bk978 0 7503 3455 6ch1

This content has been downloaded from IOPscience. Please scroll down to see the full text.
Download details:
IP Address: 140.109.103.203
This content was downloaded on 07/06/2024 at 06:18
Please note that terms and conditions apply.
You may also like:
Progressive evaluation in spectroscopic sensors for non-invasive blood haemoglobin analysis—a

review
Yogesh Kumar, Ayush Dogra, Ajeet Kaushik et al.
3rd International Conference on Turbulent Mixing and Beyond

Snezhana I Abarzhi, Serge Gauthier, Christopher J Keane et al.
International Conference on Innovative Research - ICIR Euroinvent 2017

Andrei Victor SANDU, Mohd Mustafa Al Bakri ABDULLAH, Petric VIZUREANU et al.
Regional disparity in continuously measured time-domain cerebrovascular reactivity indices: a

scoping review of human literature
Amanjyot Singh Sainbhi, Izabella Marquez, Alwyn Gomez et al.
IOP Publishing
Classical Field Theory and the Stress–Energy Tensor

(Second Edition)
Mark S Swanson
Chapter 1
Geometry and physics
Properly formulating classical field theory requires some basic concepts from
differential geometry. This book will develop these ideas and techniques specifically
for application to physical systems, and so the presentation is neither complete nor
rigorous. Instead, the intent is to place physical concepts familiar from basic physics
into a more unified context. In particular, coordinate systems, coordinate trans-
formations, vectors, the metric tensor, the gradient, the Laplacian, and differential
forms are critical ingredients in formulating classical field theory and also form the
foundation of basic differential geometry. These ideas and constructs will be used
throughout this book and so familiarity with these tools is important.
1.1 Manifolds
Newtonian physics was formulated to describe the phenomenon of motion or
dynamics. The classical dynamics of Newton makes a number of assumptions, often
taken for granted in the presentation of elementary physics, regarding the properties
of the space in which motion occurs. The most basic assumption of Newtonian
physics is that the space available for motion is a smooth collection of points and
that, as a point particle moves in that space, its position is a continuous or smooth
trajectory through those points. It is therefore possible to install a coordinate system,
at least locally, that allows the definition of the calculus operations of differentiation
and integration. Loosely speaking, the term local means that there is a neighborhood
of each point where this is possible. The installation of a coordinate system is
understood to be a local mapping of the points in the space into the coordinates of
the point or a function that meets the same objective. There may be overlapping
neighborhoods with different coordinate systems, and where they overlap these
coordinate mappings must be invertible and one-to-one, which requires their
differentiability. A collection of points is referred to mathematically as a manifold,
doi:10.1088/978-0-7503-3455-6ch1 1-1 ª IOP Publishing Ltd 2022

Classical Field Theory and the Stress–Energy Tensor (Second Edition)
generically designated M, and so such a smooth collection of points is known as a

differentiable manifold. This brief description leaves out many subtleties, and the
reader interested in a more mathematically rigorous development of manifolds is
referred to one of the many excellent books on modern analysis.
Classical physics typically assumes that the spatial manifold where motion occurs
possesses three independent dimensions or directions. Although motion may be
restricted to a two-dimensional plane or a one-dimensional line, in general, the
motion of a Newtonian particle is a three-dimensional phenomenon. Modern
physics has relaxed this assumption to allow the consideration of physical manifolds
that are n-dimensional. Stated simply, this means there are n independent directions
available for motion. This step was required to accommodate the concept of four-
dimensional space-time, but has also emerged in the context of higher-dimensional
attempts to unify forces, such as Kaluza–Klein theory and string theory. The physics
of Newton and contemporaries further assumed that both space and time,
equivalently distance and duration, were absolute properties of the world that all
observers could agree upon. A significant offshoot of that worldview was the
assumption that three-dimensional Euclidean geometry correctly describes the
relationships between points in the spatial manifold of motion. A manifold where
Euclid’s axioms and theorems hold will be referred to as Euclidean. It was under-
stood that Euclidean geometry did not hold on a curved surface, such as the surface
of the Earth, and so Euclidean manifolds are often referred to as flat manifolds. The
n-dimensional extension of the Euclidean plane is known mathematically as n ,
where n copies of the real line  serve as the Euclidean degrees of freedom.
However, experimental observations and the subsequent development of the
special and general theories of relativity forced a revision of both these assumptions.
In particular, the special theory of relativity showed that the measurements of
distance and time intervals by different observers are affected by their relative
motion, and that correctly describing motion required merging space and time into
the four-dimensional manifold referred to as space-time. Further, the principle of
general relativity incorporates a description of gravitation in terms of a space-time
that is dynamic and non-Euclidean, so that distances as well as durations are no
longer governed by Euclidean geometry. These changes required a significant
revision of Newtonian physics as well as a fundamental overhaul of how physical
theories are formulated.
An effective presentation of classical field theory must therefore take into account
the changes in assumptions regarding the manifold for motion and fields that have
occurred over the last century. Although space, together with time, are still assumed
to define a differentiable manifold, understanding both the arbitrary dimensionality
and the deviation from Euclidean behavior benefit greatly from adopting the more
flexible techniques of modern differential geometry. Some of these techniques may
be unfamiliar, and so this chapter will present a simplified introduction to the more
unified approach to manifolds, coordinates, and vector spaces that basic differential
geometry provides. The standard Euclidean treatment of manifolds and vectors is
still contained in this approach, but it allows the generalizations required by modern
classical field theory.
1-2
1.2 Coordinate systems

Quantifying position and motion in a differentiable manifold requires a choice of
coordinate system. This consists of choosing an origin point and mapping the points
of the manifold onto a coordinate grid around the origin. Figure 1.1 depicts the same
point in two commonly used coordinate systems for the two-dimensional Euclidean
plane. The choice of coordinate system and origin is arbitrary, and it is stressed that
coordinate systems themselves have no physical meaning. However, once a
coordinate system is chosen, the coordinates of a point in the space around the
origin are determined by its place on the coordinate grid in relation to the origin.
In two dimensions Cartesian coordinates consist of two straight perpendicular lines
or coordinate axes through the origin and a coordinate grid consisting of straight lines
parallel to these two axes. The Cartesian coordinates (x , y ) of a point are defined by
parallel projection, and the two gridlines that determine the position of a point form a
parallelogram with the coordinate axes. The point of intersection of the line with the
axis gives the coordinate of the point. For the choice of two-dimensional polar
coordinates the grids consist of circles of radius r around the origin coupled with radial
lines at each angle θ. The polar coordinates of the point are then given by (r, θ ). In
order for the two coordinate systems to share the same neighborhood of the manifold,
they must be related by differentiable and invertible functions. If the plane is Euclidean,
these relations are given by the familiar Euclidean expressions,
x = r cos θ , y = r sin θ ⟺ r = x2 + y2 , θ = tan−1(y / x ). (1.1)
However, depending on the manifold or the choice of coordinate system, the process
of installing a coordinate system can become subtle. For example, it is possible to
choose coordinate axes that are not perpendicular as well as defining coordinates by
perpendicular projection. In addition, if the manifold is non-Euclidean, the expres-
sions of (1.1) require modification.
Figure 1.1. Cartesian and polar coordinate grids.
1-3
The upshot of choosing a coordinate system in an n-dimensional manifold is to

assign or map each point p into a set of n coordinates, denoted xj, where the
superscript j can take on any value from 1 to n. The coordinates of each point are
assumed to be unique and unambiguous. Two important quantities are derived from
the coordinates of a point, regardless of the choice of coordinate system or whether
the manifold is Euclidean or not. The first is the differential dx j , which is an
infinitesimal change in the jth coordinate of a point. The second is the partial
derivative with respect to xj, denoted ∂j ≡ ∂/∂x j , which quantifies the rate of change
along the jth coordinate axis for a multivariable function of the coordinates of the
point, f (x1, … , x n ). These two quantities allow the operations of calculus to be
defined and are clearly dependent upon the specific choice of coordinate system for
their definition.
Since coordinate systems are chosen by an observer, another observer may choose
a different coordinate system, and this is referred to as a coordinate transformation.
Doing so changes the coordinates of a point to a different set of values, denoted yj,
and changes the differentials to dy j and the partial derivatives to ∂/∂y j . However, the
change of coordinates does not change the relationship of each point to the points
around it, and so it is referred to as a passive transformation. The ratio of distances
remains the same in both coordinate systems, as do the angles in geometric shapes
such as triangles. In order to compare the results of an observation or evaluate a
mathematical expression, it is critical to be able to relate the quantities measured or
defined at a point in either coordinate system. It is therefore assumed that the
y-coordinates can be expressed as invertible and differentiable functions of the
x-coordinates, so that y j = y j (x1, … , x n ) ≡ y j (x ) gives the y-coordinates of the point
p whose x-coordinates are xj. Since these functions are invertible, there is also a set of
differentiable functions that yields the x-coordinates x j = x j (y1, … , y n ) ≡ x j (y )
given the y-coordinates. It is assumed that the transformation between the two
coordinate systems is both unique and one-to-one, which is referred to as an
isomorphism, for all the points in the manifold where the two coordinate systems are
both defined.
For the moment, it will be assumed that the two coordinate systems share the
same point as an origin, so that y j (0, … , 0) = 0. The coordinates must be
independent degrees of freedom in both coordinate systems. Stated mathematically,
this means that both coordinate systems satisfy
∂y j ∂x j ⎧1 if j = k
= = δ k
j
= ⎨ , (1.2)
∂y k ∂x k ⎩ 0 if j ≠ k
where δ j k is known as the Kronecker delta. Requirement (1.2) simply means that
varying one independent degree of freedom creates no variation in the other
independent degrees of freedom. The n-dimensional Kronecker delta has the
properties that
n n n
∂ ∂
∑ δ j k dx k = dx j , ∑ δ j k ∂ j
=
∂x k
, ∑ δ j kδ k ℓ = δ j ℓ . (1.3)
k=1 j=1
x k=1
1-4
These two sets of coordinates must also obey the chain rule, which states that
n n
∂y j (x ) ∂x k(y ) ∂y j ∂x j (y ) ∂y k (x ) ∂x j
∑ ∂x k
∂ y ℓ
=
∂y ℓ
= δ j ℓ, ∑ ∂y k
∂ x ℓ
=
∂x ℓ
= δ j ℓ. (1.4)
k=1 k=1
The proof of (1.4) follows from the definition of differentiable mappings in a

manifold and is available in the references for multivariable calculus. Two
isomorphic coordinate systems satisfying (1.2) and (1.4) are said to be diffeomorphic.
Exercise 1.1. Show explicitly that the Cartesian to polar coordinate transformation
(1.1) satisfies
∂r ∂x ∂r ∂y ∂r ∂x ∂r ∂y
+ = 0, + = 1.
∂x ∂θ ∂y ∂θ ∂x ∂r ∂y ∂r
1.3 The Jacobian

A coordinate system may result in ambiguous or undefined coordinates for some
points. An example occurs at the origin in polar coordinates, where the polar angle is
arbitrary and therefore undefined. For the sake of brevity this book will only
peripherally touch on such subtleties and instead assume that the coordinate system
provides unambiguous and unique coordinates for all but a few points. In that
regard, the coordinate transformation is invertible at all points where the Jacobian of
the coordinate transformation is nonzero.
Defining the Jacobian, as well as many other important quantities in matrix
theory, group theory, and differential geometry, requires the Levi-Civita symbol,
which is denoted ε abc…. The Levi-Civita symbol can have an arbitrary number of
superscripts, but once the number of subscripts is specified, the range of the
subscripts must match that number. For example, the Levi-Civita symbol ε ab has
two superscripts, each of which can take the values of one or two. The Levi-Civita
symbol εijk has three subscripts, each of which can take the values of one, two, or
three. The Levi-Civita symbol ε μνρσ has four subscripts, each of which can take the
values of zero, one, two, or three when used in space-time physics applications. The
Levi-Civita symbol has one of three values, given by
⎧+ 1 if abc… is an even permutation of 1, 2, 3, …
⎪
ε abc…
= ⎨− 1 if abc… is an odd permutation of 1, 2, 3, … . (1.5)
⎪
⎩ 0 if any of the abc … are the same
For the case of two superscripts the Levi-Civita symbol ε ab has the values
ε12 = −ε 21 = 1, while ε11 = ε 22 = 0. For three superscripts, ε123 = −ε132 = 1 and
ε112 = ε 223 = 0, and so on. It is worth noting that ε3ab = ε ab , so that the lower-order
versions of the Levi-Civita symbol descend from the higher-order versions. For that
reason, the Levi-Civita symbol obeys many identities.
1-5
If the Levi-Civita symbol is defined in a Euclidean manifold using Cartesian

coordinates, it is possible to define the Levi-Civita symbol with subscripts εabc… by
simply equating it:
εabc… = ε abc…. (1.6)
However, if the manifold is non-Euclidean, care must be taken, since the two forms
are not simply related. In particular, the space-time manifold that forms the basis of
modern field theory requires a more careful analysis to find the correct relationship.
This remark will be clarified as further aspects of manifolds are developed.
Exercise 1.2. Using (1.6), demonstrate the two very useful properties of the three-
dimensional Levi-Civita symbol in a Euclidean space:
3
∑ ε kℓmεkij = δ ℓ iδ m j − δ ℓ jδ mi , (1.7)
k=1
3
∑ ε ℓjkεijk = 2 δ ℓ i . (1.8)
j ,k = 1
It is tedious but easy to show both sides are the same by writing out all possible index
combinations.
It is easy to see that

2
∑ ε abεab = ε11ε11 + ε12ε12 + ε 22ε22 + ε 21ε21 = 2 = 2! , (1.9)
a, b = 1
while (1.8) gives

3
∑ ε ijkεijk = 6 = 3! . (1.10)
i, j , k = 1
Using induction shows that

n
∑ ε ijk…εijk… = n! , (1.11)
i, j , k, …= 1
where n is the number of indices. This follows by noting that the n-dimensional case
of (1.11) is simply n copies of the n − 1- dimensional case, giving n · (n − 1)! = n!,
completing the inductive proof.
For the case of a coordinate transformation from x-coordinates to y-coordinates
in n dimensions, given by the functions x j = x j (y ) and y j = y j (x ), the Jacobian J is
given by
1-6
n
∂x j1(y ) ∂x j2(y ) ∂x jn(y )
J (y ) = ∑ ε j1j2 …jn
∂y1 ∂y 2
…
∂y n
. (1.12)
j1 , j2 , … , jn = 1
The reader may recognize (1.12) as the determinant of the matrix whose elements are
∂x j /∂y k . This is discussed in more detail in chapter 7, where linear algebra is
reviewed. The properties of the Jacobian are discussed later in this chapter, in
particular, where differential forms are presented. Wherever the Jacobian is nonzero,
the relationship between the y-coordinates and the x-coordinates is invertible. For
example, (1.1) relates the Cartesian coordinates x1 = x and x 2 = y for the two-
dimensional Euclidean plane to the polar coordinates y1 = r and y 2 = θ for the same
two-dimensional Euclidean plane. The Jacobian for transforming from Cartesian to
polar coordinates is given by
2
∂x a ∂x b ∂x ∂y ∂y ∂x
J (r , θ ) = ∑ εab 1
∂y ∂y 2
= ε12
∂r ∂θ
+ ε21
∂r ∂θ
= r cos2 θ + r sin2 θ = r , (1.13)
a, b = 1
which is nonzero everywhere except at the origin, r = 0. As mentioned earlier, polar

coordinates are not invertible at the origin, since θ = tan−1(y /x ) is undefined there.
There is a second form of the Jacobian expression (1.12) that will be useful, given by
n
∂x j1(y ) ∂x j2(y ) ∂x jn(y )
∑ ε j1j2 …jn
∂y k1 ∂y k2
…
∂y kn
= ε k1k2…knJ (y ). (1.14)
j1 , j2 , … , jn = 1
The proof that (1.14) is equivalent to (1.12) follows from relabeling any two of the
summed dummy indices, say j1 and j2, and noting that rearranging the first two
factors gives
n
∂x j2(y ) ∂x j1(y ) ∂x jn(y )
∑ ε j2 j1…jn
∂y kl ∂y k2
…
∂y kn
j1 , … , jn = 1
n (1.15)
∂x j1(y ) ∂x j2(y ) ∂x jn(y )
= − ∑ ε j1j2 …jn … .
j ,…,j =1
∂y k2 ∂y k1 ∂y kn
1 n
This shows that the left-hand side of (1.14) changes sign under a permutation of the
k indices and therefore vanishes if any two of the k indices are the same. This is
precisely the property of the Levi-Civita symbol on the right-hand side of (1.14).
Examining the expression for n = 2, where ε11 = ε22 = 0, shows that
2
∂x a(y ) ∂x b(y ) ∂x1(y ) ∂x 2(y ) ∂x 2(y ) ∂x1(y )
∑ εab
∂y c ∂y d
=
∂y c ∂y d
−
∂y c ∂y d
, (1.16)
a, b = 1
which vanishes if c = d and is given by εcd J (y ) for c ≠ d , inductively corroborating (1.14).

The Jacobian also appears in volume integrations for an n-dimensional manifold.
The infinitesimal n-dimensional volume elements in the two coordinate systems,
denoted dnx and dny, are related by
1-7
dnx = J (y ) dny , (1.17)

where the absolute value preserves the positivity of the volume element, allowing the
limits on the integration to remain the same. This result is derived when differential
forms are discussed later in this chapter. Using the Jacobian (1.13) gives the well-
known relation between the Cartesian surface element dSc and the polar surface
element dSp in two dimensions,
dSc = d2x = dx dy = J (r , θ ) dr dθ = r dr dθ = d2r = dSp. (1.18)
The Leibniz property of the differential and (1.1) combine to give

∂ ∂
dx = d(r cos θ ) = dr (r cos θ ) + dθ (r cos θ )
∂r ∂θ
= dr cos θ − r sin θ dθ ,
(1.19)
∂ ∂
dy = d(r sin θ ) = dr (r sin θ ) + dθ (r sin θ )
∂r ∂θ
= dr sin θ + r cos θ dθ .
However, it is important to note that the surface element in polar coordinates is not
given by simply multiplying together the dx and dy given by (1.19). Instead, the
coordinate differentials of (1.19) will be used later in this chapter to motivate the
wedge product that gives infinitesimal surface and volume elements.
Exercise 1.3. If attention is restricted to the first quadrant, where (x, y ) > 0 in
Cartesian coordinates, hyperbolic coordinates (u , v ) are defined by
1
u= ln (x / y ), v = xy ⟺ x = v exp(u ), y = v exp( − u ). (1.20)
2
Show that the Jacobian for this coordinate transformation is J (u , v ) = 2v, so that
hyperbolic coordinates are uniquely invertible for x, y > 0.
1.4 Contravariant and covariant quantities

Result (1.19) is a specific case of how a coordinate transformation relates the
infinitesimal differentials of the two coordinate systems, dx k and dy j , through the
Leibniz property,
n
∂y j (x )
dy j = ∑ ∂x k
dx k . (1.21)
k=1
The mapping (1.21) between dy j and the original dx j is referred to as a contravariant

transformation, and it identifies the differentials dx j as contravariant quantities.
1-8
These differentials are not the only contravariant quantities of importance in field
theory. Such contravariant quantities are notationally identified using a superscript,
so that Vj transforms identically to (1.21) in another coordinate system.
Similarly, the relationship of the partial derivatives in the two coordinate systems
∂/∂x k and ∂/∂y j is given by the inverse functions from the chain rule (1.4),
n
∂ ∂x k(y ) ∂
∂y j
= ∑ ∂y j ∂x k
. (1.22)
k=1
The mapping (1.22) between ∂/∂y j and the original ∂/∂x j is referred to as a covariant
transformation, and it identifies the partial derivatives as covariant quantities. The
partial derivative is not the only covariant quantity of importance in field theory.
Such covariant quantities are notationally identified using a subscript, so that
∂/∂x j ≡ ∂j indicates the covariant transformation property of the partial derivative
with respect to xj. All other covariant quantities Vj transform according to (1.22)
when a coordinate transformation occurs.
The two types of transformations, (1.21) and (1.22), provide observers with a well-
defined mathematical recipe to compare covariant and contravariant quantities
measured in different coordinate systems. Therefore, it is a basic assumption of field
theory that covariant and contravariant quantities, and their extension to spinor
quantities (to be discussed), provide the building blocks of all physically viable theories.
As an example, using (1.21), (1.22), and the chain rule (1.4) shows that both
coordinate systems result in the same form for the differential operator d,
n
∂
n ⎛ n
∂y j ℓ⎞⎛
n
∂x k ∂ ⎞
d = ∑ dy j
∂y j
= ∑⎜⎜∑ ∂ d x ⎟⎟⎜⎜ ∑ ⎟
k⎟
j=1 j = 1⎝ ℓ = 1
xℓ j
⎠⎝ k = 1 ∂y ∂x ⎠
(1.23)
n ⎛ n ⎞ n n
ℓ⎜ ∂x k ∂y j ⎟ ∂ ∂ ∂
= ∑ dx ⎜∑ j ℓ ⎟ k = ∑ dx ℓ δ k ℓ = ∑ dx k ∂x k .
k, ℓ = 1 ⎝ j = 1 ∂y ∂x ⎠ ∂x k, ℓ = 1
∂x k k=1
Applying (1.23) to y ℓ (x ) correctly reproduces the contravariant transformation law

(1.21) for differentials. Any quantity that takes the same form in different coordinate
systems is referred to as a scalar quantity. Scalar quantities are central to field theory,
since they are independent of the choice of coordinate system and so all observers
agree upon their value.
1.5 The summation convention

Because summing over all n degrees of freedom is so common and the summation is
typically over a pair of contravariant and covariant indices, equations such as (1.23)
are greatly streamlined by employing the summation convention. This means that a
pair of identical indices, one contravariant and one covariant, implies a summation
over the entire range of that index. From this point forward, the summation
convention will be in effect unless the indices are placed in parentheses. Using this
convention, it follows that
1-9
n n n
j
α βj = ∑ α βj ,j (j )
α βj ≠ ∑α (j )
βj , δ(jj ) ≠ ∑ δjj , (1.24)
j=1 j=1 j=1
where the last two expressions stand for the product of α (j ) with βj and the element
δ(jj ), respectively, each with no sum over the index j. The demonstration of (1.23) is
repeated, step for step, using the summation convention:
∂ ⎛ ∂y j ⎞⎛ ∂x k ∂ ⎞
d = dy j = ⎜ d x ℓ
⎟⎜ j ⎟
∂y j ⎝ ∂x ℓ ⎠⎝ ∂y ∂x k ⎠
(1.25)
⎛ ∂x k ∂y j ⎞ ∂ ∂ ∂
= dx ℓ⎜ j ℓ ⎟ k = dx ℓ δℓk = dx k k .
⎝ ∂y ∂x ⎠ ∂x ∂x k
∂x
The Kronecker delta gives
n n
δkk = ∑ δkk = ∑ 1 = n, δ((kk)) = 1. (1.26)
k=1 k=1
The value of n is determined contextually from the dimension of the quantity, which
requires the reader to be aware of the dimension of the manifold under analysis. The
repeated index is a dummy index, since the choice of repeated symbol is irrelevant, so
that δ jj = δkk = n. As with the Levi-Civita symbol, it is common practice in physics to
use the lower-case Latin letters, such as j and k, to indicate two or three dimensions and
lower-case Greek letters, such as μ and ν, to indicate four and higher dimensions.
Given a contravariant quantity A j (x ) and a covariant quantity Bk (x ), it is possible
to form a scalar quantity S (x ) through contraction, which is the sum over the
product of the components,
S (x ) = A j (x )Bj (x ). (1.27)
Demonstrating that S (x ) is a scalar is identical to the demonstration (1.25),
⎛ ∂y j ⎞⎛ ∂x k ⎞
S′(y ) = A′ j (y )B j′(y ) = ⎜ ℓ Aℓ (x(y ))⎟⎜ j Bk(x(y ))⎟
⎝ ∂x ⎠⎝ ∂y ⎠
⎛ ∂x k(y ) ∂y j (x ) ⎞ (1.28)
= Aℓ (x )⎜ ⎟Bk(x ) = Aℓ (x ) δℓk Bk(x )
⎝ ∂y j ∂x ℓ ⎠
= Ak (x )Bk(x ) = S (x ).
The Kronecker delta δkj defined by (1.2) is a purely mathematical construct, and
so it must be invariant under coordinate transformations. This will be true as long as
the j index transforms contravariantly, like (1.21), and the k index transforms
covariantly, like (1.22). Denoting the Kronecker delta in the y-coordinate system as
δ′j i , it follows from the properties of the Kronecker delta and the chain rule that
∂y i (x ) ∂x ℓ(y ) k ∂y i (x ) ∂x k(y ) ∂y i
δ′j i = δ ℓ = = = δ ij , (1.29)
∂x k ∂y j ∂x k ∂y j ∂y j
1-10
demonstrating that the Kronecker delta is the same in all coordinate systems, as it
must be.
So far, there have been no restrictions on the specific form of the coordinate
transformation other than giving a nonzero value for the Jacobian (1.12). In physical
applications, there is a significant possibility that a coordinate transformation may
involve two coordinate systems whose origins are in relative motion. Understanding
the effects of relative motion for different observers requires a physical principle for
resolution. This is because relative motion involves the interplay of time and
distance measurements, and may result in effects similar to those of forces. This
will be discussed in detail later.
1.6 Vectors and direction vectors

The spatial relationship of two points in a manifold allows the definition of a vector,
a quantity with both direction and magnitude. While vectors can be defined between
any two points in the manifold, initial attention will be given to the position vector of
a point p. In the case of a Euclidean manifold, it is defined as an oriented straight line
or arrow from the origin to the point p, with the latter point referred to as the vector
tip. The position vector therefore has both a direction as well as a magnitude given
by its length. It is denoted in boldface as rp , but the p is often dropped and r is then
understood to be the position vector of an arbitrary point in the manifold. In the x-
coordinate system the position vector will be a function of the coordinates of the
point p, and so it will often be written r (xp ) or r (x ) for an arbitrary point.
It is important to note that the position vector lies in the manifold, and if the
manifold is non-Euclidean, the position vector is not a straight line in the traditional
Euclidean sense. As an example, the position vector on a sphere can be visualized as
an arrow emanating from the north pole and coinciding with a segment of a
longitude line that intersects the point of interest. The longitude gridlines are great
circles on the sphere and correspond to the spherical version of a straight line. In a
general non-Euclidean manifold the space between the origin and the point of
interest is not flat and, unlike the sphere, may not even be uniformly curved.
Determining the length between the origin and the point is therefore more difficult
and renders the finite position vector a far less useful concept for a non-Euclidean
manifold. However, in the Euclidean case, the straight-line position vector is
uniquely determined by the choice of origin and the point defining the vector tip.
It is important to note that using different coordinate systems does not alter the
Euclidean vector rp as long as the two coordinate systems share the same origin.
In order to give a mathematical representation to rp , a set of n direction vectors {ej } is
introduced at every point in the manifold. The direction vectors are a complete set of
independent directions at each point, referred to as a local basis, and are defined in a manner
consistent with the chosen coordinate system. The definition of the set {ej } is obtained from
the relationship between partial derivatives and directions. For a particular choice of
coordinate system, the position vector of a point p will be a function of the coordinates x of
the point p, so that r = r (x ). The direction vector ej (x ) at the point p with coordinates x is
defined using the position vector r (x ) and the partial derivative ∂/∂x j ,
1-11
∂r(x ) Δr
e j (x ) = = lim
∂x j Δx j → 0 Δx j
(1.30)
r(x1, … , x j + Δx j , … , x n) − r(x1, … , x j , … , x n)
= lim .
Δx j → 0 Δx j
Definition (1.30) shows that there are n basis vectors ej , since there are n coordinates.
Understanding definition (1.30) begins by noting that the numerator is the
difference in the position vector at infinitesimally separated points,
Δr = r(x1, … , x j + Δx j , … , x j ) − r(x1, … , x j , … , x n). (1.31)
Because the partial derivative ∂/∂x j in definition (1.30) holds all coordinates except
xj constant, the difference Δr can be visualized in the following way. As Δx j
changes, the tip of the position vector directed to the point p moves along the line of
constant value for all other coordinates, which is the jth coordinate gridline, and this
defines the vector Δr . The direction of Δr is therefore tangent to the jth gridline, and
this is the direction of ej .
The case of the Euclidean plane is depicted in figure 1.2 for both Cartesian and
polar coordinates with exaggerated values of Δx j . For the case of Cartesian
coordinates (x , y ) in two dimensions, the direction vector ex ∼ Δr /Δx is tangent
to gridlines of constant y, so that ex is everywhere parallel to the x-axis. Although not
depicted, ey is tangent to the gridlines of constant x and is everywhere parallel to the
y-axis. The Cartesian direction vectors therefore maintain their orientations
throughout a Euclidean manifold. However, choosing polar coordinates (r, θ ) gives
a direction vector e θ ∼ Δr /Δθ that is tangent in the limit Δθ → 0 to the circular
gridlines of constant r. Although not depicted, er is tangent to gridlines of constant θ,
which are straight lines radiating from the origin. As a result, er is radial at every
point around the origin. It should be clear that both er and e θ change orientation at
different points in the plane, and that the local directions of both are dependent on
the angular coordinate, θ. Although not always displayed, the polar coordinate
direction vectors are therefore functions of θ and should be written er(θ ) and e θ(θ ).
As a result, a general direction vector ej may depend on the point where it is defined,
Figure 1.2. Euclidean direction vectors in Cartesian and polar coordinates.
1-12
and so it will be written ej (x ) when it is important to remind the reader of this

dependence.
It is important to note that the direction vectors are tangent to the gridlines at
every point, even if the gridlines are not straight, as is the case with polar
coordinates. If Δx j has units of length, then ej has no physical dimensions and
solely represents the jth direction, an intrinsic property of space. This is the case with
Cartesian coordinates in a flat Euclidean space, where figure 1.2 shows that the
length of the arrow Δr is congruent to the length of Δx by parallel projection. The
ratio of the lengths in Δr /Δx ∼ ex is therefore one in Cartesian coordinates, and the
Cartesian direction vectors are said to be normalized. Because the gridlines are
perpendicular to each other in Cartesian coordinates, the associated direction
vectors are also perpendicular or orthogonal. The Cartesian direction vectors are
therefore orthonormal. However, in the case of polar coordinates, the arc length of
Δr is given by r Δθ , and so the ratio of the lengths in Δr /Δθ ∼ e θ is given by r. While
er is normalized, e θ is not. This will be remedied later when the gradient is
introduced. The definition (1.30) for direction vectors generalizes to non-
Euclidean manifolds, where all members of the set of direction vectors {ej (x )} are
tangent to the manifold at the point described by the coordinates x.
In a non-Euclidean manifold, where finite position vectors are not well defined,
these definitions can still be made in the neighborhood of a point p. The differential
operator d can be applied to the position vector r (x ) to find the infinitesimal change
dr (x ) at the point p for arbitrary infinitesimal coordinate changes dx j . Combining its
Leibniz property with the definition (1.30) and the summation convention yields
∂r(x )
dr ( x ) = dx j = dx j ej (x ), (1.32)
∂x j
where {ej (x )} is the same set of direction vectors that were defined by (1.30). The
differential dr (x ) is an infinitesimal vector that is a linear sum of infinitesimal
component vectors dx (j ) ej (x ). The quantity dx j is referred to as the jth component of
the vector dr (x ) and determines the magnitude of the contribution from ej . The
method used to add these component vectors so that they sum to the original vector
dr will be discussed in detail in the next section. However, the form of (1.32) shows
that the set of direction vectors {ej (x )} provides a basis set to expand a vector into a
linear sum, at least locally in the neighborhood of the point with the coordinates x. It
is important to note that the form of (1.32) allows the position vector to change
direction at a point through the choice of dx j consistent with the curvature of the
manifold.
In a Euclidean manifold the result (1.32) can be extended to a finite position
vector r (x ) by integration. For example, for the Euclidean plane the finite position
vector for the point (x , y ) in Cartesian coordinates is given by
r(x ) r(x ) x y
r (x ) = ∫0 dr = ∫0 (dx ex + dy ey) = ex ∫0 dx + e y ∫0 dy
(1.33)
= x ex + y e y .
1-13
Result (1.33) breaks into straight-line integrations from the origin to the vector tip
along the component vectors of dr , reflecting the independence of the directions.
Because the Cartesian unit vectors are constant throughout a Euclidean manifold,
they can be factored from the integrand. Result (1.33) shows that the use of
Cartesian coordinates in a Euclidean space results in a position vector with
the Cartesian coordinates of the tip as its components, so that r k (x ) = x k . Using
the Cartesian result ∂ek /∂x j = 0 shows that (1.33) satisfies
∂r(x ) ∂ ∂ej
k
= k
(x j ej ) = δkj ej + x j = ek (1.34)
∂x ∂x ∂x k
for all k, and so it is consistent with definition (1.30).
By extension of (1.33), a finite Euclidean position vector in an arbitrary
coordinate system will take the general form
r(x ) = r1(x ) e1(x ) + r 2(x ) e 2(x ) + ⋯ + r n(x ) e n(x ) = r j (x ) ej (x ), (1.35)
where {ej (x )} is the set of direction vectors at the tip of the vector. In expression (1.35)
r j (x ) are the components of the vector r (x ). In a general coordinate system, r j (x ) are
functions of the coordinates x of the point p that defines the vector. The form of
the components will also depend on the direction vectors, which are determined by
the choice of coordinate system. However, in an arbitrary coordinate system the
components rj in general do not coincide with the coordinates of the vector tip.
Definitions (1.30) and (1.32) show that the direction vectors transform identically
to the partial derivative under a coordinate transformation. Denoting the direction
vectors in the y-coordinate system as {e′j (y )}, it follows that
∂r(y ) ∂x k(y ) ∂r(y(x )) ∂x k(y )

e′j (y ) = = = ek(x(y )). (1.36)
∂y j ∂y j ∂x k ∂y j
Combining the transformation (1.36) with the chain rule shows that the differential
vector dr (x ), like the differential d itself, is invariant under coordinate
transformations,
∂r(y ) ∂y j ∂r(y ) ∂r(y(x ))
dr ( y ) = dy j j
= dx k k = dx k = dx k ek(x ) = dr(x ). (1.37)
∂y ∂x ∂y j ∂x k
An example of using (1.36) is the coordinate transformation from Cartesian to polar
coordinates given by (1.1). The polar direction vectors are given by
∂r ∂x ∂r ∂y ∂r
er = = + = cos θ ex + sin θ e y , (1.38)
∂r ∂r ∂x ∂r ∂y
∂r ∂x ∂r ∂y ∂r
eθ = = + = −r sin θ ex + r cos θ e y . (1.39)
∂θ ∂θ ∂x ∂θ ∂y
These two results can be found using Euclidean geometry and trigonometry, but
are much more easily obtained from the covariant transformation formula (1.36).
1-14
Result (1.39) shows that e θ is indeed proportional to r. It will be useful to note that
(1.38) allows the two-dimensional finite Cartesian position vector (1.33) to be
expressed in polar coordinates as
r = x ex + y e y = r cos θ ex + r sin θ e y = r(cos θ ex + sin θ e y) = r e r . (1.40)
Using the chain rule, the transformation (1.36) can be inverted to find
⎛ ∂x ℓ(y ) ∂y j (x ) ⎞
ek(x ) = δkℓ e ℓ(x ) = ⎜ ⎟e ℓ(y(x ))
⎝ ∂y j ∂x k ⎠
(1.41)
∂y j (x ) ⎛ ∂x ℓ(y ) ⎞ ∂y j (x )
= ⎜ e ℓ ( y( x )) ⎟= e′j (y(x )),
∂x k ⎝ ∂y j ⎠ ∂x k
which will be useful in understanding the behavior of the vector components.
Exercise 1.4. The relationship between Cartesian coordinates (x, y, z ) and spherical
coordinates (r, θ , ϕ ) in three dimensions is given by
x = r sin θ cos ϕ , y = r sin θ sin ϕ , z = r cos θ ⟺

z y (1.42)
r 2 = x 2 + y 2 + z 2, θ = cos−1 , ϕ = tan−1 .
r x
Show that the direction vectors in spherical coordinates are given in terms of the
Cartesian direction vectors by
e r = sin θ cos ϕ ex + sin θ sin ϕ e y + cos θ ez ,
e θ = r cos θ cos ϕ ex + r cos θ sin ϕ e y − r sin θ ez , (1.43)
e ϕ = − r sin θ sin ϕ ex + r sin θ cos ϕ e y ,
and that the Cartesian position vector (1.33) in three dimensions, r = x ex + y ey + z ez ,
is given by
r = r er . (1.44)
Because both ej and ∂j play the same directional role, one for vectors and the
other for functions, there is a directional equivalence between the two sets, denoted
by {ej } ↔ {∂j}. This directional equivalence is at the foundation of differential
geometry, and since the work of Cartan the two sets, {ej } and {∂j}, have been used
interchangeably as a direction basis. In Cartan’s notation the vector v is written
∂
v(x ) = v j (x ) , (1.45)
∂x j
and is made concrete by examining its effect on a scalar function φ(x ), which is given
by v(x )φ(x ). Because φ(x ) is an arbitrary function, this creates a vector space
at point x. This is discussed again in the sections on trajectories and tangent spaces.
1-15
It is the covariant property of {ej } that renders the combination (1.37) invariant
under coordinate changes. The basis vectors {ej } are referred to as the legs of a
vector, and so they are often referred to as zweibein, dreibein, vierbein, and vielbein,
depending on whether there are two, three, four, or many legs.
It is possible to consider a transformation between two coordinate systems that
changes the origin. An example of such a transformation is depicted for the two-
dimensional case in figure 1.3. For simplicity, the x-coordinate system will be chosen
to be Cartesian. The inverse coordinate transformation x k (y ) has the property that
x k (0) = a k , which are the x-coordinates of the origin in the y-coordinate system. For
such a case, the finite position vector can be written
r (x ) = x k (x ) e k (x ) = (x k (x ) − a k )e k (x ) + a k e k (x )
≡ y k (x ) e k (x ) + a k e k (x ) (1.46)
= r˜(x ) + a(x ).
The original vector r (x ) is now expressed as the sum of two vectors, r̃ (x ) and a(x ),
where a is the position vector of the y-coordinate origin in the x-coordinate system,
while r̃ (x ) is the position vector of the point p in the new coordinate system, where r̃
has the components y k = x k − a k . The latter follows from the fact that the
components of r̃ (x ) are zero when y j = 0, which occurs if x k = a k . The direction
vectors are defined at point p, and so r̃ is the position vector of point p defined using
the new origin. In an arbitrary second coordinate system, (1.41) gives
⎛ ∂y j (x ) ⎞
r˜(x ) = r˜ k(x ) ek(x ) = e′j (y )⎜r˜ k(x ) ⎟ = r˜ ′ j (y ) e′j (y ) = r˜(y ). (1.47)
⎝ ∂x k ⎠ x(y )
The two vectors, r̃ (x ) and r̃ (y ), are the same vector since they share the same origin
and the same tip. Result (1.47) gives
∂y j (x )
r˜ ′ j (y ) = r˜ k(x ) . (1.48)
∂x k x(y )
Figure 1.3. A coordinate transformation involving a translation of the origin.
1-16
This shows that the finite position vector components transform as contravariant
quantities, but only for the components of the vector emanating from the origin of
the new coordinate system. This is consistent with the contravariant nature of the
infinitesimal components dx j of dr . Because the origin has been moved or
translated, such a coordinate transformation is said to include a translation. If there
is no translation of the origin, the r̃ k (x ) components in (1.48) are replaced with r k (x ).
Result (1.48) can be clarified by considering the transformation in the Euclidean
plane from two-dimensional Cartesian coordinates (x1, x 2 ) to two-dimensional polar
coordinates (r̃ , θ ) for the case depicted in figure 1.3. Such a transformation is
expressed mathematically by
⎛ x2 − a2 ⎞
r˜ = (x1 − a1)2 + (x 2 − a 2 )2 , θ = tan−1 ⎜ 1 ⎟ ⟺
⎝ x − a1 ⎠ (1.49)
1 1 2 2
x = a + r˜ cos θ , x = a + r˜ sin θ .
The vector a = a j ej is a position vector from the original Cartesian coordinate origin
to the point (a1, a 2 ) in Cartesian coordinates. Applying formula (1.48) requires the
identification of x˜1 = x1 − a1 and x˜ 2 = x 2 − a 2 as the Cartesian coordinates of the
point measured with respect to the new origin. Using these in (1.48) in combination
with (1.49) gives the position vector components in the translated polar coordinate
system,
⎛ 1 ∂˜r ⎛ 1 ⎞
∂˜r ⎞ (x − a1)2 + (x 2 − a 2 )2
r˜ ′ 1 = ⎜x
˜ + x˜ 2 ⎟ = ⎜⎜ ⎟
⎟ = r˜ ,
⎝ ∂x1 ∂x 2 ⎠ x(r˜ ) ⎝ (x1 − a1)2 + (x 2 − a 2 )2 ⎠ x(r˜ )
⎛ ∂θ ∂θ ⎞ (1.50)
r˜ ′ 2 = ˜1
⎜x + x˜ 2 ⎟
⎝ ∂x1
∂x 2 ⎠ x(r˜ )
⎛ −(x1 − a1)(x 2 − a 2 ) + (x 2 − a 2 )(x1 − a1) ⎞

=⎜ ⎟ = 0.
⎝ (x1 − a1)2 + (x 2 − a 2 )2 ⎠ x(r˜ )
Exercise 1.5. Show that the polar coordinate direction vectors ej (y ) in the
translated coordinate system take the same form as (1.38) and (1.39), although r is
replaced by r̃ , while the trigonometric functions stand for cos θ = (x1 − a1)/r˜ and
sin θ = (x 2 − a 2 )/r˜ .
Using (1.50) shows that the finite position vector r in the original system is
therefore given as the sum
r(x ) = r˜(y ) + a = r˜ e r(y ) + a . (1.51)
1-17
The vector r̃ (x ) does not involve e θ , since the position vector r̃ (x ) points radially
from the new origin at a .
It is worth repeating that a finite position vector r can be expressed in the general
form (1.35) or the specific Cartesian form (1.33) only in a Euclidean manifold. This is
possible because a Euclidean manifold allows unique straight lines everywhere
throughout the manifold. As a result, specifying the origin and the point of interest
in a Euclidean manifold is sufficient to determine the unique straight line between
the two, in turn defining the finite position vector. However, in a general non-
Euclidean manifold, only the infinitesimal change in the position vector drp , which is
defined at each point p by (1.32), is available. This is because the orientation of the
unit vectors at each point will depend on the curvature of the manifold at that point.
As a result, the components of the vector will depend upon the nature of the
manifold between the origin and the final position as well as the path between the
two points. In a non-Euclidean manifold, finite position vectors become cumber-
some and so are seldom, if ever, used. Alternate methods for determining distances
and angles will be developed.
1.7 Vector addition and the scalar product

For position vectors in a Euclidean manifold, the method used to add vectors is
deduced from relation (1.31), which gives
r(x1, … , xj + Δx j , … , x j ) = r(x1, … , x j , … , x n) + Δr . (1.52)
Examining figure 1.2 for the Cartesian case shows that the two vectors on the right-
hand side of (1.52) form two sides of a parallelogram. Their sum, the vector on the
left-hand side of (1.52), is given by the diagonal of the parallelogram. The same
graphical relationship holds in figure 1.3, where the vector r = r˜+a is the diagonal of
the parallelogram formed by a and r̃ . This gives the well-known parallelogram law of
vector addition in a Euclidean manifold. The two vectors being added are treated as
the adjacent sides of a parallelogram, and the resultant vector is the diagonal of the
parallelogram. The parallelogram law can be applied to the component vectors in
(1.35) by adding them sequentially. In the case of Cartesian coordinates, the finite
result (1.33) shows that the two perpendicular components x ex and y ey sum to the
vector directed to the point (x , y ). This is consistent with the parallelogram law,
since the two component vectors define a rectangle whose diagonal is their sum.
While this process has been introduced graphically, it can now be expressed
mathematically.
The mathematical method used to implement the parallelogram law requires the
scalar product for vectors, also known as the inner product or dot product. In
Euclidean n the scalar product of two finite position vectors r1 and r2 is defined as
r1 · r2 = r2 · r1 = r1 r2 cos ϕ12 , (1.53)
where ∣r1∣ and ∣r2∣ are the lengths or magnitudes of the position vectors r1 and r2 ,
respectively, while ϕ12 is the angle between the two vectors measured in the plane
they define. When there is no chance of confusion, the magnitude of the vector r will
1-18
be written ∣r∣ = r . The scalar product of two finite vectors creates a directionless
scalar quantity from them that depends on both their magnitudes and their relative
directions, and in a Euclidean manifold is the perpendicular projection of r1 onto r2 . If
the scalar product of two nonzero vectors is zero, then the two vectors are
perpendicular or orthogonal, since cos ϕ = 0 is satisfied only if ϕ = ±π /2. Since
the angle between a vector r and itself is zero, the magnitude r of the position vector
r in n is given by
r · r = r · r · cos 0 = r 2 . (1.54)
The scalar product therefore yields the magnitude or length of a vector in n . The
length of the position vector r in a Euclidean manifold is understood as the distance
to the origin for the point defining the vector. One of the immediate ramifications
of (1.54) is that the vector ar , where a is an arbitrary real number, has the
magnitude ar .
The parallelogram law for adding two vectors in a Euclidean manifold can now
be formalized. If r1 and r2 are two position vectors emanating from a common
origin, it follows that the magnitude of the vector r = r1 + r2 that results from adding
them is given by
r 2 = r · r = (r1 + r2 ) · (r1 + r2 ) = r1 · r1 + r2 · r2 + 2 r1 · r2
(1.55)
= r12 + r22 + 2 r1r2 cos ϕ12 .
This is the familiar law of cosines for finding the third leg of an oblique triangle,
given two sides that have the angle π − ϕ12 between them. This shows that the zero
vector, defined by 0 = r − r , has zero magnitude and is therefore a point. Once r is
known, the direction of the resultant vector can obtained from the angles between
one of the original vectors and the resultant vector. For example, the angle ϕr r1
between the resultant vector r and r1 is given by
r · r1 (r + r2 ) · r1 r r
cos ϕr r1 = = 1 = 1 + 2 cos ϕ12 . (1.56)
r r1 r r1 r r
Because of the similarity between (1.55) and the law of cosines, vector addition is
often depicted as placing one vector on the tip of the other vector and drawing the
third leg of the triangle that is formed. This is consistent with the intuitive idea that
adding the two position vectors is identical to performing sequential displacements
and determining the final position from vector addition. However, both vectors in
(1.55) and (1.56) are defined with respect to the same origin. As a result, forming a
triangle, or equivalently a parallelogram, requires the second position vector to be
moved or transported to the tip of the other position vector so that it remains parallel
to itself, a process known as parallel transport. Parallel transport in a Euclidean
manifold is straightforward, while the parallel transport of vectors in a non-
Euclidean manifold will be developed in chapter 10.
For a Euclidean manifold, result (1.54) can be combined with (1.35) to obtain the
magnitude of a vector from its components,
1-19
r 2 = r · r = (r j ej ) · (r k ek) = r j r k ej · ek . (1.57)
Result (1.57) can be evaluated for the case in which the vector is in a Cartesian
coordinate system. It was established earlier that Cartesian direction vectors have a
unit length and are tangent to orthogonal coordinate gridlines, which gives
⎧1 if j = k
ej · ek = 1 · 1 · cos ϕjk = ⎨ , (1.58)
⎩ 0 if j ≠ k
so that the Cartesian direction vectors form an orthonormal set and maintain their
orientation everywhere in a Euclidean manifold. It was shown in (1.33) that a finite
Cartesian vector has the Cartesian coordinates of its tip as the components r j = x j ,
with the result that (1.57) becomes
n
r2 = ∑ (x j ) 2 . (1.59)
j=1
Result (1.59) is simply the Pythagorean theorem extended to n dimensions.

There is a second useful form of the Euclidean scalar product of two vectors in
Cartesian coordinates, r1 = x1 ex + y1 ey and r2 = x2 ex + y2 ey . Using (1.58) gives
r1 · r2 = x1x2 ex · ex + x1y2 ex · e y + y1x2 e y · ex + y1y2 e y · e y = x1x2 + y1y2 . (1.60)
Form (1.60) is equivalent to (1.53), which is seen by switching to polar coordinates

and using x1 = r1 cos θ1, y1 = r1 sin θ1, x2 = r2 cos θ2 , and y2 = r2 sin θ2 :
r1 · r2 = x1x2 + y1y2 = r1r2(cos θ1 cos θ2 + sin θ1 sin θ2 ) = r1r2 cos(θ1 − θ2 ). (1.61)
Since the angle between the two vectors is ϕ12 = θ1 − θ2 , result (1.61) is identical to
(1.53).
Even in a Euclidean manifold, a finite position vector in another coordinate
system can take a very different form. For example, it was found in (1.51) that, after
a translation and transformation from two-dimensional Cartesian coordinates into
polar coordinates, the vector r was given by r = r˜+a , where r˜ = r˜ er and
a = a j ej = a1ex + a 2 ey . Result (1.38) gives er = cos θ ex + sin θ ey in terms of the
original Cartesian direction vectors, where θ is the polar angle in the new coordinate
system. The direction vectors give er · er = 1, er · ex = cos θ , and er · ey = sin θ . This
gives
r 2 = r · r = r˜ 2 e r · e r + a 2 + 2r˜ e r · a = r˜ 2 + a 2 + 2ra
˜ j e r · ej
(1.62)
= r˜ 2 + a 2 + 2r˜ cos θ a1 + 2r˜sin θ a 2 .
The last term in (1.62) is identical to 2r˜·a in the Cartesian form (1.60), since
r˜·a = r˜1a1 + r˜ 2a 2 = r˜ cos θ a1 + r˜sin θ a 2 . (1.63)
Result (1.62) is therefore identical in form to (1.55).
1-20
1.8 The metric tensor and distance in manifolds

If a manifold is non-Euclidean, the finite component formula (1.57) and its Cartesian
version (1.59) are no longer valid. However, it is still possible to use the local length
ds of the infinitesimal position vector element dr . This is obtained by applying (1.54)
to (1.32), which gives
ds 2 ≡ dr · dr = (dx j ej (x )) · ( dx k ek(x )) = ej (x ) · ek(x ) dx j dx k

(1.64)
≡ gjk(x ) dx j dx k .
Because of their association with length, the quantities given by

gjk(x ) = ej (x ) · ek(x ) (1.65)
are referred to as the components of the metric tensor g . The metric tensor and its
relation to distance through (1.64) is the key tool used to develop non-Euclidean
geometry and general relativity. From their definition, it follows that the compo-
nents are symmetric, since the scalar product commutes:
gjk(x ) = ej (x ) · ek(x ) = ek(x ) · ej (x ) = gkj (x ). (1.66)
1
As a result, in n dimensions, the metric tensor has 1 + 2 + ⋯ + n = 2 n(n + 1)
independent elements.
The metric tensor inherits the transformation property (1.36) of the two direction
vectors that define it,
∂x k(y ) ∂x ℓ(y )
gij′(y ) = e′i (y ) · e′j (y ) = ek(x(y ))·
∂y i ∂y j
(1.67)
∂x k(y ) ∂x ℓ(y )
e ℓ(x(y )) = gkℓ(x(y )).
∂y i ∂y j
Because two occurrences of the covariant transformation factor ∂x k /∂y j appear in
(1.67), the metric tensor gjk is referred to as a second-rank covariant tensor. It is
important to note that the transformation property (1.67) combines with the
transformation property (1.21) of dx j to render the infinitesimal length (1.64)
invariant under coordinate transformations. Expression (1.64) is, in effect, a double
contraction, and is therefore invariant for the same reason that the single contraction
of (1.27) is invariant.
The distance along any continuous path between two arbitrary points in a
manifold can, in principle, be computed by integrating the form for ds obtained
from (1.64) along that path. It is important to note that (1.64) gives the infinitesimal
length along the path element associated with arbitrary values for dx j , and so it is
not restricted to the Euclidean straight-line path between two points. It includes the
effect of curvature on length through the changing orientation of the local direction
vectors along the path. A simple application of (1.64) shows that it yields the
Pythagorean expression for distance in the Euclidean plane. In Cartesian
1-21
coordinates, the use of (1.58) shows that the metric tensor for the Euclidean plane is
given by g11 = ex · ex = 1, g22 = ey · ey = 1, and g12 = g21 = ex · ey = 0, so that (1.64)
reduces to
ds = gjk dx j dx k = (dx )2 + (dy )2 . (1.68)
The length of a straight-line vector from the origin to the point (X , Y ) can be found
using the relation y = Yx /X for the straight line, so that dy = Y dx /X . This gives
the distance from the origin to the point
r X X
r= ∫0 ds = ∫0 (dx )2 + Y 2(dx )2 / X 2 = ∫0 dx 1 + Y 2 /X 2
(1.69)
2 2 2 2
= X 1 + Y /X = X +Y ,
which is the Pythagorean result (1.59) in two dimensions. It should be stressed that
integrating (1.64) requires choosing a path, and that the choice of path is arbitrary.
In the case just treated, a straight-line path was used, since that is how distance is
defined in a Euclidean manifold. However, non-Euclidean manifolds require
redefining the distance between two points as the length of the shortest path between
the two points, referred to as a geodesic. This will be revisited in the section on the
calculus of variations, where the differential equation for the geodesic will be derived
using (1.64).
Exercise 1.6. Use (1.43) to show that the metric tensor in spherical coordinates has
the components
grr = 1, gθθ = r 2, gϕϕ = r 2 sin2 θ , (1.70)
and that all other components vanish.
1.9 The metric tensor and raising and lowering indices

In addition to providing the infinitesimal element of length, the metric tensor enables
the raising and lowering of indices, in effect converting contravariant and covariant
indices into one another. This can be demonstrated by considering the scalar product
e j ( x ) · dr ,
ej (x ) · dr = ej (x ) · dx k ek(x ) = (ej (x ) · ek(x ))dx k = gjk(x ) dx k ≡ dxj . (1.71)
The definition of the scalar product (1.53) shows that dxj is the perpendicular
projection of dr onto ej . This follows from ej · dr = ej dr cos θ , where θ is by
definition the angle between ej and dr . The quantity dxj transforms covariantly,
since (1.67) and (1.21) combine to give the relationship between dyj and dxk ,
1-22
∂x i (y ) ∂x ℓ(y ) ∂y k (x )
dyj = g jk′ (y ) dy k = giℓ ( x ( y )) dx m
∂y j ∂y k ∂x m
(1.72)
∂x i (y ) ℓ m ∂x i (y ) ℓ ∂x i (y )
= δ m giℓ ( x ) d x = g ( x ) d x = dxi ,
∂y j ∂y j iℓ ∂y j
which is the covariant transformation (1.22). It is then possible to write the

infinitesimal length (1.64) as
ds 2 = gjk dx j dx k = dxj dx j , (1.73)
which is manifestly invariant under a coordinate transformation, since it is a

contraction of the form (1.27).
For a finite position vector r in a Euclidean manifold, the quantity ek · r is the
covariant component rk of the vector r ,
rk = ej · r = ek · r j ej = ek · ej r j = gkjr j , (1.74)
and this gives the length of the finite position vector as

r 2 = r · r = r kr j ek · ej = r kgkjr j = r krk . (1.75)
In the case of Cartesian coordinates, the direction vectors are both fixed in
orientation and orthonormal, resulting in the scalar product (1.58). Since the scalar
products of (1.58) define the metric tensor components, it follows that the covariant
Cartesian components of a position vector given by rj = gjkr k result in rj and rj
coinciding. In a Euclidean manifold the contravariant and covariant components of
a vector in Cartesian coordinates are therefore identical.
In order to clarify the difference between the contravariant and covariant
components of a vector, it is useful to consider position vectors in the Euclidean
plane 2 described in skew coordinates. Figure 1.4 displays an example of skew
coordinates, where the y-axes are not orthogonal, instead forming the angle ϕ ≠ π /2.
The skew coordinates of a point, designated (y1, y 2 ), are related through parallel
Figure 1.4. Contravariant and covariant position vector components in a skew coordinate system.
1-23
projection to the Cartesian coordinates of the same point, designated (x1, x 2 ), by the
coordinate transformation
x1 = y1 + cos ϕ y 2 , x 2 = sin ϕ y 2 ⟺ y1 = x1 − cot ϕ x 2 ,

(1.76)
y 2 = cscϕ x 2 .
It is useful to express the direction vectors in the skew coordinate system, e1 and e2 ,
in terms of the Cartesian direction vectors, ex and ey . This is because the latter are
both fixed in orientation and orthonormal. The skew coordinate direction vectors
are given by (1.36), so that
∂x1 ∂x 2
e1 = ex + e y = ex ,
∂y1 ∂y1
(1.77)
∂x1 ∂x 2
e 2 = 2 ex + e y = cos ϕ ex + sin ϕ e y .
∂y ∂y 2
Using the orthonormality of the Cartesian unit vectors, the direction vectors of
(1.77) give the metric tensor for the choice of skew coordinates,
g11 = e1 · e1 = 1, g12 = g21 = e1 · e 2 = cos ϕ ,

(1.78)
g22 = e 2 · e 2 = cos2 ϕ + sin2 ϕ = 1.
It should be noted that, unlike the Cartesian direction vectors, the skew direction
vectors e1 and e2 are not orthogonal, since the gridlines to which they are tangent are
not orthogonal.
Exercise 1.7. Show that the results of (1.78) can also be obtained from the Cartesian
metric tensor by applying the tensor transformation rule (1.67).
Using the metric tensor shows that the covariant components of vector r in
figure 1.4 are given by
r1 = g11r1 + g12r 2 = r1 + cos ϕ r 2 , r2 = g21r1 + g22r 2 = cos ϕ r1 + r 2 . (1.79)
Comparing results (1.79) to the diagram shows that the covariant components are
indeed the perpendicular projections of the vector onto the two skew coordinate axes
and coincide with the contravariant components only if ϕ = π /2. Finally, the
extension of (1.73) to the finite position vector case gives the magnitude of the
vector r ,
r 2 = rj r j = r1r1 + r2r 2 = (r1 + cos ϕ r 2 )r1 + (cos ϕ r1 + r 2 )r 2

(1.80)
= (r1)2 + (r 2 )2 + 2r1r 2 cos ϕ ,
which is formula (1.55) for the addition of two vectors.
1-24
Similarly, the metric tensor can be used to define the contravariant version of the
direction vectors, denoted e j , by relating them to the covariant direction vectors,
ej = gjk e k . (1.81)
Relation (1.81) can be solved for e k using the quantities h ℓj , which are chosen to have
the property
h ℓj gjk = δkℓ . (1.82)
The quantities h ℓj are readily found using matrix methods, where they are identified
as the elements of the inverse of the metric tensor treated as an n × n matrix. This
method is reviewed in the section on linear algebra in chapter 7. Using (1.82) gives
e j,
h jk ek = h jk gkℓ e ℓ = δℓj e ℓ = e j . (1.83)
The set of direction vectors {e j } forms what is called the reciprocal basis, since these
vectors are orthogonal to the covariant direction vectors {ej }. This follows from
result (1.83), the metric tensor definition (1.65), and property (1.82) of h jk , all of
which combine to give
e k · ej = (h kℓ e ℓ) · ej = h kℓ (e ℓ · ej ) = h kℓ gℓj = δ k j . (1.84)
Because δ k j is invariant under a coordinate transformation, it follows that relation

(1.84) must hold after a coordinate transformation. Using the transformation
property (1.36) of {ej } shows that {e k} obey
⎛ ∂y ℓ ⎞ ⎛ ∂y ℓ ⎞ ⎛ ∂x j ⎞ ⎛ ∂y ℓ ∂x j ⎞
⎜ i e i ⎟ · e′k = ⎜ i e i ⎟ · ⎜ k ej ⎟ = ⎜ i ⎟e i ·
⎝ ∂x ⎠ ⎝ ∂x ⎠ ⎝ ∂y ⎠ ⎝ ∂x ∂y k ⎠
(1.85)
∂y ℓ ∂x j i ∂y ℓ
ej = δ j = = δ ℓ k.
∂x i ∂y k ∂y k
In order for (1.84) to hold, it must be the case that

∂y ℓ i
e′ ℓ = e, (1.86)
∂x i
which identifies {e k} as transforming contravariantly.
It is possible to convert both indices of the metric tensor from covariant to
contravariant; combining this with (1.82) reveals
g jk = h jℓ h ki gℓi = h jℓ h ki giℓ = h jℓ δ k ℓ = h jk . (1.87)
The contravariant version of the metric tensor therefore coincides with h jk , so that
g ij gjk = δ i k . It also follows from the symmetry of gjk that g jk = g kj . Combining the
definition (1.65) of the covariant metric tensor with (1.81) gives
g jk = g jℓg kigℓi = g jℓg ki (e ℓ · ei ) = (g jℓ e ℓ) · (g ki ei ) = e j · e k , (1.88)
1-25
showing that g jk is given by the reciprocal basis scalar product.

As an example, the reciprocal basis will be found for the skew coordinate system
depicted in figure 1.4. Using matrix methods (to be reviewed in chapter 7), it can be
shown that the skew coordinate metric tensor of (1.78) has the associated g ℓk given
by g11 = g 22 = csc 2ϕ and g12 = g 21 = −csc 2ϕ cos ϕ. It is straightforward to show that
these g ℓk satisfy (1.82) for the metric tensor of (1.78). Using these forms for g ℓk gives
the reciprocal basis in terms of the original basis,
e1 = g11e1 + g12 e 2 = csc 2ϕ e1 − csc 2ϕ cos ϕ e 2 ,

(1.89)
e 2 = g 21e1 + g 22 e 2 = −csc 2ϕ cos ϕ e1 + csc 2ϕ e 2 .
It is straightforward to show that the reciprocal vectors of (1.89) satisfy (1.84). For
example,
e1 · e1 = csc 2ϕ e1 · e1 − csc 2ϕ cos ϕ e 2 · e1 = csc 2ϕ (1 − cos2 ϕ) = 1, (1.90)
e1 · e 2 = csc 2ϕ e1 · e 2 − csc 2ϕ cos ϕ e 2 · e 2 = csc 2ϕ (cos ϕ − cos ϕ) = 0. (1.91)

However, the reciprocal vectors are not normalized, since
e1 · e1 = (e1)2 = csc 4ϕ (e1 · e1 + cos2 ϕ e 2 · e 2 − 2 cos ϕ e1 · e 2) = csc 4ϕ

(1.92)
(1 − cos2 ϕ) = csc 2ϕ .
The magnitude of e1 is therefore e1 = ∣cscϕ∣ = 1/∣sin ϕ∣, which is consistent with
g11 = e1 · e1 = (e1)2 = csc 2ϕ.
The metric tensor components gjk and those of its inverse g jk can be used to
express contractions in multiple forms. For example, a position vector can be written
as
r = ej x j = gjℓ e ℓg kjxk = (g kjgjℓ ) e ℓxk = δ k ℓ e ℓxk = e kxk . (1.93)
Even if xk and xk are different, the two contractions of (1.93) are identical. This gives
an important result regarding the scalar product r · r , which can be written in three
equivalent ways,
r 2 = r · r = (ej x j ) · (ek x k ) = (ej · ek)x j x k = gjkx j x k ,

r 2 = r · r = (e j xj ) · (e k xk ) = (e j · e k)xj xk = g jkxj xk , (1.94)
r 2 = r · r = (ej x j ) · (e k xk ) = (ej · e k)x j xk = δ k jx j xk = x j xj .
All three give the same result, although the third version shows that the length of a
vector is the contraction of the covariant and contravariant components. This shows
that the scalar product of two different vectors can be written as
r · r′ = r j rj′. (1.95)
1-26
Exercise 1.8. Treating the two-dimensional plane of figure 1.4 as Euclidean, show
that the contravariant and covariant components of the vector r are given by
sin α sin(θ − α )
r1 = r , r2 = r ,
sin θ sin θ (1.96)
r1 = r cos α , r2 = r cos(θ − α ),
where r is the Euclidean magnitude of the vector r . These immediately verify (1.95),
since r1r1 + r2r 2 = r 2 .
The relationship of xk and xk carries over into partial derivatives. Using the chain
rule gives
∂ ∂x k ∂ ∂(g kℓxℓ ) ∂ ∂
= k
= k
= g kj k = g kj ∂k = ∂ j . (1.97)
∂xj ∂xj ∂x ∂xj ∂x ∂x
This shows that ∂/∂xj transforms contravariantly.

Another important result follows from considering the partial derivative of a
direction vector. An example of this is given by
∂je j = gjk ∂ k(g ℓj e ℓ) = (g ℓjgjk )∂ ke ℓ + gjk(∂ kg ℓj )e ℓ = δ ℓ k ∂ ke ℓ + (∂jg ℓj )e ℓ

(1.98)
= ∂ kek + (∂jg ℓj )e ℓ .
The two forms of the contraction, ∂je j and ∂ kek , are equal only if ∂jg ℓj = 0 for all ℓ .
The Levi-Civita symbol defined in (1.5) can undergo a self-contraction using the
metric tensor that results in zero. For example, in three dimensions
ε jjk = gjℓε jkℓ = gℓjε ℓkj = −gjℓε jkℓ = −ε jjk , (1.99)
where the relabeling of dummy indices was combined with property (1.66) for the
metric, gjℓ = gℓj , and an odd number of permutations for the Levi-Civita symbol,
ε ℓkj = −ε jkℓ . Result (1.99) shows that εjjk = 0.
Exercise 1.9 Show that ε μν…μ = 0, where the number of indices is arbitrary.
1.10 General tensors and tensor densities

The metric tensor transformation property (1.67) serves as the prototype for the
transformation property of a general tensor. A tensor may depend on position, as the
metric tensor does in a non-Euclidean manifold. In a coordinate system a general
tensor T has the components Tkℓij…
…(x ). The superscripts indicate contravariant indices
1-27
and the subscripts indicate covariant indices. The general tensor component trans-
forms with the extensions of (1.21) and (1.22),
′ ij… ∂y i (x ) ∂y j (x ) ∂x k ′(y ) ∂x ℓ ′(y )

T kℓ…(y ) = i′ j′
⋯ k ℓ
⋯T ki ′′ ℓj ′…
′…(x ) . (1.100)
∂x ∂x ∂ y ∂ y x(y )
The rank of a tensor is given by the total number of covariant and contravariant
indices. If a tensor has r contravariant indices and p covariant indices, it referred to
as a type (r, p ) tensor. In this context, the direction vector transformation property
(1.36) identifies them as first-rank covariant tensors, while scalars, such as the
differential operator d of (1.25), are zero-rank tensors.
There is a class of objects that transforms somewhat differently than (1.100);
members of this class are referred to as tensor densities. An example is the Levi-
Civita symbol of (1.5). This can be seen by contracting both sides of relationship
n
(1.14) with the sequence ∏ j = 1 ∂x kj /∂y αj and using the chain rule (1.4). The result is
1 ∂x k1 ∂x kn ∂x k1 ∂y j1 ∂x kn ∂y jn
⋯ ε k …k = ⋯ ε j …j = δ j1α1⋯
J (y ) ∂y α1 ∂y αn 1 n ∂y α1 ∂x k1 ∂y αn ∂x kn 1 n (1.101)
δ jnαnε j1…jn = εα1…αn.
Like the Kronecker delta, the Levi-Civita symbol is a mathematical construct and
therefore coordinate independent. Result (1.101) is therefore an identity that can be
used in defining coordinate invariant expressions involving the Levi-Civita symbol.
The left-hand side of identity (1.101) shows that the Levi-Civita symbol can be
treated as a covariant tensor of rank n under a transformation, but requires an
additional factor of the inverse of the Jacobian of the transformation. This is an
important aspect of defining coordinate invariant expressions involving the Levi-
Civita symbol, since the factor of the Jacobian that appears can be used to cancel
Jacobian factors generated by other quantities. A quantity whose transformation
involves the Jacobian of the transformation is referred to as a tensor density, and the
power of the Jacobian determines its weight. Result (1.17) shows that the Jacobian
J (x ) of the inverse transformation from y-coordinates to x-coordinates is the inverse
of the Jacobian J (y ) of the transformation from x-coordinates to y-coordinates. This
follows from performing the transformation and its inverse successively:
dnx = J (y )dny = J (y )J (x )dnx ⟹ J (x )J (y ) = 1. (1.102)
Result (1.102) shows that J (x ) = 1/J (y ). The covariant components of the Levi-
Civita symbol correspond to a tensor density of weight +1, which refers to the single
positive power of J (x ) appearing on the left-hand side of (1.101).
If a tensor is to represent a physically measurable quantity, it must be expressed in
a coordinate-independent way, as in the case of dr = dx j ej . For a general tensor, this
is done by using the tensor product for the direction vectors, written
ei ⊗ ej ⊗ ⋯⊗ek . The tensor product is understood to be the product of the
directional degrees of freedom. For example, a vector in n can be written x j ej ,
1-28
while a second vector in n can be written y k ek . Their tensor product, sometimes

referred to as a dyad, is written
T = x j ej ⊗ y k ek = x j y k ej ⊗ ek ≡ T jk ej ⊗ ek . (1.103)
There are n2 components, T jk = x j y k , and these components transform as a second-
rank contravariant tensor. When coupled with the transformation property (1.36) of
ej , the tensor T defined by (1.103) is coordinate independent. The tensor product of
the direction vectors ej ⊗ ek serves as the basis for linear functions that map the
vectors w into another vector v through the scalar product,
T (w) = T · w = (T jk ej ⊗ ek) · (w ℓ e ℓ) = (T jkw ℓ ej )(ek · e ℓ) = (gkℓw ℓT jk )ej

(1.104)
= T jkwk ej = v ,
where v j = T jkwk . Definition (1.104) gives T (au + bv) = a T · u + b T · v, where u
and v are any two vectors and a and b are any two real numbers. This property
classifies tensor T as a linear function of vectors. Similarly, covariant tensor
components can be combined with reciprocal basis vectors in a coordinate-
independent manner. For example, the metric tensor can be written
g = gjk e j ⊗ e k . (1.105)
The metric tensor can then be viewed as a scalar function of two vectors. Using
(1.84) gives
g (dr , dr ) = (dx i ei ) · (gjk e j ⊗ e k) · (dx ℓ e ℓ) = gjk dx i dx ℓ δ j i δℓk

(1.106)
= gjk dx j dx k = ds 2 .
In effect, the metric tensor is a linear function that maps two vectors into a distance.
1.11 Trajectories and tangent spaces

So far, attention has primarily been focused on position vectors, either the
infinitesimal version dr of (1.32) or the finite version r of (1.35). Regardless of
whether the manifold is Euclidean or not, the metric tensor relation (1.64) associates
a distance in the manifold with position vectors, i.e. either infinitesimal position
changes or, by integration, finite distance vectors. However, vectors may be present
at every point in the manifold that are independent of the choice of an origin.
Examples from Newtonian physics include the force acting on a particle or its
velocity when it is located at a specific point. Since force and velocity have both a
magnitude and direction, they must be represented by a vector at that point.
Defining vectors at an arbitrary point starts by defining a trajectory r (t ) that
passes through the point. A trajectory is a continuous sequence of points described
by a position vector r (t ) that is a differentiable function of the parameter t. This is
identical to the trajectory of a point particle in Newtonian mechanics, where the
parameter t represents time. In what follows the parameter t will be referred to as the
time, for the moment ignoring any issues around its units. The trajectory is assumed
1-29
to pass through the point p of interest in the manifold when the parameter t has the
value tp, so that r (tp ) = rp is the same position vector of the point p that was used to
define the direction vectors (1.30). The position vector is represented in the x-
coordinate system by r (x ) and the coordinates xk of the vector tip will now be treated
as time-dependent quantities x k (t ). In every other way the trajectory is arbitrary.
The rate at which r (t ) = r (x(t )) changes at p yields the velocity vector vp at the point
p, defined by
dr ( t ) ∂r(x ) dx k(t )
vp = = = ek(x(tp)) x ̇k(tp) ≡ v k (tp) ek(x(tp)), (1.107)
dt tp ∂x k dt tp
where the notations v k (t ) = x ̇k (t ) ≡ dx k (t )/dt and a k (t ) = v̇k (t ) = x ̈k (t )

≡ d2x k (t )/dt 2 will be used from this point on. The form of vk, the kth component
of the velocity vector at the time t, will depend upon the coordinate system chosen as
well as the trajectory. The definition (1.107) shows that the velocity vector at the
point p is described in terms of the same direction vectors (1.30) at the point p that
are used for the position vectors. Definition (1.107) is also consistent with the Cartan
form (1.45) for vectors, since (1.107) can be written as
∂ ⎛ ∂r k ⎞ ∂ ∂
v k (x ) = ⎜v j (x ) j ⎟ k = (v(x )r k(x )) k , (1.108)
∂x k
⎝ ∂x ⎠ ∂x ∂x
although the notation of (1.107) is more suited to Newtonian mechanics.

While the direction vectors ej were initially developed for use with position
vectors, their utility lies in the fact that they exist at every point in the manifold, where
they are tangent to the jth gridline. In turn, this means that the metric tensor of the
manifold at that point will determine the magnitude of the velocity vector,
v 2 = v · v = v j (x )v k (x ) ej (x ) · ek(x ) = gjk(x )v j (x )v k (x ) = v j (x )vj (x ). (1.109)
Statement (1.109) is true regardless of whether the manifold is Euclidean or not. Like
the relationship between direction vectors and gridlines, the velocity vector v(tp ) is
tangent to the trajectory r (tp ) at the point p. This is true since the components of v(tp )
have the same slope as the trajectory in all directions. This follows from
v j (tp) ⎛⎜ dx j (t ) ⎞⎛ k ⎞−1 j
= ⎟⎜ dx (t ) ⎟ = dx ( t p ) , (1.110)
v k (tp) ⎜⎝ dt ⎟⎜
t p ⎠⎝ dt
⎟
tp ⎠
dx k ( t p )
which is identical to the slope of the trajectory at point p for all possible values of j
and k.
Since there is a nondenumerable infinity of trajectories that pass through point p,
the set of all possible tangent vectors at point p given by (1.107) forms a space of
vectors that are tangent to the manifold M at the point p. The term vector space
means that any two vectors belonging to the space can be combined through
addition and scalar multiplication, so that a linear combination of any two tangent
vectors at a point results in a vector that is also tangent to the point. As a result, the
1-30
space of vectors defined in this manner is referred to as the tangent space at the point
p and is denoted by Tp(M). It is important to understand that the vectors in the
tangent space do not lie in the manifold and their magnitude is not correlated with
the distance from the origin to the point. However, the definition of (1.107) shows
that the vectors in the tangent space of a point can be expressed in terms of the same
basis direction vectors used to describe the position vector, either finite or
infinitesimal, at the point. The tangent space therefore has the same dimension as
the manifold, which means that any vector in the tangent space at a point can be
expanded as a linear combination of the normalized direction vectors at the point.
This is because the direction vectors are also tangent to the manifold M at the point
p and therefore form a basis for the tangent vectors. For an n-dimensional manifold
the tangent space Tp(M) at each point coincides with n , and the tangent point
serves as the origin of the tangent space.
The one-dimensional circle and two-dimensional sphere, denoted S1 and S2
respectively, have easily visualized tangent spaces. In the case of the circle, the
tangent space is the line 1 tangent to the circle at the point p, while in the case of the
sphere, it is a plane 2 tangent to the sphere at point p. These are depicted in
figure 1.5 by embedding them in a higher-dimensional space which has the position
vector r defining the direction vectors tangent at that point. In each case the tangent
vectors at point p can be visualized as an arrow in the tangent space. It should be
stressed that the tangent vectors in Tp(M) are defined only at point p and do not
extend to other points in the underlying manifold. In the case of Tp(S1), the polar
coordinate direction vector e θ from the polar coordinates of 2 serves as the single
direction vector required to describe a vector in the tangent space. In the case of
Tp(S2), the spherical coordinate direction vectors e θ and eϕ of 3 serve as orthogonal
direction vectors for the tangent plane. A vector in Tp(S1) takes the form v = v θ e θ ,
while a vector in Tp(S2) takes the form v = v θ e θ + v ϕ eϕ . It is possible to perform a
coordinate transformation in the manifold and use the new direction vectors as the
basis in the tangent space. This transformation also requires the tangent vector
components to undergo the contravariant transformation, just as the components of
position vectors are transformed. However, the tangent space and the manifold only
Figure 1.5. The tangent spaces for the circle and the sphere, Tp(S1) and Tp(S2).
1-31
share the tangent point p, which serves as the origin of the tangent space and the
point of definition of the direction vectors.
1.12 The vector product

A second form of vector multiplication, known as the cross or vector product, defines
a vector that is orthogonal to a collection of other vectors. The version of the vector
product presented in this section will be specific to three spatial dimensions. A
related definition of the vector product for higher dimensions is given in a later
section. The vector product is quite often used in three-dimensional quantities, such
as angular momentum and the Lorentz force. However, it is less useful in arbitrary
dimensions and has largely been superseded by the wedge product to be discussed
later in this chapter. The definition of the vector product in three dimensions uses the
orthonormal Cartesian unit vectors ej ,
ej × ek = εjkℓ e ℓ , (1.111)
where εjki is the covariant Levi-Civita symbol defined by (1.5) which obeys the
identity (1.101). Because εijk = −εjik , definition (1.112) has the property that
ej × ek = −ek × ej , so that ej × ej = 0.
In three dimensions the vector product gives rise to the vector product of two
Euclidean position vectors in Cartesian coordinates,
x × y = x i y j (ei × ej ) = εijkx i y j e k . (1.112)
The vector given by (1.112) is orthogonal to both x and y , since

x · (x × y ) = εijkx ix j y k = 0. If both position vectors x and y lie in the 1-2 plane,
so that x 3 = y3 = 0, their vector product is given by
x × y = x i y j εijk e k = x i y j εij3e 3 = (x1y 2 − x 2y1)e 3. (1.113)
Expressing the Cartesian components of the two vectors in polar coordinates,

x1 = r cos θ , x 2 = r sin θ , y1 = r′ cos θ′, and y 2 = r′ sin θ′, the vector product (1.113)
becomes
x × y = rr′(cos θ sin θ′ − sin θ cos θ′)e 3 = rr′ sin(θ′ − θ ) e 3. (1.114)
If the two vectors are parallel then θ′ = θ and the vector product vanishes, which is
consistent with x × x = 0. Euclidean geometry shows that the magnitude of the
vector product (1.114) is the area of the parallelogram formed by the two vectors,
while its direction is perpendicular to the plane the two vectors form. For the case of
e 3 = e z , the vector points in the +z direction if θ′ − θ > 0. This is often referred to as
the right-hand rule, since it is identical to a right-handed coordinate system where the
right thumb aligns along the z-axis when the first two fingers are aligned with the
x- and y-axes, respectively.
Defining the three unsummed infinitesimal vectors drj = dx (j )ej and using (1.111)
gives
1-32
dri × drj = dx (i ) dx (j ) ei × ej = (dx (i ) dx (j ) εijk )e k ≡ εijk dS k , (1.115)
where dS k is the kth contravariant component of the oriented surface element. This is
understood by considering the case i = 1 and j = 2 for (1.115), which shows that the
infinitesimal surface element dS (3) = dS z = dx dy is a rectangular surface element in
the x–y plane. This infinitesimal element is associated with the direction vector e z ,
which is the direction normal to the surface element in three dimensions. The
oriented surface element is written dS = dS nˆ , where n̂ is the unit vector perpen-
dicular or normal to the plane defining the infinitesimal surface element dS . For
example, using the spherical coordinate direction vectors (1.43) and ex × ey = ez ,
ez × ex = ey , and ez × ey = −ex shows that
e θ × eϕ = r 2 sin θ (sin θ cos θ ex + sin θ sin ϕ e y + cos θ ez) = r 2 sin θ e r . (1.116)
The surface element of a sphere is therefore given by

dS = dθ dϕ (e θ × eϕ) = r 2 sin θ dθ dϕ e r = dS(r ) e r , (1.117)
so that the oriented surface elements of a sphere are radial and perpendicular to the
surface of the sphere.
Exercise 1.10. Verify (1.117).
Similarly, definition (1.111) also gives the Cartesian result

ei · (ej × ek) = εjkℓ ei · e ℓ = εjkℓ δ ℓ i = εjki = εijk , (1.118)
which vanishes if any of the indices on the left-hand side are the same. Result (1.118)
is related to the three-dimensional volume element d3x = dx dy dz in the following
way. Using (1.118) gives
dri · (drj × drk ) = dx (i ) dx (j ) dx (k ) ei · (ej × ek) = εijk d3x , (1.119)
where d3x = dx (1) dx (2) dx (3) is the infinitesimal volume element in Cartesian
coordinates. Relation (1.119) is understood to be an oriented volume element, an
infinitesimal parallelepiped with an oriented base rectangle of drj × drk and a height
of dri . The relation (1.118) can be subjected to a coordinate transformation by
contracting it with factors of ∂x i /∂y ℓ , giving
∂x i ∂x j ∂x k ∂x i ∂x j ∂x k
ℓ m
e · (e j × e k ) =
n i
εijk . (1.120)
∂y ∂y ∂y ∂y ℓ ∂y m ∂y n
Using (1.36) shows that the left-hand side of (1.120) gives the product of the
transformed direction vectors, denoted by e′ℓ , while identity (1.101) can be employed
on the right-hand side. The result is
1-33
e′ℓ · (e′m × e′n) = J (y ) εℓmn, (1.121)

which shows that the product of the magnitudes of the transformed direction vectors
is given by the Jacobian of the coordinate transformation. Combining dr′j = dy (j )e′(j )
with result (1.121) gives
drℓ′ · (drm′ × drn′) = dy (ℓ ) dy (m ) dy (n) e′ℓ · (e′m × e′n) = J (y ) d3y εℓmn, (1.122)
which is the transformed version of (1.119). The appearance of the Jacobian in

(1.122) demonstrates the rule (1.17) for the transformation of three-dimensional
infinitesimal volume elements. Result (1.121) also shows that the transformed
direction vectors are normalized by the Jacobian,
e′ℓ · (e′m × e′n)
eˆ′ℓ · (eˆ′m × eˆ′n) = = εℓmn, (1.123)
J (y )
where the associated unit vectors ê′j are given by
−1 −1
eˆ′j = e′j (e′(j ) · e′(j )) 2
( )
= e′j g(jj′ ) 2. (1.124)
This shows that the Jacobian of the transformation for three dimensions is given by
3
J (y ) = ∏ e′(j ) · e′(j ) . (1.125)
j=1
Exercise 1.11. Use the spherical coordinate direction vectors (1.43) and expression
(1.125) to show that the Jacobian for transforming from three-dimensional Cartesian
coordinates to spherical coordinates (1.42) is J (r, θ , ϕ ) = r 2 sin θ . Verify that
er · (e θ × eϕ) = r 2 sin θ = J (r, θ , ϕ ).
The results of this section, such as (1.115) and (1.119), have been limited to three-
dimensional spaces. In a later section this limitation is lifted by defining the wedge
product. However, this requires further developments.
1.13 The gradient

The relationship between direction vectors and partial derivatives is encapsulated in
the gradient, a key mathematical tool in both differential geometry and field theory.
The gradient, denoted ∇, is defined by combining the contravariant reciprocal
vectors e k of (1.84) with the covariant partial derivatives ∂/∂x k ,
∂
∇ = e k (x ) ≡ e k (x )∂ k . (1.126)
∂x k
1-34
It is important to note that the gradient uses the reciprocal basis at the same point
where the partial derivatives will be evaluated. In the case of a Euclidean manifold
with Cartesian coordinates, these reciprocal vectors are the same everywhere.
However, in other coordinate systems and in non-Euclidean manifolds, the direction
vectors and their reciprocals change with location. As a result, the order of the two
components in (1.126) is critical, since it cannot be assumed that ∂ke j = 0. Because it
takes the familiar form (1.27) of a contraction, the gradient is a coordinate invariant
vector operator. The action of the gradient on a scalar function f (x ), whose
argument x is the coordinates of the point of interest, yields a vector at r (x ) given by
∂f (x )
∇f (x ) = e j (x ) = e j (x ) ∂jf (x ). (1.127)
∂x j
The gradient appears in the application of the differential operator (1.23) to a scalar
function f (x ) of several variables, which can be written
∂f (x ) ∂ ∂
df ( x ) = dx k k
= dx k δ j k j
f (x ) = dx k ek · e j j f (x ) = dr · ∇f (x ). (1.128)
∂x ∂x ∂x
Using (1.84) shows that
ek(x ) · ∇f (x ) = ek(x ) · e j (x ) ∂jf (x ) = δ j k ∂jf (x ) = ∂kf (x ), (1.129)
which is the kth covariant component of ∇f (x ).
In that regard, the differential of a single-valued multivariable function f (r )
can be combined with the concept of the tangent space to find its extremum points.
For example, the extrema of a function f (x , y ) of two variables can be
found by extending consideration to a three-dimensional Euclidean space and
three-dimensional trajectory
r(t ) = x(t )eˆ x + y(t )eˆ y + f (x(t ), y(t ))eˆ z , (1.130)
where x(t ) and y(t ) are arbitrary trajectories through the x–y plane. This trajectory
traces a path through the surface defined by f (x , y ). The function f (x , y ) has
extrema at those points where the tangent plane of the trajectory is orthogonal to êz ,
making it parallel to the x–y plane. This gives
dr ( t ) ∂f (x , y ) ∂f (x , y ) dr
eˆ z · = ẋ + ẏ = · ∇f (x , y ) = 0. (1.131)
dt ∂x ∂y dt
Since v = dr (t )/dt is arbitrary, the extrema of a multivariable function are
determined as those points where
∇f = 0 ⟹ df (r ) = dr · ∇f (r ) = 0, (1.132)
which is the usual criterion for an extremum point.
In coordinate systems other than Cartesian, a great deal of confusion regarding
units can be avoided by rewriting (1.127) in terms of unit vectors. This is done using
the metric tensor to swap the indices, as in (1.93), and then normalizing the direction
vectors by the replacement ej = e(j )eˆ j , where (ej )2 = e (j ) · e (j ). This gives
1-35
∇f (x ) = e k(x ) ∂kf (x ) = ek(x ) ∂ kf (x ) = eˆ k(x )(e(k )(x ) ∂ kf (x ))

(1.133)
= eˆ k(x )V k(x ) ≡ V (x ).
The action of the gradient on a scalar function creates a vector function
V (x ) = ∇f (x ) at the point whose coordinates are x. It is important to note that
the vector function V (x ) exists at all points in the manifold where the gradient of the
function is well defined, and so the gradient creates a vector field throughout the
manifold once the function f (x ) is specified. The components of the vector field,
V k (x ), depend on the coordinate system chosen to define the direction vectors. The
vector field V (x ) is the sum of component vectors directed along the direction
vectors, and so the vector field defined by (1.133) is also tangent to the manifold at
that point, just as the direction vectors are. This is discussed further in the next
section.
The form of the gradient depends upon the choice of coordinate system. For
consistency, the gradient (1.126) will be expressed in terms of dimensionless unit
vectors using the method of (1.133). This gives
j
∇ = e k(x ) ∂k = ej (x ) g jk(x ) ∂k = eˆ j (x )(e(j )(x ) g jk(x )∂k) ≡ eˆ j (x ) ∂ˆ . (1.134)
Definition (1.134) is one of the occasions for which there is a sum over j even though
(j ) occurs in the expression. In the case of Cartesian coordinates in a Euclidean
space, the orthonormality of the direction vectors shows that the reciprocal vectors
e k coincide with the unit vectors êk . This follows from result (1.58), which shows that
gjk = 1 if j = k and is zero otherwise. This gives both ej = (g(jj ) )1/2 = 1 and g jk = 1 if
j = k, with g jk equal to zero if j ≠ k . In three dimensions this gives the Cartesian
gradient
∂ ∂ ∂
∇ = eˆ j (e(j ) g jk ∂k) = eˆ x + eˆ y + eˆ z . (1.135)
∂x ∂y ∂z
This Cartesian form extends to a Euclidean manifold with an arbitrary dimension.
For spherical coordinates the direction vectors were found in (1.43). The metric
tensor components are
g11 = e r · e r = er2 = 1, g22 = e θ · e θ = eθ2 = r 2 ,

(1.136)
g33 = eϕ · eϕ = eϕ2 = r 2 sin2 θ ,
and all other components are zero. Because the other components are zero, the
contravariant or inverse metric tensor components can be found simply by inverting
these results. This gives
1 1
g11 = 1, g 22 = , g 33 = , (1.137)
r2 r sin2 θ
2
and all other components are zero. Using (1.136) and (1.137) in (1.134) gives the
gradient in spherical coordinates in terms of unit vectors,
1-36
∂ ∂ ∂
∇ = e k ∂k = e r + eθ + eϕ = eˆ j (e(j ) g jk ∂k)
∂r ∂θ ∂ϕ
(1.138)
∂ 1 ∂ 1 ∂
= eˆ r + eˆ θ + eˆ ϕ .
∂r r ∂θ r sin θ ∂ϕ
The spherical coordinate unit vectors are given explicitly in terms of the Cartesian
unit vectors by
eˆ r = sin θ cos ϕ eˆ x + sin θ sin ϕ eˆ y + cos θ eˆ z = e r = e r ,
e
eˆ θ = cos θ cos ϕ eˆ x + cos θ sin ϕ eˆ y − sin θ eˆ z = θ = r e θ ,
r (1.139)
eϕ
eˆ ϕ = − sin ϕ eˆ x + cos ϕ eˆ y = = r sin θ e ϕ .
r sin θ
At each angular position (θ , ϕ ) the unit vectors of (1.139) form an orthonormal set.
It is useful to note that the covariant components of the gradient in spherical
coordinates are given by
∂ ∂ ∂
∂1 = , ∂2 = , ∂3 = , (1.140)
∂r ∂θ ∂ϕ
while the contravariant components are found using the inverse metric tensor
(1.137),
∂ 1 ∂
∂ r = ∂1 = g1j ∂j = , ∂ θ = ∂ 2 = g 2j ∂j = ,
∂r r 2 ∂θ
(1.141)
1 ∂
∂ ϕ = ∂ 3 = g 3j ∂j = 2 2
.
r sin θ ∂ϕ
Exercise 1.12. Show that the contravariant components of the gradient in spherical
coordinates given by (1.141) obey the contravariant transformation law (1.21) by
considering the explicit transformation from spherical coordinates to Cartesian
coordinates.
Exercise 1.13. Cylindrical coordinates (ρ, ϕ, z ) are related to Cartesian coordinates

(x , y , z ) in three dimensions by
x = ρ cos ϕ , y = ρ sin ϕ , z=z ⟺ ρ= x2 + y2 , ϕ = tan−1(y / x ),

(1.142)
z = z.
Show that the cylindrical coordinate direction vectors are given in terms of Cartesian
unit vectors by
eρ = cos ϕ ex + sin ϕ e y , e ϕ = − ρ sin ϕ ex + ρ cos ϕ e y , ez = ez . (1.143)
1-37
Show that a finite Euclidean position vector in cylindrical coordinates is given by

r = ρ eρ + z e z , (1.144)
and that the gradient in cylindrical coordinates is given in terms of normalized unit
vectors by
∂ ∂ ∂ ∂ 1 ∂ ∂
∇ = e k ∂k = e ρ + eϕ + ez = eˆ ρ + eˆ ϕ + eˆ z . (1.145)
∂ρ ∂ϕ ∂z ∂ρ ρ ∂ϕ ∂z
1.14 The divergence, the Laplacian, and the curl

A general vector field V (x ), not necessarily given by ∇f (x ), can be combined with the
gradient in several ways. Expressing the vector field in terms of unit vectors,
V (x ) = eˆ kV k (x ), the divergence of V (x ) is a scalar quantity defined by
∇ · V (x ) = e j ∂j · (eˆ kV k(x )) = (e j · eˆ k)∂jV k(x ) + e j · (∂jeˆ k)V k(x )

1 (1.146)
= ∂kV k(x ) + e j · (∂jeˆ k)V k(x ),
e(k )
where the position dependence of the direction vectors is implicit. In that regard,
Cartesian coordinate direction vectors are constant and normalized, so that ∂jeˆ k = 0
and e(k ) = 1. The Cartesian form of the divergence in three dimensions is therefore
given by
∂V 1(x ) ∂V 2(x ) ∂V 3(x )

∇ · V (x ) = ∂ kV k (x ) = + +
∂x1 ∂x 2 ∂x 3
x y z (1.147)
∂V (x ) ∂V (x ) ∂V (x )
= + + .
∂x ∂y ∂z
The spherical coordinate version of the divergence requires the use of (1.139) to
show that
∂θeˆ r = eˆ θ , ∂θeˆ θ = −ê r , ∂ϕeˆ r = sin θ eˆ ϕ , ∂ϕeˆ θ = cos θ eˆ ϕ ,

(1.148)
∂ϕeˆ ϕ = −sin θ eˆ r − cos θ eˆ θ ,
and all other derivatives vanish. Combining (1.148) with eθ = r and eϕ = r sin θ gives
∂V r(x ) 1 ∂V θ(x ) 1 ∂V ϕ(x ) 2

∇ · V (x ) = + + + V r (x )
∂r r ∂θ r sin θ ∂ϕ r
(1.149)
cot θ θ
+ V (x ).
r
1-38
Exercise 1.14. Show that the divergence in cylindrical coordinates is given by

∂V ρ(x ) 1 ∂V ϕ (x ) ∂V z (x ) 1
∇ · V (x ) = + + + V ρ(x ). (1.150)
∂ρ ρ ∂ϕ ∂z ρ
Using the definition of the gradient (1.126), the Laplacian operator is defined as
∇2 = ∇ · ∇ = (e k∂k) · (e j ∂j). (1.151)
The Laplacian is a coordinate invariant scalar operator that acts on functions. Its
action on a function f (x ) is given by
∇2 f (x ) = ∇ · ∇f (x ) = e k · ∂k(e j ∂jf (x ))
(1.152)
= (e k · e j )∂k∂jf (x ) + (e k · ∂ke j )∂jf (x ).
In the case of Cartesian coordinates, ∂ke j = 0 and the second term in (1.152)
vanishes. The orthonormality of the Cartesian reciprocal vectors shows that the
three-dimensional Laplacian in Cartesian coordinates is
∂2 ∂2 ∂2
∇2 = 2
+ 2 + 2. (1.153)
∂x ∂y ∂z
The Cartesian result (1.153) has the same general form in lower or higher
dimensions. However, in the case of spherical coordinates, the formulas of (1.139)
show that the first term in (1.152) generates nonzero contributions from the
following cross terms,
eθ 1 eϕ 1
e θ · ∂ θe r = e θ · = , e ϕ · ∂ϕe r = e ϕ · = ,
r r r r
(1.154)
cos θ cot θ
e ϕ · ∂ϕe θ = e ϕ · 2 eϕ = 2 .
r sin θ r
Using (1.154) in the second term in (1.152) gives the spherical coordinate Laplacian,
∂ ∂
∇2 = (e k · e j )∂k∂j + (e θ · ∂θe r + e ϕ · ∂ϕe r)
+ (e ϕ · ∂ϕe θ)
∂r ∂θ
(1.155)
∂2 1 ∂2 1 ∂2 2 ∂ cot θ ∂
= 2 + 2 2 + 2 2 + + 2 .
∂r r ∂θ r sin θ ∂ϕ 2 r ∂r r ∂θ
Exercise 1.15. Use (1.143) in (1.151) to show that the Laplacian in three-dimensional
cylindrical coordinates is given by
∂2 1 ∂ 1 ∂2 ∂2
∇2 = 2
+ + 2 2 + 2. (1.156)
∂ρ ρ ∂ρ ρ ∂ϕ ∂z
1-39
In three dimensions the curl of V (x ), denoted by ∇ × V (x ), is defined using the

vector product (1.111),
∇ × V (x ) = e k∂k × eˆ j V j (x ) = (e k × eˆ j )∂kV j (x ) + (e k × ∂keˆ j )V j (x ). (1.157)
The expressions in (1.157) can be evaluated using the direction vectors in various
coordinate systems. In a three-dimensional Euclidean manifold with Cartesian
coordinates, the direction vectors obey ∂keˆ j = 0 and ej = eˆ j , and so in three-
dimensional Cartesian coordinates, expression (1.157) is given by
∇ × V (x ) = eˆ k × eˆ j ∂kV j (x ) = g ki (eˆ i × eˆ j )∂kV j = εijℓ eˆ ℓ g ki ∂kV j (x )

= εijℓ eˆ ℓ ∂ iV j (x )
(1.158)
⎛ ∂Az ∂Ay ⎞ ⎛ ∂Ax ∂Az ⎞ ⎛ ∂Ay ∂Ax ⎞
= eˆ x⎜ − ⎟ + eˆ y⎜ − ˆ z⎜
⎟ + e − ⎟.
⎝ ∂y ∂z ⎠ ⎝ ∂z ∂x ⎠ ⎝ ∂x ∂y ⎠
Exercise 1.16. Use the spherical coordinate direction vectors (1.43) to show that the
three-dimensional curl (1.157) in spherical coordinates (1.42) is given by
1 ⎛∂ ∂V θ ⎞
∇ × V (x ) = eˆ r ⎜ (sin θ V ϕ ) − ⎟
r sin θ ⎝ ∂θ ∂ϕ ⎠
⎛ 1 ∂V r 1 ∂ ⎞
+ eˆ θ⎜ − (rV ϕ )⎟ (1.159)
⎝ r sin θ ∂ϕ r ∂r ⎠
⎛
1 ∂ ∂V r ⎞
+ eˆ ϕ ⎜ (rV θ ) − ⎟.
r ⎝ ∂r ∂θ ⎠
Exercise 1.17. Use the cylindrical coordinate direction vectors (1.143) to show that
the three-dimensional curl (1.157) in the cylindrical coordinates of (1.142) is given by
⎛ 1 ∂V z ∂V ϕ ⎞ ⎛ ∂V r ∂V z ⎞
∇ × V (x ) = eˆ ρ⎜ − ⎟ + eˆ ϕ⎜ − ⎟
⎝ ρ ∂ϕ ∂z ⎠ ⎝ ∂ϕ ∂ρ ⎠
(1.160)
⎛ ∂V ϕ 1 ∂V ρ 1 ⎞
+ eˆ z⎜ − + V ϕ⎟ .
⎝ ∂ρ ρ ∂ϕ ρ ⎠
The curl of the gradient of a scalar function f is given by

∇ × ∇f = e k∂k × e j ∂jf = (e k × ∂ke j )∂ jf + εkji e i ∂ k∂ jf . (1.161)
1-40
In the case of Cartesian coordinates, ∂ke j = 0. If the function f is differentiable, then

∂ j∂ kf = ∂ k∂ jf , and it follows that
∇ × ∇f = 0. (1.162)
This requirement is met for differentiable functions. However, it is worth noting for
future reference that ∇ × V = 0 does not mean that V = ∇f . Similarly, the
divergence of the curl of a vector function is given by
∇ · (∇ × V ) = e i ∂i · εjkℓ e j ∂ kV ℓ = εjkℓ(e i · ∂ie j )∂ kV ℓ + εjkℓ ∂ j∂ kV ℓ . (1.163)
Once again, for the case of Cartesian coordinates and ∂ j∂ kV ℓ = ∂ k∂ jV ℓ , it follows

that
∇ · (∇ × V ) = 0. (1.164)
The divergence and the curl are used in two theorems of great importance in field
theory. For the moment, these theorems will be presented in their three-dimensional
versions. The first is the divergence theorem, which states that the integral of
∇ · V (x ) over a volume V bounded by the closed surface S (so that S = S (V )) is
given by
∫V d3x ∇ · V (x) = ∮S(V ) dS · V (x), (1.165)
where dS stands for the oriented surface elements that cover the closed surface S.
These were discussed in (1.115). The second is Stokes’ theorem, which relates the line
integral of V (x ) around a closed path P to the curl of V (x ) evaluated over any surface
S bounded by the path,
∫S dS · ∇ × V (x) = ∮P(S ) dr · V (r(x)). (1.166)
In both cases, (1.165) and (1.166), the integral on the left is found entirely from the
vector field at the boundary, i.e. either the surface of the volume or the perimeter of
the surface.
1.15 Differential forms and the wedge product

In order to lift the limitation of Stokes’ theorem to three dimensions, it is necessary
to define the relationship between the tangent space of vectors Tp(M) and the set of
linear functions on the tangent space, denoted by T p*(M) and referred to as the dual
space. The members of T p*(M) are scalar functions f of the tangent vectors v that
map the vectors into a real number in a linear manner. If v and w are two tangent
vectors and a and b are two real numbers, linearity requires
f (av + bw) = af (v) + bf (w). (1.167)
1-41
It is important to find a set of basis functions that allows the dual space members to
be expressed as a linear combination, in much the same way that the tangent vectors
are a linear combination of the unit vectors. This is resolved by considering a
function f (x ) of the coordinates x for the point p where the tangent space is defined.
If v is an infinitesimal tangent vector in the tangent space at some point p with the
coordinates x , then its length (1.64) is infinitesimal. As a result, the function
f (x + v) will differ from f (x ) by an amount df (v). A Taylor series expansion gives
df (v) = f (x + v) − f (x ) = v · ∇f (x ) = Vf (x ), (1.168)
where V is the vector written in terms of the basis vectors ∂/∂x j , V = v · ∇ . Although
it is derived for infinitesimal v, (1.168) is linear in v. Therefore, (1.168) is defined to
hold for all vectors in the tangent space at x and all functions f in the dual space. In
effect, the definition of (1.168) creates a generalized inner product between the
members of Tp(M) and the members of T p*(M). This inner product is a linear
relation, just as the scalar product between two position vectors is a linear relation.
In some presentations of differential geometry, (1.168) is written
(df , v) = df [v ] = v · ∇f (x ), (1.169)
which defines an inner product between the differential df in the dual space and the
vector v in the tangent space. Viewing it in this manner shows that the differential of
a linear function has a projection onto the vector v through definition (1.169).
Picking the function f in (1.168) to be the jth coordinate xj of point p gives
dx j [v ] = v · ∇x j = v k ∂k(x j ) = v k δ j k = v j . (1.170)
Expressing v in components and using the linearity (1.167), result (1.170) gives
dx j [v ] = dx j [v k ek] = v k dx j [ek] = v j , (1.171)
which can only hold if
dx j [ek] = δ j k . (1.172)
Combining (1.172) and (1.170) with the definition of the differential of a function
gives
df [v ] = ∂jf (x ) dx j [v ] = v j (x ) ∂jf (x ) = v(x ) · ∇f (x ), (1.173)
which reproduces (1.168).

An arbitrary member of the dual space can therefore be written
ω = ωj dx j , (1.174)
and is referred to as a one-form. Like the direction vectors ej in the case of the
tangent space, the dual space is spanned by the basis one-forms dx j as long as dx j
are viewed as functions of vectors defined by (1.172). Since the one-form compo-
nents ωj transform covariantly, the one-form ω is coordinate independent. Its action
on a tangent vector v is also coordinate independent,
1-42
ω[v ] = ωj v k dx j [ek] = ωj v k δ j k = ωj v j . (1.175)
Result (1.175) is manifestly linear, satisfying (1.167). The two components appearing
in the sum are implicit functions of the position vector r for the point under
consideration.
The differential of a function dφ(x ) gives a one-form with the covariant
components ωj = ∂jφ. This follows from elementary calculus,
∂φ(x )
dφ ( x ) = dx j = ∂ j φ ( x ) dx j . (1.176)
∂x j
As a result, a scalar function φ(x ) of the coordinates is referred to as a zero-form and
the one-form ω(x ) = dφ(x ) is referred to as an exact one-form. A one-form is
identical to the line element from basic calculus, since it takes the form Ω · dx, where
Ω is a vector function given by Ω = ωj (x )e j expressed in terms of the covariant
components of the vector. The action of the differential d on a zero-form will be
generalized to define the exterior derivative, which yields an (n + 1)-form of a given
n-form.
The one-form created in (1.176) is a combination of the contravariant compo-
nents of dx and the covariant components of ∇φ. Using the transformation
properties of the two types of vectors shows that the one-form dφ transforms
according to
∂φ(y ) i ∂x j ∂y i ∂φ(x ) ∂φ(x ) ∂φ(x )
dφ ( y ) = i
d y = i k
dx k = δ j k dx k = dx j
∂y ∂y ∂x ∂xj ∂xj ∂xj (1.177)
= dφ(x ).
This shows that the one-form is a scalar quantity, since it is unchanged or invariant
under coordinate transformations regardless of the number of dimensions. It is
therefore a coordinate-free quantity.
Just as the one-form is an oriented line element, forms can now be extended to
involve infinitesimal oriented area and volume elements in arbitrary dimensions.
This was motivated earlier in (1.115) by constructing oriented area elements in three
dimensions using the vector product. In differential geometry, infinitesimal oriented
area elements are defined by the wedge product, sometimes known as the exterior
product, denoted by dx i ∧ dx j . This product involves the basis elements of the dual
space at each point. The wedge product is defined to be antisymmetric, so that, like
the Levi-Civita symbol in the vector product,
dx i ∧ dx j = −dx j ∧ dx i ⟹ dx j ∧ dx j = 0. (1.178)
In three dimensions the wedge product is defined to be identical to the vector
product result (1.115),
dx i ∧ dx j = ε ijk dSk , (1.179)
1-43
where dSk is the kth component of the oriented area element perpendicular to both
directions i and j. Given two one-forms in three dimensions, ω and ζ , a two-form ξ
can be obtained from them using the wedge product, as follows:
ξ = ω ∧ ζ = ωi ζj dx i ∧ dx j = ε ijkωi ζj dSk = (ω × ζ ) · dS . (1.180)
The vector product of two three-dimensional vectors has appeared, along with the
infinitesimal oriented area element dS . This is the two-dimensional extension of the
one-form, so it is called a two-form. This can be used with the previous results of
(1.19) to find the surface element in polar coordinates,
dx ∧ dy = (cos θ dr − r sin θ dθ ) ∧ (sin θ dr + r cos θ dθ )
= r cos2 θ dr ∧ dθ − r sin2 θ dθ ∧ dr = r(cos2 θ + sin2 θ )dr (1.181)
∧ dθ = r dr dθ n ,
where n is a unit vector perpendicular to the plane of the surface element in the right-
handed sense. This has correctly reproduced the Jacobian (1.13) for the coordinate
transformation from Cartesian coordinates to polar coordinates.
While the wedge product of two one-forms gives a two-form, a general two-form
η in three dimensions is written as
η = ηij (x ) dx i ∧ dx j = ηij ε ijk dSk . (1.182)
The components ηij can be viewed as the covariant components of a second-rank

tensor. It is useful to note that only the antisymmetric part of ηij contributes to
(1.182). To see this, note that ηij can be written as the sum of an antisymmetric part
and a symmetric part,
1 1
ηij = (ηij − ηji ) + (ηij + ηji ) ≡ ηijA + ηijS , (1.183)
2 2
where ηijA = −η jiA and ηijS = η jiS . The symmetric part of ηij drops out of the two-form,
since the dummy indices can be relabeled, which gives
ηijS dx i ∧ dx j = η jiS dx j ∧ dx i = −ηijS dx i ∧ dx j ⟹ ηijS dx i ∧ dx j = 0. (1.184)
As a result, for a two-form in three dimensions, ηij have only three independent
components, since the symmetric part of ηij has six independent components. As
long as the ηij transform like the covariant components of a second-rank tensor, the
general two-form of (1.182) is invariant under coordinate transformations, regard-
less of the number of dimensions. This is demonstrated identically to the one-form of
(1.177).
So far, the Levi-Civita symbol has been employed in the definition of the vector
product and the Jacobian, and it will now play a central role in forms. It was shown
in (1.101) that the covariant components of the Levi-Civita symbol correspond to a
tensor density of weight +1. Similarly, it is possible to show that the contravariant
1-44
components correspond to a tensor density of weight −1, since they obey the identity
given by
1 ∂y a ∂y b ∂y c
ε abc… = ⋯ε a ′ b ′ c ′…. (1.185)
J (y ) ∂x a ′ ∂x b ′ ∂x c ′
Using (1.101) and (1.185) shows that both the transformed and original versions of
the Levi-Civita symbol obey the earlier result (1.11),
ε abc…εabc… = ε a ′ b ′ c ′…εa ′ b ′ c ′… = n! . (1.186)
The Jacobian factors cancel and relation (1.186) is invariant under coordinate
changes, as it must be.
Exercise 1.18. Verify result (1.186).
The identity (1.185) determines the transformation properties of the curl of a

vector A , written in three dimensions as B = ∇ × A with components given by
B k = εijk ∂iAj . Employing the identity (1.185) and the transformation properties of ∇
and A shows that B′ has the components
1 ∂y i ∂y j ∂y k npr ∂x ℓ ∂x m ∂Am
B′k (y ) = ε
J ∂x n ∂x p ∂x r ∂y i ∂y j ∂x ℓ
(1.187)
1 ∂y k ℓmr ∂Am 1 ∂y k r
= ε = B (x ),
J ∂x r ∂x ℓ J ∂x r
so that the curl transforms as a contravariant pseudovector of weight −1. In deriving
(1.187), A was treated as a true vector, i.e. one that does not pick up a factor of the
Jacobian under transformation. The presence of the Jacobian has a profound effect
on a pseudovector when the change of coordinates causes a reflection or change of
handedness, which are coordinate changes of the form yi = −x i . A true vector
changes sign under a reflection or change of handedness, since such a transformation
gives ∂yi /∂x j = −δ i j , yielding A′(y ) = −A(x ). This means that a true vector
maintains its orientation when the coordinate system changes handedness. Result
(1.187) shows that the curl of a true vector does not change sign under a reflection,
since the Jacobian in three dimensions gives J = −1, and its sign compensates for the
sign change created by ∂yi /∂x j = −δ i j . This means that a pseudovector, the curl of a
true vector, flips its direction under reflection, since the signs of its components in the
new coordinate system, which have changed handedness, are the same as in the old
coordinate system.
The development of forms continues by defining a three-form in three dimensions
as
1-45
V = Vijk dx i ∧ dx j ∧ dx k = Vijk ε ijk d3x . (1.188)
The components Vijk are antisymmetric in a manner identical to the Levi-Civita

symbol εijk , changing sign under an odd permutation of the indices. In three
dimensions a general three-form becomes V = εijk Vijk d3x . In three dimensions it is
not possible to define a four-form or higher, since such forms will have at least one
pair of identical indices and antisymmetry in the indices requires that such a pair
should vanish. Using the same two-form arguments, it follows that Vijk transforms
as a third-rank covariant tensor, while dx i ∧ dx j ∧ dx k transforms as a third-rank
contravariant tensor. It is now possible to show that these transformation
properties are consistent with the volume element d3x that picks up a factor of
the Jacobian when the coordinate transformation occurs. The first step is to use
(1.186) to write
1
d3x = εijk dx i ∧ dx j ∧ dx k . (1.189)
3!
Using the contravariant transformation property of the differential and
dy ℓ ∧ dy m ∧ dy n = ε ℓmn d3y gives
1 ∂x i ∂x j ∂x k 1 ∂x i ∂x j ∂x k 3
d3x = εijk ℓ m n
dy ℓ ∧ dy m ∧ dy n = εijkε ℓmn ℓ d y = J d3y, (1.190)
3! ∂y ∂y ∂y 3! ∂y ∂y m ∂y n
where the Jacobian (1.12) for a three-dimensional volume transformation has

appeared, justifying its definition using the exterior algebra of exterior forms.
Since the volume element d3x has no direction associated with it and is always
positive regardless of the coordinate system, it follows that
1 ∂x i ∂x j ∂x k
∣J ∣ = εijkε ℓmn ℓ . (1.191)
3! ∂y ∂y m ∂y n
The volume element is not a true scalar, instead transforming as a pseudoscalar of

weight +1.
The exterior derivative mentioned earlier in this section can now be defined. The
exterior derivative of a zero-form φ was defined to be the exact differential or one-
form dφ = ∂iφ dx i . Given the one-form ω = ωj dx j , the action of the exterior
derivative d on ω is defined as
dω = ∂jωk dx j ∧ dx k , (1.192)
so that a two-form is created from a one-form. Using dx j ∧ dx k for three dimensions
gives
dω = ε jkℓ ∂jωk dSℓ . (1.193)
The quantity B k ≡ ε jℓk ∂jAℓ was defined earlier as the kth component of the curl of
the vector field A, so that
1-46
dA = B k dSk = (∂ × A) · dS . (1.194)
The action of the exterior derivative has changed the one-form, which is a line
element, into a two-form, which is a surface or flux element of the curl of the vector
field appearing in the one-form. Due to the antisymmetry of the wedge product, it
follows that d2A = 0. This occurs because ∂i∂jAk dx i ∧ dx j ∧ dx k vanishes if
∂i∂jAk = ∂j∂iAk . Similarly, it follows that d2φ = 0. Since the action of d2 is zero, d
is referred to as a nilpotent operator. The exterior derivative can be applied to the
wedge product of forms as well, but obeys a generalized Leibniz property. This
follows by examining the exterior derivative of the wedge product of two one-forms,
d(ω ∧ χ ) = d(ωj χk dx j ∧ dx k ) = (∂iωj χk + ωj ∂iχk ) dx i ∧ dx j ∧ dx k

(1.195)
= dω ∧ χ − ω ∧ dχ ,
where dx j ∧ dx i = −dx i ∧ dx j creates the minus sign. The wedge product of two
arbitrary forms ω and χ has an exterior derivative given by
d(ω ∧ χ ) = dω ∧ χ + ( −1)nω ∧ dχ , (1.196)
where n corresponds to the degree of the form ω, which is n = 0 for a zero-form,
n = 1 for a one-form, and so on.
It is straightforward to generalize these results to higher dimensions. In four
dimensions a one-form is written
A = Aμ dx μ, (1.197)
where μ runs from 0 to 3. Continuing in four dimensions the two-form surface
element is given by
dx μ ∧ dx ν = ε μνρσ dSρσ , (1.198)
where dSρσ = −dSσρ corresponds to all the possible two-dimensional surface
elements. For example, choosing μ = 0 and ν = 1 gives the oriented surface elements
in the y–z plane. Because dSσρ is antisymmetric, it has (4 × 4 − 4)/2 = 6 distinct
components. These represent the six distinct two-dimensional surfaces possible in a
four-dimensional space. This is the four-dimensional counterpart of the three two-
dimensional oriented surface elements in three dimensions. Remaining in four
dimensions, the three-forms are written
G = Gμνρ dx μ ∧ dx ν ∧ dx ρ = Gμνρ ε μνρσ d3Σσ . (1.199)
Here, d3Σσ are the four oriented three-volume elements, where d3Σ 0 = d3x is the
spatial volume element and d3Σ1 = dx 0 dx 2 dx 3. While there is no four-form in three
dimensions, the four-form in four dimensions is given by
dx μ ∧ dx ν ∧ dx ρ ∧ dx σ = ε μνρσ d4x . (1.200)
The behavior of d4x under a change of coordinates to y μ(x ) is derived identically to
(1.190), so that
1-47
d4x = J d4y , (1.201)

where J is the four-dimensional version of expression (1.191). The exterior derivative
can be applied to an r-form, ω, to obtain an (r + 1)-form, dω, given by
dω = ∂μων⋯ρ dx μ ∧ dx ν ∧ ⋯ ∧ dx ρ. (1.202)
Higher dimensions follow the same pattern. Clearly, if r matches the dimension n of
the space, then dω = 0. Regardless of the dimension of the space, d2 = 0 because of
the antisymmetry of the wedge product and the usual assumption
∂i∂lωj⋯k = ∂l∂iωj⋯k , so that the exterior derivative remains nilpotent.
1.16 Differential forms and Stokes’ theorem

The exterior derivative allows Stokes’ theorem to be restated. In three dimensions
the standard version of Stokes’ theorem states that the vector field A satisfies
∫S (∇ × A) · dS = ∮P(S ) A · dx, (1.203)
where S is a two-dimensional surface and P (S ) is the closed path that is the

boundary of that surface. If S is a closed surface then it has no boundary, and in that
case, the right-hand side of (1.203) vanishes. Examining (1.203) shows that the one-
form A is the argument of the right-hand side, while dA is the argument of the left-
hand side. The generalization of Stokes’ theorem for an r-form ω is then given by
∫M dω = ∫∂M ω, (1.204)
where M is the (r + 1)-dimensional space over which the (r + 1)-form dω is being

integrated, while ∂M is the r-dimensional boundary of M over which the r-form ω is
being integrated. The proof of (1.204) in arbitrary dimensions for arbitrary surfaces
is beyond the scope of this book, but it is done in a manner similar to the proof of
Stokes’ theorem in three dimensions. The space M is covered with a generalized set
of triangles, referred to as a simplicial complex, that is appropriate to the dimension
of M. It is then shown that only the boundary of that covering contributes to the left
side, reducing it to the right-hand side. For a space with an unusual topology, such
as holes and handles and boundaries, (1.204) leads to a way of characterizing the
topology in terms of the classification of forms defined on the space. The interested
reader should study homology and cohomology theory.
However, in order to clarify the new version of Stokes’ theorem, it will be
demonstrated that the usual three-dimensional version of the divergence theorem is
also contained in (1.204) using the two-form F = Fij dx i ∧ dx j in the right-hand side
of (1.204). Previous results showed that the antisymmetry of Fij allows the two-form
to be written in terms of the vector V ,
1
Fij = εijℓV ℓ . (1.205)
2
1-48
This gives
1
F = Fij dx i ∧ dx j = εijℓε ijk V ℓ dSk = δ k ℓV ℓ dSk = V k dSk = V · dS . (1.206)
2
Using the same form for Fij gives
dF = ∂kFij dx k ∧ dx i ∧ dx j = ∂kFij ε kij d3x
1 (1.207)
= ε kij εijℓ ∂kV ℓ d3x = δ k ℓ ∂kV ℓ d3x = ∂kV k d3x = ∇ · V d3x .
2
Inserting (1.206) and (1.207) into (1.204) yields the divergence theorem in three
dimensions (1.165),
∫M d3x ∇ · V = ∫∂M dS · V, (1.208)
where M is a three-dimensional volume and ∂M is the two-dimensional surface

forming the boundary of the three-dimensional volume.
Exercise 1.19. In a two-dimensional Cartesian plane the exterior derivative of a one-

form ω = ωa dx a is given by dω = ∂aωb(x, y )ε abd2S , where d2S = dx dy . For the case in
which dω is integrated over a rectangular patch V of width 2w and height 2h centered
on the origin, show that the integral becomes
w h −w
∫V dω = ∫−w dx ωx(x, −h) + ∫−h dy ωy(w, y ) + ∫w dx ωx (x , h )
−h
+ ∫h dy ω y ( − w , y ) = ∫∂V ω.
The second application of Stokes’ theorem in two dimensions is trivial, consisting

of the line integral of the exterior derivative of a zero-form dφ = ∂aφ dx a . Since dφ is
an exact differential the result is
∫P dφ = φ(xf , yf ) − φ(xo, yo).

This is the sum of the zero-form at the oriented endpoints of path P.
Because d is nilpotent, ω can be replaced in (1.204) with ω + dΛ, where Λ is an
(r − 1)-form, and the left-hand side of (1.204) is unchanged, since
dω → dω + d2Λ = dω . (1.209)
This is consistent with the right-hand side being unchanged, since applying Stokes’
theorem to the right-hand side shows that its change is given by
∫∂M dΛ = ∫∂ M Λ = 0,
2
(1.210)
1-49
which vanishes, because the boundary of a boundary does not exist.

Finally, Stokes’ theorem (1.204) remains true regardless of the dimension. For an
n-dimensional space there are n − 1 nontrivial applications of the theorem, just as
there is one nontrivial application in two dimensions, given by the previous exercise,
and two nontrivial applications in three dimensions, the divergence theorem (1.165)
and Stokes’ theorem (1.166).
1.17 The Lie derivative

It is often important to find the change in a tensor function between two points
separated by an infinitesimal position-dependent vector displacement ϵX (x ), where
the limit ϵ → 0 is understood. The starting point has the coordinates x μ and the goal
is to find the change in the tensor function at the point y μ = x μ + ϵX μ(x ). Since ϵ is
infinitesimal and the tensor function is assumed to be nonsingular, the change will be
proportional to ϵ. It is straightforward to find the change in a scalar function φ(x ).
Since the scalar field does not change form at the new point, a Taylor series
expansion gives
φ(y ) = φ(x + ϵX ) = φ(x ) + ϵX ν∂νφ(x ), (1.211)
which holds to O(ϵ ). The Lie derivative LX of a scalar function is therefore defined as
φ( y ) − φ( x )
LX φ(x ) = lim = X ν(x )∂νφ(x ) = X (x )φ(x ), (1.212)
ϵ→0 ϵ
where the Cartan notation (1.45) for the vector X is being used. This is simply the
directional derivative of φ(x ), and gives the rate of change in φ(x ) along the
direction given by X .
In the case of a vector function V (x ), it may also undergo a change in its
components at the new point, and the components will depend on the coordinate
system to which they are referred. Finding the actual change of a vector’s
components between the two points requires finding its components using the
same coordinate system. This is referred to as pulling the components of the vector
V μ(y ) at y back to the original coordinate system x to find the value V ′ μ(x ) at x , and
is called the pullback of vector V . Once this is done, V ′ μ(x ) − V μ(x ) gives the true
change of the vector function between the two points. The pullback is found by
combining the contravariant vector transformation property (1.21) with
x μ = y μ − ϵX μ, which gives
∂x μ ν
V ′μ (x ) = V (y ) = (δνμ − ϵ ∂νX μ(x ) )V ν(x + ϵX ). (1.213)
∂y ν
A term of O(ϵ 2 ) has been suppressed in (1.213) to write
∂X μ(x ) ∂X μ(x )
ϵ ≈ ϵ . (1.214)
∂y ν ∂x ν
1-50
Using a Taylor series expansion in (1.213) gives the O(ϵ ) change in the vector
V ′μ (x ) − V μ(x ) = ϵX ν(x )∂νV μ(x ) − ϵV ν(x )∂νX μ(x ). (1.215)
From (1.215) the Lie derivative of a vector V along X is given by
V ′μ (x ) − V μ(x )
(LX V ) μ = lim = X ν(x )∂νV μ(x ) − V ν(x )∂νX μ(x ). (1.216)
ϵ→0 ϵ
This gives the rate of change of vector V at point x along an infinitesimal
displacement in the direction of vector X (x ). Using the Cartan notation (1.45) for
vectors, the Lie derivative of a vector can be written as a commutator,
LX V = XV − VX ≡ [X , V ]. (1.217)
This commutator is often referred to as a Lie bracket. This follows, since the
Cartan notation uses
X = X μ∂μ, Y = Y μ∂μ, (1.218)
so that the action of the Lie bracket [X , Y ] on a scalar function f defines a vector Z
given by
[X , Y ]f = X μ(∂μY ν )∂νf + X μY ν∂μ∂νf − Y ν(∂νX μ)∂μf − Y νX μ∂ν∂μf
(1.219)
= (X ν∂νY μ − Y ν∂νX μ)∂μf = Zf ,
where the vector Z has the components
Z μ = (X ν∂νY μ − Y ν∂νX μ). (1.220)
The Lie bracket has the property of antisymmetry,
[X , Y ] = −[Y , X ], (1.221)
and satisfies the Jacobi identity,
[[X , Y ], Z ] + [[Z , X ], Y ] + [[Y , Z ], X ] = 0. (1.222)
The Jacobi identity follows from direct substitution.
Result (1.217) for LX V can also be derived from the action of the Lie derivative on a
scalar function φ, given by (1.212). Since Vφ = V i ∂iφ is a scalar, associativity gives
LX (Vφ) = X (Vφ) = (XV )φ . (1.223)
The Lie derivative obeys the Leibniz property,
LX (Vφ) = (LX V )φ + V (LX φ) = (LX V )φ + (VX )φ . (1.224)
Equating (1.223) and (1.224) immediately gives
(LX V )φ = ([X , V ])φ . (1.225)
The Lie derivative can be generalized to include its action on tensors and forms. For
example, if gμν are the components of the second-rank covariant tensor g , its Lie
derivative has the components
(LX g )μν (x ) = X ρ(x )∂ρgμν(x ) + gρν(x )∂μX ρ(x ) + gμρ(x )∂νX ρ(x ). (1.226)
1-51
Exercise 1.20. Derive (1.226) using the fact that gμνY μZ ν is a scalar function, so that
LX (gμνY μZ ν ) = X ρ∂ρ(gμνY μZ ν ), (1.227)
then applying the Leibniz property of the Lie derivative LX to gμνY μZ ν , and finally
equating the two forms using result (1.225).
References and recommended further reading

Mathematical monographs that develop the topics of this chapter include
• Y Matsushima 1972 Differentiable Manifolds (New York: Dekker)
• P Bamberg, S Sternberg 1988 A Course in Mathematics for Students of
Physics Vol I (Cambridge: Cambridge University Press)
• Darling R 1994 Differential Forms and Connections (Cambridge: Cambridge
University Press)
• P Dennery, A Krzywicki 1997 Mathematics For Physicists (New York:
Dover)
• T Frankel 1997 The Geometry of Physics: An Introduction (Cambridge:
Cambridge University Press)
• B Kusse, E Westwig 1998 Mathematical Physics (New York: Wiley)
• B O’Neill 2006 Elementary Differential Geometry 2nd edn (Amsterdam:
Elsevier)
• Arfken G, Weber H, Harris F 2013 Mathematical Methods for Physicists 7th
edn (Amsterdam: Elsevier)
• Hassani S 2013 Mathematical Physics 2nd edn (Berlin: Springer–Verlag)
• Hubbard J, Hubbard B 2015 Vector Calculus, Linear Algebra, and
Differential Forms 5th edn (Ithaca: Matrix Editions)
1-52

bk978 0 7503 3455 6ch1

Uploaded by

Copyright:

Available Formats

bk978 0 7503 3455 6ch1

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

bk978 0 7503 3455 6ch1

Uploaded by

Copyright:

Available Formats

This content has been downloaded from IOPscience. Please scroll down to see the full text.

Please note that terms and conditions apply.

You may also like:

Progressive evaluation in spectroscopic sensors for non-invasive blood haemoglobin analysis—a

3rd International Conference on Turbulent Mixing and Beyond

International Conference on Innovative Research - ICIR Euroinvent 2017

Regional disparity in continuously measured time-domain cerebrovascular reactivity indices: a

Classical Field Theory and the Stress–Energy Tensor

doi:10.1088/978-0-7503-3455-6ch1 1-1 ª IOP Publishing Ltd 2022

generically designated M, and so such a smooth collection of points is known as a

1.2 Coordinate systems

x = r cos θ , y = r sin θ ⟺ r = x2 + y2 , θ = tan−1(y / x ). (1.1)

Figure 1.1. Cartesian and polar coordinate grids.

The upshot of choosing a coordinate system in an n-dimensional manifold is to

The proof of (1.4) follows from the deﬁnition of differentiable mappings in a

1.3 The Jacobian

If the Levi-Civita symbol is deﬁned in a Euclidean manifold using Cartesian

It is easy to see that

while (1.8) gives

Using induction shows that

which is nonzero everywhere except at the origin, r = 0. As mentioned earlier, polar

which vanishes if c = d and is given by εcd J (y ) for c ≠ d , inductively corroborating (1.14).

dnx = J (y ) dny , (1.17)

dSc = d2x = dx dy = J (r , θ ) dr dθ = r dr dθ = d2r = dSp. (1.18)

The Leibniz property of the differential and (1.1) combine to give

1.4 Contravariant and covariant quantities

The mapping (1.21) between dy j and the original dx j is referred to as a contravariant

Applying (1.23) to y ℓ (x ) correctly reproduces the contravariant transformation law

1.5 The summation convention

1.6 Vectors and direction vectors

Figure 1.2. Euclidean direction vectors in Cartesian and polar coordinates.

and so it will be written ej (x ) when it is important to remind the reader of this

∂r(y ) ∂x k(y ) ∂r(y(x )) ∂x k(y )

which will be useful in understanding the behavior of the vector components.

x = r sin θ cos ϕ , y = r sin θ sin ϕ , z = r cos θ ⟺

Figure 1.3. A coordinate transformation involving a translation of the origin.

⎛ −(x1 − a1)(x 2 − a 2 ) + (x 2 − a 2 )(x1 − a1) ⎞

1.7 Vector addition and the scalar product

Result (1.59) is simply the Pythagorean theorem extended to n dimensions.

Form (1.60) is equivalent to (1.53), which is seen by switching to polar coordinates

1.8 The metric tensor and distance in manifolds

ds 2 ≡ dr · dr = (dx j ej (x )) · ( dx k ek(x )) = ej (x ) · ek(x ) dx j dx k

Because of their association with length, the quantities given by

ds = gjk dx j dx k = (dx )2 + (dy )2 . (1.68)

1.9 The metric tensor and raising and lowering indices

ej (x ) · dr = ej (x ) · dx k ek(x ) = (ej (x ) · ek(x ))dx k = gjk(x ) dx k ≡ dxj . (1.71)

which is the covariant transformation (1.22). It is then possible to write the

which is manifestly invariant under a coordinate transformation, since it is a

and this gives the length of the ﬁnite position vector as

x1 = y1 + cos ϕ y 2 , x 2 = sin ϕ y 2 ⟺ y1 = x1 − cot ϕ x 2 ,

g11 = e1 · e1 = 1, g12 = g21 = e1 · e 2 = cos ϕ ,

r 2 = rj r j = r1r1 + r2r 2 = (r1 + cos ϕ r 2 )r1 + (cos ϕ r1 + r 2 )r 2

Because δ k j is invariant under a coordinate transformation, it follows that relation

In order for (1.84) to hold, it must be the case that

showing that g jk is given by the reciprocal basis scalar product.

e1 = g11e1 + g12 e 2 = csc 2ϕ e1 − csc 2ϕ cos ϕ e 2 ,

e1 · e 2 = csc 2ϕ e1 · e 2 − csc 2ϕ cos ϕ e 2 · e 2 = csc 2ϕ (cos ϕ − cos ϕ) = 0. (1.91)

e1 · e1 = (e1)2 = csc 4ϕ (e1 · e1 + cos2 ϕ e 2 · e 2 − 2 cos ϕ e1 · e 2) = csc 4ϕ

r 2 = r · r = (ej x j ) · (ek x k ) = (ej · ek)x j x k = gjkx j x k ,