to the sum of squares.
n, then there exists a linear trans-
If Q, is positive definite and of rank
gonal, under which Q, reduces
formation 2; = ¢;;6;, not necessarily ortho
to the form
E+ Stet ctl E 2.
Q,=
(16.2)
46 LINEAR VECTOR SPACES. MATRICES [CuapP. |

Under the same transformation Q, will assume the form

(16.3) O, = bi; €,8;.

Now, under a suitable orthogonal transformation &; = d,;n; on the vari-

ables &,, Q, can be reduced to the form

(16.4) QO, = Am,

and, since orthogonal transformations leave the scalar product €;¢;
invariant, the form Q, will be unchanged, and we have

(16.5) 01, = 1.0;

Now Q, and Q, are in the desired forms, and, since the product of
successive linear transformations from 2, to 7; is a linear transformation
x; = S;;N;, it follows that the simultaneous reduction can be accomplished.
The numbers A; in (16.4) are called the characteristic numbers of the
form Q, relative to Q,. We proceed to derive the equation for the charac-
teristic numbers 4,.
We recall that if the variables x; in a form Q = a,,x,x,; with a matrix
A are subjected to a linear transformation x; = s,;7; with the matrix S,
then the matrix of the resulting quadratic form is S‘AS. The determinant
of this matrix has the value |S|? |A|. Now let us construct the quadratic
form
(16.6) Q= 0, —AQ;
= (b;; — Aa,;)x;%;,

where A is an arbitrary parameter. Under successive linear transformations

from the variables x, to 7;, Q. and Q, assume the forms (16.4) and (16.5),
and hence (16.6) reduces to

(16.7) O =A? + Asn? + °° + Ayn,

=e Any? = Ane 32 See ney An,”

PACS = Ayn.

The determinant A in (16.7) is

(16.8) A= (A, — AM aye— A) 8 (A, =A);

whereas the determinant of Q in (16.6) is

(16.9) D = |b;; — ha,;|.

It follows from remarks just made regarding the value of the determinant
in the transformed quadratic form that the determinants D and A can
differ only by a constant multiple equal to the square of the determinant
Sec. 17] UNITARY TRANSFORMATIONS 47

|S| of the transformation from the initial variables x; to the final variables
yn; Since this determinant does not vanish, and since it contains no
parameter A, the roots of polynomials 16.8 and 16.9 are identical. Taking
account of the structure of expression 16.8, we conclude thatthe coeffi-
cients A, in (16.4) are the roots of the determinantal equation

Byy — Ady, yp — Ayn * °° Dan — AGin

ie boy — Ada, daz — Ada2 Don — Adon i),

Daa — Aan Dre + AA ne oe a Doe

In application of these results to the study of small vibrations of

mechanical systems about the point of equilibrium, the forms Q, and Q»
are identified with the kinetic and potential energies of the system. The
final coordinates 7; are termed normal coordinates, and the characteristic
numbers A, are related to normal modes of vibration (see Sec. 89).

17. Unitary Transformations and Hermitean Matrices

Ih a variety of circumstances arising in applied mathematics it becomes

s to vectors
necessary to extend the concept of orthogonal transformation
defined in a complex field.
If we consider a nonsingular transformation
/
A; ;%j5 (i, j => l. EOS} n),
x; =
(7-1)

the set of numbers

in which the coefficients a,,; are complex numbers and
x, the question
(a,,...,%,) represents the components of the vector
d on the matrix
naturally arises about restrictions that must be impose
The imposition of
(a;;) if the length |x| of the vector is to be preserved.
the condition of invariance of length, namely,
hp
Xz, UL, = UX,
]
leads at once to the conclusion that [cf. (11.5)

(17.2) Gi jAin = Ojns

values. We deduce from
where bars, as usual, denote conjugate complex is 1.
of the determinant |a,,|
(17.2) that the absolute value of the square called
condi tions 17.2 are
Matrices A = (a,;) whose elements satisfy the trans-
17.1 are unitary
unitary, and the corresponding transformations
form
formations. We can write (17.2) in the

(17.3) A’A = 1,
48 LINEAR VECTOR SPACES. MATRICES [CHAP.1

where A is the conjugate matrix formed by replacing every element 4;,;

in A by @,;. From (17.3) we conclude at once that A’ = A™.
A bilinear form

(17.4) Jab Es pensiees

Cn hed bali
(j= 13
«5 J),
where a,; = d;,, is called a Hermitean form, and the matrix (a;;) = A,
corresponding to it, is a Hermitean matrix. It follows from the definition
of the Hermitean matrix that the elements along its diagonal are real and
that
Ames ie peeae
We observe that the Hermitean forms can assume, for arbitrary 7;,
only real values, since

It is clear that the Hermitean forms are a generalization of real quadratic

forms.
One can raise the question of the possibility of reduction of the form
17.4 to the canonical form
(17.5) He = Ay2,8; + Ab ok ed tees
with the aid of the transformation
ta ee, or x = UE,
where U = (u,,;) is a unitary matrix. A computation similar to that
carried out in Sec. 13 leads to the solution of the matric equation
(17.6) U--AU = A,
where A is a diagonal matrix. The procedure in this case is, in every
respect, similar to the one followed in the discussion of real symmetric
matrices. We multiply (17.6) by U and obtain
(17.7) AU = UA,
which represents a system of linear homogeneous equations for the
determination of vectors u): (w;, Usg, -- +» U,,) Entering in the columns
of U. A necessary and sufficient condition that the system represented
by (17.7) have a solution is that
(17.8) |A — Al| = 0.
The possibility of constructing a unitary matrix U satisfying equation
17.6 hinges on the fact that here the roots of (17.8) are also real. The
 PROBLEMS 49

fact that the characteristic roots 2, must necessarily be real follows from
the observation that U-1AU is a Hermitean matrix whenever A is Her-
mitean and Uis unitary.” Thus A in (17.6) is Hermitean, and consequently
the elements along its diagonal are real. :
Problems
1. Reduce the matrix
(1 =)
A = (a;;) =
—1 1

to the diagonal form S by the similitude transformation C-!1AC. Show that

feeet 4 0 0
C= ,Ci= , and S =
cal I role
nie
Ome
the
Discuss the meaning of A when it is viewed as an operator characterizing
deformation of space.
2. Diagonalize the matrices:
te ae Ot | =1 1 2
= | 1 -1], DO ge 1

—{-~ —1 1 Ora OS —3

Y. Since A is Hermitean,
7 For (U-!AUY = U’A(U-1)’ and (U“-AU) = U’A(U-
(U-!AU)’ =
is unitary, U’ = U-1and (U~)’ = U. Thus we have
A’ = A, and since U
WatAUs
 yy
TENSOR THEORY

18. Scope of Tensor Analysis. Invariance

Tensor analysis is concerned with a study of abstract objects, called

tensors, whose properties are independent of the reference frames used to
describe the objects. A tensor is represented in a particular reference
frame by a set of functions, termed its components, just as a vector is
determined in a given reference frame by a set of components. Whether
a given set of functions represents a tensor depends on the law of trans-
formation of these functions from one coordinate system to another.
The situation here is identical with that already encountered in Chapter 1.
In a given reference frame a vector A is determined uniquely by a set of
components A;. If anew coordinate system is introduced, the same vector
A is determined by a set of components B,, 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 the vector idea, and the
same is true of tensors.
Since tensor analysis deals with entities and properties that are in-
dependent of the choice of reference frames it forms an ideal tool for
the study of natural laws. Indeed, whether a logical deduction based on
a conglomerate of observational facts deserves the name of a natural law
is often determined by the generality of such a deduction, and by its
validity in a sufficiently wide class of reference systems. This is intimately
bound up with the possibility of formulating the deduction in the form of
a tensor equation because tensor equations are invariant with respect to a
given category of coordinate transformations. 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
the 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 x. If the coordinate
system is changed, the point P is described by a new set of coordinates y’,
but the transformation of coordinates does nothing to the point itself.
50
Sec. 19] TRANSFORMATION OF COORDINATES 51

Again, a pair of points (P,, P.) determines a vector P,P,. This vector, in
a particular reference frame, is uniquely determined by a set of components
A;. A transformation of coordinates does nothing to the vector P, Ps, but
. . . & arta

ia
in the new reference frame P,P, is characterized by a different set of com-
ponents B,. A set of points, such as those forming a curve or surface, is
also invariant. The curve may be described in a given coordinate system
by an equation which usually changes its form when the coordinates are
changed, but the curve itself remains unaltered. We shall say, in general,
that an object, whatever its nature, is an invariant, provided that it is not
altered by a transformation of coordinates.

19. Transformation of Coordinates

In Chapter | we discussed at some length linear transformations of co-

ordinates. Here we will deal with real, single-valued, reversible functional
transformations of the form
(19.1) ea EO ee te. ot.) Gi= 1,2,...,%),
where we use superscripts to identify the variables. A particular set of n
real numbers (29!, 22, . . . , Y”) can be thought to specify a point Py in the
n-dimensional metric manifold covered by a coordinate system X. The
set of equations 19.1 will be viewed as a transformation of coordinate
systems, so that the m-tuple of numbers (YolyYor, -- +> Yo") Obtained by
substituting in (19.1) the coordinates (x ', %°,... a“”) represents the
coordinates of P, in the ¥-reference frame. Since the transformation TJ in
(19.1) was assumed to be reversible and one-to-one, we can write
(19.2) T: af = xy}, y’,...,y"), (=1,2,...,0),
where the functions! x‘(y) are single-valued. To ensure the satisfaction of
restrictions we have just imposed on the transformation of coordinates, it
will suffice to suppose that the functions y'(x) in (19.1) are continuous
mani-
together with their first partial derivatives in some region R of the
: oy’ .
at
fold V,,, and that the Jacobian determinant J = Ee does not vanish
single-
any point of the region R. It would follow then? that not only a
are also of
valued inverse (19.2) exists, but the functions x‘(y) in (19.2)
class C! in some neighborhood of the point under consideration.
zi(y!,...,y") and
1We will often use the notation x(y) and f(z) to mean
f@',2',..., x"), respectively.
, pp. 433-438. We use the
2 See, for example, I. S. Sokolnikoff, Advanced Calculus
ous together with their
symbol C” to denote the class of functions which are continu
first n partial derivatives.
 D2 TENSOR THEORY [CHAP. 2
We observe that, if the functions y‘() in (19.1) are of class C1, then, by
Taylor’s formula,
y = ay 2h Loe

where a,’ is the value of 0y'/0z/ evaluated at some point P’ of the region R.
The point P’ depends, of course, on the choice of values (2', x*,..., x”).
Thus the transformation 19.1, with stated properties, is /ocally linear. The
nonvanishing of the Jacobian guarantees that this system of linear equa-
tions has a unique solution. Throughout the rest of this book we shall
suppose that all encountered transformations of coordinates are of the
form 19.1, in which the functions y‘(z) are at least of class C! in some
oy’ ’
region R, and that = ~ Oat any point of R. For brevity we shall refer
x
to a class of coordinate transformations with these properties as admissible
transformations.
As an example of an admissible transformation consider a system of
equations specifying the relation between the spherical polar coordinates
a‘ and the rectangular cartesian coordinates y’,
y' = x! sin x? cos 2%,
T: (y? =z! sin 2? sin 23,
if ==5 COS ue
If we suppose that 1 > 0,0 < 2? < z, and 0 < 2° < 2z, thenJ ¥ O and
the inverse transformation is given by
8 I

ewer! Vy? a (y*)

7]
x® = tan? x ;

Problem
Discuss the transformations in which the coordinates y‘ are rectangular
cartesian:
2
yt = —=
/
a! + V6
— a? 4 V6
— 23,
1 1
(a) y? = —= a! — —~ 2g? 4 —_ 23
V2 V3 IS
if 1
y> = — gl i

Vy v2

(6)
SEc. 20] ADMISSIBLE TRANSFORMATIONS 53

20. Properties of Admissible Transformations of Coordinates

From a summary of certain important properties of admissible co-

ordinate transformations contained in this section, we will see that it is
quite immaterial what particular reference frame is selected to describe the
invariant entities. It will be shown that all admissible transformations of
coordinates form a group, and hence every coordinate system in the family
can be obtained from the particular one by an admissible transformation.
This fact is of great moment in the construction of a theory that lays claim
to its independence of the accidental choice of reference systems.
THEOREM I. Jf a transformation of coordinates T possesses an inverse
T- and ifJand K are the Jacobians of T and T™, respectively, thenJK = 1.
The proof is easy. We insert the values of x’ from (19.2) in (19.1) and
obtain a set of identities in y’,

ge ag Ci Caren ee Cine ie) 5

y

The differentiation with respect to y’ yields

dy! =o_ gs= — Oy’ Ox"+ = Is 8 aloe LO):

ay) aa* ay? ( )

aude |_| au'| Je] yx

Ox" Oy’ ax*| | dy*

Since |0,'| = 1, we see that J-K = 1. Incidentally, it follows from this

result that J ~ Oin R.
Consider now any two admissible transformations
Limp 7 CR Y=;
and

The transformation
Tied =a ye eee ae oe tte) |
T; = 727}. If the Jacobian
is called the product of 7; and T;, and we write
of T; is denoted by J, it follows that
‘CRC as le Ozt ar
3 | ayt ax? dy’ || Ox?
ET SI a
T, and Tj, respectively.
where J, and J, are the Jacobians of
54 TENSOR THEORY [CHAP. 2

We can state this result as a

THeoreEM II. The Jacobian of the product transformation is equal to the
product of the Jacobians of transformations entering in the product.
These theorems enable us to establish an important
THEOREM III. The set of all admissible transformations of coordinates
forms a group.
The truth of the theorem becomes obvious if one notes that
(a) The fundamental group property, namely, the product of two
admissible transformations is a transformation belonging to the set of
admissible transformations, is clearly satisfied. This property is known as
the property of closure.
(b) The product transformation possesses an inverse, since the trans-
formations appearing in the product have inverses.
(c) The identity transformation («* = y’) obviously exists.
(d) The associative law 73;(7)T,) = (T3T2)T, obviously holds.
These properties are precisely the ones entering in the definition of an
abstract group.
As noted in the beginning of this section, the fact that admissible
transformations form a group justifies us in choosing as a point of depar-
ture any convenient coordinate system, as long as it is one of those admitted
in the set.

21. Transformation by Invariance

Let F(P) be a function of the point P in the n-dimensional manifold V,,.

We will suppose that F(P) is a continuous function in some region R of V,,
and that V,, is covered by some convenient coordinate system X. The
values of F(P) depend on the point P, but not on the coordinate system
used to represent P. We call F(P) a scalar point function or simply a scalar.
In the reference frame X, F(P) may assume the form f(21,..., #"), and, if
we introduce a new reference system Y by means of a transformation
(21.1) T:> xt = xi(y',...,y"),
the functional form of F(P) in the Y-frame is

(21.2) fly... 9")... oY... yD) = oy... 9")

since the value of f(#1,..., 2") at P(w!,..., 2”) is the same? as that of
gy’, ..-,y") at Py, <, y%).
* In a specific case, F(P) may represent the temperature of some region of space and
f(@) is the form which the temperature function assumes in the X-reference frame;
&(y) is the representation of F(P) in the Y-reference frame.
Sec. 21] TRANSFORMATION BY INVARIANCE 55

We can speak of f(x) as being the component of the scalar function

F(P) in the X-coordinate system, while g(y) is the component of the same
scalar function in the Y-coordinate system. Alternatively, we can regard
the scalar function F(P) as being defined by the totality of componenfts (2),
g(y), A(z), etc., each of which is related to one another by the substitution law
typified by formula 21.2. In other words, once the representation of the
scalar F(P) in one coordinate system is known, then the form of F(P) in
any other coordinate system Y is determined by formula 21.2. We call
this substitution transformation G°: f[z(y)] = g(y) the transformation by
invariance.
We observe that, if we have three transformations 7}, T,, and T;, where

Ty*: x= xy),
te - \ y(2),
3
with T, = T,7T,, so that
Ty: « = x[y@)],
and a scalar F(P) whose component in the X-frame is f(x), we can compute
the transforms of f(z). Indeed, the component g(y) of F(P) in the Y-
frame is determined by the law
Gy»: gy) =fleM],
given by
whereas the component h(z) of F(P) in the Z-frame is
G,°: h(z) = gly@).
mation T, = 727;, we get
On the other hand, using the product transfor
GP: he =ficly@]},
from which it is clear that G,° = G,°G,°.
and the corre-
We can represent these transformations of coordinates
diagra mmatic ally as in
sponding transformation of components of F(P)
a group T of admissible
Fig. 7. Thus, as coordinates are subjected to n trans-
under go a certai
transformations, the components of a scalar T and
sive transf ormati ons
formation G®. The relation between the succes to the
ons 7,7, corres ponds
G® is such that the product of two transformati
g(y)
y

A Ts Gy, Ga’

z f(x) h(z)
x
56 TENSOR THEORY [CHAP. 2

product of two corresponding transformations G,°G,°. When sucha relation

obtains between any two groups of transformations T and G, the groups
are said to be isomorphic. The isomorphism between the transformations
of coordinates and the transformations of functions induced by the trans-
formation of coordinates is an important characteristic of a class of in-
variants called tensors.

22. Transformation by Covariance and Contravariance

In the preceding section we discussed the transformation of components
of a scalar F(P) when the coordinates of P undergo a transformation. In
this section we discuss the law of transformation of entities determined by
the sets of partial derivatives of a scalar. Sets of partial derivatives of the
component f(a!,...,") of a scalar F(P) are of interest in physics in
connection with the notion of a gradient of potential functions.
We consider a continuously differentiable function f(z',..., 2”),
representing the scalar f(P), and a transformation of coordinates
(2220) Tere, sete:

If we form a set of n partial derivatives

(22.2) Jie)
ras | ae) of ’ or {fai},
Ox’ Ou" Ox”
the question arises: What does the set {f,:} become when the coordinates
a‘ are subjected to a transformation 22.1? This question is quite without
meaning unless one specifies precisely what is to be done with the set 22.2.
These fractions do not automatically “become” anything until one states
what law he is to use in calculating the “‘corresponding functions” in the
Y-frame. In other words, it is necessary to agree on what the term
“corresponding function” is to mean in a given situation.
For example, we might calculate the corresponding functions by the
transformation of invariance G° of Sec. 21; that is, we can insert in each
function f,.(z', ..., x”), the-values of the 2’s from (22.1). This will yield
a set of n functions

(22.3) gaily’, Sener CARE gly’, ae ey y"), a ee SAY. <u 2.5 y").

On the other hand, if one has in mind the notion of a gradient of f(P), it
is necessary to say that the set of functions corresponding to (22.2) is not
(22.3), but the set of n partial derivatives,

(22.4) eat
dy!’ ay?’ *”
wh
ay”?
SEC. 22] TRANSFORMATION BY COVARIANCE 57
computed by the rule for differentiation of composite functions, namely,

of — Oj_.ax tae vine

(225 ) G}: at ace" oy : Cie sl Oe n). :

transformation
If we have a function f(«!,...,«#") and a
Tee aes ng 28,

the set of functions corresponding to (22.2), determined by the law G*

(equation 22.5), is

We can think of the sets of functions {0//0x"}, {df/dy’}, {af/dz"}, etc., as

representing the same entity in different reference frames. At any par-
ticular point Po(ap1, . . . , %") the set 22.2 determines n numbers, which
can be regarded as the components of the gradient vector, and the set 22.4
represents the same vector in the Y-coordinate system.
If we have a set of n functions A,(),..., A,() associated with the
X-coordinate system, and if we agree to calculate the corresponding
quantities B,(y),..., B,(y) in the Y-system by means of the covariant
law G', namely,

(22:6) By) = = Ax),

t vector
we say that the set {A,(x)} represents the components of a covarian
represen ts the same covaria nt
in the X-coordinate system. The set {B,(y)}
itself is the totality of sets
vector in the Y-system, and the covariant vector
nt law G’.
of such quantities each related to one another by the covaria
of vectors, which is
As an illustration of the law of transformation
tials
quite different from the law G', consider a set of n differen

(22 7) du\, dx*,..., dx”,

a 22.1. If we
where the xs are related to the variables y’ by the formul
., 2" + dx”), then the
have two points P,(a',..., #”) and P,(a.+ dzl,..
from P, to Ps.
set of n numbers 22.7 determines the displacement vector
to the Y-coordinate
The same displacement vector when referred
system has for its components
(22.8) dy), dy?,..., dy”,
where
58 TENSOR THEORY [CHap. 2

Note that the law G?, for the determination of the quantities 22.8, is
different from G!. If we have a set of quantities 4,(x), A,(x),..., 4,(2),
then the law G?, determining the corresponding quantities B,(y), B.(y),...,
B,{y), is
(22.9) B, =
The law G? is the contravariant law, and we call the sets of quantities
transforming in accordance with it the components of a contravariant vector.
The laws G®, G!, and G? play a fundamental role in the development of
tensor analysis.
Problems
1. Show that if the transformation T: y* = a;‘x’ is orthogonal, then the
distinction between the covariant and contravariant laws disappears.
2. Prove the theorem: If f(x',2?,...,2”) is a homogeneous function of
of :
degree m, then Fees ee mf.
3. Given f(z, x?,...,%2") and a set of equations of transformation 2* =
xi(y}, y?,...,y"), where.each y* = y(t). If the transform of f by invariance is
gy’, y*,...,y"), show that df/dt = dg/dt. Hint: (0f]@x*)(dx*/dt) = df/dt and
dx*|dt = (dx%/ dy?)(dy?/dt).
4. Write out the laws of transformation of components of covariant and
contravariant vectors when TJ is the transformation from rectangular cartesian
to spherical polar coordinates given in Sec. 19.

23. The Tensor Concept. Contravariant and Covariant Tensors

Consider an admissible transformation

To fm (ane,oD, (= 12, 6):
and. a set {f;} of m continuous functions
fide we oye"): (i = 1,2,...,.m),
defined in some region R of the n-dimensional space referred to the X-
system of coordinates.
We associate with the given transformation 7 a transformation G which
transforms each /,(x1, x?,..., #”) into a function

gly’, 7", 8-5 y").

Examples of the transformation G are the transformation of invariance

and the contravariant and covariant laws introduced in preceding sections.
But, whatever the nature of the transformation G, it will always depend
on T, and to emphasize this fact we shall say that G is a function of T.
We shall call G an induced transformation on the set of functions f,.
SEC. 23] THE TENSOR CONCEPT 59
Suppose further that G, regarded as a function of 7, satisfies the
following conditions:
(a) When T is an identity transformation, then G is an identity trans-
formation. This means that, if y* = 2’, then r

Fats. 252°) = f(y acre).

(b) If T,, T:, Ts are three transformations of the type T, and Gy, G2, G3
are the corresponding induced transformations G, and if T; = T2T,, then
G,; = G,G,. In other words, the sets of transformations T and G are
isomorphic. If the given set of functions {/;} satisfies conditions (a) and
(b), we shall say that the set {f,} represents the components f; of a tensor f
in the X-coordinate system, the tensor f itself being the totality of sets of
functions {f,(x)}, {g;(y)}. ete.
It should be remarked that the term tensor was used by A. Einstein*
only in connection with the sets of quantities transforming in accordance
with the contravariant and covariant laws. The formulation of contra-
variant and covariant laws, as well as an outline of the essential features of
the algebra and calculus of contravariant and covariant tensors, is due to
G. Ricci.® The much broader characterization of tensors by the iso-
morphism of transformations of coordinates and induced transformations
is essentially due to H. Weyl and O. Veblen.® Because of the usefulness
and commonness of covariant and contravariant laws of transformation
in applications of analysis to geometry and physics, the term tensor is
generally used in the sense contemplated by Einstein. This usage is
followed in the sequel. However, the isomorphism between the laws of
transformation of coordinates and the induced transformations is so
tensor
fundamental to the idea of a tensor and to the invariant nature of
calculus that it justifies the degree of emphasis placed on it in the fore-
going.
and mixed
We now turn to a consideration of covariant, contravariant,
ns
tensors. It will be convenient to introduce (with Ricci) different notatio
at a glance.
for each type of such tensors so that they can be recognized
Let us consider first a set of functions of the variables (z!,..., 2”),

{A(i; x)} or A(1; 2), AQ; 2), ..-, A(t; #).

subscript or super-
Previously we wrote the identifying index 7 either as a
use supers cripts to denote the set of functions
script, but now we agree to
4 A. Binstein, Annalen der Physik, 49 (1916).
dei Lincei, 5 (1889).
5 G. Ricci, Atti della reale accademia nazionale
(1925-1926). O. Veblen, Invariants of
® H. Weyl, Mathematische Zeitschrift, 23, 24
24 (1927), pp. 19-20.
Quadratic Differential Forms, Cambridge Tract No.
60 TENSOR THEORY [CHAP. 2

that transform in accordance with the contravariant law and subscripts

for sets that transform in the covariant manner.’ Whenever the law of
transformation is neither covariant nor contravariant, or when its nature
is in doubt, we write {A(i; x)}, {B(i; y)}, etc. We now lay down the
following definitions.
DEFINITION 1. A covariant tensor of rank one is the entire class of
sets of quantities {A(i; x)}, {B(i; y)}, {CU 2}, . . . related to one another by
the transformation of the form

Bis) =F Aes2) (ia =1,2,...,7),

where {A(i, x)} is the representation of the tensor in the X-coordinate
system, and {B(i; y)} is its representation in any coordinate system Y
related to the X-system by the transformation T.
Frequently, we speak loosely of the given set {A(i; x)} as being a tensor,
but this usage should not conceal the fact that the tensor is the fotality of
sets of quantities typified by {A(i;x)}. The last set refers to the repre-
-sentation of the tensor in a particular reference frame and can be spoken
of as the component of the tensor in the X-coordinate system. However,
we shall use the term component of a tensor to mean the individual elements
A(i; x) in the set {A(i; x)}.
We denote components of covariant tensors by subscripts and often
suppress the variables x and y entering as the arguments of A’s and B’s.
Thus
Ox
B,;= =A, (covariant law).
oy’
DEFINITION 2. A contravariant tensor of rank one is the entire class
of quantities such as {A(i; x)}, {B(i; y)}, ... related to one another by the
transformation of the form
: oy'
Blige) = oe A(a; x),

where {A(i; x)} represents the tensor in the X-coordinate system and {B(i; y)}
in the Y-coordinate system.
We denote components of contravariant tensors by superscripts. Thus
. oy '
Bie = A* — (contravariant law).
Es
’ The only exception to this convention is in the use of superscripts to identify the
variables x‘, y', etc. These quantities do not transform according to a covariant or
contravariant law unless the transformation T is affine.
SEC. 23] THE TENSOR CONCEPT 61
The definitions of contravariant and covariant tensors of rank one are
identical with those of contravariant and covariant vectors given in Sec.
22.
We speak of scalars, defined in Sec. 21, as tensors of rank zero.
We can generalize the definitions of tensors of rank one to include
tensors of any rank as follows.
DEFINITION 3. A set of n" quantities A; ;,...;,(&), associated with the
X-coordinate system, represents the components of a covariant tensor of
rank r if the corresponding set of n’ quantities B;;,...:,Y), associated
with the Y-coordinate system, is given by
att a Op
pauate a ae er eimaee
oy dy? OY uae
The tensor itself is the totality of sets of such quantities as {Aj 5, ---4,}-
DEFINITION 4. A set ofn"quantities A'\'>’**'"(x) represents the components
of a contravariant tensor of rank r in the X-coordinate system whenever the
corresponding set B's'?’’'"*(y) of n" quantities in the Y-system is given by

As an illustration we note that the components of the covariant tensor

of rank two transform according to the law
Ox* Ox?
£0) = Ajp(),
dy’ dy
whereas the components of the contravariant tensor are given by
19,4
BY(y) = oy!oy A(z):
Ox” Ox?
There are n? components in each set.
We define next the mixed tensor.
typified in the
DEFINITION 5. The totality of sets of n'** quantities,
express ions Aj1}2'' :1(x), is a mixed tensor,
X-coordinate system by the
of rank s, provid ed that the corresponding
covariantofrank rand contravariant
by the law
quantities Bix}:::}+(y) in the Y-coordinate system are given

iyig:*+tr dy" dy’? oy" Ox? Ox? Ox?s Lit ft

s A} of the
We note that this law for the transformation of component
o J
example of a mixed
mixed tensor gives Buy) = = = A®(a). Asa simple
y' Ox
62 TENSOR THEORY [CHAP. 2

tensor that already has occurred in our discussion, we cite the Kronecker
OxuOy hoy é
delta 64. Thus — —, 62 = 6/. The verification of the fact that
Oy’ ax? *dy?
the definition of covariant, contravariant, and mixed tensors satisfies
properties (a) and (b), stated in the beginning of this section, is given in
Sec. 24.
To distinguish tensors defined over a region of space from tensors whose
domain of definition is a single. point, one occasionally speaks of the
former as constituting a tensor field.

24. Tensor Character of Covariant and Contravariant Laws

We will verify that the induced transformations defined in the preceding

section satisfy the isomorphism conditions stated in Sec. 21. The fact that
the transformation of invariance (leading to tensors of rank zero) fulfills
these conditions was noted in Sec. 21. The proofs for contravariant and
covariant tensors are special cases of the proof for a mixed tensor. Accord-
ingly we consider a mixed tensor typified by the set of functions Ahja’ 114(@)
The law G for the transformation of mixed tensors is ;

(24.1) By) = eee 2 ot

and we must show that

(a) if T-=—J, then G=1,

(b) id= 7.7,, then ~G =G,G,.

Now, if T = J, then
ea = yt, a y*2, .

and hence
Oat
oy"!
a en
1
ala
Oy" tr

Moreover, T-! = J, so that

so that
SEc. 24] TENSOR CHARACTER OF LAWS 63
Inserting these values of partial derivatives in (24.1) gives

Beis
On Omi: ean. ex)
ae Ae); ’
Hence G = Jif T= I.
Suppose now that, under a transformation 7,, the variables x’ transform
into y’, and the variables y’ transform into z‘ by the transformation 7).
The corresponding induced transformations G, and G, yield
i gataee ox --ox™ oy" oy’®
(
Dee
) 1
ee
r (y)
ee dy"?
aoy" Ox? —Z—
Oxks Abr:
ay "Bs
a2),

and

(24.3) G: CHWe gan

f@) Bad oh Oy Orietiameosas
=F pate Oztt, xprite

Now, under the product transformation T; = 7,7,, the variables x*

go into y' and the y’ into 2’, so that T; carries the x’ into the z'. Inserting
the values of the B’s from (24.2) into (24.3) gives

GG Cie ela (ae Oz"!

=| ee
Oz'* ay?
a
=|
ay?s

Ox"? ee) (5 ov")Avepeeksee

aan 2
oe % era i

Ee aah Apistinamaay)
Performing the summation on «’s and f’s yields
rast is
Gag Co. ui'(2) m= i CL Ra ae RTE
dz = zt Ax = Oa’
The resulting law G; is precisely the law of transformation of the com-
ponents of a mixed tensor when the variables x’ are transformed into the
z' by the transformation T;. Thus the law of transformation G is transitive,
and this completes the proof.
The results for covariant and contravariant tensors appear as special
cases obtained by suppressing the superscripts or subscripts.
The only types of tensors with which we will deal in this book are scalars,
covariant, contravariant, mixed, and relative tensors. The last are defined
in Sec. 28.
of
We establish next a useful property of the law of the transformation
tensors, which is frequently used in the sequel.
system be
Let the components of a mixed tensor in the X-coordinate
denoted by 4j1:::4s(z) and its components in the Y-system by Bis: ///:(y).
64 TENSOR THEORY [CHAP. 2

Then, from the law of transformation of mixed tensors we can write

(24.4) Bey) = —BeOy" Ox?

hi
ame
a).

On the other hand, if we are given the components Bix" dy), the com-
ponents Af:'''/s(a) of the same tensor in the X-reference frame are
determined by the formula

Sidyt de oc
(24.5) AGL ax) — Ox™ dy aye
“i(Y).
We note that we can obtain (24.5) from (24.4) formally by treating the
partial derivatives and sums in (24.4) as though they were fractions and
products appearing in simple algebraic expressions.
From the structure of formulas (24.4) and (24.5) we 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.
This particular theorem is of profound significance in the formulation
of physical laws. It states, in effect, that, if a certain law is implied by the
vanishing of components of a tensor in one particular coordinate system,
then the rules for transformation of the tensor components guarantee that
they will vanish in all admissible coordinate systems. A physicist has
little interest in the formulation of a law that might be valid only in some
special reference frame. Indeed the notion of invariance and the universality
of physical laws is the cornerstone about which mathematical physics is
built.

25. Algebra of Tensors

In this section we establish several rules of operation with tensors,

which are algebraic in character.
THEOREM I. The sum (or difference) of two tensors which have the same
number of covariant and the.same number of contravariant indices is again
a tensor of the same type and rank as the given tensors.
Proof. Consider two tensors A(x) and A(x) of the same type and rank
defined at the same point P, and the corresponding laws of transformation:

ay)
= ine a Tee
Oat Dye
ays aa(®)s
Aan:

Ber ope ape

dy"? oy'* Oxf
te
Oxbs
11 Ba(e)
er .
Sec. 25] ALGEBRA OF TENSORS 65

Then

Be a
Beta), (ets 28 Oy (Abe
er(otalorae Oe
Ate inaBe).
Oy — dy'")— \dxF ~— ahs
It follows from this that A + 4 is a tensor, and we write

Abs Ei (ce) erate, Sn) i io 8Eee);

which is a tensor of the same type and rank as the given tensors.
It is clear from the laws of transformation of tensors that, if each
component of a tensor is multiplied by a constant, the resulting set of
functions is a tensor. This fact, in conjunction with Theorem I, permits
us to state a
CoroLiary. Any linear combination of tensors of the same type and
rank is again a tensor of the same type and rank.
THeorEM II. The equation Af... \"3(x) = Af. .f3(a) is a tensor equa-
tion; that is, if this equation is true in some coordinate system, then it is true
in all admissible systems.
Proof. It follows from Theorem I that the difference of two tensors isa
tensor. Hence
Aba.Bs_ Abs
'.fs= 0.
However, we proved in Sec. 24 that, if all components of a tensor vanish
in one coordinate system, they vanish in all admissible coordinate systems.
We shall call the tensor all whose components vanish the zero tensor.
THEOREM III. The set of quantities consisting of the product of each
element of the set A}x:: ‘4(a), representing a tensor A, by each element of
the set Aj. ...}:(x), representing a tensor A, defines the tensor %, called the
outer product. This tensor is contravariant of rank q +s and covariant of
rank p +r.
From the definition of outer product, the components of in the
X-reference frame are given by the formula
BC Oe ie A Ae
The fact that the set of functions ./}1""! i:defines a tensor follows directly
from the law of transformation of components AL Z and. iVinita oe
We will denote the outer product 7 of A and A by writing the symbols
in juxtaposition. Thus 7 = AA. It is obvious that the outer product is
distributive with respect to addition, so that
. (4+ B)C = AC + BC.
tensors.
We introduce next the operation of contraction which yields
66 TENSOR THEORY [CHAP. 2

THEOREM IV. If, in a mixed tensor, contravariant of rank s and covariant

of rank r, we equate a covariant and a contravariant index and sum with
respect to that index, then the resulting set of n’*** sums is a mixed
tensor, covariant of rank r — | and contravariant of ranks s — 1.
To avoid complications in writing we illustrate the procedure used in
the proof by considering a mixed tensor Aj,,._ We have
dy' Ox’ Ox? Ox? |,

If we equate the indices 7 and k and sum, we obtain the set of n? quantities
dy’ Ox? dx’ Ox? |,

Thus Bi, = B,, is a covariant tensor of rank two.

In this case we can obtain three different covariant tensors of rank two
by performing the operation of contraction on other covariant indices.
We observe that, when as a result of contraction of one or more pairs of
indices there remain no free indices, the resulting quantity is a scalar.
If it is possible to apply the operation of contraction to the outer product
of two tensors A and 4, the result is a tensor called the inner product of
A and A. We denote the inner product by the symbol 4: 4. The proof
that A - 4 isa tensor is immediate, for the outer product of two tensors is a
tensor, and the operation of contraction yields a tensor.
Example. Consider the tensors 4,,(x), A,(x), and A*(x). If we form
the outer product 4,;A, = A,;,, we obtain a covariant tensor of rank
three, and hence no contraction is possible here. On the other hand, the
outer product of A;; and A* gives a mixed tensor 4,,4* = A™,, and in this
case we can contract to get a covariant tensor Aj, or Az; As already
remarked, the tensor Aj,,; may be contracted in three different ways to
yield Aq,,, Aj, and Aj,,. The tensor Ay, can be contracted twice in
several ways. The contraction of 4’ yields a scalar.

26. Quotient Laws

In this section we give two useful theorems which will enable us to

establish the tensor character of sets of functions without going to the
trouble of determining the law of transformation directly.
SEC. 26] QUOTIENT LAWS 67
We use the term inner product for sums of the type A(«, ig,..., i,)Aq
(or A(a, i2,..., i,)A*) whether the set of functions A(i,,..., i,) represents
a tensor or not. We also speak of tensors of rank one as vectors.
THEOREM I. Let {A(i,, iz, . . . ,i,)} be a set of functions of the variables
az’, and let the inner product A(«, iz,..., i,)&*, with an arbitrary vector Es
be a tensor of the type AR v.42(a); then the set A(i,,..., i,) represents a
tensor of the type Azj,..°%,(2).
In order to avoid writing out long formulas for the transformation of
tensors with many covariant and contravariant indices, we will establish
this theorem for the set of n° functions A(i, j, k), which has all features of
the more involved cases. Let us suppose that the inner product A(«,j,k)&*
for an arbitrary vector &%(x) yields a tensor of the type Aj(x). We will
prove that the set A(i, j,k) is a tensor of the type Aj, By hypothesis
A(a,j,k)§* is a tensor of the type Aj; hence its transform B(«,j,k)y* is
given by the rule
Ox’ Oy’
B(«,(a, J;j, k)q ky? = ——
dy” Ox? {A(A, B, yé :

where

(x) = =y 1°).
a a

Inserting this expression for &* in the right-hand member of the above
formula and transposing all terms on one side of the equation yields

Ox* Ox’ Oy? |

|Be. i.) - — — A(A, B, ea),

However, 7*(y) is an arbitrary vector; hence the bracket must vanish, and
we obtain

Bia, j, k) =

This is precisely the law of transformation of the tensor of the type 4%.
Clearly, we can state an analogous theorem in which the vector § is a
covariant vector. For example, if A(i, /, k, «)é, is known to be a tensor of
the type 4j,, for an arbitrary vector é,, then A(i, j, k, «) = Ay. On the
hand, if A(i,j,k,«)é&, = A’, then A(i, j, k, ©) = A”, These
other
to
expressions suggest that an algorithm of division can be employed
determine the tensor character. Thus let A(i, j, k, a)&. = Aix, and write
symbolically.
A(i,j,k, «) = Axe ;
é
 TENSOR THEORY [CHapP. 2
68
the
Now, if we should regard the covariant quantities appearing below
division line as contravariant when written above the line, we have
A(i, j,k, %) = Ai,é* :

where & is the symbolic reciprocal of &,. From the product Ai,é* we see
that A(i, j,k, «) = Ajz. Similarly, if AG j,k, #)5, = A®™*, then
tik
— AiskEa giska
A(i,j,k, «) =
-
On the other hand, if A(a,j,K)E* = Ai,

In the division algorithm the contravariant quantities appearing below the

division line are to be regarded as covariant when written above the line.
THEOREM II. Let {A(i,,..., i,)} be a set of n" functions defined in the
X-coordinate system, and let {B(i,, . . . , i,)} be the corresponding quantities
in the Y-system. If, for every set of vectors with components &,, relative to
the X-coordinates and ,. relative to the Y-coordinates, we have the equality
B(B, a nebo ec ny. = A(a, = sae ls honk

(that is, the inner product is a scalar), then the set offunctions A(iy, ... , i,)
represents a contravariant tensor of rank r in the X-coordinate system.
Proof. Since the €, are the components of a covariant vector,

E(9) = dy? (3)

ae Oxi Bi

Therefore

[BOB By) = Alaa 5) SE EY nf =0. ey OY (1) (r)

However, mbes ..., § are arbitrary; hence the term in the bracket must
vanish. Therefore
f] dy?
Bb... ., 8) = oo = (x, ep te),
Ox! Ox
which shows that
Alogi. 1G ajpmderos
This particular form of the quotient law is taken by some authors as
the definition of the contravariant tensor of rank r. Thus, if the multilinear
form A(o,...,¢,)6) +++ € is an invariant, then A(o%,,..., %,) =m
A%"’*°r, provided that the €, are the components of arbitrary vectors.
|
SEC. 28] RELATIVE TENSORS 69
On the other hand, if A(a,,..., aE Cd) tee a is an invariant, for an
arbitrary choice of &’s, then
Aly, .- +5 %) = Ag, ...0,°
r >
It is obvious from proofs of Theorems I and II that many other quotient
laws can be stated. For example, if the inner product A(i, «)&,; of the set
of n? functions A(i,j) with an arbitrary tensor is a covariant tensor of
rank two, then A(i,j)represents a mixed tensor of the type 4/. The reader
can prove this fact by following the pattern used in proving Theorem I.
The tensor properties of the set A(i,7)may be surmised from the division
algorithm. Thus, if A(i, «)&,; = ~V%;;, then AQ, a) = i Now if we
i. ag :
write the symbolic reciprocal of &,; as &”, we have A(i,a)=—=- =
6° — Az, Sad

27. Symmetric and Skew-Symmetric Tensors

When an interchange of two covariant (or contravariant) indices in the

components Ar! (2) of a tensor does not alter the value of com-
ponents, the tensor A is said to be symmezric with respect to those indices.
For example, a covariant tensor 4;,(x) is symmetric if A,(x) = A,,(2).
The definition of symmetry of tensors obviously would not be satisfactory
if the symmetry of its components were not preserved under the coordinate
transformations. To see that this is indeed so, let us suppose that
Are) = Ajg,---i,(#)- Then Aj,,...4, — Aj,i,.--4, = 0. However,
the difference of two tensors is a tensor; and if a tensor vanishes in
one coordinate system, it vanishes in all admissible systems. Hence
Bi i,---ifY) = Byi,---i,Y)-
We may say that a tensor is skew-symmetric (or antisymmetric) with
respect to certain indices whenever an interchange of a pair of covariant
of
(or contravariant) indices in the components merely changes the sign
the components. The skew-symmetry of tensors is likewise an invariant
1s
property. The proof of invariance of the skew-symmetry property
However, as an exercise the reader
similar to that given for symmetry.
law of
may find it instructive to construct a proof based on the use of the
transformation of components 4}: .”.j*.
Sec. 40.
We will extend the notions of symmetry and skew-symmetry in

28. Relative Tensors

in the X-
We recall that a function f(z, ..., 2”) represents a scalar
ned by the
reference frame whenever in the Y-reference frame, determi
70 TENSOR THEORY [CHAP. 2
transformation 2 = w(y!,...,y"), the scalar is given by the formula
e(y!,...,y") =f[zy),...,2"(y)]. We will encounter functions f(z)
which transform in accordance with the more cake law, namely,
mals
(28.1) aad: bp ia borgCh) od ey) |=
i a
where denotes the Jacobian of the transformation and W is a

constant. We observe that, if the function f(x) transforms in accordance

with the law 28.1, then

We) =f) ||"

Mi Ww dy" Ww

= g(y)| =
where we have made use of Theorem II of Sec. 20. Thus the formula 28.1
determines a class of invariant functions known as relative scalars of
weight W.
A relative scalar of weight zero is the scalar defined in Sec. 21. Some-
times a scalar of weight zero is called an absolute scalar.
A relative scalar of weight 1 is called scalar density. The reason for this
terminology may be seen from the expression for the total mass of a
distribution of matter of density p(z!, 2, 23), the coordinates x’ being
rectangular cartesian. The mass contained in a volume 7 is given by the
integral M -{f{ p(x", 2, 2°) dx! dx? dx®. If the coordinates z* are
changed with the aid of the equations oftransformation x’ = x'(y', y?, y),
(i = 1, 2, 3), the mass M is given by theBee

m= foteon|= 5 |dy" dy* dy°

=| |aeJy y) dy’ dy? dy’.
It is clear that the density of distribution when referred to the Y-coordinates
a
0
is p(y) = p(x) E
=
We can also eneaiteh the law of transformation of components ofa
mixed tensor by considering the sets of quantities Ait <*:4(@) which
transform according to the formula
Ox" W Oyit vy poy
Onts Oar
(28.2) BIC R ee es ke ee eee
ay’ Catster Boren Dyt wena—
Dyke AZ)
tant
e ce
s(x),
SEC. 28] RELATIVE TENSORS 71

The sets of quantities Aj! F(x) obeying this law of transformation are
called the components of a relative tensor of weight W.
From the discussion in Sec. 24, and from the transitive property of
Jacobians, namely, »

dx'| | dx ||dy"
’

az ay! dz?

if follows that the transformation 28.2 is transitive. In addition, from the

linear and homogeneous character of this transformation it follows that
if all components of a relative tensor vanish in one coordinate system,
they vanish in every coordinate system. An immediate corollary of this
is that a tensor equation involving relative tensors when true in one co-
ordinate system is valid in all coordinate systems. In this case the relative
tensors on two sides of equations must be of the same weight.
A little reflection will convince the reader that

(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.
being
(b) Relative tensors may be multiplied, the weight of the product
the sum of the weights of tensors entering in the product.
yields a relative
(c) The operation of contraction on a relative tensor
tensor of the same weight as the original tensor.
sections, from
To distinguish mixed tensors, considered in the preceding
used to designate
relative tensors, the term absolute tensor is frequently
in applica tions of
the former. We shall encounter several relative tensors
tensor theory.
Problems
B’* is an arbitrary symmetric
1. Given the relation A(i,j,)B”* = C', where if
Hence deduce that,
tensor. Prove that A(i, j,k) + A(i, k,j) is a tensor.
A(i,j,k) is a tensor.
A(i,j,k) is symmetric in j and k, then
the relation A(i,j,k)B* = C?, where B’* is an arbitrary skew-
2. Given Hence, if
tensor.
symmetric tensor. Prove that A(i, j,k) — A, k,j) is a
then A(i,j,k) is a tensor .
A(i,j,k) is skew-symmetric inj and k, and a(i,/) is
arbitrary vector dx‘,
3. If ali,j)dx’ dx is an invariant for an
4;;.
symmetric, show that a(i,/)is a tensor
cofactor of a,; in |a;j| divided by
4. If a;; is a tensor, show that A¥, the
|a;;| # 0, is a tensor.
{224/0x* @x7} is a tensor with respect
5. If d(x}, ..., 2”) is a scalar, show that
coordinates.
to a set of Jinear transformations of 0
6. If |a;; — 4b,;| =O for’ = 4, in one set of variables, then |a,;’ — 4b;;,|=
In other words, the roots of the polynomial
for A = 4,, in the new set of variables.
|a,; — Ab;;| are invariants.
ip) TENSOR THEORY [CHAP. 2

7. Prove that a tensor with skew-symmetric components in one coordinate

system has skew-symmetric components in all coordinate systems.
8. Show that every tensor can be expressed as the sum of two tensors, one of
which is symmetric and the other skew-symmetric.
9. Show that the tensor equation a,‘A; = «4;, where « is an invariant and 4;
an arbitrary vector, demands that a;* = 6;*«.
10. Prove directly from the law of transformation of components that sym-
metry of a tensor is an invariant property.
11. The square of the element of arc ds appears in the form
ds? = g;; dx* dx’.
Let T be an admissible transformation of coordinates z* = x*(y!,..., y"); then
ds? = h;; dy dy’. Prove that |g,;| is a relative scalar of weight two. Hint:
au% dah PEA aes :
h;(y) = By ay Sap(X)s and recall the rule for multiplication of determinants.

12. How many independent components are there in a skew-symmetric tensor

of rank two?
13. If a,; is a skew-symmetric tensor and A? is a contravariant vector, then
a;j;A'A’ = 0.
14. Prove that, if A(i, j, k)A’B’C, is a scalar for arbitrary vectors A‘, B’, and
C;,, then A(i, j, k) is a tensor.

29. The Metric Tensor

In Sec. 5 we introduced the idea of n-dimensional space E,, by extending

the concepts familiar to us from our experience with ordinary Euclidean
geometry. Thus, in defining the length |x| of a vector x, we used the
generalized formula of Pythagoras, |x| = Vx'x', where the x‘ are the
components of the vector x referred to a set of orthogonal cartesian axes.
(See Sec. 5.) If we now consider a displacement vector dz’, (i = 1,...,n),
determined by a pair of points P(x) and P’(~ + dx), wherein the coordinates
x are orthogonal cartesian, the formula of Pythagoras gives for the
square of the distance between P and P’ the expression
(29.1) GS) Use, at 1 em):
We shall call ds the element of arc in E,,.
A change of coordinate system, determined by the transformation
(29.2) ae hat CT Be ohh

permits us to write the formula 29.1 as

recy
29.3
(29.3) = ay ay?dy* dy?,
SEC. 29] THE METRIC TENSOR %

since dx’ = (dx'/dy*) dy*.. We can thus write the formula for the square of
the element of arc in the Y-reference frame as a quadratic form

(29.4) AS mie pkey .

where the coefficients g,,(y) are defined by

(29.5) i
Sap(¥)
ant Bat
= eae:
ay® oy? 4

These coefficients are functions of the variables (y’), and they are obviously
symmetric with respect to the indices « and f.
Since the square of the element of arc ds is an invariant, we conclude
(see Problem 3) that the set of functions g,,(y) represents a symmetric
tensor. This tensor is called the metric tensor, because, as will be shown
in Chapter 3, all essential metric properties of Euclidean space are com-
pletely determined by this tensor.
We have obtained the formula 29.4 by starting with expression 29.1,
which is characteristic of the Euclidean space. A transformation of co-
ordinates 29.2 clearly does not alter its metric properties, and formula
when
29.4 simply enables us to calculate distances in the Euclidean space
form 29.1 and
it is covered by a coordinate system Y. By starting with the
functions 29.2
the transformation 29.2, we have shown that the set of n
in which
satisfies a system of 4n(n + 1) partial differential equations 29.5,
if the functions
the g,,(y) are known functions of the variables y. Now,
+ 1) partial differential
8a, are specified arbitrarily, the system of 4n(n
in general, will have no
equations 29.5 for 7 unknown functions x‘(y),
29.5 has a solution,
solution. In the event the g,,’s are such that the system
which reduces the quad-
the existence of a transformation of coordinates
assured. In that event the
ratic form 29.4 to the sum of squares 29.1 is
. If, on the other hand,
metric tensor g,, defines an Euclidean manifold
29.5 has no solution, then no
the functions g,,(y) are such that the system
exists which reduces the ex-
admissible transformation of coordinates
to the Pythago rean form
pression 29.4 for the square of the arc element set of
is non-Eucl idean. A
29.1. We shall say then that the manifold s Sale
integrabi lity of equation
necessary and sufficient conditions for the
will be deduced in Sec. 39.
that our tensors are defined
We suppose in the remainder of this chapter atic
ds is given by the quadr
in metric manifolds and that the element of arc
gj; are funct ions belonging to the
form ds? = g,,(x) dx‘ dx’, where the is such that
r g,,(x)
class C!. We also assume that the symmetric tenso
discussion, but do not assume
lg,;| % 0 at any point of the region under
dean.
that our manifold is necessarily Eucli
74 TENSOR THEORY [Cuap. 2
Problems

1. Let E, be covered by orthogonal cartesian coordinates x’, and consider a

transformation
xi = y'sin y? cos y°,
22 = y! sin y? sin y°,
2? = y! cos y7,
where the y* are spherical polar coordinates (y' = r, y2 = 6, y? = ¢). What are
the metric coefficients g;;(y)? .
2. Let E; be covered by orthogonal cartesian coordinates x*, and let
a) = yf COSY”,
2" = y* sin y”,
a = 3
represent a transformation to cylindrical coordinates y’. Find the expression
for ds* in cylindrical coordinates. . é *
3. Let E; be covered by orthogonal cartesian coordinates x’, and let x’ = a;‘y’,
|a;*| #0, (i,j =1, 2,3), represent a linear transformation of coordinates.
Determine the metric coefficients g;;(y). Discuss the case when the transfor-
mation is orthogonal.

30. The Fundamental and Associated Tensors

Let g,,(x) represent a symmetric tensor such that the g;,,(x) belong to
class C* and g = |g;;| # 0 at any point of the region. We construct, with
the aid of the set of functions g;,(x), a new set of functions g(x), repre-
senting a contravariant tensor, which is such that g’g,; = 6). The tensors
g,;(x) and g(a) will play an essential role in all our subsequent consider-
ations, and for that reason they will be called the fundamental tensors.
Let us form a set of n? functions
Aeeeliis
(30.1) g(i,j) = < ,

where G” is the cofactor of the element g,; in the determinant g. The

notation used in the definition 30.1 anticipates that the g(i,j) form a
contravariant tensor, and, indeed, we will prove that they define a sym-
metric, contravariant tensor g’. The symmetry of the set of functions
g(i, j) follows directly from the observation that the determinant obtained
by deleting the ith row and the jth column in a symmetric determinant
8i; has the same value as the determinant obtained by deleting the jth
row and the ith column. We prove next, by means of a quotient law, that
the g(i,j)’s transform according to a contravariant law. We first note
that, if € is an arbitrary contravariant vector, then

(30.2) & = g,:6

SEc. 31] CHRISTOFFEL’S SYMBOLS ie
is an arbitrary covariant vector, since |g,;| # 0. Now, if both sides of the
formula 30.2 are multiplied by g(6, i) = G”'/g and summed on i, we get

(30.3) e(B, i); = ca

Seis 2 ‘:

However, by (7.4), G”'g,; = g6%, so that (30.3) can be written as

a g(B iE, = &.
Since &, is arbitrary, we conclude from Theorem I of Sec. 26 that g(f, /) is
a contravariant tensor of rank two. We can thus write (30.1) as
GY
(30.4) a
g
The reciprocal relation g”g,; = 6; follows directly from the fact that
Gg,; = O,g. Incidentally, we can conclude that the set of cofactors GY
represents a contravariant tensor of weight two. This follows from
Problem 11 of Sec. 28, where it is indicated that the determinant |g,,| is a
relative scalar of weight two.
A tensor obtained by the process of inner multiplication of any tensor
Aix. with either of the fundamental tensors g,; or g” is called a tensor
associated with the given tensor.
As an illustration of this definition consider a tensor A,,, and form the
following inner products: g* Aj, = A%, 8” Ain = Aim 2g Ain, = Aj.
All these tensors are associated with the tensor A,,,. Operating on these
tensors with g” again, we can form other associated tensors. It will be
observed that the operation of inner multiplication of g,; with any tensor,
say A’! lowers the index with respect to which the summation is performed.
Thus g,,4% = Ail, while g?Ait = Ak The procedure of raising and
lowering indices is clearly reversible. In the foregoing formulas the
position occupied by the raised (or lowered) index is indicated by a dot.
In general, such systems as g"“A,, = 4;. and gA,; = A‘, are different.
They are identical whenever 4,; = 4j;.
as
It is possible to interpret all tensors associated with a given tensor
t referenc e frames. This inter-
representing the same tensor in differen
its associated
pretation is particularly simple for the covariant vector A,, and
is Euclide an. We will return to
vector g’“*A, = A’, whenever the space
this matter in Sec. 45.

31. Christoffel’s Symbols

derivatives
We introduce in this section certain combinations of partial
will prove useful in the develo pment
of the fundamental tensor g;,(x), which
76 TENSOR THEORY [CHAP. 2

of the calculus of tensors. Let us construct a set of functions denoted by

the symbol
cs Lfog;, . 02, 28) lip _
CUA
Sl oll NK IES=} arte
hey —Sik
_ —ott). (i,j
Bafeelese Lats 5 ces Ta)

and call them the Christoffel 3-index symbols of the first kind. The set of
functions

(31.2) Lil=o li,2),

where g** is the contravariant tensor, constructed with the aid of the g;,’s
in the manner described in the preceding section, are the Christoffel 3-
index symbols of the second kind.
Evidently there are n distinct Christoffel symbols of each kind for each
independent g,,, and, since the number of independent g;,,’s is 3n(7 + 1),
the number WN of independent Christoffel symbols is N = $n*(n + 1). We
proceed to deduce several properties and identities involving Christoffel’s
symbols, which will prove useful to us in the sequel.
It is clear from definitions 31.1 and 31.2 that the Christoffel symbols
are symmetric with respect to the indices 7 and 7. Thus

(31.3) [j,k] = [ji, k],

and
(31.4) ‘|Ew
i) iJ
We see from the defining formula 31.2 that we can pass from the
k
symbol of the first kind [i, «] to the symbol (‘|by forming the inner
Y
product g*[ij, x]. Now, if we multiply equation 31.2 through by Lip»
and recall that g,,2"* = 63, we get

(31.5) Bis‘4ew
Formulas 31.2 and 31.5 are easy to remember if it is noted that the op-
eration of inner multiplication of [ij, «] with g** raises the index and
replaces the square brackets by the braces. The multiplication of “|by
Y
8x On the other hand, lowers the index and replaces the braces by the
square brackets. Formally, these operations of multiplication by g** and
8x are analogous to raising and lowering the indices on tensors, but we
will see that the Christoffel symbols, in general, are not tensors.
SEC. 31] CHRISTOFFEL’S SYMBOLS at
From (31.1) we readily deduce an expression for the partial derivative
of the fundamental tensor g,; in terms of the symbols of the first kind.
It is
(31.6) 0g;; ake ea) =
ago [ik,j] + Lk, i],

which can also be written as

08;; {a | {|
(31:7) t= g,, + g.;
=v rai ye sees
if we note (31.5). An analogous formula for the partial derivatives of the
contravariant tensor g” can be obtained by differentiating the identity
£42" = 6! with respect to x". We get
O8ia ni £3 dg”!
aur ©
—— e
BH Gat
ta = 0,

or
g dg” a) OSin
* Oak Oxk™ -

To solve this system of equations for dg%//dx* we multiply both sides by

g’’ and get
; 0g” ¢ : 0g;
= iB a5 ia
8 ip Siz = Se Se ©
ox* Ox*
Since g'’g,, = 0%, we have
dg’? adn ’
Zia = —g*g([ik, «] + [ak, i]),

where we made use of the formula 31.6. Noting the definition 31.2, we
have finally
she ee ta
Arena ile Pak)”
which is the same as
31.8

PY) ed
TR,

Ox* e
aL

ak)»
—

ied
g%”

al
O3
;

of the formula for the

_ We conclude this section with a derivation
will be useful to
derivative of the logarithm of the determinant |g,,|; this
ce of a vector field, as
us in writing a compact expression for the divergen
well as in several other connections.
obtain
The determinant g = |g,;| can be expanded by minors to
summation on i or n),
lola gain te ee0 ft °° ZinG'", (no
= g,G%, (sum on « only, i fixed),
78 TENSOR THEORY [CHaP. 2

where G“ is the cofactor of the element g;;. Since the g,,’s are functions of
av, ..., a”, the G’®’s are also functions of the same variables. From (31.9)
we deduce that

dg (ZinG")

08; 08%

= gi, oe - Gix Bi (sum on « only, i fixed).

083; 08;
Since G* contains no g;;, 0G'“/dg,; = 0, and since the g,,’s are independent
variables in this formula, 0g,,/0g,;; = 6). Thus

28 _ G49} = Gi
But
Og _ 98 O8sg _ cap 8x
O* Ogi, Ox! Ox*
and, if we recall that g*? = G**/g, the foregoing formula becomes

og 0g,
—,
Ox
= ggtt Eel.
Ox

If we now insert for dg,,/0x' from (31.7), we get

rs) )
S = at (”|+ ya fa)
x al Pl),

= 29 ("|:
xi)

Therefore we can write — og = ay and hence

22 O2x* lial

(31.10) eee Jg= Ay

Ox’ 1x

We close this section with some remarks about different notations used
for the Christoffel symbols by various authors. The notation [ij, k] for the
symbol of the first kind is fairly universal, but there are several different
k
notations for the symbol til Thus, many writers use the symbol {ij, kt.
SEC. 32] TRANSFORMATION OF SYMBOLS 79

P. Appell, in Traité de mécanique rationelle, vol. 5, uses el for the symbol

ij
of the first kind and for the second kind. The followers of the Prince-

ton school generally use the symbol If, for the symbol ("|adopted in this
y
book. Although the notation It has some advantages, it suggests that
the symbol of the second kind is a tensor. This, however, is not always
true, as will be seen from the developments of Sec. 32.

Problems

05; Bix ae 2
1. Show that onk ~i Oat = Ljk, i) = lij, ae

2. Show that, if g;; =0 for i #/, then (|= 0 whenever i, j, and k are
distinct. 4
3. Show that, if g;; = 0 fori #/, then

iyial 2 fine lo i 1 agys

li = 5 ppl OBS | coda Der ant ”
| ae) ant log Zi:

where we suspend the summation convention and suppose that i # /.

4. If|z;;| 4 0, show that

a (Bp Fk B , f
raw 4 ear ae a 4 Candee age
5. If y’ = a;‘«’ is a transformation from a set of orthogonal cartesian variables
y* to a set of oblique cartesian coordinates x* covering E3, what are the metric
coefficients g,; in ds* = g;; dx‘ dx’?

32. Transformation of Christoffel’s Symbols

We have already remarked that the Christoffel symbols do not, in

general, represent tensors. In this section we deduce the laws of trans-
k ;
for the sets of functions [i,k] and bi under coordinate
formation
to
transformations y‘ = y(z!,..., 2"), which will from now on belong
the class C2. The functions g,,(x) are assumed to belong to class C1, and
their transforms to the Y-coordinate system are denoted by the symbols
h,,(y), so that
ax” dat
(32 vi ) ad marr ap:
ti)
 80 TENSOR THEORY (CHAPS
Let us construct the Christoffel symbols ,[ij, k], where the index y signifies
that they refer to the Y-coordinate system; then

(32.2)
BS Je a oy' eh
Differentiating (32.1) we get

Tp
aye
A
(atearg AE Ny aM
“*\Oy* dy! dy? — Oy" dy? dy'
dy’ dy’ dy" Ox?
Since g,5 = Zs., we can interchange the dummy indices « and f in the
second term within parentheses and obtain

Oh;; (Ox" Ox! 07x" dx*

Ox? Ox" O84
aye? \ayF ay! dy’ | dy® dy? dy’) | dy! dy! y* Ox” |
The partial derivatives 0h;,/0y’ and 0h,,/y’, entering in (32.2), can be
obtained from this formula by a cyclic permutation of indices, and the
substitution in (32.2) yields
Ox" ax? Ox? Cn oe
(32.3) ylij, k] = ele
ke B, Valens F 5 See Saps
oy’Oy? Oy dy’ Oy’ Oy
which shows that [«f, y] is not a tensor unless the second term on the right
vanishes. The second term will vanish identically if the coordinate trans-
formation is affine, that is, if y’ = c,‘x’ and the c,’’s are constants.
Similarly, we can easily show that the Christoffel symbols of the second
kind, are not tensors in general. Indeed, we note from formula 31.2 that

|7 = i Li, uh
where
he = oy" Oy" 60
On Ox"
If we multiply (32.3) (with k replaced by «) on the left by h*“ and on the
right by its equal from the formula written just above, and simplify, we get
k dy" Ox* Ox? gf? 4 oy 07.2% ob
lol Ox? dy’ ay?5 BalP y ay dy’ dy’ aL
Thus

(32.4) peers
Ke B a

wij Ox? Oy’ Oy’ zlaB) dy’ dy’ dx*

which shows that the symbols of the second kind are not tensors unless the
coordinate transformation is affine.
Sec. 33] COVARIANT DIFFERENTIATION OF TENSORS 81

The system of equations 32.4 can be solved for 0?x%/dy' dy’ as follows.
Multiply (32.4) by dx™/dy*, sum with respect to the common value K= 1;
and obtain

f o
vlij a
) t o n u t ( | ee B o a Oy! Oy? dy” dar

Since dx™/dx° = 6” and dx™/dx* = 6%", this expression yields

(32.5) ora” es fy\ on™ {f Oac* Oat

dy dy? ij) dy” —la) dy? ay’
Obviously y and x can be interchanged, and it follows from (32.5) that

(32.6) pais [>|oe= [| a oe

dx' dx? _, ij) Aa” lap Ox' Ox

The important formulas 32.5 and 32.6 were first deduced in an entirely
of
different way by E. B. Christoffel in a memoir concerned with a study
of these
equivalence of quadratic differential forms.® We will make use
formulas to define the operations of tensorial differentiation.

33. Covariant Differentiation of Tensors

0
derivatives = :
We have observed, in Sec. 22, that the set of partial x

a covari ant vector, since

of a scalar function f(z',..., a”), represents
, 0 (a
ae) of the
of = ru “as_ But if we form the set of partial derivatives oy) \oy’
Oy? dx* dy’ 3
covariant vector = , we get
oy

ipdy?a
dyi dy’ \dx" dy"

__ ofOx? dah ay |af dy’dat

dy’
~ 92% dy! dy’ aa"
0 ae
Zap%

of the term = ~~ | » shows that the set

which, because of the presence Oa Oy' OY?
: 2
form according to a tensorial
of second derivatives dy’ dy’ does not trans
the set of partial derivatives
law. It follows from this example that
(1869).
8 B. B. Christoffel, Crelle Journal, 70
82 TENSOR THEORY [CHAP. 2

of a covariant vector, in general, is not a tensor. Indeed, if we have

a covariant vector A,(x), then

By) = ue Ay,
oy’
and
OB, ex’ Ox CA, C7 a"
te aif ey dit a Onan
33.1 SS YH _——

so that the derivatives ofa vector do not form a tensor unless the coordinate
2 %

transformation x? = 2'(y) is affine. If we insert in (33.1) for : from

dy’ o7]
the Christoffel formula 32.5, we get

OB; _ 0x* Ox! 0A, |= A _ [a ) ax? ax?

dy? dy’ dy? ax ylijlay’ * clyB) dyidy? *
%
; av :
Since —— A, = B,, we have on rearranging
oy”

OB; a (4s {y |!\ 02% Ox"

332 —- B, = |—— — A ae
Coe) dy’ ip Ox? zlop )dy’ dy?
, 0A, a
from which it is clear that the set of n? functions Fai _ 1"4a obeys
se Tae ae
the law of transformation for a covariant tensor of rank two. This leads
us to formulate a
0A,
DEFINITION 1. The set of n? functions oe — "4. defines the covariant
- Y
a’ derivative (with respect to g;;) of the covariant tensor A,.
We denote the covariant 2’ derivative of A, by the symbol 4;;. Thus

(33.3) A, = oa — |,
Ox? lij

It should be noted that in order to compute the covariant derivative

it is necessary to have the set of Christoffel symbols; that is, the funda-
mental tensor g,; must be given in advance.
Similarly, if we start with a contravariant vector A*, and differentiate
f oy’
the relation By) = aa A(x), we obtain

QB! ss At ax? ay! en Olyl giles

dy? dax® dy’ da* Ax% Aa? Ay?”
Sec. 33] COVARIANT DIFFERENTIATION OF TENSORS _ 83
and making use of the formula 32.6, we find
OB" fi i \3 — et aA
a Oy’
oy’ yj dy date

Thus the set of n? quantities A(i, /) = at | I \a forms a mixed tensor

|wl
of rank two. Accordingly, we Cate a
: 0A’
DEFINITION 2. The set of n® functions —— + 4 represents the
Oxi *)
covariant x’ derivative (with respect to g,;) of the contravariant tensor A’.
We denote the covariant 2’ derivative of the contravariant tensor A’ by
the symbol 4',. Thus
(33.4) dette Acs |iY ge
The definitions 33.3 and 33.4 can be extended, in an obvious way, to
mixed tensors. Thus we define the covariant 2’ derivative (with respect to a
given tensor g;,) of the mixed tensor 4/1:::4s by the formula
Ek oe |
= CAs
(33.5) a oa
sje
aa!

— ie|e ie (?km
i Se He Agee
Ay A Site a ia oy ae
as [PlagesFe teAve. ee) [eilate ce :
al] l all
A verification of the fact that the set of functions Aj!’!:/s (~) forms a
tensor of the type indicated by the indices presents no difficulty.
If A is a tensor of rank zero, we define its covariant derivative to be the
ordinary derivative. Thus 4, = 0A/dx'. This definition is consistent
with the formula 33.5. We We note that, if the g,,’s are constants, the
Christoffel symbols vanish identically, and hence the covariant derivatives
reduce to the ordinary derivatives. This will surely be true if the gj 8 are
the metric coefficients of an Euclidean space covered by acartesian reference
system.
We remark in conclusion that the covariant x’ derivatives of relative
tensors are defined s follows. If f(x) is a relative scalar of weight W,
at W
so that g(y) =f@ |= , then

(33.6) LF sa5 TAMA

84 TENSORS THEORY [CHAP. 2

This set of functions represents a relative vector of weight W. If ADS ide

is a relative tensor of weight W, then its covariant x’ derivative is a relative
tensor of ee W, determined by the formula

Problems

1. Prove that the following expressions are tensors.

ag aAv i Ax J Aix
(a) ai aa! Bs al ie al :

(b)
ee
A; 1 Ox!
ey,
jl a + iin
:
0A,, (xil Aa; [x
(c) Ai) = aa! a = (jl A,

(d) 0Ai 5, el oe J (
7 = ae _ en <7 "hte = Tr
Liba 2
n + fi
Ava

k k
2. Prove that " = i are components of a tensor of rank three, where
ui
‘iand ‘|are the Christoffel symbols formed from the symmetric tensors
y by
a;;(x) and b;,(x). a | ay ay* ax® | ayt
3. Use the the form
formu lanG fe = ———~-,
at ax —_}
By |—
Ba and the law of transfor =

mation of relative aes of weight W to deduce formula 33.6.

34. Formulas for Covariant Differentiation

It is easy to deduce from the structure of formula 33.5 that the rules for
covariant differentiation of sums and products of tensors are identical
with those used in the ordinary differentiation. Indeed, if Ay: ‘4s(a) and
Ai 1 fe(a) are two tensors, then the formula

(Apis fe Ma ot g eal tat) aera tre

follows directly from inspection of (33.5). To prove that the derivatives

Sec. 34] FORMULAS FOR DIFFERENTIATION 85

of the outer and inner products are given by the familiar rules,

Tas G7 istee any ore Geta Ou

Ga 28> Dis fSeri:”* Je) soon
aa, ee cack =a ee pe aga ae ig
Mee ee “Bitep tal Ags.
«

Fy °* Fe—-10 J 5s41° °° 3 = Fa Is=1% of latte 2

CAS. tr ot1’
oe tpt ioe
* *tw—-14/,
ae eae ewGS drt
is+ *tw—18
hes

Tee aie ae ++
andyp—10,19

we need only insert for A in formula 33.5 the product 4.7. We illustrate
the procedure by considering the product Aid, , = Up. We have

gy = out — {a \gpinia a Kejag

ijig,l Aas li aig iol aya

4 2 ats - f ‘ats
al al

== Atta (Sus
= |. \ li
Ox! i,U) i he|
ial

oA'1? 4 Avie {|gin1a)

12
A, (—— al
rs wi Ox! a lal
—_ 5132 of Jide .
: 2D in, + A, iA
—) A

As an exercise the reader may show

This establishes the desired result.
that
1 aa A jq AS" a Aj, ‘
(Aj)

of covariant differentiation and

He can also show that the operations xy
contraction can be permuted.
g that in covariant differentiation
We conclude this section by remarkin have
constants. Indeed, from (33.5) we
the Kronecker deltas behave like
00% al we i
fase! cae OLE | 5%I

no (ies
a Ox! (1 al

Problems
to
raction of indices A%, is equivalent
1. Note that the operation of cont cont ract ion can
show that the operation of
multiplying 47; by 6/. Using this, .
r eithe r befor e or after covariant differentiation
be performed on a tenso of indic es can be perf ormed
ng or lowe ring
2. Show that the operation of raisi
t differentiation.
either before or after covarian
86 TENSOR THEORY [CHAP. 2

35. Ricci’s Theorem

We will show in this section that the fundamental tensors g,; and g”
behave in covariant differentiation as though they were constants. This
follows from
Ricci’s THEOREM. The covariant derivative of either of the fundamental
tensors is zero.
Proof. Consider first the tensor g;; and form

Sigs aes
ag., mya). (o 1
we efra\ oe
Ox! lil jl
The right-hand member of this expression vanishes identically by virtue
of (31.7), so that g;;, = 0.
We can perform a similar calculation for the tensor g”, but it may prove
more instructive to differentiate the inner product g‘*g,; = 6. Thus
81805
ied
+ 8"ix 8.5, —= 3Sin
since 6;, = 0 and g,;; = 0, we have
. q baie —

SiS. = 0.
However, since |g,;| 4 0, the only solution of this system of homogeneous
equations is gi = 0.
As an immediate corollary of Ricci’s theorem we note that the funda-
mental tensors may be taken outside the sign of covariant differentiation,
and hence the operations of lowering and raising indices are permutable
with covariant differentiation. Thus
(Lai At.) = 24:1 Ajit

36. Riemann-Christoffel Tensor

We recall that a sufficient condition for the equality of mixed partial

cies Ou 2
derivatives ——— and ——— of a function u(x, y) is that u(a, y) be of class
Oa Oy Oy Ox
C?. We will assume henceforth that the tensor components under con-
sideration belong to class C*, but this restriction alone, as we shall see
presently, is not sufficient to insure the equality of mixed covariant
derivatives. Indeed, it will be shown that, if the order of covariant differ-
entiation 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 g;,’s,vanishes. This tensor, knownas the Riemann-Christoffel
tensor, plays a basic role in many investigations of differential geometry,
dynamics of rigid and deformable bodies, electrodynamics, and relativity.
Sec. 36] RIEMANN-CHRISTOFFEL TENSOR 87
The covariant derivative of a tensor is a tensor; hence it can be differ-
entiated covariantly again to obtain a new tensor. This tensor is called the
second covariant derivative of the given tensor.
Consider the covariant x’ derivative of A; with respect to g;;,°

(36.1) Asa9A;= fa|

ee

Now, if (36.1) is differentiated covariantly with respect to x*, there results

the tensor

(36.2) A,
=»Aes " Ay,4— Mel

a6) han 24 (G)- GIR L2

Velie ~ Vad”)
Carrying out the indicated differentiation in (36.2) and (36.3) yields

a, _ 7
9 oO

a) OA, a) 0A,
(36.4) axl =
Aji. = ax Ba} aa" ieee LalOak = fe xi

OL a\ 0A; a) fy
4 (NE < lsBatt 3 ae
OL
8A, a il ec) Svea 24s

+f) —olse-
CO a Ox’ Ox Qa! Ae — ff ax? (stox"

If we subtract (36.5) from (36.4), we get

a afik
a

a) (p ow i
ah ri s A, - Ata Ay,
Asn —Avs le ae Ox* ~* ‘ife ash Ox?
88 TENSOR THEORY [CHAP. 2

and an interchange of « and f in the first terms of each preceding line gives
3 w al;

i i (P| (*\ (A [ + .
(36.6) Aj ix a Ain; =
avi ssax® ~——«CNik) lpi) Lad pad
<i: =- =< f = a

Since A, is an arbitrary covariant tensor of rank one, and since the

difference of two tensors A; ,; — A; ,; is a covariant tensor of rank three,
we know by the Quotient Theorem I of Sec. 26 that the expression in the
bracket of (36.6) is a mixed tensor of rank four; that is,

AiG)
ind 2s
ax? Gels
Ssax®* | Niki lai las) Waa
Furthermore, if the left-hand member of (36.6) is to vanish, that is, if the
order of covariant differentiation is to be immaterial, then

hea
since A, is arbitrary. In general, however, Ri, # 0, so that the order of
covariant differentiation is not immaterial. It is clear from (36.6) that a
necessary and sufficient condition for the validity of inversion of the order of
covariant differentiation is that the tensor R¥, vanishes identically.

22) 1 1 pam
The tensor \

Ox® daz! ak! \al/ -

(36.7) Ri.,=

scalvaueealtateet
is called the mixed Riemann-Christoffel tensor or the Riemann-Christoffel
tensor of the second kind.
The associated tensor

(36.8) Rij = SiaRFa

is known as the covariant ~Riemann-Christoffel tensor, or the Riemann
Christoffel tensor of the first kind.
It is not difficult to verify that the defining formula 36.8 for R;,,, can be
written in the convenient determinantal form

0 0 a a
(36.9) Rim =| O2* at |+ a 4
TG i) GLa] |tik e) file]
which will be found useful in listing properties of this tensor in Sec. 37.
SEC. 37] RIEMANN-CHRISTOFFEL TENSORS 89

We remark in conclusion that formula 36.6 is a special case of an

identity, established by Ricci, which we record here without proof,
although the nature of proof is quite clear from the proof of the case
treated previously. This identity reads .
m
h
=
45 +9 *dg—1higs1° °° ities:
A,, ++ ttm, dk a A; Sts
a=1

In the special case of a tensor of rank two it assumes the form

ax a
—<
Ay = Aisi = A Rix AR Ay Rine
Problems
1. Show that
ee O e Ae ee
Rist = 5x tft, a [jk, i] + ice fi, e)= Z [ik, «].

2. Show that

dx) oak | Oat dx! aati dat dart aurk

+ g*(Ljk, BI, «1 — Lil, Pilik, «)-
3. Using the formula of Problem 2 show that
Riga = —Rixt = —Rijye = Revi

and Rijnr + Rinrs + Rirsx = 9-

to
4. If is a scalar, then g’’¢;; is a scalar and is equal

i
ee) dad

5. Referring to Problem 4, show that g“ ;;= 0 reduces to a2d/dx? dx? = 0

referred to a cartesian frame. This
when the g;; are the metric coefficients of E;
curvilinear coordinates has the form
implies that Laplace’s equation in general
aa eT = 0, since this is a tensor equation.
Laplace’s equation in polar coordinates
6. Referring to Problem 5, show that
has the form
1 a 2 ad 1 ad
a 1 ad en a +—coty—, =0
a tht es (y')? Oy”
y? ay?
( dy")? ED(y})? (ay?) x (y! sin y?)? (ay?)? |
Tensors
37. Properties of Riemann-Christoffel
d tensor R%,), We see immediately
From defining formula 36.7 for a mixe
metric with respect to the last two
that the set of functions Ri, is skew-sym
covariant indices. Thus
(37.1) Rix = —Rijx,
i =
and hence Ri,q)«) = 9:
90 TENSOR THEORY [CHaP. 2

We have defined the covariant tensor R;,,, by the formula

Rijn = Sickie
and, if we multiply this equation through by g’” and sum, we get
(37.2) Ria = 8°Rijn
so that the Riemann-Christoffel tensor of the second kind is obtained by
raising the first covariant index in the tensor R;;,,. To determine the
properties of the set of functions defining the Riemann-Christoffel tensor
of the first kind we expand the determinants in (36.9) and insert for
Christoffel’s symbols in the first determinant the definitions 31.1. We get
after a simple calculation the formula Sy =
a ‘ }

(73) Rig = 2Q\0x)

4(Zee, — OxSee0x — Sha Fhe)
Py. 2G. ake : oe al ~ ¢

Oz Nomex 6 ide ex
+ g(Lik, BILil, x] — (jl, Blk, «]),
from which it is obvious that

(a) Ria = —Ru-

(0) Rijn = Reser
(c) Rires = Riser:
(d) Ring + Rag + Rise = 0.
The last identity can be verified by direct substitution; by raising indices
we obtain an identity analogous to (d) for the mixed tensor Rj,;,

(e) Roa + Rij + Rijx = 0.

(f) The components of a Riemann-Christoffel tensor with more than two
like indices are necessarily zero. The identities (a) and (6) state that the
tensor R;;,, is skew-symmetric with respect to the first two and last two
indices, and the identity (c) signifies that R;,,, is symmetric with respect to
groups of first two and last two indices. It follows from these identities
that distinct, nonvanishing components of R,;,, are of three types:
1. Symbols with two distinct indices, that is, symbols of the type R isis
2. Symbols with only three distinct indices, which are of the type R tjike
3. Symbols R,,,, with four distinct indices.
It is now an easy matter to verify® that the total number N of distinct non-
vanishing components of R,;,, is N = n(n? — 1)/12.
® There are nm, = n(n— 1)/2 distinct nonvanishing symbols of the type R isids
—1)\(n-2 — 1)\(n — 2)n —
Ng =e? of the type Rix, and ny = Oe) of the
type Rizr
Sec. 38] RICCI TENSOR. BIANCHI IDENTITIES 91
jx;
In a three-dimensional space, distinct, nonvanishing components R;
1212,1313 , 2323, 1213, 2123, 3132, andin two dimensio ns
have the suffixes:
from the total of 24 = 16 components there is only one distinct non-
an
vanishing component-Rj.12. We will see that this tensor characterizes
extremely important property of surfaces.

38. Ricci Tensor. Bianchi Identities. Einstein Tensor

by
We define the Ricci tensor R;; by the formula R;; = Rij,» which,
virtue of (36.7), can be written as

Ree
a al
On Or i
fe) fal
(Bi) (Bal

fy fi]
ij) \ial|
|, (0Laced
| Ais)
a -
In Sec. 31 we have shown that ani log J/g = fa so that

Sonar aes | len by f [?)Pes

azide oat \Bj) lial laf) Oa
tensor R,; is symmetric.
From inspection of this result we see that the
nents of R;; is 3n(n + 1).
Since R;; = R;,, the number of distinct compo
if we set R,; = 0, we obtain
In a four-dimensional manifold n = 4, so that,
has adopted as his equa-
ten partial differential equations, which Einstein
space in the general theory of rela-
tions of the gravitational field in free
theory anothe r tensor, introduced by
tivity.1° In the development of that
This tensor is most readily obtained
Einstein, plays an important role.
from the identity

(38.1) Roem + Rim, x er Rimkt am 0,

due to Bianchi.
tensor g;; vanishes,
Since the covariant derivative of the fundamental
the Bianchi identity can be written in the form
(38.2) Rijxtm at Rijun,k ct Rinks = 0.
make use of the skew-symmetric
If we multiply equation 38.2 by g''g’* and
We get
properties of the Riemann tensor Rijn
o Karen, = £ hm,bin g Rim = O-
10 See Problem 2.
92 TENSOR THEORY [Cuap. 2

This result can be written as

i a
as (Ry 2 Rk am 0,

where R = g”R,,, or in alternative form

(38.3) CA m m
I =

where, RYm = o7*R, aie tensor

i St a

in parentheses in equation 38.3, is known as the Einstein tensor.

Problems
1. Show that R% 7, = 0.
2. If Rj; = aie p = R/n, where R = g”R;;. (The equation R;; = pgzis
is known as the Einstein gravitational equation at points where matter is present.
It corresponds to the Poisson equation V?V =p in the Newtonian theory of
gravitation.)
3. If = 2, show that Ry/911 = Reo/ $22 = Rio/ S12 = — Riailg-
4. If n = 3, the tensor R;;,, has six distinct components, and there are six
equations Rj; = g"R,;,,;. Prove that the solutions of these equations for Rj5x;
are given by

Reger = Skin + Seka — Pak — Poh ry Sik jl — Liki»

where R = g¥R;;.
5. Verify Bianchi’s identity 38.2.

39. Riemannian and Euclidean Spaces. Existence Theorem

Let the n-dimensional space V,, be covered by a coordinate system X.

We will metrize V,, by prescribing the element of are ds, so that
(39.1) ds? = g,, dx’ dv!
is a positive definite quadratic form in the differentials dx’. The functions
8i,(%) are assumed to be of-class C! in V,. The space V,, so metrized is
called a Riemannian n-dimensional space R,,.
We will now consider in some detail the following question: What
restriction must be imposed on the symmetric tensor g;;(x) so that there be a
coordinate system Y, defined by

it y* =a ee), (FOL, 2),

with y'(x) of class C* in R,,, in which the tensor g;;(%) has constant components
h,; throughout R,,?
Sec. 39] RIEMANNIAN AND EUCLIDEAN SPACES 93
This is one of the basic problems of differential geometry, which occurs
also under a different guise in dynamics, elasticity, relativity, and other
branches of applied mathematics. .
We note first that the components of g,,(x), when referred -to the Y-
frame, are given by
at Ab
(39.2) yeoy’ es
oy’
k Reg nd
If h,;'s are constants, then the Christoffel symbols ‘ivanish identically.
y

k 1 eee Oh;
Conversely, if the iP vanish identically, h;;; = one > and, since h,;;; = 0
y\L]
by Ricci’s theorem ,we have 0h;;/0y' = 0 in R,,. Consequently, the h,; are
constants throughout R,,. This permits us to state a
TueoreM I. A necessary and sufficient condition that the metric coeffi-
cients g,,(~) reduce to constants h,, in some reference frame Y is that the
k
Christoffel symbols Eevanish identically.
WY
From this theorem we can deduce at once a system of differential
equations that must be satisfied by functions y'(z", ... x”), if there is to be
a coordinate system Y in which the h,,;’s are constants. The law of trans-
formation 32.6 demands that

_ aout am (a
vlaB) @at dx? Oxi dx’ — aij) Ox" :
m ;
and, since |4 = 0, we have the system of equations
y\a

ou _ {1 "=o,
2,,m m

(39.3) Orc0x" ij) Ox?

Y |are formed from the g;,(x). The system 39.3, of
in which the symbols
an equivalent
second-order partial differential equations, can be rewritten in
form as a system of first-order partial differential equations

it
Ox*
5
(39.4) '
gus face (vies nD’ seein)
Ox’ ij
we now turn to the
This system, in general, will be incompatible, and
ent condit ions for the existence of
determination of the necessary and suffici
solution of the system 39.4.
94 TENSOR THEORY [CHAP. 2

In order to phrase these conditions in a symmetric form, we will consider

the system

(39.5) ae = Ff) fan dash a ee” Ae =, 2),

He eae ae}
where the F,* are known functions of the f’s and x’s. Equations 39.5
specialize to (39.4) if we setf!= y, f? = u4,...,f” =u,. The functions
F,* are defined over the n-dimensional region R and for arbitrary values of
the functions f%, that is, for —co < f’ < oo. Let us refer to the region of
definition of functions F% as R’. This region consists of the region R of the
variables x’ and the set of ranges
100
<< f* < or
We will suppose that the functions F,* are of class Clin R’. Since the region
R’ is open, we will assume that the 0F7/df? are bounded in R’. The
restrictions imposed on the choice of functions F;* are clearly satisfied by
functions appearing in the right-hand members of equations 39.4.
Since the F% are of class C1 in R’, it follows that the f*’s are of class
C?, and hence
2fa 2fa
(39.6) of = of &
Ox Oz 0a 07
This is a necessary condition for the integrability of the system 39.5. Dif-
ferentiating equations 39.5 with respect to x’, we obtain
aye at , art af?
Ca" 0s Oe ee
= OFF 4 OF; FP,
oe werer
where the last step results from the substitution of the expression for
Of? /dx’ from (39.5). Now, if we form (39.6), we get as a necessary
condition for integrability the set of equations

ey ee ayPLES Me OAT re
Axlirw Ofte ta) HORS DFRd GaN Test waevein.
We see that if the system 39.5 has a solution, then either (39.7) are identities
in f* and 2* or else there are certain functional relations existing between
the f’s and a’s. If (39.7) are identities, the system of equations 39.5 is said
to be completely integrable. Itis then possible to prove that the integrability
conditions (39.7) are not only necessary but also sufficient to guarantee the
existence of solutions of the system 39.5.
SEC. 39] RIEMANNIAN AND EUCLIDEAN SPACES 95
There are several proofs of the existence of solution of complete systems
of partial differential equations; perhaps the simplest of these was given
by T. Y. Thomas in 1934 in a paper entitled “Systems of Total Differential
Equations Defined over Simply Connected Domains,” Annals of Mathe-
matics, 35, 730-734 (1934). An earlier proof, assuming the analyticity of
functions F;*, was given by Bouquet! in 1872, and there are other proofs
by G. Darboux and E. Cartan. We shall not go into a discussion of the
sufficiency of conditions 39.7, but will merely state an
EXISTENCE THEOREM. Let R be an open n-dimensional simply connected
region referred to the X-system of coordinates, and R’ the region composed
of R and the ranges —« < f' < o. If the functions F,'(x,f) are of class
C1 in R’ and have bounded derivatives OF,,'/0f’ in R’, and iffurthermore the
integrabiltiy conditions 39.7 are satisfied identically, then the system 39.5
has one and only one set of solutions
fin pe igiet ioe (a=
I... 6a5.07));
which for an arbitrary set of values (xp,..., % ") take on the arbitrarily
prescribed values C* = f*(xq', . . . , Xo").
We will now apply these results to the special case of the system 39.4 by
identifying it with (39.5).
The dependent variables in (39.4) are y, 4,...,U,, Whereas in (39.5)
they are
fs. [7.2 2.5f », Laus, we set
fi=y, fP = wy, ...,f7 = Ug

and the system 39.4 reads

and
of é | y |! (tee 28s lee ite 1),
at, F; uy, F
Ox" a—l i (iso atly pret ya):
The substitution of the expressions for F;* in the integrability conditions
39.7 gives

fa ay 12h
(39.8) ij) ji
Rj, iu, = 0.
The first of these sets of equations is satisfied identically because of the
symmetry of Christoffel symbols. The second set states that the set of
Le¢gons sur
1 J, C. Bouquet, Bull. Sci. Math. et Astron., 3, (1872) p. 265, G. Darboux,
des espaces de
les systémes othogonaux, (1910) pp. 326-335, E. Cartan, Géometrié
Riemann, (1928) pp. 54-57. The proof by T. Y. Thomas is quite close in spirit to that
given by Cartan.
96 TENSOR THEORY [CHaAP. 2

equations 39.4 will have a solution if the Riemann-Christoffel tensor Rit

vanishes identically. Since this tensor vanishes when metric coefficients
are constants, we can enunciate a basic
THEOREM II. A necessary and sufficient condition that a symmetric
tensor g,; with|g,;| 4% 0, reduce under a suitable transformation of coordinates
to a tensor h,;, where the h,;s are constants, is that the Riemann-Christoffel
tensor formed from the g,;'s be a zero tensor.
We note further that, if the quadratic form QO =h,,y'y’ is positive
definite, there exists a nonsingular /inear transformation reducing Q to the
canonical form Q = (y)?--: + (y")?. Thus, if the g,,(7) are the co-
efficients in the positive definite quadratic differential form
(39.1) ds* = g,, dx’ dz’,
characterizing metric properties of R,,, there exists a real functional trans-
formation T: y’ = y'(x) which reduces it to the form
(39.9) ds? = (dy)? + --- + (dy")’,
provided that R',, vanishes identically in R,,.
We recall that a metric manifold R,, in which it is possible to effect the
reduction of the form 39.1 to 39.9 is called an Euclidean n-dimensional
manifold E,,, and we see that R,, is Euclidean if, and only if, the Riemann
tensor of the manifold is a zero tensor.

Problems
1. Verify the substitutions in the integrability conditions 39.7 leading to
equations 39.8.
2. Referring to the system 39.5, show that it is completely equivalent to the
system of total differential equations
df* = F% dx’,
3. What are the integrability conditions for the equation
P(x, y, z) dx + Q(x, y, z) dy + R(x, y, 2) dz =0?
Consider also the system
OF OF )
te e oe ~
4. Prove a theorem: If Pdx + OQ dy + Rdz =Ois integrable, then

AP dx + 4Qdy + AR dz =0
is also integrable for any A(x, y, z) of class C}.
5S. Deduce the integrability conditions for the equation
Pia, ..., 2") de® = 0, (ES Ke
SEc. 40] THE e-SYSTEMS 97

40. The e-Systems and the Generalized Kronecker Deltas

The notions of symmetry and skew-symmetry with respect to pairs of

indices (see Sec. 27) can be extended to cover the sets of quantities that
are symmetric or skew-symmetric with respect to more than two indices.
We will consider in this section the sets of quantities 4."''* or A, ...4,,
depending on k indices, written as subscripts or superscripts, although the
quantities A may not represent tensors.
DEFINITION 1. The system of quantities A" ** (or A, ...;,), depending on
k indices, is said to be completely symmetric if the value of the symbol A is
unchanged by any permutation of the indices.
DEFINITION 2. The system A’ ** (or A, ... ;,), depending on k indices,
is said to be completely skew-symmetric if the value of the symbol A is
unchanged by any even permutation of the indices, and A merely changes sign
after an odd permutation of the indices.
We recall that any permutation of n distinct objects, say a permutation
ofn distinct integers, can be accomplished by a finite number ofinterchanges
of pairs of these objects and that the number of interchanges required to
bring about a given permutation from a prescribed order is always even or
always odd.
It follows at once from definition 2 that in any skew-symmetric system
the term containing two like indices is necessarily zero. Thus, if one has a
skew-symmetric system of quantities A;;,, where 1,), k assume values
1, 2, 3, then Ayo. = 0, Ayo3 = —Aors, Asis = Aizg, etc. In general, the com-
ponents A,;, of a skew-symmetric system satisfy the relations A;;, =
—Ayy = —Ajx = Anns = Ars = —Apsi-
Consider now a skew-symmetric system of quantities A;,...;, (or
Ai’ tn), in which the indices i,,..., 7, assume values 1,2,...,n. We
define the e-system as follows.
DEFINITION 3. Jf the value of A,,...;, (or AQ") is +1 when isin + * ip
is an even permutation of the numbers 12 +++ n, and —1 when iyi, °° i, is an
odd permutation of 12+-+n, and if it is zero in all other cases, then the
system A,...;, (or A%’’’'*) is called the e-system.
We shall use the symbols e;,...;, or e1''™ to denote the e-systems.
It will be shown in Sec. 41 that the e-systems are relative tensors.
As an illustration we note that the components of the system e,; are:
three
€4, = 0, ey = 1, op) = —1, Coe = O. If the e-system depends on
e;;, = @123 = 1
indices ijk, then e,,, = Oif any two indices are alike, whereas
odd
if ijk is an even permutation of 123 and e,;, = —¢123 = —1 if ijk is an
permutation of 123.
Closely allied to the e-systems are the generalized Kronecker deltas,
which we proceed to define.
98 TENSOR THEORY [CHAP. 2

DEFINITION 4. A symbol 65+: ::i* depending on k superscripts and k

subscripts, each of which runs from 1to n, is called a generalized Kronecker
delta provided that: (a) it is completely skew-symmetric in superscripts and
subscripts; (b) if the superscripts are distinct from each other and the sub-
scripts are the same set of numbers as the superscripts, the value of the
symbol is +1 or —1 according as an even or odd number of transpositions is
required to arrange the superscripts in the same order as the subscripts;
(c) in all other cases the value of the symbol is zero.
As an illustration consider 6%. It follows from definition 4 that if
i= jork =1, or if the set ij is not the set kl, then 6 = 0. In all other
cases 61)equals +1 or —1 according to whether k/ is an even or an odd
permutation of ij. Thus
eaten oe :

1 = 63 = 63 = 63 =-"°,
-l= op Os eet
We prove in Sec. 41 that the generalized Kronecker deltas are tensors.
From definition 3, it follows that the direct product e'"2"""'e; 5...
of the two systems e’"’' and e; ...;, is the generalized Kronecker
delta. For example, e*’’e,;, has the following values:
(a) Zero, if two or more subscripts or superscripts are alike.
(b) +1, if the difference in the number of transpositions of “Py and ijk
from 123 is an even number.
(c) —1, if the difference in the number of transpositions of “By and ijk
from 123 is an odd number.
A little reflection will show that another way of phrasing statements (6)
and (c) is the following:

(b') e**’e,,, = +1, if an even number of transposition is required to

arrange the subscripts in the same order as the superscripts.
(c’) e**’e,,, = —1, if an odd number of transpositions is required to
arrange the subscripts in the same order as the superscripts.
We can thus write
ees,
apy =
— OF.
aby

It is clear from definitions 3 and 4 that the e-symbols can be defined in

terms of the Kronecker deltas,

ele = oe and ae ; oe S => Cris tine

since e = +1 or —1 when a" set of distinct integers i,i, - - - i,, is obtained

from the set 12 -- +n, by an even or an odd permutation, and e = 0 in all
Sec. 40] THE e-SYSTEMS 99
other cases. The e-systems and generalized Kronecker deltas prove useful
in calculations involving alternating sets of quantities.
We consider next several examples which permit us to deduce a number
of identities involving operations on these symbols. £
Let us contract 6%), on k and y. The result for n = 3 is

Oxon = Oapi + Ozp2 + Sapa= Op.

We observe that this expression vanishes if i andj are equal or if « and B
are equal. If we set i =:1 and j= 2, we get 6133 = 013, and hence 633 =
0, unless «f is a permutation of 12. In the latter case 633 = 1 if af is an
even permutation of 12, and 6)3 = —1 for an odd permutation. Similar
results hold for all values of 7 and j selected from the set of numbers 1, 2, 3.
We thus see that 62, is equal to
(a) 0, if two of the subscripts or superscripts are alike, or when the
subscripts and superscripts are not formed from the same numbers.
(b) +1, if ij is an even permutation of «f.
(c) —1, if ij is an odd permutation of af.
If we contract 6%, and halve the result, we obtain a system depending on
two indices 5

6a = 464+ 02 + 053)-
es

If we set i= 1 in 6%, we getis = 4(612 + 613). This vanishes unless

a = 1, in which event 6! = 1. Similar results can be obtained by setting
i=2ori=3. Thus 6} has the values
(a) 0ifiF«, (api 2 5):
li e
(6) 1, ifi = «.
By counting the number of terms appearing in the sums it ts not difficult
to show that, in ae

(40.1) i= ; 64 0and «= 6, = n(n — 1).

n —

We can also deduce that

(n—ce k)!
k)! Oiit2*sc Srbrea es ciee
(40.2) Onis Le= (n —r)! **UK?
Giga ** Grint’

and
n!
n(n — 1)(n — De 2 (1 = tots 1) = SET :
(40.3) Tae a =

As a special case of (40.3) we have the formula

(40.4) eit
dyin tt
te, yg, se
=H!
i7ill
100 TENSOR THEORY [CHAP. 2

and from (40.2) we deduce the relation

(40.5) Ce ites *Grtpy1'

= (n — r)! 63
Consider next a set of n?* panes” Aju ..13 (the i’s ma j’s run from
1 to n), symmetric in two or more indices (which may be superscripts or
subscripts). We can show that
6;Jide
tytg
°
++
dq
ty
A T1%2°*° Tpee—
Fid2°

if Aj: ee is symmetric in two or more aaa Also

OF.
$152
Ag
Sp ie see —
iy,
if ARs. ae is symmetric in two or more superscripts.
Suppose that Aj1!°?7? is symmetric in j, and j,; then
J1I92° “Sa TA P65 OP aha ABIZS * Ja Sel e gia
On oi CART Br Pe Opei Asa "Ig
— — 62271"
iyig ++Sgrig/*ioi1-*
ye > me)
+5g°

However, j, and j, are the dummy indices; hence

62192" ° HAR, Do ae = — §i1s2"° AR, Poss ze
THE 41%2° Jid2° tjte** j132° "Iq

CS 3: Ape o car = 0.

Problems

1. (a) Show that oyk = 3! if i,j,k =1, 2,3.

(b) Show that 6¥,

,
oe Oe beras: and oF, =|0)
ue * 6}
o: %off
a B ok ok ok
)
2. Expand for n = 3:
(a) 0!OF, (b) d}Pxty), (c) Oaty?,
(d) dffatyi, (e) 63.
3. Expand for n = 2:

(a) e“aja}. (b) e*atat. (0) e“Fal al, =e lal.

4. If a set of quantities A; ...;, is skew-symmetric in the subscripts (k in
number), then
a eee ry
ie sesNabe =k! Ay abe

5. If A;;, is completely symmetric and the indices run from 1 to , show that
the number of distinct terms in the set {A4;;,.} is

n(n — 1)(n — 2)
N=n+n(n —1) + 7
Sec. 41] APPLICATION OF THE e-SYSTEMS 101

Hint: Consider the cases where the subscripts ijk are all alike, when only two
are distinct, and when all are distinct,
6. Show that the number of distinct, nonvanishing A;;,’s in Problem 5 is
n(n — 1)(n — 2) “
when 4,;,;, is completely skew-symmetric.
3!

41. Application of the e-Systems to Determinants. Tensor Character

of Generalized Kronecker Deltas

We recall that the determinant la’ of nth order, with elements a’, con-
sists of the sum of products of the elements where each term in the sum
contains one and only one element from each row and each column of
the determinant. 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 aja; --- aj, are arranged in the normal order 12---7n, then
the product will carry the plus sign if the number of transpositions neces-
sary to arrange the subscripts in the normal order is even. The sign is
minus if the required number of transpositions is odd. Since ef" "** =
Side vin and OL 3°21", = i,i,---4,» the determinant
aah at
(41.1) jaja | 2 "lea
aj ay -"* ay
can be written compactly as
Pioieog,. & 2
om oft giay + * aj,
pe i
(41.2)
SS Cs ig +s sigh C2 an

As an example consider
a, a2 43
a,
ye a2 ene43|=
| 4.
jaj) =|
3 3
ay ag ag

a= x + aiaiat,
If this determinant is expanded by columns we get the
sign is assigned to
where ijk is a permutation of 123. The plus or minus
is even or odd.
term aiajat according to whether this permutation
determinant can be written |ai| = €,;,a@, a}as. On the other
Hence this
|a| = ei ava;
hand, if it is expanded by rows, we can write
Consider next the sum
Nie . 3).
C jp VAs (i,ve k, a, B, Ve
102 TENSOR THEORY [CHAP. 2

We will show first that this system is completely skew-symmetric in «fy.

Since the indices ijk are dummy indices, we can change them at will and
write
C ijn Ayay eeAA, ies
Ap, = Cyj{
Ag i pigk
= Cy ji Fp Ay-

Ifkand iare interchanged in e,,;;, this e-symbol will change sign, and hence

C ijn AyeeAy, eS= — C5544,ay Aha.pe

This shows that an interchange of « and y changes the sign, so that the
system under consideration is skew-symmetric in « and y. Similar results
obviously hold for other indices. A special case of this system is the
determinant |a'| = e;,;,a,a3a5, and it follows from the foregoing that
Siete Ls
C 5540434, = |45| Cx,

Similarly, we can show that

‘ik aX ab aY —
e*Kataiay || ee”.
= \a;| pXPy

It follows at once from these expressions that an interchange of two columns

(or two rows) of the determinant |a,’| changes its sign, and if two columns
in it are identical, then its value is zero.
These results can be immediatley generalized to determinants of nth
order, so that for any permutation of rows we can write
ater i) —. pij--°k Br aap
(41.3) er" Y lat] = ae " Fatat ae

and for any permutation of columns

(41.4) Cig... 4l@j] = Cap... Aah > °* ay.

We use formula 41.4 to establish the formula for the product of two
determinants. The power and compactness of this notation are strikingly
demonstrated in this derivation.
Since [b'| = e,;...,b{b3 - - b*, we can write

la’| y [5}| = la’| Giz... Dib} a bk

= (€,3...,a7a8 +> + at)(b1b3 - - - be),
where we have made use of the formula 41.4. Thus

laj| - [bj] = eng... azdy)(afbs) - +» (azbh

= Icjl,
where
i
Cj =
in
a,b; = a,b;
ipl +
i p2
ayb; a pe tes + a,b;
i
Sec. 41] APPLICATION OF THE e-SYSTEMS 103

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

column and their cofactors can be written

(41.5) la;| = os Ia ee ei Gee be a, «

Se wy
aa ajA,,

where Ay = @y;,...i, aiais---a'n is the cofactor of the element aj.

We derive next the formula for the partial derivatives of a determinant
whose elements a,’ are functions of the variables a1, 2,..., 2”. From
formula 41.2 we have
—_ tials... = qin
a Cists See 1,41 ay a,

Differentiating this expression, we get

Oa _ day t2... qin i, 0a tn .

~ Az ay, + ay 3 Nl
Ox?
coetemiaid, > | Vorak3
oe ee on

da a day 42. Lop oe

Se th = A;
=
an
; z =
On?
; A;
eon ek
dap 4p
Oxi *
by formula of the type 41.5.
that the permu-
Formulas 41.3 and 41.4 permit us to establish the fact
tation symbols e1"° ™ andé;. .. 4, are relative tensors of weights +1 and
—1, respectively.
Consider an admissible transformation
Rina (apenas),
a elie
and its Jacobian J = -. If we set a; = ce in formula 41.3, and recall

that 5= |—|, we obtain at once

ig ae Onl || ....c oy. oy
oo oS
e Byrom a Pe
3y fbi eso ae
ae A a
(41.6)

which is the law of transformation of relative contravariant tensors of

deduce that
weight +1. In an entirely similar way we
dxf? Ou"! Ox? ket.
Ou
——— e
(41.7) Cr ie ming "An A yf ay? dyin ‘
dy
of weight —1.
so that e, 12;....;, n is a relative tensor
104 TENSOR THEORY [CHAP. 2

From formula 40.5,

ee tr ng | ee aa eg eae
we see that the Kronecker delta Opie is obtained by multiplying
together two e-symbols, one of which is a relative tensor of weight +1
and the other of weight —1, and contracting with respect to a number of
indices. The result is a tensor of weight zero, that is, an ordinary tensor.
Thus we have proved that the generalized Kronecker deltas are absolute
tensors. Ji wea De é

Since 6/1/''40,
1 Ch
reduces to ee
‘8
= (0 when the coordinate system X
is cartesian, we conclude that the covariant derivatives of generalized
Kronecker deltas vanish identically. Thus the Kronecker deltas behave as
constants in a covariant differentiation.

Problems
1. Verify that 5¥/,a%? = a) — a’.
2. Verify that O7F ae = qgiik — giki 4 qgiki — gitk 4 gkii — ghii,
3. If a,; satisfies the equation
ba;; 5 ca; = 0,

then either b = —c and a;; is symmetric, or b = c and a;; is skew-symmetric.

Hint: Since i andj take on values | - - - n, the equation can be written as

ba;; + ca;; = 0.

Add and obtain (6 + c)(a;; + a;;) = 0.

4. Show that (a) e,,e** = 6i, (b) e;;,e°78 = 608 — 020%. Hint: The left-hand
member in (b) is zero unless j and & are distinct and j, & is a permutation
of r, s. If j =r and k =s, then the left-hand member is +1; if j=s and
k =r, then its value is —1. Consider now the value of the right-hand member
for the same choices of indices.
5. List the values of 53, 0},3? 6'0/0* when the indices range from | to 7.
 3}
GEOMETRY

42. Non-Euclidean Geometries

There is no branch of mathematics in which the tyranny of authority

has been felt more strongly than in geometry. The traditional Euclidean
geometry, based on a set of “self-evident truths” and created largely by
the Alexandrian School of mathematicians (around 300 B.c.), dominated
the thought and shaped the development of physics and astronomy for
over 2000 years. There were a few bold souls, even among the ancient
mathematicians, to whom “self-evident truths” contained in Euclid’s
axioms did not seem convincing, but the prestige of logical structure of
Euclid’s Elements was so high and the hand of authority so heavy that
they hindered the development of mathematics for centuries.
In 1621, Sir Henry Savile raised some questions concerning what he
called “two blemishes” in geometry, the theory of proportion and the
theory of parallels. Euclid’s axiom of parallels (Postulate V in the first
book of Elements) is to the effect that any two given lines in a plane,
when produced indefinitely, will intersect if the sum of two interior angles
made by a transversal with these lines is less than two right angles. The
fact that some of Euclid’s propositions, dealing essentially with the
converse of this postulate, can be proved without invoking Postulate V
gave hope that the postulate itself might be deduced from his other
axioms. However, all attempts to prove the fifth postulate proved un-
successful, and a hope that contradictions would emerge if this postulate
a
were abrogated while others were retained led nowhere. In 1826,
Russian mathematician, Nicolai Lobachevski, presented to the mathe-
maticians faculty of the University of Kazan a paper based on an assump-
two lines
tion that it is possible to draw through any point in the plane
proved
parallel to a given line. The geometry developed by Lobachevski
Indeed,
just as devoid of inner inconsistencies as Euclidean geometry.
of
it contained the latter as a special case and implied the arbitrariness
the concept of length adopted in Euclidean geometry.
105
106 GEOMETRY [CHAP. 3

In 1831, a Hungarian mathematician, John Bolyai, published results

of his independent investigations which conceptually differ little from
those of Lobachevski, but which perhaps contain a deeper appreciation
of the metric properties of space. Bolyai pointed out, just as Lobachevski
did, that his geometry in the small is approximately Euclidean and that
only a physical experiment can decide whether Euclidean or non-Euclidean
geometry should be adopted for the purposes of physical measurement.
Thus it appears that there are no a priori reasons for preferring one
geometry to another. However, it was only after Riemann’s profound
dissertation on the hypotheses underlying the foundations of geometry
appeared in print (published posthumously in 1867) that the mathe-
matical world recognized fully the role played by the metric concepts in
geometry.
Riemann appears 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 geometries
(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. From these researches it appeared that three con-
sistent geometries are possible on surfaces of constant curvature:
the Lobachevskian, on a surface of constant negative curvature; the
Riemannian, on a surface of constant positive curvature; and the
Euclidean, on a surface of zero curvature. These geometries are also
called hyperbolic, elliptic, and parabolic, respectively. We- consider
them briefly in the next section.

43. Length of Arc

Let the n-dimensional space R be covered by a coordinate system X,

and consider a one-dimensional subspace of R determined by

(43.1) Co ges =e *(2). Gi Sol yasta,s 7),

where ¢ is a real parameter varying continuously in the interval t; < t < fy.
The one-dimensional manifold C is called an arc of a curve. In this book
we deal only with those curves for which 2‘(t) and #‘(t) = dx'/dt are
continuous functions in t; < t < fy. The definition of the arc of a curve
given here is a direct generalization of the parametric representation of
curves of elementary analytic geometry.
Let F(2t,..., 2", %,..., 2%"), viewed.as a function of t bea pre-
scribed continuous function in the interval 4; <<t<t,. We suppose
Sec. 43] LENGTH OF ARC 107
that! F(a, @) > 0, unless every #' = 0, and that for every positive number k
A Be “A : :
le eo ahem hae) SOR ee sag Re ey pa ay he)
«
The integral
“bs
(43.2) S -| F(x, x) dt
t

is called the length of C; and the space R is said to be metrized by formula

43.2.
Different choices of functions F(x, #) lead to different metric geometries.
If one chooses to define the length of are by the formula

3 ag da oite B Beaters
(43.3 ) s = {, “Sap(X) ————
dt dt dt,
t (a n),

where Lape e? is a positive definite quadratic form in the variables 2%,

then the resulting geometry is the Riemannian geometry, and the space R
metrized in this way is the Riemannian n-dimensional space R,,.
k > 0 is
1 A function F(z, #) satisfying the condition F(z, ka) = kF(x,£) for every
and
called positively homogeneous of degree | in the x". This condition is both necessary
ce of the value of the integral 43.2 of a particular
sufficient to ensure the independen
mode of parametrization OnGwl huss if rin (43.1) is replaced by some function t = ¢(s),
and we denote x‘[¢(s)] by &#(s) so that x(t) = &i(s) we have the equality
ts 82

| F(a, «) dt -| Fé, &’) ds,

ty 8}Es

where €’'(s) = dxi/ds and t, = $(s,) and t, = (52).

number, and set
To prove this theorem, suppose that k is an arbitrary positive
t = ks, so that t; = ksy, and fz = ksy. Then (43.1) becomes
Gees (ies ie)
and
dx'(ks)
os) = - = kai(ks).
a'(ks)

If these values are inserted in (43.2), we get

k8o
5 -| Flx(ks), #(ks)]k ds,

: wee RSH
and if this is to equal
S92
s -| F[&(s), &(s)] 4s,
81
= kF(x, #). Conversely, if this relation
we must have the relation F(é, §&’) = F(#, kaw)
C and each k > 0, then the equality of integrals is
is true for every line element of $(51)
= d(s), 6'(s) > 0, 51 <s <S, with 4 =
assured for every choice of parameter t
and ty = (52).
108 GEOMETRY [CHAP. 3
We recall from Sec. 39 that, if there exists an admissible transformation
of coordinates T: y* = y'(x!,..., 2”), such that the square of the element
of are ds,
(43.4) ds" = Pp pdrede
can be reduced to the form
(43.5) ds* = dy'‘dy’,
then the Riemannian manifold R,, is said to reduce to an n-dimensional
Euclidean manifold E,,. The reference frame Y in which the element of
arc of C in E,, is given by (43.5) is called an orthogonal cartesian reference
frame. Obviously, £,, is a generalization of the so-called Euclidean plane
determined by the totality of pairs of real values (y', y). If these values
(y', y*) are associated with the points of the plane referred to a pair of
orthogonal cartesian axes, then the square of the element of arc ds assumes
the familiar form ds? = (dy')? + (dy?)?.
In what follows we find it convenient to represent pairs of real values
(y', y*) as points in a cartesian plane even when the metric of the y’-
manifold is not Euclidean. To illustrate what is meant, consider a sphere
S of radius a, immersed in a three-dimensional Euclidean manifold Es,
with center at the origin (0, 0, 0) of the set of orthogonal cartesian
axes O-X1X?X%, Let T be a plane tangent to S at (0,0, —a), and let
the points of this plane be referred to a set of orthogonal cartesian
axes O’-¥1Y* as shown in Fig. 8. If we draw from O(0, 0, 0) a radial
x3

Fig. 8
SEC. 43] LENGTH OF ARC 109
line OP, intersecting the sphere S at P(a!, x, v*) and the plane T at
Oy’, y®, —a), then the points P on the lower half of the sphere S are
in one-to-one correspondence with points (y', y*) of the tangent plane T.
To obtain an explicit analytic form for this correspondence, we note
that, if P(x’, x, x) is any point on the radial line OP, then the symmetric
equations of this line furnish us with the ratios

a—O 2-0 8) =
he wipe) basa
Od we
or

(43.6) a = Ay, gee AN a i

Since we are concerned with the images Q of points P lying on S, the

variables x’ satisfy the equation of S,
at}? + a7)2 + (a2)3 — aa

Or
2U(y)?
+(2)? +a] =a.
Solving for A and substituting in (43.6), we get
1 2
oe? = ele see ae ee .
+ oF +a
5
VPP
A
VU'y + YP + a?
(43.7) es eee

a =a
Vy? + (y+ a?
These are the desired equations giving the analytical one-to-one corre-
spondence of the points Q on T and points P on the portion of S under
consideration.
Let P,(a}, x?, 2°) and P,(x! + dxt, a? + dx®, a + dx®) be two nearby
points on some curve C lying on S. The Euclidean distance P,P2, along
C, is given by the formula
(43.8) ds? = dx‘dx', (f= 19233),

and, since the variables ’ are related to y’ by (43.7),

dx = oe dy’, («=1,2).
Thus (43.8) yields a formula
ws dx" Oa! diy" dy?
S
dy" ay?
= Sap(Y) dy* dy’, (x, B — Lp 2);
110 GEOMETRY [CHAP. 3

where the g,,(y) are functions of y’ computed from (43.7) with the aid
Ox’ Ox'
of the definition g,; =
dy* dy?”
If the image K of C on 7 is given by the equations

(y= yO),
ly? = y(t); igs ES ts,

then the length of C can be computed from the integral

(oe
S =| Vo V7y dt.
ty

A straightforward calculation gives

1
(dy')? + (dy’y? + Gy" dy* — y? dy’
(43.9) as = j
|145107" 4+ 7) \°
atin j
and

(yy? + (x) + isyy — yy’)

. . 1 . 9. 9

2] Efe Pie is Eee ae EE

te

ie = (yy? + v7]
ty

We see that the resulting formulas refer to a two-dimensional manifold

determined by the variables (y', y?) in the cartesian plane 7 and that the
geometry of the surface of the sphere imbedded in a three-dimensional
Euclidean manifold can be visualized on a two-dimensional manifold R,
with metric determined by (43.9). If the radius of S is very large, we see
from (43.9) that the terms involving I/a? can be neglected, and the geometry
of the surface of the sphere is then determined approximately by the
Euclidean metric

(43.10) ds? = (dy)? + (dy*)?.

Thus, for large values of a, metric properties of the sphere S are indis-
tinguishable from those of the Euclidean plane. The sum of the angles
of a curvilinear triangle drawn on S will be nearly equal to 180°, since
the sum of the angles of the corresponding triangle on T is 180° by
Euclidean geometry. Because of the limitations of measuring devices it
SEc. 43] LENGTH OF ARC 1a
may be impossible to decide a priori whether Euclidean formula 43.10
or the more involved Riemannian formula 43.9 should be adopted as a
basis for physical measurements.
The chief point of this illustration is to indicate that the geometry of a
sphere, imbedded in a Euclidean 3-space with the element of arc in the
form 43.8, is indistinguishable from the Riemannian geometry of a two-
dimensional manifold R, with metric 43.9. The latter manifold, although
referred to a cartesian frame Y, is not
Euclidean since (43.9) cannot be reduced Q
by an admissible transformation to (43.10).
Similarly, the geometry of Lobachevski
can be visualized on a surface of a “pseu-
dosphere,” a surface of constant negative
curvature generated by revolving a tractrix,

j= a(cos t + log tan lk

¥ = asint, C

about its asymptote. Since we will have Fig.9

no occasion to study the Lobachevskian
or hyperbolic geometry, we will only indicate the main ideas leading to
the analytical expression for the square of the element of arc
1 ;
(dy’)? + (dy’)? — 3 (y' dy? — y° dy’)
ds* =
fai“ iy? + (yy
which governs the study of this geometry.
Let a circle K of radius one be drawn in the plane. The universe of
Lobachevskian geometry consists of points interior to K. The chords PO
of the circle are straight lines in this geometry. (See Fig. 9.) The length
of the segment AB of PQ is a number given by the formula
i! 2)
log daskca= |e
QA QB
Con-
whereas the magnitude of the angle ABC is determined as follows.
Project AB
struct a sphere S of radius one tangent to K at its center.
the arcs BA’
and BC on S and determine the Euclidean angle between
BC and
and B’C’ formed by the intersection of the planes passing through
an measure of
BA perpendicular to the plane of K (Fig. 10). The Euclide
Lobachevski
A'B'C’ is, by definition, the measure of the angle ABC in the
112 GEOMETRY [CHaP. 3

Fig. 10 Fig. 11

plane: A pair of lines in the Lobachevski plane are considered parallel

if their images on the sphere do not intersect. It can be shown that the
points and lines of this geometry satisfy all postulates of Euclidean
geometry except the postulate of parallels. Parallel to any given line PQ
one can draw through a point M infinitely many lines which do not
intersect PQ. These are the lines lying in the shaded region of Fig. 11
and passing through M. It is not difficult to prove that the sum of the
angles of a triangle in this geometry is less than 180°. The consistency
of Lobachevskian geometry was investigated by Cayley, Kiein, and
Poincaré.”
The discussion of this chapter is confined mainly to Euclidean geometry
and those portions of Riemannian geometry that figure in applications.

44. Curvilinear Coordinates in FE,

The apparatus of tensor analysis was developed initially as a tool

for the analytic study of geometries of diverse sorts. Because of its
invariantive character, it was found particularly adaptable to the needs
of other branches of applied mathematics. Since dynamics, mechanics
of continuous media, and relativity lean rather heavily on geometrical
properties of the three-dimensional space of physical experience, we devote
most of this chapter to an investigation of properties of curves and surfaces
imbedded in E3.
Let the point P(y), in an Euclidean 3-space Es, be referred to a set of

* For details on hyperbolic geometry we refer the reader to specialized treatises on the
subject, especially to F. Klein’s Nicht-Euklidische Geometrie, 1, pp- 161-232.
Sec. 44] CURVILINEAR COORDINATES IN E, 113
orthogonal cartesian axes Y (Fig. 12). Consider a general functional
transformation
Te ate. yf), (i= 1, 2,3), “

Ox?
such that the 2’ are of class C1, and J = |: ~ 0 in some region R of
E,. The inverse transformation, oy
. . j

T: yo = ye, 27,2), (i = 1, 2, 3),

will then be single-valued, and the transformations T and T~ establish
one-to-one correspondence between the sets of values (x', x, 2) and
(y', y*, y®). We call the triplets of numbers (a!, x, x*) the curvilinear
coordinates of the points P in R. The reason for this terminology is the
following: if we set x1 = constant in 7, then
(44.1) x'(y*, y*, *)'= constant

defines a surface. If the constant is now allowed to assume different

values, we get a one-parameter family of surfaces. Similarly,
xy), y*, y®) = constant

and x3(y!, y?, y?) = constant define two families of surfaces.

The condition that the Jacobian J # 0 in the region under consideration
expresses the fact that the surfaces

(44.2) gies Cy, P =e, = Cp

intersect in one and only one point.

We call the surfaces defined by equations 44.2 the coordinate surfaces,
and their intersections pair-by-pair are the coordinate lines. Thus the
line of intersection of x! = c, and x? = c, is the x?-coordinate line because

Fig. 12
 114 GEOMETRY [CHAP. 3

along this line the variable «* is the only one that is changing. As an
example, consider a coordinate system defined by the transformation

y= oc sin a c0s 37,

‘== 2) Sina Sin.
i= eh COS a

The surfaces ~! = constant are spheres, x? = constant are circular cones,

and x? = constant are planes passing through the Y*-axis (Fig. 13).
The inverse transformation in this case is given by

Davyyt yy +O,
a? = tan
avyy +?
2
y
a = tan? g .
y

-if 21 > 0,0 < 2 < 7,0 < 2 < 27. This is the familiar spherical coor-
dinate system.
As another illustration, the transformation

yy) = 2 cos x7,

a? = 2" Sin a,

y= x,
defines a cylindrical coordinate system (Fig. 14).

Fig. 13
Sec. 44] CURVILINEAR COORDINATES IN £, 115

Fig. 14
Let P(y}, y?, y?) and O(y! + dy’, y? + dy*, y® + dy®) be two neighboring
points in R. The Euclidean distance between a pair of such points is
determined by the quadratic form
(ds)? = (dy*)? +(dy?)? + (dy*)?
= dy' dy’,

; ay
and, since dy* = ee dx*, we have
x

(44.3) te = Bae Lt
where
Oy" Oy"
Peace oh = j, 2, 3).
ae Ox' Ox’ ie )
Obviously, g,; is symmetric. Moreover, it is a tensor, since (ds)? is an
invariant and the vector dz‘ is arbitrary. Denote by g the determinant
|g,;; this is positive in R since g,; dx‘ dx’ is a positive definite form.
Hence we can introduce the conjugate symmetric tensor g"’, defined in
Sec. 30 by the formula g’ = G"’/g, where G’” is the cofactor of the element
8, in g.
Consider now a contravariant vector A‘(x), and form the invariant
(44.4) A = (g,,A1A))*.
Since in the orthogonal cartesian frame the invariant 44.4 assumes the
of the
form [(A1)? + (42)? + (A%)?]’4, we see that A represents the length
by
vector A‘. Similarly, the length of the covariant vector 4, is defined
the formula
(44.5) As AAs)
In orthogonal cartesian coordinates gi = 6", and we get A = (A,A,)%.
116 GEOMETRY [CHAP. 3

A vector whose length is 1 is called a unit vector. From formula 44.3

we see that

so that dx‘/ds = 4‘ is a unit vector. If x‘ = y’, so that the coordinate

system is cartesian, then dx!/ds = 11, dx?/ds = 2, dx®/ds = /? are precisely
the direction cosines of the displacement vector (dz', dx*, dx*). Accord-
ingly, we take the vector A’ to define the direction in space relative to a
curvilinear coordinate system X (Fig. 15).
Consider two directions defined by the unit vectors 4* and uw’ at some
point P (Fig. 16). Since the manifold under consideration is Euclidean,
the cosine law, following from the formula of Pythagoras, gives

OR? = PQ?
+ PR? — 2PO PRcos6,

and, since A‘ and w‘ are unit vectors, PO = PR = 1, and hence

(44.6) OR? = 2(1 — cos 9).

The components of the vector joining R with Q are A‘ — uw‘. Making

use of the formula 44.4 for the length of a vector, we get

(44.7) OR? = g,,(4' — w')\(H) — pw’)

= giA'l + g,,u'p? — 2g,,A‘p)
=1+4 1 —2g,,A‘u?
= 2(1 — gijA*p’).

Fig. 15
Sec. 44] CURVILINEAR COORDINATES IN £3; 117

Fig. 16

It follows from (44.6) and (44.7) that the invariant gijA*w? is equal to
cos 6, and we can write

(44.8) cos 0 = g;,A‘p’.

We can use (44.8) to define the angle 6 between two directions Ai and
if we make an unambiguous definition of sin 6.
If A‘ and B’ are any two vectors, then from the definition of the length
of a vector, it is clear that
gijA'B?
cos
§=
V gij;A'A? V ¢,;B'B?
an invariant, which
This leads to the formula AB cos 0 = g,,A‘B’, defining
analysis.
is precisely the “scalar product” A - B of elementary vector
It follows from the expression
ds? = g,, dx‘ dz,
n P,(z1, x, x°), and
for the square of the element of arc ds betwee
the length s of the elements of arc
P,(a! + dxt, x? + dx’, a? + dx), that
system X are
measured along the coordinate lines of our curvilinear

(44.9) doa =Veude!, ds =Vga2de®, dso) = V£43 dx.

0, 0) is given by J.on dx},
Thus the length of the displacement vector (dx',
(0, 0, dx) has the length
that of (0, dz”, 0) is J Lo dx, and the vector
\ 253 de® (Fig. 17). cosines of the angles 4»,
In addition, from (44.8) we deduce that the
by
653, 9,3 between the coordinate lines are given
§23 &13 ;
=
§12
—B{, cos 423 = , cos I, =
44.10) cos 0, J 883s
ms VJ81822 ~ VJ822833
118 GEOMETRY [CHAP. 3

Fig. 17

For, if A',): (de"/dsq, 0, 0) and ui): (0, dx?/ds(2), 0) are two unit vectors
directed along the X1- and X*-coordinate lines, respectively, then
= ie S12 dz" dt « §12
COS O49 = B54 M2) =
ds(1) dS(2) aj 211832
Since 211, 222,833 never vanish (see equation 44.9), we deduce from
(44.10) a
THEOREM. A necessary and sufficient condition that a given curvilinear
coordinate system X be orthogonal is that g;;=0, for iA j, at every
point of the region R.
From the definition of the element of volume dV in curvilinear
coordinates,

rs) B
dV=+
E
oy’
x?
dx) dx dx’,

where + |
5 is the absolute value of the Jacobian J of the transformation
x

connecting the cartesian variables y’ with the curvilinear x", we can

readily deduce that
(44.11) dV = dy dy? dy = Vg dx' dx? dx’.
For,
dy'| | dy’ dy* dy*
SPs |sey) 9 pees] | ned ee ee
Ol |Oe Ox" Ox’ Isl =8
where we made use of the definition for g;; (see equation 44.3), and of
the rule for multiplication of determinants. The determinant g is a
relative scalar of weight 2 (cf. Sec. 28) since J/g is a scalar density.
From developments of this section we see that the metric properties
of £3, referred to a curvilinear coordinate system X, > are completely
Sec. 45] RECIPROCAL BASE SYSTEMS 119
determined by the tensor g,;. Accordingly, this tensor is called the metric
tensor, and the quadratic form ds* = g,, dx' dx’ is termed the fundamental
quadratic form.
~

45. Reciprocal Base Systems. Covariant and Contravariant Vectors

In this section we interpret the main results of Sec. 44 in the language

and notation of the elementary vector analysis introduced in Chapter I.
Let a cartesian system of axes (Fig. 18) be determined by a set of ortho-
normal base vectors b,, bs, bs; then the position vector r of any point
Py’, y®, y®) can be represented in the form
(45.1) r= by, (i = 1, 2, 3).
Since the base vectors b, are independent of the position of the point
Ply’, y?, y®), we deduce from (45.1) that
(45.2) de = bj dy?
By definition the square of the element of arc between the points
(y1, y?, y®) and (y! + dy’, y® + dy’, y*® + dy*) is given by the formula

(45.3) ds? = dr + dr.

The substitution from (45.2) in (45.3) gives
ds? = b; - b; dy’ dy’
= Oi dy’ dy?
= dy’ dy’,
a familiar expression for the square of the element of arc in orthogonal
cartesian coordinates.

y? x?

Fig. 18
120 GEOMETRY [Cuap. 3
Let a set of equations of transformation

a=aityjyy), (=1,2, 3),

define a curvilinear coordinate system X. The position vector r can now
be regarded as a function of coordinates x’, and we write
or’
(45.4)
3 r Bal
dae

and
Ore Ora,
ds 2 = dr- dr = an
r-dr —+-—
Da! dr‘ dz’

= g,; dx’ dx?

where
Or or
45.5 is = Ried Ree 5
( ) oe Ox* Ox?

The geometrical meaning of the vector dr/dx* is simple; it is a base vector

directed tangentially to the X*-coordinate curve. We set
or
45.6 = &,,
el) Ox*
and rewrite (45.4) and (45.5) as

(45.7) dr = a, dx’
and
Sig = A; ° 4;-
We observe that the base vectors a; are no longer independent of the
coordinates (a1, 2?,:2°).
The use of covariant notation for the base vectors a; and b; can be
justified by observing from (45.2) and (45.7) that

a; dz? = b, dy‘
= bode,
Ox?
We see that the base vectors a; transform according to the law for the
transformation of components of covariant vectors,
Oy’
ba Hantnt
since the dx’’s are arbitrary.
The components of base vectors a;, when referred to the X-coordinate
Sec. 45] RECIPROCAL BASE SYSTEMS 121
system, are
a: (a,, 0, 0), a,: (0, a, 0), az: (0, 0, as),
and we note that they are not necessarily unit vectors, since, in general
(see equation 45.5),
8u = a+ a, ~ 1 S22 = a+ a, # I, 833 = Ag* as # I.
If the curvilinear coordinate system X is orthogonal, then
21; = a,° a; = |a,| |a,| cos 0,; = 0, if i ¥j.
This is the result stated in the theorem of Sec. 44.
We note that any vector A can be written in the form A = k dr, where
k is a suitable scalar. Since dr = (dr/0x*) dx’, we have

where Ai = k dx‘. The numbers 4‘ are the contravariant components

of the vector A, and the vectors A1a,, 42a), Aa, form the edges of the
parallelepiped whose diagonal is A. Since the a, are not unit vectors in
general, we see that the lengths of edges of this parallelepiped, or the
physical components of A, are determined by the formulas
AN gy) A’| B00, A*\/B55

SINCE 211 = A, * Al, Jog = Ag* A, $33 = Ag ° a3.

Let us introduce next three noncoplanar vectors

(45.8) a re > = + ’ (re “)

[aja.az] [aja2a3] [aya.a3]

[a,a,a3]
where a, X a3, etc., denote the vector product® of a, and as, and
is the triple scalar product a, + a, X a3.
is easily
It is obvious from the definitions 45.8 that a‘- a, = 6,', and it
triple scalar
verified that [a,a,a3] = zg, where g = |g;,|, and that the
so that [a!a’a*] =
products [a'a’a*] and [a,a.a3] are reciprocally related,
1 IV.g. Moreover,
a’ x a° a® x a’ =
XA 2

(45.9) ay = ———_ ao —= ————_

[ata’a?] | S iia atau, Se ia'a'a|,

|sin (a1, a,)|, and so oriented that
3 We recall that a, X a, is a vector of length aq,
anded system. The triple scalar product [a,a,a,], on the
a,,a, and a, X a, forma right-h on the
parallelepiped constructed
other hand, is numerically equal to the volume of the
is a set of base vectors in E,, the reciprocal basis {ai} is
vectors a,, as, a3. If {a;}
determined by a;- a’ = 0’.
122 GEOMETRY [CHAP. 3

as can be readily checked with the aid of (45.8). In view of this it is

natural to call the system of vectors a', a”, a® the reciprocal base system.
We observe that if the vectors a, aj, a3 are unit vectors associated with
an orthogonal cartesian system of coordinates, then the reciprocal system
of vectors defines the same system of coordinates.
Using the reciprocal base system, we can write the differential of a
vector r in the form dr = a‘ dz;, where the dx, are the appropriate com-
ponents of dr. Then
ds® = dr + dr = (a' dx;) « (a’ dx;)
= a’-a? dz, dz,
= g'' du, dx;,
where
(45.10) gi=a’-a’ = 2".
It is not difficult to check that the coefficients g’’, defined by the formula
45.10, coincide with the quantities g’’ defined earlier. Thus, making use
of formulas 45.8 and 45.9, we can readily show that g,,g’* = 6,’, and the
solution of this system of equations for the g’* gives g’* = G’*/g, where
G’* is the cofactor of the element g;, in the determinant |g;,;|. Thus the
definition of g*’ given in Sec. 44 follows as a theorem from the definition
45.10.
The system of base vectors determined by (45.8) can be used to represent
an arbitrary vector A in the form A = a‘4,, where the A; are the covariant
components of A. If we form the scalar product of the vector 4,a‘
with the base vector a;, and note that the latter is directed along
the X’-coordinate line, we get A,a'-a; = A, 6;'= A;. Thus AN 83; (no
sum on j) is the length of the orthogonal projection of the vector A on the
tangent to the X’-coordinate curve at the point P (Fig. 19), whereas
AN 83; 18 the length of the edge of the parallelepiped whose diagonal is
the vector A.
Since
A-=_a,;A' = afA,,
we have
a,:a,A'=a'-a,A,,
or
gijA' = O14, = A,.
We see that the vector obtained by lowering the index in A’ is precisely
the covariant vector A;. The two sets of quantities 4* and A, are thus
seen to represent the same vector A referred to two different base systems.
As has already been noted, the distinction between the covariant and
Sec. 46] MEANING OF COVARIANT DERIVATIVES 123

xe

Fig. 19

contravariant components of A disappears whenever the base vectors a;

are orthonormal.
Similarly, if we consider the coefficients 4;;, in a multilinear form
A,,,a’a’a® and require that A,;,a‘a’a® = A‘*a,a,a,, then the set of quantities
{4%*} represents the same tensor A when referred to the basis {a,}. All
associated tensors (see Sec. 30) represent the same tensor A in suitable
base systems.

46. On the Meaning of Covariant Derivatives

Let A be a vector localized at some point P(y', y*, y*) of Es referred to

an orthogonal cartesian frame Y. If at every point of some region R
about P we have a uniquely defined vector A, we refer to the totality of
vectors A in R as a vector field. We suppose that the components of A
are continuously differentiable functions of y* in R, and, if we introduce
a curvilinear system of coordinates X by means of the transformation
Toe ee (a)
tiable
the corresponding components A‘(x) will be continuously differen
position vector
functions of the point (21, x?, «*) determined by the
r(z!, 2", 2°). In the notation of Sec. 45, the base vectors in the X-reference
frame are a, = Or/0x’, so that A has the representation
(46.1) A = Aéa,.
change AA
We will be concerned with the calculation of the vector
n
in A as the point P(a', x, 7°) assumes a different positio
P'(at + Aal, 2? + Ax?, a? + Xz),
124 GEOMETRY [CHAP. 3
From (46.1) we have

AA = (A? + AA*)(a; + Aa,) — A’a,

= AA‘a,; + A’ Aa; + (AA‘)(Aa,).

As in ordinary calculus we denote the principal part of the change by dA,

and write

(46.2) dA = a; dA‘ + A‘ da,.

This formula states that the differential change in A arises from two
sources:
(a) Change in the components A’ as the values (x1, 2, 2°) are changed.
(b) Change in the base vectors a; as the position of the point (2, 2°, x°)
is altered.

The partial derivative of A with respect to x’ is defined as the limit of

the quotient,
oA Boek:
in —— >= ——
Aai+0 Aa? “og Ox? ;

and it follows from the expression for the increment AA that

46.3 oA _ OA’ da,

She
pea
he mer 2 az
erat
We show next that the vector defined by formula 46.3 is identical with
the covariant derivative of the vector A‘. First we establish the identity

(46.4) sah pp
Ox? ij)

We recall that g,; = a;-a,;. Hence

08%; kee da; da; a

CE sae ae en, gen

Permuting the indices in this formula, we get

Og, _ Oa, da, se

Qu! Oxi * * ant ™
—— a = ;

O8 ixstseset
da; da
ome s 43
Oxt Ox' Met a;
Sec. 46] MEANING OF COVARIANT DERIVATIVES 125
If we assume that T is of class C?, then

da, ae da;

Ox? = Or" .

warn i+)
We form

pd c z
and obtain
Ca;
(46.5) =o “a, = [ij, ki).
It follows from (46.5) that
0a; >
ope om [ij, k]a*.

Hence
da; a 2 * k a
ae = {ij,kla°-a

— [ij, k]g"*

from which it follows that

This establishes the identity 46.4.

Inserting this result in (46.3), we get

[+f Ox

and the expression in the bracket is precisely A*,. Thus

ij

0A
au! =
fe
A* Ay:
(46.6)

A%, of the vector A*

It follows from (46.6) that the covariant derivative
of dA/dx’
is a vector whose components are precisely the components
referred to the base system a;.

4 eet Ey (GR en (ES Pe

Hots ata ss ani awi\ dat) = Awt \art dnt”
126 GEOMETRY [CHAP. 3

We can also show that if A is represented in the form

(46.7) A = A,a’,
then
0A <
(46.8) aa = A, ah.

From a‘-a,; = 6} we have

da® a
age “Oak :
Therefore

ae a,=—a'- aS
Ae oa

= —a'-a,
at jk
by (46.4). Since a‘- a, = 64, the foregoing result is epuivalent to
dat ,= ("|
Ox” 4 jk)”
Hence

(46.9) vs | at
pe jk
The differentiation of (46.7) with respect to 2* and the substitution from
(46.9) lead at once to (46.8).
We observe that, if the Christoffel symbols vanish identically in R,
the reference frame associated with these symbols is cartesian (see Theorem
I, Sec. 39), and, in this case, the base vectors a; are independent of the
0A OA?
coordinates x'. The formula 46.3 then states that —— = — a,, and
OAt Ox? Ox?

hence A’. =
Ox) ~

47. Intrinsic Differentiation

Let a vector field A(x) be defined in some region of E;, and let
Cc: 2) = 20), rat = ks,
be a curve in that region. The vectors A(x), defined over the one-dimen-
sional manifold C, depend on the parameter ¢, and if A(z) is a differentiable
vector and the x*(t) belong to the class C!, then

dA _OA dx!
dai ae dix,
SEC. 47] INTRINSIC DIFFERENTIATION 127.
By virtue of (46.6) this can be written

= [2+ bee
dite
The vector 6A*/6t, defined by the formula
a

hijlowdt
and

dA aie (« = 1, 2, 3),
(47.1)
vel z 5 x

6t dt ij) dt
is called the absolute or intrinsic derivative of A* with respect to the
parameter f.
Following McConnell’ we will make free use of intrinsic differentiation
in the treatment of geometry of curves and surfaces.
If the vector field A* is defined in the neighborhood of C, as well as
on C, we can write
a

oAot = A's ~dt ,

and it follows that the familiar rules for differentiation of sums, products,
etc., remain valid for the process of intrinsic differentiation. If A is a
scalar, then, obviously, 64/dt = dA/dt.
The extension of the process of intrinsic differentiation to tensors of
rank greater than one is immediate. Thus we write

hin Win (Vag, tet_

af) a ear MBit
(ely, dot
at
(oy,
kph
oodt
Ot dt

Of is = 0, the fundamental tensors g,; and g’’ can

We observe that, since
Ot
be taken outside the sign of intrinsic differentiation.
Problems

d ie, OA! j «,
1. Prove that 4 SAA) = 2¢;;A° ape

aA, A;
2. Show that A, ,— 4;; = weep

d os a i hae OB!
3. Show that i SAB) are Bi + 9,;Aé ae

4. If A, = g;;4’, show that A; , = %i4%,-

5 Compare A. J. McConnell, Absolute Differential Calculus, pp. 156-162.
128 GEOMETRY [CHAP. 3

a a agi
5. Show that =,(g,,4‘B’) = A,B + A'B,,,.
6. Prove that if A is the magnitude of A‘, then A , = A, ,A‘/A.
7. If y’ are rectangular cartesian coordinates, show that in E3
i ay? ay? y a ay? Oa?

[28,7] = Sa agP Oa? ap) dx dx? dyt ”

These formulas are often found to be more convenient for the computation of
Christoffel’s symbols than the defining formulas 31.1 and 31.2.

48. Parallel Vector Fields

Consider a curve (Fig. 20),

C: ac ar a(t), ty < t = tp, G a 1; as aos

drawn in some region of £3, and a vector A localized at some point P of C.

We suppose that the functions x(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 the
curve C. We will deduce a set of necessary and sufficient conditions for
a vector field to be parallel.
If A is a parallel field along C, then the vectors A do not change along
the curve and we can write dA/dt = 0. It follows, upon noting (47.1),
that the components A’ of A satisfy a set of simultaneous differential
equations 0A‘/dt = 0, or, when written out in full,

48.1 dA’ bit dx

oY dt \«p) at

Fig. 20
Sec. 48] PARALLEL VECTOR FIELDS 129

We can show, conversely, that every solution of the system 48.1 yields
a parallel vector field along C. Indeed, from the theory of differential
equations it is known that this system of three first-order differential
equations has a unique solution when the values of the components At
are specified at a given point of C. But it was shown previously that the
vector field formed by constructing a family of vectors of fixed lengths,
parallel to a given vector, satisfies the system. Hence every solution of
equation 48.1 satisfying the initial conditions must form a parallel field
along C.
Let A‘(t) and B‘(t) be any two solutions of the system 48.1. We verify
that the lengths of vectors A‘ and B* indeed do not change as we move
along the curve. Moreover, the angle 6 between the vectors A’ and
B‘ remains fixed as the parameter ¢ is allowed to change. To prove
this we note that (Sec. 44) A- B = ABcos 6 = g,;A‘B’, and, if g,,A'B’
d = Sa
is to remain constant along C, then i (¢,,A'B’) = 0. But g,,A‘B’ is an
invariant, and, since the g,; behave like constants in the process of co-
variant differentiation, we can write
d Aig ep
ne
—(g..A'B’)) = —pe tsALB" )

g,OA
== 813 Bi
St Bor 834
; OB"
Ot e

0 and
Since, by hypothesis, the fields A’ and B’ satisfy (48.1), 64*/6t =
follows
6B‘/dt = 0, and we conclude that g,,;A‘B’ is constant along C. It
that, if A‘ = B’, then g,,A'A’ = A? is constan t
directly from this result
along C, and this implies that 6 = constan t.
extended to
The notion of a parallel vector field along a curve can be
over three- dimens ional Euclid ean manifolds.
define parallel vector fields
P(x) and a vector A localiz ed at P. If we construct
Thus consider any point
magnitude and
at every point of the manifold a vector equal to A in
vector field in the
parallel to it in direction, there will result a parallel
passing through P,
space of three dimensions. If a curve C is drawn
l field along C,
the vectors 4‘ of the field lying on C will form a paralle
A‘ are defined at
and will thus satisfy (48.1). However, since vectors
every point («*) of the manifold, we can write
da’
‘dt
_ aAtde
dx" dt’
so that equations 48.1 assume the form
0A’
phale
i AZ \=
=— =)0;
ice¥" bed dt
130 GEOMETRY [CHaP. 3

This must be true for all curves passing through P, that is, for all values
of dx"/dt. Accordingly, the parallel vector field in E; satisfies the system
of equations
oA + [lato or A, =0.
Cx ak)
The converse follows, as previously, from the existence and uniqueness of
solutions of such systems of differential equations.
The condition for a parallel displacement of a covariant vector A; is

cA fs4 atnoe dake: Mhansdsboee “hy

Ox* ik)
This follows from the observation that A;;,= g,,4% whenever A; = g,,;A?.

49. Geometry of Space Curves

Let the parametric equations of the curve C in E3 be

Coat =a"), (ioiows,; (j=, 2,3):

The square of the length of an element of C is given by

(49.1) ds =e, da dx’,

and the length of arc s of C is defined by the integral

(49.2 =) te Ae er
dx dx?
Say
) Node cat
From (49.1) we see that

49.3 tee
( ) eet ds ds

and, if we set dx*/ds = 4*, equation 49.3 can be written as

(49.4) ~ByAtd? = 1.
Thus the vector A, with components A‘, is a unit vector. Moreover, A is
tangent to C, since its components 4‘, when the curve C is referred to a
rectangular cartesian reference frame Y, become A‘ = dy'/ds. These are
precisely the direction cosines of the tangent vector to the curve C. We
shall assume throughout this discussion that the curve C is of class C2,
so that it has a continuously turning tangent at all points of C.
Consider a pair of unit vectors A and p (with components 4‘ and py‘,
respectively) at any point P of C (Fig. 21). We suppose that A is tangent
SEc. 49] GEOMETRY OF SPACE CURVES 131
A+ dr

Fig. 21

to C at P. The cosine of the angle 6 between A and yp. is given by the

formula
(49.5) cos 0 = 9,471",
and, if A and pw are orthogonal, (49.5) requires that
(49.6) giA'w’? = 0.
Any vector p. satisfying equation 49.6 is said to be normal to C at P.
If we take the intrinsic derivative with respect to the arc parameter s
of the quadratic relation 49.4, and recall that the g;,’s behave in covariant
differentiation like constants, we obtain
bi on
gh Vi + PS haa ihe = 0.
ae Os fe Os
Since g,,; is symmetric, the foregoing result can be written in the form
0A OA? :
Bisa" rae 0. We see that the vector ey either vanishes or is normal
s s
to C, and if it does not vanish we denote the unit vector codirectional
or ,
with aaa by mw’, and write
s
ce

(49.7) —,
w=-[i=je)

x OS
where x > 0 is so chosen as to make yw’ a unit vector.
the principal
The vector yw’, determined by the formula 49.7, is called
e of Cc
normal vector to the curve C at the point P, and ~ is the curvatur
at the point in question.
normal
The plane determined by the tangent vector A and the principal
vector p. is called the osculat ing plane to the curve C at P.
132 GEOMETRY [CHaP. 3

Since w is a unit vector,

(49.8) gee’ = 1,

and we can treat this quadratic relation just as we did g,,4'A’ = 1 and
, WU
Ou ; : , OH j
deduce the orthogonality of vectors ae and pw’; that is, g,,u Fre 0.
Moreover, differentiating intrinsically the orthogonality relation 49.6,
we get
jae ny
8ij 5s fad T 833
rh aon ot A 3s = 9,

or

ieAne te
813 3s Sis 5 he

= — xg; jw?
— —— 0

where we used equation 49.7 and the quadratic relation 49.8. Thus

(49.9) ie,
Os
and, since g,,A‘A’? = 1, we can write (49.9) in the form

bu? ;
which shows that the vector — + xd?’ is orthogonal to A*. This result
s
shows that if we define a unit vector v, with components »’, by the for-
mula

(49.10) y= 2 (2 = xi’),

the vector v will be orthogonal to both A and w. We agree to choose the

sign of 7 in such a way that
(49.11) Vg eindiwiv® = 1,
so that the triad of unit vectors A, u,v forms, at each point P of C,a
right-handed system of axes.®
5 We deduce from (41.2), and from the definition of the triple scalar product (Sec. 45),
that
AL pt yt
CipAipiv’=| 02 pe v2] = —=A-UXY.
A us 93 &
SEC. 49] GEOMETRY OF SPACE CURVES 133
Oy’ 2
Since e,,, is a relative tensor of weight —1 (Sec. 41), and g = >

= : dx’

it follows that e€,;, = Vg é,;, 18 an absolute tensor, and hence the left-hand
member of (49.11) is an invariant. An algorithm of division suggests
that v* in (49.11) is determined by the formula

(49.12) ye = eA,

where A; and w; are the associated vectors g,,A* and g;,u*, and

elik = 1 tik

/&
is an absolute tensor. The validity of this expression follows from an
observation that (49.12) satisfies the conditions of orthogonality g,,A‘r’ =
0, g,;40°v? = 0, and the equation 49.11 determining the orientation of the
unit vector v relative to A and w. The number 7 appearing in equation
49.10 is called the torsion of C at P, and the vector v is the binormal.
In order to reconcile these definitions with the usual definitions of the
principal normal and curvature given in elementary vector analysis, we
recall the formula 46.6, 0A/0x* = A%.a,,
Fame)
and note that if the vector field
A is defined along C, we can write

@Ada’ ya da
DI ERO
: 7 ce eee OA ts
Using the definition of intrinsic derivative, —— = A%;—— , we can write
Os eas
the preceding result as
aA pA*
49.13 peony a,.
( ds Os

Let r be the position vector of the point P on C; then the tangent

vector A is determined by

ds ,

and (49.13) gives for the curvature vector

dv) —=—a,=C,
dh OMe :
14 Gai
ds me ds Os
Ce

where ¢ is a vector perpendicular’ to A.

= 0.
7SinceA-A = 1,A-dA/ds
134 GEOMETRY [CHAP. 3

With each point P of C we can associate a constant x, such that ch =

is a unit vector. We can now rewrite (49.14) in the form

wr:

= pa,
where, in the last step, we have made use of the formula 49.7.

50. Serret-Frenet Formulas

This section contains a set of three remarkable formulas, generally

known as Frenet’s formulas, which characterize, in the small, all essential
geometric properties of space curves. Two of these formulas have already
been derived in Sec. 49. They are

(50.1) Sag ge. n> \,

os
and

(50.2) as =v — xi.
Os
The first of these gives the rate of turning of the tangent vector A as the
point moves along the curve, and the second that of the principal normal
w. The third formula,

(50.3) wae —Th,

os
to be derived next, specifies the rate of turning of the binormal as the
point P moves along the curve.
If we differentiate equation
[49.12] ye == efIE]
yg,
intrinsically, we get
50.4 v®
MAES sie ease
OAs RID apeyng i OFF3
O08) Os : Os ae és
since the covariant derivatives of «*’* are zero.8 Lowering the indices in
OA; Ou;
(50.1) and (50.2) we get — = xu, and eS Tv; — x4,; and inserting
és os :
8

For the «**’s are constants in a cartesian system, hence ¢'}* = 0, and this is a tensor
: ik? . . Soh . .

equation!
Sec. 50] SERRET-FRENET FORMULAS 135

these values in (50.4) yields

Ov" = tik yuu; + iskA (ty; — xA;)

Os
ax EAL
oad
= —T’,k
since «*1,A, = «’™u,u; = 0, because the e'* are skew-symmetric, and
p® = €*')y,. This establishes (50.3).
Formulas 50.1, 50.2, and 50.3, when written out explicitly in terms of
Christoffel’s symbols, assume the forms
az ke
: dia!
2 ant
| fi\de!
= j
dx k '
xu’, or
ds \jk) ds * ds” lik/ ds ds
du'
(50.5) ot:
ds Lx ds
z k

oe [iS —rp.
ds ds
Except for position of the curve C in space, the system 50.5 determines
the curve uniquely when continuous functions x(s) and a(s) are specified
along C.
We conclude this section by considering an example illustrating the use
of Frenet’s formulas. Consider a curve, defined in cylindrical coordinates
by equations
i = a
x? ==. 0s),
Teas
of arc in
This curve is a circle of radius a. The square of the element
cylindrical coordinates is
ds? = (dx)? + (a1)*(dx?)? + (dx*)’,
and it is easy to
so that g,, = 1, goo = (#')’, 833 = i 2, = 0, fi,
verify that the nonvanishing Christoffel symbols are

bal=-# (nd bl =
22
a,
12
—
21
C are A‘ = dzx'/ds,
=—,
foe

The components of the tangent vector A to the circle

A is a unit vector, gah =
so that J} = 0, 42 = d6/ds, 42 = 0. Since
at all points of C, and this requires that
136 GEOMETRY [CHAP. 3

Therefore (d0/ds)? = 1/a?, and the first formula in (50.5) yields

wit Ey
ds
(Medds (ada
\jk \22)” ds
2,a
> dA? (2 ede aga
Se 1? — = 0,
eds | \yihamedeetmal St ees
3 ke
a = aM (2|ige 0.
ds jk ds

Since p is a unit vector, g;,4°uw? = 1, and it follows that x = 1/a and

oe=-1w=0,4=0.
An entirely analogous calculation shows that 7 = 0 and » = 0, »* = 0,
w= 1,
Problems
1. Find the curvature and torsion at any point of the circular helix C whose
equations in cylindrical coordinates are

C2] = 6 g* = KO.

Show that the tangent vector 4 at every point of C makes a constant angle with
the direction of the X-axis. Consider C also in the form y! = acos@, y? =
asin 0, y> = k0, where the coordinates y’ are rectangular cartesian.
2. Show that
OPA?
istieidss
s akae gem
Out = dr Pa
ie
pale ee deay Whas (x?2 + 7*)u
Das 8 es

St yt
a ae t(xA* — rv?) —— wt

3. Using results of Problem 1, show that the ratio of the curvature x to the
torsion + is a constant. Show from Frenet’s formulas that whenever 7/x =
constant, and the coordinates are cartesian, »* = cA? + b*, where c and b? are
constants. From this result it follows that A4*b* = constant, so that the curve
makes a constant angle with the lines whose direction ratios are b*. In other
words, the curve is a cylindrical helix. This theorem is due to Bertrand.
4. When C is specified in the form

Cc: yi =y(s),
where the y are orthogonal cartesian coordinates and s is the arc parameter,
show that
x = [(y")"F + (y)’F + [PF
SEC. 51] EQUATIONS OF A STRAIGHT LINE ibe
and
(yy (y?) (y3)’
rx = |(y4)"(y*)"(y*)” |. .
Gy) "G77"

5. Write equations of Problem 2 in cartesian coordinates y* and show that

when + = 0 and x = constant along C, the equations of C are

y’ = A*cosxs + B‘sin xs + C’,

where A*A* = B’B* = 1/x?, A*B? = 0. Thus C is a circle.

6. Let C be a cylindrical helix determined by

y’ = (0),
Cc: (y? = (0),
y* = ka, .k = constant,

where o is the arc parameter of the directrix curve C’ in the y!y?-plane, so that
(ds)? = (dy)? + (dy)?. Note that (ds)? = (1 + k?)(do)? and show that

¢ wy ik
gp” y” (9)

p” yn (0) 1

PE OA
Pies Vie
igh
Lt k* >

and verify that r/x =k.

51. Equations of a Straight Line

Let A’ be a vector field defined along a curve C in E3, where C is given

parametrically as
Gr was), Spats io, (geen 2183);

s being the arc parameter. If the vector field A’ is parallel, then it follows
from Sec. 48 that 6A‘'/ds = 0, or

dA’ i dix?
atl as ae = 0.
GED ds ap ds

We shall make use of equation 51.1 to obtain the equations of a straight

line in general curvilinear coordinates. The characteristic property of
straight lines is that the tangent vector A to a straight line is directed along
138 GEOMETRY [CHaP. 3

the straight line, so that the totality of tangent vectors A forms a parallel
vector field. Thus the field of tangent vector A* = dx‘/ds must satisfy
(51.1), and we have

Si eae
és | ds af) ds ds
The equation
ax’ |i ee dx?
(51.2)
ds” aB) ds ds
is the equation sought. In cartesian coordinates the Christoffel symbols
vanish and we obtain the familiar form of differential equations of straight
lines. From the geometric interpretation of the curvature x as a measure
of the rate of turning of the tangent line to a curve, we are led to define
the curvature of a straight line to be zero. This definition is consistent
with the first of Frenet’s formulas 50.1.

52. Curvilinear Coordinates on a Surface

In the remainder of this chapter we will study the properties of surfaces

imbedded in a three-dimensional Euclidean space. It will be shown that
certain of these 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 ofarc of a
curve drawn on the surface. All such properties of surfaces are termed
the intrinsic properties, and the geometry based on the study of this
differential quadratic form is called the intrinsic geometry of the surface.
We find it convenient to refer the space in which the surface is imbedded
to a set of orthogonal cartesian axes Y, and regard the locus of points
satisfying the equation

(52.1) Fy’, y?, y3) = 0

as an analytical definition ofa surface S. We suppose that only two of
the variables y* in (52.1) are independent and that the specification of,
say, y' and y? in some region of the Y¥1¥*-plane determines uniquely
a real number y? such that the left-hand member in (52.1) reduces to zero.
If we suppose that F(y’, y, y°), regarded as a function of three independent
variables, is of class C1 in some region R about the point Po(yo', Yo, Yo?)
be OF
with re #0 and F(y!, Yo”, Yo?) = 0, then the fundamental theorem
Po
on implicit functions guarantees the existence of a unique solution
y° = fly’, y°), such that yo* = f(yo', Yo°)-
Sec. 52] CURVILINEAR COORDINATES ON A SURFACE 139
The definition of the surface by means of a single equation 52.1 is less
convenient than the one introduced by Gauss, who defined the surface
as a locus of points satisfying three equations of the type

(52.2) y' = y'(u, 1).

where u,! < u! < u,! and u,? < uw? < u,”, and the y’ are real functions
of class C1 in the region of definition of the independent parameters
u', u®. In order to reconcile these two different definitions we shall require
that the functions y‘(u1, u?) be such that the Jacobian matrix

dur dui du}

(52.3)
ay ay? ay
Ou Ou du

be of rank two, so that not all the determinants of the second order
selected from this matrix vanish identically in the region of definition of
parameters u‘. This requirement ensures that it is possible to solve two
equations in (52.2) for wu! and u* in terms of some pair of variables y’,
and the substitution of these solutions in the remaining equation leads to
an equation of the form y* = y°(y', y*). It should be remarked that,
if any two determinants formed from the matrix 52.3 vanish identically,
then the third one also vanishes, provided that the surface S is nota
plane parallel to one of the coordinate planes.
Since u! and w? are independent variables, the locus defined by equations
52.2 is two-dimensional, and these equations give the coordinates y’ of a
point on the surface when u! and wv? are assigned particular values. This
point of view leads one to consider the surface as a two-dimensional
manifold S imbedded in a three-dimensional enveloping space E3. We
can also study surfaces without reference to the surrounding space, and
consider parameters u' and u? as coordinates of points in the surface.
A familiar example of this is the use of the latitude and longitude as
coordinates of points on the surface of the earth.
If we assign to wu! in (52.2) some fixed value u* = c (Fig. 22) we obtain
as a locus the one-dimensional manifold
y* = y(c, u’), (i aa is 2, 3):

We
which is a curve lying on the surface S defined by equations 52.2.
the v2-curve . Similarl y, setting vu? = constant in (52.2)
shall call this curve
u’ and
defines the w!-curve, along which only wu’ varies. By assigning to
on the surface,
w2 a succession of fixed values, we obtain a net of curves,
140 GEOMETRY [CHAP. 3

Fig. 22

which are termed the coordinate curves. The intersection of a pair of

coordinate curves obtained by setting u’ = u)', v2 = uy” determines a
point Py. The variables wu’, u? determining the point P on S are called
the curvilinear, or Gaussian, coordinates on the surface.
Obviously the parametric representation of a surface in the form 52.2
is not unique, and there are infinitely many curvilinear coordinate systems
which can be used to locate points on a given surface S. Thus, if one
introduces a transformation
ur = u'(i’, a”),

(52.4) : weroet
Tian igFigg Te)
where the u*(a',u”) are of class C! and are such that the Jacobian
O(u, u? esti s: ; d
J= a does not vanish in some region of the variables #*, then one
Tih
can insert the values from (52.4) in (52.2) and obtain a different set of
parametric equations,
(62:5) y’ = f*(, u?), (i = 1, 2, 3),

defining the same surface S. Equations 52.4 can be looked upon as

representing a transformation of coordinates in the surface precisely in the
same way as equations 2’ = «‘(%!, x, z), (i = 1, 2,3) were viewed as
defining a transformation of coordinates in Es.

53. Intrinsic Geometry. First Fundamental Quadratic Form.

Metric Tensor

We remarked in Sec. 52 that the properties of surfaces that can be

described without reference to the space in which the surface is imbedded
are termed intrinsic properties. A study of intrinsic properties is made
SEC. 53] INTRINSIC GEOMETRY 14]
to depend on a certain quadratic differential form describing the metric
character of the surface. We proceed to derive this quadratic form.
It will be convenient to adopt certain conventions concerning the
meaning of the indices to be used in this and remaining sections of this
chapter. We will be dealing with two distinct sets of variables: those
referring to the space EF, in which the surface is imbedded (these are
three in number), and with two curvilinear coordinates u! and uv? referring
to the two-dimensional manifold S. In order not to confuse these sets
of variables we shall use Latin letters for the indices referring to the space
variables and Greek letters for the surface variables. Thus Latin indices
will assume values 1, 2, 3, and Greek indices will have the range of values
1,2. A transformation T of space coordinates from one system X to
another system YX will be written as

a transformation of Gaussian surface coordinates, such as described by

equations 52.4, will be denoted by
u* = u*(i', U”).
A repeated Greek index in any term denotes the summation from | to 2;
a repeated Latin index represents the sum from | to 3. Unless a statement
to the contrary is made, we shall suppose that all functions appearing
in the discussion of the remainder of this chapter are of class C® in the
regions of their definition.
Consider a surface S defined by

(53.1) y =y'u', w),

where the y? are the orthogonal cartesian coordinates covering the space
E, in which the surface S is imbedded, and a curve C on S defined by

(53-2) u~ = u(t), tts bo,

where the u*’s are the Gaussian coordinates covering S. Viewed from
the surrounding space, the curve defined by (53.2) is a curve in a three-
dimensional Euclidean space and its element of arc is given by the formula
(53.3) ds? = dy’ dy’.

From (53.1) we have

; oy’
(53.4) ayy = —
Bu? au’,
where, as is clear from (53.2),
jie = dy dt.
t
142 GEOMETRY [CHAP. 3

Substituting from (53.4) in (53.3), we get

du* du?
= d,s, du du’,
where
dy’ oy’
53.5 aap = —
( ) Out au?
The expression for ds’, namely,

(53.6) ds =a, duaau.

is the square of the linear element of C lying on the surface S, and the
right-hand member of (53.6) is called the first fundamental quadratic form
of the surface. The length of arc of the curve defined by (53.2) is given
by the formula
t —<——
s= ‘Vapi? dt,
ty

where u* = du*/dt. Since in a nontrivial case ds? > 0, it follows at once

from (53.6) upon setting uw = constant and uw! = constant in turn, that
ds?) = ay,(du')? and ds?) = ayo(du?)?. Thus ay, and dy are positive
functions of ut and u?.
Consider a transformation of surface coordinates

(53.7) u* = u%(u', a),

ed

with a nonvanishing Jacobian J = 3a? |: It follows from (53.7) that

du* = a du’,
ou’
and hence (53.6) yields

If we set
: Ou duP
a.s 76 = A,,; —ya—
=F On due”
we see that the set of quantities a,, represents a symmetric covariant
tensor of rank two with respect to the admissible transformations 53.7
of surface coordinates. The fact that the a,, are components of a tensor
is also evident from (53.6), since ds* is an invariant and the quantities
a,3, are symmetric. The tensor a,; is called the covariant metric tensor
of the surface.
Sec. 53] INTRINSIC GEOMETRY 143
Since the form 53.6 is positive definite, the determinant
Gy, Aye
a= > 0,
Gs, Age “

and we can define the reciprocal tensor a*’ (see Sec. 30) by the formula
a“’a,, = 63. Thus we have
qu a 22. Ges gst
9 =a 12 a esa
a a a
The contravariant tensor a*’ is called the contravariant metric tensor.
We can repeat, almost verbatim, the contents of Sec. 44 concerning
metric properties of our two-dimensional space S. Thus the direction of a
linear element in the surface can be specified either by the direction
cosines dy'*/ds, (i = 1, 2, 3), or by the direction parameters

du*
(53.8)
53.8 jes Fe
For,
dy!_ dy‘du’
ds Ou" ds

and the du*/ds are uniquely determined when the direction cosines dy'|ds
are specified, and conversely. We define the length of the surface vector
A*, that is, the vector determined by A}(w’, u2) and A(u1, u?), by the
formula’
A = Va,,A7A?.
It follows from (53.6) that
du* du?
1 =a agp oe
ds ds
= ayaa,

so that the direction parameters 4* are components of a unit vector.

The covariant vector
(53.9) Ay = eA
from (53.9) that
is sometimes called the direction moment, and it is clear

a’, = aah" = 6,'A" = 2,

and that
(53.10) MA, = Gegh Ae
9 The oe pe ee sf of the vector A%, as viewed ees enveloping space Es,

are given by 4‘ = = A" and it is clear that 4‘4‘ = aa an A%AP = aypAtal,

144 GEOMETRY [CHAP. 3

54. Angle between Two Intersecting Curves in a Surface.

Element of Surface Area
The equations of a curve C drawn on the surface S can be written in
the form
Ge = Oy), dip Se aly
Since the u*(t) are assumed to be of class C?, the curve C has a continuously
turning tangent. Let C, and C, be two such curves intersecting at the
point P of S (Fig. 23). We take the equations of S, referred to orthogonal
cartesian axes Y, in the form
(54.1) y =y'u, wv),
and denote the direction cosines of the tangent lines to C, and C, at P
by &* and 7’, respectively. The cosine of the angle 6 between C, and C,,
calculated by a geometer in the enveloping space Es, is
(54.2) ° cos OE",
However,

a Oy! du? day’

Ou* ds, ds,
z
ni = dy’ dau” _ doy’
du’ ds, — dsq
>

where the subscripts | and 2 refer to the elements of arc of C, and C,,
respectively. Using the definition 53.8, we can write the unit vectors in
the directions of the tangents to C, and C, as
w= d,u* 2 _ dou"
ds,” # ds,

y?

Fig. 23
Sec. 54] ELEMENT OF SURFACE AREA 145
and
Oy? ay
ess B
(54.3) ohm 1
du" ine Out 5
Inserting in (54.2) the expressions from (54.3), we get

cos 9 = Oy’ 40 P
du* du?
and since
a Oy’ oy
du* du?
the foregoing expression can be written
(54.4) COs 0 = a,,A7y".
If the curves C, and C, are orthogonal,
(54.5) fAte 0.
In particular, if the surface vectors 4* and yw” are taken along the coor-
dinate curves (A) = Vay, A= Oust =); 1° 1/~/a9), then it follows
from (54.5) that the coordinate curves will form an orthogonal net if, and
only if, ay. = 0 at every point of the surface.
We can give a pictorial interpretation of these results in the manner
of Sec. 45. Thus, if r denotes the position vector of any point P on the
surface S, and the b, are the unit vectors directed along the orthogonal
coordinate axes Y, then equations 53.1 of the surface S can be written in
vector form (see Fig. 24) as
r(ui, u?) = by*(ul, uv).

Ye or

Vayl

Fig. 24
 146 GEOMETRY [CHAP. 3

It follows from this representation of S that

Or mor
ds?
= dr- dr = — - — du* du?
; : du* du? :
= 4, du* du®,
where
Or or
54.6 au = ——",
( ) ’ aut dub
Setting Or/du* = a,, where a, and a, are obviously tangent vectors to the
coordinate curves, we see that

ay, = 4° 4), yz = a, ° as, Ag = Az ° Ap.

In the notation of (54.3) the space components of a, and a, are &* and 7',
respectively.
We can define an element of area do of the surface S by the formula
do = |a; X a,| du du*,

and it is readily verified that the right-hand member of this expression

can be written
(54.7) do = / Q414429 — Ay_” du) du?
= iha du) du2.

This formula has precisely the same structure as the expression 44.11
for the volume element.
It follows from Sec. 40 that the skew-symmetric e-systems, in a two-
dimensional manifold, can be defined by the formulas

C11 = x9 = el =e” = 0, eM = —e) = ej), = —ey = |,

and, since these systems are relative tensors (see Sec. 41), the expressions

Eup = Ja Cup and 7h => tS eh

a
are absolute tensors. Using the e-symbols, we can write the sine of the
angle 0 between two unit vectors A*, u* in the form

€,pA“u? = sin 8,
which is numerically equal to the area of the parallelogram constructed
on the unit vectors 4* and u*. It follows from this result that a necessary
and sufficient condition for the orthogonality of two surface unit-vectors A*
and y* is |e,,A*u"| = 1.
Sec. 55] FUNDAMENTAL CONCEPTS OF CALCULUS 147
Problems

1. Show that the cosine of the angle 6 between the coordinate curves wand u
on S is cos 9 = ays/ V.a4,499.
2. Find the element of area of the surface of the sphere of radius r if the
equations of the surface are given in the form:
yi =rsinuwicosuv, y2=rsinuwsinuw, y? =rcosu,
where the y’ are orthogonal cartesian coordinates. (Note that in this case
ayy = fetch ay = 0, Azo = r? sin? us)

55. Fundamental Concepts of Calculus of Variations

The most celebrated problem of intrinsic geometry of surfaces is

concerned with the determination of curves of shortest length joining
two specified points on the surface. This is the problem of geodesics.
This problem has such profound implications on the formulation of the
fundamental principles of optics, dynamics, and mechanics of deformable
media that it is desirable to treat it in greater generality than one would
-if concerned solely with the geometry of surfaces imbedded in E3. To
do this we shall draw on certain concepts in the calculus of variations.
Since we will be concerned with the study of extremal properties of
integrals, we shall recall some salient facts about the problem of relative
maxima and minima of functions of several independent variables.
Let {(v1,.27,.... +, 2) bea continuous function of n independent variables
x? defined in a bounded, closed region R. We are interested in determining
a point P(x) of R at which f attains an extreme value in comparison with
the values of f in a certain neighborhood of the point P(x). There is no
doubt about the existence of maximum or minimum of f since it is known
that every function continuous in a bounded closed region attains its maximum
and minimum values either in the interior or on the boundary of the region.”
Moreover, if
COR Oe eee)
where
is a differentiable function, then at interior points of the region,
1,2,..., n). The
the function attains its extreme values, Ofjoxt = 0, =
a sufficient
vanishing of the derivatives of f(z", «*,..-, a”) obviously is not
points of the region R at
condition for an extremum. We will call the
ne”):
which 9f/@x* vanish simultaneously the stationary points of f(ainia
d calculus,
The determination of stationary points is studied in advance
to the reader.
and we assume that this subject is quite familiar
ination of
Calculus of variations is also concerned with the determ
ions, but there is an
extreme or stationary values of certain express
10 This theorem is due to Weierstrass.
148 GEOMETRY [CHAP. 3

important distinction in that in calculus of variations we deal with extremes

of functionals rather than functions of a finite number of variables. By
a functional we understand a function depending on the changes of one
or several functions, which assume the roles of the arguments. As an
example of a functional consider the formula

= | ‘V1 + (dyldx)? dx,

defining the length of a plane curve y = y(x) joining the points whose
abscissas are x) and 2,. Here the value of s depends on the behavior of
the functional argument y(x), and the class of functions y(x) on which the
functional s depends is in some measure arbitrary. Thus, one might
consider the problem of determining the extremes of s when y(x) is an
arbitrary continuous function with a piecewise continuous first derivative.
In the study of extremes of continuous functions f(z',..., 2"), of a
finite number of independent variables x’, we must specify the region
R within which fis defined, whereas in the study of extremes of functionals
we must characterize the class of admissible functional arguments. For
example, we may demand that the functional arguments possess certain
properties of continuity, or behave in some specified fashion at the end
points of the interval, and so on. We will be concerned with relative
extremes of functionals, that is, extremes relative to a certain “‘neighbor-
hood” of functional arguments for which the functional takes on an
extreme value, just as we were with relative maxima and minima of
functions. In order to make the notion of the neighborhood of a function
precise, we introduce a
DEFINITION. A function g(x!, x, ..., 2") belongs to the h-neighborhood
of the function f(z',...,%"), provided that |f—g|<h, h>0, for all
values of the independent variables x', x*,..., x” in the interior of R.
With the aid of this definition, we can formulate the fundamental
problem of the calculus of variations as follows: Find, within the class
of admissible functional arguments, those functions f that yield extreme
values for the functional under consideration in comparison with the values
given the functional by functions belonging to some h-neighborhood off.
A word concerning the difficulties inherent in this problem is in order.
We have already remarked that in the theory of maxima and minima
of continuous functions of several independent variables the existence of
extreme values is guaranteed by the theorem of Weierstrass. In the
problem of calculus of variations, on the other hand, it may happen that
the problem is formulated without internal inconsistencies, and yet it
has no solution because of the limitations imposed on the class of ad-
missible functional arguments. For example, let it be required to join
Sec. 56] EULER’S EQUATION IN THE SIMPLEST CASE 149
two given points on the X-axis by the shortest curve with continuous
curvature so that the curve is orthogonal to the X-axis at the end points.
This problem has no solution because the length of every admissible
curve is always greater than the length of the straight line foining the
given points. We can always find a curve of admissible type whose length
differs from the length of the straight line by as little as desired so that
there exists a lower bound of the functional, but this lower bound is not
the minimum attained for any curve of the class of curves under con-
sideration. It follows from this example that in each variational problem
we are confronted with the question of the existence of a solution of
the problem.
In order to deduce the differential equations furnishing a set of necessary
conditions for an extremum of a functional, we need the following
fundamental lemma of calculus of variations.
ty

If the integral | E(t)M(t) dt, where M(t) is a continuous function of

t

t in the interval t, < t < to, vanishes for every choice of the function E(t)
of class C” in ty [t < be, and which is such that &(t,) = &(t.) = 0, then
M(t) is identically zero in the interval ty < t < tp.
We shall prove the lemma by assuming that M(t) # 0 and reaching a
contradiction. Assume M(t) #0 at some point t’ of h}<< t<h, and
suppose that M(t’) > 0. Since M(t) is continuous, there exists a number
such that M(t) > 0 in the interval (¢’ — 6,t’ + 0). Define a
6>0
function &(t) as follows:
E(t)=0, int, <¢t<7, wheres7, =?’ — 6,
Ef) =0, ntzst Sh, where tr, = t’ + 0,

E(t) = (ta ce res (ee Zine > int, < tS 72.

&(t2) = 0.
The function &(t) is surely of class C” in (t, t2) and E(t;) =
For this function, however,

["somear=["someoar>
te

ty
T2

Ti
0.
Thus we reach a
since the integrand is always positive in 7, <1? < 72.
¥ 0 is not tenable.
contradiction, and hence our assumption that M(t)

56. Euler’s Equation in the Simplest Case

is concerned with
The simplest problem of the calculus of variations
the determination of extremals of a functional

(56.1) i(z) = {SRE

150 GEOMETRY [CHapP. 3

where F(t, x, £) is a prescribed real function of its real arguments /, 2,

and # = dx/dt. We shall suppose that F(t, x, z) is of class C*, in some
region R of the plane (a, t), for all values of #.11 Concerning the class
of admissible functions 2(t), we suppose that the values x(7,) and x(t) are
prescribed in advance and that 2(t) is also of class C? int, < t < fp.
Our problem is to find a function

x = f(t), hitch,

called an extremal for the integral 56.1, such that J(x) for x = f(t) assumes
an extreme value in comparison with the values given to J by the admissible
functions in a sufficiently small A-neighborhood of the function x = f(t).
In other words, admissible functions x(t) are such that |z(t) — f(t)| <A
for t; <t<t,. We shall deduce next a necessary condition for an
extremum of J. Consider a function &(t) of class C?, such that &(t,) =
&(t.) = 0, and form a set of functions

&(t) = a(t) + <&(t) = x + Oz,

where ¢ is an arbitrary numerical parameter near zero. The functions

«(t) clearly assume the same values at the end points of the interval
(ty, f2) as x(t). We shall call the #(t) the varied functions, and the quantity
e&(t) = dx the variation of the function x = f(t). For sufficiently small
values of € all varied functions %(t) will be contained in the A-neighborhood
of the extremal x = f(t). Consequently, the integral 56.1,

J(@) = Ja + &) = (0),

considered as a function of e, will have an extreme value fore = 0. A
necessary condition that this be so is ®’(0) = 0.
Because of the restrictions imposed on functions under consideration,
the integral
te
D(e) =| F(t,x + «€,% + €é) dt
th

can be differentiated under the integral sign, and we obtain as a necessary

condition for an extremum the equation

(56.2) 0'(0) = {“(let pais 0

11 These restrictions are more severe than necessary, but we have in mind certain
geometrical problems in which the continuity of second derivatives is a desirable
property.
Sec. 56] EULER’S EQUATION IN THE SIMPLEST CASE 151
which must be true for every &(t) satisfying the conditions laid down in
the definition of &(t). Integrating (56.2) by parts, we get
te te te dF;
(56.3) } F,&(t) dt + Feé(t) -| E(t) — dt = 0,
ty ty ty dt
and, since &(t,) = &(t,) = 0, the foregoing equation simplifies to
tg ;
(56.4) { co]Fe- of) dt = 0.
ne) dt

Since &(t) satisfies the restrictions imposed on &(t) in the lemma of Sec.
55, we deduce from (56.4) that a necessary condition for an extremum
of (56.1) is that a(t) satisfy the differential equation

(56.5) pe len ty,

dt
Expanding (56.5) we obtain

d’x dx
56.6 Fiz — + Fer — + Fu — Fr = 0,
ee dt? 7S tals
and &
where the subscripts denote the derivatives of F(t, x, 2) with ¢, x,
regarded as the independent variables. In order to determine x(t) we
must solve this ordinary differential equation subject to the end conditions
a(t,) = 2, and 2(t,) = 22. Equations 56.5 and 56.6 were first deduced
by Euler and are called Euler’s equations.
The expression (see equation 56.2)

eD'(0) = | EOF: + &(t)F2] dt,

ed at « = 0,
which is akin to the differential of the function D(€) evaluat
by the symbol
is called the first variation of the integral J, and is denoted
6J. Thus
6J = «D'(0).
56.3 and the
Taking into account the left-hand member of equation
definition of 6/, we can write
to
(56.7) 6J = [Fe dalizit + { (F.= 4F,)dx dt,
ty

r of (56.7) vanishes when

where 6a = «&(t). Since the right-hand membe
a(t) is an extremal, we can state a
of the functional J(«)
THEOREM. A necessary condition for an extremum
is the vanishing of its first variation.
152 GEOMETRY [CHaAP. 3

57. Euler’s Equations for a Functional of Several Arguments

Consider next the case of a functional J depending on several functional

arguments 7’, (@ = 1, 2... 771), where.
te
(57.1) 4 =| Fito) nee he os
t zy

As in Sec. 56 we assume that F is a real function of class C? in the 2m + 1-

dimensional space of the real variables ¢, x1,...,2", Z,..., 2”.
We suppose that there exists a set of functions
CDT OY MARS r8 OT Ee
whose values at the end points of the interval are known, and which are
such that (57.1) assumes an extreme value in comparison with the values
given to J by a class of admissible functions belonging to the h-neighbor-
hood of (57.2). We introduce n arbitrary functions &' = &(t),t, <t < ty,
of class C? which vanish for ¢ = t, and ¢ = fy, and construct a family
of admissible functions
Oi3) z= a(t) + f*(0),
where the parameter ¢€ is so chosen as to make the varied paths 57.3 lie
in the h-neighborhood of the curve 57.2.
As in Sec. 56 we form the function
(57.4)
to
. :
D(e) =| F(t a + eb tt" 4 ef ee le ae «&") dt,
ty
which, by hypothesis, has an extremum for « = 0; hence,

d®
Wie)
(57.5) 7 = 0.
e=0
It follows that
te
(S7.6)oy =e } [(Foté! + Fin€!) +--+ 4+ (Fon€" + Fané")] dt = 0,
and the integration by parts gives
te
$b Fand®
tg
OJ = «|Fad

+ ['e(e0—4 pe)ace [elee— en) a] na

ty ty
ty d te d

hy dt ty dt

“To ensure the independence of the integral 57.1 of special modes of parametriza-
tion, we suppose that F(t,x,) is positively homogeneous of degree one in the « (See
Sec. 43).
Sec. 57] EULER’S EQUATIONS 153
Since the &* are arbitrary and vanish at the end points of the interval,
we conclude from the fundamental lemma that

d
(57.7) Fi meine Gi= oa eee Pe -

or
Fa — @Fa3 — GF gta Fy, = 0.

This set of n ordinary differential equations of second order is called

the Euler equations for the variational problem associated with the
functional 57.1. Thus, to obtain the set of functions 57.2, we must
determine the solution of the system 57.7 satisfying the end conditions
(57.8) Pe) aie flim , 1-
2. 275
| (eal
The problem discussed in this section appears to be entirely analogous
to the simpler one treated in Sec. 56, but there is a distinction in that the
not
vanishing of the first variation of (57.1) is a necessary condition
only for an extremum but also for a mixed maximum and minimum the ,
so-called minimax. An integral J(z!,..., 2") may attain a maximum
of the
when the function z(t) is varied and a minimum in the course
d, studied
variation of x°(t). The saddle point of a hyperbolic paraboloi
illustration
in the elementary theory of maxima and minima, is a simple
equations 57.7
of this circumstance. We will call the solutions of Euler’s
the functiona l J. This
satisfying the end conditions 57.8 the extremals of
value assumed
term will be used regardless of the nature of the stationary
neither.
by the functional J, be it a maximum, minimum, or
d that the vari-
In our derivation of Euler’s equations 57.7, we assume
ained by a set of kK<n
ables x‘ are independent. When the «* are constr
functional relations of the form
AY Oe aieco Pe a SR Cee oss
deduced by considering
the set of appropriate Euler’s equations can be
uced by Lagrange.
the free extremum of a certain new functional introd
of free and constrained
To clarify the essential differences in the problems
extrema, consider the functional
t

(57.9) 4 -| F(t, 2}, 22, a, 2) dt,

t1

relation
in which the variables are constrained by the
(57.10) A(t, x}, x?) = 0.
We suppose that the extremal,
xi=ai), t<t<t G=1,2),
154 GEOMETRY [CHAP. 3

satisfies the end condition of the type 57.8. When the constraining
condition 57.10 is written in the form

(57.11) $a,
y,2)=0
by setting « = t, y = x!,z = 2”, equation 57.11 can be thought to repre-
sent a surface referred to a set of cartesian xyz-axes. The extremal must
lie on this surface and we suppose that (57.11) can be solved for z in the
neighborhood of the extremal to yield a differentiable function
(57.12) z= f(@, y).
The substitution from (57.12) in (57.9) then yields an integral of the form
2
(713) J =| F(x, y, y’) dx,
xy
in which the variables x and y are independent, and we can obtain the
Euler equation by minimizing (57.13) on a set of admissible paths that
satisfy the end conditions y(x%,) = y;, y(%2) = yy. This is the problem of
the free extremum already considered in Sec. 56.
However, such reduction of the problem of constrained extremum to
the problem of a free extremum of the functional 57.13 is usually incon-
venient because an explicit solution 57.12 of equation 57.11 may prove
unwieldy. In this event we can follow a procedure similar to that of the
Lagrange multiplier method of obtaining the relative extreme values of
functions of several variables constrained by relations of the type 57.11.
We suppose that d¢/dz ¥ 0, so that it is theoretically possible fo obtain
the solution of (57.11). If this is so, (57.9) can be rewritten in the form

(57.14) J= { Fy, 9 A Se + fy’) az,

since 2 = f5 + 7,7’. Hs
The integrand in (57.14) is a function of x, y, and y’, which we denote by
(67515) F(a,y,y') = Fy. y'fife + fy):
Thus the Euler equation associated with the integral 57.14 is
Ze 7g
(57.16) OF a oF _
Oy dxdy'
On noting (57.15), we see that
OF
Oy a Fy ar Lge a BAGay + SaeY 33

OF
ay a + Fete
Sec. 57] EULER’S EQUATIONS 155

so that (57.16) yields

1 ep ee WY Ro = 0,
Pet Fife at Say + fy) — — fy —— — Fao + fyy’)
dx dx
or G

F, + (F, dF ,
“)— r=
dF,’
0.
hh dx dx
Thus
dFy _ F,
(57.17) oe
TS
dx

since f, is assumed to be defined along the extremal.

On the other hand, the differentiation of (57.11) yields

dy + bf, = 9,
so that

(57.18) f= _ by
d:
extremal,
The expressions for f,, given by (57.18) and (57.17) along the
represent the same function of x; hence

oe ES = pp
(57.19) ee (2),
dy $.
*
where A(x) denotes the common value of the ratios.’
for the extremum
It follows from (57.19) that the necessary conditions
of the integral 57.9 are

dFy _ tp, + A(x)$y] = 0,

(57.20) a
dF,
dx
_ tr 4 A(a)b,) = 0.
by setting = ty = 2’,
When we revert to the original notation
z = x2, we obtain a pair of equations
0d -
dF xi
at at —Ath
— Fi— ow = 9,
A(t)— (i =a )
1, 2),
(57.21)
e.
0, equation 57.11 does not define a surfac
13 We note that if both ¢, = 0 and ¢, =
156 GEOMETRY [CuHaP. 3

which have the structure of Euler’s equations 57.7 for the variational
problem associated with the free extremum of the integral

}EG. 2 DADO, de.

Similar considerations apply to the problem of minimizing the integral
57.1 in which the n variables z* are constrained by a set of k < n relations
(1) Ot, oie Fs) ==, Cj = pee 2k):
If the matrix 0¢,/0x'* is of rank k, the system of differential equations
for the extremal is
(57.23)
where
Gua Fe AA tO (eel female i).
We note in conclusion that the constraining relations 57.22 do not
involve the derivatives @*. Such constraints are called holonomic to
distinguish them from constraints of the form
(57.24) $,(t, x, @) =0
which are nonholonomic. Nonholonomic constraints arise in the study
of dissipative dynamical systems, and we shall encounter them in Chapter
4. It is clear from the foregoing discussion that equations corresponding
to (57.23), when nonholonomic constraints are present, must involve
not only the multipliers A,(t) but also the derivatives /,(1).
Problems

1. Consider the variational problems in (56.1) with Euler’s equation 56.6.

Note that if F(t, x, «) does not contain x explicitly then (56.5) yields at once the
first integral in the form F; = constant. Show that when F(t, x, #) does not
contain ¢ explicitly then the first integral of (56.6) is F — F; = constant.
d
Jeli, WES) IP Sasa Ga, 20)). compute =7 7 F — “F;) and make use of (56.6).

2. Note the hint in Problem 1 ae show that the Euler equation 56.5 can be
written in the form

3. The Euler equation 57.16 is an identity whenever F (2, y, y’) = M(x, y) +

N(a, yy’, with M, = N,. In this case (57.13) becomes
x
J= M dx + N dy
wy
and this integral is independent of the path. Thus, every curve joining the given
endpoints is an extremal for (57.13).
SEC. 58] GEODESICS IN R,, 157

58. Geodesics in R,,

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. We will
carry out our calculation for the case of the n-dimensional Riemannian
manifolds, since our results will be of interest not only in connection
with the geometry of surfaces but also in the study of dynamical trajec-
tories in Chapter 4.
Let metric properties of the n-dimensional manifold R,, be determined
by
(58.1) ds* = p,, dx‘ du}, (Po yrms 1) oe),

where g,; = 2); are specified functions of the variables x’. We suppose
that the form 58.1 is positive definite and the functions g;; are of class
C2. The length of a curve C, represented in R,, by equations
C: act a a(t), ty < t & tp,

is given by
fess SS
(58.2) S “AyJ gapt?a? dt, (x; P= .1, 225,57)

cs in R,,.
The extremals of the functional 58.2 will be termed geodesi
The function F of Sec. 57 in this case is

(58.3) F=V2,,0#,
and Fy. This
and, to form Euler’s equations 57.7, we need to compute F,,
computation is straightforward. We deduce from (58.3) that

anil a+ _ OLa8 a +B
= 5 (8ap@'®') auf toe
Fai

and
Fut = (gop)
See?
BastSof
Substituting these expressions in Euler’s equations yields
os gtak
d g ge x

dtl gil W Bagi"

written in the form
Since ds/dt = yifap 2 equation 58.4 can be

28a8 ea?
4 (se) BCs SIN
dt\ds/dt 2ds/dt ;
158 GEOMETRY [CHAP. 3
and, carrying out the indicated differentiation, we obtain
a 2 2
O85 a 1 28up eegh — Sajt" s/dt .
ax? 20%) ds/dt
Since the second term in this equation can be written as a sum of two
terms, we have
; a 72 2
Ce ae (3 O86; Seng ciate? = Saji d s/dt z

: 2\ox® dx* dx’ ds/dt

But [«f,j] = ae + Hy — $2), so that the foregoing equation

assumes the form
., a's/dt*
(58.5) £25e" + [af, j] ay ds|dt
These are the desired equations of geodesics. If we choose the parameter
t to be the arc length s of the curve, that is, if we set

ab= V gapiti? = 1,
dt
the system 58.5 simplifies to read
(58.6) Gast" + [oP jan? =:
In equation 58.6, dots denote the differentiation with respect to the arc
parameter s. ;
If we multiply equation 58.6 by the tensor g*’ and sum, we obtain a
simple form of the equations of geodesics in R,,.

(58.7) ae |jee =o, (Sd. Den Sin):

aB (es ee TES ean).
We observe that the form of these equations is identical with equations
51.2 defining the straight line in £3. Since (58.7) is an ordinary second-
order differential equation it possesses a unique solution when the values
a‘(s) and the first derivatives dx‘/ds are prescribed arbitrarily at a given
point «*(s9).
If we regard a given surface S as a Riemannian two-dimensional
manifold R,, covered by Gaussian coordinates u*, then (58.7) assumes
the form
d°u? AL du?
58.8 ences, Hf —=(0, , By y = I,2).
2B) ds* «“B) ds ds ) mB
Hence at each point of S there exists a unique geodesic with an arbitrarily
prescribed direction A* = du*/ds. It is not difficult to prove that, if there
SEc. 58] GEODESICS IN R, isd

Fig. 25

exists a unique solution u%(s), passing through two given points on S,

then the curve u%(s) is the curve of shortest length joining these points."
If the manifold R,, is Euclidean, a coordinate system exists in which
the Christoffel symbols vanish. In this case equations 58.7 become
dx‘/ds? = 0. The general solution of this equation is «* = A's + Br’.
Thus the geodesics in E,, are straight lines.
As another illustration, consider the problem of determining geodesics
on an arbitrary cylinder immersed in E;. We choose the Y*-axis parallel
to the generators of the cylinder and let the trace of the cylinder on the
Y'Y?-plane be given by equations
y' = $(0),
y? = yo),
where a is the arc length of C. (See Fig. 25.) Since
(do) = (dy’)? + (dy*),
an element of arc ds of the geodesic is given by
(ds) = (do)’ + (dy*)?,
so that ay; = do. = 1, ay, = 4p, = 0. Hence equations 58.8 reduce to
do
et 0, fory = 1,

and
d’y®
= 0, for y = 2.
ds?
p. 175.
14 See, for example, L. P. Eisenhart, Differential Geometry (1940),
160 GEOMETRY [CHAP. 3

We thus obtain
o6 = As + B,
Y= As

If A ~ 0, we can write these equations in the form

sp an Cyo a C2,

where C, and C, are arbitrary constants.

The equations of the geodesics are therefore
y' = ¢(9),
y? = (0),
y* = Cyo ar Ce

and hence the curve is a helix, whose pitch is determined by C,. The
constant C, determines the origin for the arc parameter o.
We can show in a similar way that the geodesics on the surface of a
sphere are arcs of great circles. (See Problem 1.) However, as an illustra-
tion of the use of equations 57.23 we consider the functional
to —

(58.9) p=) Jee dt, (i= 1,2,3),

ty

representing the length of arc of a curve C in cartesian coordinates. If

C is to lie on a sphere of radius a, the constraining relation 57.10 has the
form
(58.10) dg=2'x'-AW=0,
when the center of the sphere is at the origin. The function G = F + Ad,
introduced in Sec. 57, is

(58.11) Ge Vee A(t)(xixt — a?)

and equations 57.23 take the form

d dx*
58.12
(58.12) ——— tO
nade +4 24()z'=0, (i 2)
(Le = 1,2,3),
where ds = V/#'## dt.
On eliminating A(t) from the first two equations in the set 58.12, we get

or
SEC. 58] GEODESICS IN &,, 161
Therefore

(58.13) a Peg he
ay 2

where C, is a constant of integration.

Similarly, the elimination of A(t) from the last two equations in (58.12)
yields on integration

(58.14) CLI NE Ao
1 3

where C, is a constant.
From (58.13) and (58.14) we find

robes3 dx 1 i
Bias 3
ee 2
dx iE Teat akee2
(xt (x)?
Or

3)\ 2

(58.15) c.a(5) = c.a(5)

0 v

and the integration of equation 58.15 gives

(58.16) Cua — OR — Cite = >

which represents a plane through the origin. Equation 58.16 together

with the constraining relation 58.10 shows that the solution of the system
58.12 is an arc of a great circle.
Problems
1. In an orthogonal cartesian frame Y, the sphere of radius a is determined
by equations
y! =acos u'cos u’,
y2 =acos u' sin u?,
y® = asin ul.

In this case ds? = a? (du)? + a? (cos u!)* (du*)? and

Ein Sel Si Realae

S= a| V1 + cos? ul(u?)? dul,
Ug?

where w2 = du?/du'. Show that the geodesics are great circles.

2. Find the geodesics on the surface

y! = uv cos u?, y2 = u' sin uv, 0.

imbedded in E3. The coordinates y’ are orthogonal cartesian.

162 GEOMETRY [CHAP. 3

3. Show that, if we set Q = a, i? where u* = du%/ds, the equations of the

geodesics 58.8 in Ry can be written
d/20\ ag
ds aa
Hence the solutions of these equations for i” should yield — 7, u“u, as can be
seen from (58.8). This suggests a different means for computing the symbols
leetin any particular coordinate system. Use this method to calculate the
a

Christoffel symbols for the coordinate system in Problem | by determining the

coefficients of 2%? in the solutions for the second derivatives of u” with respect
to s.

59. Geodesic Coordinates

We have seen (Sec. 39) that, if a Riemannian space R,, is Euclidean,

a coordinate system exists in which the components g,; of the metric
tensor are constants throughout the space. This implies that in such a
coordinate system 0g;,,/0x* = 0. The vanishing of these partial derivatives
is equivalent to the vanishing of all Christoffel symbols, since!® [ij, k] =
ae ae Bn _ af If R, is not Euclidean, then the Christoffel
DN Oc uC TnL
symbols do not vanish at all points of R,, but it is possible to find a
coordinate system, in fact infinitely many, in which they vanish at any
given point P of R,. Such coordinates are called geodesic for that par-
ticular point, or locally cartesian at P.
Thus consider some surface net with coordinates u* and consider the
point P(u!, uo”) on S. If v* are the coordinates of some other net on S,
then
(59.1) UV = uo. Be Le al).
The second derivative formula 32.5 yields the relation
Out {o\ ee [7\s
(59.2)
dv* dv" = ulBy) Av? Av" =» lAw) av”
However, if there exists a transformation of coordinates 59.1 such that the

Christoffel symbols |
ifvanish at P, then for that particular point

: ‘D2, B
(59.3) Lg poo = 0
dv* dv" — ul By) Av* dv"
15 See also Theorem I, Sec. 39.
Sec. 60] PARALLEL VECTOR FIELDS IN A SURFACE 163
We exhibit next a solution of this equation yielding a particular trans-
formation 59.1 to a coordinate system v% in which the Christoffel symbols
vanish at P. It is the second-degree polynomial

(59.4) WP YE pt ae |poh,
2 law P
where the wp* is the value of u* at P and the ff are the values of the
LM) Pp
Christoffel symbols at P. To verify that (59.4) satisfies (59.3), we compute
Ou" {x
f29).)) — = 0, — ‘
Ov" i Aw -
and
(59.6) Cie
2,,0
ee ye)
v* Ov" lau) Pe
From (59.4) we see that the point P, in new coordinates, is given by
ed

v* = 0, and hence at the point P, equation 59.5 yields ua = Of.

P
Inserting values from this equation and (59.6) in (59.3), we see that it is
satisfied at P. Hence the new variables indeed are geodesic coordinates
au P. .
We conclude this section by a remark that there is an extension of this
result by Fermi, who proved that in every Riemannian manifold R,, there
exists a coordinate system such that the coordinates are geodesic at all
points of an arbitrarily prescribed analytic curve.’®

60. Parallel Vector Fields in a Surface

The concept of parallel vector fields along a curve imbedded in E,

(Sec. 48) was generalized by Levi-Civita to curves imbedded in n-dimen-
sional Riemannian manifolds. As an illustration of the usefulness of the
concept, consider a surface S immersed in £, and a curve C on S. We
take equations of C in the form
C: u®=u%(t), totsch,
and suppose that the metric properties of S are governed by the tensor
A,3. If A* is a surface vector field defined along C, we can calculate the
surface intrinsic derivative

aOt = ae
a a y

(60.1) dt py dt
16 A derivation of explicit equations of transformation for this case, which include
(59.4) as a special case, was given by Levi-Civita in a paper entitled “Sur l’écart
géodésique,” Mathematische Annalen, 97 (1926-27), pp. 291-320.
164 GEOMETRY [CHapP. 3
This is identical in form with the left-hand member of equation 48.1
defining the parallel vector field along a space curve. Accordingly, we
take the differential equation

(60.2) dat, fae) gpdul _ 0,

di AB)” wade
which determines a unique vector field when the components of the vector
are specified at an arbitrary point of C as the definition of the parallel
vector field along a curve C on the surface S. If the parameter ¢ is chosen
as the arc length s, equation 60.2 reads
a y
(60.3) dA / & \ 4 du’ = 2

ds \ByJ ds
and if A* is taken to be the unit tangent vector to C so that

with a,,4*A° = 1, then (60.3) yields

(60.4) ci) SACO

ds? \By) ds ds
This equation is recognized as the equation of.a geodesic on S, and hence
one can enunciate a
THEOREM. The vector obtained by the parallel propagation of the tangent
vector to a geodesic always remains tangent to the geodesic.
From uniqueness of solution of (60.4) it follows that the property of
tangency of a parallel vector field to a surface curve is both a necessary
and sufficient condition for a geodesic.
In the Euclidean plane geodesics are straight lines, and the parallel
vector field formed by the tangents to a straight line traces out the same
straight line. On the surface of the sphere the geodesic is an arc of a
great circle joining two given points on the sphere, and the corresponding
vector field is the field of tangents to the geodesic. From the last example
it is clear that parallelism with respect to a surface curve differs from the
parallelism with respect to a space curve imbedded in Es, since vectors
obtained by a parallel propagation, along the surface curve C, need not
be parallel in the Euclidean sense. However, it is easy to prove that the
lengths of vectors forming a parallel field with respect to C remain
constant. Indeed, word-for-word repetition of the proof given in Sec. 48
leads to the conclusion that the angle between two vectors propagated
in parallel fashion remains unchanged, and it follows, as it did in Sec. 48,
that the vectors forming a parallel field are constant in magnitude. A
SEC. 61] ISOMETRIC SURFACES 165
corollary of this result is that the vector field obtained by a parallel
propagation of a surface vector along a geodesic makes equal angles
with the geodesic.
It should be noted that the concept of parallelism in Riemannian
manifolds is defined relative to a given curve. A surface vector A%,
specified at a point P of S, when propagated in parallel manner along a
given curve C to a point Q, need not coincide with the vector obtained
by the parallel propagation along a different path joining P and Q. More-
over, if a closed curve C, enclosing a simply connected region of S, is
drawn, and a parallel vector field is constructed starting with some point P
on C, then the vector obtained by traversing the closed path need not
coincide with the initial vector. The angle between the initial and final
vector measures another intrinsic property of S, known as the Gaussian
curvature of S. This property is introduced in a somewhat different way”
in Sec. 62.

61. Isometric Surfaces

The properties of surfaces with which we have been concerned so far

hinged entirely on the study of the first fundamental quadratic form
(61.1) ds* = a,, du* du’.
These properties constitute a body of what is known as the intrinsic
geometry of surfaces. They take no account of the distinguishing charac-
teristics of surfaces as they might appear to an observer located in the
surrounding space. Two surfaces, a cylinder and a cone, for example,
appear to be entirely different when viewed from the enveloping space,
and yet their intrinsic geometries are completely indistinguishable since
metric properties of cylinders and cones can be described by the identical
expressions for the square of the element of arc. If a coordinate system
exists on each of the two surfaces such that the linear elements on them
are characterized by the same metric coefficients a,,, the surfaces are
called isometric. Obviously 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 without changing the lengths of arc elements,
and hence without altering the measurements of angles and areas.
In the following section we introduce an important scalar invariant,
the
known as the Gaussian curvature, which will enable us to determine
circumstances under which a given surface is developable, that is, isometric
with the Euclidean plane.
L. P. Eisenhart, Introduction to Differential Geometry, Princeton
17 Compare
University Press, p. 200.
166 GEOMETRY [CuaP. 3

As an illustration of an isometric nondevelopable surface consider the

catenoid
Si: y' =v'cosv’,
y” = v'sin v’,
vi
y® = acosh *—,
a
obtained by revolving the catenary y? = cosh (y?/a) about the Y*-axis.
We will show that the surface S, is isometric with the surface of the
helicoid defined by
Su = 2 cos,
a” = Sit 2,
ie
The first fundamental form ds? = dy‘ dy’ for S, is easily found to be

(Ca 12
(61.2) ds* = a,, dv" dv® = aa” y+ (vy1y2 (doy,
77,.2)2
v)*—a
so that
(v')?

an = (v')? ene y dy. = 0, as. = (v')? =a>+ [(v)? a a’).

For the surface S,, we find

(61.3) ds” = a,g du* du® = (du')’ + [a? + (u’)*] (du°)’,

so that
a,, = 1, ai, = 0} Ao = a® + (u')?.
Now, if we set in (61.2)
(v})? — @ = (u)), v= wv,
we obtain
ds? = (du')* + [(u))? + a®](du?)*.

Since this is identical with (61.3), the surfaces S, and S, are isometric.
It follows from discussion-in the next section that these surfaces are not
developable.

62. The Riemann-Christoffel Tensor and the Gaussian Curvature

The formulas of Sec. 37 describing the properties of Riemann-Christoffel

tensors in n-dimensional manifolds simplify considerably when n is set
equal to 2. Thus, if we are given the first fundamental form,
(62.1) ds* = a,, du* du®,
SEC. 62] THE GAUSSIAN CURVATURE 167

of the surface S, we can form the Christoffel symbols with respect to this
surface, and the corresponding Riemann tensor

(62.2) Rapys =
= a st a Pare ‘
[By,«] [Bd, «] [xy, A] [«d, A]
We recall that this tensor is skew-symmetric in the first two and last
two indices, so that, for the surface S,

(62.3) Raahy — Rapyy = 0, Ryo = Royo = —Roiie = — Rj291-

Hence every nonvanishing component of the Riemann tensor is equal to

Rjo12 OF to its negative.
We define the quantity K by the formula

(62.4) K= Ria :
a
curvature
where a = |a,,|, and call it the Gaussian curvature or the total
s a,, are involved in this
of the surface S. Since only metric coefficient
of the
definition, the properties described by K are intrinsic properties
- ryt
surface S.
If we introduce the two-dimensional e-tensors,
aa ap

yerg = x/0 eng Band” “¢ap __ ©

oi)
f
Ja
rt

40), and note

7

where the e,,’s are the alternating e-systems (see Sec.

= Loft VV be’

relations 62.3, we can write equation 62.4 as

Rapys => Kegg€ys-
(62.5)

Since <*’<,,; = 2, we can solve (62.5) for K and obtain

K = LR asaee €
(62.6)

is an invariant.
These equations show that the Gaussian curvature
the Euclidean plane, there
Now, when a surface S is isometric with
respec to which ay; = 42 = its
t
exists on S a coordinate system with
R,»,5 = 0 in this particular coor-
ay. = 0. It is obvious that in this case
vanish in every coordinate
dinate system, and since R,,,9 18 a tensor, it must
system.
at all points of the surface,
Conversely, if the Riemann tensor vanishes
there exist coordinate systems
Theorem II of Sec. 39 guarantees that
1, 412 = 0.
on the surface such that ay, = 422 =
168 GEOMETRY [CHAP. 3

Thus we have a
THEOREM. A necessary and sufficient condition that a surface S be
isometric with the Euclidean plane is that the Riemann tensor (or the
Gaussian curvature of S) be identically zero.
Consider next an invariant
(62.7) R=a"R,,,
where

(62.8) Ryy = Riva = Books

is the Ricci tensor introduced in Sec. 38.18

If we multiply (62.8) by a” and sum, we get

(62.9) R= a@Rieaeo Th
and recalling (62.3) we see that (62.9) is equivalent to

(62.10) R=—2Rigd@-a" —ia an):

Since
Aub es Ag2 qe = ay ae! ae
ie a a
we have

(62.11) R= —2 Rime,
a
Comparing (62.11) with (62.4) we see that

R= —2K.

The invariant R is sometimes called the Einstein curvature of S.

We shall give a more revealing geometrical interpretation of the Gaussian
curvature in Sec. 72, where the surface S is viewed from the enveloping
space.
Problems

1. Use formulas 62.2 and 62.4 to show that, if the system of coordinates is

- sates) +2
orthogonal, then

2Va dul Va eu eu Va ou2 ;

2. Calculate the total curvature of the manifold whose quadratic form is

ds* = a" sin? uw (du®)® + a®(du)?.

** We recall that R%,, = a’*Rj_3.

SEc. 63] GEODESIC CURVATURE OF SURFACE CURVES 169
3. Determine whether the surface of a helicoid given by
y = u' cos uv’,
y2 = w sin uv,
y® = au’,
is developable.
4. Show that, for a surface of revolution defined by
y! = u cos uv’,
y? = w sin vu’,
y> = fu’),
Ps : :
en w+ (f f is of class C*.
P ’ when

5. Show that the surface defined by

y" as fx),

y¥? = flv),

y> =u,
where f; and fy are differentiable functions, is developable.
in the
6. Show that the formula for the Gaussian curvature K can be written
form
re 1 d |Ayo 9A, 1 “|
2Va\lawla,Va aw Va au
d |2 ay» 1 Gay Me au |
Se ies .
+ _——w ee

au2LVa eu Vad ayVa dui

63. The Geodesic Curvature of Surface Curves

with a
We shall conclude our study of intrinsic geometry of surfaces
ing the behavio r of the tangent vector
derivation of a formula describ
is analogo us to Frenet’ s formula 50.1.
to a surface curve. This formula
Let C be a surface curve defined parame trical ly by

(63.1) u" = us),

where s is the arc parameter. Accordingly, at every point of the curve we
have the condition
du* du? rhe
(63.2) Bigars a
ds ds
a tangent vector ie
The quantities du'/ds, du*/ds obviously determine
to C, and it is clear from (63.2) that

(63.3) I =
170 GEOMETRY [CHaP. 3

is a unit vector. If we differentiate the quadratic relation a,,/*/’ = 1

intrinsically with respect to s, we obtain a,,4%(62’/ds) = 0, from which it
follows that the surface vector 6A%/ds is orthogonal to /*. Following the
line of thought of Sec. 49, we introduce a unit surface vector 7* normal
to A*, so that

(63.4) — = x17,
where x, is a suitable scalar. In order to determine the direction of 7*
uniquely we choose 7% in the way analogous to the choice of the triad
of vectors in Sec. 49 (equation 49.11), namely, €,,A%7° = 1. This choice
of the orientation of A and y uniquely determines the sign of ,, and it
amounts to saying that the sine of the angle between A and n is +1. The
vector 7* is the unit surface vector orthogonal to the curve C, and the
scalar x, is called the geodesic curvature of C.
We recall that the equation of the geodesic on S (see equation 60.4)
can be written as 6A*/d6s = 0. Comparing this with (63.4) leads to the
conclusion that, if the geodesic curvature x, = 0, then the curve C is a
geodesic, and conversely. Hence the
THEOREM. A necessary and sufficient condition that a curve on a surface
S be a geodesic is that its geodesic curvature be zero.
As an illustration, we compute the geodesic curvature of the small
circle
Ge u*= constant =1,' > 0, u* = 7,
on the surface of the sphere (Fig. 26)
S: y' = acosu'cos v2
y? = acos ui sin uv?
y® = a sin ul.
If the arc-length s of C is measured from the plane u, = 0, we have
u® = s/(a cos up'), and the equations of C can be written in the form

(63.5) th ce Wes oh oe
a COS Uy’
From (63.5) we find that the components of the unit tangent vector
A* = du*/ds along C are
(63.6) oe ee
so that
bas dat |1 |
ga dul
ds ds ap
SEC. 64] SURFACES IN SPACE 171
y3

ey
Shona

Y 1

Fig. 26

and
6?
— =
di”
— +
Oa au
Qo) 25 —i() 2 \aaat

Os ds a ds 4 22

Since metric coefficients of S are a, = 4", dy, = 0, do, = a® cos* u', we

find, on referring to Problem 3 of Sec. 31, that ke = sin u' cos ul,
2
|= 0. Accordingly, formulas 63.4 yield
Ze
6, 1 62°
(63.7)
63.7 ee =x
On amg ee a
gl =—tanu,, —=x,7 me = 0.

7°
Since C is not a geodesic, x, ¥ 0, and we conclude that 7? = 0. But
1/a. Hence
is a unit vector so that a,,7%n° = 1, and we find that 7! =
the relation
equations 63.7 yield x, = (tan uy!)/a. In Sec. 71 we establish
of C.
between the geodesic curvature x, and the ordinary curvature x

64. Surfaces in Space

nding space,
With the exception of occasional references to the surrou
out from the point of
our study of geometry of surfaces was carried
e is determ ined by the
view of a two-dimensional being whose univers
of surface s presen ted in
surface parameters ui and u?. The treatment tic
of the first quadra
the foregoing was based entirely on the study
61, we
differential form. In the discussion of isometric surfaces in Sec.
172 GEOMETRY [CHAP. 3

remarked that a pair of isometric surfaces, a cone and a cylinder for

example, 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 the enveloping space is the normal line to the surface. The behavior
of the normal line as its foot is displaced along the surface depends on
the shape of the surface, and it occurred to Gauss to describe certain
properties of surfaces with the aid of a quadratic form that depends in
a fundamental way on the behavior of the normal line. Before we
introduce this new quadratic form, let us recall our point of departure
in the study of surfaces in Secs. 52 and 53.
A surface S imbedded in £; was defined by three parametric equations

(64.1) y=y(u,v), @=1,2,3),

where the y’* are orthogonal cartesian coordinates of the reference frame
located in the space surrounding S. An element of arc ds of a curve
lying on S is determined by the formula

(64.2) ds* = a5 Gus un",

where
2 oy' oy’
7F Out dub

The choice of cartesian variables y‘ in the space enveloping the surface

is clearly not essential, and we could have equally well referred the points
of E; to a curvilinear coordinate system X related to Y by. the trans-
formation a‘ = x'(y, y®, y’). Now, relative to the frame YX, the line
element in E; is given by
(64.3) ds* = 9, du* dei,

dy* dy*
where g;; = ant Dal” and the set of equations 64.1 for the surface S
can be written as

(64.4) Spike da 2g (eel

It follows from this representation of S that

(64.5) dx' = gS du’,

SEC. 64] SURFACES IN SPACE 173
and hence the expression for the surface element of arc (64.3) assumes
the form
ds? g, hdat dat
Ox Ox" 4
5 63 Bu—— —ul du* du?

A comparison of this with equation 64.2 leads to the conclusion that

Ou’ Ox?
(64.6) ag= 8-55,
du* du?
(ii, = 1,23), 8 = 1,2).
We note that the foregoing formulas depend on both the Latin and
Greek indices, and we recall that the Latin indices run from 1 to 3 and
refer to the surrounding space, whereas the Greek indices assume values
1 and 2 and are associated with the surface S imbedded in F;. Further-
more, the dx‘ and g;,,’s are tensors with respect to the transformations
induced on the space variables x‘, whereas such quantities as du* and a,, are
tensors with respect to the transformation of Gaussian surface coordinates
u*. Equation 64.6 is a curious one since it contains partial derivatives,
dx'/du*, depending on both Latin and Greek indices. Since both a,,
and g,; in (64.6) are tensors, this formula suggests that dx*/du* can be
regarded either as a contravariant space vector or as a covariant surface
vector. Let us investigate this set of quantities more closely.
Let us take a small displacement on the surface S, specified by the
surface vector du*. The same displacement, as is clear from (64.5), is
described by the space vector with components

[64.5] ee dae.
Ou”
The left-hand member of this expression is independent of the Greek
indices, and hence it is invariant relative to a change of the surface
coordinates u*. Since du* is an arbitrary surface vector, we conclude that
Ox'
(64.7) ou
is a covariant surface vector. On the other hand, if we change the space
coordinates, the du*, being a surface vector, is invariant relative to this
change, so that (64.7) must be a contravariant space vector. Hence we
can write (64.7) as

(64.8) “=
ou"
of
where the indices properly describe the tensor character of this set
quantities.
174 GEOMETRY [CHAP. 3
A simple geometrical significance of the set of quantities 64.8 can be
deduced from Fig. 27. Let r be the position vector of an arbitrary point
P on S. The point P is determined by a pair of Gaussian coordinates
(u!, u), or by a triplet of space coordinates (a1, x*, x*). Accordingly,
the vector r can be viewed as a function of the space variables x* satisfying
equations 64.4. Thus
(64,9) Or, aOreos
Ou* E Ox! Ou™

But dr/dx* are the base vectors b; at P, associated with the curvilinear
system X, whereas Or/du* are the base vectors a, atP relative to the
Gaussian system U.
Hence equations 64.9 yield

(64.10) a, = b; —
3 3 : Ox? ae a
It is clear from this representation that a, = — b, and a, = aA b,, so
dul u
that dx‘/du* = x,', (« = 1,2), are the contravariant components of
the surface base vectors a, referred to the base systems b;. Thus the
sets of quantities

4p ( Ox” =) ; Ox* =)
Uy: oe ae and x3: |—,-=-,->
du’ du: du du? du?’ du?

Fig. 27
SEC. 65] THE NORMAL LINE TO THE SURFACE 175
transform in a contravariant manner relative to the transformation of
space coordinates 2°.
We can also show that the three surface vectors

Ox ae pet chee teges

ioe1) Sd eam Ox"
(Oars ened 2:Em (—,—}], a: (ieee
Ou’ du dul du? dul du
transform according to the covariant law with respect to the transforma-
tion of Gaussian surface coordinates u*. Indeed, consider a transformation
u* = u*(i', a); then the equations 64.4 of S go over into peter (iu),
and

(64.11) aa!
aut
_ Ox! aut
du’ du*
But dx‘/du’ =z, and (64.11) yields, for i = 1, 2, 3, a! = (00° /du")z5.
This is the covariant law.
Let ds be an element of arc joining a pair of points P(u’, u*) and
P(u* + du}, v2 + du?) on S. The direction of the line element ds is given
by the direction parameters du*/ds = 4*. The same direction can be
specified by an observer in the enveloping space by means of three param-
eters dx'/ds = 2‘, and it follows from (64.5) that 2’ = x A*, This formula
tells us that any surface vector A* (that is, a vector lying in the tangent
plane to S) can be viewed as a space vector with components A‘ determined
by
(64.12) Ate AC
We shall refer to a vector A‘ determined by this formula as a tangent
vector to the surface S.

65. The Normal Line to the Surface

P of S
Let A and B be a pair of surface vectors drawn at some point
64.12, they can be represe nted in the form
(Fig. 28). According to formula
(65.1) Ai=afA% B= 2,8.
tangent plane
The vector product A x B is the vector normal to the
perpendicular
determined by the vectors A and B, and the unit vector n
a right-handed
to the tangent plane, so oriented that A, B, and n form
system, is
AxB > 2? AxB
n= ————__
65.2
83:2) |A x B| AB |sin 6]
where 0 is the angle between A and B.
surface S at P. Clearly,
We call the vector n the unit normal vector to the
n is a function of coordinates (u’, u2), and, as the point P(ul, u?) is
176 GEOMETRY [CHapP. 3

Fig. 28

displaced to a new position P(u’ + du’, u? + du*), the vector n undergoes

a change
(65.3) dn = — du’,
or
whereas the position vector r is changed by the amount dr = ae du*.
Let us form the scalar product
ON SeOre tees
(65.4) dn-dr = Soares du’.

If we define
ao ee ee
= 2\duz du® du® aut)’
so that (65.4) reads
(65.5) dn- dr = —b,, du* du’,
the left-hand member of (65.5), being the scalar product of two vectors,
is obviously an invariant; moreover, from symmetry with respect to «
and 8, it is clear that the coefficients of du* du’ in the right-hand member
of (65.5) define a covariant tensor of rank two. The quadratic form

(65.6) B = by, du® dub

will be shown to play an essential part in the study of surfaces when
they are viewed from the surrounding space, just as the first fundamental
quadratic form . = dr- dr, or
A = ayy du* duP
SEC. 66] TENSOR DERIVATIVES 177
did in the study of intrinsic properties of a surface. The differential
form 65.6 was introduced by Gauss, and it is called the second fundamental
quadratic form of the surface.
Since the notation for the unit normal used previously, “despite its
pictorial suggestiveness, is more cumbersome than the tensor notation,
we shall rewrite the defining formula 65.2 in terms of the components
x’ of the base vectors a,. We denote the contravariant components of n
by n' and observe that its covariant components n, are given by’®

(65.7) pees atl

AB sin 6
and (see Sec. 54)
(65.8) AB sin 0 = €,,A°B?.
Substituting in (65.7) from (65.1) and (65.8), we get

(n€28 — €:j.%225) AB’ = 0,

and, since this relation is valid for all surface vectors, we conclude that

(65.9) NE ap = €iintaXp-
Multiplying (65.9) through by «*’, and noting that e~Pe,, = 2, we get the
desired result
(65.10) iq t = ge Me 5 XaXp-
It is clear from the structure of this formula that n; is a space vector
which does not depend on the choice of surface coordinates. This fact
is also obvious from purely geometric considerations.

66. Tensor Derivatives

In Sec. 67 we shall deduce the second fundamental quadratic form

65.6 analytically by the operation of tensor differentiation of tensor fields
which are functions of both surface and space coordinates. The fruitful
19 Note that the vector product A x B depends on the lengths of the vectors A and B
and on the angle between them. If we choose an orthogonal cartesian system of axes
the Y?-axis,
Y, so that the vectors A and B lie in the Y!Y®-plane with A directed along
the cartesian component s 4‘ of A are A! = A, A? = 0, A? = 0, and the components
then
= iixs
of B are B! = Bcos 0, B? = Bsin 0, B® = 0. Since in the Y-system €;j,
C; = €:;x A’ B* = €12AB sin 0.

C=A xX B
Hence C, = 0, C, = 0, C; = ABsin §. Thus the C; define the vector
A* and
normal to the plane determined by A and B whose magnitude is AB |sin 6|. If
This result follows
BP are the surface components of A and B, then ABsin 6= €qpA=BP.
ely from the formula for the sine of the angle between two vectors given in
immediat
Sec. 54.
178 GEOMETRY [Cuap. 3
concept of tensor differentiation was introduced by A. J. McConnell,
whose elegant treatment of surfaces is followed closely in this and several
other sections of this chapter.”°
Let us consider a curve C lying on a given surface S and a vector “a
defined along C. If tis a parameter along C, we can compute the intrinsic
derivative 6A’/6t of A*, namely,

66.1
bai _ Ai [i\sda
Cae Ot dt alikI dt

In formula 66.1 the Christoffel symbols |4 refer to the space coordinates

g\J
x* and are formed from the metric coefficients g;,. This is indicated by
the prefix g on the symbol. On the other hand, if we consider a surface
vector A* defined along the same curve C, we can form the intrinsic
derivative with respect to the surface variables, namely,

0A* dA* |a AP du’

(66.2) — = ;
Ot dt a\lfy dt

In this expression the Christoffel symbols Pa are formed from the

a By
metric coefficients a,, associated with the Gaussian surface coordinates u*.
A geometric interpretation of these formulas is at hand when the fields
A‘ and A* are such that 6A*/dt = 0 and 6A*/dt = 0. In the first’ equation
the vectors A* form a parallel field with respect to C, considered as a
space curve, whereas the equation 6A*/dt = 0 defines a parallel field with
respect to C regarded as a surface curve. The corresponding formulas
for the intrinsic derivatives of the covariant vectors A; and A, are
: ; j
(66.3) 0A; _ dA; {lay
ot dt o\ij dt
and
B
(66.4) Ag dAg _ Aa
ot dt aap dt
Consider next a tensor field T,', which is a contravariant vector with
respect to a transformation of space coordinates x‘ and a covariant vector
relative to a transformation of surface coordinates u*. An example of a
field of this type is a tensor x,* = dx*/du* introduced in Sec. 64. If T.‘is
defined over a surface curve C, and the parameter along C is t, then T. ;
is a function of t. We introduce a parallel vector field A, along C, regarded
20 A. J. McConnell, Absolute Differential Calculus, Chapters XIV-XVI.
SEC. 66] TENSOR DERIVATIVES 179
as a space curve, and a parallel vector field B* along C, viewed as a surface
curve, and form an invariant

Dt) =) ALB “

The derivative of ®(t) with respect to the parameter ¢ is given by the

expression
(66.5) eee eae BTA
dt dt dt dt

which is obviously an invariant relative to both the space and surface

coordinates. But, since the fields 4,(t) and B%(t) are parallel,

ee
; j a y

g\tJ dt dt a By dt

and (66.5) becomes

i ee he ey
‘aay ae EE + Fipe 3 |r;ae
dt Fin slid ot” alay) ° dt
Since this is invariant for an arbitrary choice of parallel fields A; and B*,
the quotient law guarantees that the expression in the brackets of (66.6)
is a tensor of the same character as 7,’. We call this tensor, after
McConnell, the intrinsic tensor derivative of T,* with respect to the param-
eter ¢, and write

es
ee
ST eed Te
$f a di
i),dz® ft
(ree
@aslikl “dt calay)
ee
(6)_,dw’
* dt
If the field 7,’ is defined over the entire surface S, we can argue that,

os (tr (2)
since

dbpwmiiaul, Colikls os “aly dt

is a tensor field and du’/dt is an arbitrary surface vector (for C is arbitrary),

the expression in the bracket is a tensor of the type T,,,°.. We write
ed Ca ae
i
eeT Py (2Tee :
;
(66.8)
Te =
Cn) Ae ae ry 5

and call Te the tensor derivative of T: with respect to u’.

The extension of this definition to more complicated tensors is obvious
Obl a
from the structure of the formula 66.8. Thus the tensor derivative
with respect to u” is given by
i =a ips ve — Ti. nae ae

(eo?) ;
nay :Ou’
: ;
9g jk om
a alny
| 26 By | ;:
a|
180 GEOMETRY [CHAP. 3

If the surface coordinates at any 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. This leads us to
conclude that the operations of tensor differentiation of products and
sums follow the usual rules and that the tensor derivatives of g;;, d,,,
€ijn) €4g and their associated tensors vanish. Accordingly, they behave
as constants in the tensor differentiation.

67. The Second Fundamental Form of a Surface

The apparatus developed in the preceding section permits us to obtain

easily and in the most general form an important set of formulas due
to Gauss. We will also deduce with its aid the second fundamental
quadratic form of a surface already encountered in Sec. 65.”
We begin by calculating the tensor derivative of the tensor x,’, repre-
senting the components of the surface base vectors a,. We have

07x" a e {6
67.1 at = eat |)let \x5,
re) Fou" due 4 jk ne ala €
from which we deduce that

(67.2) oe = ee

Since the tensor derivative of a,; vanishes, we obtain, upon differentiating

the relation :

[64.6] Dag = 8 45%2%5,

(67.3) Bis%ayte + BijtoXp,y = 0.

Interchanging «, B, y cyclically leads to two formulas:

(67.4) GisX po) 2 i;%p%s,0 = 0,

(67.5) BisXy,p%a + Bij%}X3 5= 0.
If we add (67.4) and (67.5), subtract (67.3), and take into account the
symmetry relation (67.2), we obtain
(67.6) £isky pty =O
This is the orthogonality relation which states that 2a p 1S a Space vector
normal to the surface, and hence it is directed along the unit normal n‘.
Consequently, there exists a set of functions b,, such that
(67.7) hg = bag’.

*t Compare A. J. McConnell, Absolute Differential Calculus (1931), p. 200.

SEC. 67] THE SECOND FUNDAMENTAL FORM 181
The quantities b,, are the components of a symmetric surface tensor,
and the differential quadratic form

(67.8) B = by, du* du’ .

is the desired second fundamental form.

To demonstrate the equivalence of this definition of the tensor b
with that given in Sec. 65, namely,

-i(2. Or , On 2),
CPi
Ou" due ou? Ou"

note that the vectors n and a, = Or/du* are orthogonal and hence

Ses and meee)

Ou" ou?

Differentiating these two scalar products with respect to u® and u%,

respectively, and adding, we get

1(20. Or | On ot) = —n. o'r

2\auz due | dub Ou" du? due
‘Hence
o’r
i bas =
eo. ‘ du" du?
However,

e: = a, b,x;3
ou*
therefore
O’r Ox' Ob; ,;
bet ee
Ou" Ou du’ ~—-du?
pre Obs a,
“au? " Ox “Ay

= hb,
ae rat tsl) +
Ape
oe r)

where, in the last step, we made use of formula 46.4 for the derivative of
the base vector b,.
If we insert in the right-hand member of the foregoing expression from
equation 67.1, we get
(67.10) a
Parent = b,{ x a, e+ ae hal”?od
182 GEOMETRY [CHap.3

Multiplying equation 67.10 scalarly by n, and observing that the vectors

b,x = a, and n are orthogonal, we get
or i
Ne ==) ie b;X2,5
du du?
= Xz;
= Dap

by formula 67.7. This establishes the equivalence of the two definitions

of the second fundamental quadratic form.
Equations 67.7 are known as the formulas of Gauss. The importance
of the form 67.8 in differential geometry stems from the fact that the
tensors a,, and b,,, satisfying equations of Gauss and Codazzi (to be
derived in Sec. 69), determine the surface to within a rigid body motion
in space.
Problems
1. Show that b,, = g,j0),
gM” = del €55,.%%, g®X)-
2. Show that, in the notation of Sec. 65,
1 (on Or on oF
Sj ae Se || S Se
OuX duh i eub =) S iil Xp»

where nis the unit normal and r is the position vector of the point on the surface.
3. When the equation of a surface S, referred to a set of orthogonal cartesian
axes, is taken in the form y? = f(y', y”), we can write the parametric equations
of S in the form y' =uw,y? =v, y? = f(w, vu’). If partial derivatives of
fy’, y*) are denoted by fn =p, fe =] fry =" fyyt =5, fy: = ¢, then the
coefficients a,, in ds* = a, , du du are
a, =1+> p’, Q1. = pq, Qa. = 1 +4",
whereas the coefficients bap of the second fundamental form are

r Ss i
by =, SS by = —— -
Vit¢pt+”@ Vi+p?+¢@ Vitp+¢
Show this and compute a,, and b,, for the surface of the sphere

y> = Va? — (y)? — (2)?

4. If equations of S are
S: y =y'W, uw),
where the y? are orthogonal cartesian coordinates, and r is the position vector
of the point (y’, y’, y*) on S, then on using subscripts to denote partial deriv-
atives,

a8 = Ty,« ° Ty, bys = [r,%8 “Ty X r,2]/ Va.

Show this. Use these formula to compute the a,, and b,, for the surface of
revolution y! = ul cos uv, y? = ul sin uw, y® = f(u!),
SEC. 68] THE INTEGRABILITY CONDITIONS 183

68. The Integrability Conditions

In order to get insight into the significance of the tensor_b,, let us

examine more closely the Gauss formulas

ha — bap’,
(68.1)

where

[67.1] a Ox" + {ile Aid:tee — |Oley i

PF aut dub © aljk heen a8 ae
and

[65.10] Mahe epTee,

with x,’ = 0x7/du*.

If we insert these expressions in equation 68.1, we obtain a set of
second-order partial differential equations, in which the dependent vari-
ables x‘ are functions of the surface coordinates u*. The coefficients in
these differential equations are functions of metric coefficients g;; of the
manifold in which the surface S, defined by

(68.2) ai=xi(ul,wv), (i= 1,2, 3),

is immersed; they are also functions of a,, = g;,(0x'/ du*)(dx/du*), and b,,.
1,
If equations 68.2 are given, we can compute 4,, and b,, (see Problem
Sec. 67), insert the appropriate expressions in (68.1), and, of course,
the
equations 68.1 will be satisfied identically. On the other hand, if
will become
functions a,, and b,, are prescribed in advance, equations 68.1
s, and in general they will have no solution s yielding
equations of condition
a,, and b,, be
equations 68.2 of the surface S. In order that the tensors
integrability
related to some surface, it is necessary that the 2° satisfy the
conditions,
+ Saf Za t

(68.3) Bi gel ety

du’ du? = du® Ou?

the functions x,‘ are of class C®. From our discussion of

whenever
Sec. 36, it follows that
inversion of order of covariant differentiation in
the condition 68.3 is equivalent to (cf. equation 36.6)
I ° é .
(68.4) XaBy — La yB a ReapyX >

they are not essential to the

22 We dispense with the details of computation since Calculus,
course of argument. example, A. J. McConnell, Absolute Differential
See, for
p- 203.
184 GEOMETRY [CHaAP. 3

where R®,,., is the Riemann tensor of the second kind, formed with the
aid of the coefficients a,, of the first fundamental quadratic form. Equa-
tions 68.4 involve third partial derivatives of the coordinates x’, and we
shall assume from now on that the functions entering in (68.2) are of
class C*.
We shall see that the conditions of integrability 68.4 impose certain
restrictions on the possible choices of functions b,, and a,;. These
restrictive conditions are known as the equations of Gauss and Codazzi.
They will be derived in the following section.

69. Formulas of Weingarten and Equations of Gauss and Codazzi**

In order to derive the equations of Gauss and Codazzi we need an

auxiliary result, due to Weingarten, giving the expressions for the deriva-
tives of the unit normal vector n* to S. We begin with the relation
g,,n'n’ = 1, and form its tensor derivative.24 We have
Bn ,an’ + gijn'n’, = 0,
or
(69.1) gynn, = 0.
Equation 69.1 shows that n’,, considered as a space vector, is orthogonal
to the unit normal n'‘, and hence it lies in the tangent plane to the surface.
Accordingly, it can be represented as a linear form in the base vectors zt,
(69.2) ie oto
Since n‘ is normal to the surface, we have the orthogonality relation
£,;v,n’ = 0, whose tensor derivative is
(69.3) ij%q,p0? + Bi;tzn’, = 0.
But, from (68.1),

(69.4) Xp = dapn’,
so that the substitution from (69.4) and (69.2) in (69.3) yields

bap + Big%a2ich = O,
and, since a,, = 2;;%,'%,’, we have

bap = —AgyCp”.
*® The treatment given here is patterned after A. J. McConnell, Absolute Differential
Calculus, pp. 201-205.
24
We recall that n,, i = Be
ont
oo :i nak,
j.

NS ay
SEC. 69] FORMULAS OF WEINGARTEN 185
Solving this equation for c,”, we get

Cp” = —a"'D 5;

so that equation 69.2 reads

(69.5) WalDy et
These are the Weingarten formulas which we will use in deriving the
Codazzi equations.
The equations we desire follow from the integrability conditions 68.4,
namely,

(69.6) Xap — Xa.rp = Riapy

to
We form the tensor derivative of equation 69.4, and use 69.5 to obtain

(69.7) Te pee Deg st at Doll

— Dap, yn’ ae bya bgan

Substituting from (69.7) in the left-hand member of (69.6), we get

; ‘ ea ; bh :
La py oF La yp — (bap, aad Day,p)n an (bapb5, ee Daybsp)%; -

Hence

(69.8) (bap, <7 bay,p)n' na a®*(bapbs, = bay bsp)%a° = Rispy%o

To obtain the equations of Codazzi we multiply (69.8) by n,, and, since

ain, = 0, we get the desired result
(69.9) boa = Day.p = 0.

To obtain the equations of Gauss we multiply (69.8) by g,) and obtain

(69.10) bypbay as, by, bap ae Roapy:

Since «, 6 assume values 1,2 and b,, = b,,, we see that there are
two independent equations of Codazzi and only one independent equation
of Gauss.?° The independent equations of Codazzi are

(69.11) Daz.p — Papa = % (a x B), (no sum on «),

aid of
or, when the covariant derivatives are written out in full with the
Obyp 6 6
Papi = auy abea Lippe
25 We recall that

Raapy = Rapyy = 0, Ryorz = Reie = — Rez = —Rie21-

186 GEOMETRY [CHAP. 3

we get
db Obip |6 {i
69.12 —a2 _ —ah _ p, +b =0, «Ff,
Gua ous lap Maer
(no sum on @).
The equation of Gauss, on the other hand, is

(69.13) byyb22 — biz = Riore-

This equation relates the coefficients b,; and a,, in the two fundamental
quadratic forms.
The foregoing demonstration shows that if the tensors a,, and b,, are
the fundamental tensors of the surface S: x’ = x‘(u, u?), then equations
69.11 and 69.13 are satisfied. Conversely, it can be shown that if the two
sets of functions a,, and b,, satisfying equations 69.11 and 69.13 are
prescribed, and if a,, du* du’ is a positive definite form, then the surface
S is determined (locally) to within a rigid body motion in space. The
proof?® of this depends on considering the existence of a solution of a
system of differential equations of the type discussed in Sec. 39. We
conclude by remarking that if b,; = 0, then by 65.5 the surface is a plane.

70. The Mean and Total Curvatures of a Surface

If we recall the definition 62.4 of the total curvature K,

[62.4] K = as > a= A414 92 a G5"

we can write equation 69.13 in the form

b
(70.1) K —
5=
bir bo. ana bis"
7:
431499 — Ae a
Thus the Gaussian curvature is equal to the quotient of the discriminants of
the second and first fundamental quadratic forms.
We introduce next another important invariant H, called the mean
curvature of the surface. This is given by the formula

(70.2) H = 3a*b,,,
and we shall see in Sec. 72 that the invariants K and H are connected in a
remarkable way with the ordinary curvatures of certain curves formed by
taking normal sections of the surface.
*6 For a detailed discussion, see L. P. Eisenhart, Introduction to Differential Geometry,
pp. 218-221, where the case of cartesian variables x‘ is considered.
SEc. 71] CURVES ON A SURFACE 187

71. Curves on a Surface. Theorem of Meusnier

Let equations of a smooth curve C lying on the surface

(71.1) yoann et 1)
be given in the form
(/i.2) Coe 1S),
where s is the arc parameter. If the values of w*(s) are inserted in (71.1),
we obtain the space coordinates x‘ of C in the form
(71.3) ete ex'((5'),
These are the equations of C, regarded as a space curve. The properties
of C can then be studied with the aid of the Frenet-Serret formulas 50.1,
50.2, and 50.3 by analyzing the rates of change of the unit tangent vector A,
the unit principal normal p, and the unit binormal v.
On the other hand, if we regard C as a surface curve, defined by (71.2),
the components /* of the unit tangent vector A are related to the space
components 4‘ of the same vector by the formulas
Or dil:
71.4 = = 7,1",
ia du* ds
where
nee du*
med: ya cera and i
ue) ds ds
We also recall equation 63.4,
on
(71.6) — = X41",
os
where 7% is the unit normal to C in the tangent plane to the surface, and
x, is the geodesic curvature of C: (See Fig. 29.)

tS
\

Fig. 29
188 GEOMETRY [CHAP. 3

If we differentiate (71.4) intrinsically with respect to s, we obtain

OA oe pa die SOAS
+ Sa gh ;
ds ds ds
which, upon taking into account Frenet’s formula 50.1 and equation
71.6, becomes
Me ek A get
The space components 7‘ of y are 1* = x,'7*, and, if we recall Gauss’s
formula 2!,, = b,,n', the foregoing equation becomes
(71.7) sp = DBA An’ + un,
where n’ is the unit normal to the surface S.
Formula 71.7 states that the principal normal pu to C lies in the plane
of the vectors nand y. Since n, n, and A are orthonormal and n x y =A,
we have
(71.8) ej, A,
and, since A is orthogonal to the plane of nand p, w x n = —sin OA, or

where @ is the angle between w and n.

On multiplying (71.9) by x we get
€,;,xu'n® = — x sin 8: A,,
which, on substitution from (71.7), yields

Eu (DapA "Ain? 45 x_n?)n* = — *sin 0: A,.

But ¢,,,n’n* = 0 and e,;,7’n* = —A,; by (71.8), and we conclude that

(71.10) %, = x Sin 6.

On the other hand, if we form the scalar product of both members of

equation 71.7 with n, and note that n,u' = cos 6, we get

(71.11) x cos 0 = b, gd".

The invariant b,,4%A° in (71.11) has the same value for all curves on S
with the same tangent vector A at P. In particular, it has this value for the
curve formed by the intersection of the normal plane containing n and A.
But for every normal plane section the angle @ is either 0 or = radians,
so that for the normal plane section x cos 6 = x or —x; since the right-
hand member of (71.11) is an invariant, the value of x cos @ for every
curve C tangent to A is equal to the curvature x,,,, of the normal section
SEc. 71] CURVES ON A SURFACE 189
in the direction A. The curvature x,, is called the normal curvature of
the surface S in the direction A. We can thus write (71.11) as
(71.12) #(n) = by pA*A?, “

where x,,, = x cos 8. Accordingly, Eq. 71.7 can be written as

jt = #tyyn* + 27°.
This equation states that x,,, and x, are the components of the curvature
vector x‘ in the directions of the vectors n‘ and 7’.
The result embodied in formula 71.12 can be stated as
MEUSNIER’S THEOREM. The radius of curvature R = 1/x of any curve
at a given point on the surface is equal to the product of the radius of curvature
Rin) = 1[%(,) Of the corresponding 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,,) cos 6.
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 elementary geometric considerations. (See Fig. 30.)
If we recall that ds? = a,, du* du’, and du*/ds = i*, we see that formula
71.2 can be put in the form

(71.13) %(n) wa) aaiaiuh

actates Pats, pee
ae

We note that, if the surface is a plane, the normal curvature x,, = 0 at

all points of the plane, and if it is a sphere, %(,) = 1/R, where R is the
radius of the sphere. Accordingly, we conclude from (71.13) that for the
plane b,; = 0, and for the sphere biz due du’ = (1/R)a,, du* du’ so that
a,j = Rb,, at all points of the sphere.

Fig. 30
190 GEOMETRY [CHAP. 3
Problems

1. Prove that the geodesic curvature x, and the curvature « of any surface
curve C are connected by the formula x, = sin 6, where @ is the angle between
the normal to the surface and the principal normal to C.
2. Consider the surface of the right circular cone
Sag) = wicosius
y2 = u sin u?,
Te.
and the curve
C2 fsa. a OnnSs
Write equations of C in the form ul = a, u* = s/a, where s is the arc parameter
and show with the aid of (71.6) that x, = V2/2a. Verify this result by (71.10).
3. Show that the parallels u’ = constant on a sufficiently smooth surface of
revolution
y> = uw cos u,
y? = usin v2
= f(u’)
are curves of constant geodesic curvature.
4. A curve C on a surface S is called an asymptotic line if b,,A*2°= 0 along
C. Show that the principal normal p to the asymptotic line is fnipent to S and
the binormal v is normal to S.
5. Show that the normal curvatures in the directions of the coordinate curves
are by/ay, and b59/d39.
6. Prove the theorem: If a curve is a geodesic on the surface, then either it isa
straight line or its principal normal is orthogonal to the surface at every point,
and conversely.

72. The Principal Curvatures of a Surface

We will be concerned in this section with the determination of directions

4* = du*/ds on the surface such that the normal curvature x;,), given by
the formula

[71.12] Xin) — by pA’,

assumes an extreme value.

Since the vector 4* is a unit vector, ~;,, in (71.12) has to be maximized
subject to the constraining relation
G21) Aq,AA? = 1.
Following the usual procedure of determining constrained maxima and
minima, we deduce that a necessary condition for an extremum is
SEC. 72] PRINCIPAL CURVATURES OF A SURFACE 191
where A is the Lagrange multiplier. If equation 72.2 is multiplied by
A* and account is taken of relations 71.12 and 72.1, it follows at once
that A = —x,,. Thus equation 72.2, for the determination of directions
yielding extreme values of x,,,), can be written as :

(72.3) (bap aa H(n)ap)A? = 0, (x ay UF 2),

The set of homogeneous equations 72.3 will possess nontrivial solutions

for A’ if, and only if, the values of ;,,) are the roots of the determinantal
equation
(72.4) lbap Sa Bagg — 0.

The quadratic equation 72.4, when written out in expanded form, is

(7235) B — ab gd + ee 0,
a
where b = |b,,| and a = |a,,].
Since the Gaussian curvature K is given by

70.1] Kee
a
and the mean curvature H is
[70.2] H = 4a"b,,,
we see that equation 72.5 assumes the form
(72.6) — 2H + K=0.
The roots @ = xq) and 3 = x.) of (72.6) are called the principal curvatures
of the surface, and the directions 4% and 4%, corresponding to these
extreme values of x;,,, are the principal directions on the surfaces. We
leave it for the reader to show that these directions are real.
From (72.6) it is clear that the principal curvatures x.) and %.») are
related to the mean and Gaussian curvatures by the formulas
#1) + xX(2) = 2H,
(72.7) |
(1)%(2) = K.
From equation 72.3 it follows that the principal directions are determined
by
Hee = (1) Gap Aa) = 0,
(beg — %(2)4ap)A(2) = 0.
If the first of these equations is multiplied by 2%), the second by 4%),
and the results subtracted, we obtain
(72.8) (4¢2) — ay )GapAy
Alo)= 9.
192 GEOMETRY [CHAP. 3

If x1) F 2), equation 72.8 tells us that

(72.9) Gap hiy Alay = 0,

that is, the principal directions are orthogonal. If the extreme values of
%(n) are equal at a given point, then every direction is a principal direction.
We can summarize these results as a
THEOREM. Af each point of a surface there exist two mutually orthogonal
directions for which the normal curvature attains its extreme values.
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. The
differential equation for which the lines of curvature on S are the integral
curves follows directly from equations 72.3. If we eliminate ~;,,, from
these equations and set A° = du®/ds, we get

bigdu? _ bapdul
aygdu® — dog du®
or
(72.10) (41412 — byp@y1)(du')® + (611422 — 9441) du’ du?
+ (by2422 — Ay2b2)(du*)? = 0.

At each point of S where either 6,, du* du? # 0 or b,; du* du® is not
proportional to a,,du* du’, equation 72.10 specifies two orthogonal
directions

(72.11) “~=¢,u',u),
u
(«= 1,2),
which coincide with directions of the principal curvatures.?”? Each equation
in (72.11) determines a family of curves on S covering the surface without
gaps. 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)? = G,,(dii*)? + Go.(diui?)*.
Accordingly, equation 72.10 in the coordinate system a* takes the form
—by2dy1(dit?)? + (by 1429 — by94,) di! dit? + by.d2.(di?)? = 0,
and its solutions are
ui = constant, u? = constant.

If we take di! ¥ 0 and di? = 0, we see that 5,, = 0, since 4,, ¥ 0. Thus
a necessary condition for the net of lines of curvature to be orthogonal
*7 We exclude those points on S at which x, =0 or x) = %@). See concluding
remarks in Sec, 71.
SEC. 72] PRINCIPAL CURVATURES OF A SURFACE 193
is that ay. = by = 0. Conversely, if a,, = by. = 0, then (72.10) has
solutions ut = constant, vu? = constant, so that the coordinate lines are
the lines of curvature. Hence a
THEOREM. A necessary and sufficient condition for the cootdinate net
on a surface S (other than a plane or a sphere) to be the net of lines of
curvature is that ay. = by, = 0 at allpoints of S.
We note that for every orthogonal net on a plane or a sphere a,. =
by. = 0.
Formula 71.9, for the normal curvature x,,), when the coordinate
system is taken to be the net of lines of curvature becomes

oe by,(du’)? a5 byo(du*)?
- dy,(du’)” + ay(du*)”
If we set dul = 0, du? ¥ 0, and du* = 0, du ¥ 0, we get

for the curvatures of the coordinate lines u, = constant and u, = constant.

The lines of curvature on S should not be confused with the normal
sections of S. The normal sections C,, are necessarily plane curves,
whereas the lines of curvature ordinarily are not plane curves.
We conclude this section by giving several definitions.
A surface at all points of which the Gaussian curvature K is positive
is called a surface of positive curvature. In this case (see equation 70.1),
byyboo — bys? > 0, and, since %;,) = b,A%A", we see that the principal radii
Rin) = 1/%,) to all normal sections of a surface with positive curvature
do not differ in sign. If K < 0, at a given point, the principal radii differ
in sign. Then the equation
(72.12) bapAti? =
defines two directions for which the radii of curvature are infinite. A
surface at all points of which K < 0 is called a surface of negative curvature.
If K = 0 at a given point, the directions given by (72.12) coincide, and
for this direction R is infinite. ,
From geometrical considerations it is clear that ellipsoids, biparted
hyperboloids, and elliptic paraboloids are surfaces of positive curvature.
Hyperboloids of one sheet and hyperbolic paraboloids are surfaces of
negative curvature.
A point on S is said to be elliptic if the signs of the principal curvatures
1), #2) are the same. It follows then that ~;,) at an elliptic point does
not change sign for any direction of the normal section. A point is
hyperbolic if x4) and x2, have opposite signs. At a hyperbolic point there
194 GEOMETRY [CHAP. 3

are two directions for which x;,) = 0. A point is parabolic if one of the
values 4) OF xg) is zero. In the special case x1) = %(2), all values of xn)
are equal and such points are called spherical. 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 S is a sphere. In some
books spherical points are also called umbilical.
Problems

1. Given an ellipsoid of revolution, whose surface is determined by

12s :
y) =acos usin uv,
y2ey= asin
;
u'5 sinPe wv,
2

y®> = cos uv’, (ord eA ge

show that
apa si u, aig = 0, Geo = a CoS? + c* sin* Hr,
+ 42 u 12
ac sin? ac
by =, biz = 0, bos ee
V a? cos? u2 + c? sin? v2 Va? cos? v2 + c® sin? 2
and
C2
KS Ai“
M2) 2 SSS;
(a2 cos? u2 + c? sin? u2)2

Discuss the lines of curvature on this surface.

2. Find the principal curvatures of the surface defined by

Yee
yaa,
y® = f(u', uv’).
3. Show that the helicoid
y =u cos u*,
y*? = uv sin v2,
y> = au

is a surface of negative curvature.

4. Show that when at all points of the surface b.3 du* du’ is proportional to
axp du* du’, thenb,, = ka,s,k = constant. Interpret this result geometrically.
5. Prove that if every point of the surface S is parabolic, then S is developable.
6. Given a surface of revolution S,
y) = rcos¢, y? = rsin¢, ys = f(r),

with f(r) of class C*. Prove that the lines of curvature on S are the meridians
¢ = constant and the parallels r = constant.
7. Refer to Problem 6 and show that the points on a surface of revolution
S for which f’f” > 0 are elliptic; those for which f’f” < 0 are hyperbolic; and
if f” = 0, then S is a cone.
SEC. 73] PARALLEL SURFACES 195
8. Let the vector equation of a curve C drawn on a smooth surface S be
r =r(s). Ifn(s) is a unit normal to S at a given point of C, and v is a parameter
measuring the distance along n, the vector equation R(s,v) = r(s) + vn(s)
defines a ruled surface S’. Prove that S’ is developable if C is a line of curvature,
and conversely. Outline of solution: Denote the coefficients in the second
fundamental form of S’ by dg: Compute the d,, from the formula

2R
dp = ea0
Fs N,

where v! = s, v? = v, and N is the unit normal to S’. Show that d,. = 0. The
developability condition d,,dy. — dj,” = 0 implies then that d), = 0. But, along
C, N = dr/ds x n; hence
dn dr
1 cae ae P n = 0.

If dn/ds = 0 along C, then S’ is a cylinder, which is a developable surface. If

dn/ds and dr/ds are collinear, then dn = k dr, which leads to the set of equations
(b.8 — ka,,) du? = 0 of the type (72.3). Retrace steps to prove the converse.
9. Let C be a smooth curve defined by r = r(s). The tangent surface S to C
is defined by R(s, v) =r(s) + v(dr/ds), v is a parameter measured along the
tangent dr/ds. Prove that S is developable. The curve C is called the edge of
regression for S.
10. Prove Dupins theorem. The coordinate surfaces of every triply orthog-
onal curvilinear coordinate system in E; intersect along the lines of curvature
of coordinate surfaces. Hint: Consider the surface x* = constant and take
a! = y!, 22 = uv? as surface coordinates on it. Show that along the coordinate
lines ut = constant, u® = constant, by, = 0 if a;, = 0. See Problem 4, Sec. 67.

73. Parallel Surfaces

Let S be a smooth surface defined by equations

(73.1) yio=yiul,uv), (= 1,2, 3),

where the coordinates y‘ are orthogonal cartesian. A surface S determined

by equations
(i302) gi(ul, u®) = y*(ul, v2) + hn*(u', uv’),
ed along the
where n’ is the unit normal to S and h is the distance measur
normal n, is called a parallel surface to S.
plates and
Parallel surfaces figure prominently in the theory of elastic
K and the mean
shells, where relations connecting the Gaussian curvature
for the surface S
curvature H of S with the corresponding invariants
are important.
 196 GEOMETRY [CHAP. 3

We proceed to outline a derivation of such relations by recalling first

that the base vectors a,, along the curves uw, = constant, are related to
the base vectors b, along the y’-axes by

[64.10] a, = b, og
Ou"
For simplicity in writing we introduce the notations (cf. Sec. 64)

Bu? Ya But Ya»

and, on differentiating (73.2), we get

(73.3) Ge = Ya + hn’,
so that

(73.4) Jan; = Yn; + hn',n,.

But y,’n; = 0, for the a, are orthogonal to n and nn; = 0 since nn, =" 1.
Thus (73.4) reduces to

(73.5) 70, = 0,
On the other hand, the unit normal #; to S is orthogonal to the base
vectors y/, on S, so that
(73.6) ¥,n, = 0.
We conclude from (73.5) and (73.6) that the vectors n; and A, are
collinear, and, since they are unit vectors, n; = fj;.
The metric coefficients a,, of S are given by*®

ayp = Yn Gp's

which, on making use of (73.3), yield

agp a (CA ae hn‘, )(yp" “f hn‘)

= Ya'Vp + hn'yg' + hnigyg + hentn's.

The substitution in this expression from the Weingarten formula,
[69.5] n= — a bey
gives

(73.7) Gzg = Ayg — 2hbyg + h*nin',,

Since ¥,"Ys' = Ayp.
*8 See (64.6) and recall that ¢,; = 6;; since the coordinates y are orthogonal cartesian.
Sec. 73] PARALLEL SURFACES 197

The last term in the right-hand member of (73.7) can be expressed in

terms of the Gaussian curvature K and the mean curvature H as follows.
On making use of (69.5) we get
“
.

H
a

(73:8)
, nin p =
5 aise a” bys fa" bay :
==" 0 ap’ BAs

since yiy', = a,,,- On the other hand, the Gauss equations 69.10 require
that

[69.10] Rapys = Daybps — Dasbpy,

where
Rapys = Kéap€ys
by (62.5), so that
Ke,g€y5 = baybgs — bib py:
a®’, sum on 4, and note
We multiply both members of this relation by
that (cf. Sec. 62)
ae
npy3= —Apy
and find

(739) — Kaz, = a™’b,,bgs — 2Hb,,,

since
H = 3a”b,>.
[70.2]
the right-hand member of (73.9)
The substitution in the first term in
from (73.8) yields
(73.10) nin', = —Kazg + 2Hb«p,
in the form
and we can thus write (73.7)

(73.11) digg = Ayg(1 — h®K) — 2hbag(1 — hH).

les us to compute the coefficients
The important formula 73.11 enab K, and H at
P(,, U2) on 5 from the values of a,,, 6,3,
G,, at a given point
, Us) on S. «
the corresponding point P(u of 5,
in the second fundamental form
To compute the coefficients b,,
we recall that
bap = Yx,pli
.3),
by (67.7). But from (73
Vip = Yip me hn aps

so that

(73.12) bap = bap + hn’ spn.

198 GEOMETRY [CHAP. 3

Since the coordinates y‘ are rectangular cartesian, n’ = n;, nin® = 1, and

we conclude that
nine 0,
On differentiating this orthogonality relation we find that nin’ =
—n',n',, and hence (73.12) can be written as
bap = Dag — hn’ans.
On making use of (73.10), we finally obtain

(73713) big = (1 — 2hH) bag + hKazgy.

- Formulas 73.11 and 73.13 appear on pages 110-111 of a monograph
by T. Y. Thomas, Concepts from Tensor Analysis and Differential Geometry,
Academic Press (1961). They are closely related to formulas on page 272
of L. P. Eisenhart’s Differential Geometry, Princeton Press (1940).
Once the coefficients 4,,, byp are known, the Gaussian and mean curva-
tures K and H can be computed from formulas 70.1 and 70.2. The result
of somewhat lengthy computations, which will be found on pages I11 to
113 of the mentioned monograph by T. Y. Thomas, is

(73.14)
Seay Taia Fe2hH
:
°
pe H—hK
1 + h?K — 2hH
From the first of these elegant formulas it follows that when S is a develop-
able surface, then the parallel surfaces S$ are also developable.
Problems
1. If Sis a surface of revolution, show that a parallel surface S is also a surface
of revolution. Hint: Consider y! = u' cos u?, y® = wu! sin u?, u3 = f(u?).
2. Show that the principal radii R,,, of normal curvature of S are related to
the principal radii R,,) of a parallel surface S by
Ree
a
hs

74. The Gauss-Bonnet Theorem

The description of surfaces with the aid of differential equations has

local character, since relations among the derivatives describe properties
of surfaces only in the neighborhood of a point. To obtain results valid
for the entire surface one must perform integrations. Because of the
complex structure of differential equations of the theory of surfaces,
relatively few global results have been obtained, and the available results
Sec. 74] THE GAUSS-BONNET THEOREM 199

Fig. 31

in global geometry are largely concerned with a special class of convex

surfaces. There is one important classical result, however, that 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 elegant theorem of geometry of surfaces in the large.
Let D be a region bounded by a closed piecewise smooth curve C
drawn on a smooth surface S, shown in Fig. 31. We shall suppose that
a unit
D is homeomorphic to a circular disk.? We saw in Sec. 63 that
C is related to the unit vector *
tangent vector 4% to a surface curve
normal to A* by

[63.4]
63.4 ry
a geodesic,
where ~, is the geodesic curvature of C. Moreover, if C is
then x, = 0 at all points of C, and conversely.
paragraph
Since 7% is orthogonal to /®, it follows from the concluding
of Sec. 54 that ¢,97%4" = 1, or
ee €aph?.
But from (63.4)
. on
pasill ar R.
or ;
on
Kg
os €
€ap
44Os
(74.1)

curve C gives
The integration of this expression over the
On
[ dss =| [/: phi —ree ds,
€,,A°
(74.2)
if they can be mapped into one another
2° Two regions are said to be homeomorphic
in a continuous one-to-one manner.
200 GEOMETRY [CHaP. 3

A(B)

Fig. 32

and when the line integral in the right-hand member of (74.2) is trans-
formed into a surface integral by Green’s formula, we find that*°

(74.3) i}x, ds = =i K do + 2 — Lo — a),

where the «, are interior angles of the contour C shown in Fig. 31, and
do = Va du! du? is the element of the surface area of D. If Cis smooth,
the sum > (7 — a,) = 0.
Formula (74.3) embodies the statement of the Gauss-Bonnet theorem.
Instead of deducing this theorem with the aid of Green’s formula, we give
a geometric interpretation of formula (74.3) that suggests an alternative
definition of Gaussian curvature.
Consider first a sphere S of radius R and a spherical triangle P,P.Ps;
on S (Fig. 32) formed by the arcs P,P, P:P3, PsP, of three great circles.
Denote the interior angles of this triangle at the vertices P; by «; and cover
*° The details of easy calculations are found in the following books: A. V. Pogorelov,
Differential Geometry, P. Noordhoff, N.V. (1959), p. 161; L. P. Eisenhart, Differential
Geometry, Princeton (1940), p. 191; D. J. Struik, Lectures on Classical Differential
Geometry, Addison-Wesley (1950), p. 154.
SEC. 74] THE GAUSS-BONNET THEOREM 201
the sphere by some coordinate net (u,, v2). If the base vector along the
u,-coordinate line at P, is a, and A(P,) is an arbitrary surface vector at
P,, we denote the angle between a, and A(P,) by g. Let 4 be the angle
made by A(P,) with the geodesic arc P,P,. When A(P;) is propagated
in a parallel manner along the geodesic triangle P,P,P3, it assumes the
position A’(P,) shown in Fig. 32. Our immediate object is to determine
the angle my’ between a, and A’(P)).
During the parallel propagation of A(P,) along P,P,, the angle @ is
unchanged (see Sec. 60), and the vector A assumes the position A(P2)
with the geodesic arc P,P3, then
B = 7m — (a + 8).
In the course of parallel propagation of A(P,) along P,P; the vector A
continues making angle 6 with P,P; and, on reaching the point Ps, it
assumes the position A(P;). Let y be the angle between A(P;) and the
arc P,P,, then
y =a, — PB= a, — [7 — (22 + DJ = a + a+ 9 —7.
On continuing propagations of A along P3P,, the vector A maintains
the angle y with P,P; until it reaches the point P, when it assumes the
position A’(P,). Now, the angle y’ made by A’(P,) with a, is
p=ytutep—8
igo
Opt os 1 — TM

so that the angle g’ — p between A(P’) and A(P) is

(74.4) gp —pHuyt % + %— 7.
The change ’ — @ representing the difference between the sum of interior
angles of the spherical triangle P,P,P; and the sum of interior angles of
the rectilinear triangle is called the spherical excess of the spherical
triangle P,P,P3. If instead of interior angles «; we introduce the exterior
angles 0; = 7 — «,, formula 74.4 reads

gp — pada Q9,.
t=1
of n sides,
When the vector A is propagated along a geodesic polygon
spherica l excess of the polygon®*
entirely similar computations yield for the

yg — p= t=1>, — (Ww — 2)n,

or

gy —p=2r—- D9,
i=1

polygon of n sides is (n — 2)7

31 Note that the sum of interior angles of a rectilinear
radians.
202 GEOMETRY [CHAP. 3

if we use the exterior angles 0; = 7 — «,. But it is known from spherical

trigonometry that the spherical excess of a geodesic polygon is equal to
o/R®, where o is the area of the polygon and R is the radius of the sphere.
Thus

and since the Gaussian curvature K for the sphere is equal to 1/R*, we
can write
ar 2H,
(74.5) K To eet
This formula can be generalized to obtain the Gauss-Bonnet formula
74.3 for the case where C in Fig. 31 is a geodesic polygon. Thus, if the
region D is subdivided by small geodesic polygons into subregions of
areas do,, the familiar procedures of integral calculus applied to (74.5)
yield
(74.6) |i:Kdo =2n — > 6.
This formula coincides with (74.3), since x, = 0 when C is a geodesic
polygon.
‘Formula 74.6, first obtained by Gauss, was generalized by Bonnet* to
yield the result (74.3), which, as we have already noted follows directly
from (74.2) on application of Green’s formula.
The left-hand member {|K do in (74.6) is called the integral curvature
D
of D. It turns out that the integral curvature is a topological invariant.
Two surfaces 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 (74.3) that [| « do = 47 for all regular surfaces
D
topologically equivalent to a sphere and {|Kdo = 0 for all regular
D
surfaces topologically equivalent to a torus.**

75. The n-Dimensional Manifolds

It is the purpose of this section to introduce a few concepts from the

geometry of n-dimensional metric manifolds which are of interest in
applications to dynamics and relativity. Many of these concepts are
straightforward generalizations of ideas introduced in this chapter in
** O. Bonnet, Journal école polytechnique, 19 (1848), pp. 1-146.
*° See D. J. Struik, op. cit., pp. 153-159.
SEC. 75] THE n-DIMENSIONAL MANIFOLDS 203
connection with the study of surfaces imbedded in the three-dimensional
Euclidean manifolds.
We shall suppose that the element of distance between two neighboring
points in an n-dimensional manifold is given by the quadratic form
(15.1) dst=g,,dzidsi, (j=1,...,n), with |g,,| 4 0.
We extend the definition of Euclidean space, given in Sec. 29, by saying
that the space is Euclidean if there exists a transformation of coordinates
x‘ such that the transform of ds? is a quadratic form with constant
coefficients. Since every real quadratic form with constant coefficients
can be reduced by a real linear transformation to the form
(75.2) ds? = Adz’, (A, = £1),
the form 75.2 can be used to define an Euclidean n-dimensional manifold.
If, in particular, the form 75.2 is definite, we shall say that the manifold
is purely Euclidean, but if it is indefinite, the manifold will be called
pseudo-Euclidean.
A linear manifold determined by a set of m equations
C; act a a*(t), ty = t a to,

with suitable differentiability properties, will be said to define a curve Cin

an n-dimensional manifold.
If the form 75.1 is positive definite, we shall say that the positive number

s= |“Vg,Ada'ldty(dx'|dt) dt
manifolds
is the length of the curve C. There are definitions of metric
the element of arc in the form
which are not based on the expression for
75.1, but they need not concern us here (see Sec. 43).
on of the curve,
The vector 4‘ = dx'/ds will be said to define the directi
a unit vector. The
and it is clear that g,,A‘A? = 1, so that the vector Ai is
length of any vector A’ is given by the formul a
thes V 9,,A'Al

ensional manifold is a
The notion of the angular metric in an n-dim
of the angle in the three-dimensional
direct generalization of the definition
case.
the cosine of the angle
If A‘ and w* are two unit vectors, we define
between them by the formula
(75.3) cos 0 = g,,A‘u’.
204 GEOMETRY [CHAP. 3

It is not clear from this definition that the angle 6 is necessarily real. We
shall prove, however, that this is always so if the form g,; dx‘ dx’ is
positive definite. The proof follows at once from the Cauchy-Schwarz
inequality

(75.4) (g,,2°y’)? Ss (g,,2'x/)(g,y'y’),

where the form g,;x'x’ > 0. o

We first establish the inequality 75.4. Let the form Q(x) = g;;x‘x’ be
positive definite. If we replace in it x* by x* + Ay’, where / is an arbitrary
scalar, we obtain

Ole + Ay) = gia? + Aye? + Ay’)

= g,0'x) + 2g, 2'y71 + gyy'y??
= Oe) + 20(e, yA + OYY)H.
This is a quadratic expression in A with real coefficients. By hypothesis
O(x + Ay) > 0, the sign of equality holding if, and only if, * + Ay* = 0.
Hence, the equation in A,

f(A = OY)? + 20, y)A + Oz) = 0,

possesses no distinct real roots. But a necessary and sufficient condition
that this be true is that [Q(x, y)]}* — Q(y)Q(a) < 0, that is,

(g:;2°y’)? < (g;;0'x Ng uy’).

This is precisely the inequality 75.4.
If, now, in formula 75.4 we set x* = A* and y* = uw’, we get

(gisAin’)
(SijsAA Sie’)
and, since A’ and qm’ are unit vectors, we have (g,;A‘u’)® < 1, which
states that the angle @ in formula 75.3 is real.
We define the volume element in R,, by the formula
anes Vg dx! dx® ++» dx",

and the volume by the corresponding n-tuple integral.

A generalization of the concepts of curvature and torsion to curves
imbedded in the n-dimensional Riemannian manifolds is direct and
straightforward,** but matters become rapidly involved when one comes
to consider hypersurfaces.
94 See, for example, J. C. H. Gerretsen, Lectures on Tensor Calculus and Differential
Geometry, P. Noordhoff, N.V. (1962), Chapter 6.
SEc. 75] THE n-DIMENSIONAL MANIFOLDS 205
A set of n equations
ale te wa, Ue . . «1s (ae ee eee as m<n,
is said to define an admissibly parametrized m-dimensional variety (or a
hypersurface) over a neighborhood of the variables u* if (a) the x’ in (75.5)
are of class C? and (b) the Jacobian matrix (07'/du"), (a = 1, 2,..., m), is
of rank m at each point of the neighborhood.
In Secs. 64 to 73 we have studied two-dimensional Riemannian varieties
(surfaces) imbedded in £3. A question naturally arises: Under what
circumstances an m-dimensional variety R,,,m? with a Riemannian metric
(75.6) ds = a,du gue. (7,6 = 1, 2...., 7%),
can be imbedded in the n-dimensional Euclidean manifold with

(75.2) ds? = dx‘ dx‘?

Now equations 75.5 together with 75.6 and 75.7 require that
pe Ox Ox
(75.8) ne ee
eu Ou! ©
The set of 4m(m + 1) partial differential equations 75.8 in n variables ae
will not be expected to possess a solution unless n > 3m(m + 1); thus,
if m= 2,n> 3; if m=3,n> 6, and so on. This estimate, however,
does not constitute a proof that an m-dimensional variety can be imbedded
in E,, whenever n > }m(m + 1). It is possible, however, to prove that a
neighborhood of R,, can be imbedded in E,, if n > 4m(m + 1). Con-
cerning the global imbedding of the whole of R,, in E,, almost no general
results are known. There are a few special theorems on the imbedding
of two-dimensional varieties with special topological properties. These
refer almost wholly to convex two-dimensional manifolds.** The problems
on imbedding now lie at the forefront of researches in geometry.

Problem,” Comm. on Pure and Appl.

85 See L. Nirenberg, “The Weyl and Minkowski in a
of Geometry in the Large
Math., 6, 1948, and A. V. Pogorelov, Some Questions
Riemannian space, Moscow (1957).
 4,
ANALYTICAL MECHANICS

76. Basic Concepts. Kinematics

Analytical mechanics is concerned with a mathematical description of

motion of material bodies subjected to the action of forces. Its develop-
ment follows a familiar pattern. A material body is assumed to consist of
a large number of minute bits of matter connected in some way with one
another. The attention is first focused on a single particle, which is
assumed to be free of constraints, and its behavior is analyzed when it is
subjected to the action of external forces. The resulting body of knowledge
constitutes the mechanics of a particle. To pass from mechanics of a single
particle to mechanics of aggregates of particles composing a material body,
we introduce the principle of superposition of effects and make specific
assumptions concerning the nature of constraining forces, depending on
whether the body under consideration is rigid, elastic, plastic, fluid, and so
on.
We begin our study of mechanics of continua by analyzing the motion of
a single particle. The particle is assumed to be an idealized entity having
position and inertia, but no spatial extension. The measure of inertia is
mass, and thus the particle is simply a point-mass. Another basic ingredient
of mechanics is the concept of time, which arises in the assumption of
causal connection between physical events. The hypothesis of causality
implies the possibility of ordering events, and the time f¢, as it appears in the
description of the physical universe, is an independent parameter whose
range of variation is the real-number continuum.
We will suppose that physical events take place in the three-dimensional
space whose metric is Euclidean, and we refer the position of a particle
at a given time ¢ to some curvilinear reference frame X. As in the study
of geometry in Chapter 3, we denote the coordinates of the particle relative
to a set of orthogonal cartesian axes by the symbols y’. Clearly, the
position of a particle is a relative concept depending on the selection of a
reference frame. The reference system generally used in astronomy is that
206
SEC. 77] NEWTONIAN LAWS. DYNAMICS 207
determined by the so-called fixed stars. It is termed the primary inertial
system. Any system of axes moving relative to the primary inertial system
with constant translational velocity is called a secondary inertial system. In
many mechanical problems the motion of the earth relative to the primary
inertial system is so nearly negligible that the Newtonian laws (Sec. 77),
which are assumed to be valid only in the inertial reference frames, can be
applied without modification to study the motion
of particles referred to a system of axes fixed in
the earth. P, = a
When a particle changes its position ina given
reference frame, it is said to undergo a displace-
ment. Thus, suppose that the particle is at the 4 A
point P, at time ¢. Its position at this time is
given by the vectorr,; ata later instant of time
t+ At it is at P,, determined by the position
vector r,. We denote the displacement |in the
interval of time Ar by the vector P,P, = Ar
(Fig. 33) and suppose that the particle traverses
a continuous path which is represented by the Fig. 33
vector sum of the elementary displacements dr.
We define the average velocity of the particle during the displacement Ar
by the formula v,,. = Ar/Az, and we assume that this ratio has a unique
limit as At > 0. Then the instantaneous velocity v is given by the formula
dr/dt =t =v. Velocity v is, of course, a vector.
The case in which dr/dt is constant is of relatively minor interest in
mechanics, and generally we will be concerned with accelerated motions.
We define the average acceleration of the particle, during the time interval
At, by the formula a,,, = Av/Az, and the instantaneous acceleration by

Hereafter, unless otherwise specified, the words velocity and acceleration

are taken to mean the instantaneous values.
of
The velocity and acceleration are known as the kinematical concepts
those concepts that utilize the idea
mechanics to distinguish them from
of force. We consider this idea in the following section.

77. Newtonian Laws. Dynamics

laws, the first of
In 1687, Sir Isaac Newton published three axioms or
experiments
which was based on deductions from a set of remarkable
on bodies moving on inclined planes,
performed by Galileo (1564-1642)
208 ANALYTICAL MECHANICS [CHAP. 4

and the other two represent a profound crystallization of the notions

surrounding these experiments. These laws form the point of departure
in all considerations in dynamics, and we give them here in a form that is
almost a literal translation of Newton’s Latin as it appears in the 1726
edition of the Philosophia Naturalis Principia Mathematica. The present-
day formulation of analytical mechanics is essentially due to J. L. Lagrange
(1736-1813), whose greatest work, Mécanique analytique, was written in
1788, and W. R. Hamilton (1805-1865), whose celebrated principle em-
braces the whole of mechanics.
NEWTONIAN Laws. I. Every body continues in its state of rest, or of
uniform motion in a straight line, except insofar as it is compelled by im-
pressed forces to change that state.
II. The change of motion is proportional to the impressed motive force,
and takes place in the direction of the straight line in which that force is
impressed.
“III. To every action there is always an equal and contrary reaction; or
the mutual actions of two bodies are always equal and oppositely directed
along the same straight line.
The first law depends for its meaning on the dynamical concept of force
and on the kinematical idea of uniform rectilinear motion. It ascribes
anthropomorphic attributes to a particle, which is bent on continuing its
motion in a straight line but is somehow deflected from its intentions by a
push or pull. Newton doubtless felt that the idea of force is intuitively
known and requires no further explanation. We shall presently see that
the first law is in reality a corollary of the second.
The second law of motion also introduces the kinematical concept of
motion and the dynamical idea of force. To understand its meaning it
should be noted that Newton uses the term motion in the sense of momen-
tum, that is, the product of mass by velocity. Thus the “change of motion”
means the time rate of change of momentum, and hence in vector notation
the second law can be stated as a formula

(77.1) Fi ay
dt
provided that our units are so chosen as to make the proportionality
constant equal to one.
If we postulate the invariance of mass, then equation 77.1 can be
written in the familiar form
(712) F = ma.
We note from (77.1) that, if F = 0, then d(mv)/dt = 0, so that my =
constant, and hence v is a constant vector. Thus the first law is a con-
sequence of the second.
SEC. 78] EQUATIONS OF MOTION OF A PARTICLE 209
The concept of mass can obviously be defined with the aid of the second
law in terms of force and acceleration. There were numerous attempts to
define mass and force independently of one another. The most familiar
of such definitions is due to Ernst Mach,! who formulated a definition of
mass with the aid of Newton’s third law of motion. In our opinion a fine-
grained analysis of Mach’s definition of mass reveals certain logical
difficulties which cannot be resolved by appealing to the third law alone.
For this reason it seems best to leave one of the fundamental building
blocks of mechanics (mass or force) undefined and admit it in the science
of mechanics on the same basis as the “God-given integers” in mathe-
matics.
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 oppositely directed force on some other body.
Newton called the two aspects of the force action and reaction, whence the
usual statements of the law.
The entity of mass entering in the formulation of Newtonian laws is
sometimes called the inertial mass (or simply inertia) to distinguish it from
the gravitational mass M entering in 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 r between them, and is directed along the line
joining the particles. In symbols,
(77.3) F=k )
r

M, to
where k is a universal constant and r is a vector directed from mass
mass M,.
If it is assumed (as it is usually done) that the gravitational and inertial
ng
masses are equal, the law 77.3 furnishes a practical means for compari
masses with the aid of beam balances.
ng of
In order to develop the science of mechanics of a universe consisti
an laws the
more than two particles, it is necessary to adjoin to Newtoni
ions re-
principle of superposition of effects and make further assumpt
garding the nature of constraints.

78. Equations of Motion of a Particle. Work. Energy

by a vector Tr.
Let the position of a moving particle P be determined
of r are denoted by
If the curvilinear coordinates of the terminal point
survey of it is contained in
1B. Mach, The Science of Mechanics. An interesting
R. B. Lindsay and H. Margen au, Foundat ions of Physics.
210 ANALYTICAL MECHANICS [CHAP. 4

a(t), then the equations of the path C of the particle can be written in the
form
(78.1) Cr at = (5),

and we call the curve C the trajectory of the particle.

The velocity of P is a vector v == dr/dt, whose components are

182
(78.2) eea
v= 2

The acceleration a = dv/dt = d’r/dt®? has the components (see Secs. 46

and 47)

(78.3) ai = Out Ill_ d’x' [idee

ét dt? \jk) dt dt’
where 6v’/6t is the intrinsic derivative and the |J are the Christoffel
J
symbols calculated from the metric tensor g;,, associated with the reference
system X.
If the mass of P ism, Newton’s second law of motion yields the equation
F = md’r/dt?, or

(78.4) F'= m— = ma.

In orthogonal cartesian coordinates, equation 78.4 assumes the familiar

form F* = m d?y'/dt?.
We introduce next the concept of energy, which will permit us to give
a more elegant formulation of the theory. The germ of the energy concept
can be traced back at least to Galileo, who remarked, “What is gained in
power is lost in speed,” but the first clear introduction of the idea of energy
in mechanics as a quantity equal to the product of mass and the square of
velocity of the particle (vis viva) was made by Huygens in the seventeenth
century. The full use of this idea, however, and of its relation to the con-
cept of work, did not come until the nineteenth century.
We define the element of work done by the force F in producing a dis-
placement dr by the invariant dW = F - dr, and, since the components of
F and dr are, respectively, F’ and dz‘, this scalar product is equal to

(78.5) 4 Ook a
= F; dx’,
where the F; = g,,F’ are the covariant components of the vector F. We
shall suppose that, in general, the functions F'(x), defining the vector field
Sec. 78] EQUATIONS OF MOTION OF A PARTICLE 211
F, belong to class C1. The work done in displacing a particle along the
trajectory C, joining a pair of points P, and Pg, is the line integral
P2
(78.6) W=| F,dz’. :
Hea

Making use of Newton’s second law of motion 78.4, we can write 78.6
in the form
Pe i
w=| poy dl
Pi ot
(78.7) te ieee
=| ee v’ dt.
t ot

But
O(g;v'v’) , ou
Si Fieet 2
and, since g,,v'v’ is an invariant,

6(g;;v'v’) = d inj
ayes = iF (g,;0'v’),

and hence
d ee Oe
FACou, ) == 2 283;Tian
mea ron ds

Inserting from this result in the integrand of (78.7) yields

w={ m © (g.viv’) dt
te d Bade

78.8
hE, ty 2 at

sli
ane Py

=i 7, — 13,

where
2
m °% v
T=—¢g,vv’
y o4 =
; —
> .

ing the
We have the result that the work done by the force F; in displac
differen ce of the
particle from the point P, to the point P, is equal to the
the beginni ng of
values of the quantity T= jmv* at the end and at
is exactly one-
the displacement. We define the quantity T = 4mv*, which
particle.
half of the vis viva of Huygens, as the kinetic energy of the
be enunci ated as a
The statement embodied in the formula 78.8 can
The work done in displacing a particle along its trajectory
TueoreM.
e.
is equal to the change in the kinetic energy of the particl
212 ANALYTICAL MECHANICS [CHAP. 4

It may happen that the force field F; is such that the integral 78.6 is
independent of the path. In this event the integrand F, dx’ is an exact
differential,
(78.9) dW = F, dz’,
of the work function W. The negative of the work function W is called the
force potential or potential energy. We denote the potential energy by the
symbol V, and conclude from (78.9) that

(78.10) Ep
The fields of force for which potential functions exist are called conservative.
There is a simple criterion for a field of force F; to be conservative. We
state it asa
‘THEOREM. A necessary and sufficient condition that a force field F,,
defined in a simply connected region, be conservative is that F; ;= F; ;.
The proof of this theorem follows immediately from the observation
that a necessary and sufficient condition for the expression F; dx‘ to be an
exact differential of a single-valued function V is that

(78.11) a
On gx
since these derivatives are assumed to be continuous functions.2. But

Pye {lF
Ox? ij
k\ . ies
and, since biis symmetric in 7 and 7, we conclude that the condition
78.11 is completely equivalent to the one stated in the theorem.
As a corollary we observe that a parallel force field (Sec. 48) is necessarily
conservative, since the condition for a vector field F; to be parallel is
F; ; => 0.

79. Lagrangean Equations of Motion

An alternative formulation of the Newtonian law 78.4, phrased in terms

of the kinetic energy of the particle, was obtained by Lagrange from the
principle discussed in Sec. 84. We derive these equations in this section by
a direct calculation which makes use of Newton’s second law of motion.
eit the Tegion is multiply connected, the conditions 78.11 still guarantee the existence
of potential V related to F; by formula 78.10, but, in this case, the function V, in
general, is multiple valued.
Sec. 79] LAGRANGEAN EQUATIONS OF MOTION 213

The kinetic energy T = $mv? can be written as

(79.1) T= m
=
railed
eat
Bit hs
2
If we differentiate (79.1) with respect to z we obtain
since #? = v'.
OT/dx' = mg;,;%. The derivative of this expression with respect to f is

) — meet + got)
4 O8s5 oe oi
sige
d (oT) _ <j

respect to a’, namely,

If we subtract from this the derivative of (79.1) with

wi + 1 (28 O8ix 28a) aa| |

=
4(oT) _<t= m||
g,,0° aa Dad nas fina

= m{g,,4' + Lik, ee}

mga( 2+ | |e).
jk
on the right is the acceleration
But by (78.3) the expression in parentheses
write
a', and, since mg,;,a' = ma; = F,, we can
d (=) oT Fi
acegers ee
79.2
ee dt \dx* Ox"
second law in the form used
Equations 79.2 give the statement of Newton’s
by Lagrange.
For a conservative system, F; = —dV/dx', and equations 79.2 becomes

d ( =) oT OV
79.3 = || a
dt \oz* Ox Ox'
( )
or

(79.4) 4 (2) bgt ta l spay

dt \da* Ox’
V is a function of the coordinates
We recall that the potential energy
Lagrangean function
x’ alone; hence, if we introduce the
La2T—/J,

the form
we can write equation 79.4 in
Noman
CayLue(LS) et
eae All aa!
214 ANALYTICAL MECHANICS [CHAP. 4

In the application of Lagrangean equations to specific problems one

frequently deals with the physical components F” of the force vector F
instead of the tensor components F’. The physical components of F, we
recall, are the coefficients in the representation

F = F'e,,
where the e,’s are unit vectors codirectional with the base vectors a,.
(See Sec. 45.) Since F = F’a, and a,-a; = g;;, the physical components
F are related to the tensor components F’ by the formula
Fi=vV¢g,,F', (no sum).
Problems

1. Show that the covariant components of the acceleration vector in a spherical

coordinate system with ds? = (dx1)? + (x! dx)? + (21)? sin? x°(dx*)? are
C= Bae ae sin 2)",
d
Cees i [(x!)2a?] — (v1)? sin 2 cos x? (#3)?,

d
(Bi 7 [(e" sin aye].

Deduce these expressions from formula 78.3 and also from Lagrangean equations

We
mn
Hint: F; = ma, and T = *(5) Bicep
mids mgis .,.,x),
2. Use Lagrangean equations to show that, if a particle is not subjected to the
action of forces, then its trajectory is given by y* = a‘t + b*, where the a* and
b? are constants and the y* are orthogonal cartesian coordinates.
3. Find, with the aid of Lagrangean equations, the trajectory of a particle
moving in a uniform gravitational field. Hint: T = }my'y' and V = mgy’,
where 7? is normal to the plane of the earth.
4. Deduce from Newtonian equations the equation of energy, T+ V =A,
where h is a constant. Hint: Show that d7/dt = ma,v’ = —dV/dt.
5. Prove that, if a particle moves so that its velocity is constant in magnitude,
then its acceleration vector is either orthogonal to the velocity vector, or it is
zero. Hint: Compute the intrinsic derivatives of v? = g;,v'v’.
: d oT oT
6. We have shown in Sec. 79 that —— — — is a covariant vector F;
dt ox* Ox?
whenever T(x, @) is an invariant defined by (79.1). Prove more generally that if
Seger : . : d aw ow
W(a, ©) is an invariant, then both @W/ dé? and — —— — —— are covariant
dt oa Ox?
vectors. Hint: Let a = a*(q',q?,q°) be an admissible transformation of
coordinates. Compute #', show that 03*/dg? = axt/ dq’, and observe that
the invariance of W(x, #) requires that W(x, ¢) = W[x(q), #(q)] = W(q, 9).
Sec. 80] APPLICATIONS OF LAGRANGEAN EQUATIONS 215

80. Applications of Lagrangean Equations

As an illustration of the application of Lagrangean equations to the

determination of trajectories, we consider several examples, which include
the important cases of particles moving on smooth curves and surfaces.
1. Free-Moving Particle. If a particle is not subjected to the action of
forces, the right-hand member of equation 79.2 vanishes, and we have

80.1 eee an
d aT| OT
( ) aS Ox’

If the coordinates x’ are chosen to be rectangular cartesian, then T =

(m|2)y'y', and hence equation 80.1 yields my’ = 0. Integration of this
equation gives y’ = a't + b', which represents a straight line.
2. Constant Gravitational Field. Again we choose a cartesian reference
frame and take the Y-axis to be normal to the plane of the earth. The
potential V of the constant gravitational field is V = mgy?, if the positive
Y3-axis is directed upward. In this case equations 79.2 give
Oh = 0, 7 = 0, de =

so that the trajectory is determined by

oi me At 1 D", tp ==1 1,2),
y> = —hgt?+ at +b.
Y*-axis.
Thus the trajectory is a parabola whose axis is parallel to the
Curve. Leta particle be constra ined to
3. Motion of a Particle on a
move on a curve C whose equatio ns are

(80.2) corso 8), Gi lin2,3),

s being the arc parameter. We shall suppose that C has a continuously

turning tangent, so that the «‘(s) are of class C®.
particle are
The components v' of the velocity vector v of the

(80.3) va =H H= ol,

C and v = ds/dt is the

where A’ = dr'/ds is the unit tangent vector to
magnitude of v.
a are determined by
The components a‘ of the acceleration vector
(80.3) with respec t to /
computing the intrinsic derivative of
, GRD)
pias AD 6A
(80.4) Ah Gee amar
ge
216 ANALYTICAL MECHANICS [CHAP. 4

where we wrote 6v/dt = dv/dt, since v is a scalar. However,

dA’ _ 6d’ ds i i
(80.5) ohRen

[50.1] — = «"', x > 0,

defining the curvature x and the principal normal unit vector p.

On substituting from (80.5) in (80.4), we get

(80.6) ON te oad + xvu',

which states that the acceleration vector a lies in the osculating plane of the
curve. Moreover, the component of a in the tangential direction is equal
to the time rate of change of speed v, whereas the component in the
direction of the principal normal is v?/R, where R = 1/x is the radius of
curvature of C.
The force F = ma acting on a particle of mass m moving along C is
determined by
(80.7) F‘'=m a + mxvu'.
It should be observed that F" is the resultant of all external forces that act
on the particle and thus F includes the reaction R of the curve on the
particle. 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 reaction R in the general case. In mechanics the curve C is
said to be smooth if the reaction R is normal* to C, that is, if R‘A; = 0:
If R = 0 the curve C is called the natural trajectory of the particle.
As an illustration, let a bead of mass m slide under gravity along a
smooth curve C lying in the vertical Y1 Y-plane (Fig. 34).
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 « is
the angle between the direction of R and the positive Y?-axis, the com-
ponents of F in the directions of the tangent A and the principal normal
ware
Fy = —megsina, Fi, = —mgcos«+ R.
* This is equivalent to saying that the frictional force is zero. The term “smooth”
employed in mechanics is different from the term “smooth” used in geometry, where a
“smooth curve” is one with a continuously turning tangent.
Sec. 80] APPLICATIONS OF LAGRANGEAN EQUATIONS 217

>y!

Fig. 34

On referring to (80.7), we conclude that

(80.8) m . = —mgsing, mxuv? = —mgcosa + R.

But cos « = dy'/ds, sin « = dy?/ds, and dv/dt = (dv|ds)(ds/dt). Accord-

ingly, the first of equations 80.8 yields
d dy’
mv — = —mg—,
S ds
so that

(80.9) dmv? = —mgy? + constant.

path is zero,
Since in this case the component of R in the direction of the
from the energy equation
we could have written equations 80.9 directly
T + V = constant.
n of y®. The
Equation 80.9 determines the speed v along C as a functio
R as a function of the
second equation in (80.8) then serves to determine
normal to C and
curvature x. If the curve is rough, R is no longer
the angle « depends on the coefficient of friction .
move under gravity
As a concrete example, let a particle of mass m
along a smooth cycloid
y = a(O — sin 8)
(2) y? = a(1 + cos 9), Nes Se
218 ANALYTICAL MECHANICS [CHaAP. 4
shown in Fig. 35. Then the first of equations 80.8 yields
2 dy”

(b) mo = —me
where

s = | Vary + (dy)? = a V20 — cos 6) dé

6
= 2a |enon 4a(1 — cos).
0 2 2

Since cos? (6/2) = 4(1 + cos 6) we deduce, on noting the second of

equations (a), that
vam (s — 4a)? .
8a
Accordingly, (5) yields the equation

s+s=zg,
4a A
the general solution of which is

(c) S = ¢, COS (/g/4a t + cs) + 4a.

The integration constants c, and c, are determined by the initial position

and initial velocity of m on the cycloid.
It is clear from (c) that the period of motion is bidepenceen of the
amplitude c, and is equal to 2n/V g/4a.g/4a This fact was discovered by
Christian Huygens, about 300 years ago. Huygens proposed the use of
cycloidal pendulum in the construction of isochronous clocks, Calcu-
lations, making use of the second equation 80.8 show that R = 2 mg cos a.

Fig. 35
Sec. 80] APPLICATIONS OF LAGRANGEAN EQUATIONS 219
Problems
1. Deduce the differential equations for a simple pendulum of length 1, and
show that for small oscillations the period is 27/ V g/l.
2. Derive the equations of motion for a particle moving under gravity on a
smooth helix:
=) @ COS.0) y* = asin 6, OS 3 Vide).

Note that since the helix is smooth, the reaction R is normal to the helix and
hence the component of the resultant force F in the tangential direction is equal
to the component of the gravitational force mg in that direction. The latter
component can be computed from gravitational potential V = mgy®. Further-
more, by (80.9) the energy equation in this case yields dm(v? — v9”) =
mg(y® —"").
3. If a particle of mass m moves on a smooth parabola with its axis vertical
and concavity downwards, show that the reaction R = xm(v2 — v’), where v’
is the velocity of m for which this parabola is the natural trajectory and v is the
velocity in constrained motion.
4. Motion of a Particle on a Surface. Let the equations of a regular
surface S be given in a parametric form as

(80.10) S: xt=2(,2), (=1,2, 3),

and let a particle of mass m be constrained to move on S under the action
on
of the force F. The force F is the resultant of all external forces acting
the particle and thus includes the reaction R of the surface on the particle.
that
When the surface is smooth, R is normal to S and represents pressure
constrains the particle to remain on S.
are
The space components v' of the velocity vector v of the particle
related to the surface components v* by the formula ’

us on au F
p= = = 2 a a’, (a = 1,2),
din) Oulasas,
or

(80.11) p= 20%,
where v% = ui".
yields
The acceleration a’ = 6v‘/6¢; hence equation 80.11
pene ot a OD a uae
mor ot
e the base vectors a
4 See equation 64.5. The reader should take care not to confus
nents a* used in this section.
used in Chapter 3 with the acceleration compo
220 ANALYTICAL MECHANICS [CHAP. 4

or
(80.12) a’ = x,'a* + ai gv7v",
where a® = 6v*/dt.
If we make use of the Gauss formula

[67.7] r= bane
equation 80.12 reads

(80.13) a’ = x,'a* + b,gvvPn':

Thus
Oa tb Ar
and, since the normal curvature ,,) = b,,4°A", we have

a’ = 2,'a* +. v°%,)n'.
Since F* = ma’, we have

(80.14) F* = mz,'a* + mvx;,,)n".

The first term in the right-hand member of (80.14) is the component of
F in the tangent plane to S, whereas the second is the component of F
along the normal n. For, the component of F in the direction of the
normal n is

(80.15) F'n, = mz,'n,a* + mv*x,,,)n'Nn,,

2
=> 0 oa mv X(n)>

since the surface vectors x,’ are orthogonal to n; and n'n; = 1. The com-
ponents of F in the plane tangent to S, on the other hand, are given by

8450,’ ipiF* _= MY; ;%,'i,yAia + MV 2 Xn) 8i5L,'N

jpt
faxick a
= ma,,a* + 0,

since g,,v,'x,' = a,, by (64.6), and g,,x,Jn' = 0, because the surface

vectors x,’ are orthogonal to n,. If we rewrite this relation as

a, /F; = ma,,
and set F,, = x,’F;, we obtain a pair of Newtonian equations

(80.16) F, = ma,,

relating the surface force vector F,, to the surface acceleration vector a,
Sec. 80] APPLICATIONS OF LAGRANGEAN EQUATIONS 221

Equations 80.16 can be recast into equivalent Lagrangean form by

noting that the kinetic energy T = }mv” is

m m sakes
T= —a,,0°v" = — ORO
2 2
We obtain, as in Sec. 79,

(80.17) 4 (7) Sylabie 1a

dt \du",ou"

where F, is defined by (80.16). When the force field is conservative,

F, = —0V/du«, where V is the potential.
tion
We can deduce the equation analogous to (80.6) for the accelera
of the particle moving on S. The velocity v* of the
along the trajectory
particle, along the trajecto ry, is v* = vi, hence

_ aot dodt ye yoot

= dv a. + 2 OA*

dt bs
If we recall that

[71.6] OER
és
the tangent plane, and x,
where 77% is the unit normal to the trajectory in
is the geodesic curvature, we can write
fg ==, dv Ag + 2
vn a
We a")

so that

2 ds
that
If follows from this result (cf. equation 80.7)

Fe = ae + 2Tx,7",
ds
hes identically, then dT/ds = 0
where T = mv?/2. If the vector F% vanis
first of these equations states that
and x, = 0 along the trajectory. The
ctory is a geodesic by the theorem
= constant, and, ifv # 0, then the traje
of Sec. 63.
222 ANALYTICAL MECHANICS [CHAP. 4
y3

Fig. 36

As an illustration of the use of equations 80.17, we consider a particle of

mass m constrained to move under gravity on a smooth paraboloid of
revolution (Fig. 36),

(80.18) y= fe[(y')? + (y?)"], @ = constant.

If we introduce cylindrical coordinates (r, @, z) by setting

y' = rcos 8, a" = 7 sin U, ff oe

equation 80.18 becomes

(80.19) o=

and the kinetic energy T = 4my‘y’ takes the form

rao [(t+ gaye tre]

m r? +2 ;

The potential energy of the gravitational field is V = mgy?, which, on

noting (80.19), takes the form V = mgr?/4a. Since the surface is smooth,
the reaction R is normal to S and we can use equations 80.17 with F, =
—dV/du*, since the components of R in the tangent plane to S are zero.
Sec. 80] APPLICATIONS OF LAGRANGEAN EQUATIONS 223
We parametrize the surface by setting vu! = r,u® = 6, substitute in
(80.17) for T and F, = —dV/du* and obtain two equations

ee EO wae ns gr
i+; Sasiyerin GLE wera
(80.20) :
d 2h

77
—(r°0) ) = 0.

The second of equations 80.20 gives on integration the equation of

angular momentum

(80.21) ro = h, h-= constant.

The elimination of 6 from the first of equations 80.20 with the aid of
(80.21) gives

(80.22) (1+ 5)r+-5=--£,

2 2

4 a

which has a unique solution when the initial position r = 1 and the
initial velocity 7 = vy of the particle are specified.
If our particle is constrained to move ona horizontal circle r = constant,
6? =
(80.22) requires that h? = gr*/2a, and equation 80.21 then shows that
g/2a, so that the angular velocity 6 is independent of the radius of the
circle. When the path of the particle is the meridional line 6 = constant,
we get from 80.20 the equation

n R
The integration of equation 80.22 and the calculation of the reactio
to the parabol oid is tedious. To com-
required to constrain the motion
in which F =
pute the magnitude of reaction, we need equation 80.15,
mg +R.
illustration by the
If we replace the surface of the paraboloid in this
the proble m of a spheri cal pendulum.
surface of the sphere, we have
for the spheri cal pendu lum can be
The solution of equations of motion
functi ons. When the surface is a cylinder
obtained with the aid of elliptic
easy.°
r =a, the integration of equations of motion is
in E. T. Whittaker’s Analytical
5 Interested readers are referred to pp. 99-109
Press (1917), where the motion of the particle on a surface is
Mechanics, Cambridge
analyzed in a different way.
224 ANALYTICAL MECHANICS [CHAP. 4
Problem

Let a particle of mass m be constrained to move on the surface of a sphere of

radius a. Relate the orthogonal cartesian coordinates y* to the surface coordi-
nates u* by the formulas
y’ =asinu' cos uv,
y2 = asinu' sin uv’,
y® = acos uw.

Show that equations 80.17 yield

be ees
i} — (v’)* sin u cos ul = —5,
ma

ne A “4° A F.2
i sin? 2 + 2i4u? sin 2 cos ul = —> .
ma?

Solve these equations for the case when F* = 0, and show that the trajectory is
an arc of a great circle, and the speed v = const.
Hint: The first integral of the second equation is u? sin® u} = constant. Use
this result in the first equation and observe that v® = a®{(w)? + (@*)? sin® uv’).

81. The Symbol of Variation

In this section we recall the definition of the variational symbol 6, first

introduced in Sec. 56, and record several of its properties. The notation
introduced here permits one to give a concise formulation of Hamilton’s
principle and Lagrange’s principle of least action. Either of these principles
(rather than the Newtonian laws) can serve as a starting point in the
development of analytical dynamics.
Let F(x', 2,...,%") be a function of m independent variables 2° of
class C? in some region R of an n-dimensional manifold. We shall be
concerned with the behavior of the function F in a certain neighborhood
of the curve C, defined by the parametric equations

C: x = x(t), iS tsk,

where we assume that the 2‘(r) are of class C?.

Consider an A-neighborhood of the curve C, defined by the inequalities

e—h<x <a2th, (ay Carers 3 F

where / is a small positive number, and the 2’ are the coordinates of a

point on C. We introduce a class of functions
C’: Z(t, e) = x(t) + €&(r), (i = etoen))
Sec. 81] THE SYMBOL OF VARIATION iis
where —1<e<1 and the £(¢) are single-valued functions of class C?
m 4 < f< t,, such that
&(t,) = &(4.) = 0

and |é‘(t)| < A, uniformly in 4) < t < fy.

A set of n functions z‘(t, €) constitutes a varied path, and it is clear that
the curves C’ so defined can be made to belong to the h-neighborhood of
C. In the space of two dimensions the curves C’ all lie in a band of width
2h about the curve C and coincide with C at the end points of the interval
(t, ta).
The variation dz‘ was defined in Sec. 56 by the formula

(81.1) 6a = oe -e = Ete,
de e=0

and the variation 6F of the function F(z!,..., %”) is

i = (=) :
de 0

where
(=) 0) ee ee a OF i
de /o de sp MOD at|
Thus

(81.2) OF = we ba,
Ox"

Consider next the function #(t) = dx'/dt. We form

z(t, <-) = #() + €'(),

and conclude from definition 81.1 that

64° = (9) =«f(t)h= ulbx.

de 0 dt
Hence

(81.3) Ea ee
dt dt

the variation.
so that the variation of the derivative is the derivative of
, 2%", t) of 2n + 1
Clearly, if we have a function F(a!,..., 2", #,...
C2, we can write
variables x‘, #¢ = dx‘/dt, and t, which is of class
OF OF
F OF = — 6x" — dz.
oy Ox" + 5g i
226 ANALYTICAL MECHANICS [CHAP. 4

A simple calculation, analogous to that used in deducing formula 81.3, _

leads to the conclusion that
dred
81.5 O— = — OF,
Pay Gime at
and one can readily show with the aid of (81.4) that
O(F + O) = 6F+ 69,
O(FO) = FOO + DOF,
where F and ® are any functions satisfying the conditions laid down
above, and the variational symbol 6 refers to the same varied path C’.
In Sec. 57 we considered the functional
t
Pal Feet BO Pt
ti

where the functional arguments 2'(t), t, < t < t,, belonged to the h-
neighborhood of an extremal of J. That is, we considered the behavior of
the integralJ along the varied paths Z(t, ©) = 2° + e&‘(t). Making use of
equation 81.4 of this section and referring to formula 57.6, we see that
formula 57.6 can be written
te / \

nee } (= jens 53") dt,

t1 \Ox* OG aed
so that, for a pair offixed limits t, and fg,
ty te
or =| oF dt = 6| F dt.
ty J ty
When stated in words the foregoing equation reads: The variation of the
integral with fixed limits is equal to the integral of the variation of the
integrand.
We shall make use of the symbolism introduced in this section to
formulate Hamilton’s principle.

82. Hamilton’s Principle

Consider a particle of mass m moving in a three-dimensional Euclidean

manifold, referred to a curvilinear system of coordinates Y. The particle
is in motion under the influence of force F, and our problem is to determine
the trajectory
Giiteheaa (7). (= 1).2;3); a
where ¢ denotes the time.
Sec. 82] HAMILTON’S PRINCIPLE 227
The kinetic energy T of the particle (which has a physical meaning only
along the trajectory C) is given by the formula T= 3mg;,;a'%’. If we
define a family of varied paths
«“

Ce get) = w(t) + x(t),

with dx‘(t)= «€(t) and &(t,)= &(t.) = 0, belonging to the A-neighbor-

hood of C, we can speak of the variation of 7, namely,

OT Ole.
(82.1) OT = — 6a’ + — 62’,
Ox" Ox"

and. we can phrase Hamilton’s principle as follows:

HAMILTON’S PRINCIPLE. Jf a particle is at the point P, at the time t,
and at the point P, at the time t,, then the motion of the particle takes place
in such a way that
te
(82.2) { (6T+ F, 62') dt = 0,
1

where «' = x'(t) are the coordinates of the particle along the trajectory and
a’ + 6x? are the coordinates along a varied path beginning at P, at time t,
and ending at Py, at time tg.
It will be shown next that this principle is equivalent to Lagrangean
is
equations of motion 79.2, and hence to Newtonian laws. The proof
simple. Substituting (82.1) in (82.2) yields

(82.3) |PAY (= NE oe oz!)Amit

- .
te .

fe Ox"
parts,
Integrating the first term under the integral sign of (82.3) by
t

Aaiap
e
}t TE eeaa!
te

4, “fs
dtp at Oe
and &(t,) vanishing,
and, since x(t.) = dx'(t,) = 0 by virtue of & (ty)
equation 82.3 becomes

-420)da‘ dt=0.
te
(82.4) i}fat
Ox’ = at 0a"
ty

ent used in Sec. 57

Since this integral vanishes for arbitrary dx’, the argum
shows that
(82.5) ee ee Pee Gea", 2, 3):
228 ANALYTICAL MECHANICS [CHAP. 4

Conversely, if Lagrangean equations 82.5 hold, then equation 82.4, and

hence equation 82.2, is valid.
In the foregoing formulation of Hamilton’s principle no reference is
made to the nature of the force field F;. If, in particular, this field is
conservative, then there exists a potential function V(z', x?, x) such that
oV/dx' = —F;. In this case equation 82.2 reads
to
{ (or- oY x‘)dt = 0,
ty Ox’

and, since dV = (0V/0x') dx’, we have

(82.6) |CAT nao 0

t

But in Sec. 79 we defined the Lagrangean function L = T — FV, so that

ty
equation 82.6 can be written as| dL dt = 0, and, since the limits of
t Zz

integration are fixed, we have a concise formulation of Hamilton’s principle

for a conservative field in the form
te
(82.7) ) L dt=0.
t1

We can state equation 82.7, in words, as follows: In a conservative field of

hn :
force a particle moves so that the integral i L dt, evaluated along the
z ° ty
trajectory x' = x'(t), ty < t < ty, has a Stationary value in comparison with
its values for all neighboring paths beginning at the point P, at t = t, and
ending at point P, at t = tp.
Equations of motion in form 79.5, namely,

“() -4 =
dt\axz!) axt
follow at once from the formulation 82.7.

83. Integral of Energy

We establish in this section an important general

THEOREM. The motion of a particle in a conservative field of force is
such that the sum of its kinetic and potential energies is a constant.
The proof of this theorem follows from an identity which will be
established next.
SEC. 84] PRINCIPLE OF LEAST ACTION 229
Since the kinetic energy T = dmg,,c‘# is an invariant,

OME sali >|

> (g,;4°a’)
dt ot ot
Mg, (Oe 4 222)
m
2
62
(Ot
6x?
Ot
= 0D",
so that

(83.1) ola
dt
es
where v’ is the velocity and a, is the acceleration of the particle.
For a conservative field of force, ma; = F; = —0V/dx’, and we can
write (83.1) as
Tai OV dx’
dt — Ox dt’
or

(83.2) arT__ ave

dt dt

Integrating (83.2) yields the result

(eon i — 7h

where / is a constant of integration.

84. Principle of Least Action

The history of science abounds in attempts to imbed the laws of nature

in the structure of theology. Several of these, based on the minimal
concepts, such as Heron’s (100 B.C.) doctrine of the shortest path and
Fermat’s (1601-1665) principle of least time, had an innate esthetic appeal
to mathematicians. The most celebrated of such attempts, in the domain
of mechanics, is the doctrine of least action propounded by P. M. L.
Maupertuis circa 1740. Maupertuis asserted that all activities of nature
are performed with the least possible expenditure of “action,” which he
defined as the product of mass, velocity, and distance. In order to fit his
principle to the known results of mechanics, Maupertuis was obliged to
alter the definitions of the quantities entering in the product mvs so as to
suit each problem under consideration. Thus, in the anlaysis of inelastic
230 ANALYTICAL MECHANICS [CHAP. 4

collision of two particles of masses m, amd m,, moving with velocities v,

and v,, he minimized the product mvs, where s was the distance per unit
time. This made the ‘“‘action’”’ proportional to the kinetic energy. Mau-
pertuis obtained the known correct expression for the final common
velocity, v = (mv, + mv2)/(m, + mz). On the other hand, in the problem
of refraction of light passing from one optical medium to another he used
the actual distance s and got the constant (but incorrect) value for the ratio
of the sines of the angles of the incident and refracted rays. The doctrine
of Maupertuis, who believed that it furnished a scientific demonstration of
the existence of God, excited the imaginations of Daniel Bernoulli and
Euler and was defended by them. In 1744, Euler showed that the integral
fjmv ds has a stationary value along the trajectory of a particle moving ina
central field of force. In 1760, Lagrange extended Euler’s result by demon-
P,
strating that the integral 4 = i my + ds has a stationary value along the
Le)
trajectories of particles moving in a conservative force field, provided that
the constraints are not functions of the time. This led him to formulate the
principle of least action. This formulation still left a great deal to be
desired from the point of view of clarity, and Hamilton, in an attempt to
understand Lagrange’s formulation of the principle, deduced a broader
and different principle (1827) discussed in Sec. 82. The proof of the
Lagrangean principle, which put it on a secure basis, was supplied by
Jacobi.
Let us consider the integral of Lagrange

Py
(84.1) A= i my + ds,
Py

evaluated over the path

Cs) He SO we fate tes

where C is the trajectory of the particle of mass m moving in a conservative
field of force. We suppose that neither the kinetic energy T nor the
potential energy V is a function of time.
In curvilinear coordinates the integral 84.1 assumes the form

ibe dx dx?

t(P) iodibads
SEC. 84] PRINCIPLE OF LEAST ACTION 231
and, since
ein dx dx!
~ 959 Gr at’
we have
t( Ps»)
(84.2) A -| 27 at.
(Py)
This 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 P, and P,. Let us consider a particular set of admissible paths
C’ along which the function T + V, for each value of parameter ¢, has the
same constant value h. The functional A so determined is called the action
integral, and concerning it we can formulate
THE PRINCIPLE OF LEAST ACTION.® Of all curves C’ passing through P,
and P, in the neighborhood of the trajectory C, which are traversed at a rate
such that, for each C’, 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, this principle reads

5 i(Py) oT dt = 0,
t( P2)

(84.3)
with the auxiliary condition
(84.4) T+VvV-h=0), enC
It is important to recognize that in this instance we cannot determine the
extremals of the action integral by setting F in the Euler equations 57.7
equal to 27, because of the auxiliary condition (84.4). Since Tis a function
of the velocity v, and V is a function of position alone, the times t(P,) —
t(P,) required to traverse the varied paths C’ will differ in general. Thus
the upper limit ¢(P,) in the integral 84.4 is not fixed. In this case we have
the problem in the calculus of variations with variable end points and with
one auxiliary condition 84.4. The procedure employed in solving this
problem makes use of Lagrange’s method of multipliers for a problem
with nonholonomic constraints, which we briefly indicate. (Compare
Sec. 57.)
We construct a function F = 27 + Ad, where ¢ = T+ V —A, and
determine the solution of the system of four equations

oF_ a (28jicwo
dx’ = dt \0z"
eee OIENE
T+ V—h=0.

6 Strictly speaking this principle should be called the principle of stationary action.
PE ANALYTICAL MECHANICS [CHAP. 4

An investigation of this system shows that? A(t) = —1, and it follows

from this fact that the trajectory C is determined by the solution of the
system

(84.5) a (i = 1, 2, 3).
Gi\0u /e Ode Ox’
These are precisely the Lagrangean equations of motion.
A different and somewhat more illuminating mode of attack on this
problem is to reduce it to a consideration of the variational problem wiih
fixed end points by a change of variable. Since the kinetic energy

m dx‘ dz? m(ds\

WO hse Pm rae ar og Lot br
») dt dat 2 \dt

84.6
(84.6) pele
ae

= lanl ds.
2(h— V)
Consequently the action integral 84.2 can be written®

(84.7) vibes, i“nth = Vide

since along all admissible paths T = h — V. The integrand in the pre-
ceding integral is clearly independent of ¢t. We now parametrize our varied
paths C’, so that
Gi 5t= 20(u), p< 1S Be,
where P,: x‘(u,) and P,: x*(u,), and write

ds.= Vg, 520" 7 du,

where w”? = dx'/du.
This permits us to write the action integral 84.7 in the form

(84.8) A = {V2 — V)g,;x"'x"?

du,

and, since the limits of integration in (84.8) are fixed, we see that the
determination of the trajectory is equivalent to finding the geodesics in a
three-dimensional Riemannian manifold with the arc element
(84.9) dS? = 2m(h — V)g,, dx! dx’.
* See Sec. 88 below and O. Bolza, Vorlesungen iiber Variationsrechnung, p. 586.
* The form 84.7 of the action integral was used by Jacobi. See a discussion of this
integral and its generalizations in C. Carathéodory’s Variationsrechnung, pp. 255, 290.
SEc. 85] SYSTEMS OF PARTICLES 233
If we form Euler’s equations

Fa —_— a pert — 0,

with F = V2m(h — V)g,,x"'x", and take cognizance of equation 84.6 in

the form

we get the desired equations 84.5.

We see from formulas 84.8 and 84.9 that the action is equal numerically
to the length of the curve in a Riemannian manifold with metric coefficients

h,; = 2m(h — V)eg;;,

and that the trajectories in E; correspond to the geodesics in a Riemannian

space metrized by the formula dS? = h,, dx' dx’. This geometrization of
dynamics had a far-reaching effect on the developments in relativistic
dynamics.

85. Systems of Particles. Generalized Coordinates

We have already remarked (in Sec. 77) that the passage from mechanics
of a single particle to mechanics of material bodies can be accomplished by
introducing certain assumptions regarding the nature of constraining
forces operating on particles making up the body. In some dynamical
problems the change of shape of the body is so slight that one is justified in
supposing that the particles remain at fixed distances from one another.
This assumption leads to the dynamics of rigid bodies. If a body suffers
nonnegligible deformations we can postulate, with varying degrees of
realism, the nature of constraining forces and thus arrive at the dynamics
of elastic bodies, ideal fluids, viscoplastic media, and so on. The assump-
tions concerning the nature of constitutive forces permit us to characterize
the positions of a large number of material particles in terms of relatively
few descriptive parameters. Thus a thin rigid rod of length /, moving in
space, requires only five parameters for the determination of its position.
These can be taken as space coordinates of its center of mass and two
direction ratios of one of the ends relative to the center of mass. The
choice of descriptive parameters is not unique, and they clearly need not
have the dimensions of length. A bead sliding on a curved wire requires
only one parameter for the description of its location, say the distance from
some fixed point on the wire; a particle moving on the surface is located
unambiguously by a pair of Gaussian coordinates. Whatever is the nature
234 ANALYTICAL MECHANICS [CHAP. 4

of descriptive parameters, they will be termed the generalized coordinates.

Clearly, if the characterization of dynamical systems is to be complete, the
generalized coordinates must be functionally connected with the space
coordinates of particles making up the system. ;
Let there be N particles composing a system, and let a/,), (i = 1, 2, 3),
(a = 1,2,..., N), be the positional coordinates of these particles referred
to some convenient reference frame in E;. 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 2/,), and we suppose
that there are r such independent relations.
ea = > 3a, Seed 2 oe sl 2 S40 eo
(85.1) F(a) X(1ys Xiay 3Bays Bayo Lays = =» 3 Lewy Ln) xv)) = 0,
(=ee

If these r equations of constraints 83.1 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 g’. It is more convenient, how-
ever, to assume that each of the 3N coordinates is expressed in terms of
3N — r = n independent variables g', and write 3N equations

(85.2) tn ee A(een):
where we introduced the time parameter ¢ which may enter in the problem
explicitly if one deals with moving constraints.® If t does not enter ex-
plicitly in equations 85.2, the dynamical system is called a natural system.
We will suppose that the functions 2/,) = x!,(q, t) are of class C? in
the region of definition of the variables qg‘ and ¢ and that the Jacobian
matrix (02'/dq’) is of rank n [cf. (75.5)].
The velocities of the particles are given by differentiating equations
85.2 with respect to time. Thus

(85.3) = Otte) gi 4 O@lar

0q’ Ot
We shall call the time derivatives g' of generalized coordinates g'‘ the
generalized velocities.
Occasionally, for symmetry reasons, it is desirable to introduce a
number of superfluous coordinates q', and describe the system with the
aid of k > n coordinates g',...,q*. In this event there will exist certain
relations of the form
(85.4) SECM eran oe
* For example, a bead sliding on a wire while the wire itself is moving with specified
velocity.
Sec. 86] EQUATIONS IN GENERALIZED COORDINATES 235
so that the quantities q’, and hence g’, are no longer independent. There
will be relations among the q’s of the type

(85.5) of" if 5 eaten :

0q' Ot
when thef° are differentiable.
Since equations 85.5 were obtained by differentiating equations 85.4,
it is clear that they are integrable, so that one can deduce from them
equations 85.4, and use them to eliminate the superfluous coordinates.
In some problems, however, functional relations of the type

[B9.0)) (7 4", 5. 2, g°3 Gd Mei Hossa! Wdfinds Qan do, m);

arise which are nonintegrable, that is, it may be impossible! to deduce
from these differential equations solutions of the type 85.4. The behavior
of the system in such event cannot be described with the aid of fewer than
k coordinates, so that all k coordinates are independent. If nonintegrable
‘relations 85.6 occur in the problem, we shall say that the given system has
k — m degrees of freedom, where m is the number of independent non-
integrable relations 85.6 and k is the number of independent coordinates.
The dynamical systems involving nonintegrable relations 85.6 are called
nonholonomic to distinguish them from holonomic systems in which the
number of degrees of freedom is equal to the number of independent
generalized coordinates. In other words, a holonomic system is one in
which there are no nonintegrable relations involving the generalized
velocities.
In the following section we derive the Lagrangean equations for a
holonomic system, and in Sec. 88 we treat briefly one important class of
nonholonomic systems occurring frequently in applications.

86. Lagrangean Equations in Generalized Coordinates

For concreteness of presentation the definitions of Sec. 85 were intro-

duced with reference to systems consisting of a finite but, perhaps, large
number of particles. These definitions can be readily extended to apply
10 A billiard ball rolling and spinning on a rough table is an example of this situation.
To specify the position of the ball one needs five generalized coordinates; two of these
may locate its center, and three the angles describing the orientation of the ball relative
to the center. Since the table is rough, the ball cannot slip, so that both velocity
components of the point of contact must vanish. This gives two constraining relations
of the form (85.6), involving the velocity components. They are nonintegrable, since,
at any position of the center, the orientation of the ball can be changed without violating
the constraints.
236 ANALYTICAL MECHANICS [CHAP. 4

to continuous bodies, the points of which have coordinates x” relative

to some reference system X.
The particles of a continuous body are subjected to constraints of
various sorts, and we shall suppose throughout the remainder of this
chapter that the bodies under consideration are rigid, so that the material
points remain at invariable distances from one another. If the points of
the body are uniquely determined by a finite number of generalized co-
ordinates g’, we will write

gh Sata (Gas eae Gs t), (t= iy, a: 3);

and assume, as in Sec. 85, that the functions 2’(q, t) are of class C*. The
velocity «” of any point of the body is given by

oe0q’ da!
dt
BeOt
etches asa G=1, ae
0q’- Ot
where the g’ are generalized velocities.
Let the system in question be natural, holonomic, with n degrees of
freedom, so that the relations

(86.1) a” = x(q", .-.,9")

involve n independent parameters gq’. The velocities ” in this case are
given by [cf. (80.11)]

(86.2) meeeait (r = 1,2,3;7 = 1,2,...,n),

where the g’ transform under any admissible transformation

(86.3) G = O'(Gaes 2. Ue) (k=l, a5, 0);

in accordance with the contravariant law.
The kinetic energy of the system is given by the expression of the form

(86.4) T= i> Ma) Srs%(a)Ea)s (r, Sia ileZs 395

where mis the mass of the particle located at the point 2” and the summation
(or integration) is carried over the entire region occupied by the body.
The g,. in (86.4) are the components of the metric tensor associated with
the coordinate system X covering Es.
Sec. 86] EQUATIONS IN GENERALIZED COORDINATES 237
If we insert in (86.4) the values of #’ from (86.2), we obtain"

GLeon wa:
T= 4) mg,,— —4q‘q'
i Brea aaa
= ha;,g'q’,
where soe

qa; = Nene ek G(s = iP 2a

F 0q’ 0q’
7 = 1, eee),
Since

(86.5) T = 4a,,q°q'

is an invariant, and the quantities a,, are symmetric, we conclude that the
a;; are components of a covariant tensor of rank two with respect to a class
of admissible transformations 86.3 of generalized coordinates. We note
that, since the kinetic energy T is a positive form in the velocities q’,
|a;;| > 0, and we can construct the reciprocal tensor a’.
If we carry out a computation, in every detail identical with that of
Sec. 79, by using the expression for the kinetic energy in the form 86.5,
we obtain the formula

86.6 d(aT\ aaT ahdirs PENi a

£7)
ae decab agian Talnl?#
l
where the Christoffel symbols || are constructed from the tensor a,,.
a\]
We denote the expression appearing in the parentheses of the right-hand
member of (86.6) by

and write equation 86.6 in the form

(86.7) 4 (=) ‘% oie a,,Q'

dt \dq’ 0q’
=O), (lose
ees a0)

The expression in the left-hand member of (86.7) can also be computed

by starting with formula 86.4 and by taking cognizance of the dependence
of the variables x‘ on the parameters q'. A straight-forward but somewhat
lengthy computation making use of the formula 02”/0q’ = dx*/dq’ and the
symbol x.
1 For simplicity in writing we omit the subscripts « in terms affected by the
238 ANALYTICAL MECHANICS [CHAP. 4
Onna Ott SAF Oi) .d Ox"
relations — , following from equation 86.2,
Og! dq! i
ag’ ~ dq’ aq dt dq
leads to the result
d fol \e seu Ox"
86.8
oe
<()
dt \dq°
- aq’ =2
ma,—og’
in which a; = g,,a’ is the acceleration of the point P(z).
On the other hand, Newton’s second law gives
(86.9) ma, = F,,
where the F,’s are the components of force F acting on the particle located
at the point P(x). It follows from (86.9) that

> ma, Ga SS S dks ava >

a oq‘ a 0q'

and hence equations 86.8 can be written

d (7) -<
86.10 == ||=
( ) dt \dg' = =3F5
Comparing (86.7) with (86.10), we conclude ae

Ox"
Q; ty > ae eat?
a 0q

in which the vector Q, is called generalized force.

The equations

(86.11) 4%) = fe ‘
dt \dq' 0q'
are known as Lagrangean equations in generalized coordinates. They yield
a system of n second-order ordinary differential equations for the general-
ized coordinates q’. The solutions of these equations in the form
Cc: g= dt)
represent the dynamical trajectory of the system.
If there exists a function V(q1,...,q”), such that

3 Ga =O

the system is said to be conservative, and for such systems equations

86.11 assume the form
d aL) OL _
86.12 —|(—
: Ab aqi
where L = T — V is the kinetic potential.
Sec. 86] EQUATIONS IN GENERALIZED COORDINATES 239
Since L(qg,qg) is a function of both the generalized coordinates and
velocities,

ar ag? *agit
AES Ol Pamolny

Inserting in this expression from Lagrangean equations 86.12, we get

GEEwOLY, rd yeb.
(86.13) dt agi! tea)?

=.(ot '
dt\dq'
But, since L = T — V, and the potential energy V is not a function of the
qs

since T = 4a;,q‘¢’. Thus equation 86.13 can be written in the form

d(L— 2T) frag SO 2a
0,
dt dt
which implies that T+ V =h (constant). Thus, along the dynamical
trajectory, the sum of the kinetic and potential energies is a constant.
It follows from this development that the study of natural holonomic
dynamical systems with n degrees of freedom can be reduced to a study of
motion of a single particle in the n-dimensional space.
We can phrase the problem of determining the dynamical trajectory of
the system in the language of calculus of variations. Indeed, the state-
ments of Hamilton’s principle and of the least action principle, given in
Secs. 82 and 84, can be repeated word-for-word if the “point” is interpreted
to mean a set of n parameters q!,...,q", specifying the configuration of
our dynamical system in a certain n-dimensional space.
In symbols the principle of Hamilton reads

i (6T+ 0, 6q") dt = 0,
ta as AS
(86.14)
ty

and, if the force field Q, is conservative, the principle can be stated in the
form
te
) { Ldt=0.
th

These variational equations imply the satisfaction of Lagrangean equations

86.11 and 86.12.
240 ANALYTICAL MECHANICS [CHAP. 4

It follows at once from the formulation of the principle of least action in

generalized coordinates (cf. equations 84.3 and 84.4) that dynamical
trajectories in a conservative field are geodesics in the n-dimensional
Riemannian manifold with the arc element dS given by
dS? = 2(h — V)a,; dq’ dq’.

The fact that the dynamical trajectory can be regarded as a geodesic

permits one to geometrize dynamics.
Problems
Show that the dynamical equations in spherical coordinates with

ds? = (dr)? + r2(d0)? + r? sin? 6 (d¢)?

assume the form
aV
mF — r62 — rd? sin? 6) = — —,
or

ld, lav
seh,
m)r250 % 6 cos Jsant
— rd? sin
dt r 00’

d lor
—(r24 sin? 0) | = — ==
m| 5 0% si | rsin@ a’

whereas in cylindrical coordinates, with ds? = (dr)* + r°(d6)? + (dz)?, they are

oV
mi — r62) = ——,
or

Bas al
at” poe

87. Virtual Work and Generalized Forces

In the developments of the preceding sections no characterization of

forces F, acting at a point (x’) of a rigid body was made. It is customary
in the study of mechanics of continuous media to classify forces into three
categories.”
(a) Internal constitutive forces.
(6) Reactive forces produced by constraints.
(c) External impressed forces.
* The reactive forces produced by constraints are also external forces.
Sec. 87] VIRTUAL WORK AND GENERALIZED FORCES 241
We can visualize a material body as being composed of a vast number of
particles which interact with one another in a rather complicated way.
As long as the constitutive internal forces are of the action-reaction type,
they need not be taken into account in the dynamical eqtations, since
their resultant at any point P of the body vanishes. Thus the forces F,,
appearing in the formulas of Sec. 86, consist
of reactive forces produced by constraints and
external impressed forces.
To illustrate the meaning of this we can con-
sider a rigid body fixed at some point O by
a smooth pin, and subjected to the action of
impressed force F, (see Fig. 37). The pin at O
constrains the motion of a body to that of
rotation about the point O. The reactive force
R, acting at O does no work if the body is
displaced so as not to violate the constraints at O.
We shall term all reactive forces that do no work
in an arbitrary displacement which does not Fig. 37
violate the constraints workless forces. Any dis-
placement of a point of a body that is consistent with imposed constraints
is a virtual displacement, and we denote such virtual displacements at a
point 2” by dz”.
The work done by the impressed forces F, in a virtual displacement
da” is
(87.1) Wiel ar yn ree
where the summation is carried over all particles of the body; this will be
the total work if the reactive forces are of the workless type. We define W,
to be the virtual work in producing a virtual displacement 62’, provided
that the reactions are workless. Otherwise, W, will also contain contri-
butions from the working reactive forces.
not .
It should be noted carefully that a virtual displacement da” is
undergoes under
necessarily the actual displacement dzx" that the point P(a")
ement that
the action of specified forces. It is merely any conceivable displac
a body can perform without violating the constraints.
If a given natural holono mic system with n degrees of freedom is
-- +4"), and
described by the generalized coordinates q’, then 2” = a7(q',.
generalized virtual
the virtual displacements 6x’ are related linearly to the
displacements dq’, namely, ;
Ola. 5
(87.2) ba” = — dq’.
eq
also used in dynamics, especially
13 Virtual displacements that violate constraints are
forces.
if one is concerned with the computation of reactive
242 ANALYTICAL MECHANICS [CHAP. 4

In formula 87.2 the 6g’’s are arbitrary, and they are necessarily consistent
with constraints imposed on the system, since the coordinates q° are
independent.
If we insert expressions from (87.2) in (87.1), we get

Ox 59°
(87.3) W, = =F, —
q
= Q; 6q’,

where the last step makes use of the definition of the generalized force Q;.
It follows from this formula that one can calculate the generalized forces
Q;, acting on the system, by computing the work W, produced by dis-
placing the system through a virtual displacement dq’ ¥ 0, (j fixed), and
with 6g’ = 0,1 #7. Then Q, = W,/dg’. We shall resort to this method
of computing generalized forces in the illustrative examples of Sec. 89.

88. Nonholonomic Systems

The derivation of Lagrangean equations in Sec. 86 is based on the

assumption that the dynamical system is holonomic and that its config-
uration is described by n independent generalized coordinates g‘. When
the g’ are not independent, the derivation of appropriate dynamical
equations from Hamilton’s principle (86.14) hinges on general consider-
ations presented in Sec. 57. :
In dealing with nonholonomic dynamical systems it is customary to
assume that the generalized velocities g’ enter in the constraining relations
linearly. Accordingly, we shall suppose that n generalized coordinates g'
satisfy m < n conditions of the type
(88.1) ¢,,(q',-.-,9"q°
= 0, (Kea ] Peni) (= 1,...5n),

in which the coefficients c,, are continuously differentiable functions of the

variables q’.
The set of m equations 88.1 can be written in the form

c,.g' 6t = 0,

and, since g' dt = dq‘, we have m relations

(88.2) Cy; Og? = 0,
in which the variations dq‘ in general are not independent.
“We call attention to the distinction between the virtual displacements 6g and the
actual displacements dq‘ taking place along the dynamical trajectory g‘ = q(t).
SEc. 88] NONHOLONOMIC SYSTEMS 243
To deduce the dynamical equations from Hamilton’s variational
equation
“te
(88.3) | (6T+ Q;6q')dt = 0, si
ty

in which the 6q’ are constrained by m relations 88.2, we introduce (cf.

Sec. 57) m unknown functions A*(q',..., g") and form with the aid of
(88.2) the sum
(Re -4y Ae, S0g"'=0, (ig 1 tote)! (hele
oe 71):
Since 6T = (0T/0q') 5g’ + (0T/0q') 6g’, equation 88.3 yields
TOTS SOT
(88.5) i (Sg + —dg'+ 0, 6a’dt = 0.
ti \0g 0q'

But, 6g’ = (d/dt) 6g’, and the integration by parts of the first term in the
integrand of (88.5) gives (cf. 82.3)
Siang OF '
88.6 { (£5 = 0,
o “dt
( ) t, \Oq* dtdg’ 2: )04

when we recall that 59‘(t,) = 6q‘(t.) = 0 along each varied path.

We rewrite (88.6) by inserting in the integrand the term A*c,; dq° = 0,
s/OT Vadior
88.7 | ( es diene mee? == QO; + ie.)6g’ ahi = 0.
( ) ty 0q' dt 0g’

In formula 88.7, the g‘ are constrained by m relations 88.4, and, if we agree

to consider the first n — m coordinates gq‘ as independent variables, and
suppose that m functions 2*(q', . . . ,g") can be chosen so that

(88.8) ee OTTO a Me. an, fori=n—m+1l,...,n,

dq’ dtodq’
then (88.7) reduces to
Sore, dor
88.9 | (= - <5 t it aXe ‘Je‘dt= 0, i
0q' dt 0g’ 2 :
( ) ty
(Gime 1, 2k eh, B= 1):
are independent,
Since the first n — m variables q’ in the integrand of (88.9)
chosen arbitrarily, and
the variations 6g’ for i = 1,2,...,m — mcan be
we conclude that

(88:10) -“— ———— + OFF eo, = 0,9 (= 1,2,...52 - m).

e n generalized co-
The two sets of equations (88.8) and (88.10) involv
..-,9"). By adjoining
ordinates gq’ and m Lagrangean multipliers 2*(q',
244 ANALYTICAL MECHANICS [CHAP. 4

to these equations m equations 88.1, we get m + m equations for the

determination of the qg’s and 7’s.
The circumstances under which the /* can be determined so as to satisfy
(88.8) were detailed in Sec. 57; they relate to the rank of the Jacobian
matrix for (88.1).
We note that when the equations in (88.8) and (88.10) are written as a
single set
nL anol ;
(88.11) diag’ ag O; + An Cris (i=l 2 een

the right-hand member of (88.11) differs from the right-hand member of

(86.11) by the term R; = A*c,;. This term corresponds to the generalized
reactive forces produced by constraints when the Q, are generalized forces
that act on the system in the absence of constraints.
In special situations the Q; may be derived from potential V(g',. .. , q”).
As an illustration of the use of equations 88.11, we consider a homo-
geneous circular cylinder rolling under gravity down a rough inclined
plane.
Let the cylinder of radius a and mass m roll without slipping down
the plane making a fixed angle ¢ with the horizontal. The position of the
cylinder is determined by the angle of roll @ and by the distance x
through which the center of mass of the cylinder moves down the plane.
We shall take as our generalized coordinates q' = 6, g? = x, and note
that the kinetic energy T of the system is the sum of the kinetic energy of
translation of the center of mass and the kinetic energy of rotation about
the center of mass. Thus
(88.12) T = 4mi? + Imk?6,
where k is the radius of gyration of the cylinder.
Since the plane is rough, there is frictional force F acting in the plane,
and we suppose that this force is just sufficient to prevent slipping. In
this event « and @ are related by
(88.13) ado = dx,
where a is the radius of the cylinder.
The constraint (88.13) is actually holonomic since (88.13) can be inte-
grated to yield « = af, so that the problem can be reduced to the con-
sideration of one independent variable, say x. However, to illustrate the
theory of this section, we write (88.13) in the form CpG’ = 0 [cf. (88.1)],

(88.14) a eo F0
sO that Cy = Qa, Cig = —1.
SEc. 88] NONHOLONOMIC SYSTEMS 245
Equations (88.11) then yield

ar--d
ee aT ee ia
a6. di a0
(88.15) ae .
an
Biing 4toOlesd
oT 1 Ya
sya
ee CENA Qs
Now, the work W done by the gravitational force alone, when the
center of mass moves through a distance x is W = amg sin ¢. Hence V =
—a«mg sin ¢, and
= ——
oV = 0, = —
oV
— = meg
;
sin
Q; 36 Q» an & $ @.

On inserting these expressions in (88.15) and using T in the form 88.12,

we get a pair of equations

(88.16) mb = — mé = mg sind —A,

which, when compared with (88.11), show that the generalized reactions
R, are

To compute A, observe that af = x, so that § = #/a, and use this relation

to eliminate # and 6 in (88.16). The result is
_ mgsin
Leta kt
and hence equations 88.16 yield
. mgasind ve k?mg sin
(88.17) mé= 1a MT SID eee oa :

The term k2mg sin ¢/(a? + k’), in the second of equations 88.17, represents
the frictional force F opposing the component mg sin # of the gravi-
F =
tational force along the plane. If the cylinder is solid, k? = a?/2, and
force F = uN, where uu is the
img sing. The magnitude of frictional
coefficient of friction and N is the pressure of the cylinder on the plane.
Since N = mg cos ¢, we conclude that « = F/N = } tan ¢.
the
As another illustration of the use of equations 88.11 we consider
brachistochrone problem in a resistin g medium. ’®
of Variations,” American
15 See G. A. Bliss, ‘“The Problem of Lagrange in the Calculus
Calculus of Variations (1962), pp.
Journal of Mathematics, 52 (1930) and L. A. Pars,
241-243.
246 ANALYTICAL MECHANICS [CHAP. 4

Let it be required to determine an arc of a continuously differentiable

curve

(88.18) C=y=y%), YXR)=%H, YS) = Y%,

such that the time of descent of a bead of unit mass, moving on C under
gravity, is as short as possible. We suppose that the motion is opposed by
a force R(v) per unit mass, where R(v) is a continuously differentiable
function of the speed v.
We choose the positive Y-axis in the direction of gravity. Since the
work done by gravity on the particle less work done by the resisting force
R(v) is equal to the change in kinetic energy, we have
2
a = gdy — R(v) ds.

If we take x as our independent variable, this relation gives the con-

straining condition in the form

(88.19) $Y, vy’, 0’) = we" — gy’ + RWV1 + WP =0,

where primes denote derivatives with respect to x.
The integral to be minimized under the condition 88.19 is

te x Ze 1\2
(88.20) " =i] dt = {Alin Pahited
a ee
ty 2, U 2 D

We denote the integrand in (88.20) by F= Ri1 + (y’)?/v, and construct

the function

G=F+A¢

o cette! + A(x)[vo' — gy’ + RWI

+ (y'F.
If we define

(88.21) Hee + AR(v),

we can write G as :

(88.22) G = HV1 (y'?

+ + Av’ — gy’).
The equations of C are determined from Euler’s equations

88.23
(88.23) dGy
ae mie aGy
- — G,=0,
SEc. 88] NONHOLONOMIC SYSTEMS 247

and, since G does not contain y, we conclude from the first of equations
88.23 that G,, = a or

(88.24) ee Hy’Bae) —dAg=a, %:

Vi +(y'?
where a is a constant.
The second of equations 88.23 yields

ae Sa rn, lo’ = 0:
dx
or

(88.25) 5) tral
Vi+(y'?
We thus have the system of three equations 88.19, 88.24, 88.25 for the
determination of y(x), v(x), and A(x). We can rewrite them as

d dv is giv sey
ds ds

(88.26) H dy =Ag+a,
ds

te =H,
ds

by setting ds = J1 + (y' dx. On eliminating dy and ds from (88.26),

we get the equation
H(H, dv + R da) = (gd + a)g dd,

and since R = H, by (88.21), we can write it as

H(H,, dv + H, da) = (ga + a)g da

or
H dH = (gd + ag da.

The integration of this equation yields

(88.27) H? = (gd + a)? + 3’,

where b? is the constant of integration.

on in A so that
It follows from (88.21) that (88.27) is a quadratic equati
constants a, b.
Acan be regarded as a known function of v and the integration
on of C be sought in the parame tric form
This suggests that the equati

(88.28) Circ =iartv); y = y(v).

248 ANALYTICAL MECHANICS [CHAP. 4

Since dy/dv = dy/ds - ds/dv, we find with the aid of the first two equations
in (88.26) that
dy __w(Ag +a)
(88.29)
dv g(Ag +a) — RH’
the right-hand member of which is a known function of v. On performing
quadrature we then get
(88.30) y =fiv, a, b) + ¢,
where c is a constant. Equation 88.30 is one of the desired equations in
(88.28). To obtain x = 2(v), we note that dx/dv = dx/ds - ds/dv and, since
dx/ds = vie (dy/ds)? and both dy/ds and ds/dv are determined by the
first two equations in (88.26), we see that dx/ds is also a known function of
v. The reader will check that
dx bv
dv g(Ag+a)—RH’
so that

(88.31) x = f,(v, a, b) + d,
where d is a constant. The constants of integration in (88.30) and (88.31)
must be determined for the initial conditions. To make the problem
physically meaningful we must impose some restrictions on the relative
magnitudes of R(v) and the gravitational force g, as, for example, R < g
for all relevant values of v. .

Problems

1. A hollow cylindrical drum of mass m rolls under gravity down a rough

inclined plane making an angle ¢ with the horizontal. What must the coefficient
of friction u be to prevent slipping? (Answer: « > } tan ¢.)
2. A bead of mass m slides on a smooth rod rotating in a vertical plane about
one end with constant angular velocity #. Show that the equation of motion is
? — wr =g sin wt, and solve it.
3. A bead slides on a smooth circular wire of radius a, which is rotating with
constant angular velocity about the vertical diameter of the wire. Show that
6 — w* sin 0 cos @ = (g/a) sin 6, where @ is the angle made by the radius to the
particle with the diameter.

89. Illustrative Examples

We give next three examples illustrating the use of generalized co-

ordinates.
Consider first the problem of a simple pendulum, consisting of a bob of
mass m supported by a light inextensible cord of length /. We shall
SEC. 89] ILLUSTRATIVE EXAMPLES 249
3

Fig. 38

suppose that the pendulum is set in vibration in some plane which we take
as the Y1Y*-plane. (See Fig. 38.)
In order to form Lagrangean equations

(89.1) alee) ~ ag
we need the expression for the kinetic energy
(89.2) T = hmy'y.
However,

y' = Isind = Isin@,

(89.3)
y” = I(1 — cos 0) = i(1— cos),

where we take the arc-length g = /6 as our generalized coordinate. Since

y? =qsing/l and y! = g cosq/I, equation 89.2 becomes T = 3m(q)?.
The work W, done in producing a virtual displacement 6q is
W, = —mg sin 6 6q

= —mg sin 40g,

and hence the generalized force Q = —mg sing/I. Thus equation 89.1
yields
(89.4) gt sin 4=O)
250 ANALYTICAL MECHANICS [CHaP. 4

Fig. 39

and, since for small displacements sin 6 = 6, for small vibrations we have

ink diz 0:
where k? = g//. The solution of this equation is g = acos (kt + «). The
solution of (89.4) can be expressed in terms of elliptic integrals of the
first kind.
We turn next to a more interesting problem of a double pendulum.
Consider an arrangement of particles shown in Fig. 39, where we suppose
that the masses m, and m, are supported by inextensible light cords of
lengths /, and /,, respectively. The pendulum is assumed to vibrate in one
plane, and we take as our generalized coordinates the quantities 6 and ¢,
which give the angular deviations of the cords of lengths /, and /, from the
vertical.
The equations connecting the coordinates (y,, y,”) and (y,!, y”), of the
masses m, and m,, with generalized coordinates g' = @ and g* = ¢ are

yy = 1, sin q’,

y= 1, cosq’,
Yo = i; sin q + l, sin q’,

yy" = 1, cosq’ + I, cosq’.

Since
T= gmyyy'Ya' + dngya' ye", (i = 1, 2),
an easy calculation gives

T= sm (hg)? + m[(hg')? 7 2hlg'g? cos (q? — g') + (log)? ]}.

SEC. 89] ILLUSTRATIVE EXAMPLES 251
Now, the work done in a small virtual displacement 6q* when dq! = O is

W;” = —mygl, sing? dq”

so that gee pas
Q2 = —Mbylyg sin q?.
Also the work done in a displacement dq! when dq? = 0 is
WY) = —(m, + m,)gl, sing
‘ dq’.
Thus
QO, = —(m, + m,)gl, sin q.
Making use of equations 89.1, we find a pair of simultaneous ordinary
differential equations

d
7 {(m, + my)(11)°g* + melylog? cos (q? — q*)}

(89.5) : — Mol,12g°q" sin (q° — q') = —(m, + m,)gl, sin ae

at {myly log" cos(q” — q*) + m,(1p)"q°}

L + mgl,1.474? sin (q* — q') = —magl, sing’,

for the determination of the dynamical trajectory.
Instead of determining the generalized forces Q, and Q, directly, we
could have made use of the potential energy V, which is
V = mgil,(1 — cos q') + meg(l, + 1, — 1, cos g! — I, cos g?),

if we assume V = 0 when gq! = g? = 0.

For a detailed discussion of the solution of the system of differential
equations 89.5 we refer to standard treatises on analytical dynamics.
As our final example we consider the problem of small oscillations of a
conservative dynamical system about the position of stable equilibrium.
We suppose that the system is natural, holonomic, with n degrees of
freedom, and select the generalized coordinates q‘ so that the equilibrium
position is given by q’ = 0, (i = 1,..., 1). Since the equilibrium is stable,
the potential energy V(q',...,q") has a minimum value at q’ = 0, and
hence = = 0. If we choose the potential level to be zero at q' = 0,
then iis ateaesien of V(q',...,q") in Taylor’s series about q' = 0 has
the form V = 4b,,q'q’ + O(q°), where O(q°) denotes the remainder after
the second-degree terms in the q’. Since we are concerned with small
oscillations about the point q’ = 0, we shall suppose that the potential
energy is represented with sufficient accuracy by the quadratic form
(89.6) V = 3b,,9'¢', (b;; = 5;;)-
252 ANALYTICAL MECHANICS [CHAP. 4
The kinetic energy T of the system is

(89.7) T = 34399’; (4;; = 4;;);

and we suppose that, in the neighborhood of the point q' = 0, the a’”’s do
do not vary appreciably, so that they can be regarded as constants.
The Lagrangean equations 86.12 now yield the system ofn simultaneous
second-order ordinary differential equations with constant coefficients

4,4’ + big’ = 9.
Instead of integrating this coupled system directly we can simplify the
problem by introducing a new set of independent variables g”, the so-called
normal coordinates, which are related linearly to the coordinates g’ in such
a way that the quadratic forms 89.6 and 89.7 reduce simultaneously”* to a
sum of squares. We then have
T= (GEE = (gy ae oe ce (gay

(89.8) = Gey if 7% 4. ig 0 hia

All the coefficients of the q’’s in (89.8) are nonnegative since the quadratic
form 89.6 is necessarily nonnegative if the potential energy V has a
minimum at q’ = 0.
The Lagrangean equations now become

f+ 22q" = 0, — (no sum on i),

and their solutions obviously are

q’=c,(cosat +e), (G=1,...,).

Thus the oscillation of the system, in terms of the normal coordinates,
is simple harmonic with normal modes of vibration determined by the
characteristic values A; which satisfy the frequency equation
(89.9) |b;; — 2a,;|
=0.
If the roots A; are distinct, the normal coordinates g’' are determined
essentially uniquely. For multiple roots, the choice of normal coordinates
is not unique. This follows from the analysis given in Sec. 16.
The problems of small oscillations are of great technical interest, and
there is an extensive literature concerned with the study of oscillating
systems with finite and infinite number of degrees of freedom.1”
*6 This algebraic problem was considered in detail in Sec. 16.
See, for some interesting examples, Frazer, Duncan, and Collar, Elementary
Matrices and Some Applications to Dynamics and Differential Equations, Cambridge
University Press, 1938.
SEC. 89] ILLUSTRATIVE EXAMPLES 253
As a concrete illustration of our general discussion of oscillation of
dynamical systems about the position of stable equilibrium consider the
double pendulum in Fig. 39 with J, = /, =/ and m,=m,=m. The
expressions for T and V given on pages 250-251 in this cage reduce to
T = 4Pm[2(q")* + 241g? cos (q? — q') + GP")
V = megl[(1 — cos q') + (2 — cos q! — cos q?)].
If we expand 7and V in powers of q' and g' and retain only the second-
degree terms in these variables, we get
T= ae
ian).
(89.10)
yale +(q2)
To reduce (89.10) to the form 89.8 we introduce the normal coordinates
«x = q'', y = q” by a linear transformation (cf. Sec. 16)
iy AY,
(89.11) im variety
gq? = bx + boy.
The coefficients a; and b; in (89.11) must be chosen so that T and V in
(89.10) reduce to
T=} + y),
(89.12) HET)
V == ,22? 4+ A,7y/".
The substitution from (89.11) in (89.10) yields two quadratic forms in
which the cross-product terms must vanish. Thus
2b,a. + 2b,a, = 0,
4a,a, + 2b,b, = 0.
Solving these, we get

Furthermore, the comparison of coefficients of x? and y? shows that

PM 2 OP, 2 242
Manin ance Atal
Thus the desired transformation 89.11 is

ph eo
=4/8 toy),
(89.13) a
2 -<, 5 a2 PN) 9),
254 ANALYTICAL MECHANICS [CHAP. 4

under which V assumes the form

V= S12 — y2)a" + (2 + 29")

Accordingly, the Lagrangean equations in normal coordinates are

#4 +2 —/2)e¢=0, y+ FQ + ./2)y= 0.
Solving these, we get
(89.14) % =2.¢;.C0S (Ak €a)} Y = C3 COS (Ast + Cy),
where

4i= FQ SMBs aA rac Ff 2):

The independent oscillations in (89.14) have periods 7, = 27/A, and
T, = 2zm/d,. The vibration with the larger period is that of x; it is called
the grave mode. The rapid mode is that of y. If we set y = O in (89.13)
and consider the grave mode, we see that

qo — V/2 fi:

The performance of the pendulum in this case is illustrated in Fig. 40a.

On setting x = 0, we get the motion of the rapid mode for which g. =
—V24q,. This is shown in Fig. 40b. The angles shown in these figures

O O

q q

9 = ~2q, |
q, =— V2q,

(a) Grave mode q, = W/2q, (6) Rapid mode g, =—-~/2q,

Fig. 40
SEC. 89] ILLUSTRATIVE EXAMPLES 20
are exaggerated. The general motion given by (89.13) is a combination of
motions of the two characteristic modes.
One can, of course, get the normal frequencies A,, A, directly from the
frequency equation (89.9). 4
If we substitute T and V from (89.10) in the Lagrangean equations
86.12, we get a pair of equations

2g +g + “Eq =0,
(89.15)
PichiG ch Sahen,
in which the variables g,' and g,? are coupled. We assume solutions of
(89.15) in the form
(89.16) gi=a,e*” q*?=a,e%
and determine / so that equations 89.15 are satisfied. On substituting
(89.16) in (89.15) we get two homogeneous equations

a,(22 — 224) + a(—2) = 0,

a(—7*) + a,(£ — 2?)=0,

which will have nontrivial solutions for a, and a, if, and only if,
2g/i — 24? —A/?

Ao wedis 2
On expanding this determinant, we find that

P= E24 V2),
which yields two values A, = (g//)(2 — V2), A? = (g/)(2-+ /2) corre-
sponding to the grave and rapid modes found previously. Thus the solution
(89.16) can be written
qi = cet" rs c,ettt,

q@ ae HD cet42t ae af cet,

as in (89.13).
Problems
39,
1. Find the normal modes of vibration for the double pendulum in Fig.
assuming that /, = /2, but my, # mg.
surface
2. A particle of mass m oscillates about the lowest point of a smooth
orthogon al cartesian and
z = 4(ax® + 2hey + by”), where the coordinates are
256 ANALYTICAL MECHANICS [CHAP. 4
the z-axis is directed vertically up. We suppose that the vertical component of
the velocity is small, so that T = 4m(a? + y). The potential V = mgz =
(mg/2)(ax® + 2hey + by”). Obtain equations of motion, determine their solu-
tions in the form x = a,e*, y = a,e™, and conclude that if V = min at x = 0,
y =0,thena >0,b >0, ab —h? > 0.
3. Let the particle in the problem at the end of Sec. 80 be acted on by the
force of gravity, so that F; = mga sin u’, F, = 0. (Note that the work OW done
in a small displacement dy? is 6W = —mg dy? = mga sin u’ du’.) Show that the
motion, when the particle passes through the highest and lowest points on the
sphere, is along an arc of a great circle. A complete discussion of this problem
is involved. See P. Appell, Mécanique rationelle, 1, Chapter 13, especially Sec.
277. See also a discussion of the spherical pendulum in J. L. Synge and B. A.
Griffith, Principles of Mechanics.
4. Let the particle in the preceding problem execute small oscillations about
the lower pole of the sphere. Consider projection of this motion on the plane
tangent to the pole and discuss the motion.
Hint: Set u’ = 7 — (r/a), and deduce equations

- aes
Fr
Pose ide eles a?

rii + 2ru = 0.

90.- Hamilton’s Canonical Equations

Consider a conservative holonomic dynamical system with n degrees of

freedom and the integral .
(90.1) = |Frtate) de
ty

where L = T — V is the kinetic potential. We saw in Sec. 86 that the

system of Euler’s equations associated with the variational problem J =
extremum consists of a set of m simultaneous second-order ordinary
differential equations 86.12, which we write in the form

(90.2) dy =/igi== 0, - (= 100 anew

dt
by using the subscript notation for partial derivatives of L(g,q). In
a variety of considerations it is convenient to rewrite the system of 7
Lagrangean equations 90.2 in the form of an equivalent set of 2m first-
order equations, known as Hamilton’s equations.
The function L(g, 4) = T(q,4) — V(q) depends on n generalized co-
ordinates q' and n generalized velocities q'. Instead of the variables g' we
can introduce a set of n new variables p; defined by the relations
(90.3) pa Lj, q)s (i ss I; 2; OTS n),
SEC. 90] HAMILTON’S CANONICAL EQUATIONS 251
where wesuppose that the system 90.3 is solvable for the g’ in terms of the
ie and q'. This, surely, will be the case if the Jacobian determinant

~ 0. We next construct a function H(p,q) of the independent

0q’
variables g and p,
(90.4) A(p,q) = q'pi — LG.)
by expressing the q' = 4'(q, p) in the right-hand member of (90.4) in terms
of the q' and p, with the aid of (90.3).
On differentiating (90.4) with respect to q’, we get
hon, 095 0g"
Hy = age L,i a TA?

and since p; = L;; by (90.3),

(90.5) Ay = —Ly.
Similarly, we compute
peaog 0g’
1S Mt fe see ede oe
: Op; ap;
which on using (90.3) reduces to
(90.6) Hy, =.
But the Lagrangean equations 90.2 state that

ALi: = Li
dt
and, if we recall the definition 90.3 and formula 90.5, we obtain a set of n
first-order equations,
dpi
(90.7)
90.7 —if) = aC
—H,i, Mb, )
ss on Wk

which together with the n equations 90.6,

dq‘
90.6 —3 = fT. {soul
es 1),
Hag? dt aH
constitute the system of 2n first-order Hamilton’s canonical equations.
The function H(p, q), known as the Hamiltonian function, has an im-
portant physical meaning. Since L = T — V and V is a function of the
qg' alone, we can rewrite (90.4) as

(90.8) H=q——L=g——T+V.

However, T = }a,,q'4/, OT /0q° = 4:9’,

258 ANALYTICAL MECHANICS [CHAP. 4
so that

q' do Odd iy
0g’
and hence (90.8) reduces to
H=T-4+ VS.
Thus # is the total energy of the system.
The variables
1G. :
(90.9) dD; = ae = aif
0g’
are called the generalized momenta, and we note that the square of the
magnitude of the vector p, is
(90.10) Pp? = app; = A A;0;9°9'
= AGG’ = P28be

As an illustration of a simple use of Hamilton’s equation, consider a

particle of mass m moving under the influence of a central force field with
the potential V(r), r being the distance of the particle from the center of
attraction. If we choose polar coordinates r = g', 6 = g* as our generalized
coordinates, then

,
¥ nls + (r6)*] = $4;;9'9’
where
ue Ae

But H = T + V = }a"p,p; + V, by (90.10), which yields on inserting the

values of the a’’,
2 2

Ho= £4Pa V(r).

2m 2mr
Thus
ae
GH aL ay OH _ 9 Cie Er oH Pale
er mr a0 Op; m Op, mr
and hence Hamilton’s equations (90.6), (90:7) in this problem are

90.11) dritips
—=, et* —=-4,
dl p, dp, _ £2
1 Ps" _ yp), dp tet)
ae
( ) dt m dt mr dt mr ”) dt 4
The last of these equations, combined with the second, yields
d 24

a r’@) = 0,
—(mr*é)
SEC. 91] NEWTONIAN LAW OF
GRAVITATION 259
which is a statement of Kepler’s second law of planetary motion. It is
not difficult to show by using the remaining equations in (90.11) that if
V = —m/r, the orbit is a conic section (cf. Sec. 97).
”

Problems
1. If a particle of mass m is constrained to move on a smooth surface, show
that the system of Hamilton’s equations is
du* aH dp* eH
PID ape DogeTas Bee eed)

with p, = ma,,it? and H = (1/2m)a**pyp, + V.

2. Show that along the dynamical trajectory dH/dt = 0,so that H = constant
is an integral of Hamilton’s equations.
3. Show that @L/ég* + dH/dq* = 0.
4. Write Hamilton’s canonical equations for Problem 1, Sec. 89.
5. If T = $m(q)? and V = k(q)?, k > 0, show that H = p?/2m + mo*(q)/2,
where w = k/m. Deduce that g = V2h/mw? sin (wt + «).
6. Deduce Hamilton’s equations from the variational principle 6f L dt = 0.
Hint: Write L in the form L = p,(dq‘/dt) — H(p,q), treat the variations of p

i ee
and g as independent, and show that

ae
tmOp, fOP
is
Ni Pe ct eh
oq", q
0g’) dii=D,

91. Newtonian Law of Gravitation

The general formulation of dynamical equations, outlined in the pre-

ceding sections, imposes no specific restrictions on the functional form of
the fields of force. In various applications of dynamics, including those
of astronomy and atomic physics, we are concerned with the behavior of
dynamical systems subjected to the action of central fields of force and,
in particular, those fields whose intensity varies inversely as the square of
the distance of the particles from the center of attraction. The inverse
square law of attraction had its origin in Newton’s studies of motion of plan-
etary bodies in what he termed’* the “eccentric conic sections.” We state
this law as follows: i
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.
Thus the law, when stated in the form of a vector equation, reads
mM,
F=y 3 412
ri2
18 Newton’s Principia, Book I, Sec. III, Propositions 1-17.
260 ANALYTICAL MECHANICS [CHAP. 4

where m, and m, are the masses of the particles and rj, is the vector from
P, to P,. The constant of proportionality y depends on the choice of units;
in the cgs system its value is found to be 6.664 x 10-*, and its physical
dimensions are M-1! L? T-2. In our work we shall make y = 1, by a
suitable choice of units of measure, so that

(OIE) F=

We observe first that the law of gravitation 91.1 refers to two particles,
and, since in dynamics one usually deals with continuous distributions of
matter, it is necessary to generalize it. Thus one can subdivide the bodies
into small parts, replace each part by an equivalent material particle, add
the forces corresponding to discrete particles, and pass to the limit as the
number of subdivisions is increased indefinitely. This procedure for two
bodies 7, and 7, leads to the formula

(91.2) F =| i PiP2 y 5dr, dre,

2 Tie
where dv, and dr, are the volume elements of bodies 7, and 73, p, and ps
their density functions, and r,, is the position vector of dr, relative to dry.
We shall assume that p, and p, are piecewise continuous.
Since two interacting bodies ordinarily give rise not only to resultant
forces but also to resultant moments, it is necessary to verify that the
generalized law of gravitation 91.2 reduces to the parent law 91.1 and
yields no nonvanishing couples when the bodies 7, and 7, are allowed to
shrink to a point.
To show that this is indeed so, we introduce an orthogonal cartesian
reference frame Y, and denote the coordinates of points of the bodies
7, and 7, by (y,’) and (y,'), respectively (Fig. 41). We replace the distributed
mass p, Az, by the concentrated mass m, at P,(y,}, y,2, y;°), and the mass
p2 At, by mz at P(y2', Yo”, Y2°).
In accordance with the law 91.1 we have, for the components of force
AF? due to these masses,
— owes a
AF* = pyp2A7, Ata

and for the components of moments'® AL,, relative to the origin O,

AL =e.yi AF

ae Cis Y1’ PrP2 Ar, Ar,

Ye ~ 9;
r3

19 We recall that the moment of force F relative to the origin, acting at a point
determined by the position vector r, is L=rxF or, in terms of components
L; = e:;y'F*.
SEC. 91] NEWTONIAN LAW OF GRAVITATION 261

y}

Fig. 41

Adding these vectorially gives the resultant force

(91.3) ee)
pa] | PrpaYo! a— t') 4, ie

and the resultant moment

(91-4) Lm | ikpipseiayse Ya
= = Ye dt, dtp.

We prove next that, as 7, and 7, are allowed to shrink toward P, and Ps,
respectively (or, even if 7, alone is allowed to shrink to zero) the resultant
moment L, tends to zero and equation 91.3 specializes to the law in the
form 91.1;
We choose the origin O of the coordinate system at P,, and let 7, shrink
toward O and 7, toward P.(y2!, Yo”, ¥2*). Since p; and p2 in equations 91.3
and 91.4 are nonnegative functions, the first mean value theorem for
integrals is applicable and we obtain

F= oa
wine
= ie
pip2 7; dt2,

and
eye eye
pe [ein | [ | roe tee
r U6| T2

where brackets denote the values of affected quantities evaluated at certain

 262 ANALYTICAL MECHANICS [CHAP. 4

points in 7, and rz. As the dimensions of 7, are allowed to approach

zero, y,' > 0, and hence L; > 0, whereas the first of the above integrals
reduces to

This is precisely the law of gravitation 91.1 for two particles located at
(0, 0, 0) and (y2", Yo", Yo").
It follows from the foregoing that a material body interacting with a
point mass produces no resultant moment L. Moreover, direct calcu-
lations show that this is also true when the point mass is replaced by a
sphere 7 whose density p is a continuous function of the radius alone. The
resultant force F, exerted by the body on the sphere, turns out to be the
same as that produced by the body acting on a point mass m =e dr,
located at the center of the sphere.?° :
Consider next a body 7 with piecewise continuous density p and let
Py /y2, y°) be a fixed point either within or outside 7. The gravitational
potential V(P) at the point P due to 7 is defined by the integral

(91.5) V(P)= iane dr(&),

where r = J (y! — &)? + (y? — &)? + (y® — &)? is the distance between
Py’, y*, y®) and the variable point (1, €, €°) associated with the volume
element d7(&) of r. The integral 91.5, as we shall presently see, defines a
differentiable function V(y', y?, y°) for all locations of P.
If P is outside the body, the integral (91.5) is proper and we can compute
as many derivatives of V as desired by differentiating (91.5) under the
integral sign with respect to the parameters y‘. In particular,

(91.6) — = -F,,

where the F; are components of the gravitational force

(91.7) F(P) = {BOLE

Tot

exerted by the body 7 on a particle of unit mass located at P(y).

°° See, for example, I. S. Sokolnikoff and R. M. Redheffer, Mathematics of Physics

and Modern Engineering, McGraw-Hill Book Co. (1958), pp. 410-411.
Sec. 92] INTEGRAL TRANSFORMATION THEOREMS 263
If P(y) is within 7, the integral 91.5 is improper, since r = 0 when the
variable point (&1, &*, £°) coincides with (y', y®, y*). However, an improper
integral may still be differentiated under the sign when the derived integral
is uniformly convergent. In our case the uniform convergence of (91.7)
follows from the familiar test on convergence of improper integrals.”
Moreover, it follows from the uniform convergence of (91.7) that F(P) is
continuous throughout all space.
Although V(P) is of class C” whenever P is exterior to 7, more stringent
restrictions must be imposed on the continuity of p to ensure the existence
of second derivatives of V(P) at points within 7. It is a fact that if p is of
class C!, then the second derivatives of V(P) exist at all interior points of r.
A careful analysis of the difference quotients of the function F(P) shows,
moreover, that?2 V(P) satisfies the Poisson equation

(91.8) V2V = —4rp

at all points within t and Laplace’s equation

(91.9) VV =0,

at points exterior to rT.

in
Equations 91.8 and 91.9 imply that the second derivatives of V(P)
& of 7. In
general suffer discontinuities whenever P crosses the surface
aid of Gauss’
Sec. 93, we establish the validity of (91.8) and (91.9) with the
the advantages
flux theorem. A treatment based on Gauss’ flux theorem has
a purely analytic dis-
of physical suggestiveness that do not appear in
differen ce quotients.
cussion based on the aforementioned study of the
the characte r of regions
However, it imposes quite severe restrictions on
fiux theorem is a theorem
and surfaces that bound the regions. The Gauss
results 91.8 and
in the large, and it need not be used to deduce the local
in the neighbo rhood of a
91.9, which concern the properties of potentials
given point.

92. Integral Transformation Theorems

we translate the well-
To provide analytic tools for our further study,
of Gauss , Green, and Stokes in
known integral transformation theorems
the language of tensor calculus.
a
borhood of(y*), |r" p(&)/r?| < A, ii
21 Since, for all values of (€*) in the neigh
enden t of (&). For a discus sion of this test see I. S.
where A is a constant indep 2, or
w-Hill Book Co. (1939), pp. 367-37
Sokolnikoff, Advanced Calculus, McGra 6.
O. D. Kellogg, Foundations of Potent ial Theory, Springer-Verlag (1929), pp: 146-15
pp- 146-156.
22 See O. D. Kellogg, op. cit., Chapter 6,
264 ANALYTICAL MECHANICS [CHAP. 4

Let F be a vector point function of class C1 in an open region 7 bounded

by the regular®* surface © and continuous in the closed region & + 7. We
denote by n the exterior unit normal to = and state the divergence theorem
in the form
(92.1) [aivF dr =|" -ndo.

The integral with the subscript 7 is evaluated over the volume 7, whereas
the integral in the right-hand member of (92.1) measures the flux of the
vector quantity F over the surface &.
We recall from elementary vector analysis that, in orthogonal cartesian
coordinates, the divergence of F is given by the formula
OF = Or OF
222 iV Ee eae
sos Oy: andy bay?
If the components of F relative to an arbitrary curvilinear coordinate
system X are denoted by F’, then the covariant derivative of F° is

I i \ px
yaa —— OF” _ i \ps >

Ox? kj J
and we observe that the invariant F’. in cartesian coordinates reduces to
the right-hand member of (92.2), and hence it represents the divergence of
the vector field F. In addition,

F-n= 9g,,F'n' = F'n,

and hence we can rewrite equation 92.1 in the form

(92.3) Ge dt = |Fn, do.

From this theorem two other theorems (usually attributed to Green) can
be derived easily.
Let u(x, x, x°) and v(a!, 2, 2°) be two scalar functions of class C2 in +
and of class C1 in the closed region = + 7. We denote the gradients of u
and v by u; and »v,, respectively, so that

u;= = and .=—.

Ox" Ox"
If we set
F, = uv,
** We omit a rather involved discussion of the properties of surfaces to which the
divergence theorem is applicable. For a detailed treatment of this consult O. D. Kellogg,
Foundations of Potential Theory, pp. 97-121.
SEC. 92] INTEGRAL TRANSFORMATION THEOREMS 265
and form the divergence of F’, we get

Fi, = °F, = g'(uv,; + 0,u;).

We insert this in equation 92.3 and obtain the desired formula

(92.4) [ew + v,u;) dr = |uo do.

The invariant gv; ;appearing in the left-hand member of equation

92.4, when expressed in cartesian coordinates, is the Laplacian of »v,
0*v/dy' dy’, and if we denote the Laplacian operator by the symbol V’,
we can write
ity, j = V2v.

Also the inner product g”v,u; can be written as

g’vu; = Vu- Vo,
where we use the customary operator V to denote the gradient.
Hence formula 92.4 can be written in the familiar form

(92.5) [uve dt =| un+ Vv do - [vu- Vo dr,

T

where
n° Ve= yn= ae
on
Interchanging u and v in equation 92.5 and subtracting the resulting
formula from equation 92.5 yields a symmetrical form of Green’s theorem

(92.6) |(pvelleVad dr = IE(«oe

on
~- 0S) do.
on a

Theorems stated in equations 92.3, 92.4, 92.5, and 92.6 are, perhaps,
the ones most frequently used in mathematical physics.
The Laplacian of v,
(92.7) V0 = £70;55
when written out explicitly in terms of the Christoffel symbols associated

weet)
with the curvilinear coordinates x‘ covering Es, is

a > ae
and the divergence of the vector F" is
eI IGES
aoe
k\ dv
uilox

; OF’ jal
(92.9) Fi,=—+ ale
Ox j
266 ANALYTICAL MECHANICS [CHAP. 4

Formulas 92.8 and 92.9 can be written in different forms, which fre-
quently are more convenient in computations. Equation 31.10 yields

i| ) ra
31.10 || Ete
Sore lid = date
and hence the divergence F', in (92.9), can be written as

I
O. as 0
——a ‘|F,:
Ps Ox? a (2 = Vg

or ¢
(92.10) Fi,=
plet(yek)
eaaceee
If we set in this formula F’ = g”(dv/0x’), we get
1 a(/g g? dv/dx’)
(92.11) Vv = gv J ie ae Cae
Vg Ox
We turn next to a consideration of Stokes’s theorem which permits us
to express certain surface integrals in terms of line integrals.
Let a portion of regular surface & be bounded by a closed regular curve
C, and let F be any vector function of class C1 defined on & and on C.
The theorem of Stokes states that

(92.12) { n- curl F do = F-Ads,

= Cc
where A is the unit tangent vector to C, and curl F is the vector whose
components in orthogonal cartesian coordinates are determined from

ey i) e3
7) 0 0
(92.13) curl F = By) By? By ;

TEs eRe

the e; being the unit base vectors in a cartesian frame. The determinant
in 92.13 can be written as a symbolic vector product V x F.
We consider the covariant derivative F;; of the vector F; and form a
contravariant vector
(92.14) Gi = —c*F,
It is readily checked that in cartesian coordinates equation 92.14 reduces
to 92.13, and we define the vector G to be the curl of F.
Sec. 92] INTEGRAL TRANSFORMATION THEOREMS 267
Since n+ curl F = n,G' = —e’’*F, ,n,, and the components of the unit
tangent vector A are dx'/ds, we may rewrite equation 92.12 as

(92.15) -| e*F
,.n,do =| foe ds. 1
5 Cans

The integral|F, dx' is called the circulation of F along the contour C.

C
Problems

1. Prove that

}v,n' do = |v dr,
Ji T

where v; = @v/ x? is continuous on = and of class C? in 7.

2. Show that
(a) In plane polar coordinates with ds? = (dr)? + r?(d6)?,
; Ll Orr Ser,
div F =| Leer ‘4

y Lied dv d (1 av
OO = TlaN ae) 7 BON26) |
where F, and F, are the physical components of the vector F, that is,
F = Fir, + F.6,,

where r, and 9, are unit vectors.

(b) In cylindrical coordinates with ds? = (dr)? + r2(d0)? + (dz)*,
1arF,) 10F, F,
div F =
or r 00 az”

av
it. |
tere
1a eV
ar 2a Toe”
where F = F,r, + F,0, + F,2, and ry, 9), 2 are unit vectors, so that F,, F,,
and F, are the physical components of F.
(c) In spherical coordinates with ds® = (dr)? + (0)? + 9° sin? 6 (dd),
1 a(r2F,) 1 a(sin 6F,) 1 aF,
a ~ 2 ar r sin 0 ld) rsin@ a’

dv dv
: 1 7)(gee= 1 a
a(sin esS| i 1 a

ee aes sind 00 P2 sin? 0 ag?’

9,F, +
where the physical components of F are F,, F,, Fy, so that F = riF, +
$F, r,, 9,, and¢, being the unit vectors.
268 ANALYTICAL MECHANICS [CHAP. 4

3. Show that, in an orthogonal curvilinear frame X,

Ven ay V B05 ae V 833 as

1 a] fo) re)
Cun he
VJ811822833 Ox Ox Ox

ets V £9 ae V e353Ee
where the a; are the unit base vectors and F = Fla, + Fa, + Fag.
4. Show that the contravariant components of the curl of a vector F are:
i (oF, oF \ 1 perl ere =i
I OF, OF;:
lat ~ a) vp lon sat) Vg \ot axe)
5. Prove that under suitable restrictions on continuity the curl of a gradient
vector vanishes identically.
6. In orthogonal curvilinear coordinates,
- 1 1
Seu? §22 = 99> &§33 = 733°
=p" =0, i #j, and &u
Siti
fe)

If we set ds* = e,*(dz!)? + e,%(dx?)® + e,*(dx3)*, so that 23, = e17, So0 = es",
§33 = eas then

k
(a) [j,k] = 9, - = 1) i, j, k distinct,
i

ave a ey de;
—
[ii, i]= e;aa fi | ee
\i/ _ oOloge;
[ij, 1] = —[ii,j] = €% yi?

ii pay gaSt e; i ¥ d log e;

(e;)? dui’ f|i are (no sums),
f|a

1 Q €2€3 Ov 0 (ese, Ov @ (ee, Ov

(6) V2v =
C€xeq| dx! e, dat pe Cy Ox * zal é, Ppetyiik

93. Theorem of Gauss. Solution of Poisson’s Equation

In accordance with Newton’s law of gravitation, a particle P of mass m

exerts on a particle P, of unit mass, located at a distance r from P, a force
of magnitude F = m/r*. Imagine a closed regular surface © drawn around
the point P, and let 6 denote the angle between the unit exterior normal
n to 4 and the axis of a cone with its vertex at P. This cone subtends an
element of surface do. (See Fig. 42.) The flux of the gravitational field
produced by m is
[®- ag = [mses 2 dw
f coste
where do = r° do/cos 0 and dw is the solid angle subtended by do.
SEC. 93] THEOREM OF GAUSS 269

Fig. 42

We thus have

(93.1) [p-ndo=| mado = 4em.

= 2

If there are n discrete particles of masses m, located within &, then

n
m,cos6,
F-n= >
2
4=1 T

and the total flux is :

(93.2) |e -ndo = 47> m,.

t=1

continuous
The result embodied in formula 93.2 can be easily generalized to
er such distrib utions nowher e meet the
distributions of matter whenev
is a standar d one. The contrib ution to the
surface ©. The procedure
element p dr, contain ed within 7, is
flux integral from the mass

[p-n da =| 7"r
ao,
> pay

y within & is
and the contribution from all masses contained entirel

(93.3) [pe (| 20"

nd =|x\dr ) do,
x i

interior to x. Since
where |denotes the volume integral over all bodies
r never vanishes, so that the
all masses are assumed to be interior to x,
270 ANALYTICAL MECHANICS [CHAP. 4

integrand in (93.3) is continuous, and hence one can interchange the order
of integration to obtain

(93.4) | F-ndo =| o(|ee ee

2 T x r

cos 4 do
But the integral | = 47, since it represents the flux due to a
x
unit mass contained within &. Hence

(93.5) i F-ndo= 4n| pdr = 4nm,

x T
where m denotes the total mass contained within X.
We can now state
Gauss’s THEOREM. The integral of the normal component of the gravita-
tional flux computed over a regular surface & containing gravitating masses
wholly within it is equal to 4m, where m is the total mass enclosed by X.
This theorem can be extended to situations where & intersects the dis-
tributed masses with sufficiently smooth density p. Let a regular closed
surface & intersect a distribution of mass with continuously differentiable
density p. We construct two surfaces &’ and &” parallel to = (cf. Sec. 73)
such that &’ is interior to & and &” encloses & (Fig. 43). The flux pro-
duced by the gravitating masses varies continuously across &’ and &”
when these surfaces, while remaining parallel, are made to approach %.
Since &” does not intersect &, Gauss’s flux theorem can be applied to
compute the total flux over &” produced by the masses within &. Accord-
ingly,

(93.6) i
x (F -n);do = 47m,

Fig. 43
SEC, 94] GREEN’S THIRD IDENTITY 271
where m is the total mass within & and the subscript i refers to the flux
produced by the masses inside &. On the other hand, the net flux over 2’
produced by all masses outside & is

(93.7) Le -n), do = 0,
for the flux cone from any point outside & cuts &’ twice.
Now, if we let &’ and &” approach &, the right-hand members in (93.6)
and (93.7) do not change, whereas the left-hand member of (93.6) becomes
the flux integral over © produced by the masses within & and the left-hand
member of (93.7) represents the flux over & due to all masses exterior to x.
Thus the total flux produced by a distribution of masses within & is

(93.8) [e -n)do = 47m =| Airp dr.

If we further suppose that F is continuously differentiable, we can apply

the divergence theorem to the surface integral in (93.8), and get

(93.9) {(div F — 4p) dr = 0.

This relation is true for an arbitrary region 7 and, since the integrand in
(93.8) is continuous, we conclude that

(93.10) div F = 47p throughout 7.

However, formula 91.6 states that F = —VV, and thus (93.10) is equiv-
alent to
(93.11) V2V = —4rp.
Thus at all points interior to the body 7, the gravitational potential
satisfies the Poisson equation. We note in conclusion that formula
dr
[91.5] V(P) = [=
gives a solution of equation 93.11 at all points in rT.

94. Green’s Third Identity. Harmonic Functions

Green’s symmetrical formula

F)
[92.6] i(uV2v — vV°u) dr = fi(.= =) =) do
in the open region 7
is applicable to any pair of functions u, v of class C?
setu = 1/r and v = V,
and of class C! in the closed region & + 7. Let us
212, ANALYTICAL MECHANICS [CHAP. 4
where r is the distance between the points P(x1, x?, 2°) and P,(&, &, &°),
and V is the gravitational potential of a distribution of mass with con-
tinuously differentiable density p, so that V is of class C? in +.
Since 1/r has a discontinuity at (x) = (&'), we delete P(x) from 7 by
enclosing it by a sphere o of radius 6 and with center of o at P. Functions
u = 1/r and v = V then satisfy the conditions of theorem 92.6 in the
region 7 — e« bounded by 2 and o (Fig. 44). However, in the region 7 — e,
V2u = V*(1/r) = 0, and formula 92.6 yields

(94.1) lyy de = (! My—- ya) do

me L ron on

10V 2)
lel (2Nk
+ (on On dg,

where n is the unit exterior normal to the surface & + o. Since, however,
on o the normal n is directed toward P,

I | | i]
nN=)
eg
ee lie
Sade a7aS
hes S
tel
ee

I | ey.) +. = i3
Sa?
ae= Bas
seed Seo

= -o| (=) dw — 4rV,

é or r=06

Fig. 44
Sec. 94] GREEN’S THIRD IDENTITY aE:
where V is the mean value of V over the sphere o, and w denotes the solid
angle.
On letting 6 — 0, the right-hand member of (94.2) yields —47V(P), and
it follows from (94.1) that 5

(94.3) vipy= +1 | V? Kegel wh ae 4 | pe ras

An Jr rt 4n Jz Onr 47J= On
The important formula 94.3, known as Green’s third identity, states that
every function V of class Clin © + rand of class C?in 7, can be represented
as the sum of three integrals appearing in (94.3). If V(P) is regular at
infinity, that is, if for sufficiently large values of r, V is such that

(94.4) iv] < eh band av m

= Tao2
r or i

where m is a constant independent of r, then on extending the integration

in (94.3) over all space we get

(94.5) teapdley [.fine RViViegg

dr,
TT

provided that this volume integral converges. The surface integrals in

(94.3), when extended over all space, vanish by virtue of the regularity
conditions 94.4.
At all points not occupied by matter (that is, where p = 0), the gravi-
tational potential V satisfies Laplace’s equation
(94.6) VV =0.
c
A function satisfying equation 94.6 in a given region is said to be harmoni
7, formula 94.3 reduces to
in that region. If V is harmonic in the region
1 | 1 av 1 a(1/r)
(94.7) VP) = —4a LV aeneing | do,
JerOn be vba olndn
the values of
so that the values of V are completely determined in 7 when
However, these
V and of its normal derivative dV/dén are known on x.
another, and we
surface values cannot be specified independently of one
on & fully deter-
shall see that the specification of the values of V alone
specification of aV/on
mines V(P) at all points of r. On the other hand, the
, provided that
on = determines V(P) in 7 to within an arbitrary constant

(94.8) a = 0.
x On
la 92.6 on setting vu= |
The condition 94.8 follows directly from the formu
ied by every harmonic function.
and v = V. It is anecessary condition satisf
274 ANALYTICAL MECHANICS [CHAP. 4

If = in (94.7) is the surface of a sphere of radius R, with center at P,

then [0(1/r)]/@n = [0(1/r)]/er = —1/r? and (94.7) gives

(94.9) V(P) = : [vao,

4rR? Jz
when we note the condition 94.8. Formula 94.9 states an important
property of harmonic functions: The value of a harmonic function V at
the center of a sphere is equal to the mean value of V over the surface of that
sphere. This property enables us to prove the following basic theorem on
harmonic functions.
THEOREM. A function V harmonic in a closed regular region & + 7
assumes its maximum and minimum values on the boundary & of 7, with the
single exception when V = constant throughout rt.
To prove this theorem assume that V takes on its maximum (or minimum)
V, at some interior point P of tr. We construct a sphere S in 7 with center
at P and of radius R, then
I Vdo
V(P) = -——
47 R®
by (94.9). But the right-hand member of this expression is the average
value V of V over S, and the average value V can equal the maximum Vy
only if V = V, on S. Furthermore, since R is arbitrary, we conclude that
V = V, at every interior point of S. To show that V has the same constant
value V, at every point QO of 7, we connect P and Q by a curve € of finite
length and cover it by a sequence of overlapping spheres with centers on C.
Within each sphere of this sequence, V has the same constant value V, and
hence V(Q) = Vy. Thus, unless V = constant throughout 7, it takes on
its extreme values on the boundary &.
The determination of a harmonic function V in + from the specified
values of V on the boundary & of 7 is known as the Problem of Dirichlet.
If 7 is a finite region, we have an interior problem and when 7 is an infinite
region bounded by a closed surface , we have an exterior problem of
Dirichlet. ~.
It is easy to prove that the interior problem of Dirichlet for a regular
region & + 7 does not have more than one solution. For, let there be two
functions V, and V,, harmonic in 7 and which assume the same values on
the boundary &. But V = V, — V, is also harmonic, and it assumes zero
values on &. This implies, however, that V = 0 throughout 7, since
otherwise V would have to take on its positive maximum, or a negative
minimum, in the interior. In the same way we can prove the uniqueness of
solution of the exterior problem of Dirichlet if we suppose that V is
regular at infinity.
Sec. 95] FUNCTIONS OF GREEN AND NEUMANN 275

The determination of a harmonic function V in 7 which satisfies on the

boundary & of 7, the condition

(94.10) oe= f(P), with {f(P) do = 0,

is called the Problem of Neumann.

Since V = constant is a harmonic function that satisfies the condition
aV/én = 0 on &, we conclude that the solution of the Neumann problem
(if it exists) is determined to within an arbitrary constant. It is possible to
prove, although the proof is by no means easy, that the Dirichlet and
Neumann problems are solvable for finite regular regions when the
specified values on the boundary are continuous.”
Problem

Show that formula 94.7 is valid in an infinite region 7 exterior to a closed surface
= whenever Vis regular at infinity. Hint: Apply formula 94.7 to a finite region
bounded by = and by a sphere S of radius R so large that S encloses &.

95. Functions of Green and Neumann

We have just shown that the solution of the interior problem of Dirichlet
in Laplace’s equation
V7u = 0 in 7,
(95.1)
u=f(P) on,
when it exists, is necessarily unique. Also the solution of the Neumann
interior problem
V*v =0 in 7,

(2-2) ue = g(P) on &,

on
ry constant when
with {g(P) do = 0, is determined to within an arbitra
problem unique,
g(P)is continuous. To make the solution of the Neumann
we adjoin to (95.2) the normalizing conditi on

(95.3) fedo = 0.

replaced by the Poisson

When Laplace’s equations in (95.1) and (95.2) are
ms in Poisson’s
equations, we have the Dirichlet and Neumann proble
equation.
24 See O. D. Kellogg, op. cit., p. 311.
276 ANALYTICAL MECHANICS [CHAP. 4

Formula 94.7 is not directly applicable to the solution of problems 95.1

and 95.2, since it requires the knowledge of the values of the function and
of its normal derivative on &. We show next how this difficulty can be
avoided by introducing special functions that depend only on the shape of
the region and not on the assigned boundary values f(P) and g(P). We
begin with the Dirichlet problem.
Let P(x) and P’(é) be a fixed point and a variable point, respectively,
in 7 (Fig. 45). We construct a function G(P, P’) with the following
properties:

(a) G(P, P’) = -+ w(P’),

where r = PP’ and W(P’) is harmonic in r.

(d) GU P=) on &.

The condition (6) requires that

w(P’) = — : pad
so that W(P’), and hence G(P, P’), is uniquely determined by properties
(a) and (5). We call G(P, P’) Green’s function for the region r.
We show next how Green’s function can be used to construct an
explicit integral formula solving the Dirichlet problem in Poisson’s
equation
(95.4) V?V = —4rp iy;
V =f (P) on &.

Fig. 45
Sec. 95] FUNCTIONS OF GREEN AND NEUMANN 277
The integral formula will include the solution of the boundary value
problem 95.1 as a special case.
Green’s symmetrical formula,

[92.6] [ws — vV*u) dr = (u - v a) do,

r Jz\ dn on

cannot be applied to u = G(P, P’), v = V, since G(P, P’) > 00 as P’ > P.

If, however, we delete the point P by enclosing it in a sphere o of radius 6,
as in Fig. 44, the formula 92.6 is valid in the region 7 — « bounded by >2,
and o. We can write

(95.5) { (GV?V— VV?G) dr =| (6% - yo) so

pai r\ On on
OV 2)
G— —V—] do.
+|( on on id

But in 7 — «, G = 1/r + w is harmonic, so that V7G = 0 and G= 0 on

>. Also V2V = —4zp by (95.4), so that (95.5) reduces to

(95.6) oop) drhice

—4r i} roe ha [G+»)
= Mall |eeee Ih bas %a
Bet

+| 7 eal)
Baus) do.
a or

dV/dn =
In writing (95.6) we observed that, since n is an exterior normal,
.
—aV/ar and dG/@n = —AG/er on o.
= r? dw, where dw is
Since @V/@r and w are continuous on 6 and do
second integral on the
an element of solid angle, it is obvious that the
right in (95.6) tends to zero as 6—0. Similarly,

[ vo ae =0 as 60,
a or .

whereas
[ oO ae Ag V(P) as 6-0.
a if

get
Accordingly, on letting 6 — 0 in (95.6), we
aG do,
Vi"
(95.7) 4nV(P) = I4nGp dr —|
278 ANALYTICAL MECHANICS [CHAP. 4

which is the desired solution of the problem 95.4. If we set p = 0, we get

the solution of the corresponding problem in Laplace’s equation,
1 0G
95.8 V(P) = -—- — V— do.
( ) fe 47 J= On ~

To apply this formula we must first obtain Green’s function G for the
region 7, that is, we must solve the special Dirichlet problem
V’w = 0 in 7,

w= — on».

Similar considerations are applicable to the Neumann problem 95.2.

We introduce the Neumann function

N(P, P’) = ++ w(P%, r

where w(P’) is harmonic in 7 and satisfies on the boundary = of 7 the

condition”
aw 0 (*)+ constant.
On on /

Computations entirely similar to those carried out previously for the

Dirichlet problem yield for the boundary value problem 95.2, the formula

V(P) = a { gN do.
4n Jz

Physically, Green’s function G(P, P’) can be interpreted as the electro-

static potential in the interior of a grounded conducting surface & pro-
duced by a unit charge at the point P. The potential produced by a unit
charge alone is I/r, and w(P’) represents the potential produced by the
induced surface charges on &. Since = is grounded, G(P, P’) = 1/r +
w(P") = 0 on X. The Neumann function can be interpreted as steady heat
flow from a source of strength 47 placed at P, when the heat flows across
the surface X at a uniform rate.

96. Green’s Functions for Semi-infinite Space and Spherical Regions

A physical interpretation of Green’s function given in Sec. 95 enables us

to construct Green’s functions for the half-space z > 0 and for the regions
interior and exterior to the sphere.

> To make w unique we can normalize it by requiring {wdo = 0.

x
SEC. 96 GREEN’S FUNCTIONS 279

Q(x, y¥, —2)

Fig. 46

When a positive unit charge is placed at P(z, y, z) (Fig. 46) and a negative
unit charge at the mirror image Q(z, y, —2), the electrostatic potential G
produced by these charges at P’(é, 7, ¢) is

(96.1) econpep
te
where r= PP’ =V(E—2P? +(7-y? + (6 — 2
and
r= OP =VE— 2 + (g— yh + (E+ 2.
Obviously, G = 0 on the planez = 0, and since w(P’) = —1/r’ is harmonic
for z > 0, equation 96.1 gives the desired Green’s function for the region
z > 0.
On the plane z = 0,

ie E oF Dat |
on OC Ico rat ral Ico
95.8,
and after performing simple calculations and substituting in formula
we obtain the solution of the Dirichle t problem 95.1 for the region z > 0
in the form of Poisson’s integral

baka prpind be ee es aaa

Steak eae fonfr (E— a +@q— y+ 27"
 280 ANALYTICAL MECHANICS [CHAP. 4
The specified values f(&, 7) of Von z = 0, must, clearly, be such that (96.2)
has a meaning.
A similar procedure enables us to construct Green’s function for
the spherical region x? + y? + 2? < R® and obtain the solution of the
Dirichlet problem 95.1 for the sphere.
We take P(x, y, z) in the interior of the sphere S of radius R '(Fig. 47)
and construct the image Q of P with respect to S, so that OP- OO= R®.
Let P’ be a variable point in S. When P’ is on S, similar triangles OP"P
and OP"@Q yield the relation

PIO Sn ORL Amtre ? ER

P’P OP’ z eape
where p = OP. Thus

= at 5 on S
r op

If, for any interior point P’, we define

(96.3) G(P, P’) = oy: — :

Frat pw

then (96.3) gives the desired Green’s function, since w = —(R/p)(1/r’),. is

harmonic in the interior of S and G(P, P’) = 0 on S.
A simple calculation of dG/dn from (96.3) gives
dG a8 R? = p”
on S,
dn Rr°

Fig. 47
SEc. 97] THE PROBLEM OF TWO BODIES 281
and formula 95.8 yields the solution of the Dirichlet problem 95.1 for the
sphere in the form of the Poisson integral
9
r4 “

(96.4) iat }pie eiding yt

da JS Rr’I i

This integral is usually written in spherical coordinates (p, 0, hb) as

pee) 20%
7 207 2 2 ! ! 1

(96.5) Vip, 6, ¢) = + [Rosin 6 a'|0 Ried

(R® — 2pRceosy + py)?
4 Jo

where cos y = cos 6 cos 6’ + sin @ sin 6’ cos (¢’ — ¢), 9 being the co-
latitude and ¢ the longitude of P.
Green’s function for the region 2? + y? + 2 > R is obtained from
(96.3) by interchanging the roles of P and Q.
Problems

1. Show by using (96.4) that for every position ofP in the interior of the sphere

1 R? — p
(tej
2 Nhe
47R Ss r3

fixed
Hint: Take the z-axis along OP, so that r2 = R? — 2Rpcos@ + p*. For a
is fixed. Express do = R® sin 6 dé d in spherical coordinate s
position of P, p
and evaluate the integral.
S is
2. Show that the solution of the exterior problem 95.1 for the sphere
given by

—= |
VP) = 4nR f(P) eet tay
Js re ‘
where p = OP >R.
3. Deduce (96.5) from (96.4). Hint: Let y be the angle between OP and OP’
when P’ is on S.

97, The Problem of Two Bodies

Given a system of
The problem of two bodies can be stated as follows:
ance with the law of univers al gravitation,
two particles interacting in accord
? This proble m was solved by Newton
what is the trajectory of the system
basis of all considerations in
in the Principia, Book I, Sec. III. It lies at the
astronomy.
general curvilinear co-
Since there is no particular advantage in using
we refer our system to a set of orthogonal
ordinates in specific problems,
282 ANALYTICAL MECHANICS [CHAP. 4
cartesian axes. We denote the coordinates of mass points m,, m, (at any
given instant of time ¢) by (#,', 2’, 73) and (#51, 15, x*) (Fig. 48). We
also introduce another cartesian reference frame Y moving with the mass
m, in such a way that m, is always at the origin O of the Y-system, and the
axes Y’ always remain parallel to the axes X’. The coordinates of the mass
point mg, relative to the Y-axes, are denoted by y’, and we have the
relations
(97.1) y = Xe) — xy’, (ie the 23)
We choose the coordinates y’ of the mass m, as three of our generalized
coordinates, and for the remaining three generalized coordinates we take
those of the center of mass of the system. Thus

(97.2) i ce PO ig tations)
mM, + Mg
Clearly, the wu’ lie on the line joining the points (x,") and (a,"), and our
choice of the generalized coordinates is then as follows:
g=¥, g@=y, g@=y, qg = ul, g@=uw, ge =u.

If we solve equations 97.1 and 97.2 for the 2,‘ and 2’, we obtain

a ; ye mra a
m m
(97.3) 1 2
pt yt ds FL ay y?
3
m, + Ms,
and these equations enable us to determine the positional coordinates 2‘
in terms of the generalized coordinates g’.

BXac

y?

Fig. 48
SEc. 97] THE PROBLEM OF TWO BODIES 283
This particular choice of generalized coordinates is made with a view
toward obtaining a simple expression for the potential energy V of our
system of particles. Indeed, since the magnitude of the force of attraction
F is given by F = mym,/r?, where r is the distance between the particles,
the potential energy V is
MM, _ mM,
r (2) ie! ao})? a (x, on 2) ee (x,° = Nal

and it follows from (97.1) that the coordinates u’ do not appear in V, so

that V is a function of y’, y*, and y¥?.
We recall the Lagrangean equations

[86.11] g =) POPE brie OY

alae Dates Wn OGe
and compute

1 oer ee ML) a
—(m,
5 1 + m,)u'u’
2 + =ea 12 ity’.
yy

Since 0V/dq' = 0, for i = 4, 5, 6, an easy calculation makes equations

86.11 reduce to

ey be (i= d8208))
(97.4) m, + Mm,
i’ = 0, (i-==1132;,3).
of
Equations 97.4 are the differential equations characterizing the motion
the motion of the mass m, relative to m, is
our system. We note first that
toward it
the same as though the mass m, were fixed and m, attracted
l is [(m, + m,)/m,|V . This follows at once
with a force whose potentia
form
from the first three of equations 97.4 if we rewrite them in the
m, + m, 0V
(97.5) mi = — m, oy
the action of
Thus our problem is reduced to a study of motion under
97.4 states that the center of
central forces. The second set of equations
mass moves in a straight line with constant velocit y.
under the assumption
We shall carry out the integration of equations 97.4
than m, (the mass of the
that m, (the mass of the sun) is much larger
close to the mass m,
earth). If m, > m, the center of mass u’ will lie very
with those of the mass
and hence the coordinates u’ will nearly coincide
of equati ons 97.4 we conclude
m,. Thus 2,’ = u’, and from the second set
284 ANALYTICAL MECHANICS [CHAP. 4

y2

Fig. 49

that m, moves through space with constant velocity. Accordingly, we need

to examine only the motion of mass m, relative to my.
If m, > m,,
m, + My et
my, ;
and equations 97.5 become
my = — or (approximately).
y
Let us suppose that our coordinate axes are so oriented that the motion
of the mass m, relative to m, initially is in the Y*Y*-plane. Then, since the
force field is central, the motion will remain in this plane, for there is no
component of force at right angles to the plane. Let r and 6 (Fig. 49) be
the polar coordinates of mass m,, where
‘= rcos 6,
y =rsiné;
then the kinetic energy of mass m, is
T = ym_{[(y)” + Ge)"
= }m,(?? + 7°62).
Using this expression for 7, and V = —m,m,/r, in the Lagrangean
equations 86.11, with gq! = r and q? = 6, we get*®

mF
— mar? = — MyM»
r ’

d Py
or 28)
—(m,r*6)'= 0,

*6 We consider the force directed from m, to my.

SEc. 97] THE PROBLEM OF TWO BODIES 285

Or

(97.6) caer
F— rO° + ies 0,

r°é = h,

where A is a constant of integration.

Equations 97.6 are simultaneous ordinary differential equations for the
determination of the trajectory. The second of these states that the
sectorial velocities are constant. This is one of the Kepler laws.*? We can
use the relation r2?6 = h to determine the time required to describe the
orbit.
If h ¥ 0, so that the trajectory is not a straight line, we can eliminate the
time parameter ¢ by noting that 7° dO = h dt, or
1 6

rat a0.
h Jo

Since df/dt = df/d6 - d6/dt, we have the relation d/dt = (h/r*)(d/d6), and,
making use of this in the first equation in (97.6), we get

r? d6 \r? dO err

or multiplying by 7°,
d (h dr h?
Feu ee = 0),
ie!) dé \r* d6 r :

If we further change the dependent variable r in (97.7) by setting u = 1/r,

we get a simple second-order linear equation

d'u my
de" hh?”
whose solution is

v= a — ecos(§ — «)],

or
if
(97.8) r
~ 1—ecos(6 — a)

where / = h2/m,, and « and e are constants of integration.

Sec. 90.
27 See also an illustrative example at the end of
286 ANALYTICAL MECHANICS [CHAP. 4

y?

m9

>

Fig. 50

We thus see that the orbit is a conic section (Fig. 50) whose eccentricity
is e, with the position of the apse line determined by «. The constant « is
known as the perihelion constant. We shall not go to the trouble of
determining these constants”* in terms of the initial position and velocity
of mass m,, since the main object of this section is to obtain formula 97.8
for the purpose of comparing it with the corresponding equation of the
orbit in the relativistic dynamics.
8 See P. Appell, Mecanique rationelle, vol. 1, Chapter 11, and J. L- Synge and
B. A. Griffith, Principles of Mechanics (1959),pp. 160-169.
 5)
RELATIVISTIC MECHANICS

98. Invariance of Physical Laws

The formulation of the fundamental laws of classical mechanics in the

preceding chapter is based on the hypothesis that physical phenomena
take place in a three-dimensional Euclidean space. It is also assumed that
these phenomena can be ordered in the one-dimensional continuum of
the time variable ¢. The time variable ¢ is regarded to be independent
not only of the space variables x but also of the possible motion of the
space reference systems. The mass m of a body is likewise supposed to be
independent of the motion of reference systems, and, in particular, it is
invariant with respect to a group of Galilean transformations of coordinates.
By a Galilean transformation we mean a transformation that represents
a translation of one coordinate system relative to another with constant
velocity. Thus, if Y is a given cartesian frame, then a Galilean trans-
formation of this frame has the form
(98.1) g=ytut, (=1,2, 3),
origin of
where u‘ is a constant vector representing the velocity of the
suppose d in (98.1)
the Y-system relative to the cartesian system Y. It is
¢ = 0.
that the origins of the systems Y and Y coincide at the time
accelerations
From the linear character of (98.1), it is obvious that the
Y and Y, respec-
d®y'|dt? and d?y'/dt? of a particle referred to the frames
force F acting
tively, have the same value. It follows from this that the
ce systems moving
on a particle has the same value F = ma in all referen
other words, Newton ’s
relative to one another with constant velocity. In
to a group of Galilea n
second law of motion is formally invariant relative
transformations 98.1.
same in all inertial
Although the values of accelerations a’ are the
dance with the formula
systems, the estimates of velocities differ in accor
(98.2) avi tu,
1 See Sec. 76.
287
288 RELATIVISTIC MECHANICS [CHAP. 5

Hence a statement of any law that depends on the velocity relative to a

primary inertial system will not be formally invariant when expressed
in a secondary system. Consequently, the fundamental laws of electro-
dynamics and, particularly, of optics are not invariant with respect to a
group of Galilean transformations 98.1, since these laws depend on the
velocity of propagation of light. For this reason the primary inertial
system has occupied a unique position in the theory of optics. In order
to explain the observed fact of the independence of the velocity of light
from the velocity of its source, and to imbed optics in the framework of
analytical mechanics, physicists invented ether as a hypothetical carrier
of light waves. This carrier was endowed with whatever physical properties
were essential to ensure the same constant value for the velocity of propa-
gation of light in all inertial systems, even when these properties did
great violence to the established theories of elasticity and hydrodynamics.
For instance, it was supposed that ether is an all-pervading, frictionless
fluid that remains stationary relative to the primary inertial system, and
that, when physical objects are forced to move through it, they suffer
changes in shape, produced by elastic stresses that arise in a body moving
in a quiescent fluid. It was then merely necessary to assume that the
linear dimensions of measuring instruments suffer contractions depending
on the velocity u’, these contractions being of precisely the right amount
to make the velocity of light come out to be independent of the velocity
of its source.
A suitable formula expressing the dependence of the linear dimensions
of a body on its velocity relative to a primary inertial system was developed
by Lorentz, and a considerable body of the theory of relativity was
phrased by him, in 1904, in terms of the quiescent ether. Lorentz’s
mathematics appeared to fit well the observed results in the domain of
electrodynamics and provided a simple explanation of a puzzling behavior
of the electrical field of a moving spherical charge, but the physics of the
situation still remained in great doubt. However, all experimental
attempts to detect the existence of ether have led to null results, and, in
1905, Albert Einstein achieved an explanation of the so-called Lorentz-
Fitzgerald contraction by a sort of fiat which called for a profound
revision in the prevailing notions of space and time.

99. Restricted, or Special, Theory of Relativity

In 1905, Einstein proposed two postulates, one of which relates to the

formal invariance of physical laws, and the other epitomizes the results of
certain remarkable experiments on the determination of the speed of light.
* A. Einstein, Annalen der Physik, 18 (1905), p. 891.
SEC. 99] RESTRICTED THEORY OF RELATIVITY 289
These postulates can be stated as follows:

1. Physical laws and principles have the same form in all Galilean systems;
that is, reference systems that move relative to one another with uniform
velocities.
2. The speed of light in free space has the same constant value in all
inertial systems.

In a sense there is nothing startling about these pronouncements since

the ideas involved were in a state of ferment and discussion at the close
of the nineteenth century and are quite explicit in the writings of Poincaré,
Lorentz, Voigt, and others. But deductions to which Einstein was led
from these postulates served to clarify and revise our concepts of space,
time, and matter in a truly remarkable way. When viewed in the light of
the fundamental laws of dynamics of a particle, the first postulate, as
already remarked in Sec. 98, contains nothing novel. The laws of optics,
on the other hand, are not invariant under the group of transformations
98.1, and one can set out to modify them so as to achieve the invariance
of the fundamental laws of optics as well as mechanics. One way of
accomplishing this is to abrogate the hypothesis that the estimates of
time t are identical for observers located in two different Galilean reference
systems. Mathematically this puts the time variable ¢ on the same footing
with the space variables y’.
Thus, let us suppose that we have two cartesian reference frames Y
and Y, and an observer in the Y-frame recording the occurrence of some
event at the point (y’) at the time ¢, by means of four variables CEE
The four-dimensional manifold S, of the variables (y', y”, y*, t) consists
of E, and the range —0 <t< +0. The same event is recorded by an
observer in the Y-frame as a point (j', 7°, ¥°, 7), in S,, where 7 is the
estimate of time based on the clock in the Y-system of coordinates. As
yet the variables (y', y’, 9°, t) and (y', 72, 7, 7) are unrelated, but, since
we are in search of coordinate transformations which preserve the laws
of dynamics of a particle, let the word “event” mean the track of a
particle moving in the Y-frame under the action of a zero force. The
trajectory of such a particle in the Y-frame is a straight line, and we shall
suppose that the motion of the Y-system relative to the Y-system is such
that the trajectory in it also appears as a straight line.
This hypothesis implies the invariance of Newton’s first law and re-
quires that the variables (y', y*, y®, t) and (y', 7”, 7°, @) be related linearly.
Thus
g=ajy t+ Oat (i,j = 1, 2, 3),
(99.1) puvang tay
290 RELATIVISTIC MECHANICS [CHAP. 5

It follows from these equations that the origin of the system Y moves _
relative to the system Y with constant velocity. To see this, note that the
coordinates of the origin O of the system Y are (0, 0, 0), and hence the
trajectory of the origin O relative to Y is given by (99.1) as

7 = at,
C:
i = a,'t.
Hence dy'/di = «,'/«,4 = constant.
It can be shown in a similar way that the coordinate planes move with
constant velocity, so that the reference frames Y and Y are Galilean.
Let us suppose next that a spherical pulse of light is sent out from the
point P(y', y?, y?) of the system Y at the time t. According to Einstein’s
second postulate, light travels with constant speed c in all directions;
hence in dt seconds a photon starting from the point (y’) will be at the
point (y’ + dy’), and
(99.2) dy* dy =c? dt?.
Relative to an observer located in the Y-system, the light pulse originates
at the point (v1, 7’, 7°), and his equation for the spherical wave front,
di seconds later, is
(99.3) ay dy = Cat
Now if we substitute in (99.3) from (99.1) and compare the result
with (99.2), we find that a particular set of equations

7 = k(y* = vt),

y =y",
(99.4) yy

where k = 1 RI 1 — B?, B = v/c, leaves the quadratic form

(99.5) do* = c? dt? — dy' dy’
invariant. These equations correspond to the circumstance when the
system Y moves relative to Y with the velocity v along the Y?-axis.?
Equations 99.4 are known as the Lorentz-Einstein equations of trans-
formation.* We shall not launch into extensive discussion of their
* We note that for the pulse of light do = 0
* These equations have been derived in many different ways. See, for example,
J. Rice, Relativity, p. 89; R. Tolman, Theory of Relativity of Motion; A. Einstein,
Annalen der Physik, 18 (1905); Frank, Ignatowsky, and Rothe, Archiv fiir Mathematik
und Physik, 17 and 18; J. L. Synge, Relativity: The Special Theory (1959), p. 69.
SEC. 99] RESTRICTED THEORY OF ‘RELATIVITY 291
implications since most books on theoretical physics and special theory
of relativity discuss them at great length, and there is no need to duplicate
these considerations here. We shall mention only one example which has
a direct bearing on the Lorentz-Fitzgerald contraction mentioned in
Sec. 98.
Consider a rod moving with the system Y. The end points of the rod
have the coordinates (¥,", 0, 0), (y,', 0, 0), so that its length, as measured
by an observer in the Y-system, is L = ¥,! — ¥,1. Since #1 = k(y,' — vt)
and y," = k(y;" — vt),
L=y)— yy = V1 = PG 8):
Accordingly, the estimate of the length L of the rod by an observer in
Y-system is smaller than L in the ratio V1 — 62:1. Thus the observer in
the Y-system concludes that moving objects suffer a contraction in length.
The magnitude of this contraction is the same as that deduced by Lorentz
and Fitzgerald in connection with their study of the electrical field of a
moving spherical charge. Whereas Lorentz and Fitzgerald thought of
their contraction as a “real contraction” produced by the passage of
objects through a quiescent ether, in the foregoing calculation it appears
as a property of the space-time manifold subjected to a transformation
99.4, in which the space variables y’ are such that an element of arc ds
is given by the formula ds? = dy’ dy’.
If instead of cartesian variables y* we had chosen curvilinear coordinates
a’, related to cartesian coordinates y’ by the formulas

then the form 99.5 would have read

dy" ay
doc’ 2 ==c" co dt as2 g,,du' 2 dx’, (s.,
a — aaa.
eel

We note that the determinant of coefficients of this form has the value
—cg.
The foregoing formulas can be cast in a symmetric form by setting
f= 2*; then
(99.6) do® =a,,dx* dx’, (4,8 = 1,2, 3,4),
where
a;; = —fij> GJ = il, I, 3),

ay, = 0, Gay iC", and a = |d,,| = —c’'g

transformations T
If we now introduce a class of admissible functional
in the four-dimensional manifold X,

(99.7) T: g= za, 27,2°,2'), («= 1,2, 3,4),

gph RELATIVISTIC MECHANICS [CHap. 5

and require that the form 99.6 be invariant under the class of transforma-
tions 99.7, we can formulate the calculus of tensors as we did in Chapter 2.
Problems

1. Show, with the aid of equations 99.4, that events that are simultaneous
from the point of view of an observer in the Y-system are not in general simul-
taneous in the Y-system.
2. Discuss the slowing down of moving clocks.
3. Differentiate equations 99.4, and establish the relations between the com-
ponents of velocity w* of a moving point, as measured by an observer in the
Y-system, with the corresponding quantities w* measured in the Y-system.

dy} wi +0 dy* Ww
Ans. (Coo)
dt 1+ (6/c)w’ dt k(1 + (B/c)w)’
4. With the aid of the formulas given in Problem 3, show that, if # and v are
both less than c, then w/c < 1. Thus, if v = 0.9c, w = 0.9c, then w = 0.994c
instead of 1.8c given by the usual law of composition of velocities.
5. The expression arctanh w/c is sometimes called the rapidity. Show that
the usual law of composition of velocities is obeyed by the rapidities. Thus
Ww Ww v
arctanh — = arctanh — — arctanh-.
G Cc c

100. Proper or Local Coordinates

Consider a point P whose space coordinates relative to some reference

frame X are (x1, 2", x*). Let the velocity of P, relative to this frame at
the instant t, be v. We shall introduce a Galilean reference frame Y
moving with the point P so that, at each instant ¢, the point P is at rest
relative to the system Y. We shall call the system YX a local or proper
coordinate system.
Obviously the choice of local coordinate systems is not unique, since
the definition just laid down merely requires that the velocity of the
local frame be the same as that of the particle. This implies that the
estimates of time (measured by the clocks carried in two different local
coordinate frames) are the same. Hence the transformation from one
local system X to another XY’ has the form
ie= (zl, x2, 23),
i =i.
The interval do is defined by the formula
(100.1) do” = a,,.dx" dx®
=c'dt? — g,,dx' dz’,
Sec. 100] PROPER OR LOCAL COORDINATES 293

so that
day ee2 dx dx’
(100.2) (<<)
dt ee iba di
=e

where v is the magnitude of the velocity v of the point P relative to the

X-coordinate frame. If a local coordinate system X is introduced at
P, then, relative to X, v = 0 and equation 100.2 yields

(100.3) ec

in the local system. We define the Minkowski velocity vector u* by the

formula
[100.4] ue io art) 25374),
do
0, 1c),
and observe that its components in a local system are 10,0,
recipro cal tensor
Since a = |a,,| = —c?g #0, we can construct the
a*’, the Christoffel symbols
(St Oa py Sect)
Paar ae ee ED
ce
> ted
Q\da® = a*@—— Ot?
y = aeyo
ea 40 [af, 6],

c differentiation as
and define the operations of covariant and intrinsi
s us to define the Minkowski
was done in Chapters 2 and 3. This permit
acceleration vector f* by the formula
a 2a B da’
oe a
(100.5) se bo ais Z (a, B,y =1,...,4).
do? \By) do do’
so that do? = c? di? — dy’ dy’,
If our local reference frame X is cartesian
accel erati on relative to it are
the components /* of the Minkowski

h. (ar
patdo? ya = Sy
cdi\do
di\do/do
1 d’y"
ela
dx* dx 1 v®
dx* eS ER are
Sa
5 For,
dt Vey Ver —v?
do (c? dt® — gi; dx* dxi)t
yp? = yp? = 0, zt = ¢ a local system.
in
294 RELATIVISTIC MECHANICS [CHAP. 5

so that

(iain?) 3),
Cc
fi =0, since y =7.
We shall show next how Newton’s second law can be written in an
invariant form relative to all Galilean reference frames. Consider the
formula suggested by Newton’s second law,

Fe pa rg Oe OEY
(exes

where u* = dx*/do is the Minkowski velocity and m, is a constant whose

significance will appear presently. Now

a) dt
Fo
= — (mott") —
at : Line
1 | a
= ————_—(m, —
Jc? — v2 ot do,

ale
faa
Bp ae
ot Vet — v? dt

sepa $ (—Fe =)
c/1 — pr ot\/1 — g? dt]
where we made use of the relation 100.2, and set 6 = v/c. If we define

the foregoing equation can be written in the form

Tae
100.6 “fie treat ies:Sl, dx* E
( ) Pp c’ Ot dt
and since, in the local coordinate system Y,8B =Oandm= Mo;
ay”

100.7 Fe — Mo
( ) ec dé
= mof*
This has the form of Newton’s second law used in classical mechanics.
We see that the invariant m, is the mass of the particle P referred to a
local reference frame. It is called the rest (or proper) mass of the particle.
Sec. 101] EINSTEIN’S ENERGY EQUATION 295

Since equation 100.7 is a tensor equation, we can write the force equation
as
Fr = mf,
which is valid in all Galilean reference frames. ’
We shall rewrite (100.6) in the form

(100.8) gS (=),
where v* = dzx*/dt, and F* = c2/ 1 — f? F*, and shall take it as the
equation of motion of a particle in the restricted theory of relativity.

101. Einstein’s Energy Equation

d
We conclude our sketch of the rudiments of mechanics in the restricte
on between mass
theory of relativity by establishing an important connecti
and energy.
used in
For simplicity in writing we suppose that the coordinates x’
we recall that the work done
this section are rectangular cartesian; and
a displac ement dz‘ is equal to
by the force F;, (i = 1, 2, 3), in producing
the classical theory gives
the change in the kinetic energy. Indeed,

T_T x| mo dv =| m= 4(=)
vo v at dt
t i 72,0
= |meee a
dled

y in the restricted theory

If we take as our definition of the kinetic energ
of relativity the expression
Adz
P -

(101.1) T=|
Po
, we get®
and insert for the ¥, from equation 100.8
P i
zie Fax =| ee
P
Po Fed ig tf
{|4 ( 1 dv
jodx viSS i ta
— |———— ] v —_ em
=m
Joldt\/1—p dt dtvi—p?dat
derivative reduces to the
is cartesian, the intrinsic
6 Since the reference frame
ordinary derivative.
296 RELATIVISTIC MECHANICS [CHaP. 5

But
Phe MO) 2 v'Dto a ot zt
— ’ D = 3

P Cima. dt
hence vi(dx'/dt) = Bc?, and BB = (v'/c*)(dv'/dt). Substituting these ex-
pressions in the integral, we get

iyoa
A ree lia ="

mi l ca rt a el
gales
= myc!| (ema a

Thus
=n le)
Moc
+ constant.
e (1 i By

If we wish to have T = 0 when 6 = v/c = 0, the constant of integration

is —m,c”, so that
Mg 5
T= eae
=: By = mo|e

=(m — m))c’.
Thus

(101.2) m= Mp) a =
Cc

We see that the mass m depends on the kinetic energy. If this result is
assumed to hold in dissipative systems, then the decrease in mass m must
be accounted for by the loss of energy by radiation.’
We see from the foregoing that the principles of conservation of energy
and conservation of mass, which appeared to be quite distinct in the
classical theory, can be united into one law in the restricted theory.
We also see from equation 101.2 that, if a particle takes up an amount
of energy AT, then its inertial mass m is increased by an amount A7/c?.
"In vol. 41 (1935) of the Bulletin of the American Mathematical Society, Einstein
gave an elementary derivation of this mass-energy relation by basing his considerations
on the principles of conservation of energy and momentum. For a definitive treatment
see J. L. Synge, Relativity: The Special Theory (1956), Chapter VI.
SEc. 102] RESTRICTED ‘THEORY 288 |

Thus the inertial mass m can be considered a measure of the energy of

the particle, and the law of conservation of mass holds if, and only if,
the particle neither receives nor gives up its energy. Einstein associated
with every mass m an amount of energy E = mc. Then equation 101.2
can be written in the form
E=m,c* + T,

in which myc? appears as the intrinsic energy and T as the kinetic energy.

102. Restricted Theory. Retrospect and Prospect

In our development of mechanics in the manifold of the special theory

of relativity we maintained the distinction between the space coordinates
x. (i = 1, 2,3), of a particle and the time variable t = x4. The metric
of the space was assumed to be Euclidean. The novel features of the
theory lie in the abandonment of the concept of universal time and in
the demand that the mass of the particle change with velocity in a pre-
determined way, if the Newtonian law of motion is to be invariant with
respect to a group of Lorentz-Einstein transformations.
ed.
The distinction between the space and time variables can be suppress
alued reversibl e transfo rmation of the S, mani-
by introducinga single-v
fold,
x = £(2', x”, x, x*), (or == 15253;-4).
ed coordinates
where the coordinates Z* are quite analogous to the generaliz
space S, is so metrized
of analytical mechanics. We suppose that our
that the quadratic form
(102.1) do® = a,, dx" dx?

reduces to

(102.2) do? = c* dt? — dy? dy’,

ian. Since the coeffi-
when the space coordinates x are orthogonal cartes
s that the Riemann curva-
cients in the form 102.2 are constants, it follow
es, and hence the geodesics
ture tensor R,p,s of the Sy manifold vanish
in S,, determined by
d* x" (o\ a <= 0
(102.3) do® \fBy) do do
are straight lines.
100.5, that equations 102.3
We note, with reference to equations
the absence of acceleration f*.
characterize the motion of a particle in
ng the trajectories of particles,
This suggests the possibility of interpreti
298 RELATIVISTIC MECHANICS [CHAP. 4

subjected to the action of nonvanishing forces, as geodesics in some

manifold of the variables x for which the curvature tensor does not
vanish. Physically, this corresponds to the introduction of accelerated
reference frames moving in such a way that the forces acting on the
particles vanish. If this is done, the concept of force need not enter
dynamics, and dynamical trajectories can then be viewed as geodesics
determined by the metric properties of space.
In the remaining section of this chapter we discuss the problem of two
bodies from a general relativistic point of view. This portion of the
general theory of relativity was developed in the early 1920's, and its
mathematical elegance and success in explaining the advance of the peri-
helion of Mercury gave hope that the time when all mathematical physics
would be imbedded in the framework of the general theory of relativity
was not too far away. However, the researches of the following two
decades make it appear unlikely that general relativity will prove useful
in the domain of microscopic physics, because of the failure of the theory
to unify mechanics and electrodynamics. It is likely that the future
usefulness of the theory will be in whatever stimulus it may provide to
speculations in cosmology. These remarks do not detract from the pro-
found effect which Einstein’s paper,® setting forth the foundations of the
general theory of relativity, had on the revision of the concepts of space,
time, and matter.

103. Einstein’s Gravitational Equations

In order to conform to the usual notation in books on general theory

of relativity, we denote the metric coefficients of the four-dimensional
relativity manifold by g,,(x', x, 2°, a), and write the fundamental quad-
ratic form as

(103.1) ds? = g,,dv'dv’, (i, = 1, 2, 3, 4).

In the special instance of the restricted theory the form 103.1 can be
reduced by a suitable transformation to the canonical form

(103.2) ds? = c*(dt)? — dy' dy’.

* A similar situation arose in classical mechanics (Sec. 84), where we introduced a

Riemannian manifold, with the arc element dS of the form

dS = V 2m(h — V)g,; dv‘ de’,

in which the trajectories are geodesics.
* A. Einstein, Annalen der Physik, 49 (1916), p- 769.
Sec. 103] EINSTEIN’S GRAVITATIONAL EQUATIONS 299
Our hypothesis is that the coefficients g,,, which we will term potential
aa : ;
Junctions,” can be so chosen that the trajectories of particles satisfy
the equations of geodesics,

(103.3) dix! , [i | da? dat

The Riemann curvature tensor R',,, associated with the manifold of
restricted theory, vanishes, and the rectilinear geodesics of the manifold
correspond to the trajectories of particles in the absence of a gravitational
field. Consequently, if the manifold with the quadratic form 103.1 is to
account for nonrectilinear trajectories, the Riemann curvature tensor must
not vanish. We assume, with Einstein, that the field of a large gravitating
mass (the sun) is such that the potential functions g,; satisfy in vacuum
the equations
G; = R; al 4 7R = 0,

where G;’ is the Einstein tensor defined in Sec. 38. If we contract Gr.
we get the equation R — 34R = 0, so that R= 0. Accordingly,

(103.4) Ry = Rie = 0,
where R,, is the Ricci tensor. These equations include the flat manifold
of restricted theory and admit the case for which the components of the
curvature tensor do not vanish.

Equations 103.4 are analogous to Laplace’s equation, gV,; = 9, of

Newtonian potential theory, which is valid at all points outside gravitating
matter."
We recall? that the Ricci tensor R,, appearing in the left-hand member
of equation 103.4 is given by

0 Fe aol Gl be negative.
>

where we write |g| since the determinant of the form 103.1 may
of the coefficients in
10 This terminology can be justified by examining the form
equation 84.9 in a related problem in Newtoni an mechani cs.
making use of equations of
11 This equation is suggested by a chain of reasoning
=0, where GY? = —pu'w’ with uv‘ = dx'/dt. A delightful
motion in the form G,;
s, Mathema tics of Relativity
account of this approach is contained in G. Y. Rainich’
(1950). See also Problem 2, Sec. 38.
it can be shown that there are four
12 These equations are not independent, and
on, The Mathematical Theory
relations connecting them. See, for example, A. S. Eddingt the calculations
, has no bearing on
of Relativity, 2d ed. (1924), p. 115. This fact, however
given below.
300 RELATIVISTIC MECHANICS [CHaAP. 5

It is obvious from the foregoing that the system of ten nonlinear partial
differential equations (see Sec. 38)
Ry=0

for the ten unknown functions g;; is extremely complicated.’* The general
solution of this system is not known, and one is obliged to seek particular
solutions, essentially by trial, and use Newtonian mechanics as a guide
in selecting sensible forms for the coefficients g;,. Once a set of g,,’s
satisfying equations 103.4 is found, we can form the equations of geodesics
103.3, and if the solution of equations 103.3 agrees to the first order of
small quantities with the corresponding situations in Newtonian theory,
all is well.
We shall illustrate this procedure in Sec. 104, where we will obtain the
Schwarzschild" solution of the gravitational equations 103.4.
Before we proceed to that topic we note that equations 103.3 can be
written in a neat form,
(103.5) xd? = 0,
where #' = dx'/ds. If we regard the vector dx'/ds = i’ as the tangent
vector, then equations 103.5, or A',d’ = 0, are precisely the equations for
the parallel displacement of the tangent vector 4’ along a geodesic. Our
problem has thus been reduced to the solution of a deceptively simple-
looking system
R,, = 0,
7 ea

zt’ = 0,
with which we will occupy ourselves in Secs. 104 and 105.

104. Spherically Symmetric Static Field

We proceed to deduce a solution of Einstein’s equations

(104.1) KR = 0,
‘* It is interesting to note that-as an argument for adopting this system of equations
as the law of gravitation it is frequently stated that the law 103.4 represents a simple
relation involving the curvature tensor R},;, and hence a desirable one. A skeptic
might feel that the Creator was not greatly concerned with the simplicity of mathematical
physics.
“ K. Schwarzschild, Berlin Sitzungsberichte (1916), p. 189. See also some important
special solutions in G. D. Birkhoff’s Relativity and Modern Physics, pp. 219-227. There
is also the solution of H. Weyland T. Levi-Civita, corresponding to rotational symmetry.
See P. G. Bergmann, Introduction to the Theory of Relativity (1942), pp. 206-210. For
a comprehensive discussion of spherically symmetric fields see a treatise by J. L. Synge,
Relativity: The General Theory (1960), Chapter VII.
Sec. 104] SPHERICALLY SYMMETRIC STATIC FIELD 301

for the gravitational field produced by a spherically symmetric mass

particle, which will be shown to correspond to the gravitational field of
the sun fixed at the origin of our reference frame. In obtaining this
solution we will be guided by the properties of the Newtonian gravitational
field and by the form of the corresponding solution in classical mechanics.
The discussion of the two-body problem in Sec. 97 suggests that we
adopt as our reference frame a system of coordinates which at great
distance from the gravitating mass specializes to the ordinary spherical
coordinate system. Moreover, since the field is spherically symmetric,
and since the metric of the manifold is determined by the field, the metric
tensor g,; must be spherically symmetric. Thus we shall select the coor-
dinates in such a way that, at great distance from the center of attraction
(the origin),
i a0 a a =Tt,
where r, 6, ¢ are the usual spherical coordinates.
The trajectories of particles far away from gravitating matter should
be straight lines, so that Ri,, = 0. We write the limiting form for the
space-time interval as

(104.2) ds? = (dt)? — (dr? — P(d0)? — r° sin? 0(a4)",

¢ so that
where we have adopted a new unit for the velocity of light
a spheric ally
it is 1. This leads us to assume that, in the presence of
symmetric static gravitational field,
(104.3) ds? = fy((at? — f(dry? — °(d0)? — 7° sin? O49)”,
ng to unity when
where f, and f; are unknown functions of r, each reduci
r is increased indefinitely.
terms dr d0, dh dO, etc., are omitted in the form
The cross-product
d6 and d¢ because of
104.3 since ds? must be independent of the signs of
the cross-product terms
the spherical symmetry. Likewise, we reject
is static and reversible in time,
involving dt, since we assume that the field
of dt. Our procedure in
and hence must be independent of the sign
to insert the expressions for
determining the functions f; and fz will be
ational equations 104.1,
metric coefficients g,; from (104.3) in the gravit
ion at infinity.
and use equation 104.2 as a boundary condit
For the purpose of calcu latin g f, and f; it is convenient to set

hh = eos ho = Be

effects of the gravitational field

where J and yu are functions of r. Since
and “ must tend to zero when r
diminish as r— ©, the functions A
increases indefinitely.
302 RELATIVISTIC MECHANICS [CHaP. 5
We can write the form 104.3 in the new notation as

(104.4) ds? = —e*(dr)? — r°(d0)? — r* sin? 6(d¢)” + eX(dt)’,

so that the metric coefficients g,,; are

gu = —e’, 82=—r, 833 = —r° sin” 8, Sag =

&i5 = 0, i ¥ j:

The determinant g of the quadratic form 104.4 is

= * Ati! cin?
—e'*r’ sin’ 8,
& = 211822833844 =
and the contravariant tensor g” is given by the matrix

Ca 0 0 0

fe
Aoner may 0
(g’) = 1
0 0 = 0
r® sin? 6
0 0 0 ea,

In order to form equations 104.1, we construct the Christoffel symbols

ti}and, since g;; = 0 when i # /, we have

Wee5 w (Ses 4 har _ 28) (no sum on k).

Glo \Pela pg oe

A ee oe
It is easy to verify that distinct, nonvanishing Christoffel’s symbols are

flew ene Bem

il 12) +r 13) y

14 ae - \22 3 93). :
A = —rsin® Oe“, lsat= —sin 6 cos 6, ee a ae ie

where primes denote the derivatives with respect to r.

We can now insert these symbols in the formula

ron ER) fn
" aatdat at ij) * Maj) lial lil at”
Sec. 104] SPHERICALLY SYMMETRIC STATIC FIELD 303
and obtain after tedious but simple calculations the following set of
differential equations:

(1045) Ry = bw” — Hw’ + AW —*; =O,

(104.6). Reg = e7[1 + dr(p’ — 29] — 1 =0,
(104.7) R33 = sin” 0{e7[1 + 4r(u’ — a’‘)] — 1} = 0,

(104.8) Ris en) yu i ee z|

é = 0,
R;;==0, it ing4:J:
Equation 104.7 in this set is a mere repetition of equation 104.6. We
thus have only three equations on / and yu to consider.
From equations 104.5 and 104.8 we deduce that
Ae — —p',

so that
A = —yu + constant.
However, as r—> ©, A andy tend to zero; hence,

Ar) = —H(0).
Epuation 104.6 thus becomes
(104.9) e*(1 + ry’) = 1.
We set
ev=y,
and equation 104.9 becomes
ytry’ =1.
Integrating this first-order linear equation, we get

(104.10) et err oS
105,
where 2m is a constant of integration. We shall identify m, in Sec.
with the mass of the sun.
ns
It is easily checked that the solution just obtained satisfies all equatio
we get the
in our system. Inserting e-* =e" = y in equation 104.4,
desired quadratic form
(104.11) ds* = —y“(dr)e — (db)? — FP sin? (dg)? + y(d0)*
2m vanishes, y = 1,
where y = 1 — 2m/r. If the constant of integration
ted theory. For
and the resulting manifold is the flat manifold of restric
m 0, the manifold is curved.
304 RELATIVISTIC MECHANICS [CHAP. 5
The reader may feel uneasy about the Schwarzschild solution of
Einstein’s gravitational equations, since it was obtained on the basis of
several fortuitous guesses with one eye cocked on results of the classical
theory. He may feel that a different mode of attack might yield a different
solution. That this is not so was shown’? by G. D. Birkhoff, who demon-
strated that all spherically symmetric static solutions of the gravitational
equations R,; = 0, which yield a flat metric at infinity (that is, the one
characterized by equation 104.2), are equivalent to the Schwarzschild
solution. Thus the solution obtained previously is of interest because
it is the only static solution of our equations satisfying specified boundary
conditions at infinity.

105. Planetary Orbits

We are in a position now to determine the trajectory of a particle

moving in a spherically symmetric static field determined by the quadratic
form 104.11. The trajectory of the particle is a geodesic, so that we have
to solve the set of equations’®
d*a:* |i) dx* dx? =
>
ds” ap) ds ds
wheres: B= 75 wr Csr Ve Y
Making use of the table of values of Christoffel’s symbols given in
Sec. 104, we find that for i = 2, for example, we have the equation

ogds” [Nee
12) ds ds
de
21
IEE >

ds ds (33/ ds his. a
or
2 2
(105.1) a0, Behe aoe 6(42) =)
dss aTasias S
In a similar way we form equations for i = 1, 3,4. The results are
a 1 2 2 2 2
(105.2) an au( a) — yr (2)— yrsin® 0(22) + raat) = 0,
ds*> 2y dr\ds ds ds 2 dr \ds.
2
(105.3) as SMC Leta ery 8G las
ds" rds ds ds ds

(105.4) So a andar
2 -

ds’ y drds ds
*° G. D. Birkhoff, Relativity and Modern Physics, p. 253.
*® For an elegant treatment of planetary orbits by means of Lagrangean equations
see J. L. Synge, Relativity: The General Theory (1960), pp. 289-298.
Sec. 105] PLANETARY ORBITS 305
The last of these equations can be written

dt ldydt_
2

ds* yds ds
:
or
(105.5) a(
ay
“)
di
a
ds or
We will prove that the analytic solution of equation 105.1, satisfying
the initial condition d0/ds =0, when 6 = 7/2, is 0(s) = 7/2. Since
d0/ds = (d6/dt)(dt/ds), and dt/ds # 0, this is equivalent to showing that
the trajectory of the particle lies in the plane 0 = 7/2, provided that the
initial component d6/dt of the velocity, in the direction of increasing 6,
vanishes. We thus assume that the solution 6(s) can be represented by
the series

Ge ROO: laa) e* Gala?

dé d°0\ s°

Since d6/ds =0, when 6 =7/2, equation 105.1 for = 7/2 gives
(d?6/ds*)) = 0.
To obtain (d?6/ds*), we differentiate equation 105.1, and insert in the
result the values 0= 7/2, d6/ds = 0, and d?6/ds? = 0. We find d°6/ds* = 0.
In this manner we can show that 6(s) in (105.6) is 0(s) = (8)o = 7/2.
The corresponding result in the Newtonian case is obvious since, under
the assumption of the central field of force, there can be no component
of force at right angles to the plane of motion. Thus, if the motion had
once started in the plane § = 7/2, it would continue in that plane. If
we insert the solution 6 = 7/2 of equation 105.1 in equation 105.3, we get
2

(105.7) d biieeadr dp ih 0;
ds are rasds
and integrating equations 105.5 and 105.7 we obtain

(105.8) pet =h,

ds
t
(105.9) Lb
dS. Ay
where a and / are arbitrary constants.
using the
Substituting in equation 105.2 from 105.8 and 105.9, and
previously found solution 6 = m/2, we have
atom! e
“tig, ) | ee
(")2
(ar) pe ita) avia) 9
x dv(2)
( ds? 2y dr\ds me 2 dr\y
 306 RELATIVISTIC MECHANICS [CuHap. 5

The expression for (dr/ds)?, appearing in equation 105.10, can be obtained

from formula 104.11 by using equations 105.8 and 105.9 and 6 = 7/2.
We have

which, upon insertion in (105.10), gives

d’r =m =( 21)
105.11 Sie
a
\ Ose oe
since y = | — 2m/r. But

IE eMRela peds” Oad¢

ds") Vad “ds,
Se)
dg)”
where we made use of equation 105.8.
Thus equation 105.11 can be written in the form

(105.12) nies ~(e) +2 z(t - =m).

rid¢ r° \d¢ er
If we introduce a new dependent variable u = l/r,

dr 1 du ee an eng a
a7 aes a ahag) kag
and equation 105.12 reduces to

(105.13) oe tus i + 3mu?.

Equation 105.13 together with equation 105.8, which we write as

(105.14) Hage (ak :

suffices to determine the trajectory.

It is interesting to write down here the corresponding equations of
the classical theory obtained in Sec. 97:
2

(105.15 ) ah ok
Sec. 105] PLANETARY ORBITS 307
where we write ¢@ for the angular variable @ used in that section and
introduce the gravitational constant k = 6.7 x 10-° and

m, = 1.98 x 10* pr «

is the mass of the sun. Because of our choice of units for the velocity
of light, we note that far away from gravitating matter

ds* = (dt)? — dy‘ dy’,

so that
(eee dia eX
iy) are dt dt
For planetary velocities, v is very small compared with the velocity
of light, which we took to be 1, so that to a high degree of approximation
ds = dt. Thus, in both classical and relativistic sets of equations, h can
be interpreted as the sectorial velocity. The constant of integration m
corresponds to km,, so that the relativistic equation 105.13 differs from
the corresponding classical equation only in the appearance of the term
3mu?.
Now, the ratio of 3mu? to m/h? is 3h?u?, or using equation 105.14 it
is 3(r d/ds)?. For ordinary planetary speeds this ratio is small. For
example, the average radius of the earth’s orbit is r= 1.5 x 10" cm,
the angular velocity d¢/dt = 2- 10-7 rad/sec, and, if we take as a first
approximation dt/ds = 1/c, we find the value of 3r2(dd/ds)* to be of the
orderg Or.
Consequently, in ordinary planetary motion “the correction term” in
the relativistic equation 105.13 is negligible, as far as the shape of the
the
orbit is concerned, but the influence of this term on the behavior of
perihelion, as will be seen in Sec. 106, is significan t.
It will be shown in the next section that the perihelion rotates through
be too
an angle 6m?z/h? rad during each revolution. This value proves to
with the exceptio n of Mercury ,
small for all planets in the solar system
arc per century. This advance of
for which it corresponds to nearly 42" of
on the
the perihelion of Mercury has found no satisfactory explanation
will see that the calculat ions based
basis of the Newtonian theory, and we
results which agree extraord inarily
on the relativistic equation 105.13 give
well with observed values.
calculations
We conclude this section by remarking that, if the foregoing
were performed with the quadratic form

(dr) r°{(d0)? + sin’ 0(d¢)"]

ds*® = c*y(dt)’ —
?
308 RELATIVISTIC MECHANICS [CHAP. 5

as a basis, we would have arrived at the equation”

du a km, - 3kmyu"
dd? h? Ce

where m, = 1.98 x 10*8 gr (mass ofthe sun), k = 6.7 x 10-* gr-* cm?/sec?,
r= 3 eO cnt secs
For the motion of Mercury the term km,/h? is of the order 10°”,
whereas 3km,u?/c? is of the order 10-71. These estimates justify us in
attempting to solve equation 105.13 by a method of successive approxi-
mations sketched in the following section.

106. The Advance of Perihelion

A comparison of analytical results of this section with observed astro-

nomical data provides us with the best available evidence in support of
the general theory of relativity. In Sec. 107 we mention the deflection
of the light beam by the sun and the shift of the Fraunhofer lines toward
the red end of the spectrum, but the quantitative agreement for these
phenomena between observations and theoretical predictions is still in
some doubt.
The relativistic equation for the orbit of a planet

d°u
(106.1) +u= ne + 3h*u’),
d¢”
deduced in Sec. 105, can be integrated in closed form with the aid of
elliptic functions, but the solution obtained in this way does not lend
itself to a convenient comparison with the corresponding result obtained
in Sec. 97 on the basis of the Newtonian theory.
We noted in Sec. 105 that the magnitude of the term 3/?u?, appearing
in the right-hand member of equation 106.1, is small compared with
unity, and this justifies us in attempting to obtain a solution of this equation
by the method of perturbations. Accordingly, we neglect the small term
3mu* and obtain for our first approximation wu, the Newtonian equation

au, m
ples gate
the solution of which is

(106.2) “= [1 + ecos(¢ — w)],

*” In this equation the sectorial velocity h is the sectorial velocity of the classical theory.
SEc. 106] THE ADVANCE OF PERIHELION 309
where e is the eccentricity of the orbit and @ is the longitude of the peri-
helion. Inserting from equation 106.2 in the right-hand member of
equation 106.1 yields
du m Ape -
(106.3) dé? +u= jas + 3h°u,)

m . 6m°
= ie + i ecos(¢ — w)

ame. 3m?
ieee creat aie Ae ewes

Since planetary orbits are nearly circular (for Mercury, e* = 0.04),

the contribution of the perturbation term containing e? will be negligible.
Also the term 3m?/h! will not have a significant effect on the shape of the
orbit, but the second term, containing cos(# — @), may have a pro-
nounced cumulative effect on the displacement of the perihelion. Accord-
ingly, we simplify equation 106.3 to read
d°u m 3

SP Ss + SF e cos ($ — 0).
d¢” h*
The solution of this linear equation is clearly made up of the solution
u, and the solution of

du
d¢”
3
u = ST ecos ($ — 0).

up in
The result of easy calculations gives us the second approximation
the form

(106.4) w= ml
2
+ ecos(¢ — w)+ we ed sin (d — 0) }

steps in
It will suffice for our purposes to terminate the sequence of
and to regard u,
the scheme of successive approximations at this stage
ntly high degree
as representing the solution of equation 106.1 to a sufficie
of accuracy. If we set
3m?
(106.5) bo = 72

and note that

cos (¢ — w — &),
cos (¢ — w) + dw sin(d — ©) = V1 + (da)?
(106.4) as
where « = tan-! dw = 6, we can write

(106.6) Up us e [1 + ecos(¢ — w — do)I,

310 RELATIVISTIC MECHANICS [CHAP. 5

if we neglect in comparison with unity terms of the order (dw)’. It is

clear from equations 106.5 and 106.6 that when a planet moves through
one revolution, the perihelion advances through an angle
2

(106.7) a a omrad.
Equation 106.6 represents a closed orbit, only approximately elliptical
in shape, because dw is a function of ¢. Since u = 1/r, we have

TiAl 2
> Eee es
1+ ecos(¢
— w — da)’

so that the “‘semilatus rectum” / = h?/m.

Recalling from the geometry of conics that / = a(1 — e?), where a is
the major axis of the conic, we get
h? = ml = ma(1 — e?).

Inserting this result in equation 106.7 we have’®

_ 64m? 67m
"alae
a es
In this expression m is the mass of the sun.
For Mercury the quantity « works out to be 4.90 x 10-* rad. This
angle is very small, but the observational data on the location of Mercury
during the last century are available, and since this planet has a period
of 88 days, it completes 415 revolutions per century. Thus the cumulative
advance of the perihelion in 100 years should amount to 415e = 2.04 x
10-4 rad = 42" of arc. For planets other than Mercury the corresponding
advance is too small for accurate experimental determination. Thus for
Venus it is only 9”, for Earth 4”, and for Mars 1”.
The actual path of Mercury about the sun is not an ellipse, of course,
because of the perturbing effects of other planets. We are not in reality
dealing with a two-body problem. However, perturbations due to other
planets can be taken into account and the deviations from an elliptical °
path calculated. Such calculations have been performed with great care,
and it has been found that the advance of Mercury’s perihelion should
amount to about 42” of arc per century. The Newtonian theory is unable
to account for the advance of this amount, and the remarkably close
““ For a different way of deducing the value of « see J. L. Synge, Relativity: The
General Theory (1960), pp. 294-296, and G. Y. Rainich, Mathematics of Relativity
(1950), p. 162.
Sec. 107] CONCLUDING REMARKS 311
agreement between the relativistic calculations and the best observed
value can hardly be viewed as fortuitous.!®
It is worth noting that the calculations based on the restricted theory
of relativity also give a precessional effect when one assumes that a
particle moves in a field of force with potential V = km/r. However,
the precession based on such calculations yields results that are not as
close to the observed value as those furnished by the general theory.

107. Concluding Remarks

We conclude this chapter with a mention of the relativistic prediction

of deflection of light rays by the sun and of the shift toward the red
end of the spectrum of spectral lines of light originating in dense stars.”°
Since light is material in nature it must be affected by the gravitational
field of the sun, and the deviation from the rectilinear path of the light
ray from a distant star, as it grazes the sun, can be readily calculated.
The deflection of light rays passing near a large mass can be observed
during eclipses of the sun when fixed stars in the apparent neighborhood
of the sun become visible. However, because of the uncertainty about
the magnitude of experimental errors arising from the difficulty of ob-
taining sharp photographic images, it is generally conceded that these
results neither prove nor disprove the general theory. It may be remarked
that the calculations based on Newtonian theory of gravitation can be
made to account for about one-half of the observed values.
Among other experimental evidence cited in favor of the general theory
is the observed displacement of spectral lines of light emitted from the
stars toward the red end of the spectrum. Elementary considerations
indicate that the frequency of vibration of the emitted light from a distant
star is less than the corresponding frequency on the surface of the earth.”
If this frequency is associated with the emitted light from the sun, the
lines of the solar spectrum should be shifted slightly toward the long-wave
end of the spectrum as contrasted with the corresponding lines of terrestrial
19 (1947),
19 G. M. Clemence gives 42”.56 + 0.94 in Reviews of Modern Physics, vol.
General Relativity and Cosmology (1956). These
p. 361, See also G. C. McVittie,
l astronomical
authors make incisive comments on the difficulty of performing meaningfu
observatio ns.
and Secs. 36
20 See J. L. Synge, Relativity: The General Theory (1960), pp. 298-308,
Y. Rainich’s Mathemat ics of Relativity (1950). See, also, P. G. Bergmann,
and 37 of G.
Chapter XIV, and A. S. Eddingto n,
Introduction to the Theory of Relativity (1942),
Theory of Relativity (1924), pp. 90-93. A critical survey of the validity of
Mathematical
General Relativity and
predictions of Einstein’s theory is provided by G. C. McVittie,
Cosmology (1956) .
21 See references given in the preceding footnote.
312 RELATIVISTIC MECHANICS [CHAP. 5
spectra. The expected shift for the light emitted by the sun is very small,
but for the companions of Sirius it is estimated to be about thirty times
as great as for vibrating solar particles and should be observed with a
reasonable accuracy. In 1925, Adams measured the “red shift” for the
companion of Sirius”? and found it to be AA = 0.27 for the line of wave-
length A = 4000 A. From this determination the diameter of the star
can be estimated, and it is found to be of the right order of magnitude.
The evidence here is not conclusive, but it is generally regarded as
favorable.
The law of gravitation R;; = 0 was generalized by Einstein to the form
R,; = 4g,;, where A is a small “universal constant.” Solutions of the
generalized equation have led to various cosmological theories and have
given rise to speculations about the expanding universe. We refer the
reader for detailed accounts to specialized treatises on this subject.”
*? The shift of the corresponding line in the sun’s spectrum is calculated to be
AA = 0.008.
8 A. Eddington, Mathematical Theory of Relativity (1924).
R. C. Tolman, Relativity, Thermodynamics and Cosmology (1934).
P. Bergmann, Introduction to the Theory of Relativity (1942).
G. Y. Rainich, Mathematics of Relativity (1950).
L. Landau and E. Lifshitz, The Classical Theory of Fields (1951).
J. L. Synge, Relativity: The Special Theory (1956).
J. L. Synge, Relativity: The General Theory (1960).
 6
MECHANICS OF CONTINUOUS MEDIA

108. Introductory Remarks

This chapter contains a general formulation of the basic concepts of

mechanics of continua and a derivation of the fundamental equations
governing the behavior of continuous media. The treatment contained
here forms a substantial introduction to nonlinear mechanics of fluids and
elastic solids. The linearized equations of classical theory appear as
special cases of nonlinear equations, and throughout the chapter emphasis
is placed on the unified formulation of equations of mechanics of continua
in the most general tensor form.
A systematic development of tensor calculus, with an eye to applications
to mechanics of continuous media, is contained in P. Appell’s definitive,
Traité de mécanique rationnelle, vol. 5 (1926), and in A. J. McConnell’s
Applications of the Absolute Differential Calculus (1931). These are largely
concerned with the linearized cases. The landmarks in the domain of
nonlinear theory of elasticity are papers by Leon Brillouin, “Les lois de
Lélasticité sous forme tensorielle valable pour des coordonnées quelcon-
ques,” Annales de physique, 3 (1925), pp. 251-298, and F. D. Murnaghan,’
“Finite Deformations of an Elastic Solid,” American Journal of Mathe-
matics, 59 (1937), pp. 235-260. The essence of Brillouin’s contributions
appears also in his book Les tenseurs en mécanique et en élasticité, first
in
published by Masson et Cie in 1938, and reprinted by the Dover Press
1946.
y are
Among more recent contributions to nonlinear theory of elasticit
Foundat ions of Non-Lin ear Theory of
the books of V. V. Novozhilov,
Theoretical
Elasticity, Moscow (1947), A. E. Green and W. Zerna,
A. Signorin i, Question i di Elastici ta non
Elasticity, Oxford (1954), and
exhaust ive critical survey of the founda-
Linearezzata, Rome (1960). An
tions of elasticity and fluid mechanics is contained in two extensive
tions will be found
1 A brief exposition of the central ideas of Murnaghan’s contribu
Calculus (1947).
in Chapters 14 and 15 of A. D. Michal’s Matrix and Tensor
313
314 MECHANICS OF CONTINUOUS MEDIA [CHAP.6

memoirs by C. Truesdell in Journal of Rational Mechanics and Analysis,

vol. 1 (1952) and vol. 2 (1953).
A development of the foundations of continuum mechanics, primarily
within the framework of linear theories (including applications to me-
chanics of fluids, elasticity, and plasticity) is contained in W. Prager’s
Introduction to Mechanics of Continua, Boston (1961). A general unified
development of geometrically and dynamically nonlinear mechanics of
continuous media will be found in an excellent monograph by L. I. Sedov,
Introduction to Mechanics of a Continuous Medium, Moscow (1962).
Sedov’s monograph to a large extent is based on the close union of classical
mechanics and macroscopic thermodynamics. This unification permits
one to construct the general models of gases, liquids, elastic and thermo-
elastic solids, and of several types of plastic media from a single point
of view.

109. Deformation of a Continuous Medium

We consider a continuum of identifiable material points which at a

given time-t = fj fill a certain region of space t). We shall refer to f) as
the initial time and shall call 7) the initial region. With the passage of time
the points P of 7, undergo displacements and at some time ¢ fill a certain
region 7. In the course of displacement, the initial region 7» is usually
deformed, and we suppose that the deformation of 7) into 7 is fully
determined when the motion of every point P is known. To describe the
motion of points P we introduce a coordinate system X which moves with
the medium in such a way that the coordinates (21, x?, 2°) of any given
point P initially in 7) do not change with ¢. In addition to the system X
we consider a fixed reference frame Y, relative to which the coordinates of
the point P(a', x?, x3) are given by
(109.1) a? = yi(al, a, obg).
The functional form of relations (109.1) clearly depends on the nature
of deformation of rt) into 7. We shall assume that the functions y‘(z, f) in
(109.1) are single-valued, piecewise smooth, and possess for each value of
time ¢ a single-valued, piecewise smooth inverse :
(109.2) a = x'(y', y®, y’, 2).
The fixed coordinate system Y, without loss of generality, can be assumed
to be orthogonal cartesian.
A material point P in 79, relative to an orthogonal cartesian frame Y, is
determined by the position vector (Fig. 51)
(109.3) t= cys = c,y'(x, a, x, to),
Sec. 109] DEFORMATION OF A CONTINUOUS MEDIA 315

y}

Fig. 51

where the c, are the orthonormal base vectors associated with the frame Y.
The location of the same point P in the region 7 is determined by the
vector

(109.4) Tia 04/(ad a eed).

The base vectors b, in the moving frame X are given by

(109.5) gat ey Me
ES)
Ox?

and these vectors obviously depend not only on the coordinates 2’ of P,

but also on t. When P(z!, a, x?) is in 79, we denote the base vectors b; by
a; so that
or dy'(ax, to)
109.6 a,=—t =e.
—— Ov Ox’

Thus, in analyzing the deformation of a continuous medium, we can speak

of three reference frames: a fixed reference system Y determined by the
basis c,, a moving reference frame X with the basis b;, and a fixed reference
frame X with the basis a;. We emphasize that the labels (x1, x, x*) of a
given material point P in both curvilinear coordinate systems X have the
316 MECHANICS OF CONTINUOUS MEDIA [CHAP. 6

same values, but to avoid circumlocution we shall denote the point

P(x1, x2, «*) when it is located in the initial region 7) by Pp.
Let Py’ be a point in the neighborhood of Pia nea) ed ne wector
errres
P,P) = dr, can be represented in the form

(109.7) dt; =a; ax

and the square of the arc element ds, in 7, is

(ds,)* = dr, - dty — a, - aj ae dx’,

or

(109.8) (ds)? —= hi det ax,

where h,; = a;+ a, are metric coefficients in ty. Similarly, the square of
SS
the element of arc ds determined by the corresponding vector PP’ = dr =
b, dx’ in 7 is
ds* = b, - b, dz’ dx’,
or
(109.9) ds? = g,, dx dx’,
where the g,; = b,- b; are metric coefficients in +. Ordinarily the lengths
and the orientations of vectors dr, and dr will be different, and we shall
say that the medium occupying 7 is strained whenever ds) # ds. We can
take as our measure of strain the difference
(109.10) (ds)? — (dso)? = (g,; — h,,;) dx’ dz’,

and, if we set

(109.11) 25 — hy = 2¢,;

we can write (109.10) as

(109.12) (ds)? — (dso)? = 2e,; dx‘ dx’.

Since (109.12) is an invariant and €;; = e;,, we conclude that the set of |
functions ¢,,(x, ¢) represents a tensor Ey with respect to a class of ad-
missible transformations of coordinates X, with the basis a,, covering the
region tT). The same set of functions ¢;,(x, t) also determines a tensor E
with respect to a set of transformations of coordinates determined by the
basis b; of the final state 7. In the notation of the concluding paragraph
of Sec. 45, the tensor E, is specified by the multilinear form E, = e,,a‘a’,
whereas the tensor E is determined by E = ¢,,b’b’. Thus the operations of
covariant differentiation and those of raising and lowering indices on the
Sec. 110] GEOMETRIC INTERPRETATION OF FE, AND E 317

components of E, involve the metric tensor h,;, whereas the corresponding

operations on E make use of the tensor g;;. Accordingly,
’

j j = j
aj — = E's
Cin = €&k and Lex
h

However, the two sets of functions «,’ so computed in general are distinct,
and to indicate the origin of the set €,’ obtained with the use of the tensor
h,;, we shall write
hei, = ox

It will be shown in the following section that either of the tensors E, or

E can serve to characterize the state of deformation of the neighborhood
of Py. Tensors Ey and E are sometimes called, respectively, the Lagrangean
s of
and Eulerian strain tensors in accordance with the two viewpoint
or
hydrodynamics, associated with the choices of coordinates of the initial
amical
final states as independent variables in the formulation of hydrodyn
equations.

110. Geometric Interpretation of Strain Tensors Ey and E

ns «,; by the
In the preceding section we defined the set of functio
formula
(ds)? — (dso)? = 2¢€;; dx’ dx’,
where
(110.1) 26, = 83 — Na
) as
Since g,, = b,-b, and A, = a,- a;, we can write (110.1
(110.2) Qe, = by by — a; 8;
= |b,| - |b,| cos 6,; — |a,| - la, cos OF,
s b, and b,, and 6;,° is the
where 6,, is the angle between the base vector
e the change in length per unit
angle between a, and a,. If we denote by
51, so that
length of the vector dr, = PoP in Fig.
aes . .

pi ldtlisa |dro| = ds — dS
|dro| dSo
we have

(110.3) \dr| = (1 + €) lar.

we see from (110.3) that the elon-
We call e the elongation of dro and
rs a; are given by
gations e, in the directions of base vecto

(110.4) |b;| = (1 + e,) lad.

318 MECHANICS OF CONTINUOUS MEDIA [Cnap. 6

However, |b,;| = A fie and |a,| = Shite so that

(110.5) Vu = (lee ed/ hey (no sum on i),

and hence formula 110.2 can be written with the aid of (110.4) and (110.5)
as
2€;; 5
110.6 —" __. = (1 + e,(1 + e,) cos 6,; — cos G;,.
on V hiehss ;
Since 0,,° = 6,; = 0 for i = j, equation 110.6 yields
2€;3 ~
7% at =>
(
1 -- e;
) Te 1.2

or

(110.7) a= a1 + = — 1, (no sum).

When the coordinates of the initial state are rectangular cartesians,

h,; = 1, and we see from (110.7) that for 2e;;/h;; « 1,e,;= €,,;. Accordingly,
the functions «1, €g9, €33 are related to the elongations of arc elements
directed along the base vectors aj, a, a3.
The significance of the «,; for i ¥ 7 follows from (110.6) on noting that
when a, and a, are orthogonal unit vectors, 6;,° = 7/2. If we set 6,; =
7/2 — a,;, So that «,, represents the change in the initially right angle
between the pair of arc elements directed along a; and a;, formula 110.6
gives
2¢e,, = (1 + e)(1 + e;) sin a,
or

(110.8) sin gi = fae ees >

wil a 2€e/1 + 2e;;

where we recalled (110.7). If 2e,;<« 1 and the angle «,; is small, we have
an approximate equality, «,;— 2e¢,;. Thus the functions ¢«,; for i #/j
provide a measure of the decrease in the initially right angle between the
arc elements parallel to the vectors a; and a;,. The components e,; for
i ¥ j are called shearing components of the strain tensor Ep, and the com- .
ponents e,; for i = j are the normal components of E,.
Quite analogous interpretations can be provided for the functions «,;
when these are viewed as components of the tensor E = e,,b’b’. Indeed,
if we now define the elongation e as the change in length per unit final
length |dr| of the arc element so that
ds — dsp
C=.
ds
Sec. 111] STRAIN QUADRATIC 319
the calculations similar to those that have led to formulas 110.7 and 110.8
now yield

(110.9) ee
Sui
and

(110.10) sin Ba; = pe eee ' (no sums),

V1 = 2ein/L = 2e,,
where B,; = 9,,° — 7/2.
We conclude, as before, that the components e,; in (110.9) are associated
with elongations of the arc elements originally parallel to the base vectors
b;, whereas the components ¢;; for i #j measure the corresponding
shearing deformations.

111. Strain Quadric. Principal Strains

The defining formula 109.12 for components ¢;; of the strain tensor
E = «,,;b'b’ can be written as

(ds)? — (dso)” _ oe dx‘ dx’

(111.1) 2(ds)? "ds ds°
the
where dz'/ds = 2' is the unit vector determining the direction of
to determi ne those directio ns A for
vector dr in the final state. We seek
which (111.1) takes on extreme values. Accordin gly, we set

(111.2) QO(A) = «,;4°

relation
and maximize the quadratic form Q(A) subject to constraining
H(A) = gui? — 1 = 0,
requiring that 2’ be a unit vector.
values of (111.2)
The familiar procedure for determining the extreme
iers leads to the system of equations
by the method of Lagrange multipl

00 —e hs = 0,
on on
or

(111.3) (€,; =e eg; = 0,

where « is the Lagrange multiplier.

for 4’ if, and only if,
This system possesses nontrivial solutions
le,(a) — egi(@)| = 0
320 MECHANICS OF CONTINUOUS MEDIA [CHAP. 6

at each point P(x) of the region r. In order to reduce this system 111.3 to
the form 13.10 considered in Sec. 13, we multiply (111.3) by g*, sum on i,
and. obtain
(111.4) (c* — <d)# =0,
where

(111.5) ej = gike,;.
The system 111.4 has three nontrivial solutions 2/,), Aig), Ais) @ = 1, 2, 3),
corresponding to the roots e, of the cubic
(111.6) je? — €6;'| = —F + He — Be + 0; = 0.
The coefficients #,; in this cubic are the invariants

B, =e, + 2 + €s,
(117) Oy = €n€5 + €3€) + €1€0,
Oy = €1€n€5.
It was shown in Secs. 13-15 that the roots e; are necessarily real and
the directions 2{,), Aig), Ais) associated with them are orthogonal.
The quadratic form 111.2, where we regard the 4’ as the running co-
ordinates, reduces to the canonical form

(111.8) Oy) = aly")? + x(y")” + €(y*)”,

provided that the principal directions ii,), Az), Ai) are chosen as the base
vectors of a suitable orthogonal cartesian reference system Y in 7.
We can interpret these results geometrically by introducing a strain
quadric
(111.9) €;;(x)A'? = constant,

which, at each point P(«), represents a quadric surface with the 4‘ as the
running coordinates. The principal directions 2',, coincide with the axes
of the quadric 111.9, and it follows from (111.8) that the strain tensor e¢;,,
when referred to the frame Y, has the form

€y 0 0

0 €.. Dale
WET Tips
From the geometrical significance of components «;;, i # j (see equation
110.10), it follows that the principal directions are those orthogonal directions
in the undeformed state which remain orthogonal after deformation.
The strains €,, €,, €; are termed the principal strains.
Sec. 111] STRAIN QUADRATIC a2i
The invariants #,, 35, #3 defined by (111.7) play an important role in
the construction of models of continuous media. If we expand the
determinant in (111.6) and equate the coefficients of like powers of € in the
result, we find a

ye 2 pee, er:
O,=& “I €5 + €5 =e,

2 2 3 3
Si SG 4 €g € Swe a ceo 1 ml ek oe
en is od Por eal ot 2| = 5)Omer a>
icing
€ 1
(111.10)
2 3 3
1 8
en he
2 2 a (pues Sat,
Ds Sai Gq Coun Cg || ae 3!
Oe
apy~t
€ ~jpee
3 3 3 :
ey E5 €3

We will see in the following section how these invariants enter in the
expression for the ratio of the volume elements dry and dr of the initial
and deformed states.
We could have equally well considered the quadratic form

(111.11) 0, = €,jAg' Ay’,

with A,’ = dzx'/ds, specifying the direction of the vector dro of the initial
state, and with ¢,,’s regarded as components of the strain tensor Ey =
€ja'a’.
For the determination of principal directions we now have the set of
equations of the type 111.4 in which

(111.12) ef he);

and the values of « are the roots of the characteristic equation

(111.13) le, — €d;*| = 0,

can thus
in which the ¢,* are given by (111.12). The quadratic form 111.11
be reduced to a canonical form

O= OP Sg tae €o(Yo)”
€y°(Yo')® + aM

the basis of a
when the principal directions Aj,1), Jia), Aiyg) ate taken as
suitable orthogonal frame Yo in 7».
tions e,° along the
It follows from formulas 110.7 that the elonga
principal directions are

(111.14) of = ad
dS
e T+ 268 — 1,
$22 MECHANICS OF CONTINUOUS MEDIA [CHAP. 6

whereas the elongations e,, reckoned per unit length in the final state
(cf. equation 110.9), are

(111.15) Zot ee eee ees

ds*
We conclude from (111.14) and (111.15) that
0
(111.16) e,° ad
€;
and SS
€;
=
1 — 2«; 1 + 2e,°
Formulas 111.16 permit us to express the invariants #,° of the cubic
(111.13) in terms of the invariants #, given in (111.7), and the invariants
, of the cubic (111.6) in terms of the #,°.

Problem
Show that
O° + 49° + 128,°
se 29,9 + 495° + 89,9’
ore J. + 605°
2 1 +2049 + 492° + 895°’
ov 3
03 = 0
1 = 2),° = 43,° + 804° :

112. Distortion of Volume Elements

We investigate next the change in volume elements dr, and dr of the

initial and deformed states and indicate its connection with the invariants
&@, introduced in Sec. 111.
It follows from the definition of the volume element in Sec. 44 that

dt, = nf.h dx da? dz? and a e g da) dx? dx’,

where h = |h,;| and g = |g;,| are the determinants of the quadratic forms
dsj? =h,,dx' dvi and ds? = g,, dx’ dx’.
Thus

(112.1) — = Vh/g,
The set of functions h,,(x) can be regarded as components of the tensor
H = h,,b’b’ defined in the space of the variables 2’ in the final state, so that

g" hy = hi
and
Sixhj* = hj.
Sec. 112] DISTORTION OF VOLUME ELEMENTS Eve
We conclude that
|gixh;"| = |h,,|,
so that
g |h,;'| = bh. 7

Consequently the ratio 112.1 assumes the form

dto Meg cay

(112.2) — = V{h;'|.

But from definition 110.1 we have

h,(«) = g;(x) — 2¢€;,(x),

which, upon raising the indices, reads

h,-= 6,' — 2e;'.

We can therefore write formula 112.2 as

dt
(112.3) V6; i 2e;|.
dr a

If we expand the determinant appearing under the radical sign, we find

(112.4) |6,¢ — 2e,/| = 1 — 20, + 48, — 80,

where the #; are the invariants 111.10.

be
In the linear theory of deformation the products of strains €,' can
ion for the ratio 112.3 is
disregarded, so that an approximate express

free
=
1 20,
— 1 = O,.

Thus approximately

(112.5) i A
dr
volume, and, for this
This represents the change in volume per unit
figure s promi nentl y in linear elasticity
reason, #, is called the dilatation. It
and hydrodynamics.
Formula 112.3 can be cast in the form

dt 1
(112.6) —————————— — oo

V1 + 20,° + 40. + 895°

06 +2€9|
—

dr
324 MECHANICS OF CONTINUOUS MEDIA [CHAP. 6

by expressing the invariants #; in terms of the #;° as in the Problem of

Sec. 111. When the deformations are small, it follows from (112.6) that

dt — dt) go 0
(T1237)
dry :
and, since for small deformations ¢,° = «;,, 3,°=%,, both formulas
112.5 and 112.7 give the same value for the dilatation.
Problem

Obtain formulas 112.3 and 112.6 directly from (111.14) and (111.15).

113. Displacements in Continuous Media

We define the displacement vector & of the point Py (Fig. 51) by

(113.1) o=r—7,

and denote the components of & relative to the basis a; by u‘ and its com-
ponents relative to the basis b; by w’. Thus

(11372) FE = wa; c= eb

From (113.1) we have

abe 2cgk Orsy nae
Oxi Ox dxf
so that

(113.3) b; = a; + =

On computing g,; = b;-b; with the aid of (113.3) and subtracting

h,; = a,° a, from the result, we find

(113.4) 0B ok +a, 0&8

6 — hy =e = ae0
zi ES en ; ee Ox"
= 2e¢,,, by (110.1).
Equations 113.4 can be regarded as a set of differential equations for the
components of § when the functions e;; are specified. This set of equations
assumes quite simple form when the displacement vector & is expressed in
terms of its covariant components u; or w;, so that
(113.5) E=ua’, §§=w,b’,
the a’ and b’ being the reciprocal base vectors introduced in Sec. 45.
Sec. 113] DISPLACEMENTS IN CONTINUOUS MEDIA 325
On differentiating (113.5) with respect to x’, we get (cf. Sec. 45)

(113.6) ge = U5),8’, ge itv b’,

a" Ox' :
where “
Ou; k
113.7 A Selita
( ) ok Ox’ Aid=

is the covariant derivative of u; with respect to the metric h,; of the initial
state and
Ow; k

( ) ie Ox" tida

is the covariant derivative of w; with respect to the metric g;; of the final
state. The prescripts on the Christoffel symbols in (113.7) and (113.8)
indicate that these symbols in (113.7) are constructed from the tensor hi;
whereas those in (113.8) from the g;,’s.
If we insert from the first of formulas 113.6 in 113.4, we get
1
2€,5 = (uj\.a 2 Up) 58") + (ag aun; dias ay) ;)
a Lk ke k
= Uj {Uz,)ja ° a + 6; Unis 7 0; Uni
I Ujk Ugg + Vays + Uji
since a’.a® = hh”
Thus

(113.9) e.g = Ugyg + Uys t UfMas-

On the other hand, when & is represented in the form § = w;b’, we recall

m
oe (petelee)
(113.3) and write

Ox' Ox’?

The substitution into this expression from the second of the formulas in
(113.6) yields
(113.10) 2€;3 = Wig. t Wis — WhWae, 5:
¢«;; from
Formulas 113.9 enable us to compute the strain components
to the basis a, of the initial state.
components u; of the vector & referred
ents of E relative
Formulas 113.10, on the other hand, involve the compon
s €,; are
to the basis b, of the final state. Alternatively, when the function
differen tial equatio ns serving to
specified, equations 113.9 and 113.10 are
determine the components of the displac ement vector =
326 MECHANICS OF CONTINUOUS MEDIA — [Cuap. 6

When the reference system X is orthogonal cartesian, we set y' = 2"

and obtain from (113.9) and (113.10)

_ Ou; , Ou; , Ou, Ou,

(113.11) 2<;;
Oyo? — Yo’ Yo’ Ayo?
Ow; , Ow; Ow, OW,
(113.12) 2€,, =
ay? |ay’ — dyt dy?”
where the labels y,’ refer to the cartesian coordinates in the initial state.
In special problems the derivatives of the displacement components may
be sufficiently small to justify one in neglecting products of these derivatives
in comparison with the first-order terms in these derivatives. In this event
equations 113.11 and 113.12 become linear and the theory of deformation
based on a study of resulting linear differential equations is called the
linear theory. In the linear theory it is usually assumed that the dis-
placement vector § is so small that one is justified in identifying the co-
ordinates y,? and y’ of the initial and final states. The resulting theory is
called the infinitesimal theory of deformation. In the infinitesimal theory,
formulas 113.11 and 113.12 coalesce and we write

(113.13) 2e;; = Uz; + U;,;

where the e,, are the infinitesimal components of the strain tensor e,;. In
classical theory of elasticity, the strain tensor e,; is taken in the form
(113.13). The strain invariant 3, = e,, + es, + e353, as follows from
(113.13), is then equal to divergence of the displacement vector u,, and
hence the dilation 3, = (dr — drt))/dr = u',.

114. Equations of Compatibility

Equations 113.10 or, in cartesian form, 113.12, can be viewed as a

system of six simultaneous partial differential equations for the deter-
mination of three components of displacement from prescribed values of
the strain tensor. Clearly, if a solution of this system is to exist, com-
ponents of the strain tensor cannot be specified arbitrarily. To ensure the -
integrability of the system it is necessary to impose certain restrictions on
the choice of functions ¢;;. Such conditions were deduced and the proof
of their necessity,” for the linearized case typified by equation 113.13, was
given by B. Saint Venant in 1860. We indicate here how these integrability,
or compatibility, conditions can be deduced in the general case.
* For a proof of necessity and sufficiency of Saint Venant’s conditions see I. S.
Sokolnikoff, Mathematical Theory of Elasticity (1946), pp. 24-25.
Sec. 115] ANALYSIS. OF STRESSED STATE o27
We recall that the space in which the deformations take place is Euclid-
ean, and hence the Riemann tensor, associated with the metric of Euclid-
ean space specified by ds)? = h,; dx’ dx’, vanishes (see Sec. 39). Thus

Chee © 0 MSC) Cay 1%lil

(>|
(114.1 ) Ree
kU A— (il,i i] — Ba! Lik, i] + \jk [il, x] f lik, a] — 0,

where the Riemann tensor R},, is formed from the metric coefficients h,;.
If we recall that (see 110.1)
hoo ae DE
tj fone) ij?

compute the Christoffel symbols needed in (114.1) in terms of the g,; and
e,;, and make use of the fact that the Riemann tensor R;;,, based on the
g,;8 also vanishes, we get the condition
ae
(114.2) €ijet + h (€ 5xp ita — € 51p€ixa) = 0,

where
Cpr fuutee ot Ck,dl,Ty Stick et
ijn = “in,31 e5,4 — “3.0
and

Mae
ap
>

h
H* being the cofactor? of h,, in |h;,|.
If we linearize (114.2) by dropping terms involving the products of the
€ijn. We get Saint Venant’s compatibility equations

(114.3) Cxj.nr + &nr.is — Cie,st — sran = 9;

familiar in the linear theory of strain.*
has
From the fact that in a three-dimensional space the Riemann tensor
that there are six in-
six independent nonzero components, if follows
dependent equations in (114.2) and (114.3) .

115. Analysis of Stressed State

is natural to use
In analyzing the state of stress in a deformed body, it
indepe ndent variabl es. We will
the variables x* of the final state as the
P(x) of a body, in equilibrium
demonstrate that the state of stress at a point
tensor of h,; with respect to
3 Note that the contravariant tensor h’ is the associated
the metric tensor gi;. See Sec. 30.
n Mathematical Monthly,
4 See in this connection a paper by W. R. Seugling, America
han, Finite Deformation of an Elastic
vol. 57 (1950), pp. 679-681; also F. D. Murnag (1962),
cs of a Continuous Medium
Solid (1951) and L. I. Sedov, Introduction to Mechani
pp. 128-130.
328 MECHANICS OF CONTINUOUS MEDIA [CHAP. 6

xe T do

under prescribed surface and body forces, is characterized by a symmetric

tensor, the stress tensor.
Let a body 7 be referred to a curvilinear coordinate system X, and
consider an element of surface area at some point P’ of the body. Leta
small tetrahedral volume element dz be formed by the coordinate surfaces
at a nearby point P and by the surface element do (Fig. 52). Ifv is the unit
normal to do then the elements of area do; lying in the coordinate surfaces
are given by the formulas
(iiss) do; = v; do,
where the »; are the covariant components of v.
We denote the stress vector (force per unit area) acting on do by T
where the superscript » brings into evidence the dependence of the stress
vector on the orientation of the element do. The stress vectors acting on
i]

the surface elements do; are denoted by T, and we take as their positive
directions the directions of the exterior normals to the volume element.
We can write
(115.2) T = —71b,,
where the b, are base vectors directed along the coordinate lines and the
v

7" are the contravariant components of T.

Now, if F = F'b, denotes the force per unit volume acting on the mass
contained in dr, the first condition of equilibrium requires that

(115.3) F dr + Tdo + Tdo, = 0.

Sec. 115] ANALYSIS OF STRESSED, STATE 329
If we note the definitions 115.1 and 115.2 and observe that dr = / do,
where / is the appropriate factor depending on the linear dimension of the
volume element, the equilibrium condition 115.3 becomes
» £b,l do + T’b; do — rv, dob; = 0, ,
where T’b; = T.
If the point P’ is now made to approach P so that the direction of »
remains fixed, /—» 0, and the first term in the above relation will surely
vanish whenever the body force F is bounded. This leads to the result that
the components T” of the stress T, acting on a surface element with the
orientation vy, are given by the formula
(115.4) Ti = rly,
Since T’ is a vector and »; is an arbitrary covariant vector, we conclude
that the r” are the contravariant components of a tensor, the stress tensor.
Formula 115.4 permits us to calculate the stress vector acting on a surface
element with the specified orientation whenever a set of nine functions 7”
is known. We will see in Sec. 116 that the application of the remaining
condition of equilibrium leads to the conclusion that the stress tensor is
symmetric.
We can obviously write (115.4) in the form
(1.1555) Spare i he

The component N of the vector T in the direction of the normal v is

T-v = 7,’, so that, using (115.5),
(115.6) N =i7,,V'%
In regard to the quadratic form 115.6, we can raise the question of
determining directions v’ such that N takes on the extreme values. As in
Sec. 111, this leads to the consideration of the characteristic equation
(11521) [7,6 — 76;'| = —7? + 0,7? — Opt + Og = 0,
where
OQ, = 7 + Tz + Ts,

Oz = T2T3 + 7371 + 71725

O3 = T1727,
s »’
the 7, being the roots of the cubic (115.7). The orthogonal direction
the set of
corresponding to the principal stresses 7; are determined from
linear equations (cf. equation 111.4)
(115.8) (7;* — 76;")v’ = 0,
330 MECHANICS OF CONTINUOUS MEDIA [CHAP.6

and are called the principal directions of stress. If we choose an orthog-

onal cartesian frame Y whose axes coincide with the principal directions
at P, the quadric surface

(115.9) 7,’ = constant

assumes the form

(115.10) T1(y))? + 72(y?)? + 73(y*?)? = constant.

The quadric surface 115.9 was introduced by Cauchy, and it is called the
stress quadric.
It is obvious from (115.10) that the components 7,;;, for i ¥ j, vanish
when a suitable reference frame is chosen at P. The components 7},
T22, T33 are called the normal components of stress, and the remaining ones
are shears.
By analogy with formulas 111.10, we can write down the expressions
for the stress invariants ©,. They are

(115.11) O,=7, O,= = Outersn) On= = OnpyTi

TS TH

116. Differential Equations of Equilibrium

Let a body 7 be in a state of equilibrium under the action of prescribed

body and surface forces. Since every portion of the body is in equilibrium,
the resultant of all forces and the resultant moment of these forces acting
on every subregion V of 7 must vanish. The condition that the resultant
force in every direction vanishes yields the equation

(116.1) i Fi), dr +| TA, do = 0,

Vv S

where 4, is the unit vector in an arbitrarily fixed direction.

We assume that the components of body force F(x) are continuous
functions and that the components T° of the stress vector are of class C1.
The substitution for T' from (115.4) and the application of divergence,
theorem 92.3 to the surface integral in (116.1) yields the equation

| tra + @%), Jar =0.

Since A; is a parallel vector field, A; ;= 0, so that the preceding equation
can be written
(116.2) { (Fi + 7%)A,dr = 0.
V
Sec. 116] EQUATIONS OF EQUILIBRIUM 331
Since the integrand in (116.2) is continuous and the direction of A, is
arbitrary, we conclude that, at every point P of 7,
(116.3) vi + Ft = 0. “

We apply next the condition that the resultant moment of the body and
surface forces vanishes. If r= /'b, is the position vector of the point P’(«)
relative to some point P, the component of the moment (F x r) dr in the
direction of the unit vector Ais F x r- A dr. The component of the moment
due to the surface forces T is T x r-Ado. Recalling (Sec. 49) the ex-
pression for the triple scalar product
ACB EC: a Apic*
enables us to write

| EeE'lia® dt + i éieT'l'*do = 0.
i S

The substitution in the surface integral from (115.4) and the application of
the divergence theorem yields

| eae le a) 2|dro 0,
V
SINCE €;4,m = 0.
If we carry out the indicated covariant differentiation and make use of
equations 116.3, we get
i Ak Loh OT = U,
Vv

and, since’ /’,, = 63, and V is arbitrary, we conclude that

€,4Tidk = 0.
(116.4)

Noting that €;, = —€jix enables us to write this in the form

(116.5) he,g(t? — T)A¥ = 0.

Since €,;, = V g ijn, and J g ~ 0, we have, upon expanding (116.5),

(723 — 732)A2 4 (731 — 718)A? + (7? — 771)A? = 0.
Inasmuch as the direction of A is arbitrary, we conclude that
(116.6) (oa ler nh
Thus the stress tensor is symmetric.

5 For: by = = = [',p, by (46.6). Hence I; = 65. i cis

Q
7
3o2 MECHANICS OF CONTINUOUS MEDIA [CHAP. 6

We summarize these results in a

THEOREM. Jf a body is in equilibrium, under the action of prescribed
body and surface forces, then the components of the stress tensor 7” at each
point of the body satisfy the system of partial differential equations
ae +t =),

where 7? = 7. On the surface X of the body where stress vectors T' are
assigned,
7), le

y, being the exterior unit normal ot X.

We can write down at once the equations of motion by invoking the
principle of D’Alembert. We merely have to add to the body force F'
the inertial force —pa’, where p is the density and a’ is the acceleration.
Thus the equations of motion are
(116.7) Ty + F* = pa',
where F’ is the body force per unit volume. If F’ represents the force per
unit mass, the equations of motion read
(116.8) 4 = p(a’ — F’).

Since all equations in Secs. 115 and 116 appear in tensor form, they are
valid in all admissible reference frames. In particular, in the reference
frame X of the initial state 7), the covariant derivatives in (116.8) are
taken with respect to the metric coefficients /,; and the a’ and F’ are com-
ponents of the acceleration and force vectors relative to the basis of the
initial state.

117. Virtual Work

Let a continuous medium be maintained in the state of equilibrium by

—>
the body forces F' and surface forces T’. If PoP = &(x, t) is the displace-
ment vector of the point Pin (Fig. 51), we can consider a point P’ in the
—_> %

neighborhood of P, and denote the vector P,P’ (not shown in Fig. 51) by
Ge yetine
B(x, t) = Ga, t) + OE(@, 0),
where the variation
(117.1) 6B =F —&
ed
or the virtual displacement of P, is an arbitrary vector PP’ in the neighbor-
hood of P. We consider the variation of vectors only in the final state +
Sec. 117] VIRTUAL WORK 5
and we shall say that the variations of vectors and tensors associated with
the points Py of the initial state 75 is zero.
We suppose that & is of class C? and define the variation of 0&/dx’ by

(117.2) (S) Se een eG) eh

Ox' 2 Ea Fe Ox" Ox’
so that the variation of the derivative 0&/0x' is equal to the derivative of
the variation (cf. Sec. 81). Since § = r — ro, and

LES AS MUL — a;
fi oie Or .
we obtain, on utilizing the distributive property of the symbol 6,

6 (=)
Ox"
= (bam a.) = dbp
for da, = 0, since points in the initial state are not varied. Thus

airy wo, = 0(28) = 208)

Ox" Ox"

The metric coefficients of the final state are given by g,; = b,*b; and
we find, as in Sec. 81,

ogi; = 6(b; - b;) = b,;- 6b; + b;° ob,,

so that
(117.4) 6g;;= b; +e bp, : by (117.3).

The strain tensor e,; was defined by

[109.11] 2€,, = £3 — Ny
and hence
(P17.35) 26¢€,;; = bLii

since 6(h,;) = 6(a;° aj) = 9.

On substituting from (117. 4) in (117.5), we get
0(08) 0(08)
(117.6) 26e,; — b; a
Ox) a Oxt *
components of the
But 6& = (68),b’, where the (08), are the covariant
vector 6&, and, since
A) =- (68),,,b'
334 MECHANICS OF CONTINUOUS MEDIA [CHAP. 6

by (46.8), we conclude that (117.6) can be written as

(117a7) 20€;; a (08); ; a (08); 4

If we form the inner product ofthe vector (0&),; with both members of the
equilibrium equation
(117.8) 74 = —pF",
where F’ is the body force per unit mass (cf. 116.3), and integrate over the
body 7, we get

(117.9) |(08) ;d7 = -| pF'(0&), dr,

But 74(68); = [7"(68),],; — 77(68),;, so that (117.9) can be written as

[Basar — |408). .dr = — |FOR) dr,

or

i768) ,»; do -| a (0b). dt = — |pF (08), dr,

where we transformed the volume integral over 7 into the surface integral
over surface & bounding vr.
Since ry; = T’ by (115.4) and

7(8E),4 = 3771008): 5+ (08),] = TY Oey

by (117.7), we have finally

(117.10) ir8e,, dr =/) T'(0§); do +| pF'(0&),

dr.
By definition, the surface integral in (117.10) represents the virtual work
performed by the external surface forces J’ in a virtual displacement (68),.
The volume integral in the right-hand member of (117.10), on the other
hand, represents the virtual work done by the body forces F*. If we denote
the virtual work done by body and surface forces by

CPE ieLL) OW =i T' (08); do +| pF'(0&), dr,

we can write (117.10) as :

CLigei2) OW = | Oe,;dr.

If in the foregoing calculation instead of the equilibrium equations (117.8)

we considered the dynamical equations,
[116.8] ri = p(a‘ — F),
Sec. 117] VIRTUAL WORK 335
we would have obtained in the left-hand member of (117.10) the additional
term

(117,13) OK =| pa‘(6&); dr. .

This term has a simple mechanical interpretation when the virtual dis-
placements (68), are the actual displacements (d&), that take place in a body
whose motion is governed by equations 116.8. In this event we write
(117.3) as

(117.14) dK =f pa‘(d&),
dr.

But the velocity of a point P in 7 is

= Bs,
dt
and we can thus write (117.14)

dK -| pa'v, dr dt.

Now, in the orthogonal cartesian coordinates,

; Ne, ro 1 d(v)?
SACS
tae AC
and hence
dK = i4d(v)*(p dz).
The integrand in this integral represents an increment in kinetic energy of
time
the element of mass dm = pdr acquired by it in the interval of
Thus dK represents an increment of kinetic energy K =
(t,t + dt).
equations
pv? dr. Accordingly, for the motion of a body 7 governed by
116.8, we have an important result:
(117.15) dK + dA = dW.
where
(117.16)

dA= |7 de, dr etand= ~aW = I,T'(d&), do +] pF'(d&), dr.

In the static case dK = 0 and dA = dW.

thermodynamic con-
The results of the section, coupled with some
the theore tical models of
siderations, form the basis for constructing
elastic bodies, viscous fluids, and so on.
336 MECHANICS OF CONTINUOUS MEDIA — [Cuap. 6

118. Laws of Thermodynamics

The construction of mathematical models of different types of continuous

media hinges on the use of certain energy concepts that enter in the
structures of mechanics and thermodynamics. We borrow from mechanics
the notions of potential and kinetic energy and from thermodynamics the
somewhat less sharply defined concepts of chemical energy, heat energy,
electrical energy, and so on. We shall suppose that functions defining
various kinds of energies depend on a number of parameters, some of
which are variables (positional coordinates, temperature, densities, strain
tensors, and so on), whereas others are physical or universal constants.
The totality of constant parameters c, and variable parameters g‘ chosen to
describe a given function need not be unique. But whatever particular
choices of a set of parameters is made, we shall assume that the g' (i = 1,
. ,”) are independent.
In some special situations the g’ may be determined as functions of a
scalar ¢ (usually time) so that one can regard them as defining a curve
C:. g'= q'(t),
characterizing a certain process.
In the preceding section we introduced the notion of work or mechanical
energy by considering linear forms of the type
(118.1) ONG Ogee.2 a Cie oe, Co) OG

The line integral [a dq’ then represents the work done along the path
y
C by the generalized forces Q,;. Ordinarily, such integrals depend on the
path C associated with a given process.
We shall suppose that a particle of mass dm = p dz, where p is the density
and dris the volume element, may acquire energies other than mechanical,
and we shall represent such accretions of energy in the form
(118.2) OE = ELG Soon, Gas uer ee wy Ce). 0g"
If dE includes all energies other than mechanical, the total amount of
energy acquired by the particle is determined by the integral ;

(118.3) le(OW+ OE) =i (F, + Q,) dq’.

From the principle of conservation of total energy we conclude that the
integral 118.3 must vanish for an arbitrary closed path C, and hence the
integrand (F; + Q,) dq’ is an exact differential of some function Ta sentats
q"; Cy+++,5C,) determined to within a constant of integration. We shall
Sec. 118] LAWS OF THERMODYNAMICS ao1
call U the total energy per unit mass and define the internal energy U per unit
mass by the formula

(118.4) (i fh tle «

where v is the velocity of the element of mass dm. The amount K = 3v”
represents the kinetic energy per unit mass, so that the total energy

OK:

We can thus formulate the basic law of conservation of total energy in

the form

(118.5) 6K + 6U = 6W + OE,

where the left-hand member in (118.5) is the sum of the increments of

kinetic energy K and internal energy U acquired by the unit mass.
When OE consists only of the heat energy 6Q, we have the statement of
the First Law of Thermodynamics:
(118.6) 6K + 6U = 6W + 60.
The heat energy 6Q, as shown in works on thermodynamics, can be
determined by specifying the temperature 7. Experiments show that heat
invariably passes from bodies with higher temperature to those with lower
is
temperature and that the transfer of heat from one body to another
parameter s
wholly determined by T and, of course, by certain physical
depending on constitutive properties of the bodies. Experiments further
the
show that it is impossible to construct a machine which transforms
temperatu re.
heat energy into mechanical energy from a body with the least
every
It is a consequence of this Second Law of Thermodynamics that for
S, called entropy,
reversible thermodynamic process there exists a function
such that

(118.7) 6S dm = 2
of heat acquired
T being the absolute temperature and 5Q an increment
by the element of mass dm.
brium, the kinetic
When the medium is in the state of mechanical equili
the form
energy K vanishes, and the law (118.6) assumes

(118.8) 6U = 6W+4+ 00.

118.8 in Sec. 119 to construct a
We shall make use of the laws 118.7 and
mathematical model of an elastic body.
338 MECHANICS OF CONTINUOUS MEDIA [CHAP.6

119. Elastic Media

Some bodies possess the property of recovering their original size and
shape when the impressed forces producing deformations are removed.
The media of which such bodies are composed are called elastic. In con-
structing a model of an elastic body we shall suppose that all processes
taking place in such a body are reversible, but we do not assume that the
body is necessarily in the state of thermal equilibrium. Thus our thermo-
elastic model will take account of the effects of temperature on defor-
mations.
As our points of departure we take the First Law of Thermodynamics in
the form [cf. (118.8)]

(119.1) 6U = 60+6W

in which

(119.2) OW =|. de,;dt

is given by (117.12).
We also write the relation 118.7 in the form

6Q =| TOS dm
or 7
(119.3) dQ =|rosp dr.

If uw denotes the internal energy U per unit mass of the body, then

(119.4) dU =| du dm =p du dr,

where OU stands for the increment of internal energy acquired by 7.

The substitution in (119.1) from (119.2), (119.3), and (119.4) gives

(119.5) [pdu dt =| eros dt + [> de,; dr.

We suppose that the integrands in (119.5) are continuous functions and,

since the equality 119.5 holds in an arbitrary subregion of 7, we conclude
that

(119.6) du = TOS + PS de;;

‘ P
at all points of r.
Sec. 119] ELASTIC MEDIA 339
The formula 119.6 suggests that we regard wu as a function of the in-
dependent variable S and of the nine independent parameters e,;. Since
the components e,; of the stress tensor Ey = €,,a'a’ usually depend on the
choices of a coordinate system X, the function uw may also contain
explicitly the metric tensor /,; and the coordinates x’. And, of course, u
must depend on an assortment of parameters {c} associated with the
physical properties of the medium. Thus we are led to consider uw in the
form
(119.7) U = this, Ere yp 4Cts te)
where the arguments of u are deemed independent. The relation (119.6)
then permits us to assert that
OD. s10 ress ou
= and —=
OG; p chy
The first of these relations

(119.8) bive Nite

de

connects components 7’ of the stress tensor with components e;; of the

strain tensor. It thus yields a set of stress-strain relations, in which the
internal energy density wu serves as a potential function.
A different potential function can be constructed by defining a function
¢, known as the free energy, by
(119.9) o=u— TS.

From (119.9), the increment dd of ¢ is

6d = du — TOS — SOT

and the substitution in this expression for du from (119.6) gives

(119.10) 55 Sere aes'o 7:

pP

Because of the appearance of de,; and 67 in the right-hand member of

and
(119.10), we are now led to regard T and e,, as independent variables
consider ¢ in the form [cf. (1 19.7)]

(119.11) b = (hij, €i5T, {Cc}, v’)-

We conclude, then, from (119.10) that

340 MECHANICS OF CONTINUOUS MEDIA [CHAP. 6

so that the stress-strain relation now has the form

a rs)
(119.12) Ti=p ae

Thus either uw or ¢ (when they exist) can be used to deduce the stress-
strain relations. When the process is adiabatic, S = constant and hence
6Q =0 by (119.3). It is then more convenient to use u as a stress-
potential. In the isothermal case, T = constant and ¢ appears to be more
suitable.
We say that an elastic medium is homogeneous whenever the coordinates
a do not appear explicitly in (119.7) or (119.11). The medium is isotropic
when all parameters in the set {c} are scalars, so that the values in {c} are
independent of the choices of the reference frames ¥. When the medium is
both homogeneous and isotropic, the parameters {c} have constant values
throughout the medium.
If we consider a homogeneous elastic medium and suppose that ¢(e€,;, T)
is an analytic function of the e,,; and of AT = T — Ty, where Ty is the
temperature of the initial state, we can expand ¢ in powers of e,; and AT.
When the initial state of the body is that corresponding to «,;; = 0 and
T,; = 0, the expansions will begin with the second-order terms, so that

$ = cM lee, + k%ejAT + x(AT? + ---

For small deformations, the terms of order higher than two can be
neglected, and we obtain with the aid of (119.12) a linear stress-strain
relation that includes the effects of temperature on the stress tensor 7”.
It is
(119.13) ri = pice, + k(T — Tp],
where we replaced p by p)—the density of the initial state—and wrote
€,, for the linearized components e,,. The tensor c’’*! characterizes the
elastic properties of the medium and the k” are related to the coefficients of
thermal expansion. For a given medium the tensors c”’*! and k”% must be
determined from experiments. When T = 7y, the relation 119.13 reduces
to the familiar generalized-Hooke’s law of linear elasticity,®
(119.14) ri = clikle
In the next section we deduce a special form of stress-strain relations for
large deformations for a homogeneous isotropic elastic medium and get
from it the familiar Hooke’s law of linear theory of elasticity.
* See I. S. Sokolnikoff, Mathematical Theory of Elasticity (1956), pp. 58-67, where it
is shown that the number of independent elastic coefficients c‘i*' in the most general
anisotropic case is 21.
Sec. 120] STRESS-STRAIN RELATIONS 341

120. Stress-strain Relations in Isotropic Elastic Media

When the orientation of coordinate axes is immaterial, the arguments of

the potential ¢ in (119.11) are scalars or tensors that depend only on the
metric tensor /,;. In this event the scalar invariants of tensors h,; and «,;
can be considered as functions of the invariants ?;, defined in Sec. 111,
and can be taken in the form
d= eh), Ua Vs 15 (ch 2).
If the medium is both homogeneous and isotropic, ¢ assumes the form
(120.1) ee Clee al eae Malads
in which all parameters in {c} are constants. The formula
ri 0
[119.12] Ti =p Mas
d€;;

with ¢ specified by (120.1) can be written as’

(120.2) tj) = p(d,' — 2«,) ae

de,”
where 7,’ = g,,;7°* and €;; = Zix€;"-
If we now suppose that ¢ in (120.1) with T = constant can be ex-
panded in a power series in the #; and consider the case when there is no
initial stress, so that 7,’ = 0 when e,’ = 0, the expansion takes the form
(120.3) poh = C12 + CoP. + Cg913 + CHD, + C503 + °°°-
If in this expression we retain only the terms of third order in the «,’, we
see from (120.2) that the expression for the stresses 7,’ in terms of the strains
e;' will contain five elastic coefficients c,, From the mass-conservation
principle it follows that
Po 47) = pdr,
7 Note that
Gpaeree”
aS
veel ee a
©
0€;5 de,” 0€;; p
Since

Sap = 2€ap — hap» Gey Ben =

and conclude that
Compute 0c,¢/0¢,; from €gg = Suyep’> use the above result,
den”
are = Qe," A IL
0€;;

(a).
Formula 120.2 then follows on substituting this result in
342 MECHANICS OF CONTINUOUS MEDIA [CHAP. 6

and formulas 112.3 and 112.4 yield the result

p => reel = 20; -- 4, — 805

= poll 0 oa 30,” + 204),

if we discard the third-order terms in the ¢,’. The substitution from this
formula and (120.3) in (120.2) gives the following expression for the stress-
strain relation, where we retain only the second-order terms in the strains
(120.4) 7;¢ = [2c,0, + (3c3 — 2c1)01" + C499] 6,
(Geet (Ca Co) 94] Ove? as 4c, ye; ;

25 3C5 Ojbtey"€ a — 2Cs One.

a

These involve five elastic constants. If, however, we retain in (120.4) es

the first-degree terms in the «,’, we get the linear law
(120.5) T;) = (2c, + ¢2)9, 0;' — c2¢;'.
We identify this result with the generalized Hooke’s law for isotropic media
(120.6) 7; = Ad, 0; + Qype, 0, = e,',
where A and uw are Lamé’s constants, related to Young’s modulus E and
Poisson’s ratio « by
5 Eo E
A= ——————__, p= ——_.
(1 + o)(1 — 20) 2(1 + @)
We see that
cy = 3(4 + 2), Co = —2u.
If we replace c, and c, in (120.4) by these values and set c; = /, cy = m,
Cs =n, we can write it in the form
(120.7) 7; = [A0, + 31 + m — 4d? + md] 6;
+ [2u — (m+ 24 + 2u)d,]e,; aa due,'e;* = nds;,
where ¢,' is defined by the formula
: 1
$= 39, meer°

The new elastic constants /, m, and n appearing in (120.7) are subject to

experimental determination, just as Lamé’s constants 4 and wu are.8 .
* Assumptions, of varying degrees of plausibility, about the possible relations that
might exist between the new constants (/, m, m) and the old ones (A, 44) have been made
by several authors. Murnaghan obtained a good agreement with experimental results
(for solids subjected to high hydrostatic pressures) by setting / = m = n = 0 in formu-
las 120. A discussion of this appears in a Peper by F. D. Murnaghan, ‘The Compress-
ibility of Solids under Extreme Pressures,” Th. v. Karman Anniversary Volume (1941),
pp. 112-136. See also P. Riz, Comptes rendus (Doklady) Acad. Sci, U.R.S.S., 20 (1938),
and P. M. Riz and N. V. Zvolinsky, Journal of Applied Mathematics and Mechanics,
Acad. Sci. U.S.S.R., 2 (1939).
Sec. 121] EQUATIONS OF ELASTICITY 343

An excellent discussion of a model of a thermoelastic isotropic medium

is contained on pages 234-241 of L. I. Sedov’s Introduction to Mechanics
of a Continuous Medium, Moscow (1962).

121. Equations of Elasticity

If we write the stress-strain relations 120.6 in the form

(21,1) Tig = AgiO + 2ye;;,

where 0 = ge,; = e;', and use the equilibrium equations 116.3 in the
form
(121.2) gry, + F; = 0.
we can write down the linearized differential equations of equilibrium, in
terms of the displacement vector u’, by recalling that (equation 113.13)
(121.3) C5 = (Ui, + Y;,,)-
The computation proceeds as follows. The substitution from (121.1)
into (121.2) yields

o* (teu a 210.4) + F,;=0,

x
or

(121.4) A— + Que e;;, + F; = 0.

Ox"
But from (121.3)
= 4g”*(Uy, ix + U5 ix)
100
a

since gu; ,= uy, and ul, = 0. Thus (121.4) becomes

a0 ;
(121.5) A+twait pe yg, + Fp= 0.

If we recall the notation 92.7,

SU ne= V7Ui,
we get
ao
(121.6) A+, + uV'u, + F, = 0.
in the classical theory of elasticity.
These are the celebrated Navier equations
344 MECHANICS OF CONTINUOUS MEDIA [CuapP. 6

The equations of motion,

0d 2 t.
(121.7) (A+ #) ant + pV"u, + F; = pay
x
follow at once from (121.6) upon application of the D’Alembert principle.
The differential equations 121.6 and 121.7 for the displacement vector
u; can be shown to yield unique solutions when suitable boundary and
initial conditions are specified. We refer interested readers to treatises on
the mathematical theory of elasticity where such boundary value problems
are discussed in detail.®

122. Fluid Mechanics. Equations of Continuity

We now turn to the formulation of equations governing the flow of

liquids and gases. From the point of view of mechanics, fluids are con-
tinuous distributions of matter which cannot support shearing stresses
when at rest. If follows from this definition that the stress vector T’ on a
surface element do of a fluid at rest is normal to the element. In symbols,
I? =) =p",
where v’ is the unit normal to the surface element and p(21, x*, x, t) is the
invariant called the hydrostatic or fluid pressure. In general the pressure p
is a function of the time ¢ as well as of the coordinates 2’.
Since the vector J’ is expressible in terms of the stress tensor 7”, and
vy’ = gy, we see that
Tt = ry, = —pg"y,.
Hence

(12221) 7! = —pg",
It follows from (122.1) that the hydrostatic pressure p is related to the
stress invariant @ = g,,7"’ (see equation 115.11) by the formula
(122.2) p= —eyjr'.

When the fluid is set in motion, however, in addition to the normal

stresses, new oblique stresses, produced by the interaction of moving
particles, arise. For instance, if a fluid at rest is placed between two large
parallel plates and one of the plates is caused to move parallel to the other
plate (Fig. 53), the fluid particles adhering to the moving plate transmit
® See, for example, I. S. Sokolnikoff, Mathematical Theory of Elasticity, New York
(1956).
A. E. H. Love, A Treatise on the Mathematical Theory of Elasticity, Cambridge
(1927).
SEC, 122) FLUID MECHANICS 345

——,

VLLLILLLLLL LLL LLL LLLILLIA LL LLL LLL LLL LL

Fig. 53

their momentum to the particles in the interior. In this way the fluid
between the plates is set in motion, and experiments show that the re-
tarding force per unit area of the plate, exerted on the plate by the fluid,
is proportional to its velocity and inversely proportional to the distance
between the plates. The proportionality constant in this relation is the
measure of viscosity of the fluid.
We shall say that the fluid is viscous if the stress tensor for a fluid in
motion has the form

zi = —pg" =- 7,
(122.3)

where the nonvanishing tensor ft” is the tensor of viscous stresses. The
fluid is called ideal if t’ = 0.
The mass-conservation principle of mechanics requires that

dm = po a7) = p dr,

where po(2, to) is the density of matter in the volume element dt) =
Vh dx! dx? dx® of the initial state and p(2, 1) is the density in the volume
element dr = J/g dx! da dx at time t. Thus

(122.4)
which we can also write as

= dt»
= pl®, to) _ dt a
_ p(&, t)cee
(122.5)
t) — p(%, to),
The numerator in the left-hand member of (122.5), Ap = p(x,
Ar = t — fo,
represents the change in density in the small interval of time
whereas the right-hand member

— dt. go
dt py (112.7)
dt
346 MECHANICS OF CONTINUOUS MEDIA [CHAP. 6

is the corresponding small change in volume per unit volume. Since

0,9 = div & = u', (see Sec. 113), we can write (122.5) as

(122.6) —_-f,.-.-#
and since u* = v’At, where the v’ denote the components of velocity d&/dt,
we conclude from (122.6) that —(1/p)(dp/dt) = v';. We thus get the
continuity equation in the form

(227) 1d dtig ; i)
thal
pdt
We recall that p(z, t) is a function of the coordinates x’ in the reference
frame X in which the x’ are independent of t. If Y is a fixed reference
frame (cf. Sec. 109) in which the coordinates y’ of the particle are given by

ye = (a, 27, 23, 2),

then the chain rule of differentiation gives for p(y, t)

—dp = (2)
{— Op
ee
dt Ot /yifixed dy’
Equation 122.7 then assumes the form
Opa eens
—F 4 —F vi + pri, =0,
Dihendivlaty bert
or

ap .
(T2988 ) ed ‘) =0.
y + (pv'),

In this formula, the covariant differentiation is performed with respect to

the metric tensor in the frame Y and v‘ = dy'/dt. Formula 122.8 special-
izes to (122.7) when the system Y moves with the particle.

123. Ideal Fluids. Euler’s Equations

In this section we deduce a set of equations governing the behavior of

ideal fluids. We recall from Sec. 122 that in an ideal fluid the stress tensor
has a simple form
(123.1) ri = —pgit,
in which the scalar p is the pressure.
On substituting from (123.1) in the general dynamical equations
[1 16.8] p(a’ — F*) = rT,
SEC. 123] IDEAL FLUIDS 347
we get three Euler’s equations
(123.2) pla’ — F*) = —g"p,,
or in vector form,
p(a — F) = —grad p.
Equations (123.2) involve five unknowns: the density p(«, t), the pressure
p(x, t), and three components of velocity v‘(x, t), since a’ = dv'/dt. The
system of three equations (123.2) for the determination of the five un-
knowns, thus, is not complete, and we need two additional independent
equations to complete the system. One such equation is the continuity
equation

(123.3)
1
nant ; ae):
pdt
deduced in Sec. 122. The remaining equation, known as the equation of
state, is furnished by the thermodynamical equations 118.7 and 118.8,
which, for reversible processes in the fluid, we write in the form

(123.4) dQ ==) (OWS ANTE

i
(12325) dU =dW+ dQ.
The work
a de,
(123.6) dw = —— dm,
p
performed by internal stresses 7 on an element of mass dm = p dz can
be written as [cf. (119.2)]
uj

(123.7) Fj te
LC
pe at
when we note equations 123.1.
We shall suppose that the components e,; of the deformation tensor ina
fluid are so small that they are represented with sufficiently high accuracy
by linearized formulas 113.9 or 113.10, which coalesce in the infinitesimal
theory. Thus we write
2e55= Wig + Wy,
and
de;; d
2—2 = —(w;.; + W;.1)
dt ai! STE
com-
where the e,; are the linearized components ¢,;. Since the velocity
conclud e from the equation just written that
ponents v; = dw,/dt, we
de;; I
(123.8) B(0;,5 + 03,1):
dt
348 MECHANICS OF CONTINUOUS MEDIA _ [CuapP. 6

The substitution for de,,/dt from (123.8) in (123.7) then gives

dW = — "vi,dtdm,
P

and, since v'; = —(1/p)(dp/dt) by (123.3), we have

dWw= LP atam = -pa(*) dm.

p dt p
On making use of this result in (123.5), we get for the amount of heat dQ
acquired by an element of mass dm,
1
(123.9) dQ=dU+ pa(*) dm.
p
On the other hand, formula 123.4 states that
(123710) dO'= TdS dm.
If we let dq stand for the change in heat per unit mass, so that dg = dQ/dm,
and denote the change in internal energy U per unit mass by du, we can
write (123.9) and (123.10) as
1
(123.11) te
=du+ pa(*),
dg= Tuas.
In a variety of problems the absolute temperature T, the internal energy
density u, and entropy S appear to depend only on the pressure p and
density p, so that!®
(123.12) T=T(p,p), S=S(p,p), u=u(p, p).
It follows then from (123.11) that 7, S, and u are not independent, since
equations 123.11 require that

(123.13) TdS = du pa(?),

p
When 7, S, and w in (123.12) are determined (either experimentally or
from theoretical considerations), the differential equation 123.13, if
integrable, specifies pas some function of p: :

(123.14) p =f).
Equation 123.14 is the desired equation of state needed to complete the
system of four equations 123.2, 123.4 for the determination of the five
unknown functions v', v?, v3, p, and p.
*° These functions may (and usually do) depend on physical or chemical constants
which characterize properties of a specific fluid.
SEc. 124] VISCOUS FLUIDS 349

124. Viscous Fluids. Navier’s Equations

When viscous fluid is in motion, the components 7,; of the stress tensor
have the form
(124.1) 7) = —pg + 2%,
where the t”’, as noted in Sec. 122, are associated with viscous stresses.
As in Sec. 123, we limit ourselves to the consideration of small displace-
ments and write formula 123.8 as

(124.2) 6: = 3(0,5 + 2;
where é,; = de,,/dt are components of the strain velocity tensor.
The construction of models of viscous fluids and the formulation of
complete systems of equations now call for the introduction of additional
assumptions about the nature of viscous stresses. The latter must ob-
viously depend on the strain velocities é;; and, to a first-order approxi-
mation, it is natural to suppose that
(124.3) pi = ciklg
The coefficients c’”*' are the coefficients of viscosity, which depend on the
properties of a specific fluid under consideration. The linear law (124.3)
is quite analogous to the generalized Hooke’s law (119.14).
If the fluid is both homogeneous and isotropic, the number of inde-
pendent viscosity coefficients reduces to two, and the relation 124.3
assumes the form [cf. (121.1)]
(124.4) 13 = Avi
gt + Que",
where 4 and yw are constants and v'", is the divergence of the velocity field.
Accordingly, the complete stress tensor which includes the effects of
viscosity and hydrostatic pressure can be written in the form
(124.5) Ti; = —P8ig + Bis + 2Meis,
where # = v', = gié,,.
We recall next the equations of motion 116.8, and write them in the
covariant form
(124.6) Brie = pla; — Fi),
and substitute in (124.6) for the 7,; from (124.5). The result" is Navier’s
equations offluid motion
(124.7) (A+ md, + mein — Px = PCAs — F;),
dv' aut dv
11 Note that a i= —Raima;
= — + p'F yi, so that a; = i
ar v;,;0?.
350 MECHANICS OF CONTINUOUS MEDIA [CHAP. 6
or, in vector form,
(A+ u)V3 + uV2v — Vp = pla — F).
The set of three Navier’s equations 124.7 involves five unknowns: vt"(z, f),
(i = 1, 2, 3), p(x, t), and p(x, t). To complete the system, we adjoin (as in
the case of ideal fluids) the equation of state and the continuity equation
123.3. For incompressible fluids dp/dt = 0, and hence v'; = # = 0 by
(123.3). Accordingly, for incompressible fluids, equation 124.7 yields
(124.8) MEV; i — p= pla; — F;).
Furthermore if the fluid is ideal, « = 0 and equations 124.8 reduce to
Euler’s equations (123.2).
Stokes simplified equations 124.7 by introducing a hypothesis to the
effect that the mean pressurep in a viscous fluid is given by the same formula
122.2 as in the case of fluids at rest. This assumption leads to the con-
clusion that the constants A and w are not independent. Indeed, from
(124.1)
tig = Tig + P8is;
hence
Bolg = 8 ta Pees
= —3p + 3p = 0,
if we use formula 122.2. But since 1,; is given by (124.4), we have upon
multiplying those equations by g”,
Agitg,0 + u2g%é,, = 0,
or

(3A + 2u)d = 0.
Thus
(124.9) 3A + 2u =0.
As a consequence of this relation, equations 124.7 depend only on one
viscosity coefficient “, and the substitution from (124.9) in (124.7) yields
the set of Navier-Stokes’s hydrodynamical equations
0;
(124.10) Ue vzp + i 98 = ees = p(a; os F;).
SOs eos
If the fluid is ideal we get, on setting w = 0 and
Ov; ‘
a; = — a v; iv’,
a
the Eulerian hydrodynamical equations

(124.11) EP a ee
Ot NPL
Ot p Ox" =
for ideal compressible fluids.
Sec. 124] VISCOUS FLUIDS 351
If the motion is slow, the term v,,v’ can be disregarded, and then
a = Ov'lor.
Problems

1. Show that the equation characterizing an incompressible ffuid can be

written in the form
a Vg v)

i
? =
garg
— —— =
— Oaladwbereis. le
0 /h c= Pile

2. Show that the Navier-Stokes equations can be written

ay? fi dv! e av! { | avt

hak ik
srituert S ax) dak x Ik| Ox) Es Uj} ax | jk) eet

in
14 We
posi leaee es ed old lagu ka bare
where » = //p is the kinematic viscosity.
3. Show that the equation of continuity can be written

a . Apv')
apn ant + pv Aver
_, Alog Vg = 0

Hint: Use the expression for vt, in Problem 1.

4. Show that the equation of continuity in cylindrical coordinates
[gu = 1, 222 = (a:})?, 533 tan 1)

Op A(pv*) yi
a :
or ox?

and in spherical polar coordinates [gy, = 1, g22 = (a1)?, g33 = (x1)? sin? x*] is
a Apu’) 2v}
— “Sa at v2 cot xe] = 0.
pr aire te)
5. The curly of the velocity field v is equal to twice the angular velocity of
rotation. The vector w such that curly = 2w is called the vorticity vector.
Show that w', = 0. Hint: wt = —}e'!0;». .
6. If the vorticity vector w* = 0, the motion is called irrotational. Show that,
if the motion is irrotational, the velocity vector v is the gradient of the velocity
potential ®.
7. Write out the approximate equations of motion of a viscous fluid when the
motion is slow.
352 MECHANICS OF CONTINUOUS MEDIA [Cnap. 6

125. Remarks on Turbulent Flows and Dissipative Media

We conclude our brief survey of the elements of mechanics of continua

with a few remarks on turbulent flows of fluids and on construction of
models for media, in which the processes are irreversible.
Fluid flows in which the velocity components v‘ experience complicated
pulsating changes are called turbulent. In dealing with turbulent flows of
liquids and gases it is natural to represent the velocity components in the
form v' = o' + v’, where &' is the mean value of v’ over a suitable period
of time and v’ is the pulsating component of v’. Similar resolutions into
mean and pulsating components can be made for the pressure p and density
p, so that p =p +p’ and p= p+ p’. The development of the theory
of turbulent flow crucially depends on the character of averaging proc-
esses used to compute é', f, and p and on the formulation of relations
among these average quantities.
If one assumes, for example, that the pulsating components v’, p’, and
p are governed by the Navier-Stokes equations for an incompressible
fluid, then one averaging process applied to Navier’s equation leads to a
set of equations obtained by Reynolds.” These equations involve not
only the od’, but also the mean values of the pulsating components of
velocity. Because of the presence of these latter components, the system
of Reynolds equations is incomplete and new hypotheses, based on experi-
mental evidence, must be introduced to complete the system.
It appears unlikely that a unified formulation of satisfactory models for
turbulent flows of compressible viscous fluids or for viscoelastic and plastic
solids can be constructed within the framework of classical mechanics
and thermodynamics. The development of such models is likely to be
based on statistical mechanics in which mechanical characteristics are
viewed as probabilities and their values appear as mathematical expecta-
tions.
A discussion of models of plastic and viscoelastic materials, utilizing
the principles of thermodynamics of irreversible processes, is contained in
a monograph by L. I. Sedov, cited in footnote 12 and in A. Cemal
Eringen’s Non-linear Theory of Continuous Media (New York), 1962.
™ See, for example, H. Schlichting, Boundary Layer Theory, New York (1955),
Chapter XVIII, and L. I. Sedov, Introduction to Mechanics of a Continuous Medium,
Moscow (1962), pp. 213-217.
 BIBLIOGRAPHY

P. Appell, Traité de méchanique rationelle, vol. 5 (Paris, 1926).

L. P. Eisenhart, Riemannian Geometry (Princeton, 1926).
T. Levi-Civita, The Absolute Differential Calculus (London, 1927).
A. S. Eddington, The Mathematical Theory of Relativity (Cambridge, 1930).
A. J. McConnell, Applications of the Absolute Differential Calculus (London, 1931).
O. Veblen, Invariants of Quadratic Differential Forms (Cambridge, 1933).
T. Y. Thomas, Differential Invariants of Generalized Spaces (Cambridge, 1934).
R. B. Lindsay and H. Margenau, Foundations of Physics (New York, 1936).
L. Brillouin, Les tenseurs en mécanique et en élasticité (Paris, 1938).
C. E. Weatherburn, Riemannian Geometry and the Tensor Calculus (Cambridge, 1938).
L. P. Eisenhart, An Introduction to Differential Geometry (Princeton, 1940).
P. G. Bergmann, An Introduction to the Theory of Relativity (New York, 1942).
A. D. Michal, Matrix and Tensor Calculus (New York, 1947).
J. L. Synge and A. Schild, Tensor Calculus (Toronto, 1949).
G. Y. Rainich, Mathematics of Relativity (New York, 1950).
J Struik, Lectures on Classical Differential Geometry (Cambridge, Mass., 1950).
D. J.
F. D D. Murnaghan, Finite Deformation of an Elastic Solid (New York, 1951).
A nile Green and W. Zerna, Theoretical Elasticity (Oxford, 1954).
I. S. Sokolnikoff, Mathematical Theory of Elasticity (New York, 1956).
J. L. Synge, Relativity: The Special Theory (Amsterdam, 1956).
J. L. Synge, Relativity: The General Theory (Amsterdam, 1960).
W. Prager, Introduction to Mechanics of Continua (Boston, 1961).
T. Y. Thomas, Concepts from Tensor Analysis and Differential Geometry (New York,
1961).
A. C. Eringen, Nonlinear Theory of Continuous Media (New York, 1962).
J. C. H. Gerretsen, Lectures on Tensor Calculus and Differential Geometry (Groningen,
1962).
L. I. Sedov, Introduction to Mechanics of a Continuous Medium (Moscow, 1962).

353
 r i Cee! ‘ee
a7 de

= a + x Ls ; § : Lies

~ J a | ais
thal < <3 (RE gree te tna
~ ~ j ‘4 neat = , a ‘ Poa
oA¥s mAs Ribahen b refs :

7
"teinals lh sd vhiehons ae thyc%emer
t alors 4
- ey « wea © ii a b

net essaane
4.4 a
 INDEX

Absolute derivative, 127 Brachistochrone, 245

Absolute tensor, 71 Brilouineelee, 31350353
Acceleration, 207, 287, 350
Action integral, 232 Calculus of variations, 147-156
Action, principle of least, 229 fundamental lemma in, 149
Admissible functional arguments, 148 fundamental problem of, 148
Admissible transformations, 52 Cantor, M., 1
Affine transformation, 10, 80 Carathéodory, C., 232
Algebra of tensors, 64 Cartan, E., 95
Angle, between coordinate lines, 118 Cauchy, A. L., 330
between directions in space, 117 Cauchy-Schwarz inequality, 204
between directions on a surface, 144 Cayley, A., 112
Anisotropic media, 340 Characteristic values of matrices, 32,
Appell) 792565 286.1313353 36
Arc length, along a curve in space, 130 Christoffel, E. B., 81
along a curve on a surface, 142 Christoffel symbols, 79
along coordinate lines, 117 transformation of, 80
element of, 72, 92, 106, 142, 203 Clemence, G. M., 311
Area, element of, 146 Closure, property of, 54
Associated tensors, 74 Codazzi equations, 185
Axiom, of dimensionality, 10 Collar, A. R., 252
of parallels, 105 Compatibility, equations of, 326
Axioms for linear vector spaces, 10 Components of tensors, 50, 60
laws of transformation for, 58-62
Beltrami, E., 106 Components of vectors, 7, 13
Bergmann, P. G., 300, 311, 312, 353 physical, 8; 121, 214
Bernoulli, D., 230 Conservation, of energy, 214, 228, 239,
Bertrand, J. L. F., 136 297.
Bianchi’s identities, 91 of mass, 297, 345
Binormal, 133 Conservative force fields, 212, 217
Birkhoff, Garrett, 33 Constraints, nonholonomic, 156, 242
Birkhoff, G. D., 300, 304 Continuity, equation of, 344, 346
Bliss, G. A., 245 Contraction, in relativity, 288
Bolyai, J., 106 of tensors, 65
Bonnet, O., 202 Contravariant and covariant laws, 59,
Bolza, O., 232 62
Bouquet, J. C., 95 tensor character of, 62
355
356 INDEX
Contravariant tensor, 61 Curvilinear coordinates, in space, 112
Contravariant vector, 60 on a surface, 138
Coordinate curves (or lines), 113 Cycloidal pendulum, 218
Coordinate surfaces, 113
Coordinate systems, 1, 9 D’Alembert’s principle, 332
construction of, 1 Darboux, G., 95
oblique cartesian, 3 Dedekind, J. W. R., 1
orthogonal cartesian, 3, 12 Deflection of light rays, 311
Coordinates, curvilinear, 112, 138 Deformation, of space, 25, 314
cylindrical, 114 analysis of, 314-327
Gaussian, 140 Deltas, Kronecker, 13, 18, 98, 104
generalized, 233 Density, scalar, 70 Mwy
geodesic, 162 Derivative, absolute, 127 ~
local, 292 covariant, 81, 84
normal, 47, 252 intrinsic, 127
orthogonal, 118, 145 of a base vector, 126
proper, 292 of a vector, 81, 124
spherical, 52, 114 of an invariant, 81
transformation of, 10, 51, 140 tensor, 177
Correspondence, one-to-one, 1, 9 Descartes, R., 1
Cosine of an angle, 203 Determinants, 17, 101
Covariant and contravariant laws, 59-— differentiation of, 103
62 expansion of, 18, 103
tensor character of, 62 multiplication of, 17, 102
Covariant differentiation, 81—89 Vandermondian, 33
inversion of order of, 88 Differentiation, covariant, 81
Covariant tensor, 58 intrinsic, 127
Covariant vector, 57 tensor, 177
Cramer’s rule, 18 Dilatation, 323
Curl of a vector, in cartesian coordi-
Dimensionality of space, axiom for, 10
nates, 266
Direction,in space, 116, 203
in curvilinear coordinates, 268
on a surface, 143
Curvature, Einstein, 168
principal, 191
Gaussian, 167, 186
Direction moment, 143
geodesic, 170, 188
Dirichlet’s problem, 274
integral, 202
lines of, 192 Displacement vector, 4, 207, 324
mean, 186 Distance, Euclidean, 115, 203
normal, of a surface, 189 Distortion of volume elements, 322
of a curve, 131, 136 Divergence of a vector, in cartesian co-
radius of, 189 ordinates, 264 ORY Py &:
total, 167, 186 in curvilinear coordinates, 266 Wey
Curvature vector, 133 in cylindrical coordinates, 267
Curvatures, principal, 191 in plane polar coordinates, 267
Curve, motion of particle on a, 215 in spherical coordinates, 267
Curves, coordinate, 113 Divergence theorem, 264
in space, 130, 203 Duncan, W. J., 252
on a surface, 187 Dupin’s theorem, 195
smooth, 216 Dynamics, of a particle, 207
 INDEX WwWNn|

Dynamics, of n particles, 233 Fluid, ideal, 346, 350

_ of rigid bodies, 233 Fluid, incompressible, 350
viscous, 345
e-systems, 97, 146 Flux of a gravitational field, 268
application of, to determinants, 101 Force, 208 “
e-systems, 133, 146 Forces, external and internal, 240
derivatives of, 134, 180 generalized, 238
tensor character of, 133 reactive, 240
EddingtonseAueS..299) 0310. 312, 353 workless, 241
Eigenvalues and eigenvectors, 32 Frazer, R. A., 252
Einstein, A., 59, 92, 288, 290, 296, Free indices, 17
298 Frenet formulas, 134-136
Einstein curvature, 168 Frequency equation, 252
Einstein’s energy equation, 295 Functional, 148
Einstein’s gravitational equations, 298 Functions, linear vector, 24
Einstein’s postulates, 289 of class C”, 51
Einstein’s tensor, 92 scalar point, 54
Eisenhart, L. P., 159, 166, 186, 198, Fundamental quadratic form, first,
200, 353 140, 142
Elastic constants, 342 second, 180
Elasticity, equations of, 338, 343 Fundamental tensor, 74
Energy, 209, 336
conservation of, 214, 228, 239, 297 Galilean transformations, 287
equation of, 214, 217, 228 Galileo, 207
free, 339 Gauss-Bonnet theorem, 198
integral of, 228 Gauss, equation of, 185
internal, 337 formulas of, 182, 184
kinetic, 211, 335 Gauss, K. F., 106, 177, 263
potential, 212, 339 Gauss’ equations of a surface, 139
Entropy, 337 Gauss’ flux theorem, 268
Equilibrium, differential equations of, Gaussian curvature, 167, 186
330 Generalized coordinates, 234
Eringen, A. C., 353 Generalized force, 238
Euclidean space, 4, 25, 72, 92, 108
Generalized momentum, 258
Euclid’s axiom of parallels, 105 Generalized velocities, 234
Generalized virtual displacements, 241
Euclid’s Elements, 105
Euler, L., 152, 230 Geodesic coordinates, 162
Eulerian hydrodynamical equations, Geodesic curvature, 170
346, 350 Geodesics, 157
trajectories as, 233
Euler’s equations, 152-156
Extremals of functionals, 150, 153 Geometrization of dynamics, 233
Geometry, Lobachevskian, 111
- Extremum, constrained, 153, 242
metric, 107
Fermat’s principle, 229 non-Euclidean, 105
Riemannian, 107
Fermi, E., 163
Gerretsen, J. C. H., 204, 353
Field, conservative, 212
Gravitation, Einstein’s law of, 298
tensor, 62
Newton’s law of, 259
vector, 123
Green, A. E., 313, 353
Fitzgerald, G. F., 219
358 INDEX

Green, G., 263 Lagrange, J. L., 208, 230

Green’s function, 275, 278 Lagrangean equations of motion, 212,
Green’s theorems, 264, 273 235, 242
Griffith, B. A., 256, 286 Lagrangean function, 213
Group, abstract, 54 Lamé’s constants, 342
Groups, isomorphic, 56 Landau, L., 312
Laplace’s equation, 89
Hamilton, W. R., 208, 230 Laplacian, 265
Hamiltonian function, 257 in cartesian coordinates, 89
Hamilton’s equations, 256 n curvilinear coordinates, 89
Hamilton’s principle, 226 n cylindrical coordinates, 267
Harmonic function, 273 n plane polar coordinates, 267
Helix, 137, 160 ayn spherical coordinates, 267
Hermitean matrices, 47 Length, element of, 73, 96, 107, 203
Holonomic systems, 156, 235, 242 Of aavector 1 e203
Hooke’s law, 340 Levi-Civita, T., 163, 300, 353
Huygens, C., 218 Lifshitz, E., 312
Hydrodynamics, equations of, 344-351 Light, velocity of, 288, 289
Hydrostatic pressure, 344 Light rays, deflection of, 311
Lindsay, R. B., 209, 353
Ideal fluid, 346 Line, straight, 137
Incompressible fluid, 350 Line-element, in space, 73, 96, 107,
Indices, free, 17 203
summation, 16 on a surface, 142
Inertial systems, 207 Linear dependence, 6, 15
Infinitesimal strains, 326 of vectors, 6, 15
Inner product of tensors, 66 Linear transformations, 19, 28
Integrability conditions, 95, Linear vector spaces, complex, 14
Interval, 292 real, 10
Intrinsic differentiation, 126 Lobachevskian geometry, 111
Intrinsic geometry, 138 Lobachevsky, N., 105
Invariance, concept of, 50
Local coordinates, 292
of physical laws, 287
Lorentz, H. A., 288, 289, 291
transformation by, 54
Lorentz-Einstein transformation, 290
Invariants, 51
Lorentz-Fitzgerald contraction, 288,
Irrotational motion, 351
290
Isometric surfaces, 164
Loves AZ EoHe.344
Isotropic media, 340

Mach, E., 209

Jacobi aGuGealew2 3 0res2
MacLane, S., 33
Jacobian determinants, 53
Manifold, 9
Kellogg, O. D., 263, 264 n-dimensional, 10, 202
Kepler’s law, 259, 285 non-Euclidean, 203
Kinetic energy, 211, 335 Riemannian, 92
Kinetic potential, 238 Margenau, H., 209, 353
Kleine etl2 Mass, conservation of, 297, 345
Kronecker deltas, 13, 19, 98 gravitational, 209
derivatives of, 104 inertial, 209
tensor character of, 101 rest of proper, 294
 INDEX 309

Mass-energy relationship, 297 Normal coordinates, 47, 252

Matrices, 20 Normal curvature, of a surface, 189
algebra of, 20-24 principal, 190
characteristic equation of, 36 Normal line to a surface, 175
characteristic values of, 32, 36 Normal modes of vibration, 47, 254
diagonal, 21 Normal vector, to a curve, 132
Hermitean, 47 LO -aesuTTace, W175
inverse, 23 to a surface curve, 170
orthogonal, 28 Novozhilov, V. V., 313
real symmetric, 34
reduction to diagonal form, 30 Orthogonal curvilinear coordinates, con-
similar, 29 dition for, 118
singular, 22 Orthogonal transformations, 27, 29
unitary, 47 Orthogonality of vectors, 11, 145
Maupertuis, P. M. L., 229 Ortho-normal systems of vectors, 8, 11,
McConnell, A. J., 127, 178, 179, 180, 12
183. 16,- 33,7353 Osculating plane, 131
MeVittie, G. C., 311
Mean curvature, 186 iMares. IE Us, SIs
Measure numbers of a vector, 7 Parabolic points, 194
Mechanics of a particle, 206 Parallel postulate, 105
Metric space, 9, 107 Parallel vector fields, along a curve,
Metric tensor, 72, 142 128
Meusnier’s theorem, 187 along a surface curve, 163
Michal, A. D., 313, 353 Parallel surfaces, 195
Minimum principles, 229 Parallelogram law of addition, 4
Minkowski’s acceleration, 293 Pars, L. A., 245
Minkowski’s velocity, 293 Particles, dynamics of, 207, 233
Moment of force, 260 relativistic dynamics of, 298
Momentum, 208 Pendulum, cycloidal, 218
Pendulum, double, 250
Motion, equations of, for a continuous
medium, 332 simple, 219, 249
irrotational, 351 spherical, 223, 256
Perihelion, advance of, 308
Motion of a particle on a curve, 215
Perihelion constant, 286
Motion of a particle on a surface, 219
Perihelion of Mercury, 307, 310
Murnaghan, F. D., 33, 313, 327, 342,
Physical components of a vector, 8,
353
12 Tez i
Planetary orbits, 304
Natural system, 234
Natural trajectory, 216 Pogorelov, A. V., 200, 205
Poincare, H., 112, 289
Navier equations, 342
Poisson’s equation, 263, 271
of fluid motion, 349
hydrodynamical equa- Poisson’s integral, 281
Navier-Stokes’
Poisson’s ratio, 342
tions, 350
Potential, elastic, 339
Neumann’s problem, 275
gravitational, 262
Newton, I., 207, 259, 281
259 kinetic, 238
Newtonian law of gravitation,
velocity, 351
Newtonian laws, 207
Potential energy, 212
Nirenberg, L., 205
Prager, W., 314, 353 4c
Non-holonomic systems, 235, 242
Lv)
—\oN
Pee pee
360 INDEX
Primary inertial system, 207 Scalar product, 5, 13, 117
Principal curvatures of a surface, 191 triple, 121
Principal directions of strain, 320 Schild, A., 353
Principal directions of stress, 329 Schlichting, H., 352
Principal directions on a surface, 191 Schwarzschild, K., 300
Principal strains, 320 Schwarzschild’s line element, 301
Principal stress, 329 Sedov. L.. E., 314, 327, 3433952.835)35
Principle of least action, 231 Serret-Frenet formulas, 139
Problem of two bodies, 281 Seugling, W. R., 327
Proper mass, 294 Shearing strains, 318
Pythagoras, formula of, 3, 13 Signorini, A., 313
Similar transformations, 26
Quadratic forms, 34 Skew-symmetric systems, 97
characteristic values of, 32, 37, 48 Skew-symmetric tensors, 69
classification and properties of, 44 Small oscillations, 253
index of, 44 Sokolarkoff, 417 Sz) 51,5 262,.26320320:
rank of, 44 340, 344, 353
Quadric of Cauchy, strain, 319 Space, dimensionality of, 6, 9
stress, 330 Euclidean, 4, 92, 202
Quotient laws of tensors, 66 metric, 10, 107
Riemannian, 92, 202
Rainich,= Gey) 299 3 1 Osest2 eS 55 Space curves, geometry of, 130
Rank of a tensor, 61 Space-time manifold, 289
Rapidity, 292 Spaces, complex linear vector, 14
Reciprocal base systems, 119 Euclidean, 4, 92, 202
Redheffer, R. M., 262 linear vector, 6
Regression, edge of, 195 Spectral lines, shift of, 311
Relative scalar, 70 Spherical excess, 201
Relative tensors, 69, 103 Spherical points, 194
Relativistic dynamics, 298 Spherically symmetric static field, 300
Relativity, general theory of, 298-304 State, equation of, 347
restricted theory of, 288-274 Stevinus, S., 4
Reynold’s equations, 352 Stokes, G. G., 263
Ricci, G., 59 Stokes’ theorem, 266
Ricci tensor, 91 Straight line, equation of, 137
Ricci’s identity, 89 Strain, in cartesian coordinates, 326
Ricci’s theorem, 86 infinitesimal, 326
Rice, J., 290 interpretation of, 317
Riemann-Christoffel tensor, 86, 88 principal directions of, 320
properties of, 89 velocity, 349
Riemannian geometry, 107 Strain invariants, 320
Riemannian space, 92, 107 Strain quadric, 319
Riemann’s dissertation, 106 Strain tensor, 316, 326
Rizo Stress, analysis of, 327-332
principal, 329
Saint Venant, B., 326 types of, 330
Savile, H., 105 Stress invariants, 330
Scalar, 54 Stress quadric, 330
Scalar density, 70 Stress-strain relation, 339, 340, 341
 INDEX 361
Stress tensor, 328 Total curvature, 167, 186
symmetry of, 331 Trajectories as geodesics, 232
Stress vector, 328 Trajectory, of a dynamical system, 238
Struiks DJ." 2005 2022 353 of a particle, 210, 216
Summation convention, 16 Transformation theorems, 263
Surface, curves on, 187 Transformations, admissible, 52
element of, 146 affine, 10
equations of, 139 Galilean, 287
intrinsic geometry of, 140 induced, 58
particle on a, 219 of rotation, 28
Surfaces, isometric, 165 orthogonal, 28
parallel, 195 similar, 26
tangent, 195 unitary, 47
topologically equivalent, 202 Truesdell, C., 314
Symmetric systems, 97 Turbulent flow, 352
Symmetric tensors, 69
Synge; J. L.; 256, 286, 290, 296, 300; Umbilical points, 194
3040310" 301, S12-6353 Unitary transformations, 47

Tangent surfaces, 195 Vandermondian determinant, 33

Tensor derivatives, 177 Variation, symbol of, 224
Tensor equations, 64 Variation, of strain tensor, 334
Tensor fields, 62 Veblen, O., 9, 59, 353
Tensors, absolute, 71 Vector spaces, n-dimensional, linear, 10
algebra of, 64 Velocity of a particle, 207
Tensors, associated, 74 Velocity strains, 349
Virtual displacement, 241, 332
calculus of, 81-86
Virtual work, 240
components of, 50, 60
Viscosity, coefficients of, 349
contraction of, 65
kinematic, 351
contravariant, 61
Viscous fluid, 345, 349
covariant, 58
Voigt, W., 289
covariant differentiation of, 81-86
Volume, element of, 118, 204
fundamental, 74, 181
Vorticity vector, 351
intrinsic differentiation of, 127
metric, 73 Weatherburn, C. E., 353
mixed, 61 Weight of a tensor, 71
quotient laws for, 66 Weierstrass, K., 147
rank of, 61 Weingarten’s formulas, 185
relative, 71 Weyl, H., 59, 300
Riemann-Christoffel, 86 Whittaker, E. T., 223
symmetric and skew-symmetric, 69 Work, definition of, 210
tensor differentiation of, 177 function, 212
types of, 59 virtual, 241, 332, 334
Thermodynamic laws, 336
Thermoelastic equations, 340 Young’s modulus, 342
Thomas, T. Y., 95, 198, 353
Tolman, R., 290, 312 Zerna, W., 313, 353
Torsion, 133, 137 Zvolinsky, N. V., 318

APES:

pPust grit
 *

TY

7 ~ x
“
<2
2 ity Mey
; ‘ "4 a
fi : im|
ut ' ' if 10; 28 ty Sictoedie
Uh <vitiviley x lo wwiisois¥
_— ey okt on oF
ee a ag. OY ae ey . rac}Rericar ne
fielded a
ter Rl’ Weteestineia lary
Magnesite
an
0856, oF, pee ao ee
Os; ive feutsy +
vit 2 =
ied .
Jo andbe
7

‘iietiens eet) = —_,

lad pe eh wa
Rgetw vont stv eecoeed iy@erisst paras}
aa
Seas, ile Bes
vy, po Bes
; ted spurawenlt |
Wire
wh
 about the author...
|. S. SOKOLNIKOFF is Professor
of Mathematics at the University of
RICHARD C. FREY California,
Los Angeles,a position
he has held since 1946. He received
|003 SUNSET
his B.S. from Idaho University and
CINCINNAT! 5, OHIO his Ph.D. from the University of Wis-
consin where he taught from 1927
to 1946.
He worked with the National De-
fense Research Committee and in
the Office of Scientific Research and
Development in a number of posi-
tions from 1941 to 1946. The recipi-
ent of two Guggenheim fellowships,
Dr. Sokolnikoff has also conducted
research under three grants from
the Research Corporation of New
York. He is a member of the Ameri-
can Mathematical Society.
The author of numerous journal
articles in the mathematical theory
of elasticity, he has also written a
number of books on this as well as
other topics. He is the editor of the
Quarterly of Applied Mathematics.
Advanced Engineering
Mathematics
By ERWIN KREYSZIG, The Ohio State Univefsity. “The
book is well written, and both the exposition and the exten-
sive sets of problems reflect a balance between mathe-
matical theory and emphasis on applications. The book is
a suitable textbook for a three- or a four-semester course.
or, by omitting certain sections, it can be used for separate
one-semester courses in several areas. It is an excellent
reference book for engineers and furnishes a handy guide
to the more commonly used mathematical theory, with
references to more detailed treatments for those that are
interested.’”’—Leon W. Rutland in Science.
856 pages. i

Introduction to Vector
and Tensor Analysis
By ROBERT C. WREDE, San Jose State College. This
book is a careful presentation of the fundame ideas
used in developing geometry, analysis, and linear
It is centered around four pivotal ideas: historic:
tive and motivation; the interrelationships of
geometry; a structure that stresses proof, and centr
the idea of transformation; and the use of sur
ventions and other notational devices. These
troduced early and referred to all through thet
possible the section on special relativity, in
geometric structure of the theory.
sa,

John Wiley « Sons

| New York * London *