Preprint typeset in JHEP style - PAPER VERSION Michaelmas Term, 2006 and 2007

Quantum Field Theory
University of Cambridge Part III Mathematical Tripos
Dr David Tong
Department of Applied Mathematics and Theoretical Physics,
Centre for Mathematical Sciences,
Wilberforce Road,
Cambridge, CB3 OWA, UK
http://www.damtp.cam.ac.uk/user/tong/qft.html
d.tong@damtp.cam.ac.uk
– 1 –
Recommended Books and Resources
• M. Peskin and D. Schroeder, An Introduction to Quantum Field Theory
This is a very clear and comprehensive book, covering everything in this course at the
right level. It will also cover everything in the “Advanced Quantum Field Theory”
course, much of the “Standard Model” course, and will serve you well if you go on to
do research. To a large extent, our course will follow the first section of this book.
There is a vast array of further Quantum Field Theory texts, many of them with
redeeming features. Here I mention a few very different ones.
• S. Weinberg, The Quantum Theory of Fields, Vol 1
This is the first in a three volume series by one of the masters of quantum field theory.
It takes a unique route to through the subject, focussing initially on particles rather
than fields. The second volume covers material lectured in “AQFT”.
• L. Ryder, Quantum Field Theory
This elementary text has a nice discussion of much of the material in this course.
• A. Zee, Quantum Field Theory in a Nutshell
This is charming book, where emphasis is placed on physical understanding and the
author isn’t afraid to hide the ugly truth when necessary. It contains many gems.
• M Srednicki, Quantum Field Theory
A very clear and well written introduction to the subject. Both this book and Zee’s
focus on the path integral approach, rather than canonical quantization that we develop
in this course.
There are also resources available on the web. Some particularly good ones are listed
on the course webpage: http://www.damtp.cam.ac.uk/user/tong/qft.html
Contents
0. Introduction 1
0.1 Units and Scales 4
1. Classical Field Theory 7
1.1 The Dynamics of Fields 7
1.1.1 An Example: The Klein-Gordon Equation 8
1.1.2 Another Example: First Order Lagrangians 9
1.1.3 A Final Example: Maxwell’s Equations 10
1.1.4 Locality, Locality, Locality 10
1.2 Lorentz Invariance 11
1.3 Symmetries 13
1.3.1 Noether’s Theorem 13
1.3.2 An Example: Translations and the Energy-Momentum Tensor 14
1.3.3 Another Example: Lorentz Transformations and Angular Mo-
mentum 16
1.3.4 Internal Symmetries 18
1.4 The Hamiltonian Formalism 19
2. Free Fields 21
2.1 Canonical Quantization 21
2.1.1 The Simple Harmonic Oscillator 22
2.2 The Free Scalar Field 23
2.3 The Vacuum 25
2.3.1 The Cosmological Constant 26
2.3.2 The Casimir Effect 27
2.4 Particles 29
2.4.1 Relativistic Normalization 31
2.5 Complex Scalar Fields 33
2.6 The Heisenberg Picture 35
2.6.1 Causality 36
2.7 Propagators 38
2.7.1 The Feynman Propagator 38
2.7.2 Green’s Functions 40
2.8 Non-Relativistic Fields 41
2.8.1 Recovering Quantum Mechanics 43
– 1 –
3. Interacting Fields 47
3.1 The Interaction Picture 50
3.1.1 Dyson’s Formula 51
3.2 A First Look at Scattering 53
3.2.1 An Example: Meson Decay 55
3.3 Wick’s Theorem 56
3.3.1 An Example: Recovering the Propagator 56
3.3.2 Wick’s Theorem 58
3.3.3 An Example: Nucleon Scattering 58
3.4 Feynman Diagrams 60
3.4.1 Feynman Rules 61
3.5 Examples of Scattering Amplitudes 62
3.5.1 Mandelstam Variables 66
3.5.2 The Yukawa Potential 67
3.5.3 φ
4
Theory 69
3.5.4 Connected Diagrams and Amputated Diagrams 70
3.6 What We Measure: Cross Sections and Decay Rates 71
3.6.1 Fermi’s Golden Rule 71
3.6.2 Decay Rates 73
3.6.3 Cross Sections 74
3.7 Green’s Functions 75
3.7.1 Connected Diagrams and Vacuum Bubbles 77
3.7.2 From Green’s Functions to S-Matrices 79
4. The Dirac Equation 81
4.1 The Spinor Representation 83
4.1.1 Spinors 85
4.2 Constructing an Action 87
4.3 The Dirac Equation 90
4.4 Chiral Spinors 91
4.4.1 The Weyl Equation 91
4.4.2 γ
5
93
4.4.3 Parity 94
4.4.4 Chiral Interactions 95
4.5 Majorana Fermions 96
4.6 Symmetries and Conserved Currents 98
4.7 Plane Wave Solutions 100
4.7.1 Some Examples 102
– 2 –
4.7.2 Some Useful Formulae: Inner and Outer Products 103
5. Quantizing the Dirac Field 106
5.1 A Glimpse at the Spin-Statistics Theorem 106
5.1.1 The Hamiltonian 107
5.2 Fermionic Quantization 109
5.2.1 Fermi-Dirac Statistics 110
5.3 Dirac’s Hole Interpretation 110
5.4 Propagators 112
5.5 The Feynman Propagator 114
5.6 Yukawa Theory 115
5.6.1 An Example: Putting Spin on Nucleon Scattering 115
5.7 Feynman Rules for Fermions 117
5.7.1 Examples 118
5.7.2 The Yukawa Potential Revisited 121
5.7.3 Pseudo-Scalar Coupling 122
6. Quantum Electrodynamics 124
6.1 Maxwell’s Equations 124
6.1.1 Gauge Symmetry 125
6.2 The Quantization of the Electromagnetic Field 128
6.2.1 Coulomb Gauge 128
6.2.2 Lorentz Gauge 131
6.3 Coupling to Matter 136
6.3.1 Coupling to Fermions 136
6.3.2 Coupling to Scalars 138
6.4 QED 139
6.4.1 Naive Feynman Rules 141
6.5 Feynman Rules 143
6.5.1 Charged Scalars 144
6.6 Scattering in QED 144
6.6.1 The Coulomb Potential 147
6.7 Afterword 149
– 3 –
Acknowledgements
These lecture notes are far from original. My primary contribution has been to borrow,
steal and assimilate the best discussions and explanations I could find from the vast
literature on the subject. I inherited the course from Nick Manton, whose notes form the
backbone of the lectures. I have also relied heavily on the sources listed at the beginning,
most notably the book by Peskin and Schroeder. In several places, for example the
discussion of scalar Yukawa theory, I followed the lectures of Sidney Coleman, using
the notes written by Brian Hill and a beautiful abridged version of these notes due to
Michael Luke.
My thanks to the many who helped in various ways during the preparation of this
course, including Joe Conlon, Nick Dorey, Marie Ericsson, Eyo Ita, Ian Drummond,
Jerome Gauntlett, Matt Headrick, Ron Horgan, Nick Manton, Hugh Osborn and Jenni
Smillie. My thanks also to the students for their sharp questions and sharp eyes in
spotting typos. I am supported by the Royal Society.
– 4 –
0. Introduction
“There are no real one-particle systems in nature, not even few-particle
systems. The existence of virtual pairs and of pair fluctuations shows that
the days of fixed particle numbers are over.”
Viki Weisskopf
The concept of wave-particle duality tells us that the properties of electrons and
photons are fundamentally very similar. Despite obvious differences in their mass and
charge, under the right circumstances both suffer wave-like diffraction and both can
pack a particle-like punch.
Yet the appearance of these objects in classical physics is very different. Electrons
and other matter particles are postulated to be elementary constituents of Nature. In
contrast, light is a derived concept: it arises as a ripple of the electromagnetic field. If
photons and particles are truely to be placed on equal footing, how should we reconcile
this difference in the quantum world? Should we view the particle as fundamental,
with the electromagnetic field arising only in some classical limit from a collection of
quantum photons? Or should we instead view the field as fundamental, with the photon
appearing only when we correctly treat the field in a manner consistent with quantum
theory? And, if this latter view is correct, should we also introduce an “electron field”,
whose ripples give rise to particles with mass and charge? But why then didn’t Faraday,
Maxwell and other classical physicists find it useful to introduce the concept of matter
fields, analogous to the electromagnetic field?
The purpose of this course is to answer these questions. We shall see that the second
viewpoint above is the most useful: the field is primary and particles are derived
concepts, appearing only after quantization. We will show how photons arise from the
quantization of the electromagnetic field and how massive, charged particles such as
electrons arise from the quantization of matter fields. We will learn that in order to
describe the fundamental laws of Nature, we must not only introduce electron fields,
but also quark fields, neutrino fields, gluon fields, W and Z-boson fields, Higgs fields
and a whole slew of others. There is a field associated to each type of fundamental
particle that appears in Nature.
Why Quantum Field Theory?
In classical physics, the primary reason for introducing the concept of the field is to
construct laws of Nature that are local. The old laws of Coulomb and Newton involve
“action at a distance”. This means that the force felt by an electron (or planet) changes
– 1 –
immediately if a distant proton (or star) moves. This situation is philosophically un-
satisfactory. More importantly, it is also experimentally wrong. The field theories of
Maxwell and Einstein remedy the situation, with all interactions mediated in a local
fashion by the field.
The requirement of locality remains a strong motivation for studying field theories
in the quantum world. However, there are further reasons for treating the quantum
field as fundamental
1
. Here I’ll give two answers to the question: Why quantum field
theory?
Answer 1: Because the combination of quantum mechanics and special relativity
implies that particle number is not conserved.
Particles are not indestructible objects, made at the
Figure 1:
beginning of the universe and here for good. They can be
created and destroyed. They are, in fact, mostly ephemeral
and fleeting. This experimentally verified fact was first
predicted by Dirac who understood how relativity implies
the necessity of anti-particles. An extreme demonstra-
tion of particle creation is shown in the picture, which
comes from the Relativistic Heavy Ion Collider (RHIC) at
Brookhaven, Long Island. This machine crashes gold nu-
clei together, each containing 197 nucleons. The resulting
explosion contains up to 10,000 particles, captured here in
all their beauty by the STAR detector.
We will review Dirac’s argument for anti-particles later in this course, together with
the better understanding that we get from viewing particles in the framework of quan-
tum field theory. For now, we’ll quickly sketch the circumstances in which we expect
the number of particles to change. Consider a particle of mass m trapped in a box
of size L. Heisenberg tells us that the uncertainty in the momentum is ∆p ≥ /L.
In a relativistic setting, momentum and energy are on an equivalent footing, so we
should also have an uncertainty in the energy of order ∆E ≥ c/L. However, when
the uncertainty in the energy exceeds ∆E = 2mc
2
, then we cross the barrier to pop
particle anti-particle pairs out of the vacuum. We learn that particle-anti-particle pairs
are expected to be important when a particle of mass m is localized within a distance
of order
λ =

mc
1
A concise review of the underlying principles and major successes of quantum field theory can be
found in the article by Frank Wilczek, http://arxiv.org/abs/hep-th/9803075
– 2 –
At distances shorter than this, there is a high probability that we will detect particle-
anti-particle pairs swarming around the original particle that we put in. The distance λ
is called the Compton wavelength. It is always smaller than the de Broglie wavelength
λ
dB
= h/[ p[. If you like, the de Broglie wavelength is the distance at which the wavelike
nature of particles becomes apparent; the Compton wavelength is the distance at which
the concept of a single pointlike particle breaks down completely.
The presence of a multitude of particles and antiparticles at short distances tells us
that any attempt to write down a relativistic version of the one-particle Schr¨odinger
equation (or, indeed, an equation for any fixed number of particles) is doomed to failure.
There is no mechanism in standard non-relativistic quantum mechanics to deal with
changes in the particle number. Indeed, any attempt to naively construct a relativistic
version of the one-particle Schr¨odinger equation meets with serious problems. (Negative
probabilities, infinite towers of negative energy states, or a breakdown in causality are
the common issues that arise). In each case, this failure is telling us that once we we
enter the relativistic regime we need a new formalism in order to treat states with an
unspecified number of particles. This formalism is quantum field theory (QFT).
Answer 2: Because all particles of the same type are the same
This sound rather dumb. But it’s not! What I mean by this is that two electrons
are identical in every way, regardless of where they came from and what they’ve been
through. The same is true of every other fundamental particle. Let me illustrate this
through a rather prosaic story. Suppose we capture a proton from a cosmic ray which
we identify as coming from a supernova lying 8 billion lightyears away. We compare
this proton with one freshly minted in a particle accelerator here on Earth. And the
two are exactly the same! How is this possible? Why aren’t there errors in proton
production? How can two objects, manufactured so far apart in space and time, be
identical in all respects? One explanation that might be offered is that there’s a sea
of proton “stuff” filling the universe and when we make a proton we somehow dip our
hand into this stuff and from it mould a proton. Then it’s not surprising that protons
produced in different parts of the universe are identical: they’re made of the same stuff.
It turns out that this is roughly what happens. The “stuff” is the proton field or, if
you look closely enough, the quark field.
In fact, there’s more to this tale. Being the “same” in the quantum world is not
like being the “same” in the classical world: quantum particles that are the same are
truely indistinguishable. Swapping two particles around leaves the state completely
unchanged — apart from a possible minus sign. This minus sign determines the statis-
tics of the particle. In quantum mechanics you have to put these statistics in by hand
– 3 –
and, to agree with experiment, should choose Bose statistics (no minus sign) for integer
spin particles, and Fermi statistics (yes minus sign) for half-integer spin particles. In
quantum field theory, this relationship between spin and statistics is not something
that you have to put in by hand. Rather, it is a consequence of the framework.
What is Quantum Field Theory?
Having told you why QFT is necessary, I should really tell you what it is. The clue is in
the name: it is the quantization of a classical field, the most familiar example of which
is the electromagnetic field. In standard quantum mechanics, we’re taught to take the
classical degrees of freedom and promote them to operators acting on a Hilbert space.
The rules for quantizing a field are no different. Thus the basic degrees of freedom in
quantum field theory are operator valued functions of space and time. This means that
we are dealing with an infinite number of degrees of freedom — at least one for every
point in space. This infinity will come back to bite on several occasions.
It will turn out that the possible interactions in quantum field theory are governed
by a few basic principles: locality, symmetry and renormalization group flow (the
decoupling of short distance phenomena from physics at larger scales). These ideas
make QFT a very robust framework: given a set of fields there is very often an almost
unique way to couple them together.
What is Quantum Field Theory Good For?
The answer is: almost everything. As I have stressed above, for any relativistic system
it is a necessity. But it is also a very useful tool in non-relativistic systems with many
particles. Quantum field theory has had a major impact in condensed matter, high-
energy physics, cosmology, quantum gravity and pure mathematics. It is literally the
language in which the laws of Nature are written.
0.1 Units and Scales
Nature presents us with three fundamental dimensionful constants; the speed of light c,
Planck’s constant (divided by 2π) and Newton’s constant G. They have dimensions
[c] = LT
−1
[] = L
2
MT
−1
[G] = L
3
M
−1
T
−2
Throughout this course we will work with “natural” units, defined by
c = = 1 (0.1)
– 4 –
which allows us to express all dimensionful quantities in terms of a single scale which
we choose to be mass or, equivalently, energy (since E = mc
2
has become E = m).
The usual choice of energy unit is eV , the electron volt or, more often GeV = 10
9
eV or
TeV = 10
12
eV . To convert the unit of energy back to a unit of length or time, we need
to insert the relevant powers of c and . For example, the length scale λ associated to
a mass m is the Compton wavelength
λ =

mc
With this conversion factor, the electron mass m
e
= 10
6
eV translates to a length scale
λ
e
= 2 10
−12
m.
Throughout this course we will refer to the dimension of a quantity, meaning the
mass dimension. If X has dimensions of (mass)
d
we will write [X] = d. In particular,
the surviving natural quantity G has dimensions [G] = −2 and defines a mass scale,
G =
c
M
2
p
=
1
M
2
p
(0.2)
where M
p
≈ 10
19
GeV is the Planck scale. It corresponds to a length l
p
≈ 10
−33
cm. The
Planck scale is thought to be the smallest length scale that makes sense: beyond this
quantum gravity effects become important and it’s no longer clear that the concept
of spacetime makes sense. The largest length scale we can talk of is the size of the
cosmological horizon, roughly 10
60
l
p
.
−33
10
28
10
19
10 = GeV
12
10 10
11 −3
10
−33
10
= 1 TeV
cm
Planck Scale
Energy
eV − eV eV
Constant
Cosmological
eV
Observable
Universe
~ 20 billion light years
length
Earth 10
10
cm
Atoms
−8
10
Nuclei
10
LHC
cm cm
−13
Figure 2: Energy and Distance Scales in the Universe
Some useful scales in the universe are shown in the figure. This is a logarithmic plot,
with energy increasing to the right and, correspondingly, length increasing to the left.
The smallest and largest scales known are shown on the figure, together with other
relevant energy scales. The standard model of particle physics is expected to hold up
– 5 –
to about the TeV . This is precisely the regime that is currently being probed by the
Large Hadron Collider (LHC) at CERN. There is a general belief that the framework
of quantum field theory will continue to hold to energy scales only slightly below the
Planck scale — for example, there are experimental hints that the coupling constants
of electromagnetism, and the weak and strong forces unify at around 10
18
GeV.
For comparison, the rough masses of some elementary (and not so elementary) par-
ticles are shown in the table,
Particle Mass
neutrinos ∼ 10
−2
eV
electron 0.5 MeV
Muon 100 MeV
Pions 140 MeV
Proton, Neutron 1 GeV
Tau 2 GeV
W,Z Bosons 80-90 GeV
Higgs Boson 120-200 GeV??
– 6 –
1. Classical Field Theory
In this first section we will discuss various aspects of classical fields. We will cover only
the bare minimum ground necessary before turning to the quantum theory, and will
return to classical field theory at several later stages in the course when we need to
introduce new ideas.
1.1 The Dynamics of Fields
A field is a quantity defined at every point of space and time (x, t). While classical
particle mechanics deals with a finite number of generalized coordinates q
a
(t), indexed
by a label a, in field theory we are interested in the dynamics of fields
φ
a
(x, t) (1.1)
where both a and x are considered as labels. Thus we are dealing with a system with an
infinite number of degrees of freedom — at least one for each point x in space. Notice
that the concept of position has been relegated from a dynamical variable in particle
mechanics to a mere label in field theory.
An Example: The Electromagnetic Field
The most familiar examples of fields from classical physics are the electric and magnetic
fields,

E(x, t) and

B(x, t). Both of these fields are spatial 3-vectors. In a more sophis-
ticated treatement of electromagnetism, we derive these two 3-vectors from a single
4-component field A
µ
(x, t) = (φ,

A) where µ = 0, 1, 2, 3 shows that this field is a vector
in spacetime. The electric and magnetic fields are given by

E = −∇φ −


A
∂t
and

B = ∇

A (1.2)
which ensure that two of Maxwell’s equations, ∇

B = 0 and d

B/dt = −∇

E, hold
immediately as identities.
The Lagrangian
The dynamics of the field is governed by a Lagrangian which is a function of φ(x, t),
˙
φ(x, t) and ∇φ(x, t). In all the systems we study in this course, the Lagrangian is of
the form,
L(t) =
_
d
3
x L(φ
a
, ∂
µ
φ
a
) (1.3)
– 7 –
where the official name for L is the Lagrangian density, although everyone simply calls
it the Lagrangian. The action is,
S =
_
t
2
t
1
dt
_
d
3
x L =
_
d
4
x L (1.4)
Recall that in particle mechanics L depends on q and ˙ q, but not ¨ q. In field theory
we similarly restrict to Lagrangians L depending on φ and
˙
φ, and not
¨
φ. In principle,
there’s nothing to stop L depending on ∇φ, ∇
2
φ, ∇
3
φ, etc. However, with an eye to
later Lorentz invariance, we will only consider Lagrangians depending on ∇φ and not
higher derivatives.
We can determine the equations of motion by the principle of least action. We vary
the path, keeping the end points fixed and require δS = 0,
δS =
_
d
4
x
_
∂L
∂φ
a
δφ
a
+
∂L
∂(∂
µ
φ
a
)
δ(∂
µ
φ
a
)
_
=
_
d
4
x
_
∂L
∂φ
a
−∂
µ
_
∂L
∂(∂
µ
φ
a
)
__
δφ
a
+ ∂
µ
_
∂L
∂(∂
µ
φ
a
)
δφ
a
_
(1.5)
The last term is a total derivative and vanishes for any δφ
a
(x, t) that decays at spatial
infinity and obeys δφ
a
(x, t
1
) = δφ
a
(x, t
2
) = 0. Requiring δS = 0 for all such paths
yields the Euler-Lagrange equations of motion for the fields φ
a
,

µ
_
∂L
∂(∂
µ
φ
a
)
_

∂L
∂φ
a
= 0 (1.6)
1.1.1 An Example: The Klein-Gordon Equation
Consider the Lagrangian for a real scalar field φ(x, t),
L =
1
2
η
µν

µ
φ∂
ν
φ −
1
2
m
2
φ
2
(1.7)
=
1
2
˙
φ
2

1
2
(∇φ)
2

1
2
m
2
φ
2
where we are using the Minkowski space metric
η
µν
= η
µν
=
_
+1
−1
−1
−1
_
(1.8)
Comparing (1.7) to the usual expression for the Lagrangian L = T −V , we identify the
kinetic energy of the field as
T =
_
d
3
x
1
2
˙
φ
2
(1.9)
– 8 –
and the potential energy of the field as
V =
_
d
3
x
1
2
(∇φ)
2
+
1
2
m
2
φ
2
(1.10)
The first term in this expression is called the gradient energy, while the phrase “poten-
tial energy”, or just “potential”, is usually reserved for the last term.
To determine the equations of motion arising from (1.7), we compute
∂L
∂φ
= −m
2
φ and
∂L
∂(∂
µ
φ)
= ∂
µ
φ ≡ (
˙
φ, −∇φ) (1.11)
The Euler-Lagrange equation is then
¨
φ −∇
2
φ + m
2
φ = 0 (1.12)
which we can write in relativistic form as

µ

µ
φ + m
2
φ = 0 (1.13)
This is the Klein-Gordon Equation. The Laplacian in Minkowski space is sometimes
denoted by . In this notation, the Klein-Gordon equation reads φ + m
2
φ = 0.
An obvious generalization of the Klein-Gordon equation comes from considering the
Lagrangian with arbitrary potential V (φ),
L =
1
2

µ
φ∂
µ
φ −V (φ) ⇒ ∂
µ

µ
φ +
∂V
∂φ
= 0 (1.14)
1.1.2 Another Example: First Order Lagrangians
We could also consider a Lagrangian that is linear in time derivatives, rather than
quadratic. Take a complex scalar field ψ whose dynamics is defined by the real La-
grangian
L =
i
2


˙
ψ −
˙
ψ

ψ) −∇ψ

∇ψ −mψ

ψ (1.15)
We can determine the equations of motion by treating ψ and ψ

as independent objects,
so that
∂L
∂ψ

=
i
2
˙
ψ −mψ and
∂L

˙
ψ

= −
i
2
ψ and
∂L
∂∇ψ

= −∇ψ (1.16)
This gives us the equation of motion
i
∂ψ
∂t
= −∇
2
ψ + mψ (1.17)
This looks very much like the Schr¨odinger equation. Except it isn’t! Or, at least, the
interpretation of this equation is very different: the field ψ is a classical field with none
of the probability interpretation of the wavefunction. We’ll come back to this point in
Section 2.8.
– 9 –
The initial data required on a Cauchy surface differs for the two examples above.
When L ∼
˙
φ
2
, both φ and
˙
φ must be specified to determine the future evolution;
however when L ∼ ψ

˙
ψ, only ψ and ψ

are needed.
1.1.3 A Final Example: Maxwell’s Equations
We may derive Maxwell’s equations in the vacuum from the Lagrangian,
L = −
1
2
(∂
µ
A
ν
) (∂
µ
A
ν
) +
1
2
(∂
µ
A
µ
)
2
(1.18)
Notice the funny minus signs! This is to ensure that the kinetic terms for A
i
are positive
using the Minkowski space metric (1.8), so L ∼
1
2
˙
A
2
i
. The Lagrangian (1.18) has no
kinetic term
˙
A
2
0
for A
0
. We will see the consequences of this in Section 6. To see that
Maxwell’s equations indeed follow from (1.18), we compute
∂L
∂(∂
µ
A
ν
)
= −∂
µ
A
ν
+ (∂
ρ
A
ρ
) η
µν
(1.19)
from which we may derive the equations of motion,

µ
_
∂L
∂(∂
µ
A
ν
)
_
= −∂
2
A
ν
+ ∂
ν
(∂
ρ
A
ρ
) = −∂
µ
(∂
µ
A
ν
−∂
ν
A
µ
) ≡ −∂
µ
F
µν
(1.20)
where the field strength is defined by F
µν
= ∂
µ
A
ν
− ∂
ν
A
µ
. You can check using (1.2)
that this reproduces the remaining two Maxwell’s equations in a vacuum: ∇

E = 0
and ∂

E/∂t = ∇

B. Using the notation of the field strength, we may rewrite the
Maxwell Lagrangian (up to an integration by parts) in the compact form
L = −
1
4
F
µν
F
µν
(1.21)
1.1.4 Locality, Locality, Locality
In each of the examples above, the Lagrangian is local. This means that there are no
terms in the Lagrangian coupling φ(x, t) directly to φ(y, t) with x ,= y. For example,
there are no terms that look like
L =
_
d
3
xd
3
y φ(x)φ(y) (1.22)
A priori, there’s no reason for this. After all, x is merely a label, and we’re quite
happy to couple other labels together (for example, the term ∂
3
A
0

0
A
3
in the Maxwell
Lagrangian couples the µ = 0 field to the µ = 3 field). But the closest we get for the
x label is a coupling between φ(x) and φ(x + δx) through the gradient term (∇φ)
2
.
This property of locality is, as far as we know, a key feature of all theories of Nature.
Indeed, one of the main reasons for introducing field theories in classical physics is to
implement locality. In this course, we will only consider local Lagrangians.
– 10 –
1.2 Lorentz Invariance
The laws of Nature are relativistic, and one of the main motivations to develop quantum
field theory is to reconcile quantum mechanics with special relativity. To this end, we
want to construct field theories in which space and time are placed on an equal footing
and the theory is invariant under Lorentz transformations,
x
µ
−→(x

)
µ
= Λ
µ
ν
x
ν
(1.23)
where Λ
µ
ν
satisfies
Λ
µ
σ
η
στ
Λ
ν
τ
= η
µν
(1.24)
For example, a rotation by θ about the x
3
-axis, and a boost by v < 1 along the x
1
-axis
are respectively described by the Lorentz transformations
Λ
µ
ν
=
_
_
_
_
_
_
1 0 0 0
0 cos θ −sin θ 0
0 sin θ cos θ 0
0 0 0 1
_
_
_
_
_
_
and Λ
µ
ν
=
_
_
_
_
_
_
γ −γv 0 0
−γv γ 0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
_
(1.25)
with γ = 1/

1 −v
2
. The Lorentz transformations form a Lie group under matrix
multiplication. You’ll learn more about this in the “Symmetries and Particle Physics”
course.
The Lorentz transformations have a representation on the fields. The simplest ex-
ample is the scalar field which, under the Lorentz transformation x →Λx, transforms
as
φ(x) →φ

(x) = φ(Λ
−1
x) (1.26)
The inverse Λ
−1
appears in the argument because we are dealing with an active trans-
formation in which the field is truly shifted. To see why this means that the inverse
appears, it will suffice to consider a non-relativistic example such as a temperature field.
Suppose we start with an initial field φ(x) which has a hotspot at, say, x = (1, 0, 0).
After a rotation x →Rx about the z-axis, the new field φ

(x) will have the hotspot at
x = (0, 1, 0). If we want to express φ

(x) in terms of the old field φ, we need to place
ourselves at x = (0, 1, 0) and ask what the old field looked like where we’ve come from
at R
−1
(0, 1, 0) = (1, 0, 0). This R
−1
is the origin of the inverse transformation. (If we
were instead dealing with a passive transformation in which we relabel our choice of
coordinates, we would have instead φ(x) →φ

(x) = φ(Λx)).
– 11 –
The definition of a Lorentz invariant theory is that if φ(x) solves the equations of
motion then φ(Λ
−1
x) also solves the equations of motion. We can ensure that this
property holds by requiring that the action is Lorentz invariant. Let’s look at our
examples:
Example 1: The Klein-Gordon Equation
For a real scalar field we have φ(x) → φ

(x) = φ(Λ
−1
x). The derivative of the scalar
field transforms as a vector, meaning
(∂
µ
φ)(x) →(Λ
−1
)
ν
µ
(∂
ν
φ)(y)
where y = Λ
−1
x. This means that the derivative terms in the Lagrangian density
transform as
L
deriv
(x) = ∂
µ
φ(x)∂
ν
φ(x)η
µν
−→ (Λ
−1
)
ρ
µ
(∂
ρ
φ)(y) (Λ
−1
)
σ
ν
(∂
σ
φ)(y) η
µν
= (∂
ρ
φ)(y) (∂
σ
φ)(y) η
ρσ
= L
deriv
(y) (1.27)
The potential terms transform in the same way, with φ
2
(x) → φ
2
(y). Putting this all
together, we find that the action is indeed invariant under Lorentz transformations,
S =
_
d
4
x L(x) −→
_
d
4
x L(y) =
_
d
4
y L(y) = S (1.28)
where, in the last step, we need the fact that we don’t pick up a Jacobian factor when
we change integration variables from
_
d
4
x to
_
d
4
y. This follows because det Λ = 1.
(At least for Lorentz transformation connected to the identity which, for now, is all we
deal with).
Example 2: First Order Dynamics
In the first-order Lagrangian (1.15), space and time are not on the same footing. (L
is linear in time derivatives, but quadratic in spatial derivatives). The theory is not
Lorentz invariant.
In practice, it’s easy to see if the action is Lorentz invariant: just make sure all
the Lorentz indices µ = 0, 1, 2, 3 are contracted with Lorentz invariant objects, such
as the metric η
µν
. Other Lorentz invariant objects you can use include the totally
antisymmetric tensor ǫ
µνρσ
and the matrices γ
µ
that we will introduce when we come
to discuss spinors in Section 4.
– 12 –
Example 3: Maxwell’s Equations
Under a Lorentz transformation A
µ
(x) →Λ
µ
ν
A
ν

−1
x). You can check that Maxwell’s
Lagrangian (1.21) is indeed invariant. Of course, historically electrodynamics was the
first Lorentz invariant theory to be discovered: it was found even before the concept of
Lorentz invariance.
1.3 Symmetries
The role of symmetries in field theory is possibly even more important than in particle
mechanics. There are Lorentz symmetries, internal symmetries, gauge symmetries,
supersymmetries.... We start here by recasting Noether’s theorem in a field theoretic
framework.
1.3.1 Noether’s Theorem
Every continuous symmetry of the Lagrangian gives rise to a conserved current j
µ
(x)
such that the equations of motion imply

µ
j
µ
= 0 (1.29)
or, in other words, ∂j
0
/∂t +∇

j = 0.
A Comment: A conserved current implies a conserved charge Q, defined as
Q =
_
R
3
d
3
x j
0
(1.30)
which one can immediately see by taking the time derivative,
dQ
dt
=
_
R
3
d
3
x
∂j
∂t
0
= −
_
R
3
d
3
x ∇

j = 0 (1.31)
assuming that

j → 0 sufficiently quickly as [x[ → ∞. However, the existence of a
current is a much stronger statement than the existence of a conserved charge because
it implies that charge is conserved locally. To see this, we can define the charge in a
finite volume V ,
Q
V
=
_
V
d
3
x j
0
(1.32)
Repeating the analysis above, we find that
dQ
V
dt
= −
_
V
d
3
x ∇

j = −
_
A

j d

S (1.33)
– 13 –
where A is the area bounding V and we have used Stokes’ theorem. This equation
means that any charge leaving V must be accounted for by a flow of the current 3-
vector

j out of the volume. This kind of local conservation of charge holds in any local
field theory.
Proof of Noether’s Theorem: We’ll prove the theorem by working infinitesimally.
We may always do this if we have a continuous symmetry. We say that the transfor-
mation
δφ
a
(x) = X
a
(φ) (1.34)
is a symmetry if the Lagrangian changes by a total derivative,
δL = ∂
µ
F
µ
(1.35)
for some set of functions F
µ
(φ). To derive Noether’s theorem, we first consider making
an arbitrary transformation of the fields δφ
a
. Then
δL =
∂L
∂φ
a
δφ
a
+
∂L
∂(∂
µ
φ
a
)

µ
(δφ
a
)
=
_
∂L
∂φ
a
−∂
µ
∂L
∂(∂
µ
φ
a
)
_
δφ
a
+ ∂
µ
_
∂L
∂(∂
µ
φ
a
)
δφ
a
_
(1.36)
When the equations of motion are satisfied, the term in square brackets vanishes. So
we’re left with
δL = ∂
µ
_
∂L
∂(∂
µ
φ
a
)
δφ
a
_
(1.37)
But for the symmetry transformation δφ
a
= X
a
(φ), we have by definition δL = ∂
µ
F
µ
.
Equating this expression with (1.37) gives us the result

µ
j
µ
= 0 with j
µ
=
∂L
∂(∂
µ
φ
a
)
X
a
(φ) −F
µ
(φ) (1.38)
1.3.2 An Example: Translations and the Energy-Momentum Tensor
Recall that in classical particle mechanics, invariance under spatial translations gives
rise to the conservation of momentum, while invariance under time translations is
responsible for the conservation of energy. We will now see something similar in field
theories. Consider the infinitesimal translation
x
ν
→x
ν
−ǫ
ν
⇒ φ
a
(x) →φ
a
(x) + ǫ
ν

ν
φ
a
(x) (1.39)
– 14 –
(where the sign in the field transformation is plus, instead of minus, because we’re doing
an active, as opposed to passive, transformation). Similarly, the Lagrangian transforms
as
L(x) →L(x) + ǫ
ν

ν
L(x) (1.40)
(This is the correct transformation for a Lagrangian that has no explicit x dependence,
but only depends on x through the fields φ
a
(x). All theories that we consider in this
course will have this property). Since the change in the Lagrangian is a total derivative,
we may invoke Noether’s theorem which gives us four conserved currents (j
µ
)
ν
, one for
each of the translations ǫ
ν
with ν = 0, 1, 2, 3,
(j
µ
)
ν
=
∂L
∂(∂
µ
φ
a
)

ν
φ
a
−δ
µ
ν
L ≡ T
µ
ν
(1.41)
T
µ
ν
is called the energy-momentum tensor. It satisfies

µ
T
µ
ν
= 0 (1.42)
The four conserved quantities are given by
E =
_
d
3
x T
00
and P
i
=
_
d
3
x T
0i
(1.43)
where E is the total energy of the field configuration, while P
i
is the total momentum
of the field configuration.
An Example of the Energy-Momentum Tensor
Consider the simplest scalar field theory with Lagrangian (1.7). From the above dis-
cussion, we can compute
T
µν
= ∂
µ
φ ∂
ν
φ −η
µν
L (1.44)
One can verify using the equation of motion for φ that this expression indeed satisfies

µ
T
µν
= 0. For this example, the conserved energy and momentum are given by
E =
_
d
3
x
1
2
˙
φ
2
+
1
2
(∇φ)
2
+
1
2
m
2
φ
2
(1.45)
P
i
=
_
d
3
x
˙
φ∂
i
φ (1.46)
Notice that for this example, T
µν
came out symmetric, so that T
µν
= T
νµ
. This
won’t always be the case. Nevertheless, there is typically a way to massage the energy
momentum tensor of any theory into a symmetric form by adding an extra term
Θ
µν
= T
µν
+ ∂
ρ
Γ
ρµν
(1.47)
– 15 –
where Γ
ρµν
is some function of the fields that is anti-symmetric in the first two indices so
Γ
ρµν
= −Γ
µρν
. This guarantees that ∂
µ

ρ
Γ
ρµν
= 0 so that the new energy-momentum
tensor is also a conserved current.
A Cute Trick
One reason that you may want a symmetric energy-momentum tensor is to make con-
tact with general relativity: such an object sits on the right-hand side of Einstein’s
field equations. In fact this observation provides a quick and easy way to determine a
symmetric energy-momentum tensor. Firstly consider coupling the theory to a curved
background spacetime, introducing an arbitrary metric g
µν
(x) in place of η
µν
, and re-
placing the kinetic terms with suitable covariant derivatives using “minimal coupling”.
Then a symmetric energy momentum tensor in the flat space theory is given by
Θ
µν
= −
2

−g
∂(

−gL)
∂g
µν
¸
¸
¸
¸
gµν=ηµν
(1.48)
It should be noted however that this trick requires a little more care when working
with spinors.
1.3.3 Another Example: Lorentz Transformations and Angular Momentum
In classical particle mechanics, rotational invariance gave rise to conservation of angular
momentum. What is the analogy in field theory? Moreover, we now have further
Lorentz transformations, namely boosts. What conserved quantity do they correspond
to? To answer these questions, we first need the infinitesimal form of the Lorentz
transformations
Λ
µ
ν
= δ
µ
ν
+ ω
µ
ν
(1.49)
where ω
µ
ν
is infinitesimal. The condition (1.24) for Λ to be a Lorentz transformation
becomes

µ
σ
+ ω
µ
σ
)(δ
ν
τ
+ ω
ν
τ
) η
στ
= η
µν
⇒ ω
µν
+ ω
νµ
= 0 (1.50)
So the infinitesimal form ω
µν
of the Lorentz transformation must be an anti-symmetric
matrix. As a check, the number of different 44 anti-symmetric matrices is 43/2 = 6,
which agrees with the number of different Lorentz transformations (3 rotations + 3
boosts). Now the transformation on a scalar field is given by
φ(x) →φ

(x) = φ(Λ
−1
x)
= φ(x
µ
−ω
µ
ν
x
ν
)
= φ(x
µ
) −ω
µ
ν
x
ν

µ
φ(x) (1.51)
– 16 –
from which we see that
δφ = −ω
µ
ν
x
ν

µ
φ (1.52)
By the same argument, the Lagrangian density transforms as
δL = −ω
µ
ν
x
ν

µ
L = −∂
µ

µ
ν
x
ν
L) (1.53)
where the last equality follows because ω
µ
µ
= 0 due to anti-symmetry. Once again,
the Lagrangian changes by a total derivative so we may apply Noether’s theorem (now
with F
µ
= −ω
µ
ν
x
ν
L) to find the conserved current
j
µ
= −
∂L
∂(∂
µ
φ)
ω
ρ
ν
x
ν

ρ
φ + ω
µ
ν
x
ν
L
= −ω
ρ
ν
_
∂L
∂(∂
µ
φ)
x
ν

ρ
φ −δ
µ
ρ
x
ν
L
_
= −ω
ρ
ν
T
µ
ρ
x
ν
(1.54)
Unlike in the previous example, I’ve left the infinitesimal choice of ω
µ
ν
in the expression
for this current. But really, we should strip it out to give six different currents, i.e. one
for each choice of ω
µ
ν
. We can write them as

µ
)
ρσ
= x
ρ
T
µσ
−x
σ
T
µρ
(1.55)
which satisfy ∂
µ

µ
)
ρσ
= 0 and give rise to 6 conserved charges. For ρ, σ = 1, 2, 3,
the Lorentz transformation is a rotation and the three conserved charges give the total
angular momentum of the field.
Q
ij
=
_
d
3
x (x
i
T
0j
−x
j
T
0i
) (1.56)
But what about the boosts? In this case, the conserved charges are
Q
0i
=
_
d
3
x (x
0
T
0i
−x
i
T
00
) (1.57)
The fact that these are conserved tells us that
0 =
dQ
0i
dt
=
_
d
3
x T
0i
+ t
_
d
3
x
∂T
0i
∂t

d
dt
_
d
3
x x
i
T
00
= P
i
+ t
dP
i
dt

d
dt
_
d
3
x x
i
T
00
(1.58)
But we know that P
i
is conserved, so dP
i
/dt = 0, leaving us with the following conse-
quence of invariance under boosts:
d
dt
_
d
3
x x
i
T
00
= constant (1.59)
This is the statement that the center of energy of the field travels with a constant
velocity. It’s kind of like a field theoretic version of Newton’s first law but, rather
surprisingly, appearing here as a conservation law.
– 17 –
1.3.4 Internal Symmetries
The above two examples involved transformations of spacetime, as well as transforma-
tions of the field. An internal symmetry is one that only involves a transformation of
the fields and acts the same at every point in spacetime. The simplest example occurs
for a complex scalar field ψ(x) = (φ
1
(x) +iφ
2
(x))/

2. We can build a real Lagrangian
by
L = ∂
µ
ψ


µ
ψ −V ([ψ[
2
) (1.60)
where the potential is a general polynomial in [ψ[
2
= ψ

ψ. To find the equations of
motion, we could expand ψ in terms of φ
1
and φ
2
and work as before. However, it’s
easier (and equivalent) to treat ψ and ψ

as independent variables and vary the action
with respect to both of them. For example, varying with respect to ψ

leads to the
equation of motion

µ

µ
ψ +
∂V (ψ

ψ)
∂ψ

= 0 (1.61)
The Lagrangian has a continuous symmetry which rotates φ
1
and φ
2
or, equivalently,
rotates the phase of ψ:
ψ →e

ψ or δψ = iαψ (1.62)
where the latter equation holds with α infinitesimal. The Lagrangian remains invariant
under this change: δL = 0. The associated conserved current is
j
µ
= i(∂
µ
ψ

)ψ −iψ

(∂
µ
ψ) (1.63)
We will later see that the conserved charges arising from currents of this type have
the interpretation of electric charge or particle number (for example, baryon or lepton
number).
Non-Abelian Internal Symmetries
Consider a theory involving N scalar fields φ
a
, all with the same mass and the La-
grangian
L =
1
2
N

a=1

µ
φ
a

µ
φ
a

1
2
N

a=1
m
2
φ
2
a
−g
_
N

a=1
φ
2
a
_
2
(1.64)
In this case the Lagrangian is invariant under the non-Abelian symmetry group G =
SO(N). (Actually O(N) in this case). One can construct theories from complex fields
in a similar manner that are invariant under an SU(N) symmetry group. Non-Abelian
symmetries of this type are often referred to as global symmetries to distinguish them
from the “local gauge” symmetries that you will meet later. Isospin is an example of
such a symmetry, albeit realized only approximately in Nature.
– 18 –
Another Cute Trick
There is a quick method to determine the conserved current associated to an internal
symmetry δφ = αφ for which the Lagrangian is invariant. Here, α is a constant real
number. (We may generalize the discussion easily to a non-Abelian internal symmetry
for which α becomes a matrix). Now consider performing the transformation but where
α depends on spacetime: α = α(x). The action is no longer invariant. However, the
change must be of the form
δL = (∂
µ
α) h
µ
(φ) (1.65)
since we know that δL = 0 when α is constant. The change in the action is therefore
δS =
_
d
4
x δL = −
_
d
4
x α(x) ∂
µ
h
µ
(1.66)
which means that when the equations of motion are satisfied (so δS = 0 for all varia-
tions, including δφ = α(x)φ) we have

µ
h
µ
= 0 (1.67)
We see that we can identify the function h
µ
= j
µ
as the conserved current. This way
of viewing things emphasizes that it is the derivative terms, not the potential terms,
in the action that contribute to the current. (The potential terms are invariant even
when α = α(x)).
1.4 The Hamiltonian Formalism
The link between the Lagrangian formalism and the quantum theory goes via the path
integral. In this course we will not discuss path integral methods, and focus instead
on canonical quantization. For this we need the Hamiltonian formalism of field theory.
We start by defining the momentum π
a
(x) conjugate to φ
a
(x),
π
a
(x) =
∂L

˙
φ
a
(1.68)
The conjugate momentum π
a
(x) is a function of x, just like the field φ
a
(x) itself. It
is not to be confused with the total momentum P
i
defined in (1.43) which is a single
number characterizing the whole field configuration. The Hamiltonian density is given
by
H = π
a
(x)
˙
φ
a
(x) −L(x) (1.69)
where, as in classical mechanics, we eliminate
˙
φ
a
(x) in favour of π
a
(x) everywhere in
H. The Hamiltonian is then simply
H =
_
d
3
x H (1.70)
– 19 –
An Example: A Real Scalar Field
For the Lagrangian
L =
1
2
˙
φ
2

1
2
(∇φ)
2
−V (φ) (1.71)
the momentum is given by π =
˙
φ, which gives us the Hamiltonian,
H =
_
d
3
x
1
2
π
2
+
1
2
(∇φ)
2
+ V (φ) (1.72)
Notice that the Hamiltonian agrees with the definition of the total energy (1.45) that
we get from applying Noether’s theorem for time translation invariance.
In the Lagrangian formalism, Lorentz invariance is clear for all to see since the action
is invariant under Lorentz transformations. In contrast, the Hamiltonian formalism is
not manifestly Lorentz invariant: we have picked a preferred time. For example, the
equations of motion for φ(x) = φ(x, t) arise from Hamilton’s equations,
˙
φ(x, t) =
∂H
∂π(x, t)
and ˙ π(x, t) = −
∂H
∂φ(x, t)
(1.73)
which, unlike the Euler-Lagrange equations (1.6), do not look Lorentz invariant. Nev-
ertheless, even though the Hamiltonian framework doesn’t look Lorentz invariant, the
physics must remain unchanged. If we start from a relativistic theory, all final answers
must be Lorentz invariant even if it’s not manifest at intermediate steps. We will pause
at several points along the quantum route to check that this is indeed the case.
– 20 –
2. Free Fields
“The career of a young theoretical physicist consists of treating the harmonic
oscillator in ever-increasing levels of abstraction.”
Sidney Coleman
2.1 Canonical Quantization
In quantum mechanics, canonical quantization is a recipe that takes us from the Hamil-
tonian formalism of classical dynamics to the quantum theory. The recipe tells us to
take the generalized coordinates q
a
and their conjugate momenta p
a
and promote them
to operators. The Poisson bracket structure of classical mechanics morphs into the
structure of commutation relations between operators, so that, in units with = 1,
[q
a
, q
b
] = [p
a
, p
b
] = 0
[q
a
, p
b
] = i δ
b
a
(2.1)
In field theory we do the same, now for the field φ
a
(x) and its momentum conjugate
π
b
(x). Thus a quantum field is an operator valued function of space obeying the com-
mutation relations

a
(x), φ
b
(y)] = [π
a
(x), π
b
(y)] = 0

a
(x), π
b
(y)] = iδ
(3)
(x −y) δ
b
a
(2.2)
Note that we’ve lost all track of Lorentz invariance since we have separated space x
and time t. We are working in the Schr¨odinger picture so that the operators φ
a
(x) and
π
a
(x) do not depend on time at all — only on space. All time dependence sits in the
states [ψ) which evolve by the usual Schr¨odinger equation
i
d[ψ)
dt
= H[ψ) (2.3)
We aren’t doing anything different from usual quantum mechanics; we’re merely apply-
ing the old formalism to fields. Be warned however that the notation [ψ) for the state
is deceptively simple: if you were to write the wavefunction in quantum field theory, it
would be a functional, that is a function of every possible configuration of the field φ.
The typical information we want to know about a quantum theory is the spectrum of
the Hamiltonian H. In quantum field theories, this is usually very hard. One reason for
this is that we have an infinite number of degrees of freedom — at least one for every
point x in space. However, for certain theories — known as free theories — we can find
a way to write the dynamics such that each degree of freedom evolves independently
– 21 –
from all the others. Free field theories typically have Lagrangians which are quadratic
in the fields, so that the equations of motion are linear. For example, the simplest
relativistic free theory is the classical Klein-Gordon (KG) equation for a real scalar
field φ(x, t),

µ

µ
φ + m
2
φ = 0 (2.4)
To exhibit the coordinates in which the degrees of freedom decouple from each other,
we need only take the Fourier transform,
φ(x, t) =
_
d
3
p
(2π)
3
e
i p·x
φ( p, t) (2.5)
Then φ( p, t) satisfies
_

2
∂t
2
+ ( p
2
+ m
2
)
_
φ( p, t) = 0 (2.6)
Thus, for each value of p, φ( p, t) solves the equation of a harmonic oscillator vibrating
at frequency
ω
p
= +
_
p
2
+ m
2
(2.7)
We learn that the most general solution to the KG equation is a linear superposition of
simple harmonic oscillators, each vibrating at a different frequency with a different am-
plitude. To quantize φ(x, t) we must simply quantize this infinite number of harmonic
oscillators. Let’s recall how to do this.
2.1.1 The Simple Harmonic Oscillator
Consider the quantum mechanical Hamiltonian
H =
1
2
p
2
+
1
2
ω
2
q
2
(2.8)
with the canonical commutation relations [q, p] = i. To find the spectrum we define
the creation and annihilation operators (also known as raising/lowering operators, or
sometimes ladder operators)
a =
_
ω
2
q +
i


p , a

=
_
ω
2
q −
i


p (2.9)
which can be easily inverted to give
q =
1


(a + a

) , p = −i
_
ω
2
(a −a

) (2.10)
– 22 –
Substituting into the above expressions we find
[a, a

] = 1 (2.11)
while the Hamiltonian is given by
H =
1
2
ω(aa

+ a

a)
= ω(a

a +
1
2
) (2.12)
One can easily confirm that the commutators between the Hamiltonian and the creation
and annihilation operators are given by
[H, a

] = ωa

and [H, a] = −ωa (2.13)
These relations ensure that a and a

take us between energy eigenstates. Let [E) be
an eigenstate with energy E, so that H[E) = E [E). Then we can construct more
eigenstates by acting with a and a

,
Ha

[E) = (E + ω)a

[E) , Ha [E) = (E −ω)a [E) (2.14)
So we find that the system has a ladder of states with energies
. . . , E −ω, E, E + ω, E + 2ω, . . . (2.15)
If the energy is bounded below, there must be a ground state [0) which satisfies a [0) = 0.
This has ground state energy (also known as zero point energy),
H[0) =
1
2
ω [0) (2.16)
Excited states then arise from repeated application of a

,
[n) = (a

)
n
[0) with H[n) = (n +
1
2
)ω [n) (2.17)
where I’ve ignored the normalization of these states so, ¸n[ n) , = 1.
2.2 The Free Scalar Field
We now apply the quantization of the harmonic oscillator to the free scalar field. We
write φ and π as a linear sum of an infinite number of creation and annihilation oper-
ators a

p
and a
p
, indexed by the 3-momentum p,
φ(x) =
_
d
3
p
(2π)
3
1
_

p
_
a
p
e
i p·x
+ a

p
e
−i p·x
_
(2.18)
π(x) =
_
d
3
p
(2π)
3
(−i)
_
ω
p
2
_
a
p
e
i p·x
−a

p
e
−i p·x
_
(2.19)
– 23 –
Claim: The commutation relations for φ and π are equivalent to the following com-
mutation relations for a
p
and a

p
[φ(x), φ(y)] = [π(x), π(y)] = 0
[φ(x), π(y)] = iδ
(3)
(x −y)

[a
p
, a
q
] = [a

p
, a

q
] = 0
[a
p
, a

q
] = (2π)
3
δ
(3)
( p −q)
(2.20)
Proof: We’ll show this just one way. Assume that [a
p
, a

q
] = (2π)
3
δ
(3)
( p −q). Then
[φ(x), π(y)] =
_
d
3
p d
3
q
(2π)
6
(−i)
2
_
ω
q
ω
p
_
−[a
p
, a

q
] e
i p·x−i q· y
+ [a

p
, a
q
] e
−i p·x+i q· y
_
=
_
d
3
p
(2π)
3
(−i)
2
_
−e
i p·(x− y)
−e
i p·( y−x)
_
(2.21)
= iδ
(3)
(x −y)
The Hamiltonian
Let’s now compute the Hamiltonian in terms of a
p
and a

p
. We have
H =
1
2
_
d
3
x π
2
+ (∇φ)
2
+ m
2
φ
2
=
1
2
_
d
3
x d
3
p d
3
q
(2π)
6
_


ω
p
ω
q
2
(a
p
e
i p·x
−a

p
e
−i p·x
)(a
q
e
i q·x
−a

q
e
−i q·x
)
+
1
2

ω
p
ω
q
(i p a
p
e
i p·x
−i p a

p
e
−i p·x
) (iq a
q
e
i q·x
−iq a

q
e
−i q·x
)
+
m
2
2

ω
p
ω
q
(a
p
e
i p·x
+ a

p
e
−i p·x
)(a
q
e
i q·x
+ a

q
e
−i q·x
)
_
=
1
4
_
d
3
p
(2π)
3
1
ω
p
_
(−ω
2
p
+ p
2
+ m
2
)(a
p
a
− p
+ a

p
a

− p
) + (ω
2
p
+ p
2
+ m
2
)(a
p
a

p
+ a

p
a
p
)
_
where in the second line we’ve used the expressions for φ and π given in (2.18) and
(2.19); to get to the third line we’ve integrated over d
3
x to get delta-functions δ
(3)
( p±q)
which, in turn, allow us to perform the d
3
q integral. Now using the expression for the
frequency ω
2
p
= p
2
+ m
2
, the first term vanishes and we’re left with
H =
1
2
_
d
3
p
(2π)
3
ω
p
_
a
p
a

p
+ a

p
a
p
_
=
_
d
3
p
(2π)
3
ω
p
_
a

p
a
p
+
1
2
(2π)
3
δ
(3)
(0)
_
(2.22)
– 24 –
Hmmmm. We’ve found a delta-function, evaluated at zero where it has its infinite
spike. Moreover, the integral over ω
p
diverges at large p. What to do? Let’s start by
looking at the ground state where this infinity first becomes apparent.
2.3 The Vacuum
Following our procedure for the harmonic oscillator, let’s define the vacuum [0) by
insisting that it is annihilated by all a
p
,
a
p
[0) = 0 ∀ p (2.23)
With this definition, the energy E
0
of the ground state comes from the second term in
(2.22),
H [0) ≡ E
0
[0) =
__
d
3
p
1
2
ω
p
δ
(3)
(0)
_
[ 0) = ∞[0) (2.24)
The subject of quantum field theory is rife with infinities. Each tells us something
important, usually that we’re doing something wrong, or asking the wrong question.
Let’s take some time to explore where this infinity comes from and how we should deal
with it.
In fact there are two different ∞’s lurking in the expression (2.24). The first arises
because space is infinitely large. (Infinities of this type are often referred to as infra-red
divergences although in this case the ∞ is so simple that it barely deserves this name).
To extract out this infinity, let’s consider putting the theory in a box with sides of
length L. We impose periodic boundary conditions on the field. Then, taking the limit
where L →∞, we get
(2π)
3
δ
(3)
(0) = lim
L→∞
_
L/2
−L/2
d
3
x e
ix· p
¸
¸
p=0
= lim
L→∞
_
L/2
−L/2
d
3
x = V (2.25)
where V is the volume of the box. So the δ(0) divergence arises because we’re computing
the total energy, rather than the energy density c
0
. To find c
0
we can simply divide by
the volume,
c
0
=
E
0
V
=
_
d
3
p
(2π)
3
1
2
ω
p
(2.26)
which is still infinite. We recognize it as the sum of ground state energies for each
harmonic oscillator. But c
0
→ ∞ due to the [ p[ → ∞ limit of the integral. This is
a high frequency — or short distance — infinity known as an ultra-violet divergence.
This divergence arises because of our hubris. We’ve assumed that our theory is valid
to arbitrarily short distance scales, corresponding to arbitrarily high energies. This is
clearly absurd. The integral should be cut-off at high momentum in order to reflect
the fact that our theory is likely to break down in some way.
– 25 –
We can deal with the infinity in (2.24) in a more practical way. In physics we’re
only interested in energy differences. There’s no way to measure E
0
directly, so we can
simply redefine the Hamiltonian by subtracting off this infinity,
H =
_
d
3
p
(2π)
3
ω
p
a

p
a
p
(2.27)
so that, with this new definition, H[0) = 0. In fact, the difference between this Hamilto-
nian and the previous one is merely an ordering ambiguity in moving from the classical
theory to the quantum theory. For example, if we defined the Hamiltonian of the har-
monic oscillator to be H = (1/2)(ωq −ip)(ωq +ip), which is classically the same as our
original choice, then upon quantization it would naturally give H = ωa

a as in (2.27).
This type of ordering ambiguity arises a lot in field theories. We’ll come across a number
of ways of dealing with it. The method that we’ve used above is called normal ordering.
Definition: We write the normal ordered string of operators φ
1
(x
1
) . . . φ
n
(x
n
) as
: φ
1
(x
1
) . . . φ
n
(x
n
) : (2.28)
It is defined to be the usual product with all annihilation operators a
p
placed to the
right. So, for the Hamiltonian, we could write (2.27) as
: H :=
_
d
3
p
(2π)
3
ω
p
a

p
a
p
(2.29)
In the remainder of this section, we will normal order all operators in this manner.
2.3.1 The Cosmological Constant
Above I wrote “there’s no way to measure E
0
directly”. There is a BIG caveat here:
gravity is supposed to see everything! The sum of all the zero point energies should
contribute to the stress-energy tensor that appears on the right-hand side of Einstein’s
equations. We expect them to appear as a cosmological constant Λ = E
0
/V ,
R
µν

1
2
g
µν
= −8πGT
µν
+ Λg
µν
(2.30)
Current observation suggests that 70% of the energy density in the universe has the
properties of a cosmological constant with Λ ∼ (10
−3
eV )
4
. This is much smaller than
other scales in particle physics. In particular, the Standard Model is valid at least up
to 10
12
eV . Why don’t the zero point energies of these fields contribute to Λ? Or, if
they do, what cancels them to such high accuracy? This is the cosmological constant
problem. No one knows the answer!
– 26 –
2.3.2 The Casimir Effect
“I mentioned my results to Niels Bohr, during a walk. That is nice, he
said, that is something new... and he mumbled something about zero-point
energy.”
Hendrik Casimir
Using the normal ordering prescription we can happily set E
0
= 0, while chanting
the mantra that only energy differences can be measured. But we should be careful, for
there is a situation where differences in the energy of vacuum fluctuations themselves
can be measured.
To regulate the infra-red divergences, we’ll make the x
1
direction periodic, with size
L, and impose periodic boundary conditions such that
φ(x) = φ(x + Ln) (2.31)
with n = (1, 0, 0). We’ll leave y and z alone, but remember
L
d
Figure 3:
that we should compute all physical quantities per unit area
A. We insert two reflecting plates, separated by a distance
d ≪ L in the x
1
direction. The plates are such that they
impose φ(x) = 0 at the position of the plates. The presence of
these plates affects the Fourier decomposition of the field and,
in particular, means that the momentum of the field inside the
plates is quantized as
p =
_

d
, p
y
, p
z
_
n ∈ Z
+
(2.32)
For a massless scalar field, the ground state energy between the plates is
E(d)
A
=

n=1
_
dp
y
dp
z
(2π)
2
1
2
_
_

d
_
2
+ p
2
y
+ p
2
z
(2.33)
while the energy outside the plates is E(L −d). The total energy is therefore
E = E(d) + E(L −d) (2.34)
which – at least naively – depends on d. If this naive guess is true, it would mean
that there is a force on the plates due to the fluctuations of the vacuum. This is the
Casimir force, first predicted in 1948 and observed 10 years later. In the real world, the
effect is due to the vacuum fluctuations of the electromagnetic field, with the boundary
conditions imposed by conducting plates. Here we model this effect with a scalar.
– 27 –
But there’s a problem. E is infinite! What to do? The problem comes from the
arbitrarily high momentum modes. We could regulate this in a number of different
ways. Physically one could argue that any real plate cannot reflect waves of arbitrarily
high frequency: at some point, things begin to leak. Mathematically, we want to find
a way to neglect modes of momentum p ≫ a
−1
for some distance scale a ≪ d, known
as the ultra-violet (UV) cut-off. One way to do this is to change the integral (2.33) to,
E(d)
A
=

n=1
_
dp
y
dp
z
(2π)
2
1
2
_
_
_

d
_
2
+ p
2
y
+ p
2
z
_
e
−a
q
(

d
)
2
+p
2
y
+p
2
z
(2.35)
which has the property that as a → 0, we regain the full, infinite, expression (2.33).
However (2.35) is finite, and gives us something we can easily work with. Of course,
we made it finite in a rather ad-hoc manner and we better make sure that any physical
quantity we calculate doesn’t depend on the UV cut-off a, otherwise it’s not something
we can really trust.
The integral (2.35) is do-able, but a little complicated. It’s a lot simpler if we look
at the problem in d = 1 + 1 dimensions, rather than d = 3 + 1 dimensions. We’ll find
that all the same physics is at play. Now the energy is given by
E
1+1
(d) =
π
2d

n=1
n (2.36)
We now regulate this sum by introducing the UV cutoff a introduced above. This
renders the expression finite, allowing us to start manipulating it thus,
E
1+1
(d) →
π
2d

n=1
ne
−anπ/d
= −
1
2

∂a

n=1
e
−anπ/d
= −
1
2

∂a
1
1 −e
−aπ/d
=
π
2d
e
aπ/d
(e
aπ/d
−1)
2
=
d
2πa
2

π
24d
+O(a
2
) (2.37)
where, in the last line, we’ve used the fact that a ≪ d. We can now compute the full
energy,
E
1+1
= E
1+1
(d) + E
1+1
(L −d) =
L
2πa
2

π
24
_
1
d
+
1
L −d
_
+O(a
2
) (2.38)
– 28 –
This is still infinite in the limit a → 0, which is to be expected. However, the force is
given by
∂E
1+1
∂d
=
π
24d
2
+ . . . (2.39)
where the . . . include terms of size d/L and a/d. The key point is that as we remove
both the regulators, and take a →0 and L →∞, the force between the plates remains
finite. This is the Casimir force
2
.
If we ploughed through the analogous calculation in d = 3 + 1 dimensions, and
performed the integral (2.35), we would find the result
1
A
∂E
∂d
=
π
2
480d
4
(2.40)
The true Casimir force is twice as large as this, due to the two polarization states of
the photon.
2.4 Particles
Having dealt with the vacuum, we can now turn to the excitations of the field. It’s
easy to verify that
[H, a

p
] = ω
p
a

p
and [H, a
p
] = −ω
p
a
p
(2.41)
which means that, just as for the harmonic oscillator, we can construct energy eigen-
states by acting on the vacuum [0) with a

p
. Let
[p) = a

p
[0) (2.42)
This state has energy
H[p) = ω
p
[ p) with ω
2
p
= p
2
+ m
2
(2.43)
But we recognize this as the relativistic dispersion relation for a particle of mass m and
3-momentum p,
E
2
p
= p
2
+ m
2
(2.44)
2
The number 24 that appears in the denominator of the one-dimensional Casimir force plays a more
famous role in string theory: the same calculation in that context is the reason the bosonic string lives
in 26 = 24 +2 spacetime dimensions. (The +2 comes from the fact the string itself is extended in one
space and one time dimension). You will need to attend next term’s “String Theory” course to see
what on earth this has to do with the Casimir force.
– 29 –
We interpret the state [ p) as the momentum eigenstate of a single particle of mass m.
To stress this, from now on we’ll write E
p
everywhere instead of ω
p
. Let’s check this
particle interpretation by studying the other quantum numbers of [ p). We may take
the classical total momentum

P given in (1.46) and turn it into an operator. After
normal ordering, it becomes

P = −
_
d
3
x π

∇φ =
_
d
3
p
(2π)
3
p a

p
a
p
(2.45)
Acting on our state [p) with

P, we learn that it is indeed an eigenstate,

P [p) = p [p) (2.46)
telling us that the state [ p) has momentum p. Another property of [ p) that we can
study is its angular momentum. Once again, we may take the classical expression for
the total angular momentum of the field (1.55) and turn it into an operator,
J
i
= ǫ
ijk
_
d
3
x (¸
0
)
jk
(2.47)
It’s not hard to show that acting on the one-particle state with zero momentum,
J
i
[p = 0) = 0, which we interpret as telling us that the particle carries no internal
angular momentum. In other words, quantizing a scalar field gives rise to a spin 0
particle.
Multi-Particle States, Bosonic Statistics and Fock Space
We can create multi-particle states by acting multiple times with a

’s. We interpret
the state in which n a

’s act on the vacuum as an n-particle state,
[p
1
, . . . , p
n
) = a

p
1
. . . a

pn
[0) (2.48)
Because all the a

’s commute among themselves, the state is symmetric under exchange
of any two particles. For example,
[ p, q) = [q, p) (2.49)
This means that the particles are bosons.
The full Hilbert space of our theory is spanned by acting on the vacuum with all
possible combinations of a

’s,
[0) , a

p
[0) , a

p
a

q
[0) , a

p
a

q
a

r
[0) . . . (2.50)
– 30 –
This space is known as a Fock space. The Fock space is simply the sum of the n-particle
Hilbert spaces, for all n ≥ 0. There is a useful operator which counts the number of
particles in a given state in the Fock space. It is called the number operator N
N =
_
d
3
p
(2π)
3
a

p
a
p
(2.51)
and satisfies N [p
1
, . . . , p
n
) = n[p
1
, . . . , p
n
). The number operator commutes with the
Hamiltonian, [N, H] = 0, ensuring that particle number is conserved. This means that
we can place ourselves in the n-particle sector, and stay there. This is a property of
free theories, but will no longer be true when we consider interactions: interactions
create and destroy particles, taking us between the different sectors in the Fock space.
Operator Valued Distributions
Although we’re referring to the states [ p) as “particles”, they’re not localized in space
in any way — they are momentum eigenstates. Recall that in quantum mechanics the
position and momentum eigenstates are not good elements of the Hilbert space since
they are not normalizable (they normalize to delta-functions). Similarly, in quantum
field theory neither the operators φ(x), nor a
p
are good operators acting on the Fock
space. This is because they don’t produce normalizable states. For example,
¸0[ a
p
a

p
[0) = ¸p[ p) = (2π)
3
δ(0) and ¸0[φ(x) φ(x) [0) = ¸x[ x) = δ(0) (2.52)
They are operator valued distributions, rather than functions. This means that al-
though φ(x) has a well defined vacuum expectation value, ¸0[ φ(x) [0) = 0, the fluc-
tuations of the operator at a fixed point are infinite, ¸0[ φ(x)φ(x) [0) = ∞. We can
construct well defined operators by smearing these distributions over space. For exam-
ple, we can create a wavepacket
[ϕ) =
_
d
3
p
(2π)
3
e
−i p·x
ϕ( p) [ p) (2.53)
which is partially localized in both position and momentum space. (A typical state
might be described by the Gaussian ϕ( p) = exp(− p
2
/2m
2
)).
2.4.1 Relativistic Normalization
We have defined the vacuum [0) which we normalize as ¸0[ 0) = 1. The one-particle
states [ p) = a

p
[0) then satisfy
¸ p[ q) = (2π)
3
δ
(3)
( p −q) (2.54)
– 31 –
But is this Lorentz invariant? It’s not obvious because we only have 3-vectors. What
could go wrong? Suppose we have a Lorentz transformation
p
µ
→(p

)
µ
= Λ
µ
ν
p
ν
(2.55)
such that the 3-vector transforms as p → p

. In the quantum theory, it would be
preferable if the two states are related by a unitary transformation,
[ p) →[p

) = U(Λ) [ p) (2.56)
This would mean that the normalizations of [p) and [p

) are the same whenever p and
p

are related by a Lorentz transformation. But we haven’t been at all careful with
normalizations. In general, we could get
[p) →λ( p, p

) [p

) (2.57)
for some unknown function λ( p, p

). How do we figure this out? The trick is to look at
an object which we know is Lorentz invariant. One such object is the identity operator
on one-particle states (which is really the projection operator onto one-particle states).
With the normalization (2.54) we know this is given by
1 =
_
d
3
p
(2π)
3
[ p) ¸ p[ (2.58)
This operator is Lorentz invariant, but it consists of two terms: the measure
_
d
3
p and
the projector [p)¸ p[. Are these individually Lorentz invariant? In fact the answer is no.
Claim The Lorentz invariant measure is,
_
d
3
p
2E
p
(2.59)
Proof:
_
d
4
p is obviously Lorentz invariant. And the relativistic dispersion relation
for a massive particle,
p
µ
p
µ
= m
2
⇒ p
2
0
= E
2
p
= p
2
+ m
2
(2.60)
is also Lorentz invariant. Solving for p
0
, there are two branches of solutions: p
0
= ±E
p
.
But the choice of branch is another Lorentz invariant concept. So piecing everything
together, the following combination must be Lorentz invariant,
_
d
4
p δ(p
2
0
−p
2
−m
2
)
¸
¸
¸
¸
p
0
>0
=
_
d
3
p
2p
0
¸
¸
¸
¸
p
0
=E
p
(2.61)
which completes the proof.
– 32 –
From this result we can figure out everything else. For example, the Lorentz invariant
δ-function for 3-vectors is
2E
p
δ
(3)
( p −q) (2.62)
which follows because
_
d
3
p
2E
p
2E
p
δ
(3)
( p −q) = 1 (2.63)
So finally we learn that the relativistically normalized momentum states are given by
[p) =
_
2E
p
[ p) =
_
2E
p
a

p
[0) (2.64)
Notice that our notation is rather subtle: the relativistically normalized momentum
state [p) differs from [ p) just by the absence of a vector over p. These states now satisfy
¸p[ q) = (2π)
3
2E
p
δ
(3)
( p −q) (2.65)
Finally, we can rewrite the identity on one-particle states as
1 =
_
d
3
p
(2π)
3
1
2E
p
[p) ¸p[ (2.66)
Some texts also define relativistically normalized creation operators by a

(p) =
_
2E
p
a

p
.
We won’t make use of this notation here.
2.5 Complex Scalar Fields
Consider a complex scalar field ψ(x) with Lagrangian
L = ∂
µ
ψ


µ
ψ −M
2
ψ

ψ (2.67)
Notice that, in contrast to the Lagrangian (1.7) for a real scalar field, there is no factor
of 1/2 in front of the Lagrangian for a complex scalar field. If we write ψ in terms
of real scalar fields by ψ = (φ
1
+ iφ
2
)/

2, we get the factor of 1/2 coming from the
1/

2’s. The equations of motion are

µ

µ
ψ + M
2
ψ = 0

µ

µ
ψ

+ M
2
ψ

= 0 (2.68)
where the second equation is the complex conjugate of the first. We expand the complex
field operator as a sum of plane waves as
ψ =
_
d
3
p
(2π)
3
1
_
2E
p
_
b
p
e
+i p·x
+ c

p
e
−i p·x
_
ψ

=
_
d
3
p
(2π)
3
1
_
2E
p
_
b

p
e
−i p·x
+ c
p
e
+i p·x
_
(2.69)
– 33 –
Since the classical field ψ is not real, the corresponding quantum field ψ is not hermitian.
This is the reason that we have different operators b and c

appearing in the positive
and negative frequency parts. The classical field momentum is π = ∂L/∂
˙
ψ =
˙
ψ

. We
also turn this into a quantum operator field which we write as,
π =
_
d
3
p
(2π)
3
i
_
E
p
2
_
b

p
e
−i p·x
−c
p
e
+i p·x
_
π

=
_
d
3
p
(2π)
3
(−i)
_
E
p
2
_
b
p
e
+i p·x
−c

p
e
−i p·x
_
(2.70)
The commutation relations between fields and momenta are given by
[ψ(x), π(y)] = iδ
(3)
(x −y) and [ψ(x), π

(y)] = 0 (2.71)
together with others related by complex conjugation, as well as the usual [ψ(x), ψ(y)] =
[ψ(x), ψ

(y)] = 0, etc. One can easily check that these field commutation relations are
equivalent to the commutation relations for the operators b
p
and c
p
,
[b
p
, b

q
] = (2π)
3
δ
(3)
( p −q)
[c
p
, c

q
] = (2π)
3
δ
(3)
( p −q) (2.72)
and
[b
p
, b
q
] = [c
p
, c
q
] = [b
p
, c
q
] = [b
p
, c

q
] = 0 (2.73)
In summary, quantizing a complex scalar field gives rise to two creation operators, b

p
and c

p
. These have the interpretation of creating two types of particle, both of mass M
and both spin zero. They are interpreted as particles and anti-particles. In contrast,
for a real scalar field there is only a single type of particle: for a real scalar field, the
particle is its own antiparticle.
Recall that the theory (2.67) has a classical conserved charge
Q = i
_
d
3
x (
˙
ψ

ψ −ψ

˙
ψ) = i
_
d
3
x (πψ −ψ

π

) (2.74)
After normal ordering, this becomes the quantum operator
Q =
_
d
3
p
(2π)
3
(c

p
c
p
−b

p
b
p
) = N
c
−N
b
(2.75)
so Q counts the number of anti-particles (created by c

) minus the number of particles
(created by b

). We have [H, Q] = 0, ensuring that Q is conserved quantity in the
quantum theory. Of course, in our free field theory this isn’t such a big deal because
both N
c
and N
b
are separately conserved. However, we’ll soon see that in interacting
theories Q survives as a conserved quantity, while N
c
and N
b
individually do not.
– 34 –
2.6 The Heisenberg Picture
Although we started with a Lorentz invariant Lagrangian, we slowly butchered it as we
quantized, introducing a preferred time coordinate t. It’s not at all obvious that the
theory is still Lorentz invariant after quantization. For example, the operators φ(x)
depend on space, but not on time. Meanwhile, the one-particle states evolve in time
by Schr¨odinger’s equation,
i
d [ p(t))
dt
= H[ p(t)) ⇒ [ p(t)) = e
−iE
p
t
[p) (2.76)
Things start to look better in the Heisenberg picture where time dependence is assigned
to the operators O,
O
H
= e
iHt
O
S
e
−iHt
(2.77)
so that
dO
H
dt
= i[H, O
H
] (2.78)
where the subscripts S and H tell us whether the operator is in the Schr¨odinger or
Heisenberg picture. In field theory, we drop these subscripts and we will denote the
picture by specifying whether the fields depend on space φ(x) (the Schr¨odinger picture)
or spacetime φ(x, t) = φ(x) (the Heisenberg picture).
The operators in the two pictures agree at a fixed time, say, t = 0. The commutation
relations (2.2) become equal time commutation relations in the Heisenberg picture,
[φ(x, t), φ(y, t)] = [π(x, t), π(y, t)] = 0
[φ(x, t), π(y, t)] = iδ
(3)
(x −y) (2.79)
Now that the operator φ(x) = φ(x, t) depends on time, we can start to study how it
evolves. For example, we have
˙
φ = i[H, φ] =
i
2
[
_
d
3
y π(y)
2
+∇φ(y)
2
+ m
2
φ(y)
2
, φ(x)]
= i
_
d
3
y π(y) (−i) δ
(3)
(y −x) = π(x) (2.80)
Meanwhile, the equation of motion for π reads,
˙ π = i[H, π] =
i
2
[
_
d
3
y π(y)
2
+∇φ(y)
2
+ m
2
φ(y)
2
, π(x)]
– 35 –
=
i
2
_
d
3
y (∇
y
[φ(y), π(x)]) ∇φ(y) +∇φ(y) ∇
y
[φ(y), π(x)]
+2im
2
φ(y) δ
(3)
(x −y)
= −
__
d
3
y
_

y
δ
(3)
(x −y)
_

y
φ(y)
_
−m
2
φ(x)
= ∇
2
φ −m
2
φ (2.81)
where we’ve included the subscript y on ∇
y
when there may be some confusion about
which argument the derivative is acting on. To reach the last line, we’ve simply inte-
grated by parts. Putting (2.80) and (2.81) together we find that the field operator φ
satisfies the Klein-Gordon equation

µ

µ
φ + m
2
φ = 0 (2.82)
Things are beginning to look more relativistic. We can write the Fourier expansion of
φ(x) by using the definition (2.77) and noting,
e
iHt
a
p
e
−iHt
= e
−iE
p
t
a
p
and e
iHt
a

p
e
−iHt
= e
+iE
p
t
a

p
(2.83)
which follows from the commutation relations [H, a
p
] = −E
p
a
p
and [H, a

p
] = +E
p
a

p
.
This then gives,
φ(x, t) =
_
d
3
p
(2π)
3
1
_
2E
p
_
a
p
e
−ip·x
+ a

p
e
+ip·x
_
(2.84)
which looks very similar to the previous expansion (2.18) t
x
O
O
2
1
Figure 4:
except that the exponent is now written in terms of 4-
vectors, p x = E
p
t − p x. (Note also that a sign has
flipped in the exponent due to our Minkowski metric con-
traction). It’s simple to check that (2.84) indeed satisfies
the Klein-Gordon equation (2.82).
2.6.1 Causality
We’re approaching something Lorentz invariant in the
Heisenberg picture, where φ(x) now satisfies the Klein-
Gordon equation. But there’s still a hint of non-Lorentz invariance because φ and π
satisfy equal time commutation relations,
[φ(x, t), π(y, t)] = iδ
(3)
(x −y) (2.85)
– 36 –
But what about arbitrary spacetime separations? In particular, for our theory to be
causal, we must require that all spacelike separated operators commute,
[O
1
(x), O
2
(y)] = 0 ∀ (x −y)
2
< 0 (2.86)
This ensures that a measurement at x cannot affect a measurement at y when x and y
are not causally connected. Does our theory satisfy this crucial property? Let’s define
∆(x −y) = [φ(x), φ(y)] (2.87)
The objects on the right-hand side of this expression are operators. However, it’s easy
to check by direct substitution that the left-hand side is simply a c-number function
with the integral expression
∆(x −y) =
_
d
3
p
(2π)
3
1
2E
p
_
e
−ip·(x−y)
−e
ip·(x−y)
_
(2.88)
What do we know about this function?
• It’s Lorentz invariant, thanks to the appearance of the Lorentz invariant measure
_
d
3
p/2E
p
that we introduced in (2.59).
• It doesn’t vanish for timelike separation. For example, taking x − y = (t, 0, 0, 0)
gives [φ(x, 0), φ(x, t)] ∼ e
−imt
−e
+imt
.
• It vanishes for space-like separations. This follows by noting that ∆(x − y) = 0
at equal times for all (x − y)
2
= −(x − y)
2
< 0, which we can see explicitly by
writing
[φ(x, t), φ(y, t)] =
1
2
_
d
3
p
(2π)
3
1
_
p
2
+ m
2
_
e
i p·(x− y)
−e
−i p·(x− y)
_
(2.89)
and noticing that we can flip the sign of p in the last exponent as it is an inte-
gration variable. But since ∆(x − y) Lorentz invariant, it can only depend on
(x −y)
2
and must therefore vanish for all (x −y)
2
< 0.
We therefore learn that our theory is indeed causal with commutators vanishing
outside the lightcone. This property will continue to hold in interacting theories; indeed,
it is usually given as one of the axioms of local quantum field theories. I should mention
however that the fact that [φ(x), φ(y)] is a c-number function, rather than an operator,
is a property of free fields only.
– 37 –
2.7 Propagators
We could ask a different question to probe the causal structure of the theory. Prepare
a particle at spacetime point y. What is the amplitude to find it at point x? We can
calculate this:
¸0[ φ(x)φ(y) [0) =
_
d
3
p d
3
p

(2π)
6
1
_
4E
p
E
p

¸0[ a
p
a

p

[ 0) e
−ip·x+ip

·y
=
_
d
3
p
(2π)
3
1
2E
p
e
−ip·(x−y)
≡ D(x −y) (2.90)
The function D(x−y) is called the propagator. For spacelike separations, (x−y)
2
< 0,
one can show that D(x −y) decays like
D(x −y) ∼ e
−m|x− y|
(2.91)
So it decays exponentially quickly outside the lightcone but, nonetheless, is non-vanishing!
The quantum field appears to leak out of the lightcone. Yet we’ve just seen that space-
like measurements commute and the theory is causal. How do we reconcile these two
facts? We can rewrite the calculation (2.89) as
[φ(x), φ(y)] = D(x −y) −D(y −x) = 0 if (x −y)
2
< 0 (2.92)
There are words you can drape around this calculation. When (x − y)
2
< 0, there
is no Lorentz invariant way to order events. If a particle can travel in a spacelike
direction from x →y, it can just as easily travel from y →x. In any measurement, the
amplitudes for these two events cancel.
With a complex scalar field, it is more interesting. We can look at the equation
[ψ(x), ψ

(y)] = 0 outside the lightcone. The interpretation now is that the amplitude
for the particle to propagate from x → y cancels the amplitude for the antiparticle
to travel from y → x. In fact, this interpretation is also there for a real scalar field
because the particle is its own antiparticle.
2.7.1 The Feynman Propagator
As we will see shortly, one of the most important quantities in interacting field theory
is the Feynman propagator,

F
(x −y) = ¸0[ Tφ(x)φ(y) [0) =
_
D(x −y) x
0
> y
0
D(y −x) y
0
> x
0
(2.93)
– 38 –
where T stands for time ordering, placing all operators evaluated at later times to the
left so,
Tφ(x)φ(y) =
_
φ(x)φ(y) x
0
> y
0
φ(y)φ(x) y
0
> x
0
(2.94)
Claim: There is a useful way of writing the Feynman propagator in terms of a 4-
momentum integral.

F
(x −y) =
_
d
4
p
(2π)
4
i
p
2
−m
2
e
−ip·(x−y)
(2.95)
Notice that this is the first time in this course that we’ve integrated over 4-momentum.
Until now, we integrated only over 3-momentum, with p
0
fixed by the mass-shell con-
dition to be p
0
= E
p
. In the expression (2.95) for ∆
F
, we have no such condition on
p
0
. However, as it stands this integral is ill-defined because, for each value of p, the
denominator p
2
−m
2
= (p
0
)
2
−p
2
−m
2
produces a pole when p
0
= ±E
p
= ±
_
p
2
+ m
2
.
We need a prescription for avoiding these singularities in the p
0
integral. To get the
Feynman propagator, we must choose the contour to be
Im(p )
Re(p ) E
p

E +
p
0
0
Figure 5: The contour for the Feynman propagator.
Proof:
1
p
2
−m
2
=
1
(p
0
)
2
−E
2
p
=
1
(p
0
−E
p
)(p
0
+ E
p
)
(2.96)
so the residue of the pole at p
0
= ±E
p
is ±1/2E
p
. When x
0
> y
0
, we close the contour
in the lower half plane, where p
0
→ −i∞, ensuring that the integrand vanishes since
e
−ip
0
(x
0
−y
0
)
→ 0. The integral over p
0
then picks up the residue at p
0
= +E
p
which
is −2πi/2E
p
where the minus sign arises because we took a clockwise contour. Hence
when x
0
> y
0
we have

F
(x −y) =
_
d
3
p
(2π)
4
−2πi
2E
p
i e
−iE
p
(x
0
−y
0
)+i p·(x− y)
– 39 –
=
_
d
3
p
(2π)
3
1
2E
p
e
−ip·(x−y)
= D(x −y) (2.97)
which is indeed the Feynman propagator for x
0
> y
0
. In contrast, when y
0
> x
0
, we
close the contour in an anti-clockwise direction in the upper half plane to get,

F
(x −y) =
_
d
3
p
(2π)
4
2πi
(−2E
p
)
i e
+iE
p
(x
0
−y
0
)+i p·(x− y)
=
_
d
3
p
(2π)
3
1
2E
p
e
−iE
p
(y
0
−x
0
)−i p·( y−x)
=
_
d
3
p
(2π)
3
1
2E
p
e
−ip·(y−x)
= D(y −x) (2.98)
where to go to from the second line to the third, we have flipped the sign of p which
is valid since we integrate over d
3
p and all other quantities depend only on p
2
. Once
again we reproduce the Feynman propagator.
Instead of specifying the contour, we may instead write
Im(p )
0
+ iε
i ε −
Re(p )
0
Figure 6:
the Feynman propagator as

F
(x −y) =
_
d
4
p
(2π)
4
ie
−ip·(x−y)
p
2
−m
2
+ iǫ
(2.99)
with ǫ > 0, and infinitesimal. This has the effect of shifting
the poles slightly off the real axis, so the integral along the
real p
0
axis is equivalent to the contour shown in Figure 5.
This way of writing the propagator is, for obvious reasons,
called the “iǫ prescription”.
2.7.2 Green’s Functions
There is another avatar of the propagator: it is a Green’s function for the Klein-Gordon
operator. If we stay away from the singularities, we have
(∂
2
t
−∇
2
+ m
2
)∆
F
(x −y) =
_
d
4
p
(2π)
4
i
p
2
−m
2
(−p
2
+ m
2
) e
−ip·(x−y)
= −i
_
d
4
p
(2π)
4
e
−ip·(x−y)
= −i δ
(4)
(x −y) (2.100)
Note that we didn’t make use of the contour anywhere in this derivation. For some
purposes it is also useful to pick other contours which also give rise to Green’s functions.
– 40 –
Im(p )
Re(p )
E +
p
0
0
E
p

Im(p )
Re(p ) Ep −
0
0
Ep
+
Figure 7: The retarded contour Figure 8: The advanced contour
For example, the retarded Green’s function ∆
R
(x −y) is defined by the contour shown
in Figure 7 which has the property

R
(x −y) =
_
D(x −y) −D(y −x) x
0
> y
0
0 y
0
> x
0
(2.101)
The retarded Green’s function is useful in classical field theory if we know the initial
value of some field configuration and want to figure out what it evolves into in the
presence of a source, meaning that we want to know the solution to the inhomogeneous
Klein-Gordon equation,

µ

µ
φ + m
2
φ = J(x) (2.102)
for some fixed background function J(x). Similarly, one can define the advanced Green’s
function ∆
A
(x − y) which vanishes when y
0
< x
0
, which is useful if we know the end
point of a field configuration and want to figure out where it came from. Given that
next term’s course is called “Advanced Quantum Field Theory”, there is an obvious
name for the current course. But it got shot down in the staff meeting. In the quantum
theory, we will see that the Feynman Green’s function is most relevant.
2.8 Non-Relativistic Fields
Let’s return to our classical complex scalar field obeying the Klein-Gordon equation.
We’ll decompose the field as
ψ(x, t) = e
−imt
˜
ψ(x, t) (2.103)
Then the KG-equation reads

2
t
ψ −∇
2
ψ + m
2
ψ = e
−imt
_
¨
˜
ψ −2im
˙
˜
ψ −∇
2
˜
ψ
_
= 0 (2.104)
with the m
2
term cancelled by the time derivatives. The non-relativistic limit of a
particle is [ p[ ≪m. Let’s look at what this does to our field. After a Fourier transform,
– 41 –
this is equivalent to saying that [
¨
˜
ψ[ ≪m[
˙
˜
ψ[. In this limit, we drop the term with two
time derivatives and the KG equation becomes,
i

˜
ψ
∂t
= −
1
2m

2
˜
ψ (2.105)
This looks very similar to the Schr¨odinger equation for a non-relativistic free particle of
mass m. Except it doesn’t have any probability interpretation — it’s simply a classical
field evolving through an equation that’s first order in time derivatives.
We wrote down a Lagrangian in section 1.1.2 which gives rise to field equations
which are first order in time derivatives. In fact, we can derive this from the relativistic
Lagrangian for a scalar field by again taking the limit ∂
t
ψ ≪ mψ. After losing the
˜
tilde, so
˜
ψ →ψ, the non-relativistic Lagrangian becomes
L = +iψ

˙
ψ −
1
2m
∇ψ

∇ψ (2.106)
where we’ve divided by 1/2m. This Lagrangian has a conserved current arising from
the internal symmetry ψ →e

ψ. The current has time and space components
j
µ
=
_
ψ

ψ,
i
2m


∇ψ −ψ∇ψ

)
_
(2.107)
To move to the Hamiltonian formalism we compute the momentum
π =
∂L

˙
ψ
= iψ

(2.108)
This means that the momentum conjugate to ψ is iψ

. The momentum does not depend
on time derivatives at all! This looks a little disconcerting but it’s fully consistent for a
theory which is first order in time derivatives. In order to determine the full trajectory
of the field, we need only specify ψ and ψ

at time t = 0: no time derivatives on the
initial slice are required.
Since the Lagrangian already contains a “p ˙ q” term (instead of the more familiar
1
2
p ˙ q
term), the time derivatives drop out when we compute the Hamiltonian. We get,
H =
1
2m
∇ψ

∇ψ (2.109)
To quantize we impose (in the Schr¨odinger picture) the canonical commutation relations
[ψ(x), ψ(y)] = [ψ

(x), ψ

(y)] = 0
[ψ(x), ψ

(y)] = δ
(3)
(x −y) (2.110)
– 42 –
We may expand ψ(x) as a Fourier transform
ψ(x) =
_
d
3
p
(2π)
3
a
p
e
i p·x
(2.111)
where the commutation relations (2.110) require
[a
p
, a

q
] = (2π)
3
δ
(3)
( p −q) (2.112)
The vacuum satisfies a
p
[0) = 0, and the excitations are a

p
1
. . . a

pn
[0). The one-particle
states have energy
H[ p) =
p
2
2m
[ p) (2.113)
which is the non-relativistic dispersion relation. We conclude that quantizing the first
order Lagrangian (2.106) gives rise to non-relativistic particles of mass m. Some com-
ments:
• We have a complex field but only a single type of particle. The anti-particle is
not in the spectrum. The existence of anti-particles is a consequence of relativity.
• A related fact is that the conserved charge Q =
_
d
3
x : ψ

ψ : is the particle
number. This remains conserved even if we include interactions in the Lagrangian
of the form
∆L = V (ψ

ψ) (2.114)
So in non-relativistic theories, particle number is conserved. It is only with rela-
tivity, and the appearance of anti-particles, that particle number can change.
• There is no non-relativistic limit of a real scalar field. In the relativistic theory,
the particles are their own anti-particles, and there can be no way to construct a
multi-particle theory that conserves particle number.
2.8.1 Recovering Quantum Mechanics
In quantum mechanics, we talk about the position and momentum operators

X and

P. In quantum field theory, position is relegated to a label. How do we get back to
quantum mechanics? We already have the operator for the total momentum of the field

P =
_
d
3
p
(2π)
3
p a

p
a
p
(2.115)
– 43 –
which, on one-particle states, gives

P [p) = p [p). It’s also easy to construct the position
operator. Let’s work in the non-relativistic limit. Then the operator
ψ

(x) =
_
d
3
p
(2π)
3
a

p
e
−i p·x
(2.116)
creates a particle with δ-function localization at x. We write [x) = ψ

(x) [0). A natural
position operator is then

X =
_
d
3
x xψ

(x) ψ(x) (2.117)
so that

X [x) = x[x).
Let’s now construct a state [ϕ) by taking superpositions of one-particle states [x),
[ϕ) =
_
d
3
x ϕ(x) [x) (2.118)
The function ϕ(x) is what we would usually call the Schr¨odinger wavefunction (in the
position representation). Let’s make sure that it indeed satisfies all the right properties.
Firstly, it’s clear that acting with the position operator

X has the right action of ϕ(x),
X
i
[ϕ) =
_
d
3
x x
i
ϕ(x) [x) (2.119)
but what about the momentum operator

P? We will now show that
P
i
[ϕ) =
_
d
3
x
_
−i
∂ϕ
∂x
i
_
[x) (2.120)
which tells us that P
i
acts as the familiar derivative on wavefunctions [ϕ). To see that
this is the case, we write
P
i
[ϕ) =
_
d
3
xd
3
p
(2π)
3
p
i
a

p
a
p
ϕ(x) ψ

(x) [0)
=
_
d
3
xd
3
p
(2π)
3
p
i
a

p
e
−i p·x
ϕ(x) [0) (2.121)
where we’ve used the relationship [a
p
, ψ

(x)] = e
−i p·x
which can be easily checked.
Proceeding with our calculation, we have
P
i
[ϕ) =
_
d
3
xd
3
p
(2π)
3
a

p
_
i

∂x
i
e
−i p·x
_
ϕ(x) [0)
=
_
d
3
xd
3
p
(2π)
3
e
−i p·x
_
−i
∂ϕ
∂x
i
_
a

p
[0)
=
_
d
3
x
_
−i
∂ϕ
∂x
i
_
[x) (2.122)
– 44 –
which confirms (2.120). So we learn that when acting on one-particle states, the oper-
ators

X and

P act as position and momentum operators in quantum mechanics, with
[X
i
, P
j
] [ϕ) = iδ
ij
[ϕ). But what about dynamics? How does the wavefunction ϕ(x, t)
change in time? The Hamiltonian (2.109) can be rewritten as
H =
_
d
3
x
1
2m
∇ψ

∇ψ =
_
d
3
p
(2π)
3
p
2
2m
a

p
a
p
(2.123)
so we find that
i
∂ϕ
∂t
= −
1
2m

2
ϕ (2.124)
But this is the same equation obeyed by the original field (2.105)! Except this time, it
really is the Schr¨odinger equation, complete with the usual probabilistic interpretation
for the wavefunction ϕ. Note in particular that the conserved charge arising from the
Noether current (2.107) is Q =
_
d
3
x[ϕ(x)[
2
which is the total probability.
Historically, the fact that the equation for the classical field (2.105) and the one-
particle wavefunction (2.124) coincide caused some confusion. It was thought that
perhaps we are quantizing the wavefunction itself and the resulting name “second quan-
tization” is still sometimes used today to mean quantum field theory. It’s important to
stress that, despite the name, we’re not quantizing anything twice! We simply quantize
a classical field once. Nonetheless, in practice it’s useful to know that if we treat the
one-particle Schr¨odinger equation as the equation for a quantum field then it will give
the correct generalization to multi-particle states.
Interactions
Often in quantum mechanics, we’re interested in particles moving in some fixed back-
ground potential V (x). This can be easily incorporated into field theory by working
with a Lagrangian with explicit x dependence,
L = iψ

˙
ψ −
1
2m
∇ψ

∇ψ −V (x) ψ

ψ (2.125)
Note that this Lagrangian doesn’t respect translational symmetry and we won’t have
the associated energy-momentum tensor. While such Lagrangians are useful in con-
densed matter physics, we rarely (or never) come across them in high-energy physics,
where all equations obey translational (and Lorentz) invariance.
– 45 –
One can also consider interactions between particles. Obviously these are only impor-
tant for n particle states with n ≥ 2. We therefore expect them to arise from additions
to the Lagrangian of the form
∆L = ψ

(x) ψ

(x) ψ(x) ψ(x) (2.126)
which, in the quantum theory, is an operator which destroys two particles before creat-
ing two new ones. Such terms in the Lagrangian will indeed lead to inter-particle forces,
both in the non-relativistic and relativistic setting. In the next section we explore these
types of interaction in detail for relativistic theories.
– 46 –
3. Interacting Fields
The free field theories that we’ve discussed so far are very special: we can determine
their spectrum, but nothing interesting then happens. They have particle excitations,
but these particles don’t interact with each other.
Here we’ll start to examine more complicated theories that include interaction terms.
These will take the form of higher order terms in the Lagrangian. We’ll start by asking
what kind of small perturbations we can add to the theory. For example, consider the
Lagrangian for a real scalar field,
L =
1
2

µ
φ ∂
µ
φ −
1
2
m
2
φ
2

n≥3
λ
n
n!
φ
n
(3.1)
The coefficients λ
n
are called coupling constants. What restrictions do we have on λ
n
to ensure that the additional terms are small perturbations? You might think that we
need simply make “λ
n
≪1”. But this isn’t quite right. To see why this is the case, let’s
do some dimensional analysis. Firstly, note that the action has dimensions of angular
momentum or, equivalently, the same dimensions as . Since we’ve set = 1, using
the convention described in the introduction, we have [S] = 0. With S =
_
d
4
xL, and
[d
4
x] = −4, the Lagrangian density must therefore have
[L] = 4 (3.2)
What does this mean for the Lagrangian (3.1)? Since [∂
µ
] = 1, we can read off the
mass dimensions of all the factors to find,
[φ] = 1 , [m] = 1 , [λ
n
] = 4 −n (3.3)
So now we see why we can’t simply say we need λ
n
≪ 1, because this statement only
makes sense for dimensionless quantities. The various terms, parameterized by λ
n
, fall
into three different categories
• [λ
3
] = 1: For this term, the dimensionless parameter is λ
3
/E, where E has
dimensions of mass. Typically in quantum field theory, E is the energy scale of
the process of interest. This means that λ
3
φ
3
/3! is a small perturbation at high
energies E ≫ λ
3
, but a large perturbation at low energies E ≪ λ
3
. Terms that
we add to the Lagrangian with this behavior are called relevant because they’re
most relevant at low energies (which, after all, is where most of the physics we see
lies). In a relativistic theory, E > m, so we can always make this perturbation
small by taking λ
3
≪m.
– 47 –
• [λ
4
] = 0: this term is small if λ
4
≪1. Such perturbations are called marginal.
• [λ
n
] < 0 for n ≥ 5: The dimensionless parameter is (λ
n
E
n−4
), which is small at
low-energies and large at high energies. Such perturbations are called irrelevant.
As you’ll see later, it is typically impossible to avoid high energy processes in quantum
field theory. (We’ve already seen a glimpse of this in computing the vacuum energy).
This means that we might expect problems with irrelevant operators. Indeed, these lead
to “non-renormalizable” field theories in which one cannot make sense of the infinities
at arbitrarily high energies. This doesn’t necessarily mean that the theory is useless;
just that it is incomplete at some energy scale.
Let me note however that the naive assignment of relevant, marginal and irrelevant
is not always fixed in stone: quantum corrections can sometimes change the character
of an operator.
An Important Aside: Why QFT is Simple
Typically in a quantum field theory, only the relevant and marginal couplings are
important. This is basically because, as we’ve seen above, the irrelevant couplings
become small at low-energies. This is a huge help: of the infinite number of interaction
terms that we could write down, only a handful are actually needed (just two in the
case of the real scalar field described above).
Let’s look at this a little more. Suppose that we some day discover the true su-
perduper “theory of everything unimportant” that describes the world at very high
energy scales, say the GUT scale, or the Planck scale. Whatever this scale is, let’s call
it Λ. It is an energy scale, so [Λ] = 1. Now we want to understand the laws of physics
down at our puny energy scale E ≪Λ. Let’s further suppose that down at the energy
scale E, the laws of physics are described by a real scalar field. (They’re not of course:
they’re described by non-Abelian gauge fields and fermions, but the same argument
applies in that case so bear with me). This scalar field will have some complicated
interaction terms (3.1), where the precise form is dictated by all the stuff that’s going
on in the high energy superduper theory. What are these interactions? Well, we could
write our dimensionful coupling constants λ
n
in terms of dimensionless couplings g
n
,
multiplied by a suitable power of the relevant scale Λ,
λ
n
=
g
n
Λ
n−4
(3.4)
The exact values of dimensionless couplings g
n
depend on the details of the high-energy
superduper theory, but typically one expects them to be of order 1: g
n
∼ O(1). This
– 48 –
means that for experiments at small energies E ≪ Λ, the interaction terms of the
form φ
n
with n > 4 will be suppressed by powers of (E/Λ)
n−4
. This is usually a
suppression by many orders of magnitude. (e.g for the energies E explored at the
LHC, E/M
pl
∼ 10
−16
). It is this simple argument, based on dimensional analysis, that
ensures that we need only focus on the first few terms in the interaction: those which
are relevant and marginal. It also means that if we only have access to low-energy
experiments (which we do!), it’s going to be very difficult to figure out the high energy
theory (which it is!), because its effects are highly diluted except for the relevant and
marginal interactions. The discussion given above is a poor man’s version of the ideas
of effective field theory and Wilson’s renormalization group, about which you can learn
more in the “Statistical Field Theory” course.
Examples of Weakly Coupled Theories
In this course we’ll study only weakly coupled field theories i.e. ones that can truly be
considered as small perturbations of the free field theory at all energies. In this section,
we’ll look at two types of interactions
1) φ
4
theory:
L =
1
2

µ
φ∂
µ
φ −
1
2
m
2
φ
2

λ
4!
φ
4
(3.5)
with λ ≪1. We can get a hint for what the effects of this extra term will be. Expanding
out φ
4
in terms of a
p
and a

p
, we see a sum of interactions that look like
a

p
a

p
a

p
a

p
and a

p
a

p
a

p
a
p
etc. (3.6)
These will create and destroy particles. This suggests that the φ
4
Lagrangian describes
a theory in which particle number is not conserved. Indeed, we could check that the
number operator N now satisfies [H, N] ,= 0.
2) Scalar Yukawa Theory
L = ∂
µ
ψ


µ
ψ +
1
2

µ
φ∂
µ
φ −M
2
ψ

ψ −
1
2
m
2
φ
2
−gψ

ψφ (3.7)
with g ≪ M, m. This theory couples a complex scalar ψ to a real scalar φ. While
the individual particle numbers of ψ and φ are no longer conserved, we do still have
a symmetry rotating the phase of ψ, ensuring the existence of the charge Q defined
in (2.75) such that [Q, H] = 0. This means that the number of ψ particles minus the
number of ψ anti-particles is conserved. It is common practice to denote the anti-
particle as
¯
ψ.
– 49 –
The scalar Yukawa theory has a slightly worrying aspect: the potential has a stable
local minimum at φ = ψ = 0, but is unbounded below for large enough −gφ. This
means we shouldn’t try to push this theory too far.
A Comment on Strongly Coupled Field Theories
In this course we restrict attention to weakly coupled field theories where we can use
perturbative techniques. The study of strongly coupled field theories is much more
difficult, and one of the major research areas in theoretical physics. For example, some
of the amazing things that can happen include
• Charge Fractionalization: Although electrons have electric charge 1, under
the right conditions the elementary excitations in a solid have fractional charge
1/N (where N ∈ 2Z + 1). For example, this occurs in the fractional quantum
Hall effect.
• Confinement: The elementary excitations of quantum chromodynamics (QCD)
are quarks. But they never appear on their own, only in groups of three (in a
baryon) or with an anti-quark (in a meson). They are confined.
• Emergent Space: There are field theories in four dimensions which at strong
coupling become quantum gravity theories in ten dimensions! The strong cou-
pling effects cause the excitations to act as if they’re gravitons moving in higher
dimensions. This is quite extraordinary and still poorly understood. It’s called
the AdS/CFT correspondence.
3.1 The Interaction Picture
There’s a useful viewpoint in quantum mechanics to describe situations where we have
small perturbations to a well-understood Hamiltonian. Let’s return to the familiar
ground of quantum mechanics with a finite number of degrees of freedom for a moment.
In the Schr¨odinger picture, the states evolve as
i
d[ψ)
S
dt
= H[ψ)
S
(3.8)
while the operators O
S
are independent of time.
In contrast, in the Heisenberg picture the states are fixed and the operators change
in time
O
H
(t) = e
iHt
O
S
e
−iHt
[ψ)
H
= e
iHt
[ψ)
S
(3.9)
– 50 –
The interaction picture is a hybrid of the two. We split the Hamiltonian up as
H = H
0
+ H
int
(3.10)
The time dependence of operators is governed by H
0
, while the time dependence of
states is governed by H
int
. Although the split into H
0
and H
int
is arbitrary, it’s useful
when H
0
is soluble (for example, when H
0
is the Hamiltonian for a free field theory).
The states and operators in the interaction picture will be denoted by a subscript I
and are given by,
[ψ(t))
I
= e
iH
0
t
[ψ(t))
S
O
I
(t) = e
iH
0
t
O
S
e
−iH
0
t
(3.11)
This last equation also applies to H
int
, which is time dependent. The interaction
Hamiltonian in the interaction picture is,
H
I
≡ (H
int
)
I
= e
iHot
(H
int
)
S
e
−iH
0
t
(3.12)
The Schr¨odinger equation for states in the interaction picture can be derived starting
from the Schr¨odinger picture
i
d[ψ)
S
dt
= H
S
[ψ)
S
⇒ i
d
dt
_
e
−iH
0
t
[ψ)
I
_
= (H
0
+ H
int
)
S
e
−iH
0
t
[ψ)
I
⇒ i
d[ψ)
I
dt
= e
iH
0
t
(H
int
)
S
e
−iH
0
t
[ψ)
I
(3.13)
So we learn that
i
d[ψ)
I
dt
= H
I
(t) [ψ)
I
(3.14)
3.1.1 Dyson’s Formula
“Well, Birmingham has much the best theoretical physicist to work with,
Peierls; Bristol has much the best experimental physicist, Powell; Cam-
bridge has some excellent architecture. You can make your choice.”
Oppenheimer’s advice to Dyson on which university position to accept.
We want to solve (3.14). Let’s write the solution as
[ψ(t))
I
= U(t, t
0
) [ψ(t
0
))
I
(3.15)
– 51 –
where U(t, t
0
) is a unitary time evolution operator such that U(t
1
, t
2
)U(t
2
, t
3
) = U(t
1
, t
3
)
and U(t, t) = 1. Then the interaction picture Schr¨odinger equation (3.14) requires that
i
dU
dt
= H
I
(t) U (3.16)
If H
I
were a function, then we could simply solve this by
U(t, t
0
)
?
= exp
_
−i
_
t
t
0
H
I
(t

) dt

_
(3.17)
But there’s a problem. Our Hamiltonian H
I
is an operator, and we have ordering
issues. Let’s see why this causes trouble. The exponential of an operator is defined in
terms of the expansion,
exp
_
−i
_
t
t
0
H
I
(t

) dt

_
= 1 −i
_
t
t
0
H
I
(t

) dt

+
(−i)
2
2
__
t
t
0
H
I
(t

) dt

_
2
+ . . .(3.18)
But when we try to differentiate this with respect to t, we find that the quadratic term
gives us

1
2
__
t
t
0
H
I
(t

) dt

_
H
I
(t) −
1
2
H
I
(t)
__
t
t
0
H
I
(t

) dt

_
(3.19)
Now the second term here looks good, since it will give part of the H
I
(t)U that we
need on the right-hand side of (3.16). But the first term is no good since the H
I
(t)
sits the wrong side of the integral term, and we can’t commute it through because
[H
I
(t

), H
I
(t)] ,= 0 when t

,= t. So what’s the way around this?
Claim: The solution to (3.16) is given by Dyson’s Formula. (Essentially first figured
out by Dirac, although the compact notation is due to Dyson).
U(t, t
0
) = T exp
_
−i
_
t
t
0
H
I
(t

) dt

_
(3.20)
where T stands for time ordering where operators evaluated at later times are placed
to the left
T (O
1
(t
1
) O
2
(t
2
)) =
_
O
1
(t
1
) O
2
(t
2
) t
1
> t
2
O
2
(t
2
) O
1
(t
1
) t
2
> t
1
(3.21)
Expanding out the expression (3.20), we now have
U(t, t
0
) = 1 −i
_
t
t
0
dt

H
I
(t

) +
(−i)
2
2
__
t
t
0
dt

_
t
t

dt
′′
H
I
(t
′′
)H
I
(t

)
+
_
t
t
0
dt

_
t

t
0
dt
′′
H
I
(t

)H
I
(t
′′
)
_
+ . . .
– 52 –
Actually these last two terms double up since
_
t
t
0
dt

_
t
t

dt
′′
H
I
(t
′′
)H
I
(t

) =
_
t
t
0
dt
′′
_
t
′′
t
0
dt

H
I
(t
′′
)H
I
(t

)
=
_
t
t
0
dt

_
t

t
0
dt
′′
H
I
(t

)H
I
(t
′′
) (3.22)
where the range of integration in the first expression is over t
′′
≥ t

, while in the second
expression it is t

≤ t
′′
which is, of course, the same thing. The final expression is the
same as the second expression by a simple relabelling. This means that we can write
U(t, t
0
) = 1 −i
_
t
t
0
dt

H
I
(t

) + (−i)
2
_
t
t
0
dt

_
t

t
0
dt
′′
H
I
(t

)H
I
(t
′′
) + . . . (3.23)
Proof: The proof of Dyson’s formula is simpler than explaining what all the nota-
tion means! Firstly observe that under the T sign, all operators commute (since their
order is already fixed by the T sign). Thus
i

∂t
T exp
_
−i
_
t
t
0
dt

H
I
(t

)
_
= T
_
H
I
(t) exp
_
−i
_
t
t
0
dt

H
I
(t

)
__
= H
I
(t) T exp
_
−i
_
t
t
0
dt

H
I
(t

)
_
(3.24)
since t, being the upper limit of the integral, is the latest time so H
I
(t) can be pulled
out to the left.
Before moving on, I should confess that Dyson’s formula is rather formal. It is
typically very hard to compute time ordered exponentials in practice. The power of
the formula comes from the expansion which is valid when H
I
is small and is very easily
computed.
3.2 A First Look at Scattering
Let us now apply the interaction picture to field theory, starting with the interaction
Hamiltonian for our scalar Yukawa theory,
H
int
= g
_
d
3
x ψ

ψφ (3.25)
Unlike the free theories discussed in Section 2, this interaction doesn’t conserve particle
number, allowing particles of one type to morph into others. To see why this is, we use
– 53 –
the interaction picture and follow the evolution of the state: [ψ(t)) = U(t, t
0
) [ψ(t
0
)),
where U(t, t
0
) is given by Dyson’s formula (3.20) which is an expansion in powers of
H
int
. But H
int
contains creation and annihilation operators for each type of particle.
In particular,
• φ ∼ a + a

: This operator can create or destroy φ particles. Let’s call them
mesons.
• ψ ∼ b + c

: This operator can destroy ψ particles through b, and create anti-
particles through c

. Let’s call these particles nucleons. Of course, in reality
nucleons are spin 1/2 particles, and don’t arise from the quantization of a scalar
field. But we’ll treat our scalar Yukawa theory as a toy model for nucleons
interacting with mesons.
• ψ

∼ b

+ c: This operator can create nucleons through b

, and destroy anti-
nucleons through c.
Importantly, Q = N
c
−N
b
remains conserved in the presence of H
int
. At first order in
perturbation theory, we find terms in H
int
like c

b

a. This kills a meson, producing a
nucleon-anti-nucleon pair. It will contribute to meson decay φ →ψ
¯
ψ.
At second order in perturbation theory, we’ll have more complicated terms in (H
int
)
2
,
for example (c

b

a)(cba

). This term will give contributions to scattering processes
ψ
¯
ψ → φ → ψ
¯
ψ. The rest of this section is devoted to computing the quantum ampli-
tudes for these processes to occur.
To calculate amplitudes we make an important, and slightly dodgy, assumption:
Initial and final states are eigenstates of the free theory
This means that we take the initial state [i) at t → −∞, and the final state [f) at
t → +∞, to be eigenstates of the free Hamiltonian H
0
. At some level, this sounds
plausible: at t →−∞, the particles in a scattering process are far separated and don’t
feel the effects of each other. Furthermore, we intuitively expect these states to be
eigenstates of the individual number operators N, which commute with H
0
, but not
H
int
. As the particles approach each other, they interact briefly, before departing again,
each going on its own merry way. The amplitude to go from [i) to [f) is
lim
t
±
→±∞
¸f[ U(t
+
, t

) [i) ≡ ¸f[ S [i) (3.26)
where the unitary operator S is known as the S-matrix. (S is for scattering). There
are a number of reasons why the assumption of non-interacting initial and final states
is shaky:
– 54 –
• Obviously we can’t cope with bound states. For example, this formalism can’t
describe the scattering of an electron and proton which collide, bind, and leave
as a Hydrogen atom. It’s possible to circumvent this objection since it turns out
that bound states show up as poles in the S-matrix.
• More importantly, a single particle, a long way from its neighbors, is never alone
in field theory. This is true even in classical electrodynamics, where the electron
sources the electromagnetic field from which it can never escape. In quantum
electrodynamics (QED), a related fact is that there is a cloud of virtual photons
surrounding the electron. This line of thought gets us into the issues of renormal-
ization — more on this next term in the “AQFT” course. Nevertheless, motivated
by this problem, after developing scattering theory using the assumption of non-
interacting asymptotic states, we’ll mention a better way.
3.2.1 An Example: Meson Decay
Consider the relativistically normalized initial and final states,
[i) =
_
2E
p
a

p
[0)
[f) =
_
4E
q
1
E
q
2
b

q
1
c

q
2
[0) (3.27)
The initial state contains a single meson of momentum p; the final state contains a
nucleon-anti-nucleon pair of momentum q
1
and q
2
. We may compute the amplitude for
the decay of a meson to a nucleon-anti-nucleon pair. To leading order in g, it is
¸f[ S [i) = −ig ¸f[
_
d
4


(x)ψ(x)φ(x) [i) (3.28)
Let’s go slowly. We first expand out φ ∼ a +a

using (2.84). (Remember that the φ in
this formula is in the interaction picture, which is the same as the Heisenberg picture
of the free theory). The a piece will turn [i) into something proportional to [0), while
the a

piece will turn [i) into a two meson state. But the two meson state will have
zero overlap with ¸f[, and there’s nothing in the ψ and ψ

operators that lie between
them to change this fact. So we have
¸f[ S [i) = −ig ¸f[
_
d
4


(x)ψ(x)
_
d
3
k
(2π)
3
_
2E
p
_
2E

k
a

k
a

p
e
−ik·x
[0)
= −ig ¸f[
_
d
4


(x)ψ(x)e
−ip·x
[0) (3.29)
where, in the second line, we’ve commuted a

k
past a

p
, picking up a δ
(3)
( p −

k) delta-
function which kills the d
3
k integral. We now similarly expand out ψ ∼ b + c

and
– 55 –
ψ

∼ b

+ c. To get non-zero overlap with ¸f[, only the b

and c

contribute, for they
create the nucleon and anti-nucleon from [0). We then have
¸f[ S [i) = −ig ¸0[
_ _
d
4
xd
3
k
1
d
3
k
2
(2π)
6
_
E
q
1
E
q
2
_
E

k
1
E

k
2
c
q
2
b
q
1
c

k
1
b

k
2
[0) e
i(q
1
+q
2
−p)·x
= −ig (2π)
4
δ
(4)
(q
1
+ q
2
−p) (3.30)
and so we get our first quantum field theory amplitude.
Notice that the δ-function puts constraints on the possible decays. In particular, the
decay only happens at all if m ≥ 2M. To see this, we may always boost ourselves
to a reference frame where the meson is stationary, so p = (m, 0, 0, 0). Then the
delta function imposes momentum conservation, telling us that q
1
= −q
2
and m =
2
_
M
2
+[q[
2
.
Later you will learn how to turn this quantum amplitude into something more phys-
ical, namely the lifetime of the meson. The reason this is a little tricky is that we must
square the amplitude to get the probability for decay, which means we get the square
of a δ-function. We’ll explain how to deal with this in Section 3.6 below, and again in
next term’s “Standard Model” course.
3.3 Wick’s Theorem
From Dyson’s formula, we want to compute quantities like ¸f[ T ¦H
I
(x
1
) . . . H
I
(x
n
)¦ [i),
where [i) and [f) are eigenstates of the free theory. The ordering of the operators is fixed
by T, time ordering. However, since the H
I
’s contain certain creation and annihilation
operators, our life will be much simpler if we can start to move all annihilation operators
to the right where they can start killing things in [i). Recall that this is the definition
of normal ordering. Wick’s theorem tells us how to go from time ordered products to
normal ordered products.
3.3.1 An Example: Recovering the Propagator
Let’s start simple. Consider a real scalar field which we decompose in the Heisenberg
picture as
φ(x) = φ
+
(x) + φ

(x) (3.31)
where
φ
+
(x) =
_
d
3
p
(2π)
3
1
_
2E
p
a
p
e
−ip·x
φ

(x) =
_
d
3
p
(2π)
3
1
_
2E
p
a

p
e
+ip·x
(3.32)
– 56 –
where the ±signs on φ
±
make little sense, but apparently you have Pauli and Heisenberg
to blame. (They come about because φ
+
∼ e
−iEt
, which is sometimes called the positive
frequency piece, while φ

∼ e
+iEt
is the negative frequency piece). Then choosing
x
0
> y
0
, we have
T φ(x)φ(y) = φ(x)φ(y)
= (φ
+
(x) + φ

(x))(φ
+
(y) + φ

(y)) (3.33)
= φ
+
(x)φ
+
(y) + φ

(x)φ
+
(y) + φ

(y)φ
+
(x) + [φ
+
(x), φ

(y)] + φ

(x)φ

(y)
where the last line is normal ordered, and for our troubles we have picked up a factor
of D(x −y) = [φ
+
(x), φ

(y)] which is the propagator we met in (2.90). So for x
0
> y
0
we have
T φ(x)φ(y) =: φ(x)φ(y) : +D(x −y) (3.34)
Meanwhile, for y
0
> x
0
, we may repeat the calculation to find
T φ(x)φ(y) =: φ(x)φ(y) : +D(y −x) (3.35)
So putting this together, we have the final expression
T φ(x)φ(y) =: φ(x)φ(y) : +∆
F
(y −x) (3.36)
where ∆
F
(x − y) is the Feynman propagator defined in (2.93), for which we have the
integral representation

F
(x −y) =
_
d
4
k
(2π)
4
ie
ik·(x−y)
k
2
−m
2
+ iǫ
(3.37)
Let me reiterate a comment from Section 2: although T φ(x)φ(y) and : φ(x)φ(y) : are
both operators, the difference between them is a c-number function, ∆
F
(x −y).
Definition: We define the contraction of a pair of fields in a string of operators
. . . φ(x
1
) . . . φ(x
2
) . . . to mean replacing those operators with the Feynman propaga-
tor, leaving all other operators untouched. We use the notation,
. . .
¸ .. ¸
φ(x
1
) . . . φ(x
2
) . . . (3.38)
to denote contraction. So, for example,
¸ .. ¸
φ(x)φ(y) = ∆
F
(x −y) (3.39)
– 57 –
A similar discussion holds for complex scalar fields. We have
Tψ(x)ψ

(y) =: ψ(x)ψ

(y) : +∆
F
(x −y) (3.40)
prompting us to define the contraction
¸ .. ¸
ψ(x)ψ

(y) = ∆
F
(x −y) and
¸ .. ¸
ψ(x)ψ(y) =
¸ .. ¸
ψ

(x)ψ

(y) = 0 (3.41)
3.3.2 Wick’s Theorem
For any collection of fields φ
1
= φ(x
1
), φ
2
= φ(x
2
), etc, we have
T(φ
1
. . . φ
n
) =: φ
1
. . . φ
n
: + : all possible contractions : (3.42)
To see what the last part of this equation means, let’s look at an example. For n = 4,
the equation reads
T(φ
1
φ
2
φ
3
φ
4
) = : φ
1
φ
2
φ
3
φ
4
: +
¸..¸
φ
1
φ
2
: φ
3
φ
4
: +
¸..¸
φ
1
φ
3
: φ
2
φ
4
: + four similar terms
+
¸..¸
φ
1
φ
2
¸..¸
φ
3
φ
4
+
¸..¸
φ
1
φ
3
¸..¸
φ
2
φ
4
+
¸..¸
φ
1
φ
4
¸..¸
φ
2
φ
3
(3.43)
Proof: The proof of Wick’s theorem proceeds by induction and a little thought. It’s
true for n = 2. Suppose it’s true for φ
2
. . . φ
n
and now add φ
1
. We’ll take x
0
1
> x
0
k
for all k = 2, . . . , n. Then we can pull φ
1
out to the left of the time ordered product,
writing
T (φ
1
φ
2
. . . φ
n
) = (φ
+
1
+ φ

1
) (: φ
2
. . . φ
n
: + : contractions :) (3.44)
The φ

1
term stays where it is since it is already normal ordered. But in order to write
the right-hand side as a normal ordered product, the φ
+
1
term has to make its way
past the crowd of φ

k
operators. Each time it moves past φ

k
, we pick up a factor of
¸..¸
φ
1
φ
k
= ∆
F
(x
1
−x
k
) from the commutator. (Try it!)
3.3.3 An Example: Nucleon Scattering
Let’s look at ψψ →ψψ scattering. We have the initial and final states
[i) =
_
2E
p
1
_
2E
p
2
b

p
1
b

p
2
[0) ≡ [p
1
, p
2
)
[f) =
_
2E
p

1
_
2E
p

2
b

p

1
b

p

2
[0) ≡ [p

1
, p

2
) (3.45)
We can then look at the expansion of ¸f[ S [i). In fact, we really want to calculate
¸f[ S − 1 [i) since we’re not interested in situations where no scattering occurs. At
order g
2
we have the term
(−ig)
2
2
_
d
4
x
1
d
4
x
2
T
_
ψ

(x
1
)ψ(x
1
)φ(x
1


(x
2
)ψ(x
2
)φ(x
2
)
_
(3.46)
– 58 –
Now, using Wick’s theorem we see there is a piece in the string of operators which looks
like
: ψ

(x
1
)ψ(x
1


(x
2
)ψ(x
2
) :
¸ .. ¸
φ(x
1
)φ(x
2
) (3.47)
which will contribute to the scattering because the two ψ fields annihilate the ψ parti-
cles, while the two ψ

fields create ψ particles. Any other way of ordering the ψ and
ψ

fields will give zero contribution. This means that we have
¸p

1
, p

2
[ : ψ

(x
1
)ψ(x
1


(x
2
)ψ(x
2
) : [p
1
, p
2
)
= ¸p

1
, p

2
[ ψ

(x
1


(x
2
) [0) ¸0[ ψ(x
1
)ψ(x
2
) [p
1
, p
2
)
=
_
e
ip

1
·x
1
+ip

2
·x
2
+ e
ip

1
·x
2
+ip

2
·x
1
_
_
e
−ip
1
·x
1
−ip
2
·x
2
+ e
−ip
1
·x
2
−ip
2
·x
1
_
= e
ix
1
·(p

1
−p
1
)+ix
2
·(p

2
−p
2
)
+ e
ix
1
·(p

2
−p
1
)+ix
2
·(p

1
−p
2
)
+ (x
1
↔x
2
) (3.48)
where, in going to the third line, we’ve used the fact that for relativistically normalized
states,
¸0[ ψ(x) [p) = e
−ip·x
(3.49)
Now let’s insert this into (3.46), to get the expression for ¸f[ S [i) at order g
2
,
(−ig)
2
2
_
d
4
x
1
d
4
x
2
_
e
i...
+ e
i...
+ (x
1
↔x
2
)
¸
_
d
4
k
(2π)
4
ie
ik·(x
1
−x
2
)
k
2
−m
2
+ iǫ
(3.50)
where the expression in square brackets is (3.48), while the final integral is the φ
propagator which comes from the contraction in (3.47). Now the (x
1
↔ x
2
) terms
double up with the others to cancel the factor of 1/2 out front. Meanwhile, the x
1
and
x
2
integrals give delta-functions. We’re left with the expression
(−ig)
2
_
d
4
k
(2π)
4
i(2π)
8
k
2
−m
2
+ iǫ
_
δ
(4)
(p

1
−p
1
+ k) δ
(4)
(p

2
−p
2
−k)

(4)
(p

2
−p
1
+ k) δ
(4)
(p

1
−p
2
−k)
¸
(3.51)
Finally, we can trivially do the d
4
k integral using the delta-functions to get
i(−ig)
2
_
1
(p
1
−p

1
)
2
−m
2
+ iǫ
+
1
(p
1
−p

2
)
2
−m
2
+ iǫ
_
(2π)
4
δ
(4)
(p
1
+ p
2
−p

1
−p

2
)
In fact, for this process we may drop the +iǫ terms since the denominator is never
zero. To see this, we can go to the center of mass frame, where p
1
= − p
2
and, by
– 59 –
momentum conservation, [ p
1
[ = [ p

1
[. This ensures that the 4-momentum of the meson
is k = (0, p −p

), so k
2
< 0. We therefore have the end result,
i(−ig)
2
_
1
(p
1
−p

1
)
2
−m
2
+
1
(p
1
−p

2
)
2
−m
2
_
(2π)
4
δ
(4)
(p
1
+ p
2
−p

1
−p

2
) (3.52)
We will see another, much simpler way to reproduce this result shortly using Feynman
diagrams. This will also shed light on the physical interpretation.
This calculation is also relevant for other scattering processes, such as
¯
ψ
¯
ψ →
¯
ψ
¯
ψ,
ψ
¯
ψ →ψ
¯
ψ. Each of these comes from the term (3.48) in Wick’s theorem. However, we
will never find a term that contributes to scattering ψψ →
¯
ψ
¯
ψ, for this would violate
the conservation of Q charge.
Another Example: Meson-Nucleon Scattering
If we want to compute ψφ →ψφ scattering at order g
2
, we would need to pick out the
term
: ψ

(x
1
)φ(x
1
)ψ(x
2
)φ(x
2
) :
¸ .. ¸
ψ(x
1


(x
2
) (3.53)
and a similar term with ψ and ψ

exchanged. Once more, this term also contributes to
similar scattering processes, including
¯
ψφ →
¯
ψφ and φφ →ψ
¯
ψ.
3.4 Feynman Diagrams
“Like the silicon chips of more recent years, the Feynman diagram was
bringing computation to the masses.”
Julian Schwinger
As the above example demonstrates, to actually compute scattering amplitudes using
Wick’s theorem is rather tedious. There’s a much better way. It requires drawing pretty
pictures. These pictures represent the expansion of ¸f[ S [i) and we will learn how to
associate numbers (or at least integrals) to them. These pictures are called Feynman
diagrams.
The object that we really want to compute is ¸f[ S−1 [i), since we’re not interested in
processes where no scattering occurs. The various terms in the perturbative expansion
can be represented pictorially as follows
• Draw an external line for each particle in the initial state [i) and each particle
in the final state [f). We’ll choose dotted lines for mesons, and solid lines for
nucleons. Assign a directed momentum p to each line. Further, add an arrow to
– 60 –
solid lines to denote its charge; we’ll choose an incoming (outgoing) arrow in the
initial state for ψ (
¯
ψ). We choose the reverse convention for the final state, where
an outgoing arrow denotes ψ.
• Join the external lines together with trivalent vertices
ψ
ψ
+
φ
Each such diagram you can draw is in 1-1 correspondence with the terms in the
expansion of ¸f[ S −1 [i).
3.4.1 Feynman Rules
To each diagram we associate a number, using the Feynman rules
• Add a momentum k to each internal line
• To each vertex, write down a factor of
(−ig) (2π)
4
δ
(4)
(

i
k
i
) (3.54)
where

k
i
is the sum of all momenta flowing into the vertex.
• For each internal dotted line, corresponding to a φ particle with momentum k,
we write down a factor of
_
d
4
k
(2π)
4
i
k
2
−m
2
+ iǫ
(3.55)
We include the same factor for solid internal ψ lines, with m replaced by the
nucleon mass M.
– 61 –
3.5 Examples of Scattering Amplitudes
Let’s apply the Feynman rules to compute the amplitudes for various processes. We
start with something familiar:
Nucleon Scattering Revisited
Let’s look at how this works for the ψψ → ψψ scattering at order g
2
. We can write
down the two simplest diagrams contributing to this process. They are shown in Figure
9.
p
2
p
1
p
1
/
p
2
/
p
2
p
1
/
p
/
2
1
+
p
k k
Figure 9: The two lowest order Feynman diagrams for nucleon scattering.
Applying the Feynman rules to these diagrams, we get
i(−ig)
2
_
1
(p
1
−p

1
)
2
−m
2
+
1
(p
1
−p

2
)
2
−m
2
_
(2π)
4
δ
(4)
(p
1
+ p
2
−p

1
−p

2
) (3.56)
which agrees with the calculation (3.51) that we performed earlier. There is a nice
physical interpretation of these diagrams. We talk, rather loosely, of the nucleons
exchanging a meson which, in the first diagram, has momentum k = (p
1
−p

1
) = (p
2
−p

2
).
This meson doesn’t satisfy the usual energy dispersion relation, because k
2
,= m
2
: the
meson is called a virtual particle and is said to be off-shell (or, sometimes, off mass-
shell). Heuristically, it can’t live long enough for its energy to be measured to great
accuracy. In contrast, the momentum on the external, nucleon legs all satisfy p
2
= M
2
,
the mass of the nucleon. They are on-shell. One final note: the addition of the two
diagrams above ensures that the particles satisfy Bose statistics.
There are also more complicated diagrams which will contribute to the scattering
process at higher orders. For example, we have the two diagrams shown in Figures
10 and 11, and similar diagrams with p

1
and p

2
exchanged. Using the Feynman rules,
each of these diagrams translates into an integral that we will not attempt to calculate
here. And so we go on, with increasingly complicated diagrams, all appearing at higher
order in the coupling constant g.
– 62 –
p
1
p
1
/
p
2
p
2
/
p
1
p
1
/
p
2
p
2
/
Figure 10: A contribution at O(g
4
). Figure 11: A contribution at O(g
6
)
Amplitudes
Our final result for the nucleon scattering amplitude ¸f[ S −1 [i) at order g
2
was
i(−ig)
2
_
1
(p
1
−p

1
)
2
−m
2
+
1
(p
1
−p

2
)
2
−m
2
_
(2π)
4
δ
(4)
(p
1
+ p
2
−p

1
−p

2
)
The δ-function follows from the conservation of 4-momentum which, in turn, follows
from spacetime translational invariance. It is common to all S-matrix elements. We will
define the amplitude /
fi
by stripping off this momentum-conserving delta-function,
¸f[ S −1 [i) = i /
fi
(2π)
4
δ
(4)
(p
F
−p
I
) (3.57)
where p
I
(p
F
) is the sum of the initial (final) 4-momenta, and the factor of i out front
is a convention which is there to match non-relativistic quantum mechanics. We can
now refine our Feynman rules to compute the amplitude i/
fi
itself:
• Draw all possible diagrams with appropriate external legs and impose 4-momentum
conservation at each vertex.
• Write down a factor of (−ig) at each vertex.
• For each internal line, write down the propagator
• Integrate over momentum k flowing through each loop
_
d
4
k/(2π)
4
.
This last step deserves a short explanation. The diagrams we’ve computed so far have
no loops. They are tree level diagrams. It’s not hard to convince yourself that in
tree diagrams, momentum conservation at each vertex is sufficient to determine the
momentum flowing through each internal line. For diagrams with loops, such as those
shown in Figures 10 and 11, this is no longer the case.
– 63 –
p
2
p
1
p
2
p
1
p
1
/
p
2
/
p
/
p
/
2
1
+
Figure 12: The two lowest order Feynman diagrams for nucleon to meson scattering.
Nucleon to Meson Scattering
Let’s now look at the amplitude for a nucleon-anti-nucleon pair to annihilate into a
pair of mesons: ψ
¯
ψ →φφ. The simplest Feynman diagrams for this process are shown
in Figure 12 where the virtual particle in these diagrams is now the nucleon ψ rather
than the meson φ. This fact is reflected in the denominator of the amplitudes which
are given by
i/ = (−ig)
2
_
i
(p
1
−p

1
)
2
−M
2
+
i
(p
1
−p

2
)
2
−M
2
_
(3.58)
As in (3.52), we’ve dropped the iǫ from the propagators as the denominator never
vanishes.
Nucleon-Anti-Nucleon Scattering
p
2
p
1
p
1
/
p
2
/
p
2
p
1
p
1
/
p
2
/
+
Figure 13: The two lowest order Feynman diagrams for nucleon-anti-nucleon scattering.
For the scattering of a nucleon and an anti-nucleon, ψ
¯
ψ → ψ
¯
ψ, the Feynman
diagrams are a little different. At lowest order, they are given by the diagrams of
Figure 13. It is a simple matter to write down the amplitude using the Feynman rules,
i/ = (−ig)
2
_
i
(p
1
−p

1
)
2
−m
2
+
i
(p
1
+ p
2
)
2
−m
2
+ iǫ
_
(3.59)
– 64 –
Notice that the momentum dependence in the sec-
Figure 14:
ond term is different from that of nucleon-nucleon
scattering (3.56), reflecting the different Feynman di-
agram that contributes to the process. In the center
of mass frame, p
1
= − p
2
, the denominator of the sec-
ond term is 4(M
2
+p
2
1
) −m
2
. If m < 2M, then this
term never vanishes and we may drop the iǫ. In con-
trast, if m > 2M, then the amplitude corresponding
to the second diagram diverges at some value of p.
In this case it turns out that we may also neglect the
iǫ term, although for a different reason: the meson
is unstable when m > 2M, a result we derived in
(3.30). When correctly treated, this instability adds
a finite imaginary piece to the denominator which
overwhelms the iǫ. Nonetheless, the increase in the scattering amplitude which we see
in the second diagram when 4(M
2
+ p
2
) = m
2
is what allows us to discover new parti-
cles: they appear as a resonance in the cross section. For example, the Figure 14 shows
the cross-section (roughly the amplitude squared) plotted vertically for e
+
e

→µ
+
µ

scattering from the ALEPH experiment in CERN. The horizontal axis shows the center
of mass energy. The curve rises sharply around 91 GeV, the mass of the Z-boson.
Meson Scattering
For φφ → φφ, the simplest diagram we can write
p
2
p
1
/
p
/
2
1
p
p
1
/
p
1
/
1
p
p
/
2
k
k+
− + k
− k
Figure 15:
down has a single loop, and momentum conservation at
each vertex is no longer sufficient to determine every
momentum passing through the diagram. We choose
to assign the single undetermined momentum k to the
right-hand propagator. All other momenta are then de-
termined. The amplitude corresponding to the diagram
shown in the figure is
(−ig)
4
_
d
4
k
(2π)
4
1
(k
2
−M
2
+ iǫ)((k + p

1
)
2
−M
2
+ iǫ)

1
((k + p

1
−p
1
)
2
−M
2
+ iǫ)((k −p

2
)
2
−M
2
+ iǫ)
These integrals can be tricky. For large k, this integral goes as
_
d
4
k/k
8
, which is at
least convergent as k →∞. But this won’t always be the case!
– 65 –
3.5.1 Mandelstam Variables
We see that in many of the amplitudes above — in particular those that include the
exchange of just a single particle — the same combinations of momenta are appear-
ing frequently in the denominators. There are standard names for various sums and
differences of momenta: they are known as Mandelstam variables. They are
s = (p
1
+ p
2
)
2
= (p

1
+ p

2
)
2
t = (p
1
−p

1
)
2
= (p
2
−p

2
)
2
(3.60)
u = (p
1
−p

2
)
2
= (p
2
−p

1
)
2
where, as in the examples above, p
1
and p
2
are the momenta of the two initial particles,
and p

1
and p

2
are the momenta of the final two particles. We can define these variables
whether the particles involved in the scattering are the same or different. To get a feel
for what these variables mean, let’s assume all four particles are the same. We sit in
the center of mass frame, so that the initial two particles have four-momenta
p
1
= (E, 0, 0, p) and p
2
= (E, 0, 0, −p) (3.61)
The particles then scatter at some angle θ and leave with momenta
p

1
= (E, 0, p sinθ, p cos θ) and p

2
= (E, 0, −p sinθ, −p cos θ) (3.62)
Then from the above definitions, we have that
s = 4E
2
and t = −2p
2
(1 −cos θ) and u = −2p
2
(1 + cos θ) (3.63)
The variable s measures the total center of mass energy of the collision, while the
variables t and u are measures of the momentum exchanged between particles. (They
are basically equivalent, just with the outgoing particles swapped around). Now the
amplitudes that involve exchange of a single particle can be written simply in terms of
the Mandelstam variables. For example, for nucleon-nucleon scattering, the amplitude
(3.56) is schematically / ∼ (t − m
2
)
−1
+ (u − m
2
)
−1
. For the nucleon-anti-nucleon
scattering, the amplitude (3.59) is / ∼ (t − m
2
)
−1
+ (s − m
2
)
−1
. We say that the
first case involves “t-channel” and “u-channel” diagrams. Meanwhile the nucleon-anti-
nucleon scattering is said to involve “t-channel” and “s-channel” diagrams. (The first
diagram indeed includes a vertex that looks like the letter “T”).
Note that there is a relationship between the Mandelstam variables. When all the
masses are the same we have s+t +u = 4M
2
. When the masses of all 4 particles differ,
this becomes s + t + u =

i
M
2
i
.
– 66 –
3.5.2 The Yukawa Potential
So far we’ve computed the quantum amplitudes for various scattering processes. But
these quantities are a little abstract. In Section 3.6 below (and again in next term’s
“Standard Model” course) we’ll see how to turn amplitudes into measurable quantities
such as cross-sections, or the lifetimes of unstable particles. Here we’ll instead show
how to translate the amplitude (3.52) for nucleon scattering into something familiar
from Newtonian mechanics: a potential, or force, between the particles.
Let’s start by asking a simple question in classical field theory that will turn out to
be relevant. Suppose that we have a fixed δ-function source for a real scalar field φ,
that persists for all time. What is the profile of φ(x)? To answer this, we must solve
the static Klein-Gordon equation,
−∇
2
φ + m
2
φ = δ
(3)
(x) (3.64)
We can solve this using the Fourier transform,
φ(x) =
_
d
3
k
(2π)
3
e
i

k·x
˜
φ(

k) (3.65)
Plugging this into (3.64) tells us that (

k
2
+ m
2
)
˜
φ(

k) = 1, giving us the solution
φ(x) =
_
d
3
k
(2π)
3
e
i

k·x

k
2
+ m
2
(3.66)
Let’s now do this integral. Changing to polar coordinates, and writing

k x = kr cos θ,
we have
φ(x) =
1
(2π)
2
_

0
dk
k
2
k
2
+ m
2
2 sin kr
kr
=
1
(2π)
2
r
_
+∞
−∞
dk
k sin kr
k
2
+ m
2
=
1
2πr
Re
__
+∞
−∞
dk
2πi
ke
ikr
k
2
+ m
2
_
(3.67)
We compute this last integral by closing the contour in the upper half plane k →+i∞,
picking up the pole at k = +im. This gives
φ(x) =
1
4πr
e
−mr
(3.68)
The field dies off exponentially quickly at distances 1/m, the Compton wavelength of
the meson.
– 67 –
Now we understand the profile of the φ field, what does this have to do with the force
between ψ particles? We do very similar calculations to that above in electrostatics
where a charged particle acts as a δ-function source for the gauge potential: −∇
2
A
0
=
δ
(3)
(x), which is solved by A
0
= 1/4πr. The profile for A
0
then acts as the potential
energy for another charged (test) particle moving in this background. Can we give the
same interpretation to our scalar field? In other words, is there a classical limit of the
scalar Yukawa theory where the ψ particles act as δ-function sources for φ, creating
the profile (3.68)? And, if so, is this profile then felt as a static potential? The answer
is essentially yes, at least in the limit M ≫ m. But the correct way to describe the
potential felt by the ψ particles is not to talk about classical fields at all, but instead
work directly with the quantum amplitudes.
Our strategy is to compare the nucleon scattering amplitude (3.52) to the corre-
sponding amplitude in non-relativistic quantum mechanics for two particles interacting
through a potential. To make this comparison, we should first take the non-relativistic
limit of (3.52). Let’s work in the center of mass frame, with p ≡ p
1
= − p
2
and
p

≡ p

1
= − p

2
. The non-relativistic limit means [ p[ ≪ M which, by momentum
conservation, ensures that [ p

[ ≪ M. In fact one can check that, for this particular
example, this limit doesn’t change the scattering amplitude (3.52): it’s given by
i/ = +ig
2
_
1
( p −p

)
2
+ m
2
+
1
( p + p

)
2
+ m
2
_
(3.69)
How do we compare this to scattering in quantum mechanics? Consider two particles,
separated by a distance r, interacting through a potential U(r). In non-relativistic
quantum mechanics, the amplitude for the particles to scatter from momentum states
± p into momentum states ± p

can be computed in perturbation theory, using the
techniques described in Section 3.1. To leading order, known in this context as the
Born approximation, the amplitude is given by
¸ p

[ U(r) [p ) = −i
_
d
3
r U(r)e
−i( p− p

)·r
(3.70)
There’s a relative factor of (2M)
2
that arises in comparing the quantum field theory
amplitude / to ¸p

[ U(r) [ p), that can be traced to the relativistic normalization of the
states [p
1
, p
2
). (It is also necessary to get the dimensions of the potential to work out
correctly). Including this factor, and equating the expressions for the two amplitudes,
we get
_
d
3
r U(r) e
−i( p− p

)·r
=
−λ
2
( p −p

)
2
+ m
2
(3.71)
– 68 –
where we’ve introduced the dimensionless parameter λ = g/2M. We can trivially invert
this to find,
U(r) = −λ
2
_
d
3
p
(2π)
3
e
i p·r
p
2
+ m
2
(3.72)
But this is exactly the integral (3.66) we just did in the classical theory. We have
U(r) =
−λ
2
4πr
e
−mr
(3.73)
This is the Yukawa potential. The force has a range 1/m, the Compton wavelength of
the exchanged particle. The minus sign tells us that the potential is attractive.
Notice that quantum field theory has given us an entirely new perspective on the
nature of forces between particles. Rather than being a fundamental concept, the force
arises from the virtual exchange of other particles, in this case the meson. In Section 6
of these lectures, we will see how the Coulomb force arises from quantum field theory
due to the exchange of virtual photons.
We could repeat the calculation for nucleon-anti-nucleon scattering. The amplitude
from field theory is given in (3.59). The first term in this expression gives the same
result as for nucleon-nucleon scattering with the same sign. The second term vanishes in
the non-relativisitic limit (it is an example of an interaction that doesn’t have a simple
Newtonian interpretation). There is no longer a factor of 1/2 in (3.70), because the
incoming/outgoing particles are not identical, so we learn that the potential between
a nucleon and anti-nucleon is again given by (3.73). This reveals a key feature of
forces arising due to the exchange of scalars: they are universally attractive. Notice
that this is different from forces due to the exchange of a spin 1 particle — such as
electromagnetism — where the sign flips when we change the charge. However, for
forces due to the exchange of a spin 2 particle — i.e. gravity — the force is again
universally attractive.
3.5.3 φ
4
Theory
Let’s briefly look at the Feynman rules and scattering amplitudes for the interaction
Hamiltonian
H
int
=
λ
4!
φ
4
(3.74)
The theory now has a single interaction vertex, which comes with a factor of (−iλ),
while the other Feynman rules remain the same. Note that we assign (−iλ) to the
– 69 –
vertex rather than (−iλ/4!). To see why this is, we can look at φφ → φφ scattering,
which has its lowest contribution at order λ, with the term
−iλ
4!
¸p

1
, p

2
[ : φ(x)φ(x)φ(x)φ(x) : [p
1
, p
2
) (3.75)
Any one of the fields can do the job of annihilation or creation. This gives 4! different
contractions, which cancels the 1/4! sitting out front.
Feynman diagrams in the φ
4
theory sometimes come with extra
−iλ
Figure 16:
combinatoric factors (typically 2 or 4) which are known as symmetry
factors that one must take into account. For more details, see the book
by Peskin and Schroeder.
Using the Feynman rules, the scattering amplitude for φφ → φφ is
simply i/ = −iλ. Note that it doesn’t depend on the angle at which
the outgoing particles emerge: in φ
4
theory the leading order two-particle scattering
occurs with equal probability in all directions. Translating this into a potential between
two mesons, we have
U(r) =
λ
(2m)
2
_
d
3
p
(2π)
3
e
+i p·r
=
λ
(2m)
2
δ
(3)
(r) (3.76)
So scattering in φ
4
theory is due to a δ-function potential. The particles don’t know
what hit them until it’s over.
3.5.4 Connected Diagrams and Amputated Diagrams
We’ve seen how one can compute scattering amplitudes by writing down all Feynman
diagrams and assigning integrals to them using the Feynman rules. In fact, there are
a couple of caveats about what Feynman diagrams you should write down. Both of
these caveats are related to the assumption we made earlier that “initial and final states
are eigenstates of the free theory” which, as we mentioned at the time, is not strictly
accurate. The two caveats which go some way towards ameliorating the problem are
the following
• We consider only connected Feynman diagrams, where every part of the diagram
is connected to at least one external line. As we shall see shortly, this will be
related to the fact that the vacuum [0) of the free theory is not the true vacuum
[Ω) of the interacting theory. An example of a diagram that is not connected is
shown in Figure 17.
– 70 –
• We do not consider diagrams with loops on external lines, for example the diagram
shown in the Figure 18. We will not explain how to take these into account in this
course, but you will discuss them next term. They are related to the fact that the
one-particle states of the free theory are not the same as the one-particle states of
the interacting theory. In particular, correctly dealing with these diagrams will
account for the fact that particles in interacting quantum field theories are never
alone, but surrounded by a cloud of virtual particles. We will refer to diagrams
in which all loops on external legs have been cut-off as “amputated”.
Figure 17: A disconnected diagram. Figure 18: An un-amputated diagram
3.6 What We Measure: Cross Sections and Decay Rates
So far we’ve learnt to compute the quantum amplitudes for particles decaying or scat-
tering. As usual in quantum theory, the probabilities for things to happen are the
(modulus) square of the quantum amplitudes. In this section we will compute these
probabilities, known as decay rates and cross sections. One small subtlety here is that
the S-matrix elements ¸f[ S − 1 [i) all come with a factor of (2π)
4
δ
(4)
(p
F
− p
I
), so we
end up with the square of a delta-function. As we will now see, this comes from the
fact that we’re working in an infinite space.
3.6.1 Fermi’s Golden Rule
Let’s start with something familiar and recall how to derive Fermi’s golden rule from
Dyson’s formula. For two energy eigenstates [m) and [n), with E
m
,= E
n
, we have to
leading order in the interaction,
¸m[ U(t) [n) = −i ¸m[
_
t
0
dt H
I
(t) [n)
= −i ¸m[ H
int
[n)
_
t
0
dt

e
iωt

= −¸m[ H
int
[n)
e
iωt
−1
ω
(3.77)
– 71 –
where ω = E
m
−E
n
. This gives us the probability for the transition from [n) to [m) in
time t, as
P
n→m
(t) = [ ¸m[ U(t) [n) [
2
= 2[ ¸m[ H
int
[n) [
2
_
1 −cos ωt
ω
2
_
(3.78)
The function in brackets is plotted in Figure 19 for fixed t.
Figure 19:
We see that in time t, most transitions happen in a region
between energy eigenstates separated by ∆E = 2π/t. As
t →∞, the function in the figure starts to approach a delta-
function. To find the normalization, we can calculate
_
+∞
−∞

_
1 −cos ωt
ω
2
_
= πt

_
1 −cos ωt
ω
2
_
→ πtδ(ω) as t →∞
Consider now a transition to a cluster of states with density
ρ(E). In the limit t →∞, we get the transition probability
P
n→m
=
_
dE
m
ρ(E
m
) 2[ ¸m[ H
int
[n) [
2
_
1 −cos ωt
ω
2
_
→ 2π [ ¸m[ H
int
[n) [
2
ρ(E
n
)t (3.79)
which gives a constant probability for the transition per unit time for states around
the same energy E
n
∼ E
m
= E.
˙
P
n→m
= 2π[ ¸m[ H
int
[n) [
2
ρ(E) (3.80)
This is Fermi’s Golden Rule.
In the above derivation, we were fairly careful with taking the limit as t → ∞.
Suppose we were a little sloppier, and first chose to compute the amplitude for the
state [n) at t →−∞ to transition to the state [m) at t →+∞. Then we get
−i ¸m[
_
t=+∞
t=−∞
H
I
(t) [n) = −i ¸m[ H
int
[n) 2πδ(ω) (3.81)
Now when squaring the amplitude to get the probability, we run into the problem of
the square of the delta-function: P
n→m
= [ ¸m[ H
int
[n) [
2
(2π)
2
δ(ω)
2
. Tracking through
the previous computations, we realize that the extra infinity is coming because P
m→n
– 72 –
is the probability for the transition to happen in infinite time t → ∞. We can write
the delta-functions as
(2π)
2
δ(ω)
2
= (2π)δ(ω) T (3.82)
where T is shorthand for t → ∞ (we used a very similar trick when looking at the
vacuum energy in (2.25)). We now divide out by this power of T to get the transition
probability per unit time,
˙
P
n→m
= 2π[ ¸m[ H
int
[n) [
2
δ(ω) (3.83)
which, after integrating over the density of final states, gives us back Fermi’s Golden
rule. The reason that we’ve stressed this point is because, in our field theory calcula-
tions, we’ve computed the amplitudes in the same way as (3.81), and the square of the
δ
(4)
-functions will just be re-interpreted as spacetime volume factors.
3.6.2 Decay Rates
Let’s now look at the probability for a single particle [i) of momentum p
I
(I=initial)
to decay into some number of particles [f) with momentum p
i
and total momentum
p
F
=

i
p
i
. This is given by
P =
[ ¸f[ S [i) [
2
¸f[ f) ¸i[ i)
(3.84)
Our states obey the relativistic normalization formula (2.65),
¸i[ i) = (2π)
3
2E
p
I
δ
(3)
(0) = 2E
p
I
V (3.85)
where we have replaced δ
(3)
(0) by the volume of 3-space. Similarly,
¸f[ f) =

final states
2E
p
i
V (3.86)
If we place our initial particle at rest, so p
I
= 0 and E
p
I
= m, we get the probability
for decay
P =
[/
fi
[
2
2mV
(2π)
4
δ
(4)
(p
I
−p
F
) V T

final states
1
2E
p
i
V
(3.87)
where, as in the second derivation of Fermi’s Golden Rule, we’ve exchanged one of the
delta-functions for the volume of spacetime: (2π)
4
δ
(4)
(0) = V T. The amplitudes /
fi
are, of course, exactly what we’ve been computing. (For example, in (3.30), we saw
– 73 –
that / = −g for a single meson decaying into two nucleons). We can now divide out
by T to get the transition function per unit time. But we still have to worry about
summing over all final states. There are two steps: the first is to integrate over all
possible momenta of the final particles: V
_
d
3
p
i
/(2π)
3
. The factors of spatial volume
V in this measure cancel those in (3.87), while the factors of 1/2E
p
i
in (3.87) conspire
to produce the Lorentz invariant measure for 3-momentum integrals. The result is an
expression for the density of final states given by the Lorentz invariant measure
dΠ = (2π)
4
δ
(4)
(p
F
−p
I
)

final states
d
3
p
i
(2π)
3
1
2E
p
i
(3.88)
The second step is to sum over all final states with different numbers (and possibly
types) of particles. This gives us our final expression for the decay probability per unit
time, Γ =
˙
P.
Γ =
1
2m

final states
_
[/
fi
[
2
dΠ (3.89)
Γ is called the width of the particle. It is equal to the reciprocal of the half-life τ = 1/Γ.
3.6.3 Cross Sections
Collide two beams of particles. Sometimes the particles will hit and bounce off each
other; sometimes they will pass right through. The fraction of the time that they collide
is called the cross section and is denoted by σ. If the incoming flux F is defined to
be the number of incoming particles per area per unit time, then the total number of
scattering events N per unit time is given by,
N = Fσ (3.90)
We would like to calculate σ from quantum field theory. In fact, we can calculate a more
sensitive quantity dσ known as the differential cross section which is the probability
for a given scattering process to occur in the solid angle (θ, φ). More precisely
dσ =
Differential Probability
Unit Time Unit Flux
=
1
4E
1
E
2
V
1
F
[/
fi
[
2
dΠ (3.91)
where we’ve used the expression for probability per unit time that we computed in the
previous subsection. E
1
and E
2
are the energies of the incoming particles. We now
need an expression for the unit flux. For simplicity, let’s sit in the center of mass frame
of the collision. We’ve been considering just a single particle per spatial volume V ,
– 74 –
meaning that the flux is given in terms of the 3-velocities v
i
as F = [v
1
−v
2
[/V . This
then gives,
dσ =
1
4E
1
E
2
1
[v
1
−v
2
[
[/
fi
[
2
dΠ (3.92)
If you want to write this in terms of momentum, then recall from your course on special
relativity that the 3-velocities v
i
are related to the momenta by v = p/m

1 −v
2
=
p/p
0
.
Equation (3.92) is our final expression relating the S-matrix to the differential cross
section. You may now take your favorite scattering amplitude, and compute the proba-
bility for particles to fly out at your favorite angles. This will involve doing the integral
over the phase space of final states, with measure dΠ. Notice that different scattering
amplitudes have different momentum dependence and will result in different angular
dependence in scattering amplitudes. For example, in φ
4
theory the amplitude for tree
level scattering was simply / = −λ. This results in isotropic scattering. In contrast,
for nucleon-nucleon scattering we have schematically / ∼ (t − m
2
)
−1
+ (u − m
2
)
−1
.
This gives rise to angular dependence in the differential cross-section, which follows
from the fact that, for example, t = −2[ p[
2
(1 −cos θ), where θ is the angle between the
incoming and outgoing particles.
3.7 Green’s Functions
So far we’ve learnt to compute scattering amplitudes. These are nice and physical (well
– they’re directly related to cross-sections and decay rates which are physical) but there
are many questions we want to ask in quantum field theory that aren’t directly related
to scattering experiments. For example, we might want to compute the viscosity of
the quark gluon plasma, or the optical conductivity in a tentative model of strange
metals, or figure out the non-Gaussianity of density perturbations arising in the CMB
from novel models of inflation. All of these questions are answered in the framework of
quantum field theory by computing elementary objects known as correlation functions.
In this section we will briefly define correlation functions, explain how to compute them
using Feynman diagrams, and then relate them back to scattering amplitudes. We’ll
leave the relationship to other physical phenomena to other courses.
We’ll denote the true vacuum of the interacting theory as [Ω). We’ll normalize H
such that
H[Ω) = 0 (3.93)
– 75 –
and ¸Ω[ Ω) = 1. Note that this is different from the state we’ve called [0) which is the
vacuum of the free theory and satisfies H
0
[0) = 0. Define
G
(n)
(x
1
, . . . , x
n
) = ¸Ω[ T φ
H
(x
1
) . . . φ
H
(x
n
) [Ω) (3.94)
where φ
H
is φ in the Heisenberg picture of the full theory, rather than the interaction
picture that we’ve been dealing with so far. The G
(n)
are called correlation functions,
or Green’s functions. There are a number of different ways of looking at these objects
which tie together nicely. Let’s start by asking how to compute G
(n)
using Feynman
diagrams. We prove the following result
Claim: We use the notation φ
1
= φ(x
1
), and write φ
1H
to denote the field in the
Heisenberg picture, and φ
1I
to denote the field in the interaction picture. Then
G
(n)
(x
1
, . . . , x
n
) = ¸Ω[ T φ
1H
. . . φ
nH
[Ω) =
¸0[ Tφ
1I
. . . φ
nI
S [0)
¸0[ S [0)
(3.95)
where the operators on the right-hand side are evaluated on [0), the vacuum of the free
theory.
Proof: Take t
1
> t
2
> . . . > t
n
. Then we can drop the T and write the numera-
tor of the right-hand side as
¸0[ U
I
(+∞, t
1

1I
U(t
1
, t
2
) φ
2I
. . . φ
nI
U
I
(t
n
, −∞) [0)
= ¸0[ U
I
(+∞, t
1

1H
. . . φ
nH
U
I
(t
n
, −∞) [0)
where we’ve used all the intermediate factors of U
I
(t
k
, t
k+1
) = T exp(−i
_
t
k+1
t
k
H
I
) to
convert φ
I
into φ
H
. Now let’s deal with the two remaining U(t, ±∞) at either end of
the string of operators. Consider an arbitrary state [Ψ)
¸Ψ[ U
I
(t, −∞) [0) = ¸Ψ[ U(t, −∞) [0) (3.96)
where U(t, −∞) is the Schr¨odinger evolution operator, and the equality above follows
because H
0
[0) = 0. Now insert a complete set of states, which we take to be energy
eigenstates of H = H
0
+ H
int
,
¸Ψ[ U(t, −∞) [0) = ¸Ψ[ U(t, −∞)
_
[Ω) ¸Ω[ +

n=0
[n) ¸n[
_
[0)
= ¸Ψ[ Ω) ¸Ω[ 0) + lim
t

→−∞

n=0
e
iEn(t

−t)
¸Ψ[ n) ¸n[ 0) (3.97)
– 76 –
But the last term vanishes. This follows from the Riemann-Lebesgue lemma which says
that for any well-behaved function
lim
µ→∞
_
b
a
dxf(x)e
iµx
= 0 (3.98)
Why is this relevant? The point is that the

n
in (3.97) is really an integral
_
dn,
because all states are part of a continuum due to the momentum. (There is a caveat
here: we want the vacuum [Ω) to be special, so that it sits on its own, away from the
continuum of the integral. This means that we must be working in a theory with a
mass gap – i.e. with no massless particles). So the Riemann-Lebesgue lemma gives us
lim
t

→−∞
¸Ψ[ U(t, t

) [0) = ¸Ψ[ Ω) ¸Ω[ 0) (3.99)
(Notice that to derive this result, Peskin and Schroeder instead send t → −∞ in a
slightly imaginary direction, which also does the job). We now apply the formula
(3.99), to the top and bottom of the right-hand side of (3.95) to find
¸0[ Ω) ¸Ω[ Tφ
1H
. . . φ
nH
[Ω) ¸Ω[ 0)
¸0[ Ω) ¸Ω[ Ω) ¸Ω[ 0)
(3.100)
which, using the normalization ¸Ω[ Ω) = 1, gives us the left-hand side, completing the
proof. .
3.7.1 Connected Diagrams and Vacuum Bubbles
We’re getting closer to our goal of computing the Green’s functions G
(n)
since we
can compute both ¸0[ Tφ
I
(x
1
) . . . φ
I
(x
n
) S [0) and ¸0[ S [0) using the same methods
we developed for S-matrix elements; namely Dyson’s formula and Wick’s theorem or,
alternatively, Feynman diagrams. But what about dividing one by the other? What’s
that all about? In fact, it has a simple interpretation. For the following discussion,
we will work in φ
4
theory. Since there is no ambiguity in the different types of line
in Feynman diagrams, we will represent the φ particles as solid lines, rather than the
dashed lines that we used previously. Then we have the diagramatic expansion for
¸0[ S [0).
¸0[ S [0) = 1 + +
(
+ +
)
+ . . . (3.101)
These diagrams are called vacuum bubbles. The combinatoric factors (as well as the
symmetry factors) associated with each diagram are such that the whole series sums
– 77 –
to an exponential,
¸0[ S [0) = exp
(
+ + + ...
)
(3.102)
So the amplitude for the vacuum of the free theory to evolve into itself is ¸0[ S [0) =
exp(all distinct vacuum bubbles). A similar combinatoric simplification occurs for generic
correlation functions. Remarkably, the vacuum diagrams all add up to give the same
exponential. With a little thought one can show that
¸0[ Tφ
1
. . . φ
n
[0) =
_

connected diagrams
_
¸0[ S [0) (3.103)
where “connected” means that every part of the diagram is connected to at least one
of the external legs. The upshot of all this is that dividing by ¸0[ S [0) has a very nice
interpretation in terms of Feynman diagrams: we need only consider the connected
Feynman diagrams, and don’t have to worry about the vacuum bubbles. Combining
this with (3.95), we learn that the Green’s functions G
(n)
(x
1
. . . , x
n
) can be calculated
by summing over all connected Feynman diagrams,
¸Ω[ T φ
H
(x
1
) . . . φ
H
(x
n
) [Ω) =

Connected Feynman Graphs (3.104)
An Example: The Four-Point Correlator: ¸Ω[ Tφ
H
(x
1
) . . . φ
H
(x
4
) [Ω)
As a simple example, let’s look at the four-point correlation function in φ
4
theory. The
sum of connected Feynman diagrams is given by,
x
1
x
2
x
3
x
4
x
1
x
2
x
3
x
4
x
1
x
2
x
3
x
4
+
+ + + 5 Similar
+
...
2 Similar
All of these are connected diagrams, even though they don’t look
x
1
x
2
x
3
x
4
Figure 20:
that connected! The point is that a connected diagram is defined
by the requirement that every line is joined to an external leg. An
example of a diagram that is not connected is shown in the figure.
As we have seen, such diagrams are taken care of in shifting the
vacuum from [0) to [Ω).
Feynman Rules
The Feynman diagrams that we need to calculate for the Green’s functions depend on
x
1
, . . . , x
n
. This is rather different than the Feynman diagrams that we calculated for
– 78 –
the S-matrix elements, where we were working primarily with momentum eigenstates,
and ended up integrating over all of space. However, it’s rather simple to adapt the
Feynman rules that we had earlier in momentum space to compute G
(n)
(x
1
. . . , x
n
).
For φ
4
theory, we have
• Draw n external points x
1
, . . . , x
n
, connected by the usual propagators and ver-
tices. Assign a spacetime position y to the end of each line.
• For each line
x y
fromx to y write down a factor of the Feynman propagator

F
(x −y).
• For each vertex y at position y, write down a factor of −iλ
_
d
4
y.
3.7.2 From Green’s Functions to S-Matrices
Having described how to compute correlation functions using Feynman diagrams, let’s
now relate them back to the S-matrix elements that we already calculated. The first
step is to perform the Fourier transform,
˜
G
(n)
(p
1
, . . . , p
n
) =
_
_
n

i=1
d
4
x
i
e
−ip
i
·x
i
_
G
(n)
(x
1
, . . . , x
n
) (3.105)
These are very closely related to the S-matrix elements that we’ve computed above. The
difference is that the Feynman rules for G
(n)
(x
1
, . . . , x
n
), effectively include propagators

F
for the external legs, as well as the internal legs. A related fact is that the 4-
momenta assigned to the external legs is arbitrary: they are not on-shell. Both of these
problems are easily remedied to allow us to return to the S-matrix elements: we need
to simply cancel off the propagators on the external legs, and place their momentum
back on shell. We have
¸p

1
, . . . , p

n
′ [ S −1 [p
1
. . . , p
n
) = (−i)
n+n

n

i=1
(p
′ 2
i
−m
2
)
n

j=1
(p
2
j
−m
2
) (3.106)

˜
G
(n+n

)
(−p

1
, . . . , −p

n
′ , p
1
, . . . , p
n
)
Each of the factors (p
2
−m
2
) vanishes once the momenta are placed on-shell. This means
that we only get a non-zero answer for diagrams contributing to G
(n)
(x
1
, . . . , x
n
) which
have propagators for each external leg. You might think they all do, but it’s not true!
Only diagrams that are fully connected, meaning each external point is connected to
each other external point, have this property. For example, of the diagrams that we
wrote down which contribute to the four-point function ¸Ω[ Tφ
H
(x
1
) . . . φ
H
(x
4
) [Ω), only
will survive the multiplication by on-shell propagators in (3.106) to contribute to
the S-matrix for meson scattering in φ
4
theory.
– 79 –
So what’s the point of all of this? We’ve understood that ignoring the connected
diagrams is related to shifting to the true vacuum [Ω). But other than that, introducing
the Green’s functions seems like a lot of bother for little reward. The important point
is that this provides a framework in which to deal with the true particle states in
the interacting theory through renormalization. Indeed, the formula (3.106), suitably
interpreted, remains true even in the interacting theory, taking into account the swarm
of virtual particles surrounding asymptotic states. This is the correct way to consider
scattering. In this context, (3.106) is known as the LSZ reduction formula. You will
derive it properly next term.
– 80 –
4. The Dirac Equation
“A great deal more was hidden in the Dirac equation than the author had
expected when he wrote it down in 1928. Dirac himself remarked in one of
his talks that his equation was more intelligent than its author. It should
be added, however, that it was Dirac who found most of the additional
insights.”
Weisskopf on Dirac
So far we’ve only discussed scalar fields such that under a Lorentz transformation
x
µ
→(x

)
µ
= Λ
µ
ν
x
ν
, the field transforms as
φ(x) →φ

(x) = φ(Λ
−1
x) (4.1)
We have seen that quantization of such fields gives rise to spin 0 particles. But most
particles in Nature have an intrinsic angular momentum, or spin. These arise naturally
in field theory by considering fields which themselves transform non-trivially under the
Lorentz group. In this section we will describe the Dirac equation, whose quantization
gives rise to fermionic spin 1/2 particles. To motivate the Dirac equation, we will start
by studying the appropriate representation of the Lorentz group.
A familiar example of a field which transforms non-trivially under the Lorentz group
is the vector field A
µ
(x) of electromagnetism,
A
µ
(x) →Λ
µ
ν
A
ν

−1
x) (4.2)
We’ll deal with this in Section 6. (It comes with its own problems!). In general, a field
can transform as
φ
a
(x) →D[Λ]
a
b
φ
b

−1
x) (4.3)
where the matrices D[Λ] form a representation of the Lorentz group, meaning that
D[Λ
1
]D[Λ
2
] = D[Λ
1
Λ
2
] (4.4)
and D[Λ
−1
] = D[Λ]
−1
and D[1] = 1. How do we find the different representations?
Typically, we look at infinitesimal transformations of the Lorentz group and study the
resulting Lie algebra. If we write,
Λ
µ
ν
= δ
µ
ν
+ ω
µ
ν
(4.5)
for infinitesimal ω, then the condition for a Lorentz transformation Λ
µ
σ
Λ
ν
ρ
η
σρ
= η
µν
becomes the requirement that ω is anti-symmetric:
ω
µν
+ ω
νµ
= 0 (4.6)
– 81 –
Note that an antisymmetric 4 4 matrix has 4 3/2 = 6 independent components,
which agrees with the 6 transformations of the Lorentz group: 3 rotations and 3 boosts.
It’s going to be useful to introduce a basis of these six 4 4 anti-symmetric matrices.
We could call them (/
A
)
µν
, with A = 1, . . . , 6. But in fact it’s better for us (although
initially a little confusing) to replace the single index A with a pair of antisymmetric
indices [ρσ], where ρ, σ = 0, . . . , 3, so we call our matrices (/
ρσ
)
µ
ν
. The antisymmetry
on the ρ and σ indices means that, for example, /
01
= −/
10
, etc, so that ρ and σ
again label six different matrices. Of course, the matrices are also antisymmetric on
the µν indices because they are, after all, antisymmetric matrices. With this notation
in place, we can write a basis of six 4 4 antisymmetric matrices as
(/
ρσ
)
µν
= η
ρµ
η
σν
−η
σµ
η
ρν
(4.7)
where the indices µ and ν are those of the 4 4 matrix, while ρ and σ denote which
basis element we’re dealing with. If we use these matrices for anything practical (for
example, if we want to multiply them together, or act on some field) we will typically
need to lower one index, so we have
(/
ρσ
)
µ
ν
= η
ρµ
δ
σ
ν
−η
σµ
δ
ρ
ν
(4.8)
Since we lowered the index with the Minkowski metric, we pick up various minus signs
which means that when written in this form, the matrices are no longer necessarily
antisymmetric. Two examples of these basis matrices are,
(/
01
)
µ
ν
=
_
0 1 0 0
1 0 0 0
0 0 0 0
0 0 0 0
_
and (/
12
)
µ
ν
=
_
0 0 0 0
0 0 −1 0
0 1 0 0
0 0 0 0
_
(4.9)
The first, /
01
, generates boosts in the x
1
direction. It is real and symmetric. The
second, /
12
, generates rotations in the (x
1
, x
2
)-plane. It is real and antisymmetric.
We can now write any ω
µ
ν
as a linear combination of the /
ρσ
,
ω
µ
ν
=
1
2

ρσ
(/
ρσ
)
µ
ν
(4.10)
where Ω
ρσ
are just six numbers (again antisymmetric in the indices) that tell us what
Lorentz transformation we’re doing. The six basis matrices /
ρσ
are called the gen-
erators of the Lorentz transformations. The generators obey the Lorentz Lie algebra
relations,
[/
ρσ
, /
τν
] = η
στ
/
ρν
−η
ρτ
/
σν
+ η
ρν
/
στ
−η
σν
/
ρτ
(4.11)
– 82 –
where we have suppressed the matrix indices. A finite Lorentz transformation can then
be expressed as the exponential
Λ = exp
_
1
2

ρσ
/
ρσ
_
(4.12)
Let me stress again what each of these objects are: the /
ρσ
are six 4 4 basis
elements of the Lorentz group; the Ω
ρσ
are six numbers telling us what kind of Lorentz
transformation we’re doing (for example, they say things like rotate by θ = π/7 about
the x
3
-direction and run at speed v = 0.2 in the x
1
direction).
4.1 The Spinor Representation
We’re interested in finding other matrices which satisfy the Lorentz algebra commuta-
tion relations (4.11). We will construct the spinor representation. To do this, we start
by defining something which, at first sight, has nothing to do with the Lorentz group.
It is the Clifford algebra,
¦γ
µ
, γ
ν
¦ ≡ γ
µ
γ
ν
+ γ
ν
γ
µ
= 2η
µν
1 (4.13)
where γ
µ
, with µ = 0, 1, 2, 3, are a set of four matrices and the 1 on the right-hand side
denotes the unit matrix. This means that we must find four matrices such that
γ
µ
γ
ν
= −γ
ν
γ
µ
when µ ,= ν (4.14)
and

0
)
2
= 1 , (γ
i
)
2
= −1 i = 1, 2, 3 (4.15)
It’s not hard to convince yourself that there are no representations of the Clifford
algebra using 2 2 or 3 3 matrices. The simplest representation of the Clifford
algebra is in terms of 4 4 matrices. There are many such examples of 4 4 matrices
which obey (4.13). For example, we may take
γ
0
=
_
0 1
1 0
_
, γ
i
=
_
0 σ
i
−σ
i
0
_
(4.16)
where each element is itself a 2 2 matrix, with the σ
i
the Pauli matrices
σ
1
=
_
0 1
1 0
_
, σ
2
=
_
0 −i
i 0
_
, σ
3
=
_
1 0
0 −1
_
(4.17)
which themselves satisfy ¦σ
i
, σ
j
¦ = 2δ
ij
.
– 83 –
One can construct many other representations of the Clifford algebra by taking
V γ
µ
V
−1
for any invertible matrix V . However, up to this equivalence, it turns out
that there is a unique irreducible representation of the Clifford algebra. The matrices
(4.16) provide one example, known as the Weyl or chiral representation (for reasons
that will soon become clear). We will soon restrict ourselves further, and consider only
representations of the Clifford algebra that are related to the chiral representation by
a unitary transformation V .
So what does the Clifford algebra have to do with the Lorentz group? Consider the
commutator of two γ
µ
,
S
ρσ
=
1
4

ρ
, γ
σ
] =
_
0 ρ = σ
1
2
γ
ρ
γ
σ
ρ ,= σ
_
=
1
2
γ
ρ
γ
σ

1
2
η
ρσ
(4.18)
Let’s see what properties these matrices have:
Claim 4.1: [S
µν
, γ
ρ
] = γ
µ
η
νρ
−γ
ν
η
ρµ
Proof: When µ ,= ν we have
[S
µν
, γ
ρ
] =
1
2

µ
γ
ν
, γ
ρ
]
=
1
2
γ
µ
γ
ν
γ
ρ

1
2
γ
ρ
γ
µ
γ
ν
=
1
2
γ
µ
¦γ
ν
, γ
ρ
¦ −
1
2
γ
µ
γ
ρ
γ
ν

1
2
¦γ
ρ
, γ
µ
¦γ
ν
+
1
2
γ
µ
γ
ρ
γ
ν
= γ
µ
η
νρ
−γ
ν
η
ρµ

Claim 4.2: The matrices S
µν
form a representation of the Lorentz algebra (4.11),
meaning
[S
µν
, S
ρσ
] = η
νρ
S
µσ
−η
µρ
S
νσ
+ η
µσ
S
νρ
−η
νσ
S
µρ
(4.19)
Proof: Taking ρ ,= σ, and using Claim 4.1 above, we have
[S
µν
, S
ρσ
] =
1
2
[S
µν
, γ
ρ
γ
σ
]
=
1
2
[S
µν
, γ
ρ

σ
+
1
2
γ
ρ
[S
µν
, γ
σ
]
=
1
2
γ
µ
γ
σ
η
νρ

1
2
γ
ν
γ
σ
η
ρµ
+
1
2
γ
ρ
γ
µ
η
νσ

1
2
γ
ρ
γ
ν
η
σµ
(4.20)
Now using the expression (4.18) to write γ
µ
γ
σ
= 2S
µσ
+ η
µσ
, we have
[S
µν
, S
ρσ
] = S
µσ
η
νρ
−S
νσ
η
ρµ
+ S
ρµ
η
νσ
−S
ρν
η
σµ
(4.21)
which is our desired expression.
– 84 –
4.1.1 Spinors
The S
µν
are 4 4 matrices, because the γ
µ
are 4 4 matrices. So far we haven’t given
an index name to the rows and columns of these matrices: we’re going to call them
α, β = 1, 2, 3, 4.
We need a field for the matrices (S
µν
)
α
β
to act upon. We introduce the Dirac spinor
field ψ
α
(x), an object with four complex components labelled by α = 1, 2, 3, 4. Under
Lorentz transformations, we have
ψ
α
(x) →S[Λ]
α
β
ψ
β

−1
x) (4.22)
where
Λ = exp
_
1
2

ρσ
/
ρσ
_
(4.23)
S[Λ] = exp
_
1
2

ρσ
S
ρσ
_
(4.24)
Although the basis of generators /
ρσ
and S
ρσ
are different, we use the same six
numbers Ω
ρσ
in both Λ and S[Λ]: this ensures that we’re doing the same Lorentz
transformation on x and ψ. Note that we denote both the generator S
ρσ
and the full
Lorentz transformation S[Λ] as “S”. To avoid confusion, the latter will always come
with the square brackets [Λ].
Both Λ and S[Λ] are 4 4 matrices. So how can we be sure that the spinor repre-
sentation is something new, and isn’t equivalent to the familiar representation Λ
µ
ν
? To
see that the two representations are truly different, let’s look at some specific transfor-
mations.
Rotations
S
ij
=
1
2
_
0 σ
i
−σ
i
0
__
0 σ
j
−σ
j
0
_
= −
i
2
ǫ
ijk
_
σ
k
0
0 σ
k
_
(4.25)
If we write the rotation parameters as Ω
ij
= −ǫ
ijk
ϕ
k
(meaning Ω
12
= −ϕ
3
, etc) then
the rotation matrix becomes
S[Λ] = exp
_
1
2

ρσ
S
ρσ
_
=
_
e
+i ϕ·σ/2
0
0 e
+i ϕ·σ/2
_
(4.26)
where we need to remember that Ω
12
= −Ω
21
= −ϕ
3
when following factors of 2.
Consider now a rotation by 2π about, say, the x
3
-axis. This is achieved by ϕ = (0, 0, 2π),
– 85 –
and the spinor rotation matrix becomes,
S[Λ] =
_
e
+iπσ
3
0
0 e
+iπσ
3
_
= −1 (4.27)
Therefore under a 2π rotation
ψ
α
(x) →−ψ
α
(x) (4.28)
which is definitely not what happens to a vector! To check that we haven’t been
cheating with factors of 2, let’s see how a vector would transform under a rotation by
ϕ = (0, 0, ϕ
3
). We have
Λ = exp
_
1
2

ρσ
/
ρσ
_
= exp
_
0 0 0 0
0 0 ϕ
3
0
0 −ϕ
3
0 0
0 0 0 0
_
(4.29)
So when we rotate a vector by ϕ
3
= 2π, we learn that Λ = 1 as you would expect. So
S[Λ] is definitely a different representation from the familiar vector representation Λ
µ
ν
.
Boosts
S
0i
=
1
2
_
0 1
1 0
__
0 σ
i
−σ
i
0
_
=
1
2
_
−σ
i
0
0 σ
i
_
(4.30)
Writing the boost parameter as Ω
i0
= −Ω
0i
= χ
i
, we have
S[Λ] =
_
e
+ χ·σ/2
0
0 e
− χ·σ/2
_
(4.31)
Representations of the Lorentz Group are not Unitary
Note that for rotations given in (4.26), S[Λ] is unitary, satisfying S[Λ]

S[Λ] = 1. But
for boosts given in (4.31), S[Λ] is not unitary. In fact, there are no finite dimensional
unitary representations of the Lorentz group. We have demonstrated this explicitly for
the spinor representation using the chiral representation (4.16) of the Clifford algebra.
We can get a feel for why it is true for a spinor representation constructed from any
representation of the Clifford algebra. Recall that
S[Λ] = exp
_
1
2

ρσ
S
ρσ
_
(4.32)
so the representation is unitary if S
µν
are anti-hermitian, i.e. (S
µν
)

= −S
µν
. But we
have
(S
µν
)

= −
1
4
[(γ
µ
)

, (γ
ν
)

] (4.33)
– 86 –
which can be anti-hermitian if all γ
µ
are hermitian or all are anti-hermitian. However,
we can never arrange for this to happen since

0
)
2
= 1 ⇒ Real Eigenvalues

i
)
2
= −1 ⇒ Imaginary Eigenvalues (4.34)
So we could pick γ
0
to be hermitian, but we can only pick γ
i
to be anti-hermitian.
Indeed, in the chiral representation (4.16), the matrices have this property: (γ
0
)

= γ
0
and (γ
i
)

= −γ
i
. In general there is no way to pick γ
µ
such that S
µν
are anti-hermitian.
4.2 Constructing an Action
We now have a new field to work with, the Dirac spinor ψ. We would like to construct
a Lorentz invariant equation of motion. We do this by constructing a Lorentz invariant
action.
We will start in a naive way which won’t work, but will give us a clue how to proceed.
Define
ψ

(x) = (ψ

)
T
(x) (4.35)
which is the usual adjoint of a multi-component object. We could then try to form a
Lorentz scalar by taking the product ψ

ψ, with the spinor indices summed over. Let’s
see how this transforms under Lorentz transformations,
ψ(x) → S[Λ] ψ(Λ
1
x)
ψ

(x) → ψ


−1
x) S[Λ]

(4.36)
So ψ

(x)ψ(x) → ψ


−1
x)S[Λ]

S[Λ]ψ(Λ
−1
x). But, as we have seen, for some Lorentz
transformation S[Λ]

S[Λ] ,= 1 since the representation is not unitary. This means that
ψ

ψ isn’t going to do it for us: it doesn’t have any nice transformation under the
Lorentz group, and certainly isn’t a scalar. But now we see why it fails, we can also see
how to proceed. Let’s pick a representation of the Clifford algebra which, like the chiral
representation (4.16), satisfies (γ
0
)

= γ
0
and (γ
i
)

= −γ
i
. Then for all µ = 0, 1, 2, 3
we have
γ
0
γ
µ
γ
0
= (γ
µ
)

(4.37)
which, in turn, means that
(S
µν
)

=
1
4
[(γ
ν
)

, (γ
µ
)

] = −γ
0
S
µν
γ
0
(4.38)
– 87 –
so that
S[Λ]

= exp
_
1
2

ρσ
(S
ρσ
)

_
= γ
0
S[Λ]
−1
γ
0
(4.39)
With this in mind, we now define the Dirac adjoint
¯
ψ(x) = ψ

(x) γ
0
(4.40)
Let’s now see what Lorentz covariant objects we can form out of a Dirac spinor ψ and
its adjoint
¯
ψ.
Claim 4.3:
¯
ψψ is a Lorentz scalar.
Proof: Under a Lorentz transformation,
¯
ψ(x) ψ(x) = ψ

(x) γ
0
ψ(x)
→ ψ


−1
x) S[Λ]

γ
0
S[Λ]ψ(Λ
−1
x)
= ψ


−1
x) γ
0
ψ(Λ
−1
x)
=
¯
ψ(Λ
−1
x) ψ(Λ
−1
x) (4.41)
which is indeed the transformation law for a Lorentz scalar.
Claim 4.4:
¯
ψ γ
µ
ψ is a Lorentz vector, which means that
¯
ψ(x) γ
µ
ψ(x) →Λ
µ
ν
¯
ψ(Λ
−1
x) γ
ν
ψ(Λ
−1
x) (4.42)
This equation means that we can treat the µ = 0, 1, 2, 3 index on the γ
µ
matrices as
a true vector index. In particular we can form Lorentz scalars by contracting it with
other Lorentz indices.
Proof: Suppressing the x argument, under a Lorentz transformation we have,
¯
ψ γ
µ
ψ →
¯
ψ S[Λ]
−1
γ
µ
S[Λ]ψ (4.43)
If
¯
ψ γ
µ
ψ is to transform as a vector, we must have
S[Λ]
−1
γ
µ
S[Λ] = Λ
µ
ν
γ
ν
(4.44)
We’ll now show this. We work infinitesimally, so that
Λ = exp
_
1
2

ρσ
/
ρσ
_
≈ 1 +
1
2

ρσ
/
ρσ
+ . . . (4.45)
S[Λ] = exp
_
1
2

ρσ
S
ρσ
_
≈ 1 +
1
2

ρσ
S
ρσ
+ . . . (4.46)
– 88 –
so the requirement (4.44) becomes
−[S
ρσ
, γ
µ
] = (/
ρσ
)
µ
ν
γ
ν
(4.47)
where we’ve suppressed the α, β indices on γ
µ
and S
µν
, but otherwise left all other
indices explicit. In fact equation (4.47) follows from Claim 4.1 where we showed that
[S
ρσ
, γ
µ
] = γ
ρ
η
σµ
− γ
σ
η
µρ
. To see this, we write the right-hand side of (4.47) by
expanding out /,
(/
ρσ
)
µ
ν
γ
ν
= (η
ρµ
δ
σ
ν
−η
σµ
δ
ρ
ν

ν
= η
ρµ
γ
σ
−η
σµ
γ
ρ
(4.48)
which means that the proof follows if we can show
−[S
ρσ
, γ
µ
] = η
ρµ
γ
σ
−η
σµ
γ
ρ
(4.49)
which is exactly what we proved in Claim 4.1.
Claim 4.5:
¯
ψγ
µ
γ
ν
ψ transforms as a Lorentz tensor. More precisely, the symmet-
ric part is a Lorentz scalar, proportional to η
µν
¯
ψψ, while the antisymmetric part is a
Lorentz tensor, proportional to
¯
ψS
µν
ψ.
Proof: As above.
We are now armed with three bilinears of the Dirac field,
¯
ψψ,
¯
ψγ
µ
ψ and
¯
ψγ
µ
γ
ν
ψ,
each of which transforms covariantly under the Lorentz group. We can try to build a
Lorentz invariant action from these. In fact, we need only the first two. We choose
S =
_
d
4
x
¯
ψ(x) (iγ
µ

µ
−m) ψ(x) (4.50)
This is the Dirac action. The factor of “i” is there to make the action real; upon complex
conjugation, it cancels a minus sign that comes from integration by parts. (Said another
way, it’s there for the same reason that the Hermitian momentum operator −i∇ in
quantum mechanics has a factor i). As we will see in the next section, after quantization
this theory describes particles and anti-particles of mass [m[ and spin 1/2. Notice that
the Lagrangian is first order, rather than the second order Lagrangians we were working
with for scalar fields. Also, the mass appears in the Lagrangian as m, which can be
positive or negative.
– 89 –
4.3 The Dirac Equation
The equation of motion follows from the action (4.50) by varying with respect to ψ and
¯
ψ independently. Varying with respect to
¯
ψ, we have
(iγ
µ

µ
−m) ψ = 0 (4.51)
This is the Dirac equation. It’s completely gorgeous. Varying with respect to ψ gives
the conjugate equation
i∂
µ
¯
ψ γ
µ
+ m
¯
ψ = 0 (4.52)
The Dirac equation is first order in derivatives, yet miraculously Lorentz invariant. If
we tried to write down a first order equation of motion for a scalar field, it would look
like v
µ

µ
φ = . . ., which necessarily includes a privileged vector in spacetime v
µ
and is
not Lorentz invariant. However, for spinor fields, the magic of the γ
µ
matrices means
that the Dirac Lagrangian is Lorentz invariant.
The Dirac equation mixes up different components of ψ through the matrices γ
µ
.
However, each individual component itself solves the Klein-Gordon equation. To see
this, write
(iγ
ν

ν
+ m)(iγ
µ

µ
−m) ψ = −
_
γ
µ
γ
ν

µ

ν
+ m
2
_
ψ = 0 (4.53)
But γ
µ
γ
ν

µ

ν
=
1
2
¦γ
µ
, γ
ν
¦∂
µ

ν
= ∂
µ

µ
, so we get
−(∂
µ

µ
+ m
2
)ψ = 0 (4.54)
where this last equation has no γ
µ
matrices, and so applies to each component ψ
α
, with
α = 1, 2, 3, 4.
The Slash
Let’s introduce some useful notation. We will often come across 4-vectors contracted
with γ
µ
matrices. We write
A
µ
γ
µ
≡ / A (4.55)
so the Dirac equation reads
(i / ∂ −m)ψ = 0 (4.56)
– 90 –
4.4 Chiral Spinors
When we’ve needed an explicit form of the γ
µ
matrices, we’ve used the chiral represen-
tation
γ
0
=
_
0 1
1 0
_
, γ
i
=
_
0 σ
i
−σ
i
0
_
(4.57)
In this representation, the spinor rotation transformation S[Λ
rot
] and boost transfor-
mation S[Λ
boost
] were computed in (4.26) and (4.31). Both are block diagonal,
S[Λ
rot
] =
_
e
+i ϕ·σ/2
0
0 e
+i ϕ·σ/2
_
and S[Λ
boost
] =
_
e
+ χ·σ/2
0
0 e
− χ·σ/2
_
(4.58)
This means that the Dirac spinor representation of the Lorentz group is reducible. It
decomposes into two irreducible representations, acting only on two-component spinors
u
±
which, in the chiral representation, are defined by
ψ =
_
u
+
u

_
(4.59)
The two-component objects u
±
are called Weyl spinors or chiral spinors. They trans-
form in the same way under rotations,
u
±
→e
i ϕ·σ/2
u
±
(4.60)
but oppositely under boosts,
u
±
→e
± χ·σ/2
u
±
(4.61)
In group theory language, u
+
is in the (
1
2
, 0) representation of the Lorentz group,
while u

is in the (0,
1
2
) representation. The Dirac spinor ψ lies in the (
1
2
, 0) ⊕ (0,
1
2
)
representation. (Strictly speaking, the spinor is a representation of the double cover of
the Lorentz group SL(2, C)).
4.4.1 The Weyl Equation
Let’s see what becomes of the Dirac Lagrangian under the decomposition (4.59) into
Weyl spinors. We have
L =
¯
ψ(i / ∂ −m)ψ = iu


σ
µ

µ
u

+ iu

+
¯ σ
µ

µ
u
+
−m(u

+
u

+ u


u
+
) = 0 (4.62)
– 91 –
where we have introduced some new notation for the Pauli matrices with a µ = 0, 1, 2, 3
index,
σ
µ
= (1, σ
i
) and ¯ σ
µ
= (1, −σ
i
) (4.63)
From (4.62), we see that a massive fermion requires both u
+
and u

, since they couple
through the mass term. However, a massless fermion can be described by u
+
(or u

)
alone, with the equation of motion
i¯ σ
µ

µ
u
+
= 0
or iσ
µ

µ
u

= 0 (4.64)
These are the Weyl equations.
Degrees of Freedom
Let me comment here on the degrees of freedom in a spinor. The Dirac fermion has
4 complex components = 8 real components. How do we count degrees of freedom?
In classical mechanics, the number of degrees of freedom of a system is equal to the
dimension of the configuration space or, equivalently, half the dimension of the phase
space. In field theory we have an infinite number of degrees of freedom, but it makes
sense to count the number of degrees of freedom per spatial point: this should at least
be finite. For example, in this sense a real scalar field φ has a single degree of freedom.
At the quantum level, this translates to the fact that it gives rise to a single type of
particle. A classical complex scalar field has two degrees of freedom, corresponding to
the particle and the anti-particle in the quantum theory.
But what about a Dirac spinor? One might think that there are 8 degrees of freedom.
But this isn’t right. Crucially, and in contrast to the scalar field, the equation of motion
is first order rather than second order. In particular, for the Dirac Lagrangian, the
momentum conjugate to the spinor ψ is given by
π
ψ
= ∂L/∂
˙
ψ = iψ

(4.65)
It is not proportional to the time derivative of ψ. This means that the phase space for
a spinor is therefore parameterized by ψ and ψ

, while for a scalar it is parameterized
by φ and π =
˙
φ. So the phase space of the Dirac spinor ψ has 8 real dimensions and
correspondingly the number of real degrees of freedom is 4. We will see in the next
section that, in the quantum theory, this counting manifests itself as two degrees of
freedom (spin up and down) for the particle, and a further two for the anti-particle.
A similar counting for the Weyl fermion tells us that it has two degrees of freedom.
– 92 –
4.4.2 γ
5
The Lorentz group matrices S[Λ] came out to be block diagonal in (4.58) because we
chose the specific representation (4.57). In fact, this is why the representation (4.57)
is called the chiral representation: it’s because the decomposition of the Dirac spinor
ψ is simply given by (4.59). But what happens if we choose a different representation
γ
µ
of the Clifford algebra, so that
γ
µ
→Uγ
µ
U
−1
and ψ →Uψ ? (4.66)
Now S[Λ] will not be block diagonal. Is there an invariant way to define chiral spinors?
We can do this by introducing the “fifth” gamma-matrix
γ
5
= −iγ
0
γ
1
γ
2
γ
3
(4.67)
You can check that this matrix satisfies
¦γ
5
, γ
µ
¦ = 0 and (γ
5
)
2
= +1 (4.68)
The reason that this is called γ
5
is because the set of matrices ˜ γ
A
= (γ
µ
, iγ
5
), with
A = 0, 1, 2, 3, 4 satisfy the five-dimensional Clifford algebra ¦˜ γ
A
, ˜ γ
B
¦ = 2η
AB
. (You
might think that γ
4
would be a better name! But γ
5
is the one everyone chooses - it’s
a more sensible name in Euclidean space, where A = 1, 2, 3, 4, 5). You can also check
that [S
µν
, γ
5
] = 0, which means that γ
5
is a scalar under rotations and boosts. Since

5
)
2
= 1, this means we may form the Lorentz invariant projection operators
P
±
=
1
2
_
1 ±γ
5
_
(4.69)
such that P
2
+
= P
+
and P
2

= P

and P
+
P

= 0. One can check that for the chiral
representation (4.57),
γ
5
=
_
1 0
0 −1
_
(4.70)
from which we see that the operators P
±
project onto the Weyl spinors u
±
. However,
for an arbitrary representation of the Clifford algebra, we may use γ
5
to define the
chiral spinors,
ψ
±
= P
±
ψ (4.71)
which form the irreducible representations of the Lorentz group. ψ
+
is often called a
“right-handed” spinor, while ψ

is “left-handed”.
– 93 –
4.4.3 Parity
The spinors ψ
±
are related to each other by parity. Let’s pause to define this concept.
The Lorentz group is defined by x
µ
→Λ
µ
ν
x
ν
such that
Λ
µ
ν
Λ
ρ
σ
η
νσ
= η
µρ
(4.72)
So far we have only considered transformations Λ which are continuously connected to
the identity; these are the ones which have an infinitesimal form. However there are
also two discrete symmetries which are part of the Lorentz group. They are
Time Reversal T : x
0
→−x
0
; x
i
→x
i
Parity P : x
0
→x
0
; x
i
→−x
i
(4.73)
We won’t discuss time reversal too much in this course. (It turns out to be represented
by an anti-unitary transformation on states. See, for example the book by Peskin and
Schroeder). But parity has an important role to play in the standard model and, in
particular, the theory of the weak interaction.
Under parity, the left and right-handed spinors are exchanged. This follows from the
transformation of the spinors under the Lorentz group. In the chiral representation, we
saw that the rotation (4.60) and boost (4.61) transformations for the Weyl spinors u
±
are
u
±
rot
−→e
i ϕ·σ/2
u
±
and u
±
boost
−→ e
± χ·σ/2
u
±
(4.74)
Under parity, rotations don’t change sign. But boosts do flip sign. This confirms that
parity exchanges right-handed and left-handed spinors, P : u
±
→u

, or in the notation
ψ
±
=
1
2
(1 ±γ
5
)ψ, we have
P : ψ
±
(x, t) →ψ

(−x, t) (4.75)
Using this knowledge of how chiral spinors transform, and the fact that P
2
= 1, we see
that the action of parity on the Dirac spinor itself can be written as
P : ψ(x, t) →γ
0
ψ(−x, t) (4.76)
Notice that if ψ(x, t) satisfies the Dirac equation, then the parity transformed spinor
γ
0
ψ(−x, t) also satisfies the Dirac equation, meaning
(iγ
0

t
+ iγ
i

i
−m)γ
0
ψ(−x, t) = γ
0
(iγ
0

t
−iγ
i

i
−m)ψ(−x, t) = 0 (4.77)
where the extra minus sign from passing γ
0
through γ
i
is compensated by the derivative
acting on −x instead of +x.
– 94 –
4.4.4 Chiral Interactions
Let’s now look at how our interaction terms change under parity. We can look at each
of our spinor bilinears from which we built the action,
P :
¯
ψψ(x, t) →
¯
ψψ(−x, t) (4.78)
which is the transformation of a scalar. For the vector
¯
ψγ
µ
ψ, we can look at the
temporal and spatial components separately,
P :
¯
ψγ
0
ψ(x, t) →
¯
ψγ
0
ψ(−x, t)
P :
¯
ψγ
i
ψ(x, t) →
¯
ψγ
0
γ
i
γ
0
ψ(−x, t) = −
¯
ψγ
i
ψ(−x, t) (4.79)
which tells us that
¯
ψγ
µ
ψ transforms as a vector, with the spatial part changing sign.
You can also check that
¯
ψS
µν
ψ transforms as a suitable tensor.
However, now we’ve discovered the existence of γ
5
, we can form another Lorentz
scalar and another Lorentz vector,
¯
ψγ
5
ψ and
¯
ψγ
5
γ
µ
ψ (4.80)
How do these transform under parity? We can check:
P :
¯
ψγ
5
ψ(x, t) →
¯
ψγ
0
γ
5
γ
0
ψ(−x, t) = −
¯
ψγ
5
ψ(−x, t) (4.81)
P :
¯
ψγ
5
γ
µ
ψ(x, t) →
¯
ψγ
0
γ
5
γ
µ
γ
0
ψ(−x, t) =
_

¯
ψγ
5
γ
0
ψ(−x, t) µ = 0
+
¯
ψγ
5
γ
i
ψ(−x, t) µ = i
which means that
¯
ψγ
5
ψ transforms as a pseudoscalar, while
¯
ψγ
5
γ
µ
ψ transforms as an
axial vector. To summarize, we have the following spinor bilinears,
¯
ψψ : scalar
¯
ψγ
µ
ψ : vector
¯
ψS
µν
ψ : tensor
¯
ψγ
5
ψ : pseudoscalar
¯
ψγ
5
γ
µ
ψ : axial vector (4.82)
The total number of bilinears is 1 + 4 + (4 3/2) + 4 + 1 = 16 which is all we could
hope for from a 4-component object.
– 95 –
We’re now armed with new terms involving γ
5
that we can start to add to our
Lagrangian to construct new theories. Typically such terms will break parity invariance
of the theory, although this is not always true. (For example, the term φ
¯
ψγ
5
ψ doesn’t
break parity if φ is itself a pseudoscalar). Nature makes use of these parity violating
interactions by using γ
5
in the weak force. A theory which treats ψ
±
on an equal
footing is called a vector-like theory. A theory in which ψ
+
and ψ

appear differently
is called a chiral theory.
4.5 Majorana Fermions
Our spinor ψ
α
is a complex object. It has to be because the representation S[Λ]
is typically also complex. This means that if we were to try to make ψ real, for
example by imposing ψ = ψ

, then it wouldn’t stay that way once we make a Lorentz
transformation. However, there is a way to impose a reality condition on the Dirac
spinor ψ. To motivate this possibility, it’s simplest to look at a novel basis for the
Clifford algebra, known as the Majorana basis.
γ
0
=
_
0 σ
2
σ
2
0
_
, γ
1
=
_

3
0
0 iσ
3
_
, γ
2
=
_
0 −σ
2
σ
2
0
_
, γ
3
=
_
−iσ
1
0
0 −iσ
1
_
These matrices satisfy the Clifford algebra. What is special about them is that they
are all pure imaginary (γ
µ
)

= −γ
µ
. This means that the generators of the Lorentz
group S
µν
=
1
4

µ
, γ
ν
], and hence the matrices S[Λ] are real. So with this basis of the
Clifford algebra, we can work with a real spinor simply by imposing the condition,
ψ = ψ

(4.83)
which is preserved under Lorentz transformation. Such spinors are called Majorana
spinors.
So what’s the story if we use a general basis for the Clifford algebra? We’ll ask only
that the basis satisfies (γ
0
)

= γ
0
and (γ
i
)

= −γ
i
. We then define the charge conjugate
of a Dirac spinor ψ as
ψ
(c)
= Cψ

(4.84)
Here C is a 4 4 matrix satisfying
C

C = 1 and C

γ
µ
C = −(γ
µ
)

(4.85)
Let’s firstly check that (4.84) is a good definition, meaning that ψ
(c)
transforms nicely
under a Lorentz transformation. We have
ψ
(c)
→CS[Λ]

ψ

= S[Λ]Cψ

= S[Λ]ψ
(c)
(4.86)
– 96 –
where we’ve made use of the properties (4.85) in taking the matrix C through S[Λ]

.
In fact, not only does ψ
(c)
transform nicely under the Lorentz group, but if ψ satisfies
the Dirac equation, then ψ
(c)
does too. This follows from,
(i / ∂ −m)ψ = 0 ⇒ (−i / ∂

−m)ψ

= 0
⇒ C(−i / ∂

−m)ψ

= (+i / ∂ −m)ψ
(c)
= 0
Finally, we can now impose the Lorentz invariant reality condition on the Dirac spinor,
to yield a Majorana spinor,
ψ
(c)
= ψ (4.87)
After quantization, the Majorana spinor gives rise to a fermion that is its own anti-
particle. This is exactly the same as in the case of scalar fields, where we’ve seen that
a real scalar field gives rise to a spin 0 boson that is its own anti-particle. (Be aware:
In many texts an extra factor of γ
0
is absorbed into the definition of C).
So what is this matrix C? Well, for a given representation of the Clifford algebra, it
is something that we can find fairly easily. In the Majorana basis, where the gamma
matrices are pure imaginary, we have simply C
Maj
= 1 and the Majorana condition
ψ = ψ
(c)
becomes ψ = ψ

. In the chiral basis (4.16), only γ
2
is imaginary, and we
may take C
chiral
= iγ
2
=
_
0 iσ
2
−iσ
2
0
_
. (The matrix iσ
2
that appears here is simply the
anti-symmetric matrix ǫ
αβ
). It is interesting to see how the Majorana condition (4.87)
looks in terms of the decomposition into left and right handed Weyl spinors (4.59).
Plugging in the various definitions, we find that u
+
= iσ
2
u


and u

= −iσ
2
u

+
. In
other words, a Majorana spinor can be written in terms of Weyl spinors as
ψ =
_
u
+
−iσ
2
u

+
_
(4.88)
Notice that it’s not possible to impose the Majorana condition ψ = ψ
(c)
at the same
time as the Weyl condition (u

= 0 or u
+
= 0). Instead the Majorana condition relates
u

and u
+
.
An Aside: Spinors in Different Dimensions: The ability to impose Majorana
or Weyl conditions on Dirac spinors depends on both the dimension and the signature of
spacetime. One can always impose the Weyl condition on a spinor in even dimensional
Minkowski space, basically because you can always build a suitable “γ
5
” projection
matrix by multiplying together all the other γ-matrices. The pattern for when the
Majorana condition can be imposed is a little more sporadic. Interestingly, although
the Majorana condition and Weyl condition cannot be imposed simultaneously in four
dimensions, you can do this in Minowski spacetimes of dimension 2, 10, 18, . . ..
– 97 –
4.6 Symmetries and Conserved Currents
The Dirac Lagrangian enjoys a number of symmetries. Here we list them and compute
the associated conserved currents.
Spacetime Translations
Under spacetime translations the spinor transforms as
δψ = ǫ
µ

µ
ψ (4.89)
The Lagrangian depends on ∂
µ
ψ, but not ∂
µ
¯
ψ, so the standard formula (1.41) gives us
the energy-momentum tensor
T
µν
= i
¯
ψγ
µ

ν
ψ −η
µν
L (4.90)
Since a current is conserved only when the equations of motion are obeyed, we don’t lose
anything by imposing the equations of motion already on T
µν
. In the case of a scalar
field this didn’t really buy us anything because the equations of motion are second
order in derivatives, while the energy-momentum is typically first order. However, for
a spinor field the equations of motion are first order: (i / ∂ − m)ψ = 0. This means we
can set L = 0 in T
µν
, leaving
T
µν
= i
¯
ψγ
µ

ν
ψ (4.91)
In particular, we have the total energy
E =
_
d
3
xT
00
=
_
d
3
xi
¯
ψγ
0
˙
ψ =
_
d
3


γ
0
(−iγ
i

i
+ m)ψ (4.92)
where, in the last equality, we have again used the equations of motion.
Lorentz Transformations
Under an infinitesimal Lorentz transformation, the Dirac spinor transforms as (4.22)
which, in infinitesimal form, reads
δψ
α
= −ω
µ
ν
x
ν

µ
ψ
α
+
1
2

ρσ
(S
ρσ
)
α
β
ψ
β
(4.93)
where, following (4.10), we have ω
µ
ν
=
1
2

ρσ
(/
ρσ
)
µ
ν
, and /
ρσ
are the generators of
the Lorentz algebra given by (4.8)
(/
ρσ
)
µ
ν
= η
ρµ
δ
σ
ν
−η
σµ
δ
ρ
ν
(4.94)
– 98 –
which, after direct substitution, tells us that ω
µν
= Ω
µν
. So we get
δψ
α
= −ω
µν
_
x
ν

µ
ψ
α

1
2
(S
µν
)
α
β
ψ
β
¸
(4.95)
The conserved current arising from Lorentz transformations now follows from the same
calculation we saw for the scalar field (1.54) with two differences: firstly, as we saw
above, the spinor equations of motion set L = 0; secondly, we pick up an extra piece
in the current from the second term in (4.95). We have

µ
)
ρσ
= x
σ
T
µρ
−x
ρ
T
µσ
−i
¯
ψγ
µ
S
ρσ
ψ (4.96)
After quantization, when (¸
µ
)
ρσ
is turned into an operator, this extra term will be
responsible for providing the single particle states with internal angular momentum,
telling us that the quantization of a Dirac spinor gives rise to a particle carrying spin
1/2.
Internal Vector Symmetry
The Dirac Lagrangian is invariant under rotating the phase of the spinor, ψ →e
−iα
ψ.
This gives rise to the current
j
µ
V
=
¯
ψγ
µ
ψ (4.97)
where “V ” stands for vector, reflecting the fact that the left and right-handed compo-
nents ψ
±
transform in the same way under this symmetry. We can easily check that
j
µ
V
is conserved under the equations of motion,

µ
j
µ
V
= (∂
µ
¯
ψ)γ
µ
ψ +
¯
ψγ
µ
(∂
µ
ψ) = im
¯
ψψ −im
¯
ψψ = 0 (4.98)
where, in the last equality, we have used the equations of motion i / ∂ψ = mψ and
i∂
µ
¯
ψγ
µ
= −m
¯
ψ. The conserved quantity arising from this symmetry is
Q =
_
d
3
x
¯
ψγ
0
ψ =
_
d
3
x ψ

ψ (4.99)
We will see shortly that this has the interpretation of electric charge, or particle number,
for fermions.
Axial Symmetry
When m = 0, the Dirac Lagrangian admits an extra internal symmetry which rotates
left and right-handed fermions in opposite directions,
ψ →e
iαγ
5
ψ and
¯
ψ →
¯
ψe
iαγ
5
(4.100)
– 99 –
Here the second transformation follows from the first after noting that e
−iαγ
5
γ
0
=
γ
0
e
+iαγ
5
. This gives the conserved current,
j
µ
A
=
¯
ψγ
µ
γ
5
ψ (4.101)
where A is for “axial” since j
µ
A
is an axial vector. This is conserved only when m = 0.
Indeed, with the full Dirac Lagrangian we may compute

µ
j
µ
A
= (∂
µ
¯
ψ)γ
µ
γ
5
ψ +
¯
ψγ
µ
γ
5

µ
ψ = 2im
¯
ψγ
5
ψ (4.102)
which vanishes only for m = 0. However, in the quantum theory things become more
interesting for the axial current. When the theory is coupled to gauge fields (in a
manner we will discuss in Section 6), the axial transformation remains a symmetry
of the classical Lagrangian. But it doesn’t survive the quantization process. It is the
archetypal example of an anomaly: a symmetry of the classical theory that is not
preserved in the quantum theory.
4.7 Plane Wave Solutions
Let’s now study the solutions to the Dirac equation
(iγ
µ

µ
−m)ψ = 0 (4.103)
We start by making a simple ansatz:
ψ = u( p) e
−ip·x
(4.104)
where u( p) is a four-component spinor, independent of spacetime x which, as the no-
tation suggests, can depend on the 3-momentum p. The Dirac equation then becomes

µ
p
µ
−m)u( p) =
_
−m p
µ
σ
µ
p
µ
¯ σ
µ
−m
_
u( p) = 0 (4.105)
where we’re again using the definition,
σ
µ
= (1, σ
i
) and ¯ σ
µ
= (1, −σ
i
) (4.106)
Claim: The solution to (4.105) is
u( p) =
_

p σ ξ

p ¯ σ ξ
_
(4.107)
– 100 –
for any 2-component spinor ξ which we will normalize to ξ

ξ = 1.
Proof: Let’s write u( p)
T
= (u
1
, u
2
). Then equation (4.105) reads
(p σ) u
2
= mu
1
and (p ¯ σ)u
1
= mu
2
(4.108)
Either one of these equations implies the other, a fact which follows from the identity
(p σ)(p ¯ σ) = p
2
0
− p
i
p
j
σ
i
σ
j
= p
2
0
− p
i
p
j
δ
ij
= p
µ
p
µ
= m
2
. To start with, let’s try
the ansatz u
1
= (p σ)ξ

for some spinor ξ

. Then the second equation in (4.108)
immediately tells us that u
2
= mξ

. So we learn that any spinor of the form
u( p) = A
_
(p σ) ξ



_
(4.109)
with constant A is a solution to (4.105). To make this more symmetric, we choose
A = 1/m and ξ

=

p ¯ σ ξ with constant ξ. Then u
1
= (p σ)

p ¯ σ ξ = m

p σ ξ. So
we get the promised result (4.107)
Negative Frequency Solutions
We get further solutions to the Dirac equation from the ansatz
ψ = v( p) e
+ip·x
(4.110)
Solutions of the form (4.104), which oscillate in time as ψ ∼ e
−iEt
, are called positive
frequency solutions. Those of the form (4.110), which oscillate as ψ ∼ e
+iEt
, are
negative frequency solutions. It’s important to note however that both are solutions to
the classical field equations and both have positive energy (4.92). The Dirac equation
requires that the 4-component spinor v( p) satisfies

µ
p
µ
+ m)v( p) =
_
m p
µ
σ
µ
p
µ
¯ σ
µ
m
_
v( p) = 0 (4.111)
which is solved by
v( p) =
_

p σ η


p ¯ σ η
_
(4.112)
for some 2-component spinor η which we take to be constant and normalized to η

η = 1.
– 101 –
4.7.1 Some Examples
Consider the positive frequency solution with mass m and 3-momentum p = 0,
u( p) =

m
_
ξ
ξ
_
(4.113)
where ξ is any 2-component spinor. Spatial rotations of the field act on ξ by (4.26),
ξ →e
+i ϕ·σ/2
ξ (4.114)
The 2-component spinor ξ defines the spin of the field. This should be familiar from
quantum mechanics. A field with spin up (down) along a given direction is described
by the eigenvector of the corresponding Pauli matrix with eigenvalue +1 (-1 respec-
tively). For example, ξ
T
= (1, 0) describes a field with spin up along the z-axis. After
quantization, this will become the spin of the associated particle. In the rest of this
section, we’ll indulge in an abuse of terminology and refer to the classical solutions to
the Dirac equations as “particles”, even though they have no such interpretation before
quantization.
Consider now boosting the particle with spin ξ
T
= (1, 0) along the x
3
direction, with
p
µ
= (E, 0, 0, p). The solution to the Dirac equation becomes
u( p) =
_
_

p σ
_
1
0
_

p ¯ σ
_
1
0
_
_
_
=
_
_
_
E −p
3
_
1
0
_
_
E + p
3
_
1
0
_
_
_
(4.115)
In fact, this expression also makes sense for a massless field, for which E = p
3
. (We
picked the normalization (4.107) for the solutions so that this would be the case). For
a massless particle we have
u( p) =

2E
_
0
0
1
0
_
(4.116)
Similarly, for a boosted solution of the spin down ξ
T
= (0, 1) field, we have
u( p) =
_
_

p σ
_
0
1
_

p ¯ σ
_
0
1
_
_
_
=
_
_
_
E + p
3
_
0
1
_
_
E −p
3
_
0
1
_
_
_
m→0
−→

2E
_
0
1
0
0
_
(4.117)
– 102 –
Helicity
The helicity operator is the projection of the angular momentum along the direction
of momentum,
h =
i
2
ǫ
ijk
ˆ p
i
S
jk
=
1
2
ˆ p
i
_
σ
i
0
0 σ
i
_
(4.118)
where S
ij
is the rotation generator given in (4.25). The massless field with spin ξ
T
=
(1, 0) in (4.116) has helicity h = 1/2: we say that it is right-handed. Meanwhile, the
field (4.117) has helicity h = −1/2: it is left-handed.
4.7.2 Some Useful Formulae: Inner and Outer Products
There are a number of identities that will be very useful in the following section,
regarding the inner (and outer) products of the spinors u( p) and v( p). It’s firstly
convenient to introduce a basis ξ
s
and η
s
, s = 1, 2 for the two-component spinors such
that
ξ
r †
ξ
s
= δ
rs
and η
r †
η
s
= δ
rs
(4.119)
for example,
ξ
1
=
_
1
0
_
and ξ
2
=
_
0
1
_
(4.120)
and similarly for η
s
. Let’s deal first with the positive frequency plane waves. The two
independent solutions are now written as
u
s
( p) =
_

p σ ξ
s

p ¯ σ ξ
s
_
(4.121)
We can take the inner product of four-component spinors in two different ways: either
as u

u, or as ¯ u u. Of course, only the latter will be Lorentz invariant, but it turns out
that the former is needed when we come to quantize the theory. Here we state both:
u
r †
( p) u
s
( p) =
_
ξ
r †

p σ , ξ
r †

p ¯ σ
_
_

p σ ξ
s

p ¯ σ ξ
s
_
= ξ
r †
p σξ
s
+ ξ
r †
p ¯ σξ
s
= 2ξ
r †
p
0
ξ
s
= 2p
o
δ
rs
(4.122)
while the Lorentz invariant inner product is
¯ u
r
( p) u
s
( p) =
_
ξ
r †

p σ , ξ
r †

p ¯ σ
_
_
0 1
1 0
__

p σ ξ
s

p ¯ σ ξ
s
_
= 2mδ
rs
(4.123)
– 103 –
We have analogous results for the negative frequency solutions, which we may write as
v
s
( p) =
_

p σ η
s


p ¯ σ η
s
_
with v
r †
( p) v
s
( p) = 2p
0
δ
rs
and ¯ v
r
( p) v
s
( p) = −2mδ
rs
(4.124)
We can also compute the inner product between u and v. We have
¯ u
r
( p) v
s
( p) =
_
ξ
r †

p σ , ξ
r †

p ¯ σ
_
γ
0
_

p σ η
s


p ¯ σ η
s
_
= ξ
r†
_
(p ¯ σ)(p σ)η
s
−ξ
r †
_
(p ¯ σ)(p σ)η
s
= 0 (4.125)
and similarly, ¯ v
r
( p) u
s
( p) = 0. However, when we come to u

v, it is a slightly
different combination that has nice properties (and this same combination appears
when we quantize the theory). We look at u
r †
( p) v
s
(− p), with the 3-momentum in
the spinor v taking the opposite sign. Defining the 4-momentum (p

)
µ
= (p
0
, − p), we
have
u
r †
( p) v
s
(− p) =
_
ξ
r †

p σ , ξ
r †

p ¯ σ
_
_

p

σ η
s


p

¯ σ η
s
_
= ξ
r†
_
(p ¯ σ)(p

σ)η
s
−ξ
r †
_
(p ¯ σ)(p

σ)η
s
(4.126)
Now the terms under the square-root are given by (p ¯ σ)(p

σ) = (p
0
+p
i
σ
i
)(p
0
−p
i
σ
i
) =
p
2
o
− p
2
= m
2
. The same expression holds for (p

¯ σ)(p σ), and the two terms cancel.
We learn
u
r †
( p) v
s
(− p) = v
r †
( p) u
s
(− p) = 0 (4.127)
Outer Products
There’s one last spinor identity that we need before we turn to the quantum theory. It is:
Claim:
2

s=1
u
s
( p) ¯ u
s
( p) = / p + m (4.128)
where the two spinors are not now contracted, but instead placed back to back to give
a 4 4 matrix. Also,
2

s=1
v
s
( p) ¯ v
s
( p) = / p −m (4.129)
– 104 –
Proof:
2

s=1
u
s
( p) ¯ u
s
( p) =
2

s=1
_

p σ ξ
s

p ¯ σ ξ
s
_
_
ξ
s †

p ¯ σ , ξ
s †

p σ
_
(4.130)
But

s
ξ
s
ξ
s †
= 1, the 2 2 unit matrix, which then gives us
2

s=1
u
s
( p)¯ u
s
( p) =
_
m p σ
p ¯ σ m
_
(4.131)
which is the desired result. A similar proof works for

s
v
s
( p)¯ v
s
( p).
– 105 –
5. Quantizing the Dirac Field
We would now like to quantize the Dirac Lagrangian,
L =
¯
ψ(x)
_
i / ∂ −m
_
ψ(x) (5.1)
We will proceed naively and treat ψ as we did the scalar field. But we’ll see that things
go wrong and we will have to reconsider how to quantize this theory.
5.1 A Glimpse at the Spin-Statistics Theorem
We start in the usual way and define the momentum,
π =
∂L

˙
ψ
= i
¯
ψγ
0
= iψ

(5.2)
For the Dirac Lagrangian, the momentum conjugate to ψ is iψ

. It does not involve
the time derivative of ψ. This is as it should be for an equation of motion that is first
order in time, rather than second order. This is because we need only specify ψ and
ψ

on an initial time slice to determine the full evolution.
To quantize the theory, we promote the field ψ and its momentum ψ

to operators,
satisfying the canonical commutation relations, which read

α
(x), ψ
β
(y)] = [ψ

α
(x), ψ

β
(y)] = 0

α
(x), ψ

β
(y)] = δ
αβ
δ
(3)
(x −y) (5.3)
It’s this step that we’ll soon have to reconsider.
Since we’re dealing with a free theory, where any classical solution is a sum of plane
waves, we may write the quantum operators as
ψ(x) =
2

s=1
_
d
3
p
(2π)
3
1
_
2E
p
_
b
s
p
u
s
( p)e
+i p·x
+ c
s †
p
v
s
( p)e
−i p·x
_
ψ

(x) =
2

s=1
_
d
3
p
(2π)
3
1
_
2E
p
_
b
s †
p
u
s
( p)

e
−i p·x
+ c
s
p
v
s
( p)

e
+i p·x
_
(5.4)
where the operators b
s †
p
create particles associated to the spinors u
s
( p), while c
s †
p
create
particles associated to v
s
( p). As with the scalars, the commutation relations of the
fields imply commutation relations for the annihilation and creation operators
– 106 –
Claim: The field commutation relations (5.3) are equivalent to
[b
r
p
, b
s †
q
] = (2π)
3
δ
rs
δ
(3)
( p −q)
[c
r
p
, c
s †
q
] = −(2π)
3
δ
rs
δ
(3)
( p −q) (5.5)
with all other commutators vanishing. Note the strange minus sign in the [c, c

] term.
It’s not yet obvious that it’s bad, but we should be aware of it. For now, let’s just carry
on.
Proof: Let’s show that the [b, b

] and [c, c

] commutators reproduce the field com-
mutators (5.3),
[ψ(x), ψ

(y)] =

r,s
_
d
3
p d
3
q
(2π)
6
1
_
4E
p
E
q
_
[b
r
p
, b
s †
q
]u
r
( p)u
s
(q)

e
i(x· p− y· q)
+ [c
r †
p
, c
s
q
]v
r
( p)v
s
(q)

e
−i(x· p− y· q)
_
=

s
_
d
3
p
(2π)
3
1
2E
p
_
u
s
( p)¯ u
s
( p)γ
0
e
i p·(x− y)
+ v
s
( p)¯ v
s
( p)γ
0
e
−i p·(x− y)
_
(5.6)
At this stage we use the outer product formulae (4.128) and (4.129) which tell us

s
u
s
( p)¯ u
s
( p) = / p + m and

s
v
s
( p)¯ v
s
( p) = / p −m, so that
[ψ(x), ψ

(y)] =
_
d
3
p
(2π)
3
1
2E
p
_
( / p + m)γ
0
e
i p·(x− y)
+ ( / p −m)γ
0
e
−i p·(x− y)
_
=
_
d
3
p
(2π)
3
1
2E
p
_
(p
0
γ
0
+ p
i
γ
i
+ m)γ
0
+ (p
0
γ
0
−p
i
γ
i
−m)γ
0
_
e
+i p·(x− y)
where, in the second term, we’ve changed p → − p under the integration sign. Now,
using p
0
= E
p
we have
[ψ(x), ψ

(y)] =
_
d
3
p
(2π)
3
e
+i p·(x− y)
= δ
(3)
(x −y) (5.7)
as promised. Notice that it’s a little tricky in the middle there, making sure that
the p
i
γ
i
terms cancel. This was the reason we needed the minus sign in the [c, c

]
commutator terms in (5.5).
5.1.1 The Hamiltonian
To proceed, let’s construct the Hamiltonian for the theory. Using the momentum
π = iψ

, we have
H = π
˙
ψ −L =
¯
ψ(−iγ
i

i
+ m)ψ (5.8)
– 107 –
which means that H =
_
d
3
xH agrees with the conserved energy computed using
Noether’s theorem (4.92). We now wish to turn the Hamiltonian into an operator.
Let’s firstly look at
(−iγ
i

i
+ m)ψ =
_
d
3
p
(2π)
3
1
_
2E
p
_
b
s
p
(−γ
i
p
i
+ m)u
s
( p) e
+i p·x
+ c
s †
p

i
p
i
+ m)v
s
( p) e
−i p·x
_
where, for once we’ve left the sum over s = 1, 2 implicit. There’s a small subtlety with
the minus signs in deriving this equation that arises from the use of the Minkowski
metric in contracting indices, so that p x ≡

i
x
i
p
i
= −x
i
p
i
. Now we use the defining
equations for the spinors u
s
( p) and v
s
( p) given in (4.105) and (4.111), to replace
(−γ
i
p
i
+ m)u
s
( p) = γ
0
p
0
u
s
( p) and (γ
i
p
i
+ m)v
s
( p) = −γ
0
p
0
v
s
( p) (5.9)
so we can write
(−iγ
i

i
+ m)ψ =
_
d
3
p
(2π)
3
_
E
p
2
γ
0
_
b
s
p
u
s
( p) e
+i p·x
−c
s †
p
v
s
( p) e
−i p·x
_
(5.10)
We now use this to write the operator Hamiltonian
H =
_
d
3


γ
0
(−iγ
i

i
+ m)ψ
=
_
d
3
xd
3
p d
3
q
(2π)
6
¸
E
p
4E
q
_
b
r †
q
u
r
(q)

e
−i q·x
+ c
r
q
v
r
(q)

e
+i q·x
_

_
b
s
p
u
s
( p)e
+i p·x
−c
s †
p
v
s
( p)

e
−i p·x
_
=
_
d
3
p
(2π)
3
1
2
_
b
r †
p
b
s
p
[u
r
( p)

u
s
( p)] −c
r
p
c
s †
p
[v
r
( p)

v
s
( p)]
−b
r †
p
c
s †
− p
[u
r
( p)

v
s
(− p)] + c
r
p
b
s
− p
[v
r
( p)

v
s
(− p)]
_
where, in the last two terms we have relabelled p →− p. We now use our inner product
formulae (4.122), (4.124) and (4.127) which read
u
r
( p)

u
s
( p) = v
r
( p)

v
s
( p) = 2p
0
δ
rs
and u
r
( p)

v
s
(− p) = v
r
( p)

u
s
(− p) = 0
giving us
H =
_
d
3
p
(2π)
3
E
p
_
b
s †
p
b
s
p
−c
s
p
c
s †
p
_
(5.11)
=
_
d
3
p
(2π)
3
E
p
_
b
s †
p
b
s
p
−c
s †
p
c
s
p
+ (2π)
3
δ
(3)
(0)
_
(5.12)
– 108 –
The δ
(3)
term is familiar and easily dealt with by normal ordering. However the −c

c
term is a disaster! The Hamiltonian is not bounded below, meaning that our quantum
theory makes no sense. Taken seriously it would tell us that we could tumble to states
of lower and lower energy by continually producing c

particles. As the English would
say, it’s all gone a bit Pete Tong. (No relation).
Since the above calculation was a little tricky, you might think that it’s possible to
rescue the theory to get the minus signs to work out right. You can play around with
different things, but you’ll always find this minus sign cropping up somewhere. And,
in fact, it’s telling us something important that we missed.
5.2 Fermionic Quantization
The key piece of physics that we missed is that spin 1/2 particles are fermions, meaning
that they obey Fermi-Dirac statistics with the quantum state picking up a minus sign
upon the interchange of any two particles. This fact is embedded into the structure
of relativistic quantum field theory: the spin-statistics theorem says that integer spin
fields must be quantized as bosons, while half-integer spin fields must be quantized
as fermions. Any attempt to do otherwise will lead to an inconsistency, such as the
unbounded Hamiltonian we saw in (5.12).
So how do we go about quantizing a field as a fermion? Recall that when we quantized
the scalar field, the resulting particles obeyed bosonic statistics because the creation
and annihilation operators satisfied the commutation relations,
[a

p
, a

q
] = 0 ⇒ a

p
a

q
[0) ≡ [p, q) = [q, p) (5.13)
To have states obeying fermionic statistics, we need anti-commutation relations, ¦A, B¦ ≡
AB + BA. Rather than (5.3), we will ask that the spinor fields satisfy
¦ψ
α
(x), ψ
β
(y)¦ = ¦ψ

α
(x), ψ

β
(y)¦ = 0
¦ψ
α
(x), ψ

β
(y)¦ = δ
αβ
δ
(3)
(x −y) (5.14)
We still have the expansion (5.4) of ψ and ψ

in terms of b, b

, c and c

. But now the
same proof that led us to (5.5) tells us that
¦b
r
p
, b
s †
q
¦ = (2π)
3
δ
rs
δ
(3)
( p −q)
¦c
r
p
, c
s †
q
¦ = (2π)
3
δ
rs
δ
(3)
( p −q) (5.15)
with all other anti-commutators vanishing,
¦b
r
p
, b
s
q
¦ = ¦c
r
p
, c
s
q
¦ = ¦b
r
p
, c
s †
q
¦ = ¦b
r
p
, c
s
q
¦ = . . . = 0 (5.16)
– 109 –
The calculation of the Hamiltonian proceeds as before, all the way through to the
penultimate line (5.11). At that stage, we get
H =
_
d
3
p
(2π)
3
E
p
_
b
s †
p
b
s
p
−c
s
p
c
s †
p
_
=
_
d
3
p
(2π)
3
E
p
_
b
s †
p
b
s
p
+ c
s †
p
c
s
p
−(2π)
3
δ
(3)
(0)
_
(5.17)
The anti-commutators have saved us from the indignity of an unbounded Hamiltonian.
Note that when normal ordering the Hamiltonian we now throw away a negative contri-
bution −(2π)
3
δ
(3)
(0). In principle, this could partially cancel the positive contribution
from bosonic fields. Cosmological constant problem anyone?!
5.2.1 Fermi-Dirac Statistics
Just as in the bosonic case, we define the vacuum [0) to satisfy,
b
s
p
[0) = c
s
p
[0) = 0 (5.18)
Although b and c obey anti-commutation relations, the Hamiltonian (5.17) has nice
commutation relations with them. You can check that
[H, b
r
p
] = −E
p
b
r
p
and [H, b
r †
p
] = E
p
b
r †
p
[H, c
r
p
] = −E
p
c
r
p
and [H, c
r †
p
] = E
p
c
r †
p
(5.19)
This means that we can again construct a tower of energy eigenstates by acting on the
vacuum by b
r †
p
and c
r,†
p
to create particles and antiparticles, just as in the bosonic case.
For example, we have the one-particle states
[p, r) = b
r †
p
[0) (5.20)
The two particle states now satisfy
[p
1
, r
1
; p
2
, r
2
) ≡ b
r
1

p
1
b
r
2

p
2
[0) = −[p
2
, r
2
; p
1
, r
1
) (5.21)
confirming that the particles do indeed obey Fermi-Dirac statistics. In particular, we
have the Pauli-Exclusion principle [p, r; p, r) = 0. Finally, if we wanted to be sure
about the spin of the particle, we could act with the angular momentum operator
(4.96) to confirm that a stationary particle [ p = 0, r) does indeed carry intrinsic angular
momentum 1/2 as expected.
5.3 Dirac’s Hole Interpretation
“In this attempt, the success seems to have been on the side of Dirac rather
than logic”
Pauli on Dirac
– 110 –
Let’s pause our discussion to make a small historical detour. Dirac originally viewed
his equation as a relativistic version of the Schr¨odinger equation, with ψ interpreted
as the wavefunction for a single particle with spin. To reinforce this interpretation, he
wrote (i / ∂ −m)ψ = 0 as
i
∂ψ
∂t
= −i α

∇ψ + mβψ ≡
ˆ
Hψ (5.22)
where α = −γ
0
γ and β = γ
0
. Here the operator
ˆ
H is interpreted as the one-particle
Hamiltonian. This is a very different viewpoint from the one we now have, where ψ
is a classical field that should be quantized. In Dirac’s view, the Hamiltonian of the
system is
ˆ
H defined above, while for us the Hamiltonian is the field operator (5.17).
Let’s see where Dirac’s viewpoint leads.
With the interpretation of ψ as a single-particle wavefunction, the plane-wave solu-
tions (4.104) and (4.110) to the Dirac equation are thought of as energy eigenstates,
with
ψ = u( p) e
−ip·x
⇒ i
∂ψ
∂t
= E
p
ψ
ψ = v( p) e
+ip·x
⇒ i
∂ψ
∂t
= −E
p
ψ (5.23)
which look like positive and negative energy solutions. The spectrum is once again
unbounded below; there are states v( p) with arbitrarily low energy −E
p
. At first
glance this is disastrous, just like the unbounded field theory Hamiltonian (5.12). Dirac
postulated an ingenious solution to this problem: since the electrons are fermions (a
fact which is put in by hand to Dirac’s theory) they obey the Pauli-exclusion principle.
So we could simply stipulate that in the true vacuum of the universe, all the negative
energy states are filled. Only the positive energy states are accessible. These filled
negative energy states are referred to as the Dirac sea. Although you might worry about
the infinite negative charge of the vacuum, Dirac argued that only charge differences
would be observable (a trick reminiscent of the normal ordering prescription we used
for field operators).
Having avoided disaster by floating on an infinite sea comprised of occupied negative
energy states, Dirac realized that his theory made a shocking prediction. Suppose that
a negative energy state is excited to a positive energy state, leaving behind a hole.
The hole would have all the properties of the electron, except it would carry positive
charge. After flirting with the idea that it may be the proton, Dirac finally concluded
that the hole is a new particle: the positron. Moreover, when a positron comes across
– 111 –
an electron, the two can annihilate. Dirac had predicted anti-matter, one of the greatest
achievements of theoretical physics. It took only a couple of years before the positron
was discovered experimentally in 1932.
Although Dirac’s physical insight led him to the right answer, we now understand
that the interpretation of the Dirac spinor as a single-particle wavefunction is not really
correct. For example, Dirac’s argument for anti-matter relies crucially on the particles
being fermions while, as we have seen already in this course, anti-particles exist for
both fermions and bosons. What we really learn from Dirac’s analysis is that there is
no consistent way to interpret the Dirac equation as describing a single particle. It is
instead to be thought of as a classical field which has only positive energy solutions
because the Hamiltonian (4.92) is positive definite. Quantization of this field then gives
rise to both particle and anti-particle excitations.
This from Julian Schwinger:
“Until now, everyone thought that the Dirac equation referred directly to
physical particles. Now, in field theory, we recognize that the equations
refer to a sublevel. Experimentally we are concerned with particles, yet the
old equations describe fields.... When you begin with field equations, you
operate on a level where the particles are not there from the start. It is
when you solve the field equations that you see the emergence of particles.”
5.4 Propagators
Let’s now move to the Heisenberg picture. We define the spinors ψ(x, t) at every point
in spacetime such that they satisfy the operator equation
∂ψ
∂t
= i[H, ψ] (5.24)
We solve this by the expansion
ψ(x) =
2

s=1
_
d
3
p
(2π)
3
1
_
2E
p
_
b
s
p
u
s
( p)e
−ip·x
+ c
s †
p
v
s
( p)e
+ip·x
_
ψ

(x) =
2

s=1
_
d
3
p
(2π)
3
1
_
2E
p
_
b
s †
p
u
s
( p)

e
+ip·x
+ c
s
p
v
s
( p)

e
−ip·x
_
(5.25)
Let’s now look at the anti-commutators of these fields. We define the fermionic prop-
agator to be
iS
αβ
= ¦ψ
α
(x),
¯
ψ
β
(y)¦ (5.26)
– 112 –
In what follows we will often drop the indices and simply write iS(x−y) = ¦ψ(x),
¯
ψ(y)¦,
but you should remember that S(x−y) is a 44 matrix. Inserting the expansion (5.25),
we have
iS(x −y) =
_
d
3
p d
3
q
(2π)
6
1
_
4E
p
E
q
_
¦b
s
p
, b
r †
q
¦u
s
( p)¯ u
r
(q)e
−i(p·x−q·y)
+¦c
s †
p
, c
r
q
¦v
s
( p)¯ v
r
(q)e
+i(p·x−q·y)
_
=
_
d
3
p
(2π)
3
1
2E
p
_
u
s
( p)¯ u
s
( p)e
−ip·(x−y)
+ v
s
( p)¯ v
s
( p)e
+ip·(x−y)
¸
=
_
d
3
p
(2π)
3
1
2E
p
_
( / p + m)e
−ip·(x−y)
+ ( / p −m)e
+ip·(x−y)
¸
(5.27)
where to reach the final line we have used the outer product formulae (4.128) and
(4.129). We can then write
iS(x −y) = (i / ∂
x
+ m)(D(x −y) −D(y −x)) (5.28)
in terms of the propagator for a real scalar field D(x −y) which, recall, can be written
as (2.90)
D(x −y) =
_
d
3
p
(2π)
3
1
2E
p
e
−ip·(x−y)
(5.29)
Some comments:
• For spacelike separated points (x−y)
2
< 0, we have already seen that D(x−y) −
D(y −x) = 0. In the bosonic theory, we made a big deal of this since it ensured
that
[φ(x), φ(y)] = 0 (x −y)
2
< 0 (5.30)
outside the lightcone, which we trumpeted as proof that our theory was causal.
However, for fermions we now have
¦ψ
α
(x), ψ
β
(y)¦ = 0 (x −y)
2
< 0 (5.31)
outside the lightcone. What happened to our precious causality? The best that
we can say is that all our observables are bilinear in fermions, for example the
Hamiltonian (5.17). These still commute outside the lightcone. The theory re-
mains causal as long as fermionic operators are not observable. If you think this is
a little weak, remember that no one has ever seen a physical measuring apparatus
come back to minus itself when you rotate by 360 degrees!
– 113 –
• At least away from singularities, the propagator satisfies
(i / ∂
x
−m)S(x −y) = 0 (5.32)
which follows from the fact that ( / ∂
2
x
+ m
2
)D(x − y) = 0 using the mass shell
condition p
2
= m
2
.
5.5 The Feynman Propagator
By a similar calculation to that above, we can determine the vacuum expectation value,
¸0[ ψ
α
(x)
¯
ψ
β
(y) [0) =
_
d
3
p
(2π)
3
1
2E
p
( / p + m)
αβ
e
−ip·(x−y)
¸0[
¯
ψ
α
(y)ψ
β
(x) [0) =
_
d
3
p
(2π)
3
1
2E
p
( / p −m)
αβ
e
+ip·(x−y)
(5.33)
We now define the Feynman propagator S
F
(x − y), which is again a 4 4 matrix, as
the time ordered product,
S
F
(x −y) = ¸0[ Tψ(x)
¯
ψ(y) [0) ≡
_
¸0[ ψ(x)
¯
ψ(y) [0) x
0
> y
0
¸0[ −
¯
ψ(y)ψ(x) [0) y
0
> x
0
(5.34)
Notice the minus sign! It is necessary for Lorentz invariance. When (x−y)
2
< 0, there is
no invariant way to determine whether x
0
> y
0
or y
0
> x
0
. In this case the minus sign is
necessary to make the two definitions agree since ¦ψ(x),
¯
ψ(y)¦ = 0 outside the lightcone.
We have the 4-momentum integral representation for the Feynman propagator,
S
F
(x −y) = i
_
d
4
p
(2π)
4
e
−ip·(x−y)
γ p + m
p
2
−m
2
+ iǫ
(5.35)
which satisfies (i / ∂
x
−m)S
F
(x −y) = iδ
(4)
(x −y), so that S
F
is a Green’s function for
the Dirac operator.
The minus sign that we see in (5.34) also occurs for any string of operators inside
a time ordered product T(. . .). While bosonic operators commute inside T, fermionic
operators anti-commute. We have this same behaviour for normal ordered products as
well, with fermionic operators obeying : ψ
1
ψ
2
:= − : ψ
2
ψ
1
:. With the understanding
that all fermionic operators anti-commute inside T and ::, Wick’s theorem proceeds
just as in the bosonic case. We define the contraction
¸ .. ¸
ψ(x)
¯
ψ(y) = T(ψ(x)
¯
ψ(y)) − : ψ(x)
¯
ψ(y) : = S
F
(x −y) (5.36)
– 114 –
5.6 Yukawa Theory
The interaction between a Dirac fermion of mass m and a real scalar field of mass µ is
governed by the Yukawa theory,
L =
1
2

µ
φ∂
µ
φ −
1
2
µ
2
φ
2
+
¯
ψ(iγ
µ

µ
−m)ψ −λφ
¯
ψψ (5.37)
which is the proper version of the baby scalar Yukawa theory we looked at in Section 3.
Couplings of this type appear in the standard model, between fermions and the Higgs
boson. In that context, the fermions can be leptons (such as the electron) or quarks.
Yukawa originally proposed an interaction of this type as an effective theory of nuclear
forces. With an eye to this, we will again refer to the φ particles as mesons, and the
ψ particles as nucleons. Except, this time, the nucleons have spin. (This is still not
a particularly realistic theory of nucleon interactions, not least because we’re omitting
isospin. Moreover, in Nature the relevant mesons are pions which are pseudoscalars, so
a coupling of the form φ
¯
ψγ
5
ψ would be more appropriate. We’ll turn to this briefly in
Section 5.7.3).
Note the dimensions of the various fields. We still have [φ] = 1, but the kinetic
terms require that [ψ] = 3/2. Thus, unlike in the case with only scalars, the coupling
is dimensionless: [λ] = 0.
We’ll proceed as we did in Section 3, firstly computing the amplitude of a particular
scattering process then, with that calculation as a guide, writing down the Feynman
rules for the theory. We start with:
5.6.1 An Example: Putting Spin on Nucleon Scattering
Let’s study ψψ →ψψ scattering. This is the same calculation we performed in Section
(3.3.3) except now the fermions have spin. Our initial and final states are
[i) =
_
4E
p
E
q
b
s †
p
b
r †
q
[0) ≡ [p, s; q, r)
[f) =
_
4E
p
′ E
q
′ b
s


p

b
r


q

[0) ≡ [ p

, s

; q

, r

) (5.38)
We need to be a little cautious about minus signs, because the b

’s now anti-commute.
In particular, we should be careful when we take the adjoint. We have
¸f[ =
_
4E
p
′ E
q
′ ¸0[ b
r

q
′ b
s

p
′ (5.39)
We want to calculate the order λ
2
terms from the S-matrix element ¸f[ S −1 [i).
(−iλ)
2
2
_
d
4
x
1
d
4
x
2
T
_
¯
ψ(x
1
)ψ(x
1
)φ(x
1
)
¯
ψ(x
2
)ψ(x
2
)φ(x
2
)
_
(5.40)
– 115 –
where, as usual, all fields are in the interaction picture. Just as in the bosonic calcula-
tion, the contribution to nucleon scattering comes from the contraction
:
¯
ψ(x
1
)ψ(x
1
)
¯
ψ(x
2
)ψ(x
2
) :
¸ .. ¸
φ(x
1
)φ(x
2
) (5.41)
We just have to be careful about how the spinor indices are contracted. Let’s start
by looking at how the fermionic operators act on [i). We expand out the ψ fields,
leaving the
¯
ψ fields alone for now. We may ignore the c

pieces in ψ since they give no
contribution at order λ
2
. We have
:
¯
ψ(x
1
)ψ(x
1
)
¯
ψ(x
2
)ψ(x
2
) : b
s †
p
b
r †
q
[0) = −
_
d
3
k
1
d
3
k
2
(2π)
6
[
¯
ψ(x
1
) u
m
(

k
1
)] [
¯
ψ(x
2
) u
n
(

k
2
)]
e
−ik
1
·x
1
−ik
2
·x
2
_
4E

k
1
E

k
2
b
m

k
1
b
n

k
2
b
s †
p
b
r †
q
[0) (5.42)
where we’ve used square brackets [] to show how the spinor indices are contracted. The
minus sign that sits out front came from moving ψ(x
1
) past
¯
ψ(x
2
). Now anti-commuting
the b’s past the b

’s, we get
=
−1
2
_
E
p
E
q
_
[
¯
ψ(x
1
) u
r
(q)] [
¯
ψ(x
2
) u
s
( p)]e
−ip·x
2
−iq·x
1
− [
¯
ψ(x
1
) u
s
( p)] [
¯
ψ(x
2
) u
r
(q)]e
−ip·x
1
−iq·x
2
_
[0) (5.43)
Note, in particular, the relative minus sign that appears between these two terms. Now
let’s see what happens when we hit this with ¸f[. We look at
¸0[ b
r

q
′ b
s

p
′ [
¯
ψ(x
1
) u
r
(q)] [
¯
ψ(x
2
) u
s
( p)] [0) =
e
+ip

·x
1
+iq

·x
2
2
_
E
p
′ E
q

[¯ u
s

( p

) u
r
(q)] [¯ u
r

(q

) u
s
( p)]

e
+ip

·x
2
+iq

·x
1
2
_
E
p
′ E
q

[¯ u
r

(q

) u
r
(q)] [¯ u
s

( p

) u
s
( p)]
The [
¯
ψ(x
1
) u
s
( p)] [
¯
ψ(x
2
) u
r
(q)] term in (5.43) doubles up with this, cancelling the
factor of 1/2 in front of (5.40). Meanwhile, the 1/

E terms cancel the relativistic
state normalization. Putting everything together, we have the following expression for
¸f[ S −1 [i)
(−iλ)
2
_
d
4
x
1
d
4
x
2
d
4
k
(2π)
4
ie
ik·(x
1
−x
2
)
k
2
−µ
2
+ iǫ
_
[¯ u
s

( p

) u
s
( p)] [¯ u
r

(q

) u
r
(q)]e
+ix
1
·(q

−q)+ix
2
·(p

−p)
− [¯ u
s

( p

) u
r
(q)] [¯ u
r

(q

) u
s
( p)]e
ix
1
·(p

−q)+ix
2
·(q

−p)
_
– 116 –
where we’ve put the φ propagator back in. Performing the integrals over x
1
and x
2
,
this becomes,
_
d
4
k
(2π)
4
i(−iλ)
2
k
2
−µ
2
+ iǫ
_
[¯ u
s

( p

) u
s
( p)] [¯ u
r

(q

) u
r
(q)]δ
(4)
(q

−q + k)δ
(4)
(p

−p −k)
− [¯ u
s

( p

) u
r
(q)] [¯ u
r

(q

) u
s
( p)]δ
(4)
(p

−q + k)δ
(4)
(q

−p −k)
_
And we’re almost there! Finally, writing the S-matrix element in terms of the amplitude
in the usual way, ¸f[ S −1 [i) = i/(2π)
4
δ
(4)
(p + q −p

−q

), we have
/ = (−iλ)
2
_
[¯ u
s

( p

) u
s
( p)] [¯ u
r

(q

) u
r
(q)]
(p

−p)
2
−µ
2
+ iǫ

[¯ u
s

( p

) u
r
(q)] [¯ u
r

(q

) u
s
( p)]
(q

−p)
2
−µ
2
+ iǫ
_
which is our final answer for the amplitude.
5.7 Feynman Rules for Fermions
It’s important to bear in mind that the calculation we just did kind of blows. Thankfully
the Feynman rules will once again encapsulate the combinatoric complexities and make
life easier for us. The rules to compute amplitudes are the following
• To each incoming fermion with momentum p and spin r, we associate a spinor
u
r
( p). For outgoing fermions we associate ¯ u
r
( p).
u (p)
r
p
u (p)
r
p
Figure 21: An incoming fermion Figure 22: An outgoing fermion
• To each incoming anti-fermion with momentum p and spin r, we associate a spinor
¯ v
r
( p). For outgoing anti-fermions we associate v
r
( p).
v (p)
r
p
r
v (p)
p
Figure 23: An incoming anti-fermion Figure 24: An outgoing anti-fermion
• Each vertex gets a factor of −iλ.
– 117 –
• Each internal line gets a factor of the relevant propagator.
p
i
p
2
−µ
2
+ iǫ
for scalars
p
i( / p + m)
p
2
−m
2
+ iǫ
for fermions (5.44)
The arrows on the fermion lines must flow consistently through the diagram (this
ensures fermion number conservation). Note that the fermionic propagator is a
44 matrix. The matrix indices are contracted at each vertex, either with further
propagators, or with external spinors u, ¯ u, v or ¯ v.
• Impose momentum conservation at each vertex, and integrate over undetermined
loop momenta.
• Add extra minus signs for statistics. Some examples will be given below.
5.7.1 Examples
Let’s run through the same examples we did for the scalar Yukawa theory. Firstly, we
have
Nucleon Scattering
For the example we worked out previously, the two lowest order Feynman diagrams are
shown in Figure 25. We’ve drawn the second Feynman diagram with the legs crossed
/
p,s
/
q,r
/ /
/
p,s
/
q,r
/ /
+
q,r
p,s
q,r
p,s
Figure 25: The two Feynman diagrams for nucleon scattering
to emphasize the fact that it picks up a minus sign due to statistics. (Note that the
way the legs point in the Feynman diagram doesn’t tell us the direction in which the
particles leave the scattering event: the momentum label does that. The two diagrams
above are different because the incoming legs are attached to different outgoing legs).
Using the Feynman rules we can read off the amplitude.
/ = (−iλ)
2
_
[¯ u
s

( p

) u
s
( p)] [¯ u
r

(q

) u
r
(q)]
(p −p

)
2
−µ
2

[¯ u
s

( p

) u
r
(q)] [¯ u
r

(q

) u
s
( p)]
(p −q

)
2
−µ
2
_
(5.45)
– 118 –
The denominators in each term are due to the meson propagator, with the momen-
tum determined by conservation at each vertex. This agrees with the amplitude we
computed earlier using Wick’s theorem.
Nucleon to Meson Scattering
Let’s now look at ψ
¯
ψ → φφ. The two lowest order Feynman diagrams are shown in
Figure 26.
/
p,s
/
q,r
/ /
/
p,s
/
q,r
/ /
+
q,r
p,s
q,r
p,s
Figure 26: The two Feynman diagrams for nucleon to meson scattering
Applying the Feynman rules, we have
/ = (−iλ)
2
_
¯ v
r
(q)[γ
µ
(p
µ
−p

µ
) + m]u
s
( p)
(p −p

)
2
−m
2
+
¯ v
r
(q)[γ
µ
(p
µ
−q

µ
) + m]u
s
( p)
(p −q

)
2
−m
2
_
Since the internal line is now a fermion, the propagator contains γ
µ
(p
µ
−p

µ
)+m factors.
This is a 4 4 matrix which sits on the top, sandwiched between the two external
spinors. Now the exchange statistics applies to the final meson states. These are
bosons and, correspondingly, there is no relative minus sign between the two diagrams.
Nucleon-Anti-Nucleon Scattering
For ψ
¯
ψ → ψ
¯
ψ, the two lowest order Feynman diagrams are of two distinct types, just
like in the bosonic case. They are shown in Figure 27. The corresponding amplitude
is given by,
/ = (−iλ)
2
_

[¯ u
s

( p

) u
s
( p)] [¯ v
r
(q) v
r

(q

)]
(p −p

)
2
−µ
2
+
[¯ v
r
(q) u
s
( p)] [¯ u
s

( p

) v
r

(q

)]
(p + q)
2
−µ
2
+ iǫ
_
(5.46)
As in the bosonic diagrams, there is again the difference in the momentum dependence
in the denominator. But now the difference in the diagrams is also reflected in the
spinor contractions in the numerator.
– 119 –
/
p,s
/
q,r
/ /
/
p,s
/
q,r
/ /
+
p,s
q,r
q,r
p,s
Figure 27: The two Feynman diagrams for nucleon-anti-nucleon scattering
More subtle are the minus signs. The fermionic statistics mean that the first diagram
has an extra minus sign relative to the ψψ scattering of Figure 25. Since this minus sign
will be important when we come to figure out whether the Yukawa force is attractive
or repulsive, let’s go back to basics and see where it comes from. The initial and final
states for this scattering process are
[i) =
_
4E
p
E
q
b
s †
p
c
r †
q
[0) ≡ [ p, s; q, r)
[f) =
_
4E
p
′ E
q
′ b
s


p

c
r


q

[0) ≡ [p

, s

; q

, r

) (5.47)
The ordering of b

and c

in these states is crucial and reflects the scattering ψ
¯
ψ →ψ
¯
ψ,
as opposed to ψ
¯
ψ →
¯
ψψ which would differ by a minus sign. The first diagram in
Figure 27 comes from the term in the perturbative expansion,
¸f[ :
¯
ψ(x
1
)ψ(x
1
)
¯
ψ(x
2
)ψ(x
2
) : b
s †
p
c
r †
q
[0) ∼ ¸f[ [¯ v
m
(

k
1
) ψ(x
1
)] [
¯
ψ(x
2
) u
n
(

k
2
)]c
m

k
1
b
n

k
2
b
s †
p
c
r †
q
[0)
where we’ve neglected a bunch of objects in this equation like
_
d
4
k
i
and exponential
factors because we only want to keep track of the minus signs. Moving the annihilation
operators past the creation operators, we have
+¸f[ [¯ v
r
(q) ψ(x
1
)] [
¯
ψ(x
2
) u
s
( p)] [0) (5.48)
Repeating the process by expanding out the ψ(x
1
) and
¯
ψ(x
2
) fields and moving them
to the left to annihilate ¸f[, we have
¸0[ c
r

q
′ b
s

p
′ c
m†

l
1
b
n†

l
2
[¯ v
r
(q) v
m
(

l
1
)] [¯ u
n
(

l
2
) u
s
( p)] [0) ∼ −[¯ v
r
(q) v
r

(q

)] [¯ u
s

( p

) u
s
( p)]
where the minus sign has appeared from anti-commuting c
m†

l
1
past b
s

p

. This is the
overall minus sign found in (5.46). One can also follow similar contractions to compute
the second diagram in Figure 27.
Meson Scattering
– 120 –
Finally, we can also compute the scattering of φφ → φφ
p
2
p
1
/
p
/
2
1
p
Figure 28:
which, as in the bosonic case, picks up its leading contribution
at one-loop. The amplitude for the diagram shown in the figure
is
i/ = −(−iλ)
4
_
d
4
k
(2π)
4
Tr
/ k + m
(k
2
−m
2
+ iǫ)
/ k + / p

1
+ m
((k + p

1
)
2
−m
2
+ iǫ)

/ k + / p

1
− / p
1
+ m
((k + p

1
−p
1
)
2
−m
2
+ iǫ)
/ k − / p

2
+ m
((k −p

2
)
2
−m
2
+ iǫ)
Notice that the high momentum limit of the integral is
_
d
4
k/k
4
, which is no longer
finite, but diverges logarithmically. You will have to wait until next term to make sense
of this integral.
There’s an overall minus sign sitting in front of this amplitude. This is a generic
feature of diagrams with fermions running in loops: each fermionic loop in a diagram
gives rise to an extra minus sign. We can see this rather simply in the diagram
which involves the expression
¸ .. ¸
¯
ψ
α
(x)
¸ .. ¸
ψ
α
(x)
¯
ψ
β
(y) ψ
β
(y) = −
¸ .. ¸
ψ
β
(y)
¯
ψ
α
(x)
¸ .. ¸
ψ
α
(x)
¯
ψ
β
(y)
= −Tr (S
F
(y −x)S
F
(x −y))
After passing the fermion fields through each other, a minus sign appears, sitting in
front of the two propagators.
5.7.2 The Yukawa Potential Revisited
We saw in Section 3.5.2, that the exchange of a real scalar particle gives rise to a
universally attractive Yukawa potential between two spin zero particles. Does the
same hold for the spin 1/2 particles?
Recall that the strategy to compute the potential is to take the non-relativistic limit
of the scattering amplitude, and compare with the analogous result from quantum
mechanics. Our new amplitude now also includes the spinor degrees of freedom u( p)
– 121 –
and v( p). In the non-relativistic limit, p →(m, p), and
u( p) =
_

p σξ

p ¯ σξ
_


m
_
ξ
ξ
_
v( p) =
_

p σξ


p ¯ σξ
_


m
_
ξ
−ξ
_
(5.49)
In this limit, the spinor contractions in the amplitude for ψψ → ψψ scattering (5.45)
become ¯ u
s

u
s
= 2mδ
ss

and the amplitude is
/
p,s
/
q,r
/ /
q,r
p,s
= −i(−iλ)
2
(2m)
_
δ
s

s
δ
r

r
( p −p

) + µ
2

δ
s

r
δ
r

s
( p −q

) + µ
2
_
(5.50)
The δ symbols tell us that spin is conserved in the non-relativistic limit, while the
momentum dependence is the same as in the bosonic case, telling us that once again
the particles feel an attractive Yukawa potential,
U(r) = −
λ
2
e
−µr
4πr
(5.51)
Repeating the calculation for ψ
¯
ψ → ψ
¯
ψ, there are two minus signs which cancel each
other. The first is the extra overall minus sign in the scattering amplitude (5.46),
due to the fermionic nature of the particles. The second minus sign comes from the
non-relativistic limit of the spinor contraction for anti-particles in (5.46), which is
¯ v
s

v
s
= −2mδ
ss

. These two signs cancel, giving us once again an attractive Yukawa
potential (5.51).
5.7.3 Pseudo-Scalar Coupling
Rather than the standard Yukawa coupling, we could instead consider
L
Yuk
= −λφ
¯
ψγ
5
ψ (5.52)
This still preserves parity if φ is a pseudoscalar, i.e.
P : φ(x, t) →−φ(−x, t) (5.53)
We can compute in this theory very simply: the Feynman rule for the interaction vertex
is now changed to a factor of −iλγ
5
. For example, the Feynman diagrams for ψψ →ψψ
scattering are again given by Figure 25, with the amplitude now
/ = (−iλ)
2
_
[¯ u
s

( p


5
u
s
( p)] [¯ u
r

(q


5
u
r
(q)]
(p −p

)
2
−µ
2

[¯ u
s

( p


5
u
r
(q)] [¯ u
r

(q


5
u
s
( p)]
(p −q

)
2
−µ
2
_
– 122 –
We could again try to take the non-relativistic limit for this amplitude. But this
time, things work a little differently. Using the expressions for the spinors (5.49), we
have ¯ u
s

γ
5
u
s
→ 0 in the non-relativistic limit. To find the non-relativistic amplitude,
we must go to next to leading order. One can easily check that ¯ u
s

( p


5
u
s
( p) →

s

T
( p− p

) σξ
s
. So, in the non-relativistic limit, the leading order amplitude arising
from pseudoscalar exchange is given by a spin-spin coupling,
/
p,s
/
q,r
/ /
q,r
p,s
→+im(−iλ)
2

s

T
( p −p

) σξ
s
] [ξ
r

T
( p −p

) σξ
r
]
( p −p

)
2
+ µ
2
(5.54)
– 123 –
6. Quantum Electrodynamics
In this section we finally get to quantum electrodynamics (QED), the theory of light
interacting with charged matter. Our path to quantization will be as before: we start
with the free theory of the electromagnetic field and see how the quantum theory gives
rise to a photon with two polarization states. We then describe how to couple the
photon to fermions and to bosons.
6.1 Maxwell’s Equations
The Lagrangian for Maxwell’s equations in the absence of any sources is simply
L = −
1
4
F
µν
F
µν
(6.1)
where the field strength is defined by
F
µν
= ∂
µ
A
ν
−∂
ν
A
µ
(6.2)
The equations of motion which follow from this Lagrangian are

µ
_
∂L
∂(∂
µ
A
ν
)
_
= −∂
µ
F
µν
= 0 (6.3)
Meanwhile, from the definition of F
µν
, the field strength also satisfies the Bianchi
identity

λ
F
µν
+ ∂
µ
F
νλ
+ ∂
ν
F
λµ
= 0 (6.4)
To make contact with the form of Maxwell’s equations you learn about in high school,
we need some 3-vector notation. If we define A
µ
= (φ,

A), then the electric field

E and
magnetic field

B are defined by

E = −∇φ −


A
∂t
and

B = ∇

A (6.5)
which, in terms of F
µν
, becomes
F
µν
=
_
0 Ex Ey Ez
−Ex 0 −Bz By
−Ey Bz 0 −Bx
−Ez −By Bx 0
_
(6.6)
The Bianchi identity (6.4) then gives two of Maxwell’s equations,


B = 0 and

B
∂t
= −∇

E (6.7)
– 124 –
These remain true even in the presence of electric sources. Meanwhile, the equations
of motion give the remaining two Maxwell equations,


E = 0 and

E
∂t
= ∇

B (6.8)
As we will see shortly, in the presence of charged matter these equations pick up extra
terms on the right-hand side.
6.1.1 Gauge Symmetry
The massless vector field A
µ
has 4 components, which would naively seem to tell us that
the gauge field has 4 degrees of freedom. Yet we know that the photon has only two
degrees of freedom which we call its polarization states. How are we going to resolve
this discrepancy? There are two related comments which will ensure that quantizing
the gauge field A
µ
gives rise to 2 degrees of freedom, rather than 4.
• The field A
0
has no kinetic term
˙
A
0
in the Lagrangian: it is not dynamical. This
means that if we are given some initial data A
i
and
˙
A
i
at a time t
0
, then the field
A
0
is fully determined by the equation of motion ∇

E = 0 which, expanding out,
reads

2
A
0
+∇


A
∂t
= 0 (6.9)
This has the solution
A
0
(x) =
_
d
3
x

(∇ ∂

A/∂t)(x

)
4π[x −x

[
(6.10)
So A
0
is not independent: we don’t get to specify A
0
on the initial time slice. It
looks like we have only 3 degrees of freedom in A
µ
rather than 4. But this is still
one too many.
• The Lagrangian (6.3) has a very large symmetry group, acting on the vector
potential as
A
µ
(x) →A
µ
(x) + ∂
µ
λ(x) (6.11)
for any function λ(x). We’ll ask only that λ(x) dies off suitably quickly at spatial
x →∞. We call this a gauge symmetry. The field strength is invariant under the
gauge symmetry:
F
µν
→∂
µ
(A
ν
+ ∂
ν
λ) −∂
ν
(A
µ
+ ∂
µ
λ) = F
µν
(6.12)
– 125 –
So what are we to make of this? We have a theory with an infinite number of
symmetries, one for each function λ(x). Previously we only encountered symme-
tries which act the same at all points in spacetime, for example ψ → e

ψ for a
complex scalar field. Noether’s theorem told us that these symmetries give rise
to conservation laws. Do we now have an infinite number of conservation laws?
The answer is no! Gauge symmetries have a very different interpretation than
the global symmetries that we make use of in Noether’s theorem. While the
latter take a physical state to another physical state with the same properties,
the gauge symmetry is to be viewed as a redundancy in our description. That is,
two states related by a gauge symmetry are to be identified: they are the same
physical state. (There is a small caveat to this statement which is explained in
Section 6.3.1). One way to see that this interpretation is necessary is to notice
that Maxwell’s equations are not sufficient to specify the evolution of A
µ
. The
equations read,

µν
(∂
ρ

ρ
) −∂
µ

ν
] A
ν
= 0 (6.13)
But the operator [η
µν
(∂
ρ

ρ
)−∂
µ

ν
] is not invertible: it annihilates any function of
the form ∂
µ
λ. This means that given any initial data, we have no way to uniquely
determine A
µ
at a later time since we can’t distinguish between A
µ
and A
µ
+∂
µ
λ.
This would be problematic if we thought that A
µ
is a physical object. However,
if we’re happy to identify A
µ
and A
µ
+∂
µ
λ as corresponding to the same physical
state, then our problems disappear.
Since gauge invariance is a redundancy of the system,
Gauge Orbits Gauge
Fixing
Figure 29:
we might try to formulate the theory purely in terms of
the local, physical, gauge invariant objects

E and

B. This
is fine for the free classical theory: Maxwell’s equations
were, after all, first written in terms of

E and

B. But it is
not possible to describe certain quantum phenomena, such
as the Aharonov-Bohm effect, without using the gauge
potential A
µ
. We will see shortly that we also require the
gauge potential to describe classically charged fields. To
describe Nature, it appears that we have to introduce quantities A
µ
that we can never
measure.
The picture that emerges for the theory of electromagnetism is of an enlarged phase
space, foliated by gauge orbits as shown in the figure. All states that lie along a given
– 126 –
line can be reached by a gauge transformation and are identified. To make progress,
we pick a representative from each gauge orbit. It doesn’t matter which representative
we pick — after all, they’re all physically equivalent. But we should make sure that we
pick a “good” gauge, in which we cut the orbits.
Different representative configurations of a physical state are called different gauges.
There are many possibilities, some of which will be more useful in different situations.
Picking a gauge is rather like picking coordinates that are adapted to a particular
problem. Moreover, different gauges often reveal slightly different aspects of a problem.
Here we’ll look at two different gauges:
• Lorentz Gauge: ∂
µ
A
µ
= 0
To see that we can always pick a representative configuration satisfying ∂
µ
A
µ
= 0,
suppose that we’re handed a gauge field A

µ
satisfying ∂
µ
(A

)
µ
= f(x). Then we
choose A
µ
= A

µ
+ ∂
µ
λ, where

µ

µ
λ = −f (6.14)
This equation always has a solution. In fact this condition doesn’t pick a unique
representative from the gauge orbit. We’re always free to make further gauge
transformations with ∂
µ

µ
λ = 0, which also has non-trivial solutions. As the
name suggests, the Lorentz gauge
3
has the advantage that it is Lorentz invariant.
• Coulomb Gauge: ∇

A = 0
We can make use of the residual gauge transformations in Lorentz gauge to pick


A = 0. (The argument is the same as before). Since A
0
is fixed by (6.10), we
have as a consequence
A
0
= 0 (6.15)
(This equation will no longer hold in Coulomb gauge in the presence of charged
matter). Coulomb gauge breaks Lorentz invariance, so may not be ideal for some
purposes. However, it is very useful to exhibit the physical degrees of freedom:
the 3 components of

A satisfy a single constraint: ∇

A = 0, leaving behind just
2 degrees of freedom. These will be identified with the two polarization states of
the photon. Coulomb gauge is sometimes called radiation gauge.
3
Named after Lorenz who had the misfortune to be one letter away from greatness.
– 127 –
6.2 The Quantization of the Electromagnetic Field
In the following we shall quantize free Maxwell theory twice: once in Coulomb gauge,
and again in Lorentz gauge. We’ll ultimately get the same answers and, along the way,
see that each method comes with its own subtleties.
The first of these subtleties is common to both methods and comes when computing
the momentum π
µ
conjugate to A
µ
,
π
0
=
∂L

˙
A
0
= 0
π
i
=
∂L

˙
A
i
= −F
0i
≡ E
i
(6.16)
so the momentum π
0
conjugate to A
0
vanishes. This is the mathematical consequence of
the statement we made above: A
0
is not a dynamical field. Meanwhile, the momentum
conjugate to A
i
is our old friend, the electric field. We can compute the Hamiltonian,
H =
_
d
3
x π
i
˙
A
i
−L
=
_
d
3
x
1
2

E

E +
1
2

B

B −A
0
(∇

E) (6.17)
So A
0
acts as a Lagrange multiplier which imposes Gauss’ law


E = 0 (6.18)
which is now a constraint on the system in which

A are the physical degrees of freedom.
Let’s now see how to treat this system using different gauge fixing conditions.
6.2.1 Coulomb Gauge
In Coulomb gauge, the equation of motion for

A is

µ

µ

A = 0 (6.19)
which we can solve in the usual way,

A =
_
d
3
p
(2π)
3

ξ( p) e
ip·x
(6.20)
with p
2
0
= [ p[
2
. The constraint ∇

A = 0 tells us that

ξ must satisfy

ξ p = 0 (6.21)
– 128 –
which means that

ξ is perpendicular to the direction of motion p. We can pick

ξ( p) to
be a linear combination of two orthonormal vectors ǫ
r
, r = 1, 2, each of which satisfies
ǫ
r
( p) p = 0 and
ǫ
r
( p) ǫ
s
( p) = δ
rs
r, s = 1, 2 (6.22)
These two vectors correspond to the two polarization states of the photon. It’s worth
pointing out that you can’t consistently pick a continuous basis of polarization vectors
for every value of p because you can’t comb the hair on a sphere. But this topological
fact doesn’t cause any complications in computing QED scattering processes.
To quantize we turn the Poisson brackets into commutators. Naively we would write
[A
i
(x), A
j
(y)] = [E
i
(x), E
j
(y)] = 0
[A
i
(x), E
j
(y)] = iδ
j
i
δ
(3)
(x −y) (6.23)
But this can’t quite be right, because it’s not consistent with the constraints. We
still want to have ∇

A = ∇

E = 0, now imposed on the operators. But from the
commutator relations above, we see
[∇

A(x), ∇

E(y)] = i∇
2
δ
(3)
(x −y) ,= 0 (6.24)
What’s going on? In imposing the commutator relations (6.23) we haven’t correctly
taken into account the constraints. In fact, this is a problem already in the classical
theory, where the Poisson bracket structure is already altered
4
. The correct Poisson
bracket structure leads to an alteration of the last commutation relation,
[A
i
(x), E
j
(y)] = i
_
δ
ij


i

j

2
_
δ
(3)
(x −y) (6.25)
To see that this is now consistent with the constraints, we can rewrite the right-hand
side of the commutator in momentum space,
[A
i
(x), E
j
(y)] = i
_
d
3
p
(2π)
3
_
δ
ij

p
i
p
j
[ p[
2
_
e
i p·(x− y)
(6.26)
which is now consistent with the constraints, for example
[∂
i
A
i
(x), E
j
(y)] = i
_
d
3
p
(2π)
3
_
δ
ij

p
i
p
j
[ p[
2
_
ip
i
e
i p·(x− y)
= 0 (6.27)
4
For a nice discussion of the classical and quantum dynamics of constrained systems, see the small
book by Paul Dirac, “Lectures on Quantum Mechanics”
– 129 –
We now write

A in the usual mode expansion,

A(x) =
_
d
3
p
(2π)
3
1
_
2[ p[
2

r=1
ǫ
r
( p)
_
a
r
p
e
i p·x
+ a
r †
p
e
−i p·x
_

E(x) =
_
d
3
p
(2π)
3
(−i)
_
[ p[
2
2

r=1
ǫ
r
( p)
_
a
r
p
e
i p·x
−a
r †
p
e
−i p·x
_
(6.28)
where, as before, the polarization vectors satisfy
ǫ
r
( p) p = 0 and ǫ
r
( p) ǫ
s
( p) = δ
rs
(6.29)
It is not hard to show that the commutation relations (6.25) are equivalent to the usual
commutation relations for the creation and annihilation operators,
[a
r
p
, a
s
q
] = [a
r †
p
, a
s †
q
] = 0
[a
r
p
, a
s †
q
] = (2π)
3
δ
rs
δ
(3)
( p −q) (6.30)
where, in deriving this, we need the completeness relation for the polarization vectors,
2

r=1
ǫ
i
r
( p)ǫ
j
r
( p) = δ
ij

p
i
p
j
[ p[
2
(6.31)
You can easily check that this equation is true by acting on both sides with a basis of
vectors (ǫ
1
( p), ǫ
2
( p), p).
We derive the Hamiltonian by substituting (6.28) into (6.17). The last term vanishes
in Coulomb gauge. After normal ordering, and playing around with ǫ
r
polarization
vectors, we get the simple expression
H =
_
d
3
p
(2π)
3
[ p[
2

r=1
a
r †
p
a
r
p
(6.32)
The Coulomb gauge has the advantage that the physical degrees of freedom are man-
ifest. However, we’ve lost all semblance of Lorentz invariance. One place where this
manifests itself is in the propagator for the fields A
i
(x) (in the Heisenberg picture). In
Coulomb gauge the propagator reads
D
tr
ij
(x −y) ≡ ¸0[ TA
i
(x)A
j
(y) [0) =
_
d
4
p
(2π)
4
i
p
2
+ iǫ
_
δ
ij

p
i
p
j
[ p[
2
_
e
−ip·(x−y)
(6.33)
The tr superscript on the propagator refers to the “transverse” part of the photon.
When we turn to the interacting theory, we will have to fight to massage this propagator
into something a little nicer.
– 130 –
6.2.2 Lorentz Gauge
We could try to work in a Lorentz invariant fashion by imposing the Lorentz gauge
condition ∂
µ
A
µ
= 0. The equations of motion that follow from the action are then

µ

µ
A
ν
= 0 (6.34)
Our approach to implementing Lorentz gauge will be a little different from the method
we used in Coulomb gauge. We choose to change the theory so that (6.34) arises directly
through the equations of motion. We can achieve this by taking the Lagrangian
L = −
1
4
F
µν
F
µν

1
2
(∂
µ
A
µ
)
2
(6.35)
The equations of motion coming from this action are

µ
F
µν
+ ∂
ν
(∂
µ
A
µ
) = ∂
µ

µ
A
ν
= 0 (6.36)
(In fact, we could be a little more general than this, and consider the Lagrangian
L = −
1
4
F
µν
F
µν

1

(∂
µ
A
µ
)
2
(6.37)
with arbitrary α and reach similar conclusions. The quantization of the theory is
independent of α and, rather confusingly, different choices of α are sometimes also
referred to as different “gauges”. We will use α = 1, which is called “Feynman gauge”.
The other common choice, α = 0, is called “Landau gauge”.)
Our plan will be to quantize the theory (6.36), and only later impose the constraint

µ
A
µ
= 0 in a suitable manner on the Hilbert space of the theory. As we’ll see, we will
also have to deal with the residual gauge symmetry of this theory which will prove a
little tricky. At first, we can proceed very easily, because both π
0
and π
i
are dynamical:
π
0
=
∂L

˙
A
0
= −∂
µ
A
µ
π
i
=
∂L

˙
A
i
= ∂
i
A
0

˙
A
i
(6.38)
Turning these classical fields into operators, we can simply impose the usual commu-
tation relations,
[A
µ
(x), A
ν
(y)] = [π
µ
(x), π
ν
(y)] = 0
[A
µ
(x), π
ν
(y)] = iη
µν
δ
(3)
(x −y) (6.39)
– 131 –
and we can make the usual expansion in terms of creation and annihilation operators
and 4 polarization vectors (ǫ
µ
)
λ
, with λ = 0, 1, 2, 3.
A
µ
(x) =
_
d
3
p
(2π)
3
1
_
2[ p[
3

λ=0
ǫ
λ
µ
( p)
_
a
λ
p
e
i p·x
+ a
λ†
p
e
−i p·x
_
π
µ
(x) =
_
d
3
p
(2π)
3
_
[ p[
2
(+i)
3

λ=0

µ
)
λ
( p)
_
a
λ
p
e
i p·x
−a
λ†
p
e
−i p·x
_
(6.40)
Note that the momentum π
µ
comes with a factor of (+i), rather than the familiar (−i)
that we’ve seen so far. This can be traced to the fact that the momentum (6.38) for the
classical fields takes the form π
µ
= −
˙
A
µ
+ . . .. In the Heisenberg picture, it becomes
clear that this descends to (+i) in the definition of momentum.
There are now four polarization 4-vectors ǫ
λ
( p), instead of the two polarization 3-
vectors that we met in the Coulomb gauge. Of these four 4-vectors, we pick ǫ
0
to be
timelike, while ǫ
1,2,3
are spacelike. We pick the normalization
ǫ
λ
ǫ
λ

= η
λλ

(6.41)
which also means that

µ
)
λ

ν
)
λ

η
λλ
′ = η
µν
(6.42)
The polarization vectors depend on the photon 4-momentum p = ([ p[, p). Of the two
spacelike polarizations, we will choose ǫ
1
and ǫ
2
to lie transverse to the momentum:
ǫ
1
p = ǫ
2
p = 0 (6.43)
The third vector ǫ
3
is the longitudinal polarization. For example, if the momentum lies
along the x
3
direction, so p ∼ (1, 0, 0, 1), then
ǫ
0
=
_
1
0
0
0
_
, ǫ
1
=
_
0
1
0
0
_
, ǫ
2
=
_
0
0
1
0
_
, ǫ
3
=
_
0
0
0
1
_
(6.44)
For other 4-momenta, the polarization vectors are the appropriate Lorentz transforma-
tions of these vectors, since (6.43) are Lorentz invariant.
We do our usual trick, and translate the field commutation relations (6.39) into those
for creation and annihilation operators. We find [a
λ
p
, a
λ

q
] = [a
λ†
p
, a
λ


q
] = 0 and
[a
λ
p
, a
λ


q
] = −η
λλ

(2π)
3
δ
(3)
( p −q) (6.45)
– 132 –
The minus signs here are odd to say the least! For spacelike λ = 1, 2, 3, everything
looks fine,
[a
λ
p
, a
λ


q
] = δ
λλ

(2π)
3
δ
(3)
( p −q) λ, λ

= 1, 2, 3 (6.46)
But for the timelike annihilation and creation operators, we have
[a
0
p
, a
0 †
q
] = −(2π)
3
δ
(3)
( p −q) (6.47)
This is very odd! To see just how strange this is, we take the Lorentz invariant vacuum
[0) defined by
a
λ
p
[0) = 0 (6.48)
Then we can create one-particle states in the usual way,
[ p, λ) = a
λ†
p
[0) (6.49)
For spacelike polarization states, λ = 1, 2, 3, all seems well. But for the timelike
polarization λ = 0, the state [ p, 0) has negative norm,
¸p, 0[ q, 0) = ¸0[ a
0
p
a
0 †
q
[0) = −(2π)
3
δ
(3)
( p −q) (6.50)
Wtf? That’s very very strange. A Hilbert space with negative norm means negative
probabilities which makes no sense at all. We can trace this negative norm back to the
wrong sign of the kinetic term for A
0
in our original Lagrangian: L = +
1
2
˙

A
2

1
2
˙
A
2
0
+. . ..
At this point we should remember our constraint equation, ∂
µ
A
µ
= 0, which, until
now, we’ve not imposed on our theory. This is going to come to our rescue. We will see
that it will remove the timelike, negative norm states, and cut the physical polarizations
down to two. We work in the Heisenberg picture, so that

µ
A
µ
= 0 (6.51)
makes sense as an operator equation. Then we could try implementing the constraint
in the quantum theory in a number of different ways. Let’s look at a number of
increasingly weak ways to do this
• We could ask that ∂
µ
A
µ
= 0 is imposed as an equation on operators. But this
can’t possibly work because the commutation relations (6.39) won’t be obeyed
for π
0
= −∂
µ
A
µ
. We need some weaker condition.
– 133 –
• We could try to impose the condition on the Hilbert space instead of directly
on the operators. After all, that’s where the trouble lies! We could imagine that
there’s some way to split the Hilbert space up into good states [Ψ) and bad states
that somehow decouple from the system. With luck, our bad states will include
the weird negative norm states that we’re so disgusted by. But how can we define
the good states? One idea is to impose

µ
A
µ
[Ψ) = 0 (6.52)
on all good, physical states [Ψ). But this can’t work either! Again, the condition
is too strong. For example, suppose we decompose A
µ
(x) = A
+
µ
(x) +A

µ
(x) with
A
+
µ
(x) =
_
d
3
p
(2π)
3
1
_
2[ p[
3

λ=0
ǫ
λ
µ
a
λ
p
e
−ip·x
A

µ
(x) =
_
d
3
p
(2π)
3
1
_
2[ p[
3

λ=0
ǫ
λ
µ
a
λ†
p
e
+ip·x
(6.53)
Then, on the vacuum A
+
µ
[0) = 0 automatically, but ∂
µ
A

µ
[0) , = 0. So not even
the vacuum is a physical state if we use (6.52) as our constraint
• Our final attempt will be the correct one. In order to keep the vacuum as a good
physical state, we can ask that physical states [Ψ) are defined by

µ
A
+
µ
[Ψ) = 0 (6.54)
This ensures that
¸Ψ

[ ∂
µ
A
µ
[Ψ) = 0 (6.55)
so that the operator ∂
µ
A
µ
has vanishing matrix elements between physical states.
Equation (6.54) is known as the Gupta-Bleuler condition. The linearity of the
constraint means that the physical states [Ψ) span a physical Hilbert space H
phys
.
So what does the physical Hilbert space H
phys
look like? And, in particular, have we
rid ourselves of those nasty negative norm states so that H
phys
has a positive definite
inner product defined on it? The answer is actually no, but almost!
Let’s consider a basis of states for the Fock space. We can decompose any element
of this basis as [Ψ) = [ψ
T
) [φ), where [ψ
T
) contains only transverse photons, created by
– 134 –
a
1,2 †
p
, while [φ) contains the timelike photons created by a
0 †
p
and longitudinal photons
created by a
3 †
p
. The Gupta-Bleuler condition (6.54) requires
(a
3
p
−a
0
p
) [φ) = 0 (6.56)
This means that the physical states must contain combinations of timelike and longi-
tudinal photons. Whenever the state contains a timelike photon of momentum p, it
must also contain a longitudinal photon with the same momentum. In general [φ) will
be a linear combination of states [φ
n
) containing n pairs of timelike and longitudinal
photons, which we can write as
[φ) =

n=0
C
n

n
) (6.57)
where [φ
0
) = [0) is simply the vacuum. It’s not hard to show that although the condition
(6.56) does indeed decouple the negative norm states, all the remaining states involving
timelike and longitudinal photons have zero norm
¸φ
m
[ φ
n
) = δ
n0
δ
m0
(6.58)
This means that the inner product on H
phys
is positive semi-definite. Which is an
improvement. But we still need to deal with all these zero norm states.
The way we cope with the zero norm states is to treat them as gauge equivalent
to the vacuum. Two states that differ only in their timelike and longitudinal photon
content, [φ
n
) with n ≥ 1 are said to be physically equivalent. We can think of the gauge
symmetry of the classical theory as descending to the Hilbert space of the quantum
theory. Of course, we can’t just stipulate that two states are physically identical unless
they give the same expectation value for all physical observables. We can check that
this is true for the Hamiltonian, which can be easily computed to be
H =
_
d
3
p
(2π)
3
[ p[
_
3

i=1
a
i †
p
a
i
p
−a
0 †
p
a
0
p
_
(6.59)
But the condition (6.56) ensures that ¸Ψ[ a
3 †
p
a
3
p
[Ψ) = ¸Ψ[ a
0 †
p
a
0
p
[Ψ) so that the contri-
butions from the timelike and longitudinal photons cancel amongst themselves in the
Hamiltonian. This also renders the Hamiltonian positive definite, leaving us just with
the contribution from the transverse photons as we would expect.
In general, one can show that the expectation values of all gauge invariant operators
evaluated on physical states are independent of the coefficients C
n
in (6.57).
– 135 –
Propagators
Finally, it’s a simple matter to compute the propagator in Lorentz gauge. It is given
by
¸0[ T A
µ
(x)A
ν
(y) [0) =
_
d
4
p
(2π)
4
−iη
µν
p
2
+ iǫ
e
−ip·(x−y)
(6.60)
This is a lot nicer than the propagator we found in Coulomb gauge: in particular, it’s
Lorentz invariant. We could also return to the Lagrangian (6.37). Had we pushed
through the calculation with arbitrary coefficient α, we would find the propagator,
¸0[ T A
µ
(x)A
ν
(y) [0) =
_
d
4
p
(2π)
4
−i
p
2
+ iǫ
_
η
µν
+ (α −1)
p
µ
p
ν
p
2
_
e
−ip·(x−y)
(6.61)
6.3 Coupling to Matter
Let’s now build an interacting theory of light and matter. We want to write down
a Lagrangian which couples A
µ
to some matter fields, either scalars or spinors. For
example, we could write something like
L = −
1
4
F
µν
F
µν
−j
µ
A
µ
(6.62)
where j
µ
is some function of the matter fields. The equations of motion read

µ
F
µν
= j
ν
(6.63)
so, for consistency, we require

µ
j
µ
= 0 (6.64)
In other words, j
µ
must be a conserved current. But we’ve got lots of those! Let’s look
at how we can couple two of them to electromagnetism.
6.3.1 Coupling to Fermions
The Dirac Lagrangian
L =
¯
ψ(i / ∂ −m)ψ (6.65)
has an internal symmetry ψ → e
−iα
ψ and
¯
ψ → e
+iα
¯
ψ, with α ∈ R. This gives rise to
the conserved current j
µ
V
=
¯
ψγ
µ
ψ. So we could look at the theory of electromagnetism
coupled to fermions, with the Lagrangian,
L = −
1
4
F
µν
F
µν
+
¯
ψ(i / ∂ −m)ψ −e
¯
ψγ
µ
A
µ
ψ (6.66)
– 136 –
where we’ve introduced a coupling constant e. For the free Maxwell theory, we have
seen that the existence of a gauge symmetry was crucial in order to cut down the
physical degrees of freedom to the requisite 2. Does our interacting theory above still
have a gauge symmetry? The answer is yes. To see this, let’s rewrite the Lagrangian
as
L = −
1
4
F
µν
F
µν
+
¯
ψ(i / D −m)ψ (6.67)
where D
µ
ψ = ∂
µ
ψ + ieA
µ
ψ is called the covariant derivative. This Lagrangian is
invariant under gauge transformations which act as
A
µ
→A
µ
+ ∂
µ
λ and ψ →e
−ieλ
ψ (6.68)
for an arbitrary function λ(x). The tricky term is the derivative acting on ψ, since this
will also hit the e
−ieλ
piece after the transformation. To see that all is well, let’s look
at how the covariant derivative transforms. We have
D
µ
ψ = ∂
µ
ψ + ieA
µ
ψ
→ ∂
µ
(e
−ieλ
ψ) + ie(A
µ
+ ∂
µ
λ)(e
−ieλ
ψ)
= e
−ieλ
D
µ
ψ (6.69)
so the covariant derivative has the nice property that it merely picks up a phase under
the gauge transformation, with the derivative of e
−ieλ
cancelling the transformation
of the gauge field. This ensures that the whole Lagrangian is invariant, since
¯
ψ →
e
+ieλ(x)
¯
ψ.
Electric Charge
The coupling e has the interpretation of the electric charge of the ψ particle. This
follows from the equations of motion of classical electromagnetism ∂
µ
F
µν
= j
ν
: we
know that the j
0
component is the charge density. We therefore have the total charge
Q given by
Q = e
_
d
3
x
¯
ψ(x)γ
0
ψ(x) (6.70)
After treating this as a quantum equation, we have
Q = e
_
d
3
p
(2π)
3
2

s=1
(b
s †
p
b
s
p
−c
s †
p
c
s
p
) (6.71)
which is the number of particles, minus the number of antiparticles. Note that the
particle and the anti-particle are required by the formalism to have opposite electric
– 137 –
charge. For QED, the theory of light interacting with electrons, the electric charge
is usually written in terms of the dimensionless ratio α, known as the fine structure
constant
α =
e
2
4πc

1
137
(6.72)
Setting = c = 1, we have e =

4πα ≈ 0.3.
There’s a small subtlety here that’s worth elaborating on. I stressed that there’s a
radical difference between the interpretation of a global symmetry and a gauge symme-
try. The former takes you from one physical state to another with the same properties
and results in a conserved current through Noether’s theorem. The latter is a redun-
dancy in our description of the system. Yet in electromagnetism, the gauge symmetry
ψ →e
+ieλ(x)
ψ seems to lead to a conservation law, namely the conservation of electric
charge. This is because among the infinite number of gauge symmetries parameterized
by a function λ(x), there is also a single global symmetry: that with λ(x) = constant.
This is a true symmetry of the system, meaning that it takes us to another physical
state. More generally, the subset of global symmetries from among the gauge symme-
tries are those for which λ(x) → α = constant as x → ∞. These take us from one
physical state to another.
Finally, let’s check that the 4 4 matrix C that we introduced in Section 4.5 really
deserves the name “charge conjugation matrix”. If we take the complex conjugation of
the Dirac equation, we have
(iγ
µ

µ
−eγ
µ
A
µ
−m)ψ = 0 ⇒ (−i(γ
µ
)


µ
−e(γ
µ
)

A
µ
−m)ψ

= 0
Now using the defining equation C

γ
µ
C = −(γ
µ
)

, and the definition ψ
(c)
= Cψ

, we
see that the charge conjugate spinor ψ
(c)
satisfies
(iγ
µ

µ
+ eγ
µ
A
µ
−m)ψ
(c)
= 0 (6.73)
So we see that the charge conjugate spinor ψ
(c)
satisfies the Dirac equation, but with
charge −e instead of +e.
6.3.2 Coupling to Scalars
For a real scalar field, we have no suitable conserved current. This means that we can’t
couple a real scalar field to a gauge field.
– 138 –
Let’s now consider a complex scalar field ϕ. (For this section, I’ll depart from our
previous notation and call the scalar field ϕ to avoid confusing it with the spinor). We
have a symmetry ϕ → e
−iα
ϕ. We could try to couple the associated current to the
gauge field,
L
int
= −i((∂
µ
ϕ

)ϕ −ϕ


µ
ϕ)A
µ
(6.74)
But this doesn’t work because
• The theory is no longer gauge invariant
• The current j
µ
that we coupled to A
µ
depends on ∂
µ
ϕ. This means that if we
try to compute the current associated to the symmetry, it will now pick up a
contribution from the j
µ
A
µ
term. So the whole procedure wasn’t consistent.
We solve both of these problems simultaneously by remembering the covariant deriva-
tive. In this scalar theory, the combination
T
µ
ϕ = ∂
µ
ϕ + ieA
µ
ϕ (6.75)
again transforms as T
µ
ϕ → e
−ieλ
T
µ
ϕ under a gauge transformation A
µ
→ A
µ
+ ∂
µ
λ
and ϕ → e
−ieλ
ϕ. This means that we can construct a gauge invariant action for a
charged scalar field coupled to a photon simply by promoting all derivatives to covariant
derivatives
L = −
1
4
F
µν
F
µν
+T
µ
ϕ

T
µ
ϕ −m
2
[ϕ[
2
(6.76)
In general, this trick works for any theory. If we have a U(1) symmetry that we wish to
couple to a gauge field, we may do so by replacing all derivatives by suitable covariant
derivatives. This procedure is known as minimal coupling.
6.4 QED
Let’s now work out the Feynman rules for the full theory of quantum electrodynamics
(QED) – the theory of electrons interacting with light. The Lagrangian is
L = −
1
4
F
µν
F
µν
+
¯
ψ(i / D −m)ψ (6.77)
where D
µ
= ∂
µ
+ ieA
µ
.
The route we take now depends on the gauge choice. If we worked in Lorentz gauge
previously, then we can jump straight to Section 6.5 where the Feynman rules for QED
are written down. If, however, we worked in Coulomb gauge, then we still have a bit of
work in front of us in order to massage the photon propagator into something Lorentz
invariant. We will now do that.
– 139 –
In Coulomb gauge ∇

A = 0, the equation of motion arising from varying A
0
is now
−∇
2
A
0
= eψ

ψ ≡ ej
0
(6.78)
which has the solution
A
0
(x, t) = e
_
d
3
x

j
0
(x

, t)
4π[x −x

[
(6.79)
In Coulomb gauge we can rewrite the Maxwell part of the Lagrangian as
L
Maxwell
=
_
d
3
x
1
2

E
2

1
2

B
2
=
_
d
3
x
1
2
(
˙

A +∇A
0
)
2

1
2

B
2
=
_
d
3
x
1
2
˙

A
2
+
1
2
(∇A
0
)
2

1
2

B
2
(6.80)
where the cross-term has vanished using ∇

A = 0. After integrating the second term
by parts and inserting the equation for A
0
, we have
L
Maxwell
=
_
d
3
x
_
1
2
˙

A
2

1
2

B
2

e
2
2
_
d
3
x

j
0
(x)j
0
(x

)
4π[x −x

[
_
(6.81)
We find ourselves with a nonlocal term in the action. This is exactly the type of
interaction that we boasted in Section 1.1.4 never arises in Nature! It appears here as
an artifact of working in Coulomb gauge: it does not mean that the theory of QED is
nonlocal. For example, it wouldn’t appear if we worked in Lorentz gauge.
We now compute the Hamiltonian. Changing notation slightly from previous chap-
ters, we have the conjugate momenta,

Π =
∂L

˙

A
=
˙

A
π
ψ
=
∂L

˙
ψ
= iψ

(6.82)
which gives us the Hamiltonian
H =
_
d
3
x
_
1
2
˙

A
2
+
1
2

B
2
+
¯
ψ(−iγ
i

i
+ m)ψ −e

j

A+
e
2
2
_
d
3
x

j
0
(x)j
0
(x

)
4π[x −x

[
_
where

j =
¯
ψγψ and j
0
=
¯
ψγ
0
ψ.
– 140 –
6.4.1 Naive Feynman Rules
We want to determine the Feynman rules for this theory. For fermions, the rules are
the same as those given in Section 5. The new pieces are:
• We denote the photon by a wavy line. Each end of the line comes with an i, j =
1, 2, 3 index telling us the component of

A. We calculated the transverse photon
propagator in (6.33): it is and contributes D
tr
ij
=
i
p
2
+ iǫ
_
δ
ij

p
i
p
j
[ p[
2
_
• The vertex contributes −ieγ
i
. The index on γ
i
contracts with the
index on the photon line.
• The non-local interaction which, in position space, is given by
x y
contributes a factor of
i(eγ
0
)
2
δ(x
0
−y
0
)
4π[x −y[
These Feynman rules are rather messy. This is the price we’ve paid for working in
Coulomb gauge. We’ll now show that we can massage these expressions into something
much more simple and Lorentz invariant. Let’s start with the offending instantaneous
interaction. Since it comes from the A
0
component of the gauge field, we could try to
redefine the propagator to include a D
00
piece which will capture this term. In fact, it
fits quite nicely in this form: if we look in momentum space, we have
δ(x
0
−y
0
)
4π[x −y[
=
_
d
4
p
(2π)
4
e
ip·(x−y)
[ p[
2
(6.83)
so we can combine the non-local interaction with the transverse photon propagator by
defining a new photon propagator
D
µν
(p) =
_
¸
¸
¸
¸
_
¸
¸
¸
¸
_
+
i
[ p[
2
µ, ν = 0
i
p
2
+ iǫ
_
δ
ij

p
i
p
j
[ p[
2
_
µ = i ,= 0, ν = j ,= 0
0 otherwise
(6.84)
With this propagator, the wavy photon line now carries a µ, ν = 0, 1, 2, 3 index, with
the extra µ = 0 component taking care of the instantaneous interaction. We now need
to change our vertex slightly: the −ieγ
i
above gets replaced by −ieγ
µ
which correctly
accounts for the (eγ
0
)
2
piece in the instantaneous interaction.
– 141 –
The D
00
piece of the propagator doesn’t look a whole lot different from the transverse
photon propagator. But wouldn’t it be nice if they were both part of something more
symmetric! In fact, they are. We have the following:
Claim: We can replace the propagator D
µν
(p) with the simpler, Lorentz invariant
propagator
D
µν
(p) = −i
η
µν
p
2
(6.85)
Proof: There is a general proof using current conservation. Here we’ll be more pedes-
trian and show that we can do this for certain Feynman diagrams. In particular, we
focus on a particular tree-level diagram that contributes to e

e

→e

e

scattering,
/
q
p
/
p
q
∼ e
2
[¯ u(p


µ
u(p)] D
µν
(k) [¯ u(q


ν
u(q)] (6.86)
where k = p −p

= q

−q. Recall that u( p) satisfies the equation
( / p −m)u( p) = 0 (6.87)
Let’s define the spinor contractions α
µ
= ¯ u( p


µ
u( p) and β
ν
= ¯ u(q


ν
u(q). Then
since k = p −p

= q

−q, we have
k
µ
α
µ
= ¯ u( p

)( / p − / p

)u( p) = ¯ u( p

)(m−m)u( p) = 0 (6.88)
and, similarly, k
ν
β
ν
= 0. Using this fact, the diagram can be written as
α
µ
D
µν
β
ν
= i
_
α

β
k
2

( α

k)(

β

k)
k
2
[

k[
2
+
α
0
β
0
[

k[
2
_
= i
_
α

β
k
2

k
2
0
α
0
β
0
k
2
[

k[
2
+
α
0
β
0
[

k[
2
_
= −
_
α

β
k
2

1
k
2
[

k[
2
(k
2
0
−k
2
) α
0
β
0
_
= −
i
k
2
α β = α
µ
_


µν
k
2
_
β
ν
(6.89)
– 142 –
which is the claimed result. You can similarly check that the same substitution is legal
in the diagram
p
q
p
q
/
/
∼ e
2
[¯ v(q)γ
µ
u( p)]D
µν
(k)[¯ u( p


ν
v(q

)] (6.90)
In fact, although we won’t show it here, it’s a general fact that in every Feynman dia-
gram we may use the very nice, Lorentz invariant propagator D
µν
= −iη
µν
/p
2
.
Note: This is the propagator we found when quantizing in Lorentz gauge (using the
Feynman gauge parameter). In general, quantizing the Lagrangian (6.37) in Lorentz
gauge, we have the propagator
D
µν
= −
i
p
2
_
η
µν
+ (α −1)
p
µ
p
ν
p
2
_
(6.91)
Using similar arguments to those given above, you can show that the p
µ
p
ν
/p
2
term
cancels in all diagrams. For example, in the following diagrams the p
µ
p
ν
piece of the
propagator contributes as
∼ ¯ u(p


µ
u(p) k
µ
= ¯ u(p

)( / p − / p

)u(p) = 0
∼ ¯ v(p)γ
µ
u(q) k
µ
= ¯ u(p)( / p + / q

)u(q) = 0 (6.92)
6.5 Feynman Rules
Finally, we have the Feynman rules for QED. For vertices and internal lines, we write
• Vertex: −ieγ
µ
• Photon Propagator: −

µν
p
2
+ iǫ
• Fermion Propagator:
i( / p + m)
p
2
−m
2
+ iǫ
For external lines in the diagram, we attach
• Photons: We add a polarization vector ǫ
µ
in

µ
out
for incoming/outgoing photons.
In Coulomb gauge, ǫ
0
= 0 and ǫ p = 0.
• Fermions: We add a spinor u
r
( p)/¯ u
r
( p) for incoming/outgoing fermions. We add
a spinor ¯ v
r
( p)/v
r
( p) for incoming/outgoing anti-fermions.
– 143 –
6.5.1 Charged Scalars
“Pauli asked me to calculate the cross section for pair creation of scalar
particles by photons. It was only a short time after Bethe and Heitler
had solved the same problem for electrons and positrons. I met Bethe
in Copenhagen at a conference and asked him to tell me how he did the
calculations. I also inquired how long it would take to perform this task;
he answered, “It would take me three days, but you will need about three
weeks.” He was right, as usual; furthermore, the published cross sections
were wrong by a factor of four.”
Viki Weisskopf
The interaction terms in the Lagrangian for charged scalars come from the covariant
derivative terms,
L = T
µ
ψ

T
µ
ψ = ∂
µ
ψ


µ
ψ −ieA
µ



µ
ψ −ψ∂
µ
ψ

) + e
2
A
µ
A
µ
ψ

ψ (6.93)
This gives rise to two interaction vertices. But the cubic vertex is something we haven’t
seen before: it contains kinetic terms. How do these appear in the Feynman rules?
After a Fourier transform, the derivative term means that the interaction is stronger
for fermions with higher momentum, so we include a momentum factor in the Feynman
rule. There is also a second, “seagull” graph. The two Feynman rules are
p
q
−ie(p + q)
µ
and + 2ie
2
η
µν
The factor of two in the seagull diagram arises because of the two identical particles
appearing in the vertex. (It’s the same reason that the 1/4! didn’t appear in the
Feynman rules for φ
4
theory).
6.6 Scattering in QED
Let’s now calculate some amplitudes for various processes in quantum electrodynamics,
with a photon coupled to a single fermion. We will consider the analogous set of
processes that we saw in Section 3 and Section 5. We have
– 144 –
Electron Scattering
Electron scattering e

e

→ e

e

is described by the two leading order Feynman dia-
grams, given by
/ /
q,r
p,s
p,s
/
q,r
/
+
p,s
/ /
/ /
q,r
q,r
p,s
= −i(−ie)
2
_
[¯ u
s

( p


µ
u
s
( p)] [¯ u
r

(q


µ
u
r
(q)]
(p

−p)
2

[¯ u
s

( p


µ
u
r
(q)] [¯ u
r

(q


µ
u
s
( p)]
(p −q

)
2
_
The overall −i comes from the −iη
µν
in the propagator, which contract the indices on
the γ-matrices (remember that it’s really positive for µ, ν = 1, 2, 3).
Electron Positron Annihilation
Let’s now look at e

e
+
→ 2γ, two gamma rays. The two lowest order Feynman
diagrams are,
p
/
/
q
q,r
p,s
ε
ε
µ
2
ν
1
+
ε1
ν
p
/
ε
µ
2
/
q
q,r
p,s
= i(−ie)
2
¯ v
r
(q)
_
γ
µ
( / p − / p

+ m)γ
ν
(p −p

)
2
−m
2
+
γ
ν
( / p − / q

+ m)γ
µ
(p −q

)
2
−m
2
_
u
s
( p)ǫ
ν
1
( p


µ
2
(q

)
Electron Positron Scattering
For e

e
+
→ e

e
+
scattering (sometimes known as Bhabha scattering) the two lowest
order Feynman diagrams are
/ /
q,r
p,s
p,s
/
q,r
/
+
p,s
/ /
/ /
q,r
p,s
q,r
= −i(−ie)
2
_

[¯ u
s

( p


µ
u
s
( p)] [¯ v
r
(q)γ
µ
v
r

(q

)]
(p −p

)
2
+
[¯ v
r
(q)γ
µ
u
s
( p)] [¯ u
s

( p


µ
v
r

(q

)]
(p + q)
2
_
Compton Scattering
The scattering of photons (in particular x-rays) off electrons e

γ → e

γ is known
as Compton scattering. Historically, the change in wavelength of the photon in the
– 145 –
scattering process was one of the conclusive pieces of evidence that light could behave
as a particle. The amplitude is given by,
out
ε
/
q
ε
in
q ( ) (
)
u(p) u(p)
/ −
+
ε
in
q) ( out
ε
/
q
( )
u(p) u(p)
/ −
= i(−ie)
2
¯ u
r

( p

)
_
γ
µ
( / p + / q + m)γ
ν
(p + q)
2
−m
2
+
γ
ν
( / p − / q

+ m)γ
µ
(p −q

)
2
−m
2
_
u
s
( p) ǫ
ν
in
ǫ
µ
out
This amplitude vanishes for longitudinal photons. For example, suppose ǫ
in
∼ q. Then,
using momentum conservation p + q = p

+ q

, we may write the amplitude as
i/ = i(−ie)
2
¯ u
r

( p

)
_
/ ǫ
out
( / p + / q + m) / q
(p + q)
2
−m
2
+
/ q( / p

− / q + m) / ǫ
out
(p

−q)
2
−m
2
_
u
s
( p)
= i(−ie)
2
¯ u
r

( p

) / ǫ
out
u
s
( p)
_
2p q
(p + q)
2
−m
2
+
2p

q
(p

−q)
2
−m
2
_
(6.94)
where, in going to the second line, we’ve performed some γ-matrix manipulations,
and invoked the fact that q is null, so / q / q = 0, together with the spinor equations
( / p−m)u( p) and ¯ u( p

)( / p

−m) = 0. We now recall the fact that q is a null vector, while
p
2
= (p

)
2
= m
2
since the external legs are on mass-shell. This means that the two
denominators in the amplitude read (p+q)
2
−m
2
= 2pq and (p

−q)−m
2
= −2p

q. This
ensures that the combined amplitude vanishes for longitudinal photons as promised. A
similar result holds when / ǫ
out
∼ q

.
Photon Scattering
In QED, photons no longer pass through each other unimpeded.
Figure 30:
At one-loop, there is a diagram which leads to photon scattering.
Although naively logarithmically divergent, the diagram is actually
rendered finite by gauge invariance.
Adding Muons
Adding a second fermion into the mix, which we could identify as a
muon, new processes become possible. For example, we can now have processes such
as e

µ

→e

µ

scattering, and e
+
e

annihilation into a muon anti-muon pair. Using
our standard notation of p and q for incoming momenta, and p

and q

for outgoing
– 146 –
momenta, we have the amplitudes given by
µ

µ

e

e


1
(p −p

)
2
and
µ
µ
e
e

+ +


1
(p + q)
2
(6.95)
6.6.1 The Coulomb Potential
We’ve come a long way. We’ve understood how to compute quantum amplitudes in
a large array of field theories. To end this course, we use our newfound knowledge to
rederive a result you learnt in kindergarten: Coulomb’s law.
To do this, we repeat our calculation that led us to the Yukawa force in Sections
3.5.2 and 5.7.2. We start by looking at e

e

→e

e

scattering. We have
/ /
q,r
p,s
p,s
/
q,r
/
= −i(−ie)
2
[¯ u( p


µ
u( p)] [¯ u(q


µ
u(q)]
(p

−p)
2
(6.96)
Following (5.49), the non-relativistic limit of the spinor is u(p) →

m
_
ξ
ξ
_
. This
means that the γ
0
piece of the interaction gives a term ¯ u
s
( p)γ
0
u
r
(q) → 2mδ
rs
, while
the spatial γ
i
, i = 1, 2, 3 pieces vanish in the non-relativistic limit: ¯ u
s
( p)γ
i
u
r
(q) → 0.
Comparing the scattering amplitude in this limit to that of non-relativistic quantum
mechanics, we have the effective potential between two electrons given by,
U(r) = +e
2
_
d
3
p
(2π)
3
e
i p·r
[ p[
2
= +
e
2
4πr
(6.97)
We find the familiar repulsive Coulomb potential. We can trace the minus sign that
gives a repulsive potential to the fact that only the A
0
component of the intermediate
propagator ∼ −iη
µν
contributes in the non-relativistic limit.
For e

e
+
→e

e
+
scattering, the amplitude is
/ /
q,r
p,s
p,s
/
q,r
/
= +i(−ie)
2
[¯ u( p


µ
u( p)] [¯ v(q


µ
v(q)]
(p

−p)
2
(6.98)
– 147 –
The overall + sign comes from treating the fermions correctly: we saw the same minus
sign when studying scattering in Yukawa theory. The difference now comes from looking
at the non-relativistic limit. We have ¯ vγ
0
v → 2m, giving us the potential between
opposite charges,
U(r) = −e
2
_
d
3
p
(2π)
3
e
i p·r
[ p[
2
= −
e
2
4πr
(6.99)
Reassuringly, we find an attractive force between an electron and positron. The differ-
ence from the calculation of the Yukawa force comes again from the zeroth component
of the gauge field, this time in the guise of the γ
0
sandwiched between ¯ vγ
0
v → 2m,
rather than the ¯ vv →−2m that we saw in the Yukawa case.
The Coulomb Potential for Scalars
There are many minus signs in the above calculation which somewhat obscure the
crucial one which gives rise to the repulsive force. A careful study reveals the offending
sign to be that which sits in front of the A
0
piece of the photon propagator −iη
µν
/p
2
.
Note that with our signature (+−−−), the propagating A
i
have the correct sign, while
A
0
comes with the wrong sign. This is simpler to see in the case of scalar QED, where
we don’t have to worry about the gamma matrices. From the Feynman rules of Section
6.5.1, we have the non-relativistic limit of scalar e

e

scattering,
/ /
q,r
p,s
p,s
/
q,r
/
= −iη
µν
(−ie)
2
(p + p

)
µ
(q + q

)
ν
(p

−p)
2
→ −i(−ie)
2
(2m)
2
−( p −p

)
2
where the non-relativistic limit in the numerator involves (p+p

) (q+q

) ≈ (p+p

)
0
(q+
q

)
0
≈ (2m)
2
and is responsible for selecting the A
0
part of the photon propagator rather
than the A
i
piece. This shows that the Coulomb potential for spin 0 particles of the
same charge is again repulsive, just as it is for fermions. For e

e
+
scattering, the
amplitude picks up an extra minus sign because the arrows on the legs of the Feynman
rules in Section 6.5.1 are correlated with the momentum arrows. Flipping the arrows
on one pair of legs in the amplitude introduces the relevant minus sign to ensure that
the non-relativistic potential between e

e
+
is attractive as expected.
– 148 –
6.7 Afterword
In this course, we have laid the foundational framework for quantum field theory. Most
of the developments that we’ve seen were already in place by the middle of the 1930s,
pioneered by people such as Jordan, Dirac, Heisenberg, Pauli and Weisskopf
5
.
Yet by the end of the 1930s, physicists were ready to give up on quantum field theory.
The difficulty lies in the next terms in perturbation theory. These are the terms that
correspond to Feynamn diagrams with loops in them, which we have scrupulously
avoided computing in this course. The reason we’ve avoided them is because they
typically give infinity! And, after ten years of trying, and failing, to make sense of this,
the general feeling was that one should do something else. This from Dirac in 1937,
Because of its extreme complexity, most physicists will be glad to see the
end of QED
But the leading figures of the day gave up too soon. It took a new generation of postwar
physicists — people like Schwinger, Feynman, Tomonaga and Dyson — to return to
quantum field theory and tame the infinities. The story of how they did that will be
told in next term’s course.
5
For more details on the history of quantum field theory, see the excellent book “QED and the
Men who Made it” by Sam Schweber.
– 149 –

Recommended Books and Resources

• M. Peskin and D. Schroeder, An Introduction to Quantum Field Theory This is a very clear and comprehensive book, covering everything in this course at the right level. It will also cover everything in the “Advanced Quantum Field Theory” course, much of the “Standard Model” course, and will serve you well if you go on to do research. To a large extent, our course will follow the first section of this book.

There is a vast array of further Quantum Field Theory texts, many of them with redeeming features. Here I mention a few very different ones. • S. Weinberg, The Quantum Theory of Fields, Vol 1 This is the first in a three volume series by one of the masters of quantum field theory. It takes a unique route to through the subject, focussing initially on particles rather than fields. The second volume covers material lectured in “AQFT”. • L. Ryder, Quantum Field Theory This elementary text has a nice discussion of much of the material in this course. • A. Zee, Quantum Field Theory in a Nutshell This is charming book, where emphasis is placed on physical understanding and the author isn’t afraid to hide the ugly truth when necessary. It contains many gems. • M Srednicki, Quantum Field Theory A very clear and well written introduction to the subject. Both this book and Zee’s focus on the path integral approach, rather than canonical quantization that we develop in this course.

There are also resources available on the web. Some particularly good ones are listed on the course webpage: http://www.damtp.cam.ac.uk/user/tong/qft.html

Contents
0. Introduction 0.1 Units and Scales 1. Classical Field Theory 1.1 The Dynamics of Fields 1.1.1 An Example: The Klein-Gordon Equation 1.1.2 Another Example: First Order Lagrangians 1.1.3 A Final Example: Maxwell’s Equations 1.1.4 Locality, Locality, Locality 1.2 Lorentz Invariance 1.3 Symmetries 1.3.1 Noether’s Theorem 1.3.2 An Example: Translations and the Energy-Momentum Tensor 1.3.3 Another Example: Lorentz Transformations and Angular Momentum 1.3.4 Internal Symmetries 1.4 The Hamiltonian Formalism 2. Free Fields 2.1 Canonical Quantization 2.1.1 The Simple Harmonic Oscillator 2.2 The Free Scalar Field 2.3 The Vacuum 2.3.1 The Cosmological Constant 2.3.2 The Casimir Effect 2.4 Particles 2.4.1 Relativistic Normalization 2.5 Complex Scalar Fields 2.6 The Heisenberg Picture 2.6.1 Causality 2.7 Propagators 2.7.1 The Feynman Propagator 2.7.2 Green’s Functions 2.8 Non-Relativistic Fields 2.8.1 Recovering Quantum Mechanics 1 4 7 7 8 9 10 10 11 13 13 14 16 18 19 21 21 22 23 25 26 27 29 31 33 35 36 38 38 40 41 43

–1–

5.1 Dyson’s Formula 3.6.2 A First Look at Scattering 3.1 The Spinor Representation 4.3 An Example: Nucleon Scattering 3.3 The Dirac Equation 4.6.4 Feynman Diagrams 3.1 Some Examples 47 50 51 53 55 56 56 58 58 60 61 62 66 67 69 70 71 71 73 74 75 77 79 81 83 85 87 90 91 91 93 94 95 96 98 100 102 –2– .1 Mandelstam Variables 3.3 φ4 Theory 3. Interacting Fields 3.7.2 From Green’s Functions to S-Matrices 4.4 Connected Diagrams and Amputated Diagrams 3.2 Wick’s Theorem 3.2 Decay Rates 3.1 Spinors 4.1 Feynman Rules 3.4 Chiral Spinors 4.2 The Yukawa Potential 3.6 Symmetries and Conserved Currents 4.4.4.5 Examples of Scattering Amplitudes 3.3.3.4.2 γ 5 4.1 The Weyl Equation 4.1.4.6 What We Measure: Cross Sections and Decay Rates 3.1 An Example: Meson Decay 3.3 Wick’s Theorem 3.4.7 Green’s Functions 3.3.7.3 Cross Sections 3.1 Connected Diagrams and Vacuum Bubbles 3.6.3 Parity 4.5.4 Chiral Interactions 4.2 Constructing an Action 4.1 Fermi’s Golden Rule 3.5.2.1 The Interaction Picture 3.7.5.5 Majorana Fermions 4.3.1 An Example: Recovering the Propagator 3.7 Plane Wave Solutions 4.1. The Dirac Equation 4.

5 Feynman Rules 6. Quantizing the Dirac Field 5.1 Examples 5.2 Coupling to Scalars 6.7. Quantum Electrodynamics 6.2.1.1 A Glimpse at the Spin-Statistics Theorem 5.1 An Example: Putting Spin on Nucleon Scattering 5.3 Dirac’s Hole Interpretation 5.3.1 Naive Feynman Rules 6.1 Coulomb Gauge 6.6 Yukawa Theory 5.2.5 The Feynman Propagator 5.1 Gauge Symmetry 6.2 Lorentz Gauge 6.4 QED 6.1.1 The Hamiltonian 5.2 The Quantization of the Electromagnetic Field 6.6.1 Maxwell’s Equations 6.7 Afterword –3– .2 Some Useful Formulae: Inner and Outer Products 103 106 106 107 109 110 110 112 114 115 115 117 118 121 122 124 124 125 128 128 131 136 136 138 139 141 143 144 144 147 149 5.6.7.4 Propagators 5.7.3 Coupling to Matter 6.1 Coupling to Fermions 6.1 Charged Scalars 6.4.2 Fermionic Quantization 5.3 Pseudo-Scalar Coupling 6.1 The Coulomb Potential 6.5.7 Feynman Rules for Fermions 5.2.6 Scattering in QED 6.7.1 Fermi-Dirac Statistics 5.2 The Yukawa Potential Revisited 5.3.4.

I inherited the course from Nick Manton. I am supported by the Royal Society. Hugh Osborn and Jenni Smillie. for example the discussion of scalar Yukawa theory.Acknowledgements These lecture notes are far from original. whose notes form the backbone of the lectures. My primary contribution has been to borrow. My thanks to the many who helped in various ways during the preparation of this course. Ian Drummond. Eyo Ita. –4– . Ron Horgan. Nick Manton. In several places. Jerome Gauntlett. most notably the book by Peskin and Schroeder. I followed the lectures of Sidney Coleman. I have also relied heavily on the sources listed at the beginning. Matt Headrick. including Joe Conlon. steal and assimilate the best discussions and explanations I could find from the vast literature on the subject. Nick Dorey. using the notes written by Brian Hill and a beautiful abridged version of these notes due to Michael Luke. My thanks also to the students for their sharp questions and sharp eyes in spotting typos. Marie Ericsson.

Electrons and other matter particles are postulated to be elementary constituents of Nature. We will show how photons arise from the quantization of the electromagnetic field and how massive. analogous to the electromagnetic field? The purpose of this course is to answer these questions. with the photon appearing only when we correctly treat the field in a manner consistent with quantum theory? And. In contrast. Why Quantum Field Theory? In classical physics.” Viki Weisskopf The concept of wave-particle duality tells us that the properties of electrons and photons are fundamentally very similar. The old laws of Coulomb and Newton involve “action at a distance”. the primary reason for introducing the concept of the field is to construct laws of Nature that are local. how should we reconcile this difference in the quantum world? Should we view the particle as fundamental. appearing only after quantization. we must not only introduce electron fields. Yet the appearance of these objects in classical physics is very different. gluon fields. if this latter view is correct. There is a field associated to each type of fundamental particle that appears in Nature. We shall see that the second viewpoint above is the most useful: the field is primary and particles are derived concepts. This means that the force felt by an electron (or planet) changes –1– . Maxwell and other classical physicists find it useful to introduce the concept of matter fields. W and Z-boson fields. The existence of virtual pairs and of pair fluctuations shows that the days of fixed particle numbers are over.0. not even few-particle systems. neutrino fields. If photons and particles are truely to be placed on equal footing. but also quark fields. Despite obvious differences in their mass and charge. should we also introduce an “electron field”. Higgs fields and a whole slew of others. light is a derived concept: it arises as a ripple of the electromagnetic field. We will learn that in order to describe the fundamental laws of Nature. under the right circumstances both suffer wave-like diffraction and both can pack a particle-like punch. charged particles such as electrons arise from the quantization of matter fields. Introduction “There are no real one-particle systems in nature. whose ripples give rise to particles with mass and charge? But why then didn’t Faraday. with the electromagnetic field arising only in some classical limit from a collection of quantum photons? Or should we instead view the field as fundamental.

For now. momentum and energy are on an equivalent footing.000 particles. Heisenberg tells us that the uncertainty in the momentum is ∆p ≥ /L. Particles are not indestructible objects. http://arxiv. Figure 1: We will review Dirac’s argument for anti-particles later in this course. More importantly. made at the beginning of the universe and here for good. We learn that particle-anti-particle pairs are expected to be important when a particle of mass m is localized within a distance of order λ= 1 mc A concise review of the underlying principles and major successes of quantum field theory can be found in the article by Frank Wilczek. However. mostly ephemeral and fleeting. with all interactions mediated in a local fashion by the field. In a relativistic setting. However. The field theories of Maxwell and Einstein remedy the situation. This experimentally verified fact was first predicted by Dirac who understood how relativity implies the necessity of anti-particles.org/abs/hep-th/9803075 –2– . when the uncertainty in the energy exceeds ∆E = 2mc2 . They can be created and destroyed. so we should also have an uncertainty in the energy of order ∆E ≥ c/L. An extreme demonstration of particle creation is shown in the picture. then we cross the barrier to pop particle anti-particle pairs out of the vacuum. in fact. The requirement of locality remains a strong motivation for studying field theories in the quantum world. captured here in all their beauty by the STAR detector. This machine crashes gold nuclei together. each containing 197 nucleons. The resulting explosion contains up to 10. Consider a particle of mass m trapped in a box of size L. Here I’ll give two answers to the question: Why quantum field theory? Answer 1: Because the combination of quantum mechanics and special relativity implies that particle number is not conserved. together with the better understanding that we get from viewing particles in the framework of quantum field theory.immediately if a distant proton (or star) moves. it is also experimentally wrong. there are further reasons for treating the quantum field as fundamental1 . which comes from the Relativistic Heavy Ion Collider (RHIC) at Brookhaven. This situation is philosophically unsatisfactory. They are. Long Island. we’ll quickly sketch the circumstances in which we expect the number of particles to change.

But it’s not! What I mean by this is that two electrons are identical in every way. We compare this proton with one freshly minted in a particle accelerator here on Earth. or a breakdown in causality are the common issues that arise). there is a high probability that we will detect particleanti-particle pairs swarming around the original particle that we put in. the Compton wavelength is the distance at which the concept of a single pointlike particle breaks down completely. an equation for any fixed number of particles) is doomed to failure. the quark field. It turns out that this is roughly what happens. regardless of where they came from and what they’ve been through. Swapping two particles around leaves the state completely unchanged — apart from a possible minus sign. be identical in all respects? One explanation that might be offered is that there’s a sea of proton “stuff” filling the universe and when we make a proton we somehow dip our hand into this stuff and from it mould a proton. In each case. Indeed. The distance λ is called the Compton wavelength. Then it’s not surprising that protons produced in different parts of the universe are identical: they’re made of the same stuff. The presence of a multitude of particles and antiparticles at short distances tells us that any attempt to write down a relativistic version of the one-particle Schr¨dinger o equation (or. This formalism is quantum field theory (QFT). The “stuff” is the proton field or. There is no mechanism in standard non-relativistic quantum mechanics to deal with changes in the particle number. manufactured so far apart in space and time. If you like. this failure is telling us that once we we enter the relativistic regime we need a new formalism in order to treat states with an unspecified number of particles. Let me illustrate this through a rather prosaic story. the de Broglie wavelength is the distance at which the wavelike nature of particles becomes apparent. Being the “same” in the quantum world is not like being the “same” in the classical world: quantum particles that are the same are truely indistinguishable. any attempt to naively construct a relativistic version of the one-particle Schr¨dinger equation meets with serious problems. indeed. And the two are exactly the same! How is this possible? Why aren’t there errors in proton production? How can two objects. It is always smaller than the de Broglie wavelength λdB = h/|p|. (Negative o probabilities. In quantum mechanics you have to put these statistics in by hand –3– . In fact. Suppose we capture a proton from a cosmic ray which we identify as coming from a supernova lying 8 billion lightyears away. This minus sign determines the statistics of the particle. there’s more to this tale. infinite towers of negative energy states. Answer 2: Because all particles of the same type are the same This sound rather dumb. The same is true of every other fundamental particle.At distances shorter than this. if you look closely enough.

Thus the basic degrees of freedom in quantum field theory are operator valued functions of space and time. for any relativistic system it is a necessity. They have dimensions [c] = LT −1 [ ] = L2 MT −1 [G] = L3 M −1 T −2 Throughout this course we will work with “natural” units. this relationship between spin and statistics is not something that you have to put in by hand. should choose Bose statistics (no minus sign) for integer spin particles. Rather. Quantum field theory has had a major impact in condensed matter.1) –4– . cosmology.and. This means that we are dealing with an infinite number of degrees of freedom — at least one for every point in space. defined by c= =1 (0. In standard quantum mechanics. What is Quantum Field Theory? Having told you why QFT is necessary.1 Units and Scales Nature presents us with three fundamental dimensionful constants. and Fermi statistics (yes minus sign) for half-integer spin particles. it is a consequence of the framework. The rules for quantizing a field are no different. But it is also a very useful tool in non-relativistic systems with many particles. the most familiar example of which is the electromagnetic field. I should really tell you what it is. to agree with experiment. we’re taught to take the classical degrees of freedom and promote them to operators acting on a Hilbert space. It will turn out that the possible interactions in quantum field theory are governed by a few basic principles: locality. These ideas make QFT a very robust framework: given a set of fields there is very often an almost unique way to couple them together. the speed of light c. What is Quantum Field Theory Good For? The answer is: almost everything. As I have stressed above. highenergy physics. In quantum field theory. Planck’s constant (divided by 2π) and Newton’s constant G. quantum gravity and pure mathematics. It is literally the language in which the laws of Nature are written. symmetry and renormalization group flow (the decoupling of short distance phenomena from physics at larger scales). The clue is in the name: it is the quantization of a classical field. 0. This infinity will come back to bite on several occasions.

the electron volt or. we need to insert the relevant powers of c and . the length scale λ associated to a mass m is the Compton wavelength λ= mc With this conversion factor. more often GeV = 109 eV or T eV = 1012 eV . G= 1 c = 2 2 Mp Mp (0. The usual choice of energy unit is eV . The smallest and largest scales known are shown on the figure. For example. equivalently. It corresponds to a length lp ≈ 10−33 cm. meaning the mass dimension. length increasing to the left. correspondingly. Throughout this course we will refer to the dimension of a quantity. the surviving natural quantity G has dimensions [G] = −2 and defines a mass scale. energy (since E = mc2 has become E = m). To convert the unit of energy back to a unit of length or time. together with other relevant energy scales. The largest length scale we can talk of is the size of the cosmological horizon. The Planck scale is thought to be the smallest length scale that makes sense: beyond this quantum gravity effects become important and it’s no longer clear that the concept of spacetime makes sense. This is a logarithmic plot. the electron mass me = 106 eV translates to a length scale λe = 2 × 10−12 m.2) where Mp ≈ 1019 GeV is the Planck scale. roughly 1060 lp . The standard model of particle physics is expected to hold up –5– .which allows us to express all dimensionful quantities in terms of a single scale which we choose to be mass or. Observable Universe ~ 20 billion light years Earth 10 cm 10 Cosmological Constant Atoms −8 Planck Scale Nuclei −13 LHC 10 −33 cm 10 cm 10 cm Energy length 10 −33 eV 10 −3 eV 10 11 − 10 12 eV = 1 TeV 10 28 eV = 1019 GeV Figure 2: Energy and Distance Scales in the Universe Some useful scales in the universe are shown in the figure. If X has dimensions of (mass)d we will write [X] = d. In particular. with energy increasing to the right and.

to about the T eV .Z Bosons Higgs Boson Mass ∼ 10−2 eV 0.5 MeV 100 MeV 140 MeV 1 GeV 2 GeV 80-90 GeV 120-200 GeV?? –6– . the rough masses of some elementary (and not so elementary) particles are shown in the table. Particle neutrinos electron Muon Pions Proton. This is precisely the regime that is currently being probed by the Large Hadron Collider (LHC) at CERN. there are experimental hints that the coupling constants of electromagnetism. There is a general belief that the framework of quantum field theory will continue to hold to energy scales only slightly below the Planck scale — for example. Neutron Tau W. and the weak and strong forces unify at around 1018 GeV. For comparison.

t) and ∇φ(x. 1. A) where µ = 0.3) –7– . Thus we are dealing with a system with an infinite number of degrees of freedom — at least one for each point x in space. indexed by a label a. t) (1. E(x. The Lagrangian The dynamics of the field is governed by a Lagrangian which is a function of φ(x. An Example: The Electromagnetic Field The most familiar examples of fields from classical physics are the electric and magnetic fields. t). the Lagrangian is of the form. in field theory we are interested in the dynamics of fields φa (x. Both of these fields are spatial 3-vectors. t). 3 shows that this field is a vector in spacetime. In all the systems we study in this course. 1. L(t) = d3 x L(φa . While classical particle mechanics deals with a finite number of generalized coordinates qa (t). t). Notice that the concept of position has been relegated from a dynamical variable in particle mechanics to a mere label in field theory. and will return to classical field theory at several later stages in the course when we need to introduce new ideas. t) = (φ. t). we derive these two 3-vectors from a single 4-component field Aµ (x. In a more sophisticated treatement of electromagnetism.1) where both a and x are considered as labels. We will cover only the bare minimum ground necessary before turning to the quantum theory. 2.1. Classical Field Theory In this first section we will discuss various aspects of classical fields. ∇ · B = 0 and dB/dt = −∇ × E.2) which ensure that two of Maxwell’s equations.1 The Dynamics of Fields A field is a quantity defined at every point of space and time (x. The electric and magnetic fields are given by E = −∇φ − ∂A ∂t and B = ∇ × A (1. t) and B(x. ˙ φ(x. ∂µ φa ) (1. hold immediately as identities.

∇3 φ. there’s nothing to stop L depending on ∇φ. etc. ∇2 φ. t2 S= t1 dt d3 x L = d4 x L (1.7) where we are using the Minkowski space metric +1 η µν = ηµν = −1 −1 −1 (1. The action is. we will only consider Lagrangians depending on ∇φ and not higher derivatives. However.where the official name for L is the Lagrangian density. t1 ) = δφa (x. Requiring δS = 0 for all such paths yields the Euler-Lagrange equations of motion for the fields φa . t2 ) = 0. L= η µν ∂µ φ∂ν φ − 1 m2 φ2 2 1 2 2 ˙ = 1 φ − (∇φ) − 1 m2 φ2 2 2 2 1 2 (1. ¨ we similarly restrict to Lagrangians L depending on φ and φ.7) to the usual expression for the Lagrangian L = T − V .1. keeping the end points fixed and require δS = 0. although everyone simply calls it the Lagrangian. In field theory ˙ ¨ ˙ and not φ.9) –8– .4) Recall that in particle mechanics L depends on q and q. we identify the kinetic energy of the field as T = 1 ˙ d 3 x 2 φ2 (1.1 An Example: The Klein-Gordon Equation Consider the Lagrangian for a real scalar field φ(x.8) Comparing (1. but not q . δS = = d4 x d4 x ∂L ∂L δφa + δ(∂µ φa ) ∂φa ∂(∂µ φa ) ∂L ∂L − ∂µ δφa + ∂µ ∂φa ∂(∂µ φa ) ∂L δφa ∂(∂µ φa ) (1.5) The last term is a total derivative and vanishes for any δφa (x. t). In principle. We vary the path. with an eye to later Lorentz invariance. ∂µ ∂L ∂(∂µ φa ) − ∂L =0 ∂φa (1. We can determine the equations of motion by the principle of least action. t) that decays at spatial infinity and obeys δφa (x.6) 1.

14) 2 ∂φ 1.1.8. at least. Except it isn’t! Or. while the phrase “potential energy”. the o interpretation of this equation is very different: the field ψ is a classical field with none of the probability interpretation of the wavefunction. rather than quadratic. we compute ∂L ∂L ˙ = −m2 φ and = ∂ µ φ ≡ (φ. Take a complex scalar field ψ whose dynamics is defined by the real Lagrangian i ˙ ˙ L = (ψ ⋆ ψ − ψ ⋆ ψ) − ∇ψ ⋆ · ∇ψ − mψ ⋆ ψ (1.and the potential energy of the field as V = 1 d3 x 1 (∇φ)2 + 2 m2 φ2 2 (1.12) –9– .10) The first term in this expression is called the gradient energy. We’ll come back to this point in Section 2.2 Another Example: First Order Lagrangians We could also consider a Lagrangian that is linear in time derivatives. is usually reserved for the last term. −∇φ) ∂φ ∂(∂µ φ) The Euler-Lagrange equation is then ¨ φ − ∇2 φ + m2 φ = 0 which we can write in relativistic form as ∂µ ∂ µ φ + m2 φ = 0 (1. so that i ˙ ∂L ∂L ∂L i = ψ − mψ and = −∇ψ (1. or just “potential”.7). ∂V L = 1 ∂µ φ∂ µ φ − V (φ) ⇒ ∂µ ∂ µ φ + =0 (1. (1.13) This is the Klein-Gordon Equation. To determine the equations of motion arising from (1. The Laplacian in Minkowski space is sometimes denoted by .11) (1.17) ∂t This looks very much like the Schr¨dinger equation. An obvious generalization of the Klein-Gordon equation comes from considering the Lagrangian with arbitrary potential V (φ). In this notation.15) 2 We can determine the equations of motion by treating ψ and ψ ⋆ as independent objects. the Klein-Gordon equation reads φ + m2 φ = 0.16) = − ψ and ˙ ∂ψ ⋆ 2 2 ∂∇ψ ⋆ ∂ψ⋆ This gives us the equation of motion ∂ψ i = −∇2 ψ + mψ (1.

1 L = − 1 (∂µ Aν ) (∂ µ Aν ) + 2 (∂µ Aµ )2 2 (1. as far as we know. t) with x = y. But the closest we get for the x label is a coupling between φ(x) and φ(x + δx) through the gradient term (∇φ)2 . so L ∼ 2 A2 .1. there are no terms that look like L= d3 xd3 y φ(x)φ(y) (1. t) directly to φ(y. both φ and φ must be specified to determine the future evolution.19) where the field strength is defined by Fµν = ∂µ Aν − ∂ν Aµ . ˙ however when L ∼ ψ ⋆ ψ. there’s no reason for this. In this course. The Lagrangian (1.20) (1.4 Locality. For example. a key feature of all theories of Nature.8). one of the main reasons for introducing field theories in classical physics is to implement locality. This property of locality is. After all. x is merely a label. we will only consider local Lagrangians. You can check using (1. we compute ∂L = −∂ µ Aν + (∂ρ Aρ ) η µν ∂(∂µ Aν ) from which we may derive the equations of motion. Using the notation of the field strength. Indeed. ˙ ˙ When L ∼ φ2 . ∂µ ∂L ∂(∂µ Aν ) = −∂ 2 Aν + ∂ ν (∂ρ Aρ ) = −∂µ (∂ µ Aν − ∂ ν Aµ ) ≡ −∂µ F µν (1. – 10 – . To see that Maxwell’s equations indeed follow from (1. and we’re quite happy to couple other labels together (for example. This means that there are no terms in the Lagrangian coupling φ(x.18) has no i ˙0 kinetic term A2 for A0 . We will see the consequences of this in Section 6.18). Locality. Locality In each of the examples above. the term ∂3 A0 ∂0 A3 in the Maxwell Lagrangian couples the µ = 0 field to the µ = 3 field).22) (1.21) A priori.1. only ψ and ψ ⋆ are needed.3 A Final Example: Maxwell’s Equations We may derive Maxwell’s equations in the vacuum from the Lagrangian. 1. the Lagrangian is local. we may rewrite the Maxwell Lagrangian (up to an integration by parts) in the compact form L = − 1 Fµν F µν 4 1.18) Notice the funny minus signs! This is to ensure that the kinetic terms for Ai are positive 1 ˙ using the Minkowski space metric (1.The initial data required on a Cauchy surface differs for the two examples above.2) that this reproduces the remaining two Maxwell’s equations in a vacuum: ∇ · E = 0 and ∂ E/∂t = ∇ × B.

 0   0 0 1 (1.2 Lorentz Invariance The laws of Nature are relativistic. You’ll learn more about this in the “Symmetries and Particle Physics” course. (If we were instead dealing with a passive transformation in which we relabel our choice of coordinates. it will suffice to consider a non-relativistic example such as a temperature field. The simplest example is the scalar field which. 0. and a boost by v < 1 are respectively described by the Lorentz transformations    1 0 0 0 γ −γv 0     0 cos θ − sin θ 0    and Λµν =  −γv γ 0 Λµν =     0 1  0 sin θ cos θ 0   0 0 0 0 1 0 0 (1. 1. say. – 11 – . 0). After a rotation x → Rx about the z-axis. and one of the main motivations to develop quantum field theory is to reconcile quantum mechanics with special relativity. Suppose we start with an initial field φ(x) which has a hotspot at. 0). we need to place ourselves at x = (0. This R−1 is the origin of the inverse transformation. 0).24) along the x1 -axis  (1. a rotation by θ about the x3 -axis.1. 1. under the Lorentz transformation x → Λx.26) The inverse Λ−1 appears in the argument because we are dealing with an active transformation in which the field is truly shifted. To see why this means that the inverse appears. we would have instead φ(x) → φ′ (x) = φ(Λx)). xµ −→ (x′ )µ = Λµν xν where Λµν satisfies Λµσ η στ Λντ = η µν For example. x = (1. we want to construct field theories in which space and time are placed on an equal footing and the theory is invariant under Lorentz transformations. The Lorentz transformations form a Lie group under matrix multiplication. 0) = (1. To this end.25) The Lorentz transformations have a representation on the fields. If we want to express φ′ (x) in terms of the old field φ. 0) and ask what the old field looked like where we’ve come from at R−1 (0. 0.23) 0 √ with γ = 1/ 1 − v 2 . the new field φ′ (x) will have the hotspot at x = (0. 1. transforms as φ(x) → φ′ (x) = φ(Λ−1 x) (1.

Putting this all together.15). This follows because det Λ = 1. This means that the derivative terms in the Lagrangian density transform as Lderiv (x) = ∂µ φ(x)∂ν φ(x)η µν −→ (Λ−1 )ρµ (∂ρ φ)(y) (Λ−1 )σν (∂σ φ)(y) η µν = (∂ρ φ)(y) (∂σ φ)(y) η ρσ = Lderiv (y) (1. we find that the action is indeed invariant under Lorentz transformations. Let’s look at our examples: Example 1: The Klein-Gordon Equation For a real scalar field we have φ(x) → φ′ (x) = φ(Λ−1 x). S= d4 x L(x) −→ d4 x L(y) = d4 y L(y) = S (1. but quadratic in spatial derivatives). in the last step.28) where. Other Lorentz invariant objects you can use include the totally antisymmetric tensor ǫµνρσ and the matrices γµ that we will introduce when we come to discuss spinors in Section 4. 1. we need the fact that we don’t pick up a Jacobian factor when we change integration variables from d4 x to d4 y. The derivative of the scalar field transforms as a vector. it’s easy to see if the action is Lorentz invariant: just make sure all the Lorentz indices µ = 0. (L is linear in time derivatives. is all we deal with). with φ2 (x) → φ2 (y).The definition of a Lorentz invariant theory is that if φ(x) solves the equations of motion then φ(Λ−1 x) also solves the equations of motion. 2. for now. 3 are contracted with Lorentz invariant objects. Example 2: First Order Dynamics In the first-order Lagrangian (1. such as the metric ηµν . The theory is not Lorentz invariant. We can ensure that this property holds by requiring that the action is Lorentz invariant. (At least for Lorentz transformation connected to the identity which. space and time are not on the same footing. In practice. meaning (∂µ φ)(x) → (Λ−1 )νµ (∂ν φ)(y) where y = Λ−1 x.27) The potential terms transform in the same way. – 12 – .

Example 3: Maxwell’s Equations Under a Lorentz transformation Aµ (x) → Λµν Aν (Λ−1 x). We start here by recasting Noether’s theorem in a field theoretic framework. However. To see this. QV = V d3 x j 0 (1. There are Lorentz symmetries. supersymmetries. You can check that Maxwell’s Lagrangian (1. we can define the charge in a finite volume V ..32) Repeating the analysis above..31) assuming that j → 0 sufficiently quickly as |x| → ∞. Of course. in other words. gauge symmetries. internal symmetries. 1. ∂j 0 /∂t + ∇ · j = 0. A Comment: A conserved current implies a conserved charge Q.3 Symmetries The role of symmetries in field theory is possibly even more important than in particle mechanics. historically electrodynamics was the first Lorentz invariant theory to be discovered: it was found even before the concept of Lorentz invariance.21) is indeed invariant. defined as Q= R3 (1.1 Noether’s Theorem Every continuous symmetry of the Lagrangian gives rise to a conserved current j µ (x) such that the equations of motion imply ∂µ j µ = 0 or. 1. dQ = dt d3 x R3 ∂j 0 =− ∂t R3 d3 x ∇ · j = 0 (1.29) d3 x j 0 (1. the existence of a current is a much stronger statement than the existence of a conserved charge because it implies that charge is conserved locally. we find that dQV =− dt d3 x ∇ · j = − j · dS (1.3..30) which one can immediately see by taking the time derivative.33) V A – 13 – .

35) (1. We will now see something similar in field theories. Then δL = ∂L ∂L δφa + ∂µ (δφa ) ∂φa ∂(∂µ φa ) ∂L ∂L δφa + ∂µ − ∂µ = ∂φa ∂(∂µ φa ) ∂L δφa ∂(∂µ φa ) (1.36) When the equations of motion are satisfied.3. δL = ∂µ F µ (1. So we’re left with δL = ∂µ ∂L δφa ∂(∂µ φa ) (1. Consider the infinitesimal translation xν → xν − ǫν ⇒ φa (x) → φa (x) + ǫν ∂ν φa (x) (1. We may always do this if we have a continuous symmetry. This equation means that any charge leaving V must be accounted for by a flow of the current 3vector j out of the volume. the term in square brackets vanishes.34) for some set of functions F µ (φ). invariance under spatial translations gives rise to the conservation of momentum. Proof of Noether’s Theorem: We’ll prove the theorem by working infinitesimally. To derive Noether’s theorem. we have by definition δL = ∂µ F µ .37) But for the symmetry transformation δφa = Xa (φ). while invariance under time translations is responsible for the conservation of energy.where A is the area bounding V and we have used Stokes’ theorem.38) 1. we first consider making an arbitrary transformation of the fields δφa .37) gives us the result ∂µ j µ = 0 with j µ = ∂L Xa (φ) − F µ (φ) ∂(∂µ φa ) (1. This kind of local conservation of charge holds in any local field theory.2 An Example: Translations and the Energy-Momentum Tensor Recall that in classical particle mechanics.39) – 14 – . We say that the transformation δφa (x) = Xa (φ) is a symmetry if the Lagrangian changes by a total derivative. Equating this expression with (1.

2. there is typically a way to massage the energy momentum tensor of any theory into a symmetric form by adding an extra term Θµν = T µν + ∂ρ Γρµν (1.44) One can verify using the equation of motion for φ that this expression indeed satisfies ∂µ T µν = 0. For this example. 1. Similarly. as opposed to passive.43) (1. All theories that we consider in this course will have this property).41) T µ is called the energy-momentum tensor. An Example of the Energy-Momentum Tensor Consider the simplest scalar field theory with Lagrangian (1.(where the sign in the field transformation is plus.45) (1. the Lagrangian transforms as L(x) → L(x) + ǫν ∂ν L(x) (1. instead of minus.40) (This is the correct transformation for a Lagrangian that has no explicit x dependence. Since the change in the Lagrangian is a total derivative. so that T µν = T νµ . one for each of the translations ǫν with ν = 0.7). From the above discussion. because we’re doing an active. while P i is the total momentum of the field configuration. T µν came out symmetric. (j µ )ν = ∂L µ ∂ν φa − δν L ≡ T µ ν ∂(∂µ φa ) ∂µ T µ = 0 ν The four conserved quantities are given by E= d3 x T 00 and P i = d3 x T 0i (1. 3. Nevertheless. we may invoke Noether’s theorem which gives us four conserved currents (j µ )ν . transformation). but only depends on x through the fields φa (x).46) ˙ d3 x φ ∂ i φ Notice that for this example. we can compute T µν = ∂ µ φ ∂ ν φ − η µν L (1.47) – 15 – .42) where E is the total energy of the field configuration. the conserved energy and momentum are given by E= Pi = 1 1 ˙ d3 x 2 φ2 + 2 (∇φ)2 + 1 m2 φ2 2 (1. It satisfies ν (1. This won’t always be the case.

which agrees with the number of different Lorentz transformations (3 rotations + 3 boosts). A Cute Trick One reason that you may want a symmetric energy-momentum tensor is to make contact with general relativity: such an object sits on the right-hand side of Einstein’s field equations. What is the analogy in field theory? Moreover. the number of different 4×4 anti-symmetric matrices is 4×3/2 = 6. introducing an arbitrary metric gµν (x) in place of ηµν . Now the transformation on a scalar field is given by φ(x) → φ′ (x) = φ(Λ−1 x) = φ(xµ − ω µν xν ) = φ(xµ ) − ω µν xν ∂µ φ(x) (1.48) Θ =−√ −g ∂gµν gµν =ηµν It should be noted however that this trick requires a little more care when working with spinors.50) So the infinitesimal form ω µν of the Lorentz transformation must be an anti-symmetric matrix.where Γρµν is some function of the fields that is anti-symmetric in the first two indices so Γρµν = −Γµρν . we now have further Lorentz transformations. Firstly consider coupling the theory to a curved background spacetime.3. As a check. This guarantees that ∂µ ∂ρ Γρµν = 0 so that the new energy-momentum tensor is also a conserved current.24) for Λ to be a Lorentz transformation becomes (δ µσ + ω µσ )(δ ντ + ω ντ ) η στ = η µν ⇒ ω µν + ω νµ = 0 (1.3 Another Example: Lorentz Transformations and Angular Momentum In classical particle mechanics. we first need the infinitesimal form of the Lorentz transformations Λµν = δ µν + ω µν (1. and replacing the kinetic terms with suitable covariant derivatives using “minimal coupling”. 1. In fact this observation provides a quick and easy way to determine a symmetric energy-momentum tensor.51) – 16 – . namely boosts. rotational invariance gave rise to conservation of angular momentum. Then a symmetric energy momentum tensor in the flat space theory is given by √ 2 ∂( −gL) µν (1. The condition (1.49) where ω µν is infinitesimal. What conserved quantity do they correspond to? To answer these questions.

Once again.from which we see that δφ = −ω µν xν ∂µ φ By the same argument. so dP i/dt = 0.56) (J µ )ρσ = xρ T µσ − xσ T µρ (1.55) δL = −ω µν xν ∂µ L = −∂µ (ω µν xν L) (1. But really. – 17 – . one for each choice of ω µν . leaving us with the following consequence of invariance under boosts: d d3 x xi T 00 = constant (1. 3. the Lorentz transformation is a rotation and the three conserved charges give the total angular momentum of the field. Qij = d3 x (xi T 0j − xj T 0i ) d3 x (x0 T 0i − xi T 00 ) d3 x xi T 00 (1. It’s kind of like a field theoretic version of Newton’s first law but.58) (1. the Lagrangian density transforms as where the last equality follows because ω µµ = 0 due to anti-symmetry.59) dt This is the statement that the center of energy of the field travels with a constant velocity. I’ve left the infinitesimal choice of ω µν in the expression for this current. We can write them as which satisfy ∂µ (J µ )ρσ = 0 and give rise to 6 conserved charges. we should strip it out to give six different currents. the conserved charges are Q0i = (1.54) = −ω ρν ρ ∂(∂µ φ) Unlike in the previous example. 2.57) The fact that these are conserved tells us that ∂T 0i d dQ0i = d3 x T 0i + t d3 x − 0= dt ∂t dt i dP d = Pi + t d3 x xi T 00 − dt dt But we know that P i is conserved.52) But what about the boosts? In this case. appearing here as a conservation law. For ρ. rather surprisingly. i.e.53) (1. the Lagrangian changes by a total derivative so we may apply Noether’s theorem (now with F µ = −ω µν xν L) to find the conserved current ∂L ω ρν xν ∂ρ φ + ω µν xν L jµ = − ∂(∂µ φ) ∂L xν ∂ρ φ − δ µρ xν L = −ω ρν T µ xν (1. σ = 1.

(Actually O(N) in this case). baryon or lepton number). The simplest example occurs √ for a complex scalar field ψ(x) = (φ1 (x) + iφ2 (x))/ 2.3.60) a=1 1 ∂µ φa ∂ φa − 2 µ N N 2 m2 φ2 a a=1 −g φ2 a a=1 (1.63) We will later see that the conserved charges arising from currents of this type have the interpretation of electric charge or particle number (for example. equivalently. The associated conserved current is j µ = i(∂ µ ψ ⋆ )ψ − iψ ⋆ (∂ µ ψ) (1. it’s easier (and equivalent) to treat ψ and ψ ⋆ as independent variables and vary the action with respect to both of them. The Lagrangian remains invariant under this change: δL = 0. We can build a real Lagrangian by where the potential is a general polynomial in |ψ|2 = ψ ⋆ ψ.4 Internal Symmetries The above two examples involved transformations of spacetime. For example.61) ∂ψ ⋆ The Lagrangian has a continuous symmetry which rotates φ1 and φ2 or.1. To find the equations of motion. Non-Abelian symmetries of this type are often referred to as global symmetries to distinguish them from the “local gauge” symmetries that you will meet later. One can construct theories from complex fields in a similar manner that are invariant under an SU(N) symmetry group. all with the same mass and the Lagrangian 1 L= 2 N L = ∂µ ψ ⋆ ∂ µ ψ − V (|ψ|2 ) (1. albeit realized only approximately in Nature. rotates the phase of ψ: ψ → eiα ψ or δψ = iαψ (1. An internal symmetry is one that only involves a transformation of the fields and acts the same at every point in spacetime. However.62) where the latter equation holds with α infinitesimal. we could expand ψ in terms of φ1 and φ2 and work as before.64) In this case the Lagrangian is invariant under the non-Abelian symmetry group G = SO(N). Non-Abelian Internal Symmetries Consider a theory involving N scalar fields φa . Isospin is an example of such a symmetry. as well as transformations of the field. varying with respect to ψ ⋆ leads to the equation of motion ∂V (ψ ⋆ ψ) ∂µ ∂ µ ψ + =0 (1. – 18 – .

we eliminate φa (x) in favour of π a (x) everywhere in H. including δφ = α(x)φ) we have ∂µ hµ = 0 (1. The change in the action is therefore δS = d4 x δL = − d4 x α(x) ∂µ hµ (1. ∂L (1. α is a constant real number. just like the field φa (x) itself. The Hamiltonian is then simply H= d3 x H (1. in the action that contribute to the current. and focus instead on canonical quantization. 1. (The potential terms are invariant even when α = α(x)). The Hamiltonian density is given by ˙ where. Now consider performing the transformation but where α depends on spacetime: α = α(x). the change must be of the form δL = (∂µ α) hµ (φ) (1. We start by defining the momentum π a (x) conjugate to φa (x). This way of viewing things emphasizes that it is the derivative terms.66) which means that when the equations of motion are satisfied (so δS = 0 for all variations.65) since we know that δL = 0 when α is constant.69) – 19 – . as in classical mechanics. However. For this we need the Hamiltonian formalism of field theory. Here.67) We see that we can identify the function hµ = j µ as the conserved current.68) π a (x) = ˙ ∂ φa The conjugate momentum π a (x) is a function of x.4 The Hamiltonian Formalism The link between the Lagrangian formalism and the quantum theory goes via the path integral.43) which is a single number characterizing the whole field configuration.Another Cute Trick There is a quick method to determine the conserved current associated to an internal symmetry δφ = αφ for which the Lagrangian is invariant. The action is no longer invariant.70) ˙ H = π a (x)φa (x) − L(x) (1. (We may generalize the discussion easily to a non-Abelian internal symmetry for which α becomes a matrix). not the potential terms. In this course we will not discuss path integral methods. It is not to be confused with the total momentum P i defined in (1.

even though the Hamiltonian framework doesn’t look Lorentz invariant. all final answers must be Lorentz invariant even if it’s not manifest at intermediate steps. For example. H= 1 d3 x 1 π 2 + 2 (∇φ)2 + V (φ) 2 (1. t) (1. t) and π(x. which gives us the Hamiltonian.An Example: A Real Scalar Field For the Lagrangian 1 ˙ L = 1 φ2 − 2 (∇φ)2 − V (φ) 2 (1.45) that we get from applying Noether’s theorem for time translation invariance. We will pause at several points along the quantum route to check that this is indeed the case.73) which. If we start from a relativistic theory. do not look Lorentz invariant. the physics must remain unchanged.72) Notice that the Hamiltonian agrees with the definition of the total energy (1. Nevertheless. unlike the Euler-Lagrange equations (1. Lorentz invariance is clear for all to see since the action is invariant under Lorentz transformations. t) = ∂H ∂π(x. In the Lagrangian formalism.71) ˙ the momentum is given by π = φ. – 20 – .6). t) arise from Hamilton’s equations. the Hamiltonian formalism is not manifestly Lorentz invariant: we have picked a preferred time. In contrast. t) = − ˙ ∂H ∂φ(x. ˙ φ(x. the equations of motion for φ(x) = φ(x.

1) In field theory we do the same. in units with = 1. for certain theories — known as free theories — we can find a way to write the dynamics such that each degree of freedom evolves independently – 21 – . Thus a quantum field is an operator valued function of space obeying the commutation relations [φa (x). We are working in the Schr¨dinger picture so that the operators φa (x) and o a π (x) do not depend on time at all — only on space.2.2) Note that we’ve lost all track of Lorentz invariance since we have separated space x and time t. this is usually very hard.” Sidney Coleman 2. canonical quantization is a recipe that takes us from the Hamiltonian formalism of classical dynamics to the quantum theory. Be warned however that the notation |ψ for the state is deceptively simple: if you were to write the wavefunction in quantum field theory. pb ] = 0 b [qa . so that. The Poisson bracket structure of classical mechanics morphs into the structure of commutation relations between operators. π b(y)] = iδ (3) (x − y) δa (2. π b (y)] = 0 b [φa (x). [qa . qb ] = [pa . The typical information we want to know about a quantum theory is the spectrum of the Hamiltonian H. However. pb ] = i δa (2. In quantum field theories. All time dependence sits in the states |ψ which evolve by the usual Schr¨dinger equation o i d|ψ = H |ψ dt (2. that is a function of every possible configuration of the field φ. Free Fields “The career of a young theoretical physicist consists of treating the harmonic oscillator in ever-increasing levels of abstraction. The recipe tells us to take the generalized coordinates qa and their conjugate momenta pa and promote them to operators. now for the field φa (x) and its momentum conjugate π b (x). One reason for this is that we have an infinite number of degrees of freedom — at least one for every point x in space. φb(y)] = [π a (x).1 Canonical Quantization In quantum mechanics. it would be a functional. we’re merely applying the old formalism to fields.3) We aren’t doing anything different from usual quantum mechanics.

1 The Simple Harmonic Oscillator Consider the quantum mechanical Hamiltonian 1 1 H = 2 p2 + 2 ω 2 q 2 (2.9) which can be easily inverted to give 1 q = √ (a + a† ) . t) (2π)3 (2. or sometimes ladder operators) a= i ω q+√ p . p] = i.10) – 22 – . t) satisfies ∂2 + (p 2 + m2 ) ∂t2 φ(p.5) Thus. we need only take the Fourier transform. 2. 2 2ω a† = i ω q−√ p 2 2ω (2. t) = Then φ(p. t) = 0 (2. t) we must simply quantize this infinite number of harmonic oscillators. Let’s recall how to do this. Free field theories typically have Lagrangians which are quadratic in the fields. t). For example.1. for each value of p.7) We learn that the most general solution to the KG equation is a linear superposition of simple harmonic oscillators.8) with the canonical commutation relations [q.6) d3 p ip·x e φ(p. 2ω p = −i ω (a − a† ) 2 (2. each vibrating at a different frequency with a different amplitude. φ(p. To find the spectrum we define the creation and annihilation operators (also known as raising/lowering operators. so that the equations of motion are linear. t) solves the equation of a harmonic oscillator vibrating at frequency ωp = + p 2 + m2 (2.from all the others. ∂µ ∂ µ φ + m2 φ = 0 (2.4) To exhibit the coordinates in which the degrees of freedom decouple from each other. To quantize φ(x. the simplest relativistic free theory is the classical Klein-Gordon (KG) equation for a real scalar field φ(x. φ(x.

so that H |E = E |E . a† ] = 1 while the Hamiltonian is given by H = 1 ω(aa† + a† a) 2 1 = ω(a† a + 2 ) (2. indexed by the 3-momentum p.12) One can easily confirm that the commutators between the Hamiltonian and the creation and annihilation operators are given by [H. . E − ω. 1 H |0 = 2 ω |0 (2. φ(x) = π(x) = d3 p (2π)3 1 † ap eip·x + ap e−ip·x 2ωp ωp † ap eip·x − ap e−ip·x 2 (2. a† ] = ωa† and [H. . 2. Ha† |E = (E + ω)a† |E . . a] = −ωa (2.Substituting into the above expressions we find [a. . Let |E be an eigenstate with energy E.17) where I’ve ignored the normalization of these states so. E + 2ω. E. (2. n| n = 1.2 The Free Scalar Field We now apply the quantization of the harmonic oscillator to the free scalar field.15) If the energy is bounded below. Then we can construct more eigenstates by acting with a and a† .14) So we find that the system has a ladder of states with energies . |n = (a† )n |0 with H |n = (n + 1 )ω |n 2 (2. We write φ and π as a linear sum of an infinite number of creation and annihilation oper† ators ap and ap .11) (2.16) Excited states then arise from repeated application of a† . . there must be a ground state |0 which satisfies a |0 = 0.18) (2. This has ground state energy (also known as zero point energy). E + ω. .13) These relations ensure that a and a† take us between energy eigenstates. Ha |E = (E − ω)a |E (2.19) d3 p (−i) (2π)3 – 23 – .

Assume that [ap . to get to the third line we’ve integrated over d3 x to get delta-functions δ (3) (p±q) which.Claim: The commutation relations for φ and π are equivalent to the following com† mutation relations for ap and ap [φ(x). the first term vanishes and we’re left with H = = 1 2 d3 p † † ω ap ap + ap ap 3 p (2π) d3 p † 1 ωp ap ap + 2 (2π)3 δ (3) (0) (2π)3 (2. aq ] = (2π)3 δ (3) (p − q) (2. We have H = 1 2 1 = 2 d3 x π 2 + (∇φ)2 + m2 φ2 √ ωp ωq d3 x d3 p d3 q † † − (ap eip·x − ap e−ip·x )(aq eiq·x − aq e−iq·x ) 6 (2π) 2 1 † † (ip ap eip·x − ip ap e−ip·x ) · (iq aq eiq·x − iq aq e−iq·x ) + √ 2 ωp ωq m2 † † + √ (ap eip·x + ap e−ip·x )(aq eiq·x + aq e−iq·x ) 2 ωp ωq = 1 4 d3 p 1 † † † 2 2 (−ωp + p 2 + m2 )(ap a−p + ap a† ) + (ωp + p 2 + m2 )(ap ap + ap ap ) −p 3 ω (2π) p where in the second line we’ve used the expressions for φ and π given in (2. φ(y)] = [π(x). aq ] = (2π)3 δ (3) (p − q).22) – 24 – .18) and (2. aq ] = 0 † [ap . π(y)] = iδ (3) (x − y) ⇔ † † [ap . aq ] e−ip·x+iq·y 6 (2π) 2 ωp 3 d p (−i) −eip·(x−y) − eip·(y−x) = (2π)3 2 = iδ (3) (x − y) (2.21) The Hamiltonian † Let’s now compute the Hamiltonian in terms of ap and ap . π(y)] = 0 [φ(x).20) † Proof: We’ll show this just one way. aq ] = [ap . π(y)] = ωq d3 p d3 q (−i) † † −[ap . Now using the expression for the 2 frequency ωp = p 2 + m2 . allow us to perform the d3 q integral.19). aq ] eip·x−iq·y + [ap . in turn. Then [φ(x).

H |0 ≡ E0 |0 = d3 p In fact there are two different ∞’s lurking in the expression (2.3 The Vacuum Following our procedure for the harmonic oscillator. The first arises because space is infinitely large.22).24). (Infinities of this type are often referred to as infra-red divergences although in this case the ∞ is so simple that it barely deserves this name). What to do? Let’s start by looking at the ground state where this infinity first becomes apparent. We impose periodic boundary conditions on the field. taking the limit where L → ∞. the integral over ωp diverges at large p.25) where V is the volume of the box. The integral should be cut-off at high momentum in order to reflect the fact that our theory is likely to break down in some way. let’s consider putting the theory in a box with sides of length L. rather than the energy density E0 . usually that we’re doing something wrong. This is a high frequency — or short distance — infinity known as an ultra-violet divergence. We’ve assumed that our theory is valid to arbitrarily short distance scales. We recognize it as the sum of ground state energies for each harmonic oscillator. This is clearly absurd. 2. or asking the wrong question.24) 2 The subject of quantum field theory is rife with infinities. – 25 – . 1 ωp δ (3) (0) | 0 = ∞ |0 (2. To extract out this infinity. To find E0 we can simply divide by the volume. Let’s take some time to explore where this infinity comes from and how we should deal with it. ap |0 = 0 ∀p (2. evaluated at zero where it has its infinite spike.Hmmmm.23) With this definition.26) V (2π)3 2 which is still infinite. We’ve found a delta-function. So the δ(0) divergence arises because we’re computing the total energy. Then. But E0 → ∞ due to the |p| → ∞ limit of the integral. the energy E0 of the ground state comes from the second term in (2. corresponding to arbitrarily high energies. Each tells us something important. E0 d3 p 1 E0 = = ωp (2. This divergence arises because of our hubris. let’s define the vacuum |0 by insisting that it is annihilated by all ap . we get L/2 L/2 (2π)3 δ (3) (0) = lim L→∞ d3 x eix·p −L/2 p=0 = lim L→∞ d3 x = V −L/2 (2. Moreover.

The method that we’ve used above is called normal ordering. φn (xn ) as : φ1 (x1 ) . we could write (2.We can deal with the infinity in (2. . what cancels them to such high accuracy? This is the cosmological constant problem. if we defined the Hamiltonian of the harmonic oscillator to be H = (1/2)(ωq − ip)(ωq + ip). In fact. There’s no way to measure E0 directly. H |0 = 0. So. which is classically the same as our original choice. . .27).27) so that. 1 Rµν − 2 gµν = −8πGTµν + Λgµν (2. Definition: We write the normal ordered string of operators φ1 (x1 ) .24) in a more practical way. We’ll come across a number of ways of dealing with it. H= d3 p ω a† a 3 p p p (2π) (2. There is a BIG caveat here: gravity is supposed to see everything! The sum of all the zero point energies should contribute to the stress-energy tensor that appears on the right-hand side of Einstein’s equations. We expect them to appear as a cosmological constant Λ = E0 /V . so we can simply redefine the Hamiltonian by subtracting off this infinity. we will normal order all operators in this manner. 2. This type of ordering ambiguity arises a lot in field theories. Why don’t the zero point energies of these fields contribute to Λ? Or. In particular. This is much smaller than other scales in particle physics.3.28) It is defined to be the usual product with all annihilation operators ap placed to the right. . for the Hamiltonian. the Standard Model is valid at least up to 1012 eV . In physics we’re only interested in energy differences.29) In the remainder of this section.27) as : H := d3 p † ωp ap ap (2π)3 (2. then upon quantization it would naturally give H = ωa† a as in (2. No one knows the answer! – 26 – . with this new definition. the difference between this Hamiltonian and the previous one is merely an ordering ambiguity in moving from the classical theory to the quantum theory.30) Current observation suggests that 70% of the energy density in the universe has the properties of a cosmological constant with Λ ∼ (10−3eV )4 .1 The Cosmological Constant Above I wrote “there’s no way to measure E0 directly”. For example. φn (xn ) : (2. if they do.

.32) Figure 3: For a massless scalar field. – 27 – . for there is a situation where differences in the energy of vacuum fluctuations themselves can be measured. it would mean that there is a force on the plates due to the fluctuations of the vacuum.3. we’ll make the x1 direction periodic. and impose periodic boundary conditions such that φ(x) = φ(x + Ln) with n = (1.2. But we should be careful. but remember that we should compute all physical quantities per unit area A. To regulate the infra-red divergences. This is the Casimir force. We’ll leave y and z alone. first predicted in 1948 and observed 10 years later.31) L n ∈ Z+ (2. If this naive guess is true. means that the momentum of the field inside the plates is quantized as p= nπ . In the real world. separated by a distance d ≪ L in the x1 direction. that is something new. the ground state energy between the plates is E(d) = A dpy dpz 1 (2π)2 2 nπ d 2 + p2 + p2 y z (2.. while chanting the mantra that only energy differences can be measured. 0. py . he said.33) n=1 while the energy outside the plates is E(L − d). with the boundary conditions imposed by conducting plates. That is nice. with size L. The total energy is therefore E = E(d) + E(L − d) (2. during a walk. The plates are such that they impose φ(x) = 0 at the position of the plates. The presence of these plates affects the Fourier decomposition of the field and. We insert two reflecting plates. in particular.” Hendrik Casimir Using the normal ordering prescription we can happily set E0 = 0. pz d ∞ d (2. Here we model this effect with a scalar. and he mumbled something about zero-point energy. the effect is due to the vacuum fluctuations of the electromagnetic field. 0).2 The Casimir Effect “I mentioned my results to Niels Bohr.34) which – at least naively – depends on d.

expression (2. E(d) = A ∞ n=1 dpy dpz 1 (2π)2 2 nπ d 2 + p2 y + p2 z e q 2 −a ( nπ ) +p2 +p2 y z d (2. It’s a lot simpler if we look at the problem in d = 1 + 1 dimensions.33). rather than d = 3 + 1 dimensions. otherwise it’s not something we can really trust. things begin to leak. known as the ultra-violet (UV) cut-off.35) which has the property that as a → 0. E is infinite! What to do? The problem comes from the arbitrarily high momentum modes.38) – 28 – . we made it finite in a rather ad-hoc manner and we better make sure that any physical quantity we calculate doesn’t depend on the UV cut-off a. Mathematically. We could regulate this in a number of different ways. E1+1 = E1+1 (d) + E1+1 (L − d) = π L − 2 2πa 24 1 1 + d L−d + O(a2 ) (2. allowing us to start manipulating it thus. One way to do this is to change the integral (2. in the last line.35) is finite. but a little complicated. We’ll find that all the same physics is at play.37) 2 2πa 24d where. This renders the expression finite. We can now compute the full energy. we’ve used the fact that a ≪ d.33) to.But there’s a problem. However (2. The integral (2. E1+1 (d) → π 2d ∞ n e−anπ/d n=1 ∞ 1 ∂ = − 2 ∂a = − e−anπ/d n=1 1 1 ∂ 2 ∂a 1 − e−aπ/d eaπ/d π = 2d (eaπ/d − 1)2 d π = − + O(a2 ) (2. we regain the full. we want to find a way to neglect modes of momentum p ≫ a−1 for some distance scale a ≪ d.35) is do-able. and gives us something we can easily work with. Physically one could argue that any real plate cannot reflect waves of arbitrarily high frequency: at some point. infinite.36) We now regulate this sum by introducing the UV cutoff a introduced above. Of course. Now the energy is given by π E1+1 (d) = 2d ∞ n n=1 (2.

. Let † |p = ap |0 (2.4 Particles Having dealt with the vacuum. include terms of size d/L and a/d. If we ploughed through the analogous calculation in d = 3 + 1 dimensions. It’s easy to verify that † † [H. ap ] = ωp ap and [H. .44) The number 24 that appears in the denominator of the one-dimensional Casimir force plays a more famous role in string theory: the same calculation in that context is the reason the bosonic string lives in 26 = 24 + 2 spacetime dimensions. (The +2 comes from the fact the string itself is extended in one space and one time dimension). – 29 – . we can construct energy eigen† states by acting on the vacuum |0 with ap . ap ] = −ωp ap (2..This is still infinite in the limit a → 0. and take a → 0 and L → ∞.41) which means that. 2 Ep = p 2 + m2 2 (2. which is to be expected. You will need to attend next term’s “String Theory” course to see what on earth this has to do with the Casimir force. 2. just as for the harmonic oscillator. we can now turn to the excitations of the field.43) But we recognize this as the relativistic dispersion relation for a particle of mass m and 3-momentum p. and performed the integral (2. However. the force between the plates remains finite. the force is given by π ∂E1+1 = + . This is the Casimir force2 .39) where the . The key point is that as we remove both the regulators.40) The true Casimir force is twice as large as this. .42) This state has energy H |p = ωp |p with 2 ωp = p 2 + m2 (2. we would find the result 1 ∂E π2 = A ∂d 480d4 (2.35). due to the two polarization states of the photon. ∂d 24d2 (2.

J i = ǫijk d3 x (J 0 )jk (2. . .55) and turn it into an operator. The full Hilbert space of our theory is spanned by acting on the vacuum with all possible combinations of a† ’s. pn = ap1 . After normal ordering. . To stress this. Bosonic Statistics and Fock Space We can create multi-particle states by acting multiple times with a† ’s. we learn that it is indeed an eigenstate. Once again. Multi-Particle States. We interpret the state in which n a† ’s act on the vacuum as an n-particle state.47) It’s not hard to show that acting on the one-particle state with zero momentum. ap aq ar |0 . ap |0 . For example. p This means that the particles are bosons.46) telling us that the state |p has momentum p. . q = |q. which we interpret as telling us that the particle carries no internal angular momentum. P |p = p |p (2. . from now on we’ll write Ep everywhere instead of ωp . We may take the classical total momentum P given in (1.50) – 30 – . quantizing a scalar field gives rise to a spin 0 particle. J i |p = 0 = 0. ap aq |0 .49) (2. apn |0 (2.46) and turn it into an operator. we may take the classical expression for the total angular momentum of the field (1. In other words. the state is symmetric under exchange of any two particles.We interpret the state |p as the momentum eigenstate of a single particle of mass m. . . † † |p1 .45) Acting on our state |p with P . it becomes P =− d3 x π ∇φ = d3 p † p ap ap 3 (2π) (2. † † † † † † |0 . (2. Another property of |p that we can study is its angular momentum. |p.48) Because all the a† ’s commute among themselves. Let’s check this particle interpretation by studying the other quantum numbers of |p . .

52) They are operator valued distributions. taking us between the different sectors in the Fock space. H] = 0. pn = n |p1 . . pn . The one-particle † states |p = ap |0 then satisfy p| q = (2π)3 δ (3) (p − q) (2. The number operator commutes with the Hamiltonian. This means that we can place ourselves in the n-particle sector. For example. [N. 0| φ(x) |0 = 0. The Fock space is simply the sum of the n-particle Hilbert spaces. This is a property of free theories. It is called the number operator N N= d3 p † a a (2π)3 p p (2. There is a useful operator which counts the number of particles in a given state in the Fock space.1 Relativistic Normalization We have defined the vacuum |0 which we normalize as 0| 0 = 1. This means that although φ(x) has a well defined vacuum expectation value. they’re not localized in space in any way — they are momentum eigenstates. . 0| φ(x)φ(x) |0 = ∞. . . Recall that in quantum mechanics the position and momentum eigenstates are not good elements of the Hilbert space since they are not normalizable (they normalize to delta-functions). Similarly. (A typical state might be described by the Gaussian ϕ(p) = exp(−p 2 /2m2 )). we can create a wavepacket |ϕ = d3 p −ip·x e ϕ(p) |p (2π)3 (2. We can construct well defined operators by smearing these distributions over space. This is because they don’t produce normalizable states. ensuring that particle number is conserved. . For example. . .This space is known as a Fock space. but will no longer be true when we consider interactions: interactions create and destroy particles. for all n ≥ 0. Operator Valued Distributions Although we’re referring to the states |p as “particles”. rather than functions.53) which is partially localized in both position and momentum space.4. nor ap are good operators acting on the Fock space. the fluctuations of the operator at a fixed point are infinite. .54) – 31 – . and stay there.51) and satisfies N |p1 . † 0| ap ap |0 = p| p = (2π)3 δ(0) and 0|φ(x) φ(x) |0 = x| x = δ(0) (2. in quantum field theory neither the operators φ(x). 2.

Solving for p0 .54) we know this is given by 1= d3 p |p p| (2π)3 (2. With the normalization (2. p ′). But we haven’t been at all careful with normalizations. In the quantum theory. = p0 >0 d3 p 2p0 (2. How do we figure this out? The trick is to look at an object which we know is Lorentz invariant.But is this Lorentz invariant? It’s not obvious because we only have 3-vectors. |p → |p ′ = U(Λ) |p (2. p ′ ) |p ′ (2. And the relativistic dispersion relation for a massive particle.61) p0 =Ep – 32 – .55) such that the 3-vector transforms as p → p ′ . One such object is the identity operator on one-particle states (which is really the projection operator onto one-particle states). 2 2 pµ pµ = m2 ⇒ p0 = Ep = p 2 + m2 (2. d3 p 2Ep (2.59) Proof: d4 p is obviously Lorentz invariant.57) for some unknown function λ(p. we could get |p → λ(p. there are two branches of solutions: p0 = ±Ep . In general. it would be preferable if the two states are related by a unitary transformation. Claim The Lorentz invariant measure is. So piecing everything together. but it consists of two terms: the measure d3 p and the projector |p p|.56) This would mean that the normalizations of |p and |p ′ are the same whenever p and p ′ are related by a Lorentz transformation.58) This operator is Lorentz invariant. Are these individually Lorentz invariant? In fact the answer is no. the following combination must be Lorentz invariant. What could go wrong? Suppose we have a Lorentz transformation p µ → (p′ )µ = Λµν p ν (2. But the choice of branch is another Lorentz invariant concept. d4 p δ(p2 − p 2 − m2 ) 0 which completes the proof.60) is also Lorentz invariant.

These states now satisfy p| q = (2π)3 2Ep δ (3) (p − q) Finally.67) Notice that. If we write ψ in terms √ of real scalar fields by ψ = (φ1 + iφ2 )/ 2. (2. the Lorentz invariant δ-function for 3-vectors is 2Ep δ (3) (p − q) which follows because d3 p 2Ep δ (3) (p − q) = 1 2Ep |p = 2Ep |p = † 2Ep ap |0 (2.69) – 33 – . we can rewrite the identity on one-particle states as 1= d3 p 1 |p p| (2π)3 2Ep (2. there is no factor of 1/2 in front of the Lagrangian for a complex scalar field.63) So finally we learn that the relativistically normalized momentum states are given by (2. in contrast to the Lagrangian (1. we get the factor of 1/2 coming from the √ 1/ 2’s. For example.68) where the second equation is the complex conjugate of the first.64) Notice that our notation is rather subtle: the relativistically normalized momentum state |p differs from |p just by the absence of a vector over p.From this result we can figure out everything else.5 Complex Scalar Fields Consider a complex scalar field ψ(x) with Lagrangian L = ∂µ ψ ⋆ ∂ µ ψ − M 2 ψ ⋆ ψ (2.62) (2. 2.7) for a real scalar field. The equations of motion are ∂µ ∂ µ ψ + M 2 ψ = 0 ∂µ ∂ µ ψ ⋆ + M 2 ψ ⋆ = 0 (2. We expand the complex field operator as a sum of plane waves as ψ= ψ† = d3 p (2π)3 d3 p (2π)3 1 2Ep 1 2Ep † bp e+ip·x + cp e−ip·x † bp e−ip·x + cp e+ip·x (2.65) Some texts also define relativistically normalized creation operators by a† (p) = We won’t make use of this notation here.66) † 2Ep ap .

π= π† = d3 p i (2π)3 Ep † −ip·x − cp e+ip·x bp e 2 Ep † bp e+ip·x − cp e−ip·x 2 (2. We have [H. Of course. – 34 – . π(y)] = iδ (3) (x − y) and [ψ(x). for a real scalar field there is only a single type of particle: for a real scalar field.74) After normal ordering. Recall that the theory (2. The classical field momentum is π = ∂L/∂ ψ = ψ ⋆ . as well as the usual [ψ(x).71) together with others related by complex conjugation. cq ] = (2π)3 δ (3) (p − q) † [bp . This is the reason that we have different operators b and c† appearing in the positive ˙ ˙ and negative frequency parts.75) so Q counts the number of anti-particles (created by c† ) minus the number of particles (created by b† ). In contrast. the particle is its own antiparticle.72) and † [bp . cq ] = 0 (2.70) d3 p (−i) (2π)3 The commutation relations between fields and momenta are given by [ψ(x).Since the classical field ψ is not real. bq ] = [cp . quantizing a complex scalar field gives rise to two creation operators. bq ] = (2π)3 δ (3) (p − q) (2.67) has a classical conserved charge Q=i ˙ ˙ d3 x (ψ ⋆ ψ − ψ ⋆ ψ) = i d3 x (πψ − ψ ⋆ π ⋆ ) (2. We also turn this into a quantum operator field which we write as. etc. cq ] = [bp . while Nc and Nb individually do not. ψ(y)] = [ψ(x). in our free field theory this isn’t such a big deal because both Nc and Nb are separately conserved. They are interpreted as particles and anti-particles. this becomes the quantum operator Q= d3 p † † (cp cp − bp bp ) = Nc − Nb 3 (2π) (2. † [cp . cq ] = [bp . both of mass M and both spin zero. One can easily check that these field commutation relations are equivalent to the commutation relations for the operators bp and cp . Q] = 0. However. the corresponding quantum field ψ is not hermitian. ψ † (y)] = 0. π † (y)] = 0 (2. we’ll soon see that in interacting theories Q survives as a conserved quantity. These have the interpretation of creating two types of particle. ensuring that Q is conserved quantity in the quantum theory.73) † In summary. bp † and cp .

φ(x)] d3 y π(y) (−i) δ (3) (y − x) = π(x) (2. π(y. t)] = iδ (3) (x − y) (2. t) depends on time.2. π] = ˙ i [ 2 d3 y π(y)2 + ∇φ(y)2 + m2 φ(y)2 . The commutation relations (2. OH ] dt (2. π(x)] – 35 – . φ] = [ 2 =i d3 y π(y)2 + ∇φ(y)2 + m2 φ(y)2 . introducing a preferred time coordinate t. we can start to study how it evolves. say. For example.77) where the subscripts S and H tell us whether the operator is in the Schr¨dinger or o Heisenberg picture. we slowly butchered it as we quantized. OH = eiHt OS e−iHt so that dOH = i[H. t).2) become equal time commutation relations in the Heisenberg picture. [φ(x. the equation of motion for π reads.79) Now that the operator φ(x) = φ(x. In field theory. t)] = 0 [φ(x. π(y. t)] = [π(x.80) Meanwhile. t) = φ(x) (the Heisenberg picture).78) (2. o i d |p(t) = H |p(t) dt ⇒ |p(t) = e−iEp t |p (2. t).76) Things start to look better in the Heisenberg picture where time dependence is assigned to the operators O. the one-particle states evolve in time by Schr¨dinger’s equation. but not on time. Meanwhile. The operators in the two pictures agree at a fixed time. t). For example. we have i ˙ φ = i[H. the operators φ(x) depend on space. It’s not at all obvious that the theory is still Lorentz invariant after quantization. t = 0.6 The Heisenberg Picture Although we started with a Lorentz invariant Lagrangian. π = i[H. we drop these subscripts and we will denote the picture by specifying whether the fields depend on space φ(x) (the Schr¨dinger picture) o or spacetime φ(x. φ(y.

85) – 36 – . φ(x.1 Causality O2 O1 x Figure 4: We’re approaching something Lorentz invariant in the Heisenberg picture.= i 2 d3 y (∇y [φ(y).18) except that the exponent is now written in terms of 4vectors. But there’s still a hint of non-Lorentz invariance because φ and π satisfy equal time commutation relations. t) = d3 p (2π)3 1 2Ep † ap e−ip·x + ap e+ip·x (2. t)] = iδ (3) (x − y) (2. eiHt ap e−iHt = e−iEp t ap † † and eiHt ap e−iHt = e+iEp t ap (2.82). ap ] = +Ep ap . To reach the last line.82) Things are beginning to look more relativistic.84) t which looks very similar to the previous expansion (2.80) and (2.81) =− = ∇2 φ − m2 φ where we’ve included the subscript y on ∇y when there may be some confusion about which argument the derivative is acting on. Putting (2. We can write the Fourier expansion of φ(x) by using the definition (2. [φ(x. ap ] = −Ep ap and [H.77) and noting. (Note also that a sign has flipped in the exponent due to our Minkowski metric contraction).84) indeed satisfies the Klein-Gordon equation (2.6. This then gives. 2. t).81) together we find that the field operator φ satisfies the Klein-Gordon equation ∂µ ∂ µ φ + m2 φ = 0 (2. π(x)] +2im2 φ(y) δ (3)(x − y) d3 y ∇y δ (3) (x − y) ∇y φ(y) − m2 φ(x) (2. It’s simple to check that (2. π(y.83) † † which follows from the commutation relations [H. p · x = Ep t − p · x. where φ(x) now satisfies the KleinGordon equation. we’ve simply integrated by parts. π(x)]) ∇φ(y) + ∇φ(y) ∇y [φ(y).

87) The objects on the right-hand side of this expression are operators. • It vanishes for space-like separations.89) and noticing that we can flip the sign of p in the last exponent as it is an integration variable. I should mention however that the fact that [φ(x). it’s easy to check by direct substitution that the left-hand side is simply a c-number function with the integral expression ∆(x − y) = d3 p 1 e−ip·(x−y) − eip·(x−y) (2π)3 2Ep (2. t)] = 1 2 d3 p (2π)3 1 p 2 + m2 eip·(x−y) − e−ip·(x−y) (2. • It doesn’t vanish for timelike separation. But since ∆(x − y) Lorentz invariant.59). φ(y)] (2. thanks to the appearance of the Lorentz invariant measure d3 p/2Ep that we introduced in (2. which we can see explicitly by writing [φ(x. This follows by noting that ∆(x − y) = 0 at equal times for all (x − y)2 = −(x − y)2 < 0. – 37 – . it can only depend on (x − y)2 and must therefore vanish for all (x − y)2 < 0. it is usually given as one of the axioms of local quantum field theories. Does our theory satisfy this crucial property? Let’s define ∆(x − y) = [φ(x).88) What do we know about this function? • It’s Lorentz invariant. 0) gives [φ(x. [O1 (x). For example.86) This ensures that a measurement at x cannot affect a measurement at y when x and y are not causally connected. However. indeed. we must require that all spacelike separated operators commute.But what about arbitrary spacetime separations? In particular. φ(x. taking x − y = (t. t). This property will continue to hold in interacting theories. t)] ∼ e−imt − e+imt . 0). rather than an operator. 0. We therefore learn that our theory is indeed causal with commutators vanishing outside the lightcone. φ(y)] is a c-number function. O2 (y)] = 0 ∀ (x − y)2 < 0 (2. 0. is a property of free fields only. φ(y. for our theory to be causal.

it is more interesting. nonetheless. φ(y)] = D(x − y) − D(y − x) = 0 if (x − y)2 < 0 (2. this interpretation is also there for a real scalar field because the particle is its own antiparticle. the amplitudes for these two events cancel. Prepare a particle at spacetime point y. ψ † (y)] = 0 outside the lightcone. In fact.7. 2. When (x − y)2 < 0. What is the amplitude to find it at point x? We can calculate this: 0| φ(x)φ(y) |0 = = d3 p d3 p ′ (2π)6 1 ′ † 0| ap ap ′ | 0 e−ip·x+ip ·y 4Ep Ep ′ (2. is non-vanishing! The quantum field appears to leak out of the lightcone. If a particle can travel in a spacelike direction from x → y.1 The Feynman Propagator As we will see shortly.7 Propagators We could ask a different question to probe the causal structure of the theory.93) – 38 – . For spacelike separations. it can just as easily travel from y → x. ∆F (x − y) = 0| T φ(x)φ(y) |0 = D(x − y) D(y − x) x0 > y 0 y 0 > x0 (2.91) So it decays exponentially quickly outside the lightcone but. The interpretation now is that the amplitude for the particle to propagate from x → y cancels the amplitude for the antiparticle to travel from y → x.92) There are words you can drape around this calculation. (x − y)2 < 0. there is no Lorentz invariant way to order events. How do we reconcile these two facts? We can rewrite the calculation (2. With a complex scalar field. In any measurement. one can show that D(x − y) decays like D(x − y) ∼ e−m|x−y| (2.2.90) d3 p 1 −ip·(x−y) e ≡ D(x − y) (2π)3 2Ep The function D(x − y) is called the propagator.89) as [φ(x). Yet we’ve just seen that spacelike measurements commute and the theory is causal. one of the most important quantities in interacting field theory is the Feynman propagator. We can look at the equation [ψ(x).

In the expression (2.96) so the residue of the pole at p0 = ±Ep is ±1/2Ep . Proof: 1 1 1 = 0 = 0 2 2 p2 − m2 (p ) − Ep (p − Ep )(p0 + Ep ) (2. Until now. placing all operators evaluated at later times to the left so.95) Notice that this is the first time in this course that we’ve integrated over 4-momentum. Hence when x0 > y 0 we have ∆F (x − y) = d3 p −2πi −iEp (x0 −y0 )+ip·(x−y) ie (2π)4 2Ep – 39 – . we close the contour in the lower half plane. ensuring that the integrand vanishes since 0 0 0 e−ip (x −y ) → 0. To get the Feynman propagator. We need a prescription for avoiding these singularities in the p0 integral. ∆F (x − y) = i d4 p e−ip·(x−y) 4 p2 − m2 (2π) (2. The integral over p0 then picks up the residue at p0 = +Ep which is −2πi/2Ep where the minus sign arises because we took a clockwise contour. we integrated only over 3-momentum. with p0 fixed by the mass-shell condition to be p0 = Ep . as it stands this integral is ill-defined because. T φ(x)φ(y) = φ(x)φ(y) φ(y)φ(x) x0 > y 0 y 0 > x0 (2. we must choose the contour to be Im(p0) −E p +E p Re(p0) Figure 5: The contour for the Feynman propagator. the denominator p2 −m2 = (p0 )2 −p 2 −m2 produces a pole when p0 = ±Ep = ± p 2 + m2 . for each value of p.95) for ∆F .where T stands for time ordering. However. When x0 > y 0.94) Claim: There is a useful way of writing the Feynman propagator in terms of a 4momentum integral. where p0 → −i∞. we have no such condition on p0 .

98) where to go to from the second line to the third. If we stay away from the singularities. For some purposes it is also useful to pick other contours which also give rise to Green’s functions. we may instead write the Feynman propagator as ∆F (x − y) = d4 p ie−ip·(x−y) (2π)4 p2 − m2 + iǫ (2.100) Note that we didn’t make use of the contour anywhere in this derivation. Once again we reproduce the Feynman propagator. 2. This way of writing the propagator is. when y 0 > x0 .= d3 p 1 −ip·(x−y) e = D(x − y) (2π)3 2Ep (2. so the integral along the real p0 axis is equivalent to the contour shown in Figure 5. and infinitesimal. This has the effect of shifting the poles slightly off the real axis.97) which is indeed the Feynman propagator for x0 > y 0. we have 2 (∂t − ∇2 + m2 )∆F (x − y) = d4 p i (−p2 + m2 ) e−ip·(x−y) 4 p2 − m2 (2π) d4 p −ip·(x−y) e = −i (2π)4 = −i δ (4) (x − y) (2. we have flipped the sign of p which is valid since we integrate over d3 p and all other quantities depend only on p 2 . for obvious reasons. In contrast.99) + iε Im(p0) Re(p0) −iε with ǫ > 0. we close the contour in an anti-clockwise direction in the upper half plane to get. – 40 – .7. ∆F (x − y) = = = d3 p (2π)4 d3 p (2π)3 d3 p (2π)3 2πi 0 0 i e+iEp (x −y )+ip·(x−y) (−2Ep ) 1 −iEp (y0 −x0 )−ip·(y−x) e 2Ep 1 −ip·(y−x) e = D(y − x) 2Ep (2. called the “iǫ prescription”.2 Green’s Functions Figure 6: There is another avatar of the propagator: it is a Green’s function for the Klein-Gordon operator. Instead of specifying the contour.

2. Let’s look at what this does to our field.102) for some fixed background function J(x). The non-relativistic limit of a particle is |p| ≪ m.Im(p0) Im(p0) Re(p0) −E p +E p −E p +E p Re(p0) Figure 7: The retarded contour Figure 8: The advanced contour For example. there is an obvious name for the current course. Given that next term’s course is called “Advanced Quantum Field Theory”. the retarded Green’s function ∆R (x − y) is defined by the contour shown in Figure 7 which has the property ∆R (x − y) = D(x − y) − D(y − x) 0 x0 > y 0 y 0 > x0 (2.101) The retarded Green’s function is useful in classical field theory if we know the initial value of some field configuration and want to figure out what it evolves into in the presence of a source. Similarly. which is useful if we know the end point of a field configuration and want to figure out where it came from.8 Non-Relativistic Fields Let’s return to our classical complex scalar field obeying the Klein-Gordon equation. we will see that the Feynman Green’s function is most relevant. – 41 – . We’ll decompose the field as ˜ ψ(x. meaning that we want to know the solution to the inhomogeneous Klein-Gordon equation.103) with the m2 term cancelled by the time derivatives. In the quantum theory. one can define the advanced Green’s function ∆A (x − y) which vanishes when y 0 < x0 . t) = e−imt ψ(x. ∂µ ∂ µ φ + m2 φ = J(x) (2. t) Then the KG-equation reads ¨ ˙ 2 ˜ ˜ ˜ ∂t ψ − ∇2 ψ + m2 ψ = e−imt ψ − 2imψ − ∇2 ψ = 0 (2. After a Fourier transform. But it got shot down in the staff meeting.104) (2.

1. We wrote down a Lagrangian in section 1. In order to determine the full trajectory of the field. i (ψ ⋆ ∇ψ − ψ∇ψ ⋆ ) 2m (2. ψ(y)] = [ψ † (x). we drop the term with two time derivatives and the KG equation becomes. In fact.105) This looks very similar to the Schr¨dinger equation for a non-relativistic free particle of o mass m. we need only specify ψ and ψ ⋆ at time t = 0: no time derivatives on the initial slice are required. H= 1 ∇ψ ⋆ ∇ψ 2m (2.108) This means that the momentum conjugate to ψ is iψ ⋆ . Except it doesn’t have any probability interpretation — it’s simply a classical field evolving through an equation that’s first order in time derivatives.2 which gives rise to field equations which are first order in time derivatives.110) – 42 – . ψ † (y)] = δ (3) (x − y) (2. The momentum does not depend on time derivatives at all! This looks a little disconcerting but it’s fully consistent for a theory which is first order in time derivatives. we can derive this from the relativistic Lagrangian for a scalar field by again taking the limit ∂t ψ ≪ mψ. After losing the ˜ ˜ tilde. the non-relativistic Lagrangian becomes 1 ˙ L = +iψ ⋆ ψ − ∇ψ ⋆ ∇ψ 2m (2.107) To move to the Hamiltonian formalism we compute the momentum π= ∂L = iψ ⋆ ˙ ∂ψ (2.¨ ˙ ˜ ˜ this is equivalent to saying that |ψ| ≪ m|ψ|. 1 ˙ Since the Lagrangian already contains a “pq” term (instead of the more familiar 2 pq ˙ term). the time derivatives drop out when we compute the Hamiltonian.109) To quantize we impose (in the Schr¨dinger picture) the canonical commutation relations o [ψ(x). This Lagrangian has a conserved current arising from the internal symmetry ψ → eiα ψ. The current has time and space components jµ = ψ ⋆ ψ.106) where we’ve divided by 1/2m. We get. i ˜ ∂ψ 1 2˜ =− ∇ψ ∂t 2m (2. so ψ → ψ. In this limit. ψ † (y)] = 0 [ψ(x).

110) require † [ap . • A related fact is that the conserved charge Q = d3 x : ψ † ψ : is the particle number. that particle number can change.113) which is the non-relativistic dispersion relation. particle number is conserved.111) where the commutation relations (2.8. and the excitations are ap1 .114) So in non-relativistic theories. In the relativistic theory. the particles are their own anti-particles.112) † † The vacuum satisfies ap |0 = 0. 2. . The existence of anti-particles is a consequence of relativity. How do we get back to quantum mechanics? We already have the operator for the total momentum of the field P = d3 p p a† a (2π)3 p p (2. and the appearance of anti-particles.106) gives rise to non-relativistic particles of mass m. • There is no non-relativistic limit of a real scalar field. It is only with relativity. we talk about the position and momentum operators X and P . . The one-particle states have energy H |p = p2 |p 2m (2. position is relegated to a label. The anti-particle is not in the spectrum. Some comments: • We have a complex field but only a single type of particle. This remains conserved even if we include interactions in the Lagrangian of the form ∆L = V (ψ ⋆ ψ) (2. We conclude that quantizing the first order Lagrangian (2.1 Recovering Quantum Mechanics In quantum mechanics. In quantum field theory.115) – 43 – . and there can be no way to construct a multi-particle theory that conserves particle number.We may expand ψ(x) as a Fourier transform ψ(x) = d3 p a eip·x (2π)3 p (2. apn |0 . aq ] = (2π)3 δ (3) (p − q) (2.

it’s clear that acting with the position operator X has the right action of ϕ(x).118) d3 x x ψ † (x) ψ(x) (2.117) The function ϕ(x) is what we would usually call the Schr¨dinger wavefunction (in the o position representation).120) which tells us that P i acts as the familiar derivative on wavefunctions |ϕ .119) but what about the momentum operator P ? We will now show that P i |ϕ = d3 x −i |x (2. Firstly. It’s also easy to construct the position operator. on one-particle states.116) creates a particle with δ-function localization at x. gives P |p = p |p . X i |ϕ = d3 x xi ϕ(x) |x ∂ϕ ∂xi (2.which. Let’s now construct a state |ϕ by taking superpositions of one-particle states |x . we have P |ϕ = = = i d3 xd3 p † ∂ ap i i e−ip·x ϕ(x) |0 3 (2π) ∂x 3 3 ∂ϕ d xd p −ip·x † −i i ap |0 e 3 (2π) ∂x ∂ϕ d3 x −i i |x ∂x (2. A natural position operator is then X= so that X |x = x |x . We write |x = ψ † (x) |0 . |ϕ = d3 x ϕ(x) |x (2.122) – 44 – . Then the operator ψ † (x) = d3 p † −ip·x a e (2π)3 p (2.121) where we’ve used the relationship [ap . Let’s make sure that it indeed satisfies all the right properties. we write P i |ϕ = = d3 xd3 p i † p ap ap ϕ(x) ψ † (x) |0 (2π)3 d3 xd3 p i † −ip·x ϕ(x) |0 p ap e (2π)3 (2. ψ † (x)] = e−ip·x which can be easily checked. Proceeding with our calculation. Let’s work in the non-relativistic limit. To see that this is the case.

While such Lagrangians are useful in condensed matter physics. despite the name.123) But this is the same equation obeyed by the original field (2.105)! Except this time. – 45 – . It’s important to stress that. the operators X and P act as position and momentum operators in quantum mechanics. it really is the Schr¨dinger equation. complete with the usual probabilistic interpretation o for the wavefunction ϕ.124) coincide caused some confusion.105) and the oneparticle wavefunction (2. It was thought that perhaps we are quantizing the wavefunction itself and the resulting name “second quantization” is still sometimes used today to mean quantum field theory. in practice it’s useful to know that if we treat the one-particle Schr¨dinger equation as the equation for a quantum field then it will give o the correct generalization to multi-particle states.120). where all equations obey translational (and Lorentz) invariance.which confirms (2. t) change in time? The Hamiltonian (2. Note in particular that the conserved charge arising from the Noether current (2. the fact that the equation for the classical field (2. with [X i . we rarely (or never) come across them in high-energy physics.107) is Q = d3 x |ϕ(x)|2 which is the total probability. Historically.124) d3 x 1 ∇ψ ⋆ ∇ψ = 2m d3 p p 2 † a a (2π)3 2m p p (2. 1 ˙ L = iψ ⋆ ψ − ∇ψ ⋆ ∇ψ − V (x) ψ ⋆ ψ 2m (2.109) can be rewritten as H= so we find that i ∂ϕ 1 =− ∇2 ϕ ∂t 2m (2. Interactions Often in quantum mechanics. P j ] |ϕ = iδ ij |ϕ . Nonetheless. This can be easily incorporated into field theory by working with a Lagrangian with explicit x dependence. we’re not quantizing anything twice! We simply quantize a classical field once. So we learn that when acting on one-particle states. But what about dynamics? How does the wavefunction ϕ(x. we’re interested in particles moving in some fixed background potential V (x).125) Note that this Lagrangian doesn’t respect translational symmetry and we won’t have the associated energy-momentum tensor.

126) which. in the quantum theory.One can also consider interactions between particles. In the next section we explore these types of interaction in detail for relativistic theories. both in the non-relativistic and relativistic setting. – 46 – . We therefore expect them to arise from additions to the Lagrangian of the form ∆L = ψ ⋆ (x) ψ ⋆ (x) ψ(x) ψ(x) (2. is an operator which destroys two particles before creating two new ones. Obviously these are only important for n particle states with n ≥ 2. Such terms in the Lagrangian will indeed lead to inter-particle forces.

fall into three different categories • [λ3 ] = 1: For this term. [λn ] = 4 − n (3. [φ] = 1 .2) What does this mean for the Lagrangian (3. They have particle excitations. What restrictions do we have on λn to ensure that the additional terms are small perturbations? You might think that we need simply make “λn ≪ 1”. Firstly. The various terms. We’ll start by asking what kind of small perturbations we can add to the theory. Interacting Fields The free field theories that we’ve discussed so far are very special: we can determine their spectrum. after all. we have [S] = 0. For example. the same dimensions as . Since we’ve set = 1. let’s do some dimensional analysis. Terms that we add to the Lagrangian with this behavior are called relevant because they’re most relevant at low energies (which. equivalently. using the convention described in the introduction. [m] = 1 . note that the action has dimensions of angular momentum or. but nothing interesting then happens. and [d4 x] = −4. E is the energy scale of the process of interest. λn n 1 1 φ L = ∂µ φ ∂ µ φ − m2 φ2 − 2 2 n! n≥3 (3. because this statement only makes sense for dimensionless quantities. – 47 – .3. the dimensionless parameter is λ3 /E. E > m. parameterized by λn . Typically in quantum field theory. where E has dimensions of mass. consider the Lagrangian for a real scalar field.1) The coefficients λn are called coupling constants. With S = d4 x L. But this isn’t quite right. the Lagrangian density must therefore have [L] = 4 (3. we can read off the mass dimensions of all the factors to find. but these particles don’t interact with each other. These will take the form of higher order terms in the Lagrangian. This means that λ3 φ3 /3! is a small perturbation at high energies E ≫ λ3 . but a large perturbation at low energies E ≪ λ3 . so we can always make this perturbation small by taking λ3 ≪ m.1)? Since [∂µ ] = 1.3) So now we see why we can’t simply say we need λn ≪ 1. is where most of the physics we see lies). Here we’ll start to examine more complicated theories that include interaction terms. In a relativistic theory. To see why this is the case.

This is a huge help: of the infinite number of interaction terms that we could write down. This – 48 – . As you’ll see later. only a handful are actually needed (just two in the case of the real scalar field described above). which is small at low-energies and large at high energies. This doesn’t necessarily mean that the theory is useless. only the relevant and marginal couplings are important. This means that we might expect problems with irrelevant operators. What are these interactions? Well. just that it is incomplete at some energy scale. as we’ve seen above. but the same argument applies in that case so bear with me). It is an energy scale. • [λn ] < 0 for n ≥ 5: The dimensionless parameter is (λn E n−4 ). (We’ve already seen a glimpse of this in computing the vacuum energy). λn = gn Λn−4 (3. Let’s further suppose that down at the energy scale E. This scalar field will have some complicated interaction terms (3. so [Λ] = 1. Whatever this scale is. but typically one expects them to be of order 1: gn ∼ O(1). Now we want to understand the laws of physics down at our puny energy scale E ≪ Λ. or the Planck scale. say the GUT scale. where the precise form is dictated by all the stuff that’s going on in the high energy superduper theory. Indeed. the irrelevant couplings become small at low-energies.4) The exact values of dimensionless couplings gn depend on the details of the high-energy superduper theory. Let’s look at this a little more. Such perturbations are called marginal. these lead to “non-renormalizable” field theories in which one cannot make sense of the infinities at arbitrarily high energies. This is basically because. multiplied by a suitable power of the relevant scale Λ.1). the laws of physics are described by a real scalar field. let’s call it Λ.• [λ4 ] = 0: this term is small if λ4 ≪ 1. Such perturbations are called irrelevant. it is typically impossible to avoid high energy processes in quantum field theory. An Important Aside: Why QFT is Simple Typically in a quantum field theory. we could write our dimensionful coupling constants λn in terms of dimensionless couplings gn . (They’re not of course: they’re described by non-Abelian gauge fields and fermions. Suppose that we some day discover the true superduper “theory of everything unimportant” that describes the world at very high energy scales. marginal and irrelevant is not always fixed in stone: quantum corrections can sometimes change the character of an operator. Let me note however that the naive assignment of relevant.

we’ll look at two types of interactions 1) φ4 theory: 1 1 λ L = ∂µ φ∂ µ φ − m2 φ2 − φ4 (3. In this section. the interaction terms of the form φn with n > 4 will be suppressed by powers of (E/Λ)n−4 . we could check that the number operator N now satisfies [H.75) such that [Q. 2) Scalar Yukawa Theory 1 1 (3. E/Mpl ∼ 10−16 ). H] = 0. we see a sum of interactions that look like † † † † ap ap ap ap † † † and ap ap ap ap etc. This suggests that the φ4 Lagrangian describes a theory in which particle number is not conserved.6) These will create and destroy particles.g for the energies E explored at the LHC. we do still have a symmetry rotating the phase of ψ. It is common practice to denote the anti¯ particle as ψ. ensuring the existence of the charge Q defined in (2. that ensures that we need only focus on the first few terms in the interaction: those which are relevant and marginal. (3. We can get a hint for what the effects of this extra term will be. Examples of Weakly Coupled Theories In this course we’ll study only weakly coupled field theories i. because its effects are highly diluted except for the relevant and marginal interactions. It is this simple argument. This means that the number of ψ particles minus the number of ψ anti-particles is conserved.5) 2 2 4! with λ ≪ 1. (e. – 49 – . it’s going to be very difficult to figure out the high energy theory (which it is!).e. While the individual particle numbers of ψ and φ are no longer conserved. This is usually a suppression by many orders of magnitude. m.means that for experiments at small energies E ≪ Λ.7) L = ∂µ ψ ⋆ ∂ µ ψ + ∂µ φ∂ µ φ − M 2 ψ ⋆ ψ − m2 φ2 − gψ ⋆ ψφ 2 2 with g ≪ M. Indeed. N] = 0. This theory couples a complex scalar ψ to a real scalar φ. about which you can learn more in the “Statistical Field Theory” course. ones that can truly be considered as small perturbations of the free field theory at all energies. Expanding † out φ4 in terms of ap and ap . It also means that if we only have access to low-energy experiments (which we do!). The discussion given above is a poor man’s version of the ideas of effective field theory and Wilson’s renormalization group. based on dimensional analysis.

The scalar Yukawa theory has a slightly worrying aspect: the potential has a stable local minimum at φ = ψ = 0, but is unbounded below for large enough −gφ. This means we shouldn’t try to push this theory too far. A Comment on Strongly Coupled Field Theories In this course we restrict attention to weakly coupled field theories where we can use perturbative techniques. The study of strongly coupled field theories is much more difficult, and one of the major research areas in theoretical physics. For example, some of the amazing things that can happen include • Charge Fractionalization: Although electrons have electric charge 1, under the right conditions the elementary excitations in a solid have fractional charge 1/N (where N ∈ 2Z + 1). For example, this occurs in the fractional quantum Hall effect. • Confinement: The elementary excitations of quantum chromodynamics (QCD) are quarks. But they never appear on their own, only in groups of three (in a baryon) or with an anti-quark (in a meson). They are confined. • Emergent Space: There are field theories in four dimensions which at strong coupling become quantum gravity theories in ten dimensions! The strong coupling effects cause the excitations to act as if they’re gravitons moving in higher dimensions. This is quite extraordinary and still poorly understood. It’s called the AdS/CFT correspondence. 3.1 The Interaction Picture There’s a useful viewpoint in quantum mechanics to describe situations where we have small perturbations to a well-understood Hamiltonian. Let’s return to the familiar ground of quantum mechanics with a finite number of degrees of freedom for a moment. In the Schr¨dinger picture, the states evolve as o i d|ψ S = H |ψ dt
S

(3.8)

while the operators OS are independent of time. In contrast, in the Heisenberg picture the states are fixed and the operators change in time OH (t) = eiHt OS e−iHt |ψ
H

= eiHt |ψ

S

(3.9)

– 50 –

The interaction picture is a hybrid of the two. We split the Hamiltonian up as H = H0 + Hint (3.10)

The time dependence of operators is governed by H0 , while the time dependence of states is governed by Hint . Although the split into H0 and Hint is arbitrary, it’s useful when H0 is soluble (for example, when H0 is the Hamiltonian for a free field theory). The states and operators in the interaction picture will be denoted by a subscript I and are given by, |ψ(t)
I

OI (t) = e

= eiH0 t |ψ(t)
iH0 t

OS e

S −iH0 t

(3.11)

This last equation also applies to Hint , which is time dependent. The interaction Hamiltonian in the interaction picture is, HI ≡ (Hint )I = eiHo t (Hint )S e−iH0 t (3.12)

The Schr¨dinger equation for states in the interaction picture can be derived starting o from the Schr¨dinger picture o i d|ψ S = HS |ψ dt
S

⇒ ⇒

i

d e−iH0 t |ψ I = (H0 + Hint )S e−iH0 t |ψ dt d|ψ I = eiH0 t (Hint )S e−iH0 t |ψ I i dt

I

(3.13)

So we learn that i 3.1.1 Dyson’s Formula “Well, Birmingham has much the best theoretical physicist to work with, Peierls; Bristol has much the best experimental physicist, Powell; Cambridge has some excellent architecture. You can make your choice.” Oppenheimer’s advice to Dyson on which university position to accept. We want to solve (3.14). Let’s write the solution as |ψ(t)
I

d|ψ I = HI (t) |ψ dt

I

(3.14)

= U(t, t0 ) |ψ(t0 )

I

(3.15)

– 51 –

where U(t, t0 ) is a unitary time evolution operator such that U(t1 , t2 )U(t2 , t3 ) = U(t1 , t3 ) and U(t, t) = 1. Then the interaction picture Schr¨dinger equation (3.14) requires that o dU = HI (t) U dt If HI were a function, then we could simply solve this by i U(t, t0 ) = exp −i
? t

(3.16)

HI (t′ ) dt′
t0

(3.17)

But there’s a problem. Our Hamiltonian HI is an operator, and we have ordering issues. Let’s see why this causes trouble. The exponential of an operator is defined in terms of the expansion,
t t

exp −i

HI (t ) dt
t0

= 1−i

t0

(−i)2 HI (t ) dt + 2
′ ′

t

2

HI (t ) dt
t0

+ . . .(3.18)

But when we try to differentiate this with respect to t, we find that the quadratic term gives us − 1 2
t t0

1 HI (t′ ) dt′ HI (t) − HI (t) 2

t

HI (t′ ) dt′
t0

(3.19)

Now the second term here looks good, since it will give part of the HI (t)U that we need on the right-hand side of (3.16). But the first term is no good since the HI (t) sits the wrong side of the integral term, and we can’t commute it through because [HI (t′ ), HI (t)] = 0 when t′ = t. So what’s the way around this? Claim: The solution to (3.16) is given by Dyson’s Formula. (Essentially first figured out by Dirac, although the compact notation is due to Dyson).
t

U(t, t0 ) = T exp −i

HI (t′ ) dt′
t0

(3.20)

where T stands for time ordering where operators evaluated at later times are placed to the left T (O1 (t1 ) O2 (t2 )) = O2 (t2 ) O1 (t1 )
t t

O1 (t1 ) O2 (t2 )

t1 > t2 t2 > t1

(3.21)

Expanding out the expression (3.20), we now have
t

U(t, t0 ) = 1 − i

dt′ HI (t′ ) +
t0

(−i)2 2

dt′
t0 t′ t

dt′′ HI (t′′ )HI (t′ )
t′

+
t0

dt

′ t0

dt′′ HI (t′ )HI (t′′ ) + . . .

– 52 –

I should confess that Dyson’s formula is rather formal. This means that we can write t t t′ U(t.24) since t. . To see why this is. (3. It is typically very hard to compute time ordered exponentials in practice. The power of the formula comes from the expansion which is valid when HI is small and is very easily computed. . we use – 53 – . t0 ) = 1 − i dt HI (t ) + (−i) t0 ′ ′ 2 t0 dt ′ t0 dt′′ HI (t′ )HI (t′′ ) + . of course. Before moving on. Thus i ∂ T exp −i ∂t t t dt′ HI (t′ ) t0 = T HI (t) exp = HI (t) T exp −i −i dt′ HI (t′ ) t0 t dt′ HI (t′ ) t0 (3. Hint = g d3 x ψ † ψφ (3. the same thing.22) where the range of integration in the first expression is over t′′ ≥ t′ . all operators commute (since their order is already fixed by the T sign).2 A First Look at Scattering Let us now apply the interaction picture to field theory. being the upper limit of the integral. allowing particles of one type to morph into others. The final expression is the same as the second expression by a simple relabelling. starting with the interaction Hamiltonian for our scalar Yukawa theory. 3.23) Proof: The proof of Dyson’s formula is simpler than explaining what all the notation means! Firstly observe that under the T sign. this interaction doesn’t conserve particle number.25) Unlike the free theories discussed in Section 2.Actually these last two terms double up since t t t t′′ dt t0 ′ t′ dt HI (t )HI (t ) = t0 t ′′ ′′ ′ dt ′′ t0 t′ dt′ HI (t′′ )HI (t′ ) dt′′ HI (t′ )HI (t′′ ) t0 = t0 dt′ (3. is the latest time so HI (t) can be pulled out to the left. while in the second expression it is t′ ≤ t′′ which is.

• ψ ∼ b + c† : This operator can destroy ψ particles through b. This kills a meson. which commute with H0 . we’ll have more complicated terms in (Hint )2 . and slightly dodgy. As the particles approach each other.20) which is an expansion in powers of Hint . (S is for scattering). and the final state |f at t → +∞. Of course. • ψ † ∼ b† + c: This operator can create nucleons through b† . Q = Nc − Nb remains conserved in the presence of Hint . But Hint contains creation and annihilation operators for each type of particle. each going on its own merry way. t0 ) is given by Dyson’s formula (3. Let’s call them mesons. assumption: Initial and final states are eigenstates of the free theory This means that we take the initial state |i at t → −∞. This term will give contributions to scattering processes ¯ ¯ ψ ψ → φ → ψ ψ. To calculate amplitudes we make an important. At second order in perturbation theory. and don’t arise from the quantization of a scalar field. t− ) |i ≡ f | S |i (3. but not Hint . Furthermore. The amplitude to go from |i to |f is t± →±∞ lim f | U(t+ . and create antiparticles through c† . • φ ∼ a + a† : This operator can create or destroy φ particles. to be eigenstates of the free Hamiltonian H0 . this sounds plausible: at t → −∞. for example (c† b† a)(cba† ).the interaction picture and follow the evolution of the state: |ψ(t) = U(t. There are a number of reasons why the assumption of non-interacting initial and final states is shaky: – 54 – . In particular. But we’ll treat our scalar Yukawa theory as a toy model for nucleons interacting with mesons.26) where the unitary operator S is known as the S-matrix. At some level. It will contribute to meson decay φ → ψ ψ. in reality nucleons are spin 1/2 particles. they interact briefly. At first order in perturbation theory. where U(t. producing a ¯ nucleon-anti-nucleon pair. t0 ) |ψ(t0 ) . Importantly. we find terms in Hint like c† b† a. The rest of this section is devoted to computing the quantum amplitudes for these processes to occur. Let’s call these particles nucleons. before departing again. the particles in a scattering process are far separated and don’t feel the effects of each other. and destroy antinucleons through c. we intuitively expect these states to be eigenstates of the individual number operators N.

28) Let’s go slowly.2.1 An Example: Meson Decay Consider the relativistically normalized initial and final states. So we have f | S |i = −ig f | = −ig f | d4 x ψ † (x)ψ(x) d3 k (2π)3 2Ep 2Ek † ak ap e−ik·x |0 d4 x ψ † (x)ψ(x)e−ip·x |0 (3. We first expand out φ ∼ a + a† using (2. But the two meson state will have zero overlap with f |. a long way from its neighbors. a related fact is that there is a cloud of virtual photons surrounding the electron. motivated by this problem. it is f | S |i = −ig f | d4 x ψ † (x)ψ(x)φ(x) |i (3.27) The initial state contains a single meson of momentum p. (Remember that the φ in this formula is in the interaction picture. where the electron sources the electromagnetic field from which it can never escape.84). is never alone in field theory. we’ll mention a better way. We may compute the amplitude for the decay of a meson to a nucleon-anti-nucleon pair. This line of thought gets us into the issues of renormalization — more on this next term in the “AQFT” course. the final state contains a nucleon-anti-nucleon pair of momentum q1 and q2 . and there’s nothing in the ψ and ψ † operators that lie between them to change this fact. |i = † 2Ep ap |0 |f = † † 4Eq1 Eq2 bq1 cq2 |0 (3. in the second line. picking up a δ (3) (p − k) deltafunction which kills the d3 k integral. For example. • More importantly.29) † where. this formalism can’t describe the scattering of an electron and proton which collide. after developing scattering theory using the assumption of noninteracting asymptotic states. In quantum electrodynamics (QED). bind. we’ve commuted ak past ap . The a piece will turn |i into something proportional to |0 . a single particle. It’s possible to circumvent this objection since it turns out that bound states show up as poles in the S-matrix.• Obviously we can’t cope with bound states. and leave as a Hydrogen atom. which is the same as the Heisenberg picture of the free theory). while the a† piece will turn |i into a two meson state. We now similarly expand out ψ ∼ b + c† and – 55 – . To leading order in g. Nevertheless. This is true even in classical electrodynamics. 3.

31) (3. We’ll explain how to deal with this in Section 3. and again in next term’s “Standard Model” course. 3. .32) – 56 – . The ordering of the operators is fixed by T . for they create the nucleon and anti-nucleon from |0 . HI (xn )} |i . To get non-zero overlap with f |. we want to compute quantities like f | T {HI (x1 ) . In particular. 0. . Notice that the δ-function puts constraints on the possible decays.1 An Example: Recovering the Propagator Let’s start simple. Then the delta function imposes momentum conservation. our life will be much simpler if we can start to move all annihilation operators to the right where they can start killing things in |i .30) = −ig (2π)4 δ (4) (q1 + q2 − p) and so we get our first quantum field theory amplitude. telling us that q1 = −q2 and m = 2 M 2 + |q|2 .3 Wick’s Theorem From Dyson’s formula. only the b† and c† contribute. the decay only happens at all if m ≥ 2M. 0. we may always boost ourselves to a reference frame where the meson is stationary. We then have f | S |i = −ig 0| d4 xd3 k1 d3 k2 (2π)6 Eq1 Eq2 c b c† b† |0 ei(q1 +q2 −p)·x Ek1 Ek2 q2 q1 k1 k2 (3. However.ψ † ∼ b† + c. Wick’s theorem tells us how to go from time ordered products to normal ordered products. 3.3.6 below. since the HI ’s contain certain creation and annihilation operators. where |i and |f are eigenstates of the free theory. time ordering. Consider a real scalar field which we decompose in the Heisenberg picture as φ(x) = φ+ (x) + φ− (x) where φ+ (x) = φ− (x) = d3 p (2π)3 d3 p (2π)3 1 ap e−ip·x 2Ep 1 † ap e+ip·x 2Ep (3. namely the lifetime of the meson. Recall that this is the definition of normal ordering. 0). The reason this is a little tricky is that we must square the amplitude to get the probability for decay. so p = (m. To see this. Later you will learn how to turn this quantum amplitude into something more physical. which means we get the square of a δ-function.

we may repeat the calculation to find T φ(x)φ(y) =: φ(x)φ(y) : + D(y − x) So putting this together. φ(x)φ(y) = ∆F (x − y) (3. . for which we have the integral representation ∆F (x − y) = ieik·(x−y) d4 k (2π)4 k 2 − m2 + iǫ (3. but apparently you have Pauli and Heisenberg to blame. . . φ(x2 ) . Definition: We define the contraction of a pair of fields in a string of operators . . to mean replacing those operators with the Feynman propagator. for example. ∆F (x − y). . We use the notation. to denote contraction. .where the ± signs on φ± make little sense.36) (3. Then choosing x0 > y 0 . leaving all other operators untouched.33) = φ+ (x)φ+ (y) + φ− (x)φ+ (y) + φ− (y)φ+(x) + [φ+ (x).93). φ(x2 ) . .39) (3.37) Let me reiterate a comment from Section 2: although T φ(x)φ(y) and : φ(x)φ(y) : are both operators.34) where ∆F (x − y) is the Feynman propagator defined in (2. So. and for our troubles we have picked up a factor of D(x − y) = [φ+ (x). .35) (3. . for y 0 > x0 . . So for x0 > y 0 we have T φ(x)φ(y) =: φ(x)φ(y) : + D(x − y) Meanwhile. the difference between them is a c-number function. φ(x1 ) .90). . we have T φ(x)φ(y) = φ(x)φ(y) = (φ+ (x) + φ− (x))(φ+ (y) + φ− (y)) (3. . (They come about because φ+ ∼ e−iEt . φ(x1 ) . φ− (y)] which is the propagator we met in (2. which is sometimes called the positive frequency piece. .38) – 57 – . φ− (y)] + φ− (x)φ− (y) where the last line is normal ordered. we have the final expression T φ(x)φ(y) =: φ(x)φ(y) : + ∆F (y − x) (3. while φ− ∼ e+iEt is the negative frequency piece).

.A similar discussion holds for complex scalar fields.45) We can then look at the expansion of f | S |i . φn and now add φ1 .40) To see what the last part of this equation means. φn ) = (φ+ + φ− ) (: φ2 . . Suppose it’s true for φ2 . . But in order to write 1 the right-hand side as a normal ordered product. For n = 4. . It’s true for n = 2. . In fact. we have T (φ1 . .42) ψ(x)ψ(y) = ψ † (x)ψ † (y) = 0 (3.3 An Example: Nucleon Scattering Let’s look at ψψ → ψψ scattering. φ2 = φ(x2 ). Each time it moves past φ− . . n. the equation reads T (φ1 φ2 φ3 φ4 ) = : φ1 φ2 φ3 φ4 : + φ1 φ2 : φ3 φ4 : + φ1 φ3 : φ2 φ4 : + four similar terms + φ1 φ2 φ3 φ4 + φ1 φ3 φ2 φ4 + φ1 φ4 φ2 φ3 (3. we really want to calculate f | S − 1 |i since we’re not interested in situations where no scattering occurs.2 Wick’s Theorem For any collection of fields φ1 = φ(x1 ). (Try it!) 3. . we pick up a factor of k k φ1 φk = ∆F (x1 − xk ) from the commutator. etc.44) The φ− term stays where it is since it is already normal ordered. let’s look at an example. writing T (φ1 φ2 .46) – 58 – .43) Proof: The proof of Wick’s theorem proceeds by induction and a little thought. φn : + : contractions :) 1 1 (3. We have T ψ(x)ψ † (y) =: ψ(x)ψ † (y) : +∆F (x − y) prompting us to define the contraction ψ(x)ψ † (y) = ∆F (x − y) and 3.3. . . . φn : + : all possible contractions : (3. φn ) =: φ1 . . the φ+ term has to make its way 1 past the crowd of φ− operators. Then we can pull φ1 out to the left of the time ordered product. We have the initial and final states |i = |f = 2Ep1 2Ep′1 † † 2Ep2 bp1 bp2 |0 ≡ |p1 . p2 1 2 (3.3. We’ll take x0 > x0 1 k for all k = 2. p2 † † ′ ′ 2Ep2′ bp ′ bp ′ |0 ≡ |p1 . At order g 2 we have the term (−ig)2 2 d4 x1 d4 x2 T ψ † (x1 )ψ(x1 )φ(x1 )ψ † (x2 )ψ(x2 )φ(x2 ) (3. . .41) (3.

by – 59 – . We’re left with the expression (−ig)2 i(2π)8 d4 k (2π)4 k 2 − m2 + iǫ δ (4) (p′1 − p1 + k) δ (4) (p′2 − p2 − k) + δ (4) (p′2 − p1 + k) δ (4) (p′1 − p2 − k) (3. (3. in going to the third line.50) where the expression in square brackets is (3. p2 ′ ′ ′ ′ = p′1 . using Wick’s theorem we see there is a piece in the string of operators which looks like : ψ † (x1 )ψ(x1 )ψ † (x2 )ψ(x2 ) : φ(x1 )φ(x2 ) (3..47) which will contribute to the scattering because the two ψ fields annihilate the ψ particles.48) = eix1 ·(p1 −p1 )+ix2 ·(p2 −p2 ) + eix1 ·(p2 −p1 )+ix2 ·(p1 −p2 ) + (x1 ↔ x2 ) where. where p1 = −p2 and. while the two ψ † fields create ψ particles. 0| ψ(x) |p = e−ip·x Now let’s insert this into (3. + (x1 ↔ x2 ) d4 k ieik·(x1 −x2 ) (2π)4 k 2 − m2 + iǫ (3. This means that we have p′1 ..46).47). Meanwhile. the x1 and x2 integrals give delta-functions. (−ig)2 2 d x1 d x2 e 4 4 i. we’ve used the fact that for relativistically normalized states.48). Any other way of ordering the ψ and ψ † fields will give zero contribution. p′2 | ψ † (x1 )ψ † (x2 ) |0 0| ψ(x1 )ψ(x2 ) |p1 .. To see this. p′2 | : ψ † (x1 )ψ(x1 )ψ † (x2 )ψ(x2 ) : |p1 . to get the expression for f | S |i at order g 2 . we can go to the center of mass frame. Now the (x1 ↔ x2 ) terms double up with the others to cancel the factor of 1/2 out front.. for this process we may drop the +iǫ terms since the denominator is never zero. p2 = eip1 ·x1 +ip2 ·x2 + eip1 ·x2 +ip2 ·x1 ′ ′ ′ ′ e−ip1 ·x1 −ip2 ·x2 + e−ip1 ·x2 −ip2 ·x1 (3.49) +e i.51) Finally. while the final integral is the φ propagator which comes from the contraction in (3. we can trivially do the d4 k integral using the delta-functions to get i(−ig)2 (p1 − ′ p 1 )2 1 1 ′ ′ (2π)4 δ (4) (p1 + p2 − p1 − p2 ) + ′ 2 2 + iǫ 2 + iǫ −m (p1 − p2 ) − m In fact.Now.

52) We will see another. The various terms in the perturbative expansion can be represented pictorially as follows • Draw an external line for each particle in the initial state |i and each particle in the final state |f .4 Feynman Diagrams “Like the silicon chips of more recent years. to actually compute scattering amplitudes using Wick’s theorem is rather tedious. this term also contributes to ¯ ¯ ¯ similar scattering processes. add an arrow to – 60 – . Each of these comes from the term (3. We therefore have the end result. This will also shed light on the physical interpretation. Another Example: Meson-Nucleon Scattering If we want to compute ψφ → ψφ scattering at order g 2. and solid lines for nucleons. for this would violate the conservation of Q charge. However. i(−ig)2 1 1 + ′ 2 ′ 2 (p1 − p1 ) − m (p1 − p2 )2 − m2 (2π)4 δ (4) (p1 + p2 − p′1 − p′2 ) (3. Once more. These pictures represent the expansion of f | S |i and we will learn how to associate numbers (or at least integrals) to them. ¯ ¯ ψ ψ → ψ ψ. Assign a directed momentum p to each line. the Feynman diagram was bringing computation to the masses.48) in Wick’s theorem. including ψφ → ψφ and φφ → ψ ψ. we would need to pick out the term : ψ † (x1 )φ(x1 )ψ(x2 )φ(x2 ) : ψ(x1 )ψ † (x2 ) (3. much simpler way to reproduce this result shortly using Feynman diagrams. so k 2 < 0. p − p ′ ). ¯¯ ¯¯ This calculation is also relevant for other scattering processes. |p1 | = |p1′ |. we ¯¯ will never find a term that contributes to scattering ψψ → ψ ψ. Further. such as ψ ψ → ψ ψ. These pictures are called Feynman diagrams. There’s a much better way. since we’re not interested in processes where no scattering occurs. The object that we really want to compute is f | S −1 |i .” Julian Schwinger As the above example demonstrates. This ensures that the 4-momentum of the meson is k = (0. It requires drawing pretty pictures. 3.momentum conservation. We’ll choose dotted lines for mesons.53) and a similar term with ψ and ψ † exchanged.

4. we write down a factor of i d4 k 4 k 2 − m2 + iǫ (2π) (3. 3. with m replaced by the nucleon mass M. corresponding to a φ particle with momentum k. where an outgoing arrow denotes ψ.solid lines to denote its charge. • For each internal dotted line. using the Feynman rules • Add a momentum k to each internal line • To each vertex. write down a factor of (−ig) (2π)4 δ (4) ( i ki ) (3. ψ • Join the external lines together with trivalent vertices φ ψ+ Each such diagram you can draw is in 1-1 correspondence with the terms in the expansion of f | S − 1 |i .54) where ki is the sum of all momenta flowing into the vertex. – 61 – .55) We include the same factor for solid internal ψ lines.1 Feynman Rules To each diagram we associate a number. We choose the reverse convention for the final state. we’ll choose an incoming (outgoing) arrow in the ¯ initial state for ψ (ψ).

each of these diagrams translates into an integral that we will not attempt to calculate here. Using the Feynman rules. of the nucleons exchanging a meson which. p 1 p p1 k / 1 p k / 2 + p2 p2 p2 / p / 1 Figure 9: The two lowest order Feynman diagrams for nucleon scattering. sometimes. They are shown in Figure 9. we have the two diagrams shown in Figures 10 and 11.5 Examples of Scattering Amplitudes Let’s apply the Feynman rules to compute the amplitudes for various processes. and similar diagrams with p′1 and p′2 exchanged. in the first diagram.56) which agrees with the calculation (3. because k 2 = m2 : the meson is called a virtual particle and is said to be off-shell (or. This meson doesn’t satisfy the usual energy dispersion relation. One final note: the addition of the two diagrams above ensures that the particles satisfy Bose statistics. – 62 – . nucleon legs all satisfy p2 = M 2 . We can write down the two simplest diagrams contributing to this process.3. For example. There is a nice physical interpretation of these diagrams. In contrast. Applying the Feynman rules to these diagrams. Heuristically. We start with something familiar: Nucleon Scattering Revisited Let’s look at how this works for the ψψ → ψψ scattering at order g 2 . all appearing at higher order in the coupling constant g. we get i(−ig)2 1 1 + ′ 2 ′ 2 (p1 − p1 ) − m (p1 − p2 )2 − m2 (2π)4 δ (4) (p1 + p2 − p′1 − p′2 ) (3. off massshell). has momentum k = (p1 −p′1 ) = (p2 −p′2 ). rather loosely. We talk. There are also more complicated diagrams which will contribute to the scattering process at higher orders. with increasingly complicated diagrams. the momentum on the external. the mass of the nucleon. They are on-shell. And so we go on. it can’t live long enough for its energy to be measured to great accuracy.51) that we performed earlier.

We can now refine our Feynman rules to compute the amplitude iAf i itself: • Draw all possible diagrams with appropriate external legs and impose 4-momentum conservation at each vertex. in turn. They are tree level diagrams. This last step deserves a short explanation. f | S − 1 |i = i Af i (2π)4 δ (4) (pF − pI ) (3. and the factor of i out front is a convention which is there to match non-relativistic quantum mechanics. • For each internal line.p 1 p p1 / 1 p1 / p2 p2 p2 / p2 / Figure 10: A contribution at O(g4 ). For diagrams with loops. Figure 11: A contribution at O(g6 ) Amplitudes Our final result for the nucleon scattering amplitude f | S − 1 |i at order g 2 was i(−ig)2 1 1 + ′ 2 ′ 2 (p1 − p1 ) − m (p1 − p2 )2 − m2 ′ ′ (2π)4 δ (4) (p1 + p2 − p1 − p2 ) The δ-function follows from the conservation of 4-momentum which. It is common to all S-matrix elements. • Write down a factor of (−ig) at each vertex. It’s not hard to convince yourself that in tree diagrams. The diagrams we’ve computed so far have no loops. – 63 – .57) where pI (pF ) is the sum of the initial (final) 4-momenta. momentum conservation at each vertex is sufficient to determine the momentum flowing through each internal line. write down the propagator • Integrate over momentum k flowing through each loop d4 k/(2π)4 . follows from spacetime translational invariance. We will define the amplitude Af i by stripping off this momentum-conserving delta-function. this is no longer the case. such as those shown in Figures 10 and 11.

they are given by the diagrams of Figure 13. This fact is reflected in the denominator of the amplitudes which are given by iA = (−ig)2 i (p1 − ′ p 1 )2 As in (3. Nucleon to Meson Scattering Let’s now look at the amplitude for a nucleon-anti-nucleon pair to annihilate into a ¯ pair of mesons: ψ ψ → φφ.58) p 1 1 p1 / + p2 p2 p2 / / p2 Figure 13: The two lowest order Feynman diagrams for nucleon-anti-nucleon scattering.52). ψ ψ → ψ ψ. ¯ ¯ For the scattering of a nucleon and an anti-nucleon. iA = (−ig)2 i (p1 − ′ p 1 )2 − m2 + (p1 + p2 )2 i − m2 + iǫ (3. It is a simple matter to write down the amplitude using the Feynman rules.59) – 64 – . Nucleon-Anti-Nucleon Scattering p p1 / − M2 + i (p1 − ′ p 2 )2 − M2 (3. we’ve dropped the iǫ from the propagators as the denominator never vanishes. the Feynman diagrams are a little different. The simplest Feynman diagrams for this process are shown in Figure 12 where the virtual particle in these diagrams is now the nucleon ψ rather than the meson φ. At lowest order.p 1 p p1 / 1 p / 2 + p2 p2 p2 / p / 1 Figure 12: The two lowest order Feynman diagrams for nucleon to meson scattering.

if m > 2M. although for a different reason: the meson is unstable when m > 2M. The amplitude corresponding to the diagram shown in the figure is (−ig)4 p k + p1 / 1 p1 / / k + p1 − p1 k p2 p2 k − p2 / / Figure 15: 1 d4 k 4 (k 2 − M 2 + iǫ)((k + p′ )2 − M 2 + iǫ) (2π) 1 1 × ′ 2 − M 2 + iǫ)((k − p ′ )2 − M 2 + iǫ) ((k + p1 − p1 ) 2 d4 k/k 8 . All other momenta are then determined. reflecting the different Feynman diagram that contributes to the process. In the center of mass frame. If m < 2M. But this won’t always be the case! – 65 – . When correctly treated. the simplest diagram we can write down has a single loop. the denominator of the sec2 ond term is 4(M 2 + p1 ) − m2 . In this case it turns out that we may also neglect the iǫ term. the Figure 14 shows the cross-section (roughly the amplitude squared) plotted vertically for e+ e− → µ+ µ− scattering from the ALEPH experiment in CERN. which is at These integrals can be tricky. a result we derived in Figure 14: (3. then the amplitude corresponding to the second diagram diverges at some value of p.30). In contrast. this instability adds a finite imaginary piece to the denominator which overwhelms the iǫ. The horizontal axis shows the center of mass energy. We choose to assign the single undetermined momentum k to the right-hand propagator. the increase in the scattering amplitude which we see in the second diagram when 4(M 2 + p 2 ) = m2 is what allows us to discover new particles: they appear as a resonance in the cross section. For large k. the mass of the Z-boson. The curve rises sharply around 91 GeV.Notice that the momentum dependence in the second term is different from that of nucleon-nucleon scattering (3. For example. then this term never vanishes and we may drop the iǫ.56). p1 = −p2 . Nonetheless. and momentum conservation at each vertex is no longer sufficient to determine every momentum passing through the diagram. Meson Scattering For φφ → φφ. this integral goes as least convergent as k → ∞.

5. Now the amplitudes that involve exchange of a single particle can be written simply in terms of the Mandelstam variables. the amplitude (3. For example. 0. When all the masses are the same we have s + t + u = 4M 2 . the amplitude (3. When the masses of all 4 particles differ. −p cos θ) Then from the above definitions. We can define these variables whether the particles involved in the scattering are the same or different. They are s = (p1 + p2 )2 = (p′1 + p′2 )2 u = (p1 − t = (p1 − p′1 )2 = (p2 − p′2 )2 p′2 )2 = (p2 − (3. 0. 0.3. for nucleon-nucleon scattering. while the variables t and u are measures of the momentum exchanged between particles. p cos θ) and p′2 = (E. p sin θ. −p sin θ.1 Mandelstam Variables We see that in many of the amplitudes above — in particular those that include the exchange of just a single particle — the same combinations of momenta are appearing frequently in the denominators. so that the initial two particles have four-momenta p1 = (E. 0. We say that the first case involves “t-channel” and “u-channel” diagrams. (The first diagram indeed includes a vertex that looks like the letter “T”). To get a feel for what these variables mean. −p) The particles then scatter at some angle θ and leave with momenta p′1 = (E. as in the examples above.62) (3.61) The variable s measures the total center of mass energy of the collision.56) is schematically A ∼ (t − m2 )−1 + (u − m2 )−1 .59) is A ∼ (t − m2 )−1 + (s − m2 )−1 . We sit in the center of mass frame. p1 and p2 are the momenta of the two initial particles. Meanwhile the nucleon-antinucleon scattering is said to involve “t-channel” and “s-channel” diagrams. There are standard names for various sums and differences of momenta: they are known as Mandelstam variables.60) p′1 )2 where. we have that s = 4E 2 and t = −2p2 (1 − cos θ) and u = −2p2 (1 + cos θ) (3. (They are basically equivalent. p) and p2 = (E. For the nucleon-anti-nucleon scattering. Note that there is a relationship between the Mandelstam variables. – 66 – .63) (3. this becomes s + t + u = i Mi2 . ′ and p1 and p′2 are the momenta of the final two particles. let’s assume all four particles are the same. 0. 0. just with the outgoing particles swapped around).

3.5.2 The Yukawa Potential So far we’ve computed the quantum amplitudes for various scattering processes. But these quantities are a little abstract. In Section 3.6 below (and again in next term’s “Standard Model” course) we’ll see how to turn amplitudes into measurable quantities such as cross-sections, or the lifetimes of unstable particles. Here we’ll instead show how to translate the amplitude (3.52) for nucleon scattering into something familiar from Newtonian mechanics: a potential, or force, between the particles. Let’s start by asking a simple question in classical field theory that will turn out to be relevant. Suppose that we have a fixed δ-function source for a real scalar field φ, that persists for all time. What is the profile of φ(x)? To answer this, we must solve the static Klein-Gordon equation, −∇2 φ + m2 φ = δ (3) (x) We can solve this using the Fourier transform, φ(x) = d3 k ik·x ˜ e φ(k) (2π)3 (3.65) (3.64)

˜ Plugging this into (3.64) tells us that (k 2 + m2 )φ(k) = 1, giving us the solution φ(x) = d3 k eik·x (2π)3 k 2 + m2 (3.66)

Let’s now do this integral. Changing to polar coordinates, and writing k · x = kr cos θ, we have φ(x) =
∞ 1 k2 2 sin kr dk 2 2 2 (2π) 0 k +m kr +∞ 1 k sin kr = dk 2 (2π)2 r −∞ k + m2 +∞ 1 dk keikr = Re 2 2 2πr −∞ 2πi k + m

(3.67)

We compute this last integral by closing the contour in the upper half plane k → +i∞, picking up the pole at k = +im. This gives φ(x) = 1 −mr e 4πr (3.68)

The field dies off exponentially quickly at distances 1/m, the Compton wavelength of the meson.

– 67 –

Now we understand the profile of the φ field, what does this have to do with the force between ψ particles? We do very similar calculations to that above in electrostatics where a charged particle acts as a δ-function source for the gauge potential: −∇2 A0 = δ (3) (x), which is solved by A0 = 1/4πr. The profile for A0 then acts as the potential energy for another charged (test) particle moving in this background. Can we give the same interpretation to our scalar field? In other words, is there a classical limit of the scalar Yukawa theory where the ψ particles act as δ-function sources for φ, creating the profile (3.68)? And, if so, is this profile then felt as a static potential? The answer is essentially yes, at least in the limit M ≫ m. But the correct way to describe the potential felt by the ψ particles is not to talk about classical fields at all, but instead work directly with the quantum amplitudes. Our strategy is to compare the nucleon scattering amplitude (3.52) to the corresponding amplitude in non-relativistic quantum mechanics for two particles interacting through a potential. To make this comparison, we should first take the non-relativistic limit of (3.52). Let’s work in the center of mass frame, with p ≡ p1 = −p2 and p ′ ≡ p1′ = −p2′ . The non-relativistic limit means |p| ≪ M which, by momentum conservation, ensures that |p ′ | ≪ M. In fact one can check that, for this particular example, this limit doesn’t change the scattering amplitude (3.52): it’s given by iA = +ig 2 1 (p − p ′ )2 + m2 + 1 (p + p ′ )2 + m2 (3.69)

How do we compare this to scattering in quantum mechanics? Consider two particles, separated by a distance r, interacting through a potential U(r). In non-relativistic quantum mechanics, the amplitude for the particles to scatter from momentum states ±p into momentum states ±p ′ can be computed in perturbation theory, using the techniques described in Section 3.1. To leading order, known in this context as the Born approximation, the amplitude is given by p ′ | U(r) |p = −i d3 r U(r)e−i(p−p )·r

(3.70)

There’s a relative factor of (2M)2 that arises in comparing the quantum field theory amplitude A to p ′ | U(r) |p , that can be traced to the relativistic normalization of the states |p1 , p2 . (It is also necessary to get the dimensions of the potential to work out correctly). Including this factor, and equating the expressions for the two amplitudes, we get d3 r U(r) e−i(p−p )·r =

−λ2 (p − p ′ )2 + m2

(3.71)

– 68 –

where we’ve introduced the dimensionless parameter λ = g/2M. We can trivially invert this to find, U(r) = −λ2 eip·r d3 p (2π)3 p 2 + m2 (3.72)

But this is exactly the integral (3.66) we just did in the classical theory. We have −λ2 −mr U(r) = e 4πr (3.73)

This is the Yukawa potential. The force has a range 1/m, the Compton wavelength of the exchanged particle. The minus sign tells us that the potential is attractive. Notice that quantum field theory has given us an entirely new perspective on the nature of forces between particles. Rather than being a fundamental concept, the force arises from the virtual exchange of other particles, in this case the meson. In Section 6 of these lectures, we will see how the Coulomb force arises from quantum field theory due to the exchange of virtual photons. We could repeat the calculation for nucleon-anti-nucleon scattering. The amplitude from field theory is given in (3.59). The first term in this expression gives the same result as for nucleon-nucleon scattering with the same sign. The second term vanishes in the non-relativisitic limit (it is an example of an interaction that doesn’t have a simple Newtonian interpretation). There is no longer a factor of 1/2 in (3.70), because the incoming/outgoing particles are not identical, so we learn that the potential between a nucleon and anti-nucleon is again given by (3.73). This reveals a key feature of forces arising due to the exchange of scalars: they are universally attractive. Notice that this is different from forces due to the exchange of a spin 1 particle — such as electromagnetism — where the sign flips when we change the charge. However, for forces due to the exchange of a spin 2 particle — i.e. gravity — the force is again universally attractive. 3.5.3 φ4 Theory Let’s briefly look at the Feynman rules and scattering amplitudes for the interaction Hamiltonian Hint = λ 4 φ 4! (3.74)

The theory now has a single interaction vertex, which comes with a factor of (−iλ), while the other Feynman rules remain the same. Note that we assign (−iλ) to the

– 69 –

5. is not strictly accurate. as we mentioned at the time.76) So scattering in φ4 theory is due to a δ-function potential. In fact. This gives 4! different contractions. which cancels the 1/4! sitting out front. As we shall see shortly.vertex rather than (−iλ/4!). which has its lowest contribution at order λ. we can look at φφ → φφ scattering. – 70 – . where every part of the diagram is connected to at least one external line. this will be related to the fact that the vacuum |0 of the free theory is not the true vacuum |Ω of the interacting theory. Both of these caveats are related to the assumption we made earlier that “initial and final states are eigenstates of the free theory” which. the scattering amplitude for φφ → φφ is Figure 16: simply iA = −iλ. To see why this is. An example of a diagram that is not connected is shown in Figure 17. The two caveats which go some way towards ameliorating the problem are the following • We consider only connected Feynman diagrams. The particles don’t know what hit them until it’s over. For more details. Feynman diagrams in the φ4 theory sometimes come with extra combinatoric factors (typically 2 or 4) which are known as symmetry factors that one must take into account. there are a couple of caveats about what Feynman diagrams you should write down. 3. −iλ Using the Feynman rules. p2 | : φ(x)φ(x)φ(x)φ(x) : |p1 .4 Connected Diagrams and Amputated Diagrams We’ve seen how one can compute scattering amplitudes by writing down all Feynman diagrams and assigning integrals to them using the Feynman rules. with the term −iλ ′ ′ p1 . see the book by Peskin and Schroeder. Translating this into a potential between two mesons. Note that it doesn’t depend on the angle at which the outgoing particles emerge: in φ4 theory the leading order two-particle scattering occurs with equal probability in all directions. we have U(r) = λ (2m)2 d3 p +ip·r λ e = δ (3) (r) 3 (2π) (2m)2 (3. p2 4! (3.75) Any one of the fields can do the job of annihilation or creation.

6 What We Measure: Cross Sections and Decay Rates So far we’ve learnt to compute the quantum amplitudes for particles decaying or scattering. Figure 17: A disconnected diagram. Figure 18: An un-amputated diagram 3.1 Fermi’s Golden Rule Let’s start with something familiar and recall how to derive Fermi’s golden rule from Dyson’s formula. They are related to the fact that the one-particle states of the free theory are not the same as the one-particle states of the interacting theory. so we end up with the square of a delta-function.• We do not consider diagrams with loops on external lines. with Em = En . but you will discuss them next term. We will refer to diagrams in which all loops on external legs have been cut-off as “amputated”. In this section we will compute these probabilities. 3. t m| U(t) |n = −i m| 0 dt HI (t) |n t = −i m| Hint |n = − m| Hint |n dt′ eiωt e 0 iωt ′ −1 ω (3. this comes from the fact that we’re working in an infinite space. We will not explain how to take these into account in this course. In particular. known as decay rates and cross sections. we have to leading order in the interaction. As usual in quantum theory. for example the diagram shown in the Figure 18. but surrounded by a cloud of virtual particles. One small subtlety here is that the S-matrix elements f | S − 1 |i all come with a factor of (2π)4 δ (4) (pF − pI ). correctly dealing with these diagrams will account for the fact that particles in interacting quantum field theories are never alone.6. As we will now see. For two energy eigenstates |m and |n . the probabilities for things to happen are the (modulus) square of the quantum amplitudes.77) – 71 – .

This gives us the probability for the transition from |n to |m in time t.78) dω −∞ 1 − cos ωt ω2 = πt Figure 19: ⇒ 1 − cos ωt ω2 → πtδ(ω) as t → ∞ Consider now a transition to a cluster of states with density ρ(E). most transitions happen in a region between energy eigenstates separated by ∆E = 2π/t. we can calculate +∞ 1 − cos ωt ω2 (3. To find the normalization.81) Now when squaring the amplitude to get the probability. we realize that the extra infinity is coming because Pm→n – 72 – . as Pn→m (t) = | m| U(t) |n | 2 = 2| m| Hint |n | 2 The function in brackets is plotted in Figure 19 for fixed t. we get the transition probability Pn→m = dEm ρ(Em ) 2| m| Hint |n | 2 1 − cos ωt ω2 (3. ˙ Pn→m = 2π| m| Hint |n |2 ρ(E) This is Fermi’s Golden Rule. We see that in time t. Then we get t=+∞ (3. In the above derivation.where ω = Em − En . In the limit t → ∞. the function in the figure starts to approach a deltafunction. As t → ∞. we were fairly careful with taking the limit as t → ∞. and first chose to compute the amplitude for the state |n at t → −∞ to transition to the state |m at t → +∞. we run into the problem of the square of the delta-function: Pn→m = | m| Hint |n |2 (2π)2 δ(ω)2 .80) −i m| t=−∞ HI (t) |n = −i m| Hint |n 2πδ(ω) (3. Tracking through the previous computations. Suppose we were a little sloppier.79) → 2π | m| Hint |n | 2 ρ(En )t which gives a constant probability for the transition per unit time for states around the same energy En ∼ Em = E.

after integrating over the density of final states. i| i = (2π)3 2EpI δ (3) (0) = 2EpI V where we have replaced δ (3) (0) by the volume of 3-space. as in the second derivation of Fermi’s Golden Rule.25)). f| f = 2Epi V final states (3.85) (3. (For example. The amplitudes Af i are. in our field theory calculations. we get the probability for decay P = |Af i |2 (2π)4 δ (4) (pI − pF ) V T 2mV 1 2Epi V states (3. This is given by P = | f | S |i |2 f | f i| i (3. we’ve computed the amplitudes in the same way as (3. exactly what we’ve been computing.65).86) If we place our initial particle at rest. and the square of the δ (4) -functions will just be re-interpreted as spacetime volume factors.84) Our states obey the relativistic normalization formula (2. We now divide out by this power of T to get the transition probability per unit time.81). we’ve exchanged one of the delta-functions for the volume of spacetime: (2π)4 δ (4) (0) = V T .82) where T is shorthand for t → ∞ (we used a very similar trick when looking at the vacuum energy in (2. of course.is the probability for the transition to happen in infinite time t → ∞. 3. ˙ Pn→m = 2π| m| Hint |n |2 δ(ω) (3. in (3. The reason that we’ve stressed this point is because. so pI = 0 and EpI = m.30).2 Decay Rates Let’s now look at the probability for a single particle |i of momentum pI (I=initial) to decay into some number of particles |f with momentum pi and total momentum pF = i pi .6.87) final where. Similarly.83) which. gives us back Fermi’s Golden rule. we saw – 73 – . We can write the delta-functions as (2π)2 δ(ω)2 = (2π)δ(ω) T (3.

6.87). Γ= 1 2m final |Af i |2 dΠ (3. We’ve been considering just a single particle per spatial volume V . More precisely dσ = 1 1 Differential Probability = |Af i |2 dΠ Unit Time × Unit Flux 4E1 E2 V F (3.91) where we’ve used the expression for probability per unit time that we computed in the previous subsection. sometimes they will pass right through. But we still have to worry about summing over all final states. φ). Γ = P . then the total number of scattering events N per unit time is given by. It is equal to the reciprocal of the half-life τ = 1/Γ. If the incoming flux F is defined to be the number of incoming particles per area per unit time. The result is an expression for the density of final states given by the Lorentz invariant measure dΠ = (2π)4 δ (4) (pF − pI ) d 3 pi 1 (2π)3 2Epi states (3. We now need an expression for the unit flux. The factors of spatial volume V in this measure cancel those in (3. We can now divide out by T to get the transition function per unit time. The fraction of the time that they collide is called the cross section and is denoted by σ. N = Fσ (3. – 74 – . while the factors of 1/2Epi in (3.that A = −g for a single meson decaying into two nucleons).87) conspire to produce the Lorentz invariant measure for 3-momentum integrals. In fact. For simplicity. we can calculate a more sensitive quantity dσ known as the differential cross section which is the probability for a given scattering process to occur in the solid angle (θ. E1 and E2 are the energies of the incoming particles.88) final The second step is to sum over all final states with different numbers (and possibly types) of particles. This gives us our final expression for the decay probability per unit ˙ time. 3. There are two steps: the first is to integrate over all possible momenta of the final particles: V d3 pi /(2π)3 .90) We would like to calculate σ from quantum field theory.3 Cross Sections Collide two beams of particles. Sometimes the particles will hit and bounce off each other.89) states Γ is called the width of the particle. let’s sit in the center of mass frame of the collision.

we might want to compute the viscosity of the quark gluon plasma. where θ is the angle between the incoming and outgoing particles. then recall from your course on special √ relativity that the 3-velocities vi are related to the momenta by v = p/m 1 − v 2 = p/p 0 . Equation (3. For example. This will involve doing the integral over the phase space of final states. In contrast.meaning that the flux is given in terms of the 3-velocities vi as F = |v1 − v2 |/V . in φ4 theory the amplitude for tree level scattering was simply A = −λ. We’ll denote the true vacuum of the interacting theory as |Ω . t = −2|p|2 (1 − cos θ). Notice that different scattering amplitudes have different momentum dependence and will result in different angular dependence in scattering amplitudes.7 Green’s Functions So far we’ve learnt to compute scattering amplitudes. and then relate them back to scattering amplitudes. These are nice and physical (well – they’re directly related to cross-sections and decay rates which are physical) but there are many questions we want to ask in quantum field theory that aren’t directly related to scattering experiments. for example. which follows from the fact that. You may now take your favorite scattering amplitude. for nucleon-nucleon scattering we have schematically A ∼ (t − m2 )−1 + (u − m2 )−1 . This results in isotropic scattering. with measure dΠ. and compute the probability for particles to fly out at your favorite angles. We’ll leave the relationship to other physical phenomena to other courses. 3. We’ll normalize H such that H |Ω = 0 (3. explain how to compute them using Feynman diagrams. For example.93) – 75 – . This then gives. or the optical conductivity in a tentative model of strange metals. In this section we will briefly define correlation functions. This gives rise to angular dependence in the differential cross-section. All of these questions are answered in the framework of quantum field theory by computing elementary objects known as correlation functions. or figure out the non-Gaussianity of density perturbations arising in the CMB from novel models of inflation.92) is our final expression relating the S-matrix to the differential cross section.92) If you want to write this in terms of momentum. dσ = 1 1 |Af i |2 dΠ 4E1 E2 |v1 − v2 | (3.

or Green’s functions.94) where φH is φ in the Heisenberg picture of the full theory. . Now insert a complete set of states. . which we take to be energy eigenstates of H = H0 + Hint . ±∞) at either end of the string of operators. . −∞) |Ω Ω| + = Ψ| Ω Ω| 0 + ′ lim t →−∞ n=0 |n n| |0 ′ eiEn (t −t) Ψ| n n| 0 n=0 (3. . Now let’s deal with the two remaining U(t. . −∞) |0 t where we’ve used all the intermediate factors of UI (tk . Ψ| U(t. . t1 )φ1I U(t1 . −∞) |0 = Ψ| U(t.and Ω| Ω = 1. t1 )φ1H . .95) where the operators on the right-hand side are evaluated on |0 . . Define G(n) (x1 . rather than the interaction picture that we’ve been dealing with so far. φnI UI (tn . Note that this is different from the state we’ve called |0 which is the vacuum of the free theory and satisfies H0 |0 = 0. Let’s start by asking how to compute G(n) using Feynman diagrams. Then G(n) (x1 . . Proof: Take t1 > t2 > . . > tn . Consider an arbitrary state |Ψ Ψ| UI (t. Then we can drop the T and write the numerator of the right-hand side as 0| UI (+∞. . φnH |Ω = 0| T φ1I . xn ) = Ω| T φ1H . and the equality above follows o because H0 |0 = 0. . t2 ) φ2I . . . . and write φ1H to denote the field in the Heisenberg picture. . The G(n) are called correlation functions. . . φH (xn ) |Ω (3. and φ1I to denote the field in the interaction picture. tk+1) = T exp(−i tkk+1 HI ) to convert φI into φH . φnI S |0 0| S |0 (3. −∞) |0 = 0| UI (+∞. the vacuum of the free theory. −∞) |0 (3.96) where U(t. −∞) |0 = Ψ| U(t. −∞) is the Schr¨dinger evolution operator. .97) – 76 – . . There are a number of different ways of looking at these objects which tie together nicely. We prove the following result Claim: We use the notation φ1 = φ(x1 ). φnH UI (tn . xn ) = Ω| T φH (x1 ) .

3. so that it sits on its own. namely Dyson’s formula and Wick’s theorem or. Since there is no ambiguity in the different types of line in Feynman diagrams. to the top and bottom of the right-hand side of (3. This follows from the Riemann-Lebesgue lemma which says that for any well-behaved function b µ→∞ lim dx f (x)eiµx = 0 a (3. using the normalization Ω| Ω = 1..99).97) is really an integral dn.95) to find 0| Ω Ω| T φ1H . 0| S |0 = 1 + + ( + + ) + . For the following discussion.99) (Notice that to derive this result. t′ ) |0 = Ψ| Ω Ω| 0 (3. alternatively. (There is a caveat here: we want the vacuum |Ω to be special. . φI (xn ) S |0 and 0| S |0 using the same methods we developed for S-matrix elements. completing the proof. we will represent the φ particles as solid lines.7. with no massless particles). gives us the left-hand side. we will work in φ4 theory. rather than the dashed lines that we used previously.101) These diagrams are called vacuum bubbles. which also does the job).But the last term vanishes. . (3. But what about dividing one by the other? What’s that all about? In fact. .100) which. So the Riemann-Lebesgue lemma gives us t′ →−∞ lim Ψ| U(t. The combinatoric factors (as well as the symmetry factors) associated with each diagram are such that the whole series sums – 77 – . away from the continuum of the integral. This means that we must be working in a theory with a mass gap – i. because all states are part of a continuum due to the momentum. φnH |Ω Ω| 0 0| Ω Ω| Ω Ω| 0 (3.1 Connected Diagrams and Vacuum Bubbles We’re getting closer to our goal of computing the Green’s functions G(n) since we can compute both 0| T φI (x1 ) .e. We now apply the formula (3. .98) Why is this relevant? The point is that the n in (3. Feynman diagrams. Peskin and Schroeder instead send t → −∞ in a slightly imaginary direction. it has a simple interpretation. . Then we have the diagramatic expansion for 0| S |0 ..

. . x1 x3 x2 x1 x2 x1 x2 + 2 Similar + x4 x3 x4 + x3 x4 + 5 Similar + .. even though they don’t look x1 x2 that connected! The point is that a connected diagram is defined by the requirement that every line is joined to an external leg.95). . Feynman Rules The Feynman diagrams that we need to calculate for the Green’s functions depend on x1 . Combining this with (3. The sum of connected Feynman diagrams is given by. the vacuum diagrams all add up to give the same exponential.103) where “connected” means that every part of the diagram is connected to at least one of the external legs. . Ω| T φH (x1 ) . . 0| S |0 = exp ( + + + . such diagrams are taken care of in shifting the vacuum from |0 to |Ω . . An x3 x4 example of a diagram that is not connected is shown in the figure. . . . . Remarkably. With a little thought one can show that 0| T φ1 . . Figure 20: As we have seen... . The upshot of all this is that dividing by 0| S |0 has a very nice interpretation in terms of Feynman diagrams: we need only consider the connected Feynman diagrams. xn ) can be calculated by summing over all connected Feynman diagrams. xn .102) So the amplitude for the vacuum of the free theory to evolve into itself is 0| S |0 = exp(all distinct vacuum bubbles). φH (xn ) |Ω = Connected Feynman Graphs (3. we learn that the Green’s functions G(n) (x1 . φn |0 = connected diagrams 0| S |0 (3. ) (3. let’s look at the four-point correlation function in φ4 theory.104) An Example: The Four-Point Correlator: Ω| T φH (x1 ) . . and don’t have to worry about the vacuum bubbles. . φH (x4 ) |Ω As a simple example.to an exponential. All of these are connected diagrams. This is rather different than the Feynman diagrams that we calculated for – 78 – . A similar combinatoric simplification occurs for generic correlation functions.

. −p′n′ . have this property. . meaning each external point is connected to each other external point. . We have n′ n p′1 . . xn ). You might think they all do. pn = (−i) n+n′ i=1 (pi′ 2 −m ) 2 j=1 2 (pj − m2 ) (3. n ˜ G(n) (p1 . However. .the S-matrix elements. and place their momentum back on shell. . we have • Draw n external points x1 . and ended up integrating over all of space. . . will survive the multiplication by on-shell propagators in (3. . Both of these problems are easily remedied to allow us to return to the S-matrix elements: we need to simply cancel off the propagators on the external legs. xn ). . . . xn .2 From Green’s Functions to S-Matrices Having described how to compute correlation functions using Feynman diagrams. pn ) = i=1 d4 xi e−ipi ·xi G(n) (x1 . connected by the usual propagators and vertices. p1 . .7. write down a factor of −iλ 3. of the diagrams that we wrote down which contribute to the four-point function Ω| T φH (x1 ) . . xn ) which have propagators for each external leg. The difference is that the Feynman rules for G(n) (x1 . as well as the internal legs. . . . . .106) to contribute to – 79 – . Assign a spacetime position y to the end of each line. . This means that we only get a non-zero answer for diagrams contributing to G(n) (x1 . but it’s not true! Only diagrams that are fully connected. . at position y. . .105) These are very closely related to the S-matrix elements that we’ve computed above. . it’s rather simple to adapt the Feynman rules that we had earlier in momentum space to compute G(n) (x1 . let’s now relate them back to the S-matrix elements that we already calculated. . . pn ) Each of the factors (p2 −m2 ) vanishes once the momenta are placed on-shell. . For φ4 theory. A related fact is that the 4momenta assigned to the external legs is arbitrary: they are not on-shell. . . . . . . . • For each vertex y y from x to y write down a factor of the Feynman propagator d4 y. . . . effectively include propagators ∆F for the external legs. where we were working primarily with momentum eigenstates. For example. . • For each line x ∆F (x − y). . The first step is to perform the Fourier transform. φH (x4 ) |Ω . p′n′ | S − 1 |p1 . xn ) (3. .106) ′ ˜ × G(n+n ) (−p′1 . only the S-matrix for meson scattering in φ4 theory.

taking into account the swarm of virtual particles surrounding asymptotic states. – 80 – . This is the correct way to consider scattering. In this context. introducing the Green’s functions seems like a lot of bother for little reward. remains true even in the interacting theory. The important point is that this provides a framework in which to deal with the true particle states in the interacting theory through renormalization. the formula (3. suitably interpreted.So what’s the point of all of this? We’ve understood that ignoring the connected diagrams is related to shifting to the true vacuum |Ω .106). You will derive it properly next term. (3.106) is known as the LSZ reduction formula. Indeed. But other than that.

4) and D[Λ−1 ] = D[Λ]−1 and D[1] = 1.3) where the matrices D[Λ] form a representation of the Lorentz group. But most particles in Nature have an intrinsic angular momentum. Dirac himself remarked in one of his talks that his equation was more intelligent than its author.4.5) for infinitesimal ω. (It comes with its own problems!).6) – 81 – . How do we find the different representations? Typically. a field can transform as φa (x) → D[Λ]ab φb (Λ−1x) D[Λ1 ]D[Λ2 ] = D[Λ1 Λ2 ] (4. These arise naturally in field theory by considering fields which themselves transform non-trivially under the Lorentz group. If we write. In general. In this section we will describe the Dirac equation.2) We’ll deal with this in Section 6. then the condition for a Lorentz transformation Λµσ Λνρ η σρ = η µν becomes the requirement that ω is anti-symmetric: ω µν + ω νµ = 0 (4. the field transforms as ν µ φ(x) → φ′ (x) = φ(Λ−1 x) (4. or spin. whose quantization gives rise to fermionic spin 1/2 particles. A familiar example of a field which transforms non-trivially under the Lorentz group is the vector field Aµ (x) of electromagnetism. It should be added. we look at infinitesimal transformations of the Lorentz group and study the resulting Lie algebra.” Weisskopf on Dirac So far we’ve only discussed scalar fields such that under a Lorentz transformation x → (x′ )µ = Λµ xν . we will start by studying the appropriate representation of the Lorentz group.1) We have seen that quantization of such fields gives rise to spin 0 particles. The Dirac Equation “A great deal more was hidden in the Dirac equation than the author had expected when he wrote it down in 1928. To motivate the Dirac equation. meaning that (4. Aµ (x) → Λµν Aν (Λ−1 x) (4. that it was Dirac who found most of the additional insights. however. Λµν = δ µν + ω µν (4.

Of course. We could call them (MA )µν . so we have (Mρσ )µν = η ρµ δ σν − η σµ δ ρν (4. M01 . The six basis matrices Mρσ are called the generators of the Lorentz transformations. [Mρσ . x2 )-plane. etc. The generators obey the Lorentz Lie algebra relations. If we use these matrices for anything practical (for example. M01 = −M10 . so that ρ and σ again label six different matrices. . so we call our matrices (Mρσ )µν . if we want to multiply them together. 6. generates rotations in the (x1 . with A = 1. . . . where ρ.9) The first.Note that an antisymmetric 4 × 4 matrix has 4 × 3/2 = 6 independent components. The second. we pick up various minus signs which means that when written in this form. ν 1 ω µν = 2 Ωρσ (Mρσ )µν (4. while ρ and σ denote which basis element we’re dealing with. . σ = 0. or act on some field) we will typically need to lower one index. we can write a basis of six 4 × 4 antisymmetric matrices as (Mρσ )µν = η ρµ η σν − η σµ η ρν (4. generates boosts in the x1 direction. Mτ ν ] = η στ Mρν − η ρτ Mσν + η ρν Mστ − η σν Mρτ (4. . . It is real and antisymmetric. M12 . after all. We can now write any ω µ as a linear combination of the Mρσ . With this notation in place. 0 1 0 0 0 0 0 0 0 0 0 0 (M01 )µν = 1 0 0 0 0 0 0 0 0 0 0 0 and (M12 )µ ν = 0 0 −1 0 0 1 0 0 (4. for example. the matrices are no longer necessarily antisymmetric. But in fact it’s better for us (although initially a little confusing) to replace the single index A with a pair of antisymmetric indices [ρσ]. antisymmetric matrices. the matrices are also antisymmetric on the µν indices because they are. . It is real and symmetric.8) Since we lowered the index with the Minkowski metric.11) – 82 – . which agrees with the 6 transformations of the Lorentz group: 3 rotations and 3 boosts.10) where Ωρσ are just six numbers (again antisymmetric in the indices) that tell us what Lorentz transformation we’re doing.7) where the indices µ and ν are those of the 4 × 4 matrix. Two examples of these basis matrices are. The antisymmetry on the ρ and σ indices means that. 3. It’s going to be useful to introduce a basis of these six 4 × 4 anti-symmetric matrices.

There are many such examples of 4 × 4 matrices which obey (4. 3 (4. has nothing to do with the Lorentz group. (γ i )2 = −1 i = 1. γ ν } ≡ γ µ γ ν + γ ν γ µ = 2η µν 1 (4. σ2 = 0 −i i 0 .13). σ j } = 2δ ij . 2.16) where each element is itself a 2 × 2 matrix.12) Let me stress again what each of these objects are: the Mρσ are six 4 × 4 basis elements of the Lorentz group. For example. We will construct the spinor representation. {γ µ . with the σ i the Pauli matrices σ1 = 0 1 1 0 . A finite Lorentz transformation can then be expressed as the exponential Λ = exp 1 2 Ωρσ Mρσ (4. It is the Clifford algebra. The simplest representation of the Clifford algebra is in terms of 4 × 4 matrices. are a set of four matrices and the 1 on the right-hand side denotes the unit matrix.14) It’s not hard to convince yourself that there are no representations of the Clifford algebra using 2 × 2 or 3 × 3 matrices. This means that we must find four matrices such that γ µ γ ν = −γ ν γ µ and (γ 0 )2 = 1 . σ3 = 1 0 0 −1 (4. γi = 0 σi −σ i 0 (4. – 83 – .where we have suppressed the matrix indices. we may take γ0 = 0 1 1 0 . with µ = 0.13) where γ µ . 1.15) when µ = ν (4.11).17) which themselves satisfy {σ i . at first sight. the Ωρσ are six numbers telling us what kind of Lorentz transformation we’re doing (for example. we start by defining something which. 4.1 The Spinor Representation We’re interested in finding other matrices which satisfy the Lorentz algebra commutation relations (4. they say things like rotate by θ = π/7 about the x3 -direction and run at speed v = 0. To do this.2 in the x1 direction). 3. 2.

γ ρ ]γ σ + 2 γ ρ [S µν . The matrices (4.1: [S µν .18) Let’s see what properties these matrices have: Claim 4.21) – 84 – . known as the Weyl or chiral representation (for reasons that will soon become clear).18) to write γ µ γ σ = 2S µσ + η µσ .19) (4.1 above. We will soon restrict ourselves further.16) provide one example. γ µ }γ ν + 2 γ µ γ ρ γ ν Claim 4. S ρσ ] = 2 [S µν . γ ρ } − 2 γ µ γ ρ γ ν − 2 {γ ρ . we have 1 [S µν . However. γ ρ ] = γ µ η νρ − γ ν η ρµ Proof: When µ = ν we have 1 [S µν . (4. and consider only representations of the Clifford algebra that are related to the chiral representation by a unitary transformation V . up to this equivalence. we have [S µν .2: The matrices S µν form a representation of the Lorentz algebra (4. S ρσ ] = η νρ S µσ − η µρ S νσ + η µσ S νρ − η νσ S µρ Proof: Taking ρ = σ. γ ρ ] = 2 [γ µ γ ν . γ σ ] 2 1 1 = 1 γ µ γ σ η νρ − 2 γ ν γ σ η ρµ + 1 γ ρ γ µ η νσ − 2 γ ρ γ ν η σµ 2 2 (4. γ ] = 4 0 1 2 ρ=σ ρ=σ γ γ ρ σ 1 1 = 2 γ ρ γ σ − η ρσ 2 (4.11).One can construct many other representations of the Clifford algebra by taking V γ µ V −1 for any invertible matrix V . γ ρ γ σ ] 1 = 1 [S µν . γ ρ ] = 1 γ µγ ν γ ρ − 1 γ ργ µγ ν 2 2 = γ µ η νρ − γ ν η ρµ 1 1 1 1 = 2 γ µ {γ ν . it turns out that there is a unique irreducible representation of the Clifford algebra. S ρσ = 1 ρ σ [γ .20) Now using the expression (4. S ρσ ] = S µσ η νρ − S νσ η ρµ + S ρµ η νσ − S ρν η σµ which is our desired expression. So what does the Clifford algebra have to do with the Lorentz group? Consider the commutator of two γ µ . and using Claim 4. meaning [S µν .

3. we have ψ α (x) → S[Λ]αβ ψ β (Λ−1 x) where Λ = exp S[Λ] = exp 1 2 1 2 (4. So far we haven’t given an index name to the rows and columns of these matrices: we’re going to call them α. say. We introduce the Dirac spinor field ψ α (x). 4. an object with four complex components labelled by α = 1. Note that we denote both the generator S ρσ and the full Lorentz transformation S[Λ] as “S”. 3. Under Lorentz transformations. Rotations S ij = 1 2 0 σi −σ i 0 0 σj −σ j 0 i = − ǫijk 2 σk 0 0 σk (4. we use the same six numbers Ωρσ in both Λ and S[Λ]: this ensures that we’re doing the same Lorentz transformation on x and ψ.23) (4.25) If we write the rotation parameters as Ωij = −ǫijk ϕk (meaning Ω12 = −ϕ3 .1 Spinors The S µν are 4 × 4 matrices. So how can we be sure that the spinor representation is something new. Both Λ and S[Λ] are 4 × 4 matrices.1. 2π). Consider now a rotation by 2π about.22) Ωρσ Mρσ Ωρσ S ρσ (4. because the γ µ are 4 × 4 matrices. and isn’t equivalent to the familiar representation Λµν ? To see that the two representations are truly different. We need a field for the matrices (S µν )αβ to act upon.24) Although the basis of generators Mρσ and S ρσ are different. To avoid confusion. 2. the x3 -axis. – 85 – .26) where we need to remember that Ω12 = −Ω21 = −ϕ3 when following factors of 2. 2. etc) then the rotation matrix becomes S[Λ] = exp 1 2 Ωρσ S ρσ = e+i ϕ·σ/2 0 0 e+iϕ·σ/2 (4. let’s look at some specific transformations.4. 0. This is achieved by ϕ = (0. 4. β = 1. the latter will always come with the square brackets [Λ].

We have demonstrated this explicitly for the spinor representation using the chiral representation (4. let’s see how a vector would transform under a rotation by ϕ = (0.and the spinor rotation matrix becomes. Recall that S[Λ] = exp 1 2 Ωρσ S ρσ (4. So S[Λ] is definitely a different representation from the familiar vector representation Λµν . S[Λ] is not unitary.33) 4 – 86 – .29) So when we rotate a vector by ϕ3 = 2π.31). (γ ν )† ] (4. 0.16) of the Clifford algebra. there are no finite dimensional unitary representations of the Lorentz group. ϕ3). S[Λ] is unitary.26). (S µν )† = −S µν . satisfying S[Λ]† S[Λ] = 1.30) Writing the boost parameter as Ωi0 = −Ω0i = χi . We have 0 0 0 0 0 0 0 0 0 0 Λ = exp 1 2 Ωρσ M ρσ = exp 0 0 ϕ3 0 0 −ϕ3 (4.32) so the representation is unitary if S µν are anti-hermitian. But for boosts given in (4.e. We can get a feel for why it is true for a spinor representation constructed from any representation of the Clifford algebra. Boosts 1 S = 2 0i 0 1 1 0 0 σi −σ i 0 e+χ·σ/2 0 1 = 2 −σ i 0 0 σi (4. we learn that Λ = 1 as you would expect.27) which is definitely not what happens to a vector! To check that we haven’t been cheating with factors of 2. we have S[Λ] = 0 e−χ·σ/2 (4. But we have 1 (S µν )† = − [(γ µ )† . i. In fact.31) Representations of the Lorentz Group are not Unitary Note that for rotations given in (4. S[Λ] = Therefore under a 2π rotation ψ α (x) → −ψ α (x) (4.28) e+iπσ 0 3 0 e+iπσ 3 = −1 (4.

2. like the chiral representation (4.34) So we could pick γ 0 to be hermitian.37) – 87 – . However.35) which is the usual adjoint of a multi-component object. 3 we have γ 0 γ µ γ 0 = (γ µ )† which. for some Lorentz transformation S[Λ]† S[Λ] = 1 since the representation is not unitary. We would like to construct a Lorentz invariant equation of motion.16). we can never arrange for this to happen since (γ 0 )2 = 1 ⇒ Real Eigenvalues (γ i )2 = −1 ⇒ Imaginary Eigenvalues (4. 1. the matrices have this property: (γ 0 )† = γ 0 and (γ i )† = −γ i . but will give us a clue how to proceed. 4. Then for all µ = 0.16). and certainly isn’t a scalar. as we have seen. Define ψ † (x) = (ψ ⋆ )T (x) (4. In general there is no way to pick γ µ such that S µν are anti-hermitian. Indeed. Let’s pick a representation of the Clifford algebra which. But now we see why it fails. means that 1 (S µν )† = [(γ ν )† . but we can only pick γ i to be anti-hermitian.38) (4.which can be anti-hermitian if all γ µ are hermitian or all are anti-hermitian. Let’s see how this transforms under Lorentz transformations. in turn. We could then try to form a Lorentz scalar by taking the product ψ † ψ. This means that ψ † ψ isn’t going to do it for us: it doesn’t have any nice transformation under the Lorentz group. ψ(x) → S[Λ] ψ(Λ1x) ψ † (x) → ψ † (Λ−1 x) S[Λ]† (4. in the chiral representation (4.36) So ψ † (x)ψ(x) → ψ † (Λ−1 x)S[Λ]† S[Λ]ψ(Λ−1 x). But. satisfies (γ 0 )† = γ 0 and (γ i )† = −γ i . (γ µ )† ] = −γ 0 S µν γ 0 4 (4.2 Constructing an Action We now have a new field to work with. we can also see how to proceed. with the spinor indices summed over. the Dirac spinor ψ. We will start in a naive way which won’t work. We do this by constructing a Lorentz invariant action.

2.41) which is indeed the transformation law for a Lorentz scalar. . 2 – 88 – . We work infinitesimally.44) (4. Proof: Under a Lorentz transformation.43) (4.4: ψ γ µ ψ is a Lorentz vector. ¯ ¯ ψ γ µ ψ → ψ S[Λ]−1 γ µ S[Λ]ψ ¯ If ψ γ µ ψ is to transform as a vector. .3: ψψ is a Lorentz scalar. we now define the Dirac adjoint ¯ ψ(x) = ψ † (x) γ 0 (4. under a Lorentz transformation we have. Proof: Suppressing the x argument. (4. ¯ ψ(x) ψ(x) = ψ † (x) γ 0 ψ(x) → ψ † (Λ−1 x) S[Λ]† γ 0 S[Λ]ψ(Λ−1 x) = ψ † (Λ−1 x) γ 0 ψ(Λ−1 x) ¯ = ψ(Λ−1 x) ψ(Λ−1 x) (4.42) This equation means that we can treat the µ = 0.so that S[Λ]† = exp 1 2 Ωρσ (S ρσ )† = γ 0 S[Λ]−1 γ 0 (4. which means that ¯ ¯ ψ(x) γ µ ψ(x) → Λµν ψ(Λ−1 x) γ ν ψ(Λ−1 x) (4. so that Λ = exp S[Λ] = exp 1 2 1 2 1 Ωρσ Mρσ ≈ 1 + 2 Ωρσ Mρσ + .45) (4.40) Let’s now see what Lorentz covariant objects we can form out of a Dirac spinor ψ and ¯ its adjoint ψ. .39) With this in mind.46) Ωρσ S ρσ ≈ 1 + 1 Ωρσ S ρσ + . ¯ Claim 4. 1. we must have S[Λ]−1 γ µ S[Λ] = Λµν γ ν We’ll now show this. 3 index on the γ µ matrices as a true vector index. . ¯ Claim 4. In particular we can form Lorentz scalars by contracting it with other Lorentz indices.

Proof: As above.48) which means that the proof follows if we can show −[S ρσ . proportional to η µν ψψ. we write the right-hand side of (4. In fact equation (4. ψψ. the symmet¯ ric part is a Lorentz scalar. ¯ ¯ ¯ We are now armed with three bilinears of the Dirac field.47) where we’ve suppressed the α. β indices on γ µ and S µν . To see this.44) becomes −[S ρσ . Notice that the Lagrangian is first order. (Said another way. We can try to build a Lorentz invariant action from these. each of which transforms covariantly under the Lorentz group.1. the mass appears in the Lagrangian as m. rather than the second order Lagrangians we were working with for scalar fields. γ µ ] = (Mρσ )µν γ ν (4. – 89 – .50) (4. ψγ µ ψ and ψγ µ γ ν ψ. We choose S= ¯ d4 x ψ(x) (iγ µ ∂µ − m) ψ(x) (4. As we will see in the next section. γ µ ] = η ρµ γ σ − η σµ γ ρ which is exactly what we proved in Claim 4. In fact. The factor of “i” is there to make the action real. but otherwise left all other indices explicit.5: ψγ µ γ ν ψ transforms as a Lorentz tensor. it cancels a minus sign that comes from integration by parts. while the antisymmetric part is a ¯ Lorentz tensor.1 where we showed that [S ρσ . it’s there for the same reason that the Hermitian momentum operator −i∇ in quantum mechanics has a factor i). More precisely. Also. which can be positive or negative. we need only the first two.49) This is the Dirac action.47) by expanding out M.so the requirement (4.47) follows from Claim 4. ¯ Claim 4. upon complex conjugation. γ µ ] = γ ρ η σµ − γ σ η µρ . after quantization this theory describes particles and anti-particles of mass |m| and spin 1/2. σ ρ (Mρσ )µν γ ν = (η ρµ δν − η σµ δν )γ ν = η ρµ γ σ − η σµ γ ρ (4. proportional to ψS µν ψ.

56) (4. write (iγ ν ∂ν + m)(iγ µ ∂µ − m) ψ = − γ µ γ ν ∂µ ∂ν + m2 ψ = 0 But γ µ γ ν ∂µ ∂ν = 1 {γ µ . we have (iγ µ ∂µ − m) ψ = 0 (4.55) – 90 – . It’s completely gorgeous. However.53) where this last equation has no γ µ matrices. 2. However.51) This is the Dirac equation. If we tried to write down a first order equation of motion for a scalar field. Varying with respect to ψ gives the conjugate equation ¯ ¯ i∂µ ψ γ µ + mψ = 0 (4. To see this. Varying with respect to ψ. . the magic of the γ µ matrices means that the Dirac Lagrangian is Lorentz invariant.3 The Dirac Equation The equation of motion follows from the action (4. 3. 4.. The Dirac equation mixes up different components of ψ through the matrices γ µ . γ ν }∂µ ∂ν = ∂µ ∂ µ . so we get 2 −(∂µ ∂ µ + m2 )ψ = 0 (4. The Slash Let’s introduce some useful notation.4.54) (4. which necessarily includes a privileged vector in spacetime v µ and is not Lorentz invariant. . for spinor fields. We will often come across 4-vectors contracted with γ µ matrices. with α = 1. it would look like v µ ∂µ φ = . We write / Aµ γ µ ≡ A so the Dirac equation reads / (i ∂ − m)ψ = 0 (4. and so applies to each component ψ α .50) by varying with respect to ψ and ¯ ¯ ψ independently. each individual component itself solves the Klein-Gordon equation.52) The Dirac equation is first order in derivatives. yet miraculously Lorentz invariant.

the spinor rotation transformation S[Λrot ] and boost transformation S[Λboost ] were computed in (4. We have ¯ / L = ψ(i ∂ − m)ψ = iu† σ µ ∂µ u− + iu† σ µ ∂µ u+ − m(u† u− + u† u+ ) = 0 − +¯ + − (4. the spinor is a representation of the double cover of the Lorentz group SL(2.4 Chiral Spinors When we’ve needed an explicit form of the γ µ matrices. are defined by ψ= u+ u− (4.1 The Weyl Equation Let’s see what becomes of the Dirac Lagrangian under the decomposition (4. acting only on two-component spinors u± which. 1 1 while u− is in the (0.57) In this representation. 2 ) 2 representation.4. Both are block diagonal.26) and (4.61) (4. u+ is in the ( 2 .31). It decomposes into two irreducible representations. S[Λrot ] = e+i ϕ·σ/2 0 e 0 +iϕ·σ/2 and S[Λboost ] = e+χ·σ/2 0 e 0 −χ·σ/2 (4. we’ve used the chiral representation γ0 = 01 10 . (Strictly speaking. in the chiral representation. 1 ) representation.58) This means that the Dirac spinor representation of the Lorentz group is reducible. 0) ⊕ (0. C)).59) The two-component objects u± are called Weyl spinors or chiral spinors. The Dirac spinor ψ lies in the ( 2 . γi = 0 σi −σ i 0 (4. u± → eiϕ·σ/2 u± but oppositely under boosts.4.62) – 91 – . u± → e±χ·σ/2 u± (4. They transform in the same way under rotations.59) into Weyl spinors.60) 1 In group theory language. 0) representation of the Lorentz group. 4.

65) iσ µ ∂µ u− = 0 (4. in the quantum theory. But what about a Dirac spinor? One might think that there are 8 degrees of freedom. this counting manifests itself as two degrees of freedom (spin up and down) for the particle. This means that the phase space for a spinor is therefore parameterized by ψ and ψ † . the number of degrees of freedom of a system is equal to the dimension of the configuration space or. with the equation of motion i¯ µ ∂µ u+ = 0 σ or These are the Weyl equations. equivalently. but it makes sense to count the number of degrees of freedom per spatial point: this should at least be finite. σ i ) and σ µ = (1. −σ i ) ¯ (4.63) From (4. At the quantum level. and a further two for the anti-particle. half the dimension of the phase space. 1. 2. For example. – 92 – . However. while for a scalar it is parameterized ˙ by φ and π = φ. In particular. this translates to the fact that it gives rise to a single type of particle. in this sense a real scalar field φ has a single degree of freedom. So the phase space of the Dirac spinor ψ has 8 real dimensions and correspondingly the number of real degrees of freedom is 4. 3 index. the equation of motion is first order rather than second order. A classical complex scalar field has two degrees of freedom. the momentum conjugate to the spinor ψ is given by ˙ πψ = ∂L/∂ ψ = iψ † (4. How do we count degrees of freedom? In classical mechanics. σ µ = (1. and in contrast to the scalar field. The Dirac fermion has 4 complex components = 8 real components. But this isn’t right. corresponding to the particle and the anti-particle in the quantum theory. a massless fermion can be described by u+ (or u− ) alone. for the Dirac Lagrangian.where we have introduced some new notation for the Pauli matrices with a µ = 0.64) It is not proportional to the time derivative of ψ. A similar counting for the Weyl fermion tells us that it has two degrees of freedom. In field theory we have an infinite number of degrees of freedom. Degrees of Freedom Let me comment here on the degrees of freedom in a spinor. since they couple through the mass term. We will see in the next section that.62). we see that a massive fermion requires both u+ and u− . Crucially.

iγ 5 ). while ψ− is “left-handed”. (You γ ˜ 4 5 might think that γ would be a better name! But γ is the one everyone chooses .67) The reason that this is called γ 5 is because the set of matrices γ A = (γ µ . – 93 – . where A = 1. In fact. 4 satisfy the five-dimensional Clifford algebra {˜ .58) because we chose the specific representation (4.69) 2 2 such that P+ = P+ and P− = P− and P+ P− = 0. 1.66) Now S[Λ] will not be block diagonal. 5). with ˜ A A = 0. 3. γ µ } = 0 and (γ 5 )2 = +1 (4.59). we may use γ 5 to define the chiral spinors. so that γ µ → Uγ µ U −1 and ψ → Uψ ? (4.71) which form the irreducible representations of the Lorentz group. this is why the representation (4. γ 5 ] = 0.2 γ 5 The Lorentz group matrices S[Λ] came out to be block diagonal in (4. 4. You can also check that [Sµν . which means that γ 5 is a scalar under rotations and boosts. Since (γ 5 )2 = 1. ψ+ is often called a “right-handed” spinor. 2. However. γ5 = 1 0 0 −1 (4. ψ± = P± ψ (4.57) is called the chiral representation: it’s because the decomposition of the Dirac spinor ψ is simply given by (4. 2. this means we may form the Lorentz invariant projection operators P± = 1 1 ± γ5 2 (4. 3. But what happens if we choose a different representation γ µ of the Clifford algebra.57).57).it’s a more sensible name in Euclidean space. One can check that for the chiral representation (4. Is there an invariant way to define chiral spinors? We can do this by introducing the “fifth” gamma-matrix γ 5 = −iγ 0 γ 1 γ 2 γ 3 You can check that this matrix satisfies {γ 5 .4.68) (4.4. for an arbitrary representation of the Clifford algebra. γ B } = 2η AB .70) from which we see that the operators P± project onto the Weyl spinors u± .

4.4.3 Parity The spinors ψ± are related to each other by parity. Let’s pause to define this concept. The Lorentz group is defined by xµ → Λµν xν such that Λµν Λρσ η νσ = η µρ (4.72) So far we have only considered transformations Λ which are continuously connected to the identity; these are the ones which have an infinitesimal form. However there are also two discrete symmetries which are part of the Lorentz group. They are Time Reversal T : x0 → −x0 ; xi → xi Parity P : x0 → x0 ; xi → −xi

(4.73)

We won’t discuss time reversal too much in this course. (It turns out to be represented by an anti-unitary transformation on states. See, for example the book by Peskin and Schroeder). But parity has an important role to play in the standard model and, in particular, the theory of the weak interaction. Under parity, the left and right-handed spinors are exchanged. This follows from the transformation of the spinors under the Lorentz group. In the chiral representation, we saw that the rotation (4.60) and boost (4.61) transformations for the Weyl spinors u± are u± −→ eiϕ·σ/2 u±
rot

and

u± −→ e±χ·σ/2 u±

boost

(4.74)

Under parity, rotations don’t change sign. But boosts do flip sign. This confirms that parity exchanges right-handed and left-handed spinors, P : u± → u∓ , or in the notation 1 ψ± = 2 (1 ± γ 5 )ψ, we have P : ψ± (x, t) → ψ∓ (−x, t) (4.75) Using this knowledge of how chiral spinors transform, and the fact that P 2 = 1, we see that the action of parity on the Dirac spinor itself can be written as P : ψ(x, t) → γ 0 ψ(−x, t) (4.76)

Notice that if ψ(x, t) satisfies the Dirac equation, then the parity transformed spinor γ 0 ψ(−x, t) also satisfies the Dirac equation, meaning (iγ 0 ∂t + iγ i ∂i − m)γ 0 ψ(−x, t) = γ 0 (iγ 0 ∂t − iγ i ∂i − m)ψ(−x, t) = 0 (4.77)

where the extra minus sign from passing γ 0 through γ i is compensated by the derivative acting on −x instead of +x.

– 94 –

4.4.4 Chiral Interactions Let’s now look at how our interaction terms change under parity. We can look at each of our spinor bilinears from which we built the action, ¯ ¯ P : ψψ(x, t) → ψψ(−x, t) (4.78)

¯ which is the transformation of a scalar. For the vector ψγ µ ψ, we can look at the temporal and spatial components separately, ¯ ¯ P : ψγ 0 ψ(x, t) → ψγ 0 ψ(−x, t) ¯ ¯ ¯ P : ψγ i ψ(x, t) → ψγ 0 γ i γ 0 ψ(−x, t) = −ψγ i ψ(−x, t)

(4.79)

¯ which tells us that ψγ µ ψ transforms as a vector, with the spatial part changing sign. ¯ You can also check that ψS µν ψ transforms as a suitable tensor. However, now we’ve discovered the existence of γ 5 , we can form another Lorentz scalar and another Lorentz vector, ¯ ψγ 5 ψ ¯ and ψγ 5 γ µ ψ (4.80)

How do these transform under parity? We can check: ¯ ¯ ¯ P : ψγ 5 ψ(x, t) → ψγ 0 γ 5 γ 0 ψ(−x, t) = −ψγ 5 ψ(−x, t) ¯ −ψγ 5 γ 0 ψ(−x, t) ¯ ¯ P : ψγ 5 γ µ ψ(x, t) → ψγ 0 γ 5 γ µ γ 0 ψ(−x, t) = ¯ +ψγ 5 γ i ψ(−x, t) (4.81) µ=0 µ=i

¯ ¯ which means that ψγ 5 ψ transforms as a pseudoscalar, while ψγ 5 γ µ ψ transforms as an axial vector. To summarize, we have the following spinor bilinears, ¯ ψψ : ¯ ψγ µ ψ : ¯ ψS µν ψ : ¯ ψγ 5 ψ : ¯ ψγ 5 γ µ ψ : scalar vector tensor pseudoscalar axial vector (4.82)

The total number of bilinears is 1 + 4 + (4 × 3/2) + 4 + 1 = 16 which is all we could hope for from a 4-component object.

– 95 –

We’re now armed with new terms involving γ 5 that we can start to add to our Lagrangian to construct new theories. Typically such terms will break parity invariance ¯ of the theory, although this is not always true. (For example, the term φψγ 5 ψ doesn’t break parity if φ is itself a pseudoscalar). Nature makes use of these parity violating interactions by using γ 5 in the weak force. A theory which treats ψ± on an equal footing is called a vector-like theory. A theory in which ψ+ and ψ− appear differently is called a chiral theory. 4.5 Majorana Fermions Our spinor ψ α is a complex object. It has to be because the representation S[Λ] is typically also complex. This means that if we were to try to make ψ real, for example by imposing ψ = ψ ⋆ , then it wouldn’t stay that way once we make a Lorentz transformation. However, there is a way to impose a reality condition on the Dirac spinor ψ. To motivate this possibility, it’s simplest to look at a novel basis for the Clifford algebra, known as the Majorana basis. γ =
0

0 σ2 σ2 0

,

γ =

1

iσ 3 0 0 iσ 3

,

γ =

2

σ2

0 −σ 2 0

,

γ =

3

−iσ 1 0

0 −iσ 1

These matrices satisfy the Clifford algebra. What is special about them is that they are all pure imaginary (γ µ )⋆ = −γ µ . This means that the generators of the Lorentz group S µν = 1 [γ µ , γ ν ], and hence the matrices S[Λ] are real. So with this basis of the 4 Clifford algebra, we can work with a real spinor simply by imposing the condition, ψ = ψ⋆ (4.83)

which is preserved under Lorentz transformation. Such spinors are called Majorana spinors. So what’s the story if we use a general basis for the Clifford algebra? We’ll ask only that the basis satisfies (γ 0 )† = γ 0 and (γ i )† = −γ i . We then define the charge conjugate of a Dirac spinor ψ as ψ (c) = Cψ ⋆ Here C is a 4 × 4 matrix satisfying C † C = 1 and C † γ µ C = −(γ µ )⋆ (4.85) Let’s firstly check that (4.84) is a good definition, meaning that ψ (c) transforms nicely under a Lorentz transformation. We have ψ (c) → CS[Λ]⋆ ψ ⋆ = S[Λ]Cψ ⋆ = S[Λ]ψ (c) (4.86) (4.84)

– 96 –

. ⋆ Plugging in the various definitions. An Aside: Spinors in Different Dimensions: The ability to impose Majorana or Weyl conditions on Dirac spinors depends on both the dimension and the signature of spacetime.87) After quantization. . although the Majorana condition and Weyl condition cannot be imposed simultaneously in four dimensions.85) in taking the matrix C through S[Λ]⋆ . we can now impose the Lorentz invariant reality condition on the Dirac spinor. we find that u+ = iσ 2 u− and u− = −iσ 2 u⋆ .16). In the chiral basis (4. ψ (c) = ψ (4. Instead the Majorana condition relates u− and u+ . 18. for a given representation of the Clifford algebra. This follows from. This is exactly the same as in the case of scalar fields. and we 2 0 may take Cchiral = iγ 2 = −iσ2 iσ . .. to yield a Majorana spinor. 10. (The matrix iσ 2 that appears here is simply the 0 anti-symmetric matrix ǫαβ ).88) Notice that it’s not possible to impose the Majorana condition ψ = ψ (c) at the same time as the Weyl condition (u− = 0 or u+ = 0).59). In + other words. we have simply CMaj = 1 and the Majorana condition ψ = ψ (c) becomes ψ = ψ ⋆ . basically because you can always build a suitable “γ 5 ” projection matrix by multiplying together all the other γ-matrices. So what is this matrix C? Well. where the gamma matrices are pure imaginary. only γ 2 is imaginary. where we’ve seen that a real scalar field gives rise to a spin 0 boson that is its own anti-particle. but if ψ satisfies the Dirac equation. It is interesting to see how the Majorana condition (4. it is something that we can find fairly easily. then ψ (c) does too. not only does ψ (c) transform nicely under the Lorentz group. In the Majorana basis. The pattern for when the Majorana condition can be imposed is a little more sporadic. In fact.87) looks in terms of the decomposition into left and right handed Weyl spinors (4. a Majorana spinor can be written in terms of Weyl spinors as ψ= u+ −iσ 2 u⋆ + (4. – 97 – . you can do this in Minowski spacetimes of dimension 2. (Be aware: In many texts an extra factor of γ 0 is absorbed into the definition of C). One can always impose the Weyl condition on a spinor in even dimensional Minkowski space. Interestingly. the Majorana spinor gives rise to a fermion that is its own antiparticle.where we’ve made use of the properties (4. / / (i ∂ − m)ψ = 0 ⇒ (−i ∂ − m)ψ ⋆ = 0 ⋆ ⋆ / / ⇒ C(−i ∂ − m)ψ ⋆ = (+i ∂ − m)ψ (c) = 0 Finally.

6 Symmetries and Conserved Currents The Dirac Lagrangian enjoys a number of symmetries. while the energy-momentum is typically first order. Lorentz Transformations Under an infinitesimal Lorentz transformation. following (4.10).41) gives us the energy-momentum tensor ¯ T µν = iψγ µ ∂ ν ψ − η µν L (4.93) 1 where. In the case of a scalar field this didn’t really buy us anything because the equations of motion are second order in derivatives. However. in infinitesimal form.22) which.89) ¯ The Lagrangian depends on ∂µ ψ.91) where. but not ∂µ ψ.4.8) (Mρσ )µ = η ρµ δ σ − η σµ δ ρ ν ν ν (4.92) (4. for / a spinor field the equations of motion are first order: (i ∂ − m)ψ = 0. the Dirac spinor transforms as (4. so the standard formula (1. Here we list them and compute the associated conserved currents. we have ω µν = 2 Ωρσ (Mρσ )µν .94) – 98 – . and Mρσ are the generators of the Lorentz algebra given by (4. This means we µν can set L = 0 in T . we have again used the equations of motion. we don’t lose anything by imposing the equations of motion already on T µν . we have the total energy E= d3 x T 00 = ¯ ˙ d3 x iψγ 0 ψ = d3 x ψ † γ 0 (−iγ i ∂i + m)ψ (4. in the last equality. leaving ¯ T µν = iψγ µ ∂ ν ψ In particular. reads δψ α = −ω µν xν ∂µ ψ α + 1 Ωρσ (S ρσ )αβ ψ β 2 (4. Spacetime Translations Under spacetime translations the spinor transforms as δψ = ǫµ ∂µ ψ (4.90) Since a current is conserved only when the equations of motion are obeyed.

ψ → e−iα ψ. reflecting the fact that the left and right-handed components ψ± transform in the same way under this symmetry. secondly. µ ¯ ¯ ¯ ¯ ∂µ jV = (∂µ ψ)γ µ ψ + ψγ µ (∂µ ψ) = imψψ − imψψ = 0 (4. after direct substitution. in the last equality.100) – 99 – .99) We will see shortly that this has the interpretation of electric charge.54) with two differences: firstly. So we get 1 δψ α = −ω µν xν ∂µ ψ α − 2 (Sµν )αβ ψ β (4.98) / where. this extra term will be responsible for providing the single particle states with internal angular momentum. telling us that the quantization of a Dirac spinor gives rise to a particle carrying spin 1/2. we pick up an extra piece in the current from the second term in (4. the Dirac Lagrangian admits an extra internal symmetry which rotates left and right-handed fermions in opposite directions. the spinor equations of motion set L = 0. when (J µ )ρσ is turned into an operator. This gives rise to the current µ ¯ jV = ψγ µ ψ (4. The conserved quantity arising from this symmetry is ¯ i∂µ ψγ Q= ¯ d3 x ψγ 0 ψ = d3 x ψ † ψ (4. ψ → eiαγ ψ 5 5 ¯ ¯ and ψ → ψeiαγ (4. We can easily check that µ jV is conserved under the equations of motion. Internal Vector Symmetry The Dirac Lagrangian is invariant under rotating the phase of the spinor.96) After quantization. We have ¯ (J µ )ρσ = xσ T µρ − xρ T µσ − iψγ µ S ρσ ψ (4.95) The conserved current arising from Lorentz transformations now follows from the same calculation we saw for the scalar field (1. for fermions. as we saw above.95). tells us that ω µν = Ωµν .which.97) where “V ” stands for vector. or particle number. Axial Symmetry When m = 0. we have used the equations of motion i ∂ ψ = mψ and ¯ µ = −mψ.

104) (4. with the full Dirac Lagrangian we may compute µ ¯ ¯ ¯ ∂µ jA = (∂µ ψ)γ µ γ 5 ψ + ψγ µ γ 5 ∂µ ψ = 2imψγ 5 ψ (4.107) (4. in the quantum theory things become more interesting for the axial current. However.Here the second transformation follows from the first after noting that e−iαγ γ 0 = 5 γ 0 e+iαγ .105) is u(p) = √ √ p·σξ p·σξ ¯ (4.105) – 100 – . This is conserved only when m = 0. Indeed.102) which vanishes only for m = 0. can depend on the 3-momentum p. σ i ) and σ µ = (1. σ µ = (1. This gives the conserved current. It is the archetypal example of an anomaly: a symmetry of the classical theory that is not preserved in the quantum theory. independent of spacetime x which. 4. µ ¯ jA = ψγ µ γ 5 ψ 5 (4. as the notation suggests. When the theory is coupled to gauge fields (in a manner we will discuss in Section 6). But it doesn’t survive the quantization process. the axial transformation remains a symmetry of the classical Lagrangian. −σ i ) ¯ Claim: The solution to (4.101) µ where A is for “axial” since jA is an axial vector. The Dirac equation then becomes (γ pµ − m)u(p) = where we’re again using the definition.103) where u(p) is a four-component spinor.7 Plane Wave Solutions Let’s now study the solutions to the Dirac equation (iγ µ ∂µ − m)ψ = 0 We start by making a simple ansatz: ψ = u(p) e−ip·x (4.106) µ pµ σ µ −m ¯ −m pµ σ µ u(p) = 0 (4.

108) Either one of these equations implies the other.105). a fact which follows from the identity (p · σ)(p · σ ) = p2 − pi pj σ i σ j = p2 − pi pj δ ij = pµ pµ = m2 .108) immediately tells us that u2 = mξ ′ . So we learn that any spinor of the form u(p) = A (p · σ) ξ ′ mξ ′ (4.107) Negative Frequency Solutions We get further solutions to the Dirac equation from the ansatz ψ = v(p) e+ip·x (4.for any 2-component spinor ξ which we will normalize to ξ † ξ = 1. Then the second equation in (4. u2 ).105) reads (p · σ) u2 = mu1 and (p · σ )u1 = mu2 ¯ (4. Then u1 = (p · σ) p · σ ξ = m p · σ ξ.104). To make this more symmetric.110) Solutions of the form (4. So we get the promised result (4.111) for some 2-component spinor η which we take to be constant and normalized to η † η = 1. The Dirac equation requires that the 4-component spinor v(p) satisfies (γ µ pµ + m)v(p) = which is solved by v(p) = √ p·ση √ ¯ − p·ση (4. Proof: Let’s write u(p)T = (u1. are called positive frequency solutions. – 101 – .92). let’s try ¯ 0 0 ′ the ansatz u1 = (p · σ)ξ for some spinor ξ ′ . which oscillate in time as ψ ∼ e−iEt . Then equation (4. are negative frequency solutions.112) m pµ σ µ ¯ pµ σ µ m v(p) = 0 (4.109) with constant A is a solution to (4. which oscillate as ψ ∼ e+iEt . To start with. we choose √ √ √ ¯ ¯ A = 1/m and ξ ′ = p · σ ξ with constant ξ. It’s important to note however that both are solutions to the classical field equations and both have positive energy (4.110). Those of the form (4.

ξ → e+iϕ·σ/2 ξ (4.1 Some Examples Consider the positive frequency solution with mass m and 3-momentum p = 0. u(p) = √ m ξ ξ (4. 1) field. ξ T = (1.115) In fact. For a massless particle we have u(p) = √ 0 2E 0 1 0 (4. Consider now boosting the particle with spin ξ T = (1. for which E = p3 . 0. this expression also makes sense for a massless field. p). for a boosted solution of the spin down ξ T = (0. even though they have no such interpretation before quantization.117) – 102 – . we’ll indulge in an abuse of terminology and refer to the classical solutions to the Dirac equations as “particles”. A field with spin up (down) along a given direction is described by the eigenvector of the corresponding Pauli matrix with eigenvalue +1 (-1 respectively). we have √ p·σ p·σ ¯ 0 1 0 1 u(p) =  √  =  E + p3 E− p3 0 1 0 1   m→0 −→ √ 0 2E 1 0 0 (4. 0. (We picked the normalization (4.7. Spatial rotations of the field act on ξ by (4. After quantization. In the rest of this section. this will become the spin of the associated particle. 0) along the x3 direction.4. This should be familiar from quantum mechanics.26).107) for the solutions so that this would be the case). The solution to the Dirac equation becomes √ p·σ u(p) =  √ p·σ ¯ 1 0 1 0  =  E − p3 E+ p3 1 0 1 0   (4. with pµ = (E.116) Similarly.114) The 2-component spinor ξ defines the spin of the field. 0) describes a field with spin up along the z-axis. For example.113) where ξ is any 2-component spinor.

Here we state both: √ p · σ ξs r †√ r† s r† √ p·σ ¯ p·σ. ξ p·σ ¯ 01 10 √ √ p · σ ξs p · σ ξs ¯ = 2mδ rs (4.7.Helicity The helicity operator is the projection of the angular momentum along the direction of momentum.25). only the latter will be Lorentz invariant. Let’s deal first with the positive frequency plane waves. It’s firstly convenient to introduce a basis ξ s and η s . 2 for the two-component spinors such that ξ r † ξ s = δ rs for example. Of course. ξ u (p) · u (p) = ξ √ p · σ ξs ¯ = ξ r † p · σξ s + ξ r † p · σ ξ s = 2ξ r † p0 ξ s = 2po δ rs ¯ √ r †√ (4.119) and similarly for η s .123) – 103 – .116) has helicity h = 1/2: we say that it is right-handed. 1 i ˆ ˆ h = 2 ǫijk pi S jk = 2 pi σi 0 0 σi (4. The two independent solutions are now written as √ p · σ ξs s u (p) = √ (4.122) while the Lorentz invariant inner product is u (p) · u (p) = ξ ¯ r s r† p·σ. The massless field with spin ξ T = (1. 4. 0) in (4. regarding the inner (and outer) products of the spinors u(p) and v(p). ξ1 = 1 0 and ξ 2 = 0 1 (4. but it turns out ¯ that the former is needed when we come to quantize the theory.120) and η r † η s = δ rs (4.121) p · σ ξs ¯ We can take the inner product of four-component spinors in two different ways: either as u† · u. Meanwhile. or as u · u. s = 1.117) has helicity h = −1/2: it is left-handed. the field (4.118) where S ij is the rotation generator given in (4.2 Some Useful Formulae: Inner and Outer Products There are a number of identities that will be very useful in the following section.

It is: Claim: 2 (4. when we come to u† · v.128) where the two spinors are not now contracted. ξr † p · σ γ0 ¯ (p · σ )(p · σ)η s − ξ r † ¯ = ξ r† (p · σ )(p · σ)η s = 0 ¯ p · σ ηs √ ¯ − p · σ ηs (4.127) / us (p) us (p) = p + m ¯ s=1 (4. but instead placed back to back to give a 4 × 4 matrix. which we may write as v (p) = s p · σ ηs √ ¯ − p · σ ηs √ with and v r (p) · v s (p) = −2mδ rs ¯ √ v r † (p) · v s (p) = 2p0 δ rs (4. Also. ξ p·σ ¯ u (p) · v (−p) = ξ √ ¯ − p′ · σ η s = ξ r† (p · σ )(p′ · σ)η s − ξ r † ¯ (p · σ )(p′ · σ)η s ¯ (4. −p).We have analogous results for the negative frequency solutions.124) We can also compute the inner product between u and v. We have u (p) · v (p) = ξ ¯ r s r† √ √ p · σ . we have √ ′ p · σ ηs r †√ r† s r† √ p·σ.129) – 104 – . v r (p) · us (p) = 0. However. Defining the 4-momentum (p′ )µ = (p0 . 2 s=1 / v s (p) v s (p) = p − m ¯ (4. it is a slightly ¯ different combination that has nice properties (and this same combination appears when we quantize the theory). The same expression holds for (p · σ )(p · σ).125) and similarly. and the two terms cancel. with the 3-momentum in the spinor v taking the opposite sign. We look at ur † (p) · v s (−p). ¯ We learn ur † (p) · v s (−p) = v r † (p) · us (−p) = 0 Outer Products There’s one last spinor identity that we need before we turn to the quantum theory.126) Now the terms under the square-root are given by (p· σ )(p′ ·σ) = (p0 +pi σ i )(p0 −pi σ i ) = ¯ 2 2 2 ′ po − p = m .

v (4.Proof: 2 2 u (p) u (p) = ¯ s=1 s=1 s s √ √ p · σ ξs p·σξ ¯ s ξs † √ √ p · σ .131) which is the desired result. the 2 × 2 unit matrix. A similar proof works for s – 105 – .130) But s ξ s ξ s † = 1. which then gives us 2 us (p)¯s (p) = u s=1 m p·σ ¯ p·σ m v s (p)¯s (p). ξs † p · σ ¯ (4.

ψβ (y)] = δαβ δ (3) (x − y) (5. π= ∂L ¯ = iψγ 0 = iψ † ˙ ∂ψ (5. 5. the momentum conjugate to ψ is iψ † . Quantizing the Dirac Field We would now like to quantize the Dirac Lagrangian. But we’ll see that things go wrong and we will have to reconsider how to quantize this theory. the commutation relations of the fields imply commutation relations for the annihilation and creation operators – 106 – . which read † † [ψα (x). we may write the quantum operators as 2 ψ(x) = s=1 2 d3 p (2π)3 d3 p (2π)3 1 s s bp us (p)e+ip·x + cp † v s (p)e−ip·x 2Ep 1 s s bp † us (p)† e−ip·x + cp v s (p)† e+ip·x 2Ep (5. satisfying the canonical commutation relations.1) We will proceed naively and treat ψ as we did the scalar field.5. ψβ (y)] = [ψα (x). ¯ / L = ψ(x) i ∂ − m ψ(x) (5. As with the scalars.4) ψ † (x) = s=1 s s where the operators bp † create particles associated to the spinors us (p). ψβ (y)] = 0 † [ψα (x).2) For the Dirac Lagrangian. This is because we need only specify ψ and ψ † on an initial time slice to determine the full evolution. where any classical solution is a sum of plane waves. rather than second order. while cp † create particles associated to v s (p).3) It’s this step that we’ll soon have to reconsider. Since we’re dealing with a free theory. This is as it should be for an equation of motion that is first order in time.1 A Glimpse at the Spin-Statistics Theorem We start in the usual way and define the momentum. To quantize the theory. we promote the field ψ and its momentum ψ † to operators. It does not involve the time derivative of ψ.

so that [ψ(x).3) are equivalent to r s [cp . using p0 = Ep we have [ψ(x).5). ψ † (y)] = d3 p +ip·(x−y) e = δ (3) (x − y) (2π)3 (5. c† ] commutator terms in (5. It’s not yet obvious that it’s bad. [ψ(x). c† ] commutators reproduce the field commutators (5.3). in the second term.Claim: The field commutation relations (5.8) – 107 – . ψ † (y)] = = d3 p 1 / / ( p + m)γ 0 eip·(x−y) + ( p − m)γ 0 e−ip·(x−y) (2π)3 2Ep d3 p 1 (p0 γ 0 + pi γ i + m)γ 0 + (p0 γ 0 − pi γ i − m)γ 0 e+ip·(x−y) 3 2E (2π) p where. b† ] and [c. Using the momentum π = iψ † . let’s just carry on.6) u v (2π)3 2Ep At this stage we use the outer product formulae (4. Proof: Let’s show that the [b. Now. ψ † (y)] = r.129) which tell us s s / / us vs s u (p)¯ (p) = p + m and s v (p)¯ (p) = p − m. bq † ]ur (p)us (q)† ei(x·p−y·q) r s + [cp † . cq ]v r (p)v s (q)† e−i(x·p−y·q) = s d3 p 1 us (p)¯s (p)γ 0 eip·(x−y) + v s (p)¯s (p)γ 0 e−ip·(x−y) (5. Notice that it’s a little tricky in the middle there. This was the reason we needed the minus sign in the [c.5) with all other commutators vanishing. let’s construct the Hamiltonian for the theory. cq † ] = −(2π)3 δ rs δ (3) (p − q) r s [bp . bq † ] = (2π)3 δ rs δ (3) (p − q) (5. making sure that the pi γ i terms cancel.1. Note the strange minus sign in the [c. For now. c† ] term. 5. we’ve changed p → −p under the integration sign.1 The Hamiltonian To proceed. but we should be aware of it.s d3 p d3 q (2π)6 1 4Ep Eq r s [bp . we have ˙ ¯ H = π ψ − L = ψ(−iγ i ∂i + m)ψ (5.7) as promised.128) and (4.

We now wish to turn the Hamiltonian into an operator.11) (5. to replace (−γ i pi + m)us (p) = γ 0 p0 us (p) and (γ i pi + m)v s (p) = −γ 0 p0 v s (p) so we can write (−iγ i ∂i + m)ψ = d3 p (2π)3 Ep 0 s s s γ bp u (p) e+ip·x − cp † v s (p) e−ip·x 2 (5.12) and ur (p)† · v s (−p) = v r (p)† · us (−p) = 0 – 108 – .105) and (4.124) and (4.9) We now use this to write the operator Hamiltonian H = = d3 x ψ † γ 0 (−iγ i ∂i + m)ψ d3 x d3 p d3 q (2π)6 Ep r † r † −iq·x r b u (q) e + cq v r (q)† e+iq·x · 4Eq q s s bp us (p)e+ip·x − cp † v s (p)† e−ip·x = d3 p 1 r † s r † s r s b b [u (p) · u (p)] − cp cp † [v r (p)† · v s (p)] (2π)3 2 p p r r −bp † cs † [ur (p)† · v s (−p)] + cp bs [v r (p)† · v s (−p)] −p −p where.111). Now we use the defining equations for the spinors us (p) and v s (p) given in (4. for once we’ve left the sum over s = 1.92). so that p · x ≡ i xi pi = −xi pi . We now use our inner product formulae (4.10) (5.which means that H = d3 x H agrees with the conserved energy computed using Noether’s theorem (4. 2 implicit. There’s a small subtlety with the minus signs in deriving this equation that arises from the use of the Minkowski metric in contracting indices. in the last two terms we have relabelled p → −p.127) which read ur (p)† · us (p) = v r (p)† · v s (p) = 2p0 δ rs giving us H = = d3 p s s s s E bp † bp − cp cp † 3 p (2π) d3 p s s s s Ep bp † bp − cp † cp + (2π)3 δ (3) (0) (2π)3 (5. Let’s firstly look at (−iγ i ∂i + m)ψ = d3 p (2π)3 1 s s bp (−γ i pi + m)us (p) e+ip·x + cp † (γ i pi + m)v s (p) e−ip·x 2Ep where. (4.122).

you might think that it’s possible to rescue the theory to get the minus signs to work out right. r s r s r s r s {bp . † † † † [ap . it’s all gone a bit Pete Tong.4) of ψ and ψ † in terms of b. 5. q = |q.3). but you’ll always find this minus sign cropping up somewhere. Any attempt to do otherwise will lead to an inconsistency. But now the same proof that led us to (5. cq † } = {bp . p (5. So how do we go about quantizing a field as a fermion? Recall that when we quantized the scalar field. cq } = . You can play around with different things. (No relation). As the English would say.15) with all other anti-commutators vanishing. in fact.14) We still have the expansion (5. b† . aq ] = 0 ⇒ ap aq |0 ≡ |p.16) – 109 – . cq } = {bp . bq } = {cp . Since the above calculation was a little tricky. cq † } = (2π)3 δ rs δ (3) (p − q) (5. . bq † } = (2π)3 δ rs δ (3) (p − q) r s {cp . However the −c† c term is a disaster! The Hamiltonian is not bounded below.2 Fermionic Quantization The key piece of physics that we missed is that spin 1/2 particles are fermions. ψβ (y)} = {ψα (x). it’s telling us something important that we missed. {A. ψβ (y)} = δαβ δ (3) (x − y) (5. And. c and c† . ψβ (y)} = 0 † {ψα (x). Rather than (5. we need anti-commutation relations.5) tells us that r s {bp . such as the unbounded Hamiltonian we saw in (5. . the resulting particles obeyed bosonic statistics because the creation and annihilation operators satisfied the commutation relations. we will ask that the spinor fields satisfy † † {ψα (x).The δ (3) term is familiar and easily dealt with by normal ordering. This fact is embedded into the structure of relativistic quantum field theory: the spin-statistics theorem says that integer spin fields must be quantized as bosons. meaning that they obey Fermi-Dirac statistics with the quantum state picking up a minus sign upon the interchange of any two particles. = 0 (5. meaning that our quantum theory makes no sense.13) To have states obeying fermionic statistics.12). Taken seriously it would tell us that we could tumble to states of lower and lower energy by continually producing c† particles. while half-integer spin fields must be quantized as fermions. B} ≡ AB + BA.

18) Although b and c obey anti-commutation relations. we get H = = d3 p s s s s Ep bp † bp − cp cp † (2π)3 d3 p s s s s Ep bp † bp + cp † cp − (2π)3 δ (3) (0) (2π)3 (5. r does indeed carry intrinsic angular momentum 1/2 as expected. r = bp † |0 (5.17) The anti-commutators have saved us from the indignity of an unbounded Hamiltonian. if we wanted to be sure about the spin of the particle.96) to confirm that a stationary particle |p = 0. You can check that r r [H. Cosmological constant problem anyone?! 5. the Hamiltonian (5.20) The two particle states now satisfy r r |p1 .3 Dirac’s Hole Interpretation “In this attempt.1 Fermi-Dirac Statistics Just as in the bosonic case. In principle. we have the Pauli-Exclusion principle |p. we define the vacuum |0 to satisfy. r2 . s s bp |0 = cp |0 = 0 (5.2. p2 . p1 . Finally. just as in the bosonic case. At that stage. p. this could partially cancel the positive contribution from bosonic fields. In particular. we could act with the angular momentum operator (4. r1 . the success seems to have been on the side of Dirac rather than logic” Pauli on Dirac – 110 – .19) This means that we can again construct a tower of energy eigenstates by acting on the r r. r1 1 2 (5. For example.† vacuum by bp † and cp to create particles and antiparticles. r = 0. 5.21) confirming that the particles do indeed obey Fermi-Dirac statistics.The calculation of the Hamiltonian proceeds as before.11). Note that when normal ordering the Hamiltonian we now throw away a negative contribution −(2π)3 δ (3) (0). r. cp † ] = Ep cp † r r [H. r2 ≡ bp1 † bp2 † |0 = − |p2 . cp ] = −Ep cp (5. we have the one-particle states r |p. all the way through to the penultimate line (5. bp † ] = Ep bp † r r [H.17) has nice commutation relations with them. bp ] = −Ep bp and and r r [H.

all the negative energy states are filled. he / wrote (i ∂ − m)ψ = 0 as i ∂ψ ˆ = −iα · ∇ψ + mβψ ≡ Hψ ∂t (5. the plane-wave solutions (4.12). Moreover.104) and (4. the Hamiltonian of the ˆ system is H defined above. when a positron comes across – 111 – . Dirac postulated an ingenious solution to this problem: since the electrons are fermions (a fact which is put in by hand to Dirac’s theory) they obey the Pauli-exclusion principle. there are states v(p) with arbitrarily low energy −Ep . except it would carry positive charge.Let’s pause our discussion to make a small historical detour. Dirac finally concluded that the hole is a new particle: the positron. Dirac realized that his theory made a shocking prediction. Here the operator H is interpreted as the one-particle Hamiltonian. Suppose that a negative energy state is excited to a positive energy state. The hole would have all the properties of the electron.17). These filled negative energy states are referred to as the Dirac sea. The spectrum is once again unbounded below. To reinforce this interpretation. while for us the Hamiltonian is the field operator (5. Having avoided disaster by floating on an infinite sea comprised of occupied negative energy states. Dirac argued that only charge differences would be observable (a trick reminiscent of the normal ordering prescription we used for field operators). Let’s see where Dirac’s viewpoint leads. With the interpretation of ψ as a single-particle wavefunction. At first glance this is disastrous. with ψ = u(p) e−ip·x ψ = v(p) e+ip·x ⇒ ⇒ ∂ψ = Ep ψ ∂t ∂ψ i = −Ep ψ ∂t i (5. just like the unbounded field theory Hamiltonian (5. After flirting with the idea that it may be the proton. Dirac originally viewed his equation as a relativistic version of the Schr¨dinger equation. where ψ is a classical field that should be quantized. So we could simply stipulate that in the true vacuum of the universe.22) ˆ where α = −γ 0 γ and β = γ 0 . In Dirac’s view.110) to the Dirac equation are thought of as energy eigenstates. leaving behind a hole.23) which look like positive and negative energy solutions. with ψ interpreted o as the wavefunction for a single particle with spin. This is a very different viewpoint from the one we now have. Only the positive energy states are accessible. Although you might worry about the infinite negative charge of the vacuum.

ψ] ∂t We solve this by the expansion 2 (5. We define the spinors ψ(x. Quantization of this field then gives rise to both particle and anti-particle excitations. Experimentally we are concerned with particles.26) – 112 – . yet the old equations describe fields. t) at every point in spacetime such that they satisfy the operator equation ∂ψ = i[H.. When you begin with field equations. Dirac had predicted anti-matter. we recognize that the equations refer to a sublevel.4 Propagators Let’s now move to the Heisenberg picture. everyone thought that the Dirac equation referred directly to physical particles.” 5.. you operate on a level where the particles are not there from the start. the two can annihilate.an electron. ψβ (y)} (5.92) is positive definite.24) ψ(x) = s=1 2 d3 p (2π)3 d3 p (2π)3 1 s s bp us (p)e−ip·x + cp † v s (p)e+ip·x 2Ep 1 s s bp † us (p)† e+ip·x + cp v s (p)† e−ip·x 2Ep (5. we now understand that the interpretation of the Dirac spinor as a single-particle wavefunction is not really correct. It is instead to be thought of as a classical field which has only positive energy solutions because the Hamiltonian (4. This from Julian Schwinger: “Until now.25) ψ † (x) = s=1 Let’s now look at the anti-commutators of these fields. It is when you solve the field equations that you see the emergence of particles. For example. What we really learn from Dirac’s analysis is that there is no consistent way to interpret the Dirac equation as describing a single particle. anti-particles exist for both fermions and bosons.. in field theory. Dirac’s argument for anti-matter relies crucially on the particles being fermions while. It took only a couple of years before the positron was discovered experimentally in 1932. We define the fermionic propagator to be ¯ iSαβ = {ψα (x). as we have seen already in this course. one of the greatest achievements of theoretical physics. Although Dirac’s physical insight led him to the right answer. Now.

for example the Hamiltonian (5. The theory remains causal as long as fermionic operators are not observable. cq }v s (p)¯r (q)e+i(p·x−q·y) v = = d3 p 1 us (p)¯s (p)e−ip·(x−y) + v s (p)¯s (p)e+ip·(x−y) u v (2π)3 2Ep d3 p 1 / / ( p + m)e−ip·(x−y) + ( p − m)e+ip·(x−y) (2π)3 2Ep (5. we have already seen that D(x − y) − D(y − x) = 0. If you think this is a little weak. ψβ (y)} = 0 (x − y)2 < 0 (5.29) outside the lightcone.30) d3 p 1 −ip·(x−y) e (2π)3 2Ep (5. We can then write / iS(x − y) = (i ∂ x + m)(D(x − y) − D(y − x)) (5.27) where to reach the final line we have used the outer product formulae (4. φ(y)] = 0 (x − y)2 < 0 (5.31) outside the lightcone.90) D(x − y) = Some comments: • For spacelike separated points (x − y)2 < 0. Inserting the expansion (5. but you should remember that S(x−y) is a 4×4 matrix.128) and (4. can be written as (2. recall.¯ In what follows we will often drop the indices and simply write iS(x−y) = {ψ(x). which we trumpeted as proof that our theory was causal. These still commute outside the lightcone. What happened to our precious causality? The best that we can say is that all our observables are bilinear in fermions. bq † }us (p)¯r (q)e−i(p·x−q·y) u 4Ep Eq s r +{cp † .28) in terms of the propagator for a real scalar field D(x − y) which. we have iS(x − y) = d3 p d3 q (2π)6 1 s r {bp .17). ψ(y)}. we made a big deal of this since it ensured that [φ(x).25). In the bosonic theory.129). for fermions we now have {ψα (x). However. remember that no one has ever seen a physical measuring apparatus come back to minus itself when you rotate by 360 degrees! – 113 – .

so that SF is a Green’s function for the Dirac operator.34) also occurs for any string of operators inside a time ordered product T (. In this case the minus sign is ¯ necessary to make the two definitions agree since {ψ(x). The minus sign that we see in (5. there is no invariant way to determine whether x0 > y 0 or y 0 > x0 . 5.34) Notice the minus sign! It is necessary for Lorentz invariance.32) / which follows from the fact that ( ∂ x + m2 )D(x − y) = 0 using the mass shell 2 2 condition p = m . With the understanding that all fermionic operators anti-commute inside T and ::. When (x−y)2 < 0.35) / which satisfies (i ∂ x − m)SF (x − y) = iδ (4) (x − y). we can determine the vacuum expectation value. as the time ordered product.• At least away from singularities. . fermionic operators anti-commute.5 The Feynman Propagator By a similar calculation to that above. We have this same behaviour for normal ordered products as well. the propagator satisfies / (i ∂ x − m)S(x − y) = 0 2 (5.).36) – 114 – . ¯ SF (x − y) = 0| T ψ(x)ψ(y) |0 ≡ ¯ 0| ψ(x)ψ(y) |0 ¯ 0| − ψ(y)ψ(x) |0 x0 > y 0 y 0 > x0 (5. While bosonic operators commute inside T . SF (x − y) = i d4 p −ip·(x−y) γ · p + m e (2π)4 p2 − m2 + iǫ (5. with fermionic operators obeying : ψ1 ψ2 := − : ψ2 ψ1 :. We define the contraction ¯ ¯ ¯ ψ(x)ψ(y) = T (ψ(x)ψ(y)) − : ψ(x)ψ(y) : = SF (x − y) (5. . which is again a 4 × 4 matrix. Wick’s theorem proceeds just as in the bosonic case.33) We now define the Feynman propagator SF (x − y). ψ(y)} = 0 outside the lightcone. ¯ 0| ψα (x)ψβ (y) |0 = ¯ 0| ψα (y)ψβ (x) |0 = d3 p 1 / ( p + m)αβ e−ip·(x−y) (2π)3 2Ep d3 p 1 / ( p − m)αβ e+ip·(x−y) (2π)3 2Ep (5. We have the 4-momentum integral representation for the Feynman propagator.

with that calculation as a guide. s. Thus. the fermions can be leptons (such as the electron) or quarks. the coupling is dimensionless: [λ] = 0.6 Yukawa Theory The interaction between a Dirac fermion of mass m and a real scalar field of mass µ is governed by the Yukawa theory. we will again refer to the φ particles as mesons. the nucleons have spin. 1 ¯ ¯ L = 1 ∂µ φ∂ µ φ − 2 µ2 φ2 + ψ(iγ µ ∂µ − m)ψ − λφψψ 2 (5.38) We need to be a little cautious about minus signs. This is the same calculation we performed in Section (3.5. We still have [φ] = 1. between fermions and the Higgs boson. Note the dimensions of the various fields. r ′ (5. We’ll turn to this briefly in Section 5. ¯ ¯ d4 x1 d4 x2 T ψ(x1 )ψ(x1 )φ(x1 ) ψ(x2 )ψ(x2 )φ(x2 ) (5.3. Moreover. Yukawa originally proposed an interaction of this type as an effective theory of nuclear forces.3) except now the fermions have spin. firstly computing the amplitude of a particular scattering process then. this time. We have f| = (−iλ)2 2 r s 4Ep′ Eq′ 0| bq ′ bp ′ ′ ′ (5. (This is still not a particularly realistic theory of nucleon interactions.39) We want to calculate the order λ2 terms from the S-matrix element f | S − 1 |i .40) – 115 – . r ′ ′ s r 4Ep′ Eq′ bp ′† bq ′ † |0 ≡ |p ′ . but the kinetic terms require that [ψ] = 3/2.37) which is the proper version of the baby scalar Yukawa theory we looked at in Section 3. Our initial and final states are |i = |f = r s 4Ep Eq bp † bq † |0 ≡ |p. because the b† ’s now anti-commute. in Nature the relevant mesons are pions which are pseudoscalars. In particular. s′ . not least because we’re omitting isospin. and the ψ particles as nucleons. We’ll proceed as we did in Section 3.6. unlike in the case with only scalars. q. we should be careful when we take the adjoint. writing down the Feynman rules for the theory. q ′ .1 An Example: Putting Spin on Nucleon Scattering Let’s study ψψ → ψψ scattering. With an eye to this.3). Except. We start with: 5. In that context. Couplings of this type appear in the standard model. so ¯ a coupling of the form φψγ 5 ψ would be more appropriate.7.

the contribution to nucleon scattering comes from the contraction ¯ ¯ : ψ(x1 )ψ(x1 )ψ(x2 )ψ(x2 ) : φ(x1 )φ(x2 ) (5. Now let’s see what happens when we hit this with f |. all fields are in the interaction picture. the 1/ E terms cancel the relativistic state normalization. The ¯ minus sign that sits out front came from moving ψ(x1 ) past ψ(x2 ). ¯ leaving the ψ fields alone for now.43) Note. Now anti-commuting † the b’s past the b ’s. Putting everything together.where.41) We just have to be careful about how the spinor indices are contracted. Meanwhile. we have the following expression for f | S − 1 |i (−iλ)2 d4 x1 d4 x2 d4 k ieik·(x1 −x2 ) (2π)4 k 2 − µ2 + iǫ [¯s (p ′ ) · us (p)] [¯r (q ′ ) · ur (q)]e+ix1 ·(q −q)+ix2 ·(p −p) u u − [¯s (p ′ ) · ur (q)] [¯r (q ′ ) · us (p)]eix1 ·(p −q)+ix2 ·(q −p) u u ′ ′ ′ ′ ′ ′ ′ ′ – 116 – . Just as in the bosonic calculation. cancelling the √ factor of 1/2 in front of (5. the relative minus sign that appears between these two terms. we get = −1 ¯ ¯ [ψ(x1 ) · ur (q)] [ψ(x2 ) · us (p)]e−ip·x2 −iq·x1 2 Ep Eq ¯ ¯ − [ψ(x1 ) · us (p)] [ψ(x2 ) · ur (q)]e−ip·x1 −iq·x2 |0 (5.43) doubles up with this.40). in particular. We have s r ¯ ¯ : ψ(x1 )ψ(x1 ) ψ(x2 )ψ(x2 ) : bp † bq † |0 = − d3 k1 d3 k2 ¯ ¯ [ψ(x1 ) · um (k1 )] [ψ(x2 ) · un (k2 )] (2π)6 e−ik1 ·x1 −ik2 ·x2 m n s † r † b b b b |0 (5.42) 4Ek1 Ek2 k1 k2 p q where we’ve used square brackets [·] to show how the spinor indices are contracted. We expand out the ψ fields. as usual. We may ignore the c† pieces in ψ since they give no contribution at order λ2 . We look at r s ¯ ¯ 0| bq′ bp′ [ψ(x1 ) · ur (q)] [ψ(x2 ) · us (p)] |0 = ′ ′ e+ip ·x1 +iq ·x2 s′ ′ ′ [¯ (p ) · ur (q)] [¯r (q ′ ) · us (p)] u u 2 Ep′ Eq′ − e+ip ·x2 +iq ·x1 r′ ′ ′ [¯ (q ) · ur (q)] [¯s (p ′ ) · us (p)] u u 2 Ep′ Eq′ ′ ′ ′ ′ ¯ ¯ The [ψ(x1 ) · us (p)] [ψ(x2 ) · ur (q)] term in (5. Let’s start by looking at how the fermionic operators act on |i .

For outgoing anti-fermions we associate v r (p).where we’ve put the φ propagator back in.7 Feynman Rules for Fermions It’s important to bear in mind that the calculation we just did kind of blows. d4 k (2π)4 i(−iλ)2 k 2 − µ2 + iǫ [¯s (p ′ ) · us (p)] [¯r (q ′ ) · ur (q)]δ (4) (q ′ − q + k)δ (4) (p′ − p − k) u u − [¯s (p ′ ) · ur (q)] [¯r (q ′ ) · us (p)]δ (4) (p′ − q + k)δ (4) (q ′ − p − k) u u ′ ′ ′ ′ And we’re almost there! Finally. Performing the integrals over x1 and x2 . we associate a spinor v r (p). – 117 – . we have A = (−iλ) 2 [¯s (p ′) · us (p)] [¯r (q ′ ) · ur (q)] [¯s (p ′ ) · ur (q)] [¯r (q ′ ) · us (p)] u u u u − ′ − p)2 − µ2 + iǫ ′ − p)2 − µ2 + iǫ (p (q ′ ′ ′ ′ which is our final answer for the amplitude. Thankfully the Feynman rules will once again encapsulate the combinatoric complexities and make life easier for us. we associate a spinor ur (p). 5. ¯ p ur(p) p ur(p) Figure 21: An incoming fermion Figure 22: An outgoing fermion • To each incoming anti-fermion with momentum p and spin r. For outgoing fermions we associate ur (p). this becomes. writing the S-matrix element in terms of the amplitude in the usual way. f | S − 1 |i = iA(2π)4 δ (4) (p + q − p′ − q ′ ). ¯ p vr(p) p vr(p) Figure 23: An incoming anti-fermion Figure 24: An outgoing anti-fermion • Each vertex gets a factor of −iλ. The rules to compute amplitudes are the following • To each incoming fermion with momentum p and spin r.

s / p. Using the Feynman rules we can read off the amplitude. or with external spinors u.s/ p. either with further propagators. i p for scalars 2 − µ2 + iǫ p p / i( p + m) for fermions 2 − m2 + iǫ p (5. A = (−iλ) 2 u u [¯s (p ′ ) · us (p)] [¯r (q ′ ) · ur (q)] [¯s (p ′ ) · ur (q)] [¯r (q ′ ) · us (p)] u u − ′ )2 − µ 2 ′ )2 − µ 2 (p − p (p − q ′ ′ ′ ′ (5.7. The matrix indices are contracted at each vertex.s/ / p. Firstly. • Add extra minus signs for statistics. The two diagrams above are different because the incoming legs are attached to different outgoing legs).r / / q. Some examples will be given below.r / q. v or v. 5. ¯ ¯ • Impose momentum conservation at each vertex.44) The arrows on the fermion lines must flow consistently through the diagram (this ensures fermion number conservation). the two lowest order Feynman diagrams are shown in Figure 25. (Note that the way the legs point in the Feynman diagram doesn’t tell us the direction in which the particles leave the scattering event: the momentum label does that. We’ve drawn the second Feynman diagram with the legs crossed p.r q. and integrate over undetermined loop momenta.r Figure 25: The two Feynman diagrams for nucleon scattering to emphasize the fact that it picks up a minus sign due to statistics. Note that the fermionic propagator is a 4×4 matrix.1 Examples Let’s run through the same examples we did for the scalar Yukawa theory. we have Nucleon Scattering For the example we worked out previously.45) – 118 – .s + / q. u.• Each internal line gets a factor of the relevant propagator.

The denominators in each term are due to the meson propagator, with the momentum determined by conservation at each vertex. This agrees with the amplitude we computed earlier using Wick’s theorem. Nucleon to Meson Scattering ¯ Let’s now look at ψ ψ → φφ. The two lowest order Feynman diagrams are shown in Figure 26.

p,s
/ p,s/

p,s
/ p,s/

+
/ q,r / / q,r /

q,r

q,r

Figure 26: The two Feynman diagrams for nucleon to meson scattering

Applying the Feynman rules, we have A = (−iλ)2
′ v r(q)[γ µ (pµ − p′µ ) + m]us (p) v r(q)[γ µ (pµ − qµ ) + m]us (p) ¯ ¯ + (p − p′ )2 − m2 (p − q ′ )2 − m2

Since the internal line is now a fermion, the propagator contains γµ (pµ −p′µ )+m factors. This is a 4 × 4 matrix which sits on the top, sandwiched between the two external spinors. Now the exchange statistics applies to the final meson states. These are bosons and, correspondingly, there is no relative minus sign between the two diagrams. Nucleon-Anti-Nucleon Scattering ¯ ¯ For ψ ψ → ψ ψ, the two lowest order Feynman diagrams are of two distinct types, just like in the bosonic case. They are shown in Figure 27. The corresponding amplitude is given by, A = (−iλ)
2

v u [¯s (p ′ ) · us (p)] [¯r (q) · v r (q ′ )] [¯r(q) · us (p)] [¯s (p ′ ) · v r (q ′ )] u v + − ′ )2 − µ 2 2 − µ2 + iǫ (p − p (p + q)

(5.46)

As in the bosonic diagrams, there is again the difference in the momentum dependence in the denominator. But now the difference in the diagrams is also reflected in the spinor contractions in the numerator.

– 119 –

p,s
/ p,s /

p,s

/ p,s

/

+
/ q,r / / q,r /

q,r

q,r

Figure 27: The two Feynman diagrams for nucleon-anti-nucleon scattering

More subtle are the minus signs. The fermionic statistics mean that the first diagram has an extra minus sign relative to the ψψ scattering of Figure 25. Since this minus sign will be important when we come to figure out whether the Yukawa force is attractive or repulsive, let’s go back to basics and see where it comes from. The initial and final states for this scattering process are |i = |f =
r s 4Ep Eq bp † cq † |0 ≡ |p, s; q, r
′ ′

s r 4Ep′ Eq′ bp ′† cq ′ † |0 ≡ |p ′ , s′; q ′ , r ′

(5.47)

¯ ¯ The ordering of b† and c† in these states is crucial and reflects the scattering ψ ψ → ψ ψ, ¯ ¯ as opposed to ψ ψ → ψψ which would differ by a minus sign. The first diagram in Figure 27 comes from the term in the perturbative expansion,
s r s r ¯ ¯ ¯ f | : ψ(x1 )ψ(x1 ) ψ(x2 )ψ(x2 ) : bp † cq † |0 ∼ f | [¯m (k1 ) · ψ(x1 )] [ψ(x2 ) · un (k2 )]cm bn2 bp † cq † |0 v k1 k

where we’ve neglected a bunch of objects in this equation like d4 ki and exponential factors because we only want to keep track of the minus signs. Moving the annihilation operators past the creation operators, we have ¯ + f | [¯r (q) · ψ(x1 )] [ψ(x2 ) · us (p)] |0 v (5.48)

¯ Repeating the process by expanding out the ψ(x1 ) and ψ(x2 ) fields and moving them to the left to annihilate f |, we have
r s 0| cq ′ bp ′ cm † bn † [¯r (q) · v m (l1 )] [¯n (l2 ) · us (p)] |0 ∼ −[¯r (q) · v r (q ′ )] [¯s (p ′) · us (p)] v u v u l l
1 2 ′ ′ ′ ′

s where the minus sign has appeared from anti-commuting cm † past bp ′ . This is the l1 overall minus sign found in (5.46). One can also follow similar contractions to compute the second diagram in Figure 27.

Meson Scattering

– 120 –

Finally, we can also compute the scattering of φφ → φφ which, as in the bosonic case, picks up its leading contribution at one-loop. The amplitude for the diagram shown in the figure is iA = −(−iλ)4

p

1

p1

/

p2 p2

/

/ / /′ k+m k + p1 + m d4 k Tr 2 Figure 28: (2π)4 (k − m2 + iǫ) ((k + p′1 )2 − m2 + iǫ) / /′ / / /′ k + p1 − p1 + m k − p2 + m × ′ ′ ((k + p1 − p1 )2 − m2 + iǫ) ((k − p2 )2 − m2 + iǫ)

Notice that the high momentum limit of the integral is d4 k/k 4 , which is no longer finite, but diverges logarithmically. You will have to wait until next term to make sense of this integral. There’s an overall minus sign sitting in front of this amplitude. This is a generic feature of diagrams with fermions running in loops: each fermionic loop in a diagram gives rise to an extra minus sign. We can see this rather simply in the diagram

which involves the expression ¯ ¯ ¯ ¯ ψα (x) ψα (x)ψβ (y) ψβ (y) = − ψβ (y)ψα (x) ψα (x)ψβ (y)

= −Tr (SF (y − x)SF (x − y))

After passing the fermion fields through each other, a minus sign appears, sitting in front of the two propagators. 5.7.2 The Yukawa Potential Revisited We saw in Section 3.5.2, that the exchange of a real scalar particle gives rise to a universally attractive Yukawa potential between two spin zero particles. Does the same hold for the spin 1/2 particles? Recall that the strategy to compute the potential is to take the non-relativistic limit of the scattering amplitude, and compare with the analogous result from quantum mechanics. Our new amplitude now also includes the spinor degrees of freedom u(p)

– 121 –

s / p.46). which is ′ ′ v s · v s = −2mδ ss . t) (5. we could instead consider ¯ LYuk = −λφψγ 5 ψ This still preserves parity if φ is a pseudoscalar. p). the Feynman diagrams for ψψ → ψψ scattering are again given by Figure 25. U(r) = − λ2 e−µr 4πr (5.52) We can compute in this theory very simply: the Feynman rule for the interaction vertex is now changed to a factor of −iλγ 5 . while the momentum dependence is the same as in the bosonic case. with the amplitude now A = (−iλ) 2 u u [¯s (p ′)γ 5 us (p)] [¯r (q ′ )γ 5 ur (q)] [¯s (p ′ )γ 5 ur (q)] [¯r (q ′ )γ 5 us (p)] u u − ′ )2 − µ 2 ′ )2 − µ 2 (p − p (p − q ′ ′ ′ ′ – 122 – . These two signs cancel. For example.49) In this limit.45) ′ ′ become us · us = 2mδ ss and the amplitude is ¯ p. p → (m.50) The δ symbols tell us that spin is conserved in the non-relativistic limit. telling us that once again the particles feel an attractive Yukawa potential.3 Pseudo-Scalar Coupling Rather than the standard Yukawa coupling.e. In the non-relativistic limit. 5.and v(p).r = −i(−iλ) (2m) 2 δs sδr r δs r δr s − (p − p′ ) + µ2 (p − q ′ ) + µ2 ′ ′ ′ ′ (5. t) → −φ(−x. i.s/ / q. and √ √ p · σξ u(p) = √ → m p · σξ ¯ √ √ p · σξ → m v(p) = √ ¯ − p · σξ ξ ξ ξ −ξ (5. the spinor contractions in the amplitude for ψψ → ψψ scattering (5. there are two minus signs which cancel each other.53) (5.46).51) ¯ ¯ Repeating the calculation for ψ ψ → ψ ψ. giving us once again an attractive Yukawa ¯ potential (5. due to the fermionic nature of the particles. The second minus sign comes from the non-relativistic limit of the spinor contraction for anti-particles in (5. P : φ(x.7.51).r / q. The first is the extra overall minus sign in the scattering amplitude (5.

¯ ′ we must go to next to leading order.49).r / q.s / p. p. But this time. So. Using the expressions for the spinors (5.54) – 123 – .We could again try to take the non-relativistic limit for this amplitude. To find the non-relativistic amplitude.r [ξ s T (p − p ′ ) · σξ s ] [ξ r T (p − p ′ ) · σξ r ] → +im(−iλ) (p − p ′ )2 + µ2 2 ′ ′ (5. things work a little differently. in the non-relativistic limit.s/ / q. we ′ have us γ 5 us → 0 in the non-relativistic limit. the leading order amplitude arising from pseudoscalar exchange is given by a spin-spin coupling. One can easily check that us (p ′ )γ 5 us (p) → ¯ ′ m ξ s T (p − p ′ ) · σξ s .

If we define Aµ = (φ. A). We then describe how to couple the photon to fermions and to bosons. becomes 0 Ex 0 Bz Ey −Bz 0 Bx Ez By −Bx 0 ∂A ∂t and B = ∇ × A (6.1 Maxwell’s Equations The Lagrangian for Maxwell’s equations in the absence of any sources is simply 1 L = − Fµν F µν 4 where the field strength is defined by Fµν = ∂µ Aν − ∂ν Aµ The equations of motion which follow from this Lagrangian are ∂µ ∂L ∂(∂µ Aν ) = −∂µ F µν = 0 (6.6) −Ez −By The Bianchi identity (6. Quantum Electrodynamics In this section we finally get to quantum electrodynamics (QED). the field strength also satisfies the Bianchi identity ∂λ Fµν + ∂µ Fνλ + ∂ν Fλµ = 0 (6.4) To make contact with the form of Maxwell’s equations you learn about in high school. Our path to quantization will be as before: we start with the free theory of the electromagnetic field and see how the quantum theory gives rise to a photon with two polarization states.5) Fµν = −Ex −Ey (6.3) (6. ∇·B =0 and ∂B = −∇ × E ∂t (6. in terms of Fµν . the theory of light interacting with charged matter.6. then the electric field E and magnetic field B are defined by E = −∇φ − which.2) (6.4) then gives two of Maxwell’s equations. 6.7) – 124 – .1) Meanwhile. we need some 3-vector notation. from the definition of Fµν .

∇·E =0 and ∂E =∇×B ∂t (6.1.1 Gauge Symmetry The massless vector field Aµ has 4 components. then the field A0 is fully determined by the equation of motion ∇ · E = 0 which. 6. expanding out. the equations of motion give the remaining two Maxwell equations. The field strength is invariant under the gauge symmetry: Fµν → ∂µ (Aν + ∂ν λ) − ∂ν (Aµ + ∂µ λ) = Fµν (6. It looks like we have only 3 degrees of freedom in Aµ rather than 4.10) ∂A =0 ∂t (6.8) As we will see shortly.3) has a very large symmetry group. We’ll ask only that λ(x) dies off suitably quickly at spatial x → ∞. How are we going to resolve this discrepancy? There are two related comments which will ensure that quantizing the gauge field Aµ gives rise to 2 degrees of freedom.9) So A0 is not independent: we don’t get to specify A0 on the initial time slice. ˙ • The field A0 has no kinetic term A0 in the Lagrangian: it is not dynamical. Meanwhile. acting on the vector potential as Aµ (x) → Aµ (x) + ∂µ λ(x) (6.12) – 125 – . We call this a gauge symmetry. • The Lagrangian (6.11) for any function λ(x). which would naively seem to tell us that the gauge field has 4 degrees of freedom. in the presence of charged matter these equations pick up extra terms on the right-hand side. Yet we know that the photon has only two degrees of freedom which we call its polarization states. But this is still one too many. rather than 4.These remain true even in the presence of electric sources. This ˙ means that if we are given some initial data Ai and Ai at a time t0 . reads ∇2 A0 + ∇ · This has the solution A0 (x) = d3 x′ (∇ · ∂ A/∂t)(x ′ ) 4π|x − x ′ | (6.

We will see shortly that we also require the Figure 29: gauge potential to describe classically charged fields. Since gauge invariance is a redundancy of the system. foliated by gauge orbits as shown in the figure. without using the gauge potential Aµ . we have no way to uniquely determine Aµ at a later time since we can’t distinguish between Aµ and Aµ + ∂µ λ. two states related by a gauge symmetry are to be identified: they are the same physical state. That is.3. However. such as the Aharonov-Bohm effect. then our problems disappear. if we’re happy to identify Aµ and Aµ + ∂µ λ as corresponding to the same physical state. it appears that we have to introduce quantities Aµ that we can never measure. Gauge Orbits Gauge Fixing we might try to formulate the theory purely in terms of the local. To describe Nature. for example ψ → eiα ψ for a complex scalar field. But it is not possible to describe certain quantum phenomena. One way to see that this interpretation is necessary is to notice that Maxwell’s equations are not sufficient to specify the evolution of Aµ . (There is a small caveat to this statement which is explained in Section 6. the gauge symmetry is to be viewed as a redundancy in our description. This is fine for the free classical theory: Maxwell’s equations were. This means that given any initial data. While the latter take a physical state to another physical state with the same properties. All states that lie along a given – 126 – . Do we now have an infinite number of conservation laws? The answer is no! Gauge symmetries have a very different interpretation than the global symmetries that we make use of in Noether’s theorem. after all.13) But the operator [ηµν (∂ ρ ∂ρ )−∂µ ∂ν ] is not invertible: it annihilates any function of the form ∂µ λ. This would be problematic if we thought that Aµ is a physical object.So what are we to make of this? We have a theory with an infinite number of symmetries. gauge invariant objects E and B. The picture that emerges for the theory of electromagnetism is of an enlarged phase space. Previously we only encountered symmetries which act the same at all points in spacetime. Noether’s theorem told us that these symmetries give rise to conservation laws. first written in terms of E and B. physical. The equations read.1). one for each function λ(x). [ηµν (∂ ρ ∂ρ ) − ∂µ ∂ν ] Aν = 0 (6.

Coulomb gauge is sometimes called radiation gauge. so may not be ideal for some purposes. To make progress. Coulomb gauge breaks Lorentz invariance. 3 Named after Lorenz who had the misfortune to be one letter away from greatness.line can be reached by a gauge transformation and are identified.15) (This equation will no longer hold in Coulomb gauge in the presence of charged matter). Picking a gauge is rather like picking coordinates that are adapted to a particular problem. they’re all physically equivalent. (The argument is the same as before). it is very useful to exhibit the physical degrees of freedom: the 3 components of A satisfy a single constraint: ∇ · A = 0. These will be identified with the two polarization states of the photon.10). Since A0 is fixed by (6. Then we ′ choose Aµ = Aµ + ∂µ λ. some of which will be more useful in different situations. It doesn’t matter which representative we pick — after all. Different representative configurations of a physical state are called different gauges. There are many possibilities. In fact this condition doesn’t pick a unique representative from the gauge orbit.14) This equation always has a solution. – 127 – . leaving behind just 2 degrees of freedom. which also has non-trivial solutions. in which we cut the orbits. the Lorentz gauge3 has the advantage that it is Lorentz invariant. We’re always free to make further gauge transformations with ∂µ ∂ µ λ = 0. • Coulomb Gauge: ∇ · A = 0 We can make use of the residual gauge transformations in Lorentz gauge to pick ∇ · A = 0. Moreover. different gauges often reveal slightly different aspects of a problem. we have as a consequence A0 = 0 (6. Here we’ll look at two different gauges: • Lorentz Gauge: ∂µ Aµ = 0 To see that we can always pick a representative configuration satisfying ∂µ Aµ = 0. suppose that we’re handed a gauge field A′µ satisfying ∂µ (A′ )µ = f (x). we pick a representative from each gauge orbit. But we should make sure that we pick a “good” gauge. As the name suggests. where ∂µ ∂ µ λ = −f (6. However.

the equation of motion for A is ∂µ ∂ µ A = 0 which we can solve in the usual way. 6. the momentum conjugate to Ai is our old friend. along the way. Let’s now see how to treat this system using different gauge fixing conditions.16) so the momentum π 0 conjugate to A0 vanishes.6. π0 = ∂L =0 ˙ ∂ A0 ∂L πi = = −F 0i ≡ E i ˙i ∂A (6.2 The Quantization of the Electromagnetic Field In the following we shall quantize free Maxwell theory twice: once in Coulomb gauge.2. see that each method comes with its own subtleties. The constraint ∇ · A = 0 tells us that ξ must satisfy 0 ξ·p=0 (6. H = = ˙ d3 x π i Ai − L 1 1 d3 x 2 E · E + 2 B · B − A0 (∇ · E) (6. the electric field. The first of these subtleties is common to both methods and comes when computing the momentum π µ conjugate to Aµ .18) which is now a constraint on the system in which A are the physical degrees of freedom.19) with p2 = |p|2 . We’ll ultimately get the same answers and.20) (6. Meanwhile.17) So A0 acts as a Lagrange multiplier which imposes Gauss’ law ∇·E =0 (6.1 Coulomb Gauge In Coulomb gauge. A= d3 p ξ(p) eip·x (2π)3 (6. and again in Lorentz gauge. We can compute the Hamiltonian. This is the mathematical consequence of the statement we made above: A0 is not a dynamical field.21) – 128 – .

now imposed on the operators. Aj (y)] = [E i (x). “Lectures on Quantum Mechanics” – 129 – . But from the commutator relations above. The correct Poisson bracket structure leads to an alteration of the last commutation relation. this is a problem already in the classical theory. We can pick ξ(p) to be a linear combination of two orthonormal vectors ǫr . E j (y)] = 0 j [Ai (x). Ej (y)] = i δij − ∂i ∂j ∇2 δ (3) (x − y) (6. see the small book by Paul Dirac. each of which satisfies ǫr (p) · p = 0 and ǫr (p) · ǫs (p) = δrs r.24) What’s going on? In imposing the commutator relations (6. Ej (y)] = i 4 d3 p (2π)3 δij − pi pj |p| 2 ipi eip·(x−y) = 0 (6. we can rewrite the right-hand side of the commutator in momentum space. Ej (y)] = i d3 p (2π)3 δij − pi pj |p| 2 eip·(x−y) (6.26) which is now consistent with the constraints.23) we haven’t correctly taken into account the constraints. Naively we would write [Ai (x). [Ai (x). for example [∂i Ai (x). r = 1. To quantize we turn the Poisson brackets into commutators. 2. But this topological fact doesn’t cause any complications in computing QED scattering processes.25) To see that this is now consistent with the constraints. because it’s not consistent with the constraints. We still want to have ∇ · A = ∇ · E = 0.which means that ξ is perpendicular to the direction of motion p. E j (y)] = iδi δ (3) (x − y) (6. It’s worth pointing out that you can’t consistently pick a continuous basis of polarization vectors for every value of p because you can’t comb the hair on a sphere. where the Poisson bracket structure is already altered4 . 2 (6. ∇ · E(y)] = i∇2 δ (3) (x − y) = 0 (6. In fact. s = 1. [Ai (x). we see [∇ · A(x).27) For a nice discussion of the classical and quantum dynamics of constrained systems.22) These two vectors correspond to the two polarization states of the photon.23) But this can’t quite be right.

we get the simple expression H= d3 p |p| (2π)3 2 r r ap † ap r=1 (6. aq † ] = (2π)3 δ rs δ (3) (p − q) (6.30) where.28) r=1 where. we need the completeness relation for the polarization vectors.31) You can easily check that this equation is true by acting on both sides with a basis of vectors (ǫ1 (p). we will have to fight to massage this propagator into something a little nicer. However. One place where this manifests itself is in the propagator for the fields Ai (x) (in the Heisenberg picture).29) It is not hard to show that the commutation relations (6. A(x) = E(x) = d3 p (2π)3 3 1 2|p| 2 r r ǫr (p) ap eip·x + ap † e−ip·x r=1 dp (−i) (2π)3 |p| 2 2 r r ǫr (p) ap eip·x − ap † e−ip·x (6.32) The Coulomb gauge has the advantage that the physical degrees of freedom are manifest. 2 r=1 ǫi (p)ǫj (p) = δ ij − r r pi pj |p| 2 (6. in deriving this. After normal ordering. We derive the Hamiltonian by substituting (6.33) The tr superscript on the propagator refers to the “transverse” part of the photon.17).25) are equivalent to the usual commutation relations for the creation and annihilation operators. we’ve lost all semblance of Lorentz invariance. In Coulomb gauge the propagator reads tr Dij (x − y) ≡ 0| T Ai (x)Aj (y) |0 = d4 p i (2π)4 p2 + iǫ δij − pi pj |p|2 e−ip·(x−y) (6.28) into (6. When we turn to the interacting theory. – 130 – . and playing around with ǫr polarization vectors. aq † ] = 0 s r [ap . The last term vanishes in Coulomb gauge. p). aq ] = [ap † . ǫ2 (p).We now write A in the usual mode expansion. r s r s [ap . as before. the polarization vectors satisfy ǫr (p) · p = 0 and ǫr (p) · ǫs (p) = δrs (6.

we will also have to deal with the residual gauge symmetry of this theory which will prove a little tricky. We will use α = 1.36) (6. We choose to change the theory so that (6. we could be a little more general than this. different choices of α are sometimes also referred to as different “gauges”. α = 0.34) arises directly through the equations of motion. and consider the Lagrangian 1 L = − 4 Fµν F µν − 1 (∂µ Aµ )2 2α (6.39) – 131 – . [Aµ (x).) Our plan will be to quantize the theory (6.2. π ν (y)] = 0 [Aµ (x). The equations of motion that follow from the action are then ∂µ ∂ µ Aν = 0 (6.6. As we’ll see. which is called “Feynman gauge”.2 Lorentz Gauge We could try to work in a Lorentz invariant fashion by imposing the Lorentz gauge condition ∂µ Aµ = 0. Aν (y)] = [π µ (x). we can proceed very easily. is called “Landau gauge”. The other common choice.36). rather confusingly.34) Our approach to implementing Lorentz gauge will be a little different from the method we used in Coulomb gauge.35) (In fact. and only later impose the constraint ∂µ Aµ = 0 in a suitable manner on the Hilbert space of the theory. At first. because both π 0 and π i are dynamical: π0 = ∂L = −∂µ Aµ ˙ ∂ A0 ∂L ˙ πi = = ∂ i A0 − Ai ˙i ∂A (6. πν (y)] = iηµν δ (3) (x − y) (6.38) Turning these classical fields into operators. We can achieve this by taking the Lagrangian 1 1 L = − Fµν F µν − (∂µ Aµ )2 4 2 The equations of motion coming from this action are ∂µ F µν + ∂ ν (∂µ Aµ ) = ∂µ ∂ µ Aν = 0 (6. we can simply impose the usual commutation relations. The quantization of the theory is independent of α and.37) with arbitrary α and reach similar conclusions.

1. instead of the two polarization 3vectors that we met in the Coulomb gauge.2. Aµ (x) = π (x) = µ d3 p (2π)3 dp (2π)3 3 1 2|p| 3 λ λ ǫλ (p) ap eip·x + ap † e−ip·x µ λ=0 3 λ λ (ǫµ )λ (p) ap eip·x − ap † e−ip·x |p| (+i) 2 (6.39) into those λ λ′ λ λ′ for creation and annihilation operators. We find [ap . since (6.and we can make the usual expansion in terms of creation and annihilation operators and 4 polarization vectors (ǫµ )λ . we pick ǫ0 to be timelike. rather than the familiar (−i) that we’ve seen so far. aq † ] = −η λλ (2π)3 δ (3) (p − q) ′ ′ (6. In the Heisenberg picture.41) (6. then 1 0 0 0 ǫ = 0 0 0 0 . it becomes clear that this descends to (+i) in the definition of momentum. while ǫ1.43) The third vector ǫ3 is the longitudinal polarization. and translate the field commutation relations (6. p). aq † ] = 0 and λ λ [ap . 0. 3.. Of the two spacelike polarizations.42) The polarization vectors depend on the photon 4-momentum p = (|p|. with λ = 0. ǫ = 3 0 0 1 (6.43) are Lorentz invariant. . so p ∼ (1.38) for the ˙ classical fields takes the form π µ = −Aµ + . . For example. 1). 2. We pick the normalization ǫλ · ǫλ = η λλ which also means that (ǫµ )λ (ǫν )λ ηλλ′ = ηµν ′ ′ ′ (6. if the momentum lies along the x3 direction.44) For other 4-momenta.40) λ=0 Note that the momentum π µ comes with a factor of (+i). ǫ = 1 1 0 0 . the polarization vectors are the appropriate Lorentz transformations of these vectors. aq ] = [ap † . There are now four polarization 4-vectors ǫλ (p). This can be traced to the fact that the momentum (6. 0.45) – 132 – . we will choose ǫ1 and ǫ2 to lie transverse to the momentum: ǫ1 · p = ǫ2 · p = 0 (6. Of these four 4-vectors. ǫ = 2 0 1 0 .3 are spacelike. We do our usual trick.

The minus signs here are odd to say the least! For spacelike λ = 1. Then we could try implementing the constraint in the quantum theory in a number of different ways. we take the Lorentz invariant vacuum |0 defined by λ ap |0 = 0 (6. 0 has negative norm.48) Then we can create one-particle states in the usual way. so that ∂µ Aµ = 0 (6. . λ = ap † |0 (6. we have 0 0 [ap . aq † ] = −(2π)3 δ (3) (p − q) (6. But this can’t possibly work because the commutation relations (6. 0| q.47) This is very odd! To see just how strange this is. 3. 3. 2. 0 2 At this point we should remember our constraint equation. 0 = 0| ap aq † |0 = −(2π)3 δ (3) (p − q) (6. until now.51) makes sense as an operator equation. We will see that it will remove the timelike. λ = 1. .46) But for the timelike annihilation and creation operators. 3 (6. 2. 2. the state |p. We need some weaker condition. – 133 – . But for the timelike polarization λ = 0. λ λ [ap . 0 0 p. A Hilbert space with negative norm means negative probabilities which makes no sense at all. everything looks fine.39) won’t be obeyed for π 0 = −∂µ Aµ . ∂µ Aµ = 0..49) For spacelike polarization states. We can trace this negative norm back to the ˙ 1 ˙ wrong sign of the kinetic term for A0 in our original Lagrangian: L = + 1 A2 − 2 A2 +. and cut the physical polarizations down to two.50) Wtf? That’s very very strange. we’ve not imposed on our theory. aq † ] = δ λλ (2π)3 δ (3) (p − q) ′ ′ λ. This is going to come to our rescue. negative norm states. λ′ = 1. which. We work in the Heisenberg picture. all seems well. Let’s look at a number of increasingly weak ways to do this • We could ask that ∂µ Aµ = 0 is imposed as an equation on operators. λ |p.

physical states |Ψ . created by – 134 – . We can decompose any element of this basis as |Ψ = |ψT |φ . But this can’t work either! Again.54) is known as the Gupta-Bleuler condition. but ∂µ A− |0 = 0. the condition is too strong. In order to keep the vacuum as a good physical state.53) Then. we can ask that physical states |Ψ are defined by ∂ µ A+ |Ψ = 0 µ This ensures that Ψ′ | ∂µ Aµ |Ψ = 0 (6. but almost! Let’s consider a basis of states for the Fock space. So what does the physical Hilbert space Hphys look like? And. After all. The linearity of the constraint means that the physical states |Ψ span a physical Hilbert space Hphys . For example. where |ψT contains only transverse photons. have we rid ourselves of those nasty negative norm states so that Hphys has a positive definite inner product defined on it? The answer is actually no.52) as our constraint • Our final attempt will be the correct one.52) on all good.• We could try to impose the condition on the Hilbert space instead of directly on the operators. in particular. suppose we decompose Aµ (x) = A+ (x) + A− (x) with µ µ A+ (x) = µ A− (x) = µ d3 p (2π)3 dp (2π)3 3 1 2|p| 1 2|p| 3 λ ǫλ ap e−ip·x µ λ=0 3 λ ǫλ ap † e+ip·x µ λ=0 (6. that’s where the trouble lies! We could imagine that there’s some way to split the Hilbert space up into good states |Ψ and bad states that somehow decouple from the system. But how can we define the good states? One idea is to impose ∂µ Aµ |Ψ = 0 (6. our bad states will include the weird negative norm states that we’re so disgusted by. on the vacuum A+ |0 = 0 automatically. So not even µ µ the vacuum is a physical state if we use (6. With luck. Equation (6.55) (6.54) so that the operator ∂µ Aµ has vanishing matrix elements between physical states.

one can show that the expectation values of all gauge invariant operators evaluated on physical states are independent of the coefficients Cn in (6.0 1. which can be easily computed to be H= d3 p |p| (2π)3 3 i i 0 0 ap† ap − ap † ap (6. The Gupta-Bleuler condition (6.56) ensures that Ψ| ap † ap |Ψ = Ψ| ap † ap |Ψ so that the contributions from the timelike and longitudinal photons cancel amongst themselves in the Hamiltonian.2 ap † . We can think of the gauge symmetry of the classical theory as descending to the Hilbert space of the quantum theory. But we still need to deal with all these zero norm states.56) does indeed decouple the negative norm states. Whenever the state contains a timelike photon of momentum p. The way we cope with the zero norm states is to treat them as gauge equivalent to the vacuum. it must also contain a longitudinal photon with the same momentum. In general |φ will be a linear combination of states |φn containing n pairs of timelike and longitudinal photons. |φn with n ≥ 1 are said to be physically equivalent.59) i=1 3 3 0 0 But the condition (6. all the remaining states involving timelike and longitudinal photons have zero norm φm | φn = δn0 δm0 (6. which we can write as ∞ |φ = n=0 Cn |φn (6.54) requires 3 0 (ap − ap ) |φ = 0 (6.56) This means that the physical states must contain combinations of timelike and longitudinal photons. we can’t just stipulate that two states are physically identical unless they give the same expectation value for all physical observables. while |φ contains the timelike photons created by ap † and longitudinal photons 3 created by ap † . leaving us just with the contribution from the transverse photons as we would expect.57).58) This means that the inner product on Hphys is positive semi-definite. Of course. Which is an improvement. – 135 – . In general.57) where |φ0 = |0 is simply the vacuum. Two states that differ only in their timelike and longitudinal photon content. It’s not hard to show that although the condition (6. This also renders the Hamiltonian positive definite. We can check that this is true for the Hamiltonian.

0| T Aµ (x)Aν (y) |0 = 6.64) (6. it’s Lorentz invariant. 6.Propagators Finally. But we’ve got lots of those! Let’s look at how we can couple two of them to electromagnetism. 1 ¯ / ¯ L = − 4 Fµν F µν + ψ(i ∂ − m)ψ − eψγ µ Aµ ψ (6.1 Coupling to Fermions The Dirac Lagrangian ¯ / L = ψ(i ∂ − m)ψ (6.3. for consistency. we require ∂µ j µ = 0 (6. with the Lagrangian.3 Coupling to Matter Let’s now build an interacting theory of light and matter. For example. It is given by 0| T Aµ (x)Aν (y) |0 = d4 p −iηµν −ip·(x−y) e (2π)4 p2 + iǫ (6.60) This is a lot nicer than the propagator we found in Coulomb gauge: in particular. The equations of motion read ∂µ F µν = j ν so.61) (6. we would find the propagator. Had we pushed through the calculation with arbitrary coefficient α.65) ¯ ¯ has an internal symmetry ψ → e−iα ψ and ψ → e+iα ψ. This gives rise to µ ¯ the conserved current jV = ψγ µ ψ. We could also return to the Lagrangian (6.37). We want to write down a Lagrangian which couples Aµ to some matter fields.66) – 136 – . either scalars or spinors. with α ∈ R. j µ must be a conserved current. it’s a simple matter to compute the propagator in Lorentz gauge.63) In other words. So we could look at the theory of electromagnetism coupled to fermions. we could write something like 1 L = − 4 Fµν F µν − j µ Aµ d4 p −i 4 p2 + iǫ (2π) ηµν + (α − 1) pµ pν p2 e−ip·(x−y) (6.62) where j µ is some function of the matter fields.

Note that the particle and the anti-particle are required by the formalism to have opposite electric – 137 – .68) for an arbitrary function λ(x). The tricky term is the derivative acting on ψ. For the free Maxwell theory. let’s look at how the covariant derivative transforms. Does our interacting theory above still have a gauge symmetry? The answer is yes. To see that all is well. To see this. we have Q=e d3 p (2π)3 2 s s s s (bp † bp − cp † cp ) (6. since ψ → +ieλ(x) ¯ e ψ.70) After treating this as a quantum equation.67) where Dµ ψ = ∂µ ψ + ieAµ ψ is called the covariant derivative. we have seen that the existence of a gauge symmetry was crucial in order to cut down the physical degrees of freedom to the requisite 2. let’s rewrite the Lagrangian as 1 ¯ / L = − 4 Fµν F µν + ψ(i D − m)ψ (6. Electric Charge The coupling e has the interpretation of the electric charge of the ψ particle.where we’ve introduced a coupling constant e. This ensures that the whole Lagrangian is invariant. since this will also hit the e−ieλ piece after the transformation.69) so the covariant derivative has the nice property that it merely picks up a phase under the gauge transformation. We therefore have the total charge Q given by Q=e ¯ d3 x ψ(x)γ 0 ψ(x) (6. with the derivative of e−ieλ cancelling the transformation ¯ of the gauge field.71) s=1 which is the number of particles. This Lagrangian is invariant under gauge transformations which act as Aµ → Aµ + ∂µ λ and ψ → e−ieλ ψ (6. This follows from the equations of motion of classical electromagnetism ∂µ F µν = j ν : we know that the j 0 component is the charge density. We have Dµ ψ = ∂µ ψ + ieAµ ψ → ∂µ (e−ieλ ψ) + ie(Aµ + ∂µ λ)(e−ieλ ψ) = e−ieλ Dµ ψ (6. minus the number of antiparticles.

Finally. The latter is a redundancy in our description of the system. the gauge symmetry ψ → e+ieλ(x) ψ seems to lead to a conservation law. let’s check that the 4 × 4 matrix C that we introduced in Section 4. These take us from one physical state to another. This means that we can’t couple a real scalar field to a gauge field.5 really deserves the name “charge conjugation matrix”.3. 6.charge. the subset of global symmetries from among the gauge symmetries are those for which λ(x) → α = constant as x → ∞. For QED. we have no suitable conserved current.72) 4πα ≈ 0. The former takes you from one physical state to another with the same properties and results in a conserved current through Noether’s theorem. Yet in electromagnetism. known as the fine structure constant α= Setting = c = 1. we have (iγ µ ∂µ − eγ µ Aµ − m)ψ = 0 ⇒ (−i(γ µ )⋆ ∂µ − e(γ µ )⋆ Aµ − m)ψ ⋆ = 0 Now using the defining equation C † γ µ C = −(γ µ )⋆ . namely the conservation of electric charge. If we take the complex conjugation of the Dirac equation. More generally. This is a true symmetry of the system. meaning that it takes us to another physical state. we have e = √ 1 e2 ≈ 4π c 137 (6. we see that the charge conjugate spinor ψ (c) satisfies (iγ µ ∂µ + eγ µ Aµ − m)ψ (c) = 0 (6. but with charge −e instead of +e.2 Coupling to Scalars For a real scalar field. and the definition ψ (c) = Cψ ⋆ . there is also a single global symmetry: that with λ(x) = constant. – 138 – .3.73) So we see that the charge conjugate spinor ψ (c) satisfies the Dirac equation. the electric charge is usually written in terms of the dimensionless ratio α. There’s a small subtlety here that’s worth elaborating on. I stressed that there’s a radical difference between the interpretation of a global symmetry and a gauge symmetry. the theory of light interacting with electrons. This is because among the infinite number of gauge symmetries parameterized by a function λ(x).

it will now pick up a contribution from the j µ Aµ term. We could try to couple the associated current to the gauge field. This means that we can construct a gauge invariant action for a charged scalar field coupled to a photon simply by promoting all derivatives to covariant derivatives 1 (6. So the whole procedure wasn’t consistent. Lint = −i((∂µ ϕ⋆ )ϕ − ϕ⋆ ∂µ ϕ)Aµ But this doesn’t work because • The theory is no longer gauge invariant • The current j µ that we coupled to Aµ depends on ∂µ ϕ. this trick works for any theory. then we still have a bit of work in front of us in order to massage the photon propagator into something Lorentz invariant. This means that if we try to compute the current associated to the symmetry. (6.Let’s now consider a complex scalar field ϕ. If we worked in Lorentz gauge previously. We solve both of these problems simultaneously by remembering the covariant derivative. In this scalar theory. The Lagrangian is 1 ¯ / L = − Fµν F µν + ψ(i D − m)ψ 4 where Dµ = ∂µ + ieAµ . however.76) L = − Fµν F µν + Dµ ϕ⋆ D µ ϕ − m2 |ϕ|2 4 In general.74) again transforms as Dµ ϕ → e−ieλ Dµ ϕ under a gauge transformation Aµ → Aµ + ∂µ λ and ϕ → e−ieλ ϕ. I’ll depart from our previous notation and call the scalar field ϕ to avoid confusing it with the spinor). we may do so by replacing all derivatives by suitable covariant derivatives.75) (6. then we can jump straight to Section 6. This procedure is known as minimal coupling. If. We will now do that. If we have a U(1) symmetry that we wish to couple to a gauge field.5 where the Feynman rules for QED are written down. We have a symmetry ϕ → e−iα ϕ. 6. we worked in Coulomb gauge.4 QED Let’s now work out the Feynman rules for the full theory of quantum electrodynamics (QED) – the theory of electrons interacting with light. (For this section. the combination Dµ ϕ = ∂µ ϕ + ieAµ ϕ (6.77) – 139 – . The route we take now depends on the gauge choice.

we have LMaxwell = d3 x 1 ˙ 2 A 2 − 1B 2 − 2 e2 2 d3 x′ j0 (x)j0 (x ′ ) 4π|x − x ′ | (6. Changing notation slightly from previous chapters. we have the conjugate momenta.82) e2 ¯ + 1 B 2 + ψ(−iγ i ∂i + m)ψ − ej · A + 2 2 d3 x′ j 0 (x)j 0 (x ′ ) 4π|x − x ′ | ¯ ¯ where j = ψγψ and j 0 = ψγ 0 ψ. t) 4π|x − x ′ | (6. the equation of motion arising from varying A0 is now −∇2 A0 = eψ † ψ ≡ ej 0 which has the solution A0 (x. This is exactly the type of interaction that we boasted in Section 1. ∂L ˙ =A ˙ ∂A ∂L πψ = = iψ † ˙ ∂ψ Π= which gives us the Hamiltonian H= d3 x 1 ˙ 2 A 2 (6. it wouldn’t appear if we worked in Lorentz gauge.4 never arises in Nature! It appears here as an artifact of working in Coulomb gauge: it does not mean that the theory of QED is nonlocal.80) where the cross-term has vanished using ∇ · A = 0.78) In Coulomb gauge we can rewrite the Maxwell part of the Lagrangian as LMaxwell = = = 1 d3 x 2 E 2 − 1 B 2 2 1 ˙ d3 x 2 (A + ∇A0 )2 − 1 B 2 2 1 1 ˙ d3 x 2 A 2 + 2 (∇A0 )2 − 1 B 2 2 (6.81) We find ourselves with a nonlocal term in the action. t) = e d3 x′ j 0 (x ′ .1.79) (6. For example. After integrating the second term by parts and inserting the equation for A0 . We now compute the Hamiltonian.In Coulomb gauge ∇ · A = 0. – 140 – .

33): it is and contributes Dij = 2 p + iǫ |p| • The vertex contributes −ieγ i . Let’s start with the offending instantaneous interaction. the wavy photon line now carries a µ. in position space. Each end of the line comes with an i. ν = j = 0  p2 + iǫ δij − |p|2     0 otherwise – 141 – . the rules are the same as those given in Section 5. We now need to change our vertex slightly: the −ieγ i above gets replaced by −ieγ µ which correctly accounts for the (eγ 0 )2 piece in the instantaneous interaction. ν = 0   |p|  pi pj i Dµν (p) = (6.1 Naive Feynman Rules We want to determine the Feynman rules for this theory. we could try to redefine the propagator to include a D00 piece which will capture this term.83) With this propagator. j = 1. The index on γ i contracts with the index on the photon line. 3 index. ν = 0. Since it comes from the A0 component of the gauge field. 3 index telling us the component of A. it fits quite nicely in this form: if we look in momentum space. In fact. 2. This is the price we’ve paid for working in Coulomb gauge. 2. is given by i(eγ 0 )2 δ(x0 − y 0 ) contributes a factor of 4π|x − y| x y These Feynman rules are rather messy. We calculated the transverse photon pi pj i tr δij − 2 propagator in (6.6. The new pieces are: • We denote the photon by a wavy line. we have δ(x0 − y 0 ) = 4π|x − y| d4 p eip·(x−y) (2π)4 |p|2 (6. • The non-local interaction which.4. so we can combine the non-local interaction with the transverse photon propagator by defining a new photon propagator  i  + 2  µ.84) µ = i = 0. We’ll now show that we can massage these expressions into something much more simple and Lorentz invariant. 1. with the extra µ = 0 component taking care of the instantaneous interaction. For fermions.

Lorentz invariant propagator Dµν (p) = −i ηµν p2 (6.88) 2 α · β k0 α0 β0 α0 β0 − + k2 k 2 |k|2 |k|2 1 α·β 2 (k0 − k 2 ) α0 β0 − 2 k k 2 |k|2 iηµν i = − 2 α · β = αµ − 2 β ν k k (6. the diagram can be written as αµ Dµν β ν = i =i =− α · β (α · k)(β · k) α0 β 0 + − k2 k 2 |k|2 |k|2 (6. Here we’ll be more pedestrian and show that we can do this for certain Feynman diagrams. they are. similarly. we focus on a particular tree-level diagram that contributes to e− e− → e− e− scattering. Using this fact.89) – 142 – . In particular.86) where k = p − p′ = q ′ − q.The D00 piece of the propagator doesn’t look a whole lot different from the transverse photon propagator. But wouldn’t it be nice if they were both part of something more symmetric! In fact. We have the following: Claim: We can replace the propagator Dµν (p) with the simpler.87) Let’s define the spinor contractions αµ = u(p ′ )γ µ u(p) and β ν = u(q ′ )γ ν u(q).85) Proof: There is a general proof using current conservation. kν β ν = 0. p p/ ∼ e2 [¯(p′ )γ µ u(p)] Dµν (k) [¯(q ′ )γ ν u(q)] u u q q / (6. Then ¯ ¯ ′ ′ since k = p − p = q − q. we have / /′ ¯ kµ αµ = u(p ′ )( p − p )u(p) = u(p ′ )(m − m)u(p) = 0 ¯ and. Recall that u(p) satisfies the equation / ( p − m)u(p) = 0 (6.

it’s a general fact that in every Feynman diagram we may use the very nice. ǫ0 = 0 and ǫ · p = 0. in the following diagrams the pµ pν piece of the propagator contributes as / / ∼ u(p′ )γ µ u(p) kµ = u(p′ )( p − p ′ )u(p) = 0 ¯ ¯ / / ∼ v (p)γ µ u(q) kµ = u(p)( p + q ′ )u(q) = 0 ¯ ¯ 6. We add u r r a spinor v (p)/v (p) for incoming/outgoing anti-fermions.which is the claimed result. Note: This is the propagator we found when quantizing in Lorentz gauge (using the Feynman gauge parameter).92) p2 p2 For external lines in the diagram. we write • Vertex: • Photon Propagator: • Fermion Propagator: −ieγ µ − iηµν + iǫ (6. You can similarly check that the same substitution is legal in the diagram p p/ q/ q ∼ e2 [¯(q)γ µ u(p)]Dµν (k)[¯(p ′ )γ ν v(q ′ )] v u (6.5 Feynman Rules Finally. quantizing the Lagrangian (6. ¯ – 143 – . For example.37) in Lorentz gauge. In general. although we won’t show it here.90) In fact. For vertices and internal lines. you can show that the pµ pν /p2 term cancels in all diagrams.91) p p Using similar arguments to those given above. we have the Feynman rules for QED. in out In Coulomb gauge. we attach / i( p + m) − m2 + iǫ • Photons: We add a polarization vector ǫµ /ǫµ for incoming/outgoing photons. Lorentz invariant propagator Dµν = −iηµν /p2 . we have the propagator i pµ pν Dµν = − 2 ηµν + (α − 1) 2 (6. • Fermions: We add a spinor ur (p)/¯r (p) for incoming/outgoing fermions.

There is also a second.6 Scattering in QED Let’s now calculate some amplitudes for various processes in quantum electrodynamics. I also inquired how long it would take to perform this task. (It’s the same reason that the 1/4! didn’t appear in the Feynman rules for φ4 theory). I met Bethe in Copenhagen at a conference and asked him to tell me how he did the calculations. How do these appear in the Feynman rules? After a Fourier transform. so we include a momentum factor in the Feynman rule.1 Charged Scalars “Pauli asked me to calculate the cross section for pair creation of scalar particles by photons. The two Feynman rules are p q − ie(p + q)µ and + 2ie2 ηµν The factor of two in the seagull diagram arises because of the two identical particles appearing in the vertex.5. but you will need about three weeks. But the cubic vertex is something we haven’t seen before: it contains kinetic terms. “seagull” graph. It was only a short time after Bethe and Heitler had solved the same problem for electrons and positrons. 6. with a photon coupled to a single fermion. furthermore. he answered. the derivative term means that the interaction is stronger for fermions with higher momentum.93) This gives rise to two interaction vertices. as usual.” Viki Weisskopf The interaction terms in the Lagrangian for charged scalars come from the covariant derivative terms. We will consider the analogous set of processes that we saw in Section 3 and Section 5. L = Dµ ψ † D µ ψ = ∂µ ψ † ∂ µ ψ − ieAµ (ψ † ∂ µ ψ − ψ∂ µ ψ † ) + e2 Aµ Aµ ψ † ψ (6.6. the published cross sections were wrong by a factor of four.” He was right. “It would take me three days. We have – 144 – .

Historically.s p.r Compton Scattering The scattering of photons (in particular x-rays) off electrons e− γ → e− γ is known as Compton scattering.r / + / q.r − The overall −i comes from the −iηµν in the propagator. which contract the indices on the γ-matrices (remember that it’s really positive for µ. 3).s εν 1 + q/ q q.s / + q.s / q.r / / / = −i(−ie) 2 [¯s (p ′ )γ µ us (p)] [¯r (q)γµ v r (q ′ )] u v − (p − p′ )2 [¯r(q)γ µ us (p)] [¯s (p ′ )γµ v r (q ′ )] v u + (p + q)2 ′ ′ ′ ′ q.s / p.s / p.s / / p.r / / / γµ ( p − p ′ + m)γν = i(−ie) v (q) ¯ (p − p′ )2 − m2 2 r ε2 µ q. given by p.r / p.r q. Electron Positron Annihilation Let’s now look at e− e+ → 2γ.Electron Scattering Electron scattering e− e− → e− e− is described by the two leading order Feynman diagrams. εν 1 p.r / q. the change in wavelength of the photon in the – 145 – . The two lowest order Feynman diagrams are.s p/ p/ p. ν = 1.r µ ε2 + / / γν ( p − q ′ + m)γµ (p − q ′ )2 − m2 us (p)ǫν (p ′ )ǫµ (q ′ ) 1 2 Electron Positron Scattering For e− e+ → e− e+ scattering (sometimes known as Bhabha scattering) the two lowest order Feynman diagrams are p. 2.r / = −i(−ie) q.s / p.s / 2 [¯s (p ′ )γ µ us (p)] [¯r (q ′ )γµ ur (q)] u u ′ − p)2 (p [¯s (p ′ )γ µ ur (q)] [¯r (q ′ )γµ us (p)] u u (p − q ′ )2 ′ ′ ′ ′ q. two gamma rays.

Adding Muons Figure 30: Adding a second fermion into the mix. Then. suppose ǫin ∼ q. which we could identify as a muon. ε in (q ) εout (q /) ε in (q) εout (q /) + u(p) − / u(p) u(p) − / u(p) = i(−ie) u (p ) ¯ 2 r′ ′ / / / / γ µ ( p + q + m)γ ν γ ν ( p − q ′ + m)γ µ + (p + q)2 − m2 (p − q ′ )2 − m2 us (p) ǫν ǫµ in out This amplitude vanishes for longitudinal photons. For example. Using our standard notation of p and q for incoming momenta. At one-loop. there is a diagram which leads to photon scattering. so q q = 0. we may write the amplitude as iA = i(−ie) u (p ) ¯ ′ 2 r′ ′ / / / / / /′ / / ǫout ( p + q + m) q q ( p − q + m) ǫout + 2 − m2 ′ − q)2 − m2 (p + q) (p 2p′ · q 2p · q + ′ (p + q)2 − m2 (p − q)2 − m2 us (p) (6. The amplitude is given by. the diagram is actually rendered finite by gauge invariance. and e+ e− annihilation into a muon anti-muon pair. and p′ and q ′ for outgoing – 146 – . we can now have processes such as e− µ− → e− µ− scattering. using momentum conservation p + q = p′ + q ′ . We now recall the fact that q is a null vector. For example.94) / = i(−ie)2 ur (p ′ ) ǫout us (p) ¯ where. together with the spinor equations ′ /′ / ( p − m)u(p) and u(p )( p − m) = 0. photons no longer pass through each other unimpeded. we’ve performed some γ-matrix manipulations. This ensures that the combined amplitude vanishes for longitudinal photons as promised. in going to the second line. This means that the two denominators in the amplitude read (p+q)2 −m2 = 2p·q and (p′ −q)−m2 = −2p′ ·q.scattering process was one of the conclusive pieces of evidence that light could behave as a particle. // and invoked the fact that q is null. Photon Scattering In QED. new processes become possible. while ¯ 2 ′ 2 2 p = (p ) = m since the external legs are on mass-shell. Although naively logarithmically divergent. A / similar result holds when ǫout ∼ q ′ .

We’ve understood how to compute quantum amplitudes in a large array of field theories.r .s / = −i(−ie)2 / q.1 The Coulomb Potential We’ve come a long way.momenta. the amplitude is p. We have p. We start by looking at e− e− → e− e− scattering.s / p.r / [¯(p ′ )γ µ u(p)] [¯(q ′ )γµ v(q)] u v = +i(−ie) ′ − p)2 (p 2 (6.s / p.2. ¯ Comparing the scattering amplitude in this limit to that of non-relativistic quantum mechanics. i = 1. the non-relativistic limit of the spinor is u(p) → √ m ξ We find the familiar repulsive Coulomb potential. For e− e+ → e− e+ scattering.97) / q. This ξ means that the γ 0 piece of the interaction gives a term us (p)γ 0 ur (q) → 2mδ rs .5. we have the amplitudes given by e− e− µ− 1 ∼ (p − p′ )2 µ− µ− e− and e+ µ+ ∼ 1 (p + q)2 (6. we use our newfound knowledge to rederive a result you learnt in kindergarten: Coulomb’s law.r – 147 – . 2.6.49).96) q. we repeat our calculation that led us to the Yukawa force in Sections 3. To do this. we have the effective potential between two electrons given by. To end this course.s / e2 d3 p eip·r =+ (2π)3 |p|2 4πr (6. 3 pieces vanish in the non-relativistic limit: us (p)γ i ur (q) → 0. U(r) = +e 2 Following (5. while ¯ i the spatial γ .2 and 5.98) q. We can trace the minus sign that gives a repulsive potential to the fact that only the A0 component of the intermediate propagator ∼ −iηµν contributes in the non-relativistic limit.7.r / [¯(p ′ )γ µ u(p)] [¯(q ′ )γµ u(q)] u u ′ − p)2 (p (6.95) 6.

5. This is simpler to see in the case of scalar QED.1 are correlated with the momentum arrows. Flipping the arrows on one pair of legs in the amplitude introduces the relevant minus sign to ensure that the non-relativistic potential between e− e+ is attractive as expected. Note that with our signature (+ − −−). where we don’t have to worry about the gamma matrices. while A0 comes with the wrong sign. the amplitude picks up an extra minus sign because the arrows on the legs of the Feynman rules in Section 6. we have the non-relativistic limit of scalar e− e− scattering. For e− e+ scattering. p. ¯ The Coulomb Potential for Scalars There are many minus signs in the above calculation which somewhat obscure the crucial one which gives rise to the repulsive force. just as it is for fermions. this time in the guise of the γ 0 sandwiched between v γ 0 v → 2m. We have v γ 0 v → 2m. the propagating Ai have the correct sign. ¯ rather than the v v → −2m that we saw in the Yukawa case. we find an attractive force between an electron and positron. The difference now comes from looking at the non-relativistic limit.99) Reassuringly. A careful study reveals the offending sign to be that which sits in front of the A0 piece of the photon propagator −iηµν /p2 . The difference from the calculation of the Yukawa force comes again from the zeroth component of the gauge field. U(r) = −e2 e2 d3 p eip·r =− (2π)3 |p|2 4πr (6. From the Feynman rules of Section 6.r where the non-relativistic limit in the numerator involves (p+p′ )·(q +q ′ ) ≈ (p+p′ )0 (q + q ′ )0 ≈ (2m)2 and is responsible for selecting the A0 part of the photon propagator rather than the Ai piece.1.5.The overall + sign comes from treating the fermions correctly: we saw the same minus sign when studying scattering in Yukawa theory. giving us the potential between ¯ opposite charges.s / p. This shows that the Coulomb potential for spin 0 particles of the same charge is again repulsive. – 148 – .s / = −iηµν (−ie)2 / q.r / (p + p′ )µ (q + q ′ )ν (2m)2 → −i(−ie)2 (p′ − p)2 −(p − p ′ )2 q.

see the excellent book “QED and the Men who Made it” by Sam Schweber.7 Afterword In this course. It took a new generation of postwar physicists — people like Schwinger. and failing. The difficulty lies in the next terms in perturbation theory. 5 – 149 – . physicists were ready to give up on quantum field theory. These are the terms that correspond to Feynamn diagrams with loops in them. For more details on the history of quantum field theory. the general feeling was that one should do something else. most physicists will be glad to see the end of QED But the leading figures of the day gave up too soon. Heisenberg. The story of how they did that will be told in next term’s course. This from Dirac in 1937. Dirac. Tomonaga and Dyson — to return to quantum field theory and tame the infinities. The reason we’ve avoided them is because they typically give infinity! And. after ten years of trying. which we have scrupulously avoided computing in this course. pioneered by people such as Jordan. Yet by the end of the 1930s.6. to make sense of this. Most of the developments that we’ve seen were already in place by the middle of the 1930s. we have laid the foundational framework for quantum field theory. Because of its extreme complexity. Pauli and Weisskopf 5 . Feynman.

Sign up to vote on this title
UsefulNot useful