Notes on Gauge Theories

Leonardo Almeida Lessa

Abril de 2019

Contents
1 Motivating Examples 2
1.1 Classical Electromagnetism . . . . . . . . . . . . . . . . . . . 2
1.2 Quantum Electrodynamics . . . . . . . . . . . . . . . . . . . . 5
1.3 Dirac Monopole . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.1 Magnetic Charges . . . . . . . . . . . . . . . . . . . . 6
1.3.2 Electromagnetic Interaction in Quantum Mechanics . 8
1.3.3 QM and the Dirac Monopole . . . . . . . . . . . . . . 9

2 Fibre Bundles 11
2.1 Motivation: Tangent Bundle . . . . . . . . . . . . . . . . . . . 11
2.2 Fibre Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Triviality of Bundles . . . . . . . . . . . . . . . . . . . 16
2.3 Principal Bundles . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.1 Associated Vector Bundle . . . . . . . . . . . . . . . . 18

3 Connections in Fibre Bundles 20

3.1 Fundamental Vector Field . . . . . . . . . . . . . . . . . . . . 20
3.2 Connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Parallel Transport . . . . . . . . . . . . . . . . . . . . . . . . 22
3.4 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.5 Local Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.5.1 Gauge Potential . . . . . . . . . . . . . . . . . . . . . 24
3.5.2 Field Strength . . . . . . . . . . . . . . . . . . . . . . 27
3.6 Covariant Derivative . . . . . . . . . . . . . . . . . . . . . . . 28

4 The Examples from Another View 30

4.1 QED and Yang-Mills Gauge Theories . . . . . . . . . . . . . . 30
4.2 Dirac Monopole . . . . . . . . . . . . . . . . . . . . . . . . . . 31

Here are some notes on gauge theories. We begin by discussing some

examples from Physics and their common features. Then, we detour to the
theory of fibre bundles and connections on principal bundles, which form
the mathematical basis of gauge theory. Finally, we review our previous
examples in this new framework.

1
 Leonardo A. Lessa

I personally find fascinating that many phenomena that appear uncorre-

lated at first can be explained in an unified way. In the case of gauge theory,
we see another intersection of Geometry and Topology with Physics.
The text is heavily based on “Geometry Topology and Physics”, by Mikio
Nakahara [1]. Other references are included at the end.

1 Motivating Examples
The notion of gauge appears when we have more degrees of freedom
in the description of the theory than there is in the physics of the problem.
Sometimes, the underlying mathematical objects, confined to our abstraction
of the world, are redundant and unphysical, even though derived observable
results are unambiguous.
We artificially distinguish quantities that are physically equivalent. Thus,
the observables of our theory have to be symmetric upon gauge transforma-
tions in these equivalent quantities. They possess gauge symmetry described
by gauge groups, similarly to the familiar symmetries of spacetime, but qual-
itatively different. A gauge transformation changes our description of the
modeled system while a spacetime symmetry changes the modeled system
itself, although not modifying the physical laws.
Before presenting the mathematical details, we will now discuss some
examples. Our goal now is to create a clear image of what we mean by a
gauge theory, to then formalize it in more generality.

1.1 Classical Electromagnetism

The first contact one may have with gauge transformations is with Elec-
tromagnetism, whose dynamics are determined by Maxwell’s equations. In
natural (Lorentz-Heaviside) units, by which we set c = 0 = µ0 = ~ = 1,
Maxwell’s equations can be written as

∂B
∇×E=− , ∇ · B = 0, (1.1)
∂t
∂E
∇ · E = ρ, ∇×B= + J. (1.2)
∂t
Poincaré’s Lemma applied to the homogeneous equations (1.1) tells us
there exists at least locally1 a scalar field V and a vector field A, also known
as potentials, satisfying (1.1)
Exercise 1.1
In R3 vector calculus, Poincaré’s
Lemma amounts to ∂A
E=− − ∇V, (1.3)
∂t
∇ × v = 0 ⇒ v = ∇f,
B = ∇ × A. (1.4)
∇ · v = 0 ⇒ v = ∇ × u,
for some locally defined real Although we are guaranteed of their existence, the potentials V and A
function f and vector field u.
are far from unique. For every arbitrary function Λ, we may transform the
Adapt this to prove equations
(1.3) and (1.4). 1 We will see later that the Dirac monopole is an example where the vector field A

cannot be defined in the whole space. This has to do with the topology of the problem
and has interesting consequences.

2
1.1 Classical Electromagnetism Leonardo A. Lessa

potentials by
∂Λ
V0 =V − , (1.5)
∂t
0
A = A + ∇Λ, (1.6)

but still get the same electric E0 = E and magnetic fields B0 = B(1.2) . Since
Exercise 1.2
E and B control the dynamics of charged particles via Lorentz force, we Verify this
arrive at the same physics after this transformation. Indeed, this is a case
of gauge transformation.
In the manifestly covariant formulation of Electromagnetism [2], the po-
tentials V and A are part of a (four-)potential

Aµ = (V, A), (1.7)

as well as the four-current J µ = (ρ, J), which transform like a four-vector

under Lorentz transformations. Similarly, the electric and magnetic fields
can be grouped into a antisymmetric (0, 2)-tensor Fµν , called the Faraday
tensor or electromagnetic field strength tensor. The components of Fµν in
a reference frame where the electric and magnetic fields at some point are
E = (Ex , Ey , Ez ) and B = (Bx , By , Bz ) are, in matrix notation,
 
0 Ex Ey Ez
−Ex 0 −Bz By 
Fµν = 
−Ey Bz
. (1.8)
0 −Bx 
−Ez −By Bx 0

We immediately see that the transformation laws of the electric and magnetic
fields under change of reference frame are not as simple as with the four-
potential Aµ , since they are mixed up in the coordinates of the Faraday
tensor Fµν .
Equations (1.3) and (1.4) can be written in terms of the four-potential
(1.7) and the Faraday tensor (1.8) via

Fµν = ∂µ Aν − ∂ν Aµ , (1.9)

where we have used Einstein notation to sum repeated indices and lowered
the potential with the Minkowski metric ηµν = diag(−1, +1, +1, +1) as such:

Aµ = ηµν Aν = (−V, A).

Furthermore, the gauge transformation of (1.6) and (1.5) simplifies to

A0µ = Aµ + ∂µ Λ. (1.10)

Since the electric and magnetic fields remain the same under a gauge
transformation, the same happens with Fµν . Thus, the Faraday tensor is
gauge invariant. Accordingly, we can write Maxwell’s equations in terms of
Fµν and they are also gauge invariant:

∂µ (∗F µν ) = 0, (1.11)
µν ν
∂µ F =J , (1.12)

3
 1.1 Classical Electromagnetism Leonardo A. Lessa

where ∗F is the dual field strength tensor, defined by ∗F µν = 12 µναβ Fαβ (1.3) .
Exercise 1.3
Interpret equation (1.11) know- We can solve equations for V and A and work with them, but the final
ing that the dual field strength observable results have to be gauge independent. This is not a negative
∗F µν is F µν with E and B
swapped. aspect of our theory, quite the contrary. The labor of dealing with non-
gauge-invariant quantities like Aµ is compensated by the manifest Lorentz
invariance of the theory. Unitarity and locality are also manifest when we
keep this redundancy in the quantum realm (See [8]).
Even fixing the gauge is a useful idea. It is often the case that simplifi-
cation in the calculation occurs when we choose a particular gauge. In the
case of Electromagnetism, we can set up a differential equation for Λ so as
the potentials have some property we want, as long as it doesn’t contradict
Maxwell’s equations.

Figure 1: Illustration of the space of all potentials Aµ . A point in this gauge

is connected to other points (dashed lines) by gauge transformations, and
to fix a gauge is to choose independent representatives among those (bold
line).

For example, we may choose the Lorenz Gauge ∂µ Aµ = 0(1.4) to arrive

Exercise 1.4
What differential equation does at a wave equation for the potential,
Λ have to satisfy?
(1.12)
Jν = ∂µ F µν ,
= ∂µ (∂ µ Aν − ∂ ν Aµ ),
= ∂µ ∂ µ Aν . (1.13)

Although the potential Aµ has four components, by solving (1.13), we

get two independent solutions, corresponding to the possible polarizations
of a electromagnetic wave. This is also a manifestation of gauge symmetry.

4
1.2 Quantum Electrodynamics Leonardo A. Lessa

1.2 Quantum Electrodynamics

Having already seen how classical electromagnetism has gauge symmetry,
it may not be clear what is the corresponding gauge group. In Quantum Elec-
trodynamics (QED), however, the gauge group appears naturally. In fact,
we won’t even quantize anything, (fortunately!). The only “new” concept
that comes from the quantum world is spinors. If the reader is unfamiliar
with this formalism, we recommend [6].
The main idea is to express our previous results in the Lagrangian for-
malism. The potential Aµ (x) is our dynamical field, whose dynamics is
controlled by the following Lagrangian
1
LEM = − Fµν F µν − Aµ J µ . (1.14)
4
More precisely, by the Euler-Lagrange equations associated to LEM (1.5)
Exercise 1.5
∂L ∂ ∂L Why is (1.11) automatically sat-
− = 0. (1.15) isfied?
∂Aµ ∂xν ∂(∂ν Aµ )
One may argue that LEM is not gauge invariant, R because of the term
Aµ J µ , and it is true. However, the action SEM = LEM d4 x is gauge in-
variant, because the gauge transformation of SEM gives off a surface term,
which goes to zero as we integrate over the entire spacetime M = R4 :
Z Z
− Aµ J µ d4 x → − (Aµ + ∂µ Λ)J µ d4 x,
M
ZM Z
=− Aµ J µ d4 x − [∂µ (ΛJ µ ) − Λ∂µ J µ ]d4 x,
M M
Z Z
µ µ 4
=− A J d x− ΛJ µ d3 x,
M ∂M
Z
=− Aµ J µ d4 x,
M

where in the third line we used that J µ is conserved and in the fourth, that
the integrand goes to zero at infinity. Since the Euler-Lagrange equations
(1.15) come from the variation of the action, then we still have a gauge
invariant theory.
In QED, the fermion field of electrons and their antiparticles, the positrons,
is coupled with the electromagnetic field of photons, the force carriers, in a
quite natural way. First, we begin by writing the free Lagrangian for a Dirac
spinor ψ,
LDirac = ψ(iγ µ ∂µ − m)ψ. (1.16)
It describes a fermion (spin 12 particle) with mass m by a Dirac spinor
field ψ (1.6) , which has four complex components ψα , α ∈ {0, 1, 2, 3}, and
Exercise 1.6
transforms under an irreducible representation of the Lorentz group2 . Electrons are charged particles.
The Lagrangian LDirac has a global internal symmetry, Then why does (1.16) not have
the elementary charge constant
ψ(x) → eieλ ψ(x), ψ(x) → ψ(x)e−ieλ , (1.17) e in it?
2 Technically, ψ transforms under a projective representation of the Lorentz group,

which is a proper representation of the double cover group SL(2, C). See [3, 5].

5
 1.3 Dirac Monopole Leonardo A. Lessa

where global means λ ∈ R is independent of the spacetime point x and

internal means the transformation doesn’t change the point x where the
fields are evaluated. Since λ is arbitrary and ψ is multiplied by a pure phase
eieλ , our gauge group is U(1), complex numbers of modulus one endowed
with complex multiplication.
Let’s now see what happens if we require this symmetry to be local. With
that, the gauge transformation (1.17) becomes
ψ(x) → e−ieλ(x) ψ(x), ψ(x) → ψ(x)eieλ(x) , (1.18)
This comes at a price, since LDirac is no longer gauge invariant, but trans-
forms as
LDirac = ψ(iγ µ ∂µ − m)ψ → ψ[iγ µ (∂µ − ie∂µ λ) − m]ψ,
which looks similar to the gauge transformation (1.10) for the electromag-
netic potential Aµ . In fact, if we define a covariant derivative ∇µ by
∇µ = ∂µ + ieAµ (1.19)
then we can modify our Lagrangian as
L0Dirac = ψ(iγ µ ∇µ − m)ψ. (1.20)
so we have a theory that is invariant under the local gauge transformations
ψ(x) → e−ieλ(x) ψ(x), ψ(x) → ψ(x)eieλ(x) ,
Aµ → Aµ + ∂µ λ,
and couples the electromagnetic field Aµ to the fermionic field ψ. Finally,
we add the electromagnetic Lagrangian term − 14 Fµν F µν to (1.20) to get our
final QED Lagrangian
1
LQED = − Fµν F µν + ψ(iγ µ ∇µ − m)ψ (1.21)
4
In view of LEM from (1.14), we may also rearrange LQED as
LQED = LDirac + LEM ,
1
= ψ(iγ µ ∂µ − m)ψ − Fµν F µν − Aµ J µ ,
4
with J µ = eψγ µ ψ.(1.7)
Exercise 1.7
Prove that the conserved We did not justify why is it necessary that we have local gauge invariance
Noether current associated with (1.18) instead of the weaker assumption of global gauge invariance (1.17).
the global symmetry (1.17) of One a posteriori reason is that, from this assumption, the electromagnetic
the Lagrangian LDirac is J µ .
field and its coupling to the Dirac field follow naturally. A more thorough
discussion about the need for local gauge invariance can be found in [7].

1.3 Dirac Monopole

1.3.1 Magnetic Charges
The interplay between topology, geometry and gauge theory will become
apparent once we introduce the notion of fibre bundles. Before that, the

6
1.3 Dirac Monopole Leonardo A. Lessa

example of the Dirac magnetic monopole illustrates the far-reaching conse-

quences of the geometry of spacetime.
µ
We include a magnetic four-current JB = (ρB , JB ) in Maxwell’s homo-
geneous equation (1.11) in the same way the electric four-current J µ ≡ JEµ
appears in (1.12):

∂µ (∗F µν ) = JB
ν
,
∂µ F µν = JEν .

In terms of E and B fields,

∂B
∇×E=− − JB , ∇ · B = ρB ,
∂t
∂E
∇ · E = ρE , ∇×B= + JE .
∂t
Consider a point magnetic charge of strength g in the origin. Its magnetic
charge density is ρB (r) = gδ 3 (r), so, similarly to the electrostatic case, the
magnetic field produced is
g r̂
B= . (1.22)
4π r2
We used the then homogeneous Maxwell’s equations (1.1) to derive the
potentials V and A satisfying (1.3) and (1.4). Since they are not homo-
geneous if magnetic charges are present, we cannot have globally defined
potentials as before. If we could, then B = ∇ × A would imply ∇ · B = 0
everywhere.
However, in the case of a point magnetic charge, there is a vector potential
AN whose curl almost equals the magnetic field (1.22). By “almost”, we mean
∇ × AN (r) = B(r) for r ∈ R3 \ S, for a “small set” S ⊂ R3 . In spherical
coordinates (r, θ, φ), AN is given by

g 1 − cos θ
AN (r) = êφ , (1.23)
4π r sin θ
which is well-defined for r ∈ R3 \ S, with S = {(x, y, z) ∈ R3 |z ≤ 0, x = y =
0}, a region called the Dirac string. Not only is AN singular at the origin,
where the monopole sits, but also in a line starting on the origin and going
to infinity in the −z direction.(1.8) For r ∈ R3 \ S,
Exercise 1.8
Why is AN not well-defined on
1 ∂ 1 ∂
∇ × AN (r) = (AN sin θ)r̂ − (rAN
φ )θ̂,
S?
r sin θ ∂θ φ r ∂r
g r̂
= ,
4π r2
= B(r).

Can we also cover the S region? We know we cannot do this with just
one vector potential. However, we can imitate AN by defining AS on R3 \S 0 ,
with S 0 = −S = {(x, y, z) ∈ R3 |z ≥ 0, x = y = 0},

g 1 + cos θ
AS (r) = − êφ , (1.24)
4π r sin θ

7
1.3 Dirac Monopole Leonardo A. Lessa

Figure 2: Illustration of the magnetic field B produced by a magnetic

monopole g. The ray S is called the Dirac String, where the vector po-
tential AN is not defined. I thank Gabriel Solis for helping me render this
scene.

You may have noticed that the superscripts N and S mean north and south
hemispheres, where the vector potentials AN and AS are well-defined.
Let us pause for a moment to ponder the geometric meaning of AN and
S
A . Outside the origin there is no magnetic charge, thus we can use the usual
Maxwell’s equations (1.1) and (1.2) to find find our solutions. Indeed, for
r 6= 0, we can find vector potentials for the magnetic field. But because we
are now solving Maxwell’s equations in R3 \0, a non simply connect subset of
R3 , Poincaré’s lemma does not guarantee a globally defined vector potential,
and so we need at least two vector fields to cover all R3 \ 03 . Effectively,
the magnetic monopole is altering our space topology. This close connection
with the geometry of the underlying space M will be made precise when we
study connections (Section 3) and the cohomology group H 2 (M ) (Section
4.2).

1.3.2 Electromagnetic Interaction in Quantum Mechanics

When we join Quantum Mechanics with the Dirac monopole, we get a
surprising result: the quantization of electrical charge. Before doing
so, let’s see how electromagnetism enters in Quantum Mechanics.
3 We will see this is closely related to the fact that we need at least to coordinate

patches to cover S 2 .

8
1.3 Dirac Monopole Leonardo A. Lessa

Without electromagnetism, Schrödinger’s equation for a particle of wave

function ψ(t, r) in a potential V (r) is (~ = 1)

∂ψ
Hψ = i , with
∂t
2
|p|
H= + V (r).
2m
The quantum-mechanical momentum is p = −i∇, a space derivative.
Imitating what we did for the electromagnetic coupling of the Dirac field, we
will use equation (1.19) of the covariant derivative to redefine our momentum
as
p → p + qA.
This redefinition of the momentum to account for electromagnetic interaction
is called minimal coupling. It couples a charged particle with charge q (with
no spin) to the electromagnetic field.
Although A is not gauge invariant, this coupling gives gauge invariant
physical results! To see this, let us pick a wave function ψ(t, r) which is a
solution to the Schrödinger’s equation in a region with vector potential A.
Its Hamiltonian is
1
H= (p + qA)2 + V (r). (1.25)
2m
If we do a gauge transformation A → A + ∇Λ, the Hamiltonian H changes
and so does its eigenfunction ψ(t, r) → ψ̃(t, r), which can be expressed in
terms of the old solution ψ by(1.9)
Exercise 1.9
Verify this by calculating the ef-
ψ̃(t, r) := e−ieΛ(r) ψ(t, r). (1.26) fect of the operator (p + eA +
e∇Λ) acting on ψ̃.
We know from Quantum Mechanics that global complex phases multiply-
ing the wave function are not physical, since all measurable quantities come
from taking the squared of the absolute value of inner products. With this
in mind, we might think the complex phase factor in (1.26) can be removed
without repercussions. Quite the contrary, this complex factor is responsible
for a plethora of counterintuive effects. They all have in common the ap-
pearance of a gauge invariant measurable quantity in terms of a not gauge
invariant quantity, like A. One of these effects is what we will discuss now:
when we interact a quantum particle with a magnetic monopole. The other
is the Aharanov-Bohm effect, by which we can detect the changes in phase
of a wave function in a B = 0 (but A 6= 0) environment by interferometry.

1.3.3 QM and the Dirac Monopole

We will keep the complex phase of (1.26) and see where it leads us to.
Remember that the electromagnetic field around a magnetic charge of
strength g can be described by two vector potentials, AN and AS . The
domains of definition of the vector potentials AN and AS intersect each
other in R3 \ (S ∪ S 0 ), which is the whole space minus the z line. In this
region of intersection, both vector potentials have to give the same electro-
magnetic fields. One way to achieve this is if they are equal up to a gauge

9
1.3 Dirac Monopole Leonardo A. Lessa

transformation. To see if this is the case, we calculate the difference

g 1
AN − AS = êφ ,
2π r sin θ
= ∇(gφ/2π).

Mixed feelings. The expression gφ/2π is not a function, for a 2π rotation

in the angle φ doesn’t change a point in space, but changes the expression.
On the other side, it is painful to not be able to say that AN and AS
are connected by a gauge transformation. In either way, will treat for now
Λ = gφ/2π as our gauge transformation between AS and AN , postponing a
more rigorous discussion.4
Here comes the main argument. If we have a quantum particle in a
space with a magnetic monopole at the origin, then its wave function ψ in
the equator5 may be the solution of the Hamiltonian with minimal coupling
(1.25) and A = AS .
Applying a gauge transformation Λ = gφ/2π will turn AS into AN and
ψ into qg
ψ̃(r) = e−ieΛ ψ = e−i 2π φ ψ(r). (1.27)
Here, we demand ψ(r) is a well-defined wave function, and so is single-valued.
This is equivalent to demanding that the complex phase from (1.27) satisfy
eg qg
e−i 2π φ = e−i 2π (φ+2π) ,

which is equivalent to
qg
= n ∈ Z. (1.28)
2π
In words, If there exists a magnetic monopole with strength g,
then electrical charges are quantized. Moreover, magnetic charges also
come in discrete amounts by equation (1.28).
From the Standard Model of Particle Physics, we know every charged
body has a charge which is a integer multiple of 31 e, where e is the elementary
charge. Most particles (elementary or not) have charges that are integer
multiples of e, such as electrons, protons, photons. The 13 includes quarks,
which come in a family with charge + 23 e – up, charm and top – and a family
with charge − 13 e – down, strange and bottom.

4 If the reader is unsettled, we anticipate the solution. We will treat the vector potential
g
as a one-form, so this gauge transformation is really a one-form 2π dφ, which is well-
defined.
5 We look for wave functions in the equator to stay away from the Dirac strings S and

S 0 , so both AS and AN are well-defined

10
 Leonardo A. Lessa

2 Fibre Bundles
We now introduce the concept of fibre bundle, the backbone of all the
mathematical formalism for gauge theory. Fibre bundles have a special inter-
est in itself as a mathematical construct and its subtopics include principal
bundles, holonomy, etc, many of which we will treat here. The main refer-
ence for the theory of principal bundles and connections is Kobayashi and
Nomizu’s “Foundations of Diferential Geometry” [4]. For the introduction of
fibre bundles, we will use Steenrod’s “The Topology of Fibre Bundles” [9].
For our first example of fibre bundle, the tangent bundle, we assume
the reader has some familiarity with the theory of manifolds. Although
this example is independent of the subsequent sections, it already has many
features we want to explore in the general theory.

2.1 Motivation: Tangent Bundle

One very useful prototypical example of a fibre bundle is the tangent
bundle T M of a manifold M of dimension dim(M ) = n. It is the disjoint
union of the tangent spaces Tp M , p ∈ M , as such
G [
T M := Tp M ≡ (p, Tp M ).
p∈M p∈M

3
If M is a two-dimensional surface of R , one can imagine Tp M is a plane
tangent to a point p ∈ M of the surface, and T M is the set of all those
planes, separated by which points they are tangent to.
The tangent bundle T M is itself a manifold of dimension 2n, called the
total space if viewed as a fibre bundle. The charts of T M are described by
a chart of M , the base space, and an open set of the vector space Tp M .
This arrangement may seem like T M is the product space M × V , with V an
n-dimensional vector space, isomorphic to Tp M , ∀p ∈ M , but this is not the
case for every manifold M . In fact, if T M ' M × V (is trivial), then M is a
parallelizable manifold. For the spheres S m , only the m = 1, 3 and 7 ones are
parallelizable. There is a entire area of study dedicated to the non-triviality,
or twisting, of bundles, called Chern-Weil theory of Characteristic Classes
[1](2.1)
Exercise 2.1
The tangent bundle T M is not necessarily a direct product, but it is Do these numbers remind you of
locally a direct product or, in other words, locally trivial. More precisely, for something?
each chart (U, xi ) of M , we can consider the restricted bundle T U , treating
the coordinate neighbourhood U as a manifold. From the definition of
T M , we know that in the local coordinates x = (x1 , . . . , xn ) : U → Rn of U ,
every element u = (p, V ) ∈ T U can be decomposed in its base point p ∈ M
and a vector V ∈ Tp U , which can be written as

µ ∂ 

V =V ,
∂xµ p

for (V 1 , . . . , V n ) ∈ Rn , since { ∂x∂α p }nα=1 is a base for Tp U . Naturally, we
can define a diffeomorphism φU : T U → U × Rn by
φU (u) = (p, (V 1 , . . . , V n )),

11
2.1 Motivation: Tangent Bundle Leonardo A. Lessa

Figure 3: Illustration of the tangent bundle T M . A vector v ∈ Tp M in the

fiber Tp M is projected to p ∈ M by π.

which is precisely what we meant by saying T M is locally a direct product

space. We will call φU the local trivialization map.
Another common feature of fibre bundles is the projection to the base
space. In the case of T M , we can define π : T M → M simply by π(u) = p,
where u = (p, V ) ∈ T M . Clearly, T U = π −1 (U ) and Tp M = π −1 (p), the
latter called the fibre of the bundle. All fibres are isomorphic to one another,
but are associated to different points in the base manifold M .
Given another open chart (V, y i ) of M , one may ask how the local trivi-
alizations φU : π −1 (U ) → U × Rn and φV : π −1 (V ) → V × Rn are connected,
just as the coordinates xµ : U → R and y ν : V → R are related by
∂(y ◦ x−1 )ν 

ν
y (p) = xµ (p),
∂xµ 
x(p)

6
for p ∈ U ∩ V . Likewise, the vector components of V ∈ Tp M satisfy

µ ∂ 

V = Vx µ
,
∂x p
∂y ν ∂ 

= Vxµ µ ,
∂x ∂y µ p

ν ∂ 

= Vy ν
,
∂y p
∂y ν µ
so Vyν = ∂xµ Vx and thus the local trivializations φU and φV are related by
ν
φ−1 µ −1 ν −1 ∂y µ
U (p, Vx ) = φV (p, Vy ) = φV (p, ∂xµ Vx ).

6 We ∂y ν µ
will often shorten the notation and write y ν = ∂xµ
x when the point p ∈ U ∩V ⊆
M is implicit from the context.

12
2.2 Fibre Bundles Leonardo A. Lessa

ν
∂y
The matrix Gν µ = ∂x µ is nonsingular (det(G) 6= 0) because we demand
−1
that y ◦ x be a diffeomorphism for M to be smooth. It is this matrix that
mixes the components of the vector V when we change coordinates. The set
of all real nonsingular matrices of order n is called the general linear group
GL(n, R).(2.2) Thus, we say the structure group of T M is GL(n, R) and
Exercise 2.2
Gν µ is a transition function between charts. Given a coordinate chart (U, xi )
The bold terms above are the essential components of a fibre bundle, and a matrix G ∈ GL(n, R),
which are beautifully exemplified by the tangent bundle T M . Other cor- can we create another coordinate
chart (U, y j ) such that the tran-
related concepts will appear throughout this presentation, such as cross
sition function from x to y is
section and connection, generalizations of vector field and parallel trans- given by G? This type of ques-
portation, respectively. tion is related to the problem of
reducing the structure group of a
principal bundle (See Sec. 2.3).
2.2 Fibre Bundles
With the elements highlighted in Section 2.1 for T M , we enunciate the
definition of a coordinate bundle, and then of a fibre bundle, following Steen-
rod [9].
Definition 2.1. A coordinate bundle (E, π, M, F, G, {Ui }, {φi }) consists of
(i) A topological space E called the total space or bundle space,
(ii) a topological space M called the base space,
(iii) a topological space F called the fibre,
(iv) a continuous surjection π : E → M called the projection,
(v) a topological group G called the structure group, which acts freely
on F on the left,
(vi) a open covering {Ui }i∈I of M indexed by I, consisting of open neigh-
bourhoods Ui ,
(vii) a homeomorphism φi : Ui × F → π −1 (Ui )7 for each i ∈ I, called a
local trivialization.
satisfying the following relations:
(i’) For each p ∈ M , the inverse image π −1 (p) =: Fp is homeomorphic to
the fibre F .
(ii’) Each local trivialization φi : Ui × F → π −1 (Ui ) is constrained by the
projection as such: π ◦ φi (p, f ) = p ∈ Ui .
(iii’) If we define φi,p : F → Fp as φi,p (f ) = φi (p, f ) for p ∈ Ui ∩ Uj , then
we require that the transition function tij (p) := φ−1 i,p ◦ φj,p : F → F
coincides with the operation of an element of G on F . Furthermore,
we require tij : Ui ∩ Uj → G, the map from p to the corresponding
group element tij (p) of the transition function – also denoted by tij by
abuse of notation – to be continuous.
7 To comply with the notation of Nakahara’s and Steenrod’s books, we switched domain

with the codomain of the local trivializations from our previous definition in Section 2.1.
This change is not restrictive since the trivializations are invertible.

13
 2.2 Fibre Bundles Leonardo A. Lessa

The transition function connects the trivializations by

φj (p, f ) = φi (p, tij (p)f ),

where here we treated tij (p) ∈ G as an element of G acting of f ∈ F . (2.3)

Exercise 2.3
Prove that It will also be useful to define a function pi : π −1 (Ui ) → F that as-
signs the elements of the bundle to their fibre representatives via the local
tij tjk = tik
trivialization, such that, for u ∈ E,
tii = e
tij = [tji ]−1 pi := [φi,π(u) ]−1 (2.1)

Analogous to the definition of manifolds by equivalences of atlases, we

start with an open cover {Ui } to define a coordinate bundle to, then, form
a fibre bundle out of a equivalence condition between coordinate bundles:
Definition 2.2. Two coordinate bundles (E, π, M, F, G, {Ui }, {φi }) and
(E, π, M, F, G, {Vj }, {ψj }) – differing only on the open covers and the local
trivializations – are equivalent if (E, π, M, F, G, {Ui } ∪ {Vj }, {φi } ∪ {ψj }) is
π
a coordinate bundle (2.4) . A fibre bundle, denoted by (E, π, M, F, G), E →
Exercise 2.4
Convince yourself that this M or simply E, when the rest of the structure is implicit, is an equivalence
amounts to requiring compati- class of coordinate bundles by the relation defined above.
bility between local trivializa-
tions φi and ψj as we did in Just as with manifolds, we think about fibre bundles in terms of their
condition (iii’) of Definition 2.1. representatives – the coordinate bundles – and prove theorems about these
which are invariant by the equivalence relation introduced in Definition 2.2.
Thus, hereafter we will not distinguish the equivalence class with its repre-
sentatives, hopefully not confusing the reader in the process.
In the following, we will only work with smooth fibre bundles, whose defi-
nition is the same as the fibre bundle one, but we strengthen the topological
requisites with smooth ones. More specifically, all the topological spaces
E, M , F and G are smooth manifolds, with G being a Lie group, and all
continuous (homeomorphic) functions are smooth (diffeomorphic).
The generalization of a vector field X ∈ X(M ) is a section X ∈ Γ(M, E),
defined by
Definition 2.3. A (cross) section X : M → E is a smooth map that sends
points p ∈ M of the base space to an element of its fibre X(p) ∈ Fp . In
other words, π ◦ X = id. The set of all sections from M to E is Γ(M, E), or
simply Γ(E).
A local section is a section X ∈ Γ(U, E) restricted to a open set U ⊂ M .
Example 2.1. Product bundle. For M and F topological spaces, we can
construct the product bundle (E = M × F, π, M, F, G) such that π = pr1 :
M × F → M and G = {e} is the trivial group, as every local trivialization
is the identity.
Example 2.2. Vector bundle. A vector bundle is a fibre bundle (E, π, M, V, G)
where the fibre is a vector space V . Because of this, it is common the action
of G on V to be a linear representation, i.e. an action by linear operators.
The tangent bundle T M is an example of a vector bundle with structure
group G = GL(n, R), as shown in Section 2.1. Other vector bundles will

14
2.2 Fibre Bundles Leonardo A. Lessa

come along our way. For example, the quantum mechanical wavefunction
and the Dirac field are sections in a vector bundle associated to a principal
bundle (See Section 2.3).
Example 2.3. Principal bundle. One way to define principal bundles is
to say they are fibre bundles whose typical fibre F is equal to the structure
group G. We will see the consequences of this in Section 2.3.
Another way is to start with a manifold E and a Lie group G that acts
freely on E on the right, and then define the base space M to be the quotient
space of the action of G on E

M = P/G.

Furthermore, we require E to be locally trivial in the sense that there exists

an open covering {Ui } of M and diffeomorphisms φi : Ui × G → π −1 (Ui )
such that pi (ua) = pi (u)a, where pi is defined by (2.1).
Can we loosen the requirements to construct a fibre bundle via Defi-
nitions 2.1 and 2.2? Fortunately, we can spare the specification of local
trivializations:
Theorem 2.1. Let G be a Lie Group acting freely on a manifold F , M
a manifold, {Ui } an open covering of M and {tij } a set of functions tij :
Ui ∩ Uj → G for each intersecting open sets Ui and Uj satisfying tkj tji = tki
π
and tii = e. Then, there exists a fibre bundle E → M with transition
functions {tij } and coordinate neighbourhoods {Ui }.
Proof. First, we define G
X := Ui × F
i∈I

This is our prototype for the total space. We still need to connect the
fibres from different intersecting parts of the union. To do this, we assign a
equivalence relation between points (p, fi ) ∈ Ui × F and (q, fj ) ∈ Uj × F of
X:
(p, fi ) ∼ (q, fj ) ⇐⇒ p = q and fi = tij (p)fj .
Now, we can define our total space to be E = X/ ∼, the quotient space,
formed by equivalence classes. Naturally, our projection π is defined by
π([(p, f )]) = p and the local trivialization associated to a open neighbour-
hood Ui is defined by φi (p, fi ) = [(p, fi )].(2.5)
Exercise 2.5
I glossed over some details, like
With this in mind, we can play with cylinders and Möbius strips: what is the topology of X and if
π is continuous and so on. If you
Example 2.4. Cylinder and Möbius strip. What are the possible (topo- care about those, try for yourself
logical) fibre bundles with base spaces M = S 1 and fibres F = [−1, 1], the to complete them! All the details
unit interval? One natural candidate is the cylinder M × F . For us to use are done in [9].
Theorem 2.1, it suffices to find an open covering {Ui } of S1 , a structure
group G acting on F and a set of transition functions {tij }.
The circumference S 1 ⊂ R2 is naturally covered by UN = S 1 \ {(0, −1)}
and US = S 1 \ {(0, 1)}. Their intersection has two connected components
UN ∩ US = U+ ∪ U− , defined by U± := {(x, y) ∈ S 1 | ± x ≥ 0}. Thus, we
have only one continuous transition function tSN : U+ ∪ U− → G.

15
 2.2 Fibre Bundles Leonardo A. Lessa

If we assume G is discrete, then tSN has to be constant on U+ and on

U− . Hence, the simplest cases are if G has one element, the trivial group
G = {e}, or two elements, the binary numbers group G = Z2 . In the
former case, we can have G = {e} acting on F with the identity, and then
tSN = id and E = S 1 × [−1, 1] is the cylinder. In the latter case, we can
have G = Z2 = {0, 1} acting on F by

0·f =f 1 · f = −f,

and then
tSN |U+ = 0 tSN |U− = 1.
This results in the fibre bundle of the Möbius band!
π π0
Definition 2.4. Let E → M and E 0 → M 0 be fibre bundles. A smooth
map f : E → E 0 is a bundle map if it maps each fibre Fp ⊂ E onto the
corresponding fibre Ff0 (p) ⊂ E 0 .

2.2.1 Triviality of Bundles

One way to determine the triviality of a fibre bundle is if the base space
is contractible to a point. Before enunciating the exact theorem for this, we
need to introduce bundle maps and pullback bundles.
A bundle map f : E → E 0 induces a map f¯ : M → M 0 on the base
manifolds, since f preserves the fibres.(2.6)
Exercise 2.6
Prove that f¯◦π = π 0 ◦f . Express
this in terms of a commuting di-
Lemma 2.1. Two coordinate bundles are equivalent if, and only if, they have
agram. the same base space and there exists a diffeomorphic bundle map between
them.

Proof. See Steenrod [9], Lemmas 2.6, 2.7 and 2.8.

Given a fibre bundle (E, π, M, F, G) and a smooth map f : N → M , we

can construct a bundle (f ∗ E, π1 , N, F, G) over N , called the pullback bundle.

Definition 2.5. Given a fibre bundle (E, π, M, F, G) and a smooth map

f : N → M , the total space of the pullback bundle (f ∗ E, π1 , N, F, G) is
defined as
f ∗ E = {(p, u) ∈ N × E|f (p) = π(u)},
a closed subspace of the manifold N × E; the projection π1 : f ∗ E → N is
just π1 (p, u) = p; the fibre is the same of E, as π1−1 (p) = (p, Ff (p) ) ' F ;
given an open covering of M {Ui }, the open neighbourhoods of f ∗ E are
Ui0 = f −1 (Ui ); and the local trivializations are

φ0i : Ui0 × F → π1−1 (Ui0 ),

φ0i (p, f ) = (p, φi (f (p), f )).

There is a natural bundle map π2 : (p, u) 7→ u between the pullback

bundle and the original bundle. If M = N and f = id, then, by Lemma 2.1,
f ∗ E ' E.
We are now ready to enunciate the homotopy theorem.

16
2.3 Principal Bundles Leonardo A. Lessa

Definition 2.6. Two maps f, g : X 0 → X between topological spaces X

and X 0 are said to be homotopic if there exists a map F : X 0 × [0, 1] → X,
called a homotopy, satisfying F (p, 0) = f (p) and F (p, 1) = g(p).
π
Theorem 2.2. Let E → M be a fibre bundle and f and g be homotopic
maps from a Cσ 8 space N to M , then the pullback bundles f ∗ E and g ∗ E are
equivalent.
Proof. See Steenrod [9], Theorem 11.4.
If M is contractible to a point p0 ∈ M , then id : M → M is homotopic
to the constant map e0 : M → {p0 } ⊂ M . Since id∗ E = E and e∗0 E =
M × F (2.7) , then Theorem 2.2 says
Exercise 2.7
π If this is not clear, try to prove
Corollary 2.1. The bundle E → M is trivial if M is contractible to a point.
that the pullback of E via e0
is the same as the pullback of
2.3 Principal Bundles {p0 } × F via e0 , since it is the
constant map.
Principal bundles will be crucial in the mathematical formalism of gauge
theory. One definition of principal bundles was given in Example 2.3
Definition 2.7. A principal bundle is a fibre bundle (P, π, M, F, G) where
F = G, also called the G-bundle over M , denoted only by (P, π, M, G). G
acts trivially on the fibre.
Roughly, if G is the gauge group of a physical theory, the G-bundle will
be the space of all possible gauge choices. Now, back to the mathematical
details.
Since the transition functions act on the fibres on the left, then the right
action of G on the fibres F = G is independent of the local trivialization and
thus G can act on the P bundle itself.
The action of a ∈ G on a bundle element u ∈ P is defined locally via
a trivialization φi : Ui × G → π −1 (Ui ), where p = π(u) ∈ Ui . If φ−1 i (u) =
(p, gi ), then we define ua by

ua := φi (p, gi a).

Now, we prove that this definition is independent of the local trivialization

chosen. Indeed, if p = π(u) ∈ Ui ∩ Uj , then

ua = φi (p, gi a),
= φj (p, tji (gi a)),
= φj (p, (tji gi )a),
= φj (p, gj a).

The action of G on E does not change the base point of the argument
and is transitive on fibres:

∀u1 , u2 ∈ Fp , ∃a ∈ G, u1 = u2 a,

and free:(2.8)
Exercise 2.8
8A Cσ -space is a normal locally compact manifold that admits a countable open cov- Prove that the action of G on E
ering. is transitive and free.

17
 2.3 Principal Bundles Leonardo A. Lessa

(∃u ∈ E, ua = u) ⇒ a = e.
Given a local cross section σ ∈ Γ(U, P ) over U ⊂ M in a principal bundle
P , we can construct a local trivialization defining

φσ : U × G → π −1 (U ),
φσ (p, g) = σ(p)g.

This map we have constructed is invertible since the right action of G on E

is transitive: for every point u ∈ Fp , there exists g(u) ∈ G such that

u = σ(p)g = φσ (p, g)

This is so important that we have a name for g(u):(2.9)

Exercise 2.9
Prove that the transition func-
Definition 2.8. The canonical local trivialization associated to local
tion tij : Ui ∩ Uj → G that
connects the trivializations de- section σ ∈ Γ(U, P ) is a map g : P → G such that
fined by σi ∈ Γ(Ui , P ) and
σj ∈ Γ(Uj , P ) satisfies σj (p) = ∀u ∈ U, u = σ(p)g(u).
σi (p)tij (p).

We will see later that the choice of a cross section in a G-bundle is

interpreted as gauge fixing. Note that, as with the local gauge invariance of
QED encountered in Section 1.2, the cross section can vary along the base
space M – our spacetime.

2.3.1 Associated Vector Bundle

We are going to show a method to construct a vector bundle (E, πE , M, V, G)
(See Example 2.2) given a G-bundle (P, π, M, G) and a vector space V that
G acts on by a linear representation ρ : G → GL(V ).9
The group G acts on the right on the manifold P × V as follows

(u, v) · g := (ug, ρ−1 (g)v).

π
We then define the total space of the vector bundle associated to P → M to
be the quotient of P × V by the right action of G, denoted by E := P ×G V .
If we denote the action of G on V by juxtaposition gv := ρ(g)v, then a useful
notation for the elements of E is, instead of the standard equivalence class
[(u, v)] ∈ E, we denote simply by uv, where u ∈ P and v ∈ V . The principle
behind this notation is because, for every g ∈ G, we have

uv = [(u, v)] = [(ug, ρ−1 (g)v)] = (ug)(g −1 v).

The bundle structure of E follows naturally from its associated principal

bundle. Namely, we define a projection πE : E → M by πE (uv) := π(u) and,
for every open neighborhood U ⊂ M of P , we define a local trivialization
−1
φE : U × V → πE (U ) on E using the trivialization φ : U × G → π −1 (U ) on
P such that the following diagram commutes
9 Given a manifold F , We may construct a fibre bundle with fibre F associated to a

principal bundle. Since the important cases will be with vector spaces, we anticipate and
only work with them.

18
2.3 Principal Bundles Leonardo A. Lessa

id ×ρ−1
U ×G×V U ×V
φ×id φE
[ ] −1
π −1 (U ) × V πE (U )

More directly, we can define(2.10)(2.11)

Exercise 2.10
Show that this definition guaran-
φE (u, v) := [(φ(u, e), v)] tees the preceding diagram com-
mutes.
Conversely, given a vector bundle (E, πE , M, V, G), we can employ The- Exercise 2.11
orem 2.1 to define an associated principal bundle (P, π, M, G). To do this, What are the local trivializations
we just employ the same open covering, base space and transitions functions of E = P ×G V ?
π
from E → M , and require that G acts on itself, the fibre, by the usual group
operation.(2.12)
Exercise 2.12
In our previous examples, the physical states that underwent gauge trans- Convince yourself that the vec-
formations were maps from a spacetime M to a vector space V (2.13) . The tor bundle associated to the
gauge freedom made the vector representation of a physical state highly non- principal bundle we constructed
here is the original bundle we
unique. Thus, a one-to-one association of a physical state to a mathematical
started with.
object has to gauge out this unphysical freedom. This is the main idea be- Exercise 2.13
hind the construction of the associated vector bundle. We append the vector Identify M and V of Sections 1.2,
space V to the principal bundle P and take the quotient by the right action 1.3.1 and 1.3.2.
defined in (2.3.1), which is just a way to gauge transform the vector v via the
representation ρ, but also leaving a trace of the transformation in u 7→ ug.
Thus, physical states are sections of E, elements of Γ(M, E).
Example 2.5. Frame Bundle. The associated vector bundle of the tan-
gent bundle T M is the frame bundle LM . Since the structure group of
T M is GL(n, R), where n = dim(M ), then LM is locally a product space
Ui × GL(n, R), welded by the transition functions tij : Ui ∩ Uj → GL(n, R).
In matrix notation, we have

(p, X µα ) ∈ Ui × GL(n, R), (p, Y νβ ) ∈ Uj × GL(n, R),

∂xµ  ν

X µα = (tij )µν Y να = Y α.
∂y ν  p

As the notation used above may have already hinted, we can also view
the frame bundle LM as the collection of vector bases {Xα }nα=1 ⊂ Tp M ,
called frames. More specifically, given a base point p ∈ M , we can define
the frame space Lp M as the set of all frames {Xα }nα=1 ⊂ Tp M and then
G
LM = Lp M.
p∈M

Given a chart (U, x) of M , with p ∈ U , each frame {Xα } ∈ Lp M is asso-

ciated to a non-singular matrix X µα ∈ GL(n, R) whose columns are the x
coordinates of Xα : (2.14)
Exercise 2.14
  Why is X µα non-singular?
| | |
X µα = X1 X2 . . . Xn  .
| | |

19
 Leonardo A. Lessa

The manifold structure of LM is analogous to the tangent bundle one

– if U L is an open neighbourhood of (p, {Xα }) ∈ LM whose projection
to M fits inside a chart set U , then we map (p, {Xα }) 7→ (xµ (p), X µα ) ∈
2
Rn+n . The bundle structure of (LM, π, M, GL(n, R)) is then equivalent to
the construction via association with T M .(2.15)
Exercise 2.15
What the right action of Gαβ ∈
GL(n, R) does to frames in LM ?
3 Connections in Fibre Bundles
3.1 Fundamental Vector Field
Before jumping into the theory of connections, we briefly review Lie
groups and Lie algebras and prove some technical theorems that will be
of good use in later proofs. If the reader is used to Lie groups and don’t
want to dive into the details of proofs, then they can skip this section.
The Lie algebra g of the Lie group G is the algebra of all left-invariant
vector fields A ∈ X(G), (Lg )∗ A|h = A|gh , under the operation of the Lie
bracket. Alternatively, it is the Lie algebra Te G, since any vector a ∈ Te G
induces a left-invariant vector field via A|g = (Lg )∗ a.

Definition 3.1. The fundamental vector field A# ∈ X(P ) generated by

A ∈ g is the vector field associated to the flow t 7→ u exp(tA)10 . In other
words,

d
#
f ∈ C ∞ (P ).

A |u (f ) = f (u exp(tA)) ,
dt t=0

Theorem 3.1. The mapping σ : g → X(P ) which sends A to A# is a

homomorphism between Lie algebras. Furthermore, A# does not vanish at
any point of P for A 6= 0.

Proof. We first show a more intrinsic way to define σ. For every u ∈ P , we

define σu : G → P by σu (g) = ug = Rg u. Thus, σA|u = (σu )∗ (A|e ), in
accordance with Definition 3.1.(3.1)
Exercise 3.1
Prove this To prove that σ is a homomorphism, we need to show that [A, B]# =
[A# , B # ]. Since the flow t 7→ Rat , with at = exp(tA), generates A# , then

1
[A# , B # ]|u = lim [(Ra−t )∗ (B # |uat ) − B # |u ],
t→0 t
1
= lim [(Ra−t ◦ σuat )∗ (B|e ) − σu∗ (B # |e )],
t→0 t

by the definitions,

Ra−t ◦ σuat (c) = uat ca−1

t = σu Ra−t Lat ,

for any c ∈ G. Since B is left-invariant, then (Ra−t ◦σuat )∗ (B|e ) = σu∗ Ra−t ∗ (B|at )
10 For us, a flow is a 1-parameter group of local transformations. In the case of A# ,

they are global. See definition in [4].

20
3.1 Fundamental Vector Field Leonardo A. Lessa

and
1
[A# , B # ]u = lim [σu∗ Ra−t ∗ (B|at ) − σu∗ (B # |e )],
t→0 t
1
= σu∗ lim [Ra−t ∗ (B|at ) − (B # |e )],
t→0 t
= σu∗ ([A, B]e ) = [A, B]# .

Finally, if A# vanished at, say, u ∈ P , then Rat leaves u fixed for every
t ∈ R (WHY?). Since G acts freely on P , then at = e for every t ∈ R, which
implies A = 0.
Definition 3.2. The vertical subspace Vu P is the subspace of Tu P tan-
gent to the fibre Gp 11 , where p = π(u).
Since the flow t 7→ u exp(tA) is entirely contained in Gp , with p = π(u),
then A# |u ∈ Vu P . Moreover, as G acts freely on P , A# never vanishes on
P , and the dimension of g is the same as of the fibre Gp ' G, the association
A 7→ A# is an isomorphism of Lie algebras g and Vu P . In other words, Vu P
is generated by A# , for A ∈ g.
Definition 3.3. Let Ad(a) : G → G be the adjoint map for a ∈ G, Ad(a)g =
aga−1 . Then the adjoint representation ad : G → Aut(g) of G in g is the
pushforward map ad(a) := (Ad(a))∗ .(3.2)
Exercise 3.2
For G a matrix group, ad(a) = Ad(a), if we view g as a set of matrices Prove that ad(a) = (Ra−1 )∗ .
with the same order as G’s.(3.3) . Since we will work with matrix group in
Exercise 3.3
our examples from physics, then we will often write ad(a)A = aAa−1 . Prove this
Proposition 3.1. Let φ : M → M be a transformation of a manifold M .
If the flow t 7→ φt generates a vector field X ∈ X(P ), then the flow t 7→
φ ◦ φt ◦ φ−1 generates φ∗ X.
Proof. Let p ∈ M , q = φ−1 p and f ∈ C ∞ (M ). The flow of the 1-parameter
group of local transformations φ ◦ φt ◦ φ−1 generates the following vector
field

d d
(φ ◦ φt ◦ φ ) (f ) = f (φ ◦ φt ◦ φ−1 p),
−1 

dt p dt
d
= f (φ[φt (q)]),
dt
= φ∗ (X|q ),
= (φ∗ X)|p .
Thus, φ ◦ φt ◦ φ−1 generates φ∗ X.
Theorem 3.2. If A# corresponds to A ∈ g, then (Ra )∗ A# corresponds to
ad(a−1 )A ∈ g.
Proof. Since A# is generated by the flow Rat , where at = exp(tA), then, by
Proposition 3.1, (Ra )∗ A# is generated by the flow Ra Rat Ra−1 = Ra−1 at a =
RAd(a−1 )at . Again, since t 7→ at generates the vector field A, then t 7→
Ad(a−1 )at generates Ad(a−1 )∗ A = ad(a−1 )A, as desired.
11 The fibre Gp = π −1 (p) at p ∈ M is a closed submanifold of P .

21
 3.2 Connection Leonardo A. Lessa

3.2 Connection
π
The geometric interpretation of a connection on a principal bundle P →
M is a smooth separation of each tangent space Tu P , u ∈ P , into the vertical
subspace Vu P and a horizontal subspace Hu P such that

(i) Tu P = Hu P ⊕ Vu P for all u ∈ P

(ii) (smoothness) A smooth vector field X ∈ X(P ) is decomposed into

a horizontal part X H |u ∈ Hu P and a vertical part X V |u ∈ Vu P as
X = XH + XV .

(iii) Hug P = Rg∗ (Hu P ) for every u ∈ P and g ∈ G.

The preferred way to generate this separation is by a connection form

ω ∈ g ⊗ Ω1 (P ), which is a Lie-algebra-valued one form on P .(3.4) Besides
Exercise 3.4
Given a basis {Tα } of g, prove being an element of this tensor product space ω can also be viewed as a
there existPone-forms {ω α } such map that smoothly associates to each u ∈ P a linear transformation ωu :
that ω = α Tα ⊗ ω α . Tu P → g, or as a linear map ω : X(P ) → C ∞ (M, g) between modules over
C ∞ (M ) functions. With this last form, it is more clear what the connection
does to vectors fields X ∈ X(P ). It projects X|u ∈ Tu P to Vu P , with
ker(ωu ) = Hu P . Now, let us present a proper definition.

Definition 3.4. Connection form or Ehresmann connection is a one-

form ω with values on g, denoted by ω ∈ g ⊗ Ω1 (P ), satisfying

(i) ω(A# ) = A, for every A ∈ g,

(ii) Rg∗ ω = adg−1 ω, for every g ∈ G. That is, for every X ∈ Tu P ,

ωug (Rg∗ X) = adg−1 (ωu (X)).

There is a one-to-one correspondence between separations of type Tu P =

Hu P ⊕ Vu P and projectors ωu with im(ωu ) = Vu P and ker(ωu ) = Hu P , so
there is no loss in generality for considering ω ∈ g ⊗ Ω1 (P ) instead of the
former geometrical definition.(3.5)(3.6)
Exercise 3.5
Prove that
Rg∗ Hu P = Hug P 3.3 Parallel Transport
for Hu P = ker(ωu ). With a connection, we can distinguish horizontal spaces in the tangent
Exercise 3.6 spaces of the principal bundle. This enables us to lift a curve γ : [0, 1] → M
How would you define X H and
in the base space to a curve γ̃ : [0, 1] → P in the principal bundle such that
X V given ω?
γ̃ 0 (t) ∈ Hγ(t) P . Intuitively, γ specifies the direction of γ̃ between fibers, and
the connection tells how the elevation changes along fibers.

Definition 3.5. Let γ : [0, 1] → M be a curve on M . A horizontal lift

γ̃ : [0, 1] → P of γ is a curve on P such that

(i) γ̃ 0 (t) ∈ Hγ(t) P ,

(ii) π ◦ γ̃ = γ.

The theorem for existence and uniqueness of a horizontal lift is the fol-
lowing

22
3.3 Parallel Transport Leonardo A. Lessa

Theorem 3.3. Let γ : [0, 1] → M be a curve on M , and u ∈ P such that

π(u) = γ(0). Then there exists an unique horizontal lift γ̃ : [0, 1] → M of γ
such that γ̃(0) = u.
We postpone the proof to Section 3.5
Corollary 3.1. If γ̃ is a horizontal lift of a curve γ on M , then any other
horizontal lift of γ is γ̃g, for some g ∈ G.
Proof. Let γ̃ be a horizontal lift of γ. Then for every g ∈ G, γ̃g is also a
horizontal lift of γ (See exercise 3.5). If γ̂ is another horizontal lift of γ,
then there exists a g ∈ G such that γ̂(0) = γ̃(0)g. From the uniqueness of
the horizontal lift given a starting point u = γ̂(0) = γ̃g ∈ P , then we have
γ̂ = γ̃g
Given a curve γ on M with endpoints γ(0) = p and γ(1) = q, we can now
construct a map Γ(γ) : Gp → Gq that parallel transports elements u ∈ Gp
of the fibre at p to elements Γ(γ)(u) of the fibre at Gq via a horizontal lift
starting at u. (3.7)
Exercise 3.7
There is a connection between parallel transport and the fundamental Using Corollary 3.1, prove that
group π1 (M ) of the base space M . To see this, we first note how Γ behaves
Γ(γ) ◦ Rg = Rg ◦ Γ(γ)
under composition of curves.
Theorem 3.4. Let α : [0, 1] → M and β : [0, 1] → M be curves satisfying
α(1) = β(0) and α ∗ β : [0, 1] → M be their composition, then
Γ(α ∗ β) = Γ(α) ◦ Γ(β),
Γ(α−1 ) = Γ(α)−1

What happens when γ is a loop? If the endpoints of γ are the same,

then Γ(γ) maps the fiber at p to itself. Thus, Γ(γ)(u) = u · τγ (u) for some
τγ (u) ∈ G(3.8) . If we span all loops γ based on p ∈ M , we get the holonomy
Exercise 3.8
group (3.9) (3.10)(3.11) What is the analogous of Theo-
rem 3.4 to τγ ?
Holu = {τγ (u)|γ is a loop based on p = π(u)}.
Exercise 3.9
Prove that Holu is a subgroup of
Analogously, we can define the restricted holonomy group restricting to count
G using Theorem 3.4.
only the contractible curves:
Exercise 3.10
Using exercise 3.7, prove that
Hol0u = {τγ (u)|γ is contractible to p = π(u)}.
Holug = g −1 Hu g.
These groups are related to the fundamental group π1 (M, p) based on p
Thus, all holonomy groups Holu
by the following theorem with the same base point p =
Theorem 3.5. There is a epimorphism (surjective homomorphism) Φ : π(u) are isomorphic.
π1 (M, p) → Holu / Hol0u , with p = π(u). Exercise 3.11
If M is connected, prove that
Proof. For each equivalence class of curves α ∈ π1 (M, p), we pick a repre- Holp is the same up to a conju-
gation:
sentative curve γ ∈ α and assign Φ(α) = τγ / Hol0u . This map is independent
of the representative γ since if we had picked another one γ̂ ∈ α, then Holq = τγ Holp τγ−1 ,
where γ is a curve from p to q.
τγ̂ (u)/ Hol0u = τγ (u) · τγ −1 γ̂ (u)/ Hol0u ,
= τγ̂ (u)/ Hol0u ,

23
3.4 Curvature Leonardo A. Lessa

since γ̂ −1 γ is contractible.
That the map Φ is a epimorphism comes from Theorem 3.4 and the
definition of Holu .

3.4 Curvature
In differential geometry, the Riemann curvature tensor measures the de-
gree to which a vector parallel transported in a loop fails to come back to
itself. We have seen that there is also a notion of parallel transport in gauge
theories, so it is natural to think that there exists a notion of curvature
dependent only on the connection of a principal bundle.

Definition 3.6. The Covariant Derivative of a vector-valued k−form

φ ∈ V ⊗ Ωk (P ) is a vector-valued (k + 1)−form Dφ ∈ V ⊗ Ωk+1 (P ) defined
by
Dφ(X1 , . . . , Xk+1 ) := (dP φ)(X1H , . . . , Xk+1
H
),
where dP is the exterior derivative that acts on the k−forms of φ and XiH
are the horizontal parts of the vectors Xi ∈ X(P ).

Definition 3.7. The Curvature two-form Ω ∈ g ⊗ Ω2 (P ) is the covariant

derivative of the connection one-form ω:

Ω := Dω.

The connection of the curvature form to the non-commutativity of par-

allel transport is summarized in the following theorem, by Ambrose and
Singer.

Theorem 3.6 (Ambrose-Singer). The Lie Algebra of the holonomy group

Holu0 , u0 ∈ P , is equal to the subalgebra of g spanned by elements Ωu (X, Y ),
for all X, Y ∈ Hu P and all u connected to u0 by a horizontal lift.

For the proof, see Theorem 8.1, page 89 of [4].

3.5 Local Forms

3.5.1 Gauge Potential
Let σ ∈ Γ(U, P ) be a local section defined on a open neighbourhood U .
Like a choice of gauge, we can use σ to introduce a local expression of the
connection form ω ∈ g ⊗ Ω1 (P ).

Definition 3.8. Given a connection form ω ∈ g ⊗ Ω1 (P ) and a local section

σ ∈ Γ(U, P ), the (local) gauge potential is the Lie-algebra-value one-form
A ∈ g ⊗ Ω1 (U ) on U , given by

A := σ ∗ ω.

In our examples with electromagnetism, A is actually the four-potential.

We saw in the Dirac monopole case that there are cases where A cannot be
globally defined. It is clear now that this corresponds to the fact that we
cannot always find a section σ on the whole base manifold M .

24
3.5 Local Forms Leonardo A. Lessa

Note that A is a one-form over U ⊂ M , the base manifold, with values

in the Lie algebra g. It is easier to do calculations in the base manifold M
than in the principal bundle P , since the latter can have an unknown twisted
topology, when the former is usually well known from the start.
Given that A is better to work with, can we go from A to ω? More than
that, what is the compatibility condition between gauge potentials Ai and
Aj that arise from two different sections σi ∈ Γ(Ui , P ) and σj ∈ Γ(Uj , P )
defined in intersecting open neighbourhoods Ui and Uj ? The two theorems
below answer these questions if G is a matrix group.12

Theorem 3.7. Given a g-valued one form A ∈ g⊗Ω1 (U ) defined on U ⊂ M

and a local section σ ∈ Γ(U, P ), there exists a local connection one-form
ω ∈ g × Ω1 (π −1 (U )) such that A = σ ∗ ω. The connection is given by

ω = g −1 (π ∗ A)g + g −1 dP g,

where g is the canonical local trivialization associated to σ (See Definition

2.8).(3.12)
Exercise 3.12
Prove that π ∗ A# = 0. What is
For the proof of this theorem, we refer to Theorem 10.1 of [1]. π ∗ doing?

Theorem 3.8. Two local gauge potentials Ai ∈ g × Ω1 (Ui ) and Aj ∈ g ×

Ω1 (Uj ) arise from the same connection form if, and only if, they satisfy

Aj = t−1 −1
ij Ai tij + tij dtij .

Proof. Before we prove the theorem, we need the following technical lemma.

Lemma 3.1. Let P (M, G) be a principal bundle and σi ∈ Γ(Ui , P ), σj ∈

Γ(Uj , P ) be local sections, where Ui ∩ Uj 6= ∅. For X ∈ Tp M , σi∗ X and
σj∗ X are connected by

σj∗ X = Rtij ∗ (σi∗ X) + (t−1 #

ij dtij (X)) .

For the proof of this lemma, we refer to Lemma 10.1 of [1].

Example 3.1. Gauge Transformation. The gauge potential A1 ∈ g ⊗

Ω1 (U ) associated to σ1 ∈ Γ(U, P ) changes via a gauge transformation σ2 :=
σ1 (p) · g(p), where g : U → G, as(3.13)
Exercise 3.13
Use Lemma 3.1 to prove the
A2 = g −1 A1 g + g −1 dg. transformation law of the gauge
potential.
If G is abelian, then we have simply

A2 = A1 + g −1 dg.
12 If the reader wants to see the general versions of Theorems 3.7 and 3.8, in which it is

not assumed that G is a matrix group, see Proposition 1.4, page 66 of [4].

25
3.5 Local Forms Leonardo A. Lessa

Example 3.2. U(1) Gauge Transformation. If G = U (1), then gauge

transformations are of the form g(p) = e−iΛ(p) . Then, the gauge potential
transforms as

A0 = A + eiΛ (−idΛ)e−iΛ ,
= A − idΛ.

If we define A := −iA, then, in components,

A0µ = Aµ + ∂µ Λ,

the same transformation law of the electromagnetic four-potential (See Sec-

tion 1.2). This is a good hint that electromagnetism is a U (1) gauge theory
as described in this chapter.

Example 3.3. Frame transformation. The Lie algebra of GL(n, R) is

Mat(n, R), the set of real square matrices of order n. Thus, the gauge
potentials of LM , the frame bundle, are matrix-valued one-forms Aαβ ∈
Mat(n, R) ⊗ Ω1 (U ). Let (U, x) and (V, x̃) be intersecting charts on M and
Aαβ and ˜(A)ρσ be their respective local gauge potentials of a common con-
nection. Then Theorem 3.8 tells us they are connected by
β −1 ρ
Ãρσ = t−1 ρ α λ
xx̃ α A β txx̃ σ + txx̃ λ dtxx̃ σ
∂ x̃ρ ∂xβ α ∂ x̃ρ
 λ
∂x
= α σ
A β + λ
d ,
∂x ∂ x̃ ∂x ∂ x̃σ
α
for txx̃αβ = ∂x
∂ x̃β
If we write the gauge potentials as Ãρσ = Ãρνσ dx̃ν and
µ
Aαβ = Aαµβ dxµ = Aαµβ ∂x ν
∂ x̃ν dx̃ , then in the x̃ components, the equation
above becomes

∂ x̃ρ ∂xβ ∂xµ α ∂ x̃ρ ∂ 2 xλ

Ãρνσ = α σ ν
A µβ + ,
∂x ∂ x̃ ∂ x̃ ∂xλ ∂ x̃ν ∂ x̃σ
which is the same transformation law of Christoffel symbols, if we assign
Aαµβ = Γαµβ . In section 3.6, we will see a stronger reason for the validity
of this association.

We are now capable of proving Theorem 3.3

Proof. We will use the compactness of the interval [0, 1] to construct γ̃ locally
on open sets {Uα }, to then cover [0, 1] with finitely many of {Uα }.
We first want to define γ̃(t) in a neighbourhood of t = 0. Suppose we
did that. From Condition (ii) of Definition 3.5, we know that γ̃(t) ∈ Gγ(t) =
π −1 (γ(t)). Thus, we take an open neighbourhood U of γ(t) with a local
cross section σ ∈ Γ(U, P ), which sets our local trivialization, and satisfies
σ(γ(0)) = u. Calling σ(t) = σ(γ(t)), for t ∈ V in a neighbourhood of
t = 0, then there exists a function g : V → G such that γ̃(t) = σ(t)g(t) and
g(0) = e. Since g(t) is much more tractable than γ̃, we will now construct
an equation for g(t) to find γ̃.

26
3.5 Local Forms Leonardo A. Lessa

The equation will be a differential equation, and will come from Condition
(i), which tells us that ω(γ̃ 0 ) = 0. Since γ̃ 0 = γ̃∗ γ 0 if we treat γ̃(t) = σ(t)g(t)
as a section in Γ(im(γ), P ), then, by Lemma 3.1,

0 = ω(γ̃(t)),
= ω(Rg(t)∗ (σ∗ γ 0 ) + [g(t)−1 dg(γ 0 )]# ),
dg(t)
= g(t)−1 ω(σ∗ γ 0 )g(t) + g(t)−1 .
dt

Since ω(σ∗ γ 0 ) = σ ∗ ω(γ 0 ) = A(γ 0 ), then

dg(t)
= −A(γ 0 )g(t), (3.1)
dt
whose formal solution is
 Z t   Z 
g(t) = P exp − A(γ 0 )dt = P exp − A
0 γ

where P, the path-ordering symbol, indicates that the product of the inte-
grals in the exponential function expansion is time-ordered.
This solution is valid for each open neighbourhood U where a local section
σ can be defined. Using the compactness of the interval [0, 1], we only need
finitely made of these sets to cover the entire curve. Finally, to go from open
set to the other, we just transform the initial point of the horizontal lift as
γ̃(0) → γ̃(0)g(1). The uniqueness of γ̃ is guaranteed by the uniqueness of
the solution of the ODE (3.1). (3.14)
Exercise 3.14
Prove that the path-ordered
exponential transforms under
3.5.2 Field Strength gauge transformations by conju-
gation. Thus, the trace of g(t)
Similarly to the connection one-form, we can pullback the curvature two- is gauge invariant. It is called
form on P via a local section σ ∈ Γ(U, P ) to a two-form on M : the Wilson loop in the context
of QFT.
Definition 3.9. Given a curvature two-form Ω = Dω ∈ g ⊗ Ω2 (P ) and a
local section σ ∈ Γ(U, P ), the (local) field strength is the Lie-algebra-value
two-form F ∈ g ⊗ Ω2 (U ) on U , given by

F := σ ∗ Ω.

The field strength can be expressed in terms of the gauge potential A by

F = dA + A ∧ A, (3.2)

where for A = Tα ⊗ Aα , we define(3.15)

Exercise 3.15
Prove all the relations in equa-
1
A∧A := (Tα Tβ )⊗(Aα ∧Aβ ) = fαβ γ Tγ ⊗(Aα ∧Aβ ) = [Tα , Tβ ]⊗(Aα ⊗Aβ ). tion (3.3).
2
(3.3)
To prove this, we first need to relate Ω to ω with the exterior differenti-
ation explicit:

27
 3.6 Covariant Derivative Leonardo A. Lessa

Theorem 3.9 (Cartan’s structure equation). Let X, Y ∈ Tu P , then

Ω(X, Y ) = dP ω(X, Y ) + [ω(X), ω(Y )],

in other words, Ω and ω are related by

Ω = dP ω + ω ∧ ω

For the proof of this Theorem, we refer to Theorem 10.3 of [1].

Applying the pullback σ ∗ to Cartan’s structure equation, we get equation
(3.2). In components F = 21 Fµν dxµ ∧ dxν , we have

Fµν = ∂µ Aν − ∂ν Aµ + [Aµ , Aν ], (3.4)

where each component Fµν or Aµ is a function of type U → g. If we pick a

basis {Tα } of g with structure constants fαβ γ , we can expand the equation
even further with

Fµν = Fµν α Tα ,
Aµ = Aµβ Tβ ,

such that (3.4) becomes an equation with real functions defined on U .

Fµν γ = ∂µ Aν γ − ∂ν Aµγ + fαβ γ Aµα Aν β (3.5)

In electromagnetism, the gauge group is U (1). This group has a one-

dimensional Lie Algebra, so we can drop the algebra indices on Aµα and
Fµν α . Another consequence of this is that U (1) is an abelian group, so
[Aµ , Aν ] = 0. Finally, equation (3.5) becomes Fµν = ∂µ Aν − ∂ν Aµ with Fµν
naturally interpreted as the Faraday tensor.
Under a gauge transformation, the field strength changes under conjuga-
tion:
Theorem 3.10. The field strength F1 ∈ g ⊗ Ω2 (U ) associated to σ1 ∈
Γ(U, P ) changes via a gauge transformation σ2 := σ1 (p) · g(p), where g :
U → G, as(3.16)
Exercise 3.16
Use Example 3.1 to prove the F2 = g −1 F1 g.
transformation law of the field
strength. If G is abelian, then the field strength does not change at all! This
explains the gauge invariance of the electric and magnetic fields. In the case
where G is a matrix group, we can form gauge invariant quantities taking
the traces, which are inherently invariant under conjugation. For example, a
gauge invariant scalar field used in the Yang-Mills Lagrangian (See Section
4.1) is Tr{Fµν F µν }(3.17)
Exercise 3.17
Prove Tr{Fµν F µν } is gauge in-
variant. 3.6 Covariant Derivative
As we saw in Section 1.2, we may need to modify the notion of deriva-
tive to include local gauge invariance. For this, we introduced the notion
of covariant derivative (See equation (1.19)), which included the four po-
tential and made the action gauge invariant. We know that physical states
are sections on an associated vector bundle, so we need to define a notion

28
3.6 Covariant Derivative Leonardo A. Lessa

of derivative on these sections that take account of the connection in the

associated principal bundle.
First, let us remember the structure of the associated vector bundle. If
(P, π, M, G) is a principle bundle, then we can construct a vector bundle
(E, πE , M, V, G) for any vector space V upon which G acts. The total space
is E made of equivalence classes [(u, v)], with u ∈ P and v ∈ V , and [(u, v)] =
[(ug, g −1 v)].
Given a section s ∈ Γ(M, E) on the associated vector bundle (physical
state) and a local section σ ∈ Γ(U, P ) on the principal bundle (gauge choice),
then s can be represented by a vector-valued function ξ : U → V , where

s(p) = [(σ(p), ξ(p))].

Now, we define the parallel transport of a section.

Definition 3.10. Let s ∈ Γ(M, E) be a section on E and γ : [0, 1] → M
be a curve on M with a horizontal lift γ̃. The horizontal lift implies the
existence of η : [0, 1] → V such that

s(γ(t)) = [(γ̃(t), η(t))].

Then s is said to be parallel transported along γ if η is constant.

This definition is independent of the particular horizontal lift and only
depends on the curve γ.(3.18)
Exercise 3.18
The covariant derivative of a section s is just the infinitesimal version of Use Corollary 3.1 to prove that
parallel transport Definition 3.10 is independent of
the horizontal lift.
Definition 3.11. Let s ∈ Γ(M, E) be a section on E and γ : [0, 1] → M a
curve on M . The covariant derivative of s along γ at p = γ(0) is
  
d 
∇X s := γ̃(0), η(γ(t)) ,
dt t=0

where X = γ 0 (0), with the same notation from Definition 3.10.

The covariant derivative is also independent of the horizontal lift. In fact,
it only depends on the vector X = γ 0 (0).(3.19)
Exercise 3.19
The definition of parallel transport and of covariant derivative are in Prove this.
harmony with the ones from differential geometry if P = LM .
If X ∈ X(M ) is a vector field, then we naturally have a map p 7→
∇X(p) s ∈ Fp , which is again a section on E. Thus, the covariant derivative
can be viewed as a linear map ∇ : X(M ) ⊗ Γ(M, E) → Γ(M, E), treating
X(M ) as a module over the space of C ∞ functions and Γ(M, E) as a vec-
tor space over R. We do not have C ∞ linearity on Γ(M, E) because of the
product rule
∇(f s) = (df )s + f ∇s.
Now let us see what is the local form of the covariant derivative, by local
we mean writing everything in terms of a local section σ ∈ Γ(U, P ):

γ̃(t) = σ(γ(t))g(t),
eα (p) = [(σ(p), eVα )],

29
 Leonardo A. Lessa

where g(t) := g(γ(t)) is the canonical local trivialization and {eVα } is a

basis for V . We only need to calculate the covariant derivatives of the form
∇∂/∂xµ eα , since any section s ∈ Γ(M, E) is written as

s(p) = [(σ(p), ξ(p))],

= [(σ(p), ξ α (p)eVα )],
= ξ α (p)eα ≡ sα (p)eα ,

and any X ∈ X(M ) is written as X = X µ ∂x∂ µ , so

∇X s = ∇X µ ∂/∂xµ (sα eα ),
 α 
µ ∂s α
=X eα + s ∇∂/∂xµ eα .
∂xµ

Since eα (γ(t)) = [(σ(γ(t)), eVα )] = [(γ̃(t), g(t)−1 eVα )], then the covariant
derivative of eα in the direction of ∂x∂ µ = γ 0 (0) is
  
d
γ̃(0), g(t)−1 eVα 

∇∂/∂xµ eα = ,
dt t=0
  
−1 dg(t)  −1 V

= γ̃(0), −g(0) g(0) eα ,
dt t=0
= [(γ̃(0)g(0)−1 , Aµ eVα )],

where we have used equation (3.1) and that A = Aµ dxµ . Bear in mind
that Aµ here is in reality the representation ρ(Aµ ). In matrix notation with
respect to the basis eVα , we write Aµαβ := ρ(Aµ )αβ , so

∇∂/∂xµ = [(σ(0), Aµβ α eVβ )] = Aµβ α eβ .

Finally,
∂sβ
 
(∇X s)β = X µ + A β
µ α sα
. (3.6)
∂xµ

4 The Examples from Another View

4.1 QED and Yang-Mills Gauge Theories
As we have seen, QED is a gauge theory with gauge group U (1). The
principle bundle with this gauge group and a connection in it forms the
electromagnetic part of the theory, with the four-potential and the field
strength tensor being local expressions for the connection and the curvature,
respectively.
The matter field part comes from the associated vector bundle. In QED,
we have charged particles called electrons and positrons, which classically –
and not worrying about gauge invariance – are just fields with values on the
vector space C4 . If instead we were interested in non-relativistic quantum
mechanics with electromagnetic interactions, then the matter field would
be a wave function, a complex field. In either case, we can transport the

30
4.2 Dirac Monopole Leonardo A. Lessa

concepts from the principle bundle to a associated vector bundle via a repre-
sentation on the vector space in question. This approach incorporates gauge
invariance intrinsically, since a physical field is a section on the associated
bundle that has, at each point of the base space, many representations in
the original vector space, corresponding to gauge transformation. However
so, these representations do not change the section itself. Because of this,
the section removes gauge redundancy at the cost of a definition based on
harder to work with conjugacy classes. Furthermore, we can parallel trans-
port the matter fields, which have noticeable physical consequences, like the
Aharanov-Bohm effect.
If we allow the gauge group G to be non-abelian, then we still get the
terms involving the structure constants. This case is not just some purely
mathematical generalization, but it in fact happens in Quantum Chromo-
dynamics (QCD), the theory of strong interactions. In QCD, the gauge
group is SU (3), with gauge potential corresponding to the gluon fields, and
the matter fields are the quark, which are composed of three Dirac fields,
corresponding to the three colors they have.
With SU (3) being a non-abelian group, the field strength F = dA+A∧A
is now quadratic in the gauge potential. This has an impact in the dynamics
of QCD, since its Lagrangian contains a term proportional to Tr[Fµν F µν ],
which is now quartic with A. This means that the quantum perturbation
theory has an vertex (interaction) with 4 gluons. This contrasts with QED,
that does not exhibit photon-photon interaction at low energies. The proper
quantum theory of QCD is much more complicated than this, but gauge
theory by itself already predicts a variety of complex phenomena.

4.2 Dirac Monopole

We saw in Section 1.3 that the existence of a single magnetic charge has
profound consequences, one of them being the quantization of the electric
charge. Also, we noted that this solution is essentially topological, since we
can exclude the point of the space where the magnetic monopole sits and
thus analyze the vacuum solutions to Maxwell’s equations. We will see later
how we can expect to have nontrivial solution to this problem just from
topological properties of the base space M = R3 \ {0}, but first we will
connect the arguments given earlier with Gauge Theory.
The interesting effects of the Dirac monopole came from coupling non-
relativistic quantum mechanics with electromagnetism. This suggests that
the associated gauge theory will have the quantum mechanical gauge group
G = U (1). Since gauge theory predicts the Faraday tensor F = dA is a
closed form, then it does not accommodate magnetic charges, so we are
forced to consider our base space to be M = R3 \ {0}, meaning the magnetic
charge is at the origin. From Theorem 2.2, we know that fibre bundles with
homotopic base spaces are equivalent, thus we can simplify our base space
to M = S 2 , the sphere.
The sphere cannot be covered by a single coordinate map, and this reflects
on the fact that we cannot use a single gauge potential A to describe the
solution to the magnetic monopole problem on the whole sphere. We already
knew this from our explicit vector potential solutions, which had a singularity

31
 4.2 Dirac Monopole Leonardo A. Lessa

on a ray starting at the monopole, called Dirac string.

For simplicity, let us cover the sphere with the north and south hemi-
spheres, plus small strips around the equator on which these open neigh-
bourhoods intersect. In spherical coordinates,

HN = {(θ, φ) | 0 ≤ θ < π/2 + ε, 0 ≤ φ < 2π},

HS = {(θ, φ) | π/2 − ε ≤ θ < π, 0 ≤ φ < 2π}.

As before, we can define a gauge potential on each of hemispheres as (See

equations (1.23) and (1.24))
g
AN = i (1 − cos θ)dφ, (4.1)
4π
g
AS = −i (1 + cos θ)dφ, (4.2)
4π

where we used the correspondence Aj = ieAj .(4.1) For these potentials to be

Exercise 4.1
Prove that originated from the same connection form, they have to satisfy the following
∂ relation on the intersection (See Theorem 3.8)
êφ = ,
∂φ
AN = t−1 S −1
N S A tN S + tN S dtN S ,
and thus, transforming into a
one-form via the metric, we have
the association where tN S = exp[iϕ(φ)] : HN ∩HS → U (1) is the transition function between
1 the hemispheres, which we consider to be a function only of the azimuthal
êφ → dφ. angle φ since the strip HN ∩ HS can be made as small as we want. From the
r sin θ
Remember that êφ is normal-
explicit expressions (4.1) and (4.2) and the fact that the gauge group U (1)
ized. is abelian, we have

idϕ(φ) = t−1
N S dtN S ,
= AN − AS ,
eg
= i dφ.
2π
For tN S = exp[iϕ(φ)] to be well defined, the phase ϕ : S 1 → U (1) has to
change by an integer multiple of 2π:
Z 2π
2πn = ϕ(2π) − ϕ(0) = dϕ = eg,
0

from which we get our familiar charge quantization equation.

A deeper topological reason why charge quantization follows from the
existence of a magnetic monopole comes from analyzing the transition func-
tion tN S . The different charges comes from the different ways of describing
the transition function tN S = exp[iϕ(φ)], which is essentially defined on the
equator S 1 . Thus, an equivalent problem is the classification of continuous
functions ϕ : S 1 → U (1), that is, what is the fundamental group π1 (U (1))?
And we know how to answer that! Since U (1) is homotopically equivalent
to S 1 , then π1 (U (1)) = π1 (U (1)) = Z, the integer additive group.
An even earlier prediction we could have made is the existence of nontriv-
ial vacuum solutions to Maxwell’s equations on the punctured space R3 \ {0}
that do not come from electrical charges or currents. We already know that

32
4.2 Dirac Monopole Leonardo A. Lessa

electrical charges generate fields described by a global gauge potential A and

a Faraday tensor F = dA, an exact one-form. Moreover, vacuum Maxwell’s
equations say that F is a closed form, that is, dF = 0. If we are trying to
search for solutions that do not come from electrical sources, then we want to
find classify all closed two-forms F that are not exact, which is exactly what
the cohomology group is! More specifically, the second cohomology group
of the base space M = R3 \ {0}, calculated as H 2 (M ) = H 2 (S 2 ) = R, a
non-trivial group! Note that we cannot conclude anything related to charge
quantization because we have not touched any aspect of gauge theory.

33
REFERENCES Leonardo A. Lessa

References
[1] Mikio Nakahara. Geometry, Topology and Physics. 2nd ed. Institute of
Physics Publishing, 2003. Chap. 1, 9, 10.
[2] John David Jackson. Classical electrodynamics. 3rd ed. New York, NY:
Wiley, 1999. Chap. 11.
[3] Robert M. Wald. General Relativity. The University of Chicago Press,
1984. Chap. 13, pp. 342–347.
[4] S. Kobayashi and K. Nomizu. Foundations of Differential Geometry.
Vol. 1. Wiley Classics Library. Wiley-Interscience, 1963. Chap. 1,2.
[5] R. F. Streater and A. S. Wightman. PCT, spin and statistics, and all
that. 1989. Chap. 1, pp. 9–21.
[6] N. N. Bogoliubov and D. V. Shirkov. Introduction to the Theory of
Quantized Fields. 3rd ed. John Wilet & Sons, Inc., 1980. Chap. 1, pp. 51–
63.
[7] Gerard ’t Hooft. The Conceptual Basis of Quantum Field Theory. June 8,
2016. url: http://www.staff.science.uu.nl/~hooft101/lectures/
basisqft.pdf (visited on 04/14/2019).
[8] David Tong. Gauge Theory. 2018. url: http://www.damtp.cam.ac.
uk/user/tong/gaugetheory/1em.pdf (visited on 04/16/2019).
[9] N. Steenrod. The Topology of Fibre Bundles. Princeton Landmarks in
Mathematics and Physics v. 14. Princeton University Press, 1999.

34