This action might not be possible to undo. Are you sure you want to continue?

# 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

⋆

∇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

(Λ

−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

(δ

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

(ω

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

iα

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

[φ

a

(x), φ

b

(y)] = [π

a

(x), π

b

(y)] = 0

[φ

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

√

2ω

p , a

†

=

ω

2

q −

i

√

2ω

p (2.9)

which can be easily inverted to give

q =

1

√

2ω

(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

2ω

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 =

nπ

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

nπ

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

nπ

d

2

+ p

2

y

+ p

2

z

e

−a

q

(

nπ

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

iα

ψ. The current has time and space components

j

µ

=

ψ

⋆

ψ,

i

2m

(ψ

⋆

∇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

xψ

†

(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

xψ

†

(x)ψ(x)

d

3

k

(2π)

3

2E

p

2E

k

a

k

a

†

p

e

−ik·x

[0`

= −ig 'f[

d

4

xψ

†

(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

)ψ

†

(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

)ψ

†

(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

)ψ

†

(x

2

)ψ(x

2

) : [p

1

, p

2

`

= 'p

′

1

, p

′

2

[ ψ

†

(x

1

)ψ

†

(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)

+δ

(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

)ψ

†

(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

+∞

−∞

dω

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

)φ

1I

U(t

1

, t

2

) φ

2I

. . . φ

nI

U

I

(t

n

, −∞) [0`

= '0[ U

I

(+∞, t

1

)φ

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

(Λ

−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

(Λ

−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

Ω

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

Ω

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

(γ

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

[γ

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

[γ

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

]γ

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

(Λ

−1

x) (4.22)

where

Λ = exp

1

2

Ω

ρσ

´

ρσ

(4.23)

S[Λ] = exp

1

2

Ω

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

Ω

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

Ω

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

Ω

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

(γ

0

)

2

= 1 ⇒ Real Eigenvalues

(γ

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

†

(Λ

−1

x) S[Λ]

†

(4.36)

So ψ

†

(x)ψ(x) → ψ

†

(Λ

−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

Ω

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

†

(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) (4.41)

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

Ω

ρσ

´

ρσ

≈ 1 +

1

2

Ω

ρσ

´

ρσ

+ . . . (4.45)

S[Λ] = exp

1

2

Ω

ρσ

S

ρσ

≈ 1 +

1

2

Ω

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

)γ

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

(γ

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

=

iσ

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

[γ

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

xψ

†

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

Ω

ρσ

(S

ρσ

)

α

β

ψ

β

(4.93)

where, following (4.10), we have ω

µ

ν

=

1

2

Ω

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

(γ

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

′

mξ

′

(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

(γ

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

[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) (5.3)

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

(γ

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

xψ

†

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

′

)γ

5

u

s

( p)] [¯ u

r

′

(q

′

)γ

5

u

r

(q)]

(p −p

′

)

2

−µ

2

−

[¯ u

s

′

( p

′

)γ

5

u

r

(q)] [¯ u

r

′

(q

′

)γ

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

′

)γ

5

u

s

( p) →

mξ

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

[ξ

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

iα

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,

[η

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

2α

(∂

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

(ǫ

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

(ǫ

µ

)

λ

(ǫ

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

[φ

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

′

)γ

µ

u(p)] D

µν

(k) [¯ u(q

′

)γ

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

′

)γ

µ

u( p) and β

ν

= ¯ u(q

′

)γ

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

−

iη

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

′

)γ

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

′

)γ

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: −

iη

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

/ǫ

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

(ψ

†

∂

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

′

)γ

µ

u

s

( p)] [¯ u

r

′

(q

′

)γ

µ

u

r

(q)]

(p

′

−p)

2

−

[¯ u

s

′

( p

′

)γ

µ

u

r

(q)] [¯ u

r

′

(q

′

)γ

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

′

)ǫ

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

′

)γ

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

′

)γ

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

′

)γ

µ

u( p)] [¯ u(q

′

)γ

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

′

)γ

µ

u( p)] [¯ v(q

′

)γ

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–

1 The Spinor Representation 4.4 Feynman Diagrams 3.4.7 Plane Wave Solutions 4.2 Wick’s Theorem 3.4.1 An Example: Recovering the Propagator 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– .3.6 What We Measure: Cross Sections and Decay Rates 3.2 Decay Rates 3.1.2 γ 5 4.7.1 An Example: Meson Decay 3.7 Green’s Functions 3.6.4 Connected Diagrams and Amputated Diagrams 3.3.1 Connected Diagrams and Vacuum Bubbles 3.5.3.3.5.3 The Dirac Equation 4.3 φ4 Theory 3.4.5 Majorana Fermions 4. Interacting Fields 3.4 Chiral Spinors 4.6.4.5.4.6.1 Mandelstam Variables 3.2.1 The Interaction Picture 3.1 Fermi’s Golden Rule 3.1 Feynman Rules 3. The Dirac Equation 4.2 A First Look at Scattering 3.3 Cross Sections 3.7.5 Examples of Scattering Amplitudes 3.1 Spinors 4.7.6 Symmetries and Conserved Currents 4.2 Constructing an Action 4.1 Dyson’s Formula 3.3 An Example: Nucleon Scattering 3.4 Chiral Interactions 4.5.3 Wick’s Theorem 3.2 The Yukawa Potential 3.1.3 Parity 4.2 From Green’s Functions to S-Matrices 4.1 The Weyl Equation 4.

7 Feynman Rules for Fermions 5.1 A Glimpse at the Spin-Statistics Theorem 5.2 The Quantization of the Electromagnetic Field 6.1 An Example: Putting Spin on Nucleon Scattering 5.1.1 Gauge Symmetry 6.1 The Hamiltonian 5.3.7.7 Afterword –3– .2 Lorentz Gauge 6.1 Examples 5.5 Feynman Rules 6.1 Charged Scalars 6.6.1.1 Naive Feynman Rules 6.2 Fermionic Quantization 5.1 The Coulomb Potential 6.3 Pseudo-Scalar Coupling 6.6.2 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 5.2 The Yukawa Potential Revisited 5.2.2 Coupling to Scalars 6.7.7.1 Coupling to Fermions 6.2.1 Fermi-Dirac Statistics 5.3 Coupling to Matter 6.4.6 Yukawa Theory 5.4 Propagators 5.4. Quantizing the Dirac Field 5.5.4 QED 6.1 Coulomb Gauge 6.3.1 Maxwell’s Equations 6.3 Dirac’s Hole Interpretation 5.2.5 The Feynman Propagator 5.6 Scattering in QED 6.7. Quantum Electrodynamics 6.

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

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

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

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

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

roughly 1060 lp . For example. more often GeV = 109 eV or T eV = 1012 eV . together with other relevant energy scales. we need to insert the relevant powers of c and . meaning the mass dimension. the electron volt 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. 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. To convert the unit of energy back to a unit of length or time. In particular. 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. The usual choice of energy unit is eV . If X has dimensions of (mass)d we will write [X] = d. with energy increasing to the right and.2) where Mp ≈ 1019 GeV is the Planck scale. G= 1 c = 2 2 Mp Mp (0. the electron mass me = 106 eV translates to a length scale λe = 2 × 10−12 m. This is a logarithmic plot. The smallest and largest scales known are shown on the ﬁgure. length increasing to the left. 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. correspondingly. The largest length scale we can talk of is the size of the cosmological horizon. equivalently. It corresponds to a length lp ≈ 10−33 cm. the length scale λ associated to a mass m is the Compton wavelength λ= mc With this conversion factor. energy (since E = mc2 has become E = m).

Neutron Tau W. the rough masses of some elementary (and not so elementary) particles are shown in the table. For comparison. This is precisely the regime that is currently being probed by the Large Hadron Collider (LHC) at CERN. and the weak and strong forces unify at around 1018 GeV.Z Bosons Higgs Boson Mass ∼ 10−2 eV 0. Particle neutrinos electron Muon Pions Proton. there are experimental hints that the coupling constants of electromagnetism.to about the T eV .5 MeV 100 MeV 140 MeV 1 GeV 2 GeV 80-90 GeV 120-200 GeV?? –6– . 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.

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

7) to the usual expression for the Lagrangian L = T − V . etc. with an eye to later Lorentz invariance.7) where we are using the Minkowski space metric +1 η µν = ηµν = −1 −1 −1 (1. t). However. Requiring δS = 0 for all such paths yields the Euler-Lagrange equations of motion for the ﬁelds φa .9) –8– .where the oﬃcial name for L is the Lagrangian density. ∇3 φ.1.1 An Example: The Klein-Gordon Equation Consider the Lagrangian for a real scalar ﬁeld φ(x. ∇2 φ. keeping the end points ﬁxed and require δS = 0.5) The last term is a total derivative and vanishes for any δφa (x. We vary the path. t) that decays at spatial inﬁnity and obeys δφa (x. The action is. In principle. but not q . ¨ we similarly restrict to Lagrangians L depending on φ and φ. t1 ) = δφa (x. t2 ) = 0. although everyone simply calls it the Lagrangian. ∂µ ∂L ∂(∂µ φa ) − ∂L =0 ∂φa (1. t2 S= t1 dt d3 x L = d4 x L (1. 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. there’s nothing to stop L depending on ∇φ. we identify the kinetic energy of the ﬁeld as T = 1 ˙ d 3 x 2 φ2 (1. We can determine the equations of motion by the principle of least action. In ﬁeld theory ˙ ¨ ˙ and not φ.6) 1. 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 will only consider Lagrangians depending on ∇φ and not higher derivatives.4) Recall that in particle mechanics L depends on q and q.8) Comparing (1.

or just “potential”.17) ∂t This looks very much like the Schr¨dinger equation. rather than quadratic.12) –9– .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. In this notation. while the phrase “potential energy”.14) 2 ∂φ 1. An obvious generalization of the Klein-Gordon equation comes from considering the Lagrangian with arbitrary potential V (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 (1.15) 2 We can determine the equations of motion by treating ψ and ψ ⋆ as independent objects.11) (1. we compute ∂L ∂L ˙ = −m2 φ and = ∂ µ φ ≡ (φ. We’ll come back to this point in Section 2. Except it isn’t! Or.7). The Laplacian in Minkowski space is sometimes denoted by . 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. To determine the equations of motion arising from (1.16) = − ψ and ˙ ∂ψ ⋆ 2 2 ∂∇ψ ⋆ ∂ψ⋆ This gives us the equation of motion ∂ψ i = −∇2 ψ + mψ (1. the Klein-Gordon equation reads φ + m2 φ = 0.1. ∂V L = 1 ∂µ φ∂ µ φ − V (φ) ⇒ ∂µ ∂ µ φ + =0 (1.and the potential energy of the ﬁeld as V = 1 d3 x 1 (∇φ)2 + 2 m2 φ2 2 (1. so that i ˙ ∂L ∂L ∂L i = ψ − mψ and = −∇ψ (1.13) This is the Klein-Gordon Equation.8. (1. at least. −∇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 (1.10) The ﬁrst term in this expression is called the gradient energy.

18). We will see the consequences of this in Section 6. the Lagrangian is local.1. 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 . You can check using (1. ˙ however when L ∼ ψ ⋆ ψ. there’s no reason for this. Indeed.21) A priori. This means that there are no terms in the Lagrangian coupling φ(x.1. so L ∼ 2 A2 . – 10 – . 1 L = − 1 (∂µ Aν ) (∂ µ Aν ) + 2 (∂µ Aµ )2 2 (1.19) where the ﬁeld strength is deﬁned by Fµν = ∂µ Aν − ∂ν Aµ .8). one of the main reasons for introducing ﬁeld theories in classical physics is to implement locality. t) directly to φ(y. we may rewrite the Maxwell Lagrangian (up to an integration by parts) in the compact form L = − 1 Fµν F µν 4 1.18) has no i ˙0 kinetic term A2 for A0 . both φ and φ must be speciﬁed to determine the future evolution.4 Locality. ˙ ˙ When L ∼ φ2 . we compute ∂L = −∂ µ Aν + (∂ρ Aρ ) η µν ∂(∂µ Aν ) from which we may derive the equations of motion. there are no terms that look like L= d3 xd3 y φ(x)φ(y) (1. the term ∂3 A0 ∂0 A3 in the Maxwell Lagrangian couples the µ = 0 ﬁeld to the µ = 3 ﬁeld).2) that this reproduces the remaining two Maxwell’s equations in a vacuum: ∇ · E = 0 and ∂ E/∂t = ∇ × B. The Lagrangian (1. ∂µ ∂L ∂(∂µ Aν ) = −∂ 2 Aν + ∂ ν (∂ρ Aρ ) = −∂µ (∂ µ Aν − ∂ ν Aµ ) ≡ −∂µ F µν (1. 1. as far as we know. This property of locality is. In this course. For example. only ψ and ψ ⋆ are needed. To see that Maxwell’s equations indeed follow from (1.22) (1. Locality.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. Using the notation of the ﬁeld strength. a key feature of all theories of Nature. Locality In each of the examples above. After all. t) with x = y. x is merely a label.20) (1.The initial data required on a Cauchy surface diﬀers for the two examples above. and we’re quite happy to couple other labels together (for example. we will only consider local Lagrangians.3 A Final Example: Maxwell’s Equations We may derive Maxwell’s equations in the vacuum from the Lagrangian.

we need to place ourselves at x = (0. 0. To see why this means that the inverse appears. If we want to express φ′ (x) in terms of the old ﬁeld φ. 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 new ﬁeld φ′ (x) will have the hotspot at x = (0. You’ll learn more about this in the “Symmetries and Particle Physics” course. 0). 0 0 0 1 (1. it will suﬃce to consider a non-relativistic example such as a temperature ﬁeld.23) 0 √ with γ = 1/ 1 − v 2 . 1. under the Lorentz transformation x → Λx. This R−1 is the origin of the inverse transformation. say. – 11 – . Suppose we start with an initial ﬁeld φ(x) which has a hotspot at. and one of the main motivations to develop quantum ﬁeld theory is to reconcile quantum mechanics with special relativity.2 Lorentz Invariance The laws of Nature are relativistic. To this end.24) along the x1 -axis (1. 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. x = (1. After a rotation x → Rx about the z-axis. 0). we would have instead φ(x) → φ′ (x) = φ(Λx)). 1. The simplest example is the scalar ﬁeld which. The Lorentz transformations form a Lie group under matrix multiplication. 0.26) The inverse Λ−1 appears in the argument because we are dealing with an active transformation in which the ﬁeld is truly shifted. 1. 0) and ask what the old ﬁeld looked like where we’ve come from at R−1 (0. (If we were instead dealing with a passive transformation in which we relabel our choice of coordinates.1. transforms as φ(x) → φ′ (x) = φ(Λ−1 x) (1. 0).25) The Lorentz transformations have a representation on the ﬁelds. a rotation by θ about the x3 -axis. xµ −→ (x′ )µ = Λµν xν where Λµν satisﬁes Λµσ η στ Λντ = η µν For example. 0) = (1.

space and time are not on the same footing. The theory is not Lorentz invariant.28) where. We can ensure that this property holds by requiring that the action is Lorentz invariant.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. The derivative of the scalar ﬁeld transforms as a vector. but quadratic in spatial derivatives). 1. S= d4 x L(x) −→ d4 x L(y) = d4 y L(y) = S (1. meaning (∂µ φ)(x) → (Λ−1 )νµ (∂ν φ)(y) where y = Λ−1 x.15). in the last step. This follows because det Λ = 1. 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). is all we deal with). 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. Putting this all together. such as the metric ηµν . for now. we need the fact that we don’t pick up a Jacobian factor when we change integration variables from d4 x to d4 y. 2. – 12 – . (L is linear in time derivatives. Example 2: First Order Dynamics In the ﬁrst-order Lagrangian (1. it’s easy to see if the action is Lorentz invariant: just make sure all the Lorentz indices µ = 0. we ﬁnd that the action is indeed invariant under Lorentz transformations. In practice. (At least for Lorentz transformation connected to the identity which.27) The potential terms transform in the same way. 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. 3 are contracted with Lorentz invariant objects. with φ2 (x) → φ2 (y).

32) Repeating the analysis above. we can deﬁne the charge in a ﬁnite volume V . However.29) d3 x j 0 (1. we ﬁnd that dQV =− dt d3 x ∇ · j = − j · dS (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. supersymmetries..33) V A – 13 – . We start here by recasting Noether’s theorem in a ﬁeld theoretic framework. There are Lorentz symmetries..3 Symmetries The role of symmetries in ﬁeld theory is possibly even more important than in particle mechanics. dQ = dt d3 x R3 ∂j 0 =− ∂t R3 d3 x ∇ · j = 0 (1. Of course. ∂j 0 /∂t + ∇ · j = 0. historically electrodynamics was the ﬁrst Lorentz invariant theory to be discovered: it was found even before the concept of Lorentz invariance. 1. QV = V d3 x j 0 (1. internal symmetries. To see this. You can check that Maxwell’s Lagrangian (1.30) which one can immediately see by taking the time derivative.Example 3: Maxwell’s Equations Under a Lorentz transformation Aµ (x) → Λµν Aν (Λ−1 x). deﬁned as Q= R3 (1. A Comment: A conserved current implies a conserved charge Q. 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.31) assuming that j → 0 suﬃciently quickly as |x| → ∞.21) is indeed invariant.3. in other words. gauge symmetries.

Then δL = ∂L ∂L δφa + ∂µ (δφa ) ∂φa ∂(∂µ φa ) ∂L ∂L δφa + ∂µ − ∂µ = ∂φa ∂(∂µ φa ) ∂L δφa ∂(∂µ φa ) (1.2 An Example: Translations and the Energy-Momentum Tensor Recall that in classical particle mechanics.37) But for the symmetry transformation δφa = Xa (φ).36) When the equations of motion are satisﬁed.where A is the area bounding V and we have used Stokes’ theorem. we have by deﬁnition δL = ∂µ F µ . we ﬁrst consider making an arbitrary transformation of the ﬁelds δφa . To derive Noether’s theorem. Equating this expression with (1.37) gives us the result ∂µ j µ = 0 with j µ = ∂L Xa (φ) − F µ (φ) ∂(∂µ φa ) (1. Proof of Noether’s Theorem: We’ll prove the theorem by working inﬁnitesimally. We will now see something similar in ﬁeld theories. invariance under spatial translations gives rise to the conservation of momentum.38) 1. We may always do this if we have a continuous symmetry.35) (1. the term in square brackets vanishes. This kind of local conservation of charge holds in any local ﬁeld theory.34) for some set of functions F µ (φ).39) – 14 – . This equation means that any charge leaving V must be accounted for by a ﬂow of the current 3vector j out of the volume. while invariance under time translations is responsible for the conservation of energy. Consider the inﬁnitesimal translation xν → xν − ǫν ⇒ φa (x) → φa (x) + ǫν ∂ν φa (x) (1. We say that the transformation δφa (x) = Xa (φ) is a symmetry if the Lagrangian changes by a total derivative. δL = ∂µ F µ (1.3. So we’re left with δL = ∂µ ∂L δφa ∂(∂µ φa ) (1.

From the above discussion.45) (1.40) (This is the correct transformation for a Lagrangian that has no explicit x dependence. 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. we can compute T µν = ∂ µ φ ∂ ν φ − η µν L (1. It satisﬁes ν (1. while P i is the total momentum of the ﬁeld conﬁguration. T µν came out symmetric. instead of minus.47) – 15 – .(where the sign in the ﬁeld transformation is plus. we may invoke Noether’s theorem which gives us four conserved currents (j µ )ν . but only depends on x through the ﬁelds φa (x). 3.41) T µ is called the energy-momentum tensor. as opposed to passive. For this example. so that T µν = T νµ . one for each of the translations ǫν with ν = 0. Nevertheless. transformation). the conserved energy and momentum are given by E= Pi = 1 1 ˙ d3 x 2 φ2 + 2 (∇φ)2 + 1 m2 φ2 2 (1. 2. 1. An Example of the Energy-Momentum Tensor Consider the simplest scalar ﬁeld theory with Lagrangian (1. This won’t always be the case.42) where E is the total energy of the ﬁeld conﬁguration.43) (1. because we’re doing an active.7). Similarly. the Lagrangian transforms as L(x) → L(x) + ǫν ∂ν L(x) (1. All theories that we consider in this course will have this property). (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 (1.46) ˙ d3 x φ ∂ i φ Notice that for this example. Since the change in the Lagrangian is a total derivative.44) One can verify using the equation of motion for φ that this expression indeed satisﬁes ∂µ T µν = 0.

which agrees with the number of diﬀerent Lorentz transformations (3 rotations + 3 boosts). rotational invariance gave rise to conservation of angular momentum. As a check.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. and replacing the kinetic terms with suitable covariant derivatives using “minimal coupling”. we now have further Lorentz transformations. Firstly consider coupling the theory to a curved background spacetime. 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. Then a symmetric energy momentum tensor in the ﬂat space theory is given by √ 2 ∂( −gL) µν (1. the number of diﬀerent 4×4 anti-symmetric matrices is 4×3/2 = 6. we ﬁrst need the inﬁnitesimal form of the Lorentz transformations Λµν = δ µν + ω µν (1.3. What conserved quantity do they correspond to? To answer these questions.48) Θ =−√ −g ∂gµν gµν =ηµν It should be noted however that this trick requires a little more care when working with spinors.3 Another Example: Lorentz Transformations and Angular Momentum In classical particle mechanics. What is the analogy in ﬁeld theory? Moreover. This guarantees that ∂µ ∂ρ Γρµν = 0 so that the new energy-momentum tensor is also a conserved current. The condition (1. In fact this observation provides a quick and easy way to determine a symmetric energy-momentum tensor.49) where ω µν is inﬁnitesimal. 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. introducing an arbitrary metric gµν (x) in place of ηµν .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 . 1. namely boosts.51) – 16 – .

appearing here as a conservation law. We can write them as which satisfy ∂µ (J µ )ρσ = 0 and give rise to 6 conserved charges. leaving us with the following consequence of invariance under boosts: d d3 x xi T 00 = constant (1. σ = 1. For ρ. rather surprisingly. It’s kind of like a ﬁeld theoretic version of Newton’s ﬁrst law but.58) (1.54) = −ω ρν ρ ∂(∂µ φ) Unlike in the previous example.56) (J µ )ρσ = xρ T µσ − xσ T µρ (1. – 17 – . i.53) (1. so dP i/dt = 0.e.59) dt This is the statement that the center of energy of the ﬁeld travels with a constant velocity. we should strip it out to give six diﬀerent currents. the Lorentz transformation is a rotation and the three conserved charges give the total angular momentum of the ﬁeld. 3. the Lagrangian density transforms as where the last equality follows because ω µµ = 0 due to anti-symmetry. I’ve left the inﬁnitesimal choice of ω µν in the expression for this current. 2. Qij = d3 x (xi T 0j − xj T 0i ) d3 x (x0 T 0i − xi T 00 ) d3 x xi T 00 (1. But really. 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.from which we see that δφ = −ω µν xν ∂µ φ By the same argument. the conserved charges are Q0i = (1. Once again.52) But what about the boosts? In this case. one for each choice of ω µν .55) δL = −ω µν xν ∂µ L = −∂µ (ω µν xν L) (1.57) 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.

rotates the phase of ψ: ψ → eiα ψ or δψ = iαψ (1.60) 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. The simplest example occurs √ for a complex scalar ﬁeld ψ(x) = (φ1 (x) + iφ2 (x))/ 2.3. (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. it’s easier (and equivalent) to treat ψ and ψ ⋆ as independent variables and vary the action with respect to both of them. baryon or lepton number). For example. Isospin is an example of such a symmetry.62) where the latter equation holds with α inﬁnitesimal.1. To ﬁnd the equations of motion. Non-Abelian Internal Symmetries Consider a theory involving N scalar ﬁelds φa .4 Internal Symmetries The above two examples involved transformations of spacetime. 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. An internal symmetry is one that only involves a transformation of the ﬁelds and acts the same at every point in spacetime. we could expand ψ in terms of φ1 and φ2 and work as before. We can build a real Lagrangian by where the potential is a general polynomial in |ψ|2 = ψ ⋆ ψ. The Lagrangian remains invariant under this change: δL = 0. all with the same mass and the Lagrangian 1 L= 2 N L = ∂µ ψ ⋆ ∂ µ ψ − V (|ψ|2 ) (1.64) In this case the Lagrangian is invariant under the non-Abelian symmetry group G = SO(N). However.61) ∂ψ ⋆ The Lagrangian has a continuous symmetry which rotates φ1 and φ2 or. – 18 – . The associated conserved current is j µ = i(∂ µ ψ ⋆ )ψ − iψ ⋆ (∂ µ ψ) (1. equivalently. 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. albeit realized only approximately in Nature.

67) We see that we can identify the function hµ = j µ as the conserved current. α is a constant real number. in the action that contribute to the current.43) which is a single number characterizing the whole ﬁeld conﬁguration. (The potential terms are invariant even when α = α(x)). The Hamiltonian density is given by ˙ where.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.66) which means that when the equations of motion are satisﬁed (so δS = 0 for all variations. ∂L (1.68) π a (x) = ˙ ∂ φa The conjugate momentum π a (x) is a function of x. and focus instead on canonical quantization.69) – 19 – .65) since we know that δL = 0 when α is constant. Now consider performing the transformation but where α depends on spacetime: α = α(x). just like the ﬁeld φa (x) itself. However. The Hamiltonian is then simply H= d3 x H (1.4 The Hamiltonian Formalism The link between the Lagrangian formalism and the quantum theory goes via the path integral. the change must be of the form δL = (∂µ α) hµ (φ) (1.70) ˙ H = π a (x)φa (x) − L(x) (1. This way of viewing things emphasizes that it is the derivative terms. For this we need the Hamiltonian formalism of ﬁeld theory. We start by deﬁning the momentum π a (x) conjugate to φa (x). Here. we eliminate φa (x) in favour of π a (x) everywhere in H. 1. The action is no longer invariant. including δφ = α(x)φ) we have ∂µ hµ = 0 (1. as in classical mechanics. It is not to be confused with the total momentum P i deﬁned in (1. not the potential terms. The change in the action is therefore δS = d4 x δL = − d4 x α(x) ∂µ hµ (1. In this course we will not discuss path integral methods. (We may generalize the discussion easily to a non-Abelian internal symmetry for which α becomes a matrix).

the physics must remain unchanged. If we start from a relativistic theory.45) that we get from applying Noether’s theorem for time translation invariance. In contrast. H= 1 d3 x 1 π 2 + 2 (∇φ)2 + V (φ) 2 (1. We will pause at several points along the quantum route to check that this is indeed the case. the Hamiltonian formalism is not manifestly Lorentz invariant: we have picked a preferred time. t) = − ˙ ∂H ∂φ(x. – 20 – .71) ˙ the momentum is given by π = φ. t) and π(x.6). In the Lagrangian formalism. t) arise from Hamilton’s equations. For example. all ﬁnal answers must be Lorentz invariant even if it’s not manifest at intermediate steps. do not look Lorentz invariant. unlike the Euler-Lagrange equations (1.An Example: A Real Scalar Field For the Lagrangian 1 ˙ L = 1 φ2 − 2 (∇φ)2 − V (φ) 2 (1. the equations of motion for φ(x) = φ(x. t) (1. which gives us the Hamiltonian.73) which. ˙ φ(x.72) Notice that the Hamiltonian agrees with the deﬁnition of the total energy (1. Lorentz invariance is clear for all to see since the action is invariant under Lorentz transformations. t) = ∂H ∂π(x. Nevertheless. even though the Hamiltonian framework doesn’t look Lorentz invariant.

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

from all the others. Let’s recall how to do this.4) To exhibit the coordinates in which the degrees of freedom decouple from each other. To quantize φ(x. or sometimes ladder operators) a= i ω q+√ p .1. 2. so that the equations of motion are linear. t) solves the equation of a harmonic oscillator vibrating at frequency ωp = + p 2 + m2 (2.10) – 22 – .9) which can be easily inverted to give 1 q = √ (a + a† ) . the simplest relativistic free theory is the classical Klein-Gordon (KG) equation for a real scalar ﬁeld φ(x. φ(x.8) with the canonical commutation relations [q. Free ﬁeld theories typically have Lagrangians which are quadratic in the ﬁelds. 2ω p = −i ω (a − a† ) 2 (2. t) we must simply quantize this inﬁnite number of harmonic oscillators. we need only take the Fourier transform. φ(p. each vibrating at a diﬀerent frequency with a diﬀerent amplitude. t).6) d3 p ip·x e φ(p. To ﬁnd the spectrum we deﬁne the creation and annihilation operators (also known as raising/lowering operators.1 The Simple Harmonic Oscillator Consider the quantum mechanical Hamiltonian 1 1 H = 2 p2 + 2 ω 2 q 2 (2. p] = i. ∂µ ∂ µ φ + m2 φ = 0 (2. for each value of p.5) Thus.7) We learn that the most general solution to the KG equation is a linear superposition of simple harmonic oscillators. t) (2π)3 (2. t) satisﬁes ∂2 + (p 2 + m2 ) ∂t2 φ(p. For example. t) = Then φ(p. 2 2ω a† = i ω q−√ p 2 2ω (2. t) = 0 (2.

. . Ha |E = (E − ω)a |E (2. (2. n| n = 1. E. Then we can construct more eigenstates by acting with a and a† . 1 H |0 = 2 ω |0 (2.Substituting into the above expressions we ﬁnd [a.15) If the energy is bounded below. so that H |E = E |E . .12) One can easily conﬁrm that the commutators between the Hamiltonian and the creation and annihilation operators are given by [H.14) So we ﬁnd that the system has a ladder of states with energies . 2.18) (2. E + 2ω. Ha† |E = (E + ω)a† |E . E − ω.2 The Free Scalar Field We now apply the quantization of the harmonic oscillator to the free scalar ﬁeld. Let |E be an eigenstate with energy E. a] = −ωa (2. a† ] = 1 while the Hamiltonian is given by H = 1 ω(aa† + a† a) 2 1 = ω(a† a + 2 ) (2.17) where I’ve ignored the normalization of these states so. . indexed by the 3-momentum p.19) d3 p (−i) (2π)3 – 23 – . . 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 (2.13) These relations ensure that a and a† take us between energy eigenstates. a† ] = ωa† and [H. E + ω.16) Excited states then arise from repeated application of a† . |n = (a† )n |0 with H |n = (n + 1 )ω |n 2 (2. there must be a ground state |0 which satisﬁes a |0 = 0. .11) (2. This has ground state energy (also known as zero point energy). We write φ and π as a linear sum of an inﬁnite number of creation and annihilation oper† ators ap and ap .

aq ] eip·x−iq·y + [ap . π(y)] = ωq d3 p d3 q (−i) † † −[ap . 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. aq ] = 0 † [ap . Now using the expression for the 2 frequency ωp = p 2 + m2 . Assume that [ap . π(y)] = iδ (3) (x − y) ⇔ † † [ap . Then [φ(x).19).Claim: The commutation relations for φ and π are equivalent to the following com† mutation relations for ap and ap [φ(x). φ(y)] = [π(x).22) – 24 – . 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) (2.18) and (2. aq ] = [ap .20) † Proof: We’ll show this just one way. in turn. to get to the third line we’ve integrated over d3 x to get delta-functions δ (3) (p±q) which.21) The Hamiltonian † Let’s now compute the Hamiltonian in terms of ap and ap . allow us to perform the d3 q integral. 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. aq ] = (2π)3 δ (3) (p − q). π(y)] = 0 [φ(x).

Hmmmm. 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. We recognize it as the sum of ground state energies for each harmonic oscillator. Each tells us something important. Moreover. the integral over ωp diverges at large p.3 The Vacuum Following our procedure for the harmonic oscillator.26) V (2π)3 2 which is still inﬁnite. The ﬁrst arises because space is inﬁnitely large. But E0 → ∞ due to the |p| → ∞ limit of the integral. – 25 – . This divergence arises because of our hubris. usually that we’re doing something wrong. What to do? Let’s start by looking at the ground state where this inﬁnity ﬁrst becomes apparent. To ﬁnd E0 we can simply divide by the volume.22). or asking the wrong question. We’ve found a delta-function. let’s deﬁne the vacuum |0 by insisting that it is annihilated by all ap .24). To extract out this inﬁnity. evaluated at zero where it has its inﬁnite spike. corresponding to arbitrarily high energies. 1 ωp δ (3) (0) | 0 = ∞ |0 (2. So the δ(0) divergence arises because we’re computing the total energy. Then. Let’s take some time to explore where this inﬁnity comes from and how we should deal with it. (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). This is clearly absurd. We’ve assumed that our theory is valid to arbitrarily short distance scales. let’s consider putting the theory in a box with sides of length L. taking the limit where L → ∞. We impose periodic boundary conditions on the ﬁeld.25) where V is the volume of the box. 2.23) With this deﬁnition.24) 2 The subject of quantum ﬁeld theory is rife with inﬁnities. H |0 ≡ E0 |0 = d3 p In fact there are two diﬀerent ∞’s lurking in the expression (2. This is a high frequency — or short distance — inﬁnity known as an ultra-violet divergence. rather than the energy density E0 . E0 d3 p 1 E0 = = ωp (2. the energy E0 of the ground state comes from the second term in (2. 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. ap |0 = 0 ∀p (2.

In fact. . So.3. we could write (2. what cancels them to such high accuracy? This is the cosmological constant problem.28) It is deﬁned to be the usual product with all annihilation operators ap placed to the right. Deﬁnition: We write the normal ordered string of operators φ1 (x1 ) . which is classically the same as our original choice. then upon quantization it would naturally give H = ωa† a as in (2. We’ll come across a number of ways of dealing with it. For example.27) so that.27) as : H := d3 p † ωp ap ap (2π)3 (2. H= d3 p ω a† a 3 p p p (2π) (2.27). 1 Rµν − 2 gµν = −8πGTµν + Λgµν (2. There’s no way to measure E0 directly.1 The Cosmological Constant Above I wrote “there’s no way to measure E0 directly”. the diﬀerence between this Hamiltonian and the previous one is merely an ordering ambiguity in moving from the classical theory to the quantum theory. for the Hamiltonian. We expect them to appear as a cosmological constant Λ = E0 /V . 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. 2. In particular. so we can simply redeﬁne the Hamiltonian by subtracting oﬀ this inﬁnity. if they do. . we will normal order all operators in this manner. In physics we’re only interested in energy diﬀerences. This type of ordering ambiguity arises a lot in ﬁeld theories.We can deal with the inﬁnity in (2. with this new deﬁnition.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 . The method that we’ve used above is called normal ordering. H |0 = 0. φn (xn ) as : φ1 (x1 ) . the Standard Model is valid at least up to 1012 eV . . This is much smaller than other scales in particle physics. No one knows the answer! – 26 – . if we deﬁned the Hamiltonian of the harmonic oscillator to be H = (1/2)(ωq − ip)(ωq + ip). . Why don’t the zero point energies of these ﬁelds contribute to Λ? Or. φn (xn ) : (2.29) In the remainder of this section.24) in a more practical way.

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

expression (2. Physically one could argue that any real plate cannot reﬂect waves of arbitrarily high frequency: at some point.38) – 28 – . However (2. things begin to leak. Mathematically.35) is ﬁnite. rather than d = 3 + 1 dimensions. otherwise it’s not something we can really trust. 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’ve used the fact that a ≪ d. we want to ﬁnd a way to neglect modes of momentum p ≫ a−1 for some distance scale a ≪ d. It’s a lot simpler if we look at the problem in d = 1 + 1 dimensions. We’ll ﬁnd that all the same physics is at play. Now the energy is given by π E1+1 (d) = 2d ∞ n n=1 (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. allowing us to start manipulating it thus. and gives us something we can easily work with. in the last line. We could regulate this in a number of diﬀerent ways. 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.But there’s a problem. This renders the expression ﬁnite. but a little complicated. E is inﬁnite! What to do? The problem comes from the arbitrarily high momentum modes.35) is do-able.36) We now regulate this sum by introducing the UV cutoﬀ a introduced above.33) to. E1+1 = E1+1 (d) + E1+1 (L − d) = π L − 2 2πa 24 1 1 + d L−d + O(a2 ) (2. Of course. known as the ultra-violet (UV) cut-oﬀ. we regain the full. The integral (2. inﬁnite.35) which has the property that as a → 0.37) 2 2πa 24d where.33). We can now compute the full energy. One way to do this is to change the integral (2.

– 29 – ..39) where the . and take a → 0 and L → ∞. . 2. we can now turn to the excitations of the ﬁeld. ap ] = ωp ap and [H. the force between the plates remains ﬁnite. If we ploughed through the analogous calculation in d = 3 + 1 dimensions. we would ﬁnd the result 1 ∂E π2 = A ∂d 480d4 (2. which is to be expected. . It’s easy to verify that † † [H. You will need to attend next term’s “String Theory” course to see what on earth this has to do with the Casimir force. include terms of size d/L and a/d.. (The +2 comes from the fact the string itself is extended in one space and one time dimension). just as for the harmonic oscillator. This is the Casimir force2 . the force is given by π ∂E1+1 = + .4 Particles Having dealt with the vacuum. ∂d 24d2 (2.40) The true Casimir force is twice as large as this. ap ] = −ωp ap (2. and performed the integral (2. However. 2 Ep = p 2 + m2 2 (2. The key point is that as we remove both the regulators. due to the two polarization states of the photon.41) which means that.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. we can construct energy eigen† states by acting on the vacuum |0 with ap .42) This state has energy H |p = ωp |p with 2 ωp = p 2 + m2 (2.This is still inﬁnite in the limit a → 0. Let † |p = ap |0 (2.43) But we recognize this as the relativistic dispersion relation for a particle of mass m and 3-momentum p.35).

J i |p = 0 = 0. Multi-Particle States. .46) telling us that the state |p has momentum p.47) It’s not hard to show that acting on the one-particle state with zero momentum. (2.46) and turn it into an operator. After normal ordering. which we interpret as telling us that the particle carries no internal angular momentum. p This means that the particles are bosons.55) and turn it into an operator. † † † † † † |0 . pn = ap1 . we may take the classical expression for the total angular momentum of the ﬁeld (1. from now on we’ll write Ep everywhere instead of ωp . In other words. q = |q. the state is symmetric under exchange of any two particles. J i = ǫijk d3 x (J 0 )jk (2. |p.We interpret the state |p as the momentum eigenstate of a single particle of mass m. ap aq ar |0 . apn |0 (2. We interpret the state in which n a† ’s act on the vacuum as an n-particle state. . .45) Acting on our state |p with P . ap aq |0 . † † |p1 . . Another property of |p that we can study is its angular momentum. ap |0 . we learn that it is indeed an eigenstate.49) (2. The full Hilbert space of our theory is spanned by acting on the vacuum with all possible combinations of a† ’s. To stress this. . Once again. quantizing a scalar ﬁeld gives rise to a spin 0 particle.48) Because all the a† ’s commute among themselves. it becomes P =− d3 x π ∇φ = d3 p † p ap ap 3 (2π) (2. P |p = p |p (2. Let’s check this particle interpretation by studying the other quantum numbers of |p . .50) – 30 – . . . Bosonic Statistics and Fock Space We can create multi-particle states by acting multiple times with a† ’s. We may take the classical total momentum P given in (1. For example.

52) They are operator valued distributions. The number operator commutes with the Hamiltonian. 0| φ(x) |0 = 0.4.53) which is partially localized in both position and momentum space. . This is a property of free theories. . they’re not localized in space in any way — they are momentum eigenstates.51) and satisﬁes N |p1 . It is called the number operator N N= d3 p † a a (2π)3 p p (2.This space is known as a Fock space. 2. taking us between the diﬀerent sectors in the Fock space.1 Relativistic Normalization We have deﬁned the vacuum |0 which we normalize as 0| 0 = 1. This means that we can place ourselves in the n-particle sector. . † 0| ap ap |0 = p| p = (2π)3 δ(0) and 0|φ(x) φ(x) |0 = x| x = δ(0) (2. The one-particle † states |p = ap |0 then satisfy p| q = (2π)3 δ (3) (p − q) (2. 0| φ(x)φ(x) |0 = ∞. the ﬂuctuations of the operator at a ﬁxed point are inﬁnite. we can create a wavepacket |ϕ = d3 p −ip·x e ϕ(p) |p (2π)3 (2. but will no longer be true when we consider interactions: interactions create and destroy particles. . For example. . Operator Valued Distributions Although we’re referring to the states |p as “particles”. 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). rather than functions. nor ap are good operators acting on the Fock space. We can construct well deﬁned operators by smearing these distributions over space. . ensuring that particle number is conserved. Similarly. pn . (A typical state might be described by the Gaussian ϕ(p) = exp(−p 2 /2m2 )). This means that although φ(x) has a well deﬁned vacuum expectation value. There is a useful operator which counts the number of particles in a given state in the Fock space. and stay there. H] = 0. For example. in quantum ﬁeld theory neither the operators φ(x). The Fock space is simply the sum of the n-particle Hilbert spaces. [N. pn = n |p1 . This is because they don’t produce normalizable states. for all n ≥ 0.54) – 31 – . . .

How do we ﬁgure this out? The trick is to look at an object which we know is Lorentz invariant. Solving for p0 . 2 2 pµ pµ = m2 ⇒ p0 = Ep = p 2 + m2 (2.59) Proof: d4 p is obviously Lorentz invariant. What could go wrong? Suppose we have a Lorentz transformation p µ → (p′ )µ = Λµν p ν (2.55) such that the 3-vector transforms as p → p ′ . d4 p δ(p2 − p 2 − m2 ) 0 which completes the proof.But is this Lorentz invariant? It’s not obvious because we only have 3-vectors.54) we know this is given by 1= d3 p |p p| (2π)3 (2.61) p0 =Ep – 32 – .56) This would mean that the normalizations of |p and |p ′ are the same whenever p and p ′ are related by a Lorentz transformation. In general.60) is also Lorentz invariant. One such object is the identity operator on one-particle states (which is really the projection operator onto one-particle states). In the quantum theory.58) This operator is Lorentz invariant. we could get |p → λ(p. With the normalization (2. = p0 >0 d3 p 2p0 (2. there are two branches of solutions: p0 = ±Ep . p ′). Are these individually Lorentz invariant? In fact the answer is no. p ′ ) |p ′ (2. |p → |p ′ = U(Λ) |p (2. but it consists of two terms: the measure d3 p and the projector |p p|. Claim The Lorentz invariant measure is. And the relativistic dispersion relation for a massive particle. d3 p 2Ep (2. So piecing everything together. it would be preferable if the two states are related by a unitary transformation.57) for some unknown function λ(p. But the choice of branch is another Lorentz invariant concept. But we haven’t been at all careful with normalizations. the following combination must be Lorentz invariant.

66) † 2Ep ap . These states now satisfy p| q = (2π)3 2Ep δ (3) (p − q) Finally.68) where the second equation is the complex conjugate of the ﬁrst. If we write ψ in terms √ of real scalar ﬁelds by ψ = (φ1 + iφ2 )/ 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. in contrast to the Lagrangian (1. For example.5 Complex Scalar Fields Consider a complex scalar ﬁeld ψ(x) with Lagrangian L = ∂µ ψ ⋆ ∂ µ ψ − M 2 ψ ⋆ ψ (2. there is no factor of 1/2 in front of the Lagrangian for a complex scalar ﬁeld.65) Some texts also deﬁne relativistically normalized creation operators by a† (p) = We won’t make use of this notation here.63) So ﬁnally we learn that the relativistically normalized momentum states are given by (2.7) for a real scalar ﬁeld. 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. 2.67) Notice that.From this result we can ﬁgure out everything else. 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 (2.69) – 33 – . (2. we get the factor of 1/2 coming from the √ 1/ 2’s. we can rewrite the identity on one-particle states as 1= d3 p 1 |p p| (2π)3 2Ep (2.62) (2. The equations of motion are ∂µ ∂ µ ψ + M 2 ψ = 0 ∂µ ∂ µ ψ ⋆ + M 2 ψ ⋆ = 0 (2.

π † (y)] = 0 (2. ψ † (y)] = 0.74) After normal ordering. bp † and cp . – 34 – .72) and † [bp . bq ] = [cp .71) together with others related by complex conjugation. We also turn this into a quantum operator ﬁeld which we write as. 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. the corresponding quantum ﬁeld ψ is not hermitian. π(y)] = iδ (3) (x − y) and [ψ(x). etc. cq ] = [bp .70) d3 p (−i) (2π)3 The commutation relations between ﬁelds and momenta are given by [ψ(x). cq ] = 0 (2. we’ll soon see that in interacting theories Q survives as a conserved quantity. However. This is the reason that we have diﬀerent operators b and c† appearing in the positive ˙ ˙ and negative frequency parts.Since the classical ﬁeld ψ is not real. while Nc and Nb individually do not. One can easily check that these ﬁeld commutation relations are equivalent to the commutation relations for the operators bp and cp . this becomes the quantum operator Q= d3 p † † (cp cp − bp bp ) = Nc − Nb 3 (2π) (2. in our free ﬁeld theory this isn’t such a big deal because both Nc and Nb are separately conserved. These have the interpretation of creating two types of particle. Of course. They are interpreted as particles and anti-particles. both of mass M and both spin zero. the particle is its own antiparticle.73) † In summary. The classical ﬁeld momentum is π = ∂L/∂ ψ = ψ ⋆ . Recall that the theory (2. ψ(y)] = [ψ(x). for a real scalar ﬁeld there is only a single type of particle: for a real scalar ﬁeld. bq ] = (2π)3 δ (3) (p − q) (2. † [cp . Q] = 0. quantizing a complex scalar ﬁeld gives rise to two creation operators.67) has a classical conserved charge Q=i ˙ ˙ d3 x (ψ ⋆ ψ − ψ ⋆ ψ) = i d3 x (πψ − ψ ⋆ π ⋆ ) (2. as well as the usual [ψ(x). ensuring that Q is conserved quantity in the quantum theory. cq ] = (2π)3 δ (3) (p − q) † [bp . We have [H. In contrast. cq ] = [bp .75) so Q counts the number of anti-particles (created by c† ) minus the number of particles (created by b† ).

OH = eiHt OS e−iHt so that dOH = i[H. φ(y. [φ(x.2.6 The Heisenberg Picture Although we started with a Lorentz invariant Lagrangian.80) Meanwhile.77) where the subscripts S and H tell us whether the operator is in the Schr¨dinger or o Heisenberg picture. π = i[H. π] = ˙ i [ 2 d3 y π(y)2 + ∇φ(y)2 + m2 φ(y)2 . t). t)] = [π(x. The operators in the two pictures agree at a ﬁxed time. π(y. we slowly butchered it as we quantized. π(x)] – 35 – . t) depends on time. For example. t). we have i ˙ φ = i[H. the operators φ(x) depend on space. t)] = iδ (3) (x − y) (2. t)] = 0 [φ(x. The commutation relations (2. we can start to study how it evolves. Meanwhile. t = 0.2) become equal time commutation relations in the Heisenberg picture. introducing a preferred time coordinate t. 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. the one-particle states evolve in time by Schr¨dinger’s equation. φ(x)] d3 y π(y) (−i) δ (3) (y − x) = π(x) (2. It’s not at all obvious that the theory is still Lorentz invariant after quantization.76) Things start to look better in the Heisenberg picture where time dependence is assigned to the operators O. π(y.78) (2. OH ] dt (2. φ] = [ 2 =i d3 y π(y)2 + ∇φ(y)2 + m2 φ(y)2 . o i d |p(t) = H |p(t) dt ⇒ |p(t) = e−iEp t |p (2.79) Now that the operator φ(x) = φ(x. the equation of motion for π reads. t). say. t) = φ(x) (the Heisenberg picture). In ﬁeld theory. but not on time. For example.

83) † † which follows from the commutation relations [H. We can write the Fourier expansion of φ(x) by using the deﬁnition (2. φ(x. eiHt ap e−iHt = e−iEp t ap † † and eiHt ap e−iHt = e+iEp t ap (2.82). Putting (2. It’s simple to check that (2. where φ(x) now satisﬁes the KleinGordon equation. t).1 Causality O2 O1 x Figure 4: We’re approaching something Lorentz invariant in the Heisenberg picture.18) except that the exponent is now written in terms of 4vectors.80) and (2. π(y. π(x)] +2im2 φ(y) δ (3)(x − y) d3 y ∇y δ (3) (x − y) ∇y φ(y) − m2 φ(x) (2.84) t which looks very similar to the previous expansion (2. 2. we’ve simply integrated by parts. To reach the last line. p · x = Ep t − p · x.82) Things are beginning to look more relativistic. ap ] = +Ep ap . ap ] = −Ep ap and [H. [φ(x.= i 2 d3 y (∇y [φ(y). But there’s still a hint of non-Lorentz invariance because φ and π satisfy equal time commutation relations. π(x)]) ∇φ(y) + ∇φ(y) ∇y [φ(y). t) = d3 p (2π)3 1 2Ep † ap e−ip·x + ap e+ip·x (2. This then gives. t)] = iδ (3) (x − y) (2. (Note also that a sign has ﬂipped in the exponent due to our Minkowski metric contraction).85) – 36 – .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.77) and noting.6.84) indeed satisﬁes the Klein-Gordon equation (2.81) together we ﬁnd that the ﬁeld operator φ satisﬁes the Klein-Gordon equation ∂µ ∂ µ φ + m2 φ = 0 (2.

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

91) So it decays exponentially quickly outside the lightcone but. We can look at the equation [ψ(x). nonetheless.89) as [φ(x). 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. In fact. With a complex scalar ﬁeld. the amplitudes for these two events cancel. ψ † (y)] = 0 outside the lightcone. ∆F (x − y) = 0| T φ(x)φ(y) |0 = D(x − y) D(y − x) x0 > y 0 y 0 > x0 (2. it can just as easily travel from y → x. 2.93) – 38 – . Prepare a particle at spacetime point y.1 The Feynman Propagator As we will see shortly. Yet we’ve just seen that spacelike measurements commute and the theory is causal. 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 any measurement.7. one of the most important quantities in interacting ﬁeld theory is the Feynman propagator. one can show that D(x − y) decays like D(x − y) ∼ e−m|x−y| (2. How do we reconcile these two facts? We can rewrite the calculation (2.92) There are words you can drape around this calculation. For spacelike separations. φ(y)] = D(x − y) − D(y − x) = 0 if (x − y)2 < 0 (2. it is more interesting.90) d3 p 1 −ip·(x−y) e ≡ D(x − y) (2π)3 2Ep The function D(x − y) is called the propagator. When (x − y)2 < 0. there is no Lorentz invariant way to order events. this interpretation is also there for a real scalar ﬁeld because the particle is its own antiparticle. (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 ′ (2.7 Propagators We could ask a diﬀerent question to probe the causal structure of the theory.2.

with p0 ﬁxed by the mass-shell condition to be p0 = Ep . 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 need a prescription for avoiding these singularities in the p0 integral.95) for ∆F . T φ(x)φ(y) = φ(x)φ(y) φ(y)φ(x) x0 > y 0 y 0 > x0 (2. the denominator p2 −m2 = (p0 )2 −p 2 −m2 produces a pole when p0 = ±Ep = ± p 2 + m2 .94) Claim: There is a useful way of writing the Feynman propagator in terms of a 4momentum integral. we integrated only over 3-momentum. where p0 → −i∞. To get the Feynman propagator. Hence when x0 > y 0 we have ∆F (x − y) = d3 p −2πi −iEp (x0 −y0 )+ip·(x−y) ie (2π)4 2Ep – 39 – . Until now.where T stands for time ordering. ensuring that the integrand vanishes since 0 0 0 e−ip (x −y ) → 0. In the expression (2. we close the contour in the lower half plane.95) Notice that this is the ﬁrst time in this course that we’ve integrated over 4-momentum. as it stands this integral is ill-deﬁned because.96) so the residue of the pole at p0 = ±Ep is ±1/2Ep . we must choose the contour to be Im(p0) −E p +E p Re(p0) Figure 5: The contour for the Feynman propagator. for each value of p. placing all operators evaluated at later times to the left so. we have no such condition on p0 . ∆F (x − y) = i d4 p e−ip·(x−y) 4 p2 − m2 (2π) (2. However. Proof: 1 1 1 = 0 = 0 2 2 p2 − m2 (p ) − Ep (p − Ep )(p0 + Ep ) (2. When x0 > y 0.

– 40 – . ∆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. 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. This way of writing the propagator is. so the integral along the real p0 axis is equivalent to the contour shown in Figure 5.99) + iε Im(p0) Re(p0) −iε with ǫ > 0. called the “iǫ prescription”. 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 . Once again we reproduce the Feynman propagator.7. This has the eﬀect of shifting the poles slightly oﬀ the real axis.98) where to go to from the second line to the third. In contrast. we close the contour in an anti-clockwise direction in the upper half plane to get. If we stay away from the singularities.2 Green’s Functions Figure 6: There is another avatar of the propagator: it is a Green’s function for the Klein-Gordon operator. and inﬁnitesimal. 2. Instead of specifying the contour. we may instead write the Feynman propagator as ∆F (x − y) = d4 p ie−ip·(x−y) (2π)4 p2 − m2 + iǫ (2. when y 0 > x0 .= d3 p 1 −ip·(x−y) e = D(x − y) (2π)3 2Ep (2.97) which is indeed the Feynman propagator for x0 > y 0.100) Note that we didn’t make use of the contour anywhere in this derivation. for obvious reasons. For some purposes it is also useful to pick other contours which also give rise to Green’s functions.

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

ψ † (y)] = 0 [ψ(x).¨ ˙ ˜ ˜ this is equivalent to saying that |ψ| ≪ m|ψ|.105) This looks very similar to the Schr¨dinger equation for a non-relativistic free particle of o mass m. so ψ → ψ. We get. we drop the term with two time derivatives and the KG equation becomes. In fact. we can derive this from the relativistic Lagrangian for a scalar ﬁeld by again taking the limit ∂t ψ ≪ mψ.110) – 42 – . the time derivatives drop out when we compute the Hamiltonian. i (ψ ⋆ ∇ψ − ψ∇ψ ⋆ ) 2m (2. 1 ˙ Since the Lagrangian already contains a “pq” term (instead of the more familiar 2 pq ˙ term). This Lagrangian has a conserved current arising from the internal symmetry ψ → eiα ψ. In order to determine the full trajectory of the ﬁeld. i ˜ ∂ψ 1 2˜ =− ∇ψ ∂t 2m (2.1. ψ(y)] = [ψ † (x).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.106) where we’ve divided by 1/2m.2 which gives rise to ﬁeld equations which are ﬁrst order in time derivatives. we need only specify ψ and ψ ⋆ at time t = 0: no time derivatives on the initial slice are required. We wrote down a Lagrangian in section 1. ψ † (y)] = δ (3) (x − y) (2. H= 1 ∇ψ ⋆ ∇ψ 2m (2.107) To move to the Hamiltonian formalism we compute the momentum π= ∂L = iψ ⋆ ˙ ∂ψ (2.109) To quantize we impose (in the Schr¨dinger picture) the canonical commutation relations o [ψ(x). In this limit. 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. The current has time and space components jµ = ψ ⋆ ψ. the non-relativistic Lagrangian becomes 1 ˙ L = +iψ ⋆ ψ − ∇ψ ⋆ ∇ψ 2m (2. After losing the ˜ ˜ tilde.

and the excitations are ap1 . It is only with relativity.113) which is the non-relativistic dispersion relation. and there can be no way to construct a multi-particle theory that conserves particle number. The existence of anti-particles is a consequence of relativity.111) where the commutation relations (2.1 Recovering Quantum Mechanics In quantum mechanics. particle number is conserved. 2. apn |0 . 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. the particles are their own anti-particles. Some comments: • We have a complex ﬁeld but only a single type of particle. The one-particle states have energy H |p = p2 |p 2m (2. that particle number can change.115) – 43 – . aq ] = (2π)3 δ (3) (p − q) (2. .We may expand ψ(x) as a Fourier transform ψ(x) = d3 p a eip·x (2π)3 p (2.112) † † The vacuum satisﬁes ap |0 = 0. We conclude that quantizing the ﬁrst order Lagrangian (2. we talk about the position and momentum operators X and P . • There is no non-relativistic limit of a real scalar ﬁeld. position is relegated to a label.106) gives rise to non-relativistic particles of mass m.8. • A related fact is that the conserved charge Q = d3 x : ψ † ψ : is the particle number. In quantum ﬁeld theory. . and the appearance of anti-particles. The anti-particle is not in the spectrum. This remains conserved even if we include interactions in the Lagrangian of the form ∆L = V (ψ ⋆ ψ) (2. In the relativistic theory.110) require † [ap .114) So in non-relativistic theories.

Then the operator ψ † (x) = d3 p † −ip·x a e (2π)3 p (2. Let’s now construct a state |ϕ by taking superpositions of one-particle states |x .116) creates a particle with δ-function localization at x.117) The function ϕ(x) is what we would usually call the Schr¨dinger wavefunction (in the o position representation).which.119) but what about the momentum operator P ? We will now show that P i |ϕ = d3 x −i |x (2. |ϕ = d3 x ϕ(x) |x (2. Let’s make sure that it indeed satisﬁes all the right properties. X i |ϕ = d3 x xi ϕ(x) |x ∂ϕ ∂xi (2.121) where we’ve used the relationship [ap .120) which tells us that P i acts as the familiar derivative on wavefunctions |ϕ . It’s also easy to construct the position operator. gives P |p = p |p . 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. We write |x = ψ † (x) |0 . Firstly. ψ † (x)] = e−ip·x which can be easily checked. it’s clear that acting with the position operator X has the right action of ϕ(x). To see that this is the case.118) d3 x x ψ † (x) ψ(x) (2. A natural position operator is then X= so that X |x = x |x . 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 (2.122) – 44 – . Let’s work in the non-relativistic limit. on one-particle states. Proceeding with our calculation.

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.105)! Except this time.120).109) can be rewritten as H= so we ﬁnd that i ∂ϕ 1 =− ∇2 ϕ ∂t 2m (2. the fact that the equation for the classical ﬁeld (2. despite the name. the operators X and P act as position and momentum operators in quantum mechanics. Historically. we’re interested in particles moving in some ﬁxed background potential V (x). 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. This can be easily incorporated into ﬁeld theory by working with a Lagrangian with explicit x dependence. P j ] |ϕ = iδ ij |ϕ . Nonetheless.105) and the oneparticle wavefunction (2. – 45 – . While such Lagrangians are useful in condensed matter physics. t) change in time? The Hamiltonian (2.124) d3 x 1 ∇ψ ⋆ ∇ψ = 2m d3 p p 2 † a a (2π)3 2m p p (2.124) coincide caused some confusion. It’s important to stress that. Interactions Often in quantum mechanics. it really is the Schr¨dinger equation. But what about dynamics? How does the wavefunction ϕ(x.123) But this is the same equation obeyed by the original ﬁeld (2.107) is Q = d3 x |ϕ(x)|2 which is the total probability. where all equations obey translational (and Lorentz) invariance. we’re not quantizing anything twice! We simply quantize a classical ﬁeld once. we rarely (or never) come across them in high-energy physics. Note in particular that the conserved charge arising from the Noether current (2. with [X i . 1 ˙ L = iψ ⋆ ψ − ∇ψ ⋆ ∇ψ − V (x) ψ ⋆ ψ 2m (2. complete with the usual probabilistic interpretation o for the wavefunction ϕ. So we learn that when acting on one-particle states.125) Note that this Lagrangian doesn’t respect translational symmetry and we won’t have the associated energy-momentum tensor.which conﬁrms (2.

Obviously these are only important for n particle states with n ≥ 2.126) which. – 46 – .One can also consider interactions between particles. is an operator which destroys two particles before creating two new ones. We therefore expect them to arise from additions to the Lagrangian of the form ∆L = ψ ⋆ (x) ψ ⋆ (x) ψ(x) ψ(x) (2. Such terms in the Lagrangian will indeed lead to inter-particle forces. in the quantum theory. both in the non-relativistic and relativistic setting. In the next section we explore these types of interaction in detail for relativistic theories.

the same dimensions as . where E has dimensions of mass.3. For example. is where most of the physics we see lies). and [d4 x] = −4. but these particles don’t interact with each other.1) The coeﬃcients λn are called coupling constants. we have [S] = 0. [λn ] = 4 − n (3. but nothing interesting then happens. using the convention described in the introduction. [φ] = 1 . because this statement only makes sense for dimensionless quantities. But this isn’t quite right. They have particle excitations. – 47 – . E is the energy scale of the process of interest. In a relativistic theory.2) What does this mean for the Lagrangian (3.1)? Since [∂µ ] = 1. These will take the form of higher order terms in the Lagrangian. parameterized by λn . With S = d4 x L. 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”. E > m. [m] = 1 . The various terms. Firstly. Here we’ll start to examine more complicated theories that include interaction terms. after all. Interacting Fields The free ﬁeld theories that we’ve discussed so far are very special: we can determine their spectrum. but a large perturbation at low energies E ≪ λ3 . Typically in quantum ﬁeld theory.3) So now we see why we can’t simply say we need λn ≪ 1. Since we’ve set = 1. note that the action has dimensions of angular momentum or. Terms that we add to the Lagrangian with this behavior are called relevant because they’re most relevant at low energies (which. we can read oﬀ the mass dimensions of all the factors to ﬁnd. fall into three diﬀerent categories • [λ3 ] = 1: For this term. equivalently. λn n 1 1 φ L = ∂µ φ ∂ µ φ − m2 φ2 − 2 2 n! n≥3 (3. the dimensionless parameter is λ3 /E. let’s do some dimensional analysis. consider the Lagrangian for a real scalar ﬁeld. To see why this is the case. so we can always make this perturbation small by taking λ3 ≪ m. We’ll start by asking what kind of small perturbations we can add to the theory. This means that λ3 φ3 /3! is a small perturbation at high energies E ≫ λ3 . the Lagrangian density must therefore have [L] = 4 (3.

it is typically impossible to avoid high energy processes in quantum ﬁeld theory. but the same argument applies in that case so bear with me). What are these interactions? Well. Let’s look at this a little more.1). This – 48 – . (They’re not of course: they’re described by non-Abelian gauge ﬁelds and fermions. the laws of physics are described by a real scalar ﬁeld. Let’s further suppose that down at the energy scale E. the irrelevant couplings become small at low-energies. Indeed. but typically one expects them to be of order 1: gn ∼ O(1). these lead to “non-renormalizable” ﬁeld theories in which one cannot make sense of the inﬁnities at arbitrarily high energies. (We’ve already seen a glimpse of this in computing the vacuum energy). Whatever this scale is. This doesn’t necessarily mean that the theory is useless. where the precise form is dictated by all the stuﬀ that’s going on in the high energy superduper theory. Such perturbations are called marginal. This is basically because. say the GUT scale. Let me note however that the naive assignment of relevant. or the Planck scale. let’s call it Λ. An Important Aside: Why QFT is Simple Typically in a quantum ﬁeld theory. Such perturbations are called irrelevant. only a handful are actually needed (just two in the case of the real scalar ﬁeld described above). as we’ve seen above.• [λ4 ] = 0: this term is small if λ4 ≪ 1. • [λn ] < 0 for n ≥ 5: The dimensionless parameter is (λn E n−4 ). λn = gn Λn−4 (3. multiplied by a suitable power of the relevant scale Λ. so [Λ] = 1.4) The exact values of dimensionless couplings gn depend on the details of the high-energy superduper theory. As you’ll see later. This scalar ﬁeld will have some complicated interaction terms (3. This means that we might expect problems with irrelevant operators. which is small at low-energies and large at high energies. Now we want to understand the laws of physics down at our puny energy scale E ≪ Λ. Suppose that we some day discover the true superduper “theory of everything unimportant” that describes the world at very high energy scales. It is an energy scale. we could write our dimensionful coupling constants λn in terms of dimensionless couplings gn . This is a huge help: of the inﬁnite number of interaction terms that we could write down. only the relevant and marginal couplings are important. just that it is incomplete at some energy scale. marginal and irrelevant is not always ﬁxed in stone: quantum corrections can sometimes change the character of an operator.

It also means that if we only have access to low-energy experiments (which we do!).5) 2 2 4! with λ ≪ 1. we see a sum of interactions that look like † † † † ap ap ap ap † † † and ap ap ap ap etc.6) These will create and destroy particles. This theory couples a complex scalar ψ to a real scalar φ. This means that the number of ψ particles minus the number of ψ anti-particles is conserved. This is usually a suppression by many orders of magnitude. H] = 0. about which you can learn more in the “Statistical Field Theory” course. Indeed.75) such that [Q. It is common practice to denote the anti¯ particle as ψ. Expanding † out φ4 in terms of ap and ap .g for the energies E explored at the LHC. we’ll look at two types of interactions 1) φ4 theory: 1 1 λ L = ∂µ φ∂ µ φ − m2 φ2 − φ4 (3.7) L = ∂µ ψ ⋆ ∂ µ ψ + ∂µ φ∂ µ φ − M 2 ψ ⋆ ψ − m2 φ2 − gψ ⋆ ψφ 2 2 with g ≪ M. based on dimensional analysis. ones that can truly be considered as small perturbations of the free ﬁeld theory at all energies. Examples of Weakly Coupled Theories In this course we’ll study only weakly coupled ﬁeld theories i. N] = 0. m. While the individual particle numbers of ψ and φ are no longer conserved. 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. (e. it’s going to be very diﬃcult to ﬁgure out the high energy theory (which it is!). In this section. 2) Scalar Yukawa Theory 1 1 (3. we could check that the number operator N now satisﬁes [H. We can get a hint for what the eﬀects of this extra term will be. we do still have a symmetry rotating the phase of ψ. (3. This suggests that the φ4 Lagrangian describes a theory in which particle number is not conserved. the interaction terms of the form φn with n > 4 will be suppressed by powers of (E/Λ)n−4 . The discussion given above is a poor man’s version of the ideas of eﬀective ﬁeld theory and Wilson’s renormalization group. – 49 – .e.means that for experiments at small energies E ≪ Λ. because its eﬀects are highly diluted except for the relevant and marginal interactions. It is this simple argument. E/Mpl ∼ 10−16 ).

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 –

of course. (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. Hint = g d3 x ψ † ψφ (3. 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. Before moving on.24) since t. To see why this is.22) where the range of integration in the ﬁrst expression is over t′′ ≥ t′ . t0 ) = 1 − i dt HI (t ) + (−i) t0 ′ ′ 2 t0 dt ′ t0 dt′′ HI (t′ )HI (t′′ ) + . The ﬁnal expression is the same as the second expression by a simple relabelling. this interaction doesn’t conserve particle number. The power of the formula comes from the expansion which is valid when HI is small and is very easily computed. This means that we can write t t t′ U(t. starting with the interaction Hamiltonian for our scalar Yukawa theory. is the latest time so HI (t) can be pulled out to the left.2 A First Look at Scattering Let us now apply the interaction picture to ﬁeld theory. we use – 53 – . 3. allowing particles of one type to morph into others. while in the second expression it is t′ ≤ t′′ which is. It is typically very hard to compute time ordered exponentials in practice. . I should confess that Dyson’s formula is rather formal. being the upper limit of the integral. the same thing. all operators commute (since their order is already ﬁxed by the T sign).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′ (3.25) Unlike the free theories discussed in Section 2. .

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

(Remember that the φ in this formula is in the interaction picture. a single particle.28) Let’s go slowly. the ﬁnal state contains a nucleon-anti-nucleon pair of momentum q1 and q2 .27) The initial state contains a single meson of momentum p. a long way from its neighbors. while the a† piece will turn |i into a two meson state. But the two meson state will have zero overlap with f |. It’s possible to circumvent this objection since it turns out that bound states show up as poles in the S-matrix. Nevertheless. |i = † 2Ep ap |0 |f = † † 4Eq1 Eq2 bq1 cq2 |0 (3. after developing scattering theory using the assumption of noninteracting asymptotic states. a related fact is that there is a cloud of virtual photons surrounding the electron.2. picking up a δ (3) (p − k) deltafunction which kills the d3 k integral. 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. and there’s nothing in the ψ and ψ † operators that lie between them to change this fact. We now similarly expand out ψ ∼ b + c† and – 55 – . we’ve commuted ak past ap . 3. it is f | S |i = −ig f | d4 x ψ † (x)ψ(x)φ(x) |i (3. in the second line.1 An Example: Meson Decay Consider the relativistically normalized initial and ﬁnal states. this formalism can’t describe the scattering of an electron and proton which collide.• Obviously we can’t cope with bound states.29) † where. In quantum electrodynamics (QED). bind.84). which is the same as the Heisenberg picture of the free theory). This line of thought gets us into the issues of renormalization — more on this next term in the “AQFT” course. To leading order in g. The a piece will turn |i into something proportional to |0 . and leave as a Hydrogen atom. where the electron sources the electromagnetic ﬁeld from which it can never escape. is never alone in ﬁeld theory. we’ll mention a better way. motivated by this problem. We may compute the amplitude for the decay of a meson to a nucleon-anti-nucleon pair. For example. This is true even in classical electrodynamics. • More importantly.

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

We use the notation. . . Then choosing x0 > y 0 .where the ± signs on φ± make little sense. leaving all other operators untouched.39) (3. ∆F (x − y). .35) (3. φ(x)φ(y) = ∆F (x − y) (3.33) = φ+ (x)φ+ (y) + φ− (x)φ+ (y) + φ− (y)φ+(x) + [φ+ (x). to denote contraction. and for our troubles we have picked up a factor of D(x − y) = [φ+ (x). . we have T φ(x)φ(y) = φ(x)φ(y) = (φ+ (x) + φ− (x))(φ+ (y) + φ− (y)) (3. we have the ﬁnal expression T φ(x)φ(y) =: φ(x)φ(y) : + ∆F (y − x) (3. for example. φ− (y)] which is the propagator we met in (2.90). . So for x0 > y 0 we have T φ(x)φ(y) =: φ(x)φ(y) : + D(x − y) Meanwhile. which is sometimes called the positive frequency piece. for y 0 > x0 . .34) where ∆F (x − y) is the Feynman propagator deﬁned in (2. while φ− ∼ e+iEt is the negative frequency piece). for which we have the integral representation ∆F (x − y) = ieik·(x−y) d4 k (2π)4 k 2 − m2 + iǫ (3. to mean replacing those operators with the Feynman propagator. . φ(x2 ) . . . φ(x1 ) . but apparently you have Pauli and Heisenberg to blame. Deﬁnition: We deﬁne the contraction of a pair of ﬁelds in a string of operators . φ(x1 ) . φ− (y)] + φ− (x)φ− (y) where the last line is normal ordered. (They come about because φ+ ∼ e−iEt .36) (3. . . φ(x2 ) . .37) Let me reiterate a comment from Section 2: although T φ(x)φ(y) and : φ(x)φ(y) : are both operators.38) – 57 – . we may repeat the calculation to ﬁnd T φ(x)φ(y) =: φ(x)φ(y) : + D(y − x) So putting this together. the diﬀerence between them is a c-number function. .93). So.

φn ) = (φ+ + φ− ) (: φ2 . We have the initial and ﬁnal states |i = |f = 2Ep1 2Ep′1 † † 2Ep2 bp1 bp2 |0 ≡ |p1 . we really want to calculate f | S − 1 |i since we’re not interested in situations where no scattering occurs. Then we can pull φ1 out to the left of the time ordered product. It’s true for n = 2. . etc. . . But in order to write 1 the right-hand side as a normal ordered product. For n = 4. Suppose it’s true for φ2 . In fact.2 Wick’s Theorem For any collection of ﬁelds φ1 = φ(x1 ). writing T (φ1 φ2 . p2 † † ′ ′ 2Ep2′ bp ′ bp ′ |0 ≡ |p1 .46) – 58 – . φn : + : all possible contractions : (3.A similar discussion holds for complex scalar ﬁelds. . let’s look at an example.42) ψ(x)ψ(y) = ψ † (x)ψ † (y) = 0 (3. n. We’ll take x0 > x0 1 k for all k = 2. . we have T (φ1 . .3 An Example: Nucleon Scattering Let’s look at ψψ → ψψ scattering.44) The φ− term stays where it is since it is already normal ordered. p2 1 2 (3. φ2 = φ(x2 ).43) Proof: The proof of Wick’s theorem proceeds by induction and a little thought. 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 ) (3.3. . . φn and now add φ1 . . .3. Each time it moves past φ− . the φ+ term has to make its way 1 past the crowd of φ− operators. (Try it!) 3.45) We can then look at the expansion of f | S |i . . . . 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 3.40) To see what the last part of this equation means. φn : + : contractions :) 1 1 (3. φn ) =: φ1 . we pick up a factor of k k φ1 φk = ∆F (x1 − xk ) from the commutator. .41) (3. 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.

48). while the two ψ † ﬁelds create ψ particles. p2 ′ ′ ′ ′ = p′1 ..48) = eix1 ·(p1 −p1 )+ix2 ·(p2 −p2 ) + eix1 ·(p2 −p1 )+ix2 ·(p1 −p2 ) + (x1 ↔ x2 ) where. To see this.. (3.51) Finally. Any other way of ordering the ψ and ψ † ﬁelds will give zero contribution. (−ig)2 2 d x1 d x2 e 4 4 i.47) which will contribute to the scattering because the two ψ ﬁelds annihilate the ψ particles. This means that we have p′1 . 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. 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 ) (3.49) +e i. by – 59 – . Meanwhile.. to get the expression for f | S |i at order g 2 . while the ﬁnal integral is the φ propagator which comes from the contraction in (3. in going to the third line. the x1 and x2 integrals give delta-functions. p′2 | ψ † (x1 )ψ † (x2 ) |0 0| ψ(x1 )ψ(x2 ) |p1 . for this process we may drop the +iǫ terms since the denominator is never zero. 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..47). where p1 = −p2 and.46).50) where the expression in square brackets is (3. + (x1 ↔ x2 ) d4 k ieik·(x1 −x2 ) (2π)4 k 2 − m2 + iǫ (3. we’ve used the fact that for relativistically normalized states. 0| ψ(x) |p = e−ip·x Now let’s insert this into (3. p′2 | : ψ † (x1 )ψ(x1 )ψ † (x2 )ψ(x2 ) : |p1 . Now the (x1 ↔ x2 ) terms double up with the others to cancel the factor of 1/2 out front. p2 = eip1 ·x1 +ip2 ·x2 + eip1 ·x2 +ip2 ·x1 ′ ′ ′ ′ e−ip1 ·x1 −ip2 ·x2 + e−ip1 ·x2 −ip2 ·x1 (3. we can go to the center of mass frame.

Each of these comes from the term (3. add an arrow to – 60 – . We’ll choose dotted lines for mesons. for this would violate the conservation of Q charge. These pictures represent the expansion of f | S |i and we will learn how to associate numbers (or at least integrals) to them. we would need to pick out the term : ψ † (x1 )φ(x1 )ψ(x2 )φ(x2 ) : ψ(x1 )ψ † (x2 ) (3. including ψφ → ψφ and φφ → ψ ψ. since we’re not interested in processes where no scattering occurs. such as ψ ψ → ψ ψ. However. These pictures are called Feynman diagrams.momentum conservation. we ¯¯ will never ﬁnd a term that contributes to scattering ψψ → ψ ψ. much simpler way to reproduce this result shortly using Feynman diagrams. This ensures that the 4-momentum of the meson is k = (0. This will also shed light on the physical interpretation. It requires drawing pretty pictures.” Julian Schwinger As the above example demonstrates. 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. so k 2 < 0. Assign a directed momentum p to each line. Once more.53) and a similar term with ψ and ψ † exchanged. |p1 | = |p1′ |.4 Feynman Diagrams “Like the silicon chips of more recent years. p − p ′ ). to actually compute scattering amplitudes using Wick’s theorem is rather tedious. the Feynman diagram was bringing computation to the masses. Further. this term also contributes to ¯ ¯ ¯ similar scattering processes. 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 therefore have the end result.52) We will see another. There’s a much better way. ¯¯ ¯¯ This calculation is also relevant for other scattering processes. and solid lines for nucleons. Another Example: Meson-Nucleon Scattering If we want to compute ψφ → ψφ scattering at order g 2. ¯ ¯ ψ ψ → ψ ψ. 3.48) in Wick’s theorem. The object that we really want to compute is f | S −1 |i .

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

There are also more complicated diagrams which will contribute to the scattering process at higher orders. Heuristically. – 62 – . rather loosely. nucleon legs all satisfy p2 = M 2 . We talk. And so we go on. Using the Feynman rules. 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. we have the two diagrams shown in Figures 10 and 11.3. each of these diagrams translates into an integral that we will not attempt to calculate here. We can write down the two simplest diagrams contributing to this process. There is a nice physical interpretation of these diagrams. oﬀ massshell). Applying the Feynman rules to these diagrams. of the nucleons exchanging a meson which. the momentum on the external. One ﬁnal note: the addition of the two diagrams above ensures that the particles satisfy Bose statistics. and similar diagrams with p′1 and p′2 exchanged. They are on-shell. In contrast.51) that we performed earlier. We start with something familiar: Nucleon Scattering Revisited Let’s look at how this works for the ψψ → ψψ scattering at order g 2 . sometimes. This meson doesn’t satisfy the usual energy dispersion relation. 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. all appearing at higher order in the coupling constant g. They are shown in Figure 9. it can’t live long enough for its energy to be measured to great accuracy. the mass of the nucleon. because k 2 = m2 : the meson is called a virtual particle and is said to be oﬀ-shell (or. in the ﬁrst diagram.5 Examples of Scattering Amplitudes Let’s apply the Feynman rules to compute the amplitudes for various processes. with increasingly complicated diagrams.56) which agrees with the calculation (3. has momentum k = (p1 −p′1 ) = (p2 −p′2 ). For example.

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

¯ ¯ For the scattering of a nucleon and an anti-nucleon.59) – 64 – . 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. 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. It is a simple matter to write down the amplitude using the Feynman rules. they are given by the diagrams of Figure 13. At lowest order.58) p 1 1 p1 / + p2 p2 p2 / / p2 Figure 13: The two lowest order Feynman diagrams for nucleon-anti-nucleon scattering. we’ve dropped the iǫ from the propagators as the denominator never vanishes.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.52). the Feynman diagrams are a little diﬀerent. Nucleon-Anti-Nucleon Scattering p p1 / − M2 + i (p1 − ′ p 2 )2 − M2 (3. ψ ψ → ψ ψ. 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. iA = (−ig)2 i (p1 − ′ p 1 )2 − m2 + (p1 + p2 )2 i − m2 + iǫ (3.

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

p sin θ. For the nucleon-anti-nucleon scattering. (The ﬁrst diagram indeed includes a vertex that looks like the letter “T”). 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. so that the initial two particles have four-momenta p1 = (E. (They are basically equivalent. 0. we have that s = 4E 2 and t = −2p2 (1 − cos θ) and u = −2p2 (1 + cos θ) (3. ′ and p1 and p′2 are the momenta of the ﬁnal two particles.3. p1 and p2 are the momenta of the two initial particles. while the variables t and u are measures of the momentum exchanged between particles. To get a feel for what these variables mean. We sit in the center of mass frame. 0. just with the outgoing particles swapped around). let’s assume all four particles are the same. −p sin θ. – 66 – .61) The variable s measures the total center of mass energy of the collision. Note that there is a relationship between the Mandelstam variables. 0. Meanwhile the nucleon-antinucleon scattering is said to involve “t-channel” and “s-channel” diagrams. We can deﬁne these variables whether the particles involved in the scattering are the same or diﬀerent. −p) The particles then scatter at some angle θ and leave with momenta p′1 = (E. 0.60) p′1 )2 where. p cos θ) and p′2 = (E. When the masses of all 4 particles diﬀer. for nucleon-nucleon scattering. −p cos θ) Then from the above deﬁnitions. p) and p2 = (E. this becomes s + t + u = i Mi2 .62) (3.59) is A ∼ (t − m2 )−1 + (s − m2 )−1 . Now the amplitudes that involve exchange of a single particle can be written simply in terms of the Mandelstam variables. There are standard names for various sums and diﬀerences of momenta: they are known as Mandelstam variables.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. the amplitude (3. 0.63) (3.5. When all the masses are the same we have s + t + u = 4M 2 . We say that the ﬁrst case involves “t-channel” and “u-channel” diagrams. as in the examples above. the amplitude (3. 0.56) is schematically A ∼ (t − m2 )−1 + (u − m2 )−1 . For example.

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 –

3. p2 | : φ(x)φ(x)φ(x)φ(x) : |p1 . we can look at φφ → φφ scattering. where every part of the diagram is connected to at least one external line. the scattering amplitude for φφ → φφ is Figure 16: simply iA = −iλ. 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.75) Any one of the ﬁelds can do the job of annihilation or creation. 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. As we shall see shortly. 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. To see why this is. – 70 – . which cancels the 1/4! sitting out front. is not strictly accurate. The two caveats which go some way towards ameliorating the problem are the following • We consider only connected Feynman diagrams.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. −iλ Using the Feynman rules. This gives 4! diﬀerent contractions.76) So scattering in φ4 theory is due to a δ-function potential. In fact. we have U(r) = λ (2m)2 d3 p +ip·r λ e = δ (3) (r) 3 (2π) (2m)2 (3. An example of a diagram that is not connected is shown in Figure 17. The particles don’t know what hit them until it’s over.vertex rather than (−iλ/4!). see the book by Peskin and Schroeder. For more details. as we mentioned at the time. with the term −iλ ′ ′ p1 . Both of these caveats are related to the assumption we made earlier that “initial and ﬁnal states are eigenstates of the free theory” which. which has its lowest contribution at order λ. Translating this into a potential between two mesons. there are a couple of caveats about what Feynman diagrams you should write down. p2 4! (3.5.

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

81) Now when squaring the amplitude to get the probability. the function in the ﬁgure starts to approach a deltafunction. we run into the problem of the square of the delta-function: Pn→m = | m| Hint |n |2 (2π)2 δ(ω)2 . In the limit 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). We see that in time t. This gives us the probability for the transition from |n to |m in time t. Suppose we were a little sloppier. we realize that the extra inﬁnity is coming because Pm→n – 72 – . we were fairly careful with taking the limit as t → ∞. In the above derivation.where ω = Em − En . Tracking through the previous computations. we can calculate +∞ 1 − cos ωt ω2 (3.79) → 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. most transitions happen in a region between energy eigenstates separated by ∆E = 2π/t. we get the transition probability Pn→m = dEm ρ(Em ) 2| m| Hint |n | 2 1 − cos ωt ω2 (3.80) −i m| t=−∞ HI (t) |n = −i m| Hint |n 2πδ(ω) (3. As t → ∞. To ﬁnd the normalization. ˙ Pn→m = 2π| m| Hint |n |2 ρ(E) This is Fermi’s Golden Rule. 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. and ﬁrst chose to compute the amplitude for the state |n at t → −∞ to transition to the state |m at t → +∞. Then we get t=+∞ (3.

i| i = (2π)3 2EpI δ (3) (0) = 2EpI V where we have replaced δ (3) (0) by the volume of 3-space. we get the probability for decay P = |Af i |2 (2π)4 δ (4) (pI − pF ) V T 2mV 1 2Epi V states (3.87) ﬁnal where. and the square of the δ (4) -functions will just be re-interpreted as spacetime volume factors.30). in our ﬁeld theory calculations.6. (For example. Similarly.85) (3. of course. after integrating over the density of ﬁnal states. exactly what we’ve been computing. in (3.25)).65). ˙ Pn→m = 2π| m| Hint |n |2 δ(ω) (3. The reason that we’ve stressed this point is because.84) Our states obey the relativistic normalization formula (2.is the probability for the transition to happen in inﬁnite time t → ∞. This is given by P = | f | S |i |2 f | f i| i (3. as in the second derivation of Fermi’s Golden Rule. We now divide out by this power of T to get the transition probability per unit time. we’ve computed the amplitudes in the same way as (3. so pI = 0 and EpI = m. We can write the delta-functions as (2π)2 δ(ω)2 = (2π)δ(ω) T (3.2 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 .81). f| f = 2Epi V ﬁnal states (3. The amplitudes Af i are. gives us back Fermi’s Golden rule.82) where T is shorthand for t → ∞ (we used a very similar trick when looking at the vacuum energy in (2. we’ve exchanged one of the delta-functions for the volume of spacetime: (2π)4 δ (4) (0) = V T .86) If we place our initial particle at rest. 3. we saw – 73 – .83) which.

We can now divide out by T to get the transition function per unit time.88) ﬁnal The second step is to sum over all ﬁnal states with diﬀerent numbers (and possibly types) of particles.87). We’ve been considering just a single particle per spatial volume V .6.3 Cross Sections Collide two beams of particles.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) (pF − pI ) d 3 pi 1 (2π)3 2Epi states (3. 3. φ). If the incoming ﬂux F is deﬁned to be the number of incoming particles per area per unit time. The fraction of the time that they collide is called the cross section and is denoted by σ. More precisely dσ = 1 1 Diﬀerential Probability = |Af i |2 dΠ Unit Time × Unit Flux 4E1 E2 V F (3. Γ= 1 2m ﬁnal |Af i |2 dΠ (3. – 74 – .89) states Γ is called the width of the particle. It is equal to the reciprocal of the half-life τ = 1/Γ.90) We would like to calculate σ from quantum ﬁeld theory. sometimes they will pass right through. then the total number of scattering events N per unit time is given by. We now need an expression for the unit ﬂux. The factors of spatial volume V in this measure cancel those in (3. while the factors of 1/2Epi in (3. In fact. Γ = P . let’s sit in the center of mass frame of the collision. There are two steps: the ﬁrst is to integrate over all possible momenta of the ﬁnal particles: V d3 pi /(2π)3 . Sometimes the particles will hit and bounce oﬀ each other. 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. For simplicity.that A = −g for a single meson decaying into two nucleons).91) where we’ve used the expression for probability per unit time that we computed in the previous subsection. This gives us our ﬁnal expression for the decay probability per unit ˙ time. N = Fσ (3. But we still have to worry about summing over all ﬁnal states.

For example. Notice that diﬀerent scattering amplitudes have diﬀerent momentum dependence and will result in diﬀerent angular dependence in scattering amplitudes. This results in isotropic scattering. 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. We’ll leave the relationship to other physical phenomena to other courses. You may now take your favorite scattering amplitude. This will involve doing the integral over the phase space of ﬁnal states. dσ = 1 1 |Af i |2 dΠ 4E1 E2 |v1 − v2 | (3. with measure dΠ. 3.93) – 75 – . We’ll denote the true vacuum of the interacting theory as |Ω . In contrast. and compute the probability for particles to ﬂy out at your favorite angles. in φ4 theory the amplitude for tree level scattering was simply A = −λ. For example. t = −2|p|2 (1 − cos θ). Equation (3. This then gives.92) is our ﬁnal expression relating the S-matrix to the diﬀerential cross section. In this section we will brieﬂy deﬁne correlation functions. This gives rise to angular dependence in the diﬀerential cross-section.92) If you want to write this in terms of momentum. and then relate them back to scattering amplitudes. which follows from the fact that.7 Green’s Functions So far we’ve learnt to compute scattering amplitudes.meaning that the ﬂux is given in terms of the 3-velocities vi as F = |v1 − v2 |/V . for nucleon-nucleon scattering we have schematically A ∼ (t − m2 )−1 + (u − m2 )−1 . we might want to compute the viscosity of the quark gluon plasma. or the optical conductivity in a tentative model of strange metals. We’ll normalize H such that H |Ω = 0 (3. 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 . 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. where θ is the angle between the incoming and outgoing particles. for example. explain how to compute them using Feynman diagrams.

φH (xn ) |Ω (3. −∞) |0 (3. . Then we can drop the T and write the numerator of the right-hand side as 0| UI (+∞. > tn . the vacuum of the free theory.96) where U(t.and Ω| Ω = 1. Ψ| U(t. Consider an arbitrary state |Ψ Ψ| UI (t. Then G(n) (x1 . 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. tk+1) = T exp(−i tkk+1 HI ) to convert φI into φH . .97) – 76 – . or Green’s functions. Deﬁne G(n) (x1 . −∞) |0 = Ψ| U(t.94) where φH is φ in the Heisenberg picture of the full theory. Now let’s deal with the two remaining U(t. . φnH |Ω = 0| T φ1I . xn ) = Ω| T φ1H . . . . Proof: Take t1 > t2 > . φnH UI (tn . −∞) |0 = Ψ| U(t. −∞) |Ω Ω| + = Ψ| Ω Ω| 0 + ′ lim t →−∞ n=0 |n n| |0 ′ eiEn (t −t) Ψ| n n| 0 n=0 (3. ±∞) at either end of the string of operators. . . xn ) = Ω| T φH (x1 ) . . . . which we take to be energy eigenstates of H = H0 + Hint . t1 )φ1H . and the equality above follows o because H0 |0 = 0. . −∞) |0 t where we’ve used all the intermediate factors of UI (tk . −∞) |0 = 0| UI (+∞. t1 )φ1I U(t1 . Now insert a complete set of states. The G(n) are called correlation functions. φnI UI (tn . . . and write φ1H to denote the ﬁeld in the Heisenberg picture. . . φnI S |0 0| S |0 (3. t2 ) φ2I . We prove the following result Claim: We use the notation φ1 = φ(x1 ). There are a number of diﬀerent ways of looking at these objects which tie together nicely. . −∞) is the Schr¨dinger evolution operator. Let’s start by asking how to compute G(n) using Feynman diagrams. . rather than the interaction picture that we’ve been dealing with so far. and φ1I to denote the ﬁeld in the interaction picture. . .95) where the operators on the right-hand side are evaluated on |0 .

The combinatoric factors (as well as the symmetry factors) associated with each diagram are such that the whole series sums – 77 – . We now apply the formula (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 (3.98) Why is this relevant? The point is that the n in (3. . so that it sits on its own..99) (Notice that to derive this result.7. t′ ) |0 = Ψ| Ω Ω| 0 (3. away from the continuum of the integral. namely Dyson’s formula and Wick’s theorem or.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 ) . completing the proof. φnH |Ω Ω| 0 0| Ω Ω| Ω Ω| 0 (3. (3. using the normalization Ω| Ω = 1. This means that we must be working in a theory with a mass gap – i. it has a simple interpretation. with no massless particles). Peskin and Schroeder instead send t → −∞ in a slightly imaginary direction. we will represent the φ particles as solid lines.97) is really an integral dn. . gives us the left-hand side. Since there is no ambiguity in the diﬀerent types of line in Feynman diagrams. But what about dividing one by the other? What’s that all about? In fact. to the top and bottom of the right-hand side of (3. Then we have the diagramatic expansion for 0| S |0 . For the following discussion.e. alternatively.. .But the last term vanishes. . which also does the job). .95) to ﬁnd 0| Ω Ω| T φ1H . Feynman diagrams. because all states are part of a continuum due to the momentum.99). 0| S |0 = 1 + + ( + + ) + .101) These diagrams are called vacuum bubbles. we will work in φ4 theory. So the Riemann-Lebesgue lemma gives us t′ →−∞ lim Ψ| U(t.100) which. φI (xn ) S |0 and 0| S |0 using the same methods we developed for S-matrix elements. 3. (There is a caveat here: we want the vacuum |Ω to be special. rather than the dashed lines that we used previously.

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

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

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

3) where the matrices D[Λ] form a representation of the Lorentz group. In general. the ﬁeld transforms as ν µ φ(x) → φ′ (x) = φ(Λ−1 x) (4. we look at inﬁnitesimal transformations of the Lorentz group and study the resulting Lie algebra.4) and D[Λ−1 ] = D[Λ]−1 and D[1] = 1. whose quantization gives rise to fermionic spin 1/2 particles. meaning that (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. we will start by studying the appropriate representation of the Lorentz group. Aµ (x) → Λµν Aν (Λ−1 x) (4.4. To motivate the Dirac equation. or spin.” Weisskopf on Dirac So far we’ve only discussed scalar ﬁelds such that under a Lorentz transformation x → (x′ )µ = Λµ xν .6) – 81 – . In this section we will describe the Dirac equation. It should be added.2) We’ll deal with this in Section 6. (It comes with its own problems!). however.5) for inﬁnitesimal ω. Dirac himself remarked in one of his talks that his equation was more intelligent than its author. If we write.1) We have seen that quantization of such ﬁelds gives rise to spin 0 particles. A familiar example of a ﬁeld which transforms non-trivially under the Lorentz group is the vector ﬁeld Aµ (x) of electromagnetism. a ﬁeld can transform as φa (x) → D[Λ]ab φb (Λ−1x) D[Λ1 ]D[Λ2 ] = D[Λ1 Λ2 ] (4. 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. These arise naturally in ﬁeld theory by considering ﬁelds which themselves transform non-trivially under the Lorentz group. Λµν = δ µν + ω µν (4. But most particles in Nature have an intrinsic angular momentum. How do we ﬁnd the diﬀerent representations? Typically. that it was Dirac who found most of the additional insights.

after all. M01 = −M10 . with A = 1. Mτ ν ] = η στ Mρν − η ρτ Mσν + η ρν Mστ − η σν Mρτ (4. antisymmetric matrices. for example. so that ρ and σ again label six diﬀerent matrices. 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]. the matrices are no longer necessarily antisymmetric. so we call our matrices (Mρσ )µν . M01 . We could call them (MA )µν . we pick up various minus signs which means that when written in this form. [Mρσ . . generates rotations in the (x1 . .9) The ﬁrst. The second. .11) – 82 – . We can now write any ω µ as a linear combination of the Mρσ . It’s going to be useful to introduce a basis of these six 4 × 4 anti-symmetric matrices. The antisymmetry on the ρ and σ indices means that. Of course. 6. while ρ and σ denote which basis element we’re dealing with.7) where the indices µ and ν are those of the 4 × 4 matrix. or act on some ﬁeld) we will typically need to lower one index. If we use these matrices for anything practical (for example. . . which agrees with the 6 transformations of the Lorentz group: 3 rotations and 3 boosts. σ = 0. if we want to multiply them together. It is real and antisymmetric. etc. .10) where Ωρσ are just six numbers (again antisymmetric in the indices) that tell us what Lorentz transformation we’re doing. 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. 3. It is real and symmetric. generates boosts in the x1 direction. we can write a basis of six 4 × 4 antisymmetric matrices as (Mρσ )µν = η ρµ η σν − η σµ η ρν (4. x2 )-plane. where ρ. The six basis matrices Mρσ are called the generators of the Lorentz transformations. ν 1 ω µν = 2 Ωρσ (Mρσ )µν (4.Note that an antisymmetric 4 × 4 matrix has 4 × 3/2 = 6 independent components. M12 . . Two examples of these basis matrices are. the matrices are also antisymmetric on the µν indices because they are. The generators obey the Lorentz Lie algebra relations. . so we have (Mρσ )µν = η ρµ δ σν − η σµ δ ρν (4.8) Since we lowered the index with the Minkowski metric.

where we have suppressed the matrix indices. σ2 = 0 −i i 0 . σ3 = 1 0 0 −1 (4. The simplest representation of the Cliﬀord algebra is in terms of 4 × 4 matrices.13) where γ µ . we may take γ0 = 0 1 1 0 . 4.11). they say things like rotate by θ = π/7 about the x3 -direction and run at speed v = 0. We will construct the spinor representation.17) which themselves satisfy {σ i . There are many such examples of 4 × 4 matrices which obey (4. To do this. A ﬁnite Lorentz transformation can then be expressed as the exponential Λ = exp 1 2 Ωρσ Mρσ (4. γ ν } ≡ γ µ γ ν + γ ν γ µ = 2η µν 1 (4. 3 (4. 1. 2.15) when µ = ν (4. we start by deﬁning something which. at ﬁrst sight. γi = 0 σi −σ i 0 (4. 2. – 83 – . the Ωρσ are six numbers telling us what kind of Lorentz transformation we’re doing (for example.12) Let me stress again what each of these objects are: the Mρσ are six 4 × 4 basis elements of the Lorentz group. 3.13). It is the Cliﬀord algebra. has nothing to do with the Lorentz group. are a set of four matrices and the 1 on the right-hand side denotes the unit matrix. with the σ i the Pauli matrices σ1 = 0 1 1 0 . with µ = 0. (γ i )2 = −1 i = 1. This means that we must ﬁnd four matrices such that γ µ γ ν = −γ ν γ µ and (γ 0 )2 = 1 . {γ µ . For example.14) It’s not hard to convince yourself that there are no representations of the Cliﬀord algebra using 2 × 2 or 3 × 3 matrices.2 in the x1 direction).16) where each element is itself a 2 × 2 matrix.1 The Spinor Representation We’re interested in ﬁnding other matrices which satisfy the Lorentz algebra commutation relations (4. σ j } = 2δ ij .

S ρσ = 1 ρ σ [γ . γ ρ ] = 2 [γ µ γ ν . γ ρ γ σ ] 1 = 1 [S µν .18) Let’s see what properties these matrices have: Claim 4. 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 (4.1 above. The matrices (4. and using Claim 4. γ ρ ] = γ µ η νρ − γ ν η ρµ Proof: When µ = ν we have 1 [S µν . γ µ }γ ν + 2 γ µ γ ρ γ ν Claim 4. it turns out that there is a unique irreducible representation of the Cliﬀord algebra. S ρσ ] = 2 [S µν . (4.18) to write γ µ γ σ = 2S µσ + η µσ . γ ρ } − 2 γ µ γ ρ γ ν − 2 {γ ρ . So what does the Cliﬀord algebra have to do with the Lorentz group? Consider the commutator of two γ µ . we have [S µν .1: [S µν . up to this equivalence.2: The matrices S µν form a representation of the Lorentz algebra (4. known as the Weyl or chiral representation (for reasons that will soon become clear). we have 1 [S µν .11). However.16) provide one example. γ ρ ]γ σ + 2 γ ρ [S µν .21) – 84 – . and consider only representations of the Cliﬀord algebra that are related to the chiral representation by a unitary transformation V . S ρσ ] = S µσ η νρ − S νσ η ρµ + S ρµ η νσ − S ρν η σµ which is our desired expression. γ ] = 4 0 1 2 ρ=σ ρ=σ γ γ ρ σ 1 1 = 2 γ ρ γ σ − η ρσ 2 (4. meaning [S µν . We will soon restrict ourselves further. 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 .20) Now using the expression (4.One can construct many other representations of the Cliﬀord algebra by taking V γ µ V −1 for any invertible matrix V .19) (4. S ρσ ] = η νρ S µσ − η µρ S νσ + η µσ S νρ − η νσ S µρ Proof: Taking ρ = σ.

the x3 -axis. 4. So how can we be sure that the spinor representation is something new. This is achieved by ϕ = (0. We introduce the Dirac spinor ﬁeld ψ α (x). 2π). because the γ µ are 4 × 4 matrices.25) If we write the rotation parameters as Ωij = −ǫijk ϕk (meaning Ω12 = −ϕ3 . let’s look at some speciﬁc transformations. and isn’t equivalent to the familiar representation Λµν ? To see that the two representations are truly diﬀerent. To avoid confusion. Rotations S ij = 1 2 0 σi −σ i 0 0 σj −σ j 0 i = − ǫijk 2 σk 0 0 σk (4.22) Ωρσ Mρσ Ωρσ S ρσ (4. the latter will always come with the square brackets [Λ]. So far we haven’t given an index name to the rows and columns of these matrices: we’re going to call them α. We need a ﬁeld for the matrices (S µν )αβ to act upon. Under Lorentz transformations. Note that we denote both the generator S ρσ and the full Lorentz transformation S[Λ] as “S”. Consider now a rotation by 2π about. 4. an object with four complex components labelled by α = 1. β = 1.1 Spinors The S µν are 4 × 4 matrices.4. – 85 – .23) (4. we have ψ α (x) → S[Λ]αβ ψ β (Λ−1 x) where Λ = exp S[Λ] = exp 1 2 1 2 (4. 3. 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. 3. 2. 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.26) where we need to remember that Ω12 = −Ω21 = −ϕ3 when following factors of 2. Both Λ and S[Λ] are 4 × 4 matrices.24) Although the basis of generators Mρσ and S ρσ are diﬀerent. 2.1. 0. say.

31). Recall that S[Λ] = exp 1 2 Ωρσ S ρσ (4. S[Λ] is unitary. let’s see how a vector would transform under a rotation by ϕ = (0. (S µν )† = −S µν . We have demonstrated this explicitly for the spinor representation using the chiral representation (4.27) which is deﬁnitely not what happens to a vector! To check that we haven’t been cheating with factors of 2. We can get a feel for why it is true for a spinor representation constructed from any representation of the Cliﬀord algebra. In fact. So S[Λ] is deﬁnitely a diﬀerent representation from the familiar vector representation Λµν . we learn that Λ = 1 as you would expect.e.16) of the Cliﬀord algebra. i. 0. S[Λ] is not unitary. (γ ν )† ] (4.and the spinor rotation matrix becomes.28) e+iπσ 0 3 0 e+iπσ 3 = −1 (4.33) 4 – 86 – . S[Λ] = Therefore under a 2π rotation ψ α (x) → −ψ α (x) (4.26).31) Representations of the Lorentz Group are not Unitary Note that for rotations given in (4. satisfying S[Λ]† S[Λ] = 1. 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. we have S[Λ] = 0 e−χ·σ/2 (4. But we have 1 (S µν )† = − [(γ µ )† . But for boosts given in (4.32) so the representation is unitary if S µν are anti-hermitian. 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 (4.29) So when we rotate a vector by ϕ3 = 2π. there are no ﬁnite dimensional unitary representations of the Lorentz group. ϕ3).30) Writing the boost parameter as Ωi0 = −Ω0i = χi .

the matrices have this property: (γ 0 )† = γ 0 and (γ i )† = −γ i . But now we see why it fails. in turn. the Dirac spinor ψ. Let’s pick a representation of the Cliﬀord algebra which. Deﬁne ψ † (x) = (ψ ⋆ )T (x) (4. We would like to construct a Lorentz invariant equation of motion.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. 3 we have γ 0 γ µ γ 0 = (γ µ )† which. means that 1 (S µν )† = [(γ ν )† . for some Lorentz transformation S[Λ]† S[Λ] = 1 since the representation is not unitary. Then for all µ = 0. but we can only pick γ i to be anti-hermitian. Indeed. 1. we can never arrange for this to happen since (γ 0 )2 = 1 ⇒ Real Eigenvalues (γ i )2 = −1 ⇒ Imaginary Eigenvalues (4.38) (4.16). and certainly isn’t a scalar.which can be anti-hermitian if all γ µ are hermitian or all are anti-hermitian.36) So ψ † (x)ψ(x) → ψ † (Λ−1 x)S[Λ]† S[Λ]ψ(Λ−1 x). as we have seen.37) – 87 – .34) So we could pick γ 0 to be hermitian. in the chiral representation (4. In general there is no way to pick γ µ such that S µν are anti-hermitian.2 Constructing an Action We now have a new ﬁeld to work with. (γ µ )† ] = −γ 0 S µν γ 0 4 (4. This means that ψ † ψ isn’t going to do it for us: it doesn’t have any nice transformation under the Lorentz group. satisﬁes (γ 0 )† = γ 0 and (γ i )† = −γ i . 2. But. 4. but will give us a clue how to proceed.16). We do this by constructing a Lorentz invariant action. like the chiral representation (4. with the spinor indices summed over. We will start in a naive way which won’t work. we can also see how to proceed. ψ(x) → S[Λ] ψ(Λ1x) ψ † (x) → ψ † (Λ−1 x) S[Λ]† (4. However. Let’s see how this transforms under Lorentz transformations.

42) This equation means that we can treat the µ = 0. under a Lorentz transformation we have. .40) Let’s now see what Lorentz covariant objects we can form out of a Dirac spinor ψ and ¯ its adjoint ψ.so that S[Λ]† = exp 1 2 Ωρσ (S ρσ )† = γ 0 S[Λ]−1 γ 0 (4.46) Ωρσ S ρσ ≈ 1 + 1 Ωρσ S ρσ + . 1. Proof: Suppressing the x argument. 3 index on the γ µ matrices as a true vector index.45) (4. so that Λ = exp S[Λ] = exp 1 2 1 2 1 Ωρσ Mρσ ≈ 1 + 2 Ωρσ Mρσ + . . 2 – 88 – . We work inﬁnitesimally. .4: ψ γ µ ψ is a Lorentz vector.3: ψψ is a Lorentz scalar. ¯ ¯ ψ γ µ ψ → ψ S[Λ]−1 γ µ S[Λ]ψ ¯ If ψ γ µ ψ is to transform as a vector. ¯ 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. Proof: Under a Lorentz transformation.39) With this in mind. ¯ Claim 4.43) (4. In particular we can form Lorentz scalars by contracting it with other Lorentz indices. we now deﬁne the Dirac adjoint ¯ ψ(x) = ψ † (x) γ 0 (4. . (4.41) which is indeed the transformation law for a Lorentz scalar. which means that ¯ ¯ ψ(x) γ µ ψ(x) → Λµν ψ(Λ−1 x) γ ν ψ(Λ−1 x) (4. 2. we must have S[Λ]−1 γ µ S[Λ] = Λµν γ ν We’ll now show this.44) (4. ¯ Claim 4.

it’s there for the same reason that the Hermitian momentum operator −i∇ in quantum mechanics has a factor i).49) This is the Dirac action. upon complex conjugation.1 where we showed that [S ρσ . which can be positive or negative.47) where we’ve suppressed the α. We can try to build a Lorentz invariant action from these. β indices on γ µ and S µν . each of which transforms covariantly under the Lorentz group. Also. we write the right-hand side of (4. In fact equation (4. ψγ µ ψ and ψγ µ γ ν ψ.50) (4. proportional to η µν ψψ. ψψ. ¯ Claim 4. To see this. γ µ ] = γ ρ η σµ − γ σ η µρ . As we will see in the next section. the symmet¯ ric part is a Lorentz scalar. The factor of “i” is there to make the action real. Notice that the Lagrangian is ﬁrst order.48) which means that the proof follows if we can show −[S ρσ .47) by expanding out M. – 89 – . γ µ ] = (Mρσ )µν γ ν (4.5: ψγ µ γ ν ψ transforms as a Lorentz tensor. More precisely. In fact. (Said another way. we need only the ﬁrst two. 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. We choose S= ¯ d4 x ψ(x) (iγ µ ∂µ − m) ψ(x) (4. but otherwise left all other indices explicit. γ µ ] = η ρµ γ σ − η σµ γ ρ which is exactly what we proved in Claim 4. the mass appears in the Lagrangian as m. Proof: As above. rather than the second order Lagrangians we were working with for scalar ﬁelds. while the antisymmetric part is a ¯ Lorentz tensor. it cancels a minus sign that comes from integration by parts.47) follows from Claim 4. proportional to ψS µν ψ.1.so the requirement (4.44) becomes −[S ρσ . ¯ ¯ ¯ We are now armed with three bilinears of the Dirac ﬁeld.

and so applies to each component ψ α . for spinor ﬁelds. 3. 4. yet miraculously Lorentz invariant.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 . Varying with respect to ψ. We write / Aµ γ µ ≡ A so the Dirac equation reads / (i ∂ − m)ψ = 0 (4. However. However.52) The Dirac equation is ﬁrst order in derivatives. so we get 2 −(∂µ ∂ µ + m2 )ψ = 0 (4. Varying with respect to ψ gives the conjugate equation ¯ ¯ i∂µ ψ γ µ + mψ = 0 (4.51) This is the Dirac equation. If we tried to write down a ﬁrst order equation of motion for a scalar ﬁeld. each individual component itself solves the Klein-Gordon equation. we have (iγ µ ∂µ − m) ψ = 0 (4.. 2.54) (4. We will often come across 4-vectors contracted with γ µ matrices. The Dirac equation mixes up diﬀerent components of ψ through the matrices γ µ .50) by varying with respect to ψ and ¯ ¯ ψ independently. To see this.55) – 90 – . which necessarily includes a privileged vector in spacetime v µ and is not Lorentz invariant. It’s completely gorgeous. . with α = 1. The Slash Let’s introduce some useful notation.3 The Dirac Equation The equation of motion follows from the action (4. γ ν }∂µ ∂ν = ∂µ ∂ µ . it would look like v µ ∂µ φ = . the magic of the γ µ matrices means that the Dirac Lagrangian is Lorentz invariant.4.53) where this last equation has no γ µ matrices.

u± → eiϕ·σ/2 u± but oppositely under boosts.4. u+ is in the ( 2 . we’ve used the chiral representation γ0 = 01 10 .26) and (4. are deﬁned by ψ= u+ u− (4.31). 0) ⊕ (0. 4.60) 1 In group theory language.59) The two-component objects u± are called Weyl spinors or chiral spinors. the spinor rotation transformation S[Λrot ] and boost transformation S[Λboost ] were computed in (4. acting only on two-component spinors u± which. γi = 0 σi −σ i 0 (4. 1 1 while u− is in the (0.4 Chiral Spinors When we’ve needed an explicit form of the γ µ matrices. 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 Dirac spinor ψ lies in the ( 2 . 2 ) 2 representation. u± → e±χ·σ/2 u± (4. in the chiral representation.62) – 91 – . They transform in the same way under rotations.58) This means that the Dirac spinor representation of the Lorentz group is reducible.61) (4. (Strictly speaking.4. 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 (4.57) In this representation. the spinor is a representation of the double cover of the Lorentz group SL(2.59) into Weyl spinors.1 The Weyl Equation Let’s see what becomes of the Dirac Lagrangian under the decomposition (4. C)). Both are block diagonal. 1 ) representation. It decomposes into two irreducible representations. 0) representation of the Lorentz group.

However. since they couple through the mass term. A similar counting for the Weyl fermion tells us that it has two degrees of freedom. with the equation of motion i¯ µ ∂µ u+ = 0 σ or These are the Weyl equations. corresponding to the particle and the anti-particle in the quantum theory. The Dirac fermion has 4 complex components = 8 real components. Degrees of Freedom Let me comment here on the degrees of freedom in a spinor.65) iσ µ ∂µ u− = 0 (4. 3 index. half the dimension of the phase space. σ µ = (1. σ i ) and σ µ = (1.64) It is not proportional to the time derivative of ψ. in this sense a real scalar ﬁeld φ has a single degree of freedom. We will see in the next section that. and a further two for the anti-particle. At the quantum level. So the phase space of the Dirac spinor ψ has 8 real dimensions and correspondingly the number of real degrees of freedom is 4. a massless fermion can be described by u+ (or u− ) alone. 2. the number of degrees of freedom of a system is equal to the dimension of the conﬁguration space or. this counting manifests itself as two degrees of freedom (spin up and down) for the particle. this translates to the fact that it gives rise to a single type of particle.62).where we have introduced some new notation for the Pauli matrices with a µ = 0.63) From (4. This means that the phase space for a spinor is therefore parameterized by ψ and ψ † . 1. equivalently. But this isn’t right. in the quantum theory. for the Dirac Lagrangian. and in contrast to the scalar ﬁeld. But what about a Dirac spinor? One might think that there are 8 degrees of freedom. In ﬁeld theory we have an inﬁnite number of degrees of freedom. the equation of motion is ﬁrst order rather than second order. How do we count degrees of freedom? In classical mechanics. Crucially. A classical complex scalar ﬁeld has two degrees of freedom. For example. −σ i ) ¯ (4. we see that a massive fermion requires both u+ and u− . the momentum conjugate to the spinor ψ is given by ˙ πψ = ∂L/∂ ψ = iψ † (4. In particular. – 92 – . but it makes sense to count the number of degrees of freedom per spatial point: this should at least be ﬁnite. while for a scalar it is parameterized ˙ by φ and π = φ.

iγ 5 ). 2. γ B } = 2η AB . γ µ } = 0 and (γ 5 )2 = +1 (4. One can check that for the chiral representation (4. this means we may form the Lorentz invariant projection operators P± = 1 1 ± γ5 2 (4.57) is called the chiral representation: it’s because the decomposition of the Dirac spinor ψ is simply given by (4. γ5 = 1 0 0 −1 (4. we may use γ 5 to deﬁne the chiral spinors.2 γ 5 The Lorentz group matrices S[Λ] came out to be block diagonal in (4. 2. You can also check that [Sµν . ψ± = P± ψ (4. 3.71) which form the irreducible representations of the Lorentz group. Since (γ 5 )2 = 1. (You γ ˜ 4 5 might think that γ would be a better name! But γ is the one everyone chooses . while ψ− is “left-handed”. 5). 1.70) from which we see that the operators P± project onto the Weyl spinors u± .it’s a more sensible name in Euclidean space. which means that γ 5 is a scalar under rotations and boosts. for an arbitrary representation of the Cliﬀord algebra. where A = 1. In fact.59). 4. ψ+ is often called a “right-handed” spinor.66) Now S[Λ] will not be block diagonal.68) (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 . so that γ µ → Uγ µ U −1 and ψ → Uψ ? (4. 4 satisfy the ﬁve-dimensional Cliﬀord algebra {˜ . – 93 – .58) because we chose the speciﬁc representation (4. But what happens if we choose a diﬀerent representation γ µ of the Cliﬀord algebra.69) 2 2 such that P+ = P+ and P− = P− and P+ P− = 0.4.57).67) The reason that this is called γ 5 is because the set of matrices γ A = (γ µ .57).4. this is why the representation (4. 3. γ 5 ] = 0. However. with ˜ A A = 0.

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 –

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

the Dirac spinor transforms as (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 . following (4.6 Symmetries and Conserved Currents The Dirac Lagrangian enjoys a number of symmetries.89) ¯ The Lagrangian depends on ∂µ ψ.10).41) gives us the energy-momentum tensor ¯ T µν = iψγ µ ∂ ν ψ − η µν L (4. for / a spinor ﬁeld the equations of motion are ﬁrst order: (i ∂ − m)ψ = 0. we have ω µν = 2 Ωρσ (Mρσ )µν .22) which. In the case of a scalar ﬁeld this didn’t really buy us anything because the equations of motion are second order in derivatives. so the standard formula (1.8) (Mρσ )µ = η ρµ δ σ − η σµ δ ρ ν ν ν (4. we have again used the equations of motion. in the last equality.91) where.4. Here we list them and compute the associated conserved currents. This means we µν can set L = 0 in T .93) 1 where.94) – 98 – . reads δψ α = −ω µν xν ∂µ ψ α + 1 Ωρσ (S ρσ )αβ ψ β 2 (4. 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. leaving ¯ T µν = iψγ µ ∂ ν ψ In particular. while the energy-momentum is typically ﬁrst order. However.92) (4. Spacetime Translations Under spacetime translations the spinor transforms as δψ = ǫµ ∂µ ψ (4. and Mρσ are the generators of the Lorentz algebra given by (4. Lorentz Transformations Under an inﬁnitesimal Lorentz transformation. in inﬁnitesimal form. but not ∂µ ψ.

for fermions. This gives rise to the current µ ¯ jV = ψγ µ ψ (4. telling us that the quantization of a Dirac spinor gives rise to a particle carrying spin 1/2. ψ → e−iα ψ.95) The conserved current arising from Lorentz transformations now follows from the same calculation we saw for the scalar ﬁeld (1. The conserved quantity arising from this symmetry is ¯ i∂µ ψγ Q= ¯ d3 x ψγ 0 ψ = d3 x ψ † ψ (4.which.98) / where. the Dirac Lagrangian admits an extra internal symmetry which rotates left and right-handed fermions in opposite directions. We have ¯ (J µ )ρσ = xσ T µρ − xρ T µσ − iψγ µ S ρσ ψ (4. as we saw above. or particle number. µ ¯ ¯ ¯ ¯ ∂µ jV = (∂µ ψ)γ µ ψ + ψγ µ (∂µ ψ) = imψψ − imψψ = 0 (4.95). in the last equality.97) where “V ” stands for vector. Internal Vector Symmetry The Dirac Lagrangian is invariant under rotating the phase of the spinor. after direct substitution. this extra term will be responsible for providing the single particle states with internal angular momentum. Axial Symmetry When m = 0.54) with two diﬀerences: ﬁrstly. we pick up an extra piece in the current from the second term in (4. reﬂecting the fact that the left and right-handed components ψ± transform in the same way under this symmetry. tells us that ω µν = Ωµν . We can easily check that µ jV is conserved under the equations of motion.99) We will see shortly that this has the interpretation of electric charge. ψ → eiαγ ψ 5 5 ¯ ¯ and ψ → ψeiαγ (4.100) – 99 – . the spinor equations of motion set L = 0. we have used the equations of motion i ∂ ψ = mψ and ¯ µ = −mψ. when (J µ )ρσ is turned into an operator. secondly.96) After quantization. So we get 1 δψ α = −ω µν xν ∂µ ψ α − 2 (Sµν )αβ ψ β (4.

However.Here the second transformation follows from the ﬁrst after noting that e−iαγ γ 0 = 5 γ 0 e+iαγ . Indeed. −σ i ) ¯ Claim: The solution to (4.105) is u(p) = √ √ p·σξ p·σξ ¯ (4.104) (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. in the quantum theory things become more interesting for the axial current.101) µ where A is for “axial” since jA is an axial vector. 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. independent of spacetime x which.107) (4. This is conserved only when m = 0.102) which vanishes only for m = 0.106) µ pµ σ µ −m ¯ −m pµ σ µ u(p) = 0 (4.7 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. as the notation suggests. The Dirac equation then becomes (γ pµ − m)u(p) = where we’re again using the deﬁnition. µ ¯ jA = ψγ µ γ 5 ψ 5 (4. It is the archetypal example of an anomaly: a symmetry of the classical theory that is not preserved in the quantum theory. can depend on the 3-momentum p. σ i ) and σ µ = (1. σ µ = (1.103) where u(p) is a four-component spinor. 4. But it doesn’t survive the quantization process. This gives the conserved current.105) – 100 – .

Proof: Let’s write u(p)T = (u1.105).108) immediately tells us that u2 = mξ ′ . – 101 – .111) for some 2-component spinor η which we take to be constant and normalized to η † η = 1.104). So we get the promised result (4.110) Solutions of the form (4. To make this more symmetric.92). Those of the form (4.112) m pµ σ µ ¯ pµ σ µ m v(p) = 0 (4. Then the second equation in (4.110). u2 ). are called positive frequency solutions.105) reads (p · σ) u2 = mu1 and (p · σ )u1 = mu2 ¯ (4.for any 2-component spinor ξ which we will normalize to ξ † ξ = 1.107) Negative Frequency Solutions We get further solutions to the Dirac equation from the ansatz ψ = v(p) e+ip·x (4. Then u1 = (p · σ) p · σ ξ = m p · σ ξ.109) with constant A is a solution to (4. which oscillate in time as ψ ∼ e−iEt . To start with. which oscillate as ψ ∼ e+iEt . 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. let’s try ¯ 0 0 ′ the ansatz u1 = (p · σ)ξ for some spinor ξ ′ . It’s important to note however that both are solutions to the classical ﬁeld equations and both have positive energy (4. So we learn that any spinor of the form u(p) = A (p · σ) ξ ′ mξ ′ (4. Then equation (4.108) Either one of these equations implies the other. are negative frequency solutions. 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 . we choose √ √ √ ¯ ¯ A = 1/m and ξ ′ = p · σ ξ with constant ξ.

for a boosted solution of the spin down ξ T = (0. 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 (4.116) Similarly. 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). even though they have no such interpretation before quantization. 0. with pµ = (E. This should be familiar from quantum mechanics. ξ T = (1. 1) ﬁeld.115) In fact.114) The 2-component spinor ξ deﬁnes the spin of the ﬁeld. p).117) – 102 – . for which E = p3 . u(p) = √ m ξ ξ (4.7.26).1 Some Examples Consider the positive frequency solution with mass m and 3-momentum p = 0. this expression also makes sense for a massless ﬁeld. 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 (4.4. For a massless particle we have u(p) = √ 0 2E 0 1 0 (4. 0) describes a ﬁeld with spin up along the z-axis. (We picked the normalization (4.107) for the solutions so that this would be the case). Consider now boosting the particle with spin ξ T = (1.113) where ξ is any 2-component spinor. this will become the spin of the associated particle. 0) along the x3 direction. For example. After quantization. ξ → e+iϕ·σ/2 ξ (4. 0. Spatial rotations of the ﬁeld act on ξ by (4. In the rest of this section.

120) and η r † η s = δ rs (4. 4.117) has helicity h = −1/2: it is left-handed. regarding the inner (and outer) products of the spinors u(p) and v(p). s = 1. 2 for the two-component spinors such that ξ r † ξ s = δ rs for example. ξ p·σ ¯ 01 10 √ √ p · σ ξs p · σ ξs ¯ = 2mδ rs (4. The massless ﬁeld with spin ξ T = (1.118) where S ij is the rotation generator given in (4. only the latter will be Lorentz invariant.122) while the Lorentz invariant inner product is u (p) · u (p) = ξ ¯ r s r† p·σ. It’s ﬁrstly convenient to introduce a basis ξ s and η s . or as u · u. but it turns out ¯ that the former is needed when we come to quantize the theory. 1 i ˆ ˆ h = 2 ǫijk pi S jk = 2 pi σi 0 0 σi (4.116) has helicity h = 1/2: we say that it is right-handed. Here we state both: √ p · σ ξs r †√ r† s r† √ p·σ ¯ p·σ.121) p · σ ξs ¯ We can take the inner product of four-component spinors in two diﬀerent ways: either as u† · u. Let’s deal ﬁrst with the positive frequency plane waves.Helicity The helicity operator is the projection of the angular momentum along the direction of momentum. the ﬁeld (4.7. The two independent solutions are now written as √ p · σ ξs s u (p) = √ (4.2 Some Useful Formulae: Inner and Outer Products There are a number of identities that will be very useful in the following section.123) – 103 – . 0) in (4.119) and similarly for η s . Of course. Meanwhile. 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 †√ (4.25). ξ1 = 1 0 and ξ 2 = 0 1 (4.

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 (4.129) – 104 – . The same expression holds for (p · σ )(p · σ).124) We can also compute the inner product between u and v.128) where the two spinors are not now contracted.We have analogous results for the negative frequency solutions. v r (p) · us (p) = 0. but instead placed back to back to give a 4 × 4 matrix. We have u (p) · v (p) = ξ ¯ r s r† √ √ p · σ . Deﬁning the 4-momentum (p′ )µ = (p0 . ¯ 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. However. −p). we have √ ′ p · σ ηs r †√ r† s r† √ p·σ.125) and similarly. and the two terms cancel. it is a slightly ¯ diﬀerent combination that has nice properties (and this same combination appears when we quantize the theory). 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. 2 s=1 / v s (p) v s (p) = p − m ¯ (4. We look at ur † (p) · v s (−p). 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 ¯ (4.126) 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 . with the 3-momentum in the spinor v taking the opposite sign.127) / us (p) us (p) = p + m ¯ s=1 (4. when we come to u† · v. It is: Claim: 2 (4.

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

Quantizing the Dirac Field We would now like to quantize the Dirac Lagrangian. To quantize the theory. This is because we need only specify ψ and ψ † on an initial time slice to determine the full evolution.1 A Glimpse at the Spin-Statistics Theorem We start in the usual way and deﬁne the momentum. It does not involve the time derivative of ψ. ¯ / L = ψ(x) i ∂ − m ψ(x) (5.3) It’s this step that we’ll soon have to reconsider.5.1) We will proceed naively and treat ψ as we did the scalar ﬁeld. which read † † [ψα (x). π= ∂L ¯ = iψγ 0 = iψ † ˙ ∂ψ (5. As with the scalars. the commutation relations of the ﬁelds imply commutation relations for the annihilation and creation operators – 106 – . But we’ll see that things go wrong and we will have to reconsider how to quantize this theory.4) ψ † (x) = s=1 s s where the operators bp † create particles associated to the spinors us (p). where any classical solution is a sum of plane waves. satisfying the canonical commutation relations. ψβ (y)] = δαβ δ (3) (x − y) (5. Since we’re dealing with a free theory. while cp † create particles associated to v s (p). 5. This is as it should be for an equation of motion that is ﬁrst order in time.2) For the Dirac Lagrangian. rather than second order. we promote the ﬁeld ψ and its momentum ψ † to operators. ψβ (y)] = [ψα (x). the momentum conjugate to ψ is iψ † . 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. ψβ (y)] = 0 † [ψα (x).

ψ † (y)] = r. ψ † (y)] = d3 p +ip·(x−y) e = δ (3) (x − y) (2π)3 (5.5) with all other commutators vanishing.5). we have ˙ ¯ H = π ψ − L = ψ(−iγ i ∂i + m)ψ (5.1 The Hamiltonian To proceed. Using the momentum π = iψ † .6) u v (2π)3 2Ep At this stage we use the outer product formulae (4. This was the reason we needed the minus sign in the [c. in the second term. making sure that the pi γ i terms cancel. For now.3). Notice that it’s a little tricky in the middle there.Claim: The ﬁeld commutation relations (5. c† ] term.3) are equivalent to r s [cp .7) as promised. It’s not yet obvious that it’s bad. Proof: Let’s show that the [b. bq † ]ur (p)us (q)† ei(x·p−y·q) r s + [cp † .1. so that [ψ(x). let’s construct the Hamiltonian for the theory. we’ve changed p → −p under the integration sign.s d3 p d3 q (2π)6 1 4Ep Eq r s [bp . 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.8) – 107 – . using p0 = Ep we have [ψ(x). c† ] commutators reproduce the ﬁeld commutators (5. Now. c† ] commutator terms in (5. let’s just carry on.129) which tell us s s / / us vs s u (p)¯ (p) = p + m and s v (p)¯ (p) = p − m. b† ] and [c. bq † ] = (2π)3 δ rs δ (3) (p − q) (5. Note the strange minus sign in the [c. 5. [ψ(x). but we should be aware of it.128) and (4. 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. cq † ] = −(2π)3 δ rs δ (3) (p − q) r s [bp .

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. Now we use the deﬁning equations for the spinors us (p) and v s (p) given in (4.111). for once we’ve left the sum over s = 1.122).124) and (4.10) (5. 2 implicit. 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.12) and ur (p)† · v s (−p) = v r (p)† · us (−p) = 0 – 108 – . We now use our inner product formulae (4. We now wish to turn the Hamiltonian into an operator.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.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 (5.105) and (4. (4. 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 xi pi = −xi pi .which means that H = d3 x H agrees with the conserved energy computed using Noether’s theorem (4.92). in the last two terms we have relabelled p → −p.11) (5.

16) – 109 – . † † † † [ap .5) tells us that r s {bp . Since the above calculation was a little tricky. aq ] = 0 ⇒ ap aq |0 ≡ |p. cq † } = (2π)3 δ rs δ (3) (p − q) (5. And. it’s telling us something important that we missed.The δ (3) term is familiar and easily dealt with by normal ordering.15) with all other anti-commutators vanishing. q = |q. So how do we go about quantizing a ﬁeld as a fermion? Recall that when we quantized the scalar ﬁeld. r s r s r s r s {bp . cq † } = {bp . B} ≡ AB + BA. meaning that they obey Fermi-Dirac statistics with the quantum state picking up a minus sign upon the interchange of any two particles. we will ask that the spinor ﬁelds satisfy † † {ψα (x). ψβ (y)} = 0 † {ψα (x).3).4) of ψ and ψ † in terms of b. Any attempt to do otherwise will lead to an inconsistency. Rather than (5. But now the same proof that led us to (5. ψβ (y)} = {ψα (x). such as the unbounded Hamiltonian we saw in (5. b† . while half-integer spin ﬁelds must be quantized as fermions. As the English would say. 5. it’s all gone a bit Pete Tong. = 0 (5. p (5.13) To have states obeying fermionic statistics. However the −c† c term is a disaster! The Hamiltonian is not bounded below. bq † } = (2π)3 δ rs δ (3) (p − q) r s {cp .12). Taken seriously it would tell us that we could tumble to states of lower and lower energy by continually producing c† 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. but you’ll always ﬁnd this minus sign cropping up somewhere. . cq } = {bp . {A. bq } = {cp . we need anti-commutation relations. the resulting particles obeyed bosonic statistics because the creation and annihilation operators satisﬁed the commutation relations. . 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. in fact. cq } = .14) We still have the expansion (5. (No relation). ψβ (y)} = δαβ δ (3) (x − y) (5. meaning that our quantum theory makes no sense.2 Fermionic Quantization The key piece of physics that we missed is that spin 1/2 particles are fermions. c and c† .

† vacuum by bp † and cp to create particles and antiparticles. Note that when normal ordering the Hamiltonian we now throw away a negative contribution −(2π)3 δ (3) (0). if we wanted to be sure about the spin of the particle. the Hamiltonian (5. At that stage. r2 . 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. we have the Pauli-Exclusion principle |p. Finally. cp ] = −Ep cp (5.20) The two particle states now satisfy r r |p1 . r1 . r = bp † |0 (5. cp † ] = Ep cp † r r [H. just as in the bosonic case. p2 .2.11).19) This means that we can again construct a tower of energy eigenstates by acting on the r r.17) has nice commutation relations with them. we deﬁne the vacuum |0 to satisfy. bp ] = −Ep bp and and r r [H. we could act with the angular momentum operator (4. r2 ≡ bp1 † bp2 † |0 = − |p2 .96) to conﬁrm that a stationary particle |p = 0.The calculation of the Hamiltonian proceeds as before.17) The anti-commutators have saved us from the indignity of an unbounded Hamiltonian. r = 0. this could partially cancel the positive contribution from bosonic ﬁelds.18) Although b and c obey anti-commutation relations. p1 . all the way through to the penultimate line (5. Cosmological constant problem anyone?! 5. p. bp † ] = Ep bp † r r [H.1 Fermi-Dirac Statistics Just as in the bosonic case. In principle. we have the one-particle states r |p. For example. In particular. You can check that r r [H. r.3 Dirac’s Hole Interpretation “In this attempt. 5. r does indeed carry intrinsic angular momentum 1/2 as expected. s s bp |0 = cp |0 = 0 (5. the success seems to have been on the side of Dirac rather than logic” Pauli on Dirac – 110 – .21) conﬁrming that the particles do indeed obey Fermi-Dirac statistics. r1 1 2 (5.

Suppose that a negative energy state is excited to a positive energy state. So we could simply stipulate that in the true vacuum of the universe. After ﬂirting with the idea that it may be the proton. Dirac realized that his theory made a shocking prediction. The spectrum is once again unbounded below. the plane-wave solutions (4. while for us the Hamiltonian is the ﬁeld operator (5. Dirac argued that only charge diﬀerences would be observable (a trick reminiscent of the normal ordering prescription we used for ﬁeld operators). Dirac originally viewed his equation as a relativistic version of the Schr¨dinger equation. Only the positive energy states are accessible.22) ˆ where α = −γ 0 γ and β = γ 0 . Moreover. all the negative energy states are ﬁlled. when a positron comes across – 111 – .17). Having avoided disaster by ﬂoating on an inﬁnite sea comprised of occupied negative energy states. With the interpretation of ψ as a single-particle wavefunction. This is a very diﬀerent viewpoint from the one we now have. Let’s see where Dirac’s viewpoint leads. with ψ = u(p) e−ip·x ψ = v(p) e+ip·x ⇒ ⇒ ∂ψ = Ep ψ ∂t ∂ψ i = −Ep ψ ∂t i (5.Let’s pause our discussion to make a small historical detour. Here the operator H is interpreted as the one-particle Hamiltonian. 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. To reinforce this interpretation. These ﬁlled negative energy states are referred to as the Dirac sea. Dirac ﬁnally concluded that the hole is a new particle: the positron. leaving behind a hole. he / wrote (i ∂ − m)ψ = 0 as i ∂ψ ˆ = −iα · ∇ψ + mβψ ≡ Hψ ∂t (5.104) and (4. the Hamiltonian of the ˆ system is H deﬁned above.12). In Dirac’s view.110) to the Dirac equation are thought of as energy eigenstates. there are states v(p) with arbitrarily low energy −Ep . just like the unbounded ﬁeld theory Hamiltonian (5. The hole would have all the properties of the electron. At ﬁrst glance this is disastrous. except it would carry positive charge. Although you might worry about the inﬁnite negative charge of the vacuum.23) which look like positive and negative energy solutions. where ψ is a classical ﬁeld that should be quantized. with ψ interpreted o as the wavefunction for a single particle with spin.

ψβ (y)} (5. 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. Dirac had predicted anti-matter. Although Dirac’s physical insight led him to the right answer. the two can annihilate. For example. t) at every point in spacetime such that they satisfy the operator equation ∂ψ = i[H. anti-particles exist for both fermions and bosons. We deﬁne the fermionic propagator to be ¯ iSαβ = {ψα (x). one of the greatest achievements of theoretical physics.25) ψ † (x) = s=1 Let’s now look at the anti-commutators of these ﬁelds..an electron. It is when you solve the ﬁeld equations that you see the emergence of particles.92) is positive deﬁnite. everyone thought that the Dirac equation referred directly to physical particles. It took only a couple of years before the positron was discovered experimentally in 1932.” 5.24) 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. you operate on a level where the particles are not there from the start.4 Propagators Let’s now move to the Heisenberg picture.26) – 112 – . This from Julian Schwinger: “Until now. It is instead to be thought of as a classical ﬁeld which has only positive energy solutions because the Hamiltonian (4. We deﬁne the spinors ψ(x. Quantization of this ﬁeld then gives rise to both particle and anti-particle excitations. Now. yet the old equations describe ﬁelds. we now understand that the interpretation of the Dirac spinor as a single-particle wavefunction is not really correct. Dirac’s argument for anti-matter relies crucially on the particles being fermions while. When you begin with ﬁeld equations.. in ﬁeld theory.. we recognize that the equations refer to a sublevel. ψ] ∂t We solve this by the expansion 2 (5. as we have seen already in this course. Experimentally we are concerned with particles.

17). for fermions we now have {ψα (x). for example the Hamiltonian (5. The theory remains causal as long as fermionic operators are not observable. What happened to our precious causality? The best that we can say is that all our observables are bilinear in fermions. 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. can be written as (2. we have already seen that D(x − y) − D(y − x) = 0.¯ In what follows we will often drop the indices and simply write iS(x−y) = {ψ(x). recall.31) outside the lightcone. we made a big deal of this since it ensured that [φ(x). bq † }us (p)¯r (q)e−i(p·x−q·y) u 4Ep Eq s r +{cp † .25). We can then write / iS(x − y) = (i ∂ x + m)(D(x − y) − D(y − x)) (5. ψβ (y)} = 0 (x − y)2 < 0 (5. φ(y)] = 0 (x − y)2 < 0 (5.27) where to reach the ﬁnal line we have used the outer product formulae (4. remember that no one has ever seen a physical measuring apparatus come back to minus itself when you rotate by 360 degrees! – 113 – . These still commute outside the lightcone. but you should remember that S(x−y) is a 4×4 matrix.29) outside the lightcone. Inserting the expansion (5. If you think this is a little weak. which we trumpeted as proof that our theory was causal.90) D(x − y) = Some comments: • For spacelike separated points (x − y)2 < 0. ψ(y)}. we have iS(x − y) = d3 p d3 q (2π)6 1 s r {bp .28) in terms of the propagator for a real scalar ﬁeld D(x − y) which.129). In the bosonic theory. However.30) d3 p 1 −ip·(x−y) e (2π)3 2Ep (5.128) and (4.

While bosonic operators commute inside T .36) – 114 – .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 ::.5 The Feynman Propagator By a similar calculation to that above. .34) Notice the minus sign! It is necessary for Lorentz invariance. the propagator satisﬁes / (i ∂ x − m)S(x − y) = 0 2 (5. as the time ordered product. The minus sign that we see in (5.). ¯ 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 (5.33) We now deﬁne the Feynman propagator SF (x − y). with fermionic operators obeying : ψ1 ψ2 := − : ψ2 ψ1 :. We have the 4-momentum integral representation for the Feynman propagator. there is no invariant way to determine whether x0 > y 0 or y 0 > x0 .34) also occurs for any string of operators inside a time ordered product T (. ψ(y)} = 0 outside the lightcone. so that SF is a Green’s function for the Dirac operator. fermionic operators anti-commute. We have this same behaviour for normal ordered products as well. When (x−y)2 < 0. SF (x − y) = i d4 p −ip·(x−y) γ · p + m e (2π)4 p2 − m2 + iǫ (5. Wick’s theorem proceeds just as in the bosonic case. In this case the minus sign is ¯ necessary to make the two deﬁnitions agree since {ψ(x).• At least away from singularities. we can determine the vacuum expectation value. . ¯ 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. which is again a 4 × 4 matrix.35) / which satisﬁes (i ∂ x − m)SF (x − y) = iδ (4) (x − y). 5. We deﬁne the contraction ¯ ¯ ¯ ψ(x)ψ(y) = T (ψ(x)ψ(y)) − : ψ(x)ψ(y) : = SF (x − y) (5.

(This is still not a particularly realistic theory of nucleon interactions.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. ¯ ¯ d4 x1 d4 x2 T ψ(x1 )ψ(x1 )φ(x1 ) ψ(x2 )ψ(x2 )φ(x2 ) (5. this time. not least because we’re omitting isospin.39) We want to calculate the order λ2 terms from the S-matrix element f | S − 1 |i . We start with: 5. the nucleons have spin.1 An Example: Putting Spin on Nucleon Scattering Let’s study ψψ → ψψ scattering. Moreover.3. and the ψ particles as nucleons. r ′ (5. writing down the Feynman rules for the theory. This is the same calculation we performed in Section (3. Couplings of this type appear in the standard model. because the b† ’s now anti-commute. Yukawa originally proposed an interaction of this type as an eﬀective theory of nuclear forces. In that context. In particular.40) – 115 – . 1 ¯ ¯ L = 1 ∂µ φ∂ µ φ − 2 µ2 φ2 + ψ(iγ µ ∂µ − m)ψ − λφψψ 2 (5. We still have [φ] = 1. so ¯ a coupling of the form φψγ 5 ψ would be more appropriate. We have f| = (−iλ)2 2 r s 4Ep′ Eq′ 0| bq ′ bp ′ ′ ′ (5.37) 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. between fermions and the Higgs boson.6.7. unlike in the case with only scalars.38) We need to be a little cautious about minus signs. We’ll turn to this brieﬂy in Section 5.3) except now the fermions have spin. we will again refer to the φ particles as mesons. the fermions can be leptons (such as the electron) or quarks. Thus. r ′ ′ s r 4Ep′ Eq′ bp ′† bq ′ † |0 ≡ |p ′ . with that calculation as a guide. we should be careful when we take the adjoint. in Nature the relevant mesons are pions which are pseudoscalars. ﬁrstly computing the amplitude of a particular scattering process then. Note the dimensions of the various ﬁelds. With an eye to this. but the kinetic terms require that [ψ] = 3/2. q ′ . We’ll proceed as we did in Section 3. q.5. s. the coupling is dimensionless: [λ] = 0. s′ .3). Except.

Meanwhile.43) doubles up with this. in particular. all ﬁelds are in the interaction picture. Now anti-commuting † the b’s past the b ’s. We may ignore the c† pieces in ψ since they give no contribution at order λ2 . Now let’s see what happens when we hit this with f |. as usual. the 1/ E terms cancel the relativistic state normalization. Putting everything together. We expand out the ψ ﬁelds. The ¯ minus sign that sits out front came from moving ψ(x1 ) past ψ(x2 ). cancelling the √ factor of 1/2 in front of (5.where. ¯ leaving the ψ ﬁelds alone for now.40). 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. the relative minus sign that appears between these two terms. 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 (5.43) Note. the contribution to nucleon scattering comes from the contraction ¯ ¯ : ψ(x1 )ψ(x1 )ψ(x2 )ψ(x2 ) : φ(x1 )φ(x2 ) (5. Just as in the bosonic calculation. Let’s start by looking at how the fermionic operators act on |i .41) We just have to be careful about how the spinor indices are contracted. 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 (5.42) 4Ek1 Ek2 k1 k2 p q where we’ve used square brackets [·] to show how the spinor indices are contracted. 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 – .

¯ 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. we associate a spinor v r (p). Performing the integrals over x1 and x2 . 5. For outgoing anti-fermions we associate v r (p). ¯ 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. – 117 – . Thankfully the Feynman rules will once again encapsulate the combinatoric complexities and make life easier for us. we associate a spinor ur (p).7 Feynman Rules for Fermions It’s important to bear in mind that the calculation we just did kind of blows. this becomes. writing the S-matrix element in terms of the amplitude in the usual way. 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.where we’ve put the φ propagator back in. For outgoing fermions we associate ur (p). f | S − 1 |i = iA(2π)4 δ (4) (p + q − p′ − q ′ ). The rules to compute amplitudes are the following • To each incoming fermion with momentum p and spin r. 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.

We’ve drawn the second Feynman diagram with the legs crossed p. u. or with external spinors u.s/ p. Firstly. the two lowest order Feynman diagrams are shown in Figure 25.r Figure 25: The two Feynman diagrams for nucleon scattering to emphasize the fact that it picks up a minus sign due to statistics.45) – 118 – .• Each internal line gets a factor of the relevant propagator. 5.7. Note that the fermionic propagator is a 4×4 matrix. and integrate over undetermined loop momenta. v or v. Some examples will be given below.r q.r / / q.s / p. we have Nucleon Scattering For the example we worked out previously.44) The arrows on the fermion lines must ﬂow consistently through the diagram (this ensures fermion number conservation). The matrix indices are contracted at each vertex.s + / q. i p for scalars 2 − µ2 + iǫ p p / i( p + m) for fermions 2 − m2 + iǫ p (5. either with further propagators. 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. ¯ ¯ • Impose momentum conservation at each vertex. • Add extra minus signs for statistics. The two diagrams above are diﬀerent because the incoming legs are attached to diﬀerent outgoing legs). (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.r / q.s/ / p. Using the Feynman rules we can read oﬀ the amplitude.1 Examples Let’s run through the same examples we did for the scalar Yukawa theory.

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 –

r = −i(−iλ) (2m) 2 δs sδr r δs r δr s − (p − p′ ) + µ2 (p − q ′ ) + µ2 ′ ′ ′ ′ (5. For example. p → (m. and √ √ p · σξ u(p) = √ → m p · σξ ¯ √ √ p · σξ → m v(p) = √ ¯ − p · σξ ξ ξ ξ −ξ (5. giving us once again an attractive Yukawa ¯ potential (5. 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 – .46).45) ′ ′ become us · us = 2mδ ss and the amplitude is ¯ p.and v(p). The ﬁrst is the extra overall minus sign in the scattering amplitude (5.s / p.50) The δ symbols tell us that spin is conserved in the non-relativistic limit.46). The second minus sign comes from the non-relativistic limit of the spinor contraction for anti-particles in (5. while the momentum dependence is the same as in the bosonic case. These two signs cancel.7.53) (5. which is ′ ′ v s · v s = −2mδ ss .3 Pseudo-Scalar Coupling Rather than the standard Yukawa coupling. p). there are two minus signs which cancel each other. t) → −φ(−x. U(r) = − λ2 e−µr 4πr (5. telling us that once again the particles feel an attractive Yukawa potential. 5. the Feynman diagrams for ψψ → ψψ scattering are again given by Figure 25. i.s/ / q.e. P : φ(x.49) In this limit. t) (5. the spinor contractions in the amplitude for ψψ → ψψ scattering (5.r / q. In the non-relativistic limit.51).51) ¯ ¯ Repeating the calculation for ψ ψ → ψ ψ. we could instead consider ¯ LYuk = −λφψγ 5 ψ This still preserves parity if φ is a pseudoscalar. due to the fermionic nature of the particles.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 .

So.s/ / q.54) – 123 – .s / p. One can easily check that us (p ′ )γ 5 us (p) → ¯ ′ m ξ s T (p − p ′ ) · σξ s .We could again try to take the non-relativistic limit for this amplitude. p. we ′ have us γ 5 us → 0 in the non-relativistic limit. in the non-relativistic limit. Using the expressions for the spinors (5.r / q.49). the leading order amplitude arising from pseudoscalar exchange is given by a spin-spin coupling. To ﬁnd the non-relativistic amplitude. But this time. ¯ ′ we must go to next to leading order. things work a little diﬀerently.r [ξ s T (p − p ′ ) · σξ s ] [ξ r T (p − p ′ ) · σξ r ] → +im(−iλ) (p − p ′ )2 + µ2 2 ′ ′ (5.

We then describe how to couple the photon to fermions and to bosons. 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. 6.4) then gives two of Maxwell’s equations.1) Meanwhile.2) (6.4) To make contact with the form of Maxwell’s equations you learn about in high school. A). we need some 3-vector notation. the theory of light interacting with charged matter.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. ∇·B =0 and ∂B = −∇ × E ∂t (6. If we deﬁne Aµ = (φ.5) Fµν = −Ex −Ey (6. becomes 0 Ex 0 Bz Ey −Bz 0 Bx Ez By −Bx 0 ∂A ∂t and B = ∇ × A (6.3) (6.6) −Ez −By The Bianchi identity (6. the ﬁeld strength also satisﬁes the Bianchi identity ∂λ Fµν + ∂µ Fνλ + ∂ν Fλµ = 0 (6.7) – 124 – .6. Quantum Electrodynamics In this section we ﬁnally get to quantum electrodynamics (QED). in terms of Fµν . then the electric ﬁeld E and magnetic ﬁeld B are deﬁned by E = −∇φ − which. from the deﬁnition of Fµν .

˙ • The ﬁeld A0 has no kinetic term A0 in the Lagrangian: it is not dynamical. We call this a gauge symmetry. This ˙ means that if we are given some initial data Ai and Ai at a time t0 . ∇·E =0 and ∂E =∇×B ∂t (6.These remain true even in the presence of electric sources. expanding out.1 Gauge Symmetry The massless vector ﬁeld Aµ has 4 components. But this is still one too many.9) So A0 is not independent: we don’t get to specify A0 on the initial time slice. in the presence of charged matter these equations pick up extra terms on the right-hand side. reads ∇2 A0 + ∇ · This has the solution A0 (x) = d3 x′ (∇ · ∂ A/∂t)(x ′ ) 4π|x − x ′ | (6.12) – 125 – .10) ∂A =0 ∂t (6. rather than 4.1. 6. the equations of motion give the remaining two Maxwell equations.8) As we will see shortly. Meanwhile. 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. We’ll ask only that λ(x) dies oﬀ suitably quickly at spatial x → ∞. It looks like we have only 3 degrees of freedom in Aµ rather than 4. acting on the vector potential as Aµ (x) → Aµ (x) + ∂µ λ(x) (6. Yet we know that the photon has only two degrees of freedom which we call its polarization states. • The Lagrangian (6.11) for any function λ(x). The ﬁeld strength is invariant under the gauge symmetry: Fµν → ∂µ (Aν + ∂ν λ) − ∂ν (Aµ + ∂µ λ) = Fµν (6. then the ﬁeld A0 is fully determined by the equation of motion ∇ · E = 0 which. which would naively seem to tell us that the gauge ﬁeld has 4 degrees of freedom.3) has a very large symmetry group.

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

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

see that each method comes with its own subtleties. A= d3 p ξ(p) eip·x (2π)3 (6.1 Coulomb Gauge In Coulomb gauge. The ﬁrst of these subtleties is common to both methods and comes when computing the momentum π µ conjugate to Aµ . the momentum conjugate to Ai is our old friend. the electric ﬁeld. the equation of motion for A is ∂µ ∂ µ A = 0 which we can solve in the usual way.17) So A0 acts as a Lagrange multiplier which imposes Gauss’ law ∇·E =0 (6.19) with p2 = |p|2 .2 The Quantization of the Electromagnetic Field In the following we shall quantize free Maxwell theory twice: once in Coulomb gauge. 6. We’ll ultimately get the same answers and. Meanwhile.18) which is now a constraint on the system in which A are the physical degrees of freedom. H = = ˙ d3 x π i Ai − L 1 1 d3 x 2 E · E + 2 B · B − A0 (∇ · E) (6.20) (6.6. The constraint ∇ · A = 0 tells us that ξ must satisfy 0 ξ·p=0 (6.2.16) so the momentum π 0 conjugate to A0 vanishes. π0 = ∂L =0 ˙ ∂ A0 ∂L πi = = −F 0i ≡ E i ˙i ∂A (6. Let’s now see how to treat this system using diﬀerent gauge ﬁxing conditions.21) – 128 – . We can compute the Hamiltonian. This is the mathematical consequence of the statement we made above: A0 is not a dynamical ﬁeld. along the way. and again in Lorentz gauge.

we can rewrite the right-hand side of the commutator in momentum space. To quantize we turn the Poisson brackets into commutators.which means that ξ is perpendicular to the direction of motion p. 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. Ej (y)] = i d3 p (2π)3 δij − pi pj |p| 2 eip·(x−y) (6. Ej (y)] = i 4 d3 p (2π)3 δij − pi pj |p| 2 ipi eip·(x−y) = 0 (6. s = 1. [Ai (x). for example [∂i Ai (x). now imposed on the operators. where the Poisson bracket structure is already altered4 . 2.25) To see that this is now consistent with the constraints. E j (y)] = iδi δ (3) (x − y) (6. r = 1. each of which satisﬁes ǫr (p) · p = 0 and ǫr (p) · ǫs (p) = δrs r. The correct Poisson bracket structure leads to an alteration of the last commutation relation. But from the commutator relations above. E j (y)] = 0 j [Ai (x).23) we haven’t correctly taken into account the constraints. Naively we would write [Ai (x).23) But this can’t quite be right. we see [∇ · A(x). 2 (6. Ej (y)] = i δij − ∂i ∂j ∇2 δ (3) (x − y) (6. In fact.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. see the small book by Paul Dirac.24) What’s going on? In imposing the commutator relations (6. ∇ · E(y)] = i∇2 δ (3) (x − y) = 0 (6.26) which is now consistent with the constraints. We still want to have ∇ · A = ∇ · E = 0. “Lectures on Quantum Mechanics” – 129 – . We can pick ξ(p) to be a linear combination of two orthonormal vectors ǫr . But this topological fact doesn’t cause any complications in computing QED scattering processes. this is a problem already in the classical theory. because it’s not consistent with the constraints. [Ai (x). Aj (y)] = [E i (x).

25) are equivalent to the usual commutation relations for the creation and annihilation operators.31) You can easily check that this equation is true by acting on both sides with a basis of vectors (ǫ1 (p).29) It is not hard to show that the commutation relations (6.We now write A in the usual mode expansion. aq ] = [ap † .32) The Coulomb gauge has the advantage that the physical degrees of freedom are manifest. p). we get the simple expression H= d3 p |p| (2π)3 2 r r ap † ap r=1 (6. we will have to ﬁght to massage this propagator into something a little nicer. We derive the Hamiltonian by substituting (6. 2 r=1 ǫi (p)ǫj (p) = δ ij − r r pi pj |p| 2 (6. as before. and playing around with ǫr polarization vectors.17). r s r s [ap .28) r=1 where. One place where this manifests itself is in the propagator for the ﬁelds Ai (x) (in the Heisenberg picture). 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. aq † ] = (2π)3 δ rs δ (3) (p − q) (6. aq † ] = 0 s r [ap . we’ve lost all semblance of Lorentz invariance.33) The tr superscript on the propagator refers to the “transverse” part of the photon. After normal ordering.30) where. However. ǫ2 (p).28) into (6. The last term vanishes in Coulomb gauge. in deriving this. 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. When we turn to the interacting theory. we need the completeness relation for the polarization vectors. – 130 – . the polarization vectors satisfy ǫr (p) · p = 0 and ǫr (p) · ǫs (p) = δrs (6.

and only later impose the constraint ∂µ Aµ = 0 in a suitable manner on the Hilbert space of the theory.34) Our approach to implementing Lorentz gauge will be a little diﬀerent from the method we used in Coulomb gauge. As we’ll see. πν (y)] = iηµν δ (3) (x − y) (6. [Aµ (x). At ﬁrst. 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 (6.2 Lorentz Gauge We could try to work in a Lorentz invariant fashion by imposing the Lorentz gauge condition ∂µ Aµ = 0.) Our plan will be to quantize the theory (6. We will use α = 1. because both π 0 and π i are dynamical: π0 = ∂L = −∂µ Aµ ˙ ∂ A0 ∂L ˙ πi = = ∂ i A0 − Ai ˙i ∂A (6.2. π ν (y)] = 0 [Aµ (x).38) Turning these classical ﬁelds into operators. α = 0. which is called “Feynman gauge”.35) (In fact. and consider the Lagrangian 1 L = − 4 Fµν F µν − 1 (∂µ Aµ )2 2α (6. The equations of motion that follow from the action are then ∂µ ∂ µ Aν = 0 (6. diﬀerent choices of α are sometimes also referred to as diﬀerent “gauges”. we can proceed very easily. The quantization of the theory is independent of α and. we can simply impose the usual commutation relations.36) (6. is called “Landau gauge”. The other common choice. we will also have to deal with the residual gauge symmetry of this theory which will prove a little tricky.37) with arbitrary α and reach similar conclusions.34) arises directly through the equations of motion.36). Aν (y)] = [π µ (x).6. rather confusingly. we could be a little more general than this. We choose to change the theory so that (6.39) – 131 – .

44) For other 4-momenta. since (6. 3. 1. We pick the normalization ǫλ · ǫλ = η λλ which also means that (ǫµ )λ (ǫν )λ ηλλ′ = ηµν ′ ′ ′ (6.38) for the ˙ classical ﬁelds takes the form π µ = −Aµ + . instead of the two polarization 3vectors that we met in the Coulomb gauge. with λ = 0.43) The third vector ǫ3 is the longitudinal polarization. aq † ] = −η λλ (2π)3 δ (3) (p − q) ′ ′ (6. Of the two spacelike polarizations. We ﬁnd [ap . There are now four polarization 4-vectors ǫλ (p). we will choose ǫ1 and ǫ2 to lie transverse to the momentum: ǫ1 · p = ǫ2 · p = 0 (6. while ǫ1. we pick ǫ0 to be timelike. 0.43) are Lorentz invariant. 1).and we can make the usual expansion in terms of creation and annihilation operators and 4 polarization vectors (ǫµ )λ . ǫ = 2 0 1 0 . aq † ] = 0 and λ λ [ap .41) (6. it becomes clear that this descends to (+i) in the deﬁnition of momentum. This can be traced to the fact that the momentum (6. 0. aq ] = [ap † . Of these four 4-vectors. .3 are spacelike. 2. so p ∼ (1. if the momentum lies along the x3 direction. In the Heisenberg picture. then 1 0 0 0 ǫ = 0 0 0 0 . .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. rather than the familiar (−i) that we’ve seen so far. and translate the ﬁeld commutation relations (6. the polarization vectors are the appropriate Lorentz transformations of these vectors. For example.45) – 132 – . ǫ = 1 1 0 0 . We do our usual trick.42) The polarization vectors depend on the photon 4-momentum p = (|p|.. ǫ = 3 0 0 1 (6. p).40) λ=0 Note that the momentum π µ comes with a factor of (+i).

We need some weaker condition. which.46) But for the timelike annihilation and creation operators. . Then we could try implementing the constraint in the quantum theory in a number of diﬀerent ways.48) Then we can create one-particle states in the usual way.39) won’t be obeyed for π 0 = −∂µ Aµ .50) Wtf? That’s very very strange. 0 has negative norm. 2. 0| q. But this can’t possibly work because the commutation relations (6. We work in the Heisenberg picture. 0 0 p. and cut the physical polarizations down to two. 3 (6. everything looks ﬁne. 3. – 133 – . 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 +.The minus signs here are odd to say the least! For spacelike λ = 1. 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. the state |p. But for the timelike polarization λ = 0. 0 2 At this point we should remember our constraint equation.. aq † ] = −(2π)3 δ (3) (p − q) (6.51) makes sense as an operator equation. aq † ] = δ λλ (2π)3 δ (3) (p − q) ′ ′ λ. . 2. λ |p. We will see that it will remove the timelike. all seems well. λ = ap † |0 (6. A Hilbert space with negative norm means negative probabilities which makes no sense at all. we’ve not imposed on our theory. we have 0 0 [ap . ∂µ Aµ = 0. 2. This is going to come to our rescue. until now. λ = 1. so that ∂µ Aµ = 0 (6. 3. 0 = 0| ap aq † |0 = −(2π)3 δ (3) (p − q) (6. λ′ = 1. we take the Lorentz invariant vacuum |0 deﬁned by λ ap |0 = 0 (6. negative norm states.47) This is very odd! To see just how strange this is.49) For spacelike polarization states. λ λ [ap .

In order to keep the vacuum as a good physical state.54) so that the operator ∂µ Aµ has vanishing matrix elements between physical states. After all. So what does the physical Hilbert space Hphys look like? And. on the vacuum A+ |0 = 0 automatically.• We could try to impose the condition on the Hilbert space instead of directly on the operators. our bad states will include the weird negative norm states that we’re so disgusted by.53) Then. But how can we deﬁne the good states? One idea is to impose ∂µ Aµ |Ψ = 0 (6. 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. Equation (6.52) as our constraint • Our ﬁnal attempt will be the correct one. but almost! Let’s consider a basis of states for the Fock space. we can ask that physical states |Ψ are deﬁned by ∂ µ A+ |Ψ = 0 µ This ensures that Ψ′ | ∂µ Aµ |Ψ = 0 (6. We can decompose any element of this basis as |Ψ = |ψT |φ . the condition is too strong.55) (6. The linearity of the constraint means that the physical states |Ψ span a physical Hilbert space Hphys . physical states |Ψ . created by – 134 – . So not even µ µ the vacuum is a physical state if we use (6. but ∂µ A− |0 = 0. With luck. 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. in particular.52) on all good.54) is known as the Gupta-Bleuler condition. But this can’t work either! Again. For example. 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. where |ψT contains only transverse photons.

But we still need to deal with all these zero norm states. We can check that this is true for the Hamiltonian.2 ap † . The Gupta-Bleuler condition (6. The way we cope with the zero norm states is to treat them as gauge equivalent to the vacuum.56) does indeed decouple the negative norm states. one can show that the expectation values of all gauge invariant operators evaluated on physical states are independent of the coeﬃcients Cn in (6.57) where |φ0 = |0 is simply the vacuum. which we can write as ∞ |φ = n=0 Cn |φn (6. In general. Whenever the state contains a timelike photon of momentum p.54) requires 3 0 (ap − ap ) |φ = 0 (6. This also renders the Hamiltonian positive deﬁnite. 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. Two states that diﬀer only in their timelike and longitudinal photon content.0 1. while |φ contains the timelike photons created by ap † and longitudinal photons 3 created by ap † . Which is an improvement. which can be easily computed to be H= d3 p |p| (2π)3 3 i i 0 0 ap† ap − ap † ap (6.57).58) This means that the inner product on Hphys is positive semi-deﬁnite. leaving us just with the contribution from the transverse photons as we would expect. we can’t just stipulate that two states are physically identical unless they give the same expectation value for all physical observables. – 135 – . Of course. It’s not hard to show that although the condition (6. |φn with n ≥ 1 are said to be physically equivalent.59) i=1 3 3 0 0 But the condition (6.56) This means that the physical states must contain combinations of timelike and longitudinal photons. We can think of the gauge symmetry of the classical theory as descending to the Hilbert space of the quantum theory.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. all the remaining states involving timelike and longitudinal photons have zero norm φm | φn = δn0 δm0 (6.

Propagators Finally.60) This is a lot nicer than the propagator we found in Coulomb gauge: in particular.1 Coupling to Fermions The Dirac Lagrangian ¯ / L = ψ(i ∂ − m)ψ (6. with the Lagrangian. But we’ve got lots of those! Let’s look at how we can couple two of them to electromagnetism.63) In other words. for consistency. The equations of motion read ∂µ F µν = j ν so.3 Coupling to Matter Let’s now build an interacting theory of light and matter. 1 ¯ / ¯ L = − 4 Fµν F µν + ψ(i ∂ − m)ψ − eψγ µ Aµ ψ (6. Had we pushed through the calculation with arbitrary coeﬃcient α. 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. 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. we require ∂µ j µ = 0 (6. 0| T Aµ (x)Aν (y) |0 = 6.65) ¯ ¯ has an internal symmetry ψ → e−iα ψ and ψ → e+iα ψ.37).64) (6. either scalars or spinors. j µ must be a conserved current. we would ﬁnd the propagator. it’s a simple matter to compute the propagator in Lorentz gauge.62) where j µ is some function of the matter ﬁelds. with α ∈ R. 6. So we could look at the theory of electromagnetism coupled to fermions.66) – 136 – . it’s Lorentz invariant. This gives rise to µ ¯ the conserved current jV = ψγ µ ψ.61) (6.3. We want to write down a Lagrangian which couples Aµ to some matter ﬁelds. We could also return to the Lagrangian (6.

let’s rewrite the Lagrangian as 1 ¯ / L = − 4 Fµν F µν + ψ(i D − m)ψ (6. minus the number of antiparticles. For the free Maxwell theory. To see that all is well. 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. 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. since ψ → +ieλ(x) ¯ e ψ. To see this. since this will also hit the e−ieλ piece after the transformation. we have Q=e d3 p (2π)3 2 s s s s (bp † bp − cp † cp ) (6.70) After treating this as a quantum equation. We have Dµ ψ = ∂µ ψ + ieAµ ψ → ∂µ (e−ieλ ψ) + ie(Aµ + ∂µ λ)(e−ieλ ψ) = e−ieλ Dµ ψ (6.71) s=1 which is the number of particles. Electric Charge The coupling e has the interpretation of the electric charge of the ψ particle.where we’ve introduced a coupling constant e. Note that the particle and the anti-particle are required by the formalism to have opposite electric – 137 – . Does our interacting theory above still have a gauge symmetry? The answer is yes. We therefore have the total charge Q given by Q=e ¯ d3 x ψ(x)γ 0 ψ(x) (6. This ensures that the whole Lagrangian is invariant. The tricky term is the derivative acting on ψ.67) where Dµ ψ = ∂µ ψ + ieAµ ψ is called the covariant derivative.68) for an arbitrary function λ(x).69) so the covariant derivative has the nice property that it merely picks up a phase under the gauge transformation. let’s look at how the covariant derivative transforms. This Lagrangian is invariant under gauge transformations which act as Aµ → Aµ + ∂µ λ and ψ → e−ieλ ψ (6. with the derivative of e−ieλ cancelling the transformation ¯ of the gauge ﬁeld.

2 Coupling to Scalars For a real scalar ﬁeld. This is a true symmetry of the system. known as the ﬁne structure constant α= Setting = c = 1. the subset of global symmetries from among the gauge symmetries are those for which λ(x) → α = constant as x → ∞. but with charge −e instead of +e. the electric charge is usually written in terms of the dimensionless ratio α. namely the conservation of electric charge. let’s check that the 4 × 4 matrix C that we introduced in Section 4. The former takes you from one physical state to another with the same properties and results in a conserved current through Noether’s theorem. More generally. we have no suitable conserved current. For QED. there is also a single global symmetry: that with λ(x) = constant.3. meaning that it takes us to another physical state.charge. There’s a small subtlety here that’s worth elaborating on. we have e = √ 1 e2 ≈ 4π c 137 (6. the gauge symmetry ψ → e+ieλ(x) ψ seems to lead to a conservation law.72) 4πα ≈ 0. This means that we can’t couple a real scalar ﬁeld to a gauge ﬁeld. 6. – 138 – . The latter is a redundancy in our description of the system.3. Yet in electromagnetism.5 really deserves the name “charge conjugation matrix”. and the deﬁnition ψ (c) = Cψ ⋆ . These take us from one physical state to another. This is because among the inﬁnite number of gauge symmetries parameterized by a function λ(x).73) So we see that the charge conjugate spinor ψ (c) satisﬁes the Dirac equation. 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 )⋆ . Finally. we see that the charge conjugate spinor ψ (c) satisﬁes (iγ µ ∂µ + eγ µ Aµ − m)ψ (c) = 0 (6. the theory of light interacting with electrons. If we take the complex conjugation of the Dirac equation. I stressed that there’s a radical diﬀerence between the interpretation of a global symmetry and a gauge symmetry.

We solve both of these problems simultaneously by remembering the covariant derivative. The Lagrangian is 1 ¯ / L = − Fµν F µν + ψ(i D − m)ψ 4 where Dµ = ∂µ + ieAµ . 6. This procedure is known as minimal coupling. We could try to couple the associated current to the gauge ﬁeld. We will now do that. (6. the combination Dµ ϕ = ∂µ ϕ + ieAµ ϕ (6. In this scalar theory. 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.76) L = − Fµν F µν + Dµ ϕ⋆ D µ ϕ − m2 |ϕ|2 4 In general. If. we worked in Coulomb gauge.5 where the Feynman rules for QED are written down.74) again transforms as Dµ ϕ → e−ieλ Dµ ϕ under a gauge transformation Aµ → Aµ + ∂µ λ and ϕ → e−ieλ ϕ. If we have a U(1) symmetry that we wish to couple to a gauge ﬁeld. (For this section. it will now pick up a contribution from the j µ Aµ term.77) – 139 – . This means that if we try to compute the current associated to the symmetry. we may do so by replacing all derivatives by suitable covariant derivatives.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.Let’s now consider a complex scalar ﬁeld ϕ. If we worked in Lorentz gauge previously. I’ll depart from our previous notation and call the scalar ﬁeld ϕ to avoid confusing it with the spinor). 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. The route we take now depends on the gauge choice. this trick works for any theory. We have a symmetry ϕ → e−iα ϕ. however.75) (6. then we still have a bit of work in front of us in order to massage the photon propagator into something Lorentz invariant. then we can jump straight to Section 6. So the whole procedure wasn’t consistent.

1.79) (6. we have LMaxwell = d3 x 1 ˙ 2 A 2 − 1B 2 − 2 e2 2 d3 x′ j0 (x)j0 (x ′ ) 4π|x − x ′ | (6.80) where the cross-term has vanished using ∇ · A = 0. ∂L ˙ =A ˙ ∂A ∂L πψ = = iψ † ˙ ∂ψ Π= which gives us the Hamiltonian H= d3 x 1 ˙ 2 A 2 (6. – 140 – . t) = e d3 x′ j 0 (x ′ . the equation of motion arising from varying A0 is now −∇2 A0 = eψ † ψ ≡ ej 0 which has the solution A0 (x.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.In Coulomb gauge ∇ · A = 0. This is exactly the type of interaction that we boasted in Section 1.81) We ﬁnd ourselves with a nonlocal term in the action. Changing notation slightly from previous chapters.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. it wouldn’t appear if we worked in Lorentz gauge. For example. t) 4π|x − x ′ | (6. After integrating the second term by parts and inserting the equation for A0 .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. we have the conjugate momenta. We now compute the Hamiltonian.

Since it comes from the A0 component of the gauge ﬁeld. Each end of the line comes with an i. In fact. 2. 1. 3 index telling us the component of A.4. ν = 0 |p| pi pj i Dµν (p) = (6. so we can combine the non-local interaction with the transverse photon propagator by deﬁning a new photon propagator i + 2 µ. Let’s start with the oﬀending instantaneous interaction. j = 1. we could try to redeﬁne the propagator to include a D00 piece which will capture this term.84) µ = i = 0. We’ll now show that we can massage these expressions into something much more simple and Lorentz invariant. We calculated the transverse photon pi pj i tr δij − 2 propagator in (6. The new pieces are: • We denote the photon by a wavy line. This is the price we’ve paid for working in Coulomb gauge. • The non-local interaction which. the rules are the same as those given in Section 5.33): it is and contributes Dij = 2 p + iǫ |p| • The vertex contributes −ieγ i . For fermions.83) With this propagator. ν = 0. The index on γ i contracts with the index on the photon line. 3 index. in position space. the wavy photon line now carries a µ.6. 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. 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. ν = j = 0 p2 + iǫ δij − |p|2 0 otherwise – 141 – . it ﬁts quite nicely in this form: if we look in momentum space. 2. we have δ(x0 − y 0 ) = 4π|x − y| d4 p eip·(x−y) (2π)4 |p|2 (6.1 Naive Feynman Rules We want to determine the Feynman rules for this theory.

Here we’ll be more pedestrian and show that we can do this for certain Feynman diagrams. In particular. Then ¯ ¯ ′ ′ since k = p − p = q − q. p p/ ∼ e2 [¯(p′ )γ µ u(p)] Dµν (k) [¯(q ′ )γ ν u(q)] u u q q / (6. similarly. we focus on a particular tree-level diagram that contributes to e− e− → e− e− scattering. they are. Using this fact.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 (6.89) – 142 – . Recall that u(p) satisﬁes the equation / ( p − m)u(p) = 0 (6. Lorentz invariant propagator Dµν (p) = −i ηµν p2 (6. We have the following: Claim: We can replace the propagator Dµν (p) with the simpler. But wouldn’t it be nice if they were both part of something more symmetric! In fact. we have / /′ ¯ kµ αµ = u(p ′ )( p − p )u(p) = u(p ′ )(m − m)u(p) = 0 ¯ and.86) where k = p − p′ = q ′ − q. 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 (6.85) Proof: There is a general proof using current conservation.The D00 piece of the propagator doesn’t look a whole lot diﬀerent from the transverse photon propagator. kν β ν = 0.87) Let’s deﬁne the spinor contractions αµ = u(p ′ )γ µ u(p) and β ν = u(q ′ )γ ν u(q).

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

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

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

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

3 pieces vanish in the non-relativistic limit: us (p)γ i ur (q) → 0. We’ve understood how to compute quantum amplitudes in a large array of ﬁeld theories. To do this. the non-relativistic limit of the spinor is u(p) → √ m ξ We ﬁnd the familiar repulsive Coulomb potential. the amplitude is p. i = 1.96) q.r / [¯(p ′ )γ µ u(p)] [¯(q ′ )γµ v(q)] u v = +i(−ie) ′ − p)2 (p 2 (6.97) / q.s / = −i(−ie)2 / q.momenta. We start by looking at e− e− → e− e− scattering.7.s / e2 d3 p eip·r =+ (2π)3 |p|2 4πr (6. ¯ Comparing the scattering amplitude in this limit to that of non-relativistic quantum mechanics.r – 147 – .5.1 The Coulomb Potential We’ve come a long way. we have the eﬀective potential between two electrons given by. 2.2 and 5.98) q. we use our newfound knowledge to rederive a result you learnt in kindergarten: Coulomb’s law. To end this course. U(r) = +e 2 Following (5.r / [¯(p ′ )γ µ u(p)] [¯(q ′ )γµ u(q)] u u ′ − p)2 (p (6.r .95) 6.s / p. 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. we repeat our calculation that led us to the Yukawa force in Sections 3. For e− e+ → e− e+ scattering. This ξ means that the γ 0 piece of the interaction gives a term us (p)γ 0 ur (q) → 2mδ rs . we have the amplitudes given by e− e− µ− 1 ∼ (p − p′ )2 µ− µ− e− and e+ µ+ ∼ 1 (p + q)2 (6.49).6.2.s / p. We have p. while ¯ i the spatial γ .

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

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