Applied Mathematical Sciences

Volume 141

Editors
S.S Antman
Department of Mathematics
and
Institute for Physical Science and Technology
University of Maryland
College Park, MD 20742-4015
USA
[email protected]

P. Holmes
Department of Mechanical and Aerospace Engineering
Princeton University
215 Fine Hall
Princeton, NJ 08544
[email protected]

L. Sirovich
Laboratory of Applied Mathematics
Department of Biomathematical Sciences
Mount Sinai School of Medicine
New York, NY 10029-6574
[email protected]

K. Sreenivasan
Department of Physics
New York University
70 Washington Square South
New York City, NY 10012
[email protected]

Advisors
L. Greengard J. Keener
J. Keller R. Laubenbacher B.J. Matkowsky
A. Mielke C.S. Peskin A. Stevens A. Stuart

For other titles published in this series, go to

http://www.springer.com/series/34
Gregory L. Naber

Topology, Geometry
and Gauge fields
Interactions

Second Edition
Gregory L. Naber
Drexel University
Department of Mathematics
Korman Center
3141 Chestnut Street
Philadelphia, Pennsylvania 19104-2875
USA
[email protected]

ISSN 0066-5452
ISBN 978-1-4419-7894-3 e-ISBN 978-1-4419-7895-0
DOI 10.1007/978-1-4419-7895-0
Springer New York Dordrecht Heidelberg London
Library of Congress Control Number: 2011923654

Mathematics Subject Classification (2010): 53-01, 53B21, 55-01, 70S15

© Springer Science+Business Media, LLC 2011

All rights reserved. This work may not be translated or copied in whole or in part without the written
permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York,
NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in
connection with any form of information storage and retrieval, electronic adaptation, computer software,
or by similar or dissimilar methodology now known or hereafter developed is forbidden.
The use in this publication of trade names, trademarks, service marks, and similar terms, even if they
are not identified as such, is not to be taken as an expression of opinion as to whether or not they are
subject to proprietary rights.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

This book is for my son, Aaron, and my stepson, Mike Scarborough.
Preface

This volume is intended to carry on the program initiated in Topology,

Geometry and Gauge Fields: Foundations (henceforth, [N4]). It is writ-
ten in much the same spirit and with precisely the same philosophical mo-
tivation: Mathematics and physics have gone their separate ways for nearly
a century now and it is time for this to end. Neither can any longer afford
to ignore the problems and insights of the other. Why are Dirac magnetic
monopoles in one-to-one correspondence with the principal U(1)-bundles over
S 2 ? Why do Higgs fields fall into topological types? What led Donaldson, in
1980, to seek in the Yang-Mills equations of physics for the key that unlocks
the mysteries of smooth 4-manifolds and what physical insights into quantum
field theory led Witten, fourteen years later, to propose the vastly simpler, but
equivalent Seiberg-Witten equations as an alternative? We do not presume to
answer these questions here, but only to promote an atmosphere in which
both mathematicians and physicists recognize the need for answers. More
succinctly, we shall endeavor to provide an exposition of elementary topology
and geometry that keeps one eye on the physics in which our concepts either
arose independently or have been found to lead to a deeper understanding of
the phenomena.
Chapter 1 provides a synopsis of the geometrical background we assume of
our readers (manifolds, Lie groups, bundles, connections, etc.). While all of
this material is discussed in detail in [N4], most of it is standard and it will
not matter where the background has been acquired. There follows a rather
long chapter on physics. The discussion here is informal and heuristic and
often anticipates topological and geometrical concepts that will be introduced
precisely only much later. We begin by describing a general mathematical
framework for the classical gauge theories of physics and then discuss in some
detail a number of specific examples. These include classical electromagnetic
theory and Dirac monopoles, the Klein-Gordon and Dirac equations and SU(2)
Yang-Mills-Higgs theory. The real purpose here is to witness such concepts as
de Rham cohomology, Chern classes and spinor structures arise of their own
accord in meaningful physics.
The mathematical development resumes in Chapter 3 where we collect some
basic technical tools and then study various types of frame bundles. Minkowski
spacetime, and then more general spacetime manifolds are introduced and
some concrete examples are discussed (the Einstein-de Sitter spacetime, de
Sitter spacetime and the Einstein cylinder). With this machinery we can define
precisely the notion of a spinor structure, seen in Chapter 2 to be the device
required to model spin 12 particles in the presence of gravity. Finding the
topological obstruction to the existence of such a structure (the 2nd Stiefel-
Whitney class) will have to wait until Chapter 6.
Chapter 4 contains a more or less standard exposition of multilinear algebra,
differential forms, integration and Stokes’ Theorem, but with rather more

vii
viii Preface

attention paid to vector-valued forms than is customary. In particular, there is

a detailed discussion of tensorial forms on principal bundles and their covariant
exterior derivatives in the presence of a connection. It is these derivatives that
appear in the field equations of physics.
de Rham cohomology is the subject of Chapter 5. Explicit calculations are
based on the Mayer-Vietoris sequence and we introduce the Brouwer degree
of a map between two compact, connected, orientable n-manifolds by showing
that such a manifold has 1-dimensional nth cohomology. We prove that the
degree is an integer by showing how to calculate it from a critical value of the
map. It is such a degree that gives rise to the topological quantum number of
the Higgs field in SU (2) Yang-Mills-Higgs theory. The chapter concludes with
a discussion of the Hopf invariant and its explicit calculation for the complex
Hopf map.
The notion of a characteristic class arises on several occasions in the phys-
ical considerations of Chapter 2 (the magnetic charge of a Dirac monopole,
topological charge of an instanton, and the obstruction to the existence of a
spinor structure on a spacetime manifold) and Chapter 6 takes up the sub-
ject in earnest. We construct the Chern-Weil homomorphism and from it the
Chern classes of a principal bundle. We prove that U (1)-bundles over S 2 are
characterized (up to equivalence) by their 1st Chern class and that SU (2)-
bundles over S 4 are similarly characterized by the 2nd Chern class. These
results complete the identification of magnetic charge and instanton number
with purely topological objects. The chapter concludes with the construction
of the Z2 -Čech cohomology of a smooth manifold from a simple cover. We
build the 1st and 2nd Stiefel-Whitney classes for a semi-Riemannian manifold
and prove that the former is the obstruction to orientability, while the latter,
for a spacetime, is the obstruction to the existence of a spinor structure.
Seiberg-Witten gauge theory is in some sense analogous to the spin
1
2 -electrodynamics discussed in Chapter 2, but sprang from quite differ-
ent soil and its significance is of a very different sort. Although much of
this story lies in greater depths than we have reached in the main body
of the text, the profound significance of the subject for both mathemat-
ics and physics would seem to make some attempt to tell at least part of
it a moral imperative. The Appendix, which is a much expanded version
of the Appendix that appeared in the first edition of this work, is a mod-
est attempt to relate as much of the story as we can with the machinery
we have available and should be considered a continuation of Appendix B
in [N4].
There are 228 Exercises in the book. Each is an integral part of the devel-
opment and has been included (rather than the equivalent term “clearly”) to
encourage active participation on the part of the reader.
Gregory L. Naber
2010
Acknowledgments

The California State University at Chico and, in particular, its College of

Natural Sciences and Department of Mathematics have provided much sup-
port over the years without which this and many similar projects would not
have been possible. For this I wish to express my sincere gratitude. Special
thanks are also due to my friends at San Bernardino State University, both
mathematicians and physicists, who have listened patiently and contributed
much to lectures in which I attempted to sort out the final form of the pre-
sentation. Finally, and as always, my greatest debt is to my wife, Debora. She
has devoted much of her life to typing these manuscripts of mine and the rest
to seeing to it that I have the sort of life in which it is possible to write them.
I can’t imagine doing this job without her.

ix
Contents

Chapter 1
Geometrical Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Smooth Manifolds and Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Matrix Lie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Principal Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.4 Connections and Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.5 Associated Bundles and Matter Fields . . . . . . . . . . . . . . . . . . . . . . 38

Chapter 2
Physical Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.1 General Framework for Classical Gauge Theories . . . . . . . . . . . . 45
2.2 Electromagnetic Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.3 Spin Zero Electrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.4 Spin One-Half Electrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.5 SU (2)-Yang-Mills-Higgs Theory on Rn . . . . . . . . . . . . . . . . . . . . . 104
2.6 Epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

Chapter 3
Frame Bundles and Spacetimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
3.1 Partitions of Unity, Riemannian Metrics and Connections . . . . . 139
3.2 Continuous Versus Smooth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
3.3 Frame Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
3.4 Minkowski Spacetime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
3.5 Spacetime Manifolds and Spinor Structures . . . . . . . . . . . . . . . . . 168

Chapter 4
Differential Forms and Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
4.1 Multilinear Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
4.2 Vector-Valued Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
4.3 Differential Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
4.4 The de Rham Complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
4.5 Tensorial Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
4.6 Integration on Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
4.7 Stokes’ Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

Chapter 5
de Rham Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
5.1 The de Rham Cohomology Groups . . . . . . . . . . . . . . . . . . . . . . . . 258
5.2 Induced Homomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
5.3 Cochain Complexes and Their Cohomology . . . . . . . . . . . . . . . . . 275
5.4 The Mayer-Vietoris Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281

xi
xii Contents

5.5 The Cohomology of Compact, Orientable Manifolds . . . . . . . . . . 290

5.6 The Brouwer Degree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
5.7 The Hopf Invariant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299

Chapter 6
Characteristic Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
6.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
6.2 Algebraic Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
6.3 The Chern-Weil Homomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . 317
6.4 Chern Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
6.5 Z2 -Čech Cohomology for Smooth Manifolds . . . . . . . . . . . . . . . . . 335

Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
Seiberg-Witten Gauge Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
A.1 Donaldson Invariants and TQFT . . . . . . . . . . . . . . . . . . . . . . . . . . 351
A.2 Clifford Algebra and Spinc -Structures . . . . . . . . . . . . . . . . . . . . . . 363
A.3 Seiberg-Witten Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
A.4 The Moduli Space and Invariant . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
A.5 The Witten Conjecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401

Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
1
Geometrical Background

In this preliminary chapter we have gathered together a brief synopsis of

those items from differential geometry upon which the main development of
the text will be built. Most of the material is standard so that more detailed
expositions are readily available, although, when references seem called for,
we will tend to favor [N4].

1.1 Smooth Manifolds and Maps

A topological manifold of dimension n is a second countable, Haus-dorff
topological space X with the property that, for each p ∈ X, there exists
an open set U in X containing p and a homeomorphism φ of U onto an
open subset φ(U ) in Rn . The pair (U, φ) is called a chart at p. If (U1 , φ1 )
and (U2 , φ2 ) are two charts with U1 ∩ U2 ̸= ∅, then the overlap functions
φ1 ◦ φ−12 : φ2 (U1 ∩ U2 ) −→ φ1 (U1 ∩ U2 ) and φ2 ◦ φ−11 : φ1 (U1 ∩ U2 ) −→
φ2 (U1 ∩ U2 ) are inverse homeomorphisms. Two charts (U1 , φ1 ) and (U2 , φ2 )
are C ∞ -related if either U1 ∩ U2 = ∅, or U1 ∩ U2 ̸= ∅ and the overlap
functions are C ∞ (i.e., have continuous partial derivatives of all orders and
types). An atlas for X is a collection
∪ {(Uα , φα )}α∈A of charts, any two of
which are C ∞ -related and with α∈A Uα = X. A chart is admissible to an
atlas if it is C ∞ -related to every chart in the atlas and the atlas is said to
be maximal if it contains every chart that is admissible to it. Every atlas is
contained in a unique maximal atlas (Theorem 5.3.1, [N4]). A maximal atlas
is called a differentiable structure for X. A topological manifold together
with some differentiable structure is called a differentiable (or smooth,
or C ∞ ) manifold (examples are forthcoming). The integer n is called the
dimension of X and is denoted dim X.
Remark: Some topological manifolds admit no differentiable structures at
all, while others admit many (see Appendix B of [N4]).
Let X and Y be smooth manifolds of dimension n and m, respectively,
and let f : X −→ Y be a continuous map. Let (U, φ) be a chart on
X and (V, ψ) a chart on Y with U ∩ f −1 (V ) ̸= ∅. Then ψ ◦ f ◦ φ−1 :
φ(U ∩f −1 (V )) −→ ψ(U ∩f −1 (V )) is called the coordinate expression for f
relative to these two charts. f is said to be smooth (or C ∞ ) if its coordinate
expressions are C ∞ for all such charts in some atlases for X and Y . A smooth
bijection with a smooth inverse is a diffeomorphism and, if a diffeomorphism
of X onto Y exists, we say that X and Y are diffeomorphic. Providing the

G.L. Naber, Topology, Geometry and Gauge fields: Interactions, 1

Applied Mathematical Sciences 141, DOI 10.1007/978-1-4419-7895-0_1,
© Springer Science+Business Media, LLC 2011
2 1. Geometrical Background

real line R with its standard differentiable structure (determined by the at-
las consisting of the single chart (R, id)) we denote by C ∞ (X) the set of all
smooth, real-valued functions on X and provide it with the obvious (point-
wise) structure of a commutative algebra with identity.
A tangent vector at p ∈ X is a real-valued function v : C ∞ (X) −→ R
that is linear and satisfies the Leibnitz Product Rule v (f g) = f (p)v (g) +
v (f )g(p) for all f, g ∈ C ∞ (X). The collection of all such is denoted Tp (X),
called the tangent space to X at p and provided with the natural point-
wise structure of a real vector space. The dimension of Tp (X) as a vector
space over R is the same as the dimension of X as a manifold. Indeed, if
(U, φ) is a chart at p in X with coordinate functions xi , i = 1, . . . , n (φ(p) =
(x1 (p), . . . , xn (p))), then the linear maps ∂i |p = ∂x ∂
i |p : C
∞
(X) −→ R defined
−1
by ∂xi |p (f ) = Di (f ◦ φ )(φ(p)) are in Tp (X) and { ∂x1 |p , . . . , ∂x∂n |p } is a ba-
∂ ∂

sis for Tp (X) (Theorem 5.5.3, [N4]). Any v ∈ Tp (X) can be uniquely written
i |p (summation convention). If (a, b) is an open interval in R
∂
as v = v (xi ) ∂x
provided with its standard differentiable structure (determined by the atlas
consisting of the single chart ((a, b), ι), where ι : (a, b) ,→ R is the inclusion
map), then a smooth map α : (a, b) −→ X is a smooth curve in X. Fix
t0 ∈ (a, b) and let p = α(t0 ). The velocity vector of α at t0 is the map
α′ (t0 ) : C ∞ (X) −→ R defined by (α′ (t0 ))(f ) = D1 (f ◦ α)(t0 ) for each f in
C ∞ (X). Then α′ (t0 ) is in Tp (X) and, indeed, every element of Tp (X) is the
velocity vector of some smooth curve in X through p (Corollary 5.5.6, [N4]).
If f : X −→ Y is a smooth map and p ∈ X, then the derivative of
f at p is the linear map f∗p : Tp (X) −→ Tf (p) (Y ) defined as follows: For
each v ∈ Tp (X), f∗p (v ) : C ∞ (Y ) −→ R is given by (f∗p (v ))(g) = v (g ◦ f )
for all g ∈ C ∞ (Y ). If v = α′ (t0 ) for some smooth curve α, then f∗p (v ) =
f∗p (α′ (t0 )) = (f ◦α)′ (t0 ). f is said to be an immersion at p if f∗p is one-to-one
and an immersion if this is true at each p ∈ X. f is a submersion at p if
f∗p is onto and a submersion if this is true at each p ∈ X. An immersion
that is also a homeomorphism onto its image is an imbedding. A point q ∈ Y
is a regular value of f if, for every p ∈ f −1 (q), f is a submersion at p (this
is the case, in particular, if f −1 (q) = ∅); otherwise, q is a critical value of f .
If X ′ is an open subspace of X and {(Uα , φα )}α∈A is an atlas for X, then
{(Uα ∩ X ′ , φα |Uα ∩ X ′ ) : α ∈ A, Uα ∩ X ′ ̸= ∅} is an atlas for X ′ and, with
the differentiable structure determined by this atlas, X ′ is called an open
submanifold of X. Note that dim X ′ = dim X. More generally, if dim X =
n and 1 ≤ k ≤ n is an integer, then a topological subspace X ′ of X is
called a k -dimensional submanifold of X if, for each p ∈ X ′ , there exists
a chart (U, φ) in the differentiable structure for X such that φ(U ∩ X ′ ) =
{(x1 , . . . , xk , xk+1 , . . . , xn ) ∈ φ(U ) : xk+1 = · · · = xn = 0}. For each such
(U, φ) one obtains a chart (U ∩ X ′ , φ′ ) for X ′ , where φ′ is φ|U ∩ X ′ followed
by the projection of Rn = Rk × Rn−k onto Rk . The collection of all such (U ∩
X ′ , φ′ ) is an atlas for X ′ and so determines a differentiable structure for X ′ . A
0-dimensional submanifold of X is a discrete subspace of X. Restrictions
of C ∞ maps on X to submanifolds are C ∞ with respect to this submanifold
 1.1. Smooth Manifolds and Maps 3

differentiable structure. According to the Inverse Function Theorem, if X and

Y have the same dimension and f : X −→ Y is smooth, then f∗p : Tp (X) −→
Tf (p) (Y ) is an isomorphism if and only if f is a local diffeomorphism at p, i.e.,
if and only if there exist open neighborhoods V of p and W of f (p) such that
f |V is a diffeomorphism of V onto W (Corollary 5.5.8, [N4]).
If f : X −→ Y is an imbedding, then its image f (X) is a submanifold of Y
and, regarded as a map of X onto f (X), f is a diffeomorphism (Corollary 5.6.6,
[N4]). On the other hand, if f : X −→ Y is any smooth map and q ∈ Y is a
regular value of f (so that, in particular, dim X ≥ dim Y ), then f −1 (q) is either
empty or a submanifold of X of dimension dim X − dim Y (Corollary 5.6.7,
[N4]).
We now pause to describe a collection of examples of smooth manifolds that
will play a central role in all that follows. Several methods of constructing
examples are available. One can, of course, explicitly specify a second count-
able, Hausdorff space X and an atlas {(Uα , φα )}α∈A for it. Alternatively, one
can produce smooth manifolds as submanifolds of known examples, images of
imbeddings, or inverse images of regular values of smooth maps. One other
technique (forming products) will be discussed after we have some manifolds
available with which to form them.
1. (Standard differentiable structure on Rn ) X = Rn with the atlas
consisting of a single global chart (Rn , id). A chart (U, φ) is in the differ-
entiable structure determined by this atlas if and only if φ : U −→ φ(U ) is a
diffeomorphism.
2. (A nonstandard differentiable structure on R) X = R with the atlas
consisting of a single global chart (R, φ), where φ : R −→ R is given by
φ(x) = x3 . The resulting differentiable structure on R is not the same as the
√
standard structure because φ−1 : R −→ R is given by φ−1 (y) = 3 y and this
is not C on R (in the standard sense). These two manifolds (standard and
∞

nonstandard R) have different differentiable structures, but are, nevertheless,

diffeomorphic (Exercise, or see page 243 of [N4]).
3. (Natural differentiable structures on real vector spaces) Let X = V be
a real vector space of dimension n. Select a basis {e1 , . . . , en } for V and let
{e1 , . . . , en } be the dual basis for V ∗ . Define φ : V −→ Rn as follows: For each
v = v a ea ∈ V, φ(v) = (e1 (v), . . . , en (v)) = (v 1 , . . . , v n ). Then φ is an isomor-
phism. Provide V with the topology for which φ is a homeomorphism (this
is independent of the choice of basis). Now provide V with the differentiable
structure determined by the single global chart (V, φ) (also independent of
the choice of basis). Note that when V = Rn this reduces to the standard dif-
ferentiable structure. Also note that, for each p ∈ V, the tangent space Tp (V)
can be canonically identified with V: For each v ∈ V, define v p ∈ Tp (V) by
v p = α′ (0), where α : R −→ V is given by α(t) = p + tv. Then v −→ v p is
the canonical isomorphism of V onto Tp (V).
4 1. Geometrical Background

4. (Standard differentiable structure on the n-sphere S n ) X = S n =

{(x1 , . . . , xn , xn+1 ) ∈ Rn+1 : (x1 )2 + · · · + (xn )2 + (xn+1 )2 = 1} with its topol-
ogy as a subspace of Rn+1 . We define two stereographic projection charts
on S n . Let N = (0, . . . , 0, 1), S = (0, . . . , 0, −1), US = S n − {N } and
UN = S n − {S}. Now define φS : US −→ Rn and φN : UN −→ Rn by
φS (x1 , . . . , xn , xn+1 ) = (1 − xn+1 )−1 (x1 , . . . , xn ) and φN (x1 , . . . , xn , xn+1 ) =
(1 + xn+1 )−1 (x1 , . . . , xn ). Then φ−1S : R −→ US and φN : R −→ UN are
n −1 n
−1 −1 1 2 −1
given by φS (y) = φS (y , . . . , y ) = (1 + ∥y∥ ) (2y , . . . , 2y , ∥y∥2 − 1) and
n 1 n

φ−1 −1 1
N (y) = φN (y , . . . , y ) = (1 + ∥y∥ )
n 2 −1
(2y 1 , . . . , 2y n , −∥y∥2 + 1). Thus, on
φN (UN ∩US ) = φS (UN ∩US ) = R −{0}, φS ◦φ−1
n −1
N (y) = φN ◦φS (y) = ∥y∥
−2
y
∞
and these are C . Thus, {(US , φS ), (UN , φN )} is an atlas for S and so de- n

termines a differentiable structure for S n .

Remarks: The structures of S 3 and S 4 are substantially elucidated through
the use of quaternions. One identifies
( R)4 with the real vector space of 2 × 2
α β
complex matrices of the form −β̄ ᾱ with the squared norm of such a
matrix taken to be its determinant (page 36, [N4]). Matrix multiplication
then corresponds to quaternion multiplication on R4 . S 3 is thus the group
of unit quaternions and can be identified with the special unitary group
SU (2) (Theorem 1.1.4, [N4]). Moreover, the overlap functions for the stere-
ographic projection charts (US , φS ) and (UN , φN ) on S 4 can be written
φS ◦ φ−1 −1
N (y) = φN ◦ φS (y) = ȳ
−1
for all y ∈ R4 − {0} = H − {0}.

5. (Projective Spaces) Let F = R, C, or H. Regard Fn as a right mod-

ule over F (a vector space if F = R or C). Let 0 = (0, . . . , 0) ∈ Fn
and define the standard bilinear form < , >: Fn × Fn −→ F on Fn
by < ξ, τ > = ξ¯1 τ 1 + · · · + ξ¯n τ n for all ξ, τ ∈ Fn . Let S = {ξ ∈ Fn :
< ξ, ξ > = 1}.
Fn has the topology of Rn , R2n , or R4n depending on whether F = R, C,
or H, so S is homeomorphic to S n−1 , S 2n−1 , or S 4n−1 . Define an equivalence
relation ∼ on S as follows: τ ∼ ξ if and only if there exists an a ∈ F with |a| = 1
such that τ = ξa. The equivalence class containing ξ is [ξ] = [ξ 1 , . . . , ξ n ] =
{(ξ 1 a, . . . , ξ n a) : a ∈ F, |a| = 1}. Each [ξ] is homeomorphic to S 0 , S 1 , or
S 3 (pages 50–51, [N4]). The quotient space S/ ∼ is the (real, complex, or
quaternionic) projective space FPn−1 . All of these are Hausdorff (page 51,
[N4]), second countable (Exercise 1.3.2, [N4]) and, being continuous images
of spheres, compact and connected.
Let P : S −→ FPn−1 be the quotient map (P(ξ) = [ξ]). For each
k = 1, . . . , n, let Uk = {[ξ] ∈ FPn−1 : ξ k ̸= 0}. Then P −1 (Uk ) =
{ξ ∈ S : ξ k ̸= 0}. Define φk : Uk −→ Fn−1 (= Rn−1 , R2n−2 , or R4n−4 ) by
φk ([ξ]) = φk ([ξ 1 , . . . , ξ k , . . . , ξ n ]) = (ξ 1 (ξ k )−1 , . . . , b
1, . . . , ξ n (ξ k )−1 ) (where b
means “deleted”). Then φk −1
: F n−1
−→ Uk is given by φ−1 k (y) =
−1 1 n−1 1 n−1
φk (y , . . . , y ) = [y , . . . , 1, . . . , y ] (with the 1 in the k th slot). These
are inverse homeomorphisms (pages 51–52, [N4]) and the overlap functions
φk ◦ φ−1 j : φj (Uk ∩ Uj ) −→ φk (Uk ∩ Uj ) are C ∞ . For example,
 1.1. Smooth Manifolds and Maps 5

φ2 ◦ φ−1 2 3 n 2 3 n 2 −1 3 2 −1
1 (y , y , . . . , y ) = φ2 ([1, y , y , . . . , y ]) = ((y ) , y (y ) , . . . ,
y (y ) ). Thus {(Uk , φk )}k=1,...,n is an atlas for FP
n 2 −1 n−1
and so determines a
differentiable structure.
Remark: When n = 2 we have RP1 ∼
= S 1 , CP1 ∼
= S 2 and HP1 ∼
= S4
(pages 53–54, [N4]).
6. (Classical Groups) Let F = R, C, or H and let GL(n, F) be the set of
all n × n matrices with entries in F (the reason for the peculiar notation will
emerge in Section 1.2). Topologically we identify GL(n, R) = Rn , GL(n, C) =
2

R2n and GL(n, H) = R4n . Let GL(n, F) denote the set of all invertible ele-
2 2

ments of GL(n, F). Then GL(n, F) is an open submanifold of GL(n, F) (pages

41–43, [N4]). Let U (n, F) = {A ∈ GL(n, F) : A−1 = ĀT }.

F = R : U (n, R) = O(n) = orthogonal group of order n

F = C : U (n, C) = U (n) = unitary group of order n
F = H : U (n, H) = Sp(n) = symplectic group of order n
U (n, F) is a submanifold of GL(n, F), but this is not obvious. For exam-
ple, O(n) is a submanifold of GL(n, R) because it is the inverse image of a
regular value under a smooth map of GL(n, R) into a Euclidean space (iden-
tify the set S(n, R) of symmetric n × n real matrices with Rn(n+1)/2 , define
f : GL(n, R) −→ S(n, R) by f (A) = AAT and note that O(n) = f −1 (id)).
For details, see [N4] (pages 256–257 for O(n), and Exercise 5.8.12 for U (n)
and Sp(n)). Next let

SO(n) = {A ∈ O(n) : det A = 1} = special orthogonal group of order n

SU (n) = {A ∈ U (n) : det A = 1} = special unitary group of order n

SO(n) and SU (n) are submanifolds of O(n) and U (n), respectively.

Remarks: SO(2) ∼ = U (1), SO(3) ∼
= RP3 (pages 395 and 399, [N4]), and
SU (2) ∼
= Sp(1) ∼
= S3.
7. (Products) Let X and Y be smooth manifolds of dimension n and m,
respectively. If (U, φ) is a chart on X and (V, ψ) is a chart on Y , then
(U × V, φ × ψ) is a chart on the topological space X × Y (the product map
φ × ψ : U × V −→ Rn × Rm = Rn+m is given by (φ × ψ)(p, q) = (φ(p),
ψ(q))). The set of all such charts is an atlas for X × Y determining the prod-
uct manifold structure on X × Y . The process obviously extends to larger
(finite) products. Important examples are the tori S 1 × S 1 , S 1 × S 1 × S 1 , . . .
and certain products of classical groups, e.g., SU (2)×U (1) (electroweak gauge
group), SU (3) × SU (2) × U (1) (standard model of elementary particles), etc.
A vector field on a smooth manifold X is a map V that assigns to each
p ∈ X a tangent vector V (p) = V p in Tp (X). If (U, φ) is a chart on X
with coordinate functions x1 , . . . , xn , then p −→ ∂x
∂
i |p is a vector field on
6 1. Geometrical Background

U for each i = 1, . . . , n. For any V and any p ∈ U, V (p) = V p (xi ) ∂x ∂

i |p

and the functions V : U −→ R defined by V (p) = V p (x ) are called the

i i i

components of V relative to (U, φ). V is continuous, smooth, etc. if its

components are continuous, smooth, etc. for all charts in some atlas for X.
The set of all smooth vector fields on X is denoted X (X) and has the obvious
structure of a module over C ∞ (X). Any V ∈ X (X) acts on any f ∈ C ∞ (X)
to give a V f = V (f ) in C ∞ (X) defined by (V f )(p) = V p (f ) for all p ∈ X.
This operator on C ∞ (X) is a derivation (i.e., V (af + bg) = aV f + bV g
and V (f g) = f V g + (V f )g for all a, b ∈ R and f, g ∈ C ∞ (X)). Conversely,
any derivation D : C ∞ (X) −→ C ∞ (X) gives rise to a smooth vector field V
on X defined by V (p)(f ) = D(f )(p). Thus, vector fields may be identified
with derivations. As an application we define the Lie bracket [V , W ] of two
smooth vector fields V and W on X to be the derivation/vector field defined
by [V , W ](f ) = V (W f ) − W (V f ). We record some useful properties of the
Lie bracket (proofs on pages 263–264, [N4]):

[W , V ] = −[V , W ]
[a1 V 1 + a2 V 2 , W ] = a1 [V 1 , W ] + a2 [V 2 , W ]
[f V , gW ] = f g[V , W ] + f (V g)W − g(W f )V
[V 1 , [V 2 , V 3 ]] + [V 3 , [V 1 , V 2 ]] + [V 2 , [V 3 , V 1 ]] = 0
( i i
)
j ∂W j ∂V ∂
[V , W ] = V j
−W j
∂x ∂x ∂xi

for all V , W , V 1 , V 2 , V 3 ∈ X (X), a1 , a2 ∈ R, f, g ∈ C ∞ (X) and, for the

last, any chart on X. An integral curve for V ∈ X (X) is a smooth curve α :
(a, b) −→ X whose velocity vector α′ (t0 ) at each t0 in (a, b) coincides with the
vector assigned to α(t0 ) by V , i.e., α′ (t0 ) = V (α(t0 )). These always exist, at
least locally (Theorem 5.7.2, [N4]).
Let V be an n-dimensional real vector space and {e1 , . . . , en } and
{ê1 , . . . , ên } two ordered bases for V. Then there exists a unique nonsingu-
lar n × n real matrix (Ai j )i,j=1,...,n such that êj = Ai j ei for j = 1, . . . , n.
Since det(Ai j ) ̸= 0 we can define an equivalence relation ∼ on the set of all
ordered bases for V by {ê1 , . . . , ên } ∼ {e1 , . . . , en } if and only if det(Ai j ) > 0
and conclude that there are precisely two equivalence classes, each of which
is called an orientation for V. The equivalence class containing {e1 , . . . , en }
is denoted [e1 , . . . , en ].
Now let X be an n-dimensional smooth manifold and U an open subset
of X. An orientation on U is a function µ that assigns to each p ∈ U an
orientation µp for Tp (X) and satisfies the following smoothness condition: For
each p0 ∈ U there is an open neighborhood W of p0 in X with W ⊆ U
and smooth vector fields V 1 , . . . , V n on W with {V 1 (p), . . . , V n (p)} ∈ µp
for each p ∈ W . For example, [ if (U, φ) ]is a chart with coordinate functions
x1 , . . . , xn , then p −→ ∂x ∂
|
1 p , . . . , ∂x∂n |p is an orientation on U . A manifold
 1.1. Smooth Manifolds and Maps 7

for which an orientation exists on all of X is said to be orientable and such

an X is said to be oriented by any specific choice of an orientation µ on
X. A connected, orientable manifold admits precisely two orientations (The-
orem 5.10.2, [N4]). If one of the orientations is µ, then the other is denoted
−µ and called the opposite orientation. If X is orientable with orientation
1 n
µ, then a chart (U,{φ) with coordinate } functions x , . . . , x is said to be con-
sistent with µ if ∂x1 |p , . . . , ∂xn |p ∈ µp for each p ∈ U . If (U, φ) and (V, ψ)
∂ ∂

are charts with coordinate functions x1 , . . . , xn and y 1 , . . . , y n , respectively,

both consistent with µ, and for which U ∩ V ̸= ∅, then the Jacobian of the
coordinate transformation (y 1 , . . . , y n ) = ψ ◦ φ−1 (x1 , . . . , xn ) must have pos-
itive determinant on φ(U ∩ V ). It follows that an orientable manifold has an
oriented atlas, i.e., an atlas whose overlap functions all have Jacobians with
positive determinant. The converse is also true, i.e., a manifold is orientable
if and only if it admits an oriented atlas (Exercise 5.10.9, [N4]). With the ex-
ception of RPn−1 for n odd, all of the manifolds introduced in Examples 1–6
above are orientable (pages 307–309, [N4]). Suppose X and Y are manifolds
with orientations µ and ν, respectively, and that f : X −→ Y is a diffeomor-
phism of X onto Y . Then, for each p ∈ X, f∗p : Tp (X) −→ Tf (p) (Y ) is an
isomorphism and so carries every basis for Tp (X) onto a basis for Tf (p) (Y ). f
is said to be orientation preserving if, for each p ∈ X, f∗p carries every
basis in µp onto a basis in νf (p) . If X is connected, then a diffeomorphism
f : X −→ Y is either orientation preserving or orientation reversing in the
sense that, for each p ∈ X, it carries every basis in µp onto a basis in −νf (p) .
The dual Tp∗ (X) of Tp (X) is called the cotangent space to X at p and its
elements are called covectors at p. A (real-valued) 1-form on X is a map
Θ that assigns to each p ∈ X a covector Θ(p) = Θ p in Tp∗ (X). For example, if
f ∈ C ∞ (X) we define its exterior derivative (or differential) df as follows:
For each p ∈ X, df (p) = dfp is the element of Tp∗ (X) whose value at v ∈ Tp (X)
is df (p)(v ) = dfp (v ) = v (f ). Then df is a 1-form on X. In particular, if
(U, φ) is a chart on X with coordinate functions x1 , . . . , xn , then each dxi is
a 1-form on U and, at each p ∈ U, {dx1 p , . . . , dxn p } is the basis for Tp∗ (X)
{ }
1 |p , . . . , ∂xn |p . If Θ is any 1-form, then, at each p ∈ U, Θ(p) =
∂ ∂
dual to ∂x
i |p ) (Exercise 5.5.14, [N4]). The
∂
Θ p = Θi (p)dxi p , where Θi (p) = Θ(p)( ∂x
functions Θi are the components of Θ relative to (U, φ) and Θ is said to be
continuous, smooth, etc. if its components are continuous, smooth, etc. for all
charts in some atlas for X. The set X ∗ (X) of all smooth 1-forms on X has
the structure of a module over C ∞ (X). Moreover, any Θ ∈ X ∗ (X) gives rise
to a C ∞ (X)-module homomorphism of X (X) to C ∞ (X) defined as follows:
For every V ∈ X (X), Θ(V ) = ΘV is given by Θ(V )(p) = (ΘV )(p) =
Θ p (V p ). Thus, relative to any chart, Θ(V )(p) = Θi (p)V i (p). Conversely,
every C ∞ (X)-module homomorphism A : X (X) −→ C ∞ (X) determines a
unique 1-form Θ with Θ(V ) = A(V ) for every V ∈ X (X) (pages 265–266),
[N4]). If the set of such homomorphisms is given its natural C ∞ (X)-module
8 1. Geometrical Background

structure, then this correspondence is an isomorphism (Exercise 5.7.11, [N4])

so we will not distinguish between the 1-form and the homomorphism.
If F : X −→ Y is smooth and Θ is a 1-form on Y , then the pullback
F ∗ Θ ∈ X ∗ (X) is defined by (F ∗ Θ)p (v ) = Θ F (p) (F∗p (v )) for each p ∈ X
and each v ∈ Tp (X). In coordinates, if (U, φ) is a chart on X with coordinate
functions x1 , . . . , xn and (V, ψ) is a chart on Y with coordinate functions
y 1 , . . . , y m and if φ(U ) ⊆ V , then Θ = Θi dy i =⇒

∂F i
F ∗Θ = (Θi ◦ F )dxj
∂xj
= Θi (F 1 (x1 , . . . , xn ), . . . , F m (x1 , . . . , xn ))d(F i (x1 , . . . , xn )),

where ψ ◦ F ◦ φ−1 (x1 , . . . , xn ) = (F 1 (x1 , . . . , xn ), . . . , F m (x1 , . . . , xn )).

As a special case, if g ∈ C ∞ (Y ), then F ∗ (dg) = d(g ◦ F ). Smooth real- valued
functions are often called 0-forms and, if one defines the pullback of a 0-form
g ∈ C ∞ (Y ) by F : X −→ Y to be F ∗ g = g ◦ F ∈ C ∞ (X), then this last result
reads F ∗ (dg) = d(F ∗ g), i.e., pullback commutes with the exterior derivative
on 0-forms. Note that the exterior derivative operator d carries 0-forms to
1-forms. For future reference we record its basic properties as follows:
d(af1 + bf2 ) = adf1 + bdf2 (a, b ∈ R and f1 , f2 ∈ C ∞ (X))
d(f1 f2 ) = f1 df2 + f2 df1 (f1 , f2 ∈ C ∞ (X))
F ∗ (dg) = d(F ∗ g) (F : X −→ Y smooth and g ∈ C ∞ (Y ))
and, if (U, φ) is a chart on X with coordinate functions x1 , . . . , xn ,

∂f i
df = dx (f ∈ C ∞ (X)).
∂xi
If X ′ is a submanifold of X and ι : X ′ ,→ X is the inclusion map, then the
restriction of Θ ∈ X ∗ (X) to X ′ is defined to be ι∗ Θ. Finally, if F : X −→ Y
and G : Y −→ Z are smooth, then
(G ◦ F )∗ = F ∗ ◦ G∗

(Exercise 5.7.15, [N4]).

If V is a finite dimensional real vector space, then its dual V ∗ (real-valued
linear maps on V) is often denoted J 1 (V) and called the space of covariant
tensors of rank one on V. We denote by J 2 (V) the set of all real-valued
bilinear maps on V ×V and refer to its elements as covariant tensors of rank
two on V. If α, β ∈ J 1 (V) we define their tensor product α ⊗ β ∈ J 2 (V)
by (α ⊗ β)(v, w) = α(v)β(w). If {e1 , . . . , en } is a basis for V and {e1 , . . . , en }
is its dual basis for J 1 (V), then {ei ⊗ ej : i, j = 1, . . . , n} is a basis for J 2 (V)
and every A ∈ J 2 (V) can be written uniquely as

A = Aij ei ⊗ ej = A (ei , ej ) ei ⊗ ej
 1.1. Smooth Manifolds and Maps 9

(Lemma 5.11.1, [N4]). An A ∈ J 2 (V) is said to symmetric if A(w, v) =

A(v, w) for all v, w ∈ V, skew-symmetric if A(w, v) = −A(v, w) for all v, w ∈
V, nondegenerate if A(v, w) = 0 for all v ∈ V implies w = 0 and positive
(respectively, negative) definite if A(v, v) ≥ 0 (respectively, A(v, v) ≤ 0)
for all v ∈ V with A(v, v) = 0 only for v = 0. A nondegenerate, symmetric
element of J 2 (V) is called an inner product on V (some sources reserve
this terminology for a nondegenerate, symmetric bilinear form that is also
positive definite, but the more general notion we have introduced will be
important in the context of relativity). The set of all skew-symmetric elements
of J 2 (V) is denoted Λ2 (V). For all α, β ∈ J 1 (V) we define the wedge product
α ∧ β ∈ Λ2 (V) by α ∧ β = α ⊗ β − β ⊗ α. If {e1 , . . . , en } is a basis for V and
{e1 , . . . , en } is its dual basis, then {ei ∧ ej : 1 ≤ i < j ≤ n} is a basis for Λ2 (V)
and each Ω in Λ2 (V) can be uniquely written as
∑ 1
Ω= Ωij ei ∧ ej = Ωij ei ∧ ej ,
i<j
2

where Ωij = Ω(ei , ej ) (Lemma 5.11.2, [N4]).

If X is a smooth manifold, then a covariant tensor field of rank two
on X is a map A that assigns to each p ∈ X an element A(p) = Ap ∈
J 2 (Tp (X)). If (U, φ) is( a chart on) X with coordinate functions x1 , . . . , xn ,
then A(p) = Ap = Ap ∂x ∂
| , ∂ | dxi ⊗ dxj p . The functions Aij : U −→ R
( i∂ p ∂xj ∂p ) p
defined by Aij (p) = Ap ∂xi |p , ∂xj |p are the components of A relative to
(U, φ) and A is said to be continuous, smooth, etc. if its components are
continuous, smooth, etc. for all charts in some atlas for X. The set of all
smooth A is denoted J 2 (X) and has the structure of a C ∞ (X)-module. A
g ∈ J 2 (X) which, at each p ∈ X, is an inner product on Tp (X) is called a
metric tensor on X. If each g (p) is a positive definite inner product, then g is
a Riemannian metric; otherwise, g is semi-Riemannian. An Ω ∈ J 2 (X)
which, at each p ∈ X, is skew-symmetric is called a 2-form on X and the set
of all such is denoted Λ2 (X). If Θ 1 , Θ 2 ∈ X ∗ (X), then Θ 1 ⊗ Θ 2 ∈ J 2 (X)
and Θ 1 ∧ Θ 2 ∈ Λ2 (X) are defined pointwise. If F : X −→ Y is smooth and
A ∈ J 2 (Y ), then the pullback F ∗ A ∈ J 2 (X) is defined, at each p ∈ X, by
(F ∗ A)p (v , w ) = AF (p) (F∗p (v ), F∗p (w )) for all v , w ∈ Tp (X). The pullback
of a metric (Riemannian or semi-Riemannian) is a metric of the same type
and the pullback, of a 2-form is another 2-form. Restrictions are, as for 1-
forms, pullbacks by the inclusion map. An element A of J 2 (X) gives rise to a
C ∞ (X)-bilinear map A : X (X)×X (X) −→ C ∞ (X) defined as follows: For all
V , W ∈ X (X) define A(V , W ) ∈ C ∞ (X) by A(V , W )(p) = Ap (V p , W p )
for all p ∈ X. Conversely, any C ∞ (X)-bilinear map A : X (X) × X (X) −→
C ∞ (X) gives rise to an A ∈ J 2 (X) whose value at each p ∈ X is the bilinear
map Ap : Tp (X) × Tp (X) −→ R defined as follows: For v , w ∈ Tp (X) select
V , W ∈ X (X) with V (p) = v and W (p) = w (Exercise 5.7.10, [N4]) and set
Ap (v p , w p ) = A(V , W )(p). This correspondence is one-to-one and preserves
10 1. Geometrical Background

the natural algebraic structures (Exercise 5.11.19, [N4]), so we may identify

the two notions (covariant tensor field of rank two and C ∞ (X)-bilinear map
of X (X) × X (X) to C ∞ (X)). In particular, if Θ is a 1-form (thought of as a
map from X (X) to C ∞ (X)), we define its exterior derivative dΘ ∈ Λ2 (X)
(thought of as a bilinear map from X (X) × X (X) to C ∞ (X)) by
dΘ(V , W ) = V (ΘW ) − W (ΘV ) − Θ([V , W ]).
We record some properties of the exterior derivative operator d : Λ1 (X) −→
Λ2 (X) (proved on pages 320–322, [N4]).
d(aΘ 1 + bΘ 2 ) = adΘ 1 + bdΘ 2 (a, b ∈ R and Θ 1 , Θ 2 ∈ X ∗ (X))
d(f Θ) = f dΘ + df ∧ Θ (f ∈ C ∞ (X) and Θ ∈ X ∗ (X))
d(df ) = 0 (f ∈ C ∞ (X))
F ∗ (dΘ) = d(F ∗ Θ) (F : X −→ Y smooth and Θ ∈ X ∗ (Y ))

and, if (U, φ) is a chart on X with coordinate functions x1 , . . . , xn ,

∂Θi
d(Θi dxi ) = dΘi ∧ dxi = dxj ∧ dxi (Θ ∈ X ∗ (X)).
∂xj
Let X be a smooth manifold, V a finite dimensional real vector space and
{e1 , . . . , ed } a basis for V. A V-valued 0-form on X is a map of X into V.
Thus, we may write ϕ = ϕi ei , where the real-valued functions ϕi are called
the components of ϕ relative to {e1 , . . . , ed }. ϕ is said to be continuous,
smooth, etc. if its components are continuous, smooth, etc. relative to some
(and therefore any) basis for V. A V-valued 1-form on X is a map ω which
assigns to every p ∈ X a linear transformation ω(p) = ω p from Tp (X) to V.
Writing ω = ω i ei , where each ω i is an ordinary (R-valued) 1-form, we say that
ω is continuous, smooth, etc. if its components ω 1 , . . . , ω d are continuous,
smooth, etc. A V-valued 2-form on X is a map Ω which assigns to each
p ∈ X a bilinear map Ω(p) = Ω p : Tp (X) × Tp (X) −→ V and may be written
Ω = Ω i ei , where each Ω i is an ordinary 2-form. Again, Ω 1 , . . . , Ω d are
the components of Ω relative to {e1 , . . . , ed } and Ω is continuous, smooth,
etc. if and only if its components are continuous, smooth, etc. Pullbacks of
V-valued forms are defined componentwise as are exterior derivatives of V-
valued 0-forms and 1-forms. All of this is independent of the choice of basis.
Wedge products of vector-valued 1-forms are not defined unless there is given
some sort of “multiplication” in the vector space in which the 1-forms take
their values. Suppose, somewhat more generally, that U , V and W are finite
dimensional real vector spaces and that one is given a bilinear map ρ : U ×
V −→ W (when U = V = W this is a bilinear pairing, or “multiplication” on
V). Now, if ω is a U -valued 1-form on X and η is a V-valued 1-form on X,
then we define the ρ-wedge product ω ∧ρ η of ω and η by
( ) ( )
(ω ∧ρ η)p (v , w ) = ρ ω p (v ), η p (w ) − ρ η p (v ), ω p (w )
 1.2. Matrix Lie Groups 11

for all p ∈ X and all v , w ∈ Tp (X). This then is a W-valued 2-form on X. If

{u1 , . . . , uc } is a basis for U and {v1 , . . . , vd } is a basis for V and if we write
ω = ω i ui , and η = η j vj , then
∑c ∑ d
( i )
ω ∧ρ η = ω ∧ η j ρ (ui , vj )
i=1 j=1
(Lemma 5.11.5, [N4]). For example, if U = V = W = C (regarded as a 2-
dimensional real vector space) and ρ : C × C −→ C is complex multiplication,
then, writing ω = ω 1 + ω 2 i and η = η 1 + η 2 i , one finds that

ω ∧ρ η = (ω 1 + ω 2 i ) ∧ρ (η 1 + η 2 i )
= (ω 1 ∧ η 1 − ω 2 ∧ η 2 ) + (ω 1 ∧ η 2 + ω 2 ∧ η 1 )i .

Thus, ω ∧ρ η is obtained by multiplying the forms exactly as one multiplies

complex numbers, but with real and imaginary parts “multiplied” by the ordi-
nary wedge. A similar result holds when U = V = W = H and ρ is quaternion
multiplication. In cases such as these one generally drops the “ρ” and writes
simply ω ∧ η. Most of the important examples of vector- valued forms arise
in the context of Lie groups and principal bundles, to which we now turn.

1.2 Matrix Lie Groups

A Lie group is a differentiable manifold G on which is defined a group structure
for which the group multiplication (x, y) −→ xy is a C ∞ map of G × G to G
(it follows that group inversion x −→ x−1 is a diffeomorphism of G onto itself;
see Lemma 5.8.1, [N4]). The left and right translation maps Lg , Rg : G −→ G
defined by Lg (x) = gx and Rg (x) = xg are diffeomorphisms for each g ∈ G.
The Lie groups of most interest to us are the following:
1. The nonzero real numbers, complex numbers, or quaternions with their
respective multiplications.
2. The circle S 1 (thought of as the complex numbers of modulus 1 in C =
R2 ) with complex multiplication.
3. The 3-sphere S 3 (thought of as the unit quaternions in H = R4 ) with
quaternion multiplication.
4. The general linear groups GL(n, R), GL(n, C) and GL(n, H) with
matrix multiplication.
5. Any subgroup H of a Lie group G that is also a submanifold of G is a Lie
group so, in particular, all of the classical groups are Lie groups:

O(n), SO(n), U (n), SU (n), Sp(n).

12 1. Geometrical Background

6. Any (finite) product of Lie groups (with the product manifold structure
and the direct product group structure) is a Lie group, e.g., SU (2) × U (1),
SU (3) ×SU (2) × U (1), or any torus S 1 × · · · × S 1 .
Two Lie groups G1 and G2 are isomorphic if there is a diffeomorphism of
G1 onto G2 that is also a group isomorphism, e.g., S 1 is isomorphic to both
U (1) and SO(2), while S 3 is isomorphic to SU (2) and Sp(1). Any subgroup of
some GL(n, C) that is also a submanifold is called a matrix Lie group and
we shall henceforth restrict our attention to these. GL(n, R) can be identified
with a subgroup of GL(n, C) that is also a submanifold. It is less obvious, but
also true, that GL(n, H) can be identified with a subgroup of GL(2n, C) that
is also a submanifold. To see this observe first that any quaternion x0 + x1 i +
x2 j + x3 k can be written as z 1 + z 2 j , where z 1 = x0 + x1 i and z 2 = x2 + x3 i ,
and so identified with a pair of complex numbers. Thus, any n×n quaternionic
matrix P can be written as P = A + Bj , where A and B are n × n complex
matrices. Define a map ϕ : GL (n, H) −→ GL(2n, C) by
( )
A B
ϕ(P ) = .
−B̄ Ā
Then ϕ is an algebra isomorphism that also preserves the conjugate transpose
(Exercise 1.1.28, [N4]). In particular, P ∈ Sp(n) if and only
( if ϕ(P) ) ∈ U (2n).
Moreover, a 2n × 2n complex matrix M has the form −B̄ A B
Ā
if and only
−1
( 0 id )
if it satisfies JM J = M̄ , where J = −id 0 and, if M is unitary, this is
T
equivalent to M JM = J. Thus, we may identify

GL(n, H) = {M ∈ GL(2n, C) : M invertible and JM J −1 = M̄ }

Sp(n) = {M ∈ U (2n) : M T JM = J}.

The real and complex special linear groups SL(n, R) and SL(n, C) are
the subgroups of GL(n, R) and GL(n, C), respectively, consisting of those el-
ements with determinant 1. Although the noncommutativity of H blocks any
obvious notion of a determinant for quaternionic matrices, one can define the
quaternionic special linear group SL(n, H ) to be the subset of GL(n, H)
consisting of all those elements P such that det ϕ(P ) = 1. This is, indeed,
a subgroup of GL(n, H) (Exercise 1.1.30, [N4]). One can show directly that
GL(n, H), Sp(n) and SL(n, H) are all submanifolds of GL(2n, C), but there
is also a general result to the effect that any closed subgroup of a complex
general linear group is necessarily a submanifold (see [Howe] or, for a still
more general result, [Warn]). We conclude then that all of the classical groups
are matrix Lie groups.
A Lie algebra is a real vector space L on which is defined a bilinear opera-
tion [ , ] : L×L −→ L, called bracket, that is skew-symmetric ([y, x] = −[x, y]
for all x, y ∈ L) and satisfies the Jacobi identity ([x, [y, z]] + [z, [x, y]] +
[y, [z, x]] = 0 for all x, y, z ∈ L). Two Lie algebras L1 and L2 with brackets
 1.2. Matrix Lie Groups 13

[ , ]1 and [ , ]2 , respectively, are isomorphic if there is a linear isomorphism

T : L1 −→ L2 of L1 onto L2 that satisfies T ([x, y]1 ) = [T (x), T (y)]2 for all
x, y ∈ L1 . We record a few important examples of Lie algebras
1. Any finite dimensional real vector space V with the trivial bracket [ , ] :
V × V −→ V defined by [x, y] = 0 ∈ V for all x, y ∈ V. Skew-symmetry implies
that any 1-dimensional Lie algebra has trivial bracket.
2. The commutator of two matrices A and B is defined by [A, B] =
AB − BA. Since [B, A] = −[A, B] and [A, [B, C]] + [C, [A, B]] + [B, [C, A]] = 0,
any set of matrices that forms a real vector space and is closed under the for-
mation of commutators is a Lie algebra with bracket given by the commutator.
Examples include GL(n, R), GL(n, C) and GL(n, H) as well as the set of all
n × n real, skew-symmetric (AT = −A) matrices, the set of all n × n complex,
skew-Hermitian (ĀT = −A) matrices and the set of all n×n complex matrices
that are both skew-Hermitian and tracefree (trace(A) = 0).
3. The set Im C of pure imaginary complex numbers ai with trivial bracket
is a 1-dimensional Lie algebra.
4. The set Im H of pure imaginary quaternions x = ai + bj + ck with bracket
[x, y] = xy − yx = 2 Im (xy) is a 3-dimensional Lie algebra. This Lie algebra
is, in fact, isomorphic to the Lie algebra consisting of R3 and the bracket
[x, y] = x × y given by the familiar cross product on R3 .
Every Lie group G has associated with it a Lie algebra G (called the Lie
algebra of G) that can be described in either of the following two equivalent
ways:
(i) A vector field V on G is said to be left-invariant if (Lg )∗ ◦ V = V ◦ Lg
for every g ∈ G, i.e., if (Lg )∗h (V (h)) = V (gh) for all g, h ∈ G. Such
vector fields are necessarily smooth (Theorem 5.8.2, [N4]). The set G of
all left-invariant vector fields on G is a linear subspace of X (G) and is
closed under the formation of Lie brackets (Theorem 5.8.4, [N4]). Since
the Lie bracket is skew-symmetric and satisfies the Jacobi identity it
provides G with the structure of a Lie algebra.
(ii) If e denotes the identity element in G (e = id for matrix Lie groups),
then the map V −→ V (e) is a linear isomorphism of G (as defined in (i))
onto Te (G). Define a bracket [ , ] on Te (G) as follows: For v, w ∈ Te (G)
there exist unique left-invariant vector fields V , W ∈ G with V (e) = v
and W (e) = w. Set [v, w] = [V , W ](e). With this, Te (G) is a Lie algebra
isomorphic to G.

For matrix Lie groups G, Tid (G) can be identified with a set of matrices
(velocity vectors at t = 0 to smooth curves t −→ (aij (t)) with (aij (0)) = id,
computed entrywise). One can show (Section 5.8, [N4]) that this set of
matrices is always closed under the formation of commutators and that the
14 1. Geometrical Background

Lie algebra consisting of Tid (G) with commutator bracket is isomorphic to

the Lie algebra G of G as defined in (ii). Thus, the problem of finding the Lie
algebra G of a matrix Lie group G reduces to identifying the set of matrices
that arise as velocity vectors to smooth curves in G through id (and defin-
ing the bracket on this set of matrices to be the commutator). We record a
number of important examples (for the proofs see pages 278–284, [N4]).
1. For F = R, C, or H, the Lie algebra of GL(n, F) is the set GL(n, F) of
all n × n matrices with entries in F.
2. The Lie algebra o(n) of the orthogonal group O(n) is the set of all n × n
real, skew-symmetric matrices.
3. The Lie algebra so (n) of the special orthogonal group SO(n) coincides
with the Lie algebra o(n) of O(n) (because SO(n) is just the connected com-
ponent of O(n) containing id so that Tid (SO(n)) = Tid (O(n))).
4. The Lie algebra u(n) of the unitary group U (n) is the set of all n × n
complex, skew-Hermitian matrices.
5. The Lie algebra su(n) of the special unitary group SU (n) is the set of all
n × n complex matrices that are skew-Hermitian and tracefree.
6. Identifying Sp(n) with the matrix Lie group {U ∈ U (2n) : U T JU = J},
the Lie algebra sp(n) of Sp(n) is given by

sp(n) = {M ∈ GL(2n, C) : M̄ T = −M and JM + M T J = 0}

{( ) }
A B
= ∈ GL (2n, C) : A ∈ u(n), B = B .
T
−B̄ Ā

Isomorphic Lie groups have isomorphic Lie algebras. For example, since
U (1) and SO(2) are isomorphic, so are u(1) and so(2) and, since u(1) is the
algebra of 1 × 1 skew-Hermitian matrices, both can be naturally identified
with the Lie algebra Im C. Similarly, SU (2) and Sp(1) are isomorphic and
therefore so are their Lie algebras su(2) and sp(1). But sp(1) is naturally
identfied with the Lie algebra Im H of pure imaginary quaternions which,
in turn, is isomorphic to R3 with the cross product as bracket. However,
non-isomorphic Lie groups can have isomorphic Lie algebras, e.g., O(n) and
SO(n) in #3 above. A less trivial example consists of SU (2) and SO(3).
These are certainly not isomorphic as Lie groups. Indeed, they are not even
homeomorphic since SU (2) ∼ = S 3 (Theorem 1.1.4, [N4]) and S 3 is simply
connected (page 119, [N4]), whereas SO(3) ∼ = RP3 (page 399, [N4]) and
π1 (R P3 ) ∼ = 2 Z (Theorem 2.4.5, [N4]). To see that the Lie algebras su(2)
and so(3) are isomorphic we introduce the notion of a Lie group’s “structure
constants.” Let G be an arbitrary Lie algebra. Select a basis {e1 , . . . , en } for
G. For all i, j = 1, . . . , n, [ei , ej ] is in G so there exist unique constants Cij k
,
k k
k = 1, . . . , n, such that [ei , ej ] = Cij ek . The constants Cij , i, j, k = 1, . . . , n are
called the structure constants of G relative to the basis {e1 , . . . , en }. Two
 1.2. Matrix Lie Groups 15

Lie algebras are clearly isomorphic if and only if there exist bases relative
to which the structure constants are the same. Now, an obvious basis for
so(3) (3 × 3 real, skew-symmetric matrices) consists of
     
0 0 0 0 0 1 0 −1 0
     
τ1 =   
0 0 −1 , τ2 =  0 0 0 , τ3 = 1
  0 0 
0 1 0 −1 0 0 0 0 0
∑3
and a simple calculation shows that [τi , τj ] = k=1 ϵijk τk , where ϵijk is the
Levi-Civita symbol (1 if ijk is an even permutation of 123, −1 if ijk is an
odd permutation of 123, and 0 otherwise). The usual basis for su(2) consists
of Tk = − 12 i σk , where
( ) ( ) ( )
0 1 0 −i 1 0
σ1 = , σ2 = , σ3 =
1 0 i 0 0 −1

are the Pauli spin matrices (page 396, [N4]) and here again one quickly
∑3
verifies that [Ti , Tj ] = k=1 ϵijk Tk . Thus, so(3) and su(2)
∑∞ are isomorphic.
For any A ∈ GL(n, C) we define exp(A) = eA = k=0 k! 1 k
A . The series
converges absolutely and uniformly on every bounded region in GL(n, C) =
Cn = R2n . Every exp(A) is invertible (because det(eA ) = etrace(A) ) so exp
2 2

is a C ∞ map of GL(n, C) to GL(n, C). If G is any matrix group, then its Lie
algebra G can be identified with a subalgebra of GL(n, C) for some n and the
restriction of exp to G maps into G (see Theorem 5.8.6, [N4], for the cases
of interest here). The map exp : G −→ G has a derivative at 0 ∈ G and, if
T0 (G) is identified with G by the canonical isomorphism, exp∗o : G −→ G is the
identity map (Lemma 5.8.5 (5), [N4]). It follows from the Inverse Function
Theorem that exp is a diffeomorphism of some neighborhood of 0 ∈ G onto a
neighborhood of exp(0) = id in G, i.e., on some neighborhood of id in G, exp−1
is a chart for G.
A 1-form Θ on a Lie group G (real, or vector-valued) is said to be
left invariant if (Lg )∗ Θ = Θ for all g ∈ G, i.e., if, for all g, h ∈ G,
Θ(h) = (Lg )∗ (Θ(gh)), or, equivalently, Θ(gh) = (Lg−1 )∗ (Θ(h)). This is the
case if and only if Θ(g) = (Lg−1 )∗ (Θ(id)) for every g ∈ G. Any such 1-form is
necessarily smooth (page 285, [N4]) and is uniquely determined by its value
at the identity. In particular, any covector at id in G uniquely determines
a left invariant 1-form taking that value at id. A particularly important ex-
ample of a left invariant 1-form on G is the Cartan (canonical) 1-form
which takes values in the vector space G and is defined as follows: For each
g ∈ G, Θ(g) = Θ g : Tg (G) −→ G = Tid (G) is given by
Θ(g)(v ) = Θ g (v ) = (Lg−1 )∗g (v ).
16 1. Geometrical Background

Equivalently, Θ is uniquely determined by the requirement that

Θ g (A(g)) = A(id) for every left invariant vector field A on G. Let {e1 , . . . , en }
be a basis for G and let {Θ 1 , . . . , Θ n } be the unique left invariant real-valued
1-forms on G for which {Θ 1 (id), . . . , Θ n (id)} is the basis dual to {e1 , . . . , en }.
Then the Cartan 1-form Θ is given by
Θ = Θ 1 e1 + · · · + Θ n en
(Lemma 5.9.1, [N4]). The Maurer-Cartan equations for Θ assert that
1 k i
dΘ k = − Cij Θ ∧ Θj , k = 1, . . . , n,
2
where Cijk
are the structure constants for G relative to {e1 , . . . , en } ((5.11.3),
[N4]). If H is a subgroup of G that is also a submanifold and ι : H ,→ G is
the inclusion map, then the Cartan 1-form Θ H of H is the restriction to H
of the Cartan 1-form Θ G of G, i.e.,

Θ H = ι∗ Θ G

((5.9.6), [N4]). Now we exhibit a number of specific examples of Cartan

1-forms for the groups of particular interest to us (details are available on
pages 292–296, [N4]).
1. G = GL(n, R), G = GL(n, R) : GL(n, R) is an open submanifold
{ R }with standard coordinate (entry) functions x , i, j = 1, . . . , n.
n2 ij
of
∂
∂xij id is a basis for GL(n, R). The corresponding dual
{ ij
i,j=1,...,n }
basis is dx (id) i,j=1,...,n . The R-valued 1-forms Θ ij defined by
∑ ( )
Θ ij (g) = k=1 xik g −1 dxkj (g) for all g ∈ GL(n, R) are left invariant and
n

satisfy Θ ij (id) = dxij (id) so Θ is given by

( )
∂ ∑
n
∂
ij ik −1 kj
Θ(g) = Θ (g) ij = x (g )dx (g) .
∂x id ∂xij id
k=1

For calculations
( ∑nit is ik
most convenient
) to identify Θ(g) with the matrix of
−1
components k=1 x (g )dxkj (g) i,j=1,...,n . Note that this is the formal
matrix product of g −1 and
 
dx11 (g) ... dx1n (g)
 
 .. .. 
dx(g) =  . . 
 
dxn1 (g) ... dxnn (g)

so one can write

Θ(g) = g −1 dx(g).
 1.2. Matrix Lie Groups 17

To evaluate Θ(g) at any v ∈ Tg (G) compute

dx(g)(v ) = (dxij (g) (v ))i,j=1,...,n = (v ij )i,j=1,...,n ,

where v = v ij ∂x∂ij |g , and form the product g −1 (v ij )i,j=1,...,n , the result being
a matrix in GL(n, R). Taking this computational ploy one step further we
will often identify Θ with ( a )matrix of R-valued 1-forms. We illustrate ( )with
GL(2, R): For each g = αγ βδ ∈ GL(2, R), g −1 = (αδ − βγ)−1 −γδ −βα and
we compute the product
( )( )
−1 −1 δ −β dx11 (g) dx12 (g)
g dx(g) = (αδ − βγ)
−γ α dx21 (g) dx22 (g)
( )
−1 δdx11 (g) − βdx21 (g) δdx12 (g) − βdx22 (g)
= (αδ − βγ) .
−γdx11 (g) + αdx21 (g) −γdx12 (g) + αdx22 (g)

Since α = x11 (g), β = x12 (g), γ = x21 (g) and δ = x22 (g) we may regard this
as the value at g of the matrix of ordinary 1-forms given by
Θ GL(2,R) =
( )
12 21 −1 x22 dx11 − x12 dx21 x22 dx12 − x12 dx22
11 22
(x x −x x )
−x21 dx11 + x11 dx21 −x21 dx12 + x11 dx22

2. G = GL(n, C), G = GL(n, C): One proceeds just as in #1 above,

identifying G = GL(n, C) with an open submanifold of Cn = R2n with
2 2

standard coordinates {x , y , . . . , x , y }, where z = x + i y . Again,

11 11 nn nn ij ij ij

one finds that, for any g ∈ G, Θ(g) can be identified with g −1 dz(g), where dz
is the matrix of complex-valued 1-forms
   
dz 11 ... dz 1n dx11 + i dy 11 ... dx1n + i dy 1n
 . ..   .. .. 
dz = 
 .
.  
. = . .


n1 nn n1 n1 nn nn
dz ... dz dx + i dy ... dx + i dy
(complex-valued 1-forms are just vector-valued 1-forms with values in C = R2
and we have used the basis {1, i } for C over R). Thus, if v ∈ Tg (G) (thought
of as an n×n matrix of complex numbers obtained by differentiating entrywise
a smooth curve in GL(n, C) through g), then dz ij picks out its ij-entry v ij
and Θ(g)(v) = g −1 (v ij )i,j=1,...,n . For n = 2 one can write Θ as a matrix of
complex-valued 1-forms just as we did for GL(2, R) in #1:

Θ GL(2,C) =
( )
z 22 dz 11 − z 12 dz 21 z 22 dz 12 − z 12 dz 22
(z 11 z 22 − z 12 z 21 )−1
−z 21 dz 11 + z 11 dz 21 −z 21 dz 12 + z 11 dz 22
18 1. Geometrical Background

3. G = GL(n, H), G = GL(n, H): The result is the same as for GL(n, R) and
GL(n, C) in #1 and #2: For every g ∈ GL(n, H),

Θ(g) = g −1 dq(g),

where  
dq 11 ... dq 1n
 . .. 
dq = 
 .
. . 


dq n1 ... dq nn
and dq ij = dxij + i dy ij + j duij + k dv ij , i, j = 1, . . . , n. This time, however,
there is no simple formula for g −1 when n = 2 since we lack a determinant
function for quaternionic matrices. However, when n = 1 and we identify
a 1 × 1 quaternionic matrix g = (q) with its sole entry q, GL(1, H) is just
the group H − {0} of nonzero quaternions and g −1 dq is just the quaternion
product
1
g −1 dq = q −1 dq = 2 q̄dq.
|q|
Identifying tangent vectors v ∈ Tq (H − {0}) with quaternions v ∈ H via the
canonical isomorphism, q −1 dq(v ) = |q|1 2 q̄v.
4. G = SO(2) ∼ = U (1), G = so(2) ∼
= u(1) ∼= Im C: First regard G as SO(2) ⊆
GL(2, R). Then Θ SO(2) is the restriction of Θ GL(2,R) from #1 to SO(2). One
finds (page 294, [N4]) that
( )
0 x22 dx12 − x12 dx22
Θ SO(2) =
−x22 dx12 + x12 dx22 0
( )
0 −1
= (−x22 dx12 + x12 dx22 ) .
1 0

Now,
( cos θ the)standard identification of SO(2) and U (1) is
−sin θ
−→ (ei θ ) and this induces (by differentiation at id) the
sin θ cos θ ( )
identification θ0 01 −10 −→ θ0 i of so(2) and u(1) ∼
= Im C. Thus,

Θ U (1) = (−x22 dx12 + x12 dx22 )i

on U (1) = {(x12 , x22 ) ∈ R2 : (x12 )2 + (x22 )2 = 1}.

5. G = SU (2) ∼ = Sp(1), G = su(2) ∼ = sp(1) ∼= Im H: First regard G as SU (2)
⊆ GL(2, C). Then Θ SU (2) is the restriction of Θ GL(2,C) from #2 to SU (2).
One finds (Exercise 5.9.5, [N4]) that
( )
z̄ 11 dz 11 + z 12 dz̄ 12 z̄ 11 dz 12 − z 12 dz̄ 11
Θ SU (2) =
−z 11 dz̄ 12 + z̄ 12 dz 11 z 11 dz̄ 11 + z̄ 12 dz 12
 1.2. Matrix Lie Groups 19

on SU (2). Now think of G as Sp(1) (identified with the unit quaternions).

Then Θ Sp(1) is the restriction of Θ GL(1,H) in #3 and (because |q| = 1 on
Sp(1)) this is
Θ Sp(1) = q −1 dq = q̄ dq
on Sp(1). Notice that, by writing q = z 11 + z 12 j , where z 11 = x + yi and
z 12 = u + vi , we have

Θ Sp(1) = (z̄ 11 − z̄ 12 j )(dz 11 + dz 12 j )

= (z̄ 11 dz 11 + z 12 dz̄ 12 ) + (z̄ 11 dz 12 − z 12 dz̄ 11 )j

on Sp(1). The standard identification of Sp(1) with SU (2) is

( )
A B
A + Bj −→
−B̄ Ā

and this carries Θ Sp(1) onto Θ SU (2) .

Now, let G be a matrix Lie group, V a finite dimensional vector space and
GL(V) the group of nonsingular linear transformations on V. A represen-
tation of G on V is a group homomorphism of G into GL(V). Choosing a
basis for V one can regard any representation of G on V as a map of G into
some general linear group and we say that the representation is continuous,
smooth, etc., if this corresponding matrix-valued map is continuous, smooth,
etc. Observe that this definition clearly does not depend on the choice of
basis for V. We will occasionally relax our terminology a bit and refer to a
homomorphism of G into a general linear group as a representation of G. An
example of considerable importance arises as follows. For each g ∈ G define a
map
Adg : G −→ G
by Adg (h) = ghg −1 for all h ∈ G. Thus, Adg = Lg ◦ Rg−1 = Rg−1 ◦ Lg is a
diffeomorphism of G onto G. Furthermore, Adg (id) = id so the derivative of
Adg at the identity carries G isomorphically onto G. We denote this map

adg : G −→ G.
Thus, adg = (Adg )∗id = (Lg )∗g−1 ◦ (Rg−1 )∗id = (Rg−1 )∗g ◦ (Lg )∗id and one can
show (Lemma 5.8.7, [N4]) that

adg (A) = g Ag −1
for any A ∈ G. The assignment g −→ adg is a homomorphism of G into GL(G)
and is called the adjoint representation of G on G. The significance of the
adjoint representation may not be immediately apparent, but will become
clear when we discuss connections on principal bundles. On an even more
concrete level, it is shown in Appendix A to [N4] that SO (3) can be identified
20 1. Geometrical Background

with SU (2)/ ± 1 and that, when su(2) is identified with R3 , the adjoint
representation of SU (2) on su (2) is just the natural representation of SO(3)
on R3 by rotation.
We outline a few general facts about Killing forms, although we will require
only special cases (all that we say is proved in [Helg]). If G is a matrix
Lie group with Lie algebra G and if A and B are two fixed elements of G,
then we can define a linear transformation KAB : G −→ G by KAB (X) =
[A, [B, X]] for all X ∈ G. Then the trace of this linear transformation is a
real number (because G is a real vector space) and the map K : G −→ R
defined by K(A, B) = trace (KAB ) is a symmetric, bilinear form on G called
the Killing form of G. K is ad (G)-invariant in the sense that, for any
g ∈ G, K(adg (A), adg (B)) = K(A, B) for all A, B ∈ G. The Lie group G
(or its Lie algebra G) is said to be semisimple if the Killing form K is
nondegenerate.
Remark: There is an algebraic characterization of semisimplicity due to
Cartan. An ideal in a Lie algebra G is a linear subspace H of G with the
property that [A, B] ∈ H whenever A ∈ H and B ∈ G. In particular, an ideal
is itself a Lie algebra under the same bracket operation. A proper ideal is
one that is neither the zero subspace nor the entire Lie algebra. Then G is
semisimple if and only if it can be written as a direct sum of ideals, each of
which (as a Lie algebra) has no proper ideals.
If G is connected and semisimple, then, by a theorem of Weyl, the Killing
form K is negative definite if and only if G is compact. In this case,
one obtains a positive definite, ad(G)-invariant inner product ⟨ , ⟩ on G by
setting ⟨A, B⟩ = −K(A, B). A Riemannian metric g on G is then ob-
tained by left translation, i.e., by defining, for each a ∈ G and all v , w ∈
Ta (G), g a (v , w ) = ⟨(La−1 )∗a (v ), (La−1 )∗a (w )⟩. This metric is left invari-
ant, i.e., L∗a g = g for all a ∈ G, because (L∗a g )b (v , w ) = g ab ((La )∗b (v ),
(La )∗b (w )) = ⟨(L(ab)−1 )∗ab ((La )∗b (v )), (L(ab)−1 )∗ab ((La )∗b (w ))⟩ = ⟨(Lb−1 )∗b
(v ), (Lb−1 )∗b (w )⟩ = g b (v , w ). It is also right invariant, i.e., Ra∗ g = g for
all a ∈ G, because

(Ra∗ g )b (v , w ) = g ba ((Ra )∗b (v ), (Ra )∗b (w ))

= ⟨(L(ba)−1 )∗ba ((Ra )∗b (v )), (L(ba)−1 )∗ba ((Ra )∗b (w ))⟩
= ⟨(La−1 b−1 ◦ Ra )∗b (v ), (La−1 b−1 ◦ Ra )∗b (w )⟩
= ⟨(Ada−1 ◦ Lb−1 )∗b (v ), (Ada−1 ◦ Lb−1 )∗b (w )⟩
= ⟨ada−1 ((Lb−1 )∗b (v )), ada−1 ((Lb−1 )∗b (w ))⟩
= ⟨(Lb−1 )∗b (v ), (Lb−1 )∗b (w )⟩
= g b (v , w ).

A metric on a Lie group that is both left invariant and right invariant is
said to be bi-invariant and we have just shown that the Killing form on
 1.2. Matrix Lie Groups 21

a compact, connected, semisimple Lie group G gives rise to a bi-invariant

Riemannian metric on G. All of this applies, in particular, to G = SU (2)
which is surely compact and connected and also happens to be semisimple
(su (2) ∼= so (3) ∼= sp (1) ∼
= Im H is the only 3-dimensional, semisimple Lie
algebra). Routine, but tedious calculations show that, if su (2) is regarded
as the algebra of 2 × 2 complex, skew-Hermitian, tracefree matrices, then
⟨A, B⟩ = −K(A, B) = −trace (AB). On the other hand, with the standard
identification
( )
a1 i a2 + a3 i
−→ a1 i + a2 j + a3 k −→ (a1 , a2 , a3 )
−a2 + a3 i −a1 i

of su(2) with Im H ∼
= R3 , ⟨ , ⟩ is just twice the usual inner product on R3 .
Next suppose that G is a Lie group and P is a differentiable manifold. A
smooth right action of G on P is a C ∞ map σ : P ×G −→ P which satisfies
1. σ(p, e) = p for all p ∈ P (e is the identity element in G), and
2. σ(p, g1 g2 ) = σ(σ(p, g1 ), g2 ) for all g1 , g2 ∈ G and all p ∈ P .
One generally writes σ(p, g) = p · g and thinks of g as “acting on” p to produce
p · g ∈ P . Then the defining properties assume the form
1. p · e = p for all p ∈ P , and
2. p · (g1 g2 ) = (p · g1 ) · g2 for all g1 , g2 ∈ G and all p ∈ P .
For each fixed g ∈ G we define σg : P −→ P by σg (p) = p · g. Then σg is
a diffeomorphism of P onto P with inverse σg−1 . Similarly, a smooth left
action of G on P is a smooth map ρ : G × P −→ P, ρ(g, p) = g · p, that
satisfies
1. e · p = p for all p ∈ P , and
2. (g1 g2 ) · p = g1 · (g2 · p) for all g1 , g2 ∈ G and all p ∈ P .
The maps ρg : P −→ P defined by ρg (p) = g · p are all diffeomorphisms.
Remark: If (g, p) −→ g · p is a left action, then (p, g) −→ p ⊙ g = g −1 · p
is a right action and, if (p, g) −→ p · g is a right action, then (g, p) −→
g ⊙ p = p · g −1 is a left action. Which sort of action one chooses to deal with
is generally a matter of personal taste, although some actions appear more
natural in one guise than another. For example, any smooth representation
ρ : G −→ GL(V) of G on the vector space V gives rise to an action of G
on the manifold V ((g, v) −→ (ρ(g))(v)) which, because of the way matrix
multiplication is defined, it is more natural to view as acting on the left. We
will formulate the remaining definitions in terms of right actions and leave the
obvious modifications required for left actions to your imagination.
22 1. Geometrical Background

A right action σ of G on P is said to be effective if p · g = p for all p ∈ P

implies g = e, i.e., if σg = idP if and only if g = e. The action is said to be
free if p · g = p for some p ∈ P implies g = e, i.e., if σg has a fixed point if and
only if g = e. A free action is effective, but the converse is false in general.
The action is said to be transitive if, given any two points p1 and p2 in P ,
there exists a g ∈ G such that p2 = p1 ·g. Given any p ∈ P we define the orbit
of p under σ to be the subset p · G = {p · g : g ∈ G} of P and its isotropy
subgroup (or stabilizer) to be the subset Gp = {g ∈ G : p · g = p} of G. Gp
is a closed subgroup of G for any p ∈ P (Exercise 1.6.10, [N4]). An action is
free if every isotropy subgroup is trivial and transitive if there is precisely one
orbit. We pause now to describe a number of important examples.
1. Any Lie group G acts on itself by right multiplication. That is, defining
σ : G × G −→ G by σ(p, g) = pg for all p, g ∈ G gives a smooth right
action of G on G that is clearly free (pg = p implies g = e) and transitive
(p2 = p1 (p−1
1 p2 )).

2. Any Lie group G acts on itself by conjugation. That is, defining

σ : G × G −→ G by σ(p, g) = g −1 pg for all p, g ∈ G gives a smooth right
action of G on G that is neither free nor transitive (unless G = {e}). Note
that Ge = G and, in general, Gg is the centralizer of g ∈ G. The action is
effective if and only if the center of G is trivial. The orbits are the conjugacy
classes of G. The corresponding left action is ρ(g, p) = g ·p = gpg −1 = Adg (p).
3. GL(n, F) acts on Fn (on the left) by A · v = Av, where elements of
Fn are written as column matrices and the product on the right is matrix
multiplication. This action is effective, but neither free nor transitive because
GL(n, F) fixes 0 ∈ Fn . The same action on Fn − {0} is transitive.
4. O(n) and SO(n) both act transitively on S n−1 by A·v = Av (pages 90–91,
[N4]). The isotropy subgroup of the north pole en = (0, . . . , 0, 1) ∈ S n−1 under
the action of O(n) (respectively, SO(n)) on S n−1 is isomorphic to O(n − 1)
(respectively, SO(n − 1)) (page 91, [N4]). In the same way (pages 91–92,
[N4]), U (n) and SU (n) act transitively on S 2n−1 and Sp(n) acts transitively
on S 4n−1 . The isotropy subgroups at the north pole are isomorphic to U (n−1),
SU (n − 1) and Sp(n − 1), respectively.
Remark: There is a general result (Theorem 1.6.6, [N4]) which asserts that,
if G is compact, (g, p) −→ g · p is a transitive left action of G on P and Gp0
is the isotropy subgroup of some fixed p0 ∈ P under this action, then the
quotient group G/Gp0 is homeomorphic to P . Example #4 above then gives
the following homeomorphisms:

S n−1 ∼
= O(n)/O(n − 1) ∼ = SO(n)/SO(n − 1)
2n−1 ∼
S = U (n)/U (n − 1) ∼
= SU (n)/SU (n − 1)
4n−1 ∼
S = Sp(n)/Sp(n − 1).
 1.3. Principal Bundles 23

5. S n−1 is the set of all (x1 , . . . , xn ) ∈ Rn with (x1 )2 + · · · + (xn )2 = 1.

Let G = Z2 be the discrete subgroup {−1, 1} of the multiplicative group of
nonzero real numbers. Define σ : S n−1 × Z2 −→ S n−1 by σ(p, g) = p · g =
(x1 , . . . , xn ) · g = (x1 g, . . . , xn g). Then σ is a smooth right action of Z2 on
S n−1 , the orbits of which are pairs {p, −p} of antipodal points in S n−1 . The
orbit space (i.e., the quotient space obtained from S n−1 by identifying each
orbit to a point) is therefore R Pn−1 . Note that, when n = 2, this is just S 1
((1.2.7), [N4]).

6. Identify S 2n−1 with the set of all (z 1 , . . . , z n ) ∈ Cn satisfying |z 1 |2 + · · · +

|z n |2 = 1 and U (1) with the group of all complex numbers of
modulus 1. Define σ : S 2n−1 × U (1) −→ S 2n−1 by σ(p, g) = p · g =
(z 1 , . . . , z n ) · g = (z 1 g, . . . , z n g). Then σ is a smooth right action of U (1) on
S 2n−1 , the orbits of which are submanifolds of S 2n−1 diffeomorphic to U (1),
i.e., to S 1 (pages 51 and 260, [N4]). The orbit space is CPn−1 (pages 51–52,
[N4]). Note that, when n = 2, this is just S 2 ((1.2.8), [N4]).

7. Identify S 4n−1 with the set of all (q 1 , . . . , q n ) ∈ Hn satisfying |q 1 |2 +

· · · +|q n |2 = 1 and Sp(1) ∼ = SU (2) with the group of unit quaternions. Define
σ : S 4n−1 × Sp(1) −→ S 4n−1 by σ(p, g) = p · g = (q 1 , . . . , q n ) · g =
(q 1 g, . . . , q n g). Then σ is a smooth right action of Sp(1) on S 4n−1 , the or-
bits of which are submanifolds of S 4n−1 diffeomorphic to Sp(1), i.e., to S 3
(Exercise 1.2.4 and page 260, [N4]). The orbit space is HPn−1 (pages 51–52,
[N4]) and, when n = 2, this is just S 4 ((1.2.9), [N4]).

1.3 Principal Bundles

Let X be a differentiable manifold and G a Lie group. A smooth principal
bundle over X with structure group G (or, simply, a G-bundle over X)
consists of a differentiable manifold P , a smooth map P : P −→ X of P onto
X and a smooth right action σ : P × G −→ P, σ(p, g) = p · g, of G on P such
that the following conditions are satisfied:
1. σ preserves the fibers of P, i.e., P(p · g) = P(p) for all p ∈ P and all
g ∈ G, and
2. (Local Triviality) For each x0 ∈ X there exists an open set V in X
containing x0 and a diffeomorphism Ψ : P −1 (V ) −→ V × G of the form

Ψ(p) = (P(p), ψ(p)),

where ψ : P −1 (V ) −→ G satisfies

ψ (p · g) = ψ(p)g

for all p ∈ P −1 (V ) and all g ∈ G.

24 1. Geometrical Background

The pair (V, Ψ) is called a local

∪ trivialization of the G-bundle and a family
{(Vj , Ψj )}j∈J of such with j∈J Vj = X is called a trivializing cover of X.
Depending on how much of this structure is clear from the context we may
refer to P : P −→ X, or even just P itself, as a principal G-bundle over X and
P
indicate this diagrammatically by writing G ,→ P −→ X or G ,→ P −→ X.
For each p ∈ P , the fiber of P above P(p) coincides with the orbit of a p
under σ (Lemma 4.1.1, [N4]) and is a submanifold of P diffeomorphic to G
(page 260, [N4]). We begin by recording a few examples.
1. The trivial G-bundle over X consists of the product
manifold P = X × G, the projection P : X × G −→ X onto the first
factor and the action σ((x, h), g) = (x, h) · g = (x, hg). In this case one
takes V in #2 above to be all of X and Ψ to be the identity map on
P −1 (V ) = P −1 (X) = X × G.

2. Let P = S n−1 , G = Z2 = {−1, 1} and σ : S n−1 × Z2 −→ S n−1 the right

action σ(p, g) = p · g = (x1 , . . . , xn ) · g = (x1 g, . . . , xn g) described in Exam-
ple #5, page 27. The orbit space is X = R Pn−1 and we let P : S n−1 −→
R Pn−1 be the quotient map, i.e., P(p) = P(x1 , . . . , xn ) = [x1 , . . . , xn ] (see
Example #5, page 5). Then P(p·g) = P(±p) = [p] = P(p) for all p ∈ S n−1 and
g ∈ Z2 . For each k = 1, . . . , n, let Vk = {[p] = [x1 , . . . , xn ] ∈ R Pn−1 : xk ̸= 0}.
Then P −1 (Vk ) = {p = (x1 , . . . , xn ) ∈ S n−1 : xk ̸= 0} and we define
Ψk : P −1 (Vk ) −→ Vk × Z2 by Ψk (p) = Ψk (x1 , . . . , xn ) = ([p], xk /|xk |). Then
Ψk is a diffeomorphism of the form Ψk (p) = (P(p), ψk (p)), where ψk (p) =
ψk (x1 , . . . , xn ) = xk /|xk | and ψk (p · g) = (xk g)/|xk g| = (xk g)/|xk | = ψk (p)g.
Thus, {(Vk , Ψk )}k=1,...,n is a trivializing cover of R Pn−1 and
P
Z2 ,→ S n−1 −→ R Pn−1
is a principal Z2 -bundle over R Pn−1 .
Remark: Since P −1 (Vk ) is a disjoint union of two open hemispheres on
S n−1 (xk > 0 and xk < 0) each of which is mapped homeomorphically onto Vk
by P, P : S n−1 −→ R Pn−1 is actually a covering space (page 81, [N4]). This
is not true of the complex and quaternionic analogues to which we now turn.
3. Let P = S 2n−1 , G = U (1) and σ : S 2n−1 ×U (1) −→ S 2n−1 the right action
σ(p, g) = p · g = (z 1 , . . . , z n ) · g = (z 1 g, . . . , z n g) described in Example #6,
page 28. The orbit space is X = C Pn−1 and we let P : S 2n−1 −→ C Pn−1
be the quotient map. Then P(p · g) = P(p) for all p ∈ S 2n−1 and g ∈ U (1).
For each k = 1, . . . , n, let Vk = {[p] = [z 1 , . . . , z n ] ∈ C Pn−1 : z k ̸= 0} and
define Ψk : P −1 (Vk ) −→ Vk ×U (1) by Ψk (p) = Ψk (z 1 , . . . , z n ) = ([p], z k /|z k |).
Then, {(Vk , Ψk )}k=1,...,n is a trivializing cover of C Pn−1 so
P
U (1) ,→ S 2n−1 −→ C Pn−1

is a principal U (l)-bundle over C Pn−1 .

 1.3. Principal Bundles 25

Remark: When n = 2 one can identify C P1 with the 2-sphere (see the
Remark on page 6) and thereby obtain a U (1)-bundle over S 2 generally
known as the complex Hopf bundle. There are, in fact, two natural ways
of identifying C P1 with S 2 and these yield bundles which are not “equiv-
alent” in a sense soon to be made precise. Since we will require both of
these bundles we shall write out their descriptions explicitly. We use the
charts (U1 , φ1 ) and (U2 , φ2 ) on C P1 (Example #5, page 5) and the stereo-
graphic projection chart (US , φS ) on S 2 , (Example #4, page 4). On its do-
main, φ−1 S ◦ φ2 is a diffeomorphism into S given by φS ◦ φ2 ([z , z ]) =
2 −1 1 2
−1 1 2 −1 −1 z 1
φS (z (z ) ) = φS ( z2 ) = (z z̄ + z̄ z , −i z z̄ + i z̄ z , |z | − |z | ). But
1 2 1 2 1 2 1 2 1 2 2 2

this last formula does not require z 2 ̸= 0 and, in fact, defines a diffeomorphism
[z 1 , z 2 ] −→ (z 1 z̄ 2 + z̄ 1 z 2 , −i z 1 z̄ 2 +i z̄ 1 z 2 , |z 1 |2 − |z 2 |2 ) of all of C P1 onto S 2 .
Composing with the projection P of S 3 onto C P1 we obtain a map

P1 : S 3 −→ S 2

given by

P1 (z 1 , z 2 ) = (z 1 z̄ 2 + z̄ 1 z 2 , −i z 1 z̄ 2 + i z̄ 1 z 2 , |z 1 |2 − |z 2 |2 )
P
and this provides a concrete realization of the bundle U (1) ,→ S 3 −→ C P1
where C P1 is identified with S 2 :
P
U (1) ,→ S 3 −→
1
S2
( 2)
Similarly, φ−1
S ◦ φ1 ([z 1 , z 2 ]) = φ−1S
z
z1 = (z̄ 1 z 2 + z 1 z̄ 2 , −i z̄ 1 z 2 + i z 1 z̄ 2 ,
|z 1 |2 − |z 2 |2 ) actually determines a global diffeomorphism of C P1 onto S 2
and thus a map
P−1 : S 3 −→ S 2
given by

P−1 (z 1 , z 2 ) = (z̄ 1 z 2 + z 1 z̄ 2 , −i z̄ 1 z 2 + i z 1 z̄ 2 , |z 1 |2 − |z 2 |2 )

and therefore another bundle

P−1
U (1) ,→ S 3 −→ S 2 .

Notice that P−1 (z 1 , z 2 ) differs from P1 (z 1 , z 2 ) only in the sign of the second
coordinate so P−1 is P1 followed by reflection across the “xz-plane.”

4. Let P = S 4n−1 , G = Sp(1) and σ : S 4n−1 ×Sp(1) −→ S 4n−1 the right action
σ(p, g) = p · g = (q 1 , . . . , q n ) · g = (q 1 g, . . . , q n g) described in Example #7,
page 28. The orbit space is X = H Pn−1 and we let P : S 4n−1 −→ H Pn−1 be
the quotient map. Then P(p · g) = P(p) for all p ∈ S 4n−1 and g ∈ Sp(1). For
each k = 1, . . . , n, let Vk = {[p] = [q 1 , . . . , q n ] ∈ H Pn−1 : q k ̸= 0} and define
Ψk : P −1 (Vk ) −→ Vk × Sp(1) by Ψk (p) = Ψk (q 1 , . . . , q n ) = ([p], q k /|q k |).
26 1. Geometrical Background

Then, {(Vk , Ψk )}k=1,...,n is a trivializing cover of H Pn−1 so

P
Sp(1) ,→ S 4n−1 −→ H Pn−1

is a principal Sp(l)-bundle over H Pn−1 .

Remark: Just as in the complex case (page 30), when n = 2 one can identify
HP1 with S 4 in two natural ways and thereby obtain two quaternionic Hopf
bundles
P1 P−1
Sp(1) ,→ S 7 −→ S 4 and Sp(1) ,→ S 7 −→ S 4 .

5. Let P : P −→ X be any principal G-bundle over X (with right action

σ) and suppose X ′ is a submanifold of X. Let P ′ = P −1 (X ′ ), P ′ = P|P ′ and
P
σ ′ = σ|P ′ × G. For each local trivialization (V, Ψ) of G ,→ P −→ X with
V ∩ X ̸= ∅ set V = V ∩ X and Ψ = Ψ|(P ) (V ). Then P : P ′ −→ X ′ is
′ ′ ′ ′ ′ −1 ′ ′
P
a principal G-bundle over X ′ , called the restriction of G ,→ P −→ X to X ′ ,
P′
and each (V ′ , Ψ′ ) is a local trivialization for G ,→ P ′ −→ X ′ (Exercise 4.1.3,
[N4]).
P
Let G ,→ P −→ X be a principal G-bundle over X and fix a trivializing
cover {(Vj , Ψj )}j∈J of X. Write each Ψj as Ψj (p) = (P(p), ψj (p)) for all
p ∈ P −1 (Vj ), where ψj (p · g) = ψj (p)g for all p ∈ P −1 (Vj ) and g ∈ G. Suppose
i, j ∈ J and Vi ∩ Vj ̸= ∅. Then, for any x ∈ Vi ∩ Vj , ψj (p)(ψi (p))−1 takes the
same value for every p ∈ P −1 (x). Thus, we may define

gji : Vi ∩ Vj −→ G

by
gji (x) = (ψj (p)) (ψi (p))−1

for any p ∈ P −1 (x). These maps are smooth and are called the transition
P
functions for G ,→ P −→ X corresponding to {(Vj , Ψj )}j∈J . They satisfy all
of the following conditions:

gii (x) = e
gij (x) = (gji (x))−1
gkj (x)gji (x) = gki (x) (cocycle condition)

whenever i, j, k ∈ J and Vi ∩ Vj ∩ Vk ̸= ∅. A (local) cross-section

P
of G ,→ P −→ X defined on an open set V ⊆ X is a smooth map
s : V −→ P −1 (V ) that satisfies P ◦ s = idV , i.e., it is a smooth selec-
tion of an element from each fiber above V . A cross-section s on V gives
rise to a trivialization (V, Ψ), where Ψ : P −1 (V ) −→ V × G is given by
Ψ(s(x) · g) = (x, g) (page 221, [N4]). Conversely, a trivialization (V, Ψ)
 1.3. Principal Bundles 27

gives rise to a cross-section s : V −→ P −1 (V ) defined by s(x) = Ψ−1 (x, e)

and this correspondence between trivializations and cross-sections is bijective
(pages 220–221, [N4]).
As concrete illustrations of these last few notions we consider the bun-
P
dle Sp(1) ,→ S 7 −→ H P1 (the n = 2 case of Example #4, page 31) and
its trivializing cover {(Vk , Ψk )}k=1,2 . Then V1 = {x = [q 1 , q 2 ] ∈ H P1 :
q 1 ̸= 0}, V2 = {x = [q 1 , q 2 ] ∈ H P1 : q 2 ̸= 0}, ψ1 (p) = ψ1 (q 1 , q 2 ) = q 1 /|q 1 |
and ψ2 (q 1 , q 2 ) = q 2 /|q 2 | so the transition functions g12 , g21 : V1 ∩V2 −→ Sp(1)
are given by

g12 (x) = g12 ([q 1 , q 2 ]) = (q 1 /|q 1 |)(q 2 /|q 2 |)−1 , and

−1
g21 (x) = g21 ([q , q ]) = (q /|q |)(q /|q |)
1 2 2 2 1 1
.

One can check that the inverses Ψ−1 k : Vk × Sp(1) −→ P −1 (Vk ), k = 1, 2,

are given by Ψ−1 1 ([q 1 2
, q ], y) = (|q 1
|y, q 2 (q 1 /|q 1 |)−1 y) ∈ S 7 ⊆ H2 and
Ψ−1
2 ([q 1 2
, q ], y) = (q 1 2
(q /|q 2 −1
|) y, |q 2
|y) ∈ S 7 ⊆ H2 so the associated
cross-sections sk : Vk −→ P −1 (Vk ), k = 1, 2, are

s1 (x) = s1 ([q 1 , q 2 ]) = (|q 1 |, q 2 (q 1 /|q 1 |)−1 ), and

−1
s2 (x) = s2 ([q , q ]) = (q (q /|q |)
1 2 1 2 2
, |q |).
2

For future reference we note that V1 and V2 are also the standard coordinate
neighborhoods on H P1 (called U1 and U2 in Example #5, page 5) and that
the corresponding diffeomorphisms φk : Vk −→ H = R4 are

φ1 (x) = φ1 ([q 1 , q 2 ]) = q 2 (q 1 )−1 and

1 2 1 2 −1
φ2 (x) = φ2 ([q , q ]) = q (q ) .

Their inverses are φ−1 −1

1 (q) = [1, q] and φ2 (q) = [q, 1] so the overlap maps are

φ2 ◦ φ−1
1 (q) = q
−1
= φ1 ◦ φ−1
2 (q)

P
for all q ∈ H − {0}. All of this is the same for U (1) ,→ S 3 −→ C P1 .
P1 P2
Let G ,→ P1 −→ X1 and G ,→ P2 −→ X2 be two principal G-bundles and,
for convenience, denote the actions of G on P1 and P2 by the same dot · . A
(principal) bundle map from P1 to P2 is a smooth map f : P1 −→ P2 which
satisfies f (p·g) = f (p)·g for all p ∈ P1 and g ∈ G. Such an f carries each fiber
of P1 diffeomorphically onto some fiber of P2 and therefore induces a smooth
map f¯ : X1 −→ X2 defined by P2 ◦f = f¯◦P1 . If X1 = X2 = X, then a bundle
map f : P1 −→ P2 is called an equivalence if it is a diffeomorphism and in-
duces the identity map on X, i.e., f¯ = idX . In this case the bundles G ,→
P1 P2
P1 −→ X and G ,→ P2 −→ X are said to be equivalent. It then follows that
−1 P
f : P2 −→ P1 is also an equivalence. If G ,→ P −→ X is a single
28 1. Geometrical Background

principal G-bundle over X, then an equivalence f : P −→ P is called an

automorphism of the bundle. A principal G-bundle over X is said to be
trivial if it is equivalent to the trivial G-bundle G ,→ X × G −→ X
over X. This is the case if and only if the bundle has a global trivial-
ization (i.e., if and only if it is possible to take V = X in #2 of the def-
inition on page 28) and this, in turn, is the case if and only if it has a
global cross-section s : X −→ P . Determining whether or not a given
principal bundle is trivial (i.e., admits a global cross-section) is generally
not a simple matter. Any principal G-bundle over a contractible manifold
X (e.g., Rn ) is necessarily trivial. Notice finally that there is a difference be-
tween “trivial” and “trivialized”. The former asserts the existence of a global
cross-section, while the latter amounts to a specific choice of such a cross-
section/trivialization.
We record now the smooth versions of a number of basic results that are
proved in the topological context (i.e., for C 0 -principal bundles) in Chapter 4
of [N4]. Extending these to smooth (i.e., C ∞ ) principal bundles involves some
technical issues that will be discussed in Section 3.2.

The Reconstruction Theorem: Let X be a smooth manifold, G a Lie

group and {Vj }j∈J an open cover of X. Suppose that, for each i, j ∈ J with
Vi ∩ Vj ̸= ∅, there is given a smooth map gji : Vi ∩ Vj −→ G and that these
maps have the property that, if Vi ∩ Vj ∩ Vk ̸= ∅, then

gkj (x)gji (x) = gki (x)

for all x ∈ Vi ∩ Vj ∩ Vk . Then there exists a principal G-bundle over X, unique

up to equivalence, which has the Vj , j ∈ J, as trivializing neighborhoods and
the gji as corresponding transition functions.

The Classification Theorem: Let G be a connected Lie group. Then the set
of equivalence classes of principal G-bundles over S n , n ≥ 2, is in one-to-one
correspondence with the elements of the homotopy group πn−1 (G). Thus, for
example, the principal U (1)-bundles over S 2 are in one-to-one correspondence
with the elements of π1 (U (1)) ∼ = π1 (S 1 ) ∼
= Z (the integers). Similarly, the
4
Sp(l)-bundles over S are in one-to-one correspondence with the elements of
π3 (Sp (1)) ∼
= π3 (S 3 ) ∼
= Z. We will find that the theory of characteristic classes
provides a natural means of associating an integer (Chern number) with a
bundle of either of these types.
P
For any smooth principal bundle G ,→ P −→ X the fibers P −1 (x), x ∈ X,
are all submanifolds of P diffeomorphic to G (page 260, [N4]) so, for each
p ∈ P, Tp (P ) contains a subspace isomorphic to the Lie algebra G of G
(all tangent vectors at p to smooth curves in the fiber containing p). We call
this the vertical subspace of Tp (P ) and denote it Vertp (P ). The elements of
Vertp (P ) are called vertical vectors at p. The action σ of G on P provides a
 1.4. Connections and Curvature 29

natural means of identifying each Vertp (P ) with G. To see this, fix an element
A ∈ G (thought of as a set of matrices). We associate with A a vector field
A# on P , called the fundamental vector field on P determined by A, as
follows: For each p ∈ P the map σp : G −→ P defined by σp (g) = σ(p, g) = p·g
is smooth and so has a derivative (σp )∗id at the identity. Then,

d
A# (p) = (σp )∗id (A) = (p · exp(tA))|t=0
dt
(page 287, [N4]). The mapping A −→ A# (p) is an isomorphism of G onto
Vertp (P ) (Corollary 5.8.9, [N4]). Furthermore, for any A, B ∈ G,

[A, B]# = [A# , B # ]

(Theorem 5.8.8, [N4]) and, for all g ∈ G,

( )#
(σg )∗ (A# ) = adg−1 (A) .

1.4 Connections and Curvature

P
A connection (gauge field) on a principal bundle G ,→ P −→ X with
action σ is a smooth G-valued 1-form ω on P which satisfies
1. (σg )∗ ω = adg−1 ◦ ω for all g ∈ G, i.e., for all g ∈ G, p ∈ P and v ∈
Tp·g−1 (P ), ( )
ω p (σg )∗p·g−1 (v ) = g −1 ω p·g−1 (v )g,

2. ω(A# ) = A for all A ∈ G, i.e., for all g ∈ G and p ∈ P ,

ω p (A# (p)) = A.

A local cross-section s : V −→ P −1 (V ) of the bundle is called a

local gauge and the pullback A = s∗ ω of ω to V ⊆ X by s is called a
local gauge potential (in gauge s). If {(Vj , Ψj )}j∈J is a trivializing cover
of X and sj : Vj −→ P −1 (Vj ) is the cross-section associated with (Vj , Ψj )
(page 32), then the family {Aj = s∗j ω}j∈J of local gauge potentials satisfies

Aj = adg−1 ◦ Ai + gij ∗ Θ
ij

for all i, j ∈ J with Vi ∩ Vj ̸= ∅, where gij is the corresponding tran-

sition function and Θ is the Cartan 1-form for G (Lemma 5.9.2, [N4]).
Conversely, given a trivializing cover {(Vj , Ψj )}j∈J for some principal G-
P
bundle G ,→ P −→ X and a G-valued 1-form Aj on Vj for each j ∈ J
with Aj and Ai related in this way whenever Vi ∩ Vj ̸= ∅, there exists
30 1. Geometrical Background

a unique connection form ω on P such that Aj = s∗j ω for each j ∈ J

(Theorem 6.1.1, [N4]). The G-valued 1-forms gij ∗ Θ on Vi ∩ Vj are readily
computable when G is a matrix group for then one can show that, for any
x ∈ Vi ∩ Vj and any v ∈ Tx (X), (gij ∗ Θ)x (v ) = (gij (x))−1 dgij (x)(v ), where
dgij is the entrywise differential of gij : Vj ∩ Vi −→ G (pages 305, [N4]).
The relationship between the gauge potentials Aj and Ai can then be writ-
ten
Aj = gij −1 Ai gij + gij −1 dgij .

Given a connection form ω on P one defines, for each p ∈ P , the horizontal

subspace Horp (P ) of Tp (P ) by Horp (P ) = {v ∈ Tp (P ) : ω p (v ) = 0}. Then

Tp (P ) = Horp (P ) ⊕ Vertp (P )

((6.1.2), [N4]) so every v ∈ Tp (P ) can be written uniquely as v = v H + v V ,

where v H ∈ Horp (P ) and v V ∈ Vertp (P ). Similarly, a smooth vector
field V on P can be written V = V H + V V, where V H and V V are
smooth and, for each p ∈ P, V H (p) ∈ Horp (P ) and V V (p) ∈ Vertp (P )
(page 347, [N4]). If p ∈ P and P(p) = x, then P∗p carries Horp (P )
isomorphically onto Tx (X) (Exercise 6.1.7, [N4]). Furthermore, the hori-
zontal subspaces are invariant under the action of G on P in the sense
that
(σg )∗p (Horp (P )) = Horp·g (P )

for all p ∈ P and g ∈ G ((6.1.3), [N4]). If dim X = n, then the as-

signment p −→ Horp (P ) is an example of a smooth n-dimensional distri-
bution on P (page 333, [N4]) and, moreover, any smooth n-dimensional
distribution p −→ D (p) on P that satisfies Tp (P ) ∼ = D(p) ⊕ Vertp (P )
and (σg )∗p (D(p)) = D(p · g) for all p ∈ P and g ∈ G is the distri-
bution of horizontal subspaces for some connection form on P
P
(Exercise 6.1.9, [N4]). One often sees a connection on G ,→ P −→ X de-
fined in any one of the three equivalent ways described above (G-valued 1-
form, collection of local gauge potentials, or distribution of horizontal sub-
spaces).
P P
If G ,→ P1 −→ 1
X and G ,→ P2 −→ 2
X are two principal G-bundles over
P2
X, f : P1 −→ P2 is a bundle map and ω is a connection form on G ,→ P2 −→
P1
X, then f ∗ ω is a connection form on G ,→ P1 −→ X (Theorem 6.1.3, [N4]).
This is true, in particular, for an automorphism of a single principal bundle.
P
If ω 1 and ω 2 are two connection forms on G ,→ P −→ X, and if there exists
an automorphism f : P −→ P such that ω 2 = f ∗ ω 1 , then ω 1 and ω 2 are
said to be gauge equivalent (the origin of the terminology is in physics and
will be discussed shortly). Notice that, if s : V −→ P −1 (V ) is a cross-section
 1.4. Connections and Curvature 31

and Ai = s∗ ω i is the corresponding gauge potential of ω i , i = 1, 2, then

A2 = s∗ ω 2 = s∗ (f ∗ ω 1 ) = (f ◦ s)∗ ω 1 . Since s′ = f◦s : V −→ P −1 (V ) is also a
cross-section, we conclude that A1 = s∗ ω 1 and A2 = (s′ )∗ ω 1 are both gauge
potentials for the same connection ω 1 (by different cross-sections).
A connection also gives rise to a “path lifting procedure” from X to P
and thereby notions of “parallel translation” and “holonomy.” Specifically,
P
if ω is a connection form on G ,→ P −→ X and α : [0, 1] −→ X
is a smooth curve in X with α(0) = x0 , then for each p0 ∈ P −1 (x0 ) there
exists a unique smooth curve α̃p0 : [0, 1] −→ P such that
1. α̃p0 (0) = p0 ,
2. P ◦ α̃p0 (t) = α(t) for all t ∈ [0, 1], and
3. α̃p′ 0 (t) ∈ Horα̃p0 (t) (P ) for all t ∈ [0, 1]

(Theorem 6.1.4, [N4]). In particular, if α(1) = x1 , then α̃p0 (1) ∈ P −1 (x1 )

and we may define a map τα : P −1 (x0 ) −→ P −1 (x1 ), called the parallel
translation along α determined by ω, by τα (p0 ) = α̃p0 (1) for each p0 ∈
P −1 (x0 ). If x1 = x0 (i.e., if α is a loop in X), then τα : P −1 (x0 ) −→ P −1 (x0 ).
Since G acts transitively on the fibers of P (Lemma 4.1.1, [N4]), for each
p0 ∈ P −1 (x0 ) there is a unique g ∈ G such that τα (p0 ) = p0 ·g. Holding p0 fixed
and allowing α to vary over all smooth loops at x0 in X one obtains a subset
H(p0 ) of G consisting of all those g ∈ G such that p0 is parallel translated
to p0 · g over some smooth loop at x0 . H(p0 ) is, in fact, a subgroup of G
(Exercise 6.1.24, [N4] ) called the holonomy group of ω at p0 . We record
now a number of important examples of connections on principal bundles.
1. (Flat connections on trivial bundles) Let X be any smooth manifold, G
P
any matrix Lie group, and G ,→ X × G −→ X the corresponding trivial G-
bundle (Example #1, page 29). Let Θ be the Cartan 1-form on G and define
a G-valued 1-form ω on X × G by ω = π ∗ Θ, where π : X × G −→ G is the
projection onto G. Then ω is a connection form on X × G whose horizontal
subspace Hor(x,g) (X × G) at any (x, g) ∈ X × G is the tangent space to the
submanifold X × {g} at (x, g) (Exercise 6.2.12, [N4]).

Remark: The reason these connections are called “flat” will emerge when
we discuss the “curvature” of a connection.

2. (Natural connection on the complex Hopf bundle) We consider the U (l)-

bundle
P
U (1) ,→ S 3 −→ C P1

(the n = 2 case of Example #3, page 30). Regard S 3 as the submanifold of C2

consisting of those (z 1 , z 2 ) with |z 1 |2 + |z 2 |2 = 1 and identify the Lie algebra
of U (1) with the algebra Im C of pure imaginary complex numbers. Define
an Im C-valued 1-form ω̃ on C2 by ω̃ = i Im (z̄ 1 dz 1 + z̄ 2 dz 2 ). Thus, for each
32 1. Geometrical Background

p = (p1 , p2 ) ∈ C2 and v = (v 1 , v 2 ) ∈ Tp (C2 ) ∼

= Tp1 (C) ⊕ Tp2 (C) ∼
= C ⊕ C,
ω̃ p (v ) = i Im( p̄ v + p̄ v ). Now let ω be the restriction of ω̃ to S 3 (i.e.,
1 1 2 2

ω = ι∗ ω̃, where ι : S 3 ,→ C2 is the inclusion map). Then ω is a connec-

P
tion form on U (1) ,→ S 3 −→ CP1 (the proof is exactly the same as that
for the quaternionic case on pages 297–299, [N4] ). For each p = (p1 , p2 ) ∈
S 3 , Vertp (S 3 ) is the tangent space to the fiber of P containing p and this
is a 1-dimensional subspace of Tp (S 3 ) since this fiber is diffeomorphic to S 1 .
The horizontal subspace Horp (S 3 ) determined by ω is just that part of the
real orthogonal complement of Vertp (S 3 ) in C2 = R4 that lies in Tp (S 3 )
(the proof is identical to that in the quaternionic case on page 334, [N4] ).
One could compute gauge potentials s∗k ω, k = 1, 2, for the cross-sections
corresponding to the standard trivializations (Vk , Ψk ), k = 1, 2, and the re-
sults would be entirely analogous to those in the quaternionic case derived
on pages 297–303, [N4]. However, in order to facilitate comparison with the
physics (Dirac monopoles) to be discussed in the next section we record in-
stead the corresponding results when CP1 is identified with S 2 (see the Re-
P
mark following Example #3, page 30). First consider U (1) ,→ S 3 −→ 1
S 2 . Here
the diffeomorphism [z , z ] −→ (z z̄ + z̄ z , −iz z̄ + iz̄ z , |z | − |z 2 |2 )
1 2 1 2 1 2 1 2 1 2 1 2

of CP1 onto S 2 identifies V1 and V2 with UN and US , respectively, and

so it is on these sets that we have standard cross-sections sN : UN −→
P1−1 (UN ) and sS : US −→ P1−1 (US ). Our interest is in the gauge poten-
tials AN = s∗N ω and AS = s∗S ω and we wish to describe them in terms
of standard spherical coordinates ϕ and θ on S 2 (page 238, [N4]), i.e., we
want (sN ◦ φ−1 )∗ ω and (sS ◦ φ−1 )∗ ω, where φ is a spherical coordinate
chart on S 2 (φ−1 (ϕ, θ) = (sin ϕ cos θ, sin ϕ sin θ, cos ϕ)). These are computed
by writing ω = ι∗ ω̃ = ι∗ (i Im(z̄ 1 dz 1 + z̄ 2 dz 2 )) = i ι∗ (Im(z̄ 1 dz 1 + z̄ 2 dz 2 )) =
iι∗ (−x2 dx1 + x1 dx2 − x4 dx3 + x3 dx4 ), where z 1 = x1 + x2 i and z 2 = x3 + x4 i
so that, for example,

(sN ◦ φ−1 )∗ ω = i(ι ◦ sN ◦ φ−1 )∗ (−x2 dx1 + x1 dx2 − x4 dx3 + x3 dx4 ).

But

(ι ◦ sN ◦ φ−1 )(ϕ, θ) = ι ◦ sN ( sin ϕ cos θ, sin ϕ sin θ, cos ϕ )

( )
ϕ ϕ ϕ
= cos , 0, sin cos θ, − sin sin θ
2 2 2

(page 269, [N4]). A simple calculation with the coordinate formula for pullbacks
on page 9 gives 1
(sN ◦ φ−1 )∗ ω = − i(1 − cos ϕ) dθ
2
(page 270, [N4]). Similarly,
1
(sS ◦ φ−1 )∗ ω = i(1 + cos ϕ) dθ.
2
 1.4. Connections and Curvature 33

P−1
Analogous calculations for U (1) ,→ S 3 −→ S 2 give results that differ from
these only by a sign.
3. (Natural connection on the quaternionic Hopf bundle) We consider the
Sp(1)-bundle
P
Sp(1) ,→ S 7 −→ H P1
(the n = 2 case of Example #4, page 31). Regard S 7 as the submanifold of H2
consisting of those (q 1 , q 2 ) with |q 1 |2 +|q 2 |2 = 1 and identify the Lie algebra of
Sp(1) with the algebra Im H of pure imaginary quaternions. Define an Im H-
valued 1-form ω̃ on H2 by ω̃ = Im (q̄ 1 dq 1 + q̄ 2 dq 2 ) and let ω be the restriction
P
of ω̃ to S 7 . Then ω is a connection form on Sp (1) ,→ S 7 −→ HP1 ((5.9.10)
and (5.9.11), [N4]). For each p ∈ S 7 , Horp (S 7 ) is that part of the real or-
thogonal complement of Vertp (S 7 ) in H2 = R8 that lies in Tp (S 7 ) (page 335,
[N4] ). The standard trivializations {(V1 , Ψ1 ), (V2 , Ψ2 )} of the bundle give
cross-sections s1 : V1 −→ P −1 (V1 ) and s2 : V2 −→ P −1 (V2 ) and we are in-
terested in the gauge potentials A1 = s∗1 ω and A2 = s∗2 ω. Since V1 and V2
are also coordinate neighborhoods for the standard charts φ1 : V1 −→ H and
φ2 : V2 −→ H we may compute these gauge potentials A1 and A2 in terms
of φ1 and φ2 coordinates. The results are as follows (pages 297–301, [N4]):
( )
q̄
(s1 ◦ φ−1 ) ∗
ω = Im dq
1
1 + |q|2
and
( )
−1 ∗ q̄
(s2 ◦ φ2 ) ω = Im dq .
1 + |q|2

Thus, for any x ∈ V1 and any X ∈ Tx (H P1 ),

(s∗1 ω )x (X ) = ((s1 ◦ φ−1 ∗

1 )) ω)φ1 (x) ((φ1 )∗x X )
( )
φ1 (x)v
= Im ,
1 + |φ1 (x)|2

where v = dq ((φ1 )∗x (X )) and, for any x ∈ V2 and any X ∈ Tx (H P1 ),

( )
(s∗2 ω )x (X ) = (s2 ◦ φ−1 ∗
2 ) ω φ2 (x) ((φ2 )∗x X )
( )
φ2 (x)w
= Im ,
1 + |φ2 (x)|2

where w = dq ((φ2 )∗x (X )). Notice that, on H − {0},

( )
q̄ |q|2
Im dq = Im (q −1 dq).
1 + |q|2 1 + |q|2
34 1. Geometrical Background

To compare the two gauge potentials A1 and A2 one expresses both of them
in terms of the same coordinates on V1 ∩ V2 . For example, for any x ∈ V1 ∩ V2
and any X ∈ Tx (H P1 ),
( )
∗ φ1 (x)v̄
(s2 ω)x (X ) = Im ,
|φ1 (x)|2 (1 + |φ1 (x)|2 )

where v = dq((φ1 )∗x (X )) (pages 301–303, [N4]). Thus, on V1 ∩ V2 ,

( ) ( −1 )
−1 ∗ q 1
(s2 ◦ φ1 ) ω = Im dq̄ = Im q̄ dq̄ .
|q|2 (1 + |q|2 ) 1 + |q|2

The gauge potentials A1 = s∗1 ω and A2 = s∗2 ω arose first in the physics
literature [BPST] as solutions to the Yang-Mills equations (Section 6.3, [N4]).
There they were called pseudoparticles. Today it is more common to refer
to them (or the natural connection ω from which they arose and which they
uniquely determine) as instantons.
P
4. (More instantons on Sp(1) ,→ S 7 −→ H P1 ) Let ω denote the natural
P
connection on Sp(1) ,→ S 7 −→ H P1 described in the previous example. For
P
any bundle map f : S 7 −→ S 7 , f ∗ ω is also a connection on Sp(1) ,→ S 7 −→
H P (page 37). By judiciously selecting bundle maps f (pages 336–341, [N4] )
1

one can produce, for each pair (λ, n) ∈ (0, ∞) × H, a connection form ω λ,n
P
on Sp(1) ,→ S 7 −→ H P1 , uniquely determined by the two gauge potentials
( )
( )∗ q̄ − n̄
s2 ◦ φ−1 ω λ,n = Im dq
2
λ2 + |q − n|2
and ( )
( )∗ (|n|2 + λ2 ) q̄ − n
s1 ◦ φ−1 ω λ,n = Im dq .
1
λ2 |q|2 + |1 − nq|2

P
Remark: Any connection η on Sp(1) ,→ S 7 −→ H P1 is uniquely deter-
mined by either one of the gauge potentials s1∗ η or s2∗ η (page 339, [N4]) so
it is customary to specify ω λ,n by giving the single potential
( )
q̄ − n̄
Aλ,n = Im dq .
λ2 + |q − n|2

Aλ,n is called the generic BPST potential with center n and scale λ
(page 357, [N4]).
P
The curvature Ω of a connection ω on G ,→ P −→ X is its covariant
exterior derivative, defined by having d ω operate only on horizontal parts,
i.e., for each p ∈ P and all v, w ∈ Tp (P ) we let
 1.4. Connections and Curvature 35

Ω p (v, w) = (dω)p (vH , wH ).

The Cartan Structure Equation (Theorem 6.2.1, [N4]) asserts that

1
Ω = dω + [ω, ω],
2
where we use [ω, η] to denote the wedge product ω ∧ρ η (page 13) in which
ρ : G × G −→ G is the pairing given by the Lie bracket (ρ(A, B) = [A, B]).
This wedge product is easy to compute for matrix groups: First define ω ∧ η
to be the wedge product ω ∧ρ′ η, where ρ′ is the pairing given by matrix
multiplication (ρ′ (A, B) = AB). We will prove in Chapter 4 that ω ∧ η is just
the matrix of ordinary R-, C-, or H-valued 2-forms obtained by regarding ω
and η as matrices of 1-forms and forming their matrix product with entries
multiplied by the ordinary wedge product. Moreover, [ω, η] = ω ∧ η + η ∧ ω
so [ω, ω] = 2ω ∧ ω and
Ω = dω + ω ∧ ω.

If s : V −→ P −1 (V ) is a local cross-section, then the pullback s∗ Ω is called

the local field strength (in gauge s) and denoted F . Writing A = s∗ ω and
F = s∗ Ω the Cartan Structure Equation becomes

F = dA + A ∧ A

(page 350, [N4]). Assuming (as we may) that the domain V of s is also a co-
ordinate neighborhood for a chart (V, φ) with coordinate functions x1 , . . . , xn ,
we can write A = Aα dxα and F = 12 Fαβ dxα ∧ dxβ , where the Aα and Fαβ
are G-valued functions on V . Then

Fαβ = ∂α Aβ − ∂β Aα + [Aα , Aβ ],

where we have written ∂α for ∂x∂α and these derivatives are computed
componentwise in G (Exercise 6.2.9, [N4]). If sj : Vj −→ P −1 (Vj ) and
si : Vi −→ P −1 (Vi ) are two local cross-sections with Vj ∩ Vi ̸= ∅ and if
gij : Vj ∩ Vi −→ G is the transition function relating the corresponding triv-
ializations, then sj (x) = si (x) · gij (x) for each x ∈ Vj ∩ Vi (Exercise 4.3.5,
−1 −1
[N4]) and, whereas A j = gij A i gij + gij dgij , we have
−1
F j = gij F i gij .

We write out a few specific examples.

P
1. (Flat connections on trivial bundles) We let G ,→ X × G −→ X be a
trivial bundle, Θ the Cartan 1-form on G and π : X ×G −→ G the projection.
As in Example #1, page 38, ω = π ∗ Θ is a connection form on X × G. The
Maurer-Cartan equations (page 19) yield (page 329, [N4]) the equation of
36 1. Geometrical Background

structure for G, i.e.,

1
dΘ + [Θ, Θ] = 0,
2
and from this it follows that
1
dω + [ω, ω] = 0
2
(page 353, [N4]). Thus, flat connections ω are “flat” because their curvature
forms Ω are identically zero.

2. (Natural connection on the complex Hopf bundle) Connections ω on

P
U (l)-bundles U (1) ,→ P −→ X have a number of very special properties and
we begin with a general discussion of a few of these. We identify the Lie algebra
u(1) with the algebra Im C of pure imaginary complex numbers. Since u(1)
is 1-dimensional, all brackets vanish so the curvature Ω of any connection ω
coincides with its exterior derivative:

Ω = dω.
−1
If s : V −→ P (V ) is any cross-section, then we can write the gauge potential
A = s∗ ω and field strength F = s∗ Ω as

A = −i A
F = d A = −i dA = −i F ,

where A and F are real-valued forms on V (the minus signs are conventional).
If sj : Vj −→ P −1 (Vj ) and si : Vi −→ P −1 (Vi ) are two local cross-sections
with Vj ∩ Vi ̸= ∅ and if gij : Vj ∩ Vi −→ U (1) is the corresponding transition
function, then
−1 −1 −1
Aj = gij Ai gij + gij dgij = Ai + gij dgij

and
−1
F j = gij F i gij = F i
on Vj ∩ Vi because U (1) is Abelian. In particular, the local field strengths,
since they agree on any intersections of their domains, piece together to give
a globally defined field strength 2-form F on X. This is a peculiarity of Abelian
gauge fields and generally is not true in the non-Abelian case. Also note that
since gij maps into U (1) it can be written as gij (x) = e−i Λ(x) for some real-
−1
valued function Λ on Vj ∩ Vi . Then gij dgij = ei Λ e−i Λ (−i dΛ) = −i dΛ so
Aj = Ai − i dΛ, i.e.,
Aj = Ai + dΛ,
which is the traditional form for the relationship between two “vector
potentials.”
 1.4. Connections and Curvature 37

Now we return to the case of the natural connection ω on the complex

P
Hopf bundle U (1) ,→ S 3 −→ CP1 . Identifying CP1 with S 2 as indicated on
P
page 39 we obtain U (1) ,→ S 3 −→1
S 2 and local gauge potentials AN and AS
which, in spherical coordinates, are given by
1
AN = − i(1 − cos ϕ)dθ
2
1
AS = i(1 + cos ϕ)dθ.
2
Each has the property that, on its domain, the field strength is given by
1
F = − i sin ϕ dϕ ∧ dθ
2
(just compute d AN and d As).

Remarks: The real 1-forms AN and AS defined by AN = −i AN and

AS = −i AS are the vector potentials for a Dirac magnetic monopole of
strength g = 12 (pages 19–20, [N4]) and the real 2-form F defined by F = −iF
is the corresponding magnetic field strength. For any integer n one can define
Im C-valued 1-forms on UN and US by
1
AN = − ni(1 − cos ϕ)dθ, and
2
1
AS = ni(1 + cos ϕ)dθ.
2
We will investigate these 1-forms in some detail a bit later. For the time
being we point out only that, while they are, indeed, gauge potentials for
a connection whose field strength represents a Dirac monopole of strength
n/2, this connection does not live on the Hopf bundle (unless n = 1), but
on some other U (1)-bundle over S 2 . That this must be the case is clear from
AN = AS −nidθ = enθi AS e−nθi +enθi de−nθi , whereas the transition function
P1
for U (1) ,→ S 3 −→ S 2 is gSN (ϕ, θ) = e−θi ((0.3.8), [N4]). We will eventually
show that the field strength 2-form F defines what is called the 1st Chern class
of the bundle on which AN and AS determine a connection and that this
characteristic class completely determines the principal U (l)-bundles over S 2 .
We will find also that the integral of this Chern class over S 2 is the magnetic
charge n of the monopole.

3. (Instantons on the quaternionic Hopf bundle) We consider the Sp (1)-

P
bundle Sp (1) ,→ S 7 −→ HP1 and, for each λ > 0 and n ∈ H, the connection
38 1. Geometrical Background

ω λ,n on it determined by the potential

( )
q̄ − n̄
Aλ,n = Im dq
λ2 + |q − n|2

(see the Remark on page 42). A calculation (pages 327–328, [N4]) gives

1 λ2
F λ,n = d Aλ,n + [Aλ,n , Aλ,n ] = 2 d q̄ ∧ dq
2 (λ + |q − n|2 )2

on H. In Chapter 4 we will find that a natural inner product on Im H, together

with the usual metric on H = R4 , gives rise to a norm for field strengths F
such as these. We will then define the Yang-Mills action of such a potential
A by ∫
1
YM (A) = ∥F ∥2
2 R4
and calculate to show that, for any λ > 0 and n ∈ H,

YM (Aλ,n ) = 8π 2 .
P
Remarks: All of the connections ω λ,n on Sp(1) ,→ S 7 −→ HP1 give rise to
potentials Aλ,n with the same Yang-Mills action YM(Aλ,n ) and this, we will
find, is no accident. The number 8π1 2 YM(Aλ,n ) is, in fact, a topological char-
acteristic of the quaternionic Hopf bundle, not unlike the Euler characteristic
of a surface. We will eventually show that this number is essentially the inte-
gral over HP1 of what is called the 2nd Chern class of the bundle. Like the
magnetic charge of a Dirac monopole, the Yang-Mills action for an instanton
can be thought of as a sort of “topological charge.”

1.5 Associated Bundles and Matter Fields

P
Let G ,→ P −→ X be a smooth principal G-bundle over X with right action
σ : P × G −→ P, σ(p, g) = p · g. Now let F be a smooth manifold on which
G acts smoothly on the left (the image of (g, ξ) ∈ G × F under this action
will be written g · ξ). Then ((p, ξ), g) −→ (p, ξ) · g = (p · g, g −1 · ξ) is a smooth
right action of G on P × F (Exercise 6.7.1, [N4]). We denote by P ×G F the
orbit space of P × F by this action. More precisely, we define an equivalance
relation ∼ on P × F as follows: (p1 , ξ1 ) ∼ (p2 , ξ2 ) if and only if there exists a
g ∈ G such that (p2 , ξ2 ) = (p1 , ξ1 ) · g. The equivalence class containing (p, ξ)
is [p, ξ] = {(p · g, g −1 · ξ) : g ∈ G}. Then, as a set, P ×G F = {[p, ξ] : (p, ξ) ∈
P × F } and we provide P ×G F with the quotient topology determined by
Q : P × F −→ P ×G F, Q(p, ξ) = [p, ξ]. Also define PG : P ×G F −→ X
−1
by PG ([p, ξ]) = P(p). Then PG is continuous and, for any x ∈ X, PG (x) =
−1
{[p, ξ] : ξ ∈ F }, where p is any point in P (x) (Exercise 6.7.2, [N4]). If (V, Ψ)
 1.5. Associated Bundles and Matter Fields 39

P
is any local trivialization of G ,→ P −→ X and s : V −→ P −1 (V ) is the
associated cross-section, then the map Φ̃ : V × F −→ PG−1 (V ) defined by
−1
Φ̃(x, ξ) = [s(x), ξ] is a homeomorphism with inverse Ψ̃ : PG (V ) −→ V × F
given by Ψ̃([s(x), ξ]) = (x, ξ) (page 381, [N4]). If (Vi , Ψi ) and (Vj , Ψj ) are two
such trivializations with Vi ∩Vj ̸= ∅ and gji : Vi ∩Vj −→ G is the corresponding
transition function, then Ψ̃j ◦ Ψ̃−1 i : (Vi ∩ Vj ) × F −→ (Vi ∩ Vj ) × F is given by
Ψ̃j ◦ Ψ̃−1
i (x, ξ) = (x, g ji (x) · ξ) and so is a diffeomorphism (page 382, [N4]).
It follows that there is a unique differentiable structure on P ×G F relative to
−1
which each Ψ̃ : PG (V ) −→ V × F is a diffeomorphism and that, relative to
this structure, PG : P ×G F −→ X is smooth. We call

PG : P ×G F −→ X
P
the fiber bundle associated with G ,→ P −→ X by the given left action
of G on F .
The special case of most interest to us arises as follows: Let F = V be a
finite dimensional vector space (with its natural differentiable structure) and
ρ : G −→ GL(V) a smooth representation of G on V. Then ρ gives rise to a
smooth left action of G on V ((g, v) −→ g · v = (ρ(g))(v)). The fiber bundle
P
associated with G ,→ P −→ X by this action is denoted

Pρ : P ×ρ V −→ X
P
and called the vector bundle associated with G ,→ P −→ X by
the representation ρ. In this case each fiber Pρ−1 (x) = {[p, v] : v ∈ V},
where p is any point in P −1 (x), is a copy of V and admits a natural vector
space structure: a1 [p, v1 ] + a2 [p, v2 ] = [p, a1 v1 + a2 v2 ] for all a1 , a2 ∈ R and
v1 , v2 ∈ V. We record a few examples of particular interest.
P
1. Let U (1) ,→ P −→ X be an arbitrary principal U (1)-bundle and
take V = C (as a 2-dimensional real vector space). If ρ : U (1) −→ GL(C)
is any representation of U (1) on C, then the associated vector bundle
Pρ : P ×ρ C −→ X has fibers that are copies of C and is called a
complex line bundle over X. An obvious choice for ρ : U (1) −→ GL(C)
is obtained by taking (ρ(g))(z) = gz for each g ∈ U (1) and z ∈ C (if
g = ei θ , 0 ≤ θ < 2π, then ρ(g) is rotation by θ). More generally, one can define,
for any integer n, a representation ρ : U (1) −→ GL(C) by (ρ(g))(z) = g n z
and thereby an associated vector bundle over X.

P
2. Let Sp(1) ,→ P −→ X be any principal Sp(1)-bundle and take V = H
(as a 4-dimensional real vector space). Then any representation ρ : Sp(1) −→
GL(H) (e.g., (ρ(g))(q) = gq, or (ρ(g))(q) = g n q for some integer n) defines a
vector bundle P ×ρ H with fibers isomorphic to H and called a quaternionic
line bundle over X.
40 1. Geometrical Background

P
3. Let SU (2) ,→ P −→ X be any principal SU (2)-bundle and take
V = C2 (as a 4-dimensional real vector space). Then any representation
ρ : SU (2) −→ GL(C2 ) defines an associated vector bundle P ×ρ C2 . A par-
ticularly useful choice (for )ρ is obtained as (
follows
) (we write the elements of
C as column vectors ξ2 ): For each g = γ δ ∈ SU (2) define
2 ξ1 αβ

( ) ( ) ( )  ( )
ξ1 ξ1 α β ξ1 αξ 1 + βξ 2
(ρ(g)) =g =  = .
ξ2 ξ2 γ δ ξ2 γξ 1 + δξ 2

1
This representation of SU (2) is generally denoted D 2 and we will find (in
Section 2.4) that it arises naturally in Pauli’s nonrelativistic theory of the
electron.
P
4. Let G be an arbitrary matrix Lie group and G ,→ P −→ X be an ar-
bitrary principal G-bundle. The adjoint representation ad : G −→ GL(G)
assigns to each g ∈ G the nonsingular linear transformation adg on the
Lie algebra G defined by
adg (A) = g A g −1
P
(page 24). The vector bundle associated with G ,→ P −→ X by ad is called
P
the adjoint bundle of G ,→ P −→ X and denoted

ad P = P ×ad G.

The fibers of ad P are copies of the Lie algebra G of G.

Remark: The significance of the adjoint bundle, which is considerable, will
emerge somewhat later. Roughly, the situation is as follows: The curvature Ω
P
of a connection on G ,→ P −→ X is a G-valued 2-form that is globally defined
on P . However, the field strengths F = s∗ Ω on X generally do not piece to-
gether into a globally defined G-valued 2-form on X (page 44). Physically, this
is unfortunate since one would like field strengths to live naturally on space
(or spacetime) and not on some abstract space of internal states. Mathemati-
cally, it is inconvenient since connections on bundles and their curvatures are
often only auxiliary devices for studying the topology and geometry of the
base manifold (see, for example, the discussion of Donaldson’s Theorem in
Appendix B of [N4]). In any case, we will find that by allowing 2-forms to
take values, not in a single, fixed copy of G, but in the “parametrized family
of G’s” that the adjoint bundle represents, we will be able to uniquely repre-
sent the field strength as a globally defined ad P -valued 2-form on X. More
generally, we will find that many locally defined V-valued objects of interest
(e.g., wavefunctions) become globally defined when thought of as taking their
values in some associated vector bundle.
 1.5. Associated Bundles and Matter Fields 41

P
Now, let G ,→ P −→ X be an arbitrary principal G-bundle, F a smooth
manifold on which G acts on the left and PG : P ×G F −→ X the as-
sociated fiber bundle. If V is an open subset of X, then a smooth map
ϕ : P −1 (V ) −→ F is said to be equivariant (with respect to the given actions
of G on P and F ) if
ϕ(p · g) = g −1 · ϕ(p)

for all p ∈ P −1 (V ) and g ∈ G. Given such a map one defines

−1
sϕ :V −→ PG (V ) by
sϕ (x) = [p, ϕ(p)],

where p is any point in P −1 (x). Then sϕ is smooth and satisfies

−1
PG ◦ sϕ = idV . Conversely, suppose s : V −→ PG (V ) is a smooth map
satisfying PG ◦ s = idV (called a local cross-section of the fiber bundle
PG : P ×G F −→ X). Define ϕs : P −1 (V ) −→ F as follows: Let p ∈ P −1 (V ).
−1
Then P(p) = x is in V so s(x) is in PG (V ) and there exists a unique element
ϕs (p) ∈ F such that
s(x) = [p, ϕs (p)].
Then ϕs is a smooth equivariant map (page 384, [N4]). Moreover, this cor-
respondence between equivariant maps ϕ : P −1 (V ) −→ F and cross-sections
−1
s : V −→ PG (V ) of the associated fiber bundle is one-to-one and onto
(Exercise 6.8.4, [N4]). All of this applies, in particular, to the special case of
P
a vector bundle Pρ : P ×ρ V −→ X associated to G ,→ P −→ X by some
representation ρ : G −→ GL(V) of G on V.
Remarks: Our interest in equivariant maps arises from the following con-
P
siderations. A connection ω on a principal bundle G ,→ P −→ X is what the
physicists call a gauge field and is thought of as something akin to a modern
version of the Newtonian concept of a “force” in that particles “respond” to
it by experiencing changes in their internal states. The particles coupled to
(i.e., experiencing the effects of) the gauge field have wavefunctions ψ on X
taking values in some real vector space V (e.g., C, C2 etc.) that are obtained
by solving various partial differential equations involving the gauge potentials
A = s∗ ω. Since the gauge potentials A are, in general, only locally defined
on X, the same is true of these wavefunctions. A change of gauge (s −→ s · g)
gives rise to a new potential Ag = adg−1 ◦ A + g ∗ Θ, a new field strength
F g = adg−1 ◦ F and thereby a new wavefunction ψ g = (ρ(g −1 ))ψ = g −1 · ψ,
where ρ is some representation of G on V that is characteristic of the partic-
ular class of particles under consideration. Now, we can identify ψ (in gauge
s) with a local cross-section of the associated vector bundle P ×ρ V:

x −→ [s(x), ψ(x)].
42 1. Geometrical Background

Then, in gauge s · g,

x −→ [s(x) · g(x), g(x)−1 · ψ(x)] = [s(x), ψ(x)]

so that the locally defined wavefunctions piece together to determine a globally

defined cross-section of P ×ρ V. But a globally defined cross-section of P ×ρ V
correspondings to a globally defined equivariant map ϕ : P −→ V. As V-
valued maps on X, the wavefunctions are only locally defined, but all of these
local wavefunctions on X can be uniquely represented by a single, globally
defined V-valued map on P (or a single, globally defined cross-section of the
associated vector bundle). With this as motivation we introduce the following
definitions.
P
Let G ,→ P −→ X be a principal G-bundle, V a real vector space and
ρ : G −→ GL(V) a representation of G on V. An equivariant, V-valued map
P
ϕ : P −→ V on P is called a matter field (of type ρ) on G ,→ P −→ X. If
V = C, then ϕ is called a complex scalar field. If V = G and ρ = ad is the
adjoint representation, then ϕ is called a Higgs field.
A matter field ϕ is, in particular, a V-valued 0-form on P and so has an
P
exterior derivative dϕ. Assuming now that G ,→ P −→ X has defined on it a
connection form ω, we define the covariant exterior derivative dω ϕ of ϕ
by having dϕ act only on horizontal parts: For each p ∈ P and v ∈ Tp (P ),

(dω ϕ)p (v) = (dϕ)p (v H ).

This is a V-valued 1-form on P (Exercise 6.8.5, [N4]) and satisfies

σg∗ (dω ϕ) = g −1 · dω ϕ

for each g ∈ G ((6.8.1), [N4]). These are the derivatives that appear in the
field equations describing the quantitative response of the particle to the gauge
field (pages 391–392, [N4]). A computational formula analogous to the Cartan
Structure Equation for curvature Ω (which is the covariant exterior derivative
of the connection form ω) is obtained as follows: For any A ∈ G and v ∈ V we
define A · v ∈ V by
d d
A·v = (exp(tA) · v)|t=0 = (ρ(exp(tA))(v))|t=0 .
dt dt
Remarks: Two special cases are worth pointing out immediately. If G is
a group of n × n matrices (with entries in F = R, C, or H), V = Fn
(thought of as column matrices) and ρ is the natural representation of G
on V (matrix multiplication), then, identifying G with an algebra of matrices,
A · v = Av (matrix multiplication). On the other hand, if V = G and ρ = ad,
then, for all A, B ∈ G, A · B = [A, B] (see the proof of Theorem 5.8.8,
[N4]).
 1.5. Associated Bundles and Matter Fields 43

Now, if ϕ is a V-valued 0-form on P and ω is a G-valued 1-form on P we can

define a V-valued 1-form ω · ϕ on P by

(ω · ϕ)p (v) = ω p (v) · ϕ(p)

for each p ∈ P and v ∈ Tp (P ). Then

dω ϕ = dϕ + ω · ϕ

((6.8.4), [N4]). We conclude by writing out a few concrete examples in local

coordinates (details are available on pages 388–391, [N4]).
P
1. Let U (1) ,→ P −→ X be a principal U (1)-bundle with connection ω
and let V = C (as a 2-dimensional real vector space). For each integer q we
let ρq : U (1) −→ GL(C) be the representation ρq (g)(z) = g q z. For simplic-
ity, we will write ϕ = ϕ(x1 , . . . , xn ) for a local coordinate expression of the
pullback by some cross-section of a matter field. Similarly the gauge poten-
tial will be written locally as A = A(x1 , . . . , xn ) = Aα (x1 , . . . , xn )dxα =
−iAα (x1 , . . . , xn )dxα . The corresponding local coordinate expression for the
pullback of dω ϕ is given by

(∂α ϕ + qAα ϕ)dxα = (∂α − iqAα )ϕdxα ,

∂
where ∂α = ∂xα .
P
2. Let SU (2) ,→ P −→ X be a principal SU (2)-bundle with connection
ω and let V = C2 (as a 4-dimensional real vector space). For each ( 1 inte-
)
ger q we let ρq : SU (2) −→ GL(C2 ) be the representation ρq (g) zz2 =
( 1) ( 1) ( 1 1 )
,...,xn )
g q zz2 . Writing ϕ = ϕϕ2 = ϕϕ2 (x 1 n
(x ,...,x )
for the local coordinate ex-
pression of the pullback by some cross-section of a matter field and A =
Aα (x1 , . . . , xn )dxα = −iBα (x1 , . . . , xn )dxα for the local gauge potential
(where the Aα are skew-Hermitian and tracefree, while the Bα are Hermi-
tian and tracefree; Exercise 6.8.5, [N4]), we have
( ( ) ( )) ( )
ϕ1 ϕ1 ϕ1
∂α + qAα dx α
= (∂ α − iqBα ) dxα
ϕ2 ϕ2 ϕ2

for the corresponding local coordinate expression for the pullback of dω ϕ.

2
Physical Motivation

2.1 General Framework for Classical

Gauge Theories
Our objective now is to use the machinery assembled in the previous chapter,
and that to be developed in subsequent chapters, to study a number of rather
specific “classical gauge theories” arising in modern physics. The discussion is
heuristic and informal and its intention is to indicate how the topology and
geometry that are our real concern here arise naturally in meaningful physics.
In order to have a context within which to place these examples we will devote
this rather brief section to an explicit enumeration of the basic mathematical
ingredients required to describe, at the classical level, the interaction of a
particle with a gauge field.

1. A smooth, oriented, (semi-) Riemannian manifold X.

Generally, this will be space (R3 ), a spacetime (e.g., R1,3 ; see Section 2.2), a
Euclidean (“Wick rotated”) version of a spacetime (e.g., R4 ), a compactifica-
tion of one of these (e.g., S 4 = R4 ∪ {∞}), or an open submanifold of one of
these. The particles “live” in X.

2. A finite dimensional vector space V.

The particles have wavefunctions that take values in V. The choice of V is
dictated by the internal structure of the particle (e.g., phase, isospin, spin,
etc.) and so V is called the internal space. Typical examples are C, C2 , C4 ,
or the Lie algebra of some Lie group, e.g., u(1), or su(2). V is equipped with
an inner product ⟨ , ⟩ by which one computes squared norms ∥ψ∥2 and thereby
the probabilities with which quantum mechanics deals.

3. A matrix Lie group G and a representation

ρ : G −→ GL(V)

of G on V that is orthogonal with respect to the inner product on V:

⟨ ρ(g)(v), ρ(g)(w) ⟩ = ⟨ v, w ⟩.
This will generally be one of the classical groups (e.g., U (1), SU (2)) or a
product of these and plays a dual role.

G.L. Naber, Topology, Geometry and Gauge fields: Interactions, 45

Applied Mathematical Sciences 141, DOI 10.1007/978-1-4419-7895-0_2,
© Springer Science+Business Media, LLC 2011
46 2. Physical Motivation

(i) The inner product on V determines a class of orthonormal bases, or

“frames” (e.g., isospin axes) and these are related by the elements of G,
i.e., g ∈ G “acts” on a frame p to give a new frame p ·g. By fixing some
frame at the outset (the elements of which will correspond to certain
“states” of the particle) one can identify the elements of G with the
frames.
(ii) G also acts on V via the representation ρ(v −→ ρ(g)(v) = g · v) and
so acts on the wavefunction ψ at each point. If ψ(p) is a value of the
wavefunction, described relative to the frame p, then

ψ (p · g) = g −1 · ψ (p)

is its description relative to the new frame p · g.

4. A smooth principal G-bundle over X:
P
G ,→ P −→ X.

Typical examples are trivial bundles (e.g., SU (2) ,→ R4 × SU (2) −→ R4 )

and Hopf bundles (e.g., U (1) ,→ S 3 −→ S 2 , SU (2) ,→ S 7 −→ S 4 ). At each
x ∈ X the fiber P −1 (x) is a copy of G thought of as the set of all frames in the
internal space V at x. A local cross-section s : V −→ P is a smooth selection
of a frame at each point in some open subset V of X (a local “gauge”) relative
to which wavefunctions can be described on V .
P
5. A connection ω on G ,→ P −→ X with curvature Ω.
If s : V −→ P is a local cross-section (local gauge), then the pullback A = s∗ ω
is the local gauge potential and F = s∗ Ω is the local field strength. Gen-
erally, these exist only locally since nontrivial principal bundles do not admit
global cross-sections and pullbacks by different local cross-sections usually
do not agree on the intersection of their domains. Particles coupled to (i.e.,
experiencing the effects of) the field determined by ω have locally defined
wavefunctions ψ taking values in V that are obtained by solving equations of
motion (see #8 below) involving the local potentials A. A change of gauge
(s −→ s · g) changes the wavefunction by the representation ρ (ψ −→ g −1 · ψ).
These local wavefunctions piece together into a globally defined object called
a matter field which can be described in two equivalent ways:
6. A global cross-section of the vector bundle P ×ρ V associated to
P
G ,→ P −→ X by ρ (equivalently, a V-valued map ϕ : P −→ V on P
that is equivariant: ϕ(p · g) = g −1 · ϕ(p)).
Physically, such a matter field has potential energy which we describe with
7. A non-negative, smooth, real-valued function

U : V −→ R
 2.1. General Framework for Classical Gauge Theories 47

that is invariant under the action of G on V:

U (g · v) = U (v).

U is to be regarded as a potential function with U ◦ ϕ describing the self-

interaction energy of the matter field ϕ. Typically, this will depend only on
∥ ϕ ∥2 , e.g., 12 m ∥ϕ ∥2 , or λ8 (∥ϕ ∥2 − 1)2 , where m and λ are non-negative con-
stants.
8. An action (energy) functional A(ω, ϕ), the stationary points of which are
the physically significant field configurations (ω, ϕ).
Typically, this functional is of the following general form:
∫ [ ]
A (ω, ϕ) = c ∥Fω ∥2 + c1 ∥ dω ϕ ∥2 + c2 U ◦ ϕ .
X

We will spell out in detail what each of these terms means in the concrete ex-
amples to follow. Briefly, c is a normalizing constant, c 1 and c 2 are “coupling
constants,” Fω is a global 2-form on X with values in the ad-joint bundle ad P
which locally pulls back to the gauge field strengths F , dω ϕ is the covariant
exterior derivative of the matter field ϕ (thought of as a cross-section of the
associated vector bundle) and the norms arise from the metric on X and the
Killing form on the Lie algebra G of G. Integrals of such objects over mani-
folds like X will be introduced in Chapter 4. The physically interesting field
configurations (ω, ϕ), are those which (at least locally) minimize the value
of the action functional. The Calculus of Variations provides necessary con-
ditions (the Euler-Lagrange differential equations) that must be satisfied by
such minima. The Euler-Lagrange equations for the action A(ω, ϕ), are the
appropriate field equations (the “equations of motion”) of our gauge theory.
One can, of course, generalize the model we have described by including more
than one matter field.
From the point-of-view of physics, one is generally interested only in finite
action configurations (ω, ϕ), i.e., those for which

A (ω, ϕ) < ∞.

This is assured if X is compact, but otherwise one must assume some sort of
appropriate asymptotic behavior for the terms in the integrand. Such asymp-
totic conditions turn out to have profound topological consequences (e.g., the
existence of “topological charge”) and investigating this link between topology
and asymptotics is one of our primary objectives.
These then are the basic ingredients required to build a classical gauge
theory. There is a special case that has gotten a great deal of attention, par-
ticularly in the mathematical community. When c1 = c2 = 0 in the action
A (ω, ϕ), (so that, effectively, there are no matter fields present), then one can
think of A as depending only on ω and, in this case, it is referred to as the
48 2. Physical Motivation

Yang-Mills action and written

∫
YM(ω) = c ∥ Fω ∥2 .
X

The Euler-Lagrange equations for YM are called the Yang-Mills equations

and can be written
d ω ∗ Fω = 0,
where ∗ Fω is the Hodge dual of Fω and d ω is the covariant exterior derivative.
Remark: We have not yet defined the Hodge dual or the covariant exterior
derivative in sufficient generality to cover the context in which we now find
ourselves, but the generalization is easy and will be provided in Chapter 4.
We will also find that, quite independently of the action, the field Fω also
satisfies a purely geometrical constraint known as the Bianchi identity

d ω Fω = 0.

These last two equations lie at the heart of what is called pure Yang-Mills
theory. The impact of this subject on low dimensional topology is discussed
at some length in [N4]. Although this special case may not appear to be in
the spirit of our announced intention here to model interactions we will find
that it leads (through a process known as “dimensional reduction”) directly to
the particular interactions of most interest to us in Section 2.5.

2.2 Electromagnetic Fields

The prototypical example of a gauge theory is classical electrodynamics. Al-
though our real interests lie elsewhere, this example will provide a nice warm-
up in familiar territory and so we will describe it in some detail. Keep in
mind that our intention in this chapter is primarily motivational so we will
feel free to adopt a rather casual attitude, occasionally anticipating concepts
and results that are introduced carefully only later in the text.
The arena within which electrodynamics is done is Minkowski space-
time R1,3 (assuming gravitational effects are neglected). As a differentiable
manifold, R1,3 is just R4 . Rather than the usual Riemannian metric on R4 ,
however, we define on R1,3 the semi-Riemannian metric given, relative to
standard coordinates x0 , x1 , x2 , x3 on R4 by ηαβ dxα ⊗ dxβ , where



 1, α=β=0
ηαβ = −1, α = β = 1, 2, 3.


 0, α ̸= β
 2.2. Electromagnetic Fields 49

Remark: For the moment we will require very little of the geometry of R1,3
and its physical significance. Simply think of the elements of R1,3 as “events”
whose standard coordinates represent the time (x0 ) and spatial (x1 , x2 , x3 ) co-
ordinates by which the event is identified in some fixed, but arbitrary inertial
frame of reference. The entire history of a (point) object can then be identified
with a continuous sequence of events (i.e., a curve) in R1,3 called its “world-
line.” Finally, since the differentiable structure of R4 is just its natural struc-
ture as a real vector space (Example #3, page 4), each tangent space Tp (R1,3 )
is canonically identified with R4 itself. Since the components of the semi-
Riemannian metric we have introduced are the same at every p ∈ R1,3 one
can think of R1,3 simply as the vector space R4 equipped with the Minkowski
inner product g(v, w) = ηαβ v α wβ = v 0 w0 − v 1 w1 − v 2 w2 − v 3 w3 . A vector
v in R1,3 is said to be spacelike, timelike, or null if g(v, v) is < 0, > 0,
or = 0, respectively. The physical origin of the terminology will emerge as we
procced.
We introduce a matrix
 
1 0 0 0
 
0 −1 0 0
η = (ηαβ ) =  
0 0 −1 0
0 0 0 −1

and note that its inverse η −1 = (η αβ ) is, in fact, equal to η.

Now, to build our gauge theory we begin by letting X denote some open
submanifold of R1,3 (the charges creating our electromagnetic field live in R1,3
and we intend to carve out their worldlines and deal only with the source free
Maxwell equations on the resulting open submanifold of R1,3 ). The gauge
group G is U (1) so we consider a principal U (1)-bundle
P
U (1) ,→ P −→ X

over X and a connection ω on it (we consider first the pure Yang-Mills theory
in which matter fields are absent). Since U (1) is Abelian, all brackets in the Lie
algebra u(1) = Im C are zero so the curvature Ω of ω is given by Ω = dω.
If s : V −→ P is a local cross-section, then we may write the local gauge
potential and field strength as
A = s∗ ω = −i A
F = s∗ Ω = d A = −i d A = −i F ,

where A and F are real-valued forms on V (the minus sign is conventional).

If si : Vi −→ P and sj : Vj −→ P are two such local cross-sections with
Vj ∩ Vi ̸= ∅ and if gij : Vj ∩ Vi −→ U (1) is the corresponding transition
function, then
Aj = gij −1 Ai gij + gij −1 dgij = Ai + gij −1 dgij
50 2. Physical Motivation

and
F j = gij −1 F i gij = F i
on Vj ∩ Vi because U (1) is Abelian. In particular, the local field strengths,
since they agree on any intersections of their domains, piece together to give
a globally defined field strength 2-form F on X. This is a peculiarity of Abelian
gauge theories and one should note that, even here, the potentials A do not
agree on the intersections of their domains and so do not give rise to a globally
defined object on X. Indeed, since the transition function gij is a map into
U (1) it can be written as
gij (x) = e−i Λij (x)
so that gij −1 dgij = −i dΛij and Aj = Ai − i dΛij . Equivalently,
Aj = Ai + dΛij ,
which is the traditional form for the relationship between two “vector potentials.”
Relative to standard coordinates x0 , x1 , x2 , x3 on R1,3 we can write, for any
s : V −→ P ,
A = Aα dxα = −i Aα dxα
and
1 1
F = Fαβ dxα ∧ dxβ = − i Fαβ dxα ∧ dxβ ,
2 2
where
Fαβ = ∂α Aβ − ∂β Aα + [Aα , Aβ ]
= ∂α Aβ − ∂β Aα (because U (1) is Abelian)
= −i (∂α Aβ − ∂β Aα )
= −i Fαβ .

The Fαβ are skew-symmetric in α and β. To make some contact with the
notation used in physics we define functions E 1 , E 2 , E 3 and B 1 , B 2 , B 3 by
Fi0 = Ei
and
Fij = εijk B k
where i, j, k = 1, 2, 3 and εijk is the Levi-Civita symbol (1 if ijk is an even
permutation of 123, −1 if ijk is an odd permutation of 123, and 0 otherwise).
Thus,  
0 −E 1 −E 2 −E 3
 1 
E 0 B 3 −B 2 

(Fαβ ) =  2 
E −B 3 0 B1 

E 3
B 2
−B 1
0
and
 2.2. Electromagnetic Fields 51

1
F = Fαβ dxα ∧ dxβ = −E 1 dx0 ∧ dx1 − E 2 dx0 ∧ dx2 − E 3 dx0 ∧ dx3
2
+ B 3 dx1 ∧ dx2 − B 2 dx1 ∧ dx3 + B 1 dx2 ∧ dx3
= (E 1 dx1 + E 2 dx2 + E 3 dx3 ) ∧ dx0
+ B 3 dx1 ∧ dx2 + B 1 dx2 ∧ dx3 + B 2 dx3 ∧ dx1 .

⃗ = (E 1 , E 2 , E 3 ) and B
One is to think of E ⃗ = (B 1 , B 2 , B 3 ) as the “electric field”
and the “magnetic field,” respectively, that correspond to F (the justification
for thinking this way will appear shortly).
Next we introduce functions F αβ on X defined by

F αβ = η αγ η βδ Fγδ , α, β = 0, 1, 2, 3

(classically this is referred to as “raising the indices” with the

Minkowski metric). Thus, for example, F 01 = η 0γ η 1δ Fγδ = η 00 η 11 F01 = (1)
(−1) F01 = −F01 = E 1 and F 12 = η 1γ η 2δ Fγδ = η 11 η 22 F12 = (−1)(−1)F12 =
F12 = B 3 , etc., so
 
0 E1 E2 E3
 
−E 1 0 B 3 −B 2 
(F αβ ) = 
−E 2 −B 3
.
 0 B1  
−E 3 B 2 −B 1 0

The Hodge dual ∗ F of F is defined to be the 2-form on X whose standard

components are given by

∗ 1
Fαβ = εαβγδ F γδ , α, β = 0, 1, 2, 3.
2
Writing these out one finds that
 
0 B1 B2 B3
 
 −B 1 0 E3 −E 2 
(∗ Fαβ ) = 
−B 2

 −E 3 0 E1 
−B 3 E2 −E 1 0

so that

∗ 1 ∗
F = F αβ dxα ∧ dxβ
2
= (−B 1 dx1 − B 2 dx2 − B 3 dx3 ) ∧ dx0
+ E 3 dx1 ∧ dx2 + E 1 dx2 ∧ dx3 + E 2 dx3 ∧ dx1 .

One verifies that

∗∗
F = −F (on R1,3 ).
52 2. Physical Motivation

We also define
∗
F = −i ∗F .
Finally, one can also “raise the indices” of ∗F and define ∗F αβ = η αγ η βδ ∗Fαβ
so that  
0 −B 1 −B 2 −B 3
 1 
B 0 E 3 −E 2 
∗ 
( Fαβ ) =  2 .
B −E 3 0 E1 

B 3
E 2
−E 1
0
One can think of the Hodge dual as the 2-form obtained by formally replacing
B⃗ by E⃗ and E ⃗ by −B.
⃗
All of this apparently ad hoc notation will eventually be seen to fit naturally
into the general scheme of things. For the time being the reader may wish to
regard all of it as simply a useful bookkeeping device. For example, one has
the following easily verified formulas:
1
Fαβ F αβ = |B|⃗ 2 − |E|
⃗ 2
2
1
Fαβ ∗ F αβ = E
⃗ ·B⃗
4
for the two scalar invariants normally associated with F in classical electro-
magnetic theory.
But what is the justification for all of these references to classical elec-
tromagnetic theory? We began by looking at an arbitrary connection ω on
an arbitrary principal U (1)-bundle over an open submanifold of Minkowski
spacetime and have deviously interjected things we have called “electric fields”
and “magnetic fields.” Is there any reason to believe that these objects have
anything whatever to do with what physicists call “electric fields” and “mag-
netic fields?” The answer lies in the Yang-Mills equations. We are, after all,
not really interested in arbitrary connections, but only in the stationary val-
ues of the Yang-Mills action (page 56). In our present circumstances we will
find that this action can be written
∫
1
YM(ω) = − Fαβ F αβ dx0 dx1 dx2 dx3
X 4
and the corresponding Yang-Mills equations are
d ∗F = 0,
while the Bianchi identity is
dF = 0
(in the Abelian case, covariant exterior derivatives are just ordinary exterior
derivatives). In standard coordinates these read
 2.2. Electromagnetic Fields 53

∂α F αβ = 0, β = 0, 1, 2, 3
and
∂α ∗ F αβ = 0, β = 0, 1, 2, 3,
respectively. Now, the remarkable part is that, if one writes these out in terms
⃗ and B’s
of the E’s ⃗ we introduced earlier the result is
⃗
∇ ⃗ − ∂ E = ⃗0
⃗ ×B and ∇
⃗ ·E
⃗ =0 (d ∗F = 0)
∂x0
and
⃗
∇ ⃗ + ∂ B = ⃗0
⃗ ×E and ∇
⃗ ·B
⃗ =0 (dF = 0),
∂x0

respectively, where ∇⃗ = (∂1 , ∂2 , ∂3 ) is the usual gradient operator and the ×

and · refer to the cross product and dot product on R3 . These, of course, are
the source free Maxwell equations in their usual guise.
Remark: One should note that the second pair of Maxwell equations cor-
responds to the Bianchi identity and is, in this sense, purely geometrical, i.e.,
due to the fact that we have chosen to model our fields as connections on
principal bundles. The proper way to look at this, however, is the other way
around. We are able to build a model of an electromagnetic field as a con-
nection on a principal bundle only because of this second pair of Maxwell
equations. As a matter of terminology, one says that a differential form whose
exterior derivative is zero is closed. Thus, the source free Maxwell equations
assert that both F and ∗F are closed.
Before moving on let us make one more bit of notational contact with
physics. The local gauge potentials A = −iA = −i Aα dxα satisfy dA = F ,
i.e., Fαβ = ∂α Aβ − ∂β Aα , which we rewrite as follows: Define Aα = η αγ Aγ
for α = 0, 1, 2, 3 (i.e., “raise the indices” of A) and write
( )
⃗ ,
(A0 , A1 , A2 , A3 ) = V, A
where V = A0 = A0 and A ⃗ = (A1 , A2 , A3 ) = (−A1 , −A2 , −A3 ). Then a brief
calculation shows that dA = F becomes
⃗
E⃗ = − ∂ A − ∇V
⃗ and B ⃗ =∇ ⃗ ×A
⃗
∂x0
which are the usual expressions from physics for the electric and magnetic
fields in terms of the scalar and vector potentials.
We would now like to write out two concrete examples (Coulomb fields
and Dirac monopoles) which illustrate all of this apparatus and which, more
importantly, make clear the difference between magnetic charge (which is
“topological”) and electric charge (which is not). We will build the examples
by specifying local gauge potentials and allowing these to determine (by the
way they are related on the intersections of their domains) the transition
54 2. Physical Motivation

functions and therefore the bundle on which the corresponding connections

are defined. In order to fully appreciate the topological nature of what is going
on here, however, we must briefly anticipate some material that we will treat
carefully only somewhat later in the text.
Principal U(1)-bundles over any manifold X are classified up to equivalence
by a certain de Rham cohomology class of X called the first Chern class
of X. Very briefly, here’s what these words mean: As mentioned earlier, a
differential form φ whose exterior derivative dφ is zero is said to be closed.
φ is said to be exact if it is the exterior derivative of some form ψ of degree
one less (φ = dψ). According to the Poincaré Lemma (Theorem 4.4.2), every
exact form is closed (d2 = 0), but the converse is not true. Two closed forms
φ1 and φ2 are said to be cohomologous if they differ by an exact form
(φ1 − φ2 = dψ). For each degree k this is an equivalence relation and the set
of equivalence classes, which admits a natural real vector space structure, is
k
denoted HdeR (X) and called the k th de Rham cohomology group of X.
P
Now let G ,→ P −→ X be any principal G-bundle over X (with G a
matrix Lie group). The first Chern class c1 (P ) of the bundle is an element
2
of HdeR (X) defined as follows: Choose any connection ω on the bundle (we
will prove later that connections exist on any principal bundle). Let Ω be the
curvature of ω. For any local cross-section s let F = s∗ Ω be the local field
strength. These generally depend on the choice of s and do not agree on the
intersections of their domains (except in the Abelian case). Indeed, if sg is
another cross-section, then F g = g −1 F g on the intersection. Notice, however,
that, since the trace of a matrix is invariant under conjugation, trace F g =
trace F on the intersection and so these piece together to give a globally
defined 2-form on X which we will denote simply trace F . A priori trace F is
complex-valued, but one can show that, in fact, it takes values in Im C so that
2π trace F is real-valued (the 2π actually forces the integrals of this 2-form
i 1

over compact, oriented surfaces in X to be integers—not obvious, but true).

It also happens that this 2-form is closed and so determines a cohomology
class
[ ]
i
c1 (P ) = trace F ∈ HdeR
2
(X).
2π

Now, the remarkable part of all this is that this cohomology class (not the
2-form, but its cohomology class) does not depend on the initial choice of
the connection ω from which it arose. It is therefore a characteristic of the
bundle itself and not of the connection. Indeed, c1 (P ) is the simplest example
of what is called a characteristic class for the bundle. We’ll encounter one
more example of such a thing (the second Chern class) in Section 2.5.
Now let us specialize to the case of U (1)-bundles. Here we have seen that
there is a globally defined field strength F for any connection. Moreover, since
u(1) consists of 1 × 1 matrices, the trace just picks out the sole entry in this
matrix so trace F = −iF and therefore
 2.2. Electromagnetic Fields 55
[ ]
i 1
c1 (P ) = (−iF ) = [F ] (G = U (1)).
2π 2π

In particular, the first Chern class for a principal U (1)-bundle over an open
1
submanifold of Minkowski spacetime is just 2π times the cohomology class
of the globally defined (electromagnetic) field F on X. Since principal U (1)-
bundles over any manifold are classified up to equivalence by their first Chern
classes (Appendix E, [FU]), we conclude that the U (1)-bundle on which an
electromagnetic field F is to be modeled as a connection is uniquely deter-
mined by the cohomology class of F .
Now we return to our concrete examples. First, the Coulomb field, i.e., a
static, purely electric field of a point charge which we assume to be located at
the (x1 , x2 , x3 )-origin in R1,3 . Thus, the worldline of our source is the x0 -axis
in R1,3 so we take
X = R1,3 − {(x0 , 0, 0, 0) ∈ R1,3 : x0 ∈ R}.
Define A = Aα dxα = (−n/ρ)dx0 , where n is an integer and ρ > 0 with
ρ2 = (x1 )2 + (x2 )2 + (x3 )2 (we measure “charge” in multiples of the charge of
the electron so it is an integer). A simple calculation shows that the functions
Fαβ = ∂α Aβ − ∂β Aα , α, β = 0, 1, 2, 3, are given by
 
0 −x1 −x2 −x3
n 
 x1 0 0 0 

(Fαβ ) = 3  2 .
ρ x 0 0 0 
x3 0 0 0
Thus,
n
F = 3 (x1 dx1 + x2 dx2 + x3 dx3 ) ∧ dx0
ρ
so
B⃗ = ⃗0 and E ⃗ = n ⃗r, ⃗r = (x1 , x2 , x3 ).
ρ3
This is, of course, the classical Coulomb field that we wish to describe.
The critical observation here is this: Our Coulomb potential
A = (−n/ρ)dx0 is defined and satisfies dA = F globally on all of X so
that F is exact on X and its cohomology class [F ] ∈ HdeR 2
(X) is zero. Thus,
the U(1)-bundle on which F is modeled by a connection with field strength
F = −iF has first Chern class zero and so must be the trivial bundle. This is
true for any charge n so that, in particular, the electric charge of the source
is not encoded in the topology of this bundle (we will find that the situation
is quite different for a Dirac magnetic monopole).
Here’s another way to look at this: Somewhat later we will calculate the
cohomology of X = R1,3 − {(x0 , 0, 0, 0) ∈ R1,3 : x0 ∈ R} and find that
{
k
R, k = 0, 2
HdeR (X) = .
0, otherwise
56 2. Physical Motivation

Now, the 2-form F describing the Coulomb field on X is cohomologically

trivial, i.e., [F ] = 0 ∈ HdeR
2
(X). After learning how to integrate 2-forms over
2-dimensional manifolds we will find that this implies that the integral of F
over any closed, smoothly embedded surface in some x0 = constant slice must
be zero. In particular, these integrals do not detect any “enclosed charge.”
The rest of the 2-dimensional cohomology of X is to be found in the dual
∗
F = 12 ∗ Fαβ dxα ∧ dxβ , where
 
0 0 0 0
n 
0 0 x3 −x2 

(∗ Fαβ ) = 3  .
ρ 0 −x 3
0 x 
1

0 x2 −x1 0
Thus, ∗ n
F = 3 (x1 dx2 ∧ dx3 − x2 dx1 ∧ dx3 + x3 dx1 ∧ dx2 ).
ρ
Notice that, on the 2-sphere ρ = 1 in any x0 = constant slice of X, ∗ F reduces
to n times x1 dx2 ∧ dx3 − x2 dx1 ∧ dx3 + x3 dx1 ∧ dx2 on S 2 . This, as we shall
see, is what is called the standard volume form of S 2 and its integral over S 2
is the area 4π of S 2 (Section 4.6). Thus, the integral of ∗ F over this sphere
is 4πn, which is nonzero and this implies that ∗ F cannot be cohomologically
trivial. Thus, [∗ F ] generates HdeR2
(X). Furthermore, we will show that the
integral of F over any 2-sphere surrounding {(x0 , 0, 0, 0) : x0 ∈ R} is the
∗

same so these integrals “detect” the charge n enclosed by the sphere. Over
any 2-sphere that does not enclose the x0 -axis, the integral of ∗ F is zero (this
will follow from “Stokes’ Theorem”).
The electric charge n is not “topological” because all Coulomb fields are
represented by connections on the trivial bundle—the charge is not encoded
in the topology of the bundle. The situation is quite different for a Dirac
monopole (which is not surprising since the Hodge dual for 2-forms on X
essentially interchanges “electric” and “magnetic” so, for a magnetic charge,
[F ] will play the role that [∗ F ] played for the Coulomb field). In more detail,
we once again let X = R1,3 − {(x0 , 0, 0, 0) ∈ R1,3 : x0 ∈ R}. For the moment
we let g denote an arbitrary real number (to be thought of as the magnetic
“charge” of the monopole whose worldline is the x0 -axis). We are interested
in the field F = 12 Fαβ dxα ∧ dxβ , where
 
0 0 0 0
g 
0 0 x3

−x2 
(Fαβ ) = 3  .
ρ 0 −x3 0 x1 
0 x2 −x1 0

Thus,
g
F = (x1 dx2 ∧ dx3 − x2 dx1 ∧ dx3 + x3 dx1 ∧ dx2 )
ρ3
 2.2. Electromagnetic Fields 57

so
E ⃗ = g ⃗r,
⃗ = ⃗0 and B ⃗r = (x1 , x2 , x3 ).
ρ3
Since F is independent of x0 this defines a 2-form on any x0 = constant
slice. Moreover, expressed in terms of standard spherical coordinates (ρ, φ, θ)
on such a slice,
F = g sin φ dφ ∧ d θ.
Notice that this is independent of ρ and so may be further restricted to the
copy ρ = 1 of S 2 in, say, the x0 = 0 slice of X. Now, if the monopole field on
X is the field strength of some connection on a U(1)-bundle over X, then its
restrictions would likewise be field strengths for connections on U(1)-bundles
over the submanifolds (restriction means pullback by the inclusion map and
the inclusion of a restricted bundle is a bundle map). Henceforth, we will
concentrate on these restrictions.
For the field F under consideration there is no globally defined potential
A satisfying dA = F (pages 2–3 of [N4]), but there are the usual local
potentials. Specifically, we define AN and AS on UN = S 2 − {(0, 0, 0, −1)}
and US = S 2 − {(0, 0, 0, 1)}, respectively, by
AN = g(1 − cos φ) dθ
and
AS = −g(1 + cos φ) dθ.
Then, on their respective domains, these satisfy dAN = F and dAS = F . Con-
sider now the corresponding u(1)-valued forms (we identify u(1) with Im C):

AN = −i AN = −i g(1 − cos φ) dθ
AS = −i AS = i g (1 + cos φ) dθ
F = −i F = −i g sin φ d φ ∧ d θ

If our monopole field is to be modeled by a connection on some principal

U (1)-bundle over S 2 , then that bundle could be trivialized over UN and US
and so would have a transition function gSN : UN ∩ US −→ U (1) for which
AN = gSN −1 AS gSN + gSN −1 dgSN . But notice that AN − AS = −2gi dθ so

AN = AS − 2gi dθ = e2gθi AS e−2gθi + e2gθi d(e−2gθi )

and this gives
gSN (φ, θ) = e−2gθi .
Next we observe that the only values of the constant g that are of any
interest are given by
n
g= , n ∈ Z.
2
One can understand this in a number of ways, all of which are instructive.
From the point of view of physics, it is just the Dirac quantization condi-
tion (page 7, [ N4]). On the other hand, if gSN (φ, θ) = e−2gθi is really the
58 2. Physical Motivation

transition function for a principal U (1)-bundle U (1) ,→ P −→ S 2 , then that

bundle is characterized by its first Chern class c1 (P ). As we pointed out earlier,
the integral of c1 (P ) over S 2 is an integer, called the first Chern number
of the bundle. But a simple calculation gives
∫ ∫ ∫
1 g
c1 (P ) = F = sin φdφ ∧ dθ = 2 g
S2 2π S 2 2π S 2

(we assume S 2 has its standard orientation). Thus, 2g ∈ Z. From yet another
perspective, the restriction of gSN (φ, θ) = e−2gθi to the equatorial circle S 1
in UN ∩ US (i.e., ei θ −→ (ei θ )−2g ) would be the “characteristic map” whose
homotopy type determines the bundle (page 228, [N4]) and this map is not
even well-defined (single-valued) on S 1 unless 2g is an integer. However you
choose to view the situation, we will henceforth restrict our attention to the
following forms:
1
AN = − ni (1 − cos φ) dθ
2
1
AS = ni (1 + cos φ)dθ
2
1
F = − ni sin φ dφ ∧ dθ.
2
For each fixed integer n, the potentials AN and AS uniquely determine a
connection ω n on the principal U (1)-bundle
P
U (1) ,→ Pn −→
n
S2

whose transition function gSN is given by

gSN (φ, θ) = e−nθi .

The globally defined field strength on S 2 is F and represents the field of a

Dirac monopole of “magnetic charge” n (which is the Chern number of the
bundle). Since the Chern number is a topological characteristic of the bundle,
magnetic charge is directly encoded into the topology of the bundle and is
therefore an instance of a “topological charge.”
Although it is not really necessary to do so (because the transition functions
and local gauge potentials contain all of the relevant information about the
bundles and the connections), it is possible to describe these Dirac monopoles
more explicitly. We will outline such a description.

n=1

This gives the natural connection on the complex Hopf bundle

P
U (1) ,→ S 3 −→
1
S2,
 2.2. Electromagnetic Fields 59

where P1 is the restriction to S 3 ⊆ C2 of the map

( ) ( 2 2
)
P1 z 1 , z 2 = z 1 z̄ 2 + z̄ 1 z 2 , −i z 1 z̄ 2 + i z̄ 1 z 2 , z 1 − z2 .

ω 1 is given by
ω 1 = i ι∗ (Im(z̄ 1 dz 1 + z̄ 2 dz 2 )),
where ι : S 3 ,→ C2 is the inclusion.

n = −1

This gives the natural connection on the alternate version of the complex Hopf
bundle
P−1
U (1) ,→ S 3 −→ S 2 ,
where P−1 is the restriction to S 3 ⊆ C2 of the map
( ) ( 2 2
)
P−1 z 1 , z 2 = z 1 z̄ 2 + z̄ 1 z 2 , i z 1 z̄ 2 − i z̄ 1 z 2 , z 1 − z2 .
ω −1 is given by
( ( ))
ω −1 = i ι∗ Im z̄ 1 dz 1 + z̄ 2 dz 2 = ω1 ,

where ι : S 3 ,→ C2 is the inclusion.

n=0

This gives the flat connection on the trivial bundle

P
U (1) ,→ S 2 × U (1) −→
0
S2,

where P0 is the projection onto the first factor. ω 0 is given by

ω 0 = π ∗ Θ,

where π : S 2 × U (1) −→ U (1) is the projection onto the second factor and Θ
is the Cartan 1-form on U (1).
n>1
P
Denote by U (1) ,→ Pn −→ n
S 2 the principal U (1)-bundle over S 2 with transi-
−nθi
tion function gSN (φ, θ) = e . We can identify Pn explicitly as a manifold as
follows: Identify the discrete group Zn of integers modulo n with the following
subgroup of U (1):

Zn = {e2kπi /n : k = 0, 1, . . . , n − 1}.
Then Zn acts on S 3 on the right (because U (1) does). We let S 3 /Zn be the
orbit space, e.g., S 3 /Z2 = RP3 . One can provide S 3 /Zn with a manifold
60 2. Physical Motivation

structure in the same way as for RP3 . The Hopf map P1 : S 3 −→ S 2 carries
each orbit of the usual U (1)-action on S 3 to a point in S 2 so it does the same
for a Zn -orbit. Moreover, each point of S 2 is the image under P1 of a Zn -orbit
(indeed, of many Zn -orbits). Thus, P1 descends to a surjective map

Pn : S 3 / Zn −→ S 2 .

Also note that the usual U (1)-action on S 3 carries any Zn -orbit onto another
Zn -orbit which is inside the same U (1)-orbit (and so has the same image under
Pn ). Thus, the U (1)-action on S 3 descends to a U (1)- action on S 3 /Zn which
preserves the fibers of Pn . Local triviality of Pn : S 3 /Zn −→ S 2 follows so
that
Pn
U (1) ,→ S 3 / Zn −→ S2
is a principal U (1)-bundle and the transition function gSN is given
by gSN (φ, θ) = e−n θ i . Since a bundle is determined by its transition functions,
Pn
we have an explicit model for U (1) ,→ Pn −→ S 2 . In particular, Pn ∼
= S 3 / Zn .
Remark: Pn = S 3 / Z n is an example of a “lens space.”
P
Since the transition function for U (1) ,→ S 3 / Z n −→ n
S 2 is
−n θ i
gSN (φ, θ) = e , AN = − 2 ni (1 − cos φ) dθ and AS = 2 ni (1 + cos φ) dθ
1 1

are gauge potentials on this bundle. The connection ω n they determine

can be described as follows: The Im C-valued 1-form ω̃ n on C2 given by
ω̃ n = i n Im(z̄ 1 dz 1 + z̄ 2 dz 2 ) restricts to an Im C-valued 1-form ι∗ ω̃ n on S 3
that is invariant under the Z n -action and so descends to an Im C -valued 1-
Pn
form ω n on S 3 / Z n . ω n is the required connection on U (1) ,→ S 3 / Z n −→
2
S .
n < −1
Here the construction is exactly the same as above for n > 1 except that P1
is replaced by P−1 .

2.3 Spin Zero Electrodynamics

We consider next our first example of a gauge theory in which matter fields
are present. The gauge field will be an electromagnetic field of the type
discussed in Section 2.2. The matter field coupled to this gauge field will
represent a charged particle experiencing the effects of the electromagnetic
field. According to the General Framework described in Section 2.1, the
description of such a particle will require a vector space V and a representation
ρ : U (1) −→ GL (V) of U (1) on V. Now, in physics, charged particles have
wavefunctions with a certain number of complex components, the number of
such components being determined by the particle’s spin s. Specifically, s is
an element of {0, 12 , 1, 32 , 2, . . . } and the wavefunction of a particle with spin s
 2.3. Spin Zero Electrodynamics 61

has 2s + 1 components. The simplest case is that of a particle of spin 0 (e.g.,

a π − meson) for which the wavefunction takes values in C. This is the case
we consider here.
Remark: Certain technical complications arise when s > 0. For example,
an electron has s = 12 and, according to Dirac, the corresponding matter field
is defined, not on a U (1)-bundle over X, but on a certain SL(2, C)-bundle
over X called a spinor bundle. Nevertheless, an electron responds to an elec-
tromagnetic, i.e., a U (1)-gauge, field. To fit this into our General Framework
would require “splicing” the two bundles together into a single bundle on which
both objects may be thought to live. We will return to these issues in the next
section.
We will adopt the notation of Section 2.1. Thus, X is an open submanifold of
R1,3 (standard coordinates x0 , x1 , x2 , x3 ) and ω is a connection on a principal
U (1)-bundle
P
U (1) ,→ P −→ X
over X with curvature Ω. For any cross-section s we write
s∗ ω = A = Aα dxα = −i A = −i Aα dxα
and 1 1
s∗ Ω = F = Fαβ dxα ∧ dxβ = −i F = − i Fαβ dxα ∧ dxβ ,
2 2
where Fαβ = ∂α Aβ − ∂β Aα = −i (∂α Aβ − ∂β Aα ). Now we let V = C (as a
2-dimensional real vector space). The usual inner product on C = R2 is given
by
1
⟨z1 , z2 ⟩ = (z1 z̄2 + z2 z̄1 ).
2
For each integer n we define a representation ρn : U (1) −→ GL(C) by

ρn (g)(z) = g · z = g n z,

where we identify an element of U (1) with a complex number of modulus 1

(these are, in fact, all of the “irreducible” representations of U (1) on C). Notice
that ρn is orthogonal with respect to ⟨ , ⟩ since

⟨ρn (g)(z1 ), ρn (g)(z2 )⟩ = ⟨g n z1 , g n z2 ⟩

1
= ((g n z1 ) (g n z2 ) + (g n z2 ) (g n z1 ))
2
1
= (g n (z1 z̄2 )g −n + g n (z2 z̄1 )g −n )
2
1
= (z1 z̄2 + z2 z̄1 )
2
= ⟨z1 , z2 ⟩.
62 2. Physical Motivation

The potential function U : C −→ R (#7 of the General Framework) is

taken to be
1 1 1
U (z) = m⟨z, z⟩ = m| z |2 = mz z̄
2 2 2
where m ≥ 0 is a constant. Since ρn is orthogonal with respect to ⟨ , ⟩, U is
invariant under the action of U (1) on C, as required.
A matter field of type ρn is a C -valued map ϕ on P that is equivariant
with respect to the actions of U (1) on P and C , i.e., that satisfies

ϕ(p · g) = g −n ϕ(p)

for each p ∈ P and g ∈ U (1). The connection ω determines a covariant

exterior derivative dω ϕ of ϕ and the action A(ω, ϕ) contains an appropri-
ate squared norm ∥ dω ϕ∥ 2 arising from the Minkowski metric on X and
the invariant inner product ⟨ , ⟩ on C. We will defer until later the gen-
eral procedure for constructing such norms and simply indicate at this point
what the result is for the case under consideration and why it is the “natu-
ral” choice. For any local cross-section s, s∗ ϕ = ϕ ◦ s and, for convenience,
we will write ϕ = ϕ(x0 , x1 , x2 , x3 ) for the standard coordinate representa-
tion for s∗ ϕ. The corresponding coordinate expression for the pullback of
dω ϕ is
(∂α ϕ − i n Aα ϕ) dxα
(Example #1, page 52). Each ∂α ϕ − i n Aα ϕ is a function with values in C.
The invariant inner product ⟨ , ⟩ on C and the Minkowski inner product
(g(v, w) = v 0 w0 − v 1 w1 − v 2 w2 − v 3 w3 = ηαβ v α wβ = η αβ vα wβ , where
vα = ηaα v a and wβ = ηbβ wb ) combine to give the squared norm
( )
∥ dω ϕ ∥2 = η αβ (∂α ϕ − i nAα ϕ) ∂β ϕ − i nAβ ϕ

= η αβ (∂α ϕ − i nAα ϕ) (∂β ϕ̄ + i nAβ ϕ̄) (Aβ is real)

= (∂α ϕ − i nAα ϕ) (η αβ (∂β ϕ̄) + i n(η αβ Aβ )ϕ̄)
= (∂α ϕ − i nAα ϕ) (∂ α ϕ̄ + i nAα ϕ̄),

where we have written

∂ α = η αβ ∂β , α = 0, 1, 2, 3

(so that ∂ 0 = ∂0 , ∂ 1 = −∂1 , ∂ 2 = −∂2 and ∂ 3 = −∂3 ) and

Aα = η αβ Aβ , α = 0, 1, 2, 3.

Although the expression we have given for ∥dω ϕ∥2 is local and appears to
depend on the choice of the cross-section s, it is, in fact, gauge invariant and
therefore determines a globally defined, real-valued function on X. To see
 2.3. Spin Zero Electrodynamics 63

this we observe the following: We have already seen (page 59) that a gauge
transformation g can be written g(x) = e−i Λ(x) and has the following effects
on the potential A and the matter field ϕ:

A −→ Ag = A + dΛ
ϕ −→ ϕg = g −1 · ϕ = ei nΛ ϕ.

Thus,

∂α ϕg − i n(Ag )α ϕg = ∂α (ei nΛ ϕ) − i n(Aα + ∂α Λ)(ei nΛ ϕ)

= ei nΛ ∂α ϕ + i nei nΛ (∂α Λ)ϕ − i nAα ei nΛ ϕ
− i n(∂α Λ)ei nΛ ϕ
= ei nΛ (∂α ϕ − i nAα ϕ)

and, similarly,

∂ α ϕg + i n(Ag )α ϕg = e−i nΛ (∂ α ϕ̄ + i nAα ϕ̄).

Consequently,
( )
(∂α ϕg − i n(Ag )α ϕg ) ∂ α ϕg + i n(Ag )α ϕg
( )
= (∂α ϕ − i nAα ϕ) ∂ α ϕ̄ + i nAα ϕ̄
as required. With this and an appropriate choice of normalizing and coupling
constants, we take our action functional to be
∫ [
1 1
A(ω, ϕ) = − Fαβ F αβ + (∂α ϕ − i nAα ϕ) (∂ α ϕ̄ + i nAα ϕ̄)
X 4 2
]
1
+ mϕϕ̄ dx0 dx1 dx2 dx3 .
2

The corresponding Euler-Lagrange equations are

(∂α − i nAα ) (∂ α − i nAα ) ϕ + m2 ϕ = 0
∂α F αβ = 0, β = 0, 1, 2, 3

(see [Bl]). The first of these is the Klein-Gordon equation (for a spin zero
particle of mass m and charge n interacting with a gauge field F = −i F =
−i dA determined by the local gauge potentials A = −i Aα dxα ). The second
equation is equivalent to d∗ F = 0 (see page 62). Since −i F is the pullback
of a curvature form, the Bianchi identity gives dF = 0 also so F satisfies
Maxwell’s equations.
Remark: Of course, it was our stated intention to model a spin zero particle
in an electromagnetic field, but the point here is that the electromagnetic
64 2. Physical Motivation

nature of the field F that appears in the action A(ω, ϕ) is necessitated by the
Euler-Lagrange equations. We do not have to impose Maxwell’s equations “by
hand.” Also note that conjugating the Klein-Gordon equation gives

(∂α + i nAα )(∂ α + i nAα )ϕ̄ + m2 ϕ̄ = 0.

Thus, if ϕ represents a Klein-Gordon field of mass m and charge n, ϕ̄ represents

a Klein-Gordon field of mass m and charge −n. Physicists view ϕ and ϕ̄ as
wavefunctions for a particle/antiparticle pair.
Taking A to be zero above (i.e., “turning off” the electromagnetic field) gives
the free Klein-Gordon equation
∂α ∂ α ϕ + m 2 ϕ = 0
for a spin zero particle.
Remark: Free objects in physics are also subject to field equations. For ex-
ample, a free particle of mass m in Newtonian mechanics satisfies
d
v ) = ⃗0, while in quantum mechanics its wavefunction ψ satisfies the
dt (m⃗
Schroedinger equation
1 ⃗2 ∂ψ
− ∇ ψ=i .
2m ∂t
Note also that if ϕ satisfies the free Klein-Gordon equation, then so
does ϕ̄.

We will not pursue the business of seeking solutions to these equations

and sorting out their physical significance. Indeed, it would seem that no
plausible physical interpretation of such solutions exists outside the context
of quantum field theory and for this the only service we can perform for our
reader is a referral to the physics literature (e.g., [Gui], [Ry], or [Wein]).
We will, however, spend a few moments describing an alternative approach
to the Klein-Gordon equation which more clearly exposes the philosophical
underpinnings of modern gauge theory. Physicists refer to this approach as
“minimal coupling” and it begins with the free particle equation

∂α ∂ α ϕ + m2 ϕ = 0.

This equation itself might be arrived at by “quantizing” the classical relativis-

tic relation E 2 = p⃗ 2 + m2 between energy and momentum (“quantization”
is the mystical process of replacing classical quantities such as these with
“corresponding operators” that act on the wavefunction). One then “couples”
the particle to the field by replacing the “ordinary derivatives” ∂α by “covari-
ant derivatives” ∂α − i nAα involving the potentials Aα for the field. This,
of course, does give the full Klein-Gordon equation, but the motivation is
no doubt obscure. Our search for motivation leads to the very early days of
quantum mechanics.
 2.3. Spin Zero Electrodynamics 65

In old-fashioned (non-relativistic) quantum mechanics a charged particle

(of mass m and charge n) in an electromagnetic field has a wavefunction ψ
that satisfies the Schroedinger equation
( )
1 ( ⃗ )2 ∂
−i ∇ − nA⃗ ψ= i − nV ψ, (2.3.1)
2m ∂t

where (V, A)⃗ = (A0 , A1 , A2 , A3 ) = (A0 , −A1 , −A2 , −A3 ) and A = Aα dxα
satisfies dA = F (see page 63).
Remark: The meaning of (−i ∇
⃗ − nA)
⃗ 2 as an operator on ψ is as follows:
( )2 ( ) ( )
−i ∇
⃗ − nA
⃗ ψ = −i ∇ ⃗ − nA⃗ · −i ∇⃗ − nA⃗ ψ
( ) ( )
= −i ∇⃗ − nA⃗ · −i ∇ψ
⃗ − n ψA ⃗

= −∇
⃗ 2 ψ + ni ∇⃗ · (ψ A) ⃗ · (∇ψ)
⃗ + ni A ⃗
+ n2 | A
⃗ |2 ψ.

Now, ψ takes values in C and so, at each point, has a modulus and a phase
(ψ = rei θ ). There is some arbitrariness in the phase, however, since, if a is an
element of U (1) (identified with a complex number of modulus 1), then aψ
satisfies (2.3.1) whenever ψ does (being a constant, a just slips outside of all
the derivatives in (2.3.1)). Moreover, aψ differs from ψ only in the phase fac-
tor (since |a| = 1) and so |aψ|2 = |ψ|2 . Since all of the physically significant
probabilities in quantum mechanics depend only on this squared modulus,
ψ and aψ should represent the same physical object. This freedom to alter
the phase of ψ is quite restricted, however. Since a must be constant, any
phase shift in the wavefunction must be accomplished at all spatial locations
simultaneously. Such a global phase shift, however, violates both the spirit
and the letter of relativistic law (you can’t do anything “at all spatial loca-
tions simultaneously”). Nevertheless, it is difficult to shake the feeling that
the physical significance of ψ “should” persist under some sort of phase shift
(again, because squared moduli will be unaffected). Notice that the relativistic
objection to a phase shift would disappear if one allowed the phase to shift
independently at each spacetime point, i.e., if one replaced ψ by aψ, where a
is now a function of (x, y, z, t) taking values in U (1). The problem with this,
of course, is that, if a is not constant, it will not simply “slip outside” of all
the derivatives in (2.3.1). Indeed, product rules will generate all sorts of new
terms that do not cancel so there is no reason to suppose that aψ will even
be a solution to (2.3.1). A bit (actually, quite a bit) of vector calculus will, in
fact, establish the following: Let ψ be a solution to (2.3.1) and let Λ(x, y, z, t)
be any smooth, real-valued function. Then

ψ ′ = ei n Λ ψ
66 2. Physical Motivation

is a solution to
( ( ))
1 ( ⃗ ( ))2 ∂ ∂Λ
−i ∇ − n A⃗ + ∇Λ
⃗ ψ′ = i −n V − ψ′ . (2.3.2)
2m ∂t ∂t

This last result is interesting from a number of different points of view.

Physicists in the early part of this century found in it some confirmation of
their long held belief that the nonuniqueness of the potential for an electro-
magnetic field and the nonuniqueness of the phase for a wavefunction were
both matters of no physical consequence. The reasoning went something like
this: In equation (2.3.1) the electromagnetic field to which ψ is responding is
described by the potential (V, A). ⃗ The corresponding 1-form is A = Aα dxα ,
where A0 = V and (A1 , A2 , A3 ) = −A ⃗ (see page 63). In equation (2.3.2), (V, A)
⃗
is replaced by (V − ∂t , A+ ∇Λ) and the corresponding 1-form is A−dΛ which,
∂Λ ⃗ ⃗

of course, describes the same electromagnetic field F = dA = d(A − dΛ) since

d2 = 0 (see page 63). Classically, the potential was regarded as a convenient
computational device, but with no physical significance of its own outside of
the fact that it gives rise via differentiation to the electromagnetic field. Thus,
(2.3.1) and (2.3.2) should, in some sense, be “equivalent” (describe the same
physics). Now, the solutions to (2.3.1) and (2.3.2) are in one-to-one correspon-
dence (ψ ↔ ei nΛ ψ) and the corresponding functions differ only by the phase
factor ei nΛ . Since |ψ|2 = |ei nΛ ψ|2 at each point, all of the usual probabilities
calculated for ψ and ei nΛ ψ in quantum mechanics are the same so that these
should be two descriptions of the same physical object. Everything seems to
fit together quite nicely.
This placid scene was disturbed in the late 1950s when Aharonov and Bohm
[AB] suggested that, while the phase of a single charge may well be un-
measurable, the relative phase of two charged particles that interact should
have observable consequences. Their proposed experiment (later confirmed by
Chambers in 1960) is described on page 6 of [N4 ]. The potential had now to be
taken seriously as a physical field and not merely a mathematical contrivance.
This being the case, one is no longer free to regard a phase shift ψ −→ ei nΛ ψ
as devoid of physical content. Indeed, physicists have gone to quite the other
extreme and elevated the invariance of electrodynamics under such local phase
transformations to the status of a basic physical principle (this is an instance
of the so-called “gauge principle” which lies at the heart of modern gauge
theory). To see more clearly the consequences of such a principle we consider
the special cases of (2.3.1) and (2.3.2) in which the electromagnetic field is
“turned off.”
If the electromagnetic field is zero one may choose the potential A for which
⃗ = (0, ⃗0) so that (2.3.1) becomes the usual free Schroedinger equation
(V, A)

1 ⃗ 2 ψ = i ∂ ψ.
(−i ∇) (2.3.3)
2m ∂t
 2.3. Spin Zero Electrodynamics 67

On the other hand, the same (trivial) electromagnetic field is described by any
potential of the form (0 − ∂Λ
∂t , 0 + ∇Λ). Begging the indulgence of the reader
⃗ ⃗
we would now like to call this (V, A)⃗ and write (2.3.2) as
( )
1 ( )2
⃗ ψ ′ = i ∂ − nV ψ ′ .
−i ∇
⃗ − nA (2.3.4)
2m ∂t

Now suppose one adopts as a basic physical principle that electrodynam-

ics should be invariant under local phase shifts. In particular, then equa-
tions (2.3.3) and (2.3.4), with (V, A)⃗ = (− ∂Λ , ∇Λ),
⃗ are equivalent. Physicists
∂t
take the following rather remarkable view of this: Even in a vacuum (elec-
tromagnetic field zero) the requirement of local phase shift invariance (gauge
invariance) necessitates the existence of a physical field A (not the electro-
magnetic field – that’s zero – but the gauge potential field) whose task is to
counteract, or balance, the effects of any phase shift and keep the physics
invariant. Algebraically, this happens by adding to the operators that ap-
pear in the Schroedinger equation terms whose sole purpose is to cancel the
extra stuff you get from product rules for ei n Λ ψ when Λ is not constant.
The marvelous thing about this way of viewing the situation is that it pro-
vides a completely mindless way of ensuring gauge invariance in all sorts
of contexts. One need only fudge into the differential operators whatever it
takes to cancel the offending extra terms that arise from the product rule.
In electrodynamics it works just as well when the field is not turned off. In-
deed, “turning the field on,” i.e., coupling a particle to an electromagnetic
field, can be viewed in exactly the same light. A free particle satisfies (2.3.3).
To couple the particle to an electromagnetic field F , write F = dA, where
A = Aα dxα , (A0 , A1 , A2 , A3 ) = (A0 , −A1 , −A2 , −A3 ) = (V, A)
⃗ and make the
substitutions
∂ ∂
−i ∇
⃗ −→ −i ∇
⃗ − nA
⃗ and i −→ i − nV
∂t ∂t
to obtain ( )
1 ( )2
⃗ ψ = i ∂ − nV ψ
−i ∇
⃗ − nA
2m ∂t
which is (2.3.1). The solutions to (2.3.1) will describe the wavefunction of the
charged particle coupled to the field F . The description is only one among
many possible descriptions, each differing from the others by a choice of phase
(gauge) determined by the particular A one has chosen for the representation
F = dA.
Finally, we show that the operator substitutions described above assume
a more elegant (and more familiar) form if we recast them in relativistic
notation. Here we let
( ) ( )
∂ ∂ ∂ ∂ ∂ ⃗
(∂0 , ∂1 , ∂2 , ∂3 ) = , , , = , ∇
∂x0 ∂x1 ∂x2 ∂x3 ∂t
68 2. Physical Motivation

and ( )
∂
(∂ , ∂ , ∂ , ∂ ) = (∂0 , −∂1 , −∂2 , −∂3 ) =
0 1 2 3
, −∇
⃗
∂t
so
( ) ( )
∂ ⃗ = i ∂ + i n V, −∇
i − n V, −i ∇
⃗ − nA ⃗ + i nA⃗
∂t ∂t
( 0
= i ∂ + i n V, ∂ 1 + i n A1 ,
)
∂ 2 + i n A 2 , ∂ 3 + i n A3 .

Since ( ) ( )
∂ ∂ ( )
i , −i ∇ = i
⃗ , −∇ = i ∂ 0 , ∂ 1 , ∂ 2 , ∂ 3 ,
⃗
∂t ∂t
the substitutions above simply amount to

∂ α −→ ∂ α + i n Aα , α = 0, 1, 2, 3,

or, lowering the indices once again,

∂α −→ ∂α + i n Aα , α = 0, 1, 2, 3

(the sign is + rather than − because of our decision to include the conventional
minus sign in A = −i A).
We will conclude our discussion of the Klein-Gordon equation by de-
scribing its two most important invariance properties: gauge invariance and
Lorentz invariance. We have already shown that the action A(ω, ϕ) is invariant
under gauge transformations and it follows that the same is true of its Euler-
Lagrange equations. Nevertheless, a direct proof is instructive. Thus, we con-
sider the equation

(∂α − i n Aα ) (∂ α − i n Aα )ϕ + m2 ϕ = 0 (2.3.5)

and a gauge transformation g(x) = e −i Λ(x) . The corresponding changes in the

potential and the matter field are, as usual,

A −→ Ag = A + d Λ

and
ϕ −→ ϕg = g −1 · ϕ = ei n Λ ϕ.
Our objective is to show that, if ϕ satisfies (2.3.5), then

(∂α − i n (Ag )α ) (∂ α − i n (Ag )α ) ϕg + m2 ϕg = 0. (2.3.6)

First, we compute
 2.3. Spin Zero Electrodynamics 69
( )
(∂ α − i n (Ag )α ) ϕg = (∂ α − i n (Aα + ∂ α Λ)) ei n Λ ϕ
= ∂ α (ei nΛ ϕ) − i n Aα ei n Λ ϕ − i n ∂ α Λ ei n Λ ϕ
= ei n Λ ∂ α ϕ + i n e i n Λ ∂ α Λ ϕ
− i n A α ei n Λ ϕ − i n ∂ α Λ ei n Λ ϕ
= ei n Λ (∂ α ϕ − i n Aα ϕ)
= ei n Λ (∂ α − i n Aα ) ϕ.

In the same way,

( )
(∂α − i n (Ag )α ) e i n Λ ϕ = ei n Λ (∂α − i n Aα ) ϕ

so

(∂α − i n (Ag )α ) (∂ α − i n (Ag )α ) ϕg

( )
= (∂α − i n (Ag )α ) ei n Λ (∂ α − i n Aα ) ϕ

= ei n Λ (∂α − i n Aα )(∂ α − i n Aα ) ϕ.

From this it is clear that (2.3.5) implies (2.3.6). The Klein-Gordon equation
is gauge invariant.
We wish to show next that the Klein-Gordon equation is “relativistically
invariant.” Roughly, this means that the equation has the same mathematical
form in all inertial frames of reference, but the precise meaning of such a
statement in general will require some discussion. In this section we will be
content to spell out explicitly what is being asserted for the Klein-Gordon
equation alone. When we turn to the Dirac equation in the next section we
will describe precisely what is meant by “relativistic invariance” in general.
Thus far we have written the Klein-Gordon equation (2.3.5) only in
standard coordinates x0 , x1 , x2 , x3 for R1,3 , i.e., only in one fixed inertial frame
of reference. To emphasize this fact we will now write the derivatives ∂α and
∂ α explicity as ∂/∂xα and η αβ ∂/∂xβ so that (2.3.5) becomes
( )(
∂ ∂
η αβ α
− i n A α (x 0
, . . . , x3
) − i n Aβ (x0 ,
∂x ∂xβ
) (2.3.7)
. . . , x3 ) ϕ (x0 , . . . , x3 ) + m2 ϕ (x0 , . . . , x3 ) = 0.

The basic postulate of Special Relativity is that one inertial frame of reference
is as good as another. The coordinates y 0 , y 1 , y 2 , y 3 for R1,3 , supplied by
another such frame of reference are assumed to be related to x0 , x1 , x2 , x3 by

y α = Λα β x β , α = 0, 1, 2, 3,
70 2. Physical Motivation

where Λ = (Λα β ) is an element of the so-called “proper, orthochronous Lorentz

group” L+↑ = {Λ : Λ η Λ = η, det Λ = 1, Λ0 0 ≥ 1}, where η is on page 58.
⊥

Remark: The “general Lorentz group” L consists of those Λ that satisfy

Λ η Λ = η, and these relate arbitrary orthonormal bases in R1,3 . The “or-
⊥

thochronous” condition Λ0 0 ≥ 1 ensures that Λ does not reverse time orienta-

tions, i.e., x0 ≥ 0 implies y 0 = Λ0 α xα ≥ 0, for timelike or null vectors, while
det Λ = 1 (“proper”) then guarantees that Λ does not reverse the orientation
of the spatial coordinates (x1 , x2 , x3 ).
The inverse of Λ is written Λ−1 = (Λα β ) so that

x β = Λα β y α , β = 0, 1, 2, 3.

Furthermore, one has

Λα γ Λβ δ η γδ = η αβ , α, β = 0, 1, 2, 3

and
Λα γ Λβ δ η αβ = η γδ , γ, δ = 0, 1, 2, 3

(physical motivation for and basic properties of the Lorentz group are dis-
cussed in some detail in the Introduction and first three sections of
Chapter 1 in [N3]).

The precise assertion we are making about the Klein-Gordon equation is as

follows: If ϕ = ϕ(x0 , . . . , x3 ) is a solution to (2.3.7) and if we define ϕ̂ =
ϕ̂(y 0 , . . . , y 3 ) by

ϕ̂ (y 0 , . . . , y 3 ) = ϕ (Λα 0 y α , . . . , Λα 3 y α ),

then ϕ̂ satisfies
( )(
∂ ∂
η αβ − i n Âα (y 0
, . . . , y 3
) − i nÂβ (y 0 ,
∂y α ∂y β
(2.3.8)
)
. . . , y 3 ) ϕ̂ (y 0 , . . . , y 3 ) + m2 ϕ̂ (y 0 , . . . , y 3 ) = 0.
( ) ( )
∂ ∂ ∂xβ
where A = Âα dy α so that Âα = A ∂y α = A ∂xβ ∂y α
=
( )
Λα β A ∂x∂ β = Λα β Aβ .
Remark: The essential, but rather camouflaged, issue here is that it is up to
us to specify what the wavefunction is to be in the new coordinates (i.e., how ϕ
transforms under L+↑ ) and that, in order to satisfy the requirements of special
relativity, we must do this in such a way that it satisfies “the same equation in
the new coordinate system.” In this case, the wavefunction transforms trivially
 2.4. Spin One-Half Electrodynamics 71

(physicists say, as a scalar), i.e., we simply rewrite ϕ in the new coordinates,

but there is no reason to expect such simplicity in general (compare this, for
example, with the transformation of a vector field on R3 under rotation of
the coordinate system).

Having spelled out exactly what we mean by relativistic invariance in this

case, the proof is just a simple calculation. First observe that
( )
∂
− i nÂβ (y , . . . , y ) ϕ̂ (y 0 , . . . , y 3 )
0 3
∂y β
∂
= ϕ̂ (y 0 , . . . , y 3 ) − i nÂβ (y 0 , . . . , y 3 ) ϕ̂ (y 0 , . . . , y 3 )
∂y β
∂
= ϕ (Λa 0 y a , . . .) − i nÂβ (y 0 , . . .) ϕ (Λa 0 y a , . . .)
∂y β
∂ ∂xδ
= ϕ (Λ a
0 a
y , . . .) − i n Λβ δ Aδ (Λa 0 y a , . . .) ϕ (Λa 0 y a , . . .)
∂xδ ∂y β
( )
∂
= Λβ δ − i nA δ (Λa
0 a
y , . . .) ϕ (Λa 0 y a , . . .).
∂xδ

Similarly,
( )( )
∂ ∂
− i nÂα (y , . . . , y )
0 3
− i nÂβ (y , . . . , y ) ϕ̂ (y 0 , . . . , y 3 )
0 3
∂y α ∂y β
( )( )
∂ ∂
γ
= Λα Λβ δ
− i nAγ (Λa y , . . .)
0 a
− i nAδ (Λa y , . . .)
0 a
∂xγ ∂xδ
× ϕ (Λa 0 y a , . . .).

Since η αβ Λα γ Λβδ = η γδ , we have

( )( )
∂ ∂
η αβ − i n Âα (y 0
, . . . , y 3
) − i n Âβ (y 0
, . . . , y 3
) ϕ̂(y 0 , . . . , y 3 )
∂y α ∂y β
( )( )
∂ ∂
= η γδ − i nA γ (x 0
, . . . , x3
) − i nA δ (x 0
, . . . , x3
) ϕ(x0 , . . . , x3 )
∂xγ ∂xδ

and from this it is clear that (2.3.7) implies (2.3.8).

2.4 Spin One-Half Electrodynamics

Electrons, protons and neutrons (unlike the π mesons of Section 2.3) have
spin s = 12 and so, according to the generally accepted scheme of (nonrela-
tivistic) quantum mechanics, should have a wavefunction with 2 ( 12 ) + 1 = 2
components. We intend to say just a few words on the rationale behind this
72 2. Physical Motivation

and then refer those who are interested to the elegant, and quite accessible, ac-
count of these phenomena in Volume III of The Feynman Lectures on Physics
[Fey].
The classical Bohr picture of an atom (negatively charged electrons revolving
around a positively charged nucleus) suggests that an orbiting electron actu-
ally constitutes a tiny current loop. Such a current loop produces a magnetic
field which, at large distances, is the same as that of a magnetic dipole (lo-
cated at the center of the loop and perpendicular to the plane of the loop).
Such a dipole has a magnetic moment (a vector describing its orientation and
strength). Consequently, an electron in an atom has associated with it an “or-
bital magnetic moment.” Now, magnetic moments behave in interesting and
well-understood ways when subjected to external magnetic fields. In 1922 (just
before the advent of quantum mechanics), Stern and Gerlach carried out an
experiment designed to detect these effects for the orbital magnetic moment
of an electron. From the point of view of classical physics (the only point of
view available at the time), the results were quite shocking. Somewhat later,
the quantum mechanics of Schroedinger and Heisenberg, when applied to the
orbital magnetic moment of the electron, provided a qualitative, but not quan-
titative explanation of the outcome. Finally, it was suggested by Uhlenbeck
and Goudsmit that this discrepancy (and various others associated with the
anomalous Zeeman effect and the splitting of certain spectral lines) could
be accounted for if one assumed that the electron had associated with it an
additional magnetic moment, not arising from its orbital motion, but rather
from a sort of “intrinsic” angular momentum or “spinning” of the electron. The
suggestion was not that an electron actually spins on some axis in the same
way that the earth does on its, but rather that it possesses some intrinsic
property (called “spin”) that manifests itself in an external magnetic field by
mimicing the behavior of the magnetic moment of a spinning charged ball.
This intrinsic magnetic moment vector, however, must be of a rather peculiar
sort that one could only encounter in quantum mechanics. The Stern-Gerlach
experiment suggested that its component in any spatial direction could take
on only one of two possible values (±1 with the proper choice of units). This
has the following consequence. Let us select (arbitrarily) some direction in
space (say, the z-direction of some coordinate system). The intrinsic mag-
netic moment of an electron has, at each point, a z-component σz that can
take on one of the two values ±1. Which value it has will determine how the
electron responds to certain magnetic fields and so a complete description of
the electron’s wavefunction must contain this information. More precisely, the
wavefunction must be regarded as a function of not only x, y, z and t, but of
σz as well.
ψ = ψ(x, y, z, t, σz )
However, since σz can assume only the two values ±1, such a wavefunc-
tion is equivalent to a pair of functions ψ1 (x, y, z, t) = ψ (x, y, z, t, 1) and
ψ2 (x, y, z, t) = ψ (x, y, z, t, −1). It is convenient to put these two together into
 2.4. Spin One-Half Electrodynamics 73

a column vector and adopt the point of view that an electron (or any spin
one-half particle) has a two-component wavefunction
( )
ψ1
ψ= .
ψ2

The probabilistic interpretations of the wavefunctions in quantum ∫ mechanics

now run as follows: For each fixed t and each region R ⊆ R3 , R ψ1 ψ̄1 is the
probability at time t that the particle∫ will be detected in R with its spin vector
directed in the positive z-direction; R ψ2 ψ̄2 is the probability∫ that it will be
found in R with its spin vector in the negative z-direction; R ψ1 ψ̄1 + ψ2 ψ̄2
is the probability that it will be found in R at all.
Pauli formulated a theory of the electron along the lines suggested above
and, although this work was later superceded by that of Dirac, it provides
an instructive warm-up and we will spend a few moments outlining some of
its essential features. We will consider only stationary states Ψ(x, y, z, t) =
ψ(x, y, z) e−i ωt and will focus our attention on the spatial part ψ (x, y, z) (a
proper treatment of time dependence should be relativistic, which Pauli’s
theory was not). Thus, we are interested in the two-component object
( )
ψ1 (x, y, z)
ψ (x, y, z) = .
ψ2 (x, y, z)

Here x, y and z presumably represent “standard” coordinates in R3 and we

(along with Pauli) will require that our theory be independent of which par-
ticular oriented, orthonormal basis for R3 gives rise to these coordinates, i.e.,
that it be invariant under the rotation group SO(3). We are therefore ( led )
ψ1
to ask the following question: How is the two-component wavefunction ψ2
transformed if the coordinate system is subjected to the rotation correspond-
ing to some element of SO(3)?
To answer this question let us first review the manner in which such issues
are addressed in classical physics and then decide what, if any, modifications
are required by quantum mechanics. If the coordinate system is subjected to
a rotation
( R ) ∈ SO(3), then the state of the electron will be described by
ψ1′
a pair ψ2′
, where each ψi′ is a function of the new coordinates x′ , y ′ and
z ′ . Now, it is conceivable that each ψi′ is simply ψi expressed in terms of
these new coordinates. This is, indeed, what we found to be the case for the
Klein-Gordon equation
( ′) (see
( ) page 83). However, it is also conceivable that the
ψ1 ψ1
dependence of ψ′ on ψ2 is more analogous to the transformation law for
2
an ordinary vector, or tensor field on R3 . We will not prejudge this issue here,
but will simply make a few tentative assumptions about this dependence. We
assume, for example, that ψ1′ and ψ2′ are linear functions of ψ1 and ψ2 . The
reason is that, presumably, any differential equations one might arrive at for
74 2. Physical Motivation

the wavefunction will be generalizations of the (linear) Schroedinger equation

and should (at least as a first guess) themselves be linear. Thus, for some 2 × 2
complex matrix T = T (R) we have
( ) ( )
ψ1′ ψ1
= T (R)
ψ2′ ψ2

(here it is understood that both sides have been written in terms of one of
the two coordinate systems, x, y, z or x′ , y ′ and z ′ ). Assuming also (for the
moment) that the wavefunction is uniquely determined in each coordinate
system we find that a rotation by R2 ∈ SO(3) followed by a rotation by
R1 ∈ SO(3) must have the same effect as the rotation R1 R2 ∈ SO(3). Thus,
we must have
T (R1 R2 ) = T (R1 )T (R2 ).
Similarly, each T (R) must be invertible and satisfy

T (R−1 ) = (T (R))−1

What we find then is that the rule T which associates ) every R ∈ SO(3)
( with
ψ1
the corresponding transformation matrix T (R) for ψ2 is a homomorphism
into the group of invertible, 2 × 2, complex matrices. Identifying this latter
group with GL(C2 ) we find that T is a representation of SO(3) on C2 . The
reason this information is useful is that all of the representations of SO(3)
are known. These are usually described somewhat indirectly as follows: In
Appendix A of [N4] it is shown that SU (2) is the (double) covering group of
SO(3). More precisely, there exists a smooth, surjective group homomorphism

Spin : SU (2) −→ SO(3)

with kernel ± ( 10 01 ) and with the property that each point of SO(3) has an
open neighborhood V whose inverse image under Spin is a disjoint union of
(two) open sets in SU (2), each of which is mapped diffeomorphically onto V
by Spin. Now, consider a representation

h : SO(3) −→ GL(V)

of SO(3). Composing with Spin then gives a representation of SU (2).

Every representation of SO(3) “comes from” a representation of SU (2).
The converse is not true, however. That is, a given representation h̃ :
SU (2) −→ GL(V) of SU (2) will not induce a representation of SO(3) un-
less h̃(−g) = h̃(g) for every g ∈ SU (2). The representations of SU (2) that do
not satisfy this condition are sometimes referred to in the physics literature
as “2-valued representations of SO(3),” although they are not representations
of SO(3) at all, of course.
 2.4. Spin One-Half Electrodynamics 75

SU(2)

Spin h=h
° Spin

SO(3) GL( )
h

Now, it is quite easy to write out some rather obvious representations of

SU (2). Let C[z1 , z2 ] be the vector space of all polynomials with complex
coefficients in the two unknowns z1 and z2 . For each k = 0, 1, . . ., let Vk
be the subspace consisting of those polynomials that are homogeneous of
degree k:
c0 z1k + c1 z1k−1 z2 + · · · + ck z2k .

( for V
k−r r
A basis )k consists of all polynomials z1 z2 , r = 0, 1, . . . , k. Each
g = αγ βδ in SU (2) gives rise to a linear transformation on Vk which
carries z1k−r z2r onto (z1′ )k−r (z2′ )r , where

( ) ( )( )
z1′ α γ z1
= .
z2′ β δ z2

These linear transformations are clearly invertible and so the assignment, to

each g ∈ SU (2), of the corresponding element of GL(Vk ) is a representation
of SU (2), usually denoted
k
D 2 : SU (2) −→ GL(Vk ),

and called the spin-j representation, where j = k2 . One can show (see
[vdW]) that each of these representations is irreducible (i.e., that there is
k
no proper subspace of Vk that is invariant under every D 2 (g), g ∈ SU (2))
and that every irreducible representation of SU (2) with complex representa-
tion space is equivalent to one of these (two representations D1 : G −→ GL(V1 )
and D2 : G −→ GL(V2 ) of a group G are equivalent if it is possible to choose
bases for V1 and V2 so that, for each g ∈ G, the matrices of D1 (g) and
D2 (g) are the same). Furthermore, any representation of SU (2) can be con-
structed from these irreducible representations by forming finite direct sums
(the direct sum of D1 : G −→ GL(V1 ) and D2 : G −→ GL(V2 ) is the repre-
sentation D1 ⊕ D2 : G −→ GL(V1 ⊕ V2 ) defined by (D1 ⊕ D2 )(g)(v1 , v2 ) =
(D1 (g)(v1 ), D2 (g)(v2 )). In effect, we now have all of the representations of
SU (2).
76 2. Physical Motivation

The polynomials have now served their purpose and it will be convenient to
note that Vk has complex dimension k + 1 and so can be identified with Ck+1
by identifying z1k−r z2r , r = 0, 1, . . . , k, with the standard basis for Ck+1 . The
k
linear transformations D 2 (g) can therefore be identified with (k + 1) × (k + 1)
complex matrices. For example, k = 0 gives the trivial representation of SU (2)
on C (D0 (g) = (1) for each g ∈ SU (2)), while k = 1 gives the identity
1
representation of SU (2) on C2 (D 2 (g) = g for every g ∈ SU (2)). Note that
1
D 2 is the only irreducible representation of SU (2) on C2 . The only other way
to get a representation of SU (2) on C2 is to form the direct sum of two copies
of D0 : ( )
1 0
(D ⊕ D )(g) = (1) ⊕ (1) =
0 0
.
0 1
1
This, of course, leaves everything in C2 alone. Finally notice that D 2 (−g) =
1
−D 2 (g) and (D0 ⊕ D0 )(−g) = (D0 ⊕ D0 )(g) so only this second example
descends to a representation of SO(3) (the trivial representation of SO(3) on
C2 ).
The situation we have just described would seem to present us with some-
thing of a dilemma. To ensure the rotational invariance of Pauli’s theory of
) SO(3) on C that would
2
the electron we were led to seek a representation
( of
ψ1
transform the two-component wavefunctions ψ2 when the coordinate sys-
tem is rotated. We find now that there is only one such (D0 ⊕ D0 ) and this
is the trivial
( ′ representation.
) Under
( this
) representation the transformed wave-
ψ
function ψ1′ would simply be ψ 1
ψ2 written in terms of the new coordinates.
2
However, this is clearly not consistent with the phenomenon (spin one-half)
which led us to two-component wavefunctions in the first place. Recall that
ψ1 (x, y, z) = ψ(x, y, z, 1) and ψ2 (x, y, z) = ψ(x, y, z, −1), where ±1 are the
possible z-components of the intrinsic magnetic moment of the electron. A
rotation which reverses the ( direction
) of the z-axis must interchange ψ1 and
ψ2 and so cannot leave ψ2 unchanged. Thus, D0 ⊕ D0 is not consistent
ψ1

with the structure we are attempting to model. Must we conclude then that
Pauli’s proposal is doomed to failure?
To extricate ourselves from this dilemma we must understand that there
is an essential feature of quantum mechanics that requires an adjustment
in the classical picture we painted earlier (pages 86–87). Our conclusion
that the transformation matrices T (R) satisfy T (R1 R2 ) = T (R1 )T (R2 ) and
(T (R))−1 = T (R−1 ) and therefore give rise to a representation of SO(3) fol-
lowed from the assumption that the wavefunction is uniquely determined in
each coordinate
( ) system. This, however, is not (quite) the case. For example,
both ± ψ ψ2
1
represent the same state of our electron since an overall sign
change has no effect on the probabilities described earlier (page 86) and all of
 2.4. Spin One-Half Electrodynamics 77

the physical content of the wavefunction is contained in such probabilities. It

follows, in particular, that for a given R ∈ SO(3), the transformation matrix
T (R) is determined only up to sign. Physicists would call T a “2-valued rep-
resentation” of SO(3). This makes no sense, of course, but we have just seen
exactly how one can make sense of it. The appropriate tactic is to “go to the
covering space,” i.e., to represent a rotation, not by an element of SO(3),
( but
)
ψ1
rather by two elements of SU (2) and seek the transformation law for ψ2
among the representations of SU (2). No longer being constrained to select a
representation of SU (2) that descends to SO(3), we have available one more
1 ( )
option, i.e., D 2 . With this choice each g = αγ βδ in SU (2) would transform
( )
the wavefunction ψ ψ2
1
as follows:
( ) ( )( )
ψ1′ α β ψ1
= .
ψ2′ γ δ ψ2
()
Notice, in particular, that if g = −i0 −i0 , then Spin (±g) is the rotation
about the x-axis through π (Appendix A, [N4]) and therefore reverses the
z-axis. Since ( )( ) ( )
0 −i ψ1 ψ2
= −i
−i 0 ψ2 ψ1
( ) ( ) 1
and −i ψ 2 ψ2
ψ1 represents the same state as ψ1 , the representation D , unlike
2

D0 ⊕ D0 , is at least consistent with our proposed model of spin one-half.

The program we have just described can leave one with the queasy feeling of
ambiguity. Suppose that one is given a frame in R3 and wants to rotate by R ∈
SO(3) to a new frame and thereby( )a new representation ( ) of the 1wavefunction.
( )
1 1
If Spin (±g) = R, then D (g) ψ2 and D (−g) ψ
2
ψ1 2
ψ2
1
= −D 2 (g)
ψ1
ψ2 are
physically equivalent and that’s fine, but it is difficult not to ask oneself, “But,
really, which is it?” The answer, interestingly enough, resides in the topologies
of SU (2), SO(3) and the spinor map. One is forced to regard a rotation of
frames in R3 not as an instantaneous jump from one to the other, but as a
physical process that begins with one frame and continuously rotates the axes
to the new position. The transformation law for the wavefunction depends not
only on the end result of the rotation, but also on “how you got there.” For
example, the matrix  
1 0 0
 
0 cos t − sin t 
 
0 sin t cos t

represents a rotation through t radians about the x-axis. As t varies from

0 to 2π one has a continuous sequence of rotations (i.e., a curve R1 (t) in
78 2. Physical Motivation

SO(3)) that represents the physical process of rotating a frame through one
complete turn (360◦ ) about its x-axis (identify each element of SO(3) with
the configuration of the axes that would result from applying that rotation
to the initial configuration). The curve R2 (t) in SO(3) defined by the same
formula, but with 0 ≤ t ≤ 4π represents a rotation about the x-axis through
720◦ . Both R1 and R2 begin and end with the same configuration, but there
is a real difference, both physically and mathematically.
Spin : SU (2) −→ SO(3) is a covering space (Exercise A.13, [N4]) and cov-
ering spaces have the property that curves in the covered space lift uniquely to
curves in the covering space once an initial point is selected (Corollary 1.5.13,
[N4]).

In particular, given a curve in SO(3) (representing a continuous rotation of

one frame into another) and a choice of either g or −g above the initial point
(frame), there is a uniquely determined element of SU (2) above the terminal
point that represents the transformation law that gives the wavefunction in
the new frame. No ambiguity at all!
For the curves R1 and R2 in SO(3), both of which begin and end at the
identity, one can write out the lifts g1 and g2 starting at ( 10 01 ) explicitly. Both
are given by
 
cos 2t − i sin 2t
 
 
−i sin 2t cos 2t

but with 0 ≤ t ≤ 2π for g1 and 0 ≤ t ≤ 4π for g2 (see Appendix A, [N4]). But

notice that g1 is a path in SU (2) from ( 10 01 ) to − ( 10 01 ), whereas g2 begins and
 2.4. Spin One-Half Electrodynamics 79

ends at ( 10 01 ). Thus, a rotation of the frame through 360◦ changes the sign
of the wavefunction, but a rotation through 720◦ leaves the sign unchanged,
even though both rotations begin and end with the same configuration of the
axes. Mathematically, the difference between R1 and R2 is that they represent
two different homotopy classes in π1 (SO(3)) ∼ = Z2 . R2 is nullhomotopic since
it is Spin ◦g2 and g2 is a loop at ( 10 01 ) in SU (2) ∼
= S 3 , but R1 is not because
it lifts to a path g1 from ( 0 1 ) to − ( 0 1 ), in SU (2).
1 0 1 0

The next step in this program ( )would be to look at the differential equa-
tions proposed by Pauli for ψ 1
ψ2 and decide whether or not they assume
the same form when the coordinate system is rotated and the wavefunction is
1
transformed by D 2 (they do!). Since Pauli’s theory was eventually abandoned
(because it is not relativistically invariant) we shall not pursue this here, but
will instead turn to the profoundly successful alternative proposed by Dirac.

Remark: Before moving forward with our discussion of spin one-half we

observe that the arguments we presented (on pages 85–86) for modeling such
a particle with a wavefunction having two components really had nothing
specific to do with spin one-half. The essential feature that led to the dou-
bling of the number of components was the existence of an “internal struc-
ture” that could be represented by a parameter (σz for the electron) that
could assume precisely two values. There are other examples of this sort
of thing, e.g., the “isotopic spin” parameter of a nucleon which determines
(in the absence of electromagnetic fields) whether the particle is a proton
or a neutron. The two components of the nucleon wavefunction then rep-
resent the “proton part” and the “neutron part” of the doublet. Indeed, it
was this example that provided the initial motivation for Yang-Mills theory
([YM]).

Dirac set out to construct a relativistically invariant equation that would

be satisfied by the wavefunction of a spin one-half particle. He reasoned
that his equation should, in some sense, “imply” the Klein-Gordon equa-
tion since, as we have noted, this is just the quantized version of the
relativistic energy-momentum relation E 2 = p⃗ 2 + m2 . However, he also
sought to remedy certain physical problems associated with the appearance
of the second time derivative in the Klein-Gordon equation (for a discus-
sion of these see Chapter 6 of [Hol]). Thus, Dirac sought a first order linear
equation which, upon iteration, yielded the Klein-Gordon equation. Some-
what more precisely, Dirac was in the market for a first order differential
operator
̸ = γ 0 ∂0 + γ 1 ∂1 + γ 2 ∂2 + γ 3 ∂3 = γ α ∂α
D (2.4.1)
∂
(∂α = ∂xα , α = 0, 1, 2, 3) such that, when applied to the equation

̸ ϕ = −i mϕ,
D (2.4.2)
80 2. Physical Motivation

the result is the free Klein-Gordon equation

η αβ ∂α ∂β ϕ = −m2 ϕ. (2.4.3)

The problem is to determine the γ α , α = 0, 1, 2, 3, so that this will be the

̸ in (2.4.1) to both sides of (2.4.2).
case. Apply the operator D
̸ (̸D ϕ) = D
D ̸ (−i mϕ)
(γ ∂α ) (γ β ∂β ϕ) = −i m D
α
̸ ϕ
γ α γ β ∂α (∂β ϕ) = −i m (−i mϕ)
(γ α γ β ∂α ∂β ) ϕ = −m2 ϕ (2.4.4)

This will agree with (2.4.3) if

γ α γ β ∂α ∂β = η αβ ∂α ∂β ,

i.e., if
γ α γ β = η αβ , α, β = 0, 1, 2, 3. (2.4.5)
Now, (2.4.5) clearly cannot be satisfied if the γ α are taken to be numbers
(none can be 0 since η αα = ±1, but γ α γ β = 0 if α ̸= β). Dirac’s idea was to
allow ϕ to have more than one complex component and interpret (2.4.5) as
matrix equations (one for each α, β = 0, 1, 2, 3). Specifically, if
 
ϕ1
 . 
ϕ=  . 
. 
ϕn

and ∂β ϕ is computed entrywise, then D ̸ ϕ = γ α ∂α ϕ would require the γ α to

have n columns (the reason we do not immediately follow Pauli’s lead and take
n = 2 will become clear shortly). To iterate the operator and define D ̸ (̸Dϕ)
requires that the γ α have n rows as well. Now, (2.4.5) must be interpreted as
matrix equations
γ α γ β = η αβ id, α, β = 0, 1, 2, 3,
where id is the n × n identity matrix. Notice, however, that each η αβ id
is a symmetric matrix. To accommodate this fact we observe that, because
∂α ∂β ϕ = ∂β ∂α ϕ, (2.4.4) can be written
1( α β )
γ γ + γ β γ α ∂α ∂β ϕ = −m2 ϕ
2
so that we might just as well have written (2.4.5) as
1 α β
(γ γ + γ β γ α ) = η αβ
2
and so the matrix conditions we are currently trying to satisfy may be taken
to be
γ α γ β + γ β γ α = 2η αβ id, α, β = 0, 1, 2, 3. (2.4.6)
 2.4. Spin One-Half Electrodynamics 81

Finding square matrices that satisfy (2.4.6) is actually a problem familiar

to algebraists. What we are looking for here is a matrix representation of the
Clifford algebra of (R4 , ⟨ , ⟩), where ⟨ , ⟩ is the Minkowski inner product.
There are many solutions, but the smallest n for which such matrices can be
found is n = 4. Thus, Pauli’s choice of n = 2 cannot succeed in this context
(we will eventually have to sort out how to reconcile this with the arguments
we presented earlier to the effect that a spin one-half wavefunction should
have two components). It is, in fact, easy to write down a set of 4 × 4 matrices
satisfying conditions (2.4.6). Letting
( ) ( ) ( )
0 1 0 −i 1 0
σ1 = , σ2 = , and σ3 =
1 0 i 0 0 −1

be the Pauli spin matrices and taking σ0 to be the 2 × 2 identity matrix, one
easily verifies the usual commutation relations

σi2 = σ0 , i = 1, 2, 3
(2.4.7)
σi σj = −σj σi , i, j = 1, 2, 3, i ̸= j.

Now, we define  
( ) 0 0 1 0
 
0 σ0 0 0 0 1
γ0 = = 
σ0 0 1 0 0 0
0 1 0 0
 
( ) 0 0 0 −1
 
0 −σ1 0 0 −1 0
γ1 = = 
σ1 0 0 1 0 0
1 0 0 0
 
( ) 0 0 0 i
 
0 −σ2 0 0 −i 0
γ2 = = 
σ2 0 0 −i 0 0
i 0 0 0
 
( ) 0 0 −1 0
 
0 −σ3 0 0 0 1
γ3 = = .
σ3 0 1 0 0 0
0 −1 0 0
It is now a simple matter to verify that the conditions in (2.4.6) are satisfied
by these matrices, e.g.,
82 2. Physical Motivation

γ1γ2 + γ2γ1
( )( ) ( )( )
0 −σ1 0 −σ2 0 −σ2 0 −σ1
= +
σ1 0 σ2 0 σ2 0 σ1 0
( ) ( )
−σ1 σ2 0 −σ2 σ1 0
= +
0 −σ1 σ2 0 −σ2 σ1
( )
−(σ1 σ2 + σ2 σ1 ) 0
=
0 −(σ1 σ2 + σ2 σ1 )
 
1 0 0 0
( )  
0 0  0 1 0 0 
= [4pt] = 2η 12 
 0 0 1

0 0  0 

0 0 0 1

and
( )( )
1 1 1 1 1 1 0 −σ1 0 −σ1
γ γ + γ γ = 2γ γ = 2
σ1 0 σ1 0
( ) ( )
−σ12 0 σ0 0
=2 = −2
0 −σ12 0 σ0
 
1 0 0 0
 

11 0 1 0 0
= 2η  

 0 0 1 0 
0 0 0 1

etc.
Remark: There are many other possible choices for γ 0 , γ 1 , γ 2 and γ 3 ,
many of which are used in the physics literature. Indeed, for any nonsingular
matrix B, one can replace each γ α by Bγ α B −1 and obtain another set of “Dirac
matrices” satisfying (2.4.6). Conversely, one can show (see, e.g., pages 104–106
of [Gre]) that any set of 4 × 4 matrices satisfying (2.4.6) differs from our
choice by such a similarity transformation. Algebraically, this means that, up
to equivalence, there is only one representation of the Clifford algebra of R1,3
by 4 × 4 matrices. The choice we have made is called the Weyl, or chiral
representation.
With γ 0 , γ 1 , γ 2 and γ 3 the 4 ×4 matrices described above, the wavefunction
for our spin one-half particle has four components
 2.4. Spin One-Half Electrodynamics 83
 
ϕ1
 
 ϕ2 
ϕ= 
 ϕ3 
ϕ4

and the Dirac equation (expressed in standard coordinates on R1,3 ) is

D̸ ϕ = −i mϕ
γ α ∂α ϕ = −i mϕ
  
0 0 ∂0 − ∂3 −∂1 + i ∂2 ϕ1
  
 0 0 −∂1 − i ∂2 ∂0 + ∂3  ϕ2 
  
 ∂0 + ∂3 ∂1 − i ∂2 0 0  ϕ3 
∂1 + i ∂2 ∂0 − ∂3 0 0 ϕ4
 
ϕ1
 
 ϕ 
= −i m  2  .
 ϕ3 
ϕ4
We will discuss the relativistic invariance of the Dirac equation shortly, but
first we would like to show that, in the massless (m = 0) case, there is a
simpler solution to the problem of finding a first order operator D = γ α ∂α
which, when applied to the equation
Dϕ = 0

yields the m = 0 Klein-Gordon equation

η αβ ∂α ∂β ϕ = 0.
For this purpose we write this last equation as
∂0 ∂0 ϕ = δ ij ∂i ∂j ϕ. (2.4.8)
Now, Dϕ = 0 can be written
γ 0 ∂0 ϕ = −γ i ∂i ϕ,
or, assuming γ 0 is invertible,
∂0 ϕ = −µi ∂i ϕ (µi = (γ 0 )−1 γ i , i = 1, 2, 3).
Then
∂0 ∂0 ϕ = ∂0 (−µi ∂i ϕ) = −µi ∂0 ∂i ϕ
= −µi ∂i ∂0 ϕ = −µi ∂i (−µj ∂j ϕ)

∂0 ∂0 ϕ = µi µj ∂i ∂j ϕ. (2.4.9)
84 2. Physical Motivation

Now compare (2.4.8) and (2.4.9). Again, µi µj = δ ij cannot be satisfied by

numbers so we rewrite (2.4.9) as
1 i j
∂0 ∂0 ϕ = (µ µ + µj µi )∂i ∂j ϕ, (2.4.10)
2
and seek matrices satisfying
µi µj + µj µi = 2δ ij id, i, j = 1, 2, 3. (2.4.11)
In the terminology of algebra, matrices satisfying the conditions (2.4.11)
constitute a matrix representation of the Clifford algebra of (R3 , ⟨ , ⟩), where
⟨ , ⟩ is the usual positive definite inner product on R3 . Notice that the con-
ditions (2.4.11) coincide with the commutation relations (2.4.7) for the Pauli
spin matrices σ1 , σ2 and σ3 . In particular, there is now a 2 ×2 solution to the
problem so, in the massless case, our wavefunction can have two components
( )
ϕ3
ϕ=
ϕ4

(the reason for the peculiar numbering will become clear soon). One obtains an
operator D =(γ α ∂)α of the required type by taking, for example,
10
γ 0 = σ0 = 0 1 and γ i = µi = −σi , i = 1, 2, 3. Thus,
( ) ( ) ( ) ( )
1 0 0 −1 0 i −1 0
D= ∂ + ∂1 + ∂ + ∂
0 1 0 −1 0 −i 0 2 0 1 3
( )
∂0 − ∂3 −∂1 + i ∂2
= .
−∂1 − i ∂2 ∂0 + ∂3

The corresponding equation Dϕ = 0 then becomes

( )( ) ( )
∂ 0 − ∂3 −∂1 + i ∂2 ϕ3 0
= (2.4.12)
−∂1 − i ∂2 ∂0 + ∂3 ϕ4 0

and is known as the Weyl neutrino equation. Notice that this is just what
one would obtain from the Dirac equation with m = 0 and a wave- function
of the form  
0
 0 
 
 ϕ3  .
ϕ4
For future reference we note also that the m = 0 Dirac equation for a wave-
function of the form  
ϕ1
 ϕ 
 2 
 0 
0
 2.4. Spin One-Half Electrodynamics 85

reduces to ( )( ) ( )
∂0 + ∂3 ∂1 − i ∂2 ϕ1 0
= (2.4.13)
∂1 + i ∂2 ∂0 − ∂3 ϕ2 0
and that the coefficient matrix in (2.4.13) is the formal conjugate, transposed
inverse of the coefficient matrix in (2.4.12). The significance of these obser-
vations will emerge in our discussion of the transformation properties of the
wavefunctions and the corresponding invariance properties of the equations.
Now we return to the issue of the Lorentz invariance of the Dirac equation.
The problem, as it was for the Klein-Gordon equation (page 82), is to show
that the Dirac equation has the same form in any other coordinate system
y 0 , y 1 , y 2 , y 3 for R1,3 , related to the standard coordinates x0 , x1 , x2 , x3 by

y α = Λαβ xβ , α = 0, 1, 2, 3,

where Λ = (Λαβ ) ∈ L+↑ . However, since the Dirac wavefunction has four
complex components, this will require finding a representation T : L+↑ −→
GL(C4 ) of L+↑ on C4 which, if taken to be the transformation law for the
wavefunction, preserves the form of the Dirac equation (cf., the discussion
of the two-component Pauli theory on pages 86–93). As was the case for
SO(3) in the Pauli theory, it so happens that all of the representations L+↑
are known, that they are most conveniently described in terms of a two- fold
covering group of L+↑ and that, because of the nature of a quantum mechanical
wavefunction, it is actually the representations of the covering group that do
not descend to L+↑ that turn out to be of most interest. We begin with a brief
summary of the relevant results (see Chapter 3 for more details).
Identifying L+↑ with a subset of R16 one finds that it is a submanifold
diffeomorphic to SO(3) × R3 and so is a 6-dimensional Lie group containing
SO(3) as a closed subgroup. We denote by SL(2, C) the group of 2×2 complex
matrices with determinant one. Identifying SL(2, C) with a subset of C4 = R8
one finds that it is a submanifold diffeomorphic to S 3 × R3 and so it is also
a 6-dimensional (simply connected) Lie group. Note that SU (2) is a closed
subgroup of SL(2, C). Now, we have already described a two-fold covering
map
Spin : SU (2) −→ SO(3)
and we wish now to show that this is, in fact, the restriction to SU (2) of a
two-fold covering map of SL(2, C) onto L+↑ , also denoted

Spin : SL(2, C) −→ L+↑ .

The construction of this map is carried out in detail in Section 1.7 of [N3]
so we will be brief. R1,3 can be identified with the linear space H of 2 × 2
86 2. Physical Motivation

complex Hermitian matrices

( )
x0 + x3 x1 − i x2
x=
x1 + i x2 x0 − x3

= x0 σ0 + x1 σ1 + x2 σ2 + x3 σ3 = xα σα ,

where the squared norm is taken to be the determinant. Observe that each of
the coordinates xα can be expressed as
1
xα = trace(σα x)
2
so that ∑3
1
x= trace(σα x)σα .
α=0
2

Now, for each g ∈ SL(2, C) we define a map Λg : H −→ H by

Λg (x) = gxḡ ⊤ .
⊤
Note that Λg (x) is, indeed, in H because Λg (x) = (gxḡ ⊤ )⊤ = (ḡx̄g ⊤ )⊤ =
gxḡ ⊤ = Λg (x). Also note that, for g ∈ SU (2) ⊆ SL(2, C), ḡ ⊤ = g −1 . Now,
Λg is surely linear and satisfies det(Λg (x)) = det(gxḡ ⊤ ) = det(x) so it pre-
serves the Minkowski inner product on H = R1,3 . Thus, Λg is an orthogonal
transformation and, with a bit more work (page 77, [N3]), one can show that
it is proper and orthochronous. Now let
Λg (x) = y α σα .
Then
1 1
yα = trace(σα Λg (x)) = trace(σα gxḡ ⊤ )
2 2
1 β ⊤
= trace(σα g(x σβ )ḡ )
2
1
= trace(σα gσβ ḡ ⊤ )xβ
2
= Λαβ xβ

where
1
Λαβ = trace(σα gσβ ḡ ⊤ ), α, β = 0, 1, 2, 3.
2
Thus, (Λαβ )α,β=0,1,2,3 is in L+↑ and we define Spin: SL(2, C) −→ L+↑ by

Spin(g) = (Λαβ )α, β = 0,1,2.3

for each g ∈ SL(2, C). One then shows that Spin is a two-fold covering group
for L+↑ and that its restriction to SU (2) agrees with the map of the same
 2.4. Spin One-Half Electrodynamics 87

name discussed on page 88. Before proceeding we will record one additional
fact that we will need shortly. Notice that
1 1
Λαβ = trace(σα gσβ ḡ ⊤ ) = trace(σβ (ḡ ⊤ σα g))
2 2
so that
∑
3
ḡ ⊤ σα g = Λαβ σβ . (2.4.14)
β=0

Now we proceed just as we did for SO(3) in our discussion of the Pauli
theory. Consider a representation
h : L+↑ −→ GL(V)
of L+↑ . Composing with Spin then gives a representation of SL(2, C).

SL(2, C)

Spin h = h ° Spin

+ GL( )
h

Thus, every representation of L+↑ comes from a representation of SL(2, C),

but conversely, a representation of SL(2, C) will descend to a representation
of L+↑ if and only if it takes the same value at ±g for each g ∈ SL(2, C) (all
others are of the “2-valued” variety).
The representations of SL(2, C) are all known and are described in some
detail in Section 3.1 of [N3]. We will limit ourselves to a brief discussion of just
those items that are relevant to our present context. There are a few obvious
representations of SL(2, C) on Cn . By sending every element of SL(2, C) to
the n × n identity matrix one obtains the trivial representation which leaves
everything in Cn fixed.( When ) n = 2 one also has the identity representation
which sends each g = γ δ in SL(2, C) to the linear transformation on C2
αβ

defined by
( ) ( )( ) ( )
z1 α β z1 αz 1 + βz 2
−→ = .
z2 γ δ z2 γz 1 + δz 2

This representation is generally denoted

D( 2 ,0) : SL(2, C) −→ GL(C2 )

1

and called the left-handed spinor representation of SL(2, C). Identifying

the linear transformation D( 2 ,0) (g) on C2 with its matrix relative to the
1
88 2. Physical Motivation

standard basis for C2 one can write

1
D( 2 ,0) (g) = g

for each g ∈ SL(2, C). Another, somewhat less obvious, representation

D(0, 2 ) : SL(2, C) −→ GL(C2 )

1

of SL(2, C) on C2 sends each g to the linear transformation whose matrix

relative to the standard basis is the inverse of the conjugate transpose ḡ ⊤
of g (called the right-handed spinor representation of SL (2, C)). For
simplicity we again write

D(0, 2 ) (g) = (ḡ ⊤ )−1 .

1

Remarks: The map that sends g to (the linear transformation on C2 whose

matrix is) (g ⊤ )−1 is also a representation
( of)SL(2, C), but it is equivalent
( ) to
( 12 ,0) 0 1 α β
D . The reason is that, if B = −1 0 , then, for any g = γ δ in
SL(2, C),
BgB −1 = (g ⊤ )−1 .
1
It follows that D(0, 2 ) is equivalent to the conjugation representation
1 1
g −→ ḡ. D( 2 ,0) and D(0, 2 ) are, however, not equivalent as representa-
tions of SL(2, C). To see this, suppose there were a matrix B such that
BgB −1 = (ḡ ⊤ )−1 for each g ∈ SL(2, C). Then, in particular, we would have
trace(g)
( = trace(ḡ
) ) for each g ∈ SL(2,(C). This,
⊤ −1
) however, is not true for
−2i 0 − 12 i
∈ SL(2, C) since (ḡ ⊤ )−1 =
0
g= 0 1 . It is interesting to note,
2i 0 2i
⊤ −1
however, that g −→ g and g −→ (ḡ ) are equivalent as representations of
SU (2) since there ḡ ⊤ = g −1 so (ḡ ⊤ )−1 = g.
There is a precise sense in which all of the representations of SL(2, C) can
1 1
be constructed from D( 2 ,0) and D(0, 2 ) . Rather than describing this procedure
in general we will be content to illustrate it in the only case of real interest
to us, i.e., the representations of SL(2, C) on C4 . The most obvious way to
construct a representation on C4 from two representations on C2 is by forming
their direct sum. For example, we can define the representation

D( 2 ,0) ⊕ D(0, 2 ) : SL(2, C) −→ GL(C4 )

1 1

by
( )
( 12 ,0) (0, 12 ) g 0
D ⊕D (g) = ⊤ −1
,
0 (ḡ )
 2.4. Spin One-Half Electrodynamics 89

where all of the entries are 2 × 2 matrices and we are again identifying a
linear transformation on C4 with its matrix relative to the standard basis.
Similarly, one can define the direct sum of any such pair. A somewhat less
obvious procedure for building a representation on C4 is the tensor product
of two representations on C2 . For instance, the representation

D( 2 , 2 ) = D( 2 ,0) ⊗ D(0, 2 ) : SL(2, C) −→ GL(C4 )

1 1 1 1

( )
can be described as follows: For each g = αγ βδ ∈ SL(2, C), we define
( )
( 12 , 12 ) α(ḡ ⊤ )−1 β(ḡ ⊤ )−1
D (g) =
γ(ḡ ⊤ )−1 δ(ḡ ⊤ )−1
 
αδ̄ −αγ̄ β δ̄ −βγ̄
 
 −αβ̄ αᾱ −β β̄ β ᾱ
= .
 γ δ̄ −γγ̄ δ δ̄ −δγ̄ 
−γ β̄ γ ᾱ −δ β̄ δ ᾱ
1 1 1 1 1 1
Note that D( 2 , 2 ) (−g) = D( 2 , 2 ) (g) so D( 2 , 2 ) descends to a representation of
L+↑ on C4 (also denoted D( 2 , 2 ) ). One can show that D( 2 , 2 ) is equivalent to
1 1 1 1

the natural (vector) representation of L+↑ on R1,3 . Similarly, one can define
1 1
such representations as D(1,0) = D( 2 ,0) ⊗ D( 2 ,0) . One can show that these
1 1
are irreducible, whereas such things as D( 2 ,0) ⊕ D(0, 2 ) , of course, are not.
From our point of view the important fact is that we have just described all
of the representations of SL(2, C) on C4 (up to equivalence). Proving the
relativistic invariance of the Dirac equation therefore amounts to searching
among these few representations of SL(2, C) on C4 for one which, if adopted
as the transformation law for the 4-component Dirac wavefunction, will lead
to a transformed wavefunction that satisfies the same (Dirac) equation in the
transformed coordinate system.
Begin with the Dirac equation in standard coordinates x = (x0 , x1 , x2 , x3 )
on R1,3 . ( )
γ β ∂β + i m ϕ(x) = 0. (2.4.15)

Introduce new coordinates y = (y 0 , y 1 , y 2 , y 3 ) on R1,3 by

y α = Λαβ xβ , α = 0, 1, 2, 3,

where Λ = (Λαβ ) ∈ L+↑ and define ∂ˆα = ∂

∂y α , α = 0, 1, 2, 3. Then

∂ ∂y α ∂
∂β = = = Λαβ ∂ˆα .
∂xβ ∂xβ ∂y α
90 2. Physical Motivation

Let g ∈ SL(2, C) be such that Spin (±g) = (Λαβ ). Thus,

1 ( )
Λαβ = trace σα gσβ ḡ ⊤ , α, β = 0, 1, 2, 3.
2
Our objective is to find a representation

ρ : SL(2, C) −→ GL(C4 )

such that, if
ϕ̂(y) = ρ(g)(ϕ(Λ−1 y)),
then (2.4.15) implies
( )
γ α ∂ˆα + i m ϕ̂(y) = 0. (2.4.16)

We begin by simply rewriting (2.4.15) in the new coordinates

γ β ∂β ϕ + imϕ = 0
( )( )
γ β Λαβ ∂ˆα (ρ(g))−1 ϕ̂(y) + im(ρ(g))−1 ϕ̂(y) = 0
( )
γ β (ρ(g))−1 Λαβ ∂ˆα ϕ̂(y) + (ρ(g))−1 imϕ̂(y) = 0.

Multiply through by ρ(g) to obtain

ρ(g)γ β (ρ(g))−1 Λαβ ∂ˆα ϕ̂(y) + imϕ̂(y) = 0

(( ) )
ρ(g)γ β (ρ(g))−1 Λαβ ∂ˆα + im ϕ̂(y) = 0

which will agree with (2.4.16) if and only if

ρ(g)γ β (ρ(g))−1 Λαβ = γ α . (2.4.17)

This then is the condition that our representation ρ must satisfy in order to
preserve the form of the Dirac equation. We obtain a more convenient form
of this condition as follows:

(ρ(g))γ β (ρ(g))−1 Λαβ = γ α

(ρ(g))γ β (ρ(g))−1 = Λαβ γ α

γ β (ρ(g))−1 = (ρ(g))−1 Λαβ γ α

 2.4. Spin One-Half Electrodynamics 91

γ β = (ρ(g))−1 Λαβ γ α (ρ(g))

( )
γ β = Λαβ (ρ(g))−1 γ α (ρ(g))

Λαβ γ β = (ρ(g))−1 γ α ρ(g)

(ρ(g))−1 γ α ρ(g) = Λαβ γ β , α = 0, 1, 2, 3. (2.4.18)

At this point one need only check each of the representations ρ of SL(2, C)
on C4 described earlier in the hope of finding one that satisfies (2.4.18). One’s
1 1
hopes are not dashed. The winner is D( 2 ,0) ⊕ D(0, 2 ) , as we now show. With
γ 0 , γ 1 , γ 2 and γ 3 as on page 96 and
( )
g 0
ρ(g) =
0 (ḡ ⊤ )−1

we compute
( )
−1 0 0 g −1 σ0 (ḡ ⊤ )−1
(ρ(g)) γ ρ(g) = (2.4.19)
ḡ ⊤ σ0 g 0

and, for i = 1, 2, 3,
( )
−1 i 0 −g −1 σi (ḡ ⊤ )−1
(ρ(g)) γ ρ(g) = . (2.4.20)
ḡ ⊤ σi g 0

On the other hand,

Λαβ γ β = Λα0 γ 0 + Λα1 γ 1 + Λα2 γ 2 + Λα3 γ 3

( ) ( )
α 0 σ0 α 0 −σ1
=Λ 0 +Λ 1
σ0 0 σ1 0
( ) ( )
α 0 −σ2 α 0 −σ3
+Λ 2 +Λ 3
σ2 0 σ3 0

so
 
∑
3

 0 Λα0 σ0 − Λαi σi 
 
Λαβ γ β 
= 3 i=1 . (2.4.21)

 ∑ Λα σ 0 
β β
β=0

Now, for any α = 0, 1, 2, 3, the 21-block in (ρ(g))−1 γ α ρ(g) is ḡ ⊤ σα g and the

∑3
21-block in Λαβ γ β is β=0 Λαβ σβ and these are equal by (2.4.14). The equality
of the 12-blocks in (2.4.18) follows by taking inverses on both sides of (2.4.14).
92 2. Physical Motivation

More precisely, one observes that

(ḡ ⊤ σα g)−1 = g −1 σα−1 (ḡ ⊤ )−1 = g −1 σα (ḡ ⊤ )−1

and, from a brief calculation that we will leave for the reader,

∑3
 −1  
 0
− Λ0i σi ,

 Λ σ
0 0 α =0
∑3
 Λ β σβ  =
α i=1
.

 ∑3
β=0  −Λα σ +
 α
 0 0 Λ i σi , α = 1, 2, 3
i=1

With this we have established the Lorentz invariance of the Dirac equation.
1 1
The emergence of the representation D( 2 ,0) ⊕ D(0, 2 ) as the appropriate
transformation law for a Dirac wavefunction has interesting and important
consequences that we will briefly explore. Let us write the Dirac wavefunction
ϕ as
 
ϕ1 ( )
 
 ϕ2  ϕL
ϕ= = ,
 ϕ3  ϕR
ϕ4
( ) ( ) 1
where ϕL = ϕ1
ϕ2 and ϕR = ϕ3
ϕ4 . Since ϕ transforms according to D( 2 ,0) ⊕
1 1 1
D(0, 2 ) , ϕL and ϕR transform according to D( 2 ,0) and D(0, 2 ) , respectively.
Furthermore, the Dirac equation becomes a pair of coupled equations for ϕL
and ϕR :
γ α ∂α ϕ = −imϕ
 
∑
3
 0 σ0 ∂0 − σi ∂i  ( ) ( )
  ϕL ϕL
 i=1  = −im
 
 σ ∂ + ∑σ ∂
3
0  ϕR ϕR
0 0 i i
i=1

 ( )

 ∑3

 σ0 ∂0 − σi ∂i ϕR = −imϕL


i=1
( ) . (2.4.22)

 ∑3

 σ0 ∂0 + σi ∂i ϕL = −imϕR


i=1

In particular, in the massless (m = 0) case one obtains two uncoupled

equations
 2.4. Spin One-Half Electrodynamics 93
( )
∑
3
σ0 ∂0 − σi ∂i ϕR = 0 (2.4.23)
i=1
( )
∑
3
σ 0 ∂0 + σi ∂i ϕL = 0 (2.4.24)
i=1

which are, of course, just equations (2.4.12) and (2.4.13). To understand the
significance of ϕR and ϕL in general, the relationship between our current
model of spin one-half particles as 4-component objects and our earlier (2-
component) view of spin one-half (pages 85–86) and just what (2.4.23) and
(2.4.24) have to do with neutrinos (see page 100) we must discuss yet another
symmetry (invariance property) of the Dirac equation.
The Dirac equation is invariant under the proper, orthochronous Lorentz
group L+↑ because we were able to find a (“2-valued”) representation of L+↑
which, if taken to be the transformation law for the wavefunction, led to a
transformed wavefunction that satisfied the Dirac equation in the transformed
coordinate system. Now we wish to consider a coordinate transformation,
called spatial inversion, that does not correspond to an element of L+↑ . Its
matrix is
 
1 0 0 0
 
 0 −1 0 0 
π=  0

 0 −1 0 
0 0 0 −1

and its effect is simply to switch the orientation of the spatial coordinate sys-
tem. Note that, since π 2 is the 4 ×4 identity matrix, π generates a group of
coordinate transformations which we may denote Z2 , since that’s what it is
isomorphic to. In order to prove the invariance of the Dirac equation under
spatial inversions we will find a representation ρ : Z2 −→ GL(C4 ) which, if
taken to be the transformation law for ϕ, leads to a transformed wavefunction
that satisfies the Dirac equation in the transformed coordinate system. Note
that, since (γ 0 )2 is the 4 ×4 identity matrix, the assignments
id −→ id
π −→ γ 0

define a representation ρ of Z2 on C4 . We will show that this does the job.

Begin with the Dirac equation (2.4.15) in standard coordinates
x = (x0 , x1 , x2 , x3 ) on R1,3 . Introduce new coordinates y = (y 0 , y 1 , y 2 , y 3 )
on R1,3 by y α = Λαβ xβ , α = 0, 1, 2, 3, where Λ = (Λαβ ) is in Z2 . Define
∂ˆα = ∂/∂y α , α = 0, 1, 2, 3. Then ∂β = Λαβ ∂ˆα . Now define ϕ̂(y) by

ϕ̂ (y) = ρ(Λ)(ϕ (Λ−1 y)).

94 2. Physical Motivation

Substituting into (2.4.15) as on pages 106–107 gives

( )
(ρ(Λ)γ β (ρ(Λ))−1 Λαβ )∂ˆα + i m ϕ̂(y) = 0

which will be the Dirac equation if

(ρ(Λ))γ β (ρ(Λ))−1 Λαβ = γ α , α = 0, 1, 2, 3,

i.e., if
(ρ(Λ))−1 γ α ρ(Λ) = Λαβ γ β , α = 0, 1, 2, 3 (2.4.25)
(see page 107). Now, (2.4.25) is obviously satisfied if Λ = id so we need only
verify that it is also satisfied if Λ = π. In this case, (2.4.25) becomes

γ 0 γ α γ 0 = π αβ γ β , α = 0, 1, 2, 3,

i. e.,
γ 0γ 0γ 0 = γ 0
and
γ 0 γ i γ 0 = −γ i i = 1, 2, 3.
Since these are all easy to verify directly we have established the invariance
of the Dirac equation under spatial inversion. ( )
Notice, in particular, that if we write our Dirac wavefunction ϕ = ϕϕRL as
on page 109, then, under a spatial inversion, it transforms as follows:
( ) ( )( ) ( )
ϕL 0 σ 0 ϕL ϕR
ϕ= −→ γ 0 ϕ = = .
ϕR σ0 0 ϕR ϕL

Switching the orientation of the spatial coordinate axes interchanges ϕL and

ϕR and one therefore thinks of these two components as having opposite
“handedness” or “chirality.” In particular, in the m = 0 case one can regard
(2.4.23) (respectively, (2.4.24)) as equations for a particle that is massless,
spin one-half and “left-handed” (respectively, “right-handed”). Since the dis-
covery that, in β-decay processes (in which a neutrino is emitted), parity is not
conserved, these equations have been regarded as potential models for the neu-
trino. Lee and Yang ([LY]) suggested (2.4.23), but Feynman and Gell-Mann
([FG-M]) showed that the experimental evidence suggests (2.4.24) (“neutri-
nos spin to the left”). In the general (massive) case one thinks of a spin one-half
particle as having two chiral components ϕL and ϕR , each of which has two
additional components representing the possible spin states.
One final invariance property of the Dirac equation is worthy of note. If θ
is some real constant so that eiθ is in U (1), then

(γ α ∂α + im)ϕ = 0
 2.4. Spin One-Half Electrodynamics 95

obviously implies
(γ α ∂α + im)(eiθ ϕ) = 0.
The Dirac equation is therefore invariant under the global U (1)-action ϕ −→
eiθ ϕ. Just as was the case for the Schroedinger and Klein-Gordon equations,
elevating this global symmetry to a local gauge symmetry in which θ is a
function of x0 , x1 , x2 and x3 requires the presence of an electromagnetic gauge
potential. In the physics literature such potentials are included by “minimal
coupling,” i.e., by replacing the ordinary derivatives ∂α in the Dirac equation
by “covariant derivatives” ∂α + inAα (see pages 76–80).
The problem of molding all of the information we have assembled thus far
into a gauge theory model of the type described in Section 2.1 is complicated
by a number of issues. Recall that in the spin zero case (Section 2.2) we be-
gan with an electromagnetic field (i.e., a connection on a U (1)-bundle over
spacetime) and the various representations of U (1) on C and, from them,
constructed complex scalar fields, an action and the corresponding Euler-
Lagrange equations. The resulting Klein-Gordon equations happened to be
Lorentz invariant. Our point of departure in this section has been to insist
at the outset on the relativistic invariance of the free particle equations. This
led us to a wavefunction taking its values in C4 whose external symmetry
(Lorentz invariance) was expressed in the form of a transformation law corre-
sponding to a specific representation of the double cover SL(2, C) of L+↑ . This
suggests that, in a corresponding gauge theory model, the wavefunction is a
matter field on some SL(2, C)-bundle over spacetime. However, electrons are
coupled to electromagnetic fields and these are connections on U (1)-bundles,
not SL(2, C)-bundles, over spacetime. Furthermore, gauge invariance refers
specifically to the internal symmetry of a particle reflected in the behavior of
its wavefunction under changes in the local gauge potentials for the electro-
magnetic field so this notion also “lives” in a U (1)-bundle. To build a proper
gauge theory model for Dirac electrons coupled to electromagnetic fields will
require the “splicing together” of the external SL(2, C)-bundle and the in-
ternal U (1)-bundle into a single SL(2, C) × U (1)-bundle on which both the
electron and the electromagnetic field may be thought to live. This turns out
to be a relatively simple thing to do and we will outline the construction
shortly.
A more delicate, and much more interesting, obstacle is one that we could
evade altogether by simply continuing to restrict our attention to the space-
time R1,3 and its open submanifolds. From the perspective of the workaday
world of particle physics this would be an entirely reasonable choice since it
amounts to ignoring gravitational effects and these are generally negligible
in elementary particle interactions in the laboratory. From the perspective of
topology (and nonperturbative quantum field theory), however, such a choice
would “evade” the best part. We will conclude this section with a brief synop-
sis of the issues involved in describing spin one-half particles that live in more
96 2. Physical Motivation

general spacetimes where gravitational effects are not neglected. We will deal
with these issues in detail in the remaining chapters of the book.
A spacetime is a 4-dimensional (second countable, Hausdorff) manifold X
with a “Lorentz metric” g (this is a semi-Riemannian metric with the prop-
erty that each tangent space Tx (X) has a basis {e0 , e1 , e2 , e3 } for which
g(x)(eα , eβ ) = ηαβ ). Thus, each Tx (X) with its inner product g(x) can be
identified with R1,3 . The general Lorentz group L therefore acts on the or-
thonormal bases of each Tx (X). We are, however, only interested in bases
related by elements of L+↑ . Although one can isolate such a collection of bases
at each Tx (X) individually just by selecting an isomorphism onto R1,3 , an un-
ambiguous choice over the entire manifold X is possible only if X is assumed
orientable and “time orientable.” We will discuss this latter condition in more
detail in Chapter 3; essentially, one assumes the existence of a vector field V
on X that is timelike (g(x)(V (x), V (x))) > 0 for each x ∈ X) and so makes
a smooth selection over X of a timelike direction at each point that we may
(arbitrarily) decree “future-directed.” We will adopt both of these assump-
tions and thereby obtain, at each point, a family of oriented, time oriented,
orthonormal bases for the tangent space related by elements of L+↑ .

Remark: We will find that compact spacetimes necessarily violate certain

rather basic notions of causality (they contain closed timelike curves). For this
reason we will henceforth restrict our attention to the noncompact variety.

Now, Lorentz invariance means invariance under L+↑ . When X = R1,3 , bun-
dles over X are trivial so gauge fields, matter fields, etc., can all be identified
with objects defined on X. Furthermore, each tangent space can be canoni-
cally identified with X itself so L+↑ acts on X. In the general case, none of
this is true. In particular, choosing Lorentz frames and acting by L+↑ on such
frames cannot take place globally on all of X, but only point by point. To
describe all of this precisely we will build (in Chapter 3) the “oriented, time
oriented, orthonormal frame bundle” of X. This is a principal L+↑ -bundle

P
L+↑ ,→ L(X) −→
L
X

over X whose fibers consist of the oriented, time oriented, orthonormal bases
for the tangent spaces to X. The action of L+↑ on L(X) will simply carry one
such basis onto another (above the same point in X) and one can make sense
of “Lorentz invariance” for matter fields associated with this bundle by some
representation of L+↑ .
The frame bundle L+↑ ,→ L(X) −→ X exists for every oriented, time ori-
ented spacetime and so presents no real obstacle to our program. However,
there is an obstacle (or, rather, “obstruction”). The fibers of L(X) are all iso-
morphic to L+↑ , but the Dirac wavefunction did not arise from a representation
of L+↑ on C4 . Rather, it was determined by the representation D( 2 ,0) ⊕ D(0, 2 )
1 1
 2.4. Spin One-Half Electrodynamics 97

of SL(2, C) on C4 . Thus, to regard a Dirac electron as a matter field in the

sense of Section 2.1 it will be necessary to “globalize” over all of X the double
cover

SL (2, C)

Spin

The frame bundle provides a copy of L+↑ above every x ∈ X so what we need
is an SL(2, C)-bundle
P
SL(2, C) ,→ S(X) −→
S
X (2.4.26)

over X and a map of S(X) onto L(X) that is, in effect, the spinor map of
PS−1 (x) onto PL−1 (x) for each x ∈ X. More precisely, a spinor structure for
X consists of a principal SL(2, C)-bundle (2.4.26) over X and a map

λ : S(X) −→ L(X)

such that
PL (λ(p)) = PS (p)

and
λ(p · g) = λ(p) · Spin(g)

for all p ∈ S(X) and all g ∈ SL(2, C). The following diagram therefore
commutes.

s
S(X) × SL(2, C) S(X) X

λ × Spin λ idX

(X) × + (X) X
98 2. Physical Motivation

It is at this point that we encounter our obstruction. Not every oriented,

time oriented spacetime X admits a spinor structure and, for those which
do not, it is simply not possible to define the Dirac wavefunction for a mas-
sive, spin one-half particle. Confidence in the Dirac equation is such that
this is generally regarded as adequate justification for dismissing as physi-
cally unacceptable any spacetime without a spinor structure. From our point
of view the interesting part of all of this is that the existence or nonexis-
tence of a spinor structure is a purely topological question about X. We will
show that there is a certain Čech cohomology class w2 (X) ∈ Ȟ2 (X; Z2 ),
called the “second Stiefel-Whitney class” of X, the vanishing of which is
a necessary and sufficient condition for the existence of a spinor structure
on X : w2 (X) is the obstruction to the existence of a spinor structure (see
Section 6.5).
We will show that all of the spacetimes of interest to us do, in fact, admit
spinor structures. With this the construction of a gauge theory model for
a free Dirac electron proceeds as follows: X is a (noncompact) spacetime
manifold with w2 (X) = 0. The vector space V in which the wavefunction will
take its values is C4 . The (external) symmetry group G is SL(2, C) and the
representation ρ of SL(2, C) on C4 is

ρ = D( 2 ,0) ⊕ D(0, 2 ) : SL(2, C) −→ GL(C4 ).

1 1

One can define an inner product ⟨ , ⟩ on C4 relative to which this repre-

sentation is orthogonal as follows: First define the “twisted” Hermitian form
H: C4 × C4 −→ C by

H((z1 , . . . , z4 ), (w1 , . . . , w4 )) = z1 w̄3 + z2 w̄4 + z3 w̄1 + z4 w̄2 .

Regarding z, w ∈ C4 as column matrices, this is equivalent to

H(z, w) = z ⊤ γ 0 w,

where
( )
0 0 σ0
γ =
σ0 0

as on page 96. A simple computation shows that

H(ρ(g)(z), ρ(g)(w)) = H(z, w)

for all g ∈ SL(2, C). The required inner product on C4 is then given by
1
⟨z, w⟩ = (H(z, w) + H(w, z)).
2
 2.4. Spin One-Half Electrodynamics 99

The principal SL(2, C)-bundle over X is a spinor bundle

P
SL(2, C) ,→ S(X) −→
S
X.

A free Dirac electron is a corresponding matter field, i.e., a smooth map

ϕ : S(X) −→ C4

that is equivariant, i.e., satisfies

( )
−1 g −1 0
ϕ(p · g) = g · ϕ(p) = ϕ(p).
0 ḡ ⊤
Remarks: These electrons are “free” in the sense that they are not coupled
to an electromagnetic field. Such an electromagnetic field does not live on
the spinor bundle, but rather (as a connection) on a U (1)-bundle over X.
Shortly we will describe how to splice the spinor bundle and the U (1)-bundle
together into a single SL(2, C) × U (1)-bundle on which an interaction can
be described. Notice, however, that our “free” electron is not entirely free if
the underlying spacetime X represents a nontrivial gravitational field. Such
influences enter these considerations in the form of a connection on the spinor
bundle that is essentially the lift of the canonical (Levi-Civita) connection on
the frame bundle (see Section 3.3). With this and the potential function U :
C4 −→ R given by U (z) = 12 m∥z∥2 = 12 m⟨z, z⟩ one can write down an action
whose Euler-Lagrange equations constitute the general spacetime version of
the Dirac equation. The details are available in Section 6.4 of [Bl]. Needless
to say, when X = R1,3 and the matter field is pulled back by the standard
global cross-section of the (necessarily) trivial spinor bundle, this reduces to
(γ α ∂α + i m)ϕ = 0.
A free Dirac electron can, of course, also be regarded as a cross-section of the
vector bundle associated to SL(2, C) ,→ S(X) −→ X by the representation
1 1
D( 2 ,0) ⊕ D(0, 2 ) (see pages 49–50). More generally, if ρ is any representation
of SL(2, C) on some Ck , then an equivariant Ck -valued map on S(X) (or,
equivalently, a cross-section of the associated vector bundle S(X) ×ρ Ck ) is
called a k-component spinor field of type ρ on X. 4-component spinor fields
1 1
of type D( 2 ,0) ⊕ D(0, 2 ) are generally called Dirac spinor fields. 2-component
1 1
spinor fields of type D( 2 ,0) , or D(0, 2 ) are called Weyl spinor fields.
Finally, we outline the “splicing” procedure for building from the spinor
bundle (where electrons live) and a U (1)-bundle (where electromagnetic fields
live) a single SL(2, C) × U (1)-bundle (where both live and therefore can in-
teract). Consider, in general, two principal bundles over the base X:
P
G1 ,→ P1 −→
1
X
P
G2 ,→ P2 −→
2
X
100 2. Physical Motivation

Let
P1 ◦ P2 = {(p1 , p2 ) ∈ P1 × P2 : P1 (p1 ) = P2 (p2 )}.

Then P1 ◦ P2 is a submanifold of P1 × P2 and will be the total space of the

spliced bundle. Define
P12 : P1 ◦ P2 −→ X

by
P12 (p1 , p2 ) = P1 (p1 ) = P2 (p2 ).
Then P12 is a smooth map of P1 ◦ P2 onto X. Define a smooth right action
of G1 × G2 on P1 ◦ P2 by

(p1 , p2 ) · (g1 , g2 ) = (p1 · g1 , p2 · g2 ).

Then
P
G1 × G2 ,→ P1 ◦ P2 −→
12
X
is a smooth principal G1 × G2 -bundle over X.
Next we define maps π1 : P1 ◦ P2 −→ P1 and π2 : P1 ◦ P2 −→ P2 by
πi (p1 , p2 ) = pi , i = 1, 2. Letting e1 and e2 denote the identities in G1 and G2
we identify {e1 } × G2 with G2 and G1 × {e2 } with G1 . We then have principal
bundles
π1
G2 ,→ P1 ◦ P2 −→ P1
and
π
G1 ,→ P1 ◦ P2 −→
2
P2
for which the following diagram commutes:

P1 ° P2
π1 π2

P1 12 P2

1 2
X

Now, suppose we have connections ω 1 on P1 : P1 −→ X and ω 2 on P2 :

P2 −→ X. Identify G1 with G1 ×{0} ⊆ G1 ⊕G2 and G2 with {0}×G2 ⊆ G1 ⊕G2 .
Then π1∗ ω 1 is a connection on π2 : P1 ◦ P2 −→ P2 , π2∗ ω 2 is a connection on
π1 : P1 ◦ P2 −→ P1 and π1∗ ω 1 ⊕ π2∗ ω 2 is a connection on P12 : P1 ◦ P2 −→ X.
Finally, let V be a vector space and let ρ1 : G1 −→ GL(V) and
ρ2 : G2 −→ GL(V) be two representations that satisfy

ρ1 (g1 ) ◦ ρ2 (g2 ) = ρ2 (g2 ) ◦ ρ1 (g1 ) (2.4.27)

 2.4. Spin One-Half Electrodynamics 101

for all g1 ∈ G1 and g2 ∈ G2 . Then we can define

ρ1 × ρ2 : G1 × G2 −→ GL(V)

by

(ρ1 × ρ2 )(g1 , g2 ) = ρ1 (g1 ) ◦ ρ2 (g2 ) = ρ2 (g2 ) ◦ ρ1 (g1 )

and obtain a representation of G1 × G2 on V with associated left action on V

given by
(g1 , g2 ) · ξ = (ρ1 (g1 ) ◦ ρ2 (g2 ))(ξ).

Remark: ρ1 and ρ2 must commute, i.e., satisfy (2.4.27), to ensure that

ρ1 × ρ2 is a representation:

(ρ1 × ρ2 ) ((g1 , g2 ) (g1′ , g2′ )) = (ρ1 × ρ2 ) ((g1 g1′ , g2 g2′ )) =

ρ1 (g1 g1′ ) ◦ ρ2 (g2 g2′ ) = ρ1 (g1 ) ◦ ρ1 (g1′ ) ◦ ρ2 (g2 ) ◦ ρ2 (g2′ )
= ρ1 (g1 ) ◦ ρ2 (g2 ) ◦ ρ1 (g1′ ) ◦ ρ2 (g2′ )
= (ρ1 × ρ2 )(g1 , g2 ) ◦ (ρ1 × ρ2 ) (g1′ , g2′ ).

P
Now, a matter field on G1 × G2 ,→ P1 ◦ P2 −→
12
X associated with ρ1 × ρ2 :
G1 × G2 −→ GL(V) is a map ϕ : P1 ◦ P2 −→ V satisfying

ϕ((p1 , p2 ) · (g1 , g2 )) = (g1−1 , g2−1 ) · ϕ (p1 , p2 ),

i.e., ( ( )) (( ))
ϕ ((p1 · g1 , p2 · g2 )) = ρ1 g1−1 ρ2 (g2−1 (ϕ (p1 , p2 ))).
Now we apply this construction to the following special case. Begin with an
oriented, time oriented spacetime X and a spinor bundle
P
SL(2, C) ,→ S(X) −→
S
X
P
on X (this will be G1 ,→ P1 −→ 1
X). Let ω 1 be the spinor connection on
S(X) referred to in the Remark on page 117. Take V = C4 and let ρ1 be the
representation D( 2 ,0) ⊕ D(0, 2 ) of SL(2, C) on C4 . Thus,
1 1

   
z1 ( ) z1
 .  g1 0  . 
(ρ1 (g1 ))  . 
 .  = 0 (ḡ ⊤ )−1  . 
 . 
1
z4 z4

P
for each g1 in SL(2, C). Next let U (1) ,→ P −→ 2
X be some princi-
pal U (1)-bundle over X and let ω 2 be a connection on it (representing
102 2. Physical Motivation

some electromagnetic field to which the Dirac electron will respond). Take
ρ2 : U (1) −→ GL(C4 ) to be the representation given by
     
z1 z1 g2 z1
 .   .   . 
(ρ2 (g2 ))  .   .   . 
 .  = g2  .  =  . 
z4 z4 g2 z4

for each g2 ∈ U (1). Note that ρ1 (g1 ) ◦ ρ2 (g2 ) = ρ2 (g2 ) ◦ ρ1 (g1 ) as required in
(2.4.27). Thus, we have a representation
ρ1 × ρ2 : SL(2, C) × U (1) −→ GL(C4 )

given by
   
zl z1
 .   .. 
(ρ1 × ρ2 )(g1 , g2 )  .  
 .  = ρ1 (g1 ) ◦ ρ2 (g2 )  . 


z4 z4
 
( ) g2 z1
g1 0  . 
= ( ⊤ )−1   .
. 

0 ḡ1
g2 z4
 
( ) z1
g1 0  . 
= g2 ( ⊤ )−1  .. .
0 ḡ1 
z4

Now we splice the two bundles together to obtain

SL (2, C) × U (1) ,→ S(X) ◦ P −→ X.
A Dirac electron (coupled to the U (1)-potential ω 2 ) is then a smooth map
ϕ : S(X) ◦ P −→ C4 satisfying
( )
−1
−1 g1 0
ϕ ((p1 , p2 ) · (g1 , g2 )) = g2 ϕ ( p1 , p2 ),
0 ḡ1⊤

i.e.,
( )
(g1 g2 )−1 0
ϕ ( p1 · g1 , p2 · g2 ) = ⊤ ϕ ( p1 , p2 ),
0 (g1 g2 )

where g1 g2 is the entrywise product of g1 = ei θ1 with g2 ∈ SL(2, C). With

the equipment now available one can write down an action functional whose
Euler-Lagrange equations describe the interaction of a massive, spin one-half
 2.4. Spin One-Half Electrodynamics 103

particle with an electromagnetic field (the details are available in Section 7.2
of [Bl]).
As one final illustration of this technique we will sketch an analogous con-
struction for the interaction of a nucleon with a classical Yang-Mills field (for
the details, see Section 7.3 of [Bl]). Here we face the same problem as in the
case of a Dirac electron coupled to an electromagnetic field. A Yang-Mills
field is given by a connection on a principal SU (2)-bundle over spacetime
(Section 6.3 of [N4]), whereas a nucleon (proton/neutron) is a massive, spin
one-half particle and therefore lives in a spinor bundle. There is an additional
complication, however. A nucleon is a proton/neutron doublet, i.e., its wave-
function has a proton component and a neutron component and so must take
its values in V = C4 ⊕ C4 = C8 .
We begin then with an oriented, time oriented spacetime X and a spinor
bundle
PS
SL(2, C) ,→ S(X) −→ X.
ω 1 is again the spinor connection referred to in the Remark on page 117. Now
take V = C4 ⊕ C4 , which we identify with the set of ( vv12 ) with v1 , v2 ∈ C4 .
Letting ρ = D( 2 ,0) ⊕ D(0, 2 ) we define ρ1 : SL(2, C) −→ C4 ⊕ C4 by
1 1

( ) ( )
v1 (ρ(g1 ))(v1 )
( ρ1 ( g1 ))(v) = ( ρ1 ( g1 )) = .
v2 (ρ(g1 ))(v2 )

P
Now let SU (2) ,→ P −→ 2
X be some principal SU (2)-bundle over X and
ω 2 some connection on it (representing the Yang-Mills potential to which
the nucleon is coupled). Define ρ2 : SU (2) −→ GL(C4 ⊕ C4 ) as follows: For
each
( )
α β
g2 =
γ δ
in SU (2),
( ) ( )( )
v1 α β v1
( ρ2 ( g2 ) )(v) = ( ρ2 ( g2 ) ) =
v2 γ δ v2
( )
αv1 + βv2
= .
γvl + δv2

A simple calculation shows that ρ1 ( g1 ) ◦ ρ2 ( g2 ) = ρ2 ( g2 ) ◦ ρ1 ( g1 ) so we have

a representation

ρ1 × ρ2 : SL(2, C) × SU (2) −→ GL(C4 ⊕ C4 ).

104 2. Physical Motivation

Letting  
z1
 .. 
( )  . 
 
v1  
v= =  z4 
v2  w1 
 . 
 . 
.
w4
we have ( )
( ρ1 × ρ2 ( g1 , g2 ) )(v) = ( ρ( g1 ) ) (αv 1 + βv 2 )
( ρ( g1 ) ) (γv1 + δv2 )
  
( )
αz1 + βw1
 g1 ( 0)  .. 
 −1  . 
 0 ḡ1⊤ 
 αz4 + βw4 
 
=   .
( ) 
 γz1 + δw1 
 g1 ( 0)  .. 
 −1  . 
0 ḡ1⊤
γz4 + δw4

A nucleon field (coupled to an SU (2)-Yang-Mills potential ω 2 ) is then a

smooth map ϕ : S(X) ◦ P −→ C4 ⊕ C4 such that

ϕ ( ( p1 , p2 ) · ( g1 , g2 ) ) = ( g1−1 , g2−1 ) · ϕ( p1 , p2 ).
( )
Writing ϕ = ϕϕ12 we have
 
( ρ (g1−1 ))(αϕ1 (p1 · g1 , p2 · g2 )
( )  
ϕ1 ( p1 · g1 , p2 · g2 )  +βϕ2 ( p1 · g1 , p2 · g2 ) ) 
 
= .
ϕ2 ( p1 · g1 , p2 · g2 )  (ρ (g −1 )) (γϕ ( p · g , p · g ) 
 1 1 1 1 2 2 
+δϕ2 ( p1 · g1 , p2 · g2 ) )

ϕ1 is called the proton component of ϕ, while ϕ2 is its neutron

component. Note the “mixing” of the proton and neutron components due
to the SU (2)-action.

2.5 SU (2)-Yang-Mills-Higgs Theory on Rn

The motivation behind our final example of a classical gauge theory is a
process known as “dimensional reduction.” The example itself is of profound
significance to physics, arises naturally from the pure Yang-Mills theory on
R4 that was the subject of [N4], provides an extraordinary insight into the
true nature of the Dirac magnetic monopole and gives rise to even deeper
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 105

connections between physics and topology than those we have encountered

thus far. We will begin by simply enumerating, without motivation, the eight
items required (in Section 2.1) for the construction of a classical gauge theory.
Then we will review the structure of pure SU (2)- Yang-Mills theory on R4
and show how our example arises from it. Finally, we will describe a number
of the remarkable properties of the model, both physical and mathematical.
The base manifold is X = Rn with its usual orientation and Riemannian
metric. For the vector space V we take the Lie algebra su(2) of 2 × 2 complex
matrices that are skew-Hermitian and tracefree. The (positive definite) inner
product on su(2) is given by ⟨A, B⟩ = −2 trace (AB) (−trace(AB) is the
Killing form of su(2) and the 2 is a matter of convenience). The Lie group G
is taken to be SU (2) and ρ : SU (2) −→ GL(su(2)) is the adjoint representation

adg (A) = gAg −1

for all g ∈ SU (2) and A ∈ su(2). Note that ⟨adg (A), adg (B)⟩ =
⟨gAg −1 , gBg −1 ⟩ = −2 trace((gAg −1 )(gBg −1 )) = −2 trace(g(AB)g −1 ) = −2
trace(AB) = ⟨A, B⟩, as required. Since every bundle over Rn is trivial, we
will trivialize at the outset and take
P
SU (2) ,→ Rn × SU (2) −→ Rn

as our bundle, where the right action of SU (2) on Rn × SU (2) is given by

p · g = (x, h) · g = (x, hg)

for all p = (x, h) ∈ Rn × SU (2) and all g ∈ SU (2). There is a natural global
cross-section s : Rn −→ Rn × SU (2) given by

s(x) = (x, e)
where, for convenience, we write e for the identity element in SU (2). Any
other global cross-section then has the form
sg : Rn −→ Rn × SU (2)
sg (x) = s(x) · g(x)
= (x, e) · g(x)
= (x, g(x))

for some smooth map g : Rn −→ SU (2) (Exercise 4.3.5 of [N4]). The cross-
section sg gives rise to an automorphism of the bundle (i.e., a global gauge
transformation) in the usual way (see page 343 of [N4]):

s(x) · h −→ sg (x) · h
(x, e) · h −→ (x, g(x)) · h
(x, h) −→ (x, g(x)h).
106 2. Physical Motivation

Thus, one can identify a gauge transformation with a smooth map

g : Rn −→ SU (2) which multiplies in the fibers on the left.
Because of the triviality of the bundle, any connection ω on SU (2) ,→
Rn × SU (2) −→ Rn is uniquely determined by its gauge potential
A = s∗ ω,

which is an su(2)-valued 1-form on Rn . Furthermore, any su(2)-valued 1-form

on Rn is the pullback by s of a unique connection on SU (2) ,→ Rn × SU (2)
−→ Rn (page 333, [N4]). Thus, we may restrict our attention entirely to
globally defined gauge potentials A on Rn . Relative to standard coordinates
on Rn we write
A = s∗ ω = Aα dxα ,
where each Aα , α = 1, . . . , n, takes values in su(2) (see (2.5.2) below). A gauge
transformation g : Rn −→ SU (2) gives a new gauge potential

Ag = (sg )∗ ω
related to A by
Ag = g −1 Ag + g −1 dg,
where dg is the entrywise exterior derivative of g : Rn −→ SU (2) and the
products are matrix products (we will do an explicit calculation of this sort
for the “t’ Hooft-Polyakov monopole” somewhat later). The curvature Ω of ω
is likewise uniquely determined by the field strength
1
F = s∗ Ω = dA + A ∧ A = Fαβ dxα ∧ dxβ ,
2
where
Fαβ = ∂α Aβ − ∂β Aα + [Aα , Aβ ], α, β = 1, . . . , n.
A gauge transformation g : Rn −→ SU (2) gives a new field strength

F g = (sg )∗ Ω

related to F by
F g = g −1 F g.
In this context a matter field is a smooth su(2)-valued map Φ on
Rn × SU (2) that satisfies
Φ( p · g) = g −1 · Φ(p)
Φ( (x, h) · g) = adg−1 (Φ(x, h))
Φ (x, hg ) = g −1 Φ(x, h)g

for all (x, h) ∈ Rn × SU (2) and all g ∈ SU (2). When, as in this case, V is the
Lie algebra G of the structure group G and ρ is the adjoint representation of
G on G, a matter field is referred to as a Higgs field. The triviality of the
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 107

bundle in our present circumstances allows us to identify the Higgs field with
its pullback by the global cross-section s:
ϕ = s∗ Φ = Φ ◦ s
ϕ(x) = Φ(x, e).
Under a gauge transformation g : Rn −→ SU (2),
ϕg = (sg )∗ Φ = g −1 ϕg
because
((sg )∗ Φ)(x) = Φ(sg (x))
= Φ(x, g)
= Φ(x, eg)
= g −1 Φ(x, e)g
= g −1 ϕ(x)g.

The next item on the agenda (#7 of Section 2.1) is the potential function
U : su(2) −→ R. This plays a rather peculiar role in the story we wish
to tell. Initially we adopt what is called the Georgi-Glashow potential
U : su(2) −→ R given by
λ
U (A) = (||A∥2 − 1)2 ,
8
where λ ≥ 0 is a constant and ∥ A ∥2 = ⟨ A, A ⟩ = −2 trace(A2 ), noting that
U (g · A) = U (gAg −1 ) = U (A) as required. Shortly, however, we will take λ to
be zero and retain only a vestige of the potential in the form of an asymptotic
boundary condition that it imposes on the Higgs field ϕ (see pages 132–133).
In order to describe the appropriate action (#8 of Section 2.1) for our exam-
ple we must anticipate a few results on differential forms that will be proved
later (Chapter 4). We will content ourselves with just a brief description of
those particular items required for the example. We have already introduced
real- and vector-valued 0-forms (pages 10 and 12), 1-forms (pages 9 and 12),
and 2-forms (pages 11 and 12). k-forms, for integers k ≥ 3, are defined analo-
gously and all of the familiar algebraic and analytic operations on 0-, 1-, and
2-forms extend to this more general context. For example, a real-valued 3-form
on a manifold X is a map α that assigns to each p ∈ X a real-valued trilin-
ear function αp on Tp (X) × Tp (X) × Tp (X) that is skew-symmetric (changes
sign when-ever two of its arguments are interchanged) and smooth in the
sense that, for any V 1 , V 2 , V 3 ∈ X (X), the function α(V 1 , V 2 , V 3 ) on
X defined by (α(V 1 , V 2 , V 3 ))(p) = αp (V 1 (p), V 2 (p), V 3 (p)) is in C ∞ (X).
These arise, for example, as exterior derivatives of 2-forms and wedge prod-
ucts of 1-forms and 2-forms, or of three 1-forms (all of which will be defined
carefully in Chapter 4).
In general, the set of k-forms on an n-dimensional manifold X is denoted
Λk (X) and admits a natural C ∞ (X)-module structure (and so in particular,
is a real vector space). If (U, φ) is a chart for X with coordinate functions
108 2. Physical Motivation

x1 , . . . , xn , then any α ∈ Λk (X) has a local coordinate expression

1
α = αi ···i dxi1 ∧ · · · ∧ dxik
k! 1 k
(summation over i1 , . . .,ik = 1, . . .,n), where each αi1 ···ik is in C ∞ (U ). The
exterior derivative dα of α ∈ Λk (X) is an element of Λk+1 (X) which, locally,
is given by
1
dα = (dαi1 ···ik ) ∧ dxi1 ∧ · · · ∧ dxik .
k!
There are no nonzero k-forms on X if k > n and, for 0 ≤ k ≤ n, we will show
that the dimension of Λk (X) as a C ∞ (X)-module is ( nk ). Since ( n−k n
) = ( nk ),
k n−k
the modules Λ (X) and Λ (X) are isomorphic. We will find that, when X
is oriented and has a metric (Riemannian or semi-Riemannian), then there is
a natural isomorphism
∗
: Λk (X) −→ Λn−k (X),

called the Hodge star operator. Moreover, dim Λn (X) = ( nn ) = 1 and, when
X is oriented and has a metric, there is a distinguished generator for Λn (X)
called the metric volume form and denoted vol (in standard coordinates on
Rn this is just dx1 ∧ · · · ∧ dxn ). In particular, for any α, β ∈ Λk (X), α ∧ ∗β is
in Λn (X) and so is a multiple, by some element of C ∞ (X), of vol. We denote
this element of C ∞ (X) by ⟨α, β ⟩:

α ∧∗ β = ⟨α, β ⟩vol.

This defines an inner product on Λk (X) and, when β = α, we will write

⟨ α, α ⟩ = ∥ α ∥2 . In the Riemannian case we will find that
∗∗
β = (−1)k(n−k) β and that it follows from this that the Hodge star op-
erator is actually an isometry. k-forms that vanish outside of a compact set
can be integrated over k-dimensional manifolds. Indeed, we will find that they
can even be integrated over k-dimensional regions with a sufficiently smooth
(k − 1)-dimensional “boundary” and it is in this context that we will prove a
version of Stokes’ Theorem relating the integral of a (k − 1)-form α over this
boundary to the integral of dα over the region it bounds.
Much of what we have just said about real-valued forms extends easily to
vector-valued forms by simply doing everything (evaluation at tangent vectors,
exterior derivative, Hodge star, etc.) componentwise relative to some basis for
the vector space (we will show that it all turns out to be independent of the
choice of basis). There are a few troublesome items (e.g., wedge products)
that we will treat carefully in Chapter 4 and simply illustrate here for the
particular vector space of interest in our example i.e., su(2). We take as a
basis for su(2) the set {T1 , T2 , T3 }, where
1 1 1
T1 = − i σ1 , T2 = − i σ2 , T3 = − i σ3 ,
2 2 2
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 109

and
( ) ( ) ( )
0 1 0 −i 1 0
σ1 = , σ2 = , σ3 =
1 0 i 0 0 −1

are the Pauli spin matrices. It is easy to see that {T1 , T2 , T3 } is orthonormal
with respect to the inner product ⟨ A, B ⟩ = −2 trace(AB) on su(2). Then
any su(2)-valued k-form φ on Rn (e.g, A or F ) can be regarded as a matrix
of complex k-forms
( )
3 2 1
1 1 φ i φ + φ i
φ = φa Ta = − φa (i σa ) = − (2.5.1)
2 2 −φ2 + φ1 i −φ3 i

where φ1 , φ2 , and φ3 are in Λk (Rn ). Alternatively one can write

1 a
φa = φ dxi1 ∧ · · · ∧ dxik , a = 1, . . . , n
k! i1 ···ik
and then
( )
1 a
φ = φ Ta =a
φ dx ∧ · · · ∧ dx
i1 ik
Ta
k! i1 ···ik
1 ( a )
= φi1 ···ik Ta dxi1 ∧ · · · ∧ dxik
k!

1
φ= φi ···i dxi1 ∧ · · · ∧ dxik , (2.5.2)
k! 1 k
where

φi1 ···ik = φai1 ···ik Ta

( )
1 φ3i1 ···ik i φ2i1 ···ik + φ1i1 ···ik i
=− (2.5.3)
2 −φ2i ···i + φ1i ···i i −φ3i1 ···ik i
1 k 1 k

is a smooth map into su(2) for each i1 . . . , ik = 1, . . . , n.

We will, in Chapter 4, discuss various natural ways of defining a wedge
product for matrix-valued forms such as these. From our point of view at the
moment, the most useful such notion can be described as follows: The wedge
product of two complex-valued forms is obtained by multiplying the forms as
if they were complex numbers, but with real and imaginary parts multiplied
by the ordinary wedge product of real-valued forms, i.e.,

(φ1 + φ2 i ) ∧ (ψ 1 + ψ 2 i ) = (φ1 ∧ ψ 1 − φ2 ∧ ψ 2 ) + (φ1 ∧ ψ 2 + φ2 ∧ ψ 1 )i

Now, if φ and ψ are both su(2)-valued and written as in (2.5.1), then φ ∧ ψ

is obtained by simply forming their matrix product, with entries multipled by
110 2. Physical Motivation

the complex wedge product described above. We will illustrate the procedure
with an example that will also allow us to write out our action functional. We
consider a k -form φ with values in su(2) and written in the form (2.5.1) and
will compute φ ∧ ∗ φ, where ∗ φ is the Hodge dual of φ, computed componen-
twise (i.e., entrywise). Thus,
( )( )
∗ 3 ∗ 2
∗ 1 φ3 i φ2 + φ1 i φ i φ + ∗ φ1 i
φ∧ φ=
4 −φ2 + φ1 i −φ3 i −∗ φ 2 + ∗ φ 1 i −∗ φ3 i

 
−φ3 ∧ ∗ φ3 (φ3 i ) ∧ (∗ φ2 + ∗ φ1 i )
 
 +(φ2 + φ1 i ) −(φ2 + φ1 i ) 
 
 
1 ∧(−∗ φ2 + ∗ φ1 i ) ∧(∗ φ3 i ) 
 
=  .
4 
 (−φ2 + φ1 i ) (−φ2 + φ1 i ) 
 
 ∧(∗ φ3 i ) − (φ3 i ) ∧(∗ φ2 + ∗ φ1 i ) 
 
∧(−∗ φ2 + ∗ φ1 i ) −φ2 ∧ ∗ φ3

Each entry is computed in the same way. For example,

−φ3 ∧ ∗ φ3 + (φ2 + φ1 i ) ∧ (−∗ φ2 ∧ ∗ φ1 i )

= −∥φ3 ∥2 vol − ∥φ2 ∥2 vol − ∥φ1 ∥2 vol
+ (⟨φ2 , φ1 ⟩vol − ⟨φ1 , φ2 ⟩vol)i
= −(∥φ1 ∥2 + ∥φ2 ∥2 + ∥φ3 ∥2 )vol

and similarly for the rest. In particular, the (2, 2)-entry is the same so

−2trace(φ ∧ ∗ φ) = (∥φ1 ∥2 + ∥φ2 ∥2 + ∥φ3 ∥2 )vol.

We define
∥φ∥2 = ∥φ1 ∥2 + ∥φ2 ∥2 + ∥φ3 ∥2

so that
−2trace(φ ∧ ∗ φ) = ∥φ∥2 vol. (2.5.4)

Remark: More generally, if φ and ψ are any two su(2)-valued k -forms

on Rn and {T1 , T2 , T3 } is any orthonormal basis for su(2) and if we write
φ = φa Ta and ψ = ψ a Ta , then we can define ⟨φ, ψ⟩ = ⟨φ1 , ψ 1 ⟩ + ⟨φ2 , ψ 2 ⟩ +
⟨φ3 , ψ 3 ⟩ and show, as above, that

−2trace(φ ∧ ∗ ψ) = ⟨φ, ψ⟩ vol.

 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 111

With the machinery we have assembled thus far we can write out the Yang-
Mills-Higgs action functional A(A, ϕ) for our example:
∫ (
A(A, ϕ) = − trace(F ∧ ∗ F ) − trace(dA ϕ ∧ dA ϕ)
Rn
λ∗ )
+ (∥ϕ∥2 − 1)2
8
∫ ( (2.5.5)
1
= ∥F ∥2 + ∥dA ϕ∥2
2 Rn
λ )
+ (∥ϕ∥2 − 1)2 dx1 ∧ · · · ∧ dxn ,
4
where dA ϕ = dϕ + [A, ϕ] is the covariant exterior derivative of the Higgs field
ϕ. We observe first that A(A, ϕ) is gauge invariant, i.e., that the effect of a
gauge transformation g : Rn −→ SU (2) (A −→ Ag , F −→ F g and ϕ −→ ϕg )
is to leave the integral unchanged. We have already seen that F g = g −1 F g
and ϕg = g −1 ϕg and we show now that
( )
dA ϕg = g −1 dA ϕ g
g
(2.5.6)

as well (we will need to use a few simple algebraic properties of d, e.g., the
product rule for matrix products, that will be proved in Chapter 4). Indeed,
dA ϕ = dϕ + [A, ϕ] implies
( )
g −1 d A ϕ g = g −1 dϕg + g −1 [A, ϕ]g

and
( ) [ ]
d A ϕg = dϕg + [Ag , ϕg ] = d g −1 ϕg + g −1 Ag + g −1 dg, g −1 ϕg
g

( ) [ ] [ ]
= d g −1 ϕg + g −1 Ag, g −1 ϕg + g −1 dg, g −1 ϕg
( ) [ ]
= g −1 [A, ϕ]g + d g −1 ϕg + g −1 dg, g −1 ϕg

= g −1 [A, ϕ]g + g −1 d(ϕg) + dg −1 ϕg

+ g −1 dg g −1 ϕg − g −1 ϕ g g −1 dg
= g −1 [A, ϕ]g + g −1 ϕdg + g −1 dϕg
+ dg −1 ϕg + g −1 dg g −1 ϕg − g −1 ϕdg
= g −1 [A, ϕ]g + g −1 dϕg + dg −1 ϕg + g −1 dg g −1 ϕg
( )
= g −1 [A, ϕ]g + g −1 dϕg = g −1 dA ϕ g
112 2. Physical Motivation

because

g −1 g = id =⇒ g −1 dg + dg −1 g = 0
=⇒ g −1 dg = −dg −1 g
=⇒ g −1 dg g −1 ϕg = −dg −1 ϕg.

Gauge invariance of the action will therefore follow if we can show that
∥g −1 φg∥2 = ∥φ∥2 for any su(2)-valued form φ. But if we write φ = φa Ta ,
then g −1 φg = φa (g −1 Ta g) and, since ⟨g −1 Ag, g −1 Bg⟩ = ⟨A, B⟩, {g −1 T1 g,
g −1 T2 g, g −1 T3 g} is also an orthonormal basis for su(2) so this is clear (see the
Remark on page 130).
We are interested in finite action, stationary configurations (A, ϕ), i.e.,
solutions to the Euler-Lagrange equations for the action (2.5.5) for which
A(A, ϕ) < ∞. As it happens, no such solutions exist when n > 4 (see [JT]).
When n = 2 such solutions do exist and they are called vortices. These are
studied exhaustively in [JT], but we will have no more to say about them.
When n = 4 any such solution is gauge equivalent to a pure Yang-Mills field
(λ = 0, ϕ = 0) of the type discussed in [N4] (we will briefly review this
material shortly). Our primary concern is with the case n = 3 where finite
action, stationary configurations are, for reasons we hope to make clear, called
monopoles. In fact, we intend to discuss only a special case in which just
a vestige of the Georgi-Glashow potential survives. This special case arises
in the following way: The requirement that A(A, ϕ) < ∞ implies that, as
|x| −→ ∞ in R3 ,

F −→ 0 (2.5.7)

dA ϕ −→ 0 (2.5.8)

and, at least if λ ̸= 0,
ϕ −→ 1. (2.5.9)

Indeed, it is shown in Chapter 4, Sections 10–15, of [JT] that each of these

limits is achieved uniformly. Now, when λ = 0 there is no reason to suppose
that finite action implies ∥ϕ∥ −→ 1 as |x| −→ ∞. However, [JT] also shows
that, even in this case, there is some constant c ≥ 0 such that ∥ϕ∥ −→ c
uniformly as |x| −→ ∞ and that if c ̸= 0, one can rescale in R3 to obtain
a new configuration (A′ (x), ϕ′ (x)) = (c−1 A(c−1 x), c−1 ϕ(c−1 x)) which is
a finite action stationary point for the action A with λ = 0 and satisfies
∥ϕ′ || −→ 1 uniformly as |x| −→ ∞. In effect, one loses nothing, even in the
λ = 0 case, by restricting attention to those configurations for which (2.5.7),
(2.5.8) and (2.5.9) are satisfied. This, then, is precisely what we intend to.
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 113

Remark: Before abandoning the self-interaction term, however, we point

out some general features of the action (2.5.5). In particular, we note that
there are some obvious absolute minima. Indeed, A(A, ϕ) is obviously zero
whenever A = 0 and ϕ = ϕ0 is a constant in su(2) with ∥ϕ0 ∥ = 1. Such an
absolute minimum is called a ground state for the system. The corresponding
quantum state of lowest energy is called a vacuum state and physicists perform
perturbation calculations about such vacuum states. The point here is that
such vacuum states are not unique (as long as A = 0, any ϕ0 ∈ S 2 ⊆ su(2)
will give rise to such a state). A specific choice of such a ϕ0 is said to break
the symmetry from SU (2) to U (1) and we wish to very briefly explain the
terminology (see Section 10.3 of [Bl] for more details). A gauge transformation
g : R3 −→ SU (2) acts on ϕ by ϕ −→ ϕg = g −1 ϕg. If the ground state is to
be gauge invariant, then we must have g −1 ϕ0 g = ϕ0 and this occurs only
for g in the isotropy subgroup of ϕ0 in SU (2) under the adjoint action. Now,
we claim that this isotropy subgroup is a copy U (1). To see this, identify
su(2) with R3 and SU (2) / ± id with SO(3) (page 88). Then the adjoint
action of SU (2) on su(2) corresponds to the natural action of SO(3) on R3
(this is proved, although not stated in these terms in Appendix A to [N4]).
But this natural action of SO(3) on R3 (rotation) is transitive on S 2 ⊆ R3
(page 27) Now, if H = {g ∈ SU (2) : g −1 ϕ0 g = ϕ0 }, then (since SU (2)
is compact), SU (2) / H ∼ = S 2 (Remark on page 27). Thus, dim H = 1. But
H is closed in SU (2) so it too is compact and H ∼ = U (1) (S 1 is the only
compact, connected 1-manifold; see Section 5–11 of [N1]). The ground states
of our SU (2) Yang-Mills-Higgs theory are therefore invariant only under a
U (1) subgroup of SU (2). This is an instance of the phenomenon of spontaneous
symmetry breaking in which a field theory with an exact symmetry group G
(e.g., SU (2)) gives rise to ground states that are invariant only under some
proper subgroup H (e.g., U (1)) of G.
With these few remarks behind us we now turn to the case (λ = 0) of most
interest to us. More precisely, we consider the action
∫ ( )
A(A, ϕ) = − trace(F ∧ ∗ F ) − trace(dA ϕ ∧ ∗ dA ϕ)
R3
∫ ( ) (2.5.10)
1
= ∥F ||2 + ∥dA ϕ∥2 dx1 ∧ dx2 ∧ dx3
2 R3

and take as our configuration space

{ }
C= (A, ϕ) : A(A, ϕ) < ∞, lim sup 1 − ∥ϕ∥ = 0 (2.5.11)
R−→∞ |x|≥R

(the second condition reflecting our decision to restrict attention to finite

action configurations for which ∥ϕ∥ −→ 1 as |x| −→ ∞ in R3 ). Writing out
the Euler-Lagrange equations for the action (2.5.11) yields what are called
114 2. Physical Motivation

the Yang-Mills-Higgs (YMH) equations

{
∗ A∗
d F = [dA ϕ, ϕ]
∗ A ∗ A
. (2.5.12)
d d ϕ=0

The configuration (A, ϕ) must also satisfy the following Bianchi identities:
{
dA F = 0
. (2.5.13)
dA dA ϕ = [F , ϕ]

Thus we are looking for solutions to (2.5.12) that live in C. We shall find
some interesting ones, but not by studying (2.5.12) directly. As it happens,
there is a simpler set of first order equations whose solutions necessarily also
satisfy (2.5.12) and, in fact, give the absolute minima of the action (2.5.10).
Remark: This is entirely analogous to the situation encountered in [N4],
where the (anti-) self-dual equations on R4 gave the absolute minima of the
Yang-Mills action. Shortly we will review that situation and find that there is
a closer connection than simple analogy.
The best way to see where these equations come from is as follows: We denote
by ⟨ , ⟩ the inner product we have defined on su(2)-valued forms on R3
(Remark, page 130). Notice that, on R3 , F and ∗ dA ϕ are both 2- forms and
(since the metric on R3 is Riemannian), ∥dA ϕ∥2 = ∥∗ dA ϕ∥2 . Now notice that
2 2
∗ A
∥F ∥2 + dA ϕ = ∥F ∥2 + d ϕ
⟨ ⟩
= ⟨F , F ⟩ + ∗ dA ϕ, ∗ dA ϕ
⟨ ⟩ ⟨ ⟩
= F − ∗ dA ϕ, F − ∗ d A ϕ + 2 F , ∗ dA ϕ
2 ⟨ ⟩
= F − ∗ dA ϕ + 2 F , ∗ dA ϕ

and, similarly,
2 2 ⟨ ⟩
∥F ∥2 + d A ϕ = F + ∗d Aϕ − 2 F , ∗d Aϕ .

It follows that (A, ϕ) achieves an absolute minimum when

F = ±∗ dA ϕ. (2.5.14)

These are the (Bogomolny) monopole equations and any configuration

(A, ϕ) which satisfies them also satisfies the YMH equations (2.5.12) (either
observe that an absolute minimum for A(A, ϕ) is necessarily a stationary
value, or simply substitute F = ±∗ d A ϕ into (2.5.12) and use the Bianchi
identities (2.5.13)). For the record we write out these equations in standard
coordinates on R3 , using a formula for the Hodge dual of a 1-form on R3
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 115

that we will prove in Chapter 4. With A = Ak dxk and F = 1

2 Fij dxi ∧ dxj ,
(2.5.14) becomes
∑
3
Fij = ± ϵijk (∂k ϕ + [Ak , ϕ]), i, j = 1, 2, 3, (2.5.15)
k=1

where ϵijk is the Levi-Civita symbol (anti-symmetric in i j k and ϵ123 = 1).

We will have something to say shortly about why these are called “monopole”
equations. First we wish to give some indication of how the example before us
now actually arises quite naturally out of the pure Yang-Mills theory on R4
discussed in [N4]. This is the n = 4 case of the model we have constructed
when, in the action (2.5.5), λ = 0 and ϕ = 0. In this special case the action
(which now depends only on A) is called the Yang-Mills-action and written
∫
YM(A) = − trace(F ∧ ∗ F )
R4
∫ (2.5.16)
1
= ∥F ∥2 dx1 ∧ dx2 ∧ dx3 ∧ dx4 .
2 R4

The Euler-Lagrange equations now become the Yang-Mills equations

dA ∗ F = 0, (2.5.17)

while the Bianchi identities become

dA F = 0. (2.5.18)

Now notice that if we have a potential A for which the field strength F is
self-dual (SD)
∗
F = F, (2.5.19)
or anti-self-dual (ASD)
∗
F = −F , (2.5.20)
then the Bianchi identity (2.5.18) implies that F necessarily satisfies the Yang-
Mills equations (2.5.17). Such potentials A are called SU (2) instantons on
R4 and, in [N4], a number of examples (called the BPST instantons) were
described. Written in quaternionic notation (i.e., identifying R4 with H and
su(2) with the Lie algebra Im H of pure imaginary quaternions) these can be
written ( )
q̄ − n̄
Aλ,n (q) = Im dq , (2.5.21)
λ2 + |q − n|2
where λ > 0 and n ∈ H are parameters called the scale and center of the
instanton, respectively. The corresponding field strengths are
λ2
F λ,n (q) = dq̄ ∧ dq. (2.5.22)
(λ2 + |q − n|2 )2
116 2. Physical Motivation

These are ASD and satisfy

1 48λ2
∥F λ,n (q)∥2 = 2 (2.5.23)
2 (λ + |q − n|2 )4
and
∫
48 λ2
YM(Aλ,n ) = dq 0 dq 1 dq 2 dq 3 = 8π 2 . (2.5.24)
R4 (λ2 + |q − n|2 )4
Notice that all of these potentials have the same Yang-Mills action (total field
strength). For a fixed center n, ∥F λ,n (q)∥2 has a maximum value of 96/λ2 at
q = n and, as λ −→ 0, this maximum value approaches infinity in such a way
that the integrals over R4 remain constant at 8π 2 .
The fact that all of these BPST potentials Aλ,n have the same Yang-Mills
action has a much more profound significance than may be apparent at first
glance. It was shown in [N4] (see also Examples 3 and 4, pages 40–42, of
Chapter 1) that each Aλ,n is the pullback to R4 via stereographic projection
of a connection ω λ,n on the Hopf bundle SU (2) ,→ S 7 −→ HP1 . Thinking
of HP1 ∼ = S 4 as the one-point compactification of R4 , one can reverse one’s
point of view here and say that, by virtue of the asymptotic behavior implicit
in the fact that YM(Aλ,n ) < ∞, each Aλ,n “extends to the point at infinity.”
Indeed, Uhlenbeck’s Removable Singularities Theorem asserts that, if A is an
SU (2) gauge potential on R4 with YM(A) < ∞, then A always “extends to
the point at infinity” in the sense that there exists a unique SU (2)-bundle over
S 4 and a connection ω on it such that A = (s ◦ φ−1 )∗ ω, where s is a cross-
section of the bundle defined on the complement of some point in S 4 and φ is
a stereographic projection to R4 . The specific bundle to which the potential
A “extends” is, moreover, determined by the value of YM(A). It is essential
to understand precisely what is being asserted here so, before returning to the
Bogomolny monopole equations, we will elaborate.
According to the Classification Theorem (page 34), the set of equivalence
classes of principal SU (2)-bundles over S 4 is in one-to-one correspondence
with the elements of the homotopy group π3 (SU (2)) ∼ = π3 (S 3 ) ∼
= Z. There
are various ways of associating with each SU (2)-bundle over S 4 an integer
that characterizes it up to equivalence, but from our point of view the most
useful of these arises in the theory of characteristic classes. We have already
had one brief encounter with characteristic classes (the first Chern class, in
Section 2.2) and will return to the general theory in Chapter 6. For the present
we will content ourselves with a brief, informal description of those aspects of
the subject relevant to characterizing a principal SU (2)-bundle
P
SU (2) ,→ P −→ S 4

over S 4 up to equivalence.
Choose any connection ω on the bundle and let Ω denote its curvature
(we will prove in Chapter 3 that connections exist on any smooth principal
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 117

bundle). The local field strengths F̃ = s∗ ω for various cross-sections generally

do not agree on the intersections of their domains and so do not piece together
into a globally defined 2-form on all of S 4 . Indeed, we know that if sg is
another cross-section, then F̃ g = g −1 F̃ g on the intersection. However, certain
algebraic combinations of the local field strengths can be found which do agree
on the intersections and so do piece together into globally defined forms on
S 4 . Essentially, all that is required is a symmetric, multilinear function on
su(2) that is ad-invariant, i.e., takes the same values at A and g −1 Ag for all
A ∈ su(2) and g ∈ SU (2). The trace is an obvious choice and this gave rise, in
Section 2.2, to the first Chern class which, for U (1)-bundles over spacetime, is
the cohomology class of the electromagnetic field strength. For SU (2)-bundles,
however, the local field strengths take values in su(2) where everything has
trace zero so this will not get us very far. The trace of the product would be
the next likely candidate and this gets us very far indeed. We define a 4-form
on all of S 4 by decreeing that, relative to any local cross-section, it is given
by
1
trace(F̃ ∧ F̃ )
8π 2
(the 8π1 2 is a normalizing constant whose purpose we will describe shortly).
Now, a priori this is a complex 4-form on S 4 , but it can be shown to be real-
valued (because F̃ is skew-Hermitian). Being a 4-form on the 4-dimensional
manifold S 4 , it is also closed (i.e., has exterior derivative zero) and therefore
determines a de Rham cohomology class
1 [ ]
c2 (P ) = trace( F̃ ∧ F̃ ) ∈ Hde
4 4
R (S ). (2.5.25)
8π 2
Remarkably, this cohomology class does not depend on the initial choice of the
connection ω from which it arose; it is a characteristic class for the bundle.
c2 (P ) is called the second Chern class of the bundle. Any representative
of the class c2 (P ) is a 4-form on the compact 4-manifold S 4 and so can be
integrated over S 4 . It will follow from Stokes’ Theorem that two forms in the
same cohomology class have the same integral so, in effect, we may integrate
c2 (P ) over S 4 . The result is written
∫ ∫
1
c2 (P )[S 4 ] = c2 (P ) = trace(F̃ ∧ F̃ ) (2.5.26)
S4 8π 2 S 4

and called the second Chern number of the bundle (physicists call
−c2 (P )[S 4 ] the topological charge or instanton number of the bundle).
The 8π1 2 ensures that c2 (P )[S 4 ] is an integer (not obvious, but true) and it
is this integer that labels the equivalence classes of SU (2)-bundles over S 4 .
More precisely, two principal SU (2)-bundles over S 4 are equivalent if and only
if their second Chern numbers are equal.
118 2. Physical Motivation

To relate this to our gauge potentials on R4 recall that there is a stereo-

graphic projection φ, defined at all but one point of S 4 , that is an orientation
preserving (conformal) diffeomorphism onto R4 . The integrals we define will
be invariant under such maps and unaffected by the omission of one point
so c2 (P )[S 4 ] can be computed by integrating over R4 the pullback of 8π1 2
trace(F̃ ∧ F̃ ) by φ. Pullback commutes with trace and the wedge product
and (φ−1 )∗ F̃ = (φ−1 )∗ (s∗ Ω) = (s ◦ φ−1 )∗ Ω = F is a gauge potential on
R4 . Thus, ∫
1
c2 (P )[S 4 ] = trace(F ∧ F ). (2.5.27)
8π 2 R4

Now here’s the good part: If A is an ASD potential on R4 , then

∗
F = −F so
∫
YM(A) = −trace(F ∧ ∗ F )
∫
R4
(2.5.28)
= trace(F ∧ F ) (ASD).
R4
If this is finite, the Removable Singularities Theorem assures us that A ex-
tends to some principal SU (2)-bundle over S 4 . The Chern number of this
bundle can be computed from any connection on it so we might as well use
the extension of A. Then, comparing (2.5.27) and (2.5.28) gives
1
c2 (P )[S 4 ] = YM(A) (ASD). (2.5.29)
8π 2
The Yang-Mills action of A (i.e., its total field strength) is directly encoded
in the topology of the bundle to which A extends as its Chern number. Since
the Yang-Mills action is determined by the asymptotic behavior of the field
strength F as |x| −→ ∞ in R4 , we find that it is this asymptotic behavior that
determines the bundle to which A extends. This phenomenon of boundary
conditions on physical fields manifesting themselves as topology will be a
recurrent theme here. Notice, in particular, that, since the Chern number
of an SU (2)-bundle over S 4 is an integer, the possible asymptotic boundary
conditions for finite action, ASD potential on R4 fall into countably many,
discrete “topological types.”

Remark: We have already seen that all of the BPST potentials Aλ,n are
ASD and have Yang-Mills action 8 π 2 . They must, of course, have the same
Yang-Mills action since they all extend to the same bundle, i.e., the Hopf
bundle (which we now see has Chern number 1).
In order to establish contact with the Bogomolny monopole equations we
consider an arbitrary potential  = Âα dxα on R4 with field strength F̂ =
2 F̂αβ dx ∧ dx and use (6.4.4) of [N4] to write the components of the Hodge
1 α β
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 119

dual ∗ F̂ as
1 ∑
4
∗
F̂ αβ = ϵαβγδ F̂αβ , α, β = 1, 2, 3, 4,
2
γ,δ=1

where ϵαβγδ is the Levi-Civita symbol (totally anti-symmetric in α β γ δ with

ϵ1234 = 1). Then the (anti-) self-duality equations take the form

1 ∑
4
F̂αβ = ∓ ϵαβγδ F̂αβ , α, β = 1, 2, 3, 4. (2.5.30)
2
γ,δ=1

Using i, j and k for indices taking the values 1, 2 and 3 one finds (by just
writing them out) that all of these equations are contained in the following:

∑
3
F̂ij = ± ϵijk F̂k4 , i, j = 1, 2, 3, (2.5.31)
k=1

where ϵijk is totally anti-symmetric in ijk and ϵ123 = 1. For example, taking
α = 3 and β = 4 in (2.5.30) gives

1 ∑ 1[ ]
4
F̂34 = ∓ ϵ34γδ F̂γδ = ∓ ϵ3412 F̂12 + ϵ3421 F̂21
2 2
γ,δ=1
1 [ ]
=∓ 2ϵ3412 F̂12 = ∓ ϵ3412 F̂12 = ± ϵ4123 F̂12
2
= ± ϵ123 F̂12

which is equivalent to
F̂12 = ± ϵ123 F̂34

and this is (2.5.31) with i = 1 and j = 2. Similarly, taking α = 1 and β = 2

in (2.5.30) gives
F̂12 = ∓ ϵ1234 F̂34 = ± ϵ123 F̂34

which is also (2.5.31) with i = 1 and j = 2.

Thus, the (anti-) self-duality equations on R4 take the form (2.5.31). Finite
action solutions to these equations are the instantons discussed earlier. We
now wish to seek solutions to (2.5.31) that are static, i.e., for which the Âα
are independent of x4 . Naturally, no such solution can have finite action on R4
(unless it is zero) and so cannot be an instanton. For Âα ’s that are independent
120 2. Physical Motivation

of x4 , (2.5.31) becomes

∑
3
F̂ij = ± ϵijk F̂k4
k=1
∑3 ( [ ])
=± ϵijk ∂k Â4 − ∂4 Âk + Âk , Â4
k=1
∑3 ( [ ])
F̂ij = ± ϵijk ∂k Â4 + Âk , Â4 , i, j = 1, 2, 3 (2.5.32)
k=1

(you may wish to glance back at (2.5.15) if you’re wondering where all of this
is going).
Now we “reduce to R3 ” as follows: Fix some value x40 of x4 and consider the
submanifold R3 ×{x40 } of R4 (which we henceforth identify with R3 ). Restrict
our trivial SU (2)-bundle over R4 to this R3 and obtain a trivial SU (2)-bundle
over R3 . Let A1 , A2 and A3 be the restrictions to R3 of Â1 , Â2 and Â3 (all
of which are assumed independent of x4 ). Then

A = A1 dx1 + A2 dx2 + A3 dx3 = Ai dxi

is a gauge potential on R3 and the corresponding field strength

F = 12 Fij dxi ∧ dxj has components Fij = ∂i Aj − ∂j Ai + [Ai , Aj ] that are just
the restrictions to R3 of the F̂ij , i, j = 1, 2, 3. Note that Â4 does not enter
into either the potential or the field strength on R3 . However, if we define
ϕ : R3 −→ su(2) by
ϕ = Â4 | R3 ,

then the time independent, (anti-) self-duality equations (2.5.32) on R4 be-

come (when restricted to R3 ) the Bogomolny monopole equations

∑
3
Fij = ± ϵijk (∂k ϕ + [Ak , ϕ]), i, j = 1, 2, 3. (2.5.33)
k=1

Thus, an SU (2) Bogomolny monopole on R3 is essentially just a static,

(anti-) self-dual gauge potential on R4 . We intend to describe the geometry
and topology of these monopoles in more detail, but first we will exhibit a
concrete example that will, in some sense, bring us full circle. We began (in
Chapter 0, [N4]) by thinking about Dirac’s magnetic monopoles and finding
that they were most naturally modeled by connections on U (1)-bundles over
S 2 . In particular, the monopole of lowest strength was identified with the
natural connection on the complex Hopf bundle. Seeking generalizations we
looked at the natural connection on the quaternionic Hopf bundle and found
lurking there a BPST instanton. Now we find that a “static instanton” is to
be identified with a particular type of Yang-Mills-Higgs field on R3 which we
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 121

have (for reasons that no doubt remain obscure) called a monopole. To under-
stand the terminology and to see Dirac monopoles in an entirely new light we
will describe now the t’Hooft-Polyakov-Prasad-Sommerfleld monopole,
which is an exact solution to the equations (2.5.33). For this we will need some
notation. First, in R3 we will write

x = (x1 , x2 , x3 )
√
r = |x| = (x1 )2 + (x2 )2 + (x3 )2
ni = xi /r, i = 1, 2, 3 (r ̸= 0)
d⃗x = (dx1 , dx2 , dx3 )

and, in su(2),
1
Ta = − i σa , a = 1, 2, 3
2
T⃗ = (T1 , T2 , T3 ).

In addition we set

⃗n · T⃗ = n1 T1 + n2 T2 + n3 T3 = na Ta
⃗n × T⃗ = (n2 T3 − n3 T2 , n3 T1 − n1 T3 , n1 T2 − n2 T1 )

and
( )
⃗n × T⃗ · d⃗x = (n2 T3 − n3 T2 ) dx1 + (n3 T1 − n1 T3 ) dx2
+ (n1 T2 − n2 T1 ) dx3 .

Our objective is to find (A(x), ϕ(x)) which satisfies (2.5.33) and has the re-
quired asymptotic behavior (∥ϕ∥ −→ 1 as r −→ ∞).
Although the monopole equations (2.5.33) are substantially less complicated
that the full Yang-Mills-Higgs equations, they are still far beyond the means
of elementary techniques. To reduce the level of difficulty a bit more requires
a guess (physicists prefer the term Ansatz) as to the form one might expect
for a solution. Here’s the one that worked for t’Hooft and Polyakov (some
rationale for the Ansatz is discussed in Section 4.2 of [GO]): We will seek
functions f (r) and h(r) satisfying

f (r)
−→ 1 as r −→ ∞ and f (0) = 0, (2.5.34)
r
and
h (r) −→ 0 as r −→ ∞ and h(0) = 1, (2.5.35)
such that
f (r) ( ⃗ ) f (r) a
ϕ (x) = ⃗n · T = 2 x Ta (2.5.36)
r r
122 2. Physical Motivation

and
1 − h(r) ( )
A (x) = ⃗n × T⃗ · d⃗x
r
(2.5.37)
1 − h(r) ∑
3
= ϵaij xj dxi Ta
r2 a=1

give a solution (A, ϕ) to (2.5.33).

Remark: The conditions f (0) = 0 and h(0) = 1 are intended to give ϕ(x)
and A (x) a chance of being smooth at the origin.
Substituting (2.5.36) and (2.5.37) into (2.5.33) (with the minus sign) gives the
following system of ordinary differential equations for f (r) and h(r):



dh
= −hf
 r
dr
(2.5.38)

 df
 r = f − (h − 1).
2
dr
Now make the change of variable
−f = 1 + rF, h = rH
to obtain the new system 


dH
 = HF
dr


 dF = H 2
dr
for F and H. Observe that it follows from these equations that F 2 − H 2 is
constant since
d dF dH
(F 2 − H 2 ) = 2F − 2H = 2F H 2 − 2H(HF ) = 0.
dr dr dr
(1 + f )2 h2 1 2 ( f ) ( f )2 1 2
But since, for r > 0, F 2 − H 2 = 2
− 2 = 2+ + − 2h
r r r r r r r
and this is to approach 1 as r −→ ∞, we must have

F 2 − H 2 = 1.

To get solutions f and h satisfying f (0) = 0 and h(0) = 1 we take for F and
H satisfying F 2 − H 2 = 1 the following:
F (r) = − coth r and H(r) = csch r.
Then
f (r) = r coth r − 1 and h(r) = r csch r
and one can verify directly that these are smooth (even at r = 0, where
they take the required boundary values) and satisfy (2.5.38). Thus, our field
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 123

configuration (A, ϕ) is given by

( ) ( )
1 1 1
ϕ (x) = coth r − ⃗n · T⃗ = coth r − xa Ta (2.5.39)
r r r
( )( )
1
A(x) = − csch r ⃗n × T⃗ · d⃗x
r
( )∑ 3 (2.5.40)
1 1
= − csch r ϵaij xj dxi Ta
r r a=1

Remark: The same calculation for (2.5.33) with the plus sign gives the
same result except that the sign of ϕ is changed.
The exact solution (A, ϕ) given by (2.5.39) and (2.5.40) is remarkable for
a number of reasons (quite aside from the fact that it is an exact solution
which is more than one generally has a right to expect). The form of the thing
itself is extraordinary for its “mixing” of the spatial and internal directions.
For example, (2.5.39) describes a Higgs field which, in the xa -direction in
R3 , a = 1, 2, 3, has only an internal T a -component. It is, in some strange way,
“radial” (Polyakov called it a “hedgehog” solution). Another feature, and one
that will be particularly significant quite soon, is that, despite appearances,
the component functions of ϕ and A are smooth (in fact, real analytic), even
at the origin. For example,
( ) ( )
1 1 r cosh r 1 3
1 r + 2! r + 4!1 5
r + ···
coth r − = −1 = −1
r r sinh r 1 3
r r + 3! r + 5!1 5
r + ···
( )
1 1 + 2! 1 2
r + ···
1 4
r + 4!
= − 1
r 1 + 3! 1 2
r + ···
1 4
r + 5!
( ( ) )
1 1 1
= 1+ − r + ··· − 1
2
r 2! 3!

(by long division) and this is indeed analytic at r = 0. Computing the deriva-
tive of coth r− 1r one finds that it is positive. Moreover, limr−→∞ (coth r− 1r ) =
1 so one obtains the following picture of ∥ϕ∥:

|| φ(r) || = coth r – 1r

0 r

The configuration (A, ϕ) is a globally defined, smooth object on all of R3 .

124 2. Physical Motivation

But how did configurations such as this come to be called “monopoles”?

This is not entirely clear from the “hedgehog” form in which we currently
have the solution written, but will become clear after an appropriate (local)
gauge transformation. We intend to define a gauge transformation on an open
subset of R3 which, at each point in space, rotates the Higgs field (in the
internal space su(2) at that point) from its “radial” direction to a direction
parallel to the internal T3 -axis. One should keep in mind that a gauge trans-
formation is assumed to change only the appearance, not the physics of field
configurations. Our gauge transformation will be a map g from an open subset
of R3 into SU (2) and its effect on ϕ and A will be, as usual,

ϕ −→ ϕg = g −1 ϕg
A −→ Ag = g −1 Ag + g −1 dg.

Specifically, in terms of spherical coordinates (r, φ, θ) on R3 , we let

 φ φ 
cos −e−i θ sin
 2 2 
g(x) =  φ φ . (2.5.41)
ei θ sin cos
2 2

To compute ϕg we proceed as follows:

) (
f (r) a f (r) ( −1 a )
ϕg = g −1 ϕg = g −1 2
x Ta g = 2
g (x Ta ) g
r r
f (r) a ( −1 ) 1 f (r) a ( −1 )
= 2 x g Ta g = − i 2 x g σa g .
r 2 r
( )
, α ∈ R, in SU (2),
α β
Now, notice that, for any g = −β̄ α
( )
α −β
g −1 = β̄ α and
( ) ( )
−1 −1 0 1 −αβ̄ − αβ α2 − β 2
g σ1 g = g g=
1 0 −β̄ 2 + α2 αβ̄ + αβ
( ) ( )
−1 −1 0 −1 αβ̄ − αβ −α2 − β 2
g σ2 g = i g g=i
1 0 β̄ 2 + α2 −αβ̄ + αβ
( )  
1 0 α2 − β β̄ 2αβ
g −1 σ3 g = g −1 g= 
0 −1 2αβ̄ β β̄ − α2
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 125
 φ φ
( ) cos −e−i θ sin
α β  2 2
Now, with g = −β̄ α
=  we have
φ φ
ei θ sin cos
2 2

φ
−αβ̄ − αβ = −α(β̄ + β) = − cos (2Re(β))
2
φ φ x1
= 2 cos sin cos θ = sin φ cos θ =
2 2 r
φ φ
α2 − β 2 = cos2 − e−2i θ sin2
2 2
φ φ
− (cos 2θ − i sin 2θ) sin2
= cos2
2 2
( φ φ ) φ
= cos2 − cos 2θ sin2 + i sin 2θ sin2
2 2 2
( ) ( )
i αβ̄ − αβ = i α β̄ − β = i α(−2i Im(β))
φ φ
= 2α Im (β) = 2 cos (sin θ sin )
2 2
x2
= sin φ sin θ =
r
φ φ φ
−α2 − β 2 = (− cos2 − cos 2θ sin2 ) + i sin 2θ sin2
2 2 2
φ φ
α2 − β β̄ = cos2 − sin2
2 2
x3
= cos φ =
r

φ ( −i θ φ)
2αβ = 2 cos −e sin = −e−i θ sin φ
2 2
= − cos θ sin φ + i sin θ sin φ.

We need just the (1,1) and (1,2) entries of ϕg so we compute as follows: The
(1,1) entry of xa (g −1 σa g) is

x1 (sin φ cos θ) + x2 (sin φ sin θ) + x3 cos φ

( 1) ( 2) ( 3)
1 x 2 x 3 x r2
=x +x +x = = r.
r r r r
126 2. Physical Motivation

The real part of the (1,2) entry of xa (g −1 σa g) is

φ φ φ
x1 (cos2 − cos 2θ sin2 ) + x2 (− sin 2θ sin2 ) + x3 (− cos θ sin φ)
2 2 2
2 φ φ
= (r sin φ cos θ) cos − (r sin φ cos θ) cos 2θ sin2
2 2
2 φ
− (r sin φ sin θ) sin 2θ sin − r cos φ cos θ sin φ
( )2
1 1
= r sin φ cos θ + cos φ
2 2
( )
1 1
− r sin φ cos θ(1 − 2 sin θ)2
− cos φ
2 2
( )
1 1
− 2r sin φ sin θ cos θ
2
− cos φ − r cos φ cos θ sin φ
2 2
1 1
= r sin φ cos θ + r sin ϕ cos φ cos θ
2 2 ( )
1 1
− (r sin φ cos θ − 2r sin φ cos θ sin θ)
2
− cos φ
2 2
( )
1 1
− 2r sin φ cos θ sin2 θ − cos φ − r cos φ cos θ sin φ
2 2
1 1
= r sin φ cos θ + r sin φ cos φ cos θ
2 2( )
1 1
− (r sin φ cos θ) − cos φ − r sin φ cos φ cos θ
2 2
1 1
= r sin φ cos θ − r sin φ cos φ cos θ
2 2
1 1
− r sin φ cos θ + r sin φ cos φ cos θ
2 2
= 0.

The imaginary part of the (1,2) entry of xa (g −1 σa g) is shown to be zero in

the same way so the (1,2) entry itself is zero. Since ϕg takes values in su(2)
we have
( ) ( )
1 f (r) r 0 1 f (r) 1 0
ϕ =− i 2
g
=− i
2 r 0 −r 2 r 0 −1
( )
1
ϕg = coth r − T3 . (2.5.42)
r
Thus, as promised, our Higgs field is now aligned along the third isospin
axis T3 in the internal space at each point of R3 . As to the gauge poten-
tial Ag , we will again need the (1,1) and (1,2) entries in g −1 Ag + g −1 dg.
First observe that
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 127
( )
1 − h(r) ∑
3
g −1 Ag = g −1 ϵaij xj dxi Ta g
r2 a=1

1 − h(r) ∑ ( )
3
= 2
(ϵaij xj dxi ) g −1 Ta g
r a=1

1 1 − h(r) ∑ ( )
3
=− i 2
(ϵaij xj dxi ) g −1 σa g
2 r a=1
1 1 − h(r) [ 3 2 ( )
=− i (x dx − x2 dx3 ) g −1 σ1 g
2 r2
+ (x1 dx3 − x3 dx1 )(g −1 σ2 g)
]
+ (x2 dx1 − x1 dx2 )(g −1 σ3 g) .

Notice that the (1,1) entry of g −1 Ag is zero:

[ ( 1) ( 2)
1 1 − h(r) x x
− i 2
(x 3
dx 2
− x2
dx 3
) + (x 1
dx 3
− x3
dx 1
)
2 r r r
( 3 )]
x
+(x2 dx1 − x1 dx2 )
r
[
1 1 − h(r) 1 3 2
=− i x x dx − x1 x2 dx3 + x1 x2 dx3 − x2 x3 dx1
2 r3
]
+ x2 x3 dx1 − x1 x3 dx2

= 0.

Thus, the (1,1) entry of Ag is the same as the (1,1) entry of g −1 dg (and so,
in particular, ) not depend on the gauge potential A).
( α βdoes
Let g = −β̄ α with α real and depending only on φ and β depending only
( α −β )
on φ and θ. Then g −1 = β̄ α and
 
αφ dφ βφ dφ + βθ dθ
dg =  
−β̄φ dφ − β̄θ dθ αφ dφ

so
  
α −β αφ dφ βφ dφ + βθ dθ
g −1
dg =   
β̄ α −β̄φ dφ − β̄θ dθ αφ dφ
128 2. Physical Motivation

The (1,1) entry of g −1 dg is

ααφ dφ + β β̄φ dφ + β β̄θ dθ

( )
φ 1 φ
= cos − sin dφ
2 2 2
( ( )
−i θ φ) iθ 1 φ
+ −e sin −e cos dφ
2 2 2
( φ ) ( φ )
+ −e−i θ sin −i ei θ sin dθ
2 2
1 φ φ 1 φ φ φ
= − sin cos dφ + sin cos dφ + i sin2 dθ
2 2 2 2 2 2 2
φ
= i sin2 dθ
2
1
= i (1 − cos φ)dθ.
2
Thus, Ag has the form
( )
1
2 i (1 − cos φ)dθ
A =g

( )
1 (−(1 − cos φ)dθ)i
=−
2

which we write as
( )
1 − 12 (2)i (1 − cos φ)dθ
Ag = − . (2.5.43)
2

A similar calculation for the (1,2) entry gives (Ag )1 and (Ag )2 as shown
below.

(ϕg )1 = (ϕg )2 = 0

f (r) 1
(ϕg )3 = = coth r −
r r

(Ag )1 = −h(r)(sin θ dφ + cos θ sin φ dθ)

= −r csch r(sin θ dφ + cos θ sin φ dθ) (2.5.44)

(Ag )2 = h(r)(cos θ dφ − sin θ sin φ dθ)

= r csch r(cos θ dφ − sin θ sin φ dθ)

(Ag )3 = −(1 − cos φ)dθ.

 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 129

Now for the good part. We have already observed that (A, ϕ) is a globally
defined, smooth field configuration on all of R3 . As r −→ ∞ (i.e., as seen
from a distance), the Higgs field approaches the constant value T3 (because
f (r)
−→ 1), whereas (Ag )1 and (Ag )2 approach 0 (because h(r) −→ 0),
r
while (Ag )3 , which does not depend on r, remains fixed at −(1 − cos φ)dθ. As
seen from infinity the potential function A in this gauge assumes the form
 
1  − 2 (2)i (1 − cos φ)dθ
1
0
− 
2 0 1
(2)i (1 − cos φ)dθ
2

which is essentially just the Im C-valued 1-form

1
− (2)i (1 − cos φ)dθ
2
and this is precisely the potential for a Dirac monopole of magnetic charge 2
(see page 68). Seen from afar, the (nonsingular) t’ Hooft-Polyakov monopole
looks like a Dirac monopole. The “string singularity” (page 4, [N4]) of the
Dirac monopole now shows up in the fact that our gauge transformation g
(defined in terms of spherical coordinates on R3 ) is singular on the nonpositive
x3 -axis.
Remark: The essential point to keep in mind here is that, in classical elec-
trodynamics, monopoles are (but certainly need not be) “put in by hand,”
whereas, in SU (2) Yang-Mills-Higgs theory, they arise of their own accord as
solutions to the field equations.
Since the t’Hooft-Polyakov monopole satisfies F = ∗ dA ϕ one can compute
the field strength F entirely from dA ϕ. In gauge sg ,

dA ϕ = dϕ + [A, ϕ]
( ) [ ]
f (r) f (r)
=d T3 + Aa Ta , T3
r r
( )
f (r) f (r) a
=d T3 + A [Ta , T3 ]
r r
( )
f (r) f (r) [ 1 ]
=d T3 + A [T1 , T3 ] + A2 [T2 , T3 ] + A3 [T3 , T3 ]
r r
( )
f (r) f (r)
=d T3 + [−A1 T2 + A2 T1 ]
r r
( )
f (r) 2 ∂ f (r)
= (A T1 − A1 T2 ) + dr T3
r ∂r r
( ) ( )
1 1
= coth r − (A2 T1 − A1 T2 ) + − csch 2
r dr T3
r r2
130 2. Physical Motivation
( )
A 1 [
d ϕ = coth r − (r csch r) (cos θ dφ − sin θ sin φ dθ)T1
r
]
+ (sin θ dφ + cos θ sin φ dθ)T2 (2.5.45)
( )
1
+ − csch r dr T3
2
r2

Let us return to an arbitrary configuration (A, ϕ) satisfying the monopole

equations F = ±∗ dA ϕ. We have already seen (page 134) that for such a
configuration
⟨ ⟩
∥ F ∥2 + dA ϕ = ±2 F , ∗ dA ϕ
2

so
∫
1 ( )
A(A, ϕ) = ∥ F ∥2 + ∥ dA ϕ∥2 dx1 ∧ dx2 ∧ dx3
2 R3
∫ ⟨ ⟩
=± F , ∗ dA ϕ dx1 ∧ dx2 ∧ dx3
R3
∫
( )
=∓ 2 trace F ∧ ∗∗dA ϕ
R3
∫
( )
=∓ 2 trace F ∧ dA ϕ
R3
which we now write as
∫
( )
A(A, ϕ) = ∓ Tr F ∧ dA ϕ (monopole), (2.5.46)
R3
where Tr = 2 trace.
∫ Computing this integral
∫ for the t’Hooft-Polyakov monopole
(by writing it as R3 Tr(∗ dA ϕ∧dA ϕ) = − R3 ∥ dA ϕ ∥2 dx1 ∧dx2 ∧dx3 and using
(2.5.45)) gives a value of 4π. For any configuration (A, ϕ) ∈ C satisfying the
monopole equations (2.5.14) we define the monopole number N (A, ϕ) by
∫
1
N (A, ϕ) = Tr(F ∧ dA ϕ). (2.5.47)
4π R3

There are alternative ways of computing N (A, ϕ) (several of which are dis-
cussed below) that make it clear that N (A, ϕ) is actually an integer. Indeed,
such an integer-valued monopole number∫can be defined in a much more gen-
A
R3 Tr(F ∧ d ϕ) is well-defined and
1
eral context. In [JT] it is shown that 4π
integer-valued for any (A, ϕ) ∈ C that is a critical point for the action A given
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 131

by (2.5.10). Then [Groi2] shows that it is not even necessary to assume (A, ϕ)
is a critical point. More precisely, if
∫
1 ( )
A(A, ϕ) = ∥ F ∥2 + ∥ dA ϕ ∥2 dx1 ∧ dx2 ∧ dx3
2 R3

and { }
C= ( A, ϕ) : A ( A, ϕ) < ∞, lim sup 1 − ∥ϕ∥ = 0 ,
R−→∞ | x |≥R

then, for any (A, ϕ) ∈ C,

∫
1 ( )
N (A, ϕ) = Tr F ∧ dA ϕ
4π R3

is a well-defined integer called the monopole number of (A, ϕ) (even though

we do not assume that (A, ϕ) is a solution to the monopole equations).
Remark: The results of [Groi2] are actually much more general than this,
but these will do for our purposes. Much of what we have to say also extends
to the λ > 0 case which we abandoned on page 133 (see [JT] and [Groi1]).
We will now describe some alternative ways of computing N (A, ϕ) ∫ that
shed much ∫ light on its topological significance. First
∫ we think of R3 as
limR−→∞ | x |≤R and apply Stokes’ Theorem to each | x |≤R . For this we first
note that
( ) ( )
Tr F ∧ dA ϕ = d Tr(ϕF ) , (2.5.48)

where ϕF is a matrix product. To see this note that Tr(dA ϕ) = Tr(dϕ +

[A, ϕ]) = Tr(dϕ)+ Tr([A, ϕ]) = Tr(dϕ) because any commutator has trace
zero (Tr(AB)= Tr(BA)). Now, using a few simple computational facts that
we will establish in Chapter 4,
( ) ( )
d (Tr(ϕF )) = Tr d(ϕF ) = Tr dA (ϕF )

= Tr(ϕdA F + dA ϕ ∧ F ) (“Product Rule”)

= Tr(dA ϕ ∧ F ) (Bianchi identity)

= Tr(F ∧ dA ϕ ),

where the last equality follows from the fact that dA ϕ ∧ F and F ∧ dA ϕ differ
by a bracket, which has trace zero. This proves (2.5.48) and Stokes’ Theorem
(Section 4.7) gives
∫ ( ) ∫ ∫
A
( )
Tr F ∧ d ϕ = d Tr( ϕF ) = Tr( ϕF ).
| x |≤R | x |≤R | x |=R
132 2. Physical Motivation

Writing SR2
for the set of points in R3 with | x | = R we obtain
∫ ∫
1 A 1
N ( A, ϕ) = Tr ( F ∧ d ϕ) = lim Tr ( F ∧ dA ϕ)
4π R3 R−→∞ 4π | x |≤R

∫
1
N ( A, ϕ) = lim Tr (ϕF ). (2.5.49)
R−→∞ 4π 2
SR

Since limR−→∞ sup| x |≥R | 1 − ∥ ϕ ∥ | = 0, there exists an R0 < ∞ such that

∥ ϕ(x) ∥ > 12 for all x with | x | > R0 . For any such x we define

ϕ̂ ( x ) = ∥ ϕ (x) ∥−1 ϕ (x)

and, if R > R0 ,
ϕ̂R = ϕ̂|SR2 .

2
These are smooth maps on their domains and they map into Ssu(2) (the unit
∼
2-sphere in su(2) = R ). One can show (see [JT] and [Groi2]) that ϕ can be
3

replaced by ϕ̂ in (2.5.49), i.e.,

∫
1
N (A, ϕ) = lim Tr( ϕ̂F )
R−→∞ 4π S 2
R
∫ (2.5.50)
1 −1
= lim ∥ϕ∥ Tr(ϕF ).
R−→∞ 4π S 2
R

The reason for preferring the maps ϕ̂ and ϕ̂R can be seen as follows: Each
ϕ̂R can be regarded as a map from S 2 to S 2 and so determines an element
[ϕ̂R ] of the homotopy group π2 (S 2 ). Since ϕ̂ is smooth for R > R0 , ϕ̂R varies
smoothly with R > R0 so this homotopy class is independent of R > R0
and we will denote it simply [ϕ̂]. We claim that [ϕ̂] is also gauge invariant,
i.e., that if g : R3 −→ SU (2) is a gauge transformation and ϕg = g −1 ϕg,
then, on | x | > R0 , ∥ ϕg ∥−1 ϕg is well-defined and homotopic to ∥ ϕ ∥−1 ϕ. It is
well-defined because ∥ ϕg ∥ = ∥ g −1 ϕg ∥ = ∥ ϕ ∥ which is nonzero on | x | > R0 .
On the other hand, since R3 is contractible, g is homotopic to the map that
sends all of R3 the identity e in SU (2) (Exercise 2.3.6, [N4]). Thus, on | x | >
R0 , ∥ ϕg ∥−1 ϕg = ∥ ϕ ∥−1 (g −1 ϕg) is homotopic to ∥ ϕ ∥−1 (e−1 ϕe) = ∥ ϕ ∥−1 ϕ
so [ϕg ] = [ϕ] as required.
Each map ϕ̂R can be regarded as a map from S 2 to S 2 and therefore has
a Brouwer degree deg(ϕ̂R ) (see Section 5.7 or Section 3.4 of [N4]). Since the
various maps ϕ̂R , R > R0 , determine the same homotopy class in π2 (S 2 ),
they have the same degree. Remarkably, this degree actually coincides with
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 133

the monopole number, i.e.,

( ) ( )
N ( A, ϕ) = deg ϕ̂R = deg ∥ ϕ ∥−1 ϕ|SR2 (R > R0 ) (2.5.51)

(see [JT] and [Groi2]). Notice that deg(∥ ϕ ∥−1 ϕ|SR2 ) depends only on ϕ
and, indeed, only on its asymptotic behavior. Even more, it depends only
on the “homotopy type of its asymptotic behavior” (if you get my drift).
The monopole number distinguishes “homotopy classes” of Higgs fields. These
classes are stable in the sense that a continuous perturbation of the field can-
not change the class (physicists would say that an infinite potential barrier
separates fields with different monopole numbers). Mathematically, there is
a natural topology on the configuration space C with path components la-
beled by the integers and such that two configurations lie in the same path
component if and only if they have the same monopole number (see [Groi2]).
We point out that there is an explicit integral formula for calculating the
degrees (monopole numbers) in (2.5.51) that is sometimes more manageable
than those in earlier formulas:
∫ ( )
1
N (A, ϕ) = − Tr ϕ̂dϕ̂ ∧ dϕ̂ (R > R0 ). (2.5.52)
4π SR2

We’ll do this calculation for the t’ Hooft-Polyakov monopole. From (2.5.39)

and the fact that ∥ ϕ(x) || = coth r − 1r we conclude that, on SR2
(R > R0 ),
 
3 2 1
xa
1  x i x + x i
ϕ̂(x) = ⃗n · T⃗ = Ta = − .
R 2R −x2 + x1 i −x3 i

Thus,
dϕ̂ ∧ dϕ̂
( )( )
1 dx3 i dx2 + dx1 i dx3 i dx2 + dx1 i
=
4R 2 −dx2 + dx1 i −dx3 i −dx2 + dx1 i −dx3 i
( )
1 dx1 ∧ dx2 i −dx1 ∧ dx3 + dx2 ∧ dx3 i
=−
2R 2 dx1 ∧ dx3 + dx2 ∧ dx3 i −dx1 ∧ dx2 i

and so
( )
1 x3 i x2 + x1 i
ϕ̂ dϕ̂ ∧ dϕ̂ =
4R 3 −x2 + x1 i −x3 i
( )
dx1 ∧ dx2 i −dx1 ∧ dx3 + dx2 ∧ dx3 i
× .
dx1 ∧ dx3 + dx2 ∧ dx3 i −dx1 ∧ dx2 i
134 2. Physical Motivation

The (1,1) entry is

1 [( 3 1 )
3
−x dx ∧ dx2 + x2 dx1 ∧ dx3 − x1 dx2 ∧ dx3
4R
( ) ]
+ x1 dx1 ∧ dx3 + x2 dx2 ∧ dx3 i .

and the (2,2) entry is

[( 2 1 )
1
4R 3 x dx ∧ dx3 − x1 dx2 ∧ dx3 − x3 dx1 ∧ dx2
( ) ]
− x1 dx1 ∧ dx3 + x2 dx2 ∧ dx3 i .

Thus,
( ) ( )
Tr ϕ̂dϕ̂ ∧ dϕ̂ = 2 trace ϕ̂dϕ̂ ∧ dϕ̂
1 ( 1 2 )
=− 3
x dx ∧ dx3 − x2 dx1 ∧ dx3 + x3 dx1 ∧ dx2 .
R
2
We will learn how to integrate such a 2-form over SR in Chapter 4 (indeed,
we will find that the restriction of x dx ∧ dx − x dx1 ∧ dx3 + x3 dx1 ∧ dx2 to
1 2 3 2

S 2 is just the standard volume (i.e., area) form on S 2 ). Once the machinery
is all in hand we will find that one can calculate such things by simply doing
what comes natural. In this case, one introduces spherical coordinates

x1 = R sin φ cos θ

x2 = R sin φ sin θ

x3 = R cos φ

with 0 ≤ θ ≤ 2π and 0 ≤ φ ≤ π, computes

x1 dx2 ∧ dx3 = (R sin φ cos θ) d (R sin φ sin θ) ∧ d (R cos φ)

= R3 sin φ cos θ(cos φ sin θdφ + sin φ cos θdθ) ∧ (− sin φdφ)

= −R3 sin3 φ cos2 θdθ ∧ dφ

and similarly for the remaining terms. The result is

∫ ( ) ∫
Tr ϕ̂ dϕ̂ ∧ dϕ̂ = − sin ϕ dϕ ∧ dθ
2
SR (0,π)×(0,2π)
∫ 2π ∫ π
=− sin ϕ dϕdθ
0 0

= −4π.
 2.5. SU (2)-Yang-Mills-Higgs Theory on Rn 135

Thus, for the t’ Hooft-Polyakov monopole,

∫ ( )
1
N (A, ϕ) = − Tr ϕ̂ dϕ̂ ∧ dϕ̂ = 1.
4π SR2

Remark: Changing the sign of ϕ (but leaving A alone) gives a configuration

which is a solution to F = ∗ dA ϕ (see page 144) and has monopole number
−1.
2
The behavior of the Higgs field on large 2-spheres SR therefore captures
the topological type of the configuration. There is yet another way of seeing
this that is reminiscent of our earlier experience with instantons (where the
topological type was to be found in the Chern class of a certain bundle). To
see this we again fix
{ }
(A, ϕ) ∈ C = (A, ϕ) : A(A, ϕ) < ∞, lim sup |1 − ∥ϕ∥ | = 0
R−→∞ |x|≥R

and select R0 < ∞ such that ∥ϕ(x)∥ > 1

2 on |x| > R0 . Define

ϕ̂ = ∥ϕ∥−1 ϕ : R3 − {x : |x| ≤ R0 } −→ Ssu(2)

2

and, for each R > R0 ,

ϕ̂R = ϕ̂ 2
SR
2
: SR −→ Ssu(2)
2
.

Now we fix some R > R0 . ϕ is the pullback by the standard cross-section of

an equivariant map Φ : R3 × SU (2) −→ su (2). The restriction of the trivial
SU (2)-bundle over R3 to SR
2 2
is the trivial SU (2)-bundle over SR :
P
2
SU (2) ,→ SR × SU (2) −→ SR
2
.

Let ΦR = ΦSR2 ×SU (2) and Φ̂R = ∥ΦR ∥−1 ΦR . Both are equivariant and Φ̂R
2 2
takes values in Ssu(2) . Furthermore, ϕ̂R is the pullback to SR by the standard
cross-section of Φ̂R . Thus, ϕ̂R is the standard gauge representation for a Higgs
2 2
field on the trivial SU (2)-bundle over SR with values in Ssu(2) .
Now, select some ϕ0 ∈ Ssu(2) (a “ground state” for the “virtual potential”;
2

see pages 132–133). The isotropy subgroup of ϕ0 (with respect to the adjoint
action of SU (2) on su(2)) is a copy of U (1) in SU (2) (pages 132–133). One
−1 2
can show that Φ̂R (ϕ0 ) is a submanifold of SR × SU (2) (because Φ̂R is a
−1
submersion at each point of Φ̂R (ϕ0 )) and, furthermore
−1
(i) for each x ∈ SR
2
, P −1 (x) ∩ Φ̂R (ϕ0 ) ̸= ∅, and
136 2. Physical Motivation
−1
(ii) for p ∈ Φ̂R (ϕ0 ) and g ∈ SU (2),
−1
p · g ∈ Φ̂R (ϕ0 ) iff g ∈ U (1) (isotropy subgroup of ϕ0 ).

From these it follows that

−1 −1
P Φ̂R (ϕ0 ) : Φ̂R (ϕ0 ) −→ SR
2

2 −1
is a principal U (l)-bundle over SR (where the action of U (1) on Φ̂R (ϕ0 )
2
is just the original SU (2)-action on SR × SU (2), but with p restricted to
−1
Φ̂R (ϕ0 ) and g restricted to U (1) ⊆ SU (2)). This U (1)-bundle over SR
2
is
P
called a reduction of the structure group of SU (2) ,→ SR 2
× SU (2) −→ SR 2
to
U (1). Recall that principal U (1)-bundles over spheres are characterized up to
equivalence by their 1st Chern number (see pages 63–64 and 68 of Section 2.2).
The result of interest to us is the following: The 1st Chern number of
−1
−1 P Φ̂R (ϕ0 )
U (1) ,→ Φ̂R (ϕ0 ) −−−−−−−→ SR
2

is the monopole number N (A, ϕ) of the configuration (A, ϕ).

Since the Chern class can be computed from any connection on the bundle,
the proof amounts to finding such a connection that arises naturally from the
original Yang-Mills-Higgs potential. We will briefly illustrate how this is done
(the following is a special case of Proposition 6.4, Chapter II, of [KN1]):
We have a connection on SU (2) ,→ R3 × SU (2) −→ R3 . Its restriction to
P
SU (2) ,→ SR 2
× SU (2) −→ SR 2
is a connection which we will denote ω. Since
U (1) is a subgroup of SU (2), u(1) is a subalgebra of su(2). Now, we have an
ad-invariant, positive definite inner product

⟨A, B⟩ = −2trace(AB)

on su(2). Let u(1)⊥ be the ⟨ , ⟩-orthogonal complement of u(1) in su(2). Then

su(2) = u(1) ⊕ u(1)⊥ and u(1)⊥ is also ad-invariant (we let µ : SU (2) −→
−1
GL(u(1)⊥ ) be the induced representation). If ι : Φ̂R (ϕ0 ) ,→ SR
2
×SU (2) is the
−1
inclusion map, then ι∗ ω is an su(2)-valued 1-form on Φ̂R (ϕ0 ) and therefore
splits
ι∗ ω = ω 0 + γ,
where ω 0 is u(1)-valued and γ is u(1)⊥ -valued. It’s easy to see that ω 0 is a
connection 1-form on
−1
−1 P Φ̂R (ϕ0 )
U (1) ,→ Φ̂R (ϕ0 ) −−−−−−−→ 2
SR
 2.6. Epilogue 137

(and γ is “tensorial of type µ”). Computing the 1st Chern number of this
bundle from ω 0 gives the expression (2.5.52) for N (A, ϕ).

2.6 Epilogue
We hope by now to have satisfied the curiosity of those who may have won-
dered how such apparently abstruse mathematical notions as spinor structures
and characteristic classes might arise in the study of the world around us. We
will have one more serious encounter with this in the Appendix, but it is time
now to put aside the informal, heuristic, discussions that have characterized
this chapter and deal honestly with these notions for their own sake. The
remainder of the book is intended to do just that. Certainly, one need not
demand any physical motivation to study and appreciate the rather beautiful
mathematics to follow. Nevertheless, it is pro-foundly satisfying that the phys-
ical motivation exists and, as the concepts are made precise and the theorems
are rigorously proved, we recommend a periodic dip in the murkier waters
of physics (the journal Communications in Mathematical Physics is fine for
browsing). It lends perspective.
3
Frame Bundles and Spacetimes

3.1 Partitions of Unity, Riemannian

Metrics and Connections
An n-dimensional smooth manifold X is locally diffeomorphic to Rn and
the study of such manifolds generally necessitates piecing together well-
understood local information about Rn into global information about X. The
basic tool for this piecing together operation is a partition of unity. In this
section we will prove that these exist in abundance on any manifold and use
them to establish the global existence of two useful objects that clearly always
exist locally (Riemannian metrics and connection forms). The same technique
will be used in Section 4.3 to obtain a convenient reformulation of the notion
of orientability while, in Section 4.6, a theory of integration is constructed on
any orientable manifold by piecing together Lebesgue integrals on coordinate
neighborhoods.
We begin with a purely topological lemma. A family {Aα : α ∈ A} of
subsets of a topological space X is said to be locally finite if each x ∈ X has
a neighborhood U such that U ∩ Aα ̸= ∅ for at most finitely many α ∈ A.
Exercise 3.1.1 Show that, if {Aα : α ∈ A} is locally finite and Āα denotes
the closure of Aα in X, then {Āα : α ∈ A} is also locally finite.
Recall (Section 1.4, [N4]) that a subset U of X is relatively compact if its
closure Ū is compact and that X itself is said to be locally compact if it is
Hausdorff and each point in X has a relatively compact open neighborhood.
If X is also second countable (i.e., has a countable basis for its topology), then
it has a countable basis of relatively compact open sets (Lemma 1.4.9, [N4]).
Manifolds are, in particular, locally compact, second countable topological
spaces. Finally, recall (Section 4.3, [N4]) that, if U and V are two covers of X,
then V is said to be a refinement of U if every V ∈ V is contained in some
U ∈ U.
Lemma 3.1.1 Let X be a locally compact, second countable topological space.
Then any open cover of X has a countable, locally finite refinement consisting
of relatively compact open sets.
Proof: Let {B1 , B2 , . . .} be a basis for the open sets of X with each B̄i com-
pact. We begin by constructing an auxiliary countable cover {W1 , W2 , . . .}
of X consisting of relatively compact open sets and with the property that
W̄i ⊆ Wi+1 for each i = 1, 2, . . . . Set W1 = B1 and assume (for the in-
ductive construction) that W1 , . . . , Wn have been defined so that Wi is open

G.L. Naber, Topology, Geometry and Gauge fields: Interactions, 139

Applied Mathematical Sciences 141, DOI 10.1007/978-1-4419-7895-0_3,
© Springer Science+Business Media, LLC 2011
140 3. Frame Bundles and Spacetimes

and relatively compact for i ≤ n and W̄i ⊆ Wi+1 for i ≤ n − 1. Since W̄n
is compact it is contained in a finite union of the Bi . Let in be the least
positive integer greater than or equal to n such that W̄n ⊆ B1 ∪ · · · ∪ Bin .
Set Wn+1 = B1 ∪ · · · ∪ Bin . Then Wn+1 is open, W̄n ⊆ Wn+1 and W̄n+1 =
B1 ∪ · · · ∪ Bin = B̄1 ∪ · · · ∪ B̄in is compact so the induction is complete and
we have a sequence {W1 , W2 , . . .} of relatively compact open sets and with
W̄i ⊆ Wi+1 for each i = 1, ∪2, . . . . Since Wi contains the union of the first i
∞
elements of {B1 , B2 , . . .}, i=1 Wi = X and {W1 , W2 , . . .} is a cover of X.

Exercise 3.1.2 Show that each of the sets W̄2 , W̄3 − W2 , W̄4 − W3 , . . . ,
W̄i − Wi−1 , . . . is compact, each of the sets W3 , W4 − W̄1 , W5 − W̄2 , . . .,
Wi+1 − W̄i−2 , . . . is open, and
W̄2 ⊆ W3
W̄3 − W2 ⊆ W4 − W̄1
W̄4 − W3 ⊆ W5 − W̄2
..
.
W̄i − Wi−1 ⊆ Wi+1 − W̄i−2
..
.
With this auxiliary cover in hand we can now prove the lemma. Let U =
{Uα : α ∈ A} be an arbitrary open cover of X. Define U2 = {Uα ∩W3 : α ∈ A}
and, for i ≥ 3, Ui = { Uα ∩ (Wi+1 − W̄i−2 ) : α ∈ A}. Then U2 covers W̄2
and, for i ≥ 3, Ui covers W̄i − Wi−1 . Let V2 be a finite collection of ele-
ments of U2 that cover W̄2 and, for i ≥ 3, let Vi be a∪finite collection of
∞
elements of Ui that cover W̄i − Wi−1 . Finally, let V = i=2 Vi . Then V is
a countable family of open sets in X. Each element of V is contained in
some Uα as well as in some W̄j so that V refines U and the elements of V
have compact closure. To see that V is a cover, let x be an arbitrary el-
ement of X. Select the least integer i ≥ 1 such that x ∈ Wi . If i = 1,
or 2, then x ∈ W̄2 and therefore in some element of V2 . If i ≥ 3, then
x ∈ W̄i − Wi−1 so x is in some element of Vi . This also implies that V is
locally finite since any x ∈ X is in one of the compact sets W̄2 , W̄3 − W2 , . . .
and the corresponding open set W3 , W4 − W̄1 , . . . is an open neighbor-
hood of x which, by construction, intersects only finitely many elements
of V. 

For any real-valued function f on a topological space X we define the sup-

port of f , denoted supp f , to be the closure in X of the set of points on which
f is nonzero, i.e.,
supp f = {x ∈ X : f (x) ̸= 0}.
Thus, f is identically zero on X − supp f , but may be zero at some points of
supp f as well. Now we return to the manifold setting. A (smooth) partition
 3.1. Partitions of Unity, Riemannian Metrics and Connections 141

of unity on the n-dimensional smooth manifold X is a collection {ϕγ : γ ∈ C}

of smooth, non-negative real-valued functions on X such that

1. Each supp ϕγ is compact.

2. {supp ϕγ : γ ∈ C} is locally finite.
∑
3. For each x ∈ X, γ∈C ϕγ (x) = 1.

Remark: Notice that, by #2, each x ∈ X has a neighborhood on which at

most finitely many of the ϕγ are nonzero so the sum in #3 is a finite sum. The
fact that the sum is 1 implies that the sets supp ϕγ , γ ∈ C, cover X. Moreover,
since the ϕγ are assumed non-negative, we must have 0 ≤ ϕγ (x) ≤ 1 for each
x ∈ X.

If U = {Uα : α ∈ A} is an open cover of X, then the partition of unity

{ϕγ : γ ∈ C} is said to be subordinate to U if each supp ϕγ is contained
in some Uα . We wish to prove that any open cover of any smooth manifold
has a (countable) partition of unity subordinate to it. For this we will need to
use the fact that manifolds have rich supplies of smooth real-valued functions.
More precisely, we will appeal to the following two results, both proved in
[N4].

Lemma 3.1.2 Let X be a smooth manifold, U a coordinate neighborhood in

X and A0 and A1 disjoint closed subsets of U. Then there exists a smooth
function f : U −→ R satisfying 0 ≤ f (x) ≤ 1 for all x ∈ U, A0 = f −1 (0) and
A1 = f −1 (1).

This is Exercise 5.4.4 of [N4]. The following is Lemma 5.4.2 of [N4].

Lemma 3.1.3 Let W be an open subset of the smooth manifold X and p a

point in W . Then there exists a non-negative, C ∞ function g on X that is
1 on an open neighborhood of p in W and 0 outside W (g is called a bump
function at p in W ).

With these we can prove our major result.

Theorem 3.1.4 Let X be a smooth manifold and U = {Uα : α ∈ A} an

open cover of X. Then there exists a countable partition of unity {ϕk }k=1,2,...
subordinate to U.
Proof: It will clearly suffice to find such a partition of unity subordinate to
any refinement of U so, by intersecting with the coordinate neighborhoods in
some atlas for X, we might as well assume at the outset that each Uα is a
coordinate neighborhood. For each x ∈ X we can select an open neighborhood
of x whose closure is contained in some Uα . Let U ′ be the open cover of X
consisting of all of these neighborhoods. By Lemma 3.1.1, there is a countable,
142 3. Frame Bundles and Spacetimes

locally finite, open refinement V = {Vk }k=1,2,... of U ′ with each V̄k compact.
Thus, V is also a refinement of U and, moreover, each V̄k is contained in
some element of U. By Exercise 3.1.1, {V̄k }k=1,2... is also locally finite so, to
complete the proof, it will suffice to find a partition of unity {ϕk }k=1,2... with
supp ϕk = V̄k for each k = 1, 2, . . . .
For each k = 1, 2, . . . select α(k) ∈ A such that V̄k ⊆ Uα(k) . Then Uα(k) − Vk
is a nonempty, proper closed subset of Uα(k) so, by Lemma 3.1.2, there is a non-
negative smooth function fk′ on Uα(k) with (fk′ )−1 (0) = Uα(k) − Vk . From this
we construct a non-negative smooth function fk on X with fk−1 (0) = X − Vk
as follows: For each p ∈ V̄k select a non-negative bump function gp which
is 1 on a neighborhood Up of p in Uα(k) and 0 on X − Uα(k) . Cover V̄k by
finitely many of these neighborhoods Up1 , . . . , Upj and let g = gp1 + · · · + gpj .
Then g is C ∞ on X, nonzero on V̄k and 0 on X − Uα(k) . Define fk on X
by 
 fk′ (x)g(x), x ∈ Uα(k)
fk (x) = .
 0, x ∈ X − Uα(k)

Exercise 3.1.3 Show that fk is a well-defined smooth map on X with

fk−1 (0) = X − Vk .
From Exercise 3.1.3 we conclude that supp fk = V̄k . Thus, {supp fk }k=1,2...
is locally finite so we can define a smooth real-valued function f on X
by
∞
∑
f (x) = fk (x)
k=1

for each x ∈ X. Moreover, since every x ∈ X is in some Vk , f (x) > 0 for each
x ∈ X. Thus, for each k = 1, 2, . . . , the function ϕk : X −→ R defined
by
fk (x)
ϕk (x) =
f (x)
is non-negative, C ∞ , has supp ϕk = supp fk
∑ = V̄k and satisfies
∞
k=1 ϕk (x) = 1 for each x ∈ X as required. 
The functions in a partition of unity are required to have compact support.
On occasion it is convenient to drop this requirement.

Corollary 3.1.5 Let X be a smooth manifold and U = {Uα : α ∈ A} an

open cover of X. Then there exists a family {ϕα : α ∈ A} of smooth,
non-negative real-valued functions on X such that supp ϕα ⊆ Uα for∑each
∞
α ∈ A, {supp ϕα : α ∈ A} it is locally finite and, for each x ∈ X, k=1
ϕk (x) = 1.
 3.1. Partitions of Unity, Riemannian Metrics and Connections 143

Proof: Choose a countable partition of unity {ϕ′k }k=1,2,... on X subordinate

to U. Define ϕα to be identically zero if no ϕ′k has support in Uα and, otherwise,
let ϕα be the sum of the ϕ′k with support in Uα .

Exercise 3.1.4 Show that {ϕα : α ∈ A} has the required properties. 

Exercise 3.1.5 Let X be a smooth manifold, U an open subset of X and A a

closed subset of X with A ⊆ U . Show that there exists a smooth, real-valued
function ϕ on X satisfying 0 ≤ ϕ(x) ≤ 1 for each x ∈ X, ϕ(x) = 1 for each
x ∈ A and ϕ(x) = 0 for each x ∈ X − U . Hint: Consider the open cover
{U, X − A} of X and use Corollary 3.1.5.

We conclude this section with two important applications of Theorem 3.1.4.

Theorem 3.1.6 Any smooth manifold X admits a Riemannian metric.

Proof: Observe first that if (U, φ) is a chart on X, then the open submanifold
U of X admits a Riemannian metric (e.g., φ∗ ḡ , where ḡ is the standard
metric on φ(U ) ⊆ Rn ). Now, let {(Uα , φα ) : α ∈ A} be an atlas for X.
By Theorem 3.1.4 there is a countable partition of unity {ϕk }k=1,2,... on
X subordinate to {Uα : α ∈ A}. For each k = 1, 2, . . . choose α(k) ∈ A
such that supp ϕk ⊆ Uα(k) . Then {Uα(k) }k=1,2,... is a countable subcover of
{Uα : α ∈ A}. On each Uα(k) select a Riemannian metric g k and define g by

∞
∑
g= ϕk g k .
k=1

More precisely, we define g as follows: For each p ∈ X and all v , w ∈ Tp (X)

we let
∞
∑
g (p)(v , w ) = ϕk (p)g k (p)(v , w ), (3.1.1)
k=1

where it is understood that, if p ̸∈ supp ϕk , then the term ϕk (p)g k (p)(v , w )

is taken to be zero (ϕk (p) is zero, of course, but g k (p) will not be defined if
p ̸∈ Uα(k) ). Thus, the sum is finite.
Each g (p) defined by (3.1.1) is clearly symmetric and bilinear. Moreover,
g (p)(v , v ) ≥ 0 and, if v ̸= 0, g (p)(v , v ) > 0 since no term in the sum (3.1.1)
is negative and at least one must be positive if w = v . Thus, each g (p)
is a positive definite inner product on Tp (X). All that remains is to prove
smoothness and this we may do locally. Fix a p ∈ X. Then p has an open
neighborhood that intersects only finitely many of the sets supp ϕk . Intersect
this neighborhood with all of the corresponding Uα(k) containing p to obtain
a coordinate neighborhood U of p. Restrict to U the coordinates x1 , . . . , xn
144 3. Frame Bundles and Spacetimes

of any
∑∞one of these coordinate neighborhoods Uα(k) . Then, on U , the sum
g = k=1 ϕk g k defining g is finite, each ϕk is a C ∞ function of x1 , . . . , xn
∂ ∂ ∞ ∂ ∂ ∞
and each g k ( ∂x i , ∂xj ) is C so g ( ∂x i , ∂xj ) is C for i, j = 1, . . . , n. Thus,
g is smooth on U as required. 

An analogous argument establishes the existence of a connection form on

any smooth principal bundle.
P
Theorem 3.1.7 Any smooth principal bundle G ,→ P −→ X admits a con-
nection.
Proof: First note that if (U, φ) is a chart for which U is a trivializing
P
neighborhood, then the trivial bundle G ,→ P −1 (U ) −→ U admits a con-
−1
nection. Indeed, if Ψ : P (U ) −→ U × G is a trivialization, π : U × G −→ G
is the projection and Θ is the Cartan 1-form for G, then π ∗ Θ is a (flat) con-
nection form on U × G (Exercise 6.2.12, [N4]) so Ψ∗ (π ∗ Θ) = (π ◦ Ψ)∗ Θ is a
connection form on P −1 (U ) (Theorem 6.1.3, [N4]).
Now let {(Uα , φα ) : α ∈ A} be an atlas for X with each Uα a trivializing
P
neighborhood for G ,→ P −→ X. Let {ϕk }k=1,2,... be a countable partition of
unity subordinate to U = {Uα : α ∈ A}. For each k, choose α(k) ∈ A such
that supp ϕk ⊆ Uα(k) . Then {Uα(k) }k=1,2,... is a countable subcover of U. For
each k, let ω k be a connection form on P −1 (Uk ) and define
∞
∑
ω= (ϕk ◦ P) ω k .
k=1

More precisely, for each p ∈ P and all v ∈ Tp (P ) we set

∞
∑
ω(p)(v) = ϕk (P(p))ω k (p)(v), (3.1.2)
k=1

where it is understood that, if P(p) ̸∈ supp ϕk , then the term ϕk (P(p))ω k (p)
(v) is taken to be zero. Note that (3.1.2) does, indeed, define a G-valued 1-
form on P since, at each p, the sum is finite and G is a real vector space.

Exercise 3.1.6 Show that ω is smooth.

We show that ω is a connection form by proving

σg∗ ω = adg−1 ◦ ω (3.1.3)

for all g ∈ G and

ω(A# ) = A (3.1.4)
 3.2. Continuous Versus Smooth 145

for all A ∈ G. (3.1.4) is easy because

∞
∑
ω(p)(A# (p)) = ϕk (P(p))ω k (p)(A# (p))
k=1
∞
( ∞
)
∑ ∑
= ϕk (P(p)) A = ϕk (P(p)) A
k=1 k=1
= 1A = A.

(3.1.3) follows from the linearity of adg−1 . Indeed, for g ∈ G, p ∈ P and

v ∈ Tp·g−1 (P ),
( )
(σg∗ ω) (p · g −1 )(v) = ω(p) (σg )∗p·g−1 (v)
∞
∑ ( )
= ϕk (P(p)) ω k (p) (σg )∗p·g−1 (v)
k=1
∑∞ ( )
= ϕk (P(p)) σg∗ ω k (v)
p·g −1
k=1
∑∞ ( )
= ϕk (P(p)) adg−1 ω k (p · g −1 )(v)
k=1
( ∞
)
∑
−1
= adg−1 ϕk (P(p)) ω k (p · g )(v)
k=1
( )
= adg−1 ω (p · g −1 )(v)

as required. 

3.2 Continuous Versus Smooth

We begin by using the material in Section 3.1 to prove that continuous maps
between open subsets of Euclidean spaces can be arbitrarily well approximated
by smooth maps. Indeed, we prove a bit more.
Theorem 3.2.1 Let U be an open subset of Rn , V an open subset of Rm
and h : U −→ V a continuous map. Suppose U0 is an open subset of U ,
A ⊆ U0 is closed in U and h | U0 is smooth. Then, for any continuous function
ϵ : U −→ (0, ∞), there exists a smooth map f : U −→ V such that
1. ∥f (x) − h(x)∥ < ϵ(x) for every x ∈ U .
2. f (x) = h(x) for all x ∈ A.
146 3. Frame Bundles and Spacetimes

Proof: We ask the reader to take the first step.

Exercise 3.2.1 Show that it is enough to prove that, for every continuous
ϵ : U −→ (0, ∞), there is a smooth map f : U −→ Rm satisfying (1) and (2).
Hint: Assume this has been proved and consider
{ }
1
ϵ′ (x) = min ϵ(x), dist (h(x), Rm − V ) ,
2
where
dist (h(x), Rm − V ) = inf {∥h(x) − y∥ : y ∈ Rm − V }.
Now, h and ϵ are continuous on U and U −A is open. For each fixed p ∈ U −A,
the function ∥h(p) − h(x)∥ is continuous and takes the value 0 at p. Since
ϵ(p) > 0 there is an open neighborhood Up of p in U − A on which ∥h(p) −
h(x)∥ < ϵ(x). Consider the open cover of U consisting of U0 together with all
of the Up , p ∈ U − A. Corollary 3.1.5 gives a family {ϕ0 } ∪ {ϕp : p ∈ U − A} of
smooth, non-negative functions on U such that supp ϕ0 ⊆ U0 , supp ϕp ⊆ Up
for each p ∈ U − A, { supp ∑ϕ0 } ∪ {supp ϕp : p ∈ U − A} is locally finite and,
for each x ∈ U, ϕ0 (x) + p∈U −A ϕp (x) = 1. Since h is smooth on U0 and
supp ϕ0 ⊆ U0 we can define a smooth function f : U −→ Rm by
∑
f (x) = ϕ0 (x) h (x) + ϕp (x) h (p).
p∈U −A

Notice that
 
∑
h(x) = ϕ0 (x) + ϕp (x) h(x)
p∈U −A
∑
= ϕ0 (x) h (x) + ϕp (x) h (x)
p∈U −A

so ∑
f (x) − h(x) = ϕp (x) (h(p) − h(x)).
p∈U −A

Exercise 3.2.2 Show that f satisfies (1) and (2). 

Corollary 3.2.2 Let U be an open subset of Rn , V an open subset of Rm
and h : U −→ V a continuous map. Then h is homotopic to a smooth map
f : U −→ V .
Proof: For each x ∈ U let ϵ(x) = 12 dist (h(x), Rm − V ) (see Exercise 3.2.1).
Then the open ball Uϵ(x) (h(x)) of radius ϵ(x) about h(x) is contained in V for
each x ∈ U . By Theorem 3.2.1, we can select a smooth map f : U −→ V with
∥f (x) − h(x)∥ < ϵ(x) for each x ∈ U . In particular, since f (x) ∈ Uϵ(x) (h(x))
and this ball is convex, (1 − t)h(x) + t f (x) is in Uϵ(x) (h(x)) and therefore in
V for each t in [0, 1]. Thus,
 3.2. Continuous Versus Smooth 147

F (x, t) = (1 − t)h(x) + t f (x), x ∈ U, t ∈ [0, 1],

defines a homotopy from h to f . 

Corollary 3.2.3 Let U be an open subset of Rn , V an open subset of Rm

and f0 , f1 : U −→ V two smooth maps that are homotopic. Then there exists
a smooth map F : U × R −→ V with F (x, 0) = f0 (x) and F (x, 1) = f1 (x) for
each x ∈ U (F is called a smooth homotopy from f0 to f1 ).

Proof: By assumption, there exists a continuous map G : U × [0, 1] −→ V

with G(x, 0) = f0 (x) and G(x, 1) = f1 (x) for all x ∈ U . By Exercise 3.1.5,
there exists a smooth function ϕ : R −→ R with 0 ≤ ϕ(t) ≤ 1 for each t ∈ R,
ϕ(t) = 0 for each t ∈ (−∞, 13 ] and ϕ(t) = 1 for each t ∈ [ 23 , ∞). Now define
H : U × R −→ V by
H(x, t) = G(x, ϕ(t))
for each (x, t) ∈ U × R. Note that H is continuous and satisfies H(x, t) = f0 (x)
for all t ≤ 13 and H(x, t) = f1 (x) for all t ≥ 23 and for all x ∈ U . In particular,
H is smooth on the open set U0 = U × (−∞, 13 ) ∪ U × ( 23 , ∞). Furthermore,
A = U × {0, 1} is a closed subset of the subspace U with A ⊆ U0 . According
to Theorem 3.2.1, there exists a smooth map F : U × R −→ V that agrees
with H on U × {0, 1} and this clearly has the required properties. 
These last few results have interesting applications, a few of which we will see
in Chapter 5.
Remarks: Each of the results we have proved in this section for open sub-
manifolds of Euclidean space can be generalized to arbitrary smooth man-
ifolds, but the proofs often require substantial machinery. For example, in
proving that any continuous map h : X −→ Y between smooth manifolds
is homotopic to a smooth map f : X −→ Y , the family of balls Uϵ(x) (h(x))
covering the image of h which appeared in the proof of Corollary 3.2.2 must
be replaced by a “tubular neighborhood” and the existence of these is nontriv-
ial. A good reference for this and many analogous approximation theorems is
[Hir]. Another important result of this same sort concerns cross-sections of
P
principal bundles. Suppose G ,→ P −→ X is a smooth principal bundle and
suppose there exists a continuous map sc : X −→ P such that P ◦ sc = idX
(i.e., a “continuous cross-section” of the bundle). Then one can show (Theo-
rem, Section 6.7, Part I, of [St]) that there must exist a smooth cross-section
s : X −→ P (“arbitrarily close” to sc and, if sc is smooth on an open set
U ⊆ X and A is closed in U , agreeing with sc on A). The proof requires
only tools that we now have at our disposal, but the argument given in [St]
is so thorough that we will simply refer the reader to this classic exposition
for details. An immediate consequence is that, if a smooth principal bundle is
trivial as a C 0 -bundle, then it is trivial as a C ∞ -bundle. Here is another: Let
P
G ,→ P −→ X be a C 0 -principal bundle in which it just so happens that G
148 3. Frame Bundles and Spacetimes

is a Lie group, X and P are smooth manifolds and P as well as the action of
P
G on P are smooth. Then G ,→ P −→ X is a smooth principal bundle (i.e.,
has smooth local trivializations). One need only apply the result from [St] to
each trivialization.
We conclude with two more items of this same sort. Given a smooth man-
ifold X, a Lie group G, an open cover {Vj }j∈J of X and a family {gji } of
smooth maps from nonempty intersections Vi ∩ Vj into G satisfying the co-
cycle condition, it is a simple exercise to trace through the proof of the (C 0 )
Reconstruction Theorem (Theorem 4.3.4, [N4]) and show that the resulting
bundle admits a natural smooth structure (if such an exercise does not appeal
to you, see Proposition 5.2, Chapter I, of [KN1]). One thereby obtains the ver-
sion of the Reconstruction Theorem recorded on page 34. Similarly, the proof
of the (C 0 ) Classification Theorem for principal G-bundles over S n (Theo-
rem 4.4.3, [N4]) shows that the equivalence class of such a bundle is uniquely
determined by the homotopy type of its characteristic map T = g12 | S n−1
from the equator S n−1 ⊆ S n into G. If G is a Lie group one can choose a
smooth map homotopic to T and use this as a characteristic map to build
a G-bundle over S n (Lemma 4.4.1, [N4]). This latter bundle is smooth (by
the smooth Reconstruction Theorem) and equivalent to the original bundle
so every equivalence class contains a smooth bundle. In this way one arrives
at the smooth version of the Classification Theorem, as stated on page 34.

3.3 Frame Bundles

We begin by constructing, for any smooth manifold X, a principal bundle
whose fiber above any x ∈ X consists of all the ordered bases for the tangent
space Tx (X) at x. The group is GL(n, R) and the action is the natural one that
carries one basis onto another. Additional structures on X (e.g., a Riemannian
metric) distinguish certain bases for the tangent spaces (e.g., orthonormal)
and “reduce” the structure group GL(n, R) to some subgroup (e.g., O(n)).
If X is an n-dimensional smooth manifold, then a frame at x ∈ X is an
ordered basis p = (b 1 , . . . , b n ) for Tx (X). Any such frame p gives rise to a
linear isomorphism
p : Rn −→ Tx (X)
e
defined by
p(ei ) = b i , i = 1, . . . , n ,
e
where e1 = (1, 0, . . . , 0), . . . , en = (0, 0, . . . , 1) is the standard basis for Rn .
Conversely, every isomorphism p : Rn −→ Tx (X) gives rise to a frame p =
e It is often convenient to identify a frame
(b 1 , . . . , b n ) = (p(e1 ), . . . , p(en )).
e e
p with its corresponding isomorphism p. We denote by L(X)x the set of all
∪
frames at x ∈ X and let L(X) = x∈XeL(X)x . For each p ∈ L(X)x ⊆ L(X)
 3.3. Frame Bundles 149

we define PL (p) = x and thereby obtain a surjective map PL : L(X) −→ X.

Next we define a right action σ : L(X) × GL(n, R) −→ L(X) of GL(n, R) on
L(X) as follows: For each (p, g) ∈ L(X) × GL(n, R), with p = (b 1 , . . . , b n ) ∈
L(X)x ⊆ L(X) and g = (g i j ) ∈ GL(n, R), we let σ(p, g) = p · g be the frame
(b̂ 1 , . . . , b̂ n ) at x, where

b̂ j = b i g i j , j = 1, . . . , n. (3.3.1)
Formally, we may write
 
g 11 ··· g 1n
 
 .. .. 
(b̂ 1 · · · b̂ n ) = (b 1 · · · b n )  . .  (3.3.2)
 
g n1 ··· g nn

and from this it is clear that p · (g1 g2 ) = (p · g1 ) · g2 and p · id = p. Moreover,

by definition, PL (p · g) = PL (p) for all p ∈ L(X) and g ∈ GL(n, R). It will
be useful to have a description of this action when a frame is identified with
an isomorphism p : Rn −→ Tx (X). Thus, we identify any g ∈ GL(n, R)
e g : Rn −→ Rn given by
with an isomorphism
e
g (ej ) = ei g ij , j = 1, . . . , n, (3.3.3)
e
so that
 
g 11 ··· g 1n
( )  
 .. .. 
g (ei ) · · · g (en ) = (e1 · · · en )  . . . (3.3.4)
e e  
g n1 ··· g nn

Then the frame p · g corresponds to the isomorphism

p · g = p ◦ g : Rn −→ Tx (X) (3.3.5)
g e e
since ( )
p ◦ g (ej ) = p(ei g ij ) = p(ei )g ij = b i g ij = b̂ j
e e e e
for j = 1, . . . , n.
Our objective now is to provide L(X) with a topology and manifold struc-
ture in such a way that, with the action σ described above,
P
GL(n, R) ,→ L(X) −→
L
X
is a smooth principal GL(n, R)-bundle over X, called the (linear) frame
bundle of X. Toward this end we let (U, φ) be any chart on X with coordinate
functions x1 , . . . , xn . Define φ̃: PL−1 (U ) −→ φ(U ) × GL(n, R) as follows: Let
150 3. Frame Bundles and Spacetimes

p = (b 1 , . . . , b n ) be a frame at x = PL (p) ∈ U . For each j = 1, . . . , n, write

∂
bj = Ai j (p).
∂xi x

The matrix ( )
A(p) = Ai j (p)

is nonsingular so we may set

φ̃(p) = (φ(x), A(p)). (3.3.6)

Now, φ̃ is clearly one-to-one and any A ∈ GL(n, R) corresponds to some frame

at each x ∈ U so φ̃ is a bijection of PL−1 (U ) onto φ(U ) × GL(n, R). Moreover,
φ(U ) × GL(n, R) is an open set in Rn × Rn ∼ = Rn+n . We define a topology
2 2

on L(X) by declaring that a subset U of L(X) is open if and only if, for each
chart (U, φ) on X, φ̃(U ∩ PL−1 (U )) is open in φ̃(PL−1 (U )) = φ(U ) × GL(n, R).

Exercise 3.3.1 Show that the collection of all such subsets U of L(X) does,
indeed, define a topology for L(X) and that, if (V, ψ) is any chart for X, then
PL−1 (V ) is open in L(X).
Next we consider two charts (U, φ) and (V, ψ) for X with coordinate func-
tions x1 , . . . , xn and y 1 , . . . , y n , respectively, and with U ∩ V ̸= ∅. Then φ̃ and
ψ̃ are both defined on PL−1 (U ∩ V ), which is open in L(X). We compute

ψ̃ ◦ φ̃−1 : φ(U ∩ V ) × GL(n, R) −→ ψ(U ∩ V ) × GL(n, R).

Let ((x1 , . . . , xn ), (Ai j )) be in φ(U ∩ V ) × GL(n, R). Then φ̃−1 ((x1 , . . . , xn ),

(Ai j )) is the frame p = (b 1 , . . . , b n ) at x = φ−1 (x1 , . . . , xn ) satisfying

∂
bj = Ai j , j = 1, . . . , n.
∂xi x

Moreover, ψ̃(p) = (ψ(x), (B ij )) = ((ψ ◦ φ−1 )(x1 , . . . , xn ), (B ij )), where

∂
bj = B ij , j = 1, . . . , n.
∂y i x

But
( ) ( )
∂ ∂ ∂y i ∂ ∂y i
bj = Akj = (x) Akj = (x)Akj
∂xk x ∂y i x ∂xk ∂y i
x ∂xk
so
∂y i
B ij = (x) Akj , i, j = 1, . . . , n.
∂xk
 3.3. Frame Bundles 151

Thus,
( )
ψ̃ ◦ φ̃−1 ((x1 , . . . , xn ), (Ai j ))
(3.3.7)
( ( i ))
∂y
= (ψ ◦ φ−1 )(x1 , . . . , xn ), (x) A k
j ,
∂xk

which, in particular, is C ∞ . Reversing the roles of ψ̃ and φ̃ we find that ψ̃◦ φ̃−1
and φ̃ ◦ ψ̃ −1 are inverse diffeomorphisms and we will use this fact to provide
L(X) with a differentiable structure.
First we show that, for any chart (U, φ) on X, the map φ̃ : PL−1 (U ) −→
φ(U ) × GL(n, R) is a homeomorphism (so that, in particular, L(X) is locally
Euclidean). We need only show that φ̃ is continuous and an open map. For the
latter, we let W be an open set in PL−1 (U ). Since PL−1 (U ) is open in L(X), W
is open in L(X) so φ̃(W ∩ PL−1 (U )) = φ̃(W ) is open in φ(U ) × GL(n, R),
as required. To prove continuity we let Z be open in φ(U ) × GL(n, R). To
show that φ̃−1 (Z) is open in PL−1 (U ) we let (V, ψ) be an arbitrary chart for
X with U ∩ V ̸= ∅. We must show that ψ̃(φ̃−1 (Z) ∩ PL−1 (V )) is open in
ψ(V ) × GL(n, R). But
( )
ψ̃(φ̃−1 (Z) ∩ PL−1 (V )) = ψ̃ φ̃−1 (Z) ∩ PL−1 (U ) ∩ PL−1 (V )
( )
= ψ̃ φ̃−1 (Z) ∩ PL−1 (U ∩ V )
( )
= ψ̃ φ̃−1 (Z) ∩ φ̃−1 (φ(U ∩ V ) × GL(n, R))
( )
= ψ̃ φ̃−1 (Z ∩ (φ(U ∩ V ) × GL(n, R)))
( )
= (ψ̃ ◦ φ̃−1 ) Z ∩ (φ(U ∩ V ) × GL(n, R))

which is open because ψ̃ ◦ φ̃−1 is a diffeomorphism.

Exercise 3.3.2 Show that L(X) is Hausdorff and second countable.
We have thus far shown that L(X) is a topological manifold of dimen-
sion n + n2 and that {(PL−1 (U ), φ̃) : (U, φ) is a chart for X} is an atlas for
L(X). We provide L(X) with the differentiable structure determined by this
atlas. Next we show that, with this differentiable structure, the projection
PL : L(X) −→ X is smooth. For this it will suffice to show that, for
any chart (U, φ) on X, the coordinate expression φ ◦ PL ◦ φ̃−1 is C ∞ . But
φ ◦ PL ◦ φ̃−1 ((x1 , . . . , xn ), (Ai j )) = (x1 , . . . , xn ) so this is clear.
Exercise 3.3.3 Show that the right action σ : L(X) × GL(n, R ) −→ L(X)
defined above is smooth.
P
To complete the proof that GL(n, R ) ,→ L(X) −−→
L
X is a smooth principal
GL(n, R)-bundle we need only exhibit local trivializations. For each chart
152 3. Frame Bundles and Spacetimes

(U, φ) on X we define

Φ : PL−1 (U ) −→ U × GL(n, R)

as follows: For p ∈ L(X)x ⊆ PL−1 (U ),

( )
Φ(p) = φ−1 × idGL(n,R) ◦ φ̃(p)
( )
= φ−1 × idGL(n,R) (φ(x), A(p))
(3.3.8)
= (x, A(p))
= (PL (p), A(p)).

Then Φ is a diffeomorphism and we need only check that A(p · g) = A(p)g

for all p ∈ PL−1 (U ) and g ∈ GL(n, R). Let p = (b 1 , . . . , b n ) ∈ L(X)x , where
x ∈ U . Then A(p) = (Ai j (p)), where

∂
bj = Ai j (p), j = 1, . . . , n.
∂xi x

Now, p · g = (b̂ 1 , . . . , b̂ n ), where b̂ j = b i g i j , j = 1, . . . , n, so A(p · g) =

(Ai j (p · g)), where
∂
b̂ j = Ai j (p · g).
∂xi x

But then
∂
b̂ j = b k g k j = Ai k (p)g k j
∂xi x

implies
Ai j (p · g) = Ai k (p)g k j , i, j = 1, . . . , n.

Since the right-hand side of this last equality is the (i, j)-entry in the matrix
product A(p)g we have A(p · g) = A(p)g, as required.
Notice that, if Φ is the trivialization arising from (U, φ), with coordinate
functions x1 , . . . , xn and Ψ is the trivialization arising from (V, ψ) with coor-
dinate functions y 1 , . . . , y n and if U ∩ V ̸= ∅, then it follows from (3.3.7) that
the transition function gV U : U ∩ V −→ GL(n, R) is just the Jacobian of the
coordinate transformation, i.e.,
( i )
∂y
gV U (x) = (x)
∂xk

(Lemma 4.2.1, [N4]).

 3.3. Frame Bundles 153

A local cross-section s : U −→ L(X) of the frame bundle assigns to each

x ∈ U a frame s(x) = (b 1 (x), . . . , b n (x)) at x and is called a (linear) frame
field, or moving frame, on U .
Exercise 3.3.4 Show that each b i (x), i = 1, . . . , n, defines a smooth
vector field on U and conclude that X is parallelizable (Exercise 5.8.17,
[N4]) if and only if its frame bundle is trivial (admits a global cross-
section).
Any Lie group is parallelizable (Exercise 5.8.17, [N4]) so, in particular, S 1
and S 3 have trivial frame bundles. S 7 is not a Lie group, but is close enough
(via Cayley multiplication) and turns out to be parallelizable. It is a deep
result of Bott and Milnor that S 1 , S 3 and S 7 are the only parallelizable
spheres.
Next we consider a number of vector bundles associated with the frame
bundle by various representations of GL(n, R). First consider the natu-
ral representation ρ : GL(n, R) −→ GL(Rn ) of GL(n, R) on Rn given
by ρ(g)(v) = g · v = gv, where v ∈ Rn is written as a column ma-
trix and gv denotes matrix multiplication. The associated vector bundle
L(X) ×ρ Rn is called the tangent bundle of X and is denoted T (X).
The terminology would seem to suggest a bundle whose fibers are the tan-
gent spaces to X and we would like to spend a moment showing that
T (X) is just that. First select (arbitrarily) some frame at each x ∈ X
and thereby identify each tangent space Tx (X) with Rn . Any other frame
at x then corresponds to a basis for Rn and any two such bases are re-
lated by a unique element of GL(n, R). Now, the vector bundle associ-
P
ated to GL(n, R ) ,→ L(X) −−→ L
X by ρ is a quotient of L(X) × Rn
(see the construction in Section 6.7, [N4]). An element of L(X) × Rn
is a pair
  (p, v), where p = (b 1 , . . . , b n ) is a frame at P(p) = x and
v1
.
v =  ..  is now identified with the tangent vector at x whose com-
vn
ponents relative to {b 1 , . . . , b n } are v 1 , . . . , v n (v = v i b i ). The action of
GL(n, R) on L(X) × Rn whose orbit space is the vector bundle T (X) is
given by
(p, v) · g = (p · g, g −1 · v).

Now, p · g is the new frame at x given by

 
g1 1 ··· g1 n
 . .. 
p · g = (b̂ 1 · · · b̂ n ) = (b 1 · · · b n ) 
 .
. . 


gn 1 ··· gn n
154 3. Frame Bundles and Spacetimes

and g −1 · v is given by
   −1  
v̂ 1 g1 1 ··· g1 n v1
.  . ..   .. 
 
. =  . 
.  . .  .
v̂ n gn 1 ··· gn n vn
  
g1 1 ··· gn 1 v1
 . ..   .. 
 
=  . 
 . .  . .
g1 n ··· gn n vn

Observe that v̂ 1 , . . . ,v̂ n are precisely the components of v i b i relative to

b̂ 1 , . . . , b̂ n since
v̂ j b̂ j = (gα j v α )(b β g β j ) = (g β j gα j )v α b β = δ β α v α b β = v α b α .
Thus, an equivalence class

[p, v] = {(p · g, g −1 · v) : g ∈ GL(n, R)}

is just the set of all possible descriptions of some fixed tangent vector at x.
Note that, despite the ultramodern attire, this is just a dressed-up version of
the “old-fashioned” view of a vector as a collection of n-tuples, one for each
basis, related by the transformation law ρ.
A vector field on an open subset U of X can now be identified with a
local cross-section V : U −→ T (X) of the tangent bundle. Equivalently (Sec-
tion 6.8, [N4]), one can identify V with an Rn -valued map on PL−1 (U ) that
is equivariant, i.e., satisfies V (p · g) = g −1 V (p) for each p ∈ PL−1 (U ) and
g ∈ GL(n, R).
Similarly, if one defines a representation ρ : GL(n, R ) −→ GL(Rn ) by
ρ(g)(θ) = g · θ = (g T )−1 θ, where θ ∈ Rn is written as a column matrix and
(g T )−1 θ denotes matrix multiplication, then the associated vector bundle is
called the cotangent bundle and denoted T ∗ (X). A 1-form can then be iden-
tified with either a cross-section of T ∗ (X) or an Rn -valued map on L(X) that
is equivariant with respect to this representation (θ(p · g) = g T θ(p)). Tensor
bundles (and their cross-sections, or equivariant maps, called tensor fields)
arise in exactly the same way by making other choices for the representation ρ.
If an n-dimensional manifold X has a Riemannian metric g defined on
it, then a frame p = (b 1 , . . . , b n ) at x ∈ X is orthonormal if g (b i , b j ) =
δij , i, j = 1, . . . , n. Such frames are related by elements of the orthogonal group
O(n) ⊆ GL(n, R) and we wish to build an “orthonormal frame bundle” with
P
group O(n) analogous to the linear frame bundle GL(n, R) ,→ L(X) −−→ L
X.
However, we will need the construction in the indefinite case as well so we
begin by generalizing what we know about the positive definite case. Most of
 3.3. Frame Bundles 155

the proofs are virtually identical and so will be left to the reader in a sequence
of exercises.
On Rn we will denote by ⟨ , ⟩k , 0 ≤ k ≤ n, the standard inner product of
index n − k. Thus, if {e1 , . . . , en } is the standard basis for Rn , x = xi ei and
y = y j ej , then
⟨x, y⟩k = x1 y 1 + · · · + xk y k − xk+1 y k+1 − · · · − xn y n = ηij xi y j ,

where
( )
idk×k 0
η = (ηij ) = .
0 −id(n−k)×(n−k)

Equipped with this (indefinite) inner product, Rn will be denoted Rk,n−k .

When k = n we naturally write Rn,0 = Rn . The case of particular interest to
us will be k = 1, n ≥ 2. R1,n−1 is called n-dimensional Minkowski space.
For reasons that we will discuss later, R1,3 is called Minkowski spacetime.
Exercise 3.3.5 Show that a linear transformation A : Rk,n−k −→ Rk,n−k
satisfies ⟨Ax, Ay⟩k = ⟨x, y⟩k for all x, y ∈ Rk,n−k if and only if its matrix with
respect to {e1 , . . . , en }, also denoted A, satisfies AT η A = η. Hint: Mimic the
proof of (1.1.24) in [N4].
Motivated by Exercise 3.3.5 we define the semi-orthogonal group
O(k, n − k) by
{ }
O(k, n − k) = A ∈ GL(n, R) : AT ηA = η .

Exercise 3.3.6 Show that O(k, n − k) is, indeed, a group under matrix mul-
tiplication and that det A = ±1 for every A ∈ O(k, n − k).
The subgroup
{ }
SO(k, n − k) = A ∈ O(k, n − k) : det A = 1
is called the special semi-orthogonal group. Of course, when k = n we
write O(n, 0) = O(n) and SO(n, 0) = SO(n). We will have more to say
later about O(1, 3), called the general Lorentz group and denoted L, and
SO(1, 3), called the proper Lorentz group and denoted L+ .
Exercise 3.3.7 Show that, for any θ ∈ R, the matrix
 
cosh θ 0 0 − sinh θ
 
 0 1 0 0 
L(θ) = 



 0 0 1 0 
− sinh θ 0 0 cosh θ

is in L+ .
156 3. Frame Bundles and Spacetimes

Exercise 3.3.8 Let Sn denote the set of symmetric elements of GL(n, R)

and define a map f : GL(n, R) −→ Sn by f (A) = AT η A. By mimicing the
arguments for O(n) on page 257 of [N4], show that f is a smooth map, η ∈ Sn
is a regular value of f and

O(k, n − k) = f −1 (η).

From Exercise 3.3.8 (and Corollary 5.6.7, [N4]) we conclude that O(k, n−k)
is a submanifold of GL(n, R) ∼ = Rn of dimension n2 − 12 n(n + 1) =
2

2 n(n − 1). Since matrix multiplication on GL(n, R) is smooth, O(k, n − k) is

1

a Lie group.
Remark: In the positive definite case (k = n), the orthogonal group O(n)
is compact (because the rows of any A ∈ O(n) must form a Euclidean or-
thonormal basis for Rn so O(n) ⊆ Rn is bounded). This is not the case
2

when 0 < k < n, e.g., the elements of O(1, 3) described in Exercise 3.3.7 form
an unbounded set.
Since the determinant function on GL(n, R) is continuous and
SO(k, n − k) = O(k, n − k) ∩ det−1 (0, ∞), the special semi-orthogonal group
is an open submanifold of O(k, n − k) and therefore is also a Lie group.
Since SO(k, n − k) is an open submanifold of O(k, n − k), the Lie algebras
so(k, n − k) and o(k, n − k) are the same. One determines this Lie algebra in
precisely the same way as for O(n) and SO(n).
Exercise 3.3.9 Mimic the arguments on pages 279–280 of [N4] to show that
so(k, n − k) = o(k, n − k) = {A ∈ GL(n, R) : AT = −ηAη}.
Now, a metric g on an n-manifold X is said to be semi-Riemannian of
index n − k if, at each x ∈ X, the inner product g x has index n − k. Then a
frame p = (b 1 , . . . , b n ) at x is orthonormal if



 1, i = j = 1, . . . , k
g x (b i , b j ) = ηij = −1, i = j = k + 1, . . . , n .


 0, i ̸= j

We denote by F (X)x the set of all orthonormal frames at x and let F (X) =
∪
x∈X F (X)x . For each p ∈ F (X)x ⊆ F (X) we let PF (p) = x and thereby
obtain a surjective map
PF : F (X) −→ X.
Next we define a right action σ : F (X)×O(k, n−k) −→ F (X) of O(k, n−k) on
F (X) as follows: For each (p, g) ∈ F (X)×O(k, n−k), with p = (b 1 , . . . , b n ) ∈
F (X)x ⊆ F (X) and g = (g i j ) ∈ O(k, n − k), we let σ(p, g) = p · g ∈ F (X)x
be the frame (b̂ 1 , . . . , b̂ n ) at x, where
b̂ j = b i g ij , j = 1, . . . , n.
 3.3. Frame Bundles 157

Notice that this frame is, indeed, orthonormal by Exercise 3.3.4 and that
σ is a right action for precisely the same reason that the analogous map on
L(X)×GL(n, R) is a right action (see (3.3.2)). By definition, PF (p·g) = PF (p)
for all (p, g) ∈ F (X)×O(k, n−k). Just as for L(X) it will sometimes be conve-
nient to identify an orthonormal frame p with the corresponding isomorphism
p : Rk,n−k −→ Tx (X).
e
We wish to show that
P
O(k, n − k) ,→ F (X) −−→
F
X

is a smooth principal O(k, n − k)-bundle over X, called the orthonormal

frame bundle of X (corresponding to the metric g ). The arguments are
essentially identical to those given for the linear frame bundle GL(n, R) ,→
P
L(X) −−→ L
X except that the local coordinate frame fields { ∂x ∂
1 , . . . , ôxn } must
∂

be replaced by local orthonormal frame fields {E 1 , . . . , E n }. The existence of

these is proved in the positive definite case in Proposition 5.11.3 of [N4]. We
will leave it to the reader to supply the necessary refinements of the argument
in the indefinite case.
Exercise 3.3.10 Let X be a smooth n-manifold and g a semi-Riemannian
metric of index n − k, 0 ≤ k ≤ n, on X. Show that, for each x0 ∈ X, there
exists a connected open neighborhood U of x0 in X and smooth vector fields
E 1 , . . . , E n on U such that

g x0 (E i (x), E j (x)) = ηij , i, j = 1, . . . , n,

for all x ∈ U . Show, furthermore, that, if X has an orientation µ, then one

can ensure {E 1 (x), . . . , E n (x)} ∈ µx for each x ∈ U. {E 1 , . . . , E n } is referred
to as an (oriented) orthonormal frame field on U .
Now, given a chart (U, φ) for X we may shrink U if necessary and
assume that it has defined on it an orthonormal frame field {E 1 , . . . , E n }.
Define φ̃ : PF−1 (U ) −→ φ(U ) × O(k, n − k) as follows: Let p = (b 1 , . . . , b n ) be
an orthonormal frame at x = PF (p) ∈ U . For each j = 1, . . . , n we write

b j = E i (x)Ai j (p).

Then the matrix A(p) = (Ai j (p)) is in O(k, n − k) so we may set

φ̃(p) = (φ(x), A(p)).

φ̃ is clearly one-to-one and any A ∈ O(k, n − k) corresponds to some

orthonormal frame at each x ∈ U so φ̃ is a bijection of PF−1 (U ) onto the
open set φ(U ) × O(k, n − k) in Rn × O(k, n − k). Now, define a topology on
F (X) by declaring that a subset U of F (X) is open if and only if, for each
(U, φ), φ̃(U ∩ PF−1 (U )) is open in φ̃(PF−1 (U )) = φ(U ) × O(k, n − k).
158 3. Frame Bundles and Spacetimes

Exercise 3.3.11 Show that the collection of all such subsets U of F (X) does,
indeed, define a topology on F (X) and that, if (V, ψ) is any chart for X, then
PF−1 (V ) is open in F (X).
Next let (U, φ) and (V, ψ) be two charts on X with U ∩ V ̸= ∅ and with
orthonormal frame fields {E 1 , . . . , E n } and {F 1 , . . . , F n }, respectively. For
each x ∈ U ∩ V we write

E i (x) = Λi j (x)F j (x), i = 1, . . . , n,

where Λi j (x) = g x (E i (x), F j (x)).

Exercise 3.3.12 Show that

ψ̃ ◦ φ̃−1 : φ(U ∩ V ) × O(k, n − k) −→ ψ(U ∩ V ) × O(k, n − k)

is given by
( )( )
ψ̃ ◦ φ̃−1
(x1 , . . . , xn ), (Ai j )
( ( ))
= (ψ ◦ φ−1 )(x1 , . . . , xn ), Λki (x)Akj ,

where x = φ−1 (x1 , . . . , xn ).

In particular, it follows from Exercise 3.3.12 that ψ̃ ◦ φ̃−1 is C ∞ so, reversing
the roles of ψ̃ and φ̃, we find that ψ̃ ◦ φ̃−1 and φ̃ ◦ ψ̃ −1 are inverse diffeomor-
phisms.
Exercise 3.3.13 Use this fact to show that φ̃ : PF−1 (U ) −→ φ(U )×O(k, n−k)
is a homeomorphism.
Since O(k, n − k) is locally Euclidean, it follows from Exercise 3.3.13 that the
same is true of F (X).
Exercise 3.3.14 Show that F (X) is Hausdorff and second countable.
Exercise 3.3.15 Provide F (X) with the differentiable structure determined
by the atlas {(PF−1 (U ), φ̃) : (U, φ) is a chart on X} and show that PF :
F (X) −→ X and σ : F (X) × O(k, n − k) −→ F (X) are smooth.
Exercise 3.3.16 For each chart (U, φ) show that the map Φ : PF−1 (U ) −→
U × O(k, n − k) defined by
( )
Φ = φ−1 ◦ idO(k,n−k) ◦ φ̃

is a local trivialization and that, if Ψ is the analogous trivialization arising

from (V, ψ) with U ∩ V ̸= ∅, then the transition function gV U : U ∩ V −→
O(k, n − k) is given by gV U (x) = (Λi j (x)).
 3.3. Frame Bundles 159

This completes the construction of the orthonormal frame bundle

P
O(k, n − k) ,→ F (X) −→
F
X

for any semi-Riemannian manifold X and we now have every confidence that
the reader can take the next step without assistance.
Exercise 3.3.17 Let X be an oriented semi-Riemannian manifold of index
n − k. Construct, in detail, the oriented, orthonormal frame bundle
PF+
SO(k, n − k) ,→ F+ (X) −→ X.
Being a Lie group S 3 has a trivial linear frame bundle. It is instructive, and
will be useful somewhat later, to explicitly construct a cross-section of the
oriented, orthonormal frame bundle, thereby showing that it is trivial as well.
Exercise 3.3.18 Show that it will suffice to define smooth vector fields
V 1 , V 2 and V 3 on S 3 with the property that, for each p ∈ S 3 , {V 1 (p),
V 2 (p), V 3 (p)} is an oriented, orthonormal basis for Tp (S 3 ).
The procedure for constructing the vector fields described in Exercise 3.3.18
is quite simple. Regard S 3 ⊆ R4 = H as the unit quaternions. For each p ∈ S 3 ,
the tangent space Tp (S 3 ) can be identified with a subspace of Tp (R4 ) and this,
in turn, is canonically identified with R4 itself. Viewed in this way, Tp (S 3 ) is
just the subspace of Tp (R4 ) = R4 consisting of those v = (v 0 , v 1 , v 2 , v 3 ) with
⟨v, p⟩ = 0, i.e., withv 0 p0 + v 1 p1 + v 2 p2 + v 3 p3 = 0. Now, select a point, say
p0 = (1, 0, 0, 0), in S 3 and let

e1 = (0, 1, 0, 0), e2 = (0, 0, 1, 0), e3 = (0, 0, 0, 1)

be the standard basis for Tp0 (S 3 ). We “rotate” this basis to each p ∈ S 3 by

quaternion multiplication, i.e., we define
V i (p) = ei · p, i = 1, 2, 3.
Evaluating these quaternion products gives the components of V i (p) in
Tp (S 3 ) ⊆ Tp (R4 ) = R4 in terms of the coordinates of p.

V 1 (p) = (−p1 , p0 , −p3 , p2 )

V 2 (p) = (−p2 , p3 , p0 , −p1 ) (3.3.9)
V 3 (p) = (−p , −p , p , p )
3 2 1 0

Notice that ⟨V i (p), p⟩ = 0, i = 1, 2, 3, so these are indeed vectors in Tp (S 3 ).

Furthermore,
⟨V i (p), V j (p)⟩ = δij , i, j = 1, 2, 3
for each p so {V 1 (p), V 2 (p), V 3 (p)} is an orthonormal basis for Tp (S 3 ) (the
metric on S 3 is the restriction to S 3 of the standard metric on R4 ). Now,
160 3. Frame Bundles and Spacetimes

an ordered basis for Tp (S 3 ) is in the standard orientation for S 3 if and only

if one obtains an oriented basis for R4 by adjoining to it (at the beginning)
the “position vector” p (this is a special case of Exercise 4.3.7 which may be
consulted, and solved, at this point). Thus, all that remains is to show that
the determinant of the matrix
 
p0 p1 p2 p3
 1 
−p p0 −p3 p2 
  (3.3.10)
−p2 p3 p0 −p1 
 
−p3 −p2 p1 p0

is positive. One can compute this directly or argue indirectly as follows: Since
the rows of (3.3.10) are orthogonal unit vectors in R4 the matrix itself is
an orthogonal matrix and so has determinant ±1 at each p ∈ S 3 . But the
determinant function is continuous and S 3 is connected so the determinant of
(3.3.10) is either 1 for all p ∈ S 3 or −1 for all p ∈ S 3 . Since this determinant
is obviously 1 when p = (1, 0, 0, 0) ∈ S 3 the result follows.
We will have occasion somewhat later to introduce yet one more frame
bundle (the “oriented, time oriented, orthonormal frame bundle” of a “space-
time” manifold). For the present we will conclude this discussion with a
few remarks on some topics we will not pursue in any depth here. Accord-
ing to Theorem 3.1.7, any principal bundle has connections defined on it.
and this is true, in particular, for the frame bundles we have constructed.
A connection on the linear frame bundle GL(n, R) ,→ L(X) −→ X is called
a linear connection on X and these are of fundamental importance in the
study of the geometry of X (see Chapter III of [KN1]). If X admits a Rie-
mannian or semi-Riemannian metric, then there is, among these linear con-
nections, a distinguished one called the Levi-Civita connection (or Rie-
mannian connection) that is adapted to the metric structure. The study
of this connection is the vast and beautiful subject of (semi-) Riemannian
geometry. Although this is not our subject here we intend to borrow one of
its results. Moreover, the Levi-Civita connection plays a role in defining the
Seiberg-Witten equations which we will sketch in the Appendix so a brief tour
of the definition is probably in order (for details one can consult [Bl], [O’N],
or [KN1]). We consider a smooth n-manifold X and its linear frame bundle
GL(n, R) ,→ L(X) −→ X. A connection ω on L(X) is a gl(n, R)-valued 1-
form on L(X) which we describe as follows. Let Eij be the n × n matrix for
which the entry in the ith row and j th column is 1 and all other entries are 0.
Then ω can be written as ω = ω ij Eij , where each ω ij is a real-valued 1-form on
L(X). Similarly, the curvature Ω = dω + 12 [ω, ω] can be written Ω = Ω ij Eij ,
where each Ω ij is a real-valued 2-form on L(X). Thus,

Ω ij = dω ij + ω ik ∧ ω kj .
 3.3. Frame Bundles 161

If U is a coordinate neighborhood in X with coordinates x1 , . . . , xn and SU :

U −→ L(X) is the corresponding cross-section

SU (x) = (∂1 (x), . . . , ∂n (x)),

then the pullback ω U = (SU )∗ ω can be written

ω U = (SU )∗ ω = ((SU )∗ ω ik )Eik = (Γjk

i
dxj )Eik

i
for some smooth functions Γjk on U (called the Christoffel symbols for ω
in the coordinate neighborhood U ). Similarly,
( )
∗ ∗ 1 i
Ω U = (SU ) Ω = ((SU ) Ω ij )Eij = (R dxk ∧ dxl )Eij
2 jkl
i
for some smooth functions Rjkl on U . Unraveling the definitions gives
i
Rjkl = ∂k Γlji − ∂l Γkj
i
+ Γljm Γkm
i
− Γkj
m i
Γlm

for i, j, k, l = 1, . . . , n.
Now suppose that X has a metric g of index n − k and consider the
orthonormal frame bundle O(k, n−k) ,→ F (X) −→ X. If X is also oriented we
have an oriented, orthonormal frame bundle SO(k, n − k) ,→ F+ (X) −→ X as
well. It is not difficult to see that any connection on F (X) or F+ (X) extends
uniquely to a connection on L(X) (page 158 of [KN1]), but it is not true
that every connection on L(X) restricts to a connection on F (X) or F+ (X)
because a connection on L(X) takes values in gl(n, R) and not necessarily in
o(k, n − k) = so(k, n − k); those that do are called metric connections on
L(X). The so-called Fundamental Theorem of Riemannian Geometry
states that there is a unique metric connection on L(X) that is also symmetric
i i
in the sense that Γjk = Γkj for all i, j, k = 1, . . . , n in any local coordinate
system. This is called the Levi-Civita connection and denoted ω LC (the
same terminology and notation is used for its restriction to either F (X) or
F+ (X)). One can show that ω LC is characterized by the fact that, in any local
coordinate system, the Christoffel symbols are given in terms of the metric
components by
1
i
Γjk = g il (∂k gjl + ∂j gkl − ∂l gjk )
2
for i, j, k, l = 1, . . . , n. All of the usual objects of study in (semi-)
Riemannian geometry are defined in terms of the Levi-Civita connection. For
i
example, Rjkl are the local components of the Riemann curvature tensor,
Rij = Rikj are the components of the Ricci tensor, and R = g ik Rik is the
k

scalar curvature.
162 3. Frame Bundles and Spacetimes

When we define the Čech cohomology groups of a manifold in Chapter 6

we will require the notion (and the existence) of “simple covers” for a smooth
manifold. The sort of cover we have in mind is illustrated for the circle S 1
below.

U1
U2

U3

Here we have an open cover U = {U1 , U2 , U3 } that is locally finite (actually,

finite in this case), each element of which has compact closure and which
has the property that any nonempty finite intersection of its elements is dif-
feomorphic to the Euclidean space of the same dimension as the manifold
(U1 , U2 , U3 , U1 ∩ U2 , U1 ∩ U3 , and U2 ∩ U3 are all diffeomorphic to R). In
general, a simple cover of an n-manifold X is a locally finite open cover
U = {Uα }α∈A such that any nonempty finite intersection of its elements is
diffeomorphic to Rn . Simple covers are generally not that easy to find, but
they always exist. Indeed, the following is essentially Theorem 3.7, Chapter IV,
of [KN1].
Theorem 3.3.1 Let X be an n-dimensional smooth manifold. Then any open
cover for X has a refinement that is a countable simple cover of X.
Exercise 3.3.19 Find a simple cover for the torus S 1 × S 1 .
If X has a finite simple cover, then it is said to be of finite type. Certainly,
any compact manifold is of finite type, but so is Rn .

3.4 Minkowski Spacetime

As we mentioned earlier the indefinite inner product space R1,3 is called
Minkowski spacetime and the associated semi-orthogonal groups O(1, 3) and
SO(1, 3), called the Lorentz group and the proper Lorentz group, respectively,
are denoted L and L+ . In this section we will examine these objects in some-
what more detail and have a bit to say about their physical interpretation
(also see Chapter 2).
 3.4. Minkowski Spacetime 163

Remark: A much more thorough study of both the mathematics and the
physics is available in [N3].
We will bow to the generally accepted conventions of physics and use
(x0 , x1 , x2 , x3 ), rather than (x1 , x2 , x3 , x4 ), for the standard coordinates in
R1,3 and will drop the subscript 1 on the inner product ⟨ , ⟩1 . Thus, if
{e0 , e1 , e2 , e3 } is the standard basis with x = xα eα and y = y β eβ , then
⟨x, y⟩ = x0 y 0 − x1 y 1 − x2 y 2 − x3 y 3 = ηαβ xα y β
where 


 1, α=β=0
ηαβ = −1, α = β = 1, 2, 3 .


 0, α ̸= β

In particular, {e0 , e1 , e2 , e3 } is an orthonormal basis for R1,3 . We will denote

by Q : R1; 3 −→ R the quadratic form Q(x) = ⟨x, x⟩ associated with ⟨ , ⟩.
The elements of R1,3 are called events and should be identified intuitively
with idealized physical events having no spatial extension and no duration,
e.g., an instantaneous collision, or explosion, or an instant in the history of
some point particle. The coordinates (x0 , x1 , x2 , x3 ) of such an event are to be
regarded as the spatial (x1 , x2 , x3 ) and time (x0 ) coordinates assigned to that
event by some inertial observer with x0 measured in units of distance, i.e.,
light travel time (1 meter of time, or 1 light meter, being the time required
by a photon to travel 1 m in vacuo). The observer can then be identified with
{e0 , e1 , e2 , e3 }. Another inertial observer is identified with another orthonor-
mal basis {ê0 , ê1 , ê2 , ê3 } (we assume the basis elements are always numbered
in such a way that ⟨êα , êβ } = ηαβ ). For any such basis there will exist a linear
transformation L : R1,3 −→ R1,3 with Lêα = eα for α = 0, 1, 2, 3 so that L is
an orthogonal transformation of R1,3 , i.e.,

⟨Lx, Ly⟩ = ⟨x, y⟩

for all x, y ∈ R1,3 . Writing

eβ = Λα β êα , β = 0, 1, 2, 3,

we obtain a matrix Λ = (Λα β )α,β=0,1,2,3 (the matrix of L−1 relative to {êα })

which must be in L. If x ∈ R1,3 and x = xα eα = x̂α êα , then

x̂α = Λα β xβ , α = 0, 1, 2, 3, (3.4.1)

is the coordinate transformation between the two observers. The condition

⟨eα , eβ ⟩ = ηαβ implies that Λ must satisfy

Λa α Λb β ηab = ηαβ , α, β = 0, 1, 2, 3. (3.4.2)

164 3. Frame Bundles and Spacetimes

In particular, when α = β = 0, this gives

(Λ0 0 )2 − (Λ1 0 )2 − (Λ2 0 )2 − (Λ3 0 )2 = 1

so
Λ0 0 ≥ 1 or Λ0 0 ≤ −1. (3.4.3)
Those elements of L for which Λ 0 ≥ 1 are called orthochronous and the
0

set
L+↑ = {Λ ∈ L+ : Λ0 0 ≥ 1}
is called the proper orthochronous Lorentz group. For physical reasons
that we will discuss shortly we will consider only those inertial observers
(orthonormal bases) related to the standard basis by an element of L+↑ . We
call such bases admissible.
If gravitational effects are assumed negligible, then the Minkowski
inner product ⟨ , ⟩ on R1,3 has something physically significant to say about
relationships between events. Consider, for example, two events x0 , x1 ∈ R1,3
and the displacement vector x = x1 − x0 between them. Suppose that
Q(x) = 0. Then, if we write x = (∆xα )eα ,

(∆x1 )2 + (△x2 )2 + (∆x3 )2 = (∆x0 )2 (3.4.4)

and the same equation is satisfied in any other orthonormal basis. Conse-
quently, the spatial separation of the two events x0 and x1 is numerically
equal to the distance light would travel during the time lapse between them.
The two events are “connectible by a light ray.” In this case, x is said to be
null, or lightlike. With x0 held fixed the set of all x1 for which x = x1 − x0
is null is called the null cone, or light cone, at x0 because of the formal re-
semblance of (3.4.4) to the equation of a right circular cone in R3 . A straight
line which lies entirely on such a null cone is called a worldline of a photon
and is thought of as the set of all events in the history of some “particle of
light.”
Suppose instead that Q(x) = Q(x1 − x0 ) > 0. In this case we say that x is
timelike and, in any orthonormal basis, we have

(∆x1 )2 + (∆x2 )2 + (∆x3 )2 < (∆x0 )2 (3.4.5)

(x is “inside” the null cone at x0 ). Then the spatial separation of x0 and x1

is less than the distance light would travel during the elapsed time between
them so it is possible for some material particle (or observer) to experience
both events without violating the speed limit imposed on such particles by
special relativity. A straight line joining two such events is called a timelike
straight line, or a worldline of a free material particle and can be
identified intuitively with the set of all events in the history of an unaccelerated
material particle that experiences both events.
 3.4. Minkowski Spacetime 165

If Q(x) = Q(x1 − x0 ) < 0, then x = x1 − x0 is said to be spacelike and

lies outside the null cone at x0 . No known physical agency can experience
both x0 and x1 since, to do so, such an agency would have to be transmitted
at a speed greater than that of light. A smooth curve in R1,3 is said to be
spacelike, timelike, or null if its tangent vector at each point is spacelike,
timelike, or null, respectively.
Theorem 3.4.1 Suppose x is timelike and y is either timelike, or null and
nonzero. Let {e0 , e1 , e2 , e3 } be any orthonormal basis for R1,3 with x = xα eα
and y = y β eβ . Then either
1. x0 y 0 > 0 in which case ⟨x, y⟩ > 0, or
2. x0 y 0 < 0 in which case ⟨x, y⟩ < 0.
Proof: By assumption, ⟨x, x⟩ = (x0 )2 − ⃗x · ⃗x > 0 and ⟨y, y⟩ =
(y 0 )2 − ⃗y · ⃗y ≥ 0, where we have written ⃗x · ⃗x for (x1 )2 + (x2 )2 + (x3 )2 and
similarly for y. Thus, (x0 y 0 )2 > (⃗x · ⃗x)(⃗y · ⃗y ) so
√
|x0 y 0 | > (⃗x · ⃗x) (⃗y · ⃗y ). (3.4.6)
Now, for any real number t, 0 ≤ (ty 1 + x1 )2 + (ty 2 + x2 )2 + (ty 3 + x3 )2 , i.e.,
(⃗y · ⃗y )t2 + 2(x̃ · ⃗y )t + (⃗x · ⃗x) ≥ 0. The discriminant of this quadratic in t must
therefore be less than or equal to zero and this gives (⃗x · ⃗x)(⃗y · ⃗y ) ≥ (⃗x · ⃗y )2 .
Combining this with (3.4.6) yields

|x0 y 0 | > |⃗x · ⃗y | = |x1 y 1 + x2 y 2 + x3 y 3 |. (3.4.7)

In particular, x0 y 0 ̸= 0. Suppose that x0 y 0 > 0. Then x0 y 0 = |x0 y 0 | >

|x1 y 1 + x2 y 2 + x3 y 3 | ≥ x1 y 1 + x2 y 2 + x3 y 3 so ⟨x, y⟩ > 0. On the other hand,
if x0 y 0 < 0, then ⟨x, −y⟩ > 0 so ⟨x, y⟩ < 0. 
Exercise 3.4.1 Show that if a nonzero vector in R1,3 is orthogonal to a
timelike vector, then it must be spacelike.
If Q(x1 − x0 ) = 0, then x0 and x1 lie on the worldline of a photon so one
of them can be regarded as the emission of a photon and the other as its
subsequent reception somewhere else. We assume that admissible observers
are related by orthochronous elements of L so that all of them will agree on
which is which.
Exercise 3.4.2 Let Λ = (Λα β )α,β=0,1,2,3 be an element of L. Show that the
following are equivalent.
(a) Λ is orthochronous (i.e., Λ0 0 ≥ 1).
(b) Λ preserves the time orientation of all nonzero null vectors, i.e., if x =
xα eα is nonzero and null, then x0 and x̂0 = Λ0 β xβ have the same sign.
(c) Λ preserves the time orientation of all timelike vectors.
166 3. Frame Bundles and Spacetimes

Notice that Exercise 3.4.2 does not assert that admissible observers agree on
the temporal order of x0 and x1 if x1 − x0 is spacelike and, indeed, they need
not.
Exercise 3.4.3 Show that if Λ ∈ L is orthochronous, but det Λ = −1 (see
Exercise 3.3.6), then
 
1 0 0 0
 
 
0 1 0 0
 Λ
 
0 0 1 0
 
0 0 0 −1

is in L+↑ .
Since the matrix in Exercise 3.4.3 simply reverses the orientation of the spatial
axes, our assumption that admissible observers are related by elements of L+↑
essentially amounts to the requirement that no one’s clock runs backwards
and no one uses “left-handed” spatial coordinates. Henceforth, we will refer to
the elements of L+↑ simply as Lorentz transformations.
Exercise 3.4.4 Show that if (Ri j )i,j=1,2,3 is an element of SO(3), then
 
1 0 0 0
 
 0 
R=
 0


 (Ri j ) 
0

is in L+↑ . Show also that the collection of all such elements of L+↑ is a subgroup
R of L+↑ (called the rotation subgroup of L+↑ ).
If two observers are related by an element of R, then they differ only in having
spatial coordinate axes that are rotated relative to each other.
Exercise 3.4.5 Let Λ = (Λα β )α,β=0,1,2,3 be an element of L+↑ . Show that the
following are equivalent.
(a) Λ ∈ R.
(b) Λ1 0 = Λ2 0 = Λ3 0 = 0.
(c) Λ0 1 = Λ0 2 = Λ0 3 = 0.
(d) Λ0 0 = 1.
From the point of view of physics, the elements of R are rather dull. On the
other hand, the matrices L(θ), θ ∈ R, described in Exercise 3.3.7 are clearly
in L+↑ (cosh θ ≥ 1) and these are not at all dull. Physically, they correspond
 3.4. Minkowski Spacetime 167

to inertial observers whose spatial coordinate axes are parallel and whose
relative motion is along their common x1 -, x̂1 -axis with speed β = tanh θ (see
Section 1.3 of [N3]). These Lorentz transformations L(θ) are called boosts
and, in some sense, contain all of the interesting kinematical information in
L+↑ .
Theorem 3.4.2 Let Λ be an element of L+↑ . Then there exists a real number
θ and two rotations R1 and R2 in R ⊆ L+↑ such that

Λ = R1 L(θ)R2 .
The intuitive content of this result is quite simple. The Lorentz transformation
Λ from the frame of reference of observer O1 to the frame of reference of
observer O2 can be accomplished in three stages: Rotate O1 ’s spatial axes so
that the x1 -axis coincides with the line along which the relative motion takes
place. Boost to a new frame whose spatial axes are parallel to the rotated
axes of O1 and at rest relative to O2 . Finally, rotate these new spatial axes
so that they coincide with those of O2 . For a detailed algebraic proof see
Theorem 1.3.5 of [N3].
Remark: One can use the decomposition in Theorem 3.4.2 to define a defor-
mation retraction of L+↑ onto SO(3) and conclude that these two have the same
homotopy type (Lemma 2.4.9, [N4]). Indeed, one can show more. L+↑ is actu-
ally homeomorphic to SO(3) × R3 (there is a nice proof of this on pages 73–74
of [Bl]). From either of these it follows that π1 (L+↑ ) ∼
= π1 (SO(3)) ∼
= Z2 (Ap-
pendix B of [N3]). We will need this fact only for motivational purposes in Sec-
tion 6.5 and so will not give the details here. However, we make the following
observation. On pages 92–93 we exhibited two smooth loops R1 (t) and R2 (t)
in SO(3) corresponding to a continuous rotation about the x-axis through
2π and 4π, respectively, and showed that they represent the two equivalence
classes in π1 (SO(3)) ∼
= Z2 . The same two curves, now thought of as loops in
R ⊆ L+ as in Exercise 3.4.4 represent the two classes in π1 (L+↑ ) ∼
↑
= Z2 .
We have already described (on pages 101–103) an alternative model of
Minkowski spacetime as the real linear space H of 2 × 2 complex Hermi-
tian matrices with Q given by the determinant (and ⟨ , ⟩ thereby determined
via the Polarization Identity). This view of R1,3 is particularly convenient for
describing the double cover
Spin : SL(2, C) −→ L+↑

of L+↑ . This, the reader may recall, was required to produce a (“2-valued”) rep-
resentation of L+↑ that is the appropriate transformation law (under change
of inertial frame of reference) for the wavefunction of a spin 12 particle
(Section 2.4). We also briefly alluded (on pages 113–115) to the difficulties
involved in generalizing these considerations to the situation in which gravi-
tational fields are not negligible so that R1,3 is no longer an accurate model
168 3. Frame Bundles and Spacetimes

of the “event world” and must be replaced by a spacetime manifold. It is to

these objects that we now turn our attention.

3.5 Spacetime Manifolds and Spinor

Structures
Unlike electromagnetic fields which, as we have seen (Sections 2.2, 2.3 and 2.4),
fit very nicely within the framework of Minkowski spacetime (as 2- forms),
gravitational fields stubbornly refuse to be modeled in R1,3 . The reason is
quite simple (once it has been pointed out to you by Einstein). An elec-
tromagnetic field is something “external” to the structure of space-time, an
additional field defined on and (apparently) not influencing the mathematical
structure of R1,3 . Einstein realized that a gravitational field has a very special
property which makes it unnatural to regard it as something external to the
nature of the event world. Since Galileo it had been known that all objects
with the same initial position and velocity respond to a given gravitational
field in the same way (i.e., have identical worldlines) regardless of their ma-
terial constitution (mass, charge, etc.). This is essentially what was verified
at the Leaning Tower of Pisa and contrasts rather sharply with the behav-
ior of electromagnetic fields. These worldlines (of particles with given initial
conditions of motion) seem almost to be natural “grooves” in spacetime which
anything will slide along if once placed there. But these “grooves” depend on
the particular gravitational field being modeled and, in any case, R1,3 simply
is not “grooved” (its structure does not distinguish any collection of curved
worldlines). One suspects then that R1,3 itself is somehow lacking, that the
appropriate mathematical structure for the event world may be more complex
when gravitational effects are non-negligible.
To see how the structure of R1,3 might be generalized to accommodate the
presence of gravitational fields let us begin with a structureless set X whose
elements we call “events.” One thing at least is clear: In regions that are dis-
tant from the source of any gravitational field, no accomodation is necessary
and X must locally “look like” R1,3 . But a great deal more is true. In his now
famous “Elevator Experiment” Einstein observed that any event has about it
a sufficiently small region of X which “looks like” R1,3 . To see this we rea-
son as follows: Imagine an elevator containing an observer and various other
objects which is under the influence of some uniform external gravitational
field. The cable snaps. The contents of the elevator are now in free fall. Since
all of the objects inside respond to the gravitational field in the same way
they will remain at relative rest throughout the fall. Indeed, if our observer
lifts an apple from the floor and places it in mid-air next to his head it will
remain there. You have witnessed these facts for yourself. While it is unlikely
you have ever had the misfortune of seeing a falling elevator you have seen
astronauts at play inside their space capsules while in orbit (i.e., free fall)
 3.5. Spacetime Manifolds and Spinor Structures 169

about the earth. The objects inside the elevator (capsule) seem then to con-
stitute an archetypical inertial frame (they satisfy Newton’s First Law). By
establishing spatial and temporal coordinate systems in the usual way our
observer thereby becomes an inertial observer, at least within the spatial and
temporal constraints imposed by his circumstances. Now picture an arbitrary
event. There are any number of vantage points from which the event can be
observed. One is from a freely falling elevator in the immediate spatial and
temporal vicinity of the event and from this vantage point the event receives
inertial coordinates. There is then a local inertial frame near any event in X.
The operative word is “local.” The “spatial and temporal constraints” to
which we alluded in the preceding paragraph arise from the nonuniformity of
any gravitational field in the real world. For example, in an elevator which
falls freely in the earth’s gravitational field, all of the objects inside are pulled
toward the earth’s center so that these objects do experience some slight
relative motion (toward each other). Such motion, of course, goes unnoticed
if the elevator falls neither too far nor too long. Indeed, by restricting our
observer to a sufficiently small region in space and time these effects become
negligible and the observer is indeed inertial. But then, what is “negligible” is
in the eye of the beholder. The availability of more sensitive measuring devices
will require further restrictions on the size of the spacetime region which “looks
like” R1,3 . Turn of the century mathematical terminology expressed this fact
by saying that any point in X has about it an “infinitesimal neighborhood”
which is identical to R1,3 . Today we prefer to say that X is a 4-dimensional
smooth manifold, each tangent space of which has the structure of R1,3 .
A spacetime is a 4-dimensional smooth manifold X with a semi-Riemannian
metric g of index 3 (called a Lorentz metric). Thus, for each x ∈ X there
exists a frame p = (b 0 , b 1 , b 2 , b 3 ) at x such that



 1, α=β=0
g x (b α , b β ) = ηαβ = −1, α = β = 1, 2, 3. (3.5.1)


 0, α ̸= β

For simplicity we will occasionally write g x as ⟨ , ⟩x , or even ⟨ , ⟩. A tangent

vector v at x is said to be spacelike, timelike, or null if ⟨v , v ⟩x is < 0, > 0,
or = 0, respectively. The null cone at x is the set of all null tangent vectors at
x. A smooth curve in X is spacelike, timelike, or null provided its tangent
vector is spacelike, timelike, or null, respectively, at each point.

Remark: Although we have thus far thought of Minkowski spacetime R1,3

as a vector space, this is a luxury we have been permitted because its struc-
ture as a spacetime manifold is so simple. The underlying 4-manifold is, of
course, R4 so that, in particular, all of its tangent spaces are canonically
identified with R4 itself. Thus, defining a constant Lorentz metric g on R1,3
by specifying that its components relative to the standard global chart are
170 3. Frame Bundles and Spacetimes

gαβ (x) = g x ( ∂x∂α |x , ∂x∂ β |x ) = ηαβ for each x ∈ R1,3 is tantamount to defining
the Minkowski inner product on R4 . We shall feel free to think of Minkowski
spacetime in whichever of these ways is convenient at the moment.
The most serious obstacle to the general study of spacetime manifolds is
their overwhelming number and diversity. Smooth 4-manifolds are, to say
the least, plentiful and almost all of them admit Lorentz metrics. Indeed, a
Lorentz metric can be defined on any noncompact 4-manifold and a compact
4-manifold admits a Lorentz metric if and only if its Euler characteristic (see
Section 5.7 or Section 3.4 of [N4]) is zero (see [O’N]).
Remark: Compact spacetimes are of no real interest anyway since they al-
ways contain closed timelike curves and these do violence to our most cherished
notions of causality (see [N2]). In effect then, anything that could possibly be
a spacetime, is a spacetime.
There are two ways around this difficulty. One can impose additional restric-
tions on the structure of a spacetime in an attempt to eliminate “unphysical”
behavior. A number of such restrictions are easily formulated (we will intro-
duce one shortly) and the study of the manifolds that satisfy these conditions
has led to some spectacular results (e.g., the Singularity Theorems of Penrose
and Hawking discussed in [HE], [Pen], [O’N] and [N2]). Alternatively, one
can restrict attention to spacetime manifolds that arise in physics by solving
the field equations of general relativity for more or less realistic distributions
of mass/energy. We will describe a few such examples shortly, but our real in-
terest here (and in Chapter 6) is in an essentially global, topological question
about spacetime manifolds. The question (introduced in Section 2.4) is this:
On which spacetime manifolds is it possible to introduce a meaningful notion
of a “spin 12 particle” ?
According to Dirac a spin 12 particle is characterized by the fact that its
wavefunction transforms under a certain (reducible) representation of the dou-
ble cover SL(2, C) of L+↑ . Now, each tangent space Tx (X) to a spacetime
manifold X has an inner product ⟨ , ⟩x of index 3 giving it the structure
of Minkowski spacetime. In particular, its orthonormal bases are related by
elements of L. We are, however, interested only in bases related by elements of
L+↑ . We must therefore choose, consistently over all of X, a family of bases for
each Tx (X) that are related by the Λ ∈ L satisfying det Λ = 1 and Λ00 ≥ 1.
This is possible only if X is orientable and has, moreover, some global notion
of “time orientation” (see Exercise 3.4.2). To define such a notion we proceed
in the manner familiar for smooth surfaces in R3 (which are orientable if and
only if they admit a smooth, nonzero field of normal vectors).
We say that a spacetime X is time orientable if one can define on it a
smooth vector field T which is everywhere timelike (⟨T (x), T (x)⟩x > 0 for
each x ∈ X). X is time oriented if a specific choice of such a vector field has
been made. One then thinks of T as “pointing toward the future” (keeping
in mind that the designation “future” is now entirely arbitrary and would be
 3.5. Spacetime Manifolds and Spinor Structures 171

reversed if we choose −T rather than T to time orient X). Motivated by

Theorem 3.4.1 we then say that a tangent vector v ∈ Tx (X) which is either
timelike, or null and nonzero, is future directed if ⟨T (x), v ⟩x > 0 and past
directed if ⟨T (x), v ⟩x < 0. The same terminology is used for smooth curves
that are either timelike or null (and, in the latter case, have nonzero tangent
vector at each point).
If X is an oriented, time oriented spacetime, then an oriented, time ori-
ented, orthonormal frame at x ∈ X is a frame {b 0 , b 1 , b 2 , b 3 } in Tx (X)
that is consistent with the orientation of X, satisfies ⟨b α , b β ⟩ = ηαβ and for
which the timelike vector b 0 is future directed. Exercise 3.3.10 guarantees the
existence of local oriented, orthonormal frame fields on any oriented semi-
Riemannian manifold so, in particular, each x0 ∈ X has about it a connected
open neighborhood U on which there exist smooth vector fields E 0 , E 1 , E 2
and E 3 with {E 0 (x), E 1 (x), E 2 (x), E 3 (x)} an oriented orthonormal frame for
each x ∈ U . Now, E 0 (x) is timelike for every x ∈ U and, if T is the vector
field that time orients X, ⟨E 0 (x), T (x)⟩ is never zero (Exercise 3.4.1) and so
has the same sign everywhere on U . If this sign is positive, {E 0 , E 1 , E 2 , E 3 }
is a local oriented, time oriented, orthonormal frame field on U . If the sign is
negative, {−E 0 , E 1 , E 2 , E 3 } is a local oriented, time oriented, orthonormal
frame field on U . In any case, such things always exist and the construction
PF+
of SO(k, n − k) ,→ F+ (X) −→ X in Exercise 3.3.17 can be repeated verbatim
to yield the oriented, time oriented, orthonormal frame bundle
P
L+↑ ,→ L(X) −→
L
X

with group L+↑ . As we argued in Section 2.4 (pages 114–115) the object re-
quired to describe wavefunctions of spin 12 particles on X is a lift of this bundle
to a principal SL(2, C)-bundle over X. More precisely, a spinor structure
for X consists of a principal SL(2, C)-bundle
P
SL(2, C) ,→ S(X) −→
S
X

over X and a map λ : S(X) −→ L(X) such that PL (λ(p)) = PS (p) and
λ(p · g) = λ(p) · Spin(g) for all g ∈ SL(2, C). In Section 6.5 we will determine
a necessary and sufficient condition (on the topology of X) for the existence
of such a spinor structure.
We will conclude this section with a number of examples of spacetime man-
ifolds. We have been guided in the selection of these examples by the desire
to illustrate the concepts we have introduced in a context as free of techni-
cal obfuscations as possible without wandering into the realm of physically
meaningless examples contrived solely for pedegogical purposes. Some of the
examples are of great physical significance, while others are primarily of histor-
ical interest, but all of them have played a role in the development of general
relativity.
172 3. Frame Bundles and Spacetimes

As we have already mentioned, the simplest example of a spacetime is R1,3

itself. Here the underlying 4-manifold is R4 . The standard coordinate func-
tions (corresponding to the chart idR4 : R4 −→ R1,3 ) are denoted x0 , x1 , x2
and x3 . We define the Lorentz metric g on R1,3 by specifying its components
gαβ relative to this global chart. For any p ∈ R1,3 , Tp (R1,3 ) is spanned by
{ ∂x
∂
0 |p , ∂x1 |p , ∂x2 |p , ∂x3 |p } and we define
∂ ∂ ∂

( )
∂ ∂
gαβ (p) = g p , = ηαβ , α, β = 0, 1, 2, 3,
∂xα p ∂xβ p

where 


 1, α=β=0
ηαβ = −1, α = β = 1, 2, 3 .


 0, α ̸= β

Thus, the gαβ are constant and, identifying each Tp (R1,3 ) with R4 via the
canonical isomorphism, we have, in effect, just introduced the Minkowski inner
product on R4 . Relative to the basis {dxα ⊗ dxβ : α, β = 0, 1, 2, 3} for the
covariant tensors of rank 2 on R1,3 , the metric for R1,3 is given by
∑
3
g = ηαβ dxα ⊗ dxβ = dx0 ⊗ dx0 − dxi ⊗ dxi .
i=1
To ease the typography we will often write the coordinate velocity vector
fields ∂x∂α as ∂α , α = 0, 1, 2, 3. The standard orientation for R1,3 is the
one that assigns to each p ∈ R1,3 the orientation for Tp (R1,3 ) contain-
ing {∂0 (p), ∂1 (p), ∂2 (p), ∂3 (p)} and we will time orient R1,3 with the vector
field ∂0 . Thus, a tangent vector v = v α ∂α (p) ∈ Tp (R1,3 ) which is either time-
like (g p (v , v ) = (v 0 )2 − (v 1 )2 − (v 2 )2 − (v 3 )2 > 0) or null (g p (v , v ) = (v 0 )2 −
(v 1 )2 − (v 2 )2 − (v 3 )2 = 0) and nonzero is future directed if g p (∂0 (p), v ) = v 0
is positive. Since any principal bundle over R4 is necessarily trivial, this is
true, in particular, for the oriented, time oriented, orthonormal frame bundle
of R1,3 .
Exercise 3.5.1 Introduce spherical coordinates (t, ρ, ϕ, θ) on R1,3 as follows:
Define a map from

(t, ρ, ϕ, θ) ∈ (−∞, ∞) × (0, ∞) × (0, π) × (−π, π)

to R1,3 by

x0 = t
x1 = ρ sin ϕ cos θ
x2 = ρ sin ϕ sin θ
x3 = ρ cos ϕ.
 3.5. Spacetime Manifolds and Spinor Structures 173

Show that the Jacobian of the map is ρ2 sin ϕ. Use this fact to show that the
map is a diffeomorphism onto an open set in R1,3 . What open set? The inverse
of this map is therefore a chart on R1,3 . Denote the coordinate velocity fields
for this chart ∂t , ∂ρ , ∂ϕ and ∂θ and compute the components of the metric in
this chart (i.e., g (∂t , ∂t ), g (∂t , ∂ρ ), etc.) to show that

g = dt ⊗ dt − dρ ⊗ dρ − ρ2 (dϕ ⊗ dϕ + sin2 ϕ dθ ⊗ dθ).

Finally, alter the domain of the mapping in such a way as to obtain charts
that cover as much of R1,3 as possible.
Exercise 3.5.2 Introduce advanced and retarded null coordinates v and w
on R1,3 by letting v = t + ρ and w = t − ρ (so that v ≥ w). Thus,

1
t= (v + w)
2
1
ρ = (v − w)
2
ϕ=ϕ
θ = θ.

Show that the Jacobian of the map is identically equal to 12 so that the trans-
formation (v, w, ϕ, θ) −→ (t, ρ, ϕ, θ) is nonsingular wherever it is defined. Show
that, in these coordinates, the metric is given by

1
g = dv ⊗ dw − (v − w)2 (dϕ ⊗ dϕ + sin2 ϕ dθ ⊗ dθ).
4

In particular, g (∂v , ∂v ) = g (∂w , ∂w ) = 0 so v and w are null coordinates.

Convince yourself that the following describes a physical procedure for ob-
taining the coordinates v(p) and w(p) for an event p ∈ R1,3 : Find a light ray
that is “incoming” to the origin and experiences p and take v(p) to be the
arrival time t of the ray at the origin. Next, find a light ray that is “outgoing”
from the origin and experiences p and take w(p) to be the departure time t
of the ray from the origin.
Our next example is still topologically trivial, but is geometrically much
more interesting. The Einstein-deSitter spacetime E is the simplest of all
the cosmological models to emerge from general relativity and is defined as
follows: As a smooth manifold, E is just the open submanifold (0, ∞) × R3 of
R4 . The restriction of the standard chart on R4 to (0, ∞) × R3 gives global
coordinate functions on E that we will continue to denote (x0 , x1 , x2 , x3 ).
Thus, at each p = (x0 , x1 , x2 , x3 ) ∈ E the tangent space Tp (E) is spanned
by {∂0 (p), ∂1 (p), ∂2 (p), ∂3 (p)} which, as for R1,3 , we take to be an oriented
174 3. Frame Bundles and Spacetimes

basis. We will define the Lorentz metric g for E by giving its components gαβ
relative to this global chart. Specifically, at each p = (x0 , x1 , x2 , x3 ) ∈ E,


 1 , α=β=0
gαβ (p) = gαβ (x , x , x , x ) = −(x0 )4/3 ,
0 1 2 3
α = β = 1, 2, 3. (3.5.2)


0 , α ̸= β

Thus,
∑
3
g = dx0 ⊗ dx0 − (x0 )4/3 dxi ⊗ dxi (3.5.3)
i=1

so, if v = v α ∂α (p) and w = wα ∂α (p) are in Tp (E), we have

g p (v , w ) = v 0 w0 − (x0 )4/3 (v 1 w1 + v 2 w2 + v 3 w3 ), (3.5.4)

where x0 is the “time coordinate” of p (the “height” of p if we picture the x0 -

axis in (0, ∞) × R3 vertically, as we generally do). One must show, of course,
that (3.5.3) does indeed define a Lorentz metric on E. Since smoothness of
the component functions (3.5.2) is clear on x0 > 0 we need only produce a
frame (b 0 , b 1 , b 2 , b 3 ) at each p ∈ E such that (3.5.1) is satisfied. But for this
one need only define

b 0 = ∂0 (p)
b i = (x0 )−2/3 ∂i (p), i = 1, 2, 3.

Observe that the vector field ∂0 on E is everywhere timelike and so may be

used to time orient E. At each p = (x0 , x1 , x2 , x3 ) ∈ E the null cone consists
of all v = v α ∂α (p) satisfying

(v 1 )2 + (v 2 )2 + (v 3 )2 = (x0 )−4/3 (v 0 )2 ,

which one can interpret geometrically as saying that the null cones in E
“get steeper” as p “gets higher,” i.e., as time goes on. Since E is diffeomor-
phic to R4 its oriented, time oriented, orthonormal frame bundle is trivial.
Remark: The physical significance of the Einstein-deSitter spacetime as
a cosmological model is discussed at great length in [SaW]. This is not our
concern here so we will content ourselves with a few observations on the proper
way to view E without any real attempt at justification.
The vertical straight lines α(t) = (t, x10 , x20 , x30 ), where 0 < t < ∞
and (x10 , x20 , x30 ) ∈ R3 is fixed, are clearly future directed, timelike curves
in E. These are to be interpreted as the worldlines of the galaxy clus-
ters of our universe. The “displacement vector” v between two events with
the same x0 on two different vertical worldlines is spacelike and satis-
1
fies ∥v ∥ = (−⟨v , v ⟩) 2 = K(x0 )2/3 , where K is a positive constant (here
 3.5. Spacetime Manifolds and Spinor Structures 175

we have in mind two “nearby” worldlines so that v , which actually lies

in a tangent space to E, can be thought of as “joining” the worldlines).
Observe that ∥v ∥, which is regarded as the distance between the two galaxy
clusters at the “instant” x0 in the given global coordinate system, satisfies
2
0 −1/3
dx0 ∥v ∥ = 3 K(u )
d 2
> 0 and (dxd 0 )2 ∥v ∥ = − 29 K(x0 )−4/3 < 0 so that these
clusters are receding from each other (the universe is expanding), but at a de-
creasing rate. Finally, we remark that one can define, on any semi-Riemannian
manifold, a real-valued function S called the “scalar curvature” which is re-
garded as a gross numerical measure of the extent to which the manifold
is “curved” at each point. In spacetime manifolds this measures the overall
strength of the gravitational field at each point. In E the scalar curvature is
given by S = 43 (x0 )−2 which becomes unbounded as x0 −→ 0, i.e., as one
recedes into the past. Moving backward in time along the worldlines of the
galaxies one approaches the “missing” 3-space x0 = 0 where the curvature
would have to be infinite. This is the mathematical representation in E of the
“big bang” (notice that it is not a point).
Exercise 3.5.3 Show that the cubic curve λ(t) = (t3/5 , 3t1/5 , 0, 0),
0 < t < ∞ is a future directed null curve in E. Note: This is actually what is
called a null “geodesic” in E and, from it, one can obtain all the null geodesics
of E; see pages 133 and 162 of [SaW].
Our first topologically nontrivial example is called de Sitter spacetime,
denoted D and defined as follows: Consider the 5-dimensional Minkowski space
R1,4 . As we did for Minkowski spacetime we now regard R1,4 as a semi-
Riemannian manifold with constant metric g̃ given, in standard coordinates,
by
∑
4
g̃ = dx0 ⊗ dx0 − dxi ⊗ dxi .
i=1

The map Q : R −→ R given by Q(x) = ⟨x, x⟩ = (x0 )2 − (x1 )2 − (x2 )2 −

1,4

(x3 )2 − (x4 )2 is obviously smooth and has −1 as a regular value so Q−1 (−1)
is a 4-dimensional smooth submanifold of R1,4 . As a manifold, D is just this
smooth submanifold of R1,4 , i.e.,
{
D = (x0 , x1 , x2 , x3 , x4 ) ∈ R1,4 :
}
(x1 )2 + (x2 )2 + (x3 )2 + (x4 )2 − (x0 )2 = 1 .

By suppressing two spatial dimensions one can visualize D as a hyperboloid

of one sheet (keeping in mind that the cross-sectional “circles” at constant x0
are really S 3 ).
Lemma 3.5.1 D is diffeomorphic to R × S 3 .
176 3. Frame Bundles and Spacetimes

Proof: The maps (f : R × S 3 −→ D and g : D −→ R × S 3 defined ) by

1 1 1 1
f (t, y 1 , y 2 , y 3 , y 4 ) = t, (1 + t2 ) 2 y 1 , (1 + t2 ) 2 y 2 , (1 + t2 ) 2 y 3 , (1 + t2 ) 2 y 4 and
(
g(x0 , x1 , x2 , x3 , x4 ) = x0 , (1 + (x0 )2 )− 2 x1 , (1 + (x0 )2 )− 2 x2 ,
1 1

)
(1 + (x0 )2 )− 2 x3 , (1 + (x0 )2 )− 2 x4
1 1

are inverse diffeomorphisms. 

To obtain a Lorentz metric g on D we will identify the tangent spaces

Tp (D), p ∈ D, with subspaces of Tp (R1,4 ) and restrict g̃ to these subspaces
(more precisely, g = ι∗ g̃ , where ι : D ,→ R1,4 is the inclusion). This restriction
g is clearly symmetric and bilinear at each point, but it is not obviously
nondegenerate nor is it clear that it has index 3. To prove this we introduce
a vector field grad Q on R1,4 defined, relative to standard coordinates, by

grad Q(p) = 2x0 ∂0 (p) − 2x1 ∂1 (p) − 2x2 ∂2 (p) − 2x3 ∂3 (p) − 2x4 ∂4 (p)

for each p = (x0 , x1 , x2 , x3 , x4 ) ∈ R1,4 .

Exercise 3.5.4 Show that, at each p ∈ D, g̃ p (grad Q(p), grad Q(p)) = −4
and g̃ p (grad Q(p), v ) = 0 for each v ∈ Tp (D) ⊆ Tp (R1,4 ).
It follows from Exercise 3.5.4 that the restriction of g̃ to each Tp (D) is non-
degenerate (if there were a vector in Tp (D) orthogonal to every other vector
in Tp (D), then, being orthogonal also to grad Q(p), it would be orthogonal to
everything in Tp (R1,4 )). Thus, g is a metric on D and we need only show that
it has index 3. But g̃ has index 4 and, at any p ∈ D, U (p) = 14 grad Q(p)
is orthogonal to Tp (D) and satisfies g̃ p (U (p), U (p)) = −1 so this is clear (an
orthonormal basis for Tp (D) together with U (p) is an orthonormal basis for
Tp (R1,4 )). The existence of the globally defined unit normal field U on D also
provides D with the orientation and time orientation we require for D.
Exercise 3.5.5 Show that the following defines an orientation µ on D: At
each p ∈ D, an ordered basis {b 0 , b 1 , b 2 , b 3 } for Tp (D) is in µp if and only if
{b 0 , b 1 , b 2 , b 3 , U (p)} is in the standard orientation for Tp (R1,4 ) = Tp (R5 ).
Exercise 3.5.6 For each p ∈ D define

V (p) = ∂0 (p) + g̃(U (p), ∂0 (p))U (p).

Show that V (p) ∈ Tp (D) and g (V (p), V (p)) > 0. Conclude that D is time
oriented by V = ∂0 + g̃(U , ∂0 )U .
Exercise 3.5.7 Use the oriented, orthonormal frame field {V 1 , V 2 , V 3 } for
S 3 constructed in Section 3.3 (see (3.3.9)) and the timelike, future directed
vector field V on D constructed in Exercise 3.5.6 to show that the oriented,
time oriented, orthonormal frame bundle of D is trivial.
 3.5. Spacetime Manifolds and Spinor Structures 177

We introduce coordinates ξ, ϕ, θ, t on D as follows: Define a map τ from

R4 into R1,4 by
x0 = sinh t
x1 = sin ξ sin ϕ cos θ cosh t
x2 = sin ξ sin ϕ sin θ cosh t
x3 = sin ξ cos ϕ cosh t
x4 = cos ξ cosh t.

Then (x1 )2 +(x2 )2 +(x3 )2 +(x4 )2 −(x0 )2 = 1 so the image of τ is in D. For each
fixed t in (−∞, ∞), x1 , x2 , x3 and x4 parametrize the 3-sphere of radius cosh t
and covers the entire 3-sphere for ξ, ϕ and θ restricted to 0 ≤ ξ ≤ π, 0 ≤ ϕ ≤ π
and 0 ≤ θ ≤ 2π. With these values of ξ, ϕ and θ and −∞ < t < ∞, τ maps
onto D.
Exercise 3.5.8 Compute the 5 × 4 Jacobian matrix of τ and show that it
has rank 4 on each of the following regions in R4 .

0 < ξ < π, 0 < ϕ < π, 0 < θ < 2π, −∞ < t < ∞

0 < ξ < π, 0 < ϕ < π, −π < θ < π, −∞ < t < ∞

Show also that τ is one-to-one when restricted to either of these regions and
conclude that the inverse of each of these restrictions is a chart on D.
Exercise 3.5.9 Show that, relative to the coordinates ξ, ϕ, θ, t on D, the
metric g takes the form
( )
g = dt ⊗ dt − cosh2 t dξ ⊗ dξ + sin2 ξ(dϕ ⊗ dϕ + sin2 ϕdθ ⊗ dθ) .

Remarks: The deSitter spacetime can be regarded as an empty space so-

lution to the Einstein field equations with nonzero cosmological constant (see
Section 5.2 of [HE]). In this universe the spatial cross-sections (t = constant)
are 3-spheres whose radii vary with time (at time t the radius is cosh t). Our
final example of a spacetime describes an analogous “static” universe. Since
the details are virtually identical to those above for D we will leave them to
the reader as exercises.
The Einstein cylinder spacetime C is the submanifold of R1,4 consisting
of all (x0 , x1 , x2 , x3 , x4 ) satisfying (x1 )2 + (x2 )2 + (x3 )2 + (x4 )2 = 1 with
the Lorentz metric g obtained by restricting the metric g̃ of R1,4 to this
submanifold.
Exercise 3.5.10 Show that, thus defined, C is, indeed, a spacetime manifold.
Exercise 3.5.11 Show that C is diffeomorphic to R × S 3 .
178 3. Frame Bundles and Spacetimes

Exercise 3.5.12 Show that C is orientable and time orientable and that its
oriented, time oriented, orthonormal frame bundle is trivial.
Exercise 3.5.13 Show that one can introduce coordinates ξ, ϕ, θ, t on C by

x0 = t
x1 = sin ξ sin ϕ cos θ
x2 = sin ξ sin ϕ sin θ
x3 = sin ξ cos ϕ
x4 = cos ξ

(same restrictions as in Exercise 3.5.8) and that, relative to these coordinates,

( )
g = dt ⊗ dt − dξ ⊗ dξ + sin2 ξ(dϕ ⊗ dϕ + sin2 ϕdθ ⊗ dθ) .

Exercise 3.5.14 Find a few spacelike, timelike and null curves in C.

4
Differential Forms and Integration

Introduction
Physics is expressed in the language of differential equations (e.g., Maxwell,
Dirac, Yang-Mills, Einstein, etc.). Differential equations live on differentiable
manifolds and differentiable manifolds have topologies that influence not only
the solutions to differential equations defined on them, but even the type of
equation that one can define on them. At some naive level then it is perhaps
not surprising that topology and physics interact. The profound depth of this
interaction in recent years, however, has made it abundantly clear that the
naive level is not the appropriate one from which to view this.
It is remarkable that the deep connection between topology and the differ-
ential equations of physics can be made quite explicit. The bridge between
the two subjects is the notion of an elliptic complex of differential operators
on a manifold and its corresponding cohomology. There is, in fact, an explicit
formula (the Atiyah-Singer Index Theorem) relating the analytic properties of
the differential operators in such a complex to the topology of the underlying
manifold. This is, unfortunately, quite beyond our level here, although the
object will arise again in the Appendix. We will, however, in this chapter con-
struct the simplest example of an elliptic complex (the de Rham complex) and,
in the next, study its cohomology. In Chapter 6 we will use the information
thus accumulated to construct the characteristic classes that have, at least
informally, put in an appearance in our earlier discussions of electromagnetic
and Yang-Mills fields (Chapter 2).

4.1 Multilinear Algebra

In this section we assemble the algebraic machinery required to build the
objects of interest to us. Since much of this is a straightforward general-
ization of results we already have available for 0-, 1- and 2-forms we in-
tend to leave many of the routine verifications to the reader in the form of
exercises.
Let E denote a real vector space and k ≥ 0 an integer (more generally, one
can take E to be a module over a commutative ring with identity). A map A
from E ×· · ·×E (k factors) to R is k -multilinear if, for each i with 1 ≤ i ≤ k
and each a ∈ R,
A(v1 , . . . , vi + vi′ , . . . , vk ) = A(v1 , . . . , vi , . . . , vk ) + A(v1 , . . . , vi′ , . . . , vk )

G.L. Naber, Topology, Geometry and Gauge fields: Interactions, 179

Applied Mathematical Sciences 141, DOI 10.1007/978-1-4419-7895-0_4,
© Springer Science+Business Media, LLC 2011
180 4. Differential Forms and Integration

and
A(v1 , . . . , avi , . . . , vk ) = aA(v1 , . . . , vi , . . . , vk )
for all v1 , . . . , vi , vi′ , . . . , vk in E. The set T k (E) of all such multilinear forms
is a real vector space with pointwise operations:

(A + B)(v1 , . . . , vk ) = A(v1 , . . . , vk ) + B(v1 , . . . , vk )

(aA)(v1 , . . . , vk ) = a(A(v1 , . . . , vk )).

Note that T 1 (E) is the dual space E ∗ . For convenience we will take T 0 (E) =
R. The elements of T k (E) are called covariant tensors of rank k (or simply
k -tensors) on E. If T : E1 −→ E2 is a linear transformation we define the
pullback map
T ∗ : T k (E2 ) −→ T k (E1 )

for any k as follows: If A ∈ T k (E2 ), the T ∗ A ∈ T k (E1 ) is given by

(T ∗ A)(v1 , . . . , vk ) = A(T (v1 ), . . . , T (vk )).

Exercise 4.1.1 Show that if A ∈ T k (E), then the following three conditions
are equivalent:
(a) A is zero whenever two of its arguments are equal, i.e., if 1 ≤ i,
j ≤ k and i ̸= j, then A(v1 , . . . , vi , . . . , vj , . . . , vk ) = 0 whenever
vi = vj .

(b) A changes sign whenever two of its arguments are interchanged (and
the remaining arguments are left fixed), i.e., A(v1 , . . . ,
vj , . . . , vi , . . . , vk ) = −A(v1 , . . . , vi , . . . , vj , . . . , vk ).
(c) If σ ∈ Sk is any permutation of {1, . . . , k} and (−1)σ is its sign (1 if
σ is an even permutation and −1 if σ is an odd permutation), then
A(vσ(1) , . . . , vσ(k) ) = (−1)σ A(v1 , . . . , vk ).

An A ∈ T k (E) that satisfies any (and therefore all) of the conditions in

Exercise 4.1.1 is said to be skew-symmetric and the set Λk (E) of all such is a
linear subspace of T k (E). Note that Λ1 (E) = T 1 (E) = E ∗ since the conditions
are satisfied vacuously. For convenience, we take Λ0 (E) = T 0 (E) = R. The
elements of Λk (E) are called k -forms on E. An A ∈ T k (E) which takes the
same value whenever two of its arguments are interchanged (or, equivalently,
whenever its arguments are permuted) is said to be symmetric and the
collection of all such is likewise a linear subspace of T k (E).
For any A ∈ T k (E) and B ∈ T l (E) we define the tensor product A ⊗ B ∈
T k+l
(E) by
(A ⊗ B)(v1 , . . . , vk , vk+1 , . . . , vk+l ) = A(v1 , . . . , vk ) B(vk+1 , . . . , vk+l ).
 4.1. Multilinear Algebra 181

Exercise 4.1.2 Show that A ⊗ B is, indeed, in T k+l (E) and prove each of
the following properties of the tensor product.

(a) (A1 + A2 ) ⊗ B = A1 ⊗ B + A2 ⊗ B
(b) A ⊗ (B1 + B2 ) = A ⊗ B1 + A ⊗ B2
(c) (aA) ⊗ B = A ⊗ (aB) = a(A ⊗ B) (a ∈ R)
(d) A ⊗ (B ⊗ C) = (A ⊗ B) ⊗ C
(e) T ∗ (A ⊗ B) = (T ∗ A) ⊗ (T ∗ B).

Also, find an example to show that B ⊗ A is generally not equal to A ⊗ B.

Because the tensor product is associative (Exercise 4.1.2 (d)) one can un-
ambiguously define the tensor product A1 ⊗ · · · ⊗ Am of any finite number of
tensors on E.
Lemma 4.1.1 Let {e1 , . . . , en } be a basis for E and {e1 , . . . , en } its dual basis
for E ∗ (ei (ej ) = δ i j ). Then, for any k, {ei1 ⊗ · · · ⊗ eik : 1 ≤ i1 , . . . , ik ≤ k} is
a basis for T k (E). Moreover, any A ∈ T k (E) can be uniquely written as
A = Ai1 ···ik ei1 ⊗ · · · ⊗ eik (summation convention)
where Ai1 ··· ik = A(ei1 , . . . , eik ). In particular, dim T k (E) = nk .
Exercise 4.1.3 Prove Lemma 4.1.1. Hint: The n = 2 case is
Lemma 5.11.1, [N4].
In particular, any α ∈ Λk (E) can be expanded as in Lemma 4.1.1. However,
the ei1 ⊗ · · · ⊗ eik are not skew-symmetric, in general, so these do not give a
basis for Λk (E). To produce such a basis we introduce a “skew-symmetrized”
tensor product. Suppose α ∈ Λk (E) and β ∈ Λl (E). We define their wedge
product α∧β as follows: If k = 0 or l = 0 (or both), we simply let α∧β = αβ.
If k > 0 and l > 0 we define
1 ∑ ( )
(α ∧ β) (v1 , . . . , vk+l ) = (−1)σ (α ⊗ β) vσ(1) , . . . , vσ(k+l)
k! l! σ
1 ∑ ( )
= (−1)σ α vσ(1) , . . . , vσ(k)
k! l! σ
( )
× β vσ(k+1) , . . . , vσ(k+l) ,

where the sum is over all permutations σ ∈ Sk+l of {1, . . . , k + l}.

Exercise 4.1.4 Write out each of the following special cases and show that,
in each case, α ∧ β is skew-symmetric.
(a) If α ∈ Λ1 (E) and β ∈ Λ1 (E), then
α ∧ β = α ⊗ β − β ⊗ α.
182 4. Differential Forms and Integration

(b) If α ∈ Λ1 (E) and β ∈ Λ2 (E), then

(α ∧ β)(v1 , v2 , v3 ) = α(v1 )β(v2 , v3 ) − α(v2 )β(v1 , v3 )

+ α(v3 )β(v1 , v2 ).

Showing that, in the general case, α ∧ β is skew-symmetric and deriving

the basic properties of the wedge product takes a bit of work. To facilitate
the arguments we introduce some notation. If σ ∈ Sk is a permutation of
{1, . . . , k} and (v1 , . . . , vk ) is any k -tuple we will write
( )
σ · (v1 , . . . , vk ) = vσ(1) , . . . , vσ(k) (4.1.1)

for the rearranged k -tuple.

Exercise 4.1.5 Show that if σ, ρ ∈ Sk are two permutations of {1, . . . , k},
then
( )
σ · ρ · (v1 , . . . , vk ) = (ρ ◦ σ) · (v1 , . . . , vk )

for any k -tuple (v1 , . . . , vk ).

Now define, for any A ∈ T k (E), a multilinear function Alt(A) on E × · · · × E
(k factors) by
1 ∑ ( )
Alt(A) (v1 , . . . , vk ) = (−1)σ A vσ(1) , . . . , vσ(k) . (4.1.2)
k!
σ ∈ Sk

Thus, in particular, if α ∈ Λk (E) and β ∈ Λl (E), then

(k + l)!
α∧β = Alt(α ⊗ β). (4.1.3)
k! l!
Lemma 4.1.2 Let k ≥ 0 be an integer. Then
1. A ∈ T k (E) ⇒ Alt(A) ∈ Λk (E).
2. α ∈ Λk (E) ⇒ Alt(α) = α.
3. A ∈ T k (E) ⇒ Alt(Alt(A)) = Alt(A).
Proof: (3) obviously follows from (1) and (2). To prove (1) we observe first
that Alt(A) is clearly multilinear so Alt(A) ∈ T k (E). To prove that Alt(A)
is skew-symmetric we verify (b) in Exercise 4.1.1. Thus, we fix i and j with
1 ≤ i, j ≤ k and show that interchanging the ith and j th arguments changes
the sign of Alt(A). Assume, without loss of generality, that 1 ≤ i < j ≤ k and
let (ij) be the permutation of {1, . . . , k} that switches i and j, but leaves the
others fixed. Thus, for any k -tuple (a, b, . . . , c), (ij) · (a, b, . . . , c) has the ith
and j th slots switched while the others are left fixed. Moreover, as σ varies
′
over all the permutations in Sk , so does σ ′ = (ij) ◦ σ, but (−1)σ = −(−1)σ .
Thus,
 4.1. Multilinear Algebra 183

Alt(A)(v1 , . . . , vj , . . . , vi , . . . , vk )
1 ∑ ( )
= (−1)σ A σ · (v1 , . . . , vj , . . . , vi , . . . , vk )
k! σ
( )
1 ∑ ( )
= (−1) A σ · (ij) · (v1 , . . . , vi , . . . , vj , . . . , vk )
σ
k! σ
1 ∑ ( )
= (−1)σ A ((ij) ◦ σ) · (v1 , . . . , vi , . . . , vj , . . . , vk )
k! σ
1 ∑ ′
( )
=− (−1)σ A σ ′ · (v1 , . . . , vi , . . . , vj , . . . , vk )
k! σ
1 ∑ ′
( )
=− (−1)σ A σ ′ · (v1 , . . . , vi , . . . , vj , . . . , vk )
k! ′
σ
= −Alt(A)(v1 , . . . , vi , . . . , vj , . . . , vk ).
1
Remark: Notice that the factor k! played no role in the proof of (1). This
is not the case, however, for the proof (2).
To prove (2) we begin with an α ∈ Λk (E). Observe first that, for any σ ∈ Sk ,
( )
α σ · (v1 , . . . , vk ) = (−1)σ α (v1 , . . . , vk ).

(Exercise 4.1.1 (c)). Thus,

1 ∑ ( )
Alt (α)(v1 , . . . , vk ) = (−1)σ α σ · (v1 , . . . , vk )
k! σ
1 ∑
= (−1)σ (−1)σ α(v1 , . . . , vk )
k! σ
1( )
= k! α(v1 , . . . , vk ) = α(v1 , . . . , vk )
k!
as required (notice that the 1
k! is essential here). 
From Lemma 4.1.2 (a) and (4.1.3) we conclude that, if α ∈ Λ (E) and β ∈ k

Λl (E), then α ∧ β ∈ Λk+l (E). To establish the basic properties of the wedge
product we need two more technical results.
Lemma 4.1.3 Let A ∈ T k (E) and B ∈ T l (E) and suppose Alt(A) = 0. Then
Alt(A ⊗ B) = Alt(B ⊗ A) = 0.
Proof: Begin by writing

(k + l)!Alt(A ⊗ B) (v1 , . . . , vk+l )

∑ ( ) ( )
= (−1)σ A vσ(1) , . . . , vσ(k) B vσ(k+1) , . . . , vσ(k+l) .
σ ∈ Sk+l
184 4. Differential Forms and Integration

Now let G = {σ ∈ Sk+l : σ(i) = i, i = k + 1, . . . , k + l}.

∑ σ
Exercise 4.1.6 Show that σ ∈ G (−1) A(vσ(1) , . . . , vσ(k) ) B(vσ(k+1) , . . . ,
vσ(k+l) ) = 0.
Next suppose σ0 ∈ Sk+l − G. Consider

σ0 ◦ G = {σ0 ◦ σ : σ ∈ G}.

Observe that G ∩ (σ0 ◦ G) = ∅ since if τ were in G ∩ (σ0 ◦ G), then τ = σ0 ◦ σ

for some σ ∈ G and therefore σ0 = τ ◦ σ −1 would imply that σ0 is in G, which
it is not. Now,
∑ ( )
(−1)τ (A ⊗ B) τ · (v1 , . . . , vk+l )
τ ∈ σ0 ◦ G
∑ ( )
= (−1)σ0 ◦ σ (A ⊗ B) (σ0 ◦ σ)(v1 , . . . , vk+l )
σ∈G
( )
∑ ( )
= (−1) σ0
(−1) (A ⊗ B) σ · σ0 · (v1 , . . . , vk+l )
σ
.
σ∈G

Let σ0 · (v1 , . . . , vk+l ) = (w1 , . . . , wk , wk+1 , . . . , wk+l ). Then this last sum
becomes
∑ ( ) ( )
(−1)σ0 (−1)σ A wσ(1) , . . . , wσ(k) B wσ(k+1) , . . . , wσ(k+l)
σ∈G

which is zero by Exercise 4.1.6. Thus,

∑ ( )
(−1)τ (A ⊗ B) τ · (v1 , . . . , vk+l ) = 0.
τ ∈ σ0 ◦ G

Next, suppose σ1 ∈ Sk+l − (G ∪ σ0 ◦ G) and let σ1 ◦ G = {σ1 ◦ σ : σ ∈ G}.

As above, G ∩ (σ1 ◦ G) = ∅.
Exercise 4.1.7 Show that, moreover, (σ1 ◦ G) ∩ (σ0 ◦ G) = ∅.
Now, exactly as above, we have
∑ ( )
(−1)τ (A ⊗ B) τ · (v1 , . . . , vk+l ) = 0.
τ ∈ σ1 ◦ G

Since Sk+l is finite we may continue in this way, splitting Sk+l into finitely
many disjoint subsets, the sum over each being zero. Thus,
∑ ( )
(−1)τ (A ⊗ B) τ · (v1 , . . . , vk+l ) = 0
τ ∈ Sk+l
 4.1. Multilinear Algebra 185

so Alt(A ⊗ B) = 0. Replacing G by G′ = {σ ∈ Sk+l : σ(i) = i, i = 1, . . . , k}

the same argument shows that Alt(A ⊗ B) = 0. 
Lemma 4.1.4 Let α ∈ Λk (E), β ∈ Λl (E) and γ ∈ Λm (E). Then

Alt(α ⊗ Alt(β ⊗ γ)) = Alt(α ⊗ β ⊗ γ) = Alt(Alt(α ⊗ β) ⊗ γ).

Proof: First observe that

Alt(Alt(β ⊗ γ) − β ⊗ γ) = Alt(Alt(β ⊗ γ)) − Alt(β ⊗ γ)

= Alt(β ⊗ γ) − Alt(β ⊗ γ) = 0

by (3) of Lemma 4.1.2. Thus, by Lemma 4.1.3

0 = Alt(α ⊗ [Alt(β ⊗ γ) − β ⊗ γ])

= Alt(α ⊗ Alt(β ⊗ γ) − α ⊗ β ⊗ γ)
= Alt(α ⊗ Alt(β ⊗ γ)) − Alt(α ⊗ β ⊗ γ)

which proves the first equality. The second equality is proved

analogously. 
Theorem 4.1.5 Let α, α1 , α2 ∈ Λk (E), β, β 1 , β 2 ∈ Λl (E), γ ∈ Λm (E) and
a ∈ R. Then

1. (α1 + α2 ) ∧ β = α1 ∧ β + α2 ∧ β

2. α ∧ (β 1 + β 2 ) = α ∧ β 1 + α ∧ β 2

3. (aα) ∧ β = α ∧ (aβ) = a(α ∧ β)

4. β ∧ α = (−1)kl α ∧ β

(k+l+m)!
5. (α ∧ β) ∧ γ = α ∧ (β ∧ γ) = k! l! m! Alt(α ⊗ β ⊗ γ)

6. If T : E1 −→ E2 is a linear transformation, α ∈ Λk (E2 ) and

β ∈ Λl (E2 ), then
T ∗ (α ∧ β) = (T ∗ α) ∧ (T ∗ β).

Proof: (1), (2) and (3) follow directly from the corresponding properties of
the tensor product (Exercise 4.1.2). To prove (4) we write

(α ∧ β) (v1 , . . . , vk , vk+1 , . . . , vk+l )

1 ∑ ( ) ( )
= (−1)σ α vσ(1) , . . . , vσ(k) β vσ(k+1) , . . . , vσ(k+l)
k! l!
σ ∈ Sk+l
186 4. Differential Forms and Integration

and

(β ∧ α) (v1 , . . . , vk , vk+1 , . . . , vk+l )

1 ∑ ( ) ( )
= (−1)σ β vσ(1) , . . . , vσ(l) α vσ(l+1) , . . . , vσ(l+k)
l! k!
σ ∈ Sk+l
1 ∑ ( ) ( )
= (−1)σ α vσ(l+1) , . . . , vσ(l+k) β vσ(1) , . . . , vσ(l) .
k! l!
σ ∈ Sk+l

Let ρ be the permutation of {1, . . . , k + l} that sends {1, . . . , l,

l + 1, . . . , l + k} in order to {l + 1, . . . , l + k, 1, . . . , l}. Then ρ is the product
of kl transpositions (move each of the last k in order over the preceding l)
so (−1)ρ = (−1)kl . Thus, for any σ ∈ Sk+l , (−1)σ ◦ ρ = (−1)kl (−1)σ .
Moreover, as σ varies over Sk+l , so does σ ′ = σ ◦ ρ. Thus,

(β ∧ α) (v1 , . . . , vk+l )
1 ∑ ( )
= (−1)σ (α ⊗ β) vσ(l+1) , . . . , vσ(l+k) , vσ(1) , . . . , vσ(l)
k! l!
σ ∈ Sk+l

1 ∑ ( ( ))
= (−1)σ (α ⊗ β) ρ · vσ(1) , . . . , vσ(l) , vσ(l+1) , . . . , vσ(l+k)
k! l!
σ ∈ Sk+l

1 ∑ ( ( ))
= (−1)σ (α ⊗ β) ρ · σ · (v1 , . . . , vk+l )
k! l!
σ ∈ Sk+l

1 ∑ ( )
= (−1)kl (−1)σ ◦ ρ (α ⊗ β) (σ ◦ ρ) · (v1 , . . . , vk+l )
k! l!
σ ∈ Sk+l

1 ∑ ′
( )
= (−1)kl (−1)σ (α ⊗ β) σ ′ · (v1 , . . . , vk+l )
k! l!
σ ′ ∈ Sk+l

= (−1)kl (α ∧ β) (v1 , . . . , vk+l )

as required.

Exercise 4.1.8 Use Lemma 4.1.4 to prove (5).

Exercise 4.1.9 Prove (6). 
A few special cases of Theorem 4.1.5 are worth pointing out explicitly. First
suppose α ∈ Λk (E) and β ∈ Λl (E). If either k or l is even, then
β ∧ α = α ∧ β,

but if both k and l are odd, then

β ∧ α = −α ∧ β.
 4.1. Multilinear Algebra 187

Thus, if k is odd, then

α ∧ α = 0.
This is true, in particular, for 1-forms. Associativity of the wedge product
(Theorem 4.1.5 (5)) permits the unambiguous definition of wedge products
for any finite number of forms. The following special case is of particular
interest.
Exercise 4.1.10 Let α1 , . . . , αk be in Λ1 (E). Show that
∑
α1 ∧ · · · ∧ αk = k! Alt(α1 ⊗ · · · ⊗ αk ) = (−1)σ ασ(1) ⊗ · · · ⊗ ασ(k) .
σ ∈ Sk

Hint: Proceed by induction using (4.1.3).

Theorem 4.1.6 Let {e1 , . . . , en } be a basis for E and {e1 , . . . , en } the
corresponding dual basis for E ∗ = Λ1 (E). Then, for each k = 1, . . . , n,
{ei1 ∧ · · · ∧ eik : 1 ≤ i1 < · · · < ik ≤ n} is a basis for Λk (E). Moreover,
any α ∈ Λk (E) can be uniquely written as
∑ 1
α= αi1 ··· ik ei1 ∧ · · · ∧ eik = αi1 ··· ik ei1 ∧ · · · ∧ eik ,
i1 <···< ik
k!
(n)
where αi1 ···ik = α(ei1 , . . . , eik ). In particular, dim Λk (E) = k .
Proof: Since α ∈ Λk (E) ⊆ T k (E), Lemma 4.1.1 implies that α = αi1 ···ik ei1 ⊗
· · · ⊗ eik , where αi1 ···ik = α(ei1 , . . . , eik ) and the sum is over all 1 ≤
i1 , . . . , ik ≤ n. Exercise 4.1.10 then gives
1
α= αi ··· i ei1 ∧ · · · ∧ eik .
k! 1 k
Now, if any term in this last sum has two of the indices i1 , . . . , ik equal the
wedge product is zero. Hence, we may sum only over those i1 , . . . , ik that are
all distinct. Each such sequence of distinct indices gives rise to k! terms in the
sum, each term differing from the others only in the order of the indices. But
permuting the indices changes both αi1 ···ik and ei1 ∧ · · · ∧ eik by the sign of
the permutation so all of these terms are the same and equal to the unique
term αi1 ···ik ei1 ∧ · · · ∧ eik with i1 < · · · < ik . Consequently,
∑
α= αi1 ··· ik ei1 ∧ · · · ∧ eik .
i1 <··· <ik

In particular, {ei1 ∧ · · · ∧ eik : 1 ≤ i1 < · · · < ik ≤ n} spans Λk (E).

Exercise 4.1.11 Complete the argument by proving linear independence.


188 4. Differential Forms and Integration

Note that, if dim E = n and k > n, then Λk (E)

( contains
) only the zero
function. Moreover, if k = n, then dim Λn (E) = nn = 1 so any nonzero
element of Λn (E) spans Λn (E). We wish to investigate this 1-dimensional
space of top forms a bit more closely.
Lemma 4.1.7 Let α1 , . . . , αk be elements of Λ1 (E). Then {α1 , . . . , αk } is
linearly independent in Λ1 (E) if and only if α1 ∧ · · · ∧ αk ̸= 0 in Λk (E).

Proof: First suppose {α1 , . . . , αk } is linearly independent. Then there is a

basis {e1 , . . . , ek , . . . , en } for E such that the dual basis {e1 , . . . , ek , . . . , en }
satisfies ei = αi for i = 1, . . . , k. Thus, α1 ∧ · · · ∧ αk is an element of a basis
for Λk (E) and so, in particular, is not zero.

Exercise 4.1.12 Prove the converse. 

Theorem 4.1.8 Let {e1 , . . . , en } be a basis for E and let α ∈ Λn (E). If vj =

Ai j ei , j = 1, . . . , n, are any n vectors in E, then

α(v1 , . . . , vn ) = det(Ai j ) α(e1 , . . . , en ).

Proof: First observe that the ordinary determinant function det can be
regarded as an element of Λn (Rn ) as follows: For v1 , . . . , vn ∈ Rn we write
vj = (A1 j , . . . , An j ), j = 1, . . . , n (components relative to the standard basis
for Rn ) and define
 
A1 1 · · · An 1
 
 . .. 
det(v1 , . . . , vn ) = det .. . .
 
A 1 ··· A n
n n

Properties of determinants ensure that this is skew-symmetric so

det ∈ Λn (Rn ). Since det is not identically zero on Rn it generates Λn (Rn ).
Now, if α ∈ Λn (E) we define α′ ∈ Λn (Rn ) by
( ) ( )
α′ (A1 1 , . . . , An 1 ), . . . , (A1 n , . . . , An n ) = α Ai 1 ei , . . . , Ai n ei .

Then α′ = c det for some constant c. Evaluating α′ at ((1, 0, . . . , 0), . . . , (0, 0, . . . , 1))
gives α′ ((1, 0, . . . , 0), . . . , (0, 0, . . . , 1)) = c det (id), i.e., α(e1 , . . . , en ) = c.
Thus,

α(v1 , . . . , vn ) = α(Ai 1 ei , . . . , Ai n ei )
( )
= α′ (A1 1 , . . . , An 1 ), . . . , (A1 n , . . . , An n )
= α(e1 , . . . , en ) det(Ai j ). 
 4.1. Multilinear Algebra 189

Corollary 4.1.9 Let {e1 , . . . , en } be a basis for E, {e1 , . . . , en } its dual basis
for Λ1 (E) and v1 , . . . , vn vectors in E with vj = Ai j ei , j = 1, . . . , n. Then

(e1 ∧ · · · ∧ en )(v1 , . . . , vn ) = det(Ai j ).

Corollary 4.1.10 Let T : E −→ E be a linear transformation. Then T ∗ :

Λn (E) −→ Λn (E) is given by

T ∗ α = (det T )α

for every α ∈ Λn (E).

Proof: If α = 0, then T ∗ α = 0 so the result is trivial. Suppose then
that α ̸= 0. Since dim Λn (E) = 1, α generates Λn (E) so T ∗ α = cα for
some constant c. Let {e1 , . . . , en } be a basis for E and write T ej = Ai j ei ,
j = 1, . . . , n. Then
(T ∗ α)(e1 , . . . , en ) = α(T e1 , . . . , T en )
= α(Ai 1 ei , . . . , Ai n ei )
= det(Ai j ) α (e1 , . . . , en )

so
cα (e1 , . . . , en ) = det(Ai j ) α (e1 , . . . , en ).
Since α ̸= 0, Theorem 4.1.8 implies that α(e1 , . . . , en ) ̸= 0 so
c = det(Ai j ). 
As another application of Theorem 4.1.8 we show that any nonzero element
of Λn (E) determines a unique orientation for E.
Theorem 4.1.11 Let E be an n-dimensional real vector space and α a
nonzero element of Λn (E). Then there is a unique orientation µ for E such
that [e1 , . . . , en ] ∈ µ if and only if α(e1 , . . . , en ) > 0.
Proof: Since α is nonzero, Theorem 4.1.8 implies that α(e1 , . . . , en ) is
nonzero for every basis {e1 , . . . , en } for E. Thus, the ordered bases for E
are divided into two disjoint classes according to whether α(e1 , . . . , en ) > 0
or α(e1 , . . . , en ) < 0. Let µ denote the set of all ordered bases for E for
which α(e1 , . . . , en ) > 0. We claim that µ is an orientation for E, i.e., that
µ is an equivalence class of the equivalence relation ∼ defined on ordered
bases as follows: {ê1 , . . . , ên } ∼ {e1 , . . . , en } if and only if det(Ai j ) > 0,
where êj = Ai j ei , j = 1, . . . , n. But this is clear from Theorem 4.1.8 since
α(ê1 , . . . , ên ) = det(Ai j )α(e1 , . . . , en ). Uniqueness is also clear since E has
precisely two orientations and the other one (i.e., −µ) consists of the ordered
bases for which α(e1 , . . . , en ) < 0. 
Exercise 4.1.13 Show that if {e1 , . . . , en } is an ordered basis for E and
{e1 , . . . , en } is its dual basis for Λ1 (E), then the orientation for E determined
by e1 ∧ · · · ∧ en is precisely the one that contains {e1 , . . . , en }.
190 4. Differential Forms and Integration

Thus, any nonzero n-form on E determines a unique orientation for E. Let

us suppose that, conversely, one is given an orientation µ for E. Then every
nonzero element of Λn (E) is either positive on all the elements of µ or neg-
ative on all the elements of µ (Theorem 4.1.8). Thus, µ divides the nonzero
elements of Λn (E) into two equivalence classes. In general, however, µ does
not single out a unique element of Λn (E). However, we now show that if E
is equipped with the additional structure of an inner product (nondegener-
ate, symmetric, bilinear form, but not necessarily positive definite), then a
distinguished element of Λn (E) is canonically determined.
Suppose then that E has an orientation µ and an inner product g. If
{e1 , . . . , en } and {ê1 , . . . , ên } are two bases for E that are orthonormal
with respect to g, then the matrix (Ai j ) that relates them (êj = Ai j ei ,
j = 1, . . . , n) satisfies det(Ai j ) = ±1. Thus, for any nonzero α ∈ Λn (E),
α(ê1 , . . . , ên ) = ±α(e1 , . . . , en ).
Now let {e1 , . . . , en } be an ordered orthonormal basis in µ (i.e., an oriented
orthonormal basis). Then there exists a nonzero element ω of Λn (E) with
ω(e1 , . . . , en ) = 1 (namely, e1 ∧ · · · ∧ en ). But observe that if {ê1 , . . . , ên } is
any other oriented orthonormal basis for E, then ω(ê1 , . . . , ên ) = 1 as well.
Thus, ω takes any oriented orthonormal basis to 1. Moreover, it carries any
orthonormal basis in −µ to −1.

Exercise 4.1.14 Show that ω is uniquely determined, i.e., that there is only
one element of Λn (E) that carries every oriented orthonormal basis for E to 1.
Theorem 4.1.12 Let E be an n-dimensional real vector space with an orien-
tation µ and an inner product g. Then there exists a unique
ω ∈ Λn (E) such that ω(e1 , . . . , en ) = 1 whenever {e1 , . . . , en } is an oriented
orthonormal basis for E.
For example, if E = Rn with its standard orientation and (positive definite)
inner product, then ω is just the determinant function det ∈ Λn (Rn ) (see
the first few lines in the proof of Theorem 4.1.8). In general, ω is called
the (metric) volume form for E determined by µ and g (recall that, in
R3 , | det(v1 , v2 , v3 )| is the volume of the parallelepiped spanned by v1 , v2 and
v3 ). We know that if {e1 , . . . , en } is an oriented orthonormal basis for E, then
ω = e1 ∧· · ·∧en , where {e1 , . . . , en } is the dual basis. We will need to compute
ω in an arbitrary oriented (but not necessarily orthonormal) basis for E.
Exercise 4.1.15 Let ω be the volume form on E determined by the orienta-
tion µ and an inner product g. Let {ê1 , . . . , ên } be an oriented basis for E and
{ê1 , . . . , ên } its dual basis. For each i, j = 1, . . . , n let ĝij = g(êi , êj ). Show that
1
ω = | det(ĝij )| 2 ê1 ∧ · · · ∧ ên .

Hint: Let {e1 , . . . , en } be an oriented orthonormal basis for E with êj =

Ai j ei , j = 1, . . . , n and show that det(Ai j ) = | det(ĝij )|1/2 .
 4.1. Multilinear Algebra 191

Next we observe the following: If dim E = n and k is any integer with

0 ≤ k ≤ n, then (by Theorem 4.1.6)
( ) ( )
n−k n n
dim Λ (E) = = = dim Λk (E).
n−k k

In particular, Λn−k (E) is isomorphic to Λk (E) for any vector space E. In

general, however, there is no “natural” isomorphism between these two. We
show next, however, that if E has an orientation µ and an inner product g,
then there is a canonical isomorphism
∗
: Λk (E) −→ Λn−k (E)
called the Hodge star operator (the image of a β ∈ Λk (E) under this
isomorphism will be denoted ∗ β ∈ Λn−k (E) and called the Hodge dual of β).
To construct the isomorphism we must first show that the inner product g on
E determines an inner product on each Λk (E).
Suppose then that E is an n-dimensional real vector space with an inner
product g (E need not have an orientation for this part of the construc-
tion). We wish to show that g induces an inner product (also denoted g)
on Λk (E) for each k = 0, 1, . . . , n. Since Λ0 (E) = R we will define, for
α, β ∈ Λ0 (E), g(α, β) = αβ. For k > 0 our objective is to define g in such
a way that, if {e1 , . . . , en } is an orthonormal basis for E with dual basis
{e1 , . . . , en }, then {ei1 ∧ · · · ∧ eik : 1 ≤ i1 < · · · < ik ≤ n} will be an orthonor-
mal basis for Λk (E). The procedure will be to first define g in terms of an
arbitrary basis for E, then show that the definition does not actually depend
on the initial choice of the basis and finally prove that the definition has the
desired property.
Let {e1 , . . . , en } be an arbitrary basis for E (not necessarily orthonormal).
Let gij = g(ei , ej ) for i, j = 1, . . . , n, and let (g ij ) denote the matrix inverse
of (gij ). Now, for α, β ∈ Λk (E), k ≥ 1, we write
1
α= αi ···i ei1 ∧ · · · ∧ eik
k! 1 k
and
1
βj ···j ej1 ∧ · · · ∧ ejk
β=
k! 1 k
as in Theorem 4.1.6. Now define
αj1 ···jk = g i1 j1 · · · g ik jk αi1 ···ik (4.1.4)
(classically, this is called “raising the indices of α with the metric g”).
Our definition of g on Λk (E) is now
1 j1 ···jk
g(α, β) = α βj1 ···jk . (4.1.5)
k!
To show that this definition does not depend on the choice of basis we let
{ê1 , . . . , ên } be another basis with êj = Ai j ei , j = 1, . . . , n. Let (Ai j ) denote
192 4. Differential Forms and Integration

the matrix inverse of (Ai j ). Then the dual bases {e1 , . . . , en } and {ê1 , . . . , ên }
are related by êj = Ai j ei , j = 1, . . . , n.
Exercise 4.1.16 Let ĝij = g(êi , êj ) and let (ĝ ij ) be the matrix inverse of
(ĝij ). Show that
ĝij = Ak i Al j gkl
and
ĝ ij = Ak i Al j g kl
for i, j = 1, . . . , n.
1 1
Now, for α ∈ Λk (E) we write α = αi ···i ei1 ∧ · · · ∧ eik = α̂i ...i êi1
k! 1 k k! 1 k
∧ · · · ∧ êik , where
( )
α̂i1 ...ik = α(êi1 , . . . , êik ) = α Aj1 i1 ej1 , . . . , Ajk ik ejk
= Aj1 i1 · · · Ajk ik αj1 ···jk .

Exercise 4.1.17 Show that α̂j1 ...jk = Al1 j1 . . . Alk jk αl1 ···lk .
Finally, we compute
1 j1 ···jk 1 ( j1 )( )
α̂ β̂j1 ···jk = Al1 · · · Alk jk αl1 ···lk Am1 j1 · · · Amk jk βm1 ···mk
k! k!
1 ( j 1 m1 ) ( )
= Al1 A j1 · · · Alk jk Amk jk αl1 ···lk βm1 ···mk
k!
1
= δl1m1 · · · δlkmk αl1 ···lk βm1 ···mk
k!
1
= αl1 ···lk βl1 ···lk
k!
1
= αj1 ···jk βj1 ···jk
k!
as required. Thus, g is well-defined on Λk (E) and it is clearly bilinear, sym-
metric and nondegenerate.
Theorem 4.1.13 Let E be an n-dimensional real vector space with an in-
ner product g and let k be an integer with 1 ≤ k ≤ n. If {e1 , . . . , en }
is an orthonormal basis for E with dual basis {e1 , . . . , en } for Λ1 (E), then
{ei1 ∧ · · · ∧ eik : 1 ≤ i1 < · · · < ik ≤ n} is an orthonomal basis for Λk (E)
relative to the induced metric g on Λk (E) (defined by (4.1.5)).

Proof: Let i1 < · · · < ik and j1 < · · · < jk be fixed increasing

sequences of indices and let α = ei1 ∧ · · · ∧ eik and β = ej1 ∧ · · · ∧ ejk . Then
α = k!1
αl1 ···lk el1 ∧ · · · ∧ elk , where αl1 ···lk is 1 if l1 · · · lk is an even permutation
of i1 · · · ik , −1 if l1 · · · lk is an odd permutation of i1 · · · ik and 0 otherwise.
Similarly for β = k! 1
βm1 ···mk em1 ∧· · ·∧emk . Since {e1 , . . . , en } is orthonormal,
 4.1. Multilinear Algebra 193

(gij ) is diagonal with gii = ±1 and therefore (g ij ) is diagonal with g ii = ±1.

We compute

1 m1 ···mk 1
g(α, β) = α βm1 ···mk = g l1 m1 · · · g lk mk αl1 ···lk βm1 ···mk .
k! k!
Notice that this will be zero unless l1 = m1 , . . . , lk = mk so

1 l1 l1
g(α, β) = g . . . g lk lk αl1 ···lk βl1 ···lk .
k!

In order for αl1 ···lk to be nonzero, {l1 , . . . , lk } must equal {i1 , . . . , ik }. Simi-
larly, βl1 ···lk will be zero unless {l1 , . . . , lk } = {j1 , . . . , jk }. Thus, g(α, β) will
be zero unless {i1 , . . . , ik } = {j1 , . . . , jk }. In particular,

g(ei1 ∧ · · · ∧ eik , ej1 ∧ · · · ∧ ejk ) = 0 if {i1 , . . . , ik } ̸= {j1 , . . . , jk }.

Now suppose {i1 , . . . , ik } = {j1 , . . . , jk }. Since 1 ≤ i1 < · · · < ik ≤ n and

1 ≤ j1 < · · · < jk ≤ n, we must then have α = β so βl1 ···lk = αl1 ···lk and
therefore
1 l1 l1
g(α, β) = g(α, α) = g · · · g lk lk (αl1 ···lk )2 .
k!

This is a sum over all permutations of {i1 , . . . , ik } and αl1 ···lk = ±1 so

all of the terms are equal and g(α, α) = ±1, i.e.,

g (ei1 ∧ · · · ∧ eik , ei1 ∧ · · · ∧ eik ) = ±1. 

Remark: The last few lines of this proof actually show that
g(ei1 ∧ · · · ∧ eik , ei1 ∧ · · · ∧ eik ) = (−1)m ,
where m is the number of indices among {i1 , . . . , ik } for which gii = −1.
Theorem 4.1.14 Let E be an n-dimensional real vector space with an ori-
entation µ and an inner product g. Let ω be the metric volume form for E
detemined by µ and g and let k be an integer with 0 ≤ k ≤ n. Then there
exists a unique isomorphism
∗
: Λk (E) −→ Λn−k (E)
such that
α ∧ ∗ β = g(α, β)ω (4.1.6)
for all α, β ∈ Λ (E).
k

Proof: For γ ∈ Λn−k (E) we define a map

φγ : Λk (E) −→ R
194 4. Differential Forms and Integration

as follows: Let α ∈ Λk (E). Then α ∧ γ ∈ Λn (E) so α ∧ γ is a real multiple of

ω. The coefficient of ω is taken to be φγ (α) so
α ∧ γ = φγ (α)ω.
φγ is clearly linear and is therefore an element of the dual (Λk (E))∗ of Λk (E).
Thus, γ −→ φγ is a linear map of Λn−k (E) to (Λk (E))∗ . We claim that it is
one-to-one, i.e., that if φγ (α) = 0 for all α ∈ Λk (E), then γ = 0.
Exercise 4.1.18 Prove this when k = 0.
Now suppose k ≥ 1 and φγ (α) = 0 for all α ∈ Λk (E). Then α ∧ γ = 0 for all
α ∈ Λk (E). Let {e1 , . . . , en } be a basis for E with dual basis {e1 , . . . , en } and
write ∑
γ= γi1 ··· in−k ei1 ∧ · · · ∧ ein−k .
i1 <··· < in−k

We show that each coeffiecient is zero. Fix an increasing sequence

j1 < · · · < jn−k . Let l1 < · · · < lk be the remaining indices in {1, . . . , n}
and consider α = el1 ∧ · · · ∧ elk ∈ Λk (E). Then, by assumption,
∑
0=α∧γ = γi1 ··· in−k el1 ∧ · · · ∧ elk ∧ ei1 ∧ · · · ∧ ein−k .
i1 <··· < in−k

Evaluating at (el1 , . . . , elk , ej1 , . . . , ejn−k ) gives γj1 ···jn−k = 0. Thus,

γ = 0.
We conclude that γ −→ φγ is one-to-one. But since dim Λn−k (E) =
dim Λk (E) = dim (Λk (E))∗ , it must be an isomorphism. Now, for each
β ∈ Λk (E), the map α −→ g(α, β) is an element of (Λk (E))∗ and so there is a
unique element of Λn−k (E), which we denote ∗ β, such that φ∗β (α) = g(α, β)
for every α ∈ Λk (E), i.e.,
α ∧∗β = g(α, β)ω
for every α ∈ Λk (E). Furthermore, ∗ β = 0 ∈ Λn−k (E) implies g(α, β) = 0 for
every α and therefore (by nondegeneracy of g), β = 0. Consequently, β −→∗β,
which is clearly linear, is an isomorphism of Λk (E) onto Λn−k (E) with the
required properties. 
Remark: For any β ∈ Λk (E) it is customary to write

∥β∥2 = g(β, β).

Thus, when α = β, (4.1.6) becomes

β ∧∗β = ∥β∥2 ω. (4.1.7)

Some caution is in order here, however, since we have not assumed that our
inner product is positive definite so that ∥β∥2 need not be positive.
 4.1. Multilinear Algebra 195

We will need some formulas for computing Hodge duals. These are
particularly simple relative to an oriented orthonormal basis {e1 , . . . , en } so we
will consider these first. First observe that if β ∈ Λ0 (E) = R, then ∗ β ∈ Λn (E)
and so is a real multiple of ω = e1 ∧ · · · ∧ en .
Exercise 4.1.19 Show that, if β ∈ Λ0 (E) = R, then
∗
β == βω = βe1 ∧ · · · ∧ en ,

where {e1 , . . . , en } is the dual of the oriented orthonormal basis {e1 , . . . , en }.

In particular,
∗
1 = ω = e1 ∧ · · · ∧ en .
By linearity of the Hodge star isomorphism it will suffice to describe
the Hodge dual of any basis element ei1 ∧ · · · ∧ eik , 1 ≤ i1 < · · · < ik ≤ n, in
Λk (E). We have already seen that ∥ei1 ∧ · · · ∧ eik ∥2 = (−1)m , where m is the
number of indices i1 , . . . , ik with gii = −1. Thus,

(ei1 ∧ · · · ∧ eik ) ∧∗ (ei1 ∧ · · · ∧ eik ) = (−1)m e1 ∧ · · · ∧ en .

Since ∗ (ei1 ∧ · · · ∧ eik ) is uniquely determined by this property it will suffice

to exhibit some γ ∈ Λn−k (E) for which (ei1 ∧ · · · ∧ eik ) ∧ γ = (−1)m ω.
In particular, if k = n,

∗
(e1 ∧ · · · ∧ en ) = (−1)s 1,

where s is the index of g.

Exercise 4.1.20 Show that, if k = n − 1, then

∗
(ei1 ∧ · · · ∧ ein−1 ) = ±ein

where {i1 , . . . , in−1 , in } = {1, . . . , n} and one chooses the plus (minus) sign if
i1 · · · in−1 in is an even (odd) permutation of 1· · · n.
Finally, if k < n − 1 one can select l1 , . . . , ln−k so that i1 · · · ik l1 · · · ln−k is an
even permutation of 1 · · · n. Then

(ei1 ∧ · · · ∧ eik ) ∧ ((−1)m el1 ∧ · · · ∧ eln−k ) = (−1)m ω

so
∗
(ei1 ∧ · · · ∧ eik ) = (−1)m el1 ∧ · · · ∧ eln−k , (4.1.8)
where, again, m is the number of indices i1 , . . . , ik with gii = −1.
Exercise 4.1.21 Verify the following concrete examples:
(a) Let E = R3 with its standard orientation and (positive definite)
inner product and let {e1 , e2 , e3 } be an oriented orthonormal basis.
196 4. Differential Forms and Integration

Then
∗ 1 ∗ 2 ∗ 3
e = e2 ∧ e3 e = −e1 ∧ e3 e = e1 ∧ e2
∗ ∗ ∗
(e1 ∧ e2 ) = e3 (e2 ∧ e3 ) = e1 (e1 ∧ e3 ) = −e2
∗
(e1 ∧ e2 ∧ e3 ) = 1.

(b) Let E = R4 with its standard orientation and (positive definite)

inner product and let {e1 , e2 , e3 , e4 } be an oriented orthonormal basis.
Then
∗ 1 ∗
e = e2 ∧ e3 ∧ e4 (e1 ∧ e3 ) = −e2 ∧ e4
∗ ∗
(e1 ∧ e3 ∧ e4 ) = e2 (e1 ∧ e2 ∧ e4 ) = −e3
∗
(e1 ∧ e2 ∧ e3 ∧ e4 ) = 1.

(c) Let E = R1,3 with its standard orientation and Minkowski inner product
and let {e0 , e1 , e2 , e3 } be an oriented orthonormal basis
(assume, as usual, that g(e0 , e0 ) = 1 and g(ei , ei ) = −1 for i = 1, 2, 3).
Then

∗ 2 ∗
e = −e0 ∧ e1 ∧ e3 (e1 ∧ e2 ) = e0 ∧ e3
∗ ∗
(e ∧ e ∧ e ) = −e
0 1 3 2
(e0 ∧ e1 ∧ e2 ∧ e3 ) = −1.

To describe the Hodge dual in an arbitrary oriented basis {ê1 , . . . , ên } with
dual basis {ê1 , . . . , ên } we will use the Levi-Civita symbol


 1, if j1 · · · jn is an even permutation of 1 · · · n

εj1 ···jn = −1, if j1 · · · jn is an odd permutation of 1 · · · n


 0, otherwise

and borrow a result from the theory of determinants: For any n × n matrix
A = (Ai j )i,j=1,...,n ,

εj1 ···jn Ai1 j1 · · · Ain jn = εi1 ···in (det A). (4.1.9)

Now let {e1 , . . . , en } be an oriented orthonormal basis for E with dual basis
{e1 , . . . , en }. We write êj = Ai j ei , j = 1, . . . , n, and denote by (Ai j ) the in-
verse of the matrix (Ai j ). Then êj = Ai j ei . In addition, we write gij = g(ei , ej )
and ĝij = g(êi , êj ) for i, j = 1, . . . , n, denote by (g ij ) and (ĝ ij ) the inverses
of the matrices (gij ) and (ĝij ), respectively, and recall (from Exercise 4.1.15)
that
1
det(Ai j ) = | det(ĝij )| 2 . (4.1.10)
 4.1. Multilinear Algebra 197

Now, for any β ∈ Λk (E) we write

1 1
β= βi ···i ei1 ∧ · · · ∧ eik = β̂i1 ···ik êi1 ∧ · · · ∧ êik ,
k! 1 k k!
and raise the indices as in (4.1.4) to obtain

β j1 ···jk = g i1 j1 · · · g ik jk βi1 ···ik and β̂ j1 ···jk = ĝ i1 j1 · · · ĝ ik jk β̂i1 ···ik .

Exercise 4.1.22 Show that,

1
∗ 1 | det(ĝij )| 2
β= εi1 ···ik j1 ···jn−k β̂ i1 ···ik êj1 ∧ · · · ∧ êjn−k .
(n − k)! k!

Hint: Convert the right-hand side to unhatted coordinates by using

Exercises 4.1.16 and 4.1.17 as well as (4.1.9) and (4.1.10).
Theorem 4.1.15 Let E be an n-dimensional real vector space with an orien-
tation µ and an inner product g of index s. Let k be an integer with 0 ≤ k ≤ n.
Then the corresponding Hodge star isomorphism satisfies
∗∗
β = (−1)k(n−k)+s β

for every β ∈ Λk (E).

Proof: Let {e1 , . . . , en } be an oriented orthonormal basis for E with dual
basis {e1 , . . . , en }. Then there are precisely s indices among {1, . . . , n} for
which gii = g(ei , ei ) = −1.
∗∗
Exercise 4.1.23 Show that if β ∈ Λ0 (E), then β = (−1)s β.
Now suppose 1 ≤ k ≤ n. By linearity it will suffice to prove our result for
β = ei1 ∧ · · · ∧ eik , where 1 ≤ i1 < · · · < ik ≤ n is some fixed increasing index
sequence.
Exercise 4.1.24 Prove the result for k = n − 1. Hint: Exercise 4.1.20.
Now suppose k < n − 1 and let m be the number of indices among {i1 , . . . , ik }
for which gii = −1. Then ∗ β = (−1)m el1 ∧ · · · ∧ eln−k , where i1 · · · ik l1 · · · ln−k
is an even permutation of 1 · · · n. Thus,
∗∗
( )
β = (−1)m ∗ el1 ∧ · · · ∧ eln−k .

But since {i1 , . . . , ik , l1 , . . . , ln−k } = {1, . . . , n},

′
∗
(el1 ∧ · · · ∧ eln−k ) = (−1)m ei1 ∧ · · · ∧ eik

for some appropriate integer m′ which we now determine. Notice that i1 · · · ik

l1 · · · ln−k is related to l1 · · · ln−k i1 · · · ik by a permutation of sign (−1)k(n−k)
(move each of the k indices i to the left over all of the n − k indices l in
198 4. Differential Forms and Integration

the order i1 , . . . , ik ). There are s − m indices among {l1 , . . . , ln−k } for which
gll = −1. Then
(
∗ l1
)
e ∧ · · · ∧ eln−k = (−1)s−m (−1)k(n−k) ei1 ∧ · · · ∧ eik

so m′ = k(n − k) + s − m. Thus
∗∗
β = (−1)m (−1)s−m (−1)k(n−k) ei1 ∧ · · · ∧ eik = (−1)k(n−k)+s β

as required. 
Corollary 4.1.16 If E is an oriented vector space with a positive definite in-
ner product g, then each Hodge star isomorphism ∗: Λk (E) −→ Λn−k (E), 0 ≤
k ≤ n, is an isometry, i.e.,

g (∗α, ∗β) = g(α, β)

for all α, β ∈ Λk (E).

Proof: If dim E = n and ω is the volume form on E, then

g (∗α, ∗β)ω = ∗α ∧ (∗∗ β) = (−1)k(n−k) ∗α ∧ β

= (−1)k(n−k) (−1)(n−k)k β ∧ ∗α
= g(β, α)ω
= g(α, β)ω.

Since {ω} is a basis for Λn (E), g (∗α, ∗β) = g(α, β). 

Exercise 4.1.25 Show that, if the index of g is s = 1, then

g (∗α, ∗β) = −g(α, β).

Exercise 4.1.26 Show that, if dim E = 3 and g is positive definite, then

∗∗
β = β for any β ∈ Λk (E), 0 ≤ k ≤ 3.
Exercise 4.1.27 Show that, if dim E = 4 and g is positive definite, then
∗∗
β = (−1)k β for any β ∈ Λk (E), 0 ≤ k ≤ 4.
∗∗
Exercise 4.1.28 Show that, if dim E = 4 and g has index one, then β=
(−1)k+1 β for any β ∈ Λk (E), 0 ≤ k ≤ 4.

4.2 Vector-Valued Forms

Most of the forms of interest to us will take values, not in R, but in
some finite dimensional real vector space V (e.g., C, H, or the Lie algebra
G of some matrix Lie group G). With the exception of wedge products, which
 4.2. Vector-Valued Forms 199

we discuss in some detail, virtually all of the material in Section 4.1 general-
izes immediately to this context by simply doing everything componentwise
with respect to some basis for V.
As before we let E denote some n-dimensional real vector space. For some
purposes (to be specified as we proceed), E will be assumed to have an orien-
tation µ and an inner product g. Our forms will be defined on E and will take
values in some m-dimensional real vector space V. When appropriate we will
also assume that V has an inner product h (e.g., when V is a Lie algebra, h
will generally arise from the Killing form). We will use {e1 , . . . , en } to denote
a generic basis for E and {T1 , . . . , Tm } will be a basis for V. A map
k
A : E × · · · × E −→ V

is multilinear if, for each i with 1 ≤ i ≤ k and each a ∈ R, A(v1 ,

. . . ,vi + vi′ , . . . , vk ) = A(v1 , . . . , vi , . . . , vk ) + A(v1 , . . . , vi′ , . . . , vk ) and
A(v1 , . . . , avi , . . . , vk ) = aA(v1 , . . . , vi , . . . , vk ) for all v1 , . . . , vi , vi′ , . . . ,
vk ∈ E. The set T k (E, V) of all such is a real vector space with the obvi-
ous pointwise operations:

(A + B)(v1 , . . . , vk ) = A(v1 , . . . , vk ) + B(v1 , . . . , vk )

( )
(aA)(v1 , . . . , vk ) = a A(v1 , . . . , vk ) .

For convenience, we will take T 0 (E, V) = V. The elements of T k (E, V) are

called covariant V-valued tensors of rank k (or simply V-valued k -
tensors) on E. If T : E1 −→ E2 is a linear transformation, then, for any
k, we define the pullback map

T ∗ : T k (E2 , V) −→ T k (E1 , V)

by
( )
(T ∗ A)(v1 , . . . , vk ) = A T (v1 ), . . . , T (vk )
for any A ∈ T k (E2 , V).
If {T1 , . . . , Tm } is any basis for V, then any A ∈ T k (E, V) can be written
uniquely as
A = A1 T1 + · · · + Am Tm = Ai Ti ,

where Ai ∈ T k (E) for each i = 1, . . . , m. The linear operations on T k (E, V)

as well as the pullback corresponding to any linear transformation can clearly
all be performed componentwise relative to any such basis for V.
An A ∈ T k (E, V) is said to be skew-symmetric if it satisfies the
three equivalent conditions (a), (b) and (c) stated in Exercise 4.1.1 and
the set Λk (E, V) of all such is a linear subspace of T k (E, V) (again, we
take Λ0 (E, V) = T 0 (E, V) = V). The elements of Λk (E, V) are called
V-valued k -forms on E. If α ∈ Λk (E, V) and α = αi Ti for some basis
200 4. Differential Forms and Integration

{T1 , . . . , Tm } for V, then each αi ∈ Λk (E). If E has an orientation µ and

an inner product g (and therefore an associated Hodge star) one can define
Hodge star operators
∗
: Λk (E, V) −→ Λn−k (E, V)

componentwise:
∗ ∗
α= (α1 T1 + · · · + αm Tm ) = ∗α1 T1 + · · · + ∗αm Tm . (4.2.1)

Exercise 4.2.1 Show that this definition does not depend on the choice of
basis for V.
If we assume that V also has an inner product h, then, together with g and
the induced inner products on each Λk (E) (also denoted g) we can define
inner products, denoted (gh), on each Λk (E, V) as follows: Let {T1 , . . . , Tm }
be a basis for V, write α, β ∈ Λk (E, V) as α = αi Ti and β = β j Tj and set
hij = h(Ti , Tj ) for i, j = 1, . . . , m. Now define

(gh)(α, β) = hij g(αi , β j ). (4.2.2)

Exercise 4.2.2 Show that this definition does not depend on the choice of
basis for V.
Exercise 4.2.3 Show that, if α, β ∈ Λ0 (E, V) = V, then (gh) coincides with h.

Exercise 4.2.4 Show that, if h is positive definite and {T1 , . . . , Tm } is

h-orthonormal, then (gh)(α, β) = g(α1 , β 1 ) + · · · + g(αm , β m ). In particular,
writing ∥α∥2 for g(α, α),

∥α∥2 = ∥α1 ∥2 + · · · + ∥αm ∥2 .

An A ∈ T k (E, V) is said to be symmetric if it takes the same value when-

ever two of its arguments are interchanged (or, equivalently, whenever its
arguments are permuted). This is clearly the case if and only if each of its
components Ai relative to any basis for V is symmetric. Moreover, the set of
all such is a linear subspace of T k (E, V).
Tensor and wedge products for real-valued forms depend, for their definition,
on the multiplicative structure of R. Unless some such bilinear pairing is avail-
able in V one cannot define such products for V-valued tensors. In the cases
of interest to us, however, such pairings present themselves quite naturally
(e.g., complex multiplication in C, or the Lie bracket in a Lie algebra G). We
now show how to use such additional structures to obtain useful notions of
the tensor and wedge product for vector-valued tensors.
Suppose U, V and W are all real vector spaces and that one is given a
bilinear map ρ : U × V −→ W. Let A ∈ T k (E, U) and B ∈ T l (E, V). We
 4.2. Vector-Valued Forms 201

define the ρ-tensor product A ⊗ρ B ∈ T k+l (E, W) by

( )( )
A ⊗ρ B v1 , . . ., vk , vk+1 , . . ., vk+l
( ) (4.2.3)
= ρ A (v1 , . . . , vk ), B(vk+1 , . . . , vk+l ) .

Note that, if k = l = 0, then A ∈ U, B ∈ V and A ⊗ρ B = ρ(A, B).

Exercise 4.2.5 Verify that A ⊗ρ B is, indeed, in T k+l (E, W) and that ⊗ρ
has all of the properties of ⊗ described in Exercise 4.1.2.
If α ∈ Λk (E, U) and β ∈ Λl (E, V), then their ρ-wedge product α ∧ρ β ∈
Λk+l (E, W) is defined by
( )( )
α ∧ρ β v1 , . . . , vk+l
1 ∑ ( )( ) (4.2.4)
= (−1)σ α ⊗ρ β vσ(1) , . . . , vσ(k+l) ,
k!l! σ

where the sum is over all permutations σ ∈ Sk+l of {1, . . . , k + l }.

Exercise 4.2.6 Let {U1 , . . . , Uq } be a basis for U with α = αi Ui and let
{T1 , . . . , Tm } be a basis for V with β = β j Tj . Show that
∑
q ∑
m
α ∧ρ β = (αi ∧ β j )ρ(Ui , Tj ).
i=1 j=1

Remark: There is no reason to expect that the ρ(Ui , Tj ), i = 1, . . . , q,

j = 1, . . . , m, should constitute a basis for W. Indeed, if U = V = W, then
this clearly cannot be the case (there are too many of them). Thus, the αi ∧β j
cannot be regarded as components of α ∧ρ β.
Exercise 4.2.7 Let U = V = W = C (thought of as a 2-dimensional real
vector space) and let ρ : C × C −→ C be complex multiplication. Let α ∈
Λk (E, C) and β ∈ Λl (E, C). As a basis for C (over R) take {T1 , T2 } = {1, i }
and write α = α1 1 + α2 i = α1 + α2 i and β = β 1 + β 2 i . Show that

α ∧ρ β = (α1 + α2 i ) ∧ρ (β 1 + β 2 i )
= (α1 ∧ β 1 − α2 ∧ β 2 ) + (α1 ∧ β 2 + α2 ∧ β 1 )i .

Remark: According to Exercise 4.2.7 one can wedge complex-valued forms

in the same way one multiplies complex numbers, but with the real and imag-
inary parts multiplied by ordinary wedge. One obtains an analogous result for
quaternion-valued forms when H is regarded as a 4-dimensional real vector
space with basis {1, i , j , k } and ρ : H × H −→ H is quaternion multiplication.
202 4. Differential Forms and Integration

Next we consider the case in which U = V = W = G, where G is the Lie

algebra of some matrix Lie group G and we identify G also with a real vector
space of (possibly complex) matrices. As the bilinear map ρ we take the Lie
bracket [ , ] : G × G −→ G, i.e., the matrix commutator:

ρ(A, B) = [A, B] = AB − BA

for all A, B ∈ G. With this choice of ρ it is customary to write

α ∧ρ β = [α, β]

for any G-valued forms α and β and we will adhere to this custom.
Thus, for any α ∈ Λk (E, G) and β ∈ Λl (E, G) and any v1 , . . . , vk+l in E
we have
( )
[α, β] v1 , . . . , vk+l
(4.2.5)
1 ∑ [ ( ) ( )]
= (−1)σ α vσ(1) , . . . , vσ(k) , β vσ(k+1) , . . . , vσ(k+l) .
k!l!
σ∈Sk+l

If {T1 , . . . , Tm } is any basis for G with structure constants Cijk

(so that
k
[Ti , Tj ] = Cij Tk , i, j = 1, . . . , m), then, by Exercise 4.2.6,

[α, β] = [αi Ti , β j Tj ]
(4.2.6)
= (αi ∧ β j )[Ti , Tj ] = Cij
k
(αi ∧ β j )Tk .

Exercise 4.2.8 Prove each of the following:

(a) [β, α] = (−1)kl+1 [α, β].

(b) For any γ ∈ Λr (E, G),

[ ] [ ] [ ]
(−1)kr [α, β], γ + (−1)rl [γ, α], β + (−1)kl [β, γ], α = 0.

Hint: For (b) use the Jacobi identity (page 15).

For computational purposes it is convenient to introduce another wedge

product on G-valued forms based on another bilinear pairing ρ and from which
[α, β] is easily obtained. Toward this end we take ρ to be simply matrix mul-
tiplication on G × G. Note that, since a matrix Lie algebra G is generally not
closed under matrix multiplication, ρ must be regarded as a map from G × G
to some GL. We will denote the corresponding wedge product simple α ∧ β so
 4.2. Vector-Valued Forms 203

that
( )
(α ∧ β) v1 , . . . , vk+l
1 ∑ ( ) ( ) (4.2.7)
σ
= (−1) α vσ(1) , . . . , vσ(k) β vσ(k+1) , . . . , vσ(k+l) .
k! l!
σ ∈Sk+l

Exercise 4.2.9 Show that, if α ∈ Λk (E, G) and β ∈ Λl (E, G), then

[α, β] = α ∧ β − (−1)kl β ∧ α. (4.2.8)

Remark: In particular, for G-valued 1-forms α and β, [α, β] = α ∧ β

+ β ∧ α so that [β, α] = [α, β]. The same is true, of course, if either α or
β has any odd rank. For a single G-valued form α of odd rank one obtains
( )
[α, α] = 2α ∧ α α ∈ Λ2k+1 (E, G) . (4.2.9)
Although it is the wedge product [ , ] that arises most naturally in the
context G-valued forms it is, as we have seen, readily obtained from ∧ and,
as we now show, α ∧ β is particularly easy to compute. The idea is as fol-
lows: Any α ∈ Λk (E, G) can be formally identified with a matrix of k -forms
that are R-, C-, or H-valued by selecting a basis {T1 , . . . , Tm } for G and for-
mally multiplying each entry of T i by αi and adding (we will illustrate with
a concrete example shortly). Then evaluating α at (v1 , . . . , vk ) amounts to
evaluating each entry k -form at (v1 , . . . , vk ). The result is a matrix in G. If we
express both α and β in this way as matrices of ordinary forms, then α ∧ β
can be computed by simply forming the matrix product of α and β, but with
the entries multiplied by the ordinary (real, complex, or quaternionic) wedge
product. To see this we compute as follows: Let the matrix representations
for α and β be (αij ) and (β ij ). Then
  
α11 α12 ... β 11 ...
   21 ( )
  β  v1 , . . . , vk+l
  
.. ..
. .
 
α11 ∧ β 11 + α12 ∧ β 21 + . . . . . . ( )
= ..  v1 , . . . , vk+l
.
∑ 1j 
α ∧ β j1 . . .
 ( )
= j  v1 , . . . , vk+l
..
.
∑ ( 1j )( ) 
α ∧ β j1 v1 , . . . , vk+l . . .
 
= j 
..
.
204 4. Differential Forms and Integration
 { } 
∑ 1 ∑ ( ) ( )
 (−1)σ α1j vσ(1) , . . . , vσ(k) β j1 vσ(k+1) , . . . , vσ(k+l) ···
= k! l! 
 j σ 
..
.
   
1 ∑ ∑ ( ) ( )
 (−1)σ α1j vσ(1) , . . . , vσ(k) β j1 vσ(k+1) , . . . , vσ(k+l) ···
   
= k! l! σ j 
 
..
.
∑ ( ) ( ) 
1j j1
α v , . . . , v β v , . . . , v ...
1 ∑
σ(1) σ(k) σ(k+1) σ(k+l)
 
= (−1)σ  j 
k! l! σ ..
.
1 ∑ ( ) ( )
= (−1)σ α vσ(1) , . . . , vσ(k) β vσ(k+1) , . . . , vσ(k+l)
k! l! σ
( )
= (α ∧ β) v1 , . . . , vk+l ,

as promised.
We mention one concrete example that has already come up in the context
of SU (2)-Yang-Mills-Higgs theory in Section 2.5. Let E be any n-dimensional
real vector space with an orientation µ and an innerproduct g. For V we take
the Lie algebra su(2) of all 2 × 2 skew-Hermitian, tracefree matrices. We
have defined a (positive definite) inner product h on su(2) by h(A, B) = −2
trace(AB). The basis {T1 , T2 , T3 } for su(2) consisting of
1 1 1
T1 = − i σ1 T2 = − i σ2 T3 = − i σ3 ,
2 2 2
where
( ) ( ) ( )
0 1 0 −i 1 0
σ1 = σ2 = σ3 =
1 0 i 0 0 −1

are the Pauli spin matrices is h-orthonormal, i.e., hij = h(Ti , Tj ) = δij for
i, j = 1, 2, 3. If α, β ∈ Λk (E, su(2)), then
 
3 2 1
1 α i α + α i
α = α1 T1 + α2 T2 + α3 T3 = −  2  (4.2.10)
2 −α + α1 i −α3 i

and similarly for β. The Hodge dual ∗ β is computed componentwise so

 
∗ 3 ∗ 2 ∗ 1
1 β i β + β i
∗
β = ∗ β T1 + ∗ β T2 + ∗ β T3 = −  ∗ 2 ∗ 1
1 2 3 .
2 − β + β i −∗ β 3 i
 4.3. Differential Forms 205

Exercise 4.2.10 Compute α ∧ ∗ β and show that

−2 trace(α ∧ ∗β) = (gh) (α, β) ω,

where ω is the volume form for E determined by µ and g. In particular,

−2 trace(α ∧ ∗α) = ∥α∥2 ω.

If α = αi Ti , then each αi is a real-valued k -form on E so, if {e1 , . . . , en } is

an oriented orthonormal basis for E with dual basis {e1 , . . . , en }, it can be
written αi = k!1
αii1 ···ik ei1 ∧ · · · ∧ eik . Thus, we can write
1
α= αi ···i ei1 ∧ · · · ∧ eik ,
k! 1 k
where
αi1 ···ik = αii1 ···ik Ti
are su(2)-valued functions. Raising the indices componentwise as in (4.1.4)
we obtain
αj1 ···jk = g i1 j1 · · · g ik jk αi1 ···ik .
Similarly, writing β = βj1 ···jk ej1 ∧ · · · ∧ ejk , where the βj1 ···jk are su(2)-valued
functions, the summation convention gives
αi1 ···ik β i1 ···ik

as a sum of matrix products, i.e., as a matrix.

Exercise 4.2.11 Show that
( )
−2 trace αi1 ··· ik β i1 ··· ik = k! (gh) (α, β)

so, in particular,
1 ( ) 1
−trace αi1 ···ik αi1 ···ik = ∥α∥2 .
k! 2
Combining this with Exercise 4.2.10 gives
1 1 ( )
−trace(α ∧ ∗ α) = ∥α∥2 ω = − trace αi1 ···ik αi1 ···ik ω. (4.2.11)
2 k!

4.3 Differential Forms

Roughly, a “differential k-form” α on a manifold X is a “smooth assignment”
to each x ∈ X of a k-form α(x) ∈ Λk (Tx (X)). We have seen that, in the
k = 1 case, there are a number of equivalent ways in which this notion of
a “smooth assignment” can be made precise. A map Θ that assigns to each
x ∈ X a covector (1-form) Θ(x) ∈ Tx∗ (X) = Λ1 (Tx (X)) is smooth if its
∞
i |x ) are C
∂
components Θi (x) = Θ(x)( ∂x for any chart (U, φ). Equivalently,
206 4. Differential Forms and Integration

Θ can be regarded as a C ∞ (X)-module homomorphism from X (X) to C ∞ (X)

so “smoothness” means that Θ operates on C ∞ vector fields to give C ∞ real-
valued functions. Yet again, a smooth 1-form can be identified with a smooth
cross-section of the cotangent bundle T ∗ (X) (or, if you prefer, a smooth,
equivariant map on the frame bundle). All of these generalize quite nicely. We
will formulate all of the appropriate versions of the definition and then leave
it to the reader to prove their equivalence.
Remark: Notice that each of the following equivalent definitions has an
obvious modification in which “smooth” is replaced by some less stringent
regularity condition (e.g., continuous, continuous almost everywhere, etc.).
For the integration theory we construct in Section 4.6 we will want to consider
“measurable” and “integrable” k-forms on a manifold.

1. A differential k -form α on a manifold X is an assignment to each

x ∈ X of a k-form α(x) = αx ∈ Λk (Tx (X)) on the tangent space at x which
is smooth in the following sense: Let (U, φ) be a chart for X with coordinate
functions x1 , . . . , xn . Then at each x ∈ U, α(x) = k! 1
αi1 ···ik (x) dxi1 ∧· · ·∧dxik ,
( ∂ )
where αi1 ···ik (x) = α(x) ∂xi1 |x , . . . , ∂xik |x . We say that α is smooth if these
∂

component functions αi1 ···ik (x) are C ∞ on U for all charts in some atlas for
X. The set of all such is denoted Λk (X) (with Λ0 (X) = C ∞ (X)) and has the
obvious pointwise structure of a C ∞ (X)-module.
k
2. Every α ∈ Λk (X) gives rise to a map α : X (X) × · · · × X (X) −→
C ∞ (X) defined by α(V1 , . . . ,Vk )(x) = α(x)(V 1 (x), . . . , V k (x)) that is skew-
symmetric and C ∞ (X)-multilinear. Conversely, any skew-symmetric, C ∞ (X)-
k
multilinear map A : X (X) × · · · × X (X) −→ C ∞ (X) determines a unique
α ∈ Λk (X) with α(V1 , . . . ,Vk ) = A(V1 , . . . ,Vk ) and this one-to-one corre-
spondence is a C ∞ (X)-module isomorphism. Thus, a differential k -form
α on X can be identified with a skew-symmetric, C ∞ (X)-multilinear map
α : X (X) × · · · × X (X) −→ C ∞ (X).
3. Let GL(n, R) ,→ L(X) −→ X be the linear frame bundle of X. Let
V be the real vector space Λk (Rn ) of skew-symmetric, k-multilinear forms
on Rn and define a representation ρ : GL(n, R) −→ GL(Λk (Rn )) by
(ρ(g)(α))(v1 , . . . , vk ) = α(g T v1 , . . . , g T vk ), where each vi ∈ Rn is written as a
column matrix and g T vi : is the matrix product. The corresponding associated
vector bundle L(X) ×ρ Λk (Rn ) is called the exterior k -bundle of X and a
differential k -form α on X can be identified with a smooth cross-section
α : X −→ L(X) ×ρ Λk (Rn ), Pρ ◦ α = idX , of L(X) ×ρ Λk (Rn ). Equiva-
lently (Section 6.8, [N4]) a differential k -form α on X can be identified
with a smooth map α : L(X) −→ Λk (Rn ) that is ρ-equivariant, i.e., satisfies
α(p · g) = (ρ(g −1 ))(α(p)) for each g ∈ GL(n, R) and each p ∈ L(X).
Exercise 4.3.1 Show that these three definitions are equivalent in the follow-
ing sense. Let Λk1 (X), Λk2 (X) and Λk3 (X) denote the C ∞ -modules specified by
 4.3. Differential Forms 207

definitions #1, 2 and 3, respectively, and find explicit isomorphisms between

Λk1 (X) and Λk2 (X) and between Λk1 (X) and Λk3 (X).

Hint: The equivalence of the first two definitions is established for 1-forms on
pages 265–266 of [N4]. The argument hinges on Lemma 5.7.1 of [N4] which
will be needed here as well. For the equivalence of the first and third definitions
proceed as we did for 1-forms in Section 3.3.
Each of these three views of a k-form on a manifold has its uses (unless
some particular emphasis is required we will generally omit the adjective “dif-
ferential”). For example, the purely algebraic operations on forms such as the
wedge product extend immediately to the manifold setting via the first def-
inition by simply doing everything pointwise: If α ∈ Λk (X) and β ∈ Λl (X),
then α ∧ β is the element of Λk+l (X) defined, at each x ∈ X, by
( )
(α ∧ β)(x) (v 1 , . . . , v k+l ) = (α(x) ∧ β(x)) (v 1 , . . . , v k+l )

for all v 1 , . . . , v k+l ∈ Tx (X). Similarly, if F : X −→ Y is a smooth map and

α ∈ Λk (Y ), then, for each x ∈ X, F∗x : Tx (X) −→ Tf (x) (Y ) is a linear map
and we can define F ∗ α ∈ Λk (X) by pulling α(F (x)) back to x by F∗x , i.e.,
( ∗ ) ( )( )
(F α)(x) (v 1 , . . . , v k ) = α(F (x)) F∗x (v 1 ), . . . , F∗x (v k )

for all v 1 , . . . , v k ∈ Tx (X). All of the purely algebraic properties of these

operations (e.g., those to be found in Theorem 4.1.5) are also satisfied by
differential forms (or continuous forms, etc.).

Exercise 4.3.2 Show that the wedge product is C ∞ (X)-bilinear, i.e., that if
α ∈ Λk (X), and β ∈ Λl (X) and f ∈ C ∞ (X), then

(f α) ∧ β = α ∧ (f β) = f (α ∧ β) .

Exercise 4.3.3 Let X and Y be n-dimensional manifolds, (U, φ) a chart on X

with coordinate functions x1 , . . . , xn and (V, ψ) a chart on Y with coordinate
functions y 1 , . . . , y n . Let F : X −→ Y be a smooth map for which F (U ) ⊆ V .
Show that, for any k ∈ C ∞ (U ),
( )
∗ ∂(y i ◦ F )
F (k dy ∧ · · · ∧ dy ) = (k ◦ F ) det
1 n
dx1 ∧ · · · ∧ dxn .
∂xj

Hint: F ∗ (k dy 1 ∧ · · · ∧ dy n ) = (k ◦ F )F ∗ (dy 1 ∧ · · · ∧ dy n ) and, at each point,

F ∗ (dy 1∧· · ·∧dy n ) is a scalar multiple of dx1∧· · ·∧dxn . Now use Corollary 4.1.9.

Exercise 4.3.4 Let X be an n-dimensional manifold, (U, φ) a chart on X

with coordinate functions x1 , . . . , xn and (V, ψ) a chart on X with coordinate
208 4. Differential Forms and Integration

functions y 1 , . . . , y n and with U ∩ V ̸= ∅. Show that if

k dy 1 ∧ · · · ∧ dy n = h dx1 ∧ · · · ∧ dxn

on U ∩ V , then ( )
∂y i
h = k det .
∂xj
Exercise 4.3.4 and Theorem 4.1.11 provide a link between differential forms
and the orientability of a smooth manifold.
Theorem 4.3.1 An n-dimensional manifold X is orientable if and only if X
admits a nowhere zero smooth n-form ω.

Proof: Suppose first that ω is an n-form on X which is nonzero at every

point of X. If (U, φ) is any chart (for which U is) connected and x1 , . . . , xn are
∂ ∂
the coordinate functions, then ω ∂x 1 , . . . , ∂xn is either everywhere positive
or everywhere negative on U . By renumbering, (or changing )the sign of a
∂ ∂
coordinate, if necessary, we may assume that ω ∂x 1 , . . . , ∂xn > 0 on U .
Obtain an atlas of such charts by selecting one at each point of X. Let (U, φ)
and (V, ψ) be two charts in this atlas with coordinate functions x1 , . . . , xn
and y 1 , . . . , y n , respectively, and with U ∩ V ̸= ∅.( On U ∩ V we ) write ω =
h dx1 ∧ · · · ∧ dxn = k dy 1 ∧ · · · ∧ dy n . Since ω ∂x ∂ ∂
1 , . . . , ∂xn = n!h and
( ∂ )
ω ∂x ∂
1 , . . . , ∂xn = n!k we conclude that h > 0 and k > 0 on U ∩ V . From
i
j ) must be positive on U ∩ V so the atlas
∂y
Exercise 4.3.4 it follows that det( ∂x
we have constructed is an oriented atlas and X is orientable.
For the converse we assume X is orientable and let {(Ui , φi )} be an oriented
atlas for X. On each Ui define ω i = dx1 ∧ · · · ∧ dxn , where x1 , . . . , xn are the
coordinate functions of φi . Each ω i is a nonzero n-form on Ui . If Ui ∩ Uj ̸= ∅,
then, on this intersection, ω j = h ω i , where, by assumption, h > 0. Select for
{Ui } a family {ϕi } of smooth real-valued functions on X with supp(ϕi ) ∈ Ui
and define an n-form ω on all of X by
∑
ω= ϕi ω i .
i

Exercise 4.3.5 Complete the proof by showing that ω is nowhere zero

on X. 
Thus, a nonzero n-form on an n-dimensional manifold X determines a
unique orientation for X. Just as was the case for vector spaces, however,
a given orientation µ for X does not single out a unique element of Λn (X),
but only two equivalence classes (consisting of those elements of Λn (X) that
determine µ and those that determine −µ). With the additional structure of
a semi-Riemannian metric g , however, a unique nonzero n-form on X is de-
termined by the requirement that it send any oriented, orthonormal basis for
any tangent space to 1.
 4.3. Differential Forms 209

Theorem 4.3.2 Let X be an n-dimensional manifold with orientation µ and

semi-Riemannian metric g. Then there exists a unique ω ∈ Λn (X) such that,
for any x ∈ X and any oriented, orthonormal basis {e1 , . . . , en } for Tx (X),
ω(e1 , . . . , en ) = 1.

Exercise 4.3.6 Prove Theorem 4.3.2. Hint: Each point of X is contained

in a connected open set U on which is defined a local oriented orthonormal
frame field {e 1 , . . . , e n } (Exercise 3.3.9). On each such U consider the n-form
e 1 ∧ · · · ∧ e n , where {e 1 , . . . , e n } are the 1-forms dual to {e 1 , . . . , e n }. Show
that any two such n-forms agree on the intersection of their domains.

The n-form ω described in Theorem 4.3.2 is called the (metric) volume

form for X determined by µ and g . Since we will have need of it some-
what later we intend to explicitly calculate the volume form for the n-sphere
S n , n ≥ 1. For this we will regard S n as a submanifold of Rn+1 and, as usual,
identify each Tp (S n ) with a subspace of Tp (Rn+1 ) which, in turn, we identify
with Rn+1 itself via the canonical isomorphism (Example #3, page 4). Rn+1
has its standard orientation and Riemannian metric and we will denote by
x1 , . . . , xn+1 the standard coordinate functions on Rn+1 . Thus, the volume
form on Rn+1 is just
dx1 ∧ · · · ∧ dxn+1 .

Notice that, if v0 = (v01 , . . . , v0n+1 ), . . . , vn = (vn1 , . . . , vnn+1 ) are tangent vectors

at some point of Rn+1 , then Corollary 4.1.9 gives
 
v01 ... v0n+1
 v1 v1n+1 
 1 ... 
(dx ∧ · · · ∧ dx
1 n+1
) (v0 , v1 , . . . , vn ) = det
 .. ..
.

 . . 
vn1 ... vnn+1

Now, the metric for S n is just the restriction to S n of the standard metric
on Rn+1 . The standard orientation for S n (Section 5.10, [N4]) is the one
for which the stereographic projection map φS : US −→ Rn (Example #4,
page 4) is an orientation preserving diffeomorphism (and φN : UN −→ Rn is
orientation reversing). A basis for the tangent space at some point p of S n is
in this orientation if and only if one obtains an oriented basis for Rn+1 by
adjoining to it (at the beginning) an “outward pointing” normal vector to S n
at p.
Exercise 4.3.7 Let p be a point in S n and {e1 , . . . , en } a basis for Tp (S n ) ⊆
Tp (Rn+1 ) = Rn+1 . Show that {e1 , . . . , en } is an oriented orthonormal basis
for Tp (S n ) if and only if {p, e1 , . . . , en } is an oriented orthonormal basis for
Tp (Rn+1 ) = Rn+1 .
210 4. Differential Forms and Integration

Exercise 4.3.7 makes it clear how to define the volume form ω on S n . At each
p ∈ S n and for any v1 , . . . , vn ∈ Tp (S n ) we set
( ) ( )( )
ω p v1 , . . . , vn = dx1 ∧ · · · ∧ dxn+1 p, v1 , . . . , vn
 
p1 . . . pn+1
 1 
 v1 . . . v1n+1 
 
= det . .. . (4.3.1)
 .. 
 . 
vn1 . . . vnn+1

This clearly defines a smooth n-form on S n (indeed, it is the restriction to

S n of a smooth n-form on Rn+1 that we will write out shortly). Moreover, if
{e1 , . . . , en } is an oriented orthonormal basis for Tp (S n ), then {p, e1 , . . . , en }
is an oriented orthonormal basis for Tp (Rn+1 ) so ω p (e1 , . . . , en ) = 1 and
ω must be the volume form for S n (Theorem 4.3.2). To obtain an explicit
representation for ω we will expand the determinant in (4.3.1) by the minors
of the first row.
v12 v13 · · · v1n+1 v11 v13 · · · v1n+1
.. .. .. .. .. ..
ω p (v1 , . . . , vn ) = p1 . . . − p2 . . . + ···
vn2 vn3 · · · vnn+1 vn1 vn3 · · · vnn+1

v11 v12 · · · v1n

.. .. ..
+ (−1)n−1 pn+1 . . .
vn1 vn2 · · · vnn
( )
= p1 dx2 ∧ dx3 ∧ · · · ∧ dxn+1 v1 , . . . , vn
( )
− p2 dx1 ∧ dx3 ∧ · · · ∧ dxn+1 v1 , . . . , vn + · · ·
( )
+ (−1)n−1 pn+1 dx1 ∧ dx2 ∧ · · · ∧ dxn v1 , . . . , vn

∑
ci ∧ · · · ∧ dxn+1 (v1 , . . . , vn )
n+1
= (−1)i−1 pi dx1 ∧ · · · ∧ dx
i=1
(n+1 )
∑ ( )
= ci ∧ · · · ∧ dxn+1
(−1)i−1 xi dx1 ∧ · · · ∧ dx v1 , . . . , vn .
i=1 p

Thus, ω is the restriction to S n of the n-form

∑
n+1
ω̃ = ci ∧ · · · ∧ dxn+1
(−1)i−1 xi dx1 ∧ · · · ∧ dx (4.3.2)
i=1
 4.3. Differential Forms 211

on Rn+1 (ω = ι∗ ω̃, where ι : S n ,→ Rn+1 ). For future reference we will write

this out for the first few values of n.

ω̃ = x1 dx2 − x2 dx1 (n = 1) (4.3.3)

ω̃ = x1 dx2 ∧ dx3 − x2 dx1 ∧ dx3 + x3 dx1 ∧ dx2 (n = 2) (4.3.4)

ω̃ = x1 dx2 ∧ dx3 ∧ dx4 − x2 dx1 ∧ dx3 ∧ dx4

(4.3.5)
+ x3 dx1 ∧ dx2 ∧ dx4 − x4 dx1 ∧ dx2 ∧ dx3 (n = 3)

An orientation and semi-Riemannian metric on an n-manifold X orients and

supplies an inner product to each tangent space and, from this, one obtains,
for each k = 0, . . . , n, a Hodge star operator
∗
: Λk (Tx (X)) −→ Λn−k (Tx (X))

at each point. Thus, one can define the Hodge dual of a k-form on X pointwise:

(∗β)x (v 1 , . . . , v n−k ) = ∗
(β x )(v 1 , . . . , v n−k ).

Exercise 4.3.8 Show that, thus defined, the Hodge dual of a smooth k-form
is a smooth (n − k)-form and that the resulting operator
∗
: Λk (X) −→ Λn−k (X)

is C ∞ (X)-linear.

Hint: Prove smoothness locally by using local oriented orthonormal frame

fields.
Because our definition is pointwise all of the algebraic material on Hodge
duals in Section 4.1 carries over at once to the setting of differential forms. In
particular, if ω is the metric volume form on X and α, β ∈ Λk (X), then

α ∧ ∗ β = g (α, β)ω,

where g (α, β) is the inner product of the forms α and β defined pointwise by
(4.1.4) and (4.1.5) (and therefore a C ∞ function on X).
We conclude this section by observing that, if V is a finite dimensional
real vector space, then a V-valued differential form α on a manifold X is
defined pointwise in the obvious way and, for any choice of a basis {T1 , . . . , Tm }
for V, can be written α = α1 T1 + · · · + αm Tm , where each αi is a real-valued
differential form on X. All of the algebraic material in Section 4.2 extends at
once to this context by simply doing everything pointwise (the reader who is
skeptical and/or scrupulously honest is encouraged to check all of this out).
212 4. Differential Forms and Integration

4.4 The de Rham Complex

The algebraic operations on differential forms introduced in the last section are
most conveniently described pointwise. For the exterior differentation operator
d to which we now turn, however, it is more efficient to regard forms as
operators on vector fields. Recall that if f ∈ Λ0 (X) = C ∞ (X), then the
exterior derivative (or differential) of f is the 1-form df ∈ Λ1 (X) given by
i dx . Recall also that if Θ ∈ Λ (X)
∂f i 1
df (V ) = Vf . In local coordinates, df = ∂x
∞
is a 1-form (thought of as a C (X)-module homomorphism from X (X) to
C ∞ (X)), then its exterior derivative dΘ is the map dΘ : X (X) × X (X) −→
C ∞ (X) defined by

dΘ(V1 , V2 ) = V1 (ΘV2 ) − V2 (ΘV1 ) − Θ([V1 ,V2 ]).

dΘ is C ∞ (X)-bilinear and skew-symmetric (page 320, [N4]) and so is an

element of Λ2 (X). In local coordinates

∂Θj i
dΘ = d(Θj dxj ) = dΘj ∧ dxj = dx ∧ dxj .
∂xi
Before proceeding with the general definition we ask the reader to take this one
step further.
Exercise 4.4.1 Let Ω ∈ Λ2 (X) be a 2-form on X (thought of as a bilinear
operator on vector fields) and define dΩ : X (X) × X (X) × X (X) −→ C ∞ (X)
by
dΩ(V1 , V2 , V3 ) = V1 (Ω(V2 ,V3 )) − V2 (Ω(V1 ,V3 ))
+ V3 (Ω(V1 ,V2 )) − Ω([V1 ,V2 ],V3 )
+ Ω([V1 ,V3 ],V2 ) − Ω([V2 ,V3 ],V1 ).

Show that dΩ ∈ Λ3 (X) and that, in local coordinates,

( )
1 1 1 ∂Ωjk i
dΩ = d Ωjk dx ∧ dx
j k
= dΩjk ∧ dxj ∧ dxk = dx ∧ dxj ∧ dxk .
2 2 2 ∂xi
Hint: To prove that dΩ ∈ Λ3 (X) follow the proof for dΘ, Θ ∈ Λ1 (X),
on page 320, [N4]. You will need the identities V(f g) = V(f )g + f V(g) and
[f V, gW ] = f g[V, W ]+f (Vg)W −g(Wf )V, which are proved in Section 5.7
of [N4]. For the coordinate expression, keep in mind that Lie brackets of
coordinate vector fields must vanish.
We record a few properties of the exterior differentiation operator d thus
defined on k-forms for k = 0, 1, 2. If ω 1 and ω 2 are two such k-forms and
a1 , a2 ∈ R, it is clear that

d(a1 ω 1 + a2 ω 2 ) = a1 dω 1 + a2 dω 2 . (4.4.1)
 4.4. The de Rham Complex 213

We know already that, if f ∈ Λ0 (X) and ω ∈ Λ1 (X), then d(f ω) =

f dω + df ∧ ω and, since wedge product by 0-forms coincides with ordinary
multiplication, this may be written

d(f ∧ ω) = df ∧ ω + f ∧ dω .
From this and the anti-commutativity of wedge products for 1-forms we obtain

d(ω ∧ f ) = dω ∧ f − ω ∧ df .
Exercise 4.4.2 Let ω 1 ∈ Λk (X) and ω 2 ∈ Λl (X) with k + l ≤ 2. Show that

d(ω 1 ∧ ω 2 ) = dω 1 ∧ ω 2 + (−1)k ω 1 ∧ dω 2 . (4.4.2)

We know also that d(df ) = 0 for any f ∈ Λ0 (X).

Exercise 4.4.3 Show that, for any ω ∈ Λ1 (X),

d(dω) = 0 . (4.4.3)

We show now that there is exactly one way to generalize all of this to higher
degree forms and retain properties (4.4.1), (4.4.2) and (4.4.3).
Theorem 4.4.1 Let X be a smooth n-dimensional manifold. Then there exists
a unique family of operators

dk : Λk (X) −→ Λk+1 (X), k = 0, . . . , n

(all written simply d unless particular emphasis is required) which satisfy
1. d(a1 ω 1 + a2 ω 2 ) = a1 dω 1 + a2 dω 2 for all a1 , a2 ∈ R and ω 1 , ω 2 ∈
Λk (X).
2. d(ω 1 ∧ ω 2 ) = dω 1 ∧ ω 2 + (−1)k ω 1 ∧ dω 2 for all ω 1 ∈ Λk (X) and
ω 2 ∈ Λl (X).
3. d(dω) = 0 for all ω ∈ Λk (X).
4. df is given by df (V) = Vf for all f ∈ Λ0 (X).
Remark: Each dk is called an exterior differentiation operator and dω
is the exterior derivative of ω.
Proof: First we show that if a family of operators satisfying (1)–(4)
exists, then it is unique. This we do by showing that if ω ∈ Λk (X), then
(1)–(4) imply that dω must be given in any local coordinate system by

( )
1
dω = d ωi1 ···ik dxi1 ∧ · · · ∧ dxik
k!
(4.4.4)
1 ( )
= dωi1 ···ik ∧ dxi1 ∧ · · · ∧ dxik .
k!
214 4. Differential Forms and Integration

This follows from

( )
1
dω = d ωi1 ···ik dxi1 ∧ · · · ∧ dxik
k!
1 ( )
= d ωi1 ···ik dxi1 ∧ · · · ∧ dxik (by (1))
k!
1 ( )
= d ωi1 ···ik ∧ dxi1 ∧ · · · ∧ dxik
k!

1 [( )
= dωi1 ···ik ∧ dxi1 ∧ · · · ∧ dxik
k! ]
+ (−1)0 ωi1 ···ik ∧ d(dxi1 ∧ · · · ∧ dxik ) (by (1) and (2))
1( )
= dωi1 ···ik ∧ dxi1 ∧ · · · ∧ dxik
k!
since each d(dxi1 ∧ · · · ∧ dxik ) = 0 by (2), (3) and induction.
Next we write out an explicit, coordinate independent formula for an op-
erator on Λk (X) which yields elements of Λk+1 (X) and satisfies (1)–(4) (and
so must, therefore, be given in local coordinates by (4.4.4)). Specifically, we
claim that, for each ω ∈ Λk (X), dω must be given by

∑
k+1 ( ( ))
dω(V1 , . . . ,Vk+1 ) = b i , . . . ,Vk+1
(−1)i+1 Vi ω V1 , . . . ,V
i=1
∑ ( ) (4.4.5)
+ b i , . . . ,V
(−1)i+j ω [Vi , Vj ], V1 , . . . ,V b j , . . . ,Vk+1 .
1≤i<j≤n

Exercise 4.4.4 Show that dω ∈ Λk+1 (X). Hint: See Exercise 4.4.1 and its
Hint.

Exercise 4.4.5 Show that when ω is f ∈ Λ0 (X), Θ ∈ Λ1 (X), or Ω ∈ Λ2 (X),

(4.4.5) reduces to our earlier definitions of df, dΘ and dΩ.

Exercise 4.4.6 Show directly from (4.4.5) that, in local coordinates, dω is

given by (4.4.4). Hint: See Exercise 4.4.1 and its Hint.

All that remains is to prove that, when dω is defined by (4.4.5), conditions

(1)–(4) are all satisfied. Now, (1) is obvious from (4.4.5) and (4) is the first
part of Exercise 4.4.5. On the other hand, (2) and (3) can be proved at each
fixed point of X and therefore locally, in coordinates. Furthermore, linearity
(1) of d and bilinearity of the wedge product imply that we may restrict our
attention to k-forms such as ω = f dxi1 ∧ · · · ∧ dxik , where f ∈ C ∞ (X).
Appealing to Exercise 4.4.6 we compute
 4.4. The de Rham Complex 215

dω = d(f dxi1 ∧ · · · ∧ dxik ) = df ∧ dxi1 ∧ · · · ∧ dxik

( n )
∑ ∂f ∑n
∂f i
= i
dx i
∧ dx i1
∧ · · · ∧ dx ik
= i
dx ∧ dxi1 ∧ · · · ∧ dxik
i=1
∂x i=1
∂x

and
( )
∑n
∂f
d(dω) = d dx ∧ dx ∧ · · · ∧ dx
i i1 ik

i=1
∂xi
∑n ( )
∂f
= d dx ∧ dx ∧ · · · ∧ dx
i i1 ik

i=1
∂xi
∑n ( )
∂f
= d ∧ dxi ∧ dxi1 ∧ · · · ∧ dxik
i=1
∂xi
 
∑n ∑n 2
∂ f
=  dxj  ∧ dxi ∧ dxi1 ∧ · · · ∧ dxik
∂x j ∂xi
i=1 j=1

∑
n ∑
n
∂2f
= dxj ∧ dxi ∧ dxi1 ∧ · · · ∧ dxik .
i=1 j=1
∂xj ∂xi

Now, every term in this sum with i = j is zero since dxi ∧ dxi = 0 and, when
i ̸= j, the terms
∂2f
dxj ∧ dxi ∧ dxi1 ∧ · · · ∧ dxik
∂xj ∂xi
and
∂2f
dxi ∧ dxj ∧ dxi1 ∧ · · · ∧ dxik
∂xi ∂xj

cancel in pairs. Thus, d(dω) = 0. Finally, we let ω 1 = f dxi1 ∧ · · · ∧ dxik and

ω 2 = gdxj1 ∧ · · · ∧ dxjl and compute
( )
d ω 1 ∧ ω 2 ) = d((f g)dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl
= d(f g) ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl
= (g df + f dg) ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧· · · ∧ dxjl
= g df ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl
+ f dg ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl
( ) ( )
= df ∧ dxi1 ∧ · · · ∧ dxik ∧ g dxj1 ∧ · · · ∧ dxjl
( ) ( )
+ (−1)k f dxi1 ∧ · · · ∧ dxik ∧ dg ∧ dxj1 ∧ · · · ∧ dxjl
= dω1 ∧ ω2 + (−1)k ω1 ∧ dω2 . 
216 4. Differential Forms and Integration

Exercise 4.4.7 Show that d commutes with pullback. More precisely, let
F : X −→ Y be a smooth map and ω ∈ Λk (Y ). Show that

d(F ∗ ω) = F ∗ (dω). (4.4.6)

Hint: This is already known when k = 0 (page 10) and k = 1 (page 12).
Assume the result for (k−1)-forms and prove (4.4.6) for ω = f dxi1 ∧· · ·∧dxik .
Exercise 4.4.8 Let X = R3 with its standard Riemannian metric and ori-
entation and let x, y and z be the standard coordinate functions on R3 (so
that { ∂x
∂ ∂
, ∂y ∂
, ∂z } is a global oriented orthonormal frame field and {dx, dy, dz}
is the dual oriented orthonormal coframe field). Let ∗ denote the correspond-
ing Hodge star operator. Finally, consider the natural isomorphism between
Λ1 (R3 ) and X (R3 ) under which a 1-form α = f dx + g dy + h dz corresponds
∂ ∂ ∂
to the vector field V = f ∂x + g ∂y + h ∂z .
(a) Show that, for any f ∈ Λ0 (R3 ), df ∈ Λ1 (R3 ) corresponds to grad f ∈
X (R3 ).
(b) Show that if α ∈ Λ1 (R3 ) corresponds to V ∈ X (R3 ), then ∗dα ∈
Λ1 (R3 ) corresponds to curl V ∈ X (R3 ).
(c) Show that if α ∈ Λ1 (R3 ) corresponds to V ∈ X (R3 ), then ∗d∗α ∈
Λ0 (R3 ) is divV .
(d) Show that, for any f ∈ Λ0 (R3 ), ∗d ∗df is the Laplacian of f (i.e.,
∗ ∗
d df = ∇2 f = div grad f ).
(e) Show that if α, β ∈ Λ1 (R3 ) correspond to V, W ∈ X (R3 ), then
∗
(α ∧ β) ∈ Λ1 (R3 ) corresponds to V × W ∈ X (R3 ).
(f) Use (3) of Theorem 4.4.1 to prove that curl(grad f ) = 0 and
div(curlV ) = 0.
(g) Use (2) of Theorem 4.4.1 to prove that curl(f V ) = grad f × V +
f curl V .
Remark: The upshot of Exercise 4.4.8 is that the calculus of forms on R3 is
just a disguised version of classical vector analysis. One reason for preferring
forms (aside from the elegant integration theory we construct in Section 4.6)
is that they make sense on higher dimensional manifolds whereas much of
vector calculus (e.g., the cross product and curl) do not. Thus, for example,
once they are written in terms of forms, Maxwell’s equations are defined on
any spacetime.
The exterior differentiation operators are linear transformations on vector
spaces of smooth forms and when all of these are collected into the sequence
d0 d1 dn−2 dn−1 dn
Λ0 (X) −→ Λ1 (X) −→ · · · −→ Λn−1 (X) −→ Λn (X) −→ 0
one obtains what is called the de Rham complex of the n-dimensional mani-
fold X. Theorem 4.4.1 (3) asserts that the composition of any two consecutive
 4.4. The de Rham Complex 217

maps in this sequence is identically zero, i.e., the image of any dk−1 is con-
tained in the kernel of dk . A differential form ω on X is said to be closed if
dω = 0 and exact if ω = dη for some form η of degree one less. Thus, one may
rephrase Theorem 4.4.1 (3) by saying that any exact form is closed. The con-
verse is generally not true and we will construct an explicit example shortly.
Indeed, we will devote what remains of this section and all of Chapter 5 to
the issue of when closed implies exact and, when it does not, the extent to
which it does not. Although not apparent at the moment, these are questions
about the topology of X.
We begin our discussion in R2 . If ω is a 1-form on R2 and standard coor-
dinates are x1 and
( ∂ω
2 1 2
) x 1, then2 we can write ω = ω1 dx + ω2 dx and compute
dω = ∂x1 − ∂x2 dx ∧ dx . Thus, if ω is closed we must have
2 ∂ω1

∂ω2 ∂ω1
= .
∂x1 ∂x2
Now define η ∈ Λ0 (R2 ) by
∫ 1 ∫ 1
1 2 1 1 2 2
η(x , x ) = x ω1 (tx , tx )dt + x ω2 (tx1 , tx2 )dt .
0 0

Observe that
∫ 1 ∫ 1
∂η ∂ω1
= x1 (tx 1
, tx 2
)t dt + ω1 (tx1 , tx2 )dt
∂x1 ∂x 1
0
∫ 1 0
2 ∂ω2 1 2
+x 1
(tx , tx )t dt
0 ∂x
∫ 1[ ( )
1 2 1 ∂ω1 1 2
= ω1 (tx , tx ) + t x (tx , tx )
0 ∂x1
( )]
2 ∂ω1 1 2
+t x (tx , tx ) dt
∂x2
∫ 1
d [ ] 1
= t ω1 (tx1 , tx2 ) dt = tω1 (tx1 , tx2 )
0 dt 0
= ω1 (x1 , x2 ).

Similarly,
∂η
= ω2 (x1 , x2 )
∂x2
so
ω = dη .
Thus, on R2 , every closed 1-form is, indeed, exact. We will generalize this
simple result quite substantially when we prove the Poincaré Lemma, but for
the moment we simply wish to contrast it with the situation on R2 − {(0, 0)}.
Here we will write down an explicit 1-form that is closed, but not exact.
218 4. Differential Forms and Integration

We construct our 1-form from two polar angular coordinate functions θ1 and
θ2 . First let L1 = { p ∈ R2 : x1 (p)≥0, x2 (p) = 0 } and U1= R2 − L1 . Define
√1 : U1 −→ R by φ1 (x1 , x2 ) = (r(x , x ), θ1 (x , x )), where r(x , x ) =
2 1 2 1 2 1 2 1 2
φ
1 2 2 2
(x ) + (x ) and θ1 (x , x )) is the unique angle in (0, 2π) such that
tan θ1 (x1 , x2 )) = x2 /x1 . More precisely,


 arctan(x2 /x1 ) , x1 > 0, x2 > 0






π
, x1 = 0, x2 > 0



 2
θ1 (x1 , x2 ) = π + arctan(x2 /x1 ) , x1 < 0





 3π
, x1 = 0, x2 < 0



 2


2 π + arctan(x2 /x1 ), x1 > 0, x2 < 0

Then φ1 (U1 ) = (0, ∞) × (0, 2π) and φ−1 1 : φ1 (U1 ) −→ U1 is given by

φ−1
1 (r, θ 1 ) = (r cos θ 1 , r sin θ1 ). Thus, φ 1 : U 1 −→ φ1 (U1 ) is a diffeomorphism
and, in particular, θ1 is smooth on U1 , i.e., θ1 is a 0-form on U1 . A simple
calculation shows that
−x2 x1
dθ1 = dx1 + 1 2 dx2 on U1 .
(x1 )2 2
+ (x ) 2 (x ) + (x2 )2

Now let L2 = { p ∈ R2 : x1 (p) ≤ 0, x2 (p) = 0 } and U2 = R2 − L2 .

Just as above we define a polar coordinate chart φ2 : U2 −→ (0, ∞) ×
(−π, π), φ2 (x1 , x2 ) = (r(x1 , x2 ), θ2 (x1 , x2 )), where θ2 (x1 , x2 )) is now the
unique angle in (−π, π) with tan θ2 (x1 , x2 )) = x2 /x1 . Thus, θ2 is a 0-form
on U2 and the same calculation as for θ1 shows that

−x2 1 x1
dθ2 = dx + dx2 on U2 .
(x1 )2 + (x2 )2 (x1 )2 + (x2 )2

In particular, dθ1 and dθ2 agree on the intersection U1 ∩ U2 and so together

define a 1-form ω on U1 ∪ U2 = R2 − {(0, 0)}. Moreover, ω is obviously closed
on R2 − {(0, 0)} since, on U1 , dω = d(dθ1 ) = 0 and, on U2 , dω = d(dθ2 ) = 0.
We claim, however, that ω is not exact on R2 − {(0, 0)}.
Remark: Once we have learned how to integrate 1-forms we will be able to
give the traditional proof of this that one finds in calculus books, i.e., we will
show that the integral of ω over the unit circle is nonzero.

Suppose to the contrary that there exists an f ∈ C ∞ (R2 − {(0, 0)}) such
that df = ω on R2 − {(0, 0)}. Then, in particular, on R2 − L1 , df = dθ1 so
d(f − θ1 ) = 0. Thus ∂x 1 = ∂x1 and ∂x2 = ∂x2 on R − L1 . Since R − L1
∂f ∂θ1 ∂f ∂θ1 2 2

is connected it follows that there is a constant c1 such that f = θ1 + c1 on

 4.4. The de Rham Complex 219

R2 − L1 . Similarly, there is a constant c2 such that f = θ2 + c2 on R2 − L2 .

But then θ2 = θ1 + (c1 − c2 ) on R2 − (L1 ∪ L2 ) and this is not the case since
θ2 = θ1 + 0 when x1 > 0 and θ2 = θ1 − 2π when x2 < 0.
Remark: This last argument actually shows that ω fails to be exact on
every neighborhood of (0, 0). On the other hand, ω clearly is exact on a
neighborhood of any point other than the origin.
A subset U of Rn is said to star-shaped with respect to p ∈ U if it contains
the entire line segment from p to anything else in U , i.e., if, for each x ∈
U, tx + (1 − t)p ∈ U for each t in [0, 1]. In particular, U is star-shaped with
respect to the origin if tx ∈ U whenever x ∈ U and 0 ≤ t ≤ 1.
Theorem 4.4.2 (The Poincaré Lemma): Let U be an open, star-shaped
subset of Rn . Then every closed form on U is exact.
Proof: We claim that we may assume U is star-shaped with respect to the
origin. Indeed, this is just a very special case of the following result.
Exercise 4.4.9 Let X be a smooth n-dimensional manifold and suppose that,
for some integer k with 1 ≤ k ≤ n, every closed k-form on X is exact. Let Y
be a smooth manifold diffeomorphic to X. Show that every closed k-form on
Y is exact. Hint: d commutes with pullback (Exercise 4.4.7).
Thus, we will assume that tx ∈ U whenever x ∈ U and 0 ≤ t ≤ 1.
The idea behind the proof of Theorem 4.4.2 invariably seems rather strange
the first time one encounters it, but it has turned out to be such a good idea
that it has acquired a name and an honored place in algebraic topology (and
will be familiar to those who have read Chapter 3 of [N4]). We will construct
a family of linear maps

hk : Λk (U ) −→ Λk−1 (U ), k ≥ 1,

such that
dk−1 ◦ hk + hk+1 ◦ dk = idΛk (U ) (4.4.7)

(such a family of maps is called an algebraic homotopy or cochain homotopy

and we will investigate them more formally in Section 5.4). Observe that if we
manage to build such maps the proof of the Poincaré Lemma will be trivial
since, if dk ω = 0, then hk+1 (dk ω) = 0 so (4.4.7) gives dk−1 (hk ω) = ω so hk ω
is a (k − 1)-form η with d η = ω.
Our definition of the maps hk will be in terms of standard coordinates
x , . . . , xn on Rn . We define hk on elements of Λk (U ) of the form
1

ω = f dxi1 ∧ · · · ∧ dxik

(no assumption about the ordering of the indices) and extend by linearity.
Since U is star-shaped with respect to the origin we may define hk ω for such
220 4. Differential Forms and Integration

an ω by
∑
k (∫ 1 )
k
(h ω)(x) = (−1) a−1
t k−1
f (tx)dt xia dxi1 ∧ · · ·
a=1 0

d (4.4.8)
∧ dx ia ∧ · · · ∧ dxik .

Exercise 4.4.10 Show that hk ω is well-defined, i.e., that the right-hand

side is the same if ω is written (−1)σ f dxσ(i1 ) ∧ · · · ∧ dxσ(ik ) for some
permutation σ of { 1, . . . , k }.
Now we compute dk−1 (hk ω) and hk+1 (dk ω).

∑
k {(∫ 1 )
d k−1 k
(h ω) = (−1) a−1
d k−1
t f (tx)dt xia dxi1 ∧ · · ·
a=1 0
}
d
∧ dx ia ∧ · · · ∧ dxik

∑
k ((∫ 1 ) )
= (−1) a−1
d tk−1
f (tx)dt x ia
∧ dxi1 ∧ · · ·
a=1 0

d
∧ dx ia ∧ · · · ∧ dxik

Exercise 4.4.11 Show that

((∫ 1 ) ) (∫ 1 )
d tk−1 f (tx)dt xia = tk−1 f (tx)dt dxia
0 0
n (∫
∑ 1 )
k ∂f
+ t (tx)dt xia dxj .
j=1 0 ∂xj

Thus,
∑
k (∫ 1 )
d k−1 k
(h ω) = (−1) a−1
t k−1
f (tx)dt dxia ∧ dxi1 ∧ · · ·
a=1 0

d
∧ dx ia ∧ · · · ∧ dxik

∑k ∑ n (∫ 1 )
k ∂f
+ (−1) a−1
t j
(tx)dt xia dxj ∧ dxi1 ∧ · · ·
a=1 j=1 0 ∂x
d
∧ dx ia ∧ · · · ∧ dxik
(∫ 1 )
=k tk−1
f (tx)dt dxi1 ∧ · · · ∧ dxik
0
∑
k ∑
n (∫ 1 )
∂f
+ (−1)a−1
t k
(tx)dt xia dxj ∧ dxi1 ∧ · · ·
a=1 j=1 0 ∂xj
d
∧ dx ia ∧ · · · ∧ dxik .
 4.4. The de Rham Complex 221

Now,

∑n
∂f
dk ω = j
dxj ∧ dxi1 ∧ · · · ∧ dxik
j=1
∂x

so, by linearity of hk+1 ,

n (∫
∑ 1 )
∂f
h k+1 k
(d ω) = k
t (tx)dt xj dxi1 ∧ · · · ∧ dxik
j=1 0 ∂xj

∑
n ∑
k (∫ 1 )
∂f
+ (−1) a
t k
(tx)dt xia dxj ∧ dxi1 ∧ · · ·
j=1 a=1 0 ∂xj
d
∧ dx ia ∧ · · · ∧ dxik .

Thus, computing dk−1 (hk ω) + hk+1 (dk ω), the two double sums cancel and we
obtain
(∫ 1 )
dk−1 (hk ω) + hk+1 (dk ω) = k tk−1 f (tx)dt dxi1 ∧ · · · ∧ dxik
0
n (∫
∑ 1 )
∂f
+ k
t (tx)dt xj dxi1 ∧ · · · ∧ dxik
j=1 0 ∂xj
{∫ 1 [
= ktk−1 f (tx)
0
∑
n
∂f ] }
+ t j
(tx)x dt dxi1 ∧ · · · ∧ dxik
k
j

j=1
∂x
{∫ 1 }
d k
= [t f (tx)]dt dxi1 ∧ · · · ∧ dxik
0 dt
= f (x)dxi1 ∧ · · · ∧ dxik = ω 

Exercise 4.4.12 Let X be a smooth n-dimensional manifold and ω ∈ Λk (X),

1 ≤ k ≤ n, with dω = 0. Show that ω is locally exact, i.e., that for each x ∈ X
there is an open neighborhood U of x in X and a (k − 1)-form η on U such
that dη = ω on U . Hint: Exercise 4.4.9.
Exercise 4.4.13 Let V be a finite dimensional real vector space and ω a
V-valued k-form on the manifold X. Let { T1 , . . . , Tm } be a basis for V and
write ω = ω 1 T1 + · · · + ω m Tm , where each ω i is in Λk (X). Define dω by

dω = dω 1 T1 + · · · + dω m Tm

and show that this definition does not depend on the choice of {T1 , . . . , Tm }.
222 4. Differential Forms and Integration

Exercise 4.4.14 Let U , V and W be real vector spaces and

ρ : U × V −→ W a bilinear map. Let ω 1 be a U-valued k-form and ω 2 a
V-valued l-form on X. Then ω 1 ∧ρ ω 2 a W-valued (k + l)-form on X. Show
that
d(ω 1 ∧ρ ω 2 ) = dω 1 ∧ρ ω 2 + (−1)k ω 1 ∧ρ dω 2 .
We denote the set of all smooth V-valued k-forms on X by Λk (X, V) and
provide it with the obvious pointwise structure of a C ∞ (X)-module.

4.5 Tensorial Forms

P
We consider now a smooth principal G-bundle G ,→ P −→ X, where G
is a matrix Lie group, and with right action σ : P × G −→ P written
σ(p, g) = p · g. V will denote a finite dimensional real vector space and
ρ : G −→ GL(V) a representation of G on V. We will write (ρ(g))(v) = g · v
for each g ∈ G and v ∈ V. Recall that a V-valued 0-form ϕ : P −→ V on P
is said to be equivariant if ϕ(p · g) = g −1 · ϕ(p) and notice that this can be
written σg∗ ϕ = g −1 · ϕ, where σg : P −→ P is the diffeomorphism σg (p) = p · g.
Recall also that the definition of a connection form ω on the bundle includes
the analogous condition σg∗ ω = g −1 · ω, where V is now the Lie algebra G of
G and ρ is the adjoint representation. In this case, the curvature Ω of the
connection ω also satisfies σg∗ Ω = g −1 · Ω (Lemma 6.2.2, [N4]). In general, a
V-valued k-form φ on P is said to be pseudotensorial of type ρ if it satisfies
σg∗ φ = g −1 · φ for each g ∈ G, i.e., if
( )
φp·g (σg )∗p (v 1 ), . . . , (σg )∗p (v k ) = g −1 · φp (v 1 , . . . , v k )

for all g ∈ G, p ∈ P and v 1 , . . . , v k ∈ Tp (P ). φ is said to be tensorial of

type ρ if it is pseudotensorial of type ρ and horizontal in the sense that
φp (v 1 , . . . , v k ) = 0 ∈ V if any one of v 1 , . . . , v k ∈ Tp (P ) is vertical (despite
the terminology this notion does not require the existence of a connection
on the bundle). A 0-form is taken to be vacuously horizontal so, for these,
tensorial and pseudotensorial are the same. They are not the same in general,
however, since a connection form ω is pseudotensorial of type ad, but not
tensorial (indeed, ker ω p = Horp (P )). The curvature Ω of any connection is
tensorial, however. We shall denote the set of all V-valued k-forms on P that
are tensorial of type ρ by Λkρ (P, V) and provide it with its obvious real vector
space structure.
Remark: The elements of Λ0ρ (P, V) are in one-to-one correspondence with
the cross-sections of the associated bundle P ×ρ V (Section 6.8, [N4]). These
cross-sections are often more convenient from the point of view of physics
since they are defined on the base manifold X (e.g., spacetime) rather than
the bundle space P . One can do something similar for the elements of any
Λkρ (P, V), but, for k ≥ 1, the objects defined on X that correspond to tensorial
 4.5. Tensorial Forms 223

k-forms on P are k-forms taking values in the bundle P ×ρ V. Thus, for

example, the curvature form Ω of some connection ω corresponds to a P ×ad
G = ad P valued k-form Fω defined on all of X (see Appendix B of [N4] for
a few more details).
Before proceeding with a general study of tensorial forms we mention one
other special case that will be of particular interest to us in Chapter 6. If the
representation ρ of G on V is the trivial one (i.e., g · v = v for all g ∈ G and
v ∈ V), then the condition σg∗ φ = g −1 · φ reduces to σg∗ φ = φ. Tensorial
forms of this type are characterized by the fact that they project (uniquely)
to forms on X, i.e., there is a unique form φ̄ on X with φ = P ∗ φ̄. It will
clearly suffice to establish this for real-valued forms.
P
Lemma 4.5.1 Let G ,→ P −→ X be a smooth principal G-bundle and φ an
element of Λk (P ) for some k ≥ 0 that satisfies

1. φp (v1 , . . . , vk ) = 0 whenever at least one of v1 , . . . , vk ∈ Tp (P ) is

vertical.

2. σg∗ φ = φ for every g ∈ G, i.e., for each p ∈ P and v1 , . . . , vk ∈ Tp (P ),

( )
φp·g (σg )∗p (v1 ), . . . , (σg )∗p (vk ) = φp (v1 , . . . , vk )

for each g ∈ G.

Then there exists a (necessarily unique) k-form φ̄ ∈ Λk (X) such that

P ∗ φ̄ = φ. Conversely, if φ ∈ Λk (P ) and if there exists a (necessarily unique)
φ̄ ∈ Λk (X) with P ∗ φ̄ = φ, then φ must satisfy #1 and #2.

Proof: We ask the reader to prove uniqueness and the necessity of the two
conditions.

Exercise 4.5.1 Show that projections, when they exist, must be unique (i.e.,
that P ∗ φ̄1 = P ∗ φ̄2 implies φ̄1 = φ̄2 ) and that any φ which does project
to X must satisfy #1 and #2. Note: When k = 0, #1 is taken to be sat-
isfied vacuously and #2 simply says that φ(p · g) = φ(p). In this case the
unique projection φ̄ is obviously given by φ̄(x) = φ(p) for any p ∈ P −1 (x).
Henceforth, we assume k ≥ 1.
Now, suppose φ is a k-form on P satisfying #1 and #2. We define φ̄ on X
as follows: Let x ∈ X and w 1 , . . . , w k ∈ Tx (X). Select some p ∈ P −1 (x) and
then select v 1 , . . . , v k ∈ Tp (P ) with P∗p (v i ) = w i for each i = 1, . . . , k. Now
define
φ̄x (w 1 , . . . , w k ) = φp (v 1 , . . . , v k ). (4.5.1)
224 4. Differential Forms and Integration

Remark: We will prove next that the definition (4.5.1) does not depend
on the choices of p and v 1 , . . . , v k . First, however, we point out the most
convenient way to actually make these choices in practice. Choose a local
cross-section s : V −→ P −1 (V ) with x ∈ V . Let p = s(x) and v i = s∗x (w i )
for i = 1, . . . , k. Then P(p) = P(s(x)) = x and P∗p (v i ) = P∗s(x) (s∗x (w i )) =
(P ◦ s)∗x (w i ) = ( idV )∗x (w i ) = w i as required. Notice that (4.5.1) now gives

φ̄x (w 1 , . . . , w k ) = φs(x) (s∗x (w 1 ), . . . , s∗x (w k )) = (s∗ φ)x (w 1 , . . . , w k )

so
φ̄ = s∗ φ (4.5.2)
on V .
To see that our definition is independent of the choices suppose that p′ ∈
P −1 (x) and v ′1 , . . . , v ′k ∈ Tp′ (P ) are such that P∗p′ (v ′i ) = w i for i = 1, . . . , k.
We must show that

φp′ (v ′1 , . . . , v ′k ) = φp (v 1 , . . . , v k ). (4.5.3)

There exists a g ∈ G such that p′ = p · g (Lemma 4.1.1, [N4]). Thus,

φp′ (v ′1 , . . . , v ′k ) = φp·g (v ′1 , . . . , v ′k ) = φp (v ′′1 , . . . , v ′′k ), (4.5.4)

where
v ′′i = (σg )−1 ′ ′
∗p (v i ) = (σg −1 )∗p·g (v i ), i = 1, . . . , k
(by assumption #2).
Exercise 4.5.2 Show that P∗p (v ′′i ) = w i for i = 1, . . . , k.
Thus, P∗p (v i − v ′′i ) = 0 so v i − v ′′i is vertical for i = 1, . . . , k. By assumption
#1, φp (v 1 − v ′′1 , v 2 , . . . , v k ) = 0 so

φp (v 1 , v 2 , . . . , v k ) = φp (v ′′1 , v 2 , . . . , v k ) .

Similarly, φp (v ′′1 , v 2 − v ′′2 , v 3 , . . . , v k ) = 0 so φp (v ′′1 , v 2 , v 3 , . . . , v k ) =

φp (v ′′1 , v ′′2 , v 3 , . . . , v k ) and therefore

φp (v 1 , v 2 , v 3 , . . . , v k ) = φp (v ′′1 , v ′′2 , v 3 , . . . , v k ).

Continuing in this way gives

φp (v 1 , v 2 , v 3 , . . . , v k ) = φp (v ′′1 , v ′′2 , v ′′3 , . . . , v ′′k )

which, by (4.5.4), proves (4.5.3).

We conclude that φ̄ is well-defined. It satisfies P ∗ φ̄ = φ by definition and
is clearly a k-form on X so we need only verify smoothness. But this can be
proved locally and so follows from (4.5.2). 
 4.5. Tensorial Forms 225

We will have quite a bit more to say about projectable forms on bundles,
but first we derive some general results on derivatives of tensorial forms. First
we show that the exterior derivative of a pseudotensorial form is itself pseu-
dotensorial.
P
Lemma 4.5.2 Let G ,→ P −→ X be a smooth principal bundle, V a finite
dimensional real vector space, ρ : G −→ GL(V) a representation of G on V
and φ a V-valued k-form on P that is pseudotensorial of type ρ. Then dφ is
pseudotensorial of type ρ.

Proof: φ satisfies σg∗ φ = g −1 · φ for every g ∈ G and we must show that

σg∗ (dφ) = g −1 · dφ. But d commutes with pullback and the action of G on V
is linear so
σg∗ (dφ) = d(σg∗ φ) = d(g −1 · φ) = g −1 · dφ. 

The exterior derivative of a tensorial form is therefore pseudotensorial,

but need not be horizontal. To obtain a differentiation operator that car-
ries tensorial forms to tensorial forms requires the existence of a connection
P
on G ,→ P −→ X. With such a connection any pseudotensorial form τ gives
rise to a tensorial form by allowing τ to operate only on horizontal parts.
P
Lemma 4.5.3 Let G ,→ P −→ X be a smooth principal bundle with
a connection ω, V a finite dimensional real vector space, ρ : G −→ GL(V) a
representation of G on V and τ a V-valued k-form on P that is pseudotensorial
of type ρ. Then the V-valued k-form τ H on P defined by
( )
τ H (p)(v1 , . . . , vk ) = τ (p) vH H
1 , . . . , vk

is tensorial of type ρ (vH is the horizontal part of v ∈ Tp (P ); see page 36).

Exercise 4.5.3 Prove Lemma 4.5.3. Hint: By (6.8.3) of [N4],

((σg )∗p (v ))H = (σg )∗p (v H ). 

P
Theorem 4.5.4 Let G ,→ P −→ X be a smooth principal bundle with a con-
nection ω, V a finite dimensional real vector space, ρ : G −→ GL(V) a repre-
sentation of G on V and φ a V-valued k-form on P that is pseu-dotensorial
of type ρ. Then the covariant exterior derivative dω φ of φ, defined by
( )
(dω φ)(p)(v1 , . . . , vk+1 ) = (dφ)H (p) v1 , . . . , vk+1
( )
= dφ(p) vH1 , . . . , vH
k+1

is a tensorial (k + 1)-form of type ρ. In particular,

dω : Λkρ (P, V) −→ Λk+1

ρ (P, V).
226 4. Differential Forms and Integration

Remark: The curvature Ω of a connection ω is, of course, just its covariant

exterior derivative (Ω = dω ω). Another special case that we have had occasion
to consider previously is the covariant exterior derivative of a matter field
(element of Λ0ρ (X, V)). The Cartan Structure Equation provides a convenient
computational formula for dω ω and (6.8.4) of [N4] gives an analogous formula
for matter fields. Shortly we will generalize both of these, but first we observe
that, for forms that project to the base manifold X, dω φ and dφ coincide.
P
Lemma 4.5.5 Let G ,→ P −→ X be a smooth principal bundle with a connec-
tion ω and φ a k-form on P that projects to a k-form φ̄ on X (i.e., φ = P ∗ φ̄).
Then
dω φ = dφ.
Proof: We simply compute

(dφ)p (v 1 , . . . , v k+1 ) = (d(P ∗ φ̄))p (v 1 , . . . , v k+1 )

= (P ∗ (dφ̄))p (v 1 , . . . , v k+1 )
( )
= (dφ̄)P(p) P∗p (v 1 ), . . . , P∗p (v k+1 )
( )
= (dφ̄)P(p) P∗p (v H 1 ), . . . , P∗p (v k+1 )
H

( )
= (P ∗ (dφ̄))p v H H
1 , . . . , v k+1
( )
= (d(P ∗ φ̄))p v H H
1 , . . . , v k+1
( )
= (dφ)p v H H
1 , . . . , v k+1

= (dω φ)p (v 1 , . . . , v k+1 ). 

Now, to find the promised computational formula for dω φ we begin as

in Section 6.8 of [N4] for the special case of matter fields by introducing yet
another wedge product for V-valued forms. We consider the bilinear map from
G × V to V which sends (A, v) ∈ G × V to A · v ∈ V defined by

d d
A·v = (exp(tA) · v)|t=0 = (ρ(exp(tA))(v))|t=0 ,
dt dt
where ρ : G −→ GL(V) is the representation relative to which our forms
are tensorial. Notice that if ρ is just the natural representation of G ⊆
GL(k, C) on Ck (matrix multiplication), then A · v = Av is also matrix mul-
tiplication for each A ∈ G (Exercise 6.8.6, [N4]). A special case of more
immediate concern to us here is the following: Suppose ρ = ad : G −→ GL(G)
is the adjoint representation. Then, for any A, B ∈ G,
 4.5. Tensorial Forms 227

d( )
A·B = adexp(tA) (B) |t=0
dt
1( )
= lim adexp(tA) (B) − B
t−→0 t
= [A, B],

where this last equality follows from the definition of the Lie bracket and is
proved on pages 288–289 of [N4].
Now, the bilinear map (A, v) −→ A·v of G×V to V determines a wedge prod-
uct for G-valued forms and V-valued forms. Specifically, if α ∈ Λk (P, G) and
β ∈ Λl (P, V), then we define α∧β
˙ ∈ Λk+l (P, V) at each point of P by
( ) 1 ∑ ( )
(α∧β)
˙ v 1 , . . . , v k+l = (−1)σ α v σ(1) , . . . , v σ(k)
k! l! σ
( )
· β v σ(k+1) , . . . , v σ(k+l) ,

where the sum is over all permutations σ ∈ Sk+l of {1, . . . , k + l}. Notice, in
particular, that when ρ is the adjoint representation of G on G, then α∧β ˙ is
just the bracket wedge product [α, β] described in Section 4.2. We are now in
a position to prove our major result.
P
Theorem 4.5.6 Let G ,→ P −→ X be a smooth principal bundle with a con-
nection ω, V a finite dimensional real vector space, ρ : G −→ GL(V) a repre-
sentation of G on V and φ ∈ Λkρ (P, V) a V-valued, tensorial k-form of type ρ.
Then
dω φ = dφ + ω ∧φ.˙

Proof: We must show that, for each p ∈ P and all v 1 , . . . , v k+1 ∈ Tp (P ),

( )
(dφ)p v H H
1 , . . . , v k+1 = (dφ)p (v 1 , . . . , v k+1 )

1 ∑ ( )
+ (−1)σ ω p v σ(1) (4.5.5)
k!
σ∈Sk+1
( )
· φp v σ(2) , . . . , v σ(k+1) .

We consider three cases separately.

I. Each of v 1 , . . . , v k+1 is horizontal.

In this case v H i = v i for i = 1, . . . , k + 1 and ω p (v i ) = 0 for i =

1, . . . , k + 1 so both sides of (4.5.5) are just (dφ)p (v 1 , . . . , v k+1 ).

II. Two or more of v 1 , . . . , v k+1 are vertical.

228 4. Differential Forms and Integration

The left-hand side of (4.5.5) is now zero since v H i = 0 for at least

two values of i. Moreover, the sum on the right-hand side is zero since φp
is horizontal and at least one of v σ(2) , . . . , v σ(k+1) is vertical for each σ. Thus,
we need only show that (d φ)p (v 1 , . . . , v k+1 ) is zero as well. Each of the verti-
cal vectors among v 1 , . . . , v k+1 can be extended to fundamental vector fields
on P (Corollary 5.8.9, [N4]). Extend the remaining v i to vector fields on P
in any manner (e.g., constant components on some coordinate neighborhood
and then use Exercise 5.7.2, [N4]). Denote these vector fields V 1 , . . . , V k+1 .
Now, (d φ)p (v 1 , . . . , v k+1 ) is the value at p of
( )
∑
k+1 ( )
d φ (V 1 , . . . ,V k+1 ) = (−1)i+1 V i b i , . . . ,V k+1
φ V 1 , . . . ,V
i=1
∑ (
+ (−1)i+j φ [V i , V j ], V 1 , . . . ,
1≤i<j ≤n
)
b i , . . . ,V
V b j , . . . ,V k+1

(componentwise in V). The first sum vanishes since at least one of

b i , . . . ,V k+1 is vertical and φ is horizontal. The same is true of the
V 1 , . . . ,V
second sum since the Lie bracket of two fundamental vector fields is another
fundamental vector field (Theorem 5.8.8, [N4]) and so is vertical.
III. Precisely one of v 1 , . . . , v k+1 is vertical and the rest are horizontal.
Remark: Establishing (4.5.5) in this case will actually complete the proof
since both sides are multilinear and any v can be written as v = v H + v V .
We may assume that v 1 is vertical and v 2 , . . . , v k+1 are horizontal. Extend
v 1 to a fundamental vector field V 1 . We wish to extend v 2 , . . . , v k+1 to hor-
izontal, (σg )∗ -invariant vector fields on P and for this we need a lemma on
horizontal lifts of vector fields.
P
Lemma 4.5.7 Let G ,→ P −→ X be a smooth principal bundle with a con-
nection ω and let W be a smooth vector field on X. Then there is a unique
smooth vector field W̃ on P that satisfies
1. W̃ (p) is horizontal for each p ∈ P .
2. P∗p (W̃ (p)) = W (P(p)) for each p ∈ P .
3. (σg )∗p (W̃ (p)) = W̃ (p · g) for each p ∈ P and g ∈ G.

Proof: We claim first that if we establish the existence of a unique W̃

satisfying (1) and (2), then (3) necessarily follows. Indeed, since σg is a
diffeomorphism and (σg )∗p (Horp (P )) = Horp·g (P ) ((6.1.3), [N4]), (σg )∗W̃
(defined by ((σg )∗W̃ )p = (σg )∗p·g−1 (W̃ p·g−1 )) is a smooth horizontal vector
field and
 4.5. Tensorial Forms 229
(( ) ) ( ( ))
P∗p (σg )∗W̃ = P∗p (σg )∗p·g−1 W̃ p·g−1
p
( )
= (P ◦ σg )∗p·g−1 W̃ p·g−1
( ) ( )
= P∗p·g−1 W̃ p·g−1 = W P(p · g −1 )
( )
= W P(p)

so uniqueness implies (σg )∗W̃ = W̃ so ((σg )∗W̃ )p·g = W̃ (p · g), i.e.,

(σg )∗p (W̃ (p)) = W̃ (p · g) as required.
Now, since P∗p : Horp (P ) −→ TP(p) (X) is an isomorphism (Exercise 6.1.7,
[N4]), there is only one possible definition of W̃ : For each p ∈ P, P(p) ∈ X
and there exists a unique W̃ (p) ∈ Horp (P ) such that P∗p (W̃ (p)) = W (P(p)).
Thus, W̃ is obviously a vector field on P satisfying (1) and (2), but is
not obviously smooth. We prove smoothness locally. Thus, we select a triv-
ialization Ψ : P −1 (V ) −→ V × G. There is obviously a smooth vector
field on V × G which lifts the restriction W |V of W to V . Transferring
this to P −1 (V ) via Ψ−1
∗ we obtain a smooth vector field on P −1 (V ) which
lifts W |V . The horizontal part of this vector field is smooth on P −1 (V )
(Exercise 6.2.2, [N4]), and lifts W |V so it must agree with W̃ on P −1 (V ).
Thus, W̃ is smooth on P −1 (V ). Since the trivialization was arbitrary, W̃ is
smooth. 
Now we return to the proof of Theorem 4.5.6 in Case III. The vec-
tors P∗p (v 2 ), . . . , P∗p (v k+1 ) extend to vector fields on X and these, by
Lemma 4.5.7, have unique, smooth, horizontal, (σg )∗ -invariant lifts V 2 , . . . ,
V k+1 to P .

Exercise 4.5.4 Show that V i (p) = v i for i = 2, . . . , k + 1.

Since V 1 is a fundamental vector field and each V 2 , . . . ,V k+1 is horizontal,

all of the Lie brackets [V 1 , V i ], i = 2, . . . , k + 1, are horizontal (page 349,
[N4]).

Exercise 4.5.5 Show that, in fact, [V 1 , V 2 ] = · · · = [V 1 , V k+1 | = 0.

From Exercise 4.5.5 and the fact that V 1 is vertical and φ is horizontal we
conclude that
( )
d φ(V 1 , V 2 , . . . ,V k+1 ) = (−1)1+1 V 1 φ (V 2 , . . . ,V k+1 ) + 0
( )
= V 1 φ (V 2 , . . . ,V k+1 ) .
230 4. Differential Forms and Integration

Since the left-hand side of (4.5.5) is obviously zero in Case III, the proof now
reduces to showing that
( ) 1 ∑ ( ) ( )
0 =V 1 φ (V 2 , . . . ,V k+1 ) + (−1)σ ω Vσ(1) · φ Vσ(2) , . . . ,Vσ(k+1) .
k!
σ ∈ Sk+1

But since ω(V σ(1) ) will be zero whenever σ(1) ̸= 1 we may rewrite this as

( ) 1 ∑ ( )
0 = V 1 φ (V 2 , . . . ,V k+1 ) + (−1)σ ω(V1 ) · φ V σ(2) , . . . ,Vσ(k+1)
k! σ∈S
k+1
σ(1)=1

 
( ) 
1 ∑ ( )

0 = V 1 φ (V 2 , . . . ,V k+1 ) +ω(V1 ) · (−1)σ φ Vσ(2) , . . . ,Vσ(k+1)

k! σ∈Sk+1 

σ(1)=1

( )
0 = V 1 φ (V 2 , . . . ,V k+1 ) + ω(V 1 ) · φ(V 2 , . . . ,V k+1 ). (4.5.6)

To prove (4.5.6) we proceed as follows: Since V 1 is a fundamental vector field

we may write V 1 = A# for some A ∈ G. Also let α(t) = exp(tA) for t ∈ R.
Then, for each p ∈ P ,

V 1 (p) = A# (p) = (σp ◦ α)′ (0),

where σp : G −→ P is given by σp (g) = p · g ((5.8.8), [N4]). Thus, for any

f ∈ C ∞ (P ),

V 1 (p) (f ) = (σp ◦ α)′ (0) (f ) = (f ◦ σp ◦ α)′ (0).

To compute V 1 (p)(φ(V 2 , . . . ,V k+1 )) we let f = φ(V 2 , . . . ,V k+1 ). Then f ◦

σp ◦ α is given by
( )( ) ( )
φ V 2 , . . . ,Vk+1 σp (α(t)) = φp·α(t) V 2 (p · α(t)), . . . ,V k+1 (p · α(t))
(( ) ( ) ( )
)
= φp·α(t) σα(t) (V 2 (p)), . . . , σα(t) Vk+1 (p)
∗p ∗p

(by Lemma 4.5.7 (3))

( )
= α(t)−1 · φp V 2 (p), . . . ,V k+1 (p)
( )
since φ ∈ Λkρ (P, V)
( )
= exp(−tA) · φp V 2 (p), . . . ,V k+1 (p) .
 4.5. Tensorial Forms 231

Thus,
( ) d( )
V 1 (p) φ (V 2 , . . . ,V k+1) = exp (t(−A)) · φp (V 2 (p), . . . ,Vk+1 (p))
dt t=0
( )
= (−A) · φp V 2 (p), . . . ,V k+1 (p)
( )
= −ω p (A# ) · φp V 2 (p), . . . ,V k+1 (p)
( )
= −ω p (V 1 (p)) · φp V 2 (p), . . . ,V k+1 (p)

which gives (4.5.6) at the arbitrary point p ∈ P and therefore completes the
proof. 
P
Corollary 4.5.8 Let G ,→ P −→ X be a smooth principal bundle with a
connection ω and ad : G −→ GL(G) the adjoint representation of G on G.
Then, for each φ ∈ Λkad (P, G),

d ω φ = d φ + [ω, φ].

Notice that the Corollary does not apply to ω itself, which is pseudotensorial
of type ad, but not horizontal. Indeed, the Cartan Structure Equation gives
d ω ω = Ω = d ω+ 12 [ω, ω]. However, the Corollary does apply to the curvature
Ω and gives d ω Ω = d Ω + [ω, Ω] which we now show is identically zero.
P
Theorem 4.5.9 (Bianchi Identity) Let G ,→ P −→ X be a smooth prin-
cipal bundle with connection ω and curvature Ω = d ω ω. Then

d ω Ω = 0. (4.5.7)
Proof: We simply compute, from Corollary 4.5.8,

d ω Ω = d Ω + [ω, Ω]
( ) [ ]
1 1
= d d ω + [ω, ω] + ω, d ω + [ω, ω]
2 2
1 1
= d (d ω) + d ([ω, ω]) + [ω, d ω] + [ω, [ω, ω]]
2 2
1
= d ([ω, ω]) + [ω, d ω]
2
1
= ([d ω, ω] − [ω, d ω]) + [ω, d ω]
2
1
= (−[ω, d ω] − [ω, d ω]) + [ω, d ω] = 0. 
2
Since [ω, Ω] = −[Ω, ω] one can write (4.5.7) as
d Ω = [Ω, ω]. (4.5.8)
232 4. Differential Forms and Integration

4.6 Integration on Manifolds

We begin by reviewing a little calculus. Consider a function f : R −→ R
which, for simplicity, we will temporarily assume to be continuous. If [a, b] is
an interval in R, then the integral
∫
f
[a,b]

of f over [a, b] is defined (either as a Riemann or Lebesgue integral) as a limit

of certain sums. It is invariably computed, however, by appealing to the Fun-
damental Theorem of Calculus which asserts that there exists a differentiable
function F : R −→ R with F ′ = f and that, for any such F ,
∫
f = F | ba = F (b) − F (a).
[a,b]

The task of finding such an antiderivative is often facilitated by the Change

of Variables Formula, the content of which is as follows: Let g be a diffeo-
morphism (actually, one-to-one and continuously differentiable will suffice) on
some open set containing an interval [α, β] with g([α, β]) = [a, b]. Then
∫ ∫
f= (f ◦ g) |g ′ |.
[a,b] [α,β]

In particular, if g is orientation preserving (increasing), then

∫ ∫
f= (f ◦ g) g ′ . (4.6.1)
[a,b] [α,β]

Calculus students are provided with ample opportunity to persuade them-

selves of the efficacy of this formula. Suppose, for example, that the problem
is to integrate over [0, 1] the function f whose standard coordinate repre-
sentation is f (x) = √1+x 1
2
. Define g : (− π2 , π2 ) −→ R by g = tan. Then
g([0, π4 ]) = [0, 1] and, on [0, π4 ], f ◦ g = cos and g ′ = sec2 so

∫ ∫ π ( √ )
4
f= sec = ln|sec + tan| = ln 1 + 2 .
[0,1] [0, π 0
4]

Of course, calculus students would generally not phrase their calculations

in just these terms. Perhaps something more along the following lines:
 4.6. Integration on Manifolds 233
∫ 1 ∫ tan−1 1
1 1
√ dx = √ sec2 u du
0 1 + x2 tan−1 0 1 + tan2 u
∫ π
x = tan u 4
= sec u du
dx = sec2 u du 0

π
4
= ln|sec u + tan u|
0
( √ )
= ln 1 + 2 .

There is much to be said for this. The student’s calculations, which, on the
surface, appear more formal and less rigorous, are, in fact, entirely rigorous
once one has faced the fact that it is not functions, but 1-forms that should
be integrated over intervals in R. Specifically, any f : R −→ R gives rise
to a unique 1-form ω = f dx on R and, if g is an orientation preserving
diffeomorphism with g([α, β]) = [a, b], then

g ∗ ω = g ∗ (f dx) = (f ◦ g) g ∗ (dx) = (f ◦ g) g ′ dx.

If we define the integral of any 1-form on R to be the integral of its standard

coordinate function, then the Change of Variables Formula becomes
∫ ∫
ω= g ∗ ω. (4.6.2)
[a,b] [α,β]

Our calculus student has computed (unwittingly, perhaps) the pullback of

1
the 1-form f (x) = √1+x 2
dx by g and found the interval corresponding to
[0, 1] under g. The Change of Variables Formula is built into the properties of
1-forms.
This is amusing enough, but scarcely sufficient justification for our claim
that “it is not functions, but 1-forms that should be integrated over inter-
vals in R .” Much more significant is the fact that the integral of a 1-form,
which we defined in terms of its standard coordinate representation, is actually
independent of coordinates and so will generalize immediately to arbitrary 1-
manifolds (where there is no “standard coordinate”). This too is a consequence
of (4.6.1), i.e., of the Change of Variables Formula. For the proof we regard R
as an abstract 1-manifold and consider two oriented charts (U, φ) and (V, ψ)
on R with coordinate functions x and y, respectively. To simplify the argu-
ment we will temporarily assume that [a, b] ⊆ U ∩ V . Write ω = h dx on
U and ω = k dy on V . The coordinate expressions for h and k are therefore
h ◦ φ−1 : φ(U ) −→ R and k ◦ ψ −1 : ψ(V ) −→ R (if (U, φ) is the standard
chart on R, then h ◦ φ−1 is the f in our earlier discussion). If we suppose that
234 4. Differential Forms and Integration

φ−1 ([α1 , β1 ]) = [a, b] and ψ −1 ([α2 , β2 ]) = [a, b], then our claim is that
∫ ∫
h ◦ φ−1 = k ◦ ψ −1
[α1 ,β1 ] [α2 ,β2 ]
∫
so that [a,b] ω could be defined as the integral of any of its coordinate ex-
pressions. To prove this we define

g = ψ ◦ φ−1 : φ (U ∩ V ) −→ ψ (U ∩ V ).

This is a diffeomorphism between open sets in R and

(k ◦ ψ −1 ) ◦ g = k ◦ φ−1 .

Furthermore, g([α1 , β1 ]) = [α2 , β2 ]. Now, on U ∩ V ,

( ) ( ) ( )
∂ ∂ ∂
h=ω = ω g′ = g′ ω = g′ k
∂x ∂y ∂y
so ( ) ( ) (( ) )
h ◦ φ−1 = g ′ k ◦ φ−1 = g ′ k ◦ ψ −1 ◦ g = k ◦ ψ −1 ◦ g g ′
and the Change of Variables Formula (4.6.1) gives
∫ ∫ ∫
(( ) )
k ◦ ψ −1 = k ◦ ψ −1 ◦ g g ′ = h ◦ φ−1
[α2 ,β2 ] [α1 ,β1 ] [α1 ,β1 ]

as required.
This we regard as rather persuasive evidence in favor of the view that it
is not functions, but 1-forms that one should integrate on 1-manifolds. There
is no standard coordinate system on a general 1-manifold and therefore no
standard coordinate expression for a function on a 1-manifold and, even in
R, it matters very much which coordinate expression one chooses to integrate
over a given subset. Coordinate expressions for 1-forms, on the other hand,
have just what it takes (according to (4.6.1)) to possess invariant integrals.
That all of this works out just as nicely for n-forms on n-manifolds will become
clear from a glance at Exercise 4.3.4 and the Change of Variables Formula for
Rn (sans a few hypotheses that we will add a bit later): Suppose U and V are
open in Rn and g : U −→ V is a diffeomorphism. If M ⊆ U and f : V −→ R,
then ∫ ∫
f= (f ◦ g) det(g ′ ) ,
g(M ) M
where g ′ is the Jacobian of g, i.e., the matrix of g∗ relative to standard coor-
dinates.
With this motivation behind us we set about building a theory of integration
for n-forms on n-dimensional manifolds. We will presume a familiarity with
the basics of Lebesgue integration on Rn .
 4.6. Integration on Manifolds 235

Remark: This decision to build upon the Lebesgue rather than the Riemann
integral smooths the theoretical development in the early stages, but has es-
sentially no practical impact on the explicit calculations we must perform. The
reader not versed in the Lebesgue theory has a number of options. Perhaps the
most sensible is to simply ignore all references to the subject and be assured
that, in the end, the integrals we must actually compute are accessible to the
tools available in the theory of the Riemann integral. Alternatively, one might
actually learn something about this indispensible part of modern mathemat-
ics. We will provide a brief synopsis, but everything we need (and more) is
covered very nicely and in fewer than fifty pages (specifically, pages 201–247)
in [Brow].
Throughout the remainder of this section X will denote a smooth, oriented,
n-dimensional manifold with n ≥ 1 and all charts will be assumed consistent
with the given orientation.
Recall that a subset A of Rn is said to have (Lebesgue) measure zero
in Rn if it can be covered by countable families of open rectangles with ar-
bitrarily small total volume. In more detail, an open rectangle in Rn is a set
of the form R = (a1 , b1 ) × · · · × (an , bn ), where ai < bi for i = 1, . . . , n, and
its volume is vol(R) = (b1 − a1 ) · · · (bn − an ). Then A ⊆ Rn has measure
zero if, for∪every ϵ > 0,∑there exists a family R1 , R2 , . . . of open rectangles
∞ ∞
with A ⊆ i=1 Ri and i=1 vol(Ri ) < ϵ. Subsets of sets of measure zero ob-
viously also have measure zero and countable unions of sets of measure zero
also have measure zero (cover the k th set with a family of rectangles of total
volume less than ϵ/2k ). It is also true, but not at all obvious, that smooth
images of sets of measure zero have measure zero, i.e., if U ⊆ Rn is open,
f : U −→ Rn is smooth and A ⊆ U has measure zero, then f (A) has measure
zero (Proposition 5–17 of [N1], or Lemma 1.1, Chapter 3, of [Hir]).
A subset A of a smooth manifold X is said to have measure zero in X if,
for every chart (U, φ) for X, the set φ(U ∩ A) has measure zero in Rn .
Exercise 4.6.1 Show that A has measure zero in X if and only if, for each
a ∈ A, there exists a chart (U, φ) at a for which φ(U ∩ A) has measure zero in
Rn . Hint: Keep in mind that our manifolds are assumed second countable.
If S is any set, then a collection A of subsets of S is called a σ-algebra if it
contains the empty set and is closed under the formation of complements and
countable
∪∞ unions, i.e., if (1) ∅ ∈ A, (2) S − A ∈ A whenever A ∈ A, and (3)
i=1 A i ∈ A whenever Ai ∈ A for i = 1, 2, . . . . Any collection C of subsets
of S is contained in various σ-algebras of subsets of S (e.g., the collection of
all subsets of S) and the intersection of all of these σ-algebras containing C is
itself a σ-algebra. This intersection is called the σ-algebra generated by C. In
any topological space S the σ-algebra generated by the collection C of closed
sets is called the Borel σ-algebra and its elements are called Borel sets in
S. Closed sets, open sets, countable unions of countable intersections of open
sets, etc., are all Borel sets.
236 4. Differential Forms and Integration

The collection of Borel sets in Rn is huge, but we need to enlarge the col-
lection still further. A subset M of Rn is said to be (Lebesgue) measurable
if it can be written as M = A ∪ B, where A ∩ B = ∅, A has measure zero in
Rn and B is a Borel set in Rn . Thus, measurable sets are those which differ
from a Borel set by a set of measure zero. Since the empty set has measure
zero, every Borel set is measurable. Since the empty set is a Borel set, every
set of measure zero is measurable.
Remark: Those of our readers who are “in the know” realize that this is
not the usual definition of Lebesgue measurable. That it is equivalent to the
usual definition follows from Theorem 9.29 of [Brow].
The collection M of Lebesgue measurable sets is itself a σ-algebra.
As the name suggests, the Lebesgue theory assigns to every M ∈ M a
“measure” m(M ) of its size. For example, m(R) = vol(R) for every rectan-
gle R and m(A) = 0 for any set A of measure zero. The ability to measure a
large collection of sets provided Lebesgue with the means to integrate a large
collection of functions. Roughly, his idea was as follows: Riemann integrated
a function f : [a, b] −→ R by partitioning [a, b] into subintervals, approximat-
ing f by a step function that was constant on each subinterval, defining the
integral of the step function in the only reasonable way and taking the limit
as the partition became finer and finer.

a b

This limit will not exist unless f is relatively nice, e.g., a bounded, real-valued
function on [a, b] must be continuous except perhaps on a set of measure zero
(Theorem 10.23 of [Brow]). This is unfortunate since, for example, limits of
even very nice functions are often not at all nice so the Riemann integral
does not react well to taking limits. Lebesgue’s idea was to partition, not the
 4.6. Integration on Manifolds 237

domain, but the range, approximate f by a step function that is constant on

the inverse image
∑ of each set in the partition, define the integral of the step
function by ci m(Mi ), where ci is the value on the inverse image Mi and
m(Mi ) is its measure and then take a limit as the partition becomes finer and
finer.

ci

Mi

For the record we will now briefly sketch the construction of the Lebesgue in-
tegral in somewhat more detail and formulate those definitions and results from
the theory that we will need. A real-valued function f on Rn is measurable
if the inverse image of every closed interval in R is Lebesgue measurable in
Rn (if f is an extended real-valued function, i.e., maps to R ∪ {±∞}, then it
is also required that f −1 (∞) and f −1 (−∞) be measurable). A (measurable)
step function is a function s that is zero except on a finite number of disjoint
measurable sets M1 , . . . , Mk with m(Mi ) < ∞ for each i = 1, . . . , k and which
takes a finite constant value ci on Mi for i = 1, . . . , k. The integral of such a
step function s over Rn is defined by
∫ ∑k
s dm = ci m(Mi ).
Rn i=1

One can show that if f is any non-negative measurable function on Rn , then

there exists a sequence 0 ≤ s1 ≤ s2 ≤ · · · of measurable step functions which
converges pointwise to f . The Lebesgue integral of f over Rn is defined
by ∫ ∫
f dm = sup s dm,
Rn Rn
238 4. Differential Forms and Integration

where the supremum is over all measurable step functions s with 0 ≤ s ≤ f .

This integral might well be infinite; if it is finite we say that f is Lebesgue
integrable (or summable) on Rn . If f is measurable, but not necessarily
non-negative, we write f = f + − f − , where f + and f − are measurable and
non-negative (e.g., f + = 12 (|f | + f ) and f − = 12 (|f | − f )) and say that f is
Lebesgue integrable on Rn if and only if f + and f − are both integrable.
In this case we define
∫ ∫ ∫
f dm = f dm −
+
f − dm.
Rn Rn Rn
It follows that f is Lebesgue integrable on Rn if and only if |f | is Lebesgue
integrable on Rn . Finally, suppose that M ⊆ Rn is measurable. Then its
characteristic function χM (1 on M and 0 on Rn −M ) is a measurable function.
If χM f is measurable (as it will be whenever f is measurable), then we define
∫ ∫
f dm = χM f dm.
M Rn
If this is finite (as it will be if f is integrable on Rn ), then we say that f is
Lebesgue integrable on M .
The Lebesgue integral has all of the properties one would expect of an
integral. The set of integrable functions is a linear space and the integral is a
linear functional on it:
∫ ∫ ∫
(af + bg)dm = a f dm + b g dm.
M M M

Sets of measure zero “don’t count,” i.e., if f is integrable and g agrees with
f except perhaps on a set of measure zero (we say that g equals f almost
everywhere and write g = f a.e.), then g is integrable and
∫ ∫
g dm = f dm (g = f a.e.).
M M

It follows that if M ′ differs from M by a set of measure zero and f is integrable

on M , then f is integrable on M ′ and
∫ ∫
f dm = f dm.
M′ M

Similarly, if M and N are measurable and M ∩ N is a set of measure zero,

then ∫ ∫ ∫
f dm = f dm + f dm (m(M ∩ N ) = 0).
M ∪N M N
The Change of Variables Formula takes exactly the same form as it does for
the Riemann integral: Let U and V be open sets in Rn , g : U −→ V a
 4.6. Integration on Manifolds 239

diffeomorphism, M ⊆ U measurable and f : V −→ R Lebesgue integrable.

Then g(M ) is measurable, (f ◦ g)| det(g ′ )| is integrable on M and
∫ ∫
f dm = (f ◦ g)| det(g ′ )|dm (4.6.3)
g(M ) M

(Theorem 10.46 of [Brow]). A very useful result for the purposes of calculation
is that if f is continuous on a closed∫ rectangle R̄ = [a1 , b1 ] × · · · × [an , bn ], then
it is Lebesgue integrable on R̄ and R̄ f dm agrees with the ordinary Riemann
integral of f over R̄. Consequently, the value of the integral can be calculated
as an iterated integral (Fubini’s Theorem) from the Fundamental Theorem of
Calculus:
∫ ∫ 1 (∫ 2 (
b b ∫ n b
) )
f dm = ··· f (x1 , x2 , . . . , xn ) dxn · · · dx2 dx1 .
R̄ a1 a2 an

Remark: Fubini’s Theorem becomes trickier when the continuity

assumption is weakened (see Sections 10.8 and 10.9 of [Brow]).
The most important properties of the Lebesgue integral, however, are those
related to limits. The so-called Monotone Convergence Theorem (Theorem
10.15 of [Brow]) asserts that if 0 ≤ f1 ≤ f2 ≤ · · · is a sequence of measurable
functions that converges pointwise (except perhaps on a set of measure zero)
to f , then f is measurable and
∫ ∫ ∫
lim fk dm = lim fk dm = f dm. (4.6.4)
k−→∞ Rn Rn k−→∞ Rn
The Lebesgue Dominated Convergence Theorem (Theorem 10.15 of [Brow])
asserts that if {fk } is a sequence of measurable functions that converges point-
wise (almost everywhere) to f and |fk | ≤ g for all k, where g is integrable,
then f is integrable and its integral is given by (4.6.4).
With this detour into Lebesgue integration behind us we are prepared to
build an integration theory on any smooth, oriented, n-dimensional man-
ifold X. We have already transferred the notion of measure zero to X
(Exercise 4.6.1). Although there is no reasonable way to move the Lebesgue
measure itself to X we can easily transfer the notion of measurability.
Indeed, X is a topological space and so has a σ-algebra of Borel sets so we may
say that a subset M of X is measurable if it can be written as M = A ∪ B,
where A ∩ B = ∅, A has measure zero in X and B is a Borel set in X.

Exercise 4.6.2 Show that M ⊆ X is measurable if and only if, for each
m ∈ M , there exists a chart (U, φ) at m for which φ(U ∩ M ) is Lebesgue
measurable in Rn . Hint: Use Exercise 4.6.1 and the fact that diffeomorphisms
on open sets in Rn preserve measure zero, Borel sets and disjoint unions.
240 4. Differential Forms and Integration

Now we consider an n-form ω on X (not necessarily smooth). We will

say that ω is measurable if, for every chart (U, φ) on X with coordinate
functions x1 , . . . , xn and ω = h dx1 ∧· · ·∧dxn on U , the coordinate expression
h ◦ φ−1 is Lebesgue measurable on φ(U ). We show now that it is enough to
check this condition for all charts in some atlas for X. Thus, we suppose that
{(Uα , φα )}α∈A is an atlas and that our condition has been verified for every
chart in this atlas. Let (V, ψ) be some other chart for X with coordinate
functions y 1 , . . . , y n and with ω = k dy 1 ∧ · · · ∧ dy n on V . We must show that
k ◦ ψ −1 is Lebesgue measurable on ψ(V ).
Exercise 4.6.3 Show that it will suffice to prove that k ◦ ψ −1 is Lebesgue
measurable on some neighborhood of each point of ψ(V ).

Any point in ψ(V ) is also in φα (Uα ) for some α ∈ A. If we let

g = φα ◦ ψ −1 : ψ(Uα ∩ V ) −→ φα (Uα ∩ V ), then g is a diffeomorphism
between open sets in Rn . If x1 , . . . , xn denote the coordinate functions for
(Uα , φα ) and we write ω = hα dx1 ∧ · · · ∧ dxn on Uα , then Exercise 4.3.4
gives, for every p ∈ Uα ∩ V ,
( i )
∂x ( )
k(p) = hα (p) det j
(p) = hα (p) det Dj (xi ◦ ψ −1 )(ψ(p)) .
∂y

Writing p = ψ −1 (y) for y ∈ ψ(Uα ∩ V ) this becomes

( ) ( )
k ◦ ψ −1 (y) = hα ◦ ψ −1 (y ) det Dj (xi ◦ ψ −1 )(y)
( ) ( )
= (hα ◦ φ−1
α ) ◦ g (y ) det Dj g (y )
i

so ( )
k ◦ ψ −1 = (hα ◦ φ−1 ′
α ) ◦ g det(g ) (4.6.5)
on ψ(Uα ∩ V ). Now, on open sets in R , smooth maps are measurable and
n

compositions and products of measurable functions are measurable. Since we

have assumed that hα ◦ φ−1 α is measurable it follows that k ◦ ψ
−1
is also
measurable.
If ω is any n-form on X we define its support supp ω to be the closure in
X of the set of points where ω is not the zero form:
supp ω = {x ∈ X : ω x ̸= 0}.

We define the integral of ω over X first in the case in which supp ω is compact
and contained in some coordinate neighborhood and then, in the general case,
use a partition of unity to define the integral as a sum of integrals of the first
type.
Suppose then that ω is a measurable n-form on X with compact support
supp ω ⊆ U , where (U, φ) is a chart on X (keep in mind that we consider
only charts consistent with the given orientation of X). If x1 , . . . , xn are the
coordinate functions of φ and ω = h dx1 ∧ · · · ∧ dxn on U , then the coordinate
expression h ◦ φ−1 is measurable on the open set φ(U ) in Rn . We say that
 4.6. Integration on Manifolds 241

ω is integrable on U (and also on X) if h ◦ φ−1 is integrable on φ(U ) (and

therefore on Rn ) and, in this case, we define
∫ ∫ ∫ ∫
ω= ω= h ◦ φ−1 dm = h o φ−1 dm. (4.6.6)
X U Rn φ(U )

Remark: Integrability will be assured, for example, if ω is continuous with

compact support supp ω ⊆ U .
Of course, we must prove that our definition does not depend on the choice
of the chart (U, φ) with supp ω ⊆ U . Thus we let (V, ψ) be another (oriented)
chart on X with supp ω ⊆ V . Let y 1 , . . . , y n be the coordinate functions for
ψ, and write ω = k dy 1 ∧ · · · ∧ dy n on V . We must show that
∫ ∫
h ◦ φ−1 dm = k ◦ ψ −1 dm.
φ(U ) ψ(V )

Since supp ω ⊆ U ∩ V, h ◦ φ−1 is zero outside of φ(U ∩ V ) and k ◦ ψ −1 is zero

outside of ψ(U ∩ V ) and we need only show that
∫ ∫
h ◦ φ−1 dm = k ◦ ψ −1 dm.
φ(U ∩V ) ψ(U ∩V )

Let g = φ ◦ ψ −1 : ψ(U ∩ V ) −→ φ(U ∩ V ). This is a diffeomorphism between

open sets in Rn and (4.6.5), without the subscript α, gives

k ◦ ψ −1 = ((h ◦ φ−1 ) ◦ g) det(g ′ ).

Now, since both (U, φ) and (V, ψ) are consistent with the orientation of X,
the Jacobian determinant det(g ′ ) is positive so we may, if we wish, write this
as
k ◦ ψ −1 = ((h ◦ φ−1 ) ◦ g) det(g ′ ) .

But now the Change of Variables Formula (4.6.3) with f = h ◦ φ−1 and
M = ψ(U ∩ V ) assures the integrability of k ◦ ψ −1 and gives
∫ ∫
h ◦ φ−1 dm = ((h ◦ φ−1 ) ◦ g) det(g ′ ) dm
φ(U ∩V ) ψ(U ∩V )
∫
= k ◦ ψ −1 dm
ψ(U ∩V )

as required.
Before proceeding with the general case we make a few observations about
this one. First notice that if ω 1 and ω 2 are two n-forms of the prescribed type
with supports both contained in U , then ω 1 + ω 2 is also of this type and
∫ ∫ ∫
ω1 + ω2 = ω1 + ω2 .
X X X
242 4. Differential Forms and Integration

Furthermore, if ω 1 and ω 2 agree except perhaps at a set of points of measure

zero in X, then ∫ ∫
ω1 = ω2 (ω 1 = ω 2 a.e.)
X X
Next suppose that ω is an n-form of the type we are considering and
let {ϕk }k=1,2,... be a partition of unity subordinate to an oriented atlas
for X. Since supp ω is compact and {supp ϕk }k=1,2,... is locally finite there
are only
∑∞finitely many values of k for which ϕk ω is not identically zero.
Thus, k=1 ϕk ω is actually a finite sum and
∞
(∞ )
∑ ∑
ϕk ω = ϕk ω = 1ω = ω.
k=1 k=1

Each ϕk ω is an n-form with compact support contained in a coordinate neigh-

borhood. If ω is integrable, then each ϕk ω is integrable and linearity implies
∫ ∞ ∫
∑
ω= ϕk ω. (4.6.7)
X k=1 X

Now let us turn to the general case and consider a measurable n-form ω
on X(supp ω need not be compact and need not be contained in a coordi-
nate neighborhood). Choose an oriented atlas for X and a partition of unity
{ϕk }k=1,2,... subordinate to it. Then each ϕk ω is an n-form with compact
support contained in ∑ a coordinate neighborhood. Since {supp ϕk }k=1,2,... is
∞
locally finite the sum k=1 ϕk ω is finite on some neighborhood of each point
so
∑∞
ω = ϕk ω
k=1

on X. We will say that ω is integrable on X if each ϕk ω is integrable on X

and if the series
∑∞ ∫
ϕk ω
k=1 X

converges absolutely. In this case we define

∫ ∞ ∫
∑
ω= ϕk ω.
X k=1 X

Note that if ω happens to have compact support contained in some coordi-

nate neighborhood, then we proved in (4.6.7) that this new definition agrees
with the old one. To show that the definition does not depend on the cho-
sen atlas or partition of unity we suppose {ϕ′j }j=1,2,... is a partition of unity
∑∞
subordinate to some other oriented atlas for X. Then ω = j=1 ϕ′j ω on X
(the sum is finite on some neighborhood of each point). Each of the n-forms
 4.6. Integration on Manifolds 243

ϕ′j ω has compact support contained in a coordinate neighborhood so we may

apply (4.6.7) to it ∫ ∞ ∫
∑
ϕ′j ω = ϕk ϕ′j ω.
X k=1 X

Similarly,
∫ ∞ ∫
∑
ϕk ω = ϕ′j ϕk ω.
X j=1 X

Thus,
∞ ∫
∑ ∞ ∫
∞ ∑
∑
ϕ′j ω= ϕk ϕ′j ω
j=1 X j=1 k=1 X

and ∞ ∫
∑ ∞ ∫
∞ ∑
∑
ϕk ω = ϕ′j ϕk ω. (4.6.8)
k=1 X k=1 j=1 X

∑∞ ∫
Exercise 4.6.4 Show that absolute convergence of k=1 X ϕk ω implies the
convergence of
∑∞ ∑∞ ∫
ϕ′j ϕk ω
k=1 j=1 X

and use this absolute convergence to interchange the order of summation in

(4.6.8) and obtain
∑∞ ∫ ∞ ∫
∑
ϕk ω = ϕ′j ω.
k=1 X j=1 X

∑∞ ∫
Finally, conclude that j=1 X ϕ′j ω converges absolutely.
Remark: If ω has compact support (not necessarily contained in a coordinate
neighborhood), then all of the sums dealt with above are finite so absolute con-
vergence is assured. In particular, this is always the case when X is compact.
To complete our sequence of definitions we let ω be a measurable n-form
on X and M ⊆ X a measurable set. We say that ω is integrable on M if
χM ω is integrable on X and in this case, we define
∫ ∫
ω= χM ω.
M X

Exercise 4.6.5 Establish each of the following properties.

(a) If ω 1 and ω 2 are integrable on X, then ω 1 + ω 2 is integrable on X and
∫ ∫ ∫
ω1 + ω2 = ω1 + ω2 .
X X X
244 4. Differential Forms and Integration

(b) If ω is integrable on X and a ∈ R, then aω is integrable on X and

∫ ∫
aω =a ω.
X X

(c) If ω 1 is integrable on X, and ω 2 = ω 1 a.e., then ω 2 is integrable on X

and ∫ ∫
ω2 = ω1 .
X X
′
(d) If ω is integrable on M and M differs from M by a set of measure zero,
then ω is integrable on M ′ and
∫ ∫
ω= ω.
M′ M

(e) If ω is integrable on M and M = M1 ∪ M2 , where M1 and M2 are

measurable and M1 ∩ M2 has measure zero, then ω is integrable on M1
and M2 and ∫ ∫ ∫
ω = ω+ ω.
M M1 M2

(f) If ω is integrable on X and U is an open submanifold X with inclusion

ι : U ,→ X, then ∫ ∫
ω = ι∗ ω.
U U
∫ ∫ ∫
Note: U ω = X χU ω, but, for U
ι∗ ω, U is regarded as a manifold
in its own right.
One last property of integrals that is particularly useful in calculations is
contained in the next result.
Theorem 4.6.1 Let X and Y be two smooth, oriented, n-dimensional man-
ifolds and f : X −→ Y an orientation preserving diffeomorphism of X onto
Y. Let ω be an integrable n-form on Y. Then f ∗ ω is an integrable n-form on
X and ∫ ∫
f ∗ω = ω.
X Y

Proof: Let {(Uα , φα )}α∈A be an oriented atlas for Y . Since f is an orientation

preserving diffeomorphism {(f −1 (Uα ), φα ◦ f )}α∈A is an oriented atlas for
X. Moreover, if {ϕk }k=1,2,... is a partition of unity subordinate to {Uα }α∈A ,
then {ϕk ◦ f }k=1,2,... is a partition of unity subordinate to {f −1 (Uα )}α∈A . By
assumption,
∫ ∑∞ ∫
ω= ϕk ω,
Y k=1 Y
 4.6. Integration on Manifolds 245

where the convergence on the right-hand side is absolute. We must show that
∞ ∫
∑ ∞ ∫
∑
∗
(ϕk ◦ f )f ω = ϕk ω
k=1 X k=1 Y

and that the convergence on the left-hand side is absolute. Both of these will
be proved if we show that
∫ ∫
∗
(ϕk ◦ f )f ω = ϕk ω
X Y
∗ ∗
for each k. Notice that (ϕk ◦ f )f ω = f (ϕk ω) so we must prove that
∫ ∫
f ∗ (ϕk ω) = ϕk ω.
X Y

Now, ϕk ω has compact support contained in some chart (Uα , φα ) for Y

so if we let x1 , . . . , xn be the coordinate functions of φα and write
ϕk ω = h dx1 ∧ · · · ∧ dxn on Uα we have
∫ ∫ ∫
ϕk ω = ϕk ω = h ◦ φ−1
α dm.
Y Uα φα (Uα )
−1
Moreover, on f (Uα ),
f ∗ (ϕk ω) = f ∗ (hdx1 ∧ · · · ∧ dxn ) = (h ◦ f )dy 1 ∧ · · · ∧ dy n ,

where y i = xi ◦ f, i = 1, . . . , n, are local coordinates on f −1 (Uα ). Since

f ∗ (ϕk ω) has compact support contained in f −1 (Uα ),
∫ ∫
f ∗ (ϕk ω) = f ∗ (ϕk ω)
X f −1 (Uα )
∫
= (h ◦ f ) ◦ (φα ◦ f )−1 dm
(φα ◦f )(f −1 (Uα ))

∫ ∫
−1
= h◦f ◦f ◦ φ−1
α dm = h ◦ φ−1
α dm
φα (Uα ) φα (Uα )
∫
= ϕk ω
Y

as required. 
Exercise 4.6.6 Show that, if f : X −→ Y is an orientation reversing diffeo-
morphism, then ∫ ∫
f ∗ω = − ω.
X Y
It’s time now to actually compute a few integrals. We recall (Theorem 4.3.2)
that any oriented Riemannian n-manifold X has defined on it a unique
246 4. Differential Forms and Integration

metric volume form ω with the property that ω x (e1 , . . . , en ) = 1 when-

ever {e1 , . . . , en } is an oriented, orthonormal basis for Tx (X). This n-form
is smooth and, if X is compact, it has compact support and therefore is inte-
grable over X.
Exercise 4.6.7 Show that, if ω is the ∫metric volume form on a compact,
oriented, Riemannian manifold X, then X ω > 0.
The value of the integral in Exercise 4.6.7 is called the volume of X. We will
compute the volumes of a few spheres. The volume form on S n is ι∗ ω̃, where
ω̃ is given by (4.3.2) and ι : S n ,→ Rn+1 is the inclusion. We begin with S 1 .
Remark: Scrupulous honesty in calculations such as these can result
in an avalanche of unnecessary notation. On the other hand, too cavalier
an attitude can leave one wondering where all of the machinery went with
which we built the integral. We will attempt to follow an intermediate path,
perhaps a bit too pedantic early on, but gaining bravado as we proceed.
The volume (length) of S 1 is (by (4.3.3)) given by
∫ ∫ ∫
∗
ω= ι ω̃ = ι∗ (x1 dx2 − x2 dx1 ).
S1 S1 S1

Since S 1 and S 1 − {(−1, 0)} differ by a set of measure zero,

∫ ∫
ω= ι∗ (x1 dx2 − x2 dx1 ).
S1 S 1 −{(−1,0)}

Remark: Technically, the integral on the right is

∫
χS 1 −{(−1,0)} ι∗ (x1 dx2 − x2 dx1 ),
S1

but we intend to appeal immediately to Exercise 4.6.5 (f) and regard it as

an integral over the open submanifold S 1 − {(−1, 0)} of S 1 . We should then
add another inclusion map S 1 − {(−1, 0)} ,→ S 1 , but we will permit ourselves
sufficient bravado even now to ignore this one. If this bothers you, by all means
put it in.
Now, define a map f : (−π, π) −→ S 1 − {(−1, 0)} by

(ι ◦ f )(θ) = (cos θ, sin θ),

where θ is the standard coordinate on (−π, π) ⊆ R.

Exercise 4.6.8 Show that f is an orientation preserving diffeomorphism.
Hint: Show first that it is a diffeomorphism. Then “orientation preserving”
need only be checked at one point. Now show that
( )
d ∂
(ι ◦ f )∗0 = .
dθ ∂x2
0 (1,0)
 4.6. Integration on Manifolds 247

According to Theorem 4.6.1 we therefore have

∫ ∫ ( )
∗ ∗ 1
ω= f ι (x dx − x dx )
2 2 1
S1 (−π,π)
∫
= (ι ◦ f )∗ (x1 dx2 − x2 dx1 ).
(−π,π)

Now,
(ι ◦ f )∗ (x1 dx2 − x2 dx1 ) = (x1 ◦ ι ◦ f ) d(x2 ◦ ι ◦ f )
− (x2 ◦ ι ◦ f ) d(x1 ◦ ι ◦ f )
= cos θ d(sin θ) − sin θ d(cos θ)
= cos2 θ dθ + sin2 θ dθ
= dθ

so ∫ ∫
ω = dθ.
S1 (−π,π)

Finally, dθ = 1 dθ is the standard coordinate expression for a continuous 1-

form on R and (−π, π) differs from [−π, π] by a set of measure zero so
∫ ∫ ∫
ω= 1 dθ = 1 dm = θ|π−π = 2π.
S1 [−π,π] [−π,π]

Exercise 4.6.9 Define h : R −→ S 1 by (ι ◦ h)(θ) = (cos θ, sin θ). Show that

if ω is any continuous 1-form on S 1 , then
∫ ∫
ω= h∗ ω
S1 [α,α+2π]

for any real number α.

The calculations for S 2 are analogous. By (4.3.4), the volume (area) of S 2
is given by
∫ ∫
ω= ι∗ (x1 dx2 ∧ dx3 − x2 dx1 ∧ dx3 + x3 dx1 ∧ dx2 ).
S2 S2

Define a map f from (0, π) × (−π, π) into S 2 by

(ι ◦ f ) (ϕ, θ) = (sin ϕ cos θ, sin ϕ sin θ, cos ϕ).

The image of f is the open submanifold of S 2 consisting of the complement

of the semicircle {(− sin ϕ, 0, cos ϕ) : 0 ≤ ϕ ≤ π} in S 2 . This semicircle has
measure zero so we may integrate ω over its complement to get the volume.
Exercise 4.6.10 Show that f is an orientation preserving diffeomorphism of
(0, π) × (−π, π) onto its image in S 2 .
248 4. Differential Forms and Integration

Thus, ∫ ∫
(
ω = (ι ◦ f )∗ x1 dx2 ∧ dx3
S2 (0,π)×(−π,π)
)
− x dx ∧ dx3 + x3 dx1 ∧ dx2 .
2 1

Now,
(ι ◦ f )∗ (x1 dx2 ∧ dx3 ) = sin ϕ cos θ (d(sin ϕ sin θ) ∧ d(cos ϕ))
= sin ϕ cos θ ((cos ϕ sin θ dϕ + sin ϕ cos θ dθ)
∧ (− sin ϕ dθ))
= sin3 ϕ cos2 θ dϕ ∧ dθ.

Exercise 4.6.11 Compute the remaining pullbacks and conclude that

∫ ∫
ω= sin ϕ dϕ ∧ dθ.
S2 (0,π)×(−π,π)

Now, sin ϕ dϕ ∧ dθ is the standard coordinate expression for a continuous 2-

form on R2 and (0, π)×(−π, π) differs from [0, π]×[−π, π] by a set of measure
zero so
∫ ∫ ∫ π ∫ π
ω= sin ϕ dm = sin ϕ dϕ dθ = 4π.
S2 [0,π]×[−π,π] −π 0

Exercise 4.6.12 Show that the volume of S 3 is 2π 2 . Hint: The appropriate

orientation preserving diffeomorphism this time is given by
(ι ◦ f )∗ (ξ, ϕ, θ) = (sin ξ sin ϕ cos θ, sin ξ sin ϕ sin θ, sin ξ cos ϕ, cos ξ),

where (ξ, ϕ, θ) ∈ (0, π) × (0, π) × (−π, π).

Exercise 4.6.13 Let ω be the closed, nonexact 1-form on R2 − {(0, 0)}
constructed from two polar coordinate functions θ1 and θ2 in Section 4.4
(page 252). Show that ∫
ω = 2π.
S1

Note: It will follow from Stokes’ Theorem (Section 4.7) that the integral of
ω over a circle in R2 − {(0, 0)} is 2π if the circle encloses the origin and 0 if
it does not.
We conclude this section with a calculation that yields a special case of
Stokes’ Theorem.
Theorem 4.6.2 Let X be a smooth, oriented, n-dimensional manifold and
suppose ω ∈ Λn−1 (X) is a smooth (n − 1)-form on X with compact support.
Then ∫
dω = 0.
X
 4.7. Stokes’ Theorem 249

Proof: Since dω is smooth and has compact support (because ω does), it

is integrable on X.
Exercise 4.6.14 Show that it is enough to prove the result for smooth
(n − 1)-forms with compact support contained in some coordinate neighbor-
hood U of X.
Thus, suppose (U, φ) is a chart with supp ω ⊆ U and let x1 , . . . , xn be its
coordinate functions. By composing with a diffeomorphism of Rn onto the
open unit ball in Rn , if necessary, we may assume φ(U ) is a bounded set in
∑ ci ∧ · · · ∧ dxn on U . Then
Rn . Write ω = ni=1 ωi dx1 ∧ · · · ∧ dx
( n )
∑
i−1 ∂ −1
dω = (−1) i
(ωi ◦ φ ) ◦ φ dx1 ∧ · · · ∧ dxn
i=1
∂x
on U so ( n )
∫ ∫ ∑ −1
i−1 ∂(ωi ◦ φ )
dω = (−1) dm
X φ(U ) i=1
∂xi
∑n ∫
i−1 ∂(ωi ◦ φ−1 )
= (−1) dm.
i=1 φ(U ) ∂xi

Take each ωi ◦ φ−1 to be zero outside φ(U ) and choose a cube C =

{(x1 , . . . , xn ) ∈ Rn : |xi | ≤ a, i = 1, . . . , n} containing φ(U ). Then
∫ ∑n ∫
∂(ωi ◦ φ−1 )
dω = (−1)i−1 dm.
X i=1 C ∂xi
∫ i ◦φ
−1
But each C ∂(ω∂x i
)
dm is zero because, by Fubini’s Theorem, it can be
computed as an iterated integral, integrating first with respect to xi and
ωi ◦ φ−1 vanishes on the boundary of the cube. 

4.7 Stokes’ Theorem

In this final section we prove a far-reaching and quite beautiful generalization
of the three classical integral theorems of vector calculus (Green’s Theorem,
Stokes’ Theorem and The Divergence Theorem). Each of these asserts the
equality of two integrals, one of them over an n-dimensional region D (n = 2, 3)
and the other over its (n − 1)-dimensional boundary ∂D. The Divergence
Theorem, for example, states that the outward flux of a smooth vector field
F⃗ on R3 through a closed surface ∂D equals the integral of the vector field’s
divergence div F⃗ over the region D that the surface bounds:
∫∫ ∫∫∫
F · N dS =
⃗ ⃗ div F⃗ dV.
∂D D
250 4. Differential Forms and Integration

With a glace back at Exercise 4.4.8 it may come as no surprise that these
theorems are really statements about (n − 1)-forms and their exterior deriva-
tives. Remarkably, these are all, in a sense, the same statement and it is our
objective here to derive the one simple, elegant equality from which they all
follow: ∫ ∫
∗
ι ω= dω. (4.7.1)
∂D D
The first order of business is to define precisely the appropriate regions of
integration.
Throughout this section X will denote a smooth, n-dimensional, oriented
manifold. A subset D of X will be called a domain with smooth boundary
in X if, for each p ∈ X, one of the following is true:
(1) There is an open neighborhood of p in X which is contained entirely in
X − D (such points are said to be in the exterior of D).
(2) There is an open neighborhood of p in X which is contained entirely in
D (these are called the interior points of D).
(3) There exists a chart (U, φ) for X at p with φ(p) = 0 ∈ Rn and φ(U ∩
D) = φ(U ) ∩ Rn+ , where Rn+ = {(x1 , . . . , xn ) ∈ Rn : xn ≥ 0} is the
closed upper half-space in Rn (these are called the boundary points
of D and the set of all such is denoted ∂D and called the boundary
of D).
Remark: These definitions make sense for any n ≥ 1, but, when n = 1,
∂D is a set of isolated points. Orientations for 0-dimensional manifolds and
integrals over such can all be defined in such a way that the results of this
section remain valid in this case, but we will have no need of them and so will
assume henceforth that n ≥ 2. Notice also that the set Int D of all interior
points of D is open in X and that there is nothing in the definition to preclude
the possibility that ∂D = ∅ (so X itself qualifies as a domain with smooth
boundary).
Exercise 4.7.1 Show that these three conditions are mutually exclusive and
that, for any chart (U, φ) of the type described in (3), φ(U ∩ ∂D) = {x ∈
φ (U ) : xn = 0}. Conclude (e.g., from Exercise 5.6.1, [N4]) that ∂D is an
(n − 1)-dimensional submanifold of X.
Exercise 4.7.2 Show that the n-dimensional disc Dn = {x ∈ Rn : ∥x∥ ≤ 1}
is a domain with smooth boundary in Rn and ∂Dn = S n−1 .
Exercise 4.7.3 Show that if D is a domain with smooth boundary in X,
then ∂D has measure zero in X and conclude that D is measurable in X.
Next we wish to show that, when D is a domain with smooth boundary in
X, the submanifold ∂D inherits a natural orientation from X.
Remark: This induced orientation on ∂D is “natural” in the somewhat
indirect sense that it supplies ∂D with the orientation it must have in order
 4.7. Stokes’ Theorem 251

for Stokes’ Theorem to be true. This sort of thing is to be expected, of course,

from the classical version of Stokes’ Theorem
∫ ∫∫
F⃗ · T⃗ ds = ⃗ · curl F⃗ dS
N
∂D
D

in which the orientation (tangent vector T⃗ ) to the closed curve ∂D must be

chosen consistent with the orientation (normal vector N ⃗ ) of the surface D.
Thus, for a given orientation of S 2 (outward normal, say), two applications of
2
Stokes’ Theorem, one to the upper hemisphere S+ and one to the lower hemi-
2
sphere S− , will necessitate opposite orientations on their common boundary
S 1 (counterclockwise and clockwise, respectively). The “natural” orientation
for ∂D depends on the rest of D.
The induced orientation on ∂D is defined at each p ∈ ∂D in terms of an
“outward” vector in Tp (X). Since ∂D is a submanifold of X we can identify
Tp (∂D) with a subspace of Tp (X) in the usual way. Choose a chart (U, φ)
at p in X, consistent with the orientation of X, and with φ(p) = 0 and
φ(U ∩ D) = φ(U ) ∩ Rn+ . Let x1 , . . . , xn−1 , xn be the coordinate functions
of φ. By Exercise 4.7.1, φ(U ∩ ∂D) = {x ∈ φ(U ) : xn = 0} is a coordinate
neighborhood at p in ∂D so Tp (∂D) is spanned by

∂ ∂
, . . . , n−1 .
∂x1 ∂x
p p

i |p in Tp (X) with a ̸ 0 is not in Tp (∂D) and

∂
A tangent vector u = ai ∂x n
=
we will say that u is outward pointing if a < 0 (think about D = Rn+ in
n

X = Rn ).
Exercise 4.7.4 Show that this definition does not depend on the choicϵ of
(U, φ), i.e., that if (V, ψ) is another chart at p in X, consistent with the
orientation of X, and with ψ(p) = 0 and ψ(V ∩ D) = ψ(V ) ∩ Rn+ and u =
i |p , where y , . . . , y
∂
bi ∂y 1 n−1
, y n are the coordinate functions of ψ, then an and
n
b have the same sign.
Now we define the induced orientation for Tp (∂D) by decreeing that a basis
{v 1 , . . . , v n−1 } for Tp (∂D) is in this orientation if and only if {v , v 1 , . . . , v n−1 }
is an oriented basis for Tp (X) for any outward pointing tangent vector
u ∈ Tp (X).

Exercise 4.7.5 Show that the induced orientation for Tp (∂D) is well-defined,
i.e., does not depend on the choice of outward pointing vector u ∈ Tp (X).

Exercise 4.7.6 Let X = Rn with its standard orientation and D = Rn+ so

that ∂D ∼ = Rn−1 . Show that, at each p ∈ ∂D, the induced orientation for
Tp (∂D) is the standard orientation of Rn−1 if and only if n is even.
252 4. Differential Forms and Integration

Exercise 4.7.7 Let X = S n (with either orientation), D1 = S+ n

= {(x1 , . . . ,
x n+1
)∈S :x
n n+1
≥ 0} and D2 = S− = {(x , . . . , x
n 1 n+1
) ∈ S : xn+1 ≤ 0}.
n

Then ∂D1 = S∼ n−1 ∼

and ∂D2 = S n−1
. Show that, at each p ∈ S n−1 , the
induced orientations on ∂D1 and ∂D2 are opposite.
Now we show that there is a global orientation (nonzero (n − 1)-form) on
∂D that determines the induced orientation on each Tp (∂D). First we define a
smooth vector field V on an open neighborhood of ∂D in X which, at each p ∈
∂D, is outward pointing. This is easy to do locally: For each p ∈ ∂D select an
oriented chart (Up , φp ) for X at p with φp (p) = 0 and φp (Up ∩ D) = φp (Up ) ∩
Rn+ . If x1 , . . . , xn−1 , xn are the corresponding coordinate functions, then
∂
Vp = −
∂xn
is a smooth vector field on Up that is outward pointing at each point of
Up ∩ ∂D. Then {Up : p ∈ ∂D} covers ∂D so we can select a count-
able subcover {Un }n=1,2,... and a partition of unity {ϕn }n=1,2,... with
supp ϕn ⊆ Un for each n = 1, 2, . . . . If V n∪is the outward pointing vector
∞
field on Un ∩ ∂D, then we define V on U = n=1 Un by
∞
∑
V = ϕn V n .
n=1

This sum is locally finite and therefore defines a smooth vector field on U ⊇ ∂D
which is outward pointing on ∂D since 0 ≤ ϕn ≤ 1.
Now, let ω be a nonzero n-form on X that determines the orientation of
X (i.e., ω p (v 1 , . . . , v n−1 , v n ) > 0 whenever p ∈ X and { v 1 , . . . , v n−1 , v n }
is an oriented basis for Tp (X)). Regard ω as a C ∞ (X)-multilinear map on
n
X (X) × · · · × X (X). The restriction (pullback) of ω to U , which we will
n
also denote ω, operates on X (U ) × · · · × X (U ). Thus, we may define an
(n − 1)-form ω̃ on U by

ω̃ (V 1 , . . . ,V n−1 ) = ω (V , V 1 , . . . ,V n−1 ),

where V i ∈ X (U ) for i = 1, . . . , n − 1 and V is the vector field on U defined

above that is outward pointing on ∂D. For p ∈ ∂D and { v 1 , . . . , v n−1 } a
basis for Tp (∂D), ω̃ p (v 1 , . . . , v n−1 ) > 0 if and only if { v 1 , . . . , v n−1 } is in the
induced orientation for Tp (∂D). The (n − 1)-form ω̃ determines the induced
orientation on ∂D.
Theorem 4.7.1 (Stokes’ Theorem) Let X be a smooth, oriented,
n-dimensional manifold, D a domain with smooth boundary in X and
ι : ∂D ,→ X the inclusion map. Provide ∂D with the induced orientation.
If ω is any smooth (n − 1)-form on X with compact support on X, then
∫ ∫
ι∗ ω = dω. (4.7.2)
∂D D
 4.7. Stokes’ Theorem 253

Proof: Observe first that, since ω is smooth with compact support, the
same is true of ι∗ ω and d ω so both of these are integrable. Now, choose a
countable, oriented atlas { (Uk , φk ) }k=1,2,... for X such that each φk (Uk ) is a
bounded subset of Rn and either Uk ∩∂D = ∅ (if Uk ∩D ̸= ∅ we assume in this
case that Uk ⊆ Int D), or, if Uk ∩∂D ̸= ∅, then φk (Uk ∩D) = φk (Uk )∩ Rn+ . Let
{ ϕk }k=1,2,... be a family of real-valued
∑∞ functions on X of the sort guaranteed
by Corollary 3.1.5. Then ω = ∑∞
k=1 ϕk ω, where the sum has ∑only finitely
∞
many nonzero terms. Thus d ω = k=1 d(ϕk ω) and ι∗ ω = k=1 ι∗ (ϕk ω)
are finite sums, as are each of the following:
∫ ∞ ∫
∑
∗
ι ω= ι∗ (ϕk ω)
∂D k=1 ∂D
∫ ∞ ∫
∑
dω = d(ϕk ω).
D k=1 D

Thus, it will suffice to prove that

∫ ∫
∗
ι (ϕk ω) = d (ϕk ω)
∂D D

for each k and this simply amounts to proving (4.7.2) in the special case in
which supp ω is contained in some Uk .
Assume then ∫that supp ω ⊆ Uk and consider first the
∫ case in which Uk ∩
∂D = ∅. Then ∂D ι∗ ω = 0 and we must show that D d ω is zero as well.
This is obvious if Uk ∩ D = ∅ since supp dω ⊆ supp ω ⊆ Uk . If Uk ∩ D ̸= ∅,
then Uk ⊆ Int D so
∫ ∫ ∫
dω = dω = dω
D Uk X

and this is zero by Theorem 4.6.2.

Finally, we consider the case in which supp ω ⊆ Uk and Uk ∩ ∂D ̸= ∅. Thus,
φk (Uk ∩ D) = φk (Uk ) ∩ Rn+ . Let x1 , . . . , xn be the coordinate functions of φk
and write
∑n
ω = (−1)i−1 ωi dx1 ∧ · · · ∧ dx ci ∧ · · · ∧ dxn
i=1

on Uk (including the (−1)i−1 here will spare us a few of these down the road).
Thus,
( n )
∑ ∂
−1
dω = i
(ωi ◦ φk ) ◦ φk dx1 ∧ · · · ∧ dxn (4.7.3)
i=1
∂x

on Uk , and, since ι∗ (dxn ) = d(xn ◦ ι) = 0 by Exercise 4.7.1,

254 4. Differential Forms and Integration

ι∗ ω = (−1)n−1 (ωn ◦ ι) ι∗ (dx1 ∧ · · · ∧ dxn−1 )

ι∗ ω = (−1)n−1 (ωn ◦ ι) d (x1 ◦ ι) ∧ · · · ∧ d(xn−1 ◦ ι) (4.7.4)

on Uk ∩ ∂D. Now, the map φ : Uk ∩ ∂D −→ Rn−1 with coordinate functions

(x1 ◦ ι, . . . , xn−1 ◦ ι) is a chart on ∂D.
Exercise 4.7.8 Show that the orientation d(x1 ◦ ι) ∧ · · · ∧ d(xn−1 ◦ ι) on
Uk ∩ ∂D is (−1)n times the orientation induced by dx1 ∧ · · · ∧ dxn−1 ∧ dxn on
Uk . Hint: dx1 ∧ · · · ∧ dxn−1 ∧ dxn = (−1)n (−dxn ) ∧ dx1 ∧ · · · ∧ dxn−1 .
Since ι∗ ω has compact support contained in Uk ∩ ∂D, Exercise 4.7.8 and
(4.7.4) give ∫ ∫
ι∗ ω = (−1)n (−1)n−1 (ωn ◦ ι) ◦ φ−1 dm′
∂D φ(Uk ∩∂D)
∫ (4.7.5)
−1 ′
=− ωn ◦ (ι ◦ φ )dm
φ(Uk ∩∂D)

where m′ denotes Lebesgue measure on Rn−1 . Notice that ι ◦ φ−1 is the φ-

coordinate expression for the inclusion map which, in φk -coordinates, is just
(x1 , . . . , xn−1 ) −→ (x1 , . . . , xn−1 , 0). Thus, (ωn ◦ (ι ◦ φ−1 ))(x1 , . . . , xn−1 ) =
(ωn ◦ φ−1 )(x1 , . . . , xn−1 , 0) and, since ι∗ ω has compact support contained in
Uk ∩ ∂D, (4.7.5) can be written
∫ ∫
ι∗ ω = (ωn ◦ φ−1 1
k )(x , . . . , x
n−1
, 0) dx1 · · · dxn−1 . (4.7.6)
∂D Rn−1
Now we turn to the integral over D of d ω. By (4.7.3), the fact that φk (Uk ∩
D) = φk (Uk ) ∩ Rn+ . and our assumption that ω (and therefore d ω) has
compact support contained in Uk , this is given by
∫ ∑ n ∫
∂(ωi ◦ φ−1
k )
dω = χRn+ dm ,
i=1 R
n ∂x i
D

where χRn+ is the characteristic function of Rn+ , i.e., 1 if xn ≥ 0 and 0 if

xn < 0, and, as in the proof of Theorem 4.6.2, we have taken ωi ◦ φ−1 k to be
zero outside of φ−1
k (Uk ). Also as in the proof of Theorem 4.6.2, we now choose
a cube C = { (x1 , . . . , xn ) ∈ Rn : |xi | ≤ a, i = 1, . . . , n } sufficiently large to
contain φk (Uk ). Thus,
∫ ∑ n ∫
∂(ωi ◦ φ−1k )
dω = χRn+ i
dm. (4.7.7)
D i=1 C
∂x

Exercise 4.7.9 Argue as in the proof of Theorem 4.6.2 to show that, for
i = 1, . . . , n − 1 (but not n),
∫
∂(ωi ◦ φ−1
k )
χRn+ i
dm = 0 (4.7.8)
C ∂x
 4.7. Stokes’ Theorem 255

and conclude that

∫ ∫
∂(ωn ◦ φ−1
k )
dω = χRn+ n
dm . (4.7.9)
D C ∂x

The argument used to establish (4.7.8) works because, if only

xi (i = 1, . . . , n − 1) is varying in φk (Uk ∩ D), then χRn+ is equal to 1 every-
where. In the xn -direction, of course, multiplying by χRn+ will yield a function
that is zero on (−∞, 0) and unchanged on [0, ∞). Thus, when we express the
right-hand side of (4.7.9) as an iterated integral, integrating first with respect
to xn , the integrand becomes
∫ a
∂(ωn ◦ φ−1
k )
n
dxn = (ωn ◦ φ−1 1
k ) (x , . . . , x
n−1
, a)
0 ∂x
− (ωn ◦ φ−1 1
k ) (x , . . . , x
n−1
, 0)
= −(ωn ◦ φ−1 1
k ) (x , . . . , x
n−1
, 0)

so
∫ ∫ ∫
dω = − (ωn ◦ φ−1 1
k ) (x , . . . , x
n−1
, 0) dx1 · · · dxn−1 = ι∗ ω
D R n−1 ∂D

as required. 
We conclude with an analogue of the familar principle of deformation of
paths for line integrals in vector calculus. Let X be a smooth, oriented
n-dimensional manifold and let N1 and N2 be two oriented k-dimensional
submanifolds, where 1 ≤ k ≤ n. We say that N1 can be smoothly deformed
into N2 if there exists a smooth map H : N × (0, 3) −→ X such that
H|N × {1} −→ N1 and H|N × {2} −→ N2 are orientation preserving diffeo-
morphisms. We claim that, if ω is any closed k-form on X and ι1 : N1 ,→ X
and ι2 : N2 ,→ X are the inclusion maps, then
∫ ∫
ι∗1 ω = ι∗2 ω . (4.7.10)
N1 N2

Exercise 4.7.10 Prove (4.7.10). Hint: H ∗ ω is a closed k-form on the (k +1)-

manifold N × (0, 3) and N × [0, 2] is a domain with smooth boundary in
N × (0, 3).
5
de Rham Cohomology

Introduction
The plane R2 and the punctured plane R2 − {(0, 0)} are not diffeomorphic,
nor even homeomorphic. There are various means by which one can prove
this, but the most instructive among these detect the “hole” in the punctured
plane (and none in R2 ) by distinguishing topologically certain “types” of cir-
cles that can live in the two spaces. One way of formalizing this idea is to show
that R2 and R2 − { (0, 0) } have different fundamental groups (see pages 107
and 133 of [N4]). Fundamental groups (and their higher dimensional gener-
alizations) are notoriously difficult to calculate, however. In this chapter we
will investigate another method of attaching to any smooth manifold certain
algebraic objects which likewise “detect holes”. The idea is quite simple and,
for R2 and R2 − { (0, 0) }, has already been hinted at in Section 4.4. There we
found that any closed 1-form on R2 is necessarily exact (Poincaré Lemma),
but constructed a closed 1-form ω on R2 − { (0, 0) } that is not exact. Stokes’
Theorem implies that the integral of any 1-form on R2 over any circle (com-
pact, connected, 1-dimensional submanifold) is zero and the same is true of
ω provided the circle does not contain (0, 0) in its interior. For a circle S 1
that does contain (0, 0) in its ∫interior (and has the orientation induced from
R2 − { (0, 0) }) we computed S 1 ω and obtained 2π. The idea then is that
the closed, nonexact 1-form ω detects the hole in R2 − { (0, 0) } via its inte-
grals over circles. The prospect then arises that one might distill topological
information from the existence of closed, nonexact forms on a manifold. Un-
fortunately, the collection of all such forms (even of some fixed degree) can be
a huge, unmanageable object. Observe, however, that another closed 1-form
ω ′ on R2 − { (0, 0) } that differs from ω by an exact form (ω ′ = ω + dη)
has, by Stokes’ Theorem, the same integral as ω over any circle and so for
our purposes, there is no reason to distinguish them. The natural thing to do
then is define an equivalence relation on the set of closed 1-forms whereby two
such forms are equivalent (we shall say “cohomologous”) if they differ by an
exact form. The resulting set of equivalence classes inherits a natural vector
space structure from the space of 1-forms and, with this structure, is called
the 1st de Rham cohomology group of the manifold. For R2 it has dimension 0
(no holes), while for R2 − { (0, 0) } it has dimension 1 (one hole). The precise
construction of this vector space and its higher dimensional analogues is the
task we set for ourselves in the next section. There are many formal similari-
ties with homology theory (Chapter 3 of [N4]), but the arguments tend to be
substantially easier.

G.L. Naber, Topology, Geometry and Gauge fields: Interactions, 257

Applied Mathematical Sciences 141, DOI 10.1007/978-1-4419-7895-0_5,
© Springer Science+Business Media, LLC 2011
258 5. de Rham Cohomology

5.1 The de Rham Cohomology Groups

Throughout this section X will denote an n-dimensional smooth manifold
and, for each integer k ≥ 1, Λk (X) is its vector space of smooth, real-valued
k-forms. Then Λk (X) = 0 for k ≥ n+1. For convenience, we will take Λ0 (X) =
C ∞ (X) and Λk (X) = 0 for k < 0. The exterior differentiation maps
dk : Λk (X) −→ Λk+1 (X), k = 0, 1, . . . , n
will, unless particular emphasis is required, all be written d and we will take
dk to be the trivial homomorphism when k < 0 and k ≥ n. Thus, we have a
sequence

· · · −→ Λ−1 (X) −→ Λ0 (X) −→ Λ1 (X) −→ · · ·

d d d d

d d d
−→ Λn−1 (X) −→ Λn (X) −→ · · ·

of vector spaces and linear maps. Since dk ◦ dk−1 = 0 for each k

(Theorem 4.1.1 (3)) we have
Image (dk−1 ) ⊆ ker (dk ).
The elements of ker (dk ) are the closed k-forms, or de Rham k -cocycles, on
X, while the elements of Image (dk−1 ) are the exact k-forms, or de Rham
k -coboundaries, on X. Both Image (dk−1 ) and ker (dk ) are linear subspaces
of Λk (X) and so Image (dk−1 ) is a subspace of ker (dk ). The quotient vector
space
Hdek R (X) = ker (dk ) / Image (dk−1 )
is called the k th de Rham cohomology group of X.
Remark: Hdek R (X) is a vector space, but the tradition of referring to it as
a cohomology group is so firmly embedded in the literature that any attempt
to correct the terminology here would be pointless.
The elements of Hdek R (X) are equivalence classes of closed k-forms on X
where the equivalence relation is defined as follows: If ω and ω ′ are closed
k-forms on X, then ω ′ ∼ ω if and only if ω ′ − ω = dη for some η ∈ Λk−1 (X)
(in this case we say that ω ′ and ω are cohomologous). The equivalence class
containing ω will be denoted [ω] and called the cohomology class of ω.
Since Hdek R (X) = 0 for k ≥ n + 1 and for k ≤ −1, the only (possibly)
nontrivial cohomology groups are
Hde0 R (X), Hde1 R (X), . . . , Hden R (X) (n = dim X).
Hde0 R (X) is easily calculated. Since Image (d−1 ) = 0, Hde0 R (X) = ker (d0 )
and this is the set of all f ∈ Λ0 (X) = C ∞ (X) for which df = 0.
Exercise 5.1.1 Show that df = 0 if and only if f is constant on each con-
nected component of X.
 5.1. The de Rham Cohomology Groups 259

In particular, if X is connected, then Hde0 R (X) is just the subspace of C ∞ (X)

consisting of constant functions and this is clearly isomorphic to R. If X
is not connected, then (since we assume all manifolds are second count-
able) it has at most countably many connected components C1 , C2 , . . .. Each
0
characteristic
∑∞function χCi is an element of Hde R (X) as is each (locally fi-
nite) sum i=1 ri χCi , where ri ∈ R. Indeed, every element of Hde0 R (X) is
such a sum. Thus, if X has only finitely many components (say, l), then
Hde0 R (X) ∼
= Rl . Otherwise, Hde0 R (X) is the infinite dimensional vector space
R × R × · · · = Rℵ0 . Combining these observations with the Poincaré Lemma
yields our first nontrivial result on cohomology groups.
Theorem 5.1.1 Let X be an open, star-shaped subset of Rn (e.g., Rn itself).
Then Hde0 R (X) ∼
= R and Hdek R (X) = 0 for all k ̸= 0.
Remark: A star-shaped subset of Rn is contractible, but the converse is
not true. We will, however, eventually show that any contractible manifold X
has Hdek R (X) = 0 for all k ̸= 0.
Consider a compact, connected, oriented n-manifold X. One of our major
results in this chapter is that Hden R (X) is 1-dimensional, i.e., isomorphic to
R. This will allow us, in particular, to define the notion of the degree of
a map between such manifolds. For the time being we will content ourselves
with showing that Hden R (X) cannot be trivial (keeping in mind the previous
Remark, this alone will prove that no compact, connected, orientable manifold
of dimension n > 0 is contractible). To find a nontrivial cohomology class
in Hden R (X) it will suffice to find a closed n-form on X that is not ∫ exact.
According to Stokes’ Theorem any exact n-form ω on X satisfies X ω = 0
and so we need only find a closed n-form on X whose integral over X is not
zero. But any n-form on an n-manifold is closed because Λn+1 (X) = 0 so we
need only find some n-form on X which does not integrate to zero over X.
Exercise
∫ 5.1.2 Let µ be an n-form determining the orientation of X. Show
that X µ ̸= 0 and conclude that the cohomology class [µ] ∈ Hden R (X) is
nontrivial.
Noting that the argument we have just given does not require connectivity,
we record the following lemma.
Lemma 5.1.2 Let X be a compact, orientable, n-manifold. Then
dim Hden R (X) ≥ 1.

The efficient calculation of cohomology groups will require that we first deal
with some of the more formal aspects of the subject. There is, however, at least
one small example accessible to us at this stage. We know that Hde0 R (S 1 ) = R
and Hdek R (S 1 ) = 0 for k ≥ 2 and k ≤ −1. Furthermore, Hde1 R (S 1 ) has
dimension at least one by Lemma 5.1.2. We show that, in fact,

Hde1 R (S 1 ) ∼
= R. (5.1.1)
260 5. de Rham Cohomology

Remark: We will show later that S 1 is a (smooth) deformation

retract of R2 − {(0, 0)} and that this implies that they have the same de
Rham cohomology groups. Consequently, it will follow from (5.1.1) that
Hdek R (R2 − {(0, 0)}) ∼
= R as we claimed in the Introduction.

Since we know from Lemma 5.1.2 that dim Hde1 R (S 1 ) ≥ 1 we need only prove
that dim Hde1 R (S 1 ) ≤ 1 in order to establish (5.1.1). For this it will suf-
fice to find a nontrivial cohomology class [η] ∈ Hde1 R (S 1 ) such that, for
any 1-form α on S 1 , there exists a constant α for which α − αη is exact
(since then [α] = [αη] = α[η]). For this we let ι : S 1 ,→ R2 − { (0, 0) }
be the inclusion of S 1 into the punctured plane and define η = ι∗ ω, where
ω is the 1-form on R2 − { (0, 0) } defined (in Section 4.4) from two angular
coordinate functions on R2 − { (0, 0) }. Specifically,
y x
ω=− dx + 2 dy
x2 +y 2 x + y2
1
so
∫ η is just the standard volume form1 on S1 . In Section 4.6 we found that
S1
η = 2π so, in particular, [η] ∈ Hde R (S ) is nontrivial. Now let α be an
arbitrary 1-form on S 1 and set
∫
1
α= α.
2π S 1

We claim that α − αη is exact. Of course,

∫
(α − αη) = 0
S1
so our result will follow from the next lemma.
∫
Lemma 5.1.3 A 1-form ν on S 1 is exact if and only if S1
ν = 0.
Proof: We have already ∫ seen that any exact form integrates to zero so
suppose conversely that S 1 ν = 0. Define h : R → S 1 by h(t) = (cos t, sin t)
and g : R → R by ∫ t
g(t) = h∗ ν.
0
∫
We have shown in Section 4.6 that S 1 ν is equal to the integral of h∗ ν over
any interval of length 2π so our assumption implies that

g(t + 2π) = g(t)

for any t ∈ R. Consequently, there is a well-defined smooth function

f : S 1 → R satisfying
g =f ◦h
(for each p ∈ S 1 , f (p) = g(t), where t is any real number with h(t) = p).
 5.1. The de Rham Cohomology Groups 261

Exercise 5.1.3 Show that df = ν. 

n
Remark: We will show ∫ later (Theorem 5.4.3) that an n-form ν on S
is exact if and only if S n ν = 0 and, later still (Corollary 5.5.5), that an
n-form
∫ ν on any compact, orientable n-manifold X is exact if and only if
X
ν = 0.
To develop efficient techniques for calculating the cohomology groups of other
manifolds we must delve more deeply into some of the more formal, algebraic
aspects of the subject. We will take this up in the next few sections. Before
doing so, however, we must point out that in the next chapter we will find
ourselves briefly in need of a rather obvious generalization of de Rham groups
as constructed here. Our objective in Chapter 6 is to associate with each
P
principal bundle G ,→ P → X certain cohomology classes of the base manifold
X called the Chern classes of the bundle. These are the cohomology classes
of certain closed forms constructed from a connection on the bundle. The
construction is simple enough, but a priori it would seem to yield complex,
rather than real forms. One can show that they are, in fact, real, but until this
is done one must carry out the development within the context of the “complex
de Rham cohomology groups Hdek R (X; C).” These are constructed in precisely
the same way as the groups Hdek R (X) except that one begins with complex-
valued forms Λk (X; C) and ends up with complex vector spaces. We leave it
to the reader to write out the details of the construction and to generalize
all of the algebraic machinery we will assemble. Here’s your first opportunity.
⊕∞
We shall denote by Hde∗ R (X) the direct sum Hdek R (X) of the cohomology
k=0
groups of X (of course, Hdek R (X) ∼
= 0 for k > dim X).
Remark: Recall that if V0 , V1 , . . . are vector spaces, then their direct prod-
∏
∞
uct Vk is the Cartesian product V0 × V1 × · · · with its coordinatewise
k=0
⊕
∞ ∏
∞
linear structure, while their direct sum Vk is the subspace of Vk con-
k=0 k=0
sisting of those elements with at most finitely many coordinates nonzero. Each
⊕
∞
Vk can then be canonically identified with a subspace of Vk and this di-
k=0
rect sum can then be viewed as the internal direct sum of these subspaces.
⊕
∞
In this way we view the elements of Vk as finite “formal sums” of ele-
k=0
ments from the various Vk , k = 0, 1, . . .. The elements of Hde∗ R (X) are there-
fore to be thought of as finite formal sums of cohomology classes of various
degrees.
Exercise 5.1.4 Show that if [ω 1 ] ∈ Hdek R (X) and [ω 2 ] ∈ Hdel R (X), then

[ω 1 ] ∧ [ω 2 ] = [ω 1 ∧ ω 2 ]
262 5. de Rham Cohomology

is a well-defined element of Hdek+l R (X). Now extend the product ∧ to all of

Hde∗ R (X) (by the “distributive law”) and show that Hde∗ R (X) thereby acquires
the structure of an algebra with identity.
In the same way one constructs the complex de Rham cohomology algebra
H ∗ (X; C).

5.2 Induced Homomorphisms

Let X and Y be smooth manifolds and f : X → Y a smooth map. For each
k, the pullback f ∗ : Λk (Y ) → Λk (X) carries closed forms to closed forms
(dω = 0 implies d(f ∗ ω) = f ∗ (d ω) = f ∗ (0) = 0) and exact forms to exact
forms (ω = d η implies f ∗ ω = f ∗ (d η) = d(f ∗ η)). Thus, we may define a
linear map
f # : Hdek R (Y ) −→ Hdek R (X)
by
f # ([ω]) = [f ∗ ω] (5.2.1)
for each [ω] ∈ Hdek R (Y). Notice that this definition makes sense because
f ∗ ω is closed and is independent of the choice of representative ω of the
cohomology class [ω] because ω ′ − ω = dη implies f ∗ ω ′ − f ∗ ω = d(f ∗ η).
Observe also that if Y = X and f = idX , then f # is the identity on each
Hdek R (X), i.e.,
(idX )# = idHdek R (X) (5.2.2)
for each k. Moreover, from the corresponding property of pullbacks we find
that, if f : X → Y and g : Y → Z are smooth, then

(g ◦ f )# = f # ◦ g # . (5.2.3)

In particular, if f : X → Y and g : Y → X are inverse diffeomorphisms, then

f # : Hdek R (Y ) −→ Hdek R (X) and g # : Hdek R (X) −→ Hdek R (Y )

are inverse isomorphisms for each k. Consequently, the de Rham cohomology

groups are diffeomorphism invariants.

Remark: The de Rham cohomology groups are actually invariants of ho-

motopy type, but this is by no means clear (see the Remark following Corol-
lary 5.2.4).
The maps f # are said to be those induced in cohomology by f . Properties
(5.2.2) and (5.2.3) of these induced maps should be compared with the anal-
ogous properties of induced maps in homotopy groups (Theorems 2.2.5 and
2.5.4 of [N4]) and singular homology groups (Section 3.2 of [N4]). They differ
only in the direction of the arrows and express the “functorial” nature of the
 5.2. Induced Homomorphisms 263

construction of the de Rham cohomology groups (in the jargon of category

theory the direction of the arrows is accounted for by referring to the ho-
motopy and homology functors as covariant and the cohomology functors as
contravariant).
Given two smooth maps f, g : X → Y we will be interested in determining
whether or not they induce the same linear maps in cohomology (i.e., f # = g #
for each k). A common method of proving that these induced maps are the
same is based on an idea that we encountered in the proof of the Poincaré
Lemma (Theorem 4.4.2) and is entirely analogous to the chain homotopies of
homology theory (Section 3.3 of [N4]). Suppose there exists a family of linear
maps
hk : Λk (Y ) −→ Λk−1 (X)
such that
hk+1 ◦ dk + dk−1 ◦ hk = f ∗ − g ∗ (5.2.4)
for each k.

hk ƒ* – g*
hk + 1

Remark: This is not intended to be a commutative diagram. Rather, f ∗ −g ∗

is the sum of the other edges in the adjacent triangles.
Then, evaluating both sides of (5.2.4) at some closed k-form ω on Y gives
hk+1 (dk ω) + dk−1 (hk ω) = f ∗ ω − g ∗ ω,
i.e.,
f ∗ ω − g ∗ ω = d(hk ω)
so f ∗ ω and g ∗ ω are cohomologous. Consequently, f # ([ω]) = g # ([ω]). Since
ω was arbitrary we conclude that f # = g # .
A family of maps hk : Λk (Y ) → Λk−1 (X) satisfying (5.2.4) is called an
algebraic homotopy (or cochain homotopy) between f and g and we
have just shown that the existence of such a thing implies that f and g induce
the same maps in cohomology. Our major result in this section will justify
the rather peculiar terminology by showing that if f and g are (smoothly)
homotopic maps of X into Y , then an algebraic homotopy between f and g
always exists.
Two smooth maps f, g : X → Y are said to be smoothly homotopic
if there exists a smooth map F : X × R → Y such that F (x, 0) = f (x)
264 5. de Rham Cohomology

and F (x, 1) = g(x) for all x ∈ X. F is then a smooth homotopy from

f to g. A map that is smoothly homotopic to a constant map is said to be
smoothly nullhomotopic.
Remark: In Section 3.2 we proved that, when X and Y are open subman-
ifolds of Euclidean spaces, two smooth maps that are homotopic in the usual
topological sense (Section 2.3 of [N4]) are also smoothly homotopic. The same
is true for any two manifolds X and Y , but the proof requires some rather
substantial machinery (see [Hir]).
The proof that f and g induce the same map in cohomology is somewhat
indirect and we begin by considering the following situation: Let U be an
open subset of Rn (with standard coordinate functions x1 , . . . , xn ) and let t
denote the standard coordinate function on R. Then x1 , . . . , xn , t are stan-
dard coordinates on U × R ⊆ Rn × R = Rn+1 . For each k = 0, 1, . . . we
define
P : Λk (U × R) −→ Λk−1 (U × R

as follows: If k = 0 we take P to be identically zero. Each ω ∈ Λk (U × R)

with k ≥ 1 is uniquely expressible as
∑ ∑
ω(x, t) = ωI (x, t) dt ∧ dxI + ωJ (x, t) dxJ
I J

where x = (x1 , . . . , xn ) and I and J are increasing index sets of length k − 1

and k, respectively (if I = (i1 , . . . , ik−1 ) with 1 ≤ i1 < · · · < ik−1 ≤ n, then
dxI = dxi1 ∧ · · · ∧ dxik−1 , etc.). Now, if a is a fixed, but arbitrary real number
we define P ω by
∑ [∫ t ]
Pω(x, t) = ωI (x, s) ds dxI . (5.2.5)
I a

∑
Note that P sends the term J ωJ (x, t)dxJ to zero and that P ω has no dt
(although its components depend on t). Since these components are clearly
C ∞ , P ω is indeed a (k − 1)-form.
Now, suppose V is another open set in Rn and ϕ : V → U is a diffeomor-
phism. Then
Φ = ϕ × idR : V × R −→ U × R
is a diffeomorphism and, for each k = 0, 1, . . .

Φ∗ : Λk (U × R) −→ Λk (V × R).

We claim that
Φ∗ (P ω) = P (Φ∗ ω) (5.2.6)
for each ω ∈ Λ (U × R).
k
 5.2. Induced Homomorphisms 265

P
Λk ( U × ) Λk – 1 ( U × )

Φ* Φ*

Λk ( V × ) Λk – 1 ( V × )
P

To prove (5.2.6) we observe first that it is trivial if k = 0 so we assume k ≥ 1.

Next observe that dt is a 1-form on both U × R and V × R and

Φ∗ (dt) = d(Φ∗ t) = d(t ◦ Φ) = dt

because Φ = ϕ × idR . On the other hand,

Φ∗ (dxi ) = d(Φ∗ xi ) = d(xi ◦ Φ) = d(xi ◦ ϕ)

∂ϕi j
= dϕi = dx .
∂xj
Thus,

Φ∗ (dxi1 ∧ · · · ∧ dxik ) = (Φ∗ (dxi1 )) ∧ · · · ∧ (Φ∗ (dxik ))

( i1 ) ( ik )
∂ϕ ∂ϕ
= dx j1
∧ ··· ∧ dx jk
∂xj1 ∂xjk

so each Φ∗ (dxI ) can be uniquely expressed as

∑
Φ∗ (dxI ) = γK (x)dxK ,
K

where K varies over increasing index sets of length k and the functions γK do
not depend on t. An analogous statement is true of each Φ∗ (dxJ ). Now, for
ω ∈ Λk (U × R),
( )
∑ ∑
Φ∗ ω(x, t) = Φ∗ ωI (x, t) dt ∧ dxI + ωJ (x, t) dxJ
I J
∑ ( ) ∑ ( )
∗
= ωI (x, t)Φ dt ∧ dx I
+ ωJ (x, t) Φ∗ dxJ
I J
∑ ( ) ∑ ( )
= ωI (x, t) dt ∧ Φ∗ dxI + ωJ (x, t) Φ∗ dxJ .
I J
266 5. de Rham Cohomology

Exercise 5.2.1 Show that

∑ [∫ t ]
P (Φ∗ ω) = ωI (x, s)ds Φ∗ (dxI ) = Φ∗ (P ω).
I a

The fact that P , defined for open subsets of Rn , commutes with pullbacks
by diffeomorphisms of the type Φ = ϕ × idR permits us to define an analogous
operation on manifolds. More precisely, we will prove the following: Let X be
a smooth n-manifold. Then, for each k = 0, 1, . . . there exists a linear map

P : Λk (X × R) −→ Λk−1 (X × R)

such that
1. If X = U , an open submanifold of Rn , then P is given by (5.2.5).
2. If Y is another smooth n-manifold, ϕ : Y → X is a diffeomorphism and
Φ = ϕ × idR : Y × R → X × R, then
Φ∗ ◦ P = P ◦ Φ∗ . (5.2.7)
To prove this we first take P to be identically zero when k = 0. Now let
ω ∈ Λk (X × R), where k ≥ 1. We will define P ω locally in coordinates and
then show that the definition is independent of the choice of coordinates. Let
(U, φ) be a chart on X. Then φ−1 : φ(U ) → U is a diffeomorphism and
therefore so is φ−1 × idR : φ(U ) × R → U × R. Thus, (φ−1 × idR )∗ ω is a
k-form on φ(U ) × R ⊆ Rn × R (technically we should write ι∗ ω rather than
ω, where ι : U × R ,→ X × R, but we will suppress this inclusion). Thus,
P ((φ−1 × idR )∗ ω) as defined above in Euclidean spaces, is a (k − 1)-form on
φ(U ) × R. We may therefore define P ω on U by
( ( )
( )∗ )
P ω = (φ × idR )∗ P φ−1 × idR ω .

Now, let (V, ψ) be another chart on X with U ∩ V ̸= ∅. To prove that P ω is

a well-defined (k − 1)-form on X × R we must show that, on U ∩ V ,
( (( )∗ )) ( (( )∗ ))
(ψ × idR )∗ P ψ −1 × idR ω = (φ × idR )∗ P φ−1 × idR ω .

Define ϕ : ψ(U ∩ V ) → φ(U ∩ V ) by ϕ = φ ◦ ψ −1 and let

Φ = ϕ × idR : ψ(U ∩ V ) × R −→ φ(U ∩ V ) × R.

Exercise 5.2.2 Show that

( ( )
∗
( −1 )∗ ) (
∗
)
Φ P φ × idR ω = P (ψ × idR ) ω .
 5.2. Induced Homomorphisms 267

Thus,
( (( )∗ ))
∗ −1
(ψ × idR ) P ψ × idR ω
(( ( )
))
∗ ∗ −1 ∗
= (ψ × idR ) Φ P (φ × idR ) ω

( )∗ ( (( )∗ )
)
−1
= Φ ◦ (ψ × idR ) P φ × idR ω
( ( )
∗
( −1 )∗ )
= (φ × idR ) P φ × idR ω

because Φ ◦ (ψ × idR ) = (ϕ × idR ) ◦ (ψ × idR ) = (ϕ ◦ ψ) × idR = φ × idR .

We conclude then that P is well-defined on X. The fact that P agrees with
(5.2.5) when X is an open submanifold of Rn is obvious from its definition.
Property # 2 on page 308 is also clear since it is an equality between two
(k − 1)-forms on Y × R which can be verified locally, in coordinates, using
(5.2.6). We put the operator P to use in our next result.
Theorem 5.2.1 Let π : X × R → X be the projection and, for any a ∈ R,
let ia : X → X × R be the embedding ia (x) = (x, a). Then, for any form ω
on X × R,
d ◦ P (ω) − P ◦ d(ω) = ω − (ia ◦ π)∗ ω, (5.2.8)

where P : Λk (X × R) → Λk−1 (X × R) is the operator defined as above

using the local coordinate operator (5.2.5) (note the presence of a in (5.2.5)).
Proof: First we observe that it is enough to prove the result when X is
an open subset of a Euclidean space. Indeed, suppose this has been proved.
Let φ : U → φ(U ) ⊆ Rn be a chart on X and set Φ = φ × idR . Then
Φ∗ : Λk (φ(U ) × R) → Λk (U × R) for each k. With π : U × R → U the
projection and ia : U → U × R the embedding we define π̃ and ĩa by the
following commutative diagram:

π ia
U× U U×

Φ Φ

π ia
268 5. de Rham Cohomology

For each k this induces a commutative diagram

π* i*a
Λk ( U × ) Λk ( U) Λk ( U × )

Φ* Φ*

π* ia*

For any ω̃ ∈ Λk (φ(U ) × R),

( )
π ∗ (i∗a (Φ∗ ω̃)) = Φ∗ π̃ ∗ (ĩa∗ ω̃) ,

i.e., ( )
(ia ◦ π)∗ (Φ∗ ω̃) = Φ∗ (ĩa ◦ π̃)∗ ω̃ .

Now let ω ∈ Λk (X × R). On U, ω = Φ∗ ω̃ for some ω̃ ∈ Λk (φ(U ) × R). Since

d commutes with any pullback and P commutes with Φ∗ ,

d ◦ P (ω) + P ◦ d (ω) = d ◦ P (Φ∗ ω̃) + P ◦ d (Φ∗ ω̃)

= Φ∗ (d ◦ P (ω) + P ◦ d (ω)).

Moreover,
ω − (ia ◦ π)∗ ω = Φ∗ (ω̃) − (ia ◦ π)∗ (Φ∗ ω̃)
= Φ∗ (ω̃ − (ĩa ◦ π̃)∗ ω̃).

Thus, assuming we have proved that d ◦ P (ω̃) + P ◦ d(ω̃) = ω̃ − (ĩa ◦ π̃)∗ ω̃,
the result for X will follow on U and therefore on all of X.
Thus, we need only prove the result when X = U is an open subman-
ifold of Rn (we will drop all of the tildas in the notation used above).
Any ω ∈ Λk (U × R) has a unique representation in standard coordinates
(x, t) = (x1 , . . . , xn , t) of the form
∑ ∑
ω(x, t) = ωI (x, t) dt ∧ dxI + ωJ (x, t) dxJ .
I J

Then P ω is given by (5.2.5). Since both sides of (5.2.8) are linear in ω we

need only consider the following two cases separately:
1. ω = ωJ (x, t) dxJ .
2. ω = ωI (x, t) dt ∧ dxI .
 5.2. Induced Homomorphisms 269

In the first case, ω = ωJ (x, t) dxJ implies P ω = 0 so d ◦ P (ω) = 0. Moreover,

dω is the sum of two terms, one of which contains no dt and therefore does
∂ω
not contribute to P ◦ d(ω). The other term is ∂tJ dt ∧ dxJ so
[∫ t ]
∂ωJ
P ◦ d (ω) = (x, s) ds dxJ = [ωJ (x, t) − ωJ (x, a)] dxJ
a ∂s
= ω − ωJ (x, a) dxJ .

Next observe that

(ia ◦ π)∗ ω = (ia ◦ π)∗ (ωJ (x, t) dxJ )
= (ia ◦ π)∗ (ωJ (x, t) dxj1 ∧ · · · ∧ dxjk )
= (ia ◦ π)∗ (ωJ (x, t))(ia ◦ π)∗ (dxj1 ) ∧ · · · ∧ (ia ◦ π)∗ (dxjk ).

But
(ia ◦ π)∗ (ωJ (x, t)) = ωJ ◦ (ia ◦ π)(x, t) = ωJ (x, a)
and
(ia ◦ π)∗ (dxj ) = d (xj ◦ (ia ◦ π)) = dxj
so
(ia ◦ π)∗ ω = ωJ (x, a) dxJ .
Thus,
d ◦ P (ω) + P ◦ d (ω) = 0 + ω − (ia ◦ π)∗ ω
as required.
For the second case one has, as above, (ia ◦ π)∗ (ωI (x, t)) = ωI (x, a) and
(ia ◦ π)∗ (dxi ) = dxi , but now

(ia ◦ π)∗ (dt) = d(t ◦ (ia ◦ π)) = d(t(x, a)) = d(a) = 0

so
(ia ◦ π)∗ ω = 0.
Next, [∫ ]
t
Pω = ωI (x, s) ds dxI
a
so
[( ∫ t ) ( ∫ t ) ]
∂ ∂
d ◦ P (ω) = ωI (x, s) ds dx + α
ωI (x, s) ds dt ∧ dxI
∂xα a a ∂t
[∫ t ]
∂ωI
= (x, s) ds dxα ∧ dxI + ωI (x, t) dt ∧ dxI
a ∂xα
[∫ t ]
∂ωI
= (x, s) ds dxα ∧ dxI + ω.
a ∂xα
270 5. de Rham Cohomology

Exercise 5.2.3 Show that dω = − ∂x

∂ωI
α (x, t)dt ∧ dx ∧ dx
α I
and conclude that

d ◦ P (ω) + P ◦ d(ω) = ω − (ia ◦ π)∗ ω. 

Corollary 5.2.2 Let π : X × R → X be the projection and, for any a ∈ R,

let ia : X → X × R be the embedding ia (x) = (x, a). Then the induced maps

a : Hde R (X × R) −→ Hde R (X)

i# k k

and
π # : Hdek R (X) −→ Hdek R (X × R)
are inverses of each other for every k = 0, 1, 2, . . .. In particular,

Hdek R (X × R) ∼
= Hdek R (X).

Proof: Since π ◦ ia = idX , ia# ◦ π # = idHdek R (X) for any k. Now, ia ◦ π

is not the identity, but we show that it nevertheless induces the identity in
cohomology. Letting P denote the operator of Theorem 5.2.1 we have

d(P ω) + P (d ω) = ω − (ia ◦ π)∗ ω.

If ω is closed, this gives

ω − (ia ◦ π)∗ ω = d(P ω)

so ω − (ia ◦ π)∗ ω is exact. Thus, [ω] = [(ia ◦ π)∗ ω] = (ia ◦ π)# ([ω]) so
(ia ◦ π)# = idHdek R (X×R) , i.e., π # ◦ ia# = idHdek R (X×R) and the result follows.


Remark: Note that it follows from Corollary 5.2.2 that, if a and b are any
two real numbers, then
ia# = ib# .

Corollary 5.2.3 Smoothly homotopic maps induce the same maps in

cohomology.
Proof: Let X and Y be manifolds, f, g : X → Y smooth maps and
F : X × R → Y a smooth map with F (x, 0) = f (x) and F (x, 1) = g(x)
for all x ∈ X. If i0 , i1 : X → X × R are the embeddings i0 (x) = (x, 0) and
i1 (x) = (x, 1), then f = F ◦ i0 and g = F ◦ i1 . Thus, f # = i0 # ◦ F # and
g # = i1 # ◦ F # . But i0 # = i1 # so the result follows. 
We will say that two manifolds X and Y are of the same smooth
homotopy type if there exist smooth maps h : X → Y and h′ : Y → X such
that h ◦ h′ and h′ ◦ h are smoothly homotopic to idY and idX , respectively. In
this case we conclude from Corollary 5.2.3 that (h′ )# ◦ h# = idHde k
R (Y )
and
 5.2. Induced Homomorphisms 271

h# ◦ (h′ )# = idHde
k
R (X)
for any k so h# and (h′ )# are inverse isomorphisms.
In particular, we have the following very useful computational device.
Corollary 5.2.4 Two manifolds of the same smooth homotopy type have the
same de Rham cohomology groups.
Remark: From the Remarks following Corollary 3.2.3 it follows that the
de Rham cohomology groups are homotopy invariants (and, in particular,
homeomorphism invariants).
A manifold X is smoothly contractible if the identity map idX is
smoothly homotopic to a constant map from X to X.
Exercise 5.2.4 Show that X is smoothly contractible if and only if it has the
same smooth homotopy type as a point (connected, 0-dimensional manifold).
Hint: The corresponding topological result is Theorem 2.3.7 of [N4].
Corollary 5.2.5 A smoothly contractible manifold X has trivial de Rham
R (X) for all k ≥ 1.
k
cohomology groups Hde
Let X be a manifold and A a smooth submanifold of X. Then the
inclusion map ι : A ,→ X is smooth (in fact, an embedding, by Lemma 5.6.1
of [N4]). A smooth map r : X → A is called a smooth retraction of X
onto A if r ◦ ι = idA . In this case, ι# ◦ r# = idHde
k
R (A)
for each k. A smooth
retraction r is called a smooth deformation retraction if ι ◦ r : X → A
is smoothly homotopic to idX . In this case, A and X obviously have the
same smooth homotopy type and therefore isomorphic cohomology groups,
but more is true.
Exercise 5.2.5 Show that, if r : X → A is a smooth deformation retraction
of X onto A, then ι# and r# are inverse isomorphisms for each k.
Exercise 5.2.6 Show that there is a smooth deformation retraction of
R2 − { (0, 0) } onto S 1 and then describe all of the de Rham cohomology
groups of R2 − { (0, 0) }.
Corollary 5.2.2 generalizes easily to the case in which R is replaced by
any smoothly contractible manifold. Specifically, we suppose X and Y are
manifolds with Y smoothly contractible. Let F : Y × R → Y be a smooth
map with F (y, 0) = y and F (y, 1) = y0 ∈ Y for every y ∈ Y . We also let
π : X × Y → X be the projection and iy0 (x) = (x, y0 ). Then π ◦ iy0 = idX .
Exercise 5.2.7 Show that iy0 ◦ π is smoothly homotopic to idX×Y .
Hint: Define H : (X × Y ) × R → X × Y by H((x, y), t) = (x, F (y, t)).
It follows that π # : Hde R (X) → Hde R (X × Y ) and iy0 R (X × Y ) →
k k # k
: Hde
k
Hde R (X) are inverse isomorphisms and, in particular, we have the following
result.
Theorem 5.2.6 Let X and Y be smooth manifolds with Y smoothly con-
∼ k
R (X × Y ) = Hde R (X) for every k.
k
tractible. Then Hde
272 5. de Rham Cohomology

One more result along these lines will be of use. We wish to consider a
smooth vector bundle over X. More precisely, we begin with a
P
smooth principal G-bundle G ,→ P → X over X and a representation
ρ : G → GL(V) of G on some finite dimensional vector space V. Now consider
the associated vector bundle

Pρ : P ×ρ V −→ X.

We will show that P ×ρ V and X have the same de Rham cohomology groups.
For this recall that each fiber Pρ−1 (x) = { [p, v] : v ∈ V }, where p is any point
in P −1 (x), has a natural vector space structure under which it is isomorphic
to V. In particular, each Pρ −1 (x), has a copy [p, 0] of the zero element in V
and we may define a map σ0 : X → P ×ρ V by σ0 (x) = [p, 0].
Exercise 5.2.8 Show that σ0 is a smooth global cross-section of P ×ρ V.
The cross-section σ0 is called the zero section of P ×ρ V and, because it is a
cross-section, it satisfies Pρ ◦ σ0 = idX . Defining F : (P ×ρ V) × R → P ×ρ V
by F ([p, v], t) = [p, tv] we obtain a smooth map with F ([p, v], 0) = [p, 0] = σ0 ◦
Pρ ([p, v]) and F ([p, v], 1) = [p, v] = idP ×ρ V ([p, v]). Thus, σ0 ◦ Pρ is homotopic
to idP ×ρ V . We conclude that σ0 # : Hde R (P ×ρ V) → Hde R (X) and Pρ
k k #
:
Hde R (X) → Hde R (P ×ρ V) are inverse isomorphisms.
k k

Theorem 5.2.7 A smooth vector bundle and its base manifold have the same
de Rham cohomology groups.

The situation is very different for principal bundles, however, as we shall see
in Section 5.4.
Corollary 5.2.4 is an important tool in the calculation of cohomology groups,
but its effective use requires a means of putting together the cohomology of
a manifold X from that of two open submanifolds of which it is the union.
For example, S 2 = UN ∪ US , where UN and US are S 2 minus the south and
north poles, respectively. Both UN and US are diffeomorphic to R2 and so
have trivial cohomology. Moreover, UN ∩US is diffeomorphic to the punctured
plane R2 −{ (0, 0) } and so we know its cohomology as well (because it has the
same homotopy type as S 1 ). What’s needed then is some relationship between
R (UN ∪US ), Hde R (UN ), Hde R (US ) and Hde R (UN ∩US ). This is provided
k k k k
Hde
by the “Mayer-Vietoris sequence” which is our next major objective. We take
one preliminary step to conclude this section, an important algebraic detour
in the next and finally prove the theorem in Section 5.4. This will all seem
quite familiar to those who have been through the proof of Mayer-Vietoris
for singular homology in Section 3.5 of [N4], but the argument here is much
simpler and one should keep an eye out to spot the precise point at which one
can account for this relative simplicity.
We consider then a smooth manifold X = U ∪ V , where U and V are open
submanifolds of X. Introduce the inclusions
 5.2. Induced Homomorphisms 273

iU : U ,→ X
iV : V ,→ X
jU : U ∩ V ,→ U
jV : U ∩ V ,→ V

and, for each k, their pullback (restriction) maps

iU ∗ : Λk (X) −→ Λk (U )
iV ∗ : Λk (X) −→ Λk (V )
jU ∗ : Λk (U ) −→ Λk (U ∩ V )
jV ∗ : Λk (V ) −→ Λk (U ∩ V ).

We also define

αk = iU ∗ ⊕ iV ∗ : Λk (X) −→ Λk (U ) ⊕ Λk (V )

by

αk (ω) = (iU ∗ ω, iV ∗ ω)
and

β k = jU ∗ − jV ∗ : Λk (U ) ⊕ Λk (V ) −→ Λk (U ∩ V )
by

β k (λ1 , λ2 ) = jU ∗ λ1 − jV ∗ λ2 .
Thus, for each k, we have a sequence

αk βk
0 −→ Λk (X) −→ Λk (U ) ⊕ Λk (V ) −→ Λk (U ∩ V ) −→ 0 (5.2.9)
which we claim is exact (i.e., the image of each map is the kernel of the next).
Exactness at Λk (X) is the statement that αk is one-to-one. To see this suppose
ω 1 and ω 2 are distinct elements of Λk (X). Since X = U ∪ V , there is a p in
U or in V (or in both) at which ω 1 (p) ̸= ω 2 (p). Since U and V are open in
X, the tangent space at p to either U or V coincides with Tp (X) so either
iU ∗ ω 1 (p) ̸= iU ∗ ω 2 (p) if p ∈ U or iV ∗ ω 1 (p) ̸= iV ∗ ω 2 (p) if p ∈ V . In either
case, αk (ω 1 ) ̸= αk (ω 2 ).
To prove exactness at Λk (U ) ⊕ Λk (V ) we must show that Image (αk ) =
ker (β k ) and for this we prove containment in each direction. Since αk (ω) =
(iU ∗ ω, iV ∗ ω), β k ◦ αk (ω) = jU ∗ (iU ∗ ω) − jV ∗ (iV ∗ ω) = 0 because both terms
are the restrictions of ω to U ∩ V . Thus, Image (αk ) ⊆ ker (β k ). Next we show
that ker (β k ) ⊆ Image (αk ). Suppose β k (λ1 , λ2 ) = 0. Then jU ∗ (λ1 ) = jV ∗ (λ2 )
so λ1 and λ2 agree on U ∩ V . Thus, we can define ω on X = U ∪ V by taking
iU ∗ ω = λ1 and iV ∗ ω = λ2 . Then αk (ω) = (λ1 , λ2 ) so (λ1 , λ2 ) ∈ Image (αk ).
274 5. de Rham Cohomology

Exactness at Λk (U ∩ V ) is the statement that β k is onto. To see this we let

{ ϕU , ϕV } be a family of functions of the sort guaranteed by Corollary 3.1.5
for the open cover { U, V } of X. Next let ω ∈ Λk (U ∩ V ) be arbitrary. Then
ϕV ω ∈ Λk (U ∩ V ) and, since supp ϕV ⊆ V we can define λ1 ∈ Λk (U ) by
{
ϕV ω on U ∩ V
λ1 = .
0 on U − (U ∩ V )

Similarly, define λ2 ∈ Λk (V ) to be −ϕU ω on U ∩ V and 0 on V − (U ∩ V ).

Then

β k (λ1 , λ2 ) = jU ∗ λ1 − jV ∗ λ2 = ϕV ω − (−ϕU ω)
= (ϕV + ϕU ) ω = ω.

so ω ∈ Image (β k ) as required.
Thus, we have an exact sequence (5.2.9) for each k. Letting d denote
the exterior differentiation operator on all of the spaces of forms (Λk (X),
Λk (U ), Λk (V ) or Λk (U ∩ V ) for any k) we define

d ⊕ d : Λk (U ) ⊕ Λk (V ) −→ Λk+1 (U ) ⊕ Λk+1 (V )

by (d ⊕ d)(λ1 , λ2 ) = (dλ1 , dλ2 ) for any k and consider the diagram

d d%d d
αk βk
0 Λk (X) Λk (U) % Λk (V) Λk (U ∩ V) 0

d d%d d

αk + 1 βk + 1
0 Λk + 1 (X) Λk + 1 (U) % Λk + 1 (V) Λk + 1 (U ∩ V) 0

d d%d d

We have just shown that the diagram has exact rows. The columns, although
not exact, in general, compose to zero at each stage because d2 = 0.
Exercise 5.2.9 Show that each square in this diagram is commutative, i.e.,
that, for each k,

(d ⊕ d) ◦ αk = αk ◦ d and d ◦ β k = β k ◦ (d ⊕ d). (5.2.10)

 5.3. Cochain Complexes and Their Cohomology 275

The major result toward which we are headed (The Mayer-Vietoris

Sequence) is a purely algebraic consequence of the structure of this last di-
agram. For this reason, and to build a structure large enough to accommo-
date yet another “cohomology theory” that we will require in Chapter 6, we
pause momentarily to introduce a few of the basic notions of cohomological
algebra.

5.3 Cochain Complexes and Their Cohomology

We recall that a principal ideal domain is a commutative ring R with unit
element e in which there are no zero divisors (i.e., ab = 0 implies a = 0 or
b = 0) and in which every ideal I is principal (i.e., I = Ra = { ra : r ∈ R }
for some a ∈ R). From our point of view the most important examples
are fields (e.g., R) and the ring Z of integers. A module over R, or R-
module, consists of an Abelian group C (whose operation we designate +)
together with a “scalar multiplication” map R × C → C which assigns to
each pair (a, x) ∈ R × C an element ax of C satisfying (a + b)x = ax + bx,
a(x + y) = ax + ay, (ab)x = a(bx), and ex = x for all a, b ∈ R and
x, y ∈ C. Thus, R-modules are simply real vector spaces and Z-modules
are Abelian groups. All of the basic notions of linear algebra have analogues
for R-modules and we will need just a few. If C1 and C2 are R-modules,
then a map f : C1 → C2 is said to be R-linear (or a homomorphism) if
f (ax + by) = a f (x) + b f (y) for all a, b ∈ R and x, y ∈ C1 . An additive sub-
group C ′ of the R-module C is called a submodule of C if it is closed under
scalar multiplication (and therefore is itself an R-module under the same op-
erations as C). In this case one defines the quotient module C/C ′ by providing
the quotient group with the obvious scalar multiplication: If [v] = v + C ′ is
in C/C ′ , then a[v] = a(v + C ′ ) = av + C ′ = [av]. Finally, if C1 and C2 are
R-modules, their direct sum C1 ⊕ C2 is obtained from the direct sum of the
underlying Abelian groups by defining scalar multiplication coordinatewise:
a(x1 , x2 ) = (ax1 , ax2 ), for all a ∈ R, x1 ∈ C1 and x2 ∈ C2 .
Now, a cochain complex C ∗ consists of a sequence of R-modules and
homomorphisms

δ k−1 δk δ k+1
· · · −→ C k−1 −→ C k −→ C k+1 −→ · · ·

defined for all integers k such that the image of each homomorphism is
contained in the kernel of the next, i.e., δ k+1 ◦δ k = 0 for each k. In the only ex-
ample we have seen thus far C k was the R-module Λk (X) of k-forms on a man-
ifold X and δ k was the exterior differentiation map dk . The homomorphism
δ k : C k → C k+1 is called the k th coboundary operator and, when it is con-
venient and no confusion will result, will be denoted simply δ. Image (δ k−1 ) is
a submodule of C k and its elements are called k -coboundaries. ker (δ k ) is a
submodule of C k and its elements are called k -cocycles. Since δ k ◦ δ k−1 = 0,
276 5. de Rham Cohomology

Image (δ k−1 ) ⊆ ker (δ k ) so we may form the quotient module

H k (C ∗ ) = ker (δ k ) / Image (δ k−1 ).

H k (C ∗ ) is called the k th cohomology group of the complex C ∗ (even though

k th cohomology module would be more appropriate). The elements of H k (C ∗ )
are equivalence classes [x] (called cohomology classes), where x is a k-
cocycle and the equivalence relation is defined as follows: If x, x′ ∈ ker (δ k ),
then x′ ∼ x if and only if x′ = x+y, where y ∈ Image (δ k−1 ) is a k-coboundary
(we then say that x′ and x are cohomologous).
If C1∗ and C2∗ are two cochain complexes of R-modules (with connecting
homomorphisms δ1k and δ2k , respectively), then a cochain map

α : C1∗ −→ C2∗

from C1∗ to C2∗ is a sequence of R-linear maps

αk : C1k −→ C2k , k = 0, ±1, . . .

such that
δ2k ◦ αk = αk+1 ◦ δ1k

for each k.

αk
C1k C2k

δ1k δ2k

αk + 1
C1k + 1 C2k + 1

Exercise 5.3.1 Show that αk (ker (δ1k )) ⊆ ker (δ2k ) and αk (Image (δ1k−1 )) ⊆
Image (δ2k−1 ) for each k and conclude that αk induces a homomorphism

(αk )# : H k (C1∗ ) −→ H k (C2∗ )

 5.3. Cochain Complexes and Their Cohomology 277

in cohomology defined by

(αk )# ([x]) = [αk (x)]

for each x ∈ ker (δ1k ).

Remark: We have seen an example of this phenomenon. If X and Y are

smooth manifolds and Λ∗ (X) and Λ∗ (Y ) denote their cochains of real-valued
forms, then any smooth map f : X → Y induces a pullback cochain map
f ∗ : Λ∗ (Y ) → Λ∗ (X) which, in turn, induces the homomorphisms

f # : Hde R (Y ) −→ Hde R (X), k = 0, ±1, . . .

k k

in de Rham cohomology.
Exercise 5.3.2 Two cochain maps α : C1∗ → C2∗ and β : C1∗ → C2∗ are said
to be algebraically homotopic (or cochain homotopic) if there exists a
family of homomorphisms

hk : C1k −→ C2k−1 , k = 0, ±1, . . .

such that
hk+1 ◦ δ2k + δ1k−1 ◦ hk = αk − β k

for each k (the family h = { hk } of maps is then called an algebraic homo-

topy, or cochain homotopy, between α and β). Show that, if α and β are
algebraically homotopic, then they induce the same maps in cohomology, i.e.,

(αk )# = (β k )#

for each k. Hint: The argument is the same as for de Rham cohomology.
The composition of two cochain maps C1∗ → C2∗ and C2∗ → C3∗ is defined by
composing each of the homomorphisms C1k → C2k and C2k → C3k and clearly
induces homomorphisms in cohomology that are just the compositions of those
induced by the two cochain maps. Denoting by 0 both the trivial R-module
and the cochain complex one can form from these we will say that a sequence
of cochain maps
β
0 −→ C1∗ −→ C2∗ −→ C3∗ −→ 0
α

forms a short exact sequence if, for each k, the sequence of R-modules

αk βk
0 −→ C1k −→ C2k −→ C3k −→ 0

is exact (i.e., the image of each map equals the kernel of the next).
278 5. de Rham Cohomology

αk βk
0 C1k C2k C3k 0

δ1k δ2k δ3k

αk + 1 βk + 1
0 C1k + 1 C2k + 1 C3k + 1 0

δ1k + 1 δ2k + 1 δ3k + 1

αk + 2 βk + 2
0 C1k + 2 C2k + 2 C3k + 2 0

Now, α and β both induce maps in cohomology, but it need not be the case
that the sequences

0 −→ H k (C1∗ ) −→ H k (C2∗ ) −→ H k (C3∗ ) −→ 0

are exact (we will construct an example in de Rham cohomology shortly).

Thus, short exact sequences of cochain maps do not induce short exact se-
quences in cohomology. They do, however, induce a certain long exact sequence
of cohomology groups which is the basis for most of our calculations of coho-
mology.
α β
Theorem 5.3.1 Let 0 → C1∗ → C2∗ → C3∗ → 0 be a short exact sequence of
cochain complexes. Then there exist homomorphisms

∂ k : H k (C3∗ ) −→ H k+1 (C1∗ ), k = 0, ±1, . . .

such that the following sequence is exact:

∂ k−1 (αk )# (β k )# ∂k
· · · H k−1 (C3∗ ) −→ H k (C1∗ ) −→ H k (C2∗ ) −→ H k (C3∗ ) −→ H k+1 (C1∗ ) −→ · · ·

Proof: To define ∂ k : H k (C3∗ ) → H k+1 (C1∗ ) we will make two choices and
then prove that the definition is independent of those choices and has the
required properties.
Any element of H k (C3∗ ) is [x] for some x ∈ C3k with δ3k (x) = 0. Since β k is
surjective, there exists a y ∈ C2k with β k (y) = x. Now, observe that
( ) ( )
β k+1 δ2k (y) = δ3k β k (y) = δ3k (x) = 0
 5.3. Cochain Complexes and Their Cohomology 279

so
δ2k (y) ∈ ker (β k+1 ) = Image (αk+1 ).
Since αk+1 is injective, there is a unique z ∈ C1k+1 such that

αk+1 (z) = δ2k (y).

We claim that δ1k+1 (z) = 0 so z determines a cohomology class

[z] ∈ H k+1 (C1∗ ). To see this note that, since αk+2 is injective, it is enough to
show that αk+1 (δ1k+1 (z)) = 0. But
( ) ( )
αk+2 δ1k+1 (z) = δ2k+1 (αk+1 (z)) = δ2k+1 δ2k (y) = 0.

Thus, we may define

∂ k ([x]) = [z].
Now we check that this definition of ∂ k ([x]) does not depend on either of
the choices we have made.
1. The representative x of [x] ∈ H k (C3∗ ).
2. The y ∈ C2k with β k (y) = x.
We deal with the second of these first. Suppose then that y, y ′ ∈ C2k with
β k (y) = β k (y ′ ) = x. Then δ2k (y) and δ2k (y ′ ) are both in Image (αk+1 ) (proved
above) so there exist unique elements z and z ′ of C1k+1 with αk+1 (z) = δ2k (y)
and αk+1 (z ′ ) = δ2k (y ′ ). We must show that z − z ′ = δ1k (w) for some w ∈ C1k
so that [z ′ ] = [z]. But β k (y − y ′ ) = β k (y) − β k (y ′ ) = x − x = 0 implies
y−y ′ ∈ ker (β k ) = Image (αk ) so there is a unique w ∈ C1k with αk (w) = y−y ′ .
Now compute
( )
αk+1 δ1k (w) = δ2k (αk (w)) = δ2k (y − y ′ )
= δ2k (y) − δ2k (y ′ ) = αk+1 (z) − αk+1 (z ′ )
= αk+1 (z − z ′ ).

But αk+1 is injective so z − z ′ = δ1k (w) as required.

Exercise 5.3.3 Use what has just been proved to show that the map x → [z]
from the cocycles in C3k to H k+1 (C1∗ ) is well-defined and R-linear.
Next we prove # 1. Thus, let x, x′ ∈ C3k with δ3k (x) = δ3k (x′ ) = 0 and
x − x′ = δ3k−1 (w) for some w ∈ C3k−1 . Let w = β k−1 (v) for v ∈ C2k−1 . Then
( )
x − x′ = δ3k−1 (w) = δ3k−1 (β k−1 (v)) = β k δ2k−1 (v)

so x − x′ is the image of y = δ2k−1 (v) ∈ C2k under β k . Now, δ2k (y) =

δ2k (δ2k−1 (v)) = 0 so the unique element of C1k+1 which αk+1 maps onto δ2k (y)
280 5. de Rham Cohomology

is 0, i.e., x − x′ → [0] ∈ H k+1 (C1∗ ). But, by Exercise 5.3.3, x − x′ → [x − x′ ] =

[x] − [x′ ] so [x] = [x′ ] as required.
We have shown then that the map ∂ k is well-defined and, by
Exercise 5.3.3, it is R-linear. All that remains is to show that the long se-
quence of cohomology groups in the theorem is exact at H k (C1∗ ), H k (C2∗ )
and H k (C3∗ ) for each k. These are all similar so we will prove the last
and leave the other two as exercises. Thus, we show that Image ((β k )# ) ⊆
ker (∂ k ) and ker (∂ k ) ⊆ Image ((β k )# ). Let [y] ∈ H k (C2∗ ) and consider
(β k )# ([y]) ∈ Image ((β k )# ). We must show that ∂ k ((β k )# ([y])) = [0]. But
∂ k ((β k )# ([y])) = ∂ k ([β k (y)]) = [z], where αk+1 (z) = δ2k (y). Now, y is a co-
cycle in C2k so δ2k (y) = 0 and therefore αk+1 (z) = 0. But αk+1 is injective
so z = 0 and the result follows. For the reverse containment we begin with
[x] ∈ H k (C3∗ ) for which ∂ k ([x]) = [0]. We must show that [x] = (β k )# ([y ′ ])
for some [y ′ ] ∈ H k (C2∗ ), i.e., [x] = [β k (y ′ )] for some y ′ ∈ C2k with δ2k (y ′ ) = 0.
Now, ∂ k ([x]) = [0] implies [z] = 0, where αk+1 (z) = δ2k (y) and β k (y) = x.
But then z is a coboundary so there exists a u ∈ C1k with δ1k (u) = z.
Thus, ( )
δ2k (y) = αk+1 (z) = αk+1 δ1k (u) = δ2k (αk (u))

and so
δ2k (y − αk (u)) = 0.
Let y ′ = y − αk (u) ∈ C2k . Then δ2k (y ′ ) = 0 and β k (y ′ ) = β k (y − αk (u)) =
β k (y) − β k (αk (U )) = β k (y) = x as required.

Exercise 5.3.4 Give analogous arguments to establish exactness at H k (C1∗ )

and H k (C2∗ ). 
With one last bit of algebraic machinery we will be in a position to calculate
some de Rham cohomology groups. Let C1∗ and C2∗ be two cochain complexes
with connecting homomorphisms δ1k and δ2k , respectively. The direct sum of
C1∗ and C2∗ is the cochain complex denoted C1∗ ⊕ C2∗ whose R-modules are
C1k ⊕ C2k and whose connecting homomorphisms

δ k = δ1k ⊕ δ2k : C1k ⊕ C2k −→ C1k+1 ⊕ C2k+1

are defined by
( ) ( )
δ k (x1 , x2 ) = δ1k ⊕ δ2k (x1 , x2 ) = δ1k (x1 ), δ2k (x2 ) .

Exercise 5.3.5 Show that C1∗ ⊕ C2∗ is, indeed, a cochain complex and that,
for each k,
H k (C1∗ ⊕ C2∗ ) ∼
= H k (C1∗ ) ⊕ H k (C2∗ ).
 5.4. The Mayer-Vietoris Sequence 281

5.4 The Mayer-Vietoris Sequence

We now apply Theorem 5.3.1 to the following situation in de Rham cohomol-
ogy: Let X be a smooth manifold with X = U ∪ V , where U and V are open
submanifolds of X. We have cochain complexes Λ∗ (X), Λ∗ (U ), Λ∗ (V ), and
Λ∗ (U ∩ V ) consisting of the R-modules of real-valued k-forms with exterior
differentiation as the connecting homomorphisms. In addition, we form the
direct sum complex Λ∗ (U ) ⊕ Λ∗ (V ) and the short exact sequence
β
0 −→ Λ∗ (X) −→ Λ∗ (U ) ⊕ Λ∗ (V ) −→ Λ∗ (U ∩ V ) −→ 0
α

as in Section 5.2 (see page 317). Theorem 5.3.1 then provides

homomorphisms
R (U ∩ V ) −→ Hde R (X)
k+1
∂ k : Hde
k

such that the following long sequence (eventually ending with zeros) is exact:
(α0 )# (β 0 )#
0 −→ Hde
0
R (X) −→ Hde R (U ) ⊕ Hde R (V ) −→ Hde R (U ∩ V )
0 0 0

∂0 ∂ k−1 (αk )#
−→ · · · Hde
k−1
R (U ∩ V ) −→ Hde R (X) −→ Hde R (U ) ⊕ Hde R (V )
k k k

(β k )# ∂k
−→ Hde
k
R (U ∩ V ) −→ Hde R (X) −→ · · · .
k+1

This is called the Mayer-Vietoris sequence and much of what remains of

this chapter will consist of an enumeration of some of its consequences. Begin
by noting that if U ∩ V = ∅, then each Hde
k
R (U ∩ V ) is trivial so we have short
exact sequences

0 −→ Hde
k
R (X) −→ Hde R (U ) ⊕ Hde R (V ) −→ 0
k k

from which it follows that

∼ k
R (X) = Hde R (U ) ⊕ Hde R (V ) (U ∩ V ≠ 0).
k k
Hde

Thus, in computing the cohomology of a manifold X one can consider each

connected component separately.
Recall that an open cover U = { Uα }α∈A of a manifold X is said to be
simple if any finite intersection Uα0 ∩ · · · ∩ Uαj of its elements is either empty
or diffeomorphic to Rn (where n = dim X) and that every smooth manifold
has a simple cover (see Section 3.3). A manifold is said to be of finite type
if it admits a finite simple cover. Any compact manifold is of finite type, but
many noncompact manifolds are as well, e.g., Rn . We show now that any such
manifold has finite dimensional cohomology.
Theorem 5.4.1 Let X be a smooth manifold of finite type. Then each
k
Hde R (X) is a finite dimensional real vector space.
282 5. de Rham Cohomology

Proof: The proof is by induction on the size of the simple cover. If X has
a simple cover consisting of just one open set, then it is diffeomorphic to
Rn (n = dim X) so the result follows from Theorem 5.1.1. Assume now that
the result has been established for manifolds admitting simple covers consist-
ing of N open sets and suppose that X has a simple cover { U1 , . . . , UN , UN +1 }
consisting of N +1 open sets. Let U = U1 ∪· · ·∪UN and V = UN +1 . Then U ∩V
is an open submanifold of X and { U1 ∩ UN +1 , . . . , UN ∩ UN +1 } is a simple
cover of it. By the induction hypothesis, all of the cohomology groups of U and
U ∩ V are finite dimensional. Obviously, the same is true of V = UN +1 ∼ = Rn .
Now, X = U ∪ V so, for each k, the Mayer-Vietoris sequence gives the exact
sequence
∂ α
· · · −→ Hde
k−1
R (U ∩ V ) −→ Hde R (X) −→ Hde R (U ) ⊕ Hde R (V ) −→ · · ·
k k k

R (U ) ⊕ Hde R (V ) is finite dimensional, so is Image (α). Since

k k
Since Hde
k
Hde R (U ∩ V ) is finite dimensional, Image (∂) = ker (α) is finite dimen-
k
sional. Since dim (Hde R (X)) = dim (ker(α)) + dim (Image (α)), it follows
that Hdek
R (X) is finite dimensional as well. 
Next we intend to calculate all of the cohomology groups of the spheres
S n , n ≥ 0. The Mayer-Vietoris sequence again provides an inductive
approach so we begin at the beginning. Recall that S 0 is the 2-point discrete
subspace {−1, 1} of R.
Exercise 5.4.1 Show that H 0 (S 0 ) ∼
= R ⊕ R and H k (S 0 ) = 0 for all k ≥ 1.
Remark: The de Rham cohomology “groups” are actually vector spaces so it
might, perhaps, be more reasonable to write H 0 (S 0 ) ∼
= R2 . We will, however,
bow to traditions that evolved in more general cohomology theo- ries and
retain the direct sum notation.
We have already calculated the cohomology of S 1 ((5.1.1)), but let’s do it
again with Mayer-Vietoris. Write S 1 = UN ∪ US , where UN is S 1 minus the
south pole (0, −1) and US is S 1 minus the north pole (0, 1). Both UN and
US are diffeomorphic to R (via a stereographic projection), while UN ∩ US
1 ∼
is the disjoint union of two copies of R. Since Hde
k
R (S ) = 0 for k ≥ 2, the
Mayer-Vietoris sequence gives

0 −→ Hde
0
R (S ) −→ Hde R (UN ) ⊕ Hde R (US ) −→ Hde R (UN ∩ US ) −→
1 0 0 0

R (S ) −→ Hde R (UN ) ⊕ Hde R (US ) −→ Hde R (UN ∩ US ) −→ 0

1 1 1 1 1
Hde

i.e.,

0 −→ R −→ R ⊕ R −→ R ⊕ R −→ H 1 (S 1 ) −→ 0 −→ 0 −→ 0

which we collapse to the exact sequence

 5.4. The Mayer-Vietoris Sequence 283

α β ∂
0 −→ R −→ R ⊕ R −→ R ⊕ R −→ H 1 (S 1 ) −→ 0.

Now, α is injective so dim (Image (α)) = 1 = dim (ker(β)). Moreover,

dim (R ⊕ R) = dim (Image (β)) + dim (ker(β)) implies dim (Image (β)) =1 =
dim (ker(∂)) and from this it follows in the same way that dim (Image
(∂)) = 1. But ∂ is surjective so dim (H 1 (S 1 )) = 1, i.e., H 1 (S 1 ) ∼
= R,
as expected.
Remark: DeRham cohomology is computationally much simpler than
singular homology and the reason is apparent in this first example. DeRham
cohomology “groups” are actually vector spaces and a vector space is com-
pletely determined by a single number (its dimension). Singular homol-
ogy groups really are just (Abelian) groups and these are not so easy to
characterize.
The sort of dimension counting employed in this last example occurs often
enough that it is worth the effort to prove a lemma that will allow us to evade
it in the future.
Lemma 5.4.2 Let

0 −→ V1 −→ V2 −→ V3 −→ · · · −→ Vk−1 −→ Vk −→ 0

be an exact sequence of finite dimensional vector spaces. Then

dim V1 − dim V2 + dim V3 − · · · + (−1)k−1 dim Vk = 0.

Proof: The proof is by induction on k. If k = 1 and 0 −→ V1 −→ 0 is exact,

then V1 is trivial so dim V1 = 0. Now assume the result for sequences of k − 1
vector spaces and consider the exact sequence

α β
0 −→ V1 −→ V2 −→ V3 −→ · · · −→ Vk−1 −→ Vk −→ 0.

Since ker (β) = Image (α), β : V2 −→ V3 induces a map

β̂ : V2 /Image (α) −→ V3

defined by β̂(v2 + Image (α)) = β(v2 ) for each v2 ∈ V2 . Observe that

β̂ is injective since β̂(v2 + Image (α)) = β̂(v2′ + Image (α)) implies
β(v2 ) = β(v2′ ) so β(v2 − v2′ ) = 0 and therefore v2 − v2′ ∈ ker (β) = Image
(α). This then implies that v2 + Image (α) = v2′ + Image (α) as required.
Thus,
βˆ
0 −→ V2 /Image (α) −→ V3
284 5. de Rham Cohomology

is exact. Since Image (β̂) = Image (β) it follows that, in fact,

βˆ
0 −→ V2 /Image (α) −→ V3 −→ · · · −→ Vk −1 −→ Vk −→ 0

is exact. The induction hypothesis therefore gives

0 = dim (V2 /Image (α)) − dim V3 + · · · + (−1)k dim Vk
= dim V2 − dim (Image (α)) − dim V3 + · · · + (−1)k dim Vk
= dim V2 − dim V1 − dim V3 + · · · + (−1)k dim Vk

since α is one-to-one. Thus

0 = dim V1 − dim V2 + dim V3 − · · · + (−1)k−1 dim Vk
as required. 
Now we will use the Mayer-Vietoris sequence to calculate the cohomology
of S n . The argument is an induction on the dimension of the sphere and
is based on the following observation: S n can be written as the union
S n = UN ∪ US of two copies of Rn for which UN ∩ US is S n minus two
points (the south and north poles). We know the cohomology of UN and US .
Furthermore, there is a smooth deformation retraction of UN ∩ US onto the
equator S n−1 in S n (Exercise 5.4.2) so the induction hypothesis will give us
the cohomology of UN ∩ US as well. Mayer-Vietoris will then put all of this
together into the cohomology of S n .
Exercise 5.4.2 Let n ≥ 2 be an integer and write S n = UN ∪ US ,
where UN is S n minus the south pole (0, . . . , 0, −1) and US is S n minus
the north pole (0, . . . , 0, 1). Let {e1 , . . . , en , en+1 } be the standard basis for
Rn+1 ⊇ S n and let S n−1 be the equatorial (n − 1)-sphere in S n , i.e.,
S n−1 = {(x1 , . . . , xn , xn+1 ) ∈ S n : xn+1 = 0}. Define r : UN ∩US −→ S n−1 by

x − xn+1 en+1
r(x) =
|x − xn+1 en+1 |

N
x ∈ UN ∩ US

xn + 1 en + 1
en + 1

r(x)

Sn – 1

x – xn + 1 en + 1

S
 5.4. The Mayer-Vietoris Sequence 285

Show that r is a smooth deformation retraction of UN ∩ US onto S n−1 .

Hint: For the required homotopy, consider the following figure, where
0 ≤ t ≤ 1.

txn + 1 en + 1 H (x, t)

r (x)

x – txn + 1 en + 1

Now, to carry out an induction one needs an induction hypothesis. Our only
information thus far concerns S 1 :
{
k 1 ∼ R, if k = 0, 1
Hde R (S ) = .
0, if k ̸= 0, k ̸= 1

To check that this is indicative of the general result (and to get a bit more
2 ∼
practice with Mayer-Vietoris) we will do S 2 as well. Of course, Hde
k
R (S ) = 0
for k ≥ 3 so the Mayer-Vietoris sequence gives

0 −→ Hde
0
R (S ) −→ Hde R (UN ) ⊕ Hde R (US ) −→ Hde R (UN ∩ US ) −→
2 0 0 0

R (S ) −→ Hde R (UN ) ⊕ Hde R (US ) −→ Hde R (UN ∩ US ) −→

1 2 1 1 1
Hde

R (S ) −→ Hde R (UN ) ⊕ Hde R (US ) −→ Hde R (UN ∩ US ) −→ 0.

2 2 2 2 2
Hde

R (UN ) ⊕ Hde R (US ) = R ⊕ R, but the remaining direct sums are

0 0
Now, Hde
trivial. Furthermore, Exercise 5.4.2 and Corollary 5.2.4 imply that UN ∩ US
has the same cohomology as S 1 . Thus, our sequence becomes
0 −→ R −→ R ⊕ R −→ R −→
R (S ) −→ −→ R −→
1 2
Hde 0
2
Hde 2
R (S ) −→ 0 −→ 0 −→ 0.
This “splits” into two exact sequences:

0 −→ R −→ R ⊕ R −→ R −→ Hde
1
R (S ) −→ 0
2

and
0 −→ R −→ Hde
2
R (S ) −→ 0.
2
286 5. de Rham Cohomology

Applying Lemma 5.4.2 to each gives

1 − 2 + 1 − dim (Hde
1 2
R (S )) = 0

and
1 − dim (Hde
2 2
R (S )) = 0

so
2 ∼ 2 ∼
R (S ) = 0 and Hde R (S ) = R.
1 2
Hde
We conclude then that
{
k 2 ∼ R, if k = 0, 2
Hde R (S ) = .
0, if k ̸= 0, k ̸= 2

Remark: Notice that, for k = 1, the sequence of forms and the corres-
ponding cohomology sequence are

0 −→ Λ1 (S 2 ) −→ Λ1 (UN ) ⊕ Λ1 (US ) −→ Λ1 (UN ∩ US ) −→ 0

which we know is exact and

0 −→ 0 −→ 0 ⊕ 0 −→ R −→ 0

which certainly is not. This provides the example promised just before
Theorem 5.3.1.
Exercise 5.4.3 Prove, by induction on n ≥ 1, that
{
k n ∼ R, if k = 0, n
Hde R (S ) = .
0, if k ̸= 0, k ̸= n

R (S ) has dimension one for every n ≥ 1 it is generated

n n
Remark: Since Hde
by the cohomology class of any nonexact n-form on S n , i.e., any n-form that
does not integrate to zero over S n (see Section 5.1). We found in Section 4.6
that the standard volume form on S n is of this type.
With this we are now in a position to generalize Lemma 5.1.3.
∫
Theorem 5.4.3 An n-form ν on S n is exact if and only if S n ν = 0.
∫
Proof: All that remains is to show that, if S n ν = 0, then ν is exact. Since
n n
ν is necessarily closed, it determines a cohomology class [ν] in Hde R (S ). But
dim Hde R (S ) = 1 so [ν] = α[ω], where ω is any nonexact n-form on S n and
n n

α is some real number. Now, [ν] = [αω] implies [ν − αω] = [0] so ν − αω is

exact and therefore integrates to zero over S n . Thus,
∫ ∫ ∫
0= (ν − αω) = ν −α ω
Sn Sn Sn
 5.4. The Mayer-Vietoris Sequence 287

so ∫ ∫
ν =α ω.
Sn Sn
∫ ∫
If S n ν = 0, then, since S n ω ̸= 0, we must have α = 0 and so [ν] = [0], i.e.,
ν is exact. 

Remark: We will eventually show that the same is true for any compact,
orientable manifold.
Exercise 5.4.4 For [ω] ∈ Hde
n n n
R (S ), define the integral of [ω] over S by
∫ ∫
[ω] = ω.
Sn Sn

Show that this is well-defined (i.e., independent of the representative ω of [ω])

and that the map ∫
R (S ) −→ R
n n
: Hde
Sn
thus defined is an isomorphism.

Remark: This too will generalize to arbitrary compact, connected,

orientable manifolds.

Exercise 5.4.5 Let 0 denote the origin (0, . . . , 0) in Rn+1 , n ≥ 1. Show that
{
R, if k = 0, n
k
Hde R (R
n+1
− {0}) ∼
= .
0, if k ̸= 0, k ̸= n

Hint: Exercise 5.4.3.

Exercise 5.4.6 Let X denote the subspace of Rn+1 , n ≥ 2, obtained by
deleting the “xn+1 -axis” (i.e., the set of all points of the form (0, . . . , 0, t) for
t ∈ R). Show that
{
k ∼ R, if k = 0, n − 1
Hde R (X) = .
0, if k ̸= 0, k ̸= n − 1
Hint: Corollary 5.2.2.
One can use the cohomology groups we have already calculated to prove
some rather nontrivial results concerning the topology of Euclidean spaces.
We illustrate by proving that Rn and Rm are homeomorphic if and only if
n = m.
Remark: Note that the corresponding assertion in the linear category (i.e.,
Rn and Rm are isomorphic if and only if n = m) is obvious and that this
renders the analogous result in the smooth category (i.e., Rn and Rm are
diffeomorphic if and only if n = m) equally obvious since diffeomorphisms
288 5. de Rham Cohomology

have derivatives that are isomorphisms. The game here is to get from the
smooth to the topological category. This was done using singular homology in
Section 3.4 of [N4], but the computations required to do it were substantially
more involved.

In Section 3.2 we proved that, if U and V are open submanifolds of

some Euclidean spaces, then any smooth map h : U −→ V is (continuously)
homotopic to a smooth map f : U −→ V and that two smooth maps f,
f ′ : U −→ V that are continuously homotopic are also smoothly homotopic.
We use these facts to prove that continuous maps from U to V induce linear
k k
maps from Hde R (V ) to Hde R (U ) for every k.
Let U ⊆ R and V ⊆ Rm be open sets and h : U −→ V a continuous
n

map. Select a smooth map f : U −→ V that is continuously homotopic to h.

Then for each k, f induces a linear map f # from Hde k k
R (Y ) to Hde R (X). If
′
f : U −→ V is any other smooth map continuously homotopic to h, then
f and f ′ are continuously homotopic and, consequently, are also smoothly
homotopic. Corollary 5.2.3 then implies that f and f ′ induce the same linear
maps in cohomology. Thus, we may unambiguously define

h# : Hde R (Y ) −→ Hde R (X)

k k

by
h# = f # ,
where f is any smooth map homotopic to h. We ask the reader to establish
all of the usual properties.
Exercise 5.4.7 Let U, V and W be open sets in Euclidean spaces and
k any integer. Prove each of the following.
1. If h0 , h1 : U −→ V are homotopic continuous maps, then
h# #
0 = h1 : Hde R (V ) −→ Hde R (U ).
k k

2. If h : U −→ V and g : V −→ W are continuous maps, then

(g ◦ h)# = h# ◦ g # : Hde
k
R (W ) −→ Hde R (U ).
k

3. If the continuous map h : U −→ V is a homotopy equivalence, then

h# : Hde R (V ) −→ Hde R (U ) is an isomorphism.
k k

It follows, in particular, from Exercise 5.4.7 (3), that if h : U −→ V is

a homeomorphism, then h# : Hde R (V ) −→ Hde R (U ) is an isomorphism for
k k

each k. Now, suppose R and R are Euclidean spaces. Certainly, if n =

n m

m, then Rn and Rm are homeomorphic. Suppose, conversely, that there is a

homeomorphism H : Rn −→ Rm . By composing with a translation if necessary
we may assume that H carries the origin 0 ∈ Rn onto the origin 0 ∈ Rm .
Consequently,
h = H|Rn −{0} : Rn − {0} −→ Rm − {0}
 5.4. The Mayer-Vietoris Sequence 289

is a homeomorphism. But then

h# : Hde R (R − {0}) −→ Hde R (R − {0})

k m k n

is an isomorphism for each k so, in particular, this is true when k = m − 1.

∼
R (R − {0}) = R ⊕ R so we must have n = 1 as well
0 n
Now, if m = 1, Hde
0 n ∼
(n ≥ 2 implies Hde R (R − {0}) = R since Rn − {0} is connected). If m ≥ 2,
∼ ∼
R (R −{0}) = R so HdeR (R −{0}) = R
m−1 m m−1 n
then Exercise 5.4.5 implies that Hde
as well. This is impossible if n = 1 and, if n ≥ 2, Exercise 5.4.5 again implies
that we must have m − 1 = n − 1 so m = n. Thus we have proved the
topological invariance of dimension for Euclidean spaces.
Theorem 5.4.4 Rn and Rm are homeomorphic if and only if n = m.
Remark: Many other classical theorems of Euclidean topology (Brouwer
Fixed Point Theorem, Jordan-Brouwer Separation Theorem, Invariance of
Domain, etc.) are accessible to similar techniques. Those interested in see-
ing some of this are referred to Chapter 7 of [MT]. Similar applications were
treated from the perspective of homology theory in Section 3.4 of [N4].
Exercise 5.4.8 Compute the de Rham cohomology of the torus
T = S 1 × S 1 . Hint: Mayer-Vietoris with the open sets U and V shown below.

U V

U∩V

Exercise 5.4.9 Let p1 , . . . , pl be l distinct points in Rn , n ≥ 2. Show that



 Rl , if k = n − 1

Hde R (R − {p1 , . . . , pl }) ∼
k n
= R, if k = 0 .


 0, if k ̸= 0, n − 1

Exercise 5.4.10 Show that if X and Y are smooth manifolds with Y

∼ k
R (X × Y ) = Hde R (X) for every k.
k
smoothly contractible, then Hde
290 5. de Rham Cohomology

Remark: The cohomology of products in general is described by the

Künneth formula (see Chapter 5, Section 6, of [GHVI]).
Exercise 5.4.11 Show that the de Rham cohomology of a smooth principal
bundle does not, in general, coincide with that of its base manifold.
Remark: For the cohomology of principal bundles, consult the Leray-Hirsch
Theorem (Theorem 5.11, [BT]).

5.5 The Cohomology of Compact,

Orientable Manifolds
Our objective here is to generalize a number of results proved in the last section
for spheres. Specifically, we will show that if X is any compact, connected,
∼
R (X) = R. This result will be used in the
n
orientable n-manifold, then Hde
next section to define the degree of a map between two such manifolds. We
begin with a sequence of lemmas. The first deals with the following situation.
Any ω ∈ Λn (Rn ) is exact. Suppose ω also has compact support. When will
there exist an η ∈ Λn−1 (Rn ), also with compact support, such that ω = dη?
The lemma asserts that the obvious necessary condition is also sufficient.
Lemma 5.5.1 Let ω ∈ Λn (Rn ) have compact support. Then there exists a
compactly supported η ∈ Λn−1 (Rn ) with dη = ω if and only if
∫
ω = 0.
Rn
Proof: Assume first that such an η exists. Then dη is also compactly sup-
ported and therefore integrable on Rn so Stokes’ Theorem gives
∫ ∫
ω= dη = 0.
Rn Rn
∫
Conversely, suppose Rn ω = 0. Let φS : US −→ Rn be the stereographic
projection from the north pole N of S n . This is an orientation preserving
diffeomorphism from US = S n − {N } onto Rn (Exercise 5.10.14, [N4]). More-
over, ω ′ = φ∗S ω is an n-form on US whose compact support is contained in
the complement of the intersection U of S n with some open ball about N in
Rn+1 . We may, via a bump function (Lemma 3.1.3) regard ω ′ as defined on
all of S n . Moreover,
∫ ∫ ∫ ∫
′ ′ ∗
ω = ω = φS ω = ω = 0.
Sn S n −{N } S n −{N } Rn

By Theorem 5.4.3, ω ′ is exact, i.e., there exists a ξ ∈ Λn−1 (S n ) such that

ω ′ = dξ. Moreover, d ξ vanishes on U and U is contractible so there exists an
(n − 2)-form ν on U with ξ = dν on U . With a bump function that is 1 on
a smaller neighborhood V ⊆ U of N in S n we can regard ν as defined on all
 5.5. The Cohomology of Compact, Orientable Manifolds 291

of S n with dν = ξ on V and dν = 0 outside U . Thus, ξ − dν ∈ Λn−1 (S n )

vanishes on V so supp (ξ − dν) ⊆ S n − V . Moreover, d(ξ − dν) = dξ = ω ′
on S n . Denoting the restriction of ξ − dν to S n − {N } by ξ − dν also we let
η = (φ−1 ∗
S ) (ξ − dν). The support of η is contained in φS (S − V ) so it is
n

bounded and therefore compact.

Exercise 5.5.1 Complete the proof by showing that dη = ω. 
Exercise 5.5.2 Use Lemma 5.5.1 to show that there exists a compactly sup-
ported n-form ω ′ on Rn with
∫
ω ′ = 1.
Rn

Lemma 5.5.2 Let X be an orientable n-manifold and U an open subset of X

that is diffeomorphic to Rn . Then there exists an n-form ω on X with compact
support contained in U such that
∫
ω = 1.
X

Proof: Let ϕ be an orientation preserving diffeomorphism of U into Rn and

let ω ′ be the n-form on Rn described in Exercise 5.5.2. Then ϕ∗ ω ′ is an n-form
on U with support supp (ϕ∗ ω ′ ) = ϕ−1 (supp ω ′ ) and this is compact. With
a bump function that is 1 on supp (ϕ∗ ω ′ ) and 0 outside U we may regard
ω = ϕ∗ ω ′ as defined on all of X. Moreover,
∫ ∫ ∫ ∫
ω = ω = ϕ∗ ω ′ = ω ′ = 1. 
X U U Rn

Lemma 5.5.3 Let X be an orientable n-manifold, U an open subset of

X diffeomorphic to R n
∫ and ω an n-form on X with compact support con-
tained in U and with X ω = 1. If ξ is any other n-f orm on X with compact
support contained in U , then there exists a c ∈ R such that ξ is cohomologous
to c ω on X.
Proof: Since ω and ξ have supports contained in U we will use the same
symbols for their restrictions to U . Let ϕ be an orientation preserving diffeo-
morphism of U onto Rn and set
∫ ∫ ∫
c= ξ= ξ= (ϕ−1 )∗ ξ.
X U Rn

Note that ∫ ∫ ∫
−1 ∗
(ϕ ) ω= ω= ω =1
Rn U X
so ∫
(ϕ−1 )∗ (c ω ) = c.
Rn
292 5. de Rham Cohomology

Thus, ∫
(ϕ−1 )∗ (ξ − c ω ) = 0.
Rn

Since (ϕ−1 )∗ (ξ − c ω ) has compact support, Lemma 5.5.1 implies that there
exists a compactly supported η ∈ Λn−1 (Rn ) such that

(ϕ−1 )∗ (ξ − c ω ) = dη

on Rn . Thus,
ξ − c ω = ϕ∗ (dη) = d(ϕ∗ η)
on U . Now, ϕ∗ η has compact support contained in U so, with a bump function
that is 1 on supp (ϕ∗ η) and 0 outside U , we may regard it as an (n − 1)-form
ν on all of X. Then
ξ − c ω = dν
on all of X (because everything is zero outside of U ) so ξ and c ω are coho-
mologous on X. 
Theorem 5.5.4 Let X be a compact. connected, orientable n-manifold. Then
∼
R (X) = R.
n
Hde

Proof: Select an open set U in X diffeomorphic to Rn ∫ and an

ω ∈ Λn (X) with compact support contained in U and such that X ω = 1
(Lemma 5.5.2). We will show for any other n-form ξ on X (which neces-
sarily has compact support since X is compact) there exists a c ∈ R such
R (X) ≤ 1 and the
n
that ξ is cohomologous to c ω. This will give dim Hde
reverse inequality is Lemma 5.1.2. If ξ happens to have its support also con-
tained in U , then our result follows from Lemma 5.5.3. In general, this need
not be the case, of course. However, we claim that we may assume, without
loss of generality, that supp ξ is contained in some open set V ⊆ X that
is diffeomorphic to Rn . To see this, cover X with finitely many coordinate
neighborhoods U1 , . . . , Uk , each diffeomorphic to Rn , and choose a family of
functions ϕ1 , . . . , ϕk for this cover of the type guaranteed by Corollary 3.1.5.
Then ξ = ϕ1 ξ + · · · + ϕk ξ and each ϕi ξ has support contained in Ui . It will
suffice to find ci ∈ R and η i ∈ Λn−1 (X) such that ϕi ξ = ci ω + dη i since then
ξ = (c1 + · · · + ck )ω + d(η 1 + · · · + η k ).
Thus, we assume supp ξ ⊆ V , where V is an open set in X that is diffeo-
morphic to Rn .
Exercise 5.5.3 Show that we may select a sequence of open sets

U = V1 , V2 , . . . , Vk−1 , Vk = V

in X, each diffeomorphic to Rn , with Vi ∩Vi+1 ̸= ∅ for i = 1, 2, . . . , k−1. Hint:

X is connected and therefore pathwise connected (Corollary 1.5.6, [N4]).
 5.5. The Cohomology of Compact, Orientable Manifolds 293

Inside Vi ∩Vi+1 select an open set Ui diffeomorphic to Rn and, by Lemma 5.5.2,

an
∫ n-form ω i on X with compact support contained in Ui and such that
X i
ω = 1. Since supp ω 1 ⊆ U1 ⊆ V1 ∩ V2 ⊆ U , Lemma 5.5.3 gives a real
number c1 such that ω 1 is cohomologous to c1 ω (we will write ω 1 ∼ c1 ω).
But supp ω 1 ⊆ V2 and supp ω 2 ⊆ V2 so the same lemma provides a c2 ∈ R
for which ω 2 ∼ c2 ω 1 and therefore ω 2 ∼ (c1 c2 )ω. Continuing inductively we
obtain
ω k−1 ∼ (c1 c2 · · · ck−1 )ω.
But then supp ξ ⊆ V = Vk and supp ω k ⊆ Vk implies that there exists a
ck ∈ R such that ξ ∼ ck ω k−1 and therefore
ξ ∼ (c1 c2 · · · ck−1 ck )ω
as required. 
Corollary 5.5.5 Let X be a compact,∫ orrientable n-manifold. Then an n-
form ω on X is exact if and only if X ω = 0.
Exercise 5.5.4 Prove Corollary 5.5.5. Hint: See the proof of
Theorem 5.4.3. 
Corollary 5.5.6 Let X be a compact, orientable n-manifold. Then
∫
R (X) −→ R
n
: Hde
X

is an isomorphism.
Exercise 5.5.5 Prove Corollary 5.5.6.
Exercise 5.5.6 Suppose X is a compact, connected orientable n-manifold,
where n = 2k for some k ≥ 1. Define QX : Hde
k
R (X) × Hde R (X) −→ R by
k

∫
QX ([ξ], [η]) = ξ ∧ η.
X

Show that QX is well-defined, bilinear and satisfies QX ([η], [ξ]) = (−1)k

QX ([ξ], [η]). In particular, when n ≡ 0 mod 4, QX is symmetric. QX is called
the intersection form of X. Another definition of the intersection form for
4-manifolds was introduced in Appendix B of [N4].
We conclude this section with a few remarks on matters of consider-
able importance, but which we do not intend to pursue in detail. We have
shown that the 0th and nth deRham cohomology groups of a compact, con-
nected, orientable n-manifold X are always isomorphic to R. In particular,
0
Hde ∼ n
R (X) = Hde R (X). This is a special case of a more general result known
as Poincaré Duality which asserts that, for any k = 0, 1, . . . , n,
n−k
Hde ∼ k
R (X) = Hde R (X).
294 5. de Rham Cohomology

Remark: The Poincaré Duality Theorem is actually much more general

than we have indicated here. Those interested in pursuing this should consult
either Chapter 13 of [MT], or Chapter 11, Volume I, of [Sp2].
Any n-manifold X of finite type has finite dimensional cohomology (Theo-
rem 5.4.1) so we can define, for each k = 0, 1, . . . , n,

bk = dim Hde
k
R (X)

and form their alternating sum to obtain

∑
n
b0 − b1 + b2 − · · · + (−1)n bn = (−1)k dim Hde
k
R (X).
k=0

Although it is not obvious, this alternating sum coincides with the Euler
characteristic of X, defined in Chapter 3 of [N4] as the alternating sum of
the ranks of the singular homology groups of X. We will have a bit more to
say about the Euler characteristic in the Appendix.

5.6 The Brouwer Degree

Suppose X and Y are two compact, connected, orientable, n-dimensional man-
ifolds and f : X −→ Y is a smooth map. ∫ By Lemma 5.5.2 we may select a (nec-
essarily closed) n-form ω 0 on Y with Y ω 0 = 1. Corollary 5.5.5 implies that
ω 0 is not exact and determines a nontrivial cohomology class [ω 0 ] ∈ Hde
n
R (Y ).
n
But Theorem 5.5.4 then implies that [ω 0 ] generates Hde R (Y ). We call ω 0 a
n
normalized generator for Hde R (Y ). Pull ω 0 back to X by f and integrate
over X to obtain a real number that we will call the (Brouwer) degree of
f and denote
∫
deg(f ) = f ∗ ω0 . (5.6.1)
X

We must show that this definition does not depend on the choice of the nor-
n ′
malized ∫generator ω 0 for Hde R (Y ). Thus, suppose ω 0 is another n-form on
′
Y with Y ω 0 = 1. Since [ω 0 ] generates Hde R (Y ) there exists an α ∈ R such
n

that
ω ′0 = αω 0 + dη

for some η ∈ Λn−1 (Y ).

Exercise 5.6.1 Show that α = 1.
Thus, ω ′0 = ω 0 + dη so f ∗ ω ′0 = f ∗ ω 0 + d(f ∗ ω 0 ) and, by Stokes’ Theorem,
∫ ∫
f ∗ ω ′0 = f ∗ ω0
X X
 5.6. The Brouwer Degree 295

as required. Now consider any other (necessarily closed) n-form ω on Y . Again,

R (Y ), there exists an α ∈ R such that
n
because [ω 0 ] generates Hde

ω = αω 0 + dη

for some η ∈ Λn−1 (Y ).

∫
Exercise 5.6.2 Show that α = Y
ω.
Thus, f ∗ ω = αf ∗ ω 0 + d(f ∗ η 0 ) so Stokes’ Theorem gives
∫ ∫
f ∗ω = α f ∗ ω 0 = α deg(f )
X X

which we write as ∫ ∫
∗
f ω = deg(f ) ω. (5.6.2)
X Y

Notice that (5.6.1) is the special case of (5.6.2) in which ω is a normalized

n
generator for Hde R (Y ).

Remark: Although it is by no means apparent from the definition we have

just given, we will prove shortly that deg(f ) is actually an integer.
Exercise 5.6.3 Let X, Y and Z be compact, connected orientable
n-manifolds and f : X −→ Y and g : Y −→ Z smooth maps. Show that

deg(g ◦ f ) = deg(g) deg(f ).

Concrete calculations are greatly facilitated by the fact that deg is a homotopy
invariant.
Theorem 5.6.1 Let X and Y be compact, connected orientable n-manifolds
and suppose f, g : X −→ Y are two smooth maps that are smoothly homotopic.
Then
deg(f ) = deg(g).

Proof: According to Corollary 5.2.3, f and g induce the same maps in

cohomology:
f # = g # : Hde R (Y ) −→ Hde R (X).
n n

n
Thus, given any normalized generator ω 0 for Hde R (Y ) we have
# # ∗ ∗
f ([ω 0 ]) = g ([ω 0 ]) so [f ω 0 ] = [g ω 0 ] and, consequently,

f ∗ ω 0 = g ∗ ω 0 + dη

for some η ∈ Λn−1 (X). Stokes’ Theorem then gives

∫ ∫
f ∗ ω0 = g∗ ω0 ,
X X
296 5. de Rham Cohomology

i.e., deg(f ) = deg(g). 

Since a constant map from X to Y obviously has degree 0 we obtain the
following consequence of Theorem 5.6.1.
Corollary 5.6.2 Let X and Y be compact, connected orientable n-manifolds.
Then any smoothly nullhomotopic map from X to Y has degree zero.
Corollary 5.6.3 Let X be a compact, connected orientable manifold. Then
idX : X −→ X is not smoothly nullhomotopic (so X is not smoothly con-
tractible).
Exercise 5.6.4 Prove Corollary 5.6.3. 
We now consider the special case in which the range of our map is a sphere.
Specifically, we let X be a compact, connected, orientable n-manifold and

f : X −→ S n

a smooth map. We claim first that if f is not surjective, then it must be

smoothly nullhomotopic and therefore have degree zero. To see this we pro-
ceed as follows: Since f is not surjective we may assume, without loss of
generality, that f (X) does not contain the north pole N of S n (just choose
the coordinate system in Rn+1 properly). Consequently, f maps into the
domain of the stereographic projection φS and we can define a smooth
map
φS ◦ f : X −→ Rn .
Exercise 5.6.5 Show that φS ◦ f is smoothly nullhomotopic. Hint: Consider
the map H(x, t) = (1 − t)(φS (f (x))).
Exercise 5.6.6 Show that f is smoothly nullhomotopic. Hint: Consider
φ−1
S ◦ H, where H is as in Exercise 5.6.5.

Theorem 5.6.4 Let X be a compact, connected orientable n-manifold and

f : X −→ S n a nonsurjective smooth map. Then f is smoothly nullhomotopic
and so deg(f ) = 0.
Exercise 5.6.7 Let X be a compact, connected orientable manifold and
f : X −→ X a diffeomorphism of X onto itself. Show that deg(f ) is 1 if f is
orientation preserving and −1 if f is orientation reversing.
Exercise 5.6.8 Any element A of O(n + 1) determines a linear map of
Rn+1 onto itself which, when restricted to the unit sphere, gives a diffeo-
morphism of S n onto S n . Show that the degree of this map is det A. Conclude
that the degree of the antipodal map A : S n −→ S n given by A(p) = −p is
(−1)n+1 .
Remark: In Section 3.4 of [N4] there is a definition of the Brouwer degree
of a continuous map from S n to S n based on the fact that the nth singular
 5.6. The Brouwer Degree 297

homology of S n is isomorphic to Z. In this case the degree is an integer by

definition. Our current definition in the smooth category does not obviously
result in an integer; it does result in an integer, but just not obviously so. We
will prove this now. What we will not prove (but is true nonetheless) is that,
for smooth maps from S n to S n , the two definitions give the same integer.
We conclude our discussion of the degree of a map by describing another way
to calculate it and, in the process, showing that it is always an integer. Thus,
we consider two compact, connected, oriented n-manifolds X and Y and a
smooth map f : X −→ Y . Let q ∈ Y be a regular value of f .
Remark: According to Theorem 3 − 14 of [Sp1], the set of critical values
of a smooth map g : A −→ Rn , A open in Rn , has measure zero in Rn .
In particular, regular values always exist. Applying this to any coordinate
expression for f we find that it too must have regular values.
Then f −1 (q) is a compact submanifold of X of dimension n − n = 0, i.e.,
a finite set of points {p1 , . . . , pk } (perhaps ∅). If f −1 (p) = {p1 , . . . , pk } ̸= ∅,
define, for each i = 1, . . . , k, the sign of f at pi by
{
1 if f∗pi is orientation preserving
sign(f, pi ) =
−1 if f∗pi is orientation reversing

(note that f∗pi : Tpi (X) −→ Tf (pi ) (Y ) is an isomorphism). We claim that

 k

 ∑

 sign (f, pi ), if f −1 (q) = {p1 , . . . , pk } ̸= ∅
deg(f ) = i=1 (5.6.3)

 .


0, if f −1 (q) = ∅

In particular, deg(f ) is an integer (and is clearly zero whenever f is not

surjective since any q ∈ Y − f (X) is a regular value). Observe also that,
although it is not obvious at the moment that the integer on the right-hand
side of (5.6.3) is independent of the choice of q, this will follow once (5.6.3) is
proved.
To prove (5.6.3) we first suppose f −1 (q) = ∅. Since X is compact, f (X)
is closed in Y so Y − f (X) is an open set. Y − f (X) is nonempty (since it
contains q) and so it contains an open set U diffeomorphic to Rn . Lemma
∫ 5.5.2
provides an n-form ω 0 on Y with support contained in U such that Y ω 0 = 1.
In particular, ω 0 is zero on f (X) so f ∗ ω 0 is the zero n-form on X. Thus,
∫
deg(f ) = f ∗ ω 0 = 0.
X

Now suppose f −1 (q) = {p1 , . . . , pk }. Choose coordinate neighborhoods

U1 , . . . , Uk in X such that pi ∈ Ui , i = 1, . . . , k, Ui ∩ Uj = ∅ if i ̸= j
298 5. de Rham Cohomology

and such that f is a diffeomorphism on each Ui (since q is a regular value

and dim X = dim Y , the Inverse Function Theorem applies at each pi ).
We now wish to find a coordinate neighborhood V of q in Y such that
f −1 (V ) = U1 ∪ · · · ∪ Uk . First choose some compact neighborhood W of q
in Y and consider

W ′ = f −1 (W ) − (U1 ∪ · · · ∪ Uk ).
This is closed in X and therefore compact. Consequently, f (W ′ ) is a closed
set in Y that does not contain q. Select a coordinate neighborhood V of q with
V ⊆ W − f (W ′ ). Then f −1 (V ) ⊆ U1 ∪ · · · ∪ Uk . Now, if necessary, replace
each Ui by Ui ∩ f −1 (V ) (still a coordinate neighborhood) so that
f −1 (V ) = U1 ∪ · · · ∪ Uk .
Since V must contain an open set diffeomorphic to R∫n , Lemma 5.5.2 gives
an n-form ω 0 on Y with support contained in V and Y ω 0 = 1. Moreover,
supp (f ∗ ω 0 ) ⊆ U1 ∪ · · · ∪ Uk so, since this union is disjoint,
∫ k ∫
∑
f ∗ ω0 = f ∗ ω0 .
X i=1 Ui

But f is a diffeomorphism on each Ui so

∫ ∫
f ∗ ω0 = ± ω0
Ui f (Ui )
∫ ∫
=± ω0 = ± ω 0 = ±1,
V Y

where the plus sign is chosen if f is orientation preserving on Ui and the minus
sign is chosen if f is orientation reversing on Ui . Thus,
∫ {
∗ 1 if f∗pi is orientation preserving
f ω0 =
Ui −1 if f∗pi is orientation reversing
= sign(f, pi )

so the result follows. 

The formula (5.6.3) for deg(f ) provides some intuitive insight into the ge-
ometrical significance of the Brouwer degree that is not apparent from our
definition (5.6.1). For any regular value q of f, f −1 (q) is a finite set of points,
but the number of elements in it (i.e., the number of times q is “covered” by
f ) generally depends on q. Thus, f generally does not cover all of its regular
values the same number of times. However, if we “count” the preimages of a
regular value properly (with multiplicity 1 when f is orientation preserving
there and with multiplicity −1 when f is orientation reversing there) the re-
sult is the same for every regular value. This counting with multiplicity simply
 5.7. The Hopf Invariant 299

cancels two preimages if f maps them onto q in “opposite directions” (picture

this for maps from S 1 to S 1 ). Thus, one thinks of deg(f ) as the “net” number
of times f covers each of its regular values.
Exercise 5.6.9 Identify S 1 with the set of complex numbers z of modulus 1
and let n be a positive integer. Define f : S 1 −→ S 1 and g : S 1 −→ S 1 by
f (z) = z n and g(z) = z̄ n . Show that deg(f ) = n and deg(g) = −n.
Remark: Identifying S 3 with the set of quaternions q of modulus 1 one can
similarly define, for any positive integer n, f : S 3 −→ S 3 and g : S 3 −→ S 3
by f (q) = q n and g(q) = q̄ n . In Section 6.4 we will compute the degrees of
these maps and find that deg(f ) = n and deg(g) = −n.

The Brouwer degree is a homotopy invariant of maps between compact,

connected, orientable n-manifolds. We will conclude this chapter with a very
brief look at another homotopy invariant for maps from S 2n−1 to S n , n ≥ 2,
which has played a significant role in the development of modern topology.

5.7 The Hopf Invariant

Consider a smooth map f : S 2n−1 −→ S n , n ≥ 2, and let ω 0 be a nor-
n n ∗ 2n−1
malized generator for Hde R (S ). Then f ω 0 is an n-form on S . But
2n−1 ∼
n
Hde R (S ) = 0 so there exists an (n − 1)-form ω on S 2n−1
such that
dω = f ∗ ω 0 . We define the Hopf invariant of f by
∫
H(f ) = ω ∧ dω.
S 2n−1

To see that the definition does not depend on the choice of ω, suppose ω ′ is
another (n − 1)-form on S 2n−1 with dω ′ = f ∗ ω 0 . Then d(ω − ω ′ ) = 0 so
∫ ∫ ∫
′ ′
ω ∧ dω − ω ∧ dω = (ω − ω ′ ) ∧ dω
S 2n−1 S 2n−1
∫ S 2n−1

= d ((ω − ω ′ ) ∧ ω ) = 0
S 2n−1

by Stokes’ Theorem.
Exercise 5.7.1 Show that if n is odd, then H(f ) = 0 for any smooth map
f : S 2n−1 −→ S n . Hint: Compute d(ω ∧ ω) for any even dimensional form ω.
We show now that the Hopf invariant is a smooth homotopy invariant.
Theorem 5.7.1 Let f0 , f1 : S 2n−1 −→ S n , n ≥ 2, be smoothly homotopic
maps. Then H(f0 ) = H(f1 ).
300 5. de Rham Cohomology

Proof: Let F : S 2n−1 × (0 − ϵ, 1 + ϵ) −→ S n , be a smooth map with

F (x, 0) = f0 (x) and F (x, 1) = f1 (x) for all x ∈ S 2n−1 . Also let

ι0 : S0 = S 2n−1 × {0} ,→ S 2n−1 × (0 − ϵ, 1 + ϵ)

and
ι1 : S1 = S 2n−1 × {1} ,→ S 2n−1 × (0 − ϵ, 1 + ϵ)
be the inclusion maps. Identifying S0 and S1 with S 2n−1 we may also identify
F ◦ ι0 with f0 and F ◦ ι1 with f1 , respectively. Now let ω 0 be a normalized
∗
generator for Hde n n
R (S ). Then F ω 0 is an n-form on S
2n−1
× (0 − ϵ, 1 + ϵ). By
n
Theorem 5.2.6, Hde R (S 2n−1
× (0 − ϵ, 1 + ϵ)) ∼ n
= Hde R (S 2n−1
) and, since n ≥ 2,
this is trivial. Consequently, there is an (n − 1)-form η on S 2n−1 × (0 − ϵ, 1 + ϵ)
such that
F ∗ ω 0 = dη.
Let ι∗0 η = ω and ι∗1 η = ω ′ . Notice that

f0∗ ω 0 = (F ◦ ι0 )∗ ω 0 = ι∗0 (F ∗ ω 0 ) = ι∗0 (dη) = d(ι∗0 η) = dω

and, similarly,
f1∗ ω 0 = dω ′ .
Moreover,

ω ∧ dω = (ι∗0 η) ∧ d(ι∗0 η) = (ι∗0 η) ∧ ι∗0 (dη) = ι∗0 (η ∧ dη)

and, similarly,
ω ′ ∧ dω ′ = ι∗1 (η ∧ dη).
We must show that ∫ ∫
ω ∧ dω = ω ′ ∧ dω ′ ,
So S1

i.e., that ∫ ∫
ι∗0 (η ∧ dη) = ι∗1 (η ∧ dη). (5.7.4)
S0 S1

Now, S 2n−1 × [0, 1] is a domain with smooth boundary in

S 2n−1 × (0 − ϵ, 1 + ϵ) and its boundary is ∂(S 2n−1 × [0, 1]) = S0 ∪ S1 . Note
also that

d(η ∧ dη) = dη ∧ dη = F ∗ ω 0 ∧ F ∗ ω 0 = F ∗ (ω 0 ∧ ω 0 ) = 0

because ω 0 ∧ ω 0 is a 2n-form on the n-dimensional manifold S n .

Exercise 5.7.2 Prove (5.7.1) by applying Stokes’ Theorem to the integral
over S 2n−1 × [0, 1] of the restriction of d(η ∧ dη). Hint: S0 and S1 acquire
opposite orientations from S 2n−1 × [0, 1]. 
 5.7. The Hopf Invariant 301

Exercise 5.7.3 Show that a nullhomotopic map f : S 2n−1 −→ S n , n ≥ 2,

has Hopf invariant zero.

Remark: As was the case for the Brouwer degree there is another, more
geometrical way of calculating the Hopf invariant which shows that H(f ) is
actually an integer. This involves the linking number of the preimages f −1 (p)
and f −1 (q) of any two distinct regular values of f (see Chapter III, Section 17,
of [BT]). Still other approaches to the Hopf invariant are described in [Huse].
We conclude with a few exercises that will lead the reader through the
calculation of the Hopf invariant for the projection map

P1 : S 3 −→ S 2

of the complex Hopf bundle. The result will be H(P1 ) = 1 so that, in partic-
ular, Exercise 5.7.3 then implies that P1 is not nullhomotopic. This was, in
fact, the first example of a homotopically nontrivial map of one sphere onto
another of smaller dimension and came as quite a shock in the 1930s when
Hopf constructed it. Recall that if S 3 is identified with the subset of C2 con-
sisting of those (z1 , z2 ) for which |z1 |2 + |z2 |2 = 1 and S 2 is identified with a
subset of R3 , then
( )
P1 (z1 , z2 ) = z 1 z̄ 2 + z̄ 1 z 2 , −i z 1 z̄ 2 + iz̄ 1 z 2 , |z 1 |2 − |z 2 |2 .

For the calculation of the Hopf invariant we will take the normalized gener-
n 2 2
ator ω 0 of Hde R (S ) to be the standard volume form for S divided by the
“volume” 4π of S (see Section 4.6). Recall that, if x , x and x3 are standard
2 1 2

coordinates on R3 and ι : S 2 ,→ R3 is the inclusion, then

1 ∗ 1 2
ω0 = ι (x dx ∧ dx3 − x2 dx1 ∧ dx3 + x3 dx1 ∧ dx2 ).
4π
Exercise 5.7.4 Denote the standard coordinates on R4 by y 1 , y 2 , y 3 , y 4
and define p : R4 −→ R3 by
(
p (y 1 , y 2 , y 3 , y 4 ) = 2y 1 y 3 + 2y 2 y 4 , 2y 2 y 3 − 2y 1 y 4 ,
)
( )2 ( )2 ( )2 ( )2
y1 + y2 − y3 − y4 .

Identify S 3 with a subset of R4 via (z 1 , z 2 ) = (y 1 + i y 2 , y 3 + i y 4 ) −→

(y 1 , y 2 , y 3 , y 4 ) and let ι′ : S 3 ,→ R4 be the inclusion. Show that p ◦ ι′ = ι ◦ P1
and therefore
1 ′ ∗( ∗( 1 2 ))
P1∗ ω 0 = (ι ) p x dx ∧ dx3 − x2 dx1 ∧ dx3 + x3 dx1 ∧ dx2 .
4π
302 5. de Rham Cohomology

Exercise 5.7.5 Show that

1
P1∗ ω 0 = − (ι′ )∗ (dy 1 ∧ dy 2 + dy 3 ∧ dy 4 ).
π
Exercise 5.7.6 Show that dy 1 ∧ dy 2 + dy 3 ∧ dy 4 = d(y 1 dy 2 + y 3 dy 4 ). Then
define ω by ω = − π1 (ι′ )∗ (y 1 dy 2 + y 3 dy 4 ) and show that dω = P1∗ ω 0 .
1 ′ ∗ 1 2
Exercise 5.7.7 Show that ω ∧ dω = (ι ) (y dy ∧ dy 3 ∧ dy 4 + y 3 dy 1 ∧
π2
dy 2 ∧ dy 4 ).
Exercise 5.7.8 Compute
∫
H(P1 ) = ω ∧ dω = 1.
S3
6
Characteristic Classes

6.1 Motivation
We have had a number of previous, albeit informal encounters with the notion
of a characteristic class for a principal bundle (in Section 2.2, 2.4 and 2.5) and
are now in possession of sufficient machinery to take up the subject in earnest.
Before plunging into the thick of the battle, however, we would like to make
clear the strategy we propose to adopt. This is perhaps best achieved by
going through what might be regarded as the “trivial” case in detail and then
indicating the modifications required when matters are not so simple.
P
We begin then by considering a principal U (1)-bundle U (1) ,→ P −→ X
over a manifold X and let ω denote a connection on the bundle. Thus,
one might have in mind an electromagnetic field on spacetime. Since U (1)
is Abelian, the curvature is given by Ω = dω. For any local cross-section
s : V −→ P −1 (V ) we write, as usual, A = s∗ ω = −i A and F =
s∗ Ω = −i F for the local gauge potential and field strength. Again be-
cause U (1) is Abelian, the field strengths F for various cross-sections s
agree on the intersections of their domains and thereby determine a glob-
ally defined u(1)-valued 2-form on X that we also denote by F .
This F is −i F for a globally defined real-valued 2-form F on X. Since
1
dF = 0, F is closed and therefore 2π F determines an element of
Hde2
R (X; R).
Now suppose ω ′ is another connection on the same bundle
P
U (1) ,→ P −→ X determining global 2-forms F ′ and F ′ in the same way. It
follows that ω − ω ′ = τ is in Λ1ad (P, u(1)). Thus, dω − dω ′ = dτ so

Ω − Ω ′ = dτ .
Since τ is tensorial of type ad and U (1) is Abelian, σg∗ τ = g −1 ·τ = gτ g −1 = τ
for every g ∈ U (1). But then Lemma 4.5.1. implies that τ projects to X, i.e.,
that there exists a unique u(1)-valued 1-form τ̄ on X such that τ = P ∗ τ̄ .
Since (in the Abelian case) curvatures also project to X, Ω = P ∗ F and
Ω ′ = P ∗ F ′ and therefore P ∗ F − P ∗ F ′ = d(P ∗ τ̄ ) so

P ∗ (F − F ′ ) = P ∗ (dτ̄ ).
But projections to X are unique when they exist so we must have

F − F ′ = dτ̄ .

G.L. Naber, Topology, Geometry and Gauge fields: Interactions, 303

Applied Mathematical Sciences 141, DOI 10.1007/978-1-4419-7895-0_6,
© Springer Science+Business Media, LLC 2011
304 6. Characteristic Classes

1
In particular, 2π 1
F and 2π F ′ are cohomologous and therefore determine the
same element of Hde R (X; R). The unique cohomology class determined in
2
P
this way from any connection on U (1) ,→ P −→ X is called the 1 st Chern
class of the bundle and is denoted
[ ]
1
F ∈ Hde R (X; R).
2
c1 (P ) =
2π

The 1 st Chern class of our U (1)-bundle is determined entirely by the bundle

itself and not by the particular connection on that bundle from which it is
constructed. It is true, but by no means obvious, that c1 (P ) is actually “char-
acteristic” of the bundle in the sense that two principal U (1)-bundles over
the same manifold X are equivalent if and only if their 1 st Chern classes are
the same. Since this is easy to prove when X is a sphere S n , we shall do so.
The only interesting case is n = 2. The reason is that, for n = 1, 3, 4, 5, . . . ,
every U (1)-bundle over S n is trivial (for n = 1, see Exercise 4.4.11 of [N4];
for n = 3, 4, 5, . . . , use the Classification Theorem for bundles over spheres
and Corollary 2.5.11 of [N4] which asserts that πn (U (1)) is trivial for n ≥ 2).
Since a trivial bundle admits a flat connection (Exercise 6.2.12, [N4]), its 1 st
Chern class must be zero so we have the (admittedly, rather silly) result that
two U (1)-bundles over S n , n = 1, 3, 4, 5, . . . , are equivalent if and only if their
1 st Chern classes are the same. The n = 2 case takes just slightly more work.
We know all of the principal U (1)-bundles over S 2 . They are in one-
to-one correspondence with the elements of π1 (U (1)) ∼ = Z (Classification
Theorem). One can describe them explicitly in terms of transition functions
as follows: Trivialize the bundle over UN and US and consider the map
gSN : UN ∩ US −→ U (1) given by gSN (φ, θ) = e−nθi , where n ∈ Z and
(φ, θ) are standard spherical coordinates on S 2 . This map determines a prin-
cipal U (1)-bundle over S 2 for each n and two such bundles are equivalent if
and only if they correspond to the same n. There are explicit descriptions of
all of these bundles in Section 2.2 where we have also written out a connec-
tion (Dirac Monopole) on each. These, in fact, exhaust all of the (equivalence
classes of) principal U (1)-bundles over S 2 . The reason is that any such bun-
dle is determined up to equivalence by the homotopy class of its characteristic
map gSN | S 1 (Theorem 4.4.2, [N4]) and every map from S 1 to S 1 is homo-
topic to e−nθi for some n (see the proof of Theorem 2.4.4, [N4]). To show
that these bundles are also determined up to equivalence by their 1 st Chern
classes we will show that the integer n can be retrieved from c1 (P ) just by
integrating it over S 2 .
Any representative of the cohomology class c1 (P ) is a smooth 2-form
on the compact manifold S 2 and so is integrable. Furthermore, all such
representatives have the same integral over S 2 by Stokes’ Theorem. Since we
know a connection on the bundle with transition function gSN (φ, θ) = e−nθi
we could simply compute the corresponding representative and integrate it.
It is much more instructive, however, to proceed indirectly. Thus, we assume
 6.1. Motivation 305

only that we have some connection on the bundle with local gauge potentials
AN and AS on UN and US , respectively. Then, on UN ∩ US ,

−1 −1
AN = gSN AS gSN + gSN dgSN

so
AN = AS − i n d θ.

The corresponding field strengths are denoted F N and F S , respectively.

To integrate 2π 1 i
F = 2π F over S 2 we will write S 2 = S+ 2
∪ S−2
as the union
of the upper (z ≥ 0) and lower (z ≤ 0) hemispheres and denote by S 1 the
equatorial (z = 0) circle with ι : S 1 ,→ S 2 the inclusion map. Provide S 1 with
2
the orientation it inherits from the standard orientation of S+ (“counterclock-
2
wise”) and note that it receives the opposite orientation from S− . Applying
Stokes’ Theorem twice then gives
∫ ∫ ∫
i i i
F= FN + FS
S 2 2π 2π S+2 2π S− 2

∫ ∫
i i
= ι∗ AN − ι∗ AS
2π S 1 2π S 1
∫ ∫
i i
= ι∗ (AS − i n d θ) − ι∗ AS
2π S 1 2π S 1
∫ ∫ ∫
i n i
= ι∗ AS + ι∗ d θ − ι∗ AS
2π S 1 2π S 1 2π S 1
∫
n
= d(θ ◦ ι).
2π S 1

The essential feature of the calculation to this point is that all references to
the connection itself have dropped out and we are left with an integral that in-
−1
volves only the transition function (gSN dgSN = −i n d θ). This is as it should
be, of course, since the Chern class (and therefore its integral) depends only
on the bundle and not on the connection. Calculating the remaining integral
is easy and, in fact, was done explicitly in Section 4.6. The result was simply
the length (“volume”) of S 1 :
∫
d(θ ◦ ι) = 2π.
S1

Thus ∫
n
c1 (P ) = (2π) = n
S2 2π
as promised.
This then concludes our discussion of the “trivial” case and the issue before
us now is whether or not it is possible to push through an analogous program
306 6. Characteristic Classes

for bundles with other structure groups. On the surface, the prospects do not
appear to be good since virtually every stage in our construction depended on
the commutativity of U (1). It is not even altogether clear how to get the ball
rolling since the very existence of the 2-form on X whose cohomology class
contained the topological information about the bundle required that local
field strengths agree on the intersections of their domains. Since, in general,
such local field strengths are related by F g = g −1 F g, this simply is not
true when the structure group is non-Abelian. The key idea here is to evade
this difficulty by considering, not the field strengths themselves, but various
functions of the field strength that cannot tell the difference between F and
g −1 F g. Identifying all of these objects with matrices there are a number of
obvious choices. For example, since the trace of a matrix is invariant under
conjugation, trace (g −1 F g) = trace F and the 2-forms defined locally on X
by taking the trace of the local field strengths will agree on the intersections
of their domains and therefore determine a globally defined 2-form trace F on
X. Perhaps trace F is closed and so determines a cohomology class. Perhaps
this cohomology class is independent of the connection from which trace F
was constructed. Perhaps the same is true for other ad-invariant functions of
F such as trace (F ∧ F ), or det F . Perhaps this is all a bit too much to ask.
We shall see.

6.2 Algebraic Preliminaries

We begin by considering a finite dimensional real vector space V (typically,
this will be a Lie algebra for some matrix Lie group). If k ≥ 1 is an integer,
then a map f˜ : V × · · · × V −→ C is k -multilinear if it is linear in each
k

variable separately and symmetric if f˜(vσ(1) , . . . , vσ(k) ) = f˜(v1 , . . . , vk ) for

all v1 , . . . , vk in V and all permutations σ ∈ Sk . The real and imaginary parts
of such an f˜ are therefore just symmetric k-tensors on V. The set S k (V) of
all such maps has the obvious pointwise structure of a complex vector space
and, setting S 0 (V) = C, we define
∞
⊕
S(V) = S k (V),
k=0

(see the Remark on page 302). If f˜ ∈ S k (V) and g̃ ∈ S l (V) we define

f˜ ⊙ g̃ by
( )
f˜ ⊙ g̃ (v1 , . . . , vk , vk+1 , . . . , vl )
(6.2.1)
1 ∑ ( ) ( )
= f˜ vτ (1) , . . . , vτ (k) g̃ vτ (k+1) , . . . , vτ (k+l) .
(k + l)!
τ ∈Sk+l
 6.2. Algebraic Preliminaries 307

Exercise 6.2.1 Show that f˜ ⊙ g̃ ∈ S k+l (V). Extend ⊙ to S(V) and show
that, with this operation, S(V) has the structure of a commutative algebra
with identity.
If {e1 , . . . , en } is a basis for V and {e1 , . . . , en } is the dual basis for V ∗ , then
for any f˜ ∈ S k (V), there exist unique complex numbers ai1 ...ik , i1 , . . . , ik =
1, . . . , n, symmetric in i1 , . . . , ik , such that

f˜ = ai1 ··· ik ei1 ⊗ · · · ⊗ eik (summation convention).

Writing v1 = xi11 ei1 , . . . , vk = xikk eik we therefore have

f˜(v1 , . . . , vk ) = ai1 ···ik xi11 · · · xikk .

Now define f : V −→ C by f (v) = f˜(v, . . . , v) so that, if v = xi ei ,

f (v) = ai1 ···ik xi1 · · · xik .

Thus, f is a homogeneous polynomial of degree k in the components of v. For

convenience we write
f = ai1 ···ik ei1 · · · eik ,
where the function on the right-hand side is defined by
( )
ai1 ···ik ei1 · · · eik (v) = ai1 ···ik ei1 (v) · · · eik (v) = ai1 ···ik xi1 · · · xik .

More generally, let us say that a map f : V −→ C is a homogeneous

polynomial of degree k on V if it can be expressed as a homogeneous
polynomial of degree k in e1 , . . . , en in the sense that there exist ai1 ···ik ∈
C, i1 , . . . , ik = 1, . . . , n, symmetric in i1 , . . . , ik , such that f = ai1 ···ik ei1 · · · eik .
Exercise 6.2.2 Show that this definition does not depend on the choice of
basis.
The collection of all such homogeneous polynomials of degree k on V, with its
obvious pointwise linear structure, is denoted P k (V). Setting P 0 (V) = C we
define
⊕∞
P (V) = P k (V).
k=0

Exercise 6.2.3 Show that, if f ∈ P (V) and g ∈ P l (V), then the product f g
k

defined by (f g)(v) = f (v)g(v) is in P k+l (V). Extend this product structure

to P (V) and show that P (V) thereby becomes a commutative algebra with
identity.
We show next that the two algebras S(V) and P (V) are, in fact, isomorphic.
Begin by fixing a k ≥ 1. Then any f˜ ∈ S k (V) gives rise to an f ∈ P k (V) defined
by f (v) = f˜(v, . . . , v). If {e1 , . . . , en } is any basis for V and {e1 , . . . , en } is the
308 6. Characteristic Classes

dual basis for V ∗ , then this assignment is given by

ai1 ···ik ei1 ⊗ · · · ⊗ eik −→ ai1 ···ik ei1 · · · eik

and is clearly linear and surjective. To prove that this is an isomorphism we

must show that f˜ is uniquely determined by f and this involves the familiar
process of polarization. There are several useful ways to view this process and
we will begin by illustrating in the k = 2 case. Thus, suppose we are given an
f ∈ P 2 (V). To show that the f˜ ∈ S 2 (V) for which f (v) = f˜(v, v) is unique we
will prove that, in fact,

1[ ]
f˜(v, w) = f (v + w) − f (v) − f (w)
2

for all v, w ∈ V. Indeed, one need only compute f˜(v+w, v+w) using bilinearity

f˜(v + w, v + w) = f˜(v, v) + 2f˜(v, w) + f˜(w, w)

to obtain
f (v + w) = f (v) + 2f˜(v, w) + f (w).

There are analogous formulas for larger k, e.g., for f ∈ P 3 (V),

1[
f˜(u, v, w) = f (u + v + w) − f (u + v) − f (u + w)
6 ]
− f (v + w) + f (u) + f (v) + f (w) .

Another way of arriving at f˜ when k = 2 is as follows: Write f = aij ei ej and

expand f (sv + tw) as a polynomial in s and t.

f (sv + tw) = aij (sv i + twi )(sv j + twj )

= (aij v i v j )s2 + (2aij v i wj )st + (aij wi wj )t2

= (f (v))s2 + (2f˜(v, w))st + (f (w))t2

Thus, f˜(v, w) is 1
2 times the coefficient of st in the expansion of f (sv + tw).
Exercise 6.2.4 Show that, in the general case, f˜(v1 , . . . , vk ) is k! 1
times the
coefficient of t1 t2 · · · tk in the expansion of f (t1 v1 + t2 v2 + · · · + tk vk ).
The linear isomorphisms f˜ −→ f of S k (V) onto P k (V), together with
the identity map of S 0 (V) onto P 0 (V), gives a linear isomorphism of S(V)
onto P (V). To show that this is, in fact, an algebra isomorphism it will suf-
fice to prove that, if f˜ ∈ S k (V) with f˜ −→ f ∈ P k (V) and g̃ ∈ S l (V)
 6.2. Algebraic Preliminaries 309

with g̃ −→ g ∈ P l (V), then f˜ ⊙ g̃ ∈ S k+l (V) is sent to f g ∈ P k+l (V).

But f˜ ⊙ g̃ is sent to the element of P k+l (V) whose value at v is, by (6.2.1),
( ) 1 [ ]
f˜ ⊙ g̃ (v, . . . , v, v, . . . , v) = (k + l)!f˜(v, . . . , v)g̃(v, . . . , v)
(k + l)!
= f (v)g(v)
= (f g)(v)

as required.
Now, suppose that G is a Lie group and ρ : G −→ GL(V) is a representation
of G on V. For each k ≥ 1, let Sρk (V) and Pρk (V) denote the subspaces of S k (V)
and P k (V), respectively, that are invariant under ρ, i.e., satisfy

f˜(ρ(g)(v1 ), . . . , ρ(g)(vk )) = f˜(v1 , . . . , vk )

and
f (ρ(g)(v)) = f (v)

for all g ∈ G and all v, v1 , . . . , vk in V. The isomorphism f˜ −→ f clearly

carries Sρk (V) onto Pρk (V). Furthermore, each of the products we have defined
⊕∞ k
preserves ρ-invariance so the subspaces Sρ (V) = k=0 Sρ (V) and Pρ (V) =
⊕∞ k
k=0 Pρ (V) of S(V) and P (V), respectively, are isomorphic subalgebras (as
usual, the k = 0 summand is just C).
The case of most immediate interest to us is that in which G is a ma-
trix Lie group, V is its Lie algebra G and ρ is the adjoint representation
ad : G −→ GL(G). Then Sad (G) and Pad (G) are, respectively, the algebras
of symmetric multilinear functions and polynomials that are invariant under
conjugation
⊕∞ by elements of G. It is customary to write I k (G) for Sad
k
(G) and
k
I(G) = k=0 I (G) for Sad (G)

Remark: The switch from G to G in this notation is not accidental. Two

Lie groups with the same Lie algebra need not have the same ad-invariant
multilinear forms since a given form can be invariant under conjugation by
the elements of one of the groups, but not the other.
We will now construct a number of examples. In each case, G will denote an
arbitrary matrix Lie group and its Lie algebra G will also be identified with
a set of matrices. Let k ≥ 1 be an integer. The symmetrized trace is the
multilinear map
symtr : G × · · · × G −→ C
k

defined by
1 ∑ ( )
symtr(A1 , . . . , Ak ) = trace Aσ(1) · · · Aσ(k) . (6.2.2)
k!
σ∈Sk
310 6. Characteristic Classes

This is clearly symmetric and is also invariant under conjugation by elements

of G since
(( ) ( )) ( ( ) )
trace g −1 Aσ(1) g · · · g −1 Aσ(k) g = trace g −1 Aσ(1) · · · Aσ(k) g
( )
= trace Aσ(1) · · · Aσ(k) .

Thus, symtr is in I k (G). The corresponding ad-invariant polynomial of degree

k on G is
symtr(A, . . . , A) = trace(Ak ). (6.2.3)

In the case of the symmetrized trace we defined the invariant multilinear

form and, from it, an invariant polynomial. For the remaining examples we will
define the polynomials directly (and, when necessary, retrieve the multilinear
forms by polarization). Let λ denote a parameter. For any fixed A ∈ G we
consider the determinant
( )
i
det λI + A
2π

(the reason for the i will emerge shortly; the reason for the 2 π, a bit later).
When expanded this determinant is a polynomial in λ of degree n if the
dimension of G is n. The coefficients of this polynomial depend on A and we
write them as follows:
( )
i
det λI + A = f0 (A)λn + f1 (A)λn−1 + · · · + fn−1 (A)λ + fn (A). (6.2.4)
2π
The functions fk (A) are invariant under conjugation since
( ) ( ( ) )
i −1 i
det λI + (g Ag) = det g −1 (λI )g + g −1 A g
2π 2π
( ( ) )
i
= det g −1 λI + A g
2π
( )
i
= det λI + A .
2π

Notice that f0 (A) = 1 for any A. To obtain convenient expressions for the
remaining fk (A) and show that they are, in fact, homogeneous polynomials
we recall a few basic facts concerning the elementary symmetric functions.
Let F denote a field of characteristic zero (e.g., R or C) and denote by
F[x1 , . . . , xn ] the algebra of polynomials in the n variables x1 , . . . , xn with
coefficients in F. For any P ∈ F[x1 , . . . , xn ] and σ ∈ Sn define σ P by
σ
P (x1 , . . . , xn ) = P (xσ(1) , . . . , xσ(n) ). Then P is symmetric if σ P = P for
every σ ∈ Sn . The elementary symmetric polynomials S0 , S1 , . . . , Sn are
the elements of F[x1 , . . . , xn ] defined by
 6.2. Algebraic Preliminaries 311

S0 (x1 , x2 , . . . , xn ) = 1

S1 (x1 , x2 , . . . , xn ) = x1 + x2 + · · · + xn

S2 (x1 , x2 , . . . , xn ) = x1 x2 + x1 x3 + · · · + x1 xn +
x2 x3 + · · · + x2 xn +
· · · + xn−1 xn
S3 (x1 , x2 , . . . , xn ) = x1 x2 x3 + x1 x2 x4 + · · · + x1 x2 xn +
x1 x3 x4 + · · · + x1 x3 xn + · · · + xn−2 xn−1 xn
..
.
Sk (x1 , x2 , . . . , xn ) = x1 x2 · · · xk + · · · + xn−k+1 xn−k+2 · · · xn
..
.
Sn (x1 , x2 , . . . , xn ) = x1 x2 · · · xn .

Remark: One writes down Sk (x1 , x2 , . . . , xn ) as follows: From

{x1 , x2 , . . . , xn } select any k distinct elements and form their product (one
can always write the product with the subscripts ascending). Form the sum
of all such products.
Another way of describing the same functions is as follows: Let
F[x1 , . . . , xn , z] denote the algebra of polynomials over F in the n+1 variables
x1 , . . . , xn , z. In F[x1 , . . . , xn , z] form the product
∏
(z + xi ) = (z + x1 )(z + x2 ) · · · (z + xn )
1≤i≤n

and write it as a polynomial in z with coefficients in F[x1 , . . . , xn ].

The result is
∏ ∑
n
(z + xi ) = Sk (x1 , . . . , xn )z n−k
1≤i≤n k=0

= S0 (x1 , . . . , xn )z n + S1 (x1 , . . . , xn )z n−1 (6.2.5)

+ · · · + Sn−1 (x1 , . . . , xn )z + Sn (x1 , . . . , xn ).

The so-called Fundamental Theorem on Symmetric Polynomials con-

sists of two assertions:
I. The elementary symmetric polynomials S1 , . . . , Sn are algebraically in-
dependent in F[x1 , . . . , xn ], i.e., for every nonzero f ∈ F[x1 , . . . , xn ] the
polynomial f (S1 , . . . , Sn ) is nonzero (S1 , . . . , Sn do not satisfy any poly-
nomial equation with coefficients in F).
312 6. Characteristic Classes

II. Every symmetric polynomial P in F[x1 , . . . , xn ] can be written as a unique

polynomial in S1 , . . . , Sn , i.e., there exists a unique fP ∈ F[x1 , . . . , xn ]
such that P (x1 , . . . , xn ) = fP (S1 , . . . , Sn ). For example, when n = 3,
P (x1 , x2 , x3 ) = x31 + x32 + x33 is symmetric and one verifies that

x31 + x32 + x33 = (x1 + x2 + x3 )3

− 3(x1 + x2 + x3 )(x1 x2 + x1 x3 + x2 x3 )
+ 3x1 x2 x3
= S13 − 3S1 S2 + 3S3

so that fP (x1 , x2 , x3 ) = x31 − 3x1 x2 + 3x3 .

Essentially any book on Algebra should contain a proof, e.g., see

Section 5.2.2 of [Fang]. The bottom line here is that the subalgebra of
F[x1 , . . . , xn ] consisting of the symmetric polynomials is just F[S1 , . . . , Sn ]
and {S1 , . . . , Sn } is a minimal generating set for this subalgebra.
Now let us return to (6.2.4). We wish to find simple formulas for the
functions fk (A). Begin by fixing an A. We have shown that fk is invari-
ant under conjugation by any nonsingular matrix so its value at A is the
same as its value at g −1 Ag for any nonsingular g. Now, one can always find
a nonsingular matrix g for which g −1 Ag is triangular (see, e.g., Corollary 2,
Chapter X, of [Lang]). Select such a g and let a1 , . . . , an be the diagonal
entries of g −1 Ag. Then
( ) ( ) ∏ n ( )
i i −1 i
det λI + A = det λI + (g Ag) = λ+ ai
2π 2π i=1
2π
( )
i i
= S0 a1 , . . . , an λn
2π 2π
( )
i i
+ S1 a1 , . . . , an λn−1 + · · ·
2π 2π
( )
i i
+ Sn−1 a1 , . . . , an λ
2π 2π
( )
i i
+ Sn a1 , . . . , an
2π 2π
( )
i
= λn + S1 (a1 , . . . , an ) λn−1 + · · ·
2π
( )n−1
i
+ Sn−1 (a1 , . . . , an )
2π
( )n
i
+ Sn (a1 , . . . , an ).
2π
 6.2. Algebraic Preliminaries 313

From (6.2.4) we therefore obtain

( )k
i
fk (A) = Sk (a1 , . . . , an ), k = 0, 1, . . . , n, (6.2.6)
2π

where a1 , . . . , an are the diagonal entries in a triangular matrix similar to A.

We show now that one need not actually find a triangular matrix similar to
A because all of the Sk (a1 , . . . , an ) can be expressed directly in terms of A.
Of course, S0 (a1 , . . . , an ) = 1 so

f0 (A) = 1. (6.2.7)

Next, observe that S1 (a1 , . . . , an ) = a1 + · · · + an = trace(g −1 Ag) = trace A

so
i
f1 (A) = trace A. (6.2.8)
2π
For f2 (A) we note that

S2 (a1 , . . . , an ) = a1 a2 + · · · + a1 an + a2 a3 + · · · + a2 an + · · · + an−1 an
1[ ]
= (a1 + · · · + an )2 − (a21 + · · · + a2n ) .
2
Again, a1 + · · · + an = trace A. Furthermore,
(( )2 ) ( )
a21 + · · · + a2n = trace g −1 Ag = trace (g −1 Ag)(g −1 Ag)
= trace(g −1 (A2 )g) = trace (A2 )

so we have
1[ ]
S2 (a1 , . . . , an ) = (trace A)2 − trace(A2 )
2
and therefore
1 [ ]
f2 (A) = − (trace A) 2
− trace(A 2
) . (6.2.9)
8π 2

Exercise 6.2.5 Show that

1[ ]
S3 (a1 , . . . , an ) = (trace A)3 − 3 trace(A2 )trace A + 2 trace(A3 )
6
and conclude that
i [
f3 (A) = − (trace A)3
48π 3 ] (6.2.10)
−3 trace(A2 ) trace A + 2 trace(A3 ) .
314 6. Characteristic Classes

Exercise 6.2.6 Show that

( )n
i
fn (A) = det A. (6.2.11)
2π

There are similar formulas for all of the fk (A), but these will suffice for our
purposes.
At this point we are going to restrict our attention to the two specific Lie
groups that are of most interest to us, i.e., U (n) and SU (n). Observe that, for
either of these, an element A of the Lie algebra is a skew-Hermitian matrix
so i A is Hermitian and therefore has n distinct real eigenvalues. Thus, A has
n distinct pure imaginary eigenvalues a1 , . . . , an . Recalling that each term in
Sk (a1 , . . . , an ) is a product of k distinct elements of {a1 , . . . , an } we have
( )k
i (−1)k
fk (A) = Sk (a1 , . . . , an ) = Sk (i a1 , . . . , i an ).
2π (2π)k

Since i a1 , . . . , i an are real we conclude that fk is real-valued on either u(n)

or su(n). Furthermore, since the elements of su(n) are skew-Hermitian and
tracefree, (6.2.8) implies that f1 (A) = 0 for all A in su(2).
Our next objective is a complete description of the algebra I(U (n)) of sym-
metric, ad-invariant multilinear forms on u(n). More precisely, we will show
that if f˜1 , . . . , f˜n are the elements of I(U (n)) corresponding (by polarization)
to the polynomials f1 , . . . , fn on u(n) described above, then the subalgebra
of I(U (n)) generated by {f˜1 , . . . , f˜n } is, in fact, all of I(U (n)). Since f˜ −→ f
is an isomorphism of I(U (n)) onto the algebra of ad-invariant polynomials on
u(n) it will suffice to show that this latter algebra is generated by {f1 , . . . , fn }.
Begin by letting G′ denote the subgroup of U (n) consisting of the diagonal
matrices in U (n). This is clearly also a submanifold and so a Lie group. The
Lie algebra G ′ of G′ is the subalgebra of u(n) consisting of all
 
i ξ1 0 ··· 0
 
 0 i ξ2 ··· 0 
 
diag(i ξ , . . . , i ξ ) = 
1 n
.. .. .. 
 
 . . . 
0 0 ··· i ξn

where ξ 1 , . . . , ξ n are real. We claim first that

u(n) = {g −1 A′ g : A′ ∈ G ′ , g ∈ U (n)}. (6.2.12)

Since G ′ ⊆ u(n) it is clear that any g −1 A′ g with A′ ∈ G ′ and g ∈

U (n) is in u(n). For the reverse containment we let A ∈ u(n). Then A is
skew-Hermitain so −i A is Hermitian. According to the Spectral Theorem
 6.2. Algebraic Preliminaries 315

(Chapter XI, [Lang]), there exists a g ∈ U (n) such that g(−i A)g −1 is diago-
nal with real entries, say,

g(−i A)g −1 = diag(ξ 1 , . . . , ξ n ).

Thus,
−i A = g −1 (diag(ξ 1 , . . . , ξ n ))g
so
A = g −1 diag(i ξ 1 , . . . , i ξ n )g
as required.
Now consider the restriction map R : I(U (n)) −→ I(G′ ), defined as follows:
If f˜: u(n) × · · · × u(n) −→ C is in I(U (n)), then Rf˜ : G ′ × · · · × G ′ −→ C
is given by (Rf˜)(A′1 , . . . , A′k ) = f˜(A′1 , . . . , A′k ) for all A′1 , . . .,A′k in G ′ ⊆ u(n).
Also let
N = {g ∈ U (n) : g −1 A′ g ∈ G ′ for all A′ ∈ G ′ } ⊇ G′
and
{
IN (G′ ) = f˜ ∈ I(G′ ) : f˜(g −1 A′1 g, . . . , g −1 A′k g) =
}
f˜(A′1 , . . . , A′k ) for all A′1 , . . . , A′k in G ′ and all g ∈ N .

We claim that R carries I(U (n)) isomorphically into IN (G′ ). Notice that R
certainly carries I(U (n)) into IN (G′ ) and is an algebra homomorphism. To
show that R is injective we show that its kernel is trivial. Thus, suppose
f˜ ∈ I(U (n)) and Rf˜ is the zero element of IN (G′ ), i.e., (Rf˜)(A′1 , . . . , A′k ) = 0
for all A′1 , . . . , A′k in G ′ . In particular, (Rf˜)(A′ , . . . , A′ ) = 0 for all A′ in G ′ .
By (6.2.12), every A ∈ u(n) can be written as g −1 A′ g for some g ∈ U (n) and
some A′ ∈ G ′ . Thus, f˜(A, . . . , A) = f˜(g −1 A′ g, . . . , g −1 A′ g) = f˜(A′ , . . . , A′ ) =
(Rf˜)(A′ , . . . , A′ ) = 0 for all A ∈ u(n). Consequently, the polynomial on u(n)
corresponding to f˜ is identically zero. Since this correspondence is an isomor-
phism, f˜ is identically zero and the kernel of R is trivial.
To prove that R maps onto IN (G′ ) we note first that every element of
IN (G′ ) gives rise to a polynomial on G ′ that is invariant under conjuga-
tion by elements of N . Identifying G ′ with the set of all diag (i ξ 1 , . . . , i ξ n )
with ξ 1 , . . . , ξ n real, we claim that such a polynomial must be a symmet-
ric function of ξ 1 , . . . , ξ n . To prove this it will suffice to show that, for ev-
ery pair (i, j) of indices with 1 ≤ i < j ≤ n, there exists a g ∈ N such
that
g −1 (diag(i ξ 1 , . . . , i ξ i , . . . , i ξ j , . . . , i ξ n ))g
(6.2.13)
= diag(i ξ 1 , . . . , i ξ j , . . . , i ξ i , . . . , i ξ n ).
Exercise 6.2.7 Show that the n × n matrix g having 1 in the (i, j)-slot, the
(j, i)-slot and the (k, k)-slot for k ̸= i, j and having 0 elsewhere is in N and
satisfies (6.2.13).
316 6. Characteristic Classes

Now consider the elements f˜1 , . . ., f˜n in I(U (n)) corresponding to the
polynomials f1 , . . . , fn on u(n) defined by (6.2.6). The polynomials on G ′
corresponding to Rf˜1 , . . . , Rf˜n are simply the restrictions to G ′ of f1 , . . . , fn
and these are given by
( )k
i
fk (diag(i ξ 1 , . . . , i ξ n )) = Sk (i ξ 1 , . . . , i ξ n )
2π
(−1)k
= Sk (ξ 1 , . . . , ξ n )
(2π)k

for k = 1, . . . , n. Since every symmetric polynomial in ξ 1 , . . . , ξ n is a poly-

nomial in the elementary symmetric functions S1 (ξ 1 , . . . , ξ k ), . . . ,
Sn (ξ 1 , . . . , ξ k ), we conclude that every element of I(U (n)) gives rise to
a polynomial on G ′ that can be written as a polynomial in the restric-
tions of f1 , . . . , fn to G ′ . Thus, every element of IN (G ′ ) is a polynomial
in Rf˜1 , . . . , Rf˜n . Since R is an algebra homomorphism, a polynomial in
Rf˜1 , . . . , Rf˜n is the image under R of a polynomial in f˜1 , . . . , f˜n so
every element in IN (G′ ) is the image under R of some element of I(U (n)).
Thus,
R : I(U (n)) −→ IN (G′ )

is an isomorphism of algebras.
Notice that we have actually proved more. Since the restrictions of f1 , . . . , fn
to G ′ are (up to constants) the elementary symmetric polynomials in ξ 1 , . . . , ξ n ,
they are algebraically independent and therefore the same is true of the ele-
ments Rf˜1 , . . . , Rf˜n in I(G′ ). Thus, f˜1 , . . . , f˜n are algebraically independent
in I(U (n)). Furthermore, Rf˜1 , . . . , Rf˜n generate the algebra IN (G′ ) and so
f˜1 , . . . , f˜n generate I(U (n)) and we have established our major result.

Theorem 6.2.1 Let f˜1 , . . . , f˜n be the elements of I(U (n)) corresponding by
polarization to the ad-invariant polynomials on u(n) defined by (6.2.4) (or
(6.2.6)). Then f˜1 , . . . , f˜n are algebraically independent and generate the alge-
bra I(U (n)).
The result corresponding to Theorem 6.2.1 for SU (n) is almost the same
except that, since su(n) consists of tracefree matrices, f1 is identically zero
and so does not appear in the generating set.
Theorem 6.2.2 Let f˜2 , . . . , f˜n be the elements of I(SU(n)) corresponding
by polarization to the ad-invariant polynomials on su(n) defined by (6.2.4)
(or (6.2.6)). Then f˜2 , . . . , f˜n are algebraically independent and generate the
algebra I(SU (n)).
Exercise 6.2.8 Prove Theorem 6.2.2 by making whatever modifications are
required in the arguments that led to Theorem 6.2.1.
 6.3. The Chern-Weil Homomorphism 317

Keep in mind that the generators of I(U (n)) and I(SU (n)) described in
Theorems 6.2.1 and 6.2.2 are actually real-valued on u(n) and su(n), respec-
tively. This will be of some interest when we use them to define the Chern
characteristic classes in the next section.

6.3 The Chern-Weil Homomorphism

With this algebra behind us we can at last begin building characteristic classes
on principal bundles. Start with a matrix Lie group G whose Lie algebra G is
also identified with a set of matrices and a principal G-bundle
P
G ,→ P −→ X

over some manifold X. Choose some connection ω on the bundle and

denote its curvature by Ω (one of the major tasks before us is to show that
the objects we are about to construct do not depend on this choice). For
every f˜ ∈ I k (G), k ≥ 1, we define a complex-valued 2k-form f˜(Ω) on P
by
( ) 1 ∑ ( ( )
f˜(Ω) (v 1 , . . . , v 2k ) = k (−1)σ f˜ Ω p v σ(1) , v σ(2) ,
p 2
σ∈S2k

(6.3.1)
( ))
. . . , Ω p v σ(2k−1) , v σ(2k)

for all v 1 , . . . , v 2k in Tp (P ). For k = 0, f˜ is an element of C and we take

f˜(Ω) to be the constant 0-form whose value is f˜. Notice that, if {e1 , . . . , en }
is a basis for G and Ω = Ω α eα , where each Ω α is a real-valued 2-form on P ,
then

f˜(Ω) = f˜(eα1 , . . . , eαk )Ω α1 ∧ · · · ∧ Ω αk (6.3.2)

for k ≥ 1. To see this we observe that

( ) 1 ∑ ( ( )
f˜(Ω) (v 1 , . . . , v 2k ) = k (−1)σ f˜ Ω α
p
1
v σ(1) , v σ(2) eα1 ,
p 2
σ∈S2k
( ) )
. . . , Ωα
p
k
v σ(2k−1) , v σ(2k) eαk

1 ∑ ( )
= k
(−1)σ Ω α
p
1
v σ(1) , v σ(2)
2
σ∈S2k
( ) ( )
. . . Ωα
p
k
v σ(2k−1) , v σ(2k) f˜ eα1 , . . . , eαk .
318 6. Characteristic Classes

But
1 ∑ ( )
(Ω α1 ∧ · · · ∧ Ω αk )p (v 1 , . . . , v 2k ) = (−1)σ Ω α 1
v σ(1) , v σ(2)
2! 2! · · · 2! p
σ∈S2k
( )
αk
. . . Ω p v σ(2k−1) , v σ(2k)
1 ∑ ( )
= k (−1)σ Ω α p
1
v σ(1) , v σ(2)
2
σ∈S2k
( )
. . . Ωα p
k
v σ(2k−1) , v σ(2k)

so the result follows.

Remark: Since any f˜ ∈ I(G) is uniquely determined (via polarization) by
the ad-invariant polynomial f on G corresponding to it, the form f˜(Ω) is also
uniquely determined by f (and Ω, of course). Thus, for any f ∈ Pad k
(G), we
˜
will often write f (Ω) for the 2k-form f (Ω).
Before proving the major results of this section we pause to write out a
number of concrete examples. Specifically, we will compute, for k = 1, 2, the
2k-form determined by Ω and the symmetrized trace (6.2.2). Begin with k = 1
so that f˜(A) = symtr(A) = trace(A). Then
( ) 1 ∑ ( ( ))
f˜(Ω) (v 1 , v 2 ) = 1 (−1)σ f˜ Ω p v σ(1) , v σ(2)
p 2
σ∈S2

1[ ]
= f˜ (Ω p (v 1 , v 2 )) − f˜ (Ω p (v 2 , v 1 ))
2
1[ ]
= trace(Ω p (v 1 , v 2 )) − trace(Ω p (v 2 , v 1 ))
2
1
= trace(Ω p (v 1 , v 2 ) − Ω p (v 2 , v 1 ))
2
1
= trace(2Ω p (v 1 , v 2 ))
2
= trace(Ω p (v 1 , v 2 ))

= (trace ◦ Ω)p (v 1 , v 2 )

so f˜(Ω) = trace ◦ Ω, which we write simply as

( )
f˜(Ω) = trace Ω f˜(A) = symtr(A) = trace(A) . (6.3.3)

When
∑ k = 2 the symmetrized trace is given by f˜(A1 , A2 ) = symtr(A1 , A2 ) =
1 1
2! σ∈S2 trace(Aσ(1) Aσ(2) ) = 2 [trace(A1 A2 )+ trace(A2 A1 )] = trace(A1 A2 ).
 6.3. The Chern-Weil Homomorphism 319

Thus,
( ) 1 ∑ ( ( )
f˜(Ω) (v 1 , v 2 , v 3 , v 4 ) = 2 (−1)σ f˜ Ω p v σ(1) , v σ(2) ,
p 2
σ∈S4
( ))
Ω p v σ(3) , v σ(4)
1 ∑ ( ( )
= (−1)σ trace Ω p v σ(1) , v σ(2) .
4
σ∈S4
( ))
Ω p v σ(3) , v σ(4)
( 1 ∑ )
= trace (−1)σ Ω p (v σ(1) , v σ(2) .
2! 2!
σ∈S4
( ))
Ω p v σ(3) , v σ(4)
( )
= trace (Ω ∧ Ω)p (v 1 , v 2 , v 3 , v 4 )

= (trace ◦ (Ω ∧ Ω))p (v 1 , v 2 , v 3 , v 4 )

so f˜(Ω) = trace ◦ (Ω ∧ Ω). Again, we will write this as

f˜(Ω) = trace(Ω ∧ Ω)
( ) (6.3.4)
f˜(A1 , A2 ) = symtr(A1 , A2 ) = trace(A1 A2 ) .

Exercise 6.3.1 Show that the forms trace Ω and trace(Ω ∧ Ω) on P both
project to closed forms on X. Hint: Consider trace F and
trace(F ∧ F ), where F = s∗ Ω is any field strength corresponding to a local
cross-section s.
Our first major result is a generalization of Exercise 6.3.1.
Theorem 6.3.1 Let ω be a connection with curvature Ω on the principal
P
G-bundle G ,→ P −→ X and let f˜ ∈ I k (G) for some k ≥ 1. Then the 2k-form
f˜(Ω) defined by (6.3.1) projects to a unique closed 2k-form f˜(Ω) on X (i.e.,
f˜(Ω) = P ∗ (f¯(Ω)) and d(f¯(Ω)) = 0).
Proof: To show that f˜(Ω) projects to a unique 2k-form on X we use
Lemma 4.5.1. Thus, we must show

1. f˜(Ω) is zero whenever at least one of its arguments is vertical.

2. σg∗ (f˜(Ω)) = f˜(Ω)) for each g ∈ G.

The first of these is clear from (6.3.2) since Ω vanishes when either of its
arguments is vertical. To prove the second we recall that σg∗ (Ω) = adg−1 ◦ Ω
320 6. Characteristic Classes

(Lemma 6.2.2, [N4]) and that f˜ satisfies f˜(adg−1 (A1 ), . . . , adg−1 (Ak )) =
f˜(A1 , . . . , Ak ) so
( ( ))
σg∗ f˜(Ω) (v 1 , . . . , v 2k )
p
( )
= f˜(Ω) ((σg )∗p (v 1 ), . . . , (σg )∗p (v 2k ))
p·g
1 ∑ ( ( )
= k (−1)τ f˜ Ω p·g (σg )∗p (v τ (1) ), (σg )∗p (v τ (2) ) , . . . ,
2
τ ∈S2k
( ))
Ω p·g (σg )∗p (v τ (2k−1) ), (σg )∗p (v τ (2k) )
1 ∑ (
= k (−1)τ f˜ (σg∗ Ω)p (v τ (1) , v τ (2) ), . . . ,
2
τ ∈S2k
)
(σg∗ Ω)p (v τ (2k−1) , v τ (2k) )

1 ∑ ( ( )
τ ˜
= (−1) f ad g −1 Ω p (v τ (1) , v τ (2) ) , . . . ,
2k
τ ∈S2k
( ))
adg−1 Ω p (v τ (2k−1) , v τ (2k) )
1 ∑ ( )
= k (−1)τ f˜ Ω p (v τ (1) , v τ (2) ), . . . , Ω p (v τ (2k−1) , v τ (2k) )
2
τ ∈S2k
( )
= f˜(Ω) (v 1 , . . . , v 2k )
p

as required.
Thus, there exists a unique 2k-form f¯(Ω) on X with P ∗ (f¯(Ω)) = f˜(Ω). We
show that f¯(Ω) is closed. Notice that d(f˜(Ω)) = d(P ∗ (f¯(Ω))) = P ∗ (d(f¯(Ω)))
so d(f˜(Ω)) projects to d(f¯(Ω)). Since projections are unique when they ex-
ist, it will suffice to show that f˜(Ω) is closed. Now, because f˜(Ω) projects,
Lemma 4.5.5 implies that d(f˜(Ω)) = dω (f˜(Ω)) so d(f˜(Ω)) vanishes when any
one of its arguments is vertical. Thus, we need only show that d(f˜(Ω)) van-
ishes when all of its arguments are horizontal. For this we let {eα } be a basis
for G and write Ω = Ω α eα , where each Ω α is a real-valued 2-form on P . Then

f˜(Ω) = f˜(eα1 , . . . , eαk )Ω α1 ∧ · · · ∧ Ω αk

so

d(f˜(Ω)) = f˜(eα1 , . . . , eαk )(dΩ α1 ∧ Ω α2 ∧ · · · ∧ Ω αk

+ Ω α1 ∧ dΩ α2 ∧ · · · ∧ Ω αk (6.3.5)
+ · · · + Ω α1 ∧ Ω α2 ∧ · · · ∧ dΩ αk ).
 6.3. The Chern-Weil Homomorphism 321

Now, the Bianchi Identity (Theorem 4.5.9) gives

dω Ω = 0
dω (Ω α eα ) = 0
(dω Ω α )eα = 0

so dω Ω α = 0 for α = 1, . . . , n. Consequently, each dΩ α vanishes when all of

its arguments are horizontal and, by (6.3.5), so does d(f˜(Ω)). 

Exercise 6.3.2 Let f˜ ∈ I k (G) and g̃ ∈ I l (G). Show that

( ) ( )
f˜ ⊙ g̃ (Ω) = P ∗ f¯(Ω) ∧ ḡ(Ω) .

Hint: Keep in mind that the wedge product is commutative on real-valued

2-forms so that Ω α1 ∧ · · · ∧ Ω αk+l is symmetric in α1 , . . . , αk+l .
Before moving on to the major result of this section we will pause to con-
struct a few of the projections f¯(Ω). Recall (from the proof of Lemma 4.5.1)
that if φ is a k-form on P which vanishes when any of its arguments is vertical
and satisfies σg∗ φ = φ for every g ∈ G, then its unique projection φ̄ to X can
be described explicitly as follows: For each x ∈ X and w 1 , . . . , w k in Tx (X)

φ̄x (w 1 , . . . , w k ) = φs(x) (s∗x (w 1 ), . . . , s∗x (w k )),

where s is any local cross-section with x in its domain. Thus, φ̄ = s∗ φ.

Now, suppose f˜(A) = symtr(A) = trace(A). Then, by (6.3.3), f˜(Ω) =
trace Ω. The projection f¯(Ω) of f˜(Ω) to X is obtained as follows: Choose a
local cross-section s : V −→ P −1 (V ). Then, for any x ∈ V and any w 1 , w 2 ∈
Tx (X),
( )
(f¯(Ω))x (w 1 , w 2 ) = f˜(Ω) (s∗x (w 1 ), s∗x (w 2 ))
s(x)

= (trace Ω)s(x) (s∗x (w 1 ), s∗x (w 2 ))

( )
= trace Ω s(x) (s∗x (w 1 ), s∗x (w 2 ))

= trace((s∗ Ω)x (w 1 , w 2 ))

= trace(F x (w 1 , w 2 ))

= (trace F )x (w 1 , w 2 ),

where F = s∗ Ω is the local field strength corresponding to Ω in gauge s.

Thus,
f (A) = trace A =⇒ f¯(Ω) = trace F . (6.3.6)
322 6. Characteristic Classes

Remark: Again we emphasize that different cross-sections give different

local field strengths, in general, but their traces agree on the intersections of
their domains so trace F only appears to be a locally defined object; it is, in
fact, a global 2-form on X.
Exercise 6.3.3 Show that

f (A) = trace(A2 ) =⇒ f¯(Ω) = trace(F ∧ F ). (6.3.7)

Hint: See (6.3.4).

As a final example we consider f˜ ⊙ f˜ ∈ I 2 (G), where f˜ ∈ I 1 (G) is given by
˜
f (A) = trace A. We compute
( ) 1 ∑ ( ) ( )
f˜ ⊙ f˜ (A1 , A2 ) = f˜ Aσ(1) f˜ Aσ(2)
(1 + 1)!
σ∈S2

1[ ]
= f˜(A1 )f˜(A2 ) + f˜(A2 )f˜(A1 )
2
= f˜(A1 )f˜(A2 )

= (trace A1 )(trace A2 ).

Note that the corresponding polynomial is (trace A)2 . Furthermore,

Exercise 6.3.2 implies that the projection of (f˜ ⊙ f˜)(Ω) onto X is
f¯(Ω) ∧ f¯(Ω) which, by (6.3.6), is (trace F ) ∧ (trace F ). Thus,

f (A) = (trace A)2 =⇒ f¯(Ω) = (trace F ) ∧ (trace F ). (6.3.8)

P
Now, suppose G ,→ P −→ X is a fixed principal G-bundle and f˜ is a
fixed element of I k (G), but that ω 0 and ω 1 are two connections on P with
curvatures Ω 0 and Ω 1 , respectively. One computes f˜(Ω 0 ) and f˜(Ω 1 ) as well
as their projections f¯(Ω 0 ) and f¯(Ω 1 ) to X. There is, of course, no reason to
expect these projections to be the same and, indeed, they are generally not.
Our next result, however, asserts that, at least in the cohomological sense,
they differ only trivially, i.e., by an exact form.
Theorem 6.3.2 Let ω 0 and ω 1 be connections on the principal G-bundle
P
G ,→ P −→ X with curvatures Ω 0 and Ω 1 , respectively, and let f˜ ∈ I k (G)
for some k ≥ 1. Then there exists a (2k − 1)-form φ on X such that f¯(Ω 1 ) −
f¯(Ω 0 ) = dφ.
Proof: Let α = ω 1 − ω 0 and for each t ∈ [0, 1], ω t = ω 0 + tα. Then α
is tensorial of type ad. Thus, each t α is tensorial of type ad so each ω t is a
connection on P . Let
1
Ω t = dωt ω t = dω t + [ω t , ω t ]
2
 6.3. The Chern-Weil Homomorphism 323

be the curvature of ω t and notice that

[ω t , ω t ] = [ω 0 + tα, ω 0 + tα]
= [ω 0 , ω 0 ] + t[α, ω 0 ] + t[ω 0 , α] + t2 [α, α]

so that
d
[ω t , ω t ] = [α, ω 0 ] + [ω 0 , α] + 2t[α, α]
dt
= ([ω 0 , α] + t[α, α]) + ([α, ω 0 ] + t[α, α])
= [ω 0 + tα, α ] + [α, ω 0 + tα]
= [ω t , α] + [α, ω t ]
= 2[α, ω t ].

Thus,
( )
d d 1
Ωt = dω t + [ω t , ω t ]
dt dt 2
( )
d 1
= dω 0 + t dα + [ω t , ω t ]
dt 2
d
Ω t = dα + [α, ω t ]. (6.3.9)
dt

Exercise 6.3.4 Let f˜ be an element of I k (G) and let φ1 , . . . , φk be

G-valued forms on P of degree q1 , . . . , qk , respectively. Define a
(q1 + · · · + qk )-form f˜(φ1 , . . . , φk ) on P by
( ) ( )
f˜(φ1 , . . . , φk ) v 1 , . . . , v q1 +···+qk
1 ∑
p ( ( )
= (−1)σ f˜ φ1 v σ(1) , . . . , v σ(q1 ) , (6.3.10)
q1 ! · · · qk !
(σ∈Sq1 +···+qk ))
. . . , φk v σ(q1 +···+qk−1 +1) , . . . , v σ(q1 +···+qk ) .

Show that, if {e1 , . . . , en } is a basis for G and φ1 = φα αk

1 eα1 , . . . , φk = φk eαk ,
1

then ( )
f˜(φ1 , . . . , φk ) = f˜ eα1 , . . . , eαk φα
1 ∧ · · · ∧ φk .
1 αk
(6.3.11)

Now consider the (2k − 1)-form f˜(α, Ω t , . . . , Ω t ) on P defined as in

Exercise 6.3.4. If {e1 , . . . , en } is a basis for G with α = αα0 eα0 and Ω t =
Ωα
t eα , (6.3.11) gives
( )
f˜(α, Ω t , . . . , Ω t ) = f˜ eα0 , eα1 , . . . , eαk−1 αα0
(6.3.12)
α
∧ Ωα
t ∧ · · · ∧ Ωt
1 k−1
.
324 6. Characteristic Classes

Exercise 6.3.5 Show that f˜(α, Ω t , . . . , Ω t ) projects to a unique (2k − 1)-

form on X. Hint: The argument is the same as for f˜(Ω) in the proof of
Theorem 6.3.1.
Next define a (2k − 1)-form Φ on P by
∫ 1
Φ=k f˜(α, Ω t , . . . , Ω t )dt. (6.3.13)
0

Since f˜(α, Ω t , . . . , Ω t ) projects to X, so does Φ (to the analogously defined

integral of the projection of f˜(α, Ω t , . . . , Ω t )). We claim that

dΦ = f˜(Ω 1 ) − f˜(Ω 0 ). (6.3.14)

First note that this will complete the proof of the Theorem since, if
Φ = P ∗ φ it gives
( ) ( )
d(P ∗ φ) = P ∗ f¯(Ω 1 ) − P ∗ f¯(Ω 0 )
( )
P ∗ (d φ) = P ∗ f¯(Ω 1 ) − f¯(Ω 0 )

and therefore, since projections are unique when they exist,

dφ = f¯(Ω 1 ) − f¯(Ω 0 ). (6.3.15)

Thus, we need only prove (6.3.14). First note that d(f˜(α, Ω t , . . . , Ω t )) =

dωt (f˜(α, Ω t , . . . , Ω t )) by Lemma 4.5.5. But, from (6.3.12), the product rule
for dωt and the Bianchi Identity,
( )
dωt f˜(α, Ω t , . . . , Ω t ) = f˜(dωt α, Ω t , . . . , Ω t )
= f˜(dα + [α, ω t ], Ω t , . . . , Ω t ).

Now we use (6.3.9) to compute

( )
kd f˜(α, Ω t , . . . , Ω t ) = k f˜(dα + [α, ω t ], Ω t , . . . , Ω t )
( )
d
= k f˜ Ωt, Ωt, . . . , Ωt
dt
( ) ( )
d d
= f˜ Ω t , Ω t , . . . , Ω t + f˜ Ω t , Ω t , . . . , Ω t
dt dt
( )
d
+ · · · + f˜ Ω t , Ω t , . . . , Ω t
dt
d ˜
= f (Ω t , Ω t , . . . , Ω t ).
dt
 6.3. The Chern-Weil Homomorphism 325

Thus,
∫ 1 ( ) ∫ 1
˜ d ˜
dΦ = k d f (α, Ω t , . . . , Ω t ) dt = f (Ω t , Ω t , . . . , Ω t )dt
0 0 dt
= f˜(Ω 1 , Ω 1 , . . . , Ω 1 ) − f˜(Ω 0 , Ω 0 , . . . , Ω 0 ) = f˜(Ω 1 ) − f˜(Ω 0 )

as required. 

The situation is now as follows: We are given a principal bundle

P
G ,→ P −→ X. Connections always exist on this bundle (Theorem 3.1.7)
so we select one. Let ω denote this connection and Ω its curvature. Now, let
f ∈ Padk
(G) be any symmetric homogeneous polynomial on the Lie algebra G
that is invariant under conjugation by elements of G. We have associated with
f a closed 2k-form f¯(Ω) on X. A different connection ω ′ will give rise to a dif-
ferent 2k-form f¯(Ω ′ ), but we now know that f¯(Ω) and f¯(Ω ′ ) are cohomologi-
cally the same, i.e., cohomologous. Thus, f¯(Ω) and f¯(Ω ′ ) determine the same
(complex) de Rham cohomology class, which we denote w(f ) ∈ Hde R (X; C).
2k

We have then a mapping

R (X; C)
∗
w : I(G) −→ Hde

which assigns to each f˜ ∈ I(G) the cohomology class [f¯(Ω)], where Ω is the
curvature of any connection on the bundle. This map is, in fact, an algebra
homomorphism since it is clearly linear and, by Exercise 6.3.2, it also satisfies
( )
w f˜ ⊙ g̃ = w(f˜) ∧ w(g̃).

R (X; C) under this homomorphism is an algebra of

∗
The image of I(G) in Hde
complex cohomology classes on X the elements of which are called (complex)
P
characteristic classes for the bundle G ,→ P −→ X. The map w itself is
called the Chern-Weil homomorphism of the bundle.
If f˜ ∈ I(G) happens to be real-valued, then any f¯(Ω) is a real-valued
closed form on X and so we can identify w(f˜) = [f¯(Ω)] with an element of
R (X; R). This is, in fact, the case for many of our most important exam-
∗
Hde
ples. If fk is one of the ad-invariant polynomials defined by (6.2.4) (or (6.2.6)),
then the corresponding characteristic class

ck (P ) = w(fk ) = [f¯k (Ω)] (6.3.16)

P
is called the k th Chern class of the bundle G ,→ P −→ X. For example,
using (6.2.8) and (6.3.6), the 1st Chern class is given by

i
c1 (P ) = [trace F ]. (6.3.17)
2π
326 6. Characteristic Classes

Similarly, from (6.2.9), (6.3.7) and (6.3.8), the 2nd Chern class is
1
c2 (P ) = − [(trace F ) ∧ (trace F ) − trace(F ∧ F )]. (6.3.18)
8π 2
There are similar formulas for the remaining ck (P ). However, since f¯k (Ω) is
a 2k-form on X these will reduce to zero when k > 12 dim X. In particular, if
dim X = 2, then c1 (P ) is the only nonzero Chern class. If dim X = 4, then
c1 (P ) and c2 (P ) are the only nontrivial Chern classes.
Let us consider in somewhat more detail the cases G = U (n) and
G = SU (n) that most interest us. We have already seen that, for these groups,
the Chern classes are real cohomology classes.

Remark: Although we will not do so, one can prove that this is actually
true for any matrix Lie group G; Chern classes are real cohomology classes.
Indeed, they are (in a sense we will describe shortly) integral cohomology
classes.
Furthermore, Theorems 6.2.1 and 6.2.2 imply that the algebra of characteristic
classes for any U (n)- or SU (n)-bundle is generated by the Chern classes.
Remark: This is not true for arbitary Lie groups. However, it is known
that, for any compact, semi-simple Lie group G, the algebra I(G) is finitely
generated so that the Chern-Weil homomorphism for any G-bundle is deter-
mined by its values on finitely many elements of I(G). The images of these
then generate the algebra of characteristic classes. For many of the most im-
portant Lie groups it is possible to write out simple, finite generating sets (see
Chapter XII of [KN2]).
In the case of SU (n), trace F is zero so many of the formulas simplify consid-
erably, e.g.,

c1 (P ) = 0 (G = SU (n)), (6.3.19)
1
c2 (P ) = [trace(F ∧ F )] (G = SU (n)). (6.3.20)
8π 2
We saw in Section 6.1 that U (1)-bundles over S 2 are classified up to equiv-
alence by their 1st Chern classes (note that, for U (1)-bundles, F is a 1 × 1
matrix so trace F is just its sole entry and the definition of c1 (P ) given in
Section 6.1 agrees with (6.3.17)). We would like to show that SU (2)-bundles
over S 4 are likewise characterized by their 2nd Chern classes. First we show
that equivalent bundles have, not only the same Chern classes, but the same
Chern-Weil homomorphisms.
P P′
Theorem 6.3.3 Let G ,→ P −→ X and G ,→ P ′ −→ X be equivalent
principal G-bundles over X. Then the Chern-Weil homomorphisms
I(G) −→ H ∗ (X; C ) they determine are equal.
 6.4. Chern Numbers 327

Proof: Let λ : P −→ P ′ be an equivalence (i.e., a diffeomorphism of P

onto P ′ satisfying λ(p · g) = λ(p) · g and P ′ (λ(p)) = P(p) for all p ∈ P
and g ∈ G). Fix f˜ ∈ I k (G), k ≥ 1. Now, select a connection ω ′ on P ′ with
curvature Ω ′ . Then ω = λ∗ ω ′ is a connection on P with curvature Ω = λ∗ Ω ′ .
If {e1 , . . . , en } is a basis for G and we write Ω ′ = ( Ω ′ )α eα , then f˜(Ω ′ ) =
f˜(eα1 , . . . , eαk )(Ω ′ )α1 ∧ · · · ∧ (Ω ′ )αk . Since pullback is linear and commutes
with wedge products we find that f˜(Ω) = λ∗ (f˜(Ω ′ )).
To show that the Chern-Weil homomorphisms agree we need only prove
that f¯(Ω) and f¯(Ω ′ ) determine the same cohomology class. We will, in fact,
show that they are equal. Moreover, since projections to X are unique, it will
suffice to prove that P ∗ (f¯(Ω ′ )) = f˜(Ω). But

P ∗ (f¯(Ω ′ )) = (P ′ ◦ λ)∗ (f¯(Ω ′ ))

( )
= λ∗ (P ′ )∗ (f¯(Ω ′ ))
( )
= λ∗ f˜(Ω ′ )
= f˜(Ω)

as required. 

Exercise 6.3.6 Show that the algebra of characteristic classes for a trivial
R (X; C). Hint: The product bundle
∗
bundle is the trivial subalgebra of Hde
π
G ,→ X × G −→ X admits a flat connection (page 353, [N4]).

6.4 Chern Numbers

Since characteristic classes are cohomology classes on the base manifold X of
the bundle, they can be integrated over submanifolds of X.
P
Exercise 6.4.1 Let G ,→ P −→ X be a principal G-bundle, f˜ ∈ I k (G), where
1 ≤ k ≤ 12 dim X, N a compact, oriented submanifold of X of dimension 2k
and ι : N ,→ X the inclusion map. Show that
∫
( )
ι∗ f¯(Ω)
N
takes the same value for any connection on P . Show, moreover, that if N1
and N2 are two such submanifolds and N1 can be deformed into N2 in X
(Exercise 4.7.10), then
∫ ∫
( )
ι∗1 (f¯(Ω)) = ι∗2 f¯(Ω) .
N1 N2

Remark: These integrals are, in general, complex numbers. The integral

of the corresponding cohomology class [f¯(Ω)] is, of course, taken to be the
328 6. Characteristic Classes

integral of any of its representatives. If f˜ is real-valued, we have identified

[f¯(Ω)] with a real cohomology class and the integral is real. This is the case,
in particular, for the Chern classes (see the Remark on page 377). One can
show that, in fact, the integral of a Chern class over a submanifold is always
an integer and, in this sense, Chern classes are integral cohomology classes.
We will verify this for the 2nd Chern class of an SU (2)-bundle over S 4 shortly.
When this is done, the reason for the strange normalizing constants in (6.2.6)
will be clear: The i makes the Chern classes real and the 2π makes them
integral.

A special case of the integrals discussed in Exercise 6.4.1 is of particular

P
interest. Suppose that G ,→ P −→ X is a principal G-bundle, where X is an
oriented compact manifold of dimension 2n. Then the nth Chern class cn (P )
R (X; R) and so can be integrated over X. The value of
2n
is an element of Hde
the integral
∫
cn (P )
X
is called the Chern number of the bundle. Chern numbers for equivalent
bundles are, of course, the same. In Section 6.1 we computed the Chern
number for each equivalence class of U (1)-bundles over S 2 and found that
the result was precisely the integer n that appeared in the transition func-
tion gSN (φ, θ) = e−nθi characterizing the bundle. Such bundles are therefore
classified up to equivalence by their Chern numbers. We would now like to
carry out a similar program for SU (2)-bundles over S 4 . Specifically, we will
let
P
SU (2) ,→ P −→ S 4

be some SU (2)-bundle over S 4 and will calculate the Chern number

∫ ∫
1
c2 (P ) = trace(F ∧ F ). (6.4.1)
S4 8π 2 S4

The general plan is essentially the same as for U (1)-bundles over S 2 in

Section 6.1. We trivialize the bundle over UN and US . Choose a connection
and pull it back to obtain local potentials AN and AS and local field strengths
F N and F S . Let S+ 4 4
and S− be the upper and lower hemispheres of S 4 ,
3
respectively, and denote by S the equatorial 3-sphere S+ 4
∩ S−4
in S 4 (with
4
the orientation it inherits from S+ ). Then
∫
1
trace(F ∧ F )
8π 2 S 4
∫ ∫ (6.4.2)
1 1
= trace(F N ∧ F N ) + 2 trace(F S ∧ F S )
8π 2 S+4 8π S− 4
 6.4. Chern Numbers 329

and we wish to apply Stokes’ Theorem to each of the integrals on the right-
hand side. To do so, of course, we need to express both trace(F N ∧ F N ) and
trace(F S ∧ F S ) as exterior derivatives of 3-forms on UN and US , respectively.
This is possible, of course, since they are closed forms on discs, but we will
require more explicit information.
P
Lemma 6.4.1 Let SU (2) ,→ P −→ X be a principal SU (2)-bundle over X, ω
a connection on it with curvature Ω, s : V −→ P −1 (V ) a local cross-section
and A = s∗ ω and F = s∗ Ω the local gauge potential and field strength. Then,
on V ,
( ( ))
2
trace(F ∧ F ) = d trace A ∧ dA + A ∧ A ∧ A . (6.4.3)
3

Proof: Throughout the proof we will work exclusively on the trivial bundle
SU (2) ,→ P −1 (V ) −→ V , but, for convenience, we will denote any relevant
restrictions by the same symbo1, e.g., P is the projection on P −1 (V ), ω is
P
the connection on SU (2) ,→ P −1 (V ) −→ V , etc. We intend to apply (6.3.14),
where f˜ is f˜2 , Ω 1 is Ω and Ω 0 is the curvature of a connection whose existence
we now ask the reader to demonstrate.
P
Exercise 6.4.2 Let G ,→ P −→ X be any principal G-bundle,
s : V −→ P −1 (V ) a local cross-section and Ψ : P −1 (V ) −→ V × G the asso-
ciated trivialization (so s(x) = Ψ−1 (x, e) for every x ∈ V ). Show that there
exists a connection ω 0 on G ,→ P −1 (V ) −→ V for which the local gauge po-
tential A0 = s∗ ω 0 is identically zero. Hint: If Θ is the Cartan 1-form for G
and π : V × G −→ G is the projection, then π ∗ Θ is a flat connection on V × G
(page 353, [N4]). Show that ω 0 = (π ◦ Ψ)∗ Θ has the required properties.
Now, (6.3.14) asserts that

f˜2 (Ω) − f˜2 (Ω 0 ) = dΦ, (6.4.4)

where
∫ 1
Φ=2 f˜2 (α, Ω t )dt
0
α = ω − ω0
ω t = ω 0 + tα
1
Ω t = dωt ω t = dω 0 + t dα + [ω t , ω t ]
2
= dω 0 + t dα + ω t ∧ ω t .

Pulling (6.4.4) back by s gives

f¯2 (Ω) − f¯2 (Ω 0 ) = d(s∗ Φ)

330 6. Characteristic Classes

from which we obtain, since F 0 = 0,

trace(F ∧ F ) = d(s∗ Φ). (6.4.5)

Now, we compute s∗ Φ as follows:

Exercise 6.4.3 Show that

f˜2 (α, Ω t ) = trace(α ∧ Ω t ).

But

α ∧ Ω t = α ∧ (dω 0 + t dα + ω t ∧ ω t )
= α ∧ dω 0 + tα ∧ dα + α ∧ (ω 0 + tα) ∧ (ω 0 + tα)
= α ∧ dω 0 + tα ∧ dα + α ∧ ω 0 ∧ ω 0 + tα ∧ ω 0 ∧ α
+ tα ∧ α ∧ ω 0 + t2 α ∧ α ∧ α

so
∫ 1 ∫ 1
Φ=2 f˜2 (α, Ω t )dt = 2 trace(α ∧ Ω t )dt
0 0
( ∫ 1 )
= trace 2 α ∧ Ω t dt
0
(
= trace 2α ∧ dω 0 + α ∧ dα + 2α ∧ ω 0 ∧ ω 0 + α ∧ ω 0 ∧ α
)
2
+α ∧ α ∧ ω 0 + α ∧ α ∧ α .
3

Noting that, by the way we have chosen ω 0 , s∗ ω 0 = A0 = 0 and

s∗ α = s∗ (ω − ω 0 ) = A we obtain
( )
∗ 2
s Φ = trace A ∧ dA + A ∧ A ∧ A
3

and the Lemma follows from (6.4.5). 

Note that, since F = dA + A ∧ A, (6.4.3) can be written
( ( ))
1
trace(F ∧ F ) = d trace A ∧ F − A ∧ A ∧ A . (6.4.6)
3

We now return to the evaluation of the integrals in (6.4.2). Applying Stokes’

Theorem twice on the right-hand side and keeping in mind that the equatorial
3-sphere S 3 is assumed to have the orientation it inherits from S+4
we obtain
 6.4. Chern Numbers 331
∫
1
trace(F ∧ F )
8π 2 S4
∫ ( ( ))
1 ∗ 1
= ι trace A N ∧ F N − AN ∧ AN ∧ AN
8π 2 S 3 3
∫ ( ( ))
1 1
− 2 ι∗ trace AS ∧ F S − AS ∧ AS ∧ AS
8π S 3 3
∫ ( (
1 1
= ι∗ trace AN ∧ F N − AN ∧ AN ∧ AN
8π 2 S 3 3
))
1
−AS ∧ F S + AS ∧ AS ∧ AS ,
3

where ι : S 3 ,→ S 4 is the inclusion.

−1 −1
Exercise 6.4.4 Substitute AN = gSN AS gSN + gSN dgSN and
−1
F N = gSN F S gSN into this last expression to show that
∫
1
trace(F ∧ F )
8π 2 S 4
∫ [
1 ∗ 1 −1 −1
= ι − trace((gSN dgSN ) ∧ (gSN dgSN )
8π 2 S 3 3
]
−1 −1
∧ (gSN dgSN )) + d(trace(AS dgSN gSN ))

and apply Stokes’ Theorem to obtain

∫
1
trace(F ∧ F )
8π 2 S 4
∫ (
1 ∗ −1
=− ι trace((gSN dgSN ) (6.4.7)
24π 2 S 3
)
−1 −1
∧ (gSN dgSN ) ∧ (gSN dgSN )) .

Notice that, as expected, the integral in (6.4.7), i.e., the Chern number,
does not depend on the connection or the curvature, but only on the transition
function gSN of the bundle. To obtain a computationally more efficient formula
−1 ∗
we expand gSN dgSN = gSN Θ in terms of the basis
( ) ( ) ( )
i 0 0 1 0 i
I = , J = , K =
0 −i −1 0 i 0

for su(2) ∼
= Im H. Thus, let
( )
−1 1 2 3 Θ1 i Θ2 + Θ3 i
gSN dgSN = Θ I + Θ J + Θ K = .
−Θ 2 + Θ 3 i − Θ1 i
332 6. Characteristic Classes

Exercise 6.4.5 Compute the matrix products and show that

−1 −1 −1
trace((gSN dgSN ) ∧ (gSN dgSN ) ∧ (gSN dgSN )) = −12Θ 1 ∧ Θ 2 ∧ Θ 3 .

Thus,
∫ ∫
1 1
trace(F ∧ F ) = ι∗ (Θ 1 ∧ Θ 2 ∧ Θ 3 ). (6.4.8)
8 π2 S4 2π 2
S3

Our objective now is to show that the Chern number captures the topolog-
ical type of an SU (2)-bundle over S 4 . For this we recall that the equivalence
class of such a bundle is uniquely determined by the homotopy type of gSN | S 3
(this is essentially the Classification Theorem for bundles over spheres, but
Lemma 4.4.1 and Theorem 4.4.2 of [N4] make the correspondence explicit).
Our procedure then will be to select one representative from each homotopy
class in π3 (SU (2)) ∼ = π3 (S 3 ) and show that the bundles with these transition
functions have different Chern numbers.
The notation is less cumbersome if we identify both S 3 and SU (2) with the
unit quaternions. Fix a base point 1 ∈ S 3 and identify π3 (S 3 ) with the set
[(S 3 , 1), (S 3 , 1)] of homotopy classes of maps g : (S 3 , 1) −→ (S 3 , 1) (see page
154, [N4]). For each n = 0, 1, 2, . . . we let

gn : (S 3 , 1) −→ (S 3 , 1)

be the restriction to S 3 of the map g̃n : H −→ H given by g̃n (q) = q n . Thus,

ι ◦ gn = g̃n ◦ ι, where ι : S 3 ,→ H is the inclusion. We will show that the degree
of gn and the Chern number of the bundle whose gSN | S 3 is gn are both n.
This is, of course, clear for n = 0 since the constant map g0 has degree 0,
while dg0 = 0 and (6.4.7) shows that the Chern number of the corresponding
bundle is also 0. Next notice that g1 is the identity map on S 3 and so has
degree 1. Furthermore, g1−1 dg1 = q −1 dq.[2pt]
Exercise 6.4.6 Show that, on H,

q −1 dq = [(x0 )2 + (x1 )2 + (x2 )2 + (x3 )2 ]−2

· [(x0 dx0 + x1 dx1 + x2 dx2 + x3 dx3 )
+ (x0 dx1 − x1 dx0 − x2 dx3 + x3 dx2 )i
+ (x0 dx2 − x2 dx0 − x3 dx1 + x1 dx3 )j
+ (x0 dx3 − x3 dx0 − x1 dx2 + x2 dx1 )k ].

Show also that, on S 3 ⊆ H, the real part of q −1 dq is zero and the wedge
product of the i , j and k components is
 6.4. Chern Numbers 333

x0 dx1 ∧ dx2 ∧ dx3 − x1 dx0 ∧ dx2 ∧ dx3 +

(6.4.9)
x2 dx0 ∧ dx1 ∧ dx3 − x3 dx0 ∧ dx1 ∧ dx2 .
Now, observe that the restriction of the 3-form in (6.4.9) to S 3 is the standard
volume form for S 3 (Section 4.6) so its integral over S 3 is 2π 2 . Thus, according
to (6.4.8), the Chern number of the bundle corresponding to g1 is 2π1 2 (2π 2 ) = 1
which, in particular, coincides with deg(g1 ).
For the general case we proceed as follows: The degree of a map
g : S 3 −→ S 3 can be computed as
∫
deg(g) = g ∗ ω,
S3

where ω is a 3-form on S 3 whose integral over S 3 is 1. Such a 3-form

on S 3 is given by 2π1 2 times the standard volume form on S 3 . Accord-
ing to Exercise 6.4.6, the volume form on S 3 is the restriction to S 3 of
Θ 1 ∧ Θ 2 ∧ Θ 3 , where Θ 1 I + Θ 2 J + Θ 3 K is the imaginary part of q −1 dq.
By Exercise 6.4.5, this, in turn, is the restriction to S 3 of
− 12
1
trace((q −1 dq) ∧ (q −1 dq) ∧ (q −1 dq)). Thus,

1 ∗( −1 −1 −1
)
ω=− ι trace((q dq) ∧ (q dq) ∧ (q dq))
24π 2
so

1 ( )
g∗ ω = − 2
(ι ◦ g)∗ trace((q −1 dq) ∧ (q −1 dq) ∧ (q −1 dq))
24π
1 (
=− 2
trace (ι ◦ g)−1 d(ι ◦ g) ∧ (ι ◦ g)−1 d(ι ◦ g)
24π )
∧ (ι ◦ g)−1 d(ι ◦ g) .

Now, for the map gn defined above we have ι ◦ gn = g̃n ◦ ι so

1 (
gn∗ ω = − 2
trace (g̃n ◦ ι)−1 d(g̃n ◦ ι)
24π )
∧ (g̃n ◦ ι)−1 d(g̃n ◦ ι) ∧ (g̃n ◦ ι)−1 d(g̃n ◦ ι)
1 ∗( −1 −1 −1
)
=− ι trace((g̃ n dg̃n ) ∧ (g̃ n dg̃n ) ∧ (g̃ n dg̃n )) .
24π 2
Integrating both sides over S 3 shows that the degree of gn is the same as the
Chern number of the corresponding bundle. To show that these are both equal
to n we need the following identity.
Exercise 6.4.7 Let g, h : S 3 −→ S 3 and define gh : S 3 −→ S 3 by (gh)(q) =
g(q)h(q), where the product on the right-hand side is the quaternion product.
334 6. Characteristic Classes

Show that

trace((gh)−1 d(gh) ∧ (gh)−1 d(gh) ∧ (gh)−1 d(gh))

= trace(g −1 dg ∧ g −1 dg ∧ g −1 dg)
(6.4.10)
+ trace(h−1 dh ∧ h−1 dh ∧ h−1 dh)
+ d(3 trace(g −1 dg ∧ h−1 dh)).

Since g2 = g1 g1 , g3 = g2 g1 , . . . and since the integral over S 3 of the last term

in (6.4.10) is zero by Stokes’ Theorem, we find from (6.4.7) and induction
that the Chern number of the bundle corresponding to gn (and therefore the
degree of gn ) is n. All that remains of our stated objective then is to extend
this to the negative integers as well. Thus, for n = −1, −2, . . . we let gn :
(S 3 , 1) −→ (S 3 , 1) be the restriction to S 3 of the map g̃n : H −{0} −→ H −{0}
given by g̃n (q) = q n .
Exercise 6.4.8 Show that deg(g−1 ) and the Chern number of the bundle
corresponding to g−1 are both −1. Hints: Show that g−1 is an orienta-
tion reversing diffeomorphism of S 3 and so has degree −1. Next note that,
−1
on S 3 , g−1 dg−1 = qdq̄ and perform calculations analogous to those outlined
in Exercise 6.4.6.
Exercise 6.4.9 Use (6.4.10) to show that, for each n = −1, −2, . . . , deg(gn )
and the Chern number of the bundle corresponding to gn are both n.
For the record, we combine what we have just proved with the results of
Section 6.1 to obtain the following theorem.
Theorem 6.4.2 The Chern number for a U (1)-bundle over S 2 or an SU (2)-
bundle over S 4 is an integer that uniquely determines the equivalence class of
the bundle.
Remark: In the physics literature a bundle with Chern number n is said to
have topological charge −n (think of the minus sign as simply an attempt
to keep life interesting). For U (1)-bundles the topological charge is also called
the magnetic charge, while for SU (2)-bundles it is known as the instanton
number of the bundle. The idea here is that such a bundle is characterized
topologically by the strength of the Dirac monopole or instanton that can live
on it (see Sections 2.2 and 2.5, respectively).
For U (n)- and SU (n)-bundles the algebra of characteristic classes is gener-
ated by the Chern classes so, in particular, any formal linear combination of
Chern classes is itself a characteristic class. For example, the sum

c(P ) = 1 + c1 (P ) + c2 (P ) + · · ·

of all the Chern classes (necessarily finite since ck (P ) = 0 when 2k > dim X)
is called the total Chern class of the bundle. For bundles with other struc-
ture groups one can often isolate analogous finite generating sets and build
 6.5. Z2 -Čech Cohomology for Smooth Manifolds 335

from them interesting and useful characteristic classes. Those interested in

pursuing these matters should proceed to Chapter XII of [KN2] and, for ap-
plications to physics, to [EGH]. We intend now to take up a rather different
means of associating with certain bundles various cohomology classes of the
base manifold, one of which will tell us whether or not a spacetime manifold
admits a spinor structure. The “cohomology” is not that of de Rham, however.

6.5 Z2-Čech Cohomology for Smooth Manifolds

The cohomology theory relevant to the question of whether or not a space-
time admits spinor fields is called Čech cohomology with coefficients in the
group Z2 . This can be viewed as a (very) special case of what is called “sheaf
cohomology,” but in this section we take a more pedestrian approach. We
construct the groups of interest explicitly from a simple cover of the manifold
X and discuss only briefly the sheaf-theoretic proof that the construction is
actually independent of the choice of the simple cover. We then show how the
orthonormal frame bundle of a semi-Riemannian manifold X gives rise to a
Čech cohomology class of X (the 1st Stiefel-Whitney class) which we show
represents the obstruction to orientability in the sense that X is orientable
if and only if this cohomology class is trivial. When X is an oriented and
time oriented spacetime, another such Čech cohomology class (the 2nd Stiefel-
Whitney class) is shown to represent the obstruction to the existence of a
spinor structure for X. Since diffeomorphic manifolds have isomorphic Čech
cohomology groups we arrive at the rather surprising conclusion that the
existence of spinor structures is a topological matter and does not depend
on the particular Lorentz metric defined on the manifold. Before immersing
ourselves in the intricacies of a new cohomology theory, however, we will try
to get some feel for what the topology of a spacetime X has to do with one’s
ability to define spinor fields on X (a quick review of Sections 2.4 and 3.5
might be in order here).
We consider then an oriented and time oriented spacetime X. Of course,
if the topology of X is trivial (X ∼ = R4 ), then any principal bundle over X
is likewise trivial so, in particular, its oriented, time oriented, orthonormal
frame bundle can be identified with the product bundle

L+↑ ,→ X × L+↑ −→ X.

In this case one can build a spinor structure for X from the product bundle

SL(2, C) ,→ X × SL(2, C) −→ X

by simply defining λ to be the spinor map Spin on each fiber {x} × SL(2, C).
Indeed, X need not be trivial for this to work. All that’s required is that
the frame bundle be trivial and this is the case whenever it has a global
336 6. Characteristic Classes

cross-section, i.e., whenever X has a global orthonormal frame field

s : X −→ L(X) (PL ◦ s = idX ).

Suppose now that the frame bundle L(X) is not trivial. It will simplify
the discussion and not obscure any of the essential ideas if we assume that
X is simply connected. In this case it can occur that the frame bundle L(X)
is also simply connected (this cannot occur when the frame bundle is trivial
since π1 (X × L+↑ ) ∼
= π1 (X) ⊕ π1 (L+ ) ∼
↑
= π1 (X) ⊕ Z2 which is just Z2 when X
is simply connected). Now, fix an x0 ∈ X and consider the fiber PL−1 (x0 ) ∼ =
↑
L+ above x0 . In Section 3.4 we constructed a curve R1 (t), 0 ≤ t ≤ 2π, in
L+↑ representing a continuous rotation of the spatial coordinate axes through
360o . We now regard R1 as a curve in the fiber PL−1 (x0 ). Within PL−1 (x0 ), R1
is not nullhomotopic. However, if L(X) is simply connected, the loop R1 (t),
0 ≤ t ≤ 2π, is nullhomotopic in L(X) (see the figure).

–1
(x0) ≅ +

R1
(X)

Homotopy in (X) from R1

to the trivial loop

X x0

But a Dirac spinor field (for example) must change sign under( a )360◦
rotation at x0 (because R1 lifts to a path in SL(2, C) from 10 01 to
( )
− 10 01 ) and obviously does not change sign under “no rotation at all.” Since
the wavefunction cannot change continuously from ψ to −ψ but would have to
do so over the homotopy in L(X) from R1 to the trivial loop, it is not possible
to unambiguously define such spinor fields on X. With this as motivation we
now set about finding a characteristic class, the vanishing of which prohibits
this sort of topological obstruction.
Recall (Section 3.3) that an open cover {Uα }α∈A of a manifold X is said
to be simple if any finite intersection Uα0 ∩ · · · ∩ Uαj of its elements is ei-
ther empty or diffeomorphic to Rn (where n = dim X). Smooth manifolds
 6.5. Z2 -Čech Cohomology for Smooth Manifolds 337

admit simple covers and, indeed, any open cover of a smooth manifold has
a countable, locally finite, simple open refinement in which each element has
compact closure. Our construction of the Čech cohomology groups begins with
the selection of a locally finite, simple open cover U = {Uk }k=1,2,... for X with
each Ūk compact. We first build the Čech cohomology groups Ȟ j (U; Z2 ) cor-
responding to this cover and then discuss how one goes about showing that
the construction is actually independent of this choice. We will identify Z2
with the multiplicative group {−1, 1}.
For each integer j ≥ 0, a j -simplex (for U) is an ordered (j + 1)-tuple
σ = (k0 , . . . , kj ) of indices for which Uk0 ∩ · · · ∩ Ukj is nonempty (and
therefore diffeomorphic to Rn ). The support of σ is the nonempty open
set |σ| = Uk0 ∩ · · · ∩ Ukj . For i = 0, . . . , j, the i th -face of the j-simplex
σ = (k0 , . . . , kj ) is the (j − 1)-simplex σ i = (k0 , . . . , ki−1 , ki+1 , . . . , kj ). The
set of all j-simplexes for j ≥ 0 is called the nerve of U and denoted N (U).
For each j ≥ 0, a (Z2 -Čech) j -cochain is a function f which assigns to each
j-simplex σ = (k0 , . . . , kj ) an element f (σ) = f (k0 , . . . , kj ) of Z2 and that is
totally symmetric, i.e.,
( )
f kτ (0) , . . . , kτ (j) = f (k0 , . . . , kj )

for all permutations τ ∈ Sj+1 . The set Č j (U; Z2 ) of all j-cochains is an Abelian
group under pointwise multiplication, i.e.,

(f g)(k0 , . . . , kj ) = f (k0 , . . . , kj ) g(k0 , . . . , kj )

for all j-simplexes (k0 , . . . , kj ). Notice that the identity element of this group
assigns the value 1 ∈ Z2 to each j-simplex and that every element of Č j (U; Z2 )
is its own inverse.
Before proceeding with the construction we would like to persuade the
reader that Čech cochains actually arise in nature, as it were. Suppose
then that X is a semi-Riemannian manifold and consider the orthonor-
P
mal frame bundle O(m, n − m) ,→ F (X) −→ X. Select a simple cover
U = {Uk }k=1,2,... for X of the sort described above. Since each Uk is dif-
feomorphic to Rn , the bundle is trivial over Uk and so we may select an
orthonormal frame field on Uk , i.e., a cross-section sk : Uk −→ F (X).
Do this for each element of U. Now, if k0 and k1 are two indices for
which Uk0 ∩ Uk1 ̸= ∅, then sk0 and sk1 are related on Uk0 ∩ Uk1 by
sk1 = sk0 · gk0 k1 , where gk0 k1 is the transition function. Since this tran-
sition function takes values in O(m, n − m), its determinant is either 1
or −1 at each point Uk0 ∩ Uk1 . But Uk0 ∩ Uk1 is diffeomorphic to Rn
(and therefore connected) so det(gk0 k1 ) is either 1 everywhere on Uk0 ∩ Uk1
or −1 everywhere on Uk0 ∩ Uk1 . Thus, we may define f (k0 , k1 ) ∈ Z2
by ( )
f (k0 , k1 ) = det gk0 k1 .
338 6. Characteristic Classes

Since det(gk1 k0 ) = det(gk0 k1 −1 ) = (det(gk0 k1 ))−1 = det(gk0 k1 ),

f (k1 , k0 ) = f (k0 , k1 )

and we have defined a Čech 1-cochain f ∈ Č(U; Z2 ).

It will be convenient to define Č j (U; Z2 ) for j < 0 to be the trivial group
{1}. Now, for each j we define a coboundary operator

δ j : Č j (U; Z2 ) −→ Č j+1 (U; Z2 )

as follows: For j < 0, δ j is the trivial homomorphism, while, for j ≥ 0,

(δ j f )(k0 , . . . , kj+1 ) = f (k1 , . . . , kj+1 )f (k0 , k2 , . . . , kj+1 )

· · · f (k0 , . . . , kj )
∏
j+1 ( )
= f k0 , . . . , k̂i , . . . , kj+1 .
i=0

Since f is totally symmetric, so is δ j f and therefore δ j f ∈ Č j+1 (U; Z2 ).

Exercise 6.5.1 Show that, because Z2 is Abelian, each δ j is a homo-
morphism.
Notice that, if f ∈ Č 0 (U; Z2 ), then

(δ 0 f )(k0 , k1 ) = f (k0 )f (k1 )

and

(δ 1 (δ 0 f ))(k0 , k1 , k2 ) = (δ 0 f )(k1 , k2 )(δ 0 f )(k0 , k2 )(δ 0 f )(k0 , k1 )

= f (k1 )f (k2 )f (k0 )f (k2 )f (k0 )f (k1 )
= (f (k0 ))2 (f (k1 ))2 (f (k2 ))2
= 1.

Thus, δ 1 ◦δ 0 is the trivial homomorphism. We claim that this is true in general.

Lemma 6.5.1 δ j+1 ◦ δ j : Č j (U; Z2 ) −→ Č j+2 (U; Z2 ) is the trivial homomor-
phism.
Remark: As is customary one generally drops the indices on the cobound-
ary operators, writing them all as δ and abbreviating Lemma 6.5.1 as δ 2 = 1.

Proof: For j < 0 the result is obvious so assume j ≥ 0. Let

f ∈ Č j (U; Z2 ) and let (k0 , . . . , kj+2 ) be a (j + 2)-simplex. We must show
that ((δ j+1 ◦ δ j )f )(k0 , . . . , kj+2 ) = 1. But
 6.5. Z2 -Čech Cohomology for Smooth Manifolds 339

∏
j+2 ( )
(δ j+1 (δ j f ))(k0 , . . . , kj+2 ) = (δ j f ) k0 , . . . , k̂i , . . . , kj+2
i=0
= (δ j f )(k1 , k2 , . . . , kj+2 )(δ j f )(k0 , k2 , . . . , kj+2 )
· · · (δ j f )(k0 , k1 , . . . , kj+1 )
[ ]
= f (k2 , k3 , . . . , kj+2 )f (k1 , k3 , . . . , kj+2 ) · · · f (k1 , k2 , . . . , kj+1 )
[ ]
· f (k2 , k3 , . . . , kj+2 )f (k0 , k3 , . . . , kj+2 ) · · · f (k0 , k2 , . . . , kj+1 ) · · · ·
[ ]
· f (k1 , k2 , . . . , kj+1 )f (k0 , k2 , . . . , kj+1 ) · · · f (k0 , k1 , . . . , kj ) .

Each of these factors has precisely two of the indices k0 , . . . , kj+2 missing.
Moreover, for any pair of indices, there are precisely two factors in which
these indices are missing. Thus, any factor that takes the value −1 occurs
precisely twice and the product must be 1. 
Lemma 6.5.1 asserts that the sequence of cochain groups and coboundary
operators forms a cochain complex (Section 5.3) and so we may build its
cohomology theory in the usual way. A Čech j -coboundary is an element
of Č j (U; Z2 ) that is in the image of δ j−1 and the set

B̌ j (U; Z2 ) = Image δ j−1

of all such is a subgroup of Č j (U; Z2 ). A Čech j -cocycle is an element of

Č j (U; Z2 ) that is in the kernel of δ j and the set

Ž j (U; Z2 ) = ker δ j

of all such is a subgroup of Č j (U; Z2 ). Lemma 6.5.1 asserts that

B̌ j (U; Z2 ) ⊆ Ž j (U; Z2 )

so B̌ j (U; Z2 ) is a (normal) subgroup of Ž j (U; Z2 ). The quotient group

Ȟ j (U; Z2 ) = Ž j (U; Z2 )/B̌ j (U; Z2 )

is the j th Z2 - Čech cohomology group of the cover U. The elements of

Ȟ j (U; Z2 ) are equivalence classes [f ] of j-cocycles (δ j f = 1), where the equiv-
alence relation is f ′ ∼ f if and only if f ′ f −1 is a coboundary, i.e.,

f ′ = (δ j−1 h)f

for some h ∈ Č j−1 (U; Z2 ).

Let us pause for a moment to compute a simple example. Thus, we consider
again a semi-Riemannian manifold X with simple cover U = {Uk }k=1,2,...
P
and orthonormal frame bundle O(m, n − m) ,→ F (X) −→ X. Choose an
340 6. Characteristic Classes

orthonormal frame field sk : Uk −→ F (X) on each Uk and consider the 1-

cochain f (k0 , k1 ) = det(gk0 k1 ) constructed on page 391. We claim that f is a
1-cocycle and therefore determines a cohomology class in Ȟ 1 (U; Z2 ). To see
this we let (k0 , k1 , k2 ) be a 2-simplex and compute

(δ 1 f ) (k0 , k1 , k2 ) = f (k1 , k2 ) f (k0 , k2 ) f (k0 , k1 )

( ) ( ) ( )
= det gk1 k2 det gk0 k2 det gk0 k1
( ) ( ) ( )
= det gk0 k1 det gk1 k2 det gk2 k0
( )
= det gk0 k1 gk1 k2 gk2 k0

=1

since, by the cocycle condition satisfied by transition functions (page 32),

gk0 k1 gk1 k2 gk2 k0 is the identity matrix. Thus, f determines

[f ] ∈ Ȟ 1 (U; Z2 ).

One disturbing aspect of this last example is that the cohomology class [f ]
is not obviously intrinsic to the topology of X as is the case, for example,
for de Rham cohomology classes. Indeed, it appears to depend not only on
the simple cover U from which it is constructed, but even on the particu-
lar orthonormal frame fields sk : Uk −→ F (X) selected on each of the el-
ements of U. Let us show that at least this last dependence on the cross-
sections sk is only apparent. Thus, we let s′k : Uk −→ F (X) be another
orthonormal frame field on Uk for each k = 1, 2, . . . . Both sk and s′k give
rise to trivializations of the bundle defined on P −1 (Uk ) and these trivializa-
tions are related by a transition function gk defined on Uk (so s′k = sk · gk
on Uk ).
Exercise 6.5.2 Let (k0 , k1 ) be a 1-simplex with sk1 = sk0 · gk0 k1 and s′k1 =
s′k0 · gk′ 0 k1 on Uk0 ∩ Uk1 . Show that gk′ 0 k1 = gk−1
0
gk0 k1 gk1 on Uk0 ∩ Uk1 .
Now, defining a 1-cocycle f ′ just as we did for f , but using the cross-section
s′ rather than s gives
( ) ( )
f ′ (k0 , k1 ) = det gk0 k′ = det gk−1
0
gk0 k1 gk1
1
( ( ) ( ))
= det gk0 det gk1 f (k0 , k1 ).

Exercise 6.5.3 Show that h(k) = det(gk ) defines an element h of Č 0 (U; Z2 )

and that
( ) ( )
(δ 0 h)(k0 , k1 ) = det gk0 det gk1 .
 6.5. Z2 -Čech Cohomology for Smooth Manifolds 341

We conclude from Exercise 6.5.3 that

( )
f ′ (k0 , k1 ) = (δ 0 h)(k0 , k1 ) f (k0 , k1 ),

i.e.,
f ′ = (δ 0 h)f

so f and f ′ differ by a 0-coboundary. Thus, [f ′ ] = [f ] and the cohomology class

we constructed above does not depend on the choice of the local orthonormal
frames.
Exercise 6.5.4 Show that, if U consists of the single open set U0 = X (so
that, in particular, X is diffeomorphic to Rn ), then Ȟ 0 (U; Z2 ) ∼
= Z2 , but
Ȟ j (U; Z2 ) is trivial for every other j.
The issue of whether or not our construction is independent of the choice
of the simple cover U is not so easily resolved. Indeed, the standard ar-
gument by which this independence is established employs the methods of
axiomatic sheaf cohomology (for a clear and concise exposition that pro-
ceeds from the basics to the result in question see Chapter IV, Section A,
and Chapter VI, Sections A, B and D of [GR]). We will content ourselves
here with a few elementary observations. Notice first that since any two
simple covers of X of the type employed in our construction have a com-
mon refinement of the same type it would suffice to prove that Ȟ j (U; Z2 ) ∼ =
Ȟ j (V; Z2 ) whenever U = {Uk }k=1,2,... and V = {Vl }l=1,2,... are two such cov-
ers and V is a refinement of U. In this case one can select a refining map
ρ : {1, 2, . . .} −→ {1, 2, . . .} with the property that Vl ⊆ Uρ(l) for each
l = 1, 2, . . . . Notice that ρ is generally not uniquely determined, nor is it
one-to-one in general. However, any such ρ must carry a j-simplex of V onto
a j-simplex of U since Vl0 ∩ · · · ∩ Vlj ̸= ∅ implies Uρ(l0 ) ∩ · · · ∩ Uρ(lj ) ̸= ∅. Thus,
we can define, for j ≥ 0, a homomorphism

ρj : Č j (U; Z2 ) −→ Č j (V; Z2 )

by
( )
(ρj f )(l0 , . . . , lj ) = f ρ(l0 ), . . . , ρ(lj )

for each f ∈ Č j (V; Z2 ) and every j-simplex (l0 , . . . , lj ) for V. For j < 0 we
take ρj : Č j (U; Z2 ) −→ Č j (V; Z2 ) to be the trivial homomorphism.
Lemma 6.5.2 For each integer j, ρj+1 ◦ δ j = δ j ◦ ρj .
342 6. Characteristic Classes

δj
2

ρj ρ j +1

Čj ( ; 2) Č j+1 ( ; 2)
δj

Proof: For j < 0 this is trivial so assume j ≥ 0. Let f ∈ Č j (U; Z2 ). Then,

for any (j + 1)-simplex (l0 , . . . , lj+1 ) in V,

((δ j ◦ ρj )f )(l0 , . . . , lj+1 ) = δ j (ρj f )(l0 , . . . , lj+1 )

∏
j+1 ( )
= (ρj f ) l0 , . . . , ˆli , . . . , lj+1
i=0

∏
j+1 ( )
= f ρ(l0 ), . . . , ρ(lc
i ), . . . , ρ(lj+1 )
i=0
( )
= (δ j f ) ρ(l0 ), . . . , ρ(lj+1 )

= (ρj+1 (δ j f ))(l0 , . . . , lj+1 )

= ((ρj+1 ◦ δ j )f )(l0 , . . . , lj+1 ).

Thus, (δ j ◦ ρj )f = (ρj+1 ◦ δ j )f for any f ∈ Č j (U; Z2 ), as required. 

According to Lemma 6.5.2 the maps ρj : Č j (U; Z2 ) −→ Č j (V; Z2 ) determine
a cochain map and so induce maps

(ρj )# : Ȟ j (U; Z2 ) −→ Ȟ j (V; Z2 )

in cohomology. These homomorphisms are actually independent of the choice

of the refining map as the reader may wish to verify in the following exercise.
Exercise 6.5.5 Let τ : {1, 2, . . .} −→ {1, 2, . . .} be another map with the
property that Vl ⊆ Uτ (l) for each l = 1, 2, . . . For each j, let
τ j : Č j (U; Z2 ) −→ Č j (V; Z2 ) be the corresponding homomorphism. Define
K j : Č j (U; Z2 ) −→ Č j−1 (V; Z2 ) for j > 0 by

∏
j−1 ( )
j
(K f )(l0 , . . . , lj−1 ) = f ρ(l0 ), . . . , ρ(li ), τ (li ), . . . , τ (lj−1 )
i=0

and let K j be the trivial homomorphism for j ≤ 0. Show that

(ρj )(τ j )−1 = (δ j−1 ◦ K j )(K j+1 ◦ δ j )

 6.5. Z2 -Čech Cohomology for Smooth Manifolds 343

and conclude that the cochain maps {ρj } and {τ j } are algebraically homotopic
and therefore induce the same maps in cohomology.
The next step (which we do not intend to take) would be to show that
the homomorphisms (ρj )# are, in fact, isomorphisms (again, we refer to
[GR]). Granting this, we will henceforth call the groups we have constructed
the Z2 - Čech cohomology groups of X and denote them Ȟ j (X; Z2 ). Any
calculation of these groups for some specific manifold X would, of course,
begin with the selection of some convenient simple cover. For example,
Exercise 6.5.4 now implies that Ȟ 0 (Rn ; Z2 ) ∼ = Z2 and Ȟ j (Rn ; Z2 ) is
trivial for every other j. Our particular concern will be with isolating cer-
tain specific cohomology classes which act as “obstructions” to the exis-
tence of various desirable structures on a manifold. For example, the coho-
mology class in Ȟ 1 (X; Z2 ) associated with the orthonormal frame bundle
P
O(m, n − m) ,→ F (X) −→ X of a semi-Riemannian manifold X in the man-
ner described above is called the 1st Stiefel-Whitney class of X and is
denoted w1 (X). The next order of business is to show that the 1st Stiefel-
Whitney class w1 (X) ∈ Ȟ 1 (X; Z2 ) is the obstruction to orientability for a
semi-Riemannian manifold X.

Theorem 6.5.3 Let X be a semi-Riemannian manifold and

w1 (X) ∈ Ȟ 1 (X; Z2 ) its 1st Stiefel-Whitney class. Then X is orientable if
and only if w1 (X) is trivial.

Proof: We select an arbitrary locally finite, simple cover U = {Uk }k=1,2,...

for X and identify Ȟ j (X; Z2 ) with Ȟ j (U; Z2 ) for each j. Suppose first
that X is orientable. Then we may select local orthonormal frame fields
sk : Uk −→ F (X) consistent with the orientation of X. In particular, the
transition functions gk0 k1 that relate these local orthonormal frames are in
SO(m, n − m) and therefore satisfy det(gk0 k1) = 1 for each 1-simplex (k0 , k1 ).
But then the 1-cocycle f defined by f (k0 , k1 ) = det(gk0 k1), whose cohomology
class is w1 (X), is identically equal to 1 so w1 (X) is the identity element of
Ȟ 1 (U; Z2 ).
Now suppose, conversely, that w1 (X) is trivial. Select local orthonormal
frame fields sk : Uk −→ F (X), k = 1, 2, . . . and let gk0 k1 be the corresponding
transition functions. Let f be the 1-cocycle defined by f (k0 , k1 ) = det(gk0 k1 ).
Since w1 (X) = [f ], f is a coboundary, i.e., f = δ 0 f0 for some f0 ∈ Č 0 (U; Z2 ).
For each index k, f0 (k) = ±1 so we can select a gk ∈ O(m, n − m) with
det(gk ) = f0 (k). Now define new local orthonormal frames s′k : Uk −→ F (X)
by s′k = sk · gk for k = 1, 2, . . . . According to Exercise 6.5.2, the tran-
sition functions gk′ 0 k1 corresponding to the new trivializations are given
by
gk′ 0 k1 = gk−1
0
gk0 k1 gk1
344 6. Characteristic Classes

so
( ) ( ) ( ) ( )
det gk′ 0 k1 = det gk0 det gk1 det gk0 k1
= f0 (k0 ) f0 (k1 ) f (k0 , k1 )
= (δ0 f0 ) (k0 , k1 ) f (k0 , k1 )
= (f (k0 , k1 ))2
= 1.

Since s′k1 = s′k0 · gk′ 0 k1 for every 1-simplex (k0 , k1 ), the orthonormal frames
s′k0 and s′k1 therefore determine the same orientation at each p ∈ Uk0 ∩ Uk1 .
Consequently, the collection of all s′k : Uk −→ F (X), k = 1, 2, . . . , determines
an orientation on X. 

We now restrict our attention to an oriented, time oriented spacetime X

and seek a Čech cohomology class which represents the obstruction to the
existence of a spinor structure for X in the same sense that the 1st Stiefel-
Whitney class represents the obstruction to the existence of an orientation for
a semi-Riemannian manifold. We recall (Section 2.4) that a spinor structure
for X consists of a principal SL(2, C)-bundle
P
SL(2, C) ,→ S(X) −→
S
X

and a map λ : S(X) −→ L(X) to the oriented, time oriented, orthonormal

frame bundle such that
SL ( 2, C ) S(X) s

λ X

+ (X)

PL ◦ λ = PS
and
λ(p · g) = λ(p) · Spin (g)

for all p ∈ S(X) and all g ∈ SL(2, C):

S (X) × SL (2, C) S(X) s

λ × Spin λ X
(X)× + (X)
 6.5. Z2 -Čech Cohomology for Smooth Manifolds 345

Let U = {Uk }k=1,2,... be a locally finite, simple open cover of X with each
Ūk compact and choose local orthonormal frame fields sk : Uk −→ L(X),
k = 1, 2, . . . . The corresponding transition functions

gk0 k1 : Uk0 ∩ Uk1 −→ L+↑

map into L+↑ for every 1-simplex (k0 , k1 ). Because each Uk0 ∩ Uk1 is diffeomor-
phic to R4 and Spin : SL(2, C) −→ L+↑ is a covering map, these transition
functions lift to SL(2, C):

SL ( 2, C )

k0 k1 Spin

U k ∩ Uk +
0 1
k0 k1

We can clearly select these lifts in such a way that

g̃k1 k0 = g̃k0 k1 −1

for all k0 , k1 = 1, 2, . . . . Furthermore, since Spin is a homomorphism,

( )
Spin g̃k0 k1 g̃k1 k2 g̃k2 k0 = gk0 k1 gk1 k2 gk2 k0 = id.

But the kernel of Spin is { ±id } so

g̃k0 k1 g̃k1 k2 g̃k2 k0 = ±id

at each point in Uk0 ∩ Uk1 ∩ Uk2 . But this intersection is connected (diffeo-
morphic to R4 ) and so this map is either id everywhere or −id everywhere.
Thus, we can define a map Z from 2-simplexes to Z2 by

g̃k0 k1 g̃k1 k2 g̃k2 k0 = Z(k0 , k1 , k2 )(id).

Exercise 6.5.6 Show that Z is a 2-cocycle and therefore determines a coho-

mology class [Z] ∈ Ȟ 2 (X; Z2 ).
Once again it would appear that the cohomology class [Z] we have defined
depends on not only the initial choice of local orthonormal frames sk , but also
on the lifts g̃k0 k1 of the corresponding transition functions. We show now that
this is not, in fact, the case. Thus, we suppose s′k : Uk −→ L(X), k = 1, 2, . . . ,
are also cross-sections. Then s′k = sk · gk on Uk for some gk : Uk −→ L+↑ . As
in Exercise 6.5.2 we let sk1 = sk0 · gk0 k1 and s′k1 = s′k0 · gk′ 0 k1 on Uk0 ∩ Uk1
346 6. Characteristic Classes

and conclude that

g̃k′ 0 k1 = gk−1
0
gk0 k1 gk1 .
Let g̃k0 k1 and g̃k′ 0 k1 be lifts and define the 2-cocycles Z and Z ′ as above. To
show that these determine the same cohomology class we must prove that

Z ′ = (δ 1 h)Z

for some 1-cochain h. Lift the maps gk to SL(2, C).

SL ( 2, C )

gk Spin

Uk gk +

Now define, for each 1-simplex (k0 , k1 ), a map

hk1 k0 : Uk0 ∩ Uk1 −→ SL(2, C)

by
hk0 k1 = g̃k0 k1 g̃k1 g̃k′ 1 k0 g̃k−1
0
.

Since Spin is a homomorphism, Spin ◦ hk0 k1 takes the value id ∈ L+↑ for each
x in Uk0 ∩ Uk1 . Thus, hk0 k1 maps into {±id} ⊆ SL(2, C) and connectivity of
Uk0 ∩ Uk1 implies that it must be constant. Thus, we may define a 1-cochain
h by
h(k0 , k) == hk0 k1 (x)

where x is any element of Uk0 ∩ Uk1 .

Exercise 6.5.7 Show that Z ′ = (δ 1 h)Z.

Thus, we have a well-defined cohomology class [Z] which we will call the 2nd
Stiefel-Whitney class of X and denote w2 (X). To unearth its revelance to
the existence of spinor structures on X we begin with the simple observation
that w2 (X) is trivial if and only if there exist lifts g̃k0 k1 of the transition
functions which satisfy

g̃k0 k1 g̃k1 k2 g̃k2 k0 = id (6.5.1)

at each point of Uk0 ∩ Uk1 ∩ Uk2 . Of course, if such lifts exist,

Z(k0 , k1 , k2 ) = 1 so w2 (X) = [Z] is trivial. Suppose, conversely, that w2 (X) is
trivial. Select lifts g̃k0 k1 and define the 2-cochain Z(k0 , k1 , k2 ) as above. Since
w2 (X) = [Z] is trivial there exists a 1-cochain f such that Z = δ 1 f , i.e.,
 6.5. Z2 -Čech Cohomology for Smooth Manifolds 347

Z(k0 , k1 , k2 ) = f (k1 , k2 ) f (k0 , k2 ) f (k0 , k1 )

for every 2-simplex (k0 , k1 , k2 ). Define

g̃k′ 0 k1 : Uk0 ∩ Uk1 −→ SL(2, C)

by
g̃k′ 0 k1 = g̃k0 k1 f (k0 , k1 ).

Then

g̃k′ 0 k1 g̃k′ 1 k2 g̃k′ 2 k0 = g̃k0 k1 f (k0 , k1 ) g̃k1 k2 f (k1 , k2 ) g̃k2 k0 f (k2 , k0 )

= (Z(k0 , k1 , k2 ))2 (id)
= id

so the primed lifts satisfy (6.5.1). Now we will prove our major result by show-
ing that the existence of lifts satisfying (6.5.1) is equivalent to the existence
of a spinor structure on X.

Theorem 6.5.4 Let X be an oriented, time oriented spacetime. Then X

admits a spinor structure if and only if the 2nd Stiefel-Whitney class w2 (X) ∈
Ȟ 2 (X; Z2 ) of X is trivial.

Proof: For both parts of the proof we fix at the outset a locally finite, simple
open cover U = {Uk }k=1,2,... for X.
Suppose first that a spinor structure

S (X) × SL (2, C) S(X) s

λ × Spin λ X
(X)× + (X)

P
for X exists. The spinor bundle SL(2, C) ,→ S(X) −→ S
X is trivial over each
Uk so we can choose a cross-section
( )
s̃k : Uk −→ PS−1 (Uk ) = λ−1 PL−1 (Uk )

for each k = 1, 2, . . . . Define

sk : Uk −→ PS−1 (Uk )

by
sk = λ ◦ s̃k .
348 6. Characteristic Classes

P
Then each sk is a cross-section of the frame bundle L+↑ ,→ L(X) −→ L
X
because PL ◦ λ = Ps . These cross-sections give rise to trivializations and
therefore transition functions and we denote the transition functions by g̃k0 k1
and gk0 k1 , respectively.

Exercise 6.5.8 Show that, on Uk0 ∩ Uk1 , Spin ◦ g̃k0 k1 = gk0 k1 .

Thus, the g̃k0 k1 are lifts of the gk0 k1 . Since they are transition functions of
a bundle, they satisfy the cocycle condition (6.5.1) and so we conclude that
w2 (X) is trivial.
For the converse we begin with local orthonormal frame fields
sk : Uk −→ PL−1 (Uk ), k = 1, 2, . . . and denote by gk0 k1 : Uk0 ∩ Uk1 −→ L+↑ the
corresponding transition function for each 1-simplex (k0 , k1 ). Our assumption
now is that w2 (X) is trivial and we have seen that it follows from this that
there exist lifts
SL ( 2, C )

k0 k1 Spin

U k ∩ Uk +
0 1
k0 k1

satisfying g̃k1 k0 = g̃k−1

0 k1
and (6.5.1). The Reconstruction Theorem therefore
guarantees the existence of a unique (up to equivalence) SL(2, C)-bundle
P
SL(2, C) ,→ S(X) −→
S
X

with the g̃k0 k1 as transition functions. All that remains is to define a map

λ : S(X) −→ L(X)

satisfying PL ◦ λ = PS and λ(p · g) = λ(p) · Spin (g). For this we will define
λ above each Uk and then show that the definitions agree on any nonempty
intersections.
The trivialization Ψk : PL−1 (Uk ) −→ Uk × L+↑ determined by sk is given by

Ψk (sk (x) · g) = (x, g)

for all x ∈ Uk and g ∈ L+↑ . Let Ψ̃k : PS−1 (Uk ) −→ Uk × SL(2, C),
k = 1, 2, . . . , be trivializations of S(X) related by the transition functions
g̃k0 k1 and let s̃k : Uk −→ PS−1 (Uk ), k = 1, 2, . . . , be the corresponding canon-
ical cross-sections. Define

λk : PS−1 (Uk ) −→ PL−1 (Uk )

 6.5. Z2 -Čech Cohomology for Smooth Manifolds 349

by
λk = Ψ−1
k ◦ (idUk × Spin) ◦ Ψ̃k

for each k = 1, 2, . . . .
Exercise 6.5.9 Show that λk (s̃k (x) · g) = sk (x) · Spin(g) for all x ∈ Uk and
all g ∈ SL(2, C) and conclude that, on PS−1 (Uk ),

PL ◦ λk = PS

and
λk (p · g) = λk (p) · Spin (g).

To show that the maps λk determine a map λ : S(X) −→ L(X) satisfying

PL ◦ λ = PS and λ(p · g) = λ(p) · Spin(g) it will be enough to show that they
agree on any nonempty intersection
( ) ( ) ( )
PS−1 Uk0 ∩ PS−1 Uk1 = PS−1 Uk0 ∩ Uk1 .

But, for x ∈ Uk0 ∩ Uk1 , we have

( )
λk1 s̃k1 (x) · g = sk1 (x) · Spin(g)

s̃k1 (x) = s̃k0 (x) · g̃k0 k1 (x)

and
sk1 (x) = sk0 (x) · gk0 k1 (x)

so
( ) ( ( ))
λk0 s̃k1 (x) · g = λk0 s̃k0 (x) · g̃k0 k1 (x)g
( )
= sk0 (x) · Spin g̃k0 k1 (x)g
( ( ))
= sk0 (x) · Spin g̃k0 k1 (x) · Spin(g)
( )
= sk0 (x) · gk0 k1 (x) · Spin(g)

= sk1 (x) · Spin(g)

( )
= λk1 s̃k1 (x) · g

as required. 
350 6. Characteristic Classes

Theorem 6.5.4 has some rather obvious consequences that are of sufficient
importance to be recorded officially. If X happens to be diffeomorphic to R4 ,
then Ȟ 2 (X; Z2 ) is trivial so w2 (X) is trivial and X admits a spinor structure.
Corollary 6.5.5 A time oriented spacetime diffeomorphic to R4 admits a
spinor structure.
This applies, in particular, to Minkowski spacetime and the Einstein-deSitter
spacetime. The deSitter spacetime and the Einstein cylinder are diffeomorphic,
not to R4 , but to S 3 × R. That they nevertheless do admit spinor structures
will follow from our next result.
Corollary 6.5.6 Let X be an oriented, time oriented spacetime and suppose
that the oriented, time oriented orthonormal frame bundle
↑ PL
L+ ,→ L(X) −→ X is trivial. Then X admits a spinor structure.
Proof: By assumption, there exists a global cross-section s : X −→ L(X).
Letting U = {Uk }k=1,2,... be a locally finite simple open cover we define sk :
Uk −→ PL−1 (Uk ) by sk = s|Uk . Since sk0 = sk1 · gk0 k1 on Uk0 ∩ Uk1 , all of
the corresponding transition functions gk0 k1 are identically equal to id ∈ L+↑ .
These surely lift to maps g̃k0 k1 : Uk0 ∩ Uk1 −→ SL(2, C) that take the value
id ∈ SL(2, C) everywhere and these satisfy (6.5.1) so w2 (X) is trivial and X
admits a spinor structure. 
P
Remark: Geroch [G1] has shown that the triviality of L+↑ ,→ L(X) −→ L
X
actually characterizes the existence of spinor structures for oriented, time
oriented (noncompact) spacetimes.
Since we have shown in Chapter 3 that both deSitter spacetime and the Ein-
stein cylinder satisfy the hypotheses of Corollary 6.5.6., they too admit spinor
structures.
Appendix

Seiberg-Witten Gauge Theory

Spin 12 -electrodynamics is a gauge theory of a U (1)-gauge field coupled to a
spinor field and, although the connection is defined on a different U (1)-bundle
and the spinor field is of a different type, so is Seiberg-Witten theory. The
significance of the latter, however, resides in an entirely different arena and
one cannot truly appreciate this significance without placing the theory in its
historical context. This we will attempt to do in Section A.l. The (rather sub-
stantial) algebraic preliminaries required just to describe the basic elements
of the theory are introduced in Section A.2 and then we move on to the field
equations and their moduli space of solutions. Much of what we have to say
lies in considerably deeper waters than the main body of the text and so we
must content ourselves with something akin to an initial geographical survey
of the terrain with signposts to the literature for those who wish to dive in.
What we offer is really a continuation of the story begun in Appendix B of
[N4] and we will need to assume that our reader is aquainted with this.

A.1 Donaldson Invariants and TQFT

The first incursion of gauge theory into topology was Donaldson’s theorem
on compact, simply connected, smooth 4-manifolds M with b+ 2 (M ) = 0 and
definite intersection form. The proof is based on an analysis of the moduli
P1
space of anti-self-dual connections on the SU (2)-bundle SU (2) ,→ P1 −→ M
over M with Chern number 1. The structure of this moduli space and Don-
aldson’s proof of his theorem are sketched (ever so briefly) in Appendix B of
[N4]. Much of the analysis described there can be carried out for the bundles
P
SU (2) ,→ Pk −→ k
M with positive Chern number k > 0 and without the
assumption that b+ 2 (M ) = 0, although the results are rather different. We
will not repeat this analysis here, but will simply ask that our reader become
aquainted with Appendix B of [N4] and state the end result.
Throughout this discussion M will denote a compact, simply connected,
oriented, smooth 4-manifold with b+ 2 (M ) > 0 (soon we will impose additional
P
2 (M ) and explain why). For each k > 0 we let SU (2) ,→ Pk −→
conditions on b+ k

M be the principal SU (2)-bundle over M with Chern number k, A(Pk ) the

(affine) space of all connection 1-forms on Pk and G(Pk ) the gauge group of
automorphisms of Pk . As usual, G(Pk ) acts on A(Pk ) by pullback and two
connections ω, ω ′ ∈ A(Pk ) are gauge equivalent if there is an f ∈ G(Pk )
such that ω ′ = f ∗ ω. The moduli space of all gauge equivalence classes [ω] is

G.L. Naber, Topology, Geometry and Gauge fields: Interactions, 351

Applied Mathematical Sciences 141, DOI 10.1007/978-1-4419-7895-0,
© Springer Science+Business Media, LLC 2011
352 Appendix

denoted B(Pk ) = A(Pk )/G(Pk ). The stabilizer of ω is the subgroup of G(Pk )

consisting of those f that leave ω fixed and always contains a copy of Z2 . If
the stabilizer is precisely Z2 , then ω is said to be irreducible (otherwise, it is
reducible). The (open) subset of A(Pk ) consisting of irreducible connections
is denoted Â(Pk ) and its moduli space is B̂(Pk ) = Â(Pk )/G(Pk ).
Now we let g be a Riemannian metric on M and consider the set Asd(Pk , g )
of all ω ∈ A(Pk ) that are g -anti-self-dual, i.e., satisfy ∗ F ω = −F ω , where ∗ is
the corresponding Hodge star operator and F ω is the curvature of ω thought
of as an ad Pk -valued 2-form on M . This might well be empty (when k = 1
and b+ 2 (M ) = 0 a theorem of Taubes prohibits this, but we are assuming
b+
2 (M ) > 0). The moduli space

M(Pk , g ) = Asd(Pk , g )/G(Pk )

of g -anti-self-dual connections on Pk could therefore also be empty. The as-
sumption that b+ 2 (M ) is positive has beneficial consequences as well, however.
Under this assumption Donaldson proved a generic metrics theorem to the
effect that there is a dense subset of the space R of Riemannian metrics on M
with the property that for any g in this set every g -anti-self-dual connection
ω on Pk is irreducible and the moduli space M(Pk , g ) is either empty or a
smooth orientable manifold of dimension
( )
8k − 3 1 + b+2 (M )

(moreover, an orientation for M(Pk , g ) is canonically determined by choosing

an orientation for the vector space H+2
(M ; R )).
Remark: The restriction on b+ 2 (M ) arises because the subset of R consist-
ing of those g for which reducible g -ASD connections exist is a countable
union of smooth submanifolds of codimension b+ +
2 (M ). If b2 (M ) = 0, then re-
ducibles are generically unavoidable (and we saw in Appendix B of [N4] that
this is a good, not a bad thing as it leads to the cone singularities required
for the cobordism proof of Donaldson’s Theorem).
With this information in hand Donaldson set about constructing a
sequence γd (M ), d = 0, 1, 2, . . ., of differential topological invariants of M
(this means that the invariants agree for two smooth 4-manifolds that are
diffeomorphic, but need not agree if the manifolds are merely homeomorphic,
but not diffeomorphic). The first γ0 (M ) is simply an integer, but for d ≥ 1,
γd (M ) : H2 (M ) −→ Z[ 21 ] is a function on the second homology of M with val-
ues that are integer multiples of 12 . They are called the Donaldson invariants
of M . Their definition and, especially, their computation is enormously com-
plex, but these computations produced remarkable insights into the structure
of smooth 4-manifolds. Eventually they were superceded by invariants arising
from Seiberg-Witten gauge theory, but to understand how this came about
one must have at least some general, intuitive sense of how they are defined
(those in search of honest definitions should proceed first to [M2] and then
 A.1. Donaldson Invariants and TQFT 353

move on to [DK]). In the brief description that follows we assume that we

have selected a generic Riemannian metric g and fixed some orientation of
H2+ (M ; R).
Notice that, by an appropriate arrangement of b+ 2 (M ) and the Chern num-
ber k, it is entirely possible for the dimension 8k−3(1+b+ 2 (M )) of M(Pk , g ) to
come out zero. Then M(Pk , g ) is either empty or an oriented, 0-dimensional
manifold, i.e., it is a set of isolated points [ω] each equipped with a sign ±1
which we will denote (−1)[ω] . As it happens, in this case (and only in this
case) M(Pk , g ) is compact, i.e., finite, so we can define


 0 , M(Pk , g ) = ∅


γ0 (M ) = ∑ .

 (−1) [ω]
, M(P , g ) ̸
= ∅

 k
[ω]∈M(Pk ,g )

Regrettably, this integer need not be an invariant and may, in fact, depend
on the choice of generic metric g . To prevent this one must ensure that not
only is it possible to choose a metric for which there are no reducible ASD
connections, but also that a generic variation of that metric (generic path in
the space R of Riemannian metrics on M ) does not introduce reducibles. For
this one needs the set of metrics in R for which there are reducibles to be
sufficiently “thin” and this, by the Remark above, means that b+2 (M ) must be
sufficiently large. One can show that

b+
2 (M ) > 1

will do and so we henceforth make this assumption about M (one more

assumption on b+ 2 (M ) is forthcoming). With this we have the 0-dimensional
Donaldson invariant γ0 (M ) ∈ Z of M .
The definition of γd (M ) for d > 0 is a very great deal more complicated.
Even the briefest sketch would require substantial effort and is not really
necessary for what we have in mind here. For our purposes it will suffice to
have a naive, intuitive picture of the idea that lies behind the definition and
some appreciation of what makes it so naive. Very roughly, the idea is that the
values of γd (M ) will be obtained by integrating over the (finite-dimensional)
moduli space certain carefully selected differential forms. To implement this
idea Donaldson defined a map

µ : H2 (M ) −→ H 2 (M(Pk , g ))

(think of this as assigning 2-forms on M(Pk , g ) to surfaces in M ) and wedged

enough of the images together to get a form with rank dim M(Pk , g ) that can
then be integrated over M(Pk , g ). Of course, any wedge product of 2-forms
has even rank so this cannot succeed unless dim M(Pk , g ) = 8k−3(1+b+ 2 (M ))
is even, i.e., unless b+
2 (M ) is odd. Thus, we arrive at our final restriction on
354 Appendix

b+
2 (M ). Henceforth, we assume

2 (M ) ≡ 1 mod 2 and b2 (M ) > 1.

b+ +

Remark: The definition of Donaldson’s µ-map is technical and involves

a number of ideas that we have not encountered (Pontryagin characteristic
classes and an operation from algebraic topology called the slant product).
The details can be found in [M2] and [DK]. In the special case in which k is
odd a rough idea of what is behind the definition can be described as follows:
Donaldson constructs a certain auxiliary SU (2)-bundle SU (2) ,→ P −→ M ×
M(Pk , g ) over the product M × M(Pk , g ). Its second Chern class c2 (P ) is an
element of H 4 (M × M(Pk , g )). There is a general result on the cohomology
of
⊕ products (called the Künneth formula) that gives H 4 (M × M(Pk , g )) ∼ =
i+j=4 H (M ) ⊗ H (M(Pk , g )) so c2 (P ) has a (2, 2)-part (α, β) ∈ H (M ) ⊗
i j 2

H 2 (M(Pk , g )) Then, if x is an element of H2 (M ) and we identify x with a

smoothly embedded surface in M representing it, we can define
(∫ )
µ(x) = α β ∈ H 2 (M(Pk , g )).
x

Now we are prepared to describe our “naive” definition of γd (M ) when

8k − 3(1 + b+ +
2 (M )) > 0. First we fix the Chern number k. Since b2 (M ) is odd
we can write ( )
8k − 3 1 + b+ 2 (M ) = 2dk

for some positive integer dk . For any x ∈ H2 (M ), µ(x) is in H 2 (M(Pk , g )) so

dk
µ(x) ∧ · · · ∧ µ(x) ∈ H 2dk (M(Pk , g )) and we set
∫
dk
γdk (M )(x) = µ(x) ∧ · · · ∧ µ(x).
M(Pk ,g )

Now for any positive integer d with

3( )
d≡− 1 + b+
2 (M ) mod 4
2
we can select a k for which 2d = 8k − 3(1 + b+
2 (M )) so that d = dk and we
have “defined” γd (M ).
Remarks: There are probably no more than a thousand things wrong
with this “definition” and setting it all straight requires a huge amount
of technical labor. Nevertheless, morally at least it represents the correct
idea and is, in fact, the way these invariants are often viewed by physi-
cists (more on this shortly). We will offer just a brief tour of what is wrong
and what it takes to make it right. Most of the fuss can be traced to the
fact that, when its dimension is positive, M(Pk , g ) is never compact and so
one can generally not integrate over it. Deep analytical results of Uhlenbeck
 A.1. Donaldson Invariants and TQFT 355

and Taubes and much labor led Donaldson to a compactification M̄(Pk , g ),

called the Uhlenbeck compactification, of M(Pk , g ) and an extension
µ̄ : H2 (M ) −→ H 2 (M̄(Pk , g )) of the µ-map to it. Unfortunately, although
it is compact, M̄(Pk , g ) is not a manifold and so integration no longer makes
sense. Algebraic topology presents the alternative of pairing the cohomology
class µ̄(x) ∧ · · · ∧ µ̄(x) with what is called the fundamental homology class
[M̄(Pk , g )] of M̄(Pk , g ). Alas, M(Pk , g ) has a fundamental class only for suf-
ficiently large k (the so-called “stable range” k > 34 (1+b+ 2 (M )) or, equivalently,
dk > 32 (1 + b+2 (M )). For d ≡ − 3
2 (1 + b +
2 (M )) mod 4 and d > 32 (1 + b+
2 (M )) one
can define γd (M ) in the manner just described. Removing the stable range
condition and the required mod 4 congruence requires another µ-map
( )
µ : H0 (M ) −→ H 4 M(Pk , g ) ,
a detour around the fact that this one does not extend fully to M̄(Pk , g ) and
a blow-up formula relating γd (M ) to γd (M # CP ), where M # CP is the
2 2

connected sum of M and CP (which is obtained by deleting an open 4-ball

2

from both M and CP and identifying the boundary spheres). Incidentally, it

2

is only in this last step that the values of γd (M ) become multiples of 12 rather
than just integers. Those inclined to find out what all of this really means are
referred to [M2] and [DK].
It is certainly not our intention here to compute Donaldson invariants or use
them to obtain topological information about 4-manifolds. Rather we would
like to sketch how, by adopting a slightly different perspective, they lead, by
way of quantum field theory, to the Seiberg-Witten theory that is our real
concern here. To keep the discussion as uncluttered as possible we will focus
most of our attention on γ0 (M ) and will economize on notation by writing P
for Pk , G for G(Pk ), A for A(Pk ), etc.
The gauge group G does not act freely on the space  of irreducible con-
nections since even irreducibles have a Z2 stabilizer. However, Ĝ = G/Z2 does
act freely on  so we have an infinite-dimensional principal bundle
Ĝ ,→ Â −→ B̂.
We build a vector bundle associated to this principal bundle as follows: We
claim that there is a smooth left action of Ĝ on the (infinite-dimensional)
vector space Λ2+ (M , ad P ) of self-dual 2-forms on M with values in the adjoint
bundle ad P . To see this we think of G as the group of sections of the nonlinear
adjoint bundle Ad P under pointwise multiplication. Since the elements of
Λ2+ (M , ad P ) take values in the su(2)-fibers of ad P, G acts on these fibers by
conjugation. Moreover, conjugation takes the same value at ±f ∈ G so this
G-action on Λ2+ (M , ad P ) descends to a G/Z2 = Ĝ-action. Thus, we have an
associated vector bundle

 ×Ĝ Λ2+ (M, ad P ),

356 Appendix

the elements of which are equivalence classes [ω, α] = [ω · f, f −1 · α ] with

ω ∈ Â, α ∈ Λ2+ (M , ad P ) and f ∈ Ĝ.
Now recall that sections of associated vector bundles can be identified with
equivariant maps from the principal bundle space to the vector space fiber.
In our case we have an obvious map from  to Λ2+ (M , ad P ), namely, the
self-dual curvature map

F + : Â −→ Λ2+ (M, ad P )
F + (ω) = F +
ω =
1
2 (F ω + ∗ F ω ) .

Since the Ĝ-action on  is by conjugation and curvature transforms by con-

jugation under a gauge transformation, F + is equivariant:
−1 +
F + (ω · f ) = F +
ω·f = f F ω f = f −1 · F +
ω =f
−1
· F + (ω).

F + can therefore be identified with the section

s+ : B̂ −→ Â ×Ĝ Λ2+ (M, ad P )

s+ ([ω]) = [ω, F +
ω]

of our vector bundle. Now notice that the moduli space M of anti-self-dual
connections (F +ω = 0) is precisely the zero set of the section s+ (for generic
g all such connections are irreducible). It is a general fact that the base of
any smooth vector bundle is diffeomorphic to the image of any cross-section
of the vector bundle (in particular, the one that picks out the zero element in
each fiber). Thus, we can identify B̂ with the image of the zero section

s0 : B̂ −→ Â ×Ĝ Λ2+ (M, ad P )

s0 ([ω]) = [ω, 0].

We conclude that ( ) ( )
M = s+ B̂ ∩ s0 B̂ .

In the case in which M is 0-dimensional each point in this intersection acquires

a sign ±1 and the Donaldson invariant γ0 (M ) is the sum of these signs.
This last interpretation of γ0 (M ) is reminiscent of a very famous result
in topology that we now digress momentarily to describe. We have seen
(Section 3.3) that a smooth vector field V on a manifold X can be iden-
tified with a cross-section V : X −→ T X of the tangent bundle of X. The
image V (X) of X is a submanifold of T X diffeomorphic to X and the same is
true of the zero cross-section (vector field) V 0 : X −→ T X. We will identify
X with V 0 (X). Then the zeros (singularities) of V are just the points of
V (X) ∩ V 0 (X). Now assume that X is compact and oriented. Then T X is
also oriented and we can give V (X) the orientation that makes V orientation
 A.1. Donaldson Invariants and TQFT 357

preserving. Now we proceed exactly as in the definition of the intersection

form (Appendix B of [N4]) to define the index Ind(V ) of V : Perturb V
slightly so that V (X) and V 0 (X) intersect transversally and therefore (since
dim T X = 2 dim X) in a finite set of isolated points. Such a point p is as-
signed the value 1 if an oriented basis for Tp (V 0 (T X)) followed by an ori-
ented basis for Tp (V (T X)) gives an oriented basis for Tp (T X); otherwise,
it is assigned the value-1. Then Ind(V ) is the sum of these values over all
points in V 0 (X) ∩ V (X). The Poincaré-Hopf Theorem then asserts that
Ind(V ) is equal to the Euler characteristic χ(X) of X (a proof of this can be
found in [MT]). The point is that we have a topological invariant of X that
is determined by cross-sections of a vector bundle over X and in much the
same way (sum of signed points in the zero set of the section) that γ0 (M ) is
determined by the section s+ of  ×Ĝ Λ2+ (M , ad P ). Notice also that one can
generalize the definition of Ind(V ) to define the index of any section s of any
vector bundle E −→ X that shares a few properties of the tangent bundle (it
should be orientable in the sense that orientations can be smoothly supplied
to the fibers and the fiber dimension should be the same as dim X).
To understand how physics enters these considerations (and we must un-
derstand this because that’s where Seiberg-Witten came from) we will take
a moment to describe another very famous result of topology/geometry that
provides yet another way to calculate χ(X), this time by integration of a cer-
tain characteristic class. Since it turns out that χ(X) is always zero when
dim X is odd we will restrict ourselves to (compact, oriented) manifolds X
of dimension 2k for some k ≥ 1. The result we need is the Gauss-Bonnet-
Chern Theorem which asserts that
∫
χ(X) = e(X),
X

where e(X) is called the Euler class of X. This is a characteristic class that
can be defined by the Chern-Weil procedure described in Section 6.3. Briefly,
the construction goes like this: Choose a Riemannian metric on X and consider
the corresponding oriented, orthonormal frame bundle

SO(2k) ,→ F+ (X) −→ X

(Exercise 3.3.17). Choose a connection ω on this bundle with curvature Ω. To

apply the Chern-Weil procedure we now need a symmetric polynomial on the
Lie algebra so(2k) of SO(2k) that is invariant under the adjoint action. The
appropriate choice this time is called the Pfaffian

P f : so(2k) −→ R.

Although there are much more elegant ways to define this (see [MS]) we will
opt for a simple-minded formula in terms of the matrix entries. Let A = (Aij )
358 Appendix

be an element of so(2k) and define

1 ∑
P f (A) = (−1)σ Aσ(1)σ(2) · · · Aσ(2k−1)σ(2k) ,
2k k!
σ∈S2k

where S2k is the group of permutations of {1, . . . 2k} and (−1)σ denotes the
sign of the permutation σ. One can check, for example, that if
 

 0 λ 



1 


 
 0

 


 −λ 

 1 0 
A= . .
 . 

 


 0 λk 


 





0 



−λk 0

then P f (A) = λ1 · · · λk (in fact, it is always the case that (P f (A))2 = det A).
One can show that P f is invariant under the adjoint action of SO(2k) on
so(2k) so Chern-Weil guarantees that if we write Ω = (Ωij ) as a matrix of
1-forms (and normalize by − 2π 1
), then
( )
1 (−1)k ∑
Pf − Ω = (−1)σ Ωσ(1)σ(2) ∧ · · · ∧ Ωσ(2k−1)σ(2k)
2π 22k π k k!
σ∈S2k
descends to a closed 2k-form on X whose cohomology class e(X) does not
depend on the choice of the connection Ω. This is the Euler class and its
integral over X is the Euler characteristic (this is proved in [MT]).
All of this can be generalized to an arbitrary oriented, real vector bundle
E −→ X of fiber dimension 2k over a compact, oriented manifold of dimension
2k. Here one chooses a fiber metric (smoothly varying positive definite inner
products on the fibers of E) to get an oriented, orthonormal frame bundle
SO(2k) ,→ F+ (E) −→ X. The Euler class e(E) is then defined just as above
and one defines the Euler number χ(E) of the bundle to be its integral over X
(this is no longer the Euler characteristic of X, of course). One can then prove
an analogue of the Poincaré-Hopf Theorem that gives χ(E) as the intersection
number for a cross-section of E (see [MT]).
What we have seen is that the Donaldson invariant γ0 (M ) is analogous to
the Poincaré-Hopf version of an Euler number (although the vector bundle
is infinite-dimensional and the Poincaré-Hopf Theorem itself is valid only in
finite dimensions). If one were to take this analogy seriously it might suggest
the possibility of an integral representation of γ0 (M ) analogous to the Gauss-
Bonnet-Chern Theorem. Notice, however, that such an integral would be over
the base of the vector bundle which is the infinite-dimensional moduli space B̂.
Now, integrals over infinite-dimensional manifolds are notoriously difficult to
make rigorous mathematical sense of, but, fortunately, this does not bother the
 A.1. Donaldson Invariants and TQFT 359

physicists at all. As Nigel Hitchin has said, “This is such stuff as quantum field
theory is made of.” Indeed, it was Edward Witten [W2] who first produced a
(formal) integral representation of γ0 (M ), not directly, but as what is called
the partition function of a certain variant of supersymmetric quantum Yang-
Mills theory. Indeed, this quantum field theory also yielded formal integral
representations for all of the Donaldson invariants and eventually led to the
Seiberg-Witten invariants. Athough we are not so presumptuous as to attempt
any sort of exegesis of Witten’s work a quick tour of a few of the ideas is the
only way to see the emergence of Seiberg-Witten.
There are various approaches to the construction of a quantum field theory,
but the only one that will concern us here is the so-called Feynmann path
integral approach. Here one begins with a classical field theory of just the
sort we discussed in Chapter 2. Thus, one is given a collection ξ of classical
fields (gauge fields, i.e., connections, and matter fields, i.e., sections of vector
bundles) and an action S(ξ). The action has various symmetries (e.g., gauge
invariance, relativistic invariance, etc.) so that the physically significant object
of study is the moduli space F/S of fields modulo symmetrics. Real-valued
functions
O : F/S −→ R
on the moduli space are called observables. In this context, “quantization” is
viewed as the process of assigning expectation values ⟨ O ⟩ to
observables O and, according to the rules of the game, this is accomplished
by a weighted integral over the moduli space which, when the underlying
manifold is Riemannian (as opposed to semi-Riemannian), is of the form
∫
e−S(ξ)/e O([ξ]) [Dξ].
2
⟨O⟩ =
F/S

Here e is a “coupling constant” and [Dξ] represents a (generally nonexistent)

measure on the infinite-dimensional moduli space. Such integrals as a rule have
no precise mathematical definition, but physicists compute with them to great
effect nevertheless. The end result is a quantum field theory; it is called
a topological quantum field theory (TQFT) if for some distinguished
set of observables these expectation values are independent of the choice of
Riemannian metric on the underlying manifold. The integral
∫
e−S(ξ)/e [Dξ]
2
Z=
F/S

is called the partition function of the quantum field theory.

Witten [W2] constructed the first example of such a TQFT with the specific
intention of exhibiting the Donaldson invariants as expectation values of cer-
tain observables. The classical gauge theory from which the quantum field
theory was built is rather more complicated than any we have encountered
and has more structure than our examples, but very briefly it looked some-
thing like this: M as usual is a compact, simply connected, oriented, smooth
360 Appendix

4-manifold with b+ 2 (M ) > 1 and odd and we choose some generic metric g
on M . The field content consists of one gauge field (connection) ω and five
matter fields ϕ, λ, η, χ and ψ all of which are forms (of various degrees) with
values in the adjoint bundle ad P . But this gauge theory of Witten’s is super-
symmetric which means that each matter field is classified as either “bosonic”
or “fermionic” and there is an additional (super) symmetry operator that in-
terchanges bosons and fermions. We will make no attempt to explain what
this means, but will simply refer the curious to [W1] and record the types of
the matter fields:

Bosonic Fermionic
ϕ ∈ Λ (M, ad P )
0
η ∈ Λ0 (M, ad P )
λ ∈ Λ0 (M, ad P ) χ ∈ Λ1 (M, ad P )
ψ ∈ Λ2+ (M, ad P )

A 6-tuple ξ = (ω, ϕ, λ, η, χ, ψ) of such fields is called a field configura-

tion for the gauge theory. Now we need an action functional. This is called
the Donaldson-Witten Action. It is somewhat more intimidating than the
others we have seen and we will make no real use of it, but since everyone
should see Witten’s Lagrangian once in their lives, here it is,
∫ (
1 1 1
SDW (ξ) = trace − F ω ∧ ∗ F ω − F ω ∧ F ω + [ψ, ψ]ϕ
M 4 4 2
+ i dω χ ∧ ψ − 2i [χ, ∗ χ] λ + i ∗ (ϕ∆ω
0 λ)
)
+ χ ∧ ∗ dω η

Remarks: A few of the terms in the action are familiar. The first is a Yang-
Mills term. The second Witten calls a topological term since it is essentially
the Chern class. The rest are to be regarded as interaction terms. The path
Witten followed to arrive at the field content, the symmetries and the terms in
the action is described in detail in [W2], but it is not a path easily traversed
by mathematicians. Remarkably, Atiyah and Jeffrey [AJ] have shown that the
fields and the action all arise naturally from purely geometrical considerations
by formally applying to the infinite-dimensional vector bundle  ×Ĝ Λ2+ (M ,
ad P ) a formula for the Euler class proved (for finite-dimensional vector bun-
dles) by Mathai and Quillen [MQ].
SDW (ξ) is a perfectly well-defined mathematical object, but now we must
quantize and this means integrating over the entire moduli space of field con-
figurations. The partition function, for example, is formally written as the
path integral ∫
e−SDW (ξ)/e [Dξ].
2
ZDW =
 A.1. Donaldson Invariants and TQFT 361

Now, a path integral is really not an integral at all, but just a suggestive
notation for a certain limit (a limit that generally does not exist). Physicists
have developed elaborate techniques for dealing with such “integrals”, but
we will say nothing about this. Rather we will simply sketch, in very broad
terms, how Witten was led to identify ZDW with γ0 (M ) in the case in which
8k − 3(1 + b+ 2 (M )) = 0. The crucial observation is that ZDW , whatever it
means, is a function of the coupling constant e. At the classical level, e plays
no real role and can simply be regarded as a rescaling of the action, but in
the quantum theory its size will determine the computability (or not) of the
relevant physical quantities. This is because the usual procedure for dealing
with these quantities is to do perturbation calculations and this involves series
expansions in e. If e is “small” (i.e., in what is called the “weak coupling
limit”) such calculations are extraordinarily effective, but if e is “large” (“strong
coupling”) they fail completely.
Witten computed ZDW is the weak coupling limit by performing what the
physicists call the “semi-classical approximation” which he concluded, based
on all of the symmetries built into the action, must, in fact, be exact.
Remark: This is an infinite-dimensional analogue of the famous
Duistermaat-Heckman Theorem on exact stationary phase approximations
(see [BV]).
From this he argued that the path integral defining ZDW “localizes” to an
integral over the moduli space M of anti-self-dual connections on P and,
when dim M = 0, this integral over the 0-dimensional moduli space is just a
sum that can be identified with the Donaldson invariant γ0 (M ).
Remark: Intuitively, this is not unlike the Residue Theorem which localizes
a contour integral around a closed path to a sum of contributions from the
singularities of the integrand inside (although there are no “symmetry” con-
siderations here). Much closer mathematical analogues are the Equivariant
Localization Theorems of Berline and Vergne (see [BV]) which extract and
generalize the essential content of the Duistermaat-Heckman Theorem.
Witten also isolated observables in his TQFT whose expectation values
formally coincide with the integrals we used in our “naive” definition of
γd (M ) for d > 0. Very briefly, it goes like this: For any field configuration
ξ = (ω, ϕ, λ, η, χ, ψ) define W = trace( 14 ψ ∧ ψ + ϕF ω ). This is an ordinary
2-form on M . For each homology class x ∈ H2 (M ), thought of as an embedded
surface, define ∫
O(x) = W.
x

Each O(x) maps a field configuration to a real number. Again due to all of the
built-in symmetries, O(x) is actually defined on the moduli space and so is
an observable. Witten associates with each O(x) a closed 2-form α(x) on the
moduli space M whose cohomology class can be identified with µ(x)(µ is the
362 Appendix

Donaldson µ-map). Finally, if dim M = 2d, d ≥ 1, he shows that Od , defined

on H2 (M ) by Od (x) = (O(x))d , is a real-valued function on the moduli space
(i.e., an observable) whose expectation value, localized to M, is
∫
d
⟨Od (x)⟩ = α(x) ∧ · · · ∧ α(x)
M

and so coincides with our integral description of γd (M ).

All of this is quite formal, of course, and the end result is not really the
Donaldson invariants but only our naive description of some of them. Never-
theless, it is remarkable. Still more remarkable is the fact that this is just the
beginning of the story and that the best part is yet to come. To relate this part
of the story we must first say a few words about another feature of Witten’s
TQFT. This is a very subtle and deep type of “duality” that we cannot do
justice to here, but which has its origins in the fact that, unlike most quan-
tum fields theories, its partition function and expectation values are actually
independent of the coupling constant e. This is not a mathematical theorem,
of course, but a consequence of formal path integral manipulations based on
the myriad symmetries of the action. Ordinarily one would expect the weak
and strong coupling regimes to describe very different physical systems, but
in the case at hand they must be entirely equivalent, although their mathe-
matical descriptions would no doubt look quite different (since, for example,
perturbation calculations are possible in one, but not the other). This suggests
the possibility of an entirely different description of the Donaldson invariants
buried in the strong coupling regime. Witten was well aware of this in 1988
when [W2] appeared, but could do nothing about it because no one knew
how to compute anything in the strong coupling (nonperturbative) regime.
And so matters stood until 1994 when Seiberg and Witten developed entirely
new techniques for doing exact calculations in strong coupling for so-called
N = 2 supersymmetric Yang-Mills theories. This done, Witten dusted off his
old TQFT, applied the new techniques and uncovered the dual version of Don-
aldson theory. This is, of course, the Seiberg-Witten theory that we have been
leading up to all along. The details of the argument leading from Donaldson-
Witten to Seiberg-Witten are at the deepest levels of theoretical physics and,
alas, quite beyond the powers of your poor author whose only service can be
to refer the stout-hearted among his readers to [W3]. The end result, however,
was what Witten was convinced must be a “substitute” for Donaldson theory.
It contained new fields (not a single SU (2)-connection, but a U (1)-connection
and a spinor field), new equations (not the anti-self-dual equations, but the
Seiberg-Witten equations which are, from the point-of-view of partial differ-
ential equations, much simpler), a new moduli space and new invariants. But
the physical equivalence of the quantum field theories from which they arose
(“duality”) left no doubt in Witten’s mind that they must contain the same
topological information. The story of how this conjecture was sprung on the
 A.2. Clifford Algebra and Spinc -Structures 363

mathematical community and the pandemonium that ensued has been told
many times, but is best heard from someone who was there so for this as well
as a lovely introduction to what is to come here and a pleasant afternoon’s
entertainment we heartily recommend [Tau2]. We will now get on with the
business of describing the mathematical side of the new classical gauge theory.
When it is possible, with the background we have at our disposal, to provide
details we will do so; when it is not, we will try to provide a sense of what is
involved, what needs to be learned and where it can be learned.

A.2 Clifford Algebra and Spinc -Structures

Seiberg-Witten invariants are defined, just as the Donaldson invariants are,
from a moduli space of solutions to certain partial differential equations. They
are much simpler to deal with than those of Donaldson, but the price one
must pay is that just writing down the equations involves a rather substantial
investment of time in various algebraic preliminaries.
Much of the algebraic background we require is most conveniently phrased
in the language of Clifford algebras. We recall that any finite dimensional, real
vector space V with an inner product ⟨ , ⟩ has a Clifford algebra Cl(V ) which
can be described abstractly as the quotient of the tensor algebra J (V ) by the
2-sided ideal I(V ) generated by elements of the form v ⊗ v + ⟨ , ⟩1 with v ∈ V .
More concretely, if {e1 , . . . , en } is an orthonormal basis for V , then Cl(V ) is
the real associative algebra with unit 1 generated by {e1 , . . . , en } and subject
to the relations
ei ej + ej ei = −2⟨ei , ej ⟩1, i, j = 1, . . . , n. (A.2.1)
We intend to be even more concrete and construct an explicit matrix model
for the Clifford algebra Cl(4) = Cl(R4 ) of R4 with its usual positive definite
inner product. The procedure will be to identify R4 with a real linear subspace
of a matrix algebra, find an orthonormal basis for this copy of R4 satisfying
the defining conditions (A.2.1), where the product is matrix multiplication
and 1 is the identity matrix, and form the subalgebra it generates.
One can, of course, identify R4 with the algebra H of quaternions
q = q 1 + q 2 i + q 3 j + q 4 k, but we wish to embed this into the real,
associative algebra H2×2 of 2 × 2 quaternionic matrices:
{( ) }
q q
H2×2 = ll 12
: qij ∈ H, i, j = 1, 2
q21 q22

Specifically, we identify R4 with the real linear subspace of H2×2 consisting

of all elements of the form
( )
0 q
x= , q∈H (A.2.2)
−q̄ 0
364 Appendix

(this is, of course, not a subalgebra of H2×2 ). Notice that det x = ∥q∥2 so,
defining a norm on the set of x given by (A.2.2) by

∥x∥2 = det x (A.2.3)

an inner product by polarization (⟨x, y⟩ = 14 (∥x + y∥2 − ∥x − y∥2 ) we find that

the subspace of H2×2 consisting of all x of the form (A.2.2) is isomorphic to
R4 as an inner product space. One easily checks that {e1 , e2 , e3 , e4 } given by
( ) ( ) ( ) ( )
0 1 0 i 0 j 0 k
e1 = , e2 = , e3 = , e4 = (A.2.4)
−1 0 i 0 j 0 k 0

is an orthonormal basis and, moreover, satisfies

ei ej + ej ei = −2⟨ei , ej ⟩1, i, j = 1, 2, 3, 4, (A.2.5)

where we use 1 generically for the identity matrix of any size (2 ×2 in this
case). Note that it follows from (A.2.5) that
xy + yx = −2⟨x, y⟩, x, y ∈ R4 . (A.2.6)
The real subalgebra of H 2×2
generated by {e1 , e2 , e3 , e4 } is the real
Clifford algebra of R4 and is denoted Cl(4). Writing out products of basis
vectors and using (A.2.5) to eliminate linear dependencies gives the following
basis for Cl(4): ( )
1 0
e0 = =1
0 1
( ) ( ) ( ) ( )
0 1 0 i 0 j 0 k
e1 = e2 = e3 = e4 =
−1 0 i 0 j 0 k 0
( ) ( ) ( )
i 0 j 0 k 0
e1 e2 = e1 e3 = e1 e4 =
0 −i 0 −j 0 −k
( ) ( ) ( )
k 0 −j 0 i 0
e2 e3 = e2 e4 = e3 e4 = (A.2.7)
0 k 0 −j 0 i
( ) ( )
0 k 0 −j
e1 e2 e3 = e1 e2 e4 =
−k 0 j 0
( ) ( )
0 i 0 −1
e1 e3 e4 = e2 e3 e4 =
−i 0 −1 0
( )
−1 0
e1 e2 e3 e4 =
0 l
 A.2. Clifford Algebra and Spinc -Structures 365

Thus, dim Cl(4) = 16. Since H2×2 itself has real dimension 16 we conclude
that, in fact,
Cl(4) = H2×2 . (A.2.8)
Notice that the basis (A.2.7) gives Cl(4) a natural Z2 -grading

Cl(4) = Cl0 (4) ⊕ Cl1 (4), (A.2.9)

where Cl0 (4) is spanned by e0 , e1 e2 , e1 e3 , e1 e4 , e2 e3 , e2 e4 , e3 e4 and e1 e2 e3 e4

and Cl1 (4) is spanned by e1 , e2 , e3 , e4 , e1 e2 e3 , e1 e2 e4 , e1 e3 e4 and e2 e3 e4 .
The elements of Cl0 (4) are said to be even, while those of Cl1 (4) are odd.
Regarding Z2 as {0, 1} with addition modulo 2,

(Cli (4)) (Clj (4)) ⊆ Cli+j (4) (A.2.10)

for i, j = 0, 1, so Cl(4) is a Z2 -graded algebra, i.e., a superalgebra. From (A.2.7)

it is clear that the decomposition (A.2.9) corresponds simply to

( ) ( ) ( )
q11 q12 q11 0 0 q12
= + .
q21 q22 0 q22 q21 0

Lemma A.2.1 The center Z(Cl(4)) of Cl(4) is Span{e0 } ∼

= R.

Proof: Since e0 = 1 it commutes with everything in Cl(4) so

Span{e0 } ⊆ Z(Cl(4)) is clear. To complete the proof it will suffice to show
that every

eI = ei1 · · · eik , 1 ≤ k ≤ 4, 1 ≤ i1 < · · · < ik ≤ 4

fails to commute with something in Cl(4). For k = 1 this is clear since ei ej =

−ej ei for i ̸= j. For k = 4, eI = e1 e2 e3 e4 so e1 eI = (e1 e1 )e2 e3 e4 = −e2 e3 e4 ,
whereas eI e1 = (e1 e2 e3 e4 )e1 = (−1)3 (e1 e1 )e2 e3 e4 = e2 e3 e4 . Now suppose
1 < k < 4. Then eI ei1 = (−1)k−1 ei1 eI and, if el is not among ei1 , . . . , eik , eI el =
(−1)k el eI . Thus, eI cannot commute with both ei1 and el . 
Lemma A.2.2 If x ∈ R4 ⊆ Cl(4) and ∥x∥ = 1, then x is a unit in Cl(4)
(i.e., is invertible) and x−1 = −x.

Proof: ⟨x, x⟩ = 1 and xx + xx = −2⟨x, x⟩1 imply xx = −1. 

We denote by Cl× (4) the multiplicative group of units in Cl(4) and by
Pin(4) the subgroup of Cl× (4) generated by all of the x ∈ R4 with ∥x∥ = 1
(see Lemma A.2.2). Now, an x of the form (A.2.2) has ∥x∥ = 1 if and only if
q ∈ Sp(1) (the Lie group of unit quaternions) and the set of all such is closed
under inversion (x−1 = −x). Thus, Pin(4) is just the set of all products of
such elements. The even elements of Pin(4) are just its diagonal elements and
366 Appendix

they form a subgroup denoted

Spin(4) = Pin(4) ∩ Cl0 (4)

{( ) }
u1 0
= : u1 , u2 ∈ Sp(1) (A.2.11)
0 u2
∼
= Sp(1) × Sp(1).

The topology and differentiable structure Spin(4) inherits from

H2×2 ∼
= H4 ∼
= R16 are the product structures from Sp(1) (which is diffeo-
morphic to S 3 ) so Spin(4) is a compact, simply connected Lie group. Since
the Lie algebra of Sp(1) can be identified with the pure imaginary quaternions
Im H, the Lie algebra of Spin(4) can be identified with
{( ) }
Q1 0
spin(4) = : Q1 , Q2 ∈ Im H . (A.2.12)
0 Q2

The significance of Spin(4) lies in the following theorem.

Theorem A.2.3 Spin(4) is a simply connected double cover of SO(4).

Proof: Since Spin(4) is a simply connected Lie group we need only show
that it is a double cover of SO(4). For this we consider the adjoint action of
Cl× (4) on Cl(4), i.e., for each u ∈ Cl× (4) we define a map
adu : Cl(4) −→ Cl(4)
by
adu (p) = upu−1 (A.2.13)

for each p ∈ Cl(4). This is clearly an algebra isomorphism that preserves the
grading (A.2.9). Note that if x ∈ R4 ⊆ Cl(4) has ∥x∥ = 1, then, for every
v ∈ R4 ⊆ Cl(4),
adx (v) = xvx−1 = xv(−x) = −xvx
so the identity vx + xv = −2⟨v, x⟩1 implies

xvx + xxv = −2⟨v, x⟩ x

xvx − v = −2⟨v, x⟩ x
xvx = v − 2⟨v, x⟩ x
xvx = (v − ⟨v, x⟩ x) − ⟨v, x} x.

Now, v − ⟨v, x⟩x is the projection of v into the hyperplane x⊥ orthogonal to

x so xvx is the reflection of v through x⊥ , written Reflx⊥ (v). Thus,

adx (v) = −Reflx⊥ (v). (A.2.14)

 A.2. Clifford Algebra and Spinc -Structures 367

In particular,
adx : R4 −→ R4 (x ∈ R4 , ∥x∥ = 1).
But any element of Pin(4) is a product of elements x ∈ R4 with ∥x∥ = 1 so

adu : R4 −→ R4 (u ∈ Pin(4)). (A.2.15)

Since any product of reflections is an orthogonal transformation, adu is an

orthogonal transformation for each u ∈ Pin(4). Since an element of Spin(4) is
a product of an even number of x ∈ R4 with ∥x∥ = 1 and any product of an
even number of reflections is a rotation we have

adu ∈ SO(R4 ) (u ∈ Spin(4)). (A.2.16)

Thus, we have a map, called the spinor map,

Spin : Spin(4) −→ SO(R4 ) ∼

= SO(4) (A.2.17)

Spin(u) = adu .
Since any reflection can clearly be written as −adx for some x ∈ R4
with ∥x∥ = 1 and since any rotation can be written as a product of an
even number of reflections, the spinor map is a surjective group homomor-
phism. Finally, to see that ker(Spin) = Z2 = {±1}, so that it is precisely
two-to-one, note that adu is the identity in SO(4) if and only if uxu−1 = x
for each x ∈ R4 . But then u must commute with everything in Cl(4), i.e.,
u ∈ Z(Cl(4)). By Lemma A.2.1, u = ae0 = a1 for some a ∈ R. By (A.2.11),
a2 = 1 and u = ±1. 
Remark: Globalizing these constructions leads to the notion of a “spin
structure” on a manifold. In the context of spacetime (as opposed to
Riemannian) manifolds this is just the “spinor structure” we introduced and
studied in Sections 3.5 and 6.5 so we will just briefly describe the defini-
tion and then explain why we need the more general concept of a “spinc
structure”. We let B denote a compact, oriented, smooth 4-manifold with a
πSO
Riemannian metric g . Let SO(4) ,→ FSO (B) −→ B denote the correspond-
ing oriented, orthonormal frame bundle. A spin structure S consists of a
principal Spin(4)-bundle
π
Spin(4) ,→ S(B) −→
S
B
over B and a smooth map
λ : S(B) −→ FSO (B)
satisfying
πSO ◦ λ = πS (A.2.18)
and
λ(p · u) = λ(p) · Spin(u) (A.2.19)
368 Appendix

for each p ∈ S(B) and each u ∈ Spin(4).

The fibers of FSO (B) are copies of SO(4) so (A.2.18) says that we have a copy
of Spin(4) “above” each of these and (A.2.19) says that the map λ of S(B)
onto FSO (B) is essentially the spinor map at each point of B. Now, unlike
the frame bundle FSO (B), which exists for any manifold of the type we have
described, there is an obstruction to the existence to a spin structure. The
arguments of Section 6.5 carry over verbatim to show that B admits a spin
structure if and only if the 2nd Stiefel-Whitney class w2 (B) ∈ Ȟ 2 (B; Z2 ) is
trivial. Unfortunately, many interesting 4-manifolds (e.g., CP2 ) do not satisfy
this condition and without a spin structure one cannot define “spinor fields”
in the usual sense. Since spinor fields are crucial to Seiberg-Witten theory and
since one would like this theory to apply to as many 4-manifolds as possible
we seek a generalized notion of both “spin structure” and “spinor field”. As it
happens, there is a very natural generalization obtained by complexifying our
previous algebraic considerations.
To define complex analogues of the algebraic objects we have introduced
we will embed Cl(4) into a complex algebra of matrices and form the complex
subalgebra it generates. The basic tool we use is the usual matrix model of
the quaternions. Specifically, we consider the map γ : H −→ C2×2 from the
quaternions to the 2 × 2 complex matrices given by
( )
α β
γ(q) = γ(α + βj) = (A.2.20)
−β̄ ᾱ
where we have written

q = q 1 + q 2 i + q 3 j + q 4 k = (q 1 + q 2 i) + (q 3 + q 4 i)j = α + βj.

One easily verifies that γ is real linear, injective, preserves products, carries q̄
⊤
to γ(q) and satisfies det(γ(q)) = ∥q∥2 so ( that we) can identify H with the set
α β
of all 2 × 2 complex matrices of the form −β̄ ᾱ . More specifically, if we let
( ) ( )
1 0 i 0
γ(1) = =1 γ(i) = =I
0 1 0 −i
( ) ( ) (A.2.21)
0 1 0 i
γ(j) = =J γ(k) = =K
−1 0 i 0
 A.2. Clifford Algebra and Spinc -Structures 369

then we can identify q = q 1 + q 2 i + q 3 j + q 4 k with

q = q 1 1 + q 2 I + q 3 J + q 4 K. (A.2.22)

Now we identify Cl(4) = H2×2 with a subset of C4×4 . Define

Γ : H2×2 −→ C4×4 by
( ) ( )
q11 q12 γ(q11 ) γ(q12 )
Γ = , (A.2.23)
q21 q22 γ(q21 ) γ(q22 )

where each γ(qij ) is a 2 × 2 block in the matrix on the right-hand side. This
map Γ is also real linear, injective and preserves products so we can identify
the real algebra Cl(4) with its image.

Cl(4) = Γ(H2×2 )

The restriction of Γ to R4 ⊆ Cl(4) is

( ) ( )
0 q 0 γ(q)
x= −→ Γ(x) = ⊤ . (A.2.24)
−q̄ 0 −γ(q) 0
Since det Γ(x) = det x = ∥x∥2 = ∥q∥2 we can define an inner product via
polarization on this copy of R4 from ∥Γ(x)∥2 = det Γ(x) and then Γ | R4
becomes an isometry. We now fully identify R4 with this copy and obtain the
basis
( ) ( )
0 1 0 I
E1 = γ(e1 ) = E2 = γ(e2 ) =
−1 0 I 0
( ) ( ) (A.2.25)
0 J 0 K
E3 = γ(e3 ) = E4 = γ(e4 ) =
J 0 K 0

satisfying

Ei Ej + Ej Ei = −2⟨Ei , Ej ⟩1, i, j = 1, 2, 3, 4. (A.2.26)

In this context, Cl(4) is the real subalgebra of C4×4 generated by {E1 , E2 , E3 ,
E4 } and a basis is as in (A.2.7), but with everything capitalized (and 1 changed
to 1). Under γ, Sp(1) is mapped to SU (2) so, in our new model of Cl(4) we
have the identifications
{( ) }
U1 0
Spin(4) = : U1 , U2 ∈ SU (2) ∼= SU (2) × SU (2) (A.2.27)
0 U2
370 Appendix

and {( ) }
A1 0
spin(4) = : A1 , A2 ∈ su(2) (A.2.28)
0 A2
corresponding to (A.2.11) and (A.2.12).
Now we regard C4×4 as a complex algebra and define the complexified
Clifford algebra Cl(4) ⊗ C to be the complex subalgebra generated by
{E1 , E2 , E3 , E4 }, i.e., by Cl(4). A basis over C is given by (A.2.7), with all of
the ei capitalized. Since C4×4 also has dimension 16 over C we conclude that

Cl(4) ⊗ C = C4×4 . (A.2.29)

Now let
SC = C4
be the complex vector space C4 with its usual Hermitian inner product
(⟨z, w⟩ = z̄ 1 w1 + z̄ 2 w2 + z̄ 3 w3 + z̄ 4 w4 )) and identify Cl(4) ⊗ C with the
vector space EndC (SC ) of complex linear transformations of SC to itself:

Cl(4) ⊗ C = EndC (SC ) (A.2.30)

Thus, the elements of Cl(4) ⊗ C (and therefore also Cl(4), R4 and Spin(4))
act as endomorphisms of SC . This action is called Clifford multiplication
and will be written with a dot ·. In particular, we have a representation of the
real Clifford algebra by endomorphisms of SC :

Cl(4) −→ EndC (SC )

(representations of algebras are by endomorphisms rather than isomorphisms

since not all elements of an algebra are units). This representation of Cl(4)
is easily seen to be irreducible by writing out the real linear combinations of
the basis E0 , . . . , E1 E2 E3 E4 for Cl(4) ⊆ Cl(4) ⊗ C. Restricting the Clifford
action further to Spin(4) ⊆ Cl(4) ⊗ C gives a group respresentation of Spin(4)
on SC :
∆C : Spin(4) −→ AutC (SC )
(by automorphisms now since the elements of Spin(4) are all units). This is
called the complex spin representation and, as we shall now see, is not
irreducible. Indeed, if we write

SC ∼ +
= SC −
⊕ SC
     
z1 z1 0
 2  2  
z  z   0 
  =   +  , (A.2.31)
z 3   0  z 3 
     
z4 0 z4
 A.2. Clifford Algebra and Spinc -Structures 371

then Clifford multiplication by elements of Cl0 (4), because they are block
+ −
diagonal, preserves SC and SC , whereas Clifford multiplication by elements
+ −
of Cl1 (4), because they are block anti-diagonal, interchanges SC and SC . In
particular, ∆C resolves into a direct sum
−
C ⊕ ∆C
∆C = ∆+
where
∆± ±
C : Spin(4) −→ SU (SC )
−
(see (A.2.27) for the “SU”). ∆+ C and ∆C are inequivalent, irreducible repre-
sentations of Spin(4). Notice also that Clifford multiplication by the elements
of R4 , which are odd, interchanges SC +
and SC−
(this will be crucial when we
define the “Dirac operator” shortly).
Recall that Spin(4) is the set of all even elements in the subgroup of mul-
tiplicative units in the Clifford algebra Cl(4) generated by the unit sphere
in R4 ⊆ Cl(4). For the complex analogue we add to the generators the unit
circle in C. More precisely, we identify U (1) with the subset
U (1) = {eθ i 1 : θ ∈ R}
of Cl(4) ⊗ C (often dropping the “ 1” and thinking of eθ i as an element of
Cl(4) ⊗ C). Then
Spinc (4)
is defined to be the subgroup of the group of multiplicative units in Cl(4) ⊗ C
generated by Spin(4) and U (1). Notice that the elements of Spinc (4) are nec-
essarily even, i.e., in Cl0 (4) ⊗ C. Since U (1) is in the center of Cl(4) ⊗ C we
have
Spinc (4) = {eθ i u : θ ∈ R, u ∈ Spin(4)} (A.2.32)
{ ( ) }
U1 0
= eθ i : θ ∈ R, U1 , U2 ∈ SU (2) .
0 U2

Note that U1 , U2 ∈ SU (2) implies eθ i U1 , eθ i U2 ∈ U (2) and

det(eθ i U1 ) = det(eθ i U2 ) = e2θ i . (A.2.33)
Since every element of U (2) can be written as eθ i U, U ∈ SU (2) (uniquely up
to a simultaneous change of sign for both eθ i and U ) we have
{( ) }
U+ 0
c
Spin (4) = : U± ∈ U (2), det U+ = det U− . (A.2.34)
0 U−

There is yet another useful way of looking at Spinc (4). The mapping

Spin(4) × U (1) −→ Spinc (4)

(u, eθ i 1) −→ eθ i u
372 Appendix

is a surjective homomorphism. Its kernel is the set of (α, α−1 ), where

α ∈ Spin(4). But Spin(4) intersects the scalars only in ±1 so this kernel
is Z2 = ±(1, 1). Thus,

Spinc (4) ∼
= Spin(4) × U (1)/Z2 . (A.2.35)

Finally, notice that, from Lemma A.2.1 and (A.2.32) it follows that the center
of Spinc (4) is
Z (Spinc (4)) = U (1). (A.2.36)
Globalizing all of this to 4-manifolds will require a few mappings which we
now introduce. First define
δ : Spinc (4) −→ U (1)
as follows: For
( ) ( )
U+ 0 eθ i U1 0
ξ= = ∈ Spinc (4),
0 U− 0 eθ i U2

δ(ξ) = det U+ = det U− = e2θ i . (A.2.37)

Then δ is a surjective homomorphism with kernel Spin(4). Next define

π : Spinc (4) −→ SO(4)

as follows: The adjoint action of Spin(4) on R4 extends to an adjoint
action of Spinc (4) on R4 . Indeed, if ξ = eθ i u ∈ Spinc (4), then, for each
x ∈ R4 , adξ (x) = ξxξ −1 = uxu−1 = adu (x) so, on R4 ,

adξ = adu = Spin(u) ∈ SO (R4 ) ∼

= SO(4)

and we may take

π(ξ) = adξ = adu = Spin(u). (A.2.38)
Finally, define
Spinc : Spinc (4) −→ SO(4) × U (1)
by
( ) ( )
Spinc (ξ) = Spinc eθ i u = (π (ξ), δ (ξ)) = Spin(u), e2θ i . (A.2.39)

Then Spinc is a surjective homomorphism whose kernel is easily seen to be

Z2 = ±1. It follows that Spinc (4) is a double cover of SO(4) × U (1). Thus,
the Lie algebra spinc (4) is so(4) × u(1) ∼
= spin(4) × u(1) and can be identified
with the subset
{( ) ( ) }
A1 0 1 0
c
spin (4) = + ti : t ∈ R, A1 , A2 ∈ su(2) (A.2.40)
0 A2 0 1
 A.2. Clifford Algebra and Spinc -Structures 373

of Cl(4) ⊗ C (in particular, spinc (4) also acts by Clifford multiplication on

±
SC , preserving both SC ).
Now, the identification Cl(4) ⊗ C = EndC (SC ) and the fact that the ele-
ments of Spinc (4) are all units implies that the complex spin representation
∆C : Spin(4) −→ AutC (SC ) extends to a representation
ˆ C : Spinc (4) −→ AutC (SC ).
∆
ˆ C also splits into
Since the elements of Spinc (4) are block diagonal, ∆
ˆ−
ˆ+ ⊕ ∆
ˆC = ∆
∆ C C
where
ˆ ± : Spinc (4) −→ U (S ± )
∆ C C (A.2.41)

(see (A.2.34)).
One of the Seiberg-Witten equations relates the self-dual part of the curva-
ture of a U (1)-connection to a certain trace free endomorphism of a positive
spinor. The last of our algebraic preliminaries describes the relationship be-
tween 2-forms and endomorphisms. ∧ We note that there is a natural linear
isomorphism from the space 2 (R4 , C) of (complex-valued) 2-forms on R4
into Cl(4) ⊗ C. Indeed, if {e1 , e2 , e3 , e4 } is the standard basis for R4 and
{e1 , e2 , e3 , e4 } is its dual, then we define
∧2
ρ: (R4 , C) −→ Cl0 (4) ⊗ C

by
 
∑ ∑
ρ(η) = ρ  ηij ei ∧ ej  = ηij Ei Ej =
i<j i<j
 
(η12 + η34 )I
 +(η13 − η24 )J 0  (A.2.42)
 
 +(η14 + η23 )K 
 .
 (−η12 + η34 )I 
 
 0 +(−η13 − η24 )J 
+(−η14 + η23 )K

Notice that, although ρ is clearly a linear isomorphism, it is not multiplicative,

e.g., e1 ∧ e1 = 0, but E1 E∧1 = −l. There is, of course, an analogous map in any
(R4 , C ) is real-valued (respectively, ImC-valued),
2
rank. Notice that if η ∈
then ρ(η) is skew-Hermitian (respectively, Hermitian). For example, if η is
374 Appendix

real-valued,
T ∑ T ∑
ρ(η) = η̄ij Ei Ej = ηij ĒjT ĒiT
i<j i<j
∑ ∑
= ηij (−Ej )(−Ei ) = ηij Ej Ei
i<j i<j
∑
= ηij (−Ei Ej )
i<j
= −ρ(η).

Note also that, in (A.2.42), {ea } can be replaced by any oriented, orthonormal
basis provided {Ea } is replaced by its image under Γ (see (A.2.25)).
±
Now, being even (i.e., block diagonal) any ρ(η) preserves the subspaces SC
±
of SC and so we obtain endomorphisms of SC by setting
±
ρ± (η) = ρ(η) | SC . (A.2.43)
+
For example, suppressing the two zero entries in SC (see (A.2.31)),
ρ+ (η) = (η12 + η34 )I + (η13 + η42 )J + (η14 + η23 )K. (A.2.44)
Thus, we have two maps
∧2
ρ± : (R4 , C) −→ EndC (SC
±
). (A.2.45)
∧2 4 ∧2 ∧2
Now let (R , C) = + (R4 , C) ⊕ − (R4 , C) be the decomposition of
∧2 4
(R , C) into self-dual and anti-self-dual 2-forms (relative to the Hodge star
∗ for the usual orientation and inner product on R4 ). We show that ρ± carries
∧2
± (R , C) isomorphically onto the space End0 (SC ) of trace free (complex)
4 ±
±
endomorphisms of SC .
∧2
Lemma A.2.4 ρ± | ± (R4 , C) is a complex linear isomorphism onto
±
End0 (SC ).
∧2 ∧2
Proof: We give the argument for ρ+ | + (R4 , C). The ρ− | − (R4 , C) case
is analogous. A simple computation from (A.2.42) gives

( )
I 0
ρ(e ∧ e + e ∧ e ) = 2
1 2 3 4
0 0
( )
J 0
ρ(e ∧ e + e ∧ e ) = 2
1 3 4 2
(A.2.46)
0 0
( )
K 0
ρ(e ∧ e + e ∧ e ) = 2
1 4 2 3
.
0 0
 A.2. Clifford Algebra and Spinc -Structures 375

Since {e1 ∧e2 +e3 ∧e4 , e1 ∧e3 +e4 ∧e2 , e1 ∧e4 +e2 ∧e3 } spans the set of self-
∧2
dual 2-forms on R4 , it is clear that ρ+ | + (R4 , C ) is a linear, injective map
+
to End(SC ). Because I, J and K are trace free, so is everything in the image
∧
of ρ | + (R4 , C). Furthermore, one can show that every 2×2 complex, trace
+ 2
∧2
free matrix is a complex linear combination of I, J and K so ρ+ | + (R4 , C )
maps onto End0 (SC +
). 
It ∧ follows, in ∧2 particular, from Lemma A.2.4 that the map
ρ+ | + (R4 , C) : + (R4 , C) −→ End0 (SC
2 +
) has an inverse that we will simply
denote ∧2
+
σ + : End0 (SC ) −→ (R4 , C). (A.2.47)
+

One can compute this inverse explicitly, but we will content ourselves with
describing its action on the particular type of trace free endomorphism that
arises in the Seiberg-Witten equations. For this we consider an element
( )
ψ1
ψ=
ψ2
+ +
of SC (temporarily suppress the two zero components in SC ). Define an en-
+
domorphism of SC by the matrix
( ) ( )
∗ ψl |ψ 1 |2 ψ 1 ψ̄ 2
ψ⊗ψ = 1 2
(ψ̄ ψ̄ ) = . (A.2.48)
ψ2 ψ̄ 1 ψ 2 |ψ 2 |2

The trace free part of this endomorphism is

1
(ψ ⊗ ψ ∗ )0 = ψ ⊗ ψ ∗ − tr(ψ ⊗ ψ ∗ ) l
2
( ) (A.2.49)
2 (|ψ | − |ψ | )
1 1 2 2 2
ψ 1 ψ̄ 2
= .
2 (|ψ | − |ψ | )
1
ψ̄ 1 ψ 2 2 2 1 2

One can verify directly that the image of (ψ ⊗ ψ ∗ )0 ∈ End0 (SC +

) under
σ + is
1 [
σ + ((ψ ⊗ ψ ∗ )0 ) = − i (|ψ 1 |2 − |ψ 2 |2 )(e1 ∧ e2 + e3 ∧ e4 )
4
−2 Im(ψ 1 ψ̄ 2 )(e1 ∧ e3 + e4 ∧ e2 )
]
−2 Re(ψ 1 ψ̄ 2 )(e1 ∧ e4 + e2 ∧ e3 )
1[ (A.2.50)
= − (ψ ∗ I ψ)(e1 ∧ e2 + e3 ∧ e4 )
4
+ (ψ ∗ Jψ)(e1 ∧ e3 + e4 ∧ e2 )
]
+ (ψ ∗ Kψ)(e1 ∧ e4 + e2 ∧ e3 )

(apply ρ+ to the right-hand side to get (ψ ⊗ ψ ∗ )0 ).

376 Appendix

Finally, we have the algebraic spadework completed and we can proceed

to the problem of globalizing all of these notions to manifolds and bundles.
Henceforth, B will denote a compact, connected, simply connected, oriented
smooth 4-manifold (simple connectivity is not essential here, but will stream-
line some of what we have to say). For any choice of Riemannian metric
g on B,
πSO
SO(4) ,→ FSO (B) −→ B
will denote the corresponding oriented, orthonormal frame bundle of B.
Should B happen to admit a spin structure (see the Remark following the
proof of Theorem A.2.3), then the representations ∆± ±
C : Spin(4) −→ SU (SC )
±
give associated spinor bundles S(B) ×△± SC whose sections are spinor fields.
C
Because such a spin structure need not exist (and because spinor fields are
essential ingredients in Seiberg-Witten theory), we formulate a complex ana-
logue of a spin structure, which always exists.
A spinc structure L on B consists of a principal Spinc (4)-bundle
π c
Spinc (4) ,→ S c (B) −→
S
B (A.2.51)

over B and a smooth map

Λ : S c (B) −→ FSO (B) (A.2.52)

satisfying
πSO ◦ Λ = πS c (A.2.53)

and

Λ(p · ξ) = Λ(p) · π(ξ) (A.2.54)

for each p ∈ S c (B) and each ξ ∈ Spinc (4). Here π : Spinc (4) −→ SO(4) is
defined by (A.2.38).

It is known that, for any B of the type we have described and any choice
of the Riemannian metric g , spinc structures exist (see [LM]). In terms of
transition functions this means that for any trivializing cover {Uα } for the
frame bundle with transition functions gαβ : Uα ∩ Uβ −→ SO(4), there exist
lifts g̃αβ : Uα ∩ Uβ −→ Spinc (4)
 A.2. Clifford Algebra and Spinc -Structures 377

Spinc(4)

αβ
π

Uα ∩ Uβ SO (4)
αβ

satisfying the cocycle condition.

Given a spinc structure L on B each of the representations ∆ ˆ C : Spinc (4)
± ±
−→ AutC (SC ), ∆C : Spin (4) −→ U (SC ) and δ : Spin (4) −→ U (1) give rise
ˆ c c

to vector bundles associated to (A.2.51) which we will write as follows.

S(L) = S c (B) ×∆
ˆ C SC

±
S ± (L) = S c (B) ×∆
ˆ ± SC
C
L(L) = S (B) ×δ C
c

S(L) is called the spinor bundle of L, S ± (L) are the positive and
negative spinor bundles of L and L(L) is the determinant line bundle
of L. The algebraic decomposition (A.2.31) persists in the bundle setting to
give a Whitney sum decomposition

S(L) = S + (L) ⊕ S − (L).

Remark: The Whitney sum of two vector bundles πi : Ei −→ X, i = 1, 2, is

just the natural vector bundle analogue of the direct sum of two vector spaces.
Its fibers are, indeed, the direct sums π1−1 (x) ⊕ π2−1 (x) of the fibers of E1 and
E2 and, if {Uα }α∈A is a trivializing cover for both E1 and E2 , its transition
functions are the direct sums of the transition functions for E1 and E2 .
We will also need the principal U (1)-bundle
π L0
U (1) ,→ L0 (L) −→ B

associated to L(L).
Remark: This bundle can be described as follows. Choose a Hermitian fiber
metric (smoothly varying Hermitian inner products on the fibers) on the com-
plex line bundle L(L). Then L0 (L) is the unit circle bundle in L(L) i.e., it
is the corresponding oriented, orthonormal frame bundle. One can retrieve
L(L) from L0 (L) as the vector bundle associated to L0 (L) by complex multi-
plication. One can show that w2 (B) = c1 (L0 (L)) mod 2, where w2 (B) is the
second Stiefel-Whitney class of B, and that, conversely, given a U (1)-bundle
L0 over B with w2 (B) = c1 (L0 ) mod 2 there is a spinc structure L on B with
378 Appendix

L0 (L) = L0 . More details on this and much of what follows are available in
[M1]
We will also require a bundle associated to the frame bundle that does not
require a spin or spinc structure. Notice that Spin(4), being contained in
Cl× (4), acts on Cl(4) by conjugation and, since (−u)p(−u)−1 = upu−1 ,
this gives an action of SO(4) = Spin(4)/Z2 on Cl(4) which clearly preserves
products (u(pq)u−1 = (upu−1 )(uqu−1 )). The Clifford bundle Cl(B) is the
bundle with typical fiber Cl(4) over B associated to the frame bundle by this
action.
Cl(B) = FSO (B)×SO(4) Cl(4)
Similarly, one has a complexified Clifford bundle
Cl(B) ⊗ C = FSO (B)×SO(4) (Cl(4) ⊗ C).
These decompose into even and odd summands, e.g., Cl(B) ∼ = Cl0 (B) ⊕
Cl1 (B). Moreover, pointwise multiplication provides the spaces of sections of
these bundles with algebra structures and such sections act on sections of the
spinor bundle by pointwise Clifford multiplication.
Now, Spinc (4) double covers SO(4) × U (1) by the map Spinc so S c (B)
.
double covers the fiber product FSO (B) × L0 (L) (this is just that part of
the product bundle SO(4) × U (1) ,→ FSO (B) × L0 (L) ,→ B × B above the
diagonal in B × B with this diagonal identified with B in the obvious way).
We will use the symbol Spinc also for this double cover.
Spinc (4) ,→ S c (B) −→ B



 Spinc (A.2.55)
y
.
SO(4) × U (1) ,→ FSO (B) × L0 (L) −→ B

Locally, this map is just

( )
(b, ξ) −→ b, Spinc (ξ)) = (b, (π(ξ), δ(ξ))
( ( )) (A.2.56)
= b, Spin(u), e2θi

where ξ = eθi u ∈ Spinc (4).

Next observe that the algebraic isomorphism σ + in (A.2.47) globalizes
as follows. The map ρ of (A.2.42) is independent of the choice of ori-
ented, orthonormal basis for R4 so, using local oriented, orthonormal frame
fields on B (i.e., sections of FSO (B)) it gives a map from 2-forms on B
to sections of Cl(B) ⊗ C. These sections in turn act on the spinor bundle
S(L) ∼= S + (L) ⊕ S − (L) of any spinc structure. Since the action is fiberwise,
the image of a self-dual 2-form preserves S + (L) and is, at each point, a trace
free endomorphism of SC+ . Thus, a self-dual 2-form on B gives rise to a sec-
tion of the trace free endomorphism bundle End0 (S + (L)) of S + (L) and we
 A.3. Seiberg-Witten Equations 379

have an isomorphism (also denoted σ + ) from the sections Γ(End0 (S + (L))) of

End0 (S + (L)) to the self-dual 2-forms on B:
( ( )) ∧2
σ + : Γ End0 S + (L) −→ (B, C) (A.2.57)
+

In particular, if ψ is a positive spinor field on B (i.e., a section of S + (L)),

then (ψ ⊗ ψ ∗ )0 ∈ Γ(End0 (S + (L))) is given pointwise by (A.2.49) and σ + ((ψ ⊗
∧2
ψ∗)0 ) ∈ + (B, C) is given pointwise relative to a local oriented, orthonormal
frame field on B by (A.2.50).

A.3 Seiberg-Witten Equations

At this point we can begin to give some idea of where all of this is going. To
write the Seiberg-Witten equations for B one chooses a Riemannian metric g
and a spinc structure L. The field content of the theory consists of a U (1)-
connection A on L0 (L) (the gauge field) and a positive spinor field ψ. They
are related by two equations, one of which requires that the self-dual 2-form
σ + ((ψ⊗ψ ∗ )0 ) should coincide with the self-dual part of the curvature of A i.e.,
∗
F+A = σ ((ψ ⊗ ψ )0 ). The other equation still requires a bit of preparation,
+

however.
Recall (Section 3.3) that the frame bundle SO(4) ,→ FSO (B) −→ B has a
distinguished (Levi-Civita) connection which we will denote ω LC . This can
be characterized locally as follows. If {e1 , e2 , e3 , e4 } is a local oriented, or-
thonormal frame field on B (i.e., a section of FSO (B)) with dual 1-form field
{e1 , e2 , e3 , e4 }, then ω LC is represented by a skew-symmetric matrix (ω ij )
of R-valued 1-forms satisfying dei = −ω ij ∧ ej , i = 1, 2, 3, 4. Now no-
tice that if B had a spin structure Spin(4) ,→ S(B) −→ B, then the map
λ : S(B) −→ FSO (B) is a double cover that respects the group actions so
that any connection on FSO (B), e.g., ω LC , automatically lifts to a connec-
tion on S(B) (think of the connection as a distribution of horizontal spaces).
However, if B has only a spinc structure Spinc (4) ,→ S c (B) −→ B, then the
map Λ of (A.2.52) is not a finite covering so ω LC alone will not determine a
connection on S c (B). However, Spinc : S c (B) −→ FSO (B)×L ˙ 0 (L) is a dou-
0
ble cover (A.2.55) and if A is any connection on L (L), then A and ω LC
together determine a connection on FSO (B)×L ˙ 0 (L) which will then lift to
c
a connection on S (B). Specifically, if prF and prL0 denote the restrictions
to FSO (B) × ˙ L0 (L) of the projections of FSO (B) × L0 (L) onto FSO (B) and
0
L (L), respectively, then
prF ∗ ω LC ⊕ prL0 ∗A
is a connection on the fiber product and, identifying spinc (4) with the subset
of Cl(4) ⊗ C given in (A.2.40),
ω A = (Spinc )∗ (prF ∗ ω LC ⊕ prL0 ∗A) (A.3.1)
380 Appendix

is a connection on S c (B). Any such connection ω A is called a spinc

connection for L and with one of these we can introduce the basic differential
operator of Seiberg-Witten theory.
Remark: The Levi-Civita connection ω LC on FSO (B) is to be regarded as
fixed. The U (1)-connection A on L0 (L), on the other hand, is the gauge field
of Seiberg-Witten theory and will be constrained only by the field equations
we eventually write down. The task of A is to produce, with ω LC , a spinc
connection ω A on S c (B).
Now any connection ω on any principal bundle G ,→ P −→ X determines a
differential operator ∇ω (called a covariant derivative) on the space Γ(P ×ρ
V) of sections of any associated vector bundle. One simply recalls (Section 6.8
and Appendix B of [N4]) that such sections s correspond bijectively to V-
valued 0-forms φ on P that are tensorial of type ρ, that the covariant exterior
derivative dω φ is also tensorial of type ρ (Theorem 4.5.4) and so it, in turn,
corresponds to a 1-form ∇ω s on X with values in P ×ρ V. Regarding the
1-form ∇ω s as acting on vector fields we can think of the covariant derivative
∇ω s of the section s ∈ Γ(P ×ρ V) as a 1-form with values in the sections
Γ(P ×ρ V) so
∧1
∇ω : Γ(P ×ρ V) −→ (X) ⊗ Γ(P ×ρ V).
π c
A spinc connection ω A on Spinc (4) ,→ S c (B) −→ S
B induces covariant
derivatives on the associated spinor bundles S(L), S + (L) and S − (L), all of
which will be denoted ∇A when thought of as operating on sections of vector
bundles and dA when operating on SC -valued equivariant maps on S c (B). We
will use ∧1
∇A : Γ(S(L)) −→ (B) ⊗ Γ(S(L))
to define the Dirac operator

̸ ˜ A : Γ(S(L)) −→ Γ(S(L))
D

as follows. Let {e1 , e2 , e3 , e4 } be a local oriented, orthonormal frame field on

U ⊆ B (i.e., a local section of FSO (B)). Each ei can be regarded as a vector
field on U and also as a section of the Clifford bundle Cl(B) which therefore
acts by Clifford multiplication on sections of S(L) defined on U . Thus, for
each Ψ ∈ Γ(S(L)) we can define D ̸ ˜ A Ψ on U by
∑
4
̸ ˜ AΨ =
D ei · ∇A Ψ(ei ). (A.3.2)
i=1

One shows that this is independent of the choice of {e1 , e2 , e3 , e4 } and so

̸ ˜ A Ψ globally. We will write out a concrete example shortly.
defines D
Since S(L) = S + (L) ⊕ S − (L) we may restrict D ̸ ˜ A to sections of either
S (L) or S (L). Since Clifford multiplication by ei switches S ± (L), so will
+ −
 A.3. Seiberg-Witten Equations 381

these restrictions. We will write these as

( ) ( )
D̸ A : Γ S + (L) −→ Γ S − (L) (A.3.3)

and
( ) ( )
̸ A∗ : Γ S − (L) −→ Γ S + (L)
D (A.3.4)

(these are, in fact, adjoints relative to the L2 inner product on sections induced
by the pointwise Hermitian inner product on fibers). We will also follow the
custom in mathematics of referring to D ̸ A also as a Dirac operator.
With this we can (at last) formulate the Seiberg-Witten equations. Thus,
we let B denote a compact, connected, simply connected, oriented, smooth
4-manifold. Select a Riemannian metric g for B and then a spinc structure L
for the corresponding oriented, orthonormal frame bundle FSO (B). A pair
(A, ψ) consisting of a U (1)-connection A on U (1) ,→ L0 (B) −→ B and
a positive spinor field ψ ∈ Γ(S + (L)) satisfies the Seiberg-Witten (SW)
equations if

̸ Aψ = 0
D (Dirac Equation) (A.3.5)

and
∗
A = σ ((ψ ⊗ ψ )0 )
F+ +
(Curvature Equation), (A.3.6)

where F +
A is the g -self-dual part of the curvature of A.

Remark: The curvature of A is actually a u(1)-valued 2-form ΩA = dA

on L0 (L), but, since U (1) is Abelian, this projects to a u(1)-valued 2-form on
B and this is what we mean by F A .
To gain some sense of what these equations actually look like we will write
them out explicitly on R4 . More precisely, we consider R4 with its usual
Riemannian metric and orientation. Since R4 is contractible, all of the relevant
bundles over it are trivial and we will work with explicit trivializations. Thus,
the oriented, orthonormal frame bundle is
SO(4) ,→ R4 × SO(4) −→ R4
and there is an essentially unique spinc structure

4
Λ 4
× Spinc(4) × SO(4)

4
382 Appendix

where Λ(b, ξ) = (b, π(ξ)) with π given by (A.2.38). The spinor bundles are
therefore also trivial so their sections can be identified with globally defined
functions on R4 which we will write
 
ψ1
 2 
 ψ 
Ψ =  3  : R4 −→ SC ∼ = C4
 ψ 
ψ4
 
ψ1
 2 
 ψ  + ∼
ψ=   : R4 −→ SC = C2
 0 
0
 
0
 
 0  − ∼
ϕ =  3  : R4 −→ SC = C2 .
 ψ 
ψ4
( )
ψ
For convenience we will often abuse the notation and write Ψ = ϕ by
1 2 3 4
suppressing the zero components. We use x , x , x , x for the standard coor-
dinates on R4 and write ∂i for ∂x ∂
i , i = 1, 2, 3, 4 (these being applied compo-

nentwise to spinor fields).

The determinant line bundle is likewise trivial, as is the corresponding prin-
cipal U (1)-bundle
U (1) ,→ R4 × U (1) −→ R4 .
A connection on this U (1)-bundle is then uniquely determined by a globally
defined u(1) = Im C-valued 1-form on R4 :
A = Ai dxi
Ai : R4 −→ Im C, i = 1, 2, 3, 4.

In orthonormal coordinates the covariant exterior derivative induced by the

Levi-Civita connection is just ordinary (componentwise) exterior differentia-
tion so the covariant derivative ∇A induced by it and the U (1)-connection A
takes the form ∇A = d + A, i.e.,
∇A = ∇i dxi = (∂i + Ai )dxi ,
so that
∇A Ψ = (∂i Ψ + Ai Ψ)dxi
 
(∂i ψ l + Ai ψ 1 )dxi
 
 (∂i ψ 2 + Ai ψ 2 )dxi 
=  (∂ ψ 3 + A ψ 3 )dxi
.

 i i 
(∂i ψ + Ai ψ 4 )dxi
4
 A.3. Seiberg-Witten Equations 383

Thus, with {ei } = {∂i } the standard oriented, orthonormal frame field on R4 ,
we have ∇A Ψ(ei ) = ∂i Ψ + Ai Ψ and, for convenience, we will write this as
 
∂i ψ 1 + Ai ψ 1
 
 ∂i ψ 2 + Ai ψ 2 
∇i Ψ = (∂i + Ai )Ψ =   ∂ ψ3 + A ψ3  .

 i i 
∂i ψ 4 + Ai ψ 4
∑4
The Dirac operator D ̸ ˜ A Ψ = i=1 ei · ∇i Ψ requires that we Clifford multiply
by the basis elements ei , i.e., matrix
( multiply
) by Ei = Γ(ei ) ∈ Cl(4) ⊗ C as
in (A.2.25). For this we write Ψ = ψϕ so that

∑
4 ∑
4
̸ ˜ AΨ =
D ei · ∇i Ψ = Ei ∇i Ψ
i=1 i=1
( )
∇1 ϕ + I∇2 ϕ + J∇3 ϕ + K∇4 ϕ
= .
−∇1 ψ + I∇2 ψ + J∇3 ψ + K∇4 ψ
( )
Note that, as expected, D ̸ ˜ A sends positive spinor fields Ψ = ψ0 to negative
̸ ˜ A to positive spinor fields will
spinor fields and vice versa. The restriction of D
be written
̸ A ψ = −∇1 ψ + I∇2 ψ + J∇3 ψ + K∇4 ψ
D (A.3.7)

by suppressing the zero components (but now one must remember that D ̸ Aψ
is a negative spinor field). The first Seiberg-Witten equation (A.3.5) then
becomes
∇1 ψ = I∇2 ψ + J∇3 ψ + K∇4 ψ (A.3.8)

or, in complete detail,

( )( ) ( )
−(∂1 + A1 ) + i(∂2 + A2 ) (∂3 + A3 ) + i(∂4 + A4 ) ψ1 0
= . (A.3.9)
−(∂3 + A3 ) + i(∂4 + A4 ) −(∂1 + A1 ) − i(∂2 + A2 ) ψ2 0

This part is linear, of course.

For the second Seiberg-Witten equation (A.3.6) we use the (pointwise)
expressions (A.2.50) for σ + ((ψ ⊗ ψ ∗ )0 ) and the following
∑ local idescrip-
tion of F + i<j Fij dx ∧ dx ,
i j
A . With A = A i dx we have F A = dA =
where Fij = ∂i Aj − ∂j Ai . A basis for the self-dual 2-forms is given by
{dx1 ∧ dx2 + dx3 ∧ dx4 , dx1 ∧ dx3 + dx4 ∧ dx2 , dx1 ∧ dx4 + dx2 ∧ dx3 } so
384 Appendix

1 1
F+
A = (F A + ∗ F A ) = (F12 + F34 )(dx1 ∧ dx2 + dx3 ∧ dx4 )
2 2
1
+ (F13 + F42 )(dx1 ∧ dx3 + dx4 ∧ dx2 )
2
1
+ (F14 + F23 )(dx1 ∧ dx4 + dx2 ∧ dx3 ).
2
Thus, (A.3.6) becomes
1
F12 + F34 = − ψ ∗ Iψ
2
1
F13 + F42 = − ψ ∗ Jψ (A.3.10)
2
1
F14 + F23 = − ψ ∗ Kψ
2
or, in more detail,
1 ( )
(∂1 A2 − ∂2 A1 ) + (∂3 A4 − ∂4 A3 ) = − i |ψ 1 |2 − |ψ 2 |2
2 ( )
(∂1 A3 − ∂3 A1 ) + (∂4 A2 − ∂2 A4 ) = −i Re ψ̄ 1 ψ 2 (A.3.11)
( )
(∂1 A4 − ∂4 A1 ) + (∂2 A3 − ∂3 A2 ) = −i Im ψ̄ 1 ψ 2 .

Note that these are only rather mildly nonlinear.

Remark: It is perhaps worth pointing out that these equations do have
nontrivial solutions. We will produce some solutions with ψ = 0 (these will
turn out to be reducible solutions with respect to the gauge action to be
introduced shortly). When ψ = 0 the Dirac equation is satisfied identically
and the curvature equation reduces to dF + A = 0 so what we need is an anti-
self-dual U (1)-connection, i.e., an A = Ai dxi for which dA is of the form
( )
F = F1 dx1 ∧ dx2 − dx3 ∧ dx4
( ) ( )
+ F2 dx1 ∧ dx3 − dx4 ∧ dx2 + F3 dx1 ∧ dx4 − dx2 ∧ dx3 .

What we will do is seek functions F1 , F2 , F3 that are independent of x4 and

for which this 2-form is closed. The Poincaré Lemma then implies that it is
exact so it must be dA for some A and this A (together with ψ = 0) gives
our solution. Now,

d(F1 (dx1 ∧ dx2 − dx3 ∧ dx4 )) = dF1 ∧ (dx1 ∧ dx2 − dx3 ∧ dx4 )
= (∂1 F1 dx1 + ∂2 F1 dx2 + ∂3 F1 dx3 ) ∧ (dx1 ∧ dx2 − dx3 ∧ dx4 )
= −(∂1 F1 )dx1 ∧ dx3 ∧ dx4 − (∂2 F1 )dx2 ∧ dx3 ∧ dx4 +(∂3 F1 )dx1 ∧ dx2 ∧ dx3 .

Computing the remaining terms similarly one finds that the coefficient of dx1 ∧
dx3 ∧ dx4 is −(∂1 F1 + ∂2 F2 + ∂3 F3 ) and this is just (minus) the divergence of
the map F = (F1 , F2 , F3 ). The remaining coefficients are just the components
 A.4. The Moduli Space and Invariant 385

of the curl of this same map F . Consequently, F will be closed if we choose

F to be a map with div F = 0 and curl F = 0 and these are a dime-a-dozen
(think about the static, source-free Maxwell equations, or just make some up).
We return now to the general development so that B is a compact, con-
nected, simply connected, oriented, smooth 4-manifold. Choosing a
Riemannian metric g on B gives an oriented, orthonormal frame bundle and
one can then select a spinc structure L. The corresponding Seiberg-Witten
configuration space A(L) consists of all pairs (A, ψ), where A is a connec-
tion on the principal U (1)-bundle L0 (L) and ψ ∈ Γ(S + (L)) is a positive spinor
field. An (A, ψ) ∈ A(L) is a Seiberg-Witten monopole (SW monopole)
if it satisfies (A.3.5) and (A.3.6) (these two equations together will henceforth
be denoted simply (SW)).

A.4 The Moduli Space and Invariant

As in the case of Donaldson theory, our real interest is in a moduli space of SW
monopoles so we begin by isolating the appropriate gauge group. This will be a
subgroup of the group of automorphisms of the spinc bundle (diffeomorphisms
σ of S c (B) onto itself satisfying σ(p · ξ) = σ(p) · ξ for each p ∈ S c (B) and each
ξ ∈ Spinc (4) and πS c ◦ σ = πS c ).
Recall that S c (B) double covers the fiber product FSO (B)×L ˙ 0 (L) via the
map Spin . Letting prF and prL0 be the projections of FSO (B)×L
c ˙ 0 (L) onto
0
FSO (B) and L (L) we obtain maps

Spinc (4) ,→ S c (B) −→ B




 prF ◦ Spinc (A.4.1)
y
SO(4) ,→ FSO (B) −→ B

and
Spinc (4) ,→ S c (B) −→ B



 prL0 ◦ Spinc (A.4.2)
y

U (1) ,→ L0 (L) −→ B

We will say that an automorphism σ : S c (B) −→ S c (B) covers the

identity on FSO (B) if

prF ◦ Spinc ◦ σ = prF ◦ Spinc . (A.4.3)

386 Appendix

The collection of all such is a group G(L) under composition which we call
the (Seiberg-Witten) gauge group and which we will show acts naturally
on the solutions to (SW).
Lemma A.4.1 If γ ∈ C ∞ (B, U (1)) is any smooth map of B into
U (1) ⊆ Spinc (4), then the map

σγ : S c (B) −→ S c (B)
σγ (p) = p · γ(πS c (p))

is an automorphism of S c (B) that covers the identity on FSO (B). Conversely,

every element of G(L) is σγ for some γ ∈ C ∞ (B, U (1)) and

G(L) ∼
= C ∞ (B, U (1))

where the group operation in C ∞ (B, U (1)) is pointwise multiplication in U (1).

Proof: First note that, because U (1) = Z(Spinc (4)), σγ (p · ξ) =
(p · ξ) · γ(πS c (p · ξ)) = (p · ξ) · γ(πS c (p)) = (p · γ(πS c (p))) · ξ = σγ (p) · ξ
so σγ is a bundle map. Since p and p · γ(πS c (p)) are in the same fiber
of πS c , σγ covers the identity on B so σγ is an automorphism. It covers
the identity on FSO (B) as well because prF ◦ Spinc is locally given by
(b, ξ) = (b, eθi u) −→ (b, Spin (u)) so (b, ξ) and (b, ξ) · eϕi = (b, e(θ+ϕ)i u)
have the same image, i.e., prF ◦ Spinc ◦ σγ = prF ◦ Spinc .
For the converse, it is easy to verify that p1 , p2 ∈ S c (B) have the same
image under prF ◦ Spinc if and only if they differ by the action of some-
thing in U (1), i.e., p2 = p1 · eϕi for some ϕ ∈ R. Thus, an automorphism
σ : S c (B) −→ S c (B) that covers the identity on FSO (B) must satisfy
σ(p) = p · (something in U (1)) for each p ∈ S c (B). We claim that this
“something” must be the same for all points in the same fiber of πS c .
Indeed, πS c (p1 ) = πS c (p2 ) implies p2 = p1 · ξ for some ξ ∈ Spinc (4) and if
σ(p1 ) = p1 · eϕi , then σ(p2 ) = σ(p1 · ξ) = σ(p1 ) · ξ = (p1 · eϕi ) · ξ = (p1 · ξ) · eϕi =
p2 · eϕi also. Thus, σ(p) = p · γ(πS c (p)) for some γ ∈ C ∞ (B, U (1)) as required
and the rest is clear. 
We will use whichever view of the gauge group G(L) is most convenient in any
particular situation.
Our goal now is to show that G(L) acts naturally on the Seiberg-Witten
configuration space A(L) and, indeed, preserves the set of
solutions to (SW). We begin by defining the action of G(L) on positive
spinor fields ψ ∈ Γ(S + (L)). For this we identify ψ with an equivariant SC+

-valued map on S (B) and define the action of σγ ∈ G(L) by pullback,

c

i.e.,

ψ · σγ = ψ · γ = σγ ∗ ψ = ψ ◦ σγ . (A.4.4)

Thus, at each p ∈ S c (B),

 A.4. The Moduli Space and Invariant 387
( ) ( ) ( )
ψ · σγ (p) = ψ σγ (p) = ψ p · γ(πS c (p))
( )−1 (A.4.5)
= γ(πS c (p)) ψ(p).

Thus, if we think instead of ψ as a section of S + (L) we have

ψ · γ = (γ ◦ πS c )−1 ψ. (A.4.6)

The same formulas define the action of G(L) on negative spinor fields.
Turning next to the connection A on U (1) ,→ L0 (L) −→ B we note that the
automorphism σγ of S c (B) induces an automorphism σ ′ γ of L0 (L) as follows:
σγ
S c (B) S c (B)

prL0 ° Spinc prL0 ° Spinc

L0( ) L0( )
σ ɂγ

σ ′ γ ◦ prL0 ◦ Spinc = prL0 ◦ Spinc ◦ σγ (A.4.7)

(we will write out an explicit local expression for σ ′ γ shortly). Now we define
the action of σγ ∈ G(L) on A by
A · σγ = A · γ = (σ ′ γ )∗ A. (A.4.8)

It will be convenient to have (A.4.8) expressed locally in terms of gauge

potentials. Thus, we let s be a local section of L0 (L) and write A = s∗ A.
Also define
( )
A · γ = s∗ (A · γ) = s∗ (σ ′ γ )∗ A = (σ ′ γ ◦ s)∗ A. (A.4.9)

Now, since prL0 ◦ Spinc is locally given by

( ) ( ( )) ( )
(b, ξ) = b, eθi u −→ b, δ eθi u = b, e2θi

it satisfies
( ) (( ) )
prL0 ◦ Spinc (p · ξ0 ) = prL0 ◦ Spinc (p) · δ(ξ0 ) (A.4.10)

so
(( ) ) ( )( )
σ′ γ prL0 ◦ Spinc (p) = prL0 ◦ Spinc σγ (p)
( )( )
= prL0 ◦ Spinc p · γ(πS c (p))
(( ) ) ( )
= prL0 ◦ Spinc (p) · δ γ(πS c (p)) .
388 Appendix

Thus, we may write

( ( )) ( ( ))2
σ ′ γ (x) = x · δ γ πL0 (x) = x · γ πL0 (x) (A.4.11)

for each x ∈ L0 (L). In particular,

( ′ )
σ γ ◦ s (b) = s(b) · (γ(b))2 . (A.4.12)

It then follows from (A.4.8) that

A · γ = (γ 2 )−1 A(γ 2 ) + (γ 2 )−1 d(γ 2 ) = A + (γ 2 )−1 (2γdγ)

and therefore
A · γ = A + 2γ −1 dγ. (A.4.13)

Applying πL∗0 to both sides of (A.4.13) gives

A · γ = A + πL∗0 (2γ −1 dγ) = A + 2(γ ◦ πL0 )−1 d(γ ◦ πL0 ). (A.4.14)

We now have an action of the group G(L) on the Seiberg-Witten configu-

ration space A(L) given by
( )
(A, ψ) · σγ = (A, ψ) · γ = (σ ′ γ )∗ A, σγ ∗ ψ
( )
= A + 2(γ ◦ πL0 )−1 d(γ ◦ πL0 ), (γ ◦ πS c )−1 ψ

or, locally on B,

(A, ψ) · γ = (A + 2γ −1 dγ, γ −1 ψ). (A.4.15)

In order to show that this action preserves solutions to (SW) we first observe
that the spinc connection corresponding to A · γ is the pullback by σγ of that
corresponding to A, i.e.,

ω A·γ = σγ ∗ ω A . (A.4.16)

To see this we note, from (A.3.1), that

σγ ∗ ω A = (Spinc ◦ σγ )∗ (prF∗ ω LC + prL∗0 A)

and
( ( ))
ω A·γ = (Spinc )∗ prF∗ ω LC + prL∗0 (σγ ′ )∗ A
= (prF ◦ Spinc )∗ ω LC + (σγ ′ ◦ prL0 ◦ Spinc )∗ A
= (prF ◦ Spinc ◦ σγ )∗ ω LC + (prL0 ◦ Spinc ◦ σγ )∗ A
 A.4. The Moduli Space and Invariant 389

(by (A.4.3) and (A.4.7))

= (Spinc ◦ σγ )∗ (prF∗ ω LC + prL∗0 A)
= σγ ∗ ω A .

Another computation of ω A·γ using (A.4.14) gives

( ( ( )))
ω A·γ = (Spinc )∗ prF∗ ω LC + prL∗0 A + πL∗0 2γ −1 dγ
( ( ( )))
= ω A + (Spinc )∗ prL∗0 πL∗0 2γ −1 dγ
( )∗ ( )
= ω A + πL0 ◦ prL0 ◦ Spinc 2γ −1 dγ
( )
= ω A + πS∗c 2γ −1 dγ
( )
ω A·γ = ω A + 2 γ ◦ πS c )−1 d(γ ◦ πS c . (A.4.17)

Theorem A.4.2 The action of G(L) on the Seiberg-Witten configuration

space A(L) carries solutions to (SW ) onto other solutions to (SW ). More
precisely, if (A, ψ) ∈ A(L) satisfies
{
̸ Aψ = 0
D
FA+ = σ + ((ψ ⊗ ψ ∗ )0 )

then for any σγ ∈ G(L), (A, ψ) · γ = (A · γ, ψ · γ) satisfies

{
̸ A·γ (ψ · γ) = 0
D
.
+
FA·γ = σ + (((ψ · γ) ⊗ (ψ · γ)∗ )0 )

Proof: For the curvature equation we observe that (A.4.16), the usual trans-
formation equation for the curvature and the fact that U (1) is Abelian imply
that
+
F A·γ = γ 2 F A+ (γ 2 )−1 = F A+ .
Similarly, the commutativity of U (1) gives

(ψ · γ) ⊗ (ψ · γ)∗ = (γ −1 ψ) ⊗ (γ −1 ψ)∗
= (γ −1 ψ) ⊗ (γ ψ ∗ )
= (γ −1 γ)(ψ ⊗ ψ ∗ )
= ψ ⊗ ψ∗ .

Thus, F A+ = σ + ((ψ ⊗ ψ ∗ )0 ) implies F A·γ

+
= σ + (((ψ · γ) ⊗ (ψ · γ)∗ )0 ). To verify
the analogous statement for the Dirac equation it will surely be enough to
show that
̸ A·γ (ψ · γ) = (̸DA ψ) · γ .
D (A.4.18)
390 Appendix

+
For this it will be convenient to identify ψ with an equivariant SC -valued map
on S (B) and compare the covariant exterior derivatives dA ψ and dA·γ (ψ · γ).
c

There are standard formulas for such derivatives (see, for example, (6.8.4) of
[N4]) which, in our case, give
1
dA ψ = dψ + ω A ψ
2
and
1
dA·γ (ψ · γ) = d(ψ · γ) + ω A·γ (ψ · γ) ,
2
where, e.g., ω A takes values in spin (4), identified with a subset of Cl(4) ⊗ C
c

as in (A.2.40), and ω A ψ is a matrix product.

Remark: The factor 12 arises from the actual identification of spinc (4)
with so(4) ⊕ u(1) via the derivative at the identity of the double cover map
Spinc : Spinc (4) −→ SO(4) × U (1).
Now we compute
( )
dA·γ (ψ · γ) = dA·γ (γ ◦ πS c )−1 ψ
( 1 ( )
= d (γ ◦ πS c )−1 ψ) + ω A·γ (γ ◦ πS c )−1 ψ
2
= (γ ◦ πS c )−1 d ψ − (γ ◦ πS c )−2 d (γ ◦ πS c ) ψ
1( )( )
+ ω A + 2(γ ◦ πS c )−1 d(γ ◦ πS c ) (γ ◦ πS c )−1 ψ
2
( 1 )
= (γ ◦ πS c )−1 d ψ + ω A ψ − (γ ◦ πS c )−2 d (γ ◦ πS c ) ψ
2
+ (γ ◦ πS c )−2 d (γ ◦ πS c ) ψ
= (γ ◦ πS c )−1 (dA ψ)

and from this conclude that

∇A·γ (ψ · γ) = (γ ◦ πS c )−1 ∇A ψ. (A.4.19)

Finally,
∑
4
̸ A·γ (ψ · γ) =
D ei · ∇A·γ (ψ · γ)(ei )
i=1
∑4
= ei · (ψ ◦ πS c )−1 ∇A ψ (ei )
i=1
∑
4
−1
= (ψ ◦ π ) Sc ei · ∇A ψ (ei )
i=1
= (ψ ◦ πS c )−1 D
̸ Aψ
= (̸DA ψ) · γ

which proves (A.4.18) and therefore Theorem 6. 

 A.4. The Moduli Space and Invariant 391

Thus, the space of solutions to the Seiberg-Witten equations is invariant

under the action of the gauge group G(L) and we may, as for the anti-self-dual
equations, consider the moduli space M(L) of gauge equivalence classes of
Seiberg-Witten monopoles:

M(L) = {(A, ψ) ∈ A(L) :̸D A ψ = 0, F A+ = σ + ((ψ ⊗ ψ ∗ )0 )}/G(L). (A.4.20)

The ensuing analysis required to manufacture a differential topological

invariant of B from M(L) is in many ways analogous to that which we out-
lined for Donaldson theory. For this reason we will simply sketch with much
broader strokes those aspects of the construction that are much the same and
linger a bit longer over those that present something new. As for Donald-
son theory we will ignore the (important) issue of replacing spaces of smooth
objects with “appropriate Sobelev completions.”
Seiberg-Witten theory is in many ways technically much simpler than Don-
aldson theory and very often the simplifications can be attributed to the fact
that U (1), unlike SU (2), is Abelian. Our first manifestation of this is the ease
with which we identify the reducible elements of the configuration space.
Lemma A.4.3 An element (A, ψ) of A(L) is left fixed by some non-identity
element σγ of G(L) if and only if ψ ≡ 0 and, in this case, γ : B −→ U (1)
must be a constant map.
Proof: It will suffice to argue locally so we let A be a gauge potential for
A. Then (A, ψ) · γ = (A, ψ) if and only if

(A + 2 γ −1 d γ, γ −1 ψ) = (A, ψ)

and this, in turn, is the case if and only if

γ −1 ψ = ψ and 2 γ −1 d γ = 0 .

Since γ ̸= 1, the first of these can be true if and only if ψ ≡ 0. The second
implies d γ = 0 and, since B is connected, γ must be constant. 
An (A, ψ) ∈ A(L) is said to be reducible if ψ ≡ 0 and irreducible
otherwise.
As was the case in Donaldson theory, the configuration space A(L) is an
affine space and therefore an infinite-dimensional manifold. The tangent space
at any (A, ψ) ∈ A(L) can be identified with
∧1
T(A,ψ) (A(L)) = (B, Im C) ⊕ Γ(S + (L)) (A.4.21)

(no adjoint bundle required in the first summand because U (1) is Abelian).
The gauge group G(L) has the structure of a Hilbert Lie group whose Lie
∧0
algebra can be identified with (B, Im C). Fixing (A, ψ) ∈ A(L), the ac-
tion of G(L) on (A, ψ) gives a map G(L) −→ A(L) whose derivative at the
392 Appendix

identity is
∧0 ∧1
(B, Im C) −→ (B, Im C) ⊕ Γ(S + (L))
a −→ (2 d a, −a · ψ) , (A.4.22)

where a · ψ is the rotation of ψ obtained by regarding each value of a as an

element of the complexified Clifford algebra and Clifford multiplying by a.
Remark: Here is a formal, i.e., C ∞ , local argument to persuade you of
(A.4.22). The map is γ ∈ G(L) −→ (A, ψ) · γ = (A + 2γ −1 d γ, γ −1 ψ). Let
∧0
a∈ (B, Im C) and write a(x) = i θ(x), where θ is a real-valued function
on B. Then α(t) = ei tθ(x) is a curve in G(L) = C ∞ (B, U (1)) with α(0) = 1
and α′ (0) = a. Thus, the derivative at t = 0 is

d( ) d( )
(A, ψ) · ei tθ(x) = A + 2i tdθ, e−i tθ(x) ψ
dt t=0 dt t=0
= (2i dθ, −iθ(x)ψ)
= (2d a, −aψ)
= (2d a, −a · ψ) .

Again as in Donaldson theory, the “large” moduli space

B(L) = A(L)/G(L)
of configurations is a smooth Banach manifold away from the reducible config-
urations (i.e., away from those [A, ψ] with ψ ≡ 0) and the monopole moduli
space M(L) is a subset of it. Define the Seiberg-Witten map
∧2
F : A(L) −→ (B, Im C) ⊕ Γ(S − (L))
+

by
F (A, ψ) = (F A+ − σ + ((ψ ⊗ ψ ∗ )0 ), D
̸ A ψ) . (A.4.23)

Then (A, ψ) satisfies (SW ) if and only if F (A, ψ) = (0, 0). The derivative of
F at (A, ψ) ∈ M(L), i.e., the linearization of the Seiberg-Witten equations
at (A, ψ), is
∧1 ∧2
F∗(A,ψ) : (B, Im C) ⊕ Γ(S + (L)) −→ (B, Im C) ⊕ Γ(S − (L))
+
( )
d+ −Dψ
F∗(A,ψ) = (A.4.24)
· 12 ψ ̸ A
D

where
∧1 ∧2
d+ : (B, Im C) −→ (B, Im C)
+
 A.4. The Moduli Space and Invariant 393

is d followed by the projection onto the self-dual part,

1 ∧1
· ψ: (B, Im C) −→ Γ(S − (L))
2
takes a 1-form α to α · 12 ψ (which is understood to mean Clifford
multiplication by the vector field dual to the 1-form α) and
∧2
Dψ : Γ(S + (L)) −→ (B, Im C)
+
is given by
( ) )
1(
∗
Dψ (η) = σ ψ ⊗ η + η ⊗ ψ −
+ ∗
⟨ψ, η⟩ + ⟨ψ, η⟩ 1
2

(the object inside the parentheses being a section of End0 (S + (L)) which σ +
identifies with a self-dual 2-form on B).
Remark: (A.4.24) can be verified with a local argument analogous to that
for (A.4.22).
Associated to any solution (A, ψ) to (SW ) is a fundamental elliptic
complex E(A, ψ):
∧0 ∧1
0 −→ (B, Im C) −→ (B, Im C) ⊕ Γ(S + (L))
∧2
−→ (B, Im C) ⊕ Γ(S − (L)) −→ 0 ,
+

where the second and third maps are, respectively, the derivative (A.4.22) of
the action of G(L) on (A, ψ), and the derivative (A.4.24) of the Seiberg-
Witten map at (A, ψ). This complex has finite-dimensional cohomology
groups H i (A, ψ), i = 0, 1, 2, which admit interpretations analogous to those
in Donaldson theory:
H 0 (A, ψ) = tangent space to the stablizer of (A, ψ) in G(L) so
H 0 (A, ψ) = 0 ⇐⇒ (A, ψ) irreducible
⇐⇒ ψ ̸≡ 0

H 1 (A, ψ) = formal tangent space to M(L) at [A, ψ]

H 2 (A, ψ) = obstruction space, i.e.,
H 2 (A, ψ) = 0 ⇐⇒ Implicit Function Theorem
gives a local manifold
structure for F −1 (0, 0) near
(A, ψ) of dimension dim H 1 (A, ψ).
394 Appendix

If H 2 (A, ψ) = 0 and H 0 (A, ψ) = 0, then the local manifold structure for

F −1 (0, 0) near (A, ψ) projects injectively into the moduli space M(L) and,
near [A, ψ], M(L) is a smooth manifold of dimension

dim H 1 (A, ψ) = − dim H 0 (A, ψ) + dim H 1 (A, ψ) − dim H 2 (A, ψ)

= −Index (E(A, ψ))
1
= (c1 (L0 )2 − 2χ(B) − 3σ(B)),
4
where the last expression
∫ comes from the Atiyah-Singer Index Theorem and
c1 (L0 )2 means B c1 (L0 ) ∧ c1 (L0 ).
As in Donaldson theory, H 0 (A, ψ) = 0 and H 2 (A, ψ) = 0 are the “generic”
situation when b+ 2 (B) > 1, but this means something slightly different here.
First, the part that is the same.
Theorem A.4.4 (Generic Metrics Theorem) Let B denote a compact,
connected, simply connected, oriented, smooth 4-manifold with b+ 2 (B) > 0.
Then there is a dense subset Gen(R) of the space R(B) of Riemannian metrics
on B with the following property: For any g ∈ Gen(R) and any corresponding
spinc structure L, any solution (A, ψ) to the Seiberg-Witten equations is irre-
ducible, i. e., satisfies H 0 (A, ψ) = 0. If b+
2 (B) > 1, then for any generic path
g(t), 0 ≤ t ≤ 1, of Riemannian metrics in R(B) there are no reducible so-
lutions to the Seiberg-Witten equations for any spinc structure corresponding
to any of the metrics g(t), 0 ≤ t ≤ 1.
For H 2 (A, ψ) = 0 there is no known generic metrics theorem of this
sort. In this case one must perturb, not the metric, but the equations them-
∧2precisely, we fix the metric g and the spin structure L. For any
c
selves. More
fixed η ∈ + (B, Im C) we introduce the (η) perturbed Seiberg-Witten
(PSW) equations
̸ Aψ = 0
D (A.4.25)
∗
A = σ ((ψ ⊗ ψ )0 ) + η.
F+ +
(A.4.26)
Remark: The motivation here is easy to understand. Solutions to (SW) are
solutions to the equation F (A, ψ) = (0, 0), where F is the Seiberg-Witten
map (A.4.23). For this to be a manifold, (0, 0) must be a regular value of F .
If it is not, the infinite-dimensional
∧ version of Sard’s Theorem suggests that
a small perturbation of (0, 0) in 2+ (B, Im C) ⊕ Γ(S − (L)) of the form (η, 0)
will be a regular value so that F (A, ψ) = (η, 0) will define a manifold of
(A, ψ). But F (A, ψ) = (η, 0) is just (PSW).
The linearized complex at any solution to (PSW) is given by the same maps
as for (SW) so the cohomology is the same. G(L) acts on solutions to (PSW) in
the same way so there is a moduli space M(L, η) of solutions and everything
we have said above for (SW) is also true for (PSW).
 A.4. The Moduli Space and Invariant 395

Theorem A.4.5 (Generic Perturbations Theorem) Let B denote a

compact, connected, simply connected, oriented, smooth 4-manifold. Fix a
Riemannian metric g and a spinc structure L for B. Then there is a dense sub-
∧2 ∧2
set Gen( + ) in the space + (B, ImC) of ImC-valued self-dual 2-forms on B
∧2
with the following properties: For η ∈ Gen( + ), every
solution (A, ψ) to the perturbed Seiberg-Witten equations (A.4.25) and (A.4.26)
has H 2 (A, ψ) = 0. If b+
∧2 2 (B) > 0 and g ∈ Gen(R), then, for any η ∈
Gen( + ), the moduli space M(L, η) is a smooth submanifold of B(L) of
dimension 14 (c1 (L0 )2 − 2χ(B) − 3σ(B)).
Remark: In particular, if c1 (L0 )2 − 2χ(B) − 3σ(B) < 0, then the moduli
space is generically empty.

Exactly as in the case of Donaldson theory one can show that, for a fixed
generic metric g and perturbation η and any associated spinc structure L,
a choice of orientation for the vector space H+ 2
(B; R ) canonically orients
all of the moduli spaces M(L, η). Likewise as in Donaldson theory, when
b+
2 (B) > 1 there is a cobordism result which roughly says that for a generic
l-parameter family g (t), 0 ≤ t ≤ 1, of metrics and a generic l-parameter
family η(t), 0 ≤ t ≤ 1, of perturbations, the moduli spaces parametrized
by t fit together to form a smooth manifold with boundary containing no
points corresponding to reducible solutions. The boundary is the disjoint union
of moduli spaces for (g (0), η(0)) and (g (1), η(1)). Moreover, selecting an
orientation for H+2
(B; R) orients this parametrized moduli space and the two
boundary moduli spaces inherit opposite orientations.

Remark: There is a technical point which we glossed over here and should
mention because it has no analogue in Donaldson theory. Changing the metric
g changes the orthonormal frame bundle and so, one would think, the spinc
structure. It would appear that the discussion above is incomplete without a
specification of a spinc structure for each t. In fact, however, one can show
that frame bundles for different metrics are naturally isomorphic and so one
can pull back spinc structures by the isomorphisms, thus effectively “fixing”
L (up to equivalence) regardless of the choice of g .
Except for a few minor simplifications and adaptations the story of the
Seiberg-Witten moduli space thus far has been virtually indistinguishable
from what we had to say about the anti-self-dual moduli space. The one
aspect of Seiberg-Witten theory that differs significantly from Donaldson the-
ory (and that accounts for its relative simplicity) is that there is no need for
an “Uhlenbeck- style compactification”:
For any metric g , and spinc structure L
and any perturbation η, the moduli space
M(L, η) is always compact.
396 Appendix

The proof of this involves what is called an elliptic “bootstrapping” argument

(which we will not describe) based on the crucial fact that the spinor field ψ
and curvature F A for any solution (A, ψ) to (PSW) satisfy uniform a pri-
ori bounds (this is categorically false for the anti-self-dual equations because
these are conformally invariant in dimension four). Because of its significance
we will sketch a proof of this but, since the perturbation adds only arithmetic
to the argument, we will do this for the unperturbed equations, written in the
form
̸ Aψ = 0
D (A.4.27)
∗
A ) = (ψ ⊗ ψ )0
ρ+ (F + (A.4.28)
(recall that ρ+ is the inverse of σ + ). We will appeal to the famous Weitzenböck
formula from differential geometry which, in our present circumstances, reads
1
̸ ∗A ◦ D
D ̸ A ψ = ∇∗A ◦ ∇A ψ + κψ + ρ+ (F + A )ψ, (A.4.29)
4
∗
where DA is the formal adjoint of D ̸ A : Γ(S + (L)) −→ Γ(S − (L)),
∗
∇A is the formal adjoint of the covariant derivative ∇A : Γ(S + (L)) −→
∧1 (B) ⊗ Γ(S + (L)) and κ is the scalar curvature of B (for the metric g ). Be-
cause (A, ψ) is a solution to the Seiberg-Witten equations, (A.4.29) reduces to
1
∇∗A ◦ ∇A ψ + κψ + (ψ ⊗ ψ ∗ )0 ψ = 0. (A.4.30)
4
Take the pointwise inner product with ψ to obtain
1 1
⟨ ∇A∗ ◦ ∇A ψ, ψ ⟩ + κ∥ ψ ∥2 + ∥ ψ ∥4 = 0 (A.4.31)
4 2
(for the last term compute (ψ ⊗ ψ ∗ )0 ψ from (A.2.49) and then take the inner
product with ψ).
Now, ∥ ψ(x) ∥ is a continuous function on the compact space B so there is
an x0 ∈ B at which it achieves an absolute maximum. We claim that, at this
point, the first term in (A.4.31) is non-negative, i.e.,

⟨ ∇A∗ ◦ ∇A ψ(x0 ), ψ(x0 ) ⟩ ≥ 0 (A.4.32)

(note that (A.4.31) implies that ⟨ ∇A∗ ◦ ∇A ψ, ψ ⟩ must be real). The proof
depends on the identity

∆g ∥ ψ ∥2 = −2∥ ∇A ψ ∥2 + 2⟨ ∇A∗ ◦ ∇A ψ, ψ ⟩ (A.4.33)

where ∆g = d∗ ◦ d is the scalar Laplacian corresponding to g . This can be

verified by writing out ∆g ∥ ψ ∥2 in a local orthonormal frame field. Now, at
x0 , (A.4.33) gives

2⟨ ∇A∗ ◦ ∇A ψ(x0 ), ψ(x0 ) ⟩ = △g ∥ ψ ∥2 (x0 ) + 2∥ ∇A ψ ∥2 (x0 ). (A.4.34)

 A.4. The Moduli Space and Invariant 397

Now, obviously 2∥ ∇A ψ ∥2 (x0 ) ≥ 0. Moreover, since ∥ ψ ∥2 achieves a maxi-

mum at x0 , △g ∥ ψ ∥2 (x0 ) ≥ 0 as well so (A.4.32) is proved. It then follows
from (A.4.31), evaluated at x0 , that
1 1
κ(x0 )∥ ψ(x0 ) ∥2 + ∥ ψ(x0 ) ∥4 ≤ 0
4 2
so
1
∥ ψ(x0 ) ∥4 ≤ − κ(x0 )∥ ψ(x0 ) ∥2 .
2
There are two possibilities. Either ∥ ψ(x0 ) ∥ = 0, in which case ψ ≡ 0 and
(A, ψ) is a reducible solution (and so ∥ ψ ∥ is certainly uniformly bounded).
Otherwise we have
1
∥ ψ(x0 ) ∥2 ≤ − κ (x0 )
2
and therefore
1
∥ ψ(x) ∥2 ≤ − κ(x0 ) (A.4.35)
2
for every x in B. Now, despite appearences, the right-hand side of (A.4.35)
depends on ψ (through x0 ) so, to get a bound on the spinor field of every
solution (A, ψ) we define k(x0 ) = max {− 12 κ(x0 ), 0} for each x0 ∈ B and

k(B) = max{k(x0 ) : x0 ∈ B}

and conclude that for any fixed metric and any spinc structure, any solution
(A, ψ) to the Seiberg-Witten equations has spinor field ψ bounded by the
geometrical constant k(B):

∥ ψ(x) ∥2 ≤ k(B) ∀x ∈ B. (A.4.36)

The second Seiberg-Witten equation (A.3.6) then gives a uniform bound on

the self-dual part of the curvature for any solution. A bit more work then
gives a bound on the anti-self-dual part of the curvature. From these one can
deduce that, for a given g (and η in the perturbed case) there are at most
finitely many (equivalence classes of) spinc structures for which the moduli
space is nonempty (Theorem 5.2.4 of [M1]).
The bounds described thus far are not sufficient to prove the compactness of
the moduli space. For this one must bound the connection parts A of solutions
(A, ψ) “up to gauge”. This is generally accomplished by a gauge fixing argu-
ment and the bootstrapping referred to above (see Section 5.3 of [M1]). In any
case, it can be done and the end result is that Seiberg-Witten moduli spaces
are always compact (and, generically, are smooth manifolds). This compact-
ness simplifies enormously the task of defining and computing “Donaldson-
like” invariants in Seiberg-Witten theory (because there is no need, as there
is in the anti-self-dual case, to compactify the moduli space before integrating
398 Appendix

cohomology classes over it). Even so we intend to consider what appears to

be only a very special case. Thus, we fix a generic metric g and perturba-
tion η and suppose that there exists a spinc structure L for which the formal
dimension of the moduli space is zero, i.e.,

c1 (L0 )2 = 2χ(B) + 3σ(B) (A.4.37)

(such an L need not exist). Assuming that an orientation for the vector space
H+2
(B; R) has been fixed, the moduli space is a finite set of isolated points each
of which is equipped with a sign ±1. The sum of these signs is an integer and,
as in Donaldson theory, when b+ 2 (B) > 1 a cobordism argument shows that
the integer is independent of the choice of (generic) metric and perturbation.
We call this integer the 0-dimensional Seiberg-Witten invariant of B
associated with L and denote it

SW0 (B, L).

This is, indeed, an invariant in the sense that, if f : B ′ −→ B is a

diffeomorphism for which the induced map f ∗ : H+ 2
(B; R) −→ H+ 2
(B ′ ; R)
preserves orientation, then the induced spinc structure f ∗ L for B ′ also satis-
fies (A.4.37) and SW0 (B ′ , f ∗ L) = SW0 (B, L).
Remark: When one goes to the trouble of defining the Seiberg-Witten in-
variant even when (A.4.37) is not satisfied (which we will not) one obtains
a map SW (B, ·) on the set of (equivalence classes of) spinc structures on
B which assigns to each such an integer SW (B, L) ∈ Z, taken to be zero
if the corresponding moduli space is empty. We have already noted that
there can be at most finitely many L for which SW (B, L) ̸= 0. The em-
pirical evidence suggests that, when b+ 2 (B) > 1, SW (B, L) ̸= 0 occurs only
for those L satisfying (A.4.37). We will say that a B with b+ 2 (B) > 1 is of
SW -simple type if SW (B, L) ̸= 0 implies that L must satisfy (A.4.37),
i.e., if nonzero Seiberg-Witten invariants occur only for 0-dimensional mod-
uli spaces. It has been conjectured that every B with b+ 2 (B) > 1 is of
SW -simple type. Finally, we shall refer to the elements of H 2 (B) which arise
as c1 (L0 ) for some spinc structure L satisfying (A.4.37) as Seiberg-Witten
(SW -) basic classes. Thus, SW -basic classes are just the first Chern classes
of complex line bundles corresponding to spinc structures for which the SW
moduli space is 0-dimensional.

A.5 The Witten Conjecture

As we saw in Section A.l, Witten was led to Seiberg-Witten gauge theory
through topological quantum field theory and saw it as a dual version of Don-
aldson theory. Indeed, Witten was led to much more. He formulated a very con-
crete conjecture on the relationship between Donaldson and Seiberg-Witten
 A.5. The Witten Conjecture 399

invariants. The conjecture was remarkable. It was a very deep and purely
mathematical statement, but one suggested entirely by physics. We will con-
clude with a very brief description of what Witten believed must be true (and
was eventually proved by Feehan and Leness [FeLe]).
The sequence γd (B), d = 0, 1, 2, . . ., of Donaldson invariants for B can be
assembled into a single formal power series
∞
∑ γd (B)(x)
DB (x) =
d!
d=0

on H 2 (B) called the Donaldson series of B. For example, all of the invariants
for K3 have been computed and one finds that
∑∞
1
DK3 (x) = (QK3 (x, x)/2)n = exp(QK3 (x, x)/2).
n=0
n!

Kronheimer and Mrowka [KM] formulated a condition on B which we will

call KM-simple type that is sufficient to guarantee that if one is somehow
given the Donaldson series DB (x) one can inductively extract from it all of
the individual Donaldson invariants γd (B).
Remark: In order to state this condition precisely one requires more de-
tailed information about how the γd (B) are “really” defined (see the Remark
following our “naive” definition in Section A.1). However, there is no known
counterexample to the conjecture that every manifold B of the type we are
considering is of KM-simple type.
They then proved a remarkable structure theorem for such manifolds the
essential content of which was that all of the information contained in the
(infinitely many) Donaldson invariants is also contained in a finite set of data
(certain cohomology classes K1 , . . . , Ks of B and corresponding rational num-
bers a1 , . . . , as ). Specifically, they proved the following theorem.

Theorem A.5.1 If B is of KM-simple type, then there exist cohomology

classes K1 , . . . , Ks ∈ H 2 (B), called KM-basic classes, and rational numbers
a1 , . . . , as ∈ Q, called KM-coefficients, such that
∑
s
DB (x) = exp(QB (x, x)/2) ar exp(Kr (x))
r=1

for every x ∈ H2 (B). Moreover, each KM-basic class reduces mod 2 to the
second Stiefel-Whitney class w2 (B) of B.
The appearance of Theorem A.5.1 in the spring of 1994 was an extraor-
dinary breakthrough in Donaldson theory. Enormously complicated calcula-
tions of an infinite set of invariants were suddenly replaced by the (certainly
400 Appendix

not easy, but at least finite) problem of determining the KM-basic classes
and coefficients. As fate would have it, however, the fall of 1994 witnessed
another event which rendered this triumph of Kronheimer and
Mrowka moot. Edward Witten, in his now famous lecture at M.I.T.
(described in [Tau2]), made the conjecture which, within weeks, brought
about the demise of Donaldson theory and initiated an entirely new approach
to the study of smooth 4-manifolds.

Witten’s Conjecture Let B be a compact, connected, simply connected,

oriented, smooth 4-manifold with b+
2 (B) > 1 and odd. Then

1. B is of SW-simple type if and only if B is of KM-simple type and, in this

case,
2. SW-basic classes coincide with KM-basic classes and
( ) ∑ m(B)
DB (x) = exp QB (x, x)/2 2 SW0 (B, L)exp(c1 (L0 )(x)),

where m(B) = 2 + 14 (7χ(B) + 11σ(B)), χ(B) is the Euler characteristic of

B, σ(B) is the signature of B and the sum is over all (equivalence classes
of ) spinc structures L for which (A.4.37) is satisffied.
The content of the conjecture is that, for manifolds of simple type, the basic
classes are just those elements of H2 (B) corresponding to spinc structures with
0-dimensional SW moduli spaces and the coefficients are, up to the topological
factor 2m(B) , just the corresponding 0-dimensional SW invariants.
There are a number of attitudes one might adopt toward a conjecture of this
sort. One might, of course, try to prove it and, although this has not been the
principal focus of work in this area, much deep and interesting mathematics
has been directed toward this end. A strategy suggested by Pidstrigatch and
Tyurin [PT] for relating the two sets of invariants by viewing the Donaldson
and Seiberg-Witten moduli spaces as singular submanifolds of a larger moduli
space of “nonabelian monopoles” has been taken up in earnest by Feehan and
Leness in a long series of difficult and technical papers (see [FeLe]). By far
the more prevalent attitude has been that, even if a proof is hard to come by,
the conjecture has been checked in every case in which all of the invariants
are known and has survived so that it would seem to make good practical
sense for the topologist to (at least provisionally) abandon the Donaldson
invariants for the much more tractable Seiberg-Witten invariants. Perhaps
the most enlightened attitude, however, and one which has been emphasized
by Atiyah, is that if physics is truly capable of casting such a penetrating light
upon mathematics at the very deepest levels, then mathematicians will want
to take heed and turn their attention once again to their historical roots in
physics.
References
[AB] Aharonov, Y. and D. Bohm, “Significance of electromagnetic po-
tentials in the quantum theory”, Phys. Rev., 115(1959), 485–491.
[AJ] Atiyah, M.F. and L. Jeffrey, “Topological Lagrangians and
Cohomology”, J. Geo. Phys., 7(1990), 119–136.
[BB] Booss, B. and D.D. Bleecker, Topology and Analysis: The Atiyah-
Singer Index Formula and Gauge-Theoretic Physics, Springer-
Verlag, New York, Berlin, 1985.
[BGV] Berline, N., E. Getzler and M. Vergne, Heat Kernels and Dirac
Operators, Springer-Verlag, New York, Berlin, 1996.
[Bl] Bleecker, D., Gauge Theory and Variational Principles, Addison-
Wesley, Reading, MA, 1981.
[BPST] Belavin, A., A. Polyakov, A. Schwartz and Y. Tyupkin,
“Pseudoparticle solutions of the Yang-Mills equations”, Phys.
Lett., 59B(1975), 85–87.
[Brow] Browder, A., Mathematical Analysis: An Introduction, Springer-
Verlag, New York, Berlin, 1996.
[BT] Bott, R. and L.W. Tu, Differential Forms in Algebraic Topology,
Springer-Verlag, New York, Berlin, 1982.
[DT] Dereli, T. and M. Tekmen, “Wu-Yang monopoles and non-Abelian
Seiberg-Witten equations”, hep-th/9806031.
[EGH] Eguchi, T., P.B. Gilkey and A.J. Hanson, “Gravitation, Gauge
Theories and Differential Geometry”, Physics Reports, 66,
No. 6(1980), 213–393.
[Fang] Fang, J., Abstract Algebra, Schaum Publishing Co., New York,
1963.
[FeLe] Feehan, P.M.N. and T.G. Leness, “Witten’s Conjecture
for Many Four-Manifolds of Simple Type”, arXiv:math/
0609530v2[math.DG].
[Fey] Feynman, R.P., R.B. Leighton and M. Sands, The Feynman Lec-
tures on Physics, Volumes I–III, Addison-Wesley, Reading, MA,
1964.
[FG-M] Feynman, R.P. and M. Gell-Mann, “Theory of Fermi interactions”,
Phys. Rev., 109, 193.

401
402 References

[FU] Freed, D.S., and K.K. Uhlenbeck, Instantons and Four-Manifolds,

MSRI Publications, Springer-Verlag, New York, Berlin, 1984.
[G1] Geroch, R.P., “Spinor structure of space-times in general relativity,
I”, J. Math. Phys., 9(1968), 1739–1744.
[G2] Geroch, R.P., “Spinor structure of space-times in general relativity,
II”, J. Math. Phys., 11(1970), 343–348.
[GHVI] Greub, W., S. Halperin and R. Vanstone, Connections, Curvature
and Cohomology, Vol. I, Academic Press, New York, 1972.
[Gre] Greiner, W., Relativistic Quantum Mechanics, Springer-Verlag,
New York, Berlin, 1993.
[GO] Goddard, P. and D.I. Olive, “Magnetic monopoles in gauge field
theories”, Rep. Prog. Phys., Vol. 41(1978), 1357–1437.
[GP] Guillemin, V. and A. Pollack, Differential Topology, Prentice-Hall,
Englewood Cliffs, NJ, 1974.
[Groi1] Groisser, D., “SU (2) Yang-Mills-Higgs Theory on R3 ”, Harvard
Ph.D. thesis, 1983.
[Groi2] Groisser, D., “Integrality of the monopole number in SU (2) Yang-
Mills-Higgs theory on R3 ”, Comm. Math. Phys., 93(1984), 367–
378.
[GR] Gunning, R.C. and H. Rossi, Analytic Functions of Several Com-
plex Variables, Prentice-Hall, Englewood Cliffs, NJ, 1965.
[Gui] Guidry, M., Gauge Field Theories, John Wiley & Sons, Inc., New
York, 1991.
[HE] Hawking, S.W. and G.F.R. Ellis, The Large Scale Structure of
Space-Time, Cambridge University Press, Cambridge, England,
1976.
[Helg] Helgason, S., Differential Geometry and Symmetric Spaces, Aca-
demic Press, New York, 1962.
[Hir] Hirsch, M.W., Differential Topology, Springer-Verlag, GTM #33,
New York, Berlin, 1976.
[Hol] Holstein, B.R., Topics in Advanced Quantum Mechanics,
Addison-Wesley, Redwood City, CA, 1992.
[Howe] Howe, R., “Very basic Lie theory”, Amer. Math. Monthly, Novem-
ber, 1983, 600–623.
[Huse] Husemoller, D., Fibre Bundles, 3rd Ed., Springer-Verlag, New
York, Berlin, 1994.
 References 403

[Jost] Jost, J., Riemannian Geometry and Geometric Analysis, Third

Edition, Springer-Verlag, New York, Berlin, 2002.
[JT] Jaffe, A. and C. Taubes, Vortices and Monopoles: Structure of
Static Gauge Theories, Birkhauser, Boston, 1980.
[KM] Kronheimer, P.B. and T.S. Mrowka, “Embedded Surfaces and the
Structure of Donaldson’s Polynomial Invariants”, J. Diff. Geo.,
3(1995), 573–734.
[KN1] Kobayashi, S. and K. Nomizu, Foundations of Differential
Geometry, Vol. 1, Wiley-Interscience, New York, 1963.
[KN2] Kobayashi, S. and K. Nomizu, Foundations of Differential
Geometry, Vol. 2, Wiley-Interscience, New York, 1969.
[Lang] Lang, S., Linear Algebra, 2nd Edition, Addison-Wesley, Reading,
MA, 1971.
[Lan] Landkof, N.S., Foundations of Modern Potential Theory,
Springer-Verlag, New York, Berlin, 1972.
[LM] Lawson, H.B. and M.L. Michelsohn, Spin Geometry, Princeton
University Press, Princeton, NJ, 1989.
[M1] Morgan, J.W., The Seiberg-Witten Equations and Applications
to the Topology of Smooth Four-Manifolds, Princeton University
Press, Princeton, NJ, 1996.
[M2] Morgan, J.W., “An Introduction to Gauge Theory”, in Gauge The-
ory and the Topology of Four-Manifolds, American Mathematical
Society, 1998.
[Miln] Milnor, J., Topology from the Differentiable Viewpoint, University
of Virginia Press, Charlottesville, VA, 1966.
[MQ] Mathai, V. and D. Quillen, “Superconnections, Thom Classes
and Equivariant Differential Forms”, Topology, 25(1986),
85–110.
[MS] Milnor, J.W. and J.D. Stasheff, Characteristic Classes, Princeton
University Press, Princeton, NJ, 1974.
[MT] Madsen, I. and J. Tornehave, From Calculus to Cohomology:
De Rham Cohomology and Characteristic Classes, Cambridge Uni-
versity Press, Cambridge, England, 1997.
[N1] Naber, G.L., Topological Methods in Euclidean Spaces, Dover Pub-
lications, Mineola, NY, 2000.
[N2] Naber, G.L., Spacetime and Singularities, Cambridge University
Press, Cambridge, England, 1988.
404 References

[N3] Naber, G.L., The Geometry of Minkowski Spacetime, Springer-

Verlag, Applied Mathematical Sciences Series #92, 2nd Edition,
New York, Berlin (to appear 2011).
[N4] Naber, G.L., Topology, Geometry and Gauge Fields: Foundations,
Springer-Verlag, Texts in Applied Mathematics 25, 2nd Edition,
New York, Berlin, 2010.
[O’N] O’Neill, B., Semi-Riemannian Geometry: with Applications to
Relativity, Academic Press, 1983.
[Pen] Penrose, R., Techniques of Differential Topology in Relativity,
S.I.A.M., Philadelphia, PA, 1972.
[PT] Pidstrigach, V and A. Tyurin, “Localization of Donaldson’s Invari-
ants along Seiberg-Witten Classes”, arXiv: dg-ga/9507004.
[Ry] Ryder, L.H., Quantum Field Theory, Cambridge University Press,
Cambridge, England, 1996.
[SaW] Sachs, R.H. and H. Wu, General Relativity for Mathematicians,
Springer-Verlag, New York, Berlin, 1977.
[Sp1] Spivak, M., Calculus on Manifolds, W. A. Benjamin, Inc., New
York, 1965.
[Sp2] Spivak, M., A Comprehensive Introduction to Differential
Geometry, Volumes I–V, Publish or Perish, Inc., Boston, 1979.
[St] Steenrod, N., The Topology of Fibre Bundles, Princeton Univ.
Press, Princeton, N.J., 1951.
[SW] Seiberg, N. and E. Witten, “Monopoles, duality and chiral
symmetry breaking in N = 2 supersymmetric QCD”, Nucl. Phys.
B 431(1994), 581–640.
[Tau1] Taubes, C.H., “Self-Dual Connections on Non-Self-Dual
4-Manifolds”, J. Diff. Geo., 17(1982), 139–170.
[Tau2] Taubes, C.H., The Seiberg-Witten Invariants (Videotape),
American Mathematical Society, Providence, RI, 1995.
[vdW] van der Waerden, B.L., Group Theory and Quantum Mechanics,
Springer-Verlag, New York, Berlin, 1974.
[W1] Witten, E., “Supersymmetry and Morse Theory”, J. Diff. Geo.,
17(1982), 661–692.
[W2] Witten, E., “Topological quantum field theory”, Comm. Math.
Phys. 117(1988), 353–386.
[W3] Witten, E., “Monopoles and 4-manifolds”, Math. Res. Letters
1(1994), 764–796.
 References 405

[Warn] Warner, F.W., Foundations of Differentiable Manifolds and Lie

Groups, Springer-Verlag, GTM #94, New York, Berlin, 1983.
[Wein] Weinberg, S., The Quantum Theory of Fields, Volumes I and II,
Cambridge University Press, Cambridge, England, 1995 and 1996.
[YM] Yang, C.N. and R.L. Mills, “Conservation of isotopic spin and
isotopic gauge invariance”, Phys. Rev., 96(1954), 191–195.
Symbols

What follows is a list of the those symbols that are used consistently through-
out the text, a brief discription of their meaning and/or a reference to the
page on which such a description can be found.

R, real numbers
C, complex numbers
H, quaternions
F = R, C, or H
⟨ , ⟩, inner product on Fn
| |, absolute value in R, modulus in C or H
dim X, dimension of the manifold X
C ∞ (X), real-valued C ∞ -functions on X
id, identity map
Tp (X), tangent space at p
Di , i th partial derivative operator
ι, inclusion map
α′ (t0 ), velocity vector to α at t0
f∗p , derivative of f at p
V ∗ , dual of vector space V
(US , φS ), (UN , φN ), stereographic projection charts
∥ ∥, norm in Euclidean space
SU (2), 5
FPn−1 , projective spaces
[ ξ ], equivalence class containing ξ
P, projection map
GL(n, F), 5
GL(n, F), 5
U (n, F), 5
O(n), 5
U (n), 5
Sp(n), 5
SO(n), 5
SU (n), 5
ĀT , conjugate transpose of A
φ × ψ, 5

407
408 Symbols

V (f ) = V f , 7
X (X), smooth vector fields on X
[V , W ], Lie bracket of vector fields
[e1 , . . . , en ], orientation class of {e1 , . . . , en }
Tp∗ (X), cotangent space at p
df , exterior derivative of f
dfp = df (p)
X ∗ (X), smooth 1-forms on X
Θ(V ) = ΘV , 7
F ∗ , pullback by F
⊗, tensor product
∧, wedge product
∧ρ , ρ-wedge product
SL(n, F), 12
[ , ], Lie bracket
trace (A), trace of the matrix A
Im C, pure imaginary complex numbers
Im H, pure imaginary quaternions
G, Lie algebra of the Lie group G
e, identity element in the group G
o(n), 14
so(n), 14
u(n), 14
su(n), 14
sp(n), 14
σ1 , σ2 , σ3 , Pauli spin matrices, 15
exp (A) = eA , 18
σ, right action
p · g, right action of g on p
σg , right action by g
ρ, left action, 21
g · p, left action of g on p
ρg , left action by g
P
G ,→ P −→ X, principal G-bundle over X
Vertp (P ), vertical vectors at p in P
A# , fundamental vector field
σp , left action on p
A, gauge potential
 Symbols 409

Horp (P ), horizontal vectors at p in P

V H , horizontal part of V , 30
V V , vertical part of V , 30
ω λ,n , BPST connection, 34
Aλ,n , BPST potential, 34
F , gauge field strength, 35
YM, Yang-Mills action, 38
P ×G F , associated bundle
[p, ξ], element of P ×G F
ad P , adjoint bundle
dω , covariant exterior derivative
A(ω, ϕ), action functional
∗
F , Hodge dual of F
R1,3 , Minkowski spacetime
ηαβ , Minkowski metric components
η, Minkowski matrix, 49
εijk , εαβγδ , Levi-Civita symbols
k th
Hde R (X), k de Rham cohomology group of X
L, Lorentz group
L+↑ , proper, orthochronous Lorentz group
Spin, the spinor map
̸ , Dirac operator
D
γ 0 , γ 1 , γ 2 , γ 3 , Dirac matrices
Λk (X), real-valued k-forms on X
vol , metric volume form
⟨α, β⟩, inner product on forms, 127
∥α∥2 = ⟨ α, α ⟩
SD, self-dual
ASD, anti-self-dual
N (A, ϕ), monopole number
suppf , support of f
p, map corresponding to the frame p, 148
e
L(X), linear frames on X
P
GL(n, R) ,→ L(X) −→ L
X, linear frame bundle
T (X), tangent bundle of X
T ∗ (X), cotangent bundle of X
Rk,n−k , 155
⟨ , ⟩k , inner product on Rk,n−k
410 Symbols

O(k, n − k), 156

SO(k, n − k), 156
o(k, n − k), 157
so(k, n − k), 157
F (X), orthonormal frames on X
P
O(k, n − k) ,→ F (X) −→
F
X, orthonormal frame bundle of X
PF+
SO(k, n − k) ,→ F+ (X) −→ X, oriented, orthonormal frame bundle of X
R, rotation subgroup of L↑+
P
L↑+ ,→ L(X) −→ L
X, oriented, time oriented, orthonormal frame bundle of X
E, Einstein-deSitter spacetime
D, deSitter spacetime
C, Einstein cylinder
T k (E), multilinear maps on E, 179
T ∗ , pullback
Λk (E), k-forms on E
Alt (A), 477
εj1 ···jn , Levi-Civita symbol
Λk (E, V), k-forms with values in V
A ⊗ρ B, ρ-tensor product
α ∧ρ β, ρ-wedge product
[α, β], 202
dk , exterior differentiation operator
dω , exterior derivative of ω
Λk (X, V), V-valued differential k-forms on X
Λkρ (P, V), tensorial forms of type ρ on P
dω φ, covariant exterior derivative of φ
α ∧˙ β, 227
vol(R), volume of the rectangle R
χm , characteristic function of M
dm, Lebesgue measure on Rn
a.e., almost everywhere, 238
supp ω, support of ω,240
Rn+ , 250
∂D, boundary of D, 250
Int D, interior of D, 250
Dn , unit disc in Rn , 250
f # , map induced in cohomology, 262
 Symbols 411

C ∗ , cochain complex
H k (C ∗ ), k th cohomology group of C ∗
Λ∗ (X), cochain of forms on X
QX , intersection form on X
deg(f ), degree of f
H(f ), Hopf invariant of f
c1 (P ), 1st Chern class of P
S k (V), complex-valued symmetric k-multilinear maps on V
S(V), direct sum of the S k (V)
P k (V), homogeneous polynomials of degree k on V
P (V), direct sum of the P k (V)
Sρk (V), ρ-invariant subspace of S k (V)
Pρk (V), ρ-invariant subspace of P k (V)
Sρ (V), direct sum of the Sρk (V)
Pρ (V), direct sum of the Pρk (V)
I k (G) = Sad
k
(G), 309
I(G), direct sum of the I k (G)
symtr, symmetrized trace
σ
P , 310
S0 , S1 , . . . , Sn , elementary symmetric polynomials
ck (P ), k th Chern class of P
| σ |, support of the simplex σ
N (U), nerve of U
Č j (U; Z2 ), Čech j-cochain group
δ j , coboundary operator
B̌ j (U; Z2 ), Čech j-coboundaries of U
Ž j (U; Z2 ), Čech j-cocycles of U
Ȟ j (U; Z2 ), j th Čech cohomology group of U
Ȟ j (X; Z2 ), j th Čech cohomology group of X
w1 (X), 1st Stiefel-Whitney class of X
w2 (X), 2nd Stiefel-Whitney class of X
γd (M), Donaldson invariants of M
 ×Ĝ Λ2+ (M , ad P ), vector bundle of Donaldson theory, 356
e(X), Euler class of the manifold X
Pf, Pfaffian, 357
e(E), Euler class of the vector bundle E
χ(E), Euler number of the vector bundle E
TQFT, topological quantum field theory
412 Symbols

SDW , Donaldson-Witten action

ZDW , Donaldson-Witten partition function
Cl(4), real Clifford algebra of R4
Cl× (4), multiplicative group of units in Cl(4)
Pin(4), 365
Spin(4), spin group of R4
Spin, spinor map
Cl(4) ⊗ C, complexified Clifford algebra of R4
∆C , complex spin representation, 370s
Spinc (4), complex spin group of R4 , 372
Spinc , 372, 378
∆ˆ C , 372
S(L), spinor bundle
L(L), determinant line bundle
L0 (L), U(1)-principal bundle of L(L)
Cl(B), Clifford bundle
Cl(B) ⊗ C, complexified Clifford bundle
∇A , covariant derivative associated with A
̸ A, D
D̃ ̸ A , Dirac operators
G(L), Seiberg-Witten gauge group
M(L), Seiberg-Witten moduli space
SW0 (B, L), 0-dimensional Seiberg-Witten invariant
DB (x), Donaldson series
Index

A C
action canonical isomorphism, 3
effective, 22 Cartan 1-form, 15
free, 22 Cartan Structure Equation, 35
left, 21 Čech cohomology group, 339, 343
right, 21 Čech j-coboundary, 339
transitive, 22 Čech j-cocycle, 339
action functional, 47 characteristic class, 303, 325
Donaldson-Witten, 360 chart, 1
gauge theory, 47 admissible, 1
spin 0-electrodynamics, 60 consistent with orientation, 7
SU (2) Yang-Mills-Higgs, 111 chiral representation, 82
Yang-Mills, 38, 48, 116 Chern-Weil homomorphism, 325
ad(G)-invariant, 20, 21 Chern number, 328
adjoint bundle, 40 Chern class, 325
total, 334
adjoint representation, 19
C ∞ -manifold, 1
admissible basis, 164
C ∞ -map, 2
algebraic homotopy, 219
C ∞ -related, 1
almost everywhere, 238
classical groups, 5
antiparticle, 64
Classification Theorem, 28
antipodal map, 296 Clifford algebra of R4 , 364
anti-self-dual (ASD), 115 center of, 365
Atiyah-Singer Index Theorem, 394 complexified, 370
atlas, 1 even elements of, 365
maximal, 1 odd elements of, 365
oriented, 7 real, 364
automorphism of bundles, 28 Clifford bundle, 378
complexified, 378
Clifford multiplication, 370
B closed form, 217
Betti numbers coboundary, 275
Bianchi Identity, 231 coboundary operator, 275
Bogomolny monopole cochain complex, 275
equations, 114 cochain homotopy, 277
boosts, 167 cochain map, 276
Borel sets, 235 cocycle, 275
BPST potential, 34, 115 cocycle condition, 26
center, 34, 115 cohomologous, 276
instanton, 115 cohomology class, 276
scale, 34, 115 commutator, 13
bump function, 141 complex line bundle, 39
bundle map, 27 complex scalar field, 42
bundle splicing, 99 conjugation representation, 88

413
414 Index

connection, 29 diffeomorphic, 1
existence of, 144 diffeomorphism, 1
Levi-Civita, 160 differentiable manifold, 1
linear, 160 differentiable structure, 1
metric, 161 differential form, 206
Riemannian, 160 vector valued, 211
coordinate expression, 1 dimension
coordinates on R1,3 , 173 of a differentiable manifold, 1
null, 173 of a submanifold, 2
spherical, 172 of a of a topological manifold, 1
cotangent bundle, 154 dimensional reduction, 104
cotangent space, 7 Dirac electron, 102
Coulomb field, 55 coupled, 102
coupling constant,47, 359 free, 99
covariant derivative, 64 Dirac equation, 83
covariant exterior derivative, 225 Dirac magnetic monopole, 37, 104
covariant tensors, 180 Dirac matrices, 82
rank, 180 Dirac quantization condition, 57
rank one, 8 Dirac spinor, 99
rank two, 8 direct sum of cochains, 280
covariant tensor field of rank domain with smooth boundary, 250
two, 9 boundary, 250
components, 9 exterior, 250
continuous, 9 interior, 250
smooth, 9 Dominated Convergence
covector, 7 Theorem, 239
critical value, 2 Donaldson invariants, 352
cross-section, 26 Donaldson µ-map, 354 355
global, 28 Donaldson series, 412
curvature, 34 duality, 293, 362
curvature equation, 381 Duistermaat-Heckman
theorem, 361
D
deformed, 255 E
degree, 294 Einstein cylinder, 177
de Rham, 258 Einstein-de Sitter spacetime, 173
coboundary, 258 electric charge, 53, 55, 56
cocycles, 258 electric field, 51, 55
cohomology group, 258 electromagnetic theory, 52
complex, 216 elementary symmetric
derivation, 6 polynomials, 310
derivative, 2 elevator experiment, 168
de Sitter spacetime, 175 elliptic complex, 393
determinant line bundle, 377 equation of structure, 35
 Index 415

equations of motion, 46, 47 future directed, 171

equivalent bundles, 27
equivariant localization G
theorems, 361 gauge equivalent, 30
equivariant map, 41 gauge field, 29
Euler characteristic, 38, 170, 294 gauge invariance, 67
Euler class, 357, 358 gauge potential, 29
Euler number, 358 gauge principle, 66
Euler-Lagrange equations, 47, 63, gauge theories, 45
115 Gauss-Bonnet-Chern Theorem, 357
events, 49, 163, 168 G-bundle, 23
exact form, 54 trivial, 24
exterior derivative, 213 generic perturbations theorem, 395
covariant, 225 Georgi-Glashow potential, 107
of real-valued function, 7 ground state, 113, 135
of real-valued 1-form, 10
exterior differentiation, 213
H
exterior k-bundle
hedgehog solution, 123
Higgs field, 42, 106
F homotopy classes of, 133
fiber bundle, 39 Hodge dual, 110, 191
finite action, 47 Hodge star operator, 191
finite type, 162 holonomy group, 31
1st Chern class, 304 homogeneous polynomial, 307
1st Chern number Hopf bundle, 25
1st Stiefel-Whitney class, 343 complex, 25
flat connection, 31, 35 quaternionic, 26
forms, 205, 206 Hopf invariant, 299
vector-valued horizontal form, 222
frame, 46, 148 horizontal lift, 228
linear, 149 horizontal subspace, 30
moving, 153
oriented, time oriented, I
orthonormal 171 imbedding, 2
orthonormal, 157, 171 immersion, 2
frame bundle immersion at p, 2
linear, 149 index of a semi-Riemannian
oriented, orthonormal 159, 171 metric, 156
oriented, time oriented, inner product, 9
orthonormal instanton, 34, 37
orthonormal, 157, 159, 171 center, 115
frame field, 153 number, 334
oriented, orthonormal, 157 scale, 115
fundamental vector field, 29 integral curve, 6
416 Index

internal space, 45 local gauge, 29

intersection form, 293, 351 Local Triviality, 23
isomorphic Lie algebras, 13 local trivialization, 24
isomorphic Lie groups, 12 locally finite, 139
isotropy subgroup, 22 Lorentz group, 70
general, 70, 155
J proper, 155, 164
Jacobi identity, 12 orthochronous, 164
j-cochain, 337 Lorentz invariance, 68
j-simplex, 337 of the Dirac equation, 85
face, 337 Lorentz metric, 169
support, 337 Lorentz transformation, 166

M
K
magnetic charge, 37, 334
k-dimensional submanifold, 2 magnetic field, 52
k-tensor magnetic monopole, see 104
skew-symmetric, 180 Dirac magnetic monopole
symmetric, 180 map induced in cohomology, 262
vector-valued, matrix Lie group, 12
Killing form, 20 matter field, 42
Klein-Gordon equation, 63 Maurer-Cartan equations, 16
KM-basic classes, 399 Maxwell’s equations, 53
KM-coefficients, 399 Mayer-Vietoris sequence, 281
KM-simple type, 399 measure zero, 235
measurable form, 243
L measurable function, 237
Laplacian, 216 measurable set, 236
Lebesgue integrable, 238 metric, 20
form, 243 bi-invariant, 20
Lebesgue integral, 237 left invariant, 20
Leaning Tower of Pisa, 168 Lorentz, 169
Leibnitz Product Rule, 2 Riemannian, 9
lens space, 60 right invariant, 20
Lie algebra, 12 semi-Riemannian, 9
ideal in, 20 metric volume form, 190, 209
of a Lie group, 14 Minkowski inner product, 164
Lie bracket, 6 Minkowski space, 155
trivial, 13 Minkowski spacetime, 155
Lie group, 11 moduli space, 351
matrix, 12 Donaldson theory, 352
semisimple, 20 Seiberg-Witten theory, 391
light cone, 164 monopole number, 130
lightlike, 164 Monotone Convergence
local field strength, 35, 46 Theorem, 239
 Index 417

moving frame, 153 P

multilinear map, 179 parallel translation, 31
vector-valued, 199 parallelizable, 153
partition of unity, 140
partition function, 359
N past directed, 171
natural connection, 31 Pauli spin matrices, 15, 81
complex Hopf bundle, 31 Poincaré Duality, 293
quaternionic Hopf bundle, 33 Poincaré-Hopf Theorem, 357
natural differentiable structures on Poincaré lemma, 219
real vector spaces, 3 polarization, 167, 308
nerve, 337 principal bundle, 23
nonstandard differentiable structure restriction of, 26
on R, 3 trivial, 28
normalized generator, 294 product manifold, 5
nucleon, 104 projective spaces, 4
nucleon field, 104 pseudoparticles, 34
neutron component, 104 pseudotensorial forms, 222
proton component, 104 pullback, 8, 9, 10, 180, 199
null, 49, 165
null cone, 164 Q
quantum field theory, 359
O observables in, 359
observable, 359 partition function of, 359
expectation value of, 359 topological, 359
1-form, 7 quaternionic line bundle, 39
left-invariant, 7
real-valued, 7 R
restriction, 8 rank
orbit, 22 Reconstruction Theorem, 28
orientable manifold, 7 refinement, 139
oriented atlas, 7 regular value, 2
oriented manifold, 7 Removable Singularities Theorem,
orientation, 6 116, 118
chart consistent with, 7 representation, 19
opposite, 7 ρ tensor product, 201
orientation form, 208 ρ-wedge product, 201
orientation preserving, 7 rotation subgroup of L+↑ , 166
orientation reversing, 7
oriented orthonormal basis, 159 S
orthochronous, 164 Schroedinger equation, 64, 65
orthogonal group, 5 2nd Chern class, 326
outward pointing vector, 251 2nd Chern number, 117 328,
overlap functions, 1 2nd Stiefel-Whitney class, 346
418 Index

Seiberg-Witten spinc -structure, 376

apriori bounds on solutions, 396 spin 0, 60, 63
basic classes, 398 spinor bundle, 377
compactness of the moduli spinor field, 99
space 397, spinor map, 367
configuration space, 385 spinor representations, 87
elliptic complex, 393 spinor structure, 97, 171
equations, 394 spontaneous symmetry
gauge group, 386 breaking, 113
invariant, 397 Standard differentiable structure, 3
irreducible configurations, 391 on Rn , 3
map, 392 on S n , 4
moduli space, 391 star-shaped, 219
monopoles, 385 step function, 237
perturbed equations, 394 stereographic projection, 4
reducible configurations, 391 Stiefel-Whitney class, 343, 346
simple type, 398 Stokes’ Theorem, 257
self-dual (SD), 115 structure constants, 14
semi-orthogonal group, 155 structure group, 23
short exact sequence, 277 reduction of, 136
σ-algebra, 235 submanifold, 2
sign of f , 297 open, 2
simple cover, 162 0-dimensional, 2
smooth curve, 2 submersion, 2
smooth deformation submersion at p, 2
retraction, 271 subordinate, 141
smooth homotopy, 147 support, 140
smooth homotopy type, 270 symmetric, 9, 180, 200, 293, 306, 310
smooth manifold, 1 symmetric polynomials, 310
smooth map, 2 Fundamental Theorem, 311
smooth retraction, 271 symmetrized trace, 309
smoothly contractible, 271 symplectic group, 5
smoothly homotopic, 263
smoothly nullhomotopic, 264
spacelike, 49, 169 T
spatial inversion, 93 tangent bundle, 153, 154
spacetime manifold, 169 tangent space, 2
special linear groups, 12 tangent vector, 2
special orthogonal group, 5 tensor bundle, 154
special semi-orthogonal tensor field, 154
group, 155 tensor product, 180, 200, 201
special unitary group, 5 tensorial forms, 222
spin-j representation, 75 t’Hooft-Polyakov-
Spin one-half, 71, 73 Prasad-Sommerfleld
spin structure, 367 monopole, 121
 Index 419

time orientable, 170 vertical subspace, 28

time oriented, 170 vertical vector, 28
timelike, 49, 165 volume, 209
topological charge, 38, 47, 117 volume form, 209
topological manifold, 1 vortices, 112
topological quantum field
theory, 359
W
transition functions, 26
wedge product, 9, 10, 181, 201
trivialization, 24
Weitzenböck formula, 396
global, 28
Weyl neutrino equation, 84
trivializing cover, 24
Weyl representation, 82
two-component wavefunction, 73
Weyl spinor, 99
2-valued representation, 74, 77
Witten’s conjecture, 400
Witten’s TQFT, 361
U worldline, 164
Uhlenbeck compactification, 355 of a free material particle, 164
unitary group, 5 of a photon, 164

V
Y
vacuum state, 113
Yang-Mills action, 38, 48, 115
vector analysis, 216
Yang-Mills equations, 48, 115
vector bundle, 39
Yang-Mills theory, 48, 49, 115
vector field, 5
Yang-Mills-Higgs action, 111
components of, 6
gauge invariance, 112
continuous, 6
Yang-Mills-Higgs equations, 114
horizontal lifts, 228
Yang-Mills-Higgs monopoles, 112
left-invariant, 13
smooth, 6
vector-valued forms, 10, 198 Z
velocity vector, 2 zero section, 272, 356