Quantum Field Theory

Lecture Notes

Joachim Kopp

April 21, 2016

 Contents

1 Introduction and Motivation 7

1.1 Do you recognize the following equations? . . . . . . . . . . . . . . . . . . 7
1.2 A Note on Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 The Klein–Gordon Field 9

2.1 Necessity of the Field Viewpoint . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Elements of Classical Field Theory . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 The Euler–Lagrange Equations . . . . . . . . . . . . . . . . . . . . 10
2.2.2 The Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.3 Noether’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Quantization of the Klein–Gordon Field . . . . . . . . . . . . . . . . . . . 13
2.3.1 Commutation Relations . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.2 The Quantized Hamiltonian . . . . . . . . . . . . . . . . . . . . . . 15
2.3.3 Time-Dependence of the Klein–Gordon Field Operator . . . . . . . 16
2.4 The Feynman Propagator for the Klein–Gordon Field . . . . . . . . . . . 17
2.4.1 Green’s Functions of the Klein–Gordon Operator . . . . . . . . . . 17
2.4.2 The Feynman Propagator . . . . . . . . . . . . . . . . . . . . . . . 19
2.4.3 Relation to Correlation Functions . . . . . . . . . . . . . . . . . . . 19

3 The Dirac Field 21

3.1 The Dirac Equation and its Solutions . . . . . . . . . . . . . . . . . . . . 21
3.1.1 The Equation and the Corresponding Lagrangian . . . . . . . . . . 21
3.1.2 Solutions of the Dirac Equation . . . . . . . . . . . . . . . . . . . . 22
3.1.3 Spin Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Quantization of the Dirac Field . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2.1 How Not to Quantize the Dirac Field . . . . . . . . . . . . . . . . 25
3.2.2 Quantizing the Dirac Field with Anticommutators . . . . . . . . . 26
3.2.3 Physical Significance of the Quantized Dirac Field . . . . . . . . . 28
3.3 The Feynman Propagator for the Dirac Field . . . . . . . . . . . . . . . . 28
3.4 Symmetries of the Dirac Theory . . . . . . . . . . . . . . . . . . . . . . . 30
3.4.1 Lorentz Invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4.2 Parity (P ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4.3 Time Reversal (T ) . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4.4 Charge Conjugation (C) . . . . . . . . . . . . . . . . . . . . . . . . 40

4 Interacting Fields and Feynman Diagrams 43

3
Contents

4.1 Time-Dependent Perturbation Theory for Correlation Functions . . . . . 43

4.1.1 φ4 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.1.2 The Vacuum State of the Interacting Theory . . . . . . . . . . . . 44
4.1.3 Correlation Functions . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.1.4 Perturbation Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 Wick’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.3 Feynman Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.3.1 Basic Idea and Application to a Simple 4-Point Function . . . . . 54
4.3.2 An Example in φ4 Theory . . . . . . . . . . . . . . . . . . . . . . . 54
4.3.3 A More Advanced Example . . . . . . . . . . . . . . . . . . . . . . 55
4.3.4 More Examples for Diagrams with Non-Trivial Symmetry Factors 56
4.3.5 Position Space Feynman Rules . . . . . . . . . . . . . . . . . . . . 57
4.3.6 Momentum Space Feynman Rules . . . . . . . . . . . . . . . . . . 58
4.3.7 Disconnected Feynman Diagrams . . . . . . . . . . . . . . . . . . . 59
4.3.8 The Denominator of the Correlation Function . . . . . . . . . . . . 61
4.4 The LSZ Reduction Formula . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.5 Computing S-Matrix Elements from Feynman Diagrams . . . . . . . . . . 65
4.6 Feynman Rules for Fermions . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.6.1 The Master Formula for Correlation Functions Involving Fermions 68
4.6.2 Wick’s Theorem for Fermions . . . . . . . . . . . . . . . . . . . . . 69
4.6.3 The LSZ Formula for Fermions . . . . . . . . . . . . . . . . . . . . 70
4.6.4 Yukawa Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.6.5 The Yukawa Potential . . . . . . . . . . . . . . . . . . . . . . . . . 78

5 Quantum Electrodynamics 81
5.1 The QED Lagrangian from Symmetry Arguments . . . . . . . . . . . . . . 81
5.2 The Feynman Rules for QED . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.3 e+ e− → µ+ µ− . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.3.1 Feynman Diagram and Squared Matrix Element . . . . . . . . . . 86
5.3.2 Trace Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.3.3 The Squared Matrix Element for e+ e− → µ+ µ− (Part II) . . . . . 89
5.3.4 The Cross Section — General Results . . . . . . . . . . . . . . . . 90
5.3.5 The Cross Section for e+ e− → µ+ µ− . . . . . . . . . . . . . . . . . 91
5.3.6 e+ e− → µ+ µ− : Summary . . . . . . . . . . . . . . . . . . . . . . . 92
5.4 More Technology for Evaluating QED Feynman Diagrams . . . . . . . . . 93
5.4.1 Scattering of Polarized Particles . . . . . . . . . . . . . . . . . . . 93
5.4.2 External Photons . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

6 Path Integrals 95
6.1 Path Integrals in Quantum Mechanics . . . . . . . . . . . . . . . . . . . . 95
6.2 The Path Integral for a Free Scalar Field . . . . . . . . . . . . . . . . . . . 97
6.3 The Feynman Propagator from the Path Integral . . . . . . . . . . . . . . 100
6.4 Wick’s Theorem from the Path Integral . . . . . . . . . . . . . . . . . . . 101

4
 Contents

6.5 Interacting Field Theories in the Path Integral Formalism . . . . . . . . . 102

6.6 Quantization of the Photon Field . . . . . . . . . . . . . . . . . . . . . . . 103
6.7 Path Integrals for Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.7.1 Grassmann Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.7.2 Partition Function, Functional Derivative and Correlation Func-
tions for Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.8 The Quantum Equations of Motion: Schwinger–Dyson Equations . . . . . 109
6.9 The Ward–Takahashi Identity . . . . . . . . . . . . . . . . . . . . . . . . . 110

7 Weyl and Majorana Fermions 115

7.1 Spinor Indices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.1.1 Left-handed spinors . . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.1.2 Raising and lowering spinor indices . . . . . . . . . . . . . . . . . . 117
7.1.3 Right-handed spinors . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.1.4 Conjugate spinors . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
7.1.5 Lorentz invariance of the Pauli matrices . . . . . . . . . . . . . . . 119
7.1.6 One more example: the vector current . . . . . . . . . . . . . . . . 120
7.2 The QED Lagrangian in 2-Component Notation . . . . . . . . . . . . . . 121
7.3 Majorana Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
7.4 Application: Majorana Neutrinos and the Seesaw Mechanism . . . . . . . 124
7.4.1 Neutrino mass terms . . . . . . . . . . . . . . . . . . . . . . . . . . 124
7.4.2 The seesaw mechanism . . . . . . . . . . . . . . . . . . . . . . . . . 125
7.4.3 Interlude: measuring neutrino masses . . . . . . . . . . . . . . . . 127
7.5 Twistors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
7.5.1 Unifying spinors and momentum 4-vectors . . . . . . . . . . . . . . 131
7.5.2 Twistor notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.5.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

Bibliography 139

5
Contents

6
 Introduction and Motivation
1
1.1 Do you recognize the following equations?

i~ψ̇ = Ĥψ (1.1)

√
a† |ni = n + 1|n + 1i (1.2)

∂t2 ψ − ∇2 ψ − m2 ψ = 0 (1.3)

i∂/µ ψ − mψ = 0 (1.4)

δL(φ, ∂µ φ) δL(φ, ∂µ φ)
∂µ − =0 (1.5)
δ(∂µ φ) δφ
dσ
= |f (θ, φ)|2 (1.6)
dΩ
δL(φ, ∂µ φ)
jµ = ∆φ − J µ (1.7)
δ(∂µ φ)

1.2 A Note on Notation

In this course, we work in natural units, where

~ = c = 1. (1.8)

This implies that we set

1 = ~ · c = 6.58 · 10−16 eV−1 sec × 3 · 108 m/sec = 197 · 10−9 eV · m (1.9)

7
Chapter 1 Introduction and Motivation

It follows for instance that in our units

1 m = 5.066 · 106 eV−1 , (1.10)

15 −1
1 sec = 1.52 · 10 eV . (1.11)

Throughout the lecture, we use the “West Coast Metric”

 
1
 −1
g µν = gµν = 

, (1.12)
 −1 
−1

which is also employed by Peskin and Schroeder [1], and by most particle physicists
today. Note that the book by Srednicki, for instance, uses the “East Coast Metric”, with
−1 in the timelike component and +1 in the spacelike components [2].

8
 The Klein–Gordon Field
2
2.1 Necessity of the Field Viewpoint
From relativistic quantum mechanics, we know how to deal with the dynamics of a single
relativistic particle: the Klein-Gordon equation

(∂µ ∂ µ + m2 )ψ(x) = 0 (2.1)

describes the time evolution of the wave function of a relativistic scalar particle in vac-
uum. (Here, x = (t, x) is the coordinate 4-vector and m is the mass of the particle.)
The Klein–Gordon equation can be easily generalized to motions in an external field by
making the replacements pµ → pµ + ieAµ . The Dirac equation

(iγ µ ∂µ − m)ψ(x) = 0 (2.2)

achieves the same for a fermionic particle. However, these equations only describe the
motion of a single particle. The relativistic equivalence of mass and energy tells us
that kinetic energy can be converted into the production of new particle (e.g. e+ e−
pair production when an ultrarelativistic electron travels through matter), therefore it
is clear that a complete description of a particles’ dynamics must provide for creation
and annihilation processes as well. Note that even in processes where too little energy
is available to actually create an e+ e− pair, the intermittent existence of such pairs for
very short time intervals is allowed by the Heisenberg uncertainty relation ∆E · ∆t ≥ ~
and will therefore happen.
Therefore a relativistic theory that can describe multiparticle processes, including
in particular particle creation and annihilation, is highly desirable. Actually, we have
encountered creation and annihilation processes before: when studying the quantum
mechanical harmonic oscillator, the operators a and a† that transformed its wave function
to a lower or higher energy state were called creation and annihilation operators, though

9
Chapter 2 The Klein–Gordon Field

at the time this seemed like a bit of a misnomer. Quantum field theory (QFT) generalizes
the concept of the harmonic oscillator to an extent that makes the terms “creation”
and “annihilation” operator appropriate. It describes particles of a given species (e.g.
electrons) as a field, i.e. a function φ(x) that maps each spacetime point x to a scalar value
(scalar fields, e.g. the Higgs boson), a Dirac spinor (fermion fields, e.g. the electron) or a
Lorentz 4-vector (vector fields, e.g. the photon). We will show below that the equation
of motion for each Fourier mode of this field has exactly the same form as the equation
of motion of the harmonic oscillator. It can therefore be quantized in exactly the same
way. The creation and annihilation operators then become ap and a†p , e.g. they carry a
momentum index p corresponding to the Fourier mode they describe. The interpretation
of a field is then the following: its ground state, where each momentum mode carries
only the zero point energy, corresponds to the vacuum |0i. If a momentum mode is in
its first excited state a†p |0i, this means that one particle with momentum p exists. The
operators a†p and ap then create and annihilate particles. The equations of motion of
the theory describe the rules according to which such creation and annihilation processes
occur.
In the following, we will first review a few concepts of classical (non-quantized) field
theory (section 2.2) and then make the above statements more precise and more mathe-
matical (section 2.3).

2.2 Elements of Classical Field Theory

2.2.1 The Euler–Lagrange Equations
In classical mechanics, the dynamics of a particle was described by its trajectory x(t),
which was obtained as the solution of its equations of motion. These, in turn, were
derived based on the principle that the action
Z
S ≡ dt L(x, ẋ) , (2.3)

is stationary, i.e.

δS = 0 , (2.4)

where L(x, ẋ) is the Lagrange functional and δS means the variation of S with respect
to x and ẋ. In field theory, the Lagrange functional becomes the Lagrange density (or
Lagrangian for short) L(φ, ∂µ φ), a functional of the field value and its spactime deriva-
tives. (It is called a density because we will see that it has units of [length]−3 [time]−1 , as
opposed to the Lagrange functional in classical mechanics, which has units of [time]−1 .)
The action is defined as
Z
S ≡ d4 x L(φ, ∂µ φ) . (2.5)

10
 2.2 Elements of Classical Field Theory

The principal of stationary action, δS = 0, then implies

Z  
4 δL δL
δS = d x δ(∂µ φ) + δφ (2.6)
δ(∂µ φ) δφ
Z  
4 δL δL
= d x − ∂µ + δφ , (2.7)
δ(∂µ φ) δφ

where in the last step we have integrated by parts. Since eq. (2.7) is required to be
satisfied for any variation δφ, the term in square brackets must vanish. This leads to the
Euler-Lagrange equations
δL δL
∂µ − = 0. (2.8)
δ(∂µ φ) δφ

As an example, consider a scalar field φ (for instance the Higgs field), for which the
Lagrangian reads
1 1
L = (∂µ φ)(∂ µ φ) − m2 φ2 . (2.9)
2 2
The Euler–Lagrange equations then lead to the equation of motion

∂µ ∂ µ φ + m2 φ = 0 , (2.10)

which is just the Klein–Gordon equation.

2.2.2 The Hamiltonian

In classical mechanics, the Lagrange functional L is related to the Hamilton functional
H through a Legendre transform

H = π ẋ − L , (2.11)

where the canonical momentum π is defined as

δL
π≡ . (2.12)
δ ẋ
In analogy, we define the Hamiltonian density (or Hamiltonian for short) in field theory
as

H ≡ π(x)φ̇(x) − L , (2.13)

where π(x) in field theory is defined as

δL
π(x) ≡ . (2.14)
δ φ̇(x)

11
Chapter 2 The Klein–Gordon Field

2.2.3 Noether’s Theorem

A crucial concept in quantum field theory is symmetries, i.e. transformations of the
system that leave the action invariant. Noether’s theorem relates symmetries to conserved
physical quantities. It applies to continuous symmetries, i.e. symmetry transformations
that depend on a parameter α, which can vary continuously from 0 to ∞, where α = 0
corresponds to the identity map. Examples include rotations or gauge transformations,
which should be familiar from classical electrodynamics. It is for our purposes sufficient
to consider infinitesimal transformations

φ(x) → φ0 (x) ≡ φ(x) + α ∆φ(x) , (2.15)

where α is an infinitesimal parameter. We call this transformation a symmetry if it leaves

the action S invariant. This requires in particular that the Lagrangian L is left invariant
up to possibly a 4-divergence ∂µ J µ , which vanishes when integrated over d4 x by virtue
of Gauss’ theorem (We always assume boundary terms to be zero in such integrals.) A
symmetry transformation thus satisfies

L → L + α∂µ J µ . (2.16)

Thus,
δL δL
α∆L ≡ α∂µ J µ = (α∆φ) + ∂µ (α∆φ) (2.17)
δφ δ(∂µ φ)
    
δL δL δL
= α∂µ ∆φ + α − ∂µ ∆φ . (2.18)
δ(∂µ φ) δφ δ(∂µ φ)

In the second step, we have used the product rule of differentiation backwards. The term
in square brackets vanishes due to the Euler–Lagrange equations. Therefore, we are left
with the conclusion that the current
 
µ δL
j ≡ ∆φ − J µ (2.19)
δ(∂µ φ)

is conserved:

∂µ j µ = 0 . (2.20)

This implies in particular for the associated charge (which, except in the case of electro-
magnetic gauge transformations has nothing to do with electric charge)
Z
Q ≡ d3 x j 0 (2.21)

is constant:
Z Z Z
Q̇ = d x∂0 j = d x∂µ j − d3 x∂k j k = 0 .
3 0 3 µ
(2.22)

12
 2.3 Quantization of the Klein–Gordon Field

Here, as usual, the greek index µ runs from 0 to 3, while the latin index k runs only from
1 to 3. The term containing ∂µ j µ vanishes by virtue of Noether’s theorem (eq. (2.20)),
the one containing ∂k j k vanishes thanks to Gauss’ theorem.
As an example, consider infinitesimal spacetime shifts

xµ → xµ − αaµ , (2.23)

for some constant 4-vector aµ , applied to the free scalar field from eq. (2.9). The corre-
sponding field transformation is

φ(x) → φ(x + αa) = φ(x) + α ∂µ φ(x) aµ .

 
(2.24)

Plugging this into the Lagrangian eq. (2.9), we obtain (remembering that α is infinitesi-
mal)

L → L + α(∂µ ∂ν φ)(∂ µ φ)aν + αm2 φ(∂ν φ)aν = L + αaν ∂ν L . (2.25)

This implies

J ν = aν L (2.26)

and
h i
j µ = (∂ µ φ) ∂ν φ(x) − L δ µν aν .
 
(2.27)

Consider in particular time-like shifts, i.e. aν = (1, 0, 0, 0). Then the term in square
brackets becomes

j 0µ ≡ (∂ µ φ) ∂0 φ(x) − L δ µ0 .
 
(2.28)

The 0-component of this expression is just the Legendre transform of the Lagrangian, i.e.
the Hamiltonian density. The d3 x integral over the Hamiltonian density (or Hamiltonian
for short) is just the energy. Thus, invariance under time-like shifts implies energy
conservation. Had we chosen aν to correspond to space-like shifts, we would in analogy
have found momentum conservation.

2.3 Quantization of the Klein–Gordon Field

2.3.1 Commutation Relations
Let us now proceed to quantized field theory. We use as an example the real scalar field
φ(x), whose Lagrangian eq. (2.9) we repeat here:
1 1
L = (∂µ φ)(∂ µ φ) − m2 φ2 . (2.29)
2 2

13
Chapter 2 The Klein–Gordon Field

When going from classical mechanics to quantum mechanics, the procedure is to promote
the position q and the canonical momentum π ≡ (δL)/(δ q̇) to operators that satisfy
canonical commutation relations

[q j , pk ] = iδ jk (2.30)
j k j k
[q , q ] = [p , p ] = 0 . (2.31)

We now do this analogously for the field φ(x) by promoting φ(x) and

δL
π(x) ≡ (2.32)
δ φ̇(x)

to operators and postulating that they satisfy

φ(x, t), π(y, t) = iδ (3) (x − y)

 
    (2.33)
φ(x, t), φ(y, t) = π(x, t), π(y, t) = 0 .

Note that the time coordinate of the spacetime points x and y is the same here.1 Since
we normally deal with particles of known energy (for instance the particles in the LHC
beams), without being too much interested in their position, it is convenient to go to
Fourier space and write

d3 p ipx
Z
φ(x, t) = e φ(p, t) , (2.34)
(2π)3
d3 p ipx
Z
π(x, t) = e π(p, t) . (2.35)
(2π)3

Note that φ(−p, t) = φ(p, t) because φ(x, t) was assumed to be real. The Klein–Gordon
equation

∂µ ∂ µ φ + m2 φ = 0 , (2.36)

then becomes
 2 
∂ 2 2
+ |p| + m φ(p, t) = 0 . (2.37)
∂t2

This is precisely the equation of motion for a harmonic oscillator with frequency
p
ωp = |p|2 + m2 . (2.38)

1
Allowing for different time coordinates would not make sense. For points with a spacelike separation,
causality dictates that all commutators must be zero because measurements at points that are outside
each other’s light cone cannot influence each other. Measurements at points within the light cone,
however, can affect each other. Thus, the commutator of fields at different t is more complicated.

14
 2.3 Quantization of the Klein–Gordon Field

We know how to quantize the harmonic oscillator: we introduce creation and annihilation
operators a†p (t) and ap (t), defined such that

1
ap (t) + a†−p (t)


φ(p, t) = p (2.39)
2ωp
r
ωp
ap (t) − a†−p (t) .


π(p, t) = −i (2.40)
2
Note that, by convention, we assign the creation operator an index −p. It will become
clear later why this choice makes sense. Note also that we work in the Heisenberg picture
here where the field operators are time dependent. We will see below in sec. 2.4 what
the explicit form of this time dependence is. From the commutation relations eq. (2.33),
we can then derive
 
† 1 √ i √ 0 i 0
[ap (t), ap0 (t)] = ωp φ(p, t) + √ π(p, t), ωp φ(−p , t) − √
0 π(−p , t)
2 ωp ωp0
(2.41)
r 
i ωp  ωp0 
r
φ(p, t), π(−p0 , t) + φ(−p0 , t), π(p, t)
 
=− (2.42)
2 ωp0 ωp
Z Z r 
i ωp  ωp0 
r
3 −ipx 3 0 ip0 x 0 0
 
=− d xe d x e φ(x, t), π(x , t) + φ(x , t), π(x, t)
2 ωp0 ωp
(2.43)
Z r 
1 ωp ωp0
r
0
= d3 x e−i(p−p )x + (2.44)
2 ωp0 ωp
= (2π)3 δ (3) (p − p0 ) . (2.45)

2.3.2 The Quantized Hamiltonian

Let us next consider the Hamilton operator H = d3 x H. According to eq. (2.13) it
R

reads
Z  
3 1 2 1 2 1 2 2
H= d x [π(x, t)] + [∇φ(x, t)] + m [φ(x, t)] (2.46)
2 2 2
√
d3 p d3 p0 i(p+p0 )x ωp ωp0
Z Z 
= d3 x ap (t) − a†−p (t) ap0 (t) − a†−p0 (t)



3 3
e −
(2π) (2π) 4
0 2

−p · p + m †
†

+ √ ap (t) + a−p (t) ap0 (t) + a−p0 (t) (2.47)
4 ωp ωp0
d3 p
Z
ωp a†p (t)ap (t) + 12 [ap , a†p ] .


= 3
(2.48)
(2π)

In the last line, we have used that ω−p = ωp . As for the harmonic oscillator in quantum
mechanics, the operator a†p (t)ap (t) is the particle number operator. When applied to a

15
Chapter 2 The Klein–Gordon Field

quantum state |ψi, it gives the number of particles with momentum p in that state:
a†p (t)ap (t)|ψi = np |ψi (2.49)

The term [ap , a†p ] is an infinite constant according to the commutation relation eq. (2.45)
and corresponds to the zero point energies of all the individual p-modes. However, exper-
iments measure only energy differences, therefore a constant (albeit infinite) contribution
to H can be dropped without changing the physics. This is what we will do from now
on.2
We now discuss the eigenstates of the Hamiltonian eq. (2.48). As before, we call the
vacuum state |0i. It isnormalized according to
h0|0i = 1 . (2.50)

A state containing exactly one particle of momentum p will be written as |pi ≡ ca†p |0i.
Here, ap (without the argument t) denotes ap (0). Remember that we are working in the
Heisenberg picture, so the states are time-independent and all the time-dependence is
included in the operators. The normalization constant c is chosen such that
hp|qi = 2Ep (2π)3 δ (3) (p − q) , (2.51)
p
where Ep = p2 + m2 . This normalization condition has the advantage of being Lorentz
invariant, as can be easily shown by applying a Lorentz transformation and using the
properties of the δ-function (Exercise!). It implies
|pi ≡ 2Ep a†p |0i .
p
(2.52)
This is easily seen by directly computing
hp|qi = 2 Ep Eq h0|ap a†q |0i
p
(2.53)
 
= 2 Ep Eq h0|a†q ap |0i + h0|[ap , a†q ]|0i
p
(2.54)
= 2Ep (2π)3 δ (3) (p − q) . (2.55)

2.3.3 Time-Dependence of the Klein–Gordon Field Operator

So far, we have not made the time dependence of the creation and annihilation operators
a†p and ap explicit. To do so, we use the Heisenberg equation of motion, which tells us
that
d
ap (t) = i[H, ap (t)] . (2.56)
dt
2
There is one situation where the absolute energy scale is relevant, and that is cosmology. The expansion
rate of the Universe depends on the total energy it contains, which includes vacuum energy (i.e. energy
that is there even if no particles exist so that all fields are in their ground states). Such vacuum
energy is what dominated the energy density of the Universe during the early phase of inflation, and
is dominating again nowadays, where it is driving the accelerated expansion of the Universe. In this
context, it is dubbed dark energy, and we have no idea what it is or what determines its magnitude.

16
 2.4 The Feynman Propagator for the Klein–Gordon Field

It is easy to show that

Z 3 0
d p
Ep0 a†p0 (t)ap0 (t), ap (t)
 
[H, ap (t)] = 3
(2.57)
(2π)
= −Ep ap (t) , (2.58)

which implies

ap (t) = e−iEp t ap . (2.59)

We again use the notation ap ≡ ap (0). In a similar way we can show that

a†p (t) = eiEp t a†p . (2.60)

Consequently, making the t-dependence explicit, the Klein–Gordon field becomes, ac-
cording to eqs. (2.34) and (2.39),

d3 p
Z
ap e−ipx + a†p eipx .


φ(x, t) = 3
p (2.61)
(2π) 2Ep

As usual, the symbols p and x denote 4-vectors, i.e. px = Ep t−px. In the term containing
a†p (t), we have substituted p → −p.
Note that the field described by eq. (2.61) contains a term with positive frequency
(proportional to e−ipx ) and a term with negative frequency (proportional toe e+ipx ).
The positive frequency term comes with an operator that destroys a positive energy
state, and the negative frequency term comes with an operator that creates a positive
energy state. So the Hilbert space contains only positive energy states, and thus has
a straightforward physical interpretation—in contrast to the wave function solutions to
the Klein–Gordon equation in relativistic quantum mechanics.

2.4 The Feynman Propagator for the Klein–Gordon Field

2.4.1 Green’s Functions of the Klein–Gordon Operator
So far, we have discussed the theory of a free (non-interacting) real scalar field. This
is admittedly a bit boring, and we would eventually like to discuss interactions among
particles. Adding interactions will make the Lagrangian more complicated. For instance
to introduce electromagnetic couplings in the canonical way, we would replace ∂µ →
∂µ +ieAµ , where e is the electromagnetic coupling constant and Aµ is the electromagnetic
4-potential. A convenient way of dealing with these extra terms is the Green’s function
method. A Green’s function D(x − y) of the Klein–Gordon operator is defined by the
requirement

(∂x2 + m2 )D(x − y) = −iδ (4) (x − y) . (2.62)

17
Chapter 2 The Klein–Gordon Field

Figure 2.1: The integration contour for the p0 integral in the Feynman propagator,
shown in the complex p0 plane. Figure taken from [1].

(In general, we should write D(x, y), but translational invariance implies that the depen-
dence can only be on x − y.) A solution of the equation

(∂x2 + m2 )φ(x) = j(x) , (2.63)

where j(x) denotes the extra terms, can then be obtained as

Z
φ(x) = d4 y iD(x − y)j(y) . (2.64)

To find D(x − y), we go to Fourier space and write

d4 p −ip(x−y)
Z
D(x − y) = e D̃(p) (2.65)
(2π)4

Equation (2.62) then becomes

d4 p −ip(x−y) d4 p −ip(x−y)
Z Z
2 2
(∂x + m ) e D̃(p) = −i e (2.66)
(2π)4 (2π)4
d4 p −ip(x−y) d4 p −ip(x−y)
Z Z
2 2
⇔ e (−p + m ) D̃(p) = −i e . (2.67)
(2π)4 (2π)4

We arrive at

d4 p ie−ip(x−y)
Z
D(x − y) = . (2.68)
(2π)4 p2 − m2

Note that in going from eq. (2.67) to eq. (2.68), we have divided by p2 − m2 , which
is only well-defined if p2 6= m2 . This, of course, need not be the case—in particular,
p2 = Ep2 − p2 = m2 is just the relativistic mass-shell condition, satisfied for “real”
particles. We therefore have to regularize eq. (2.68). We do so by shifting the poles away
from the real axis by an inifinitesimal amount . There are four ways of doing this, and
all of them lead to valid Green’s functions.

18
 2.4 The Feynman Propagator for the Klein–Gordon Field

2.4.2 The Feynman Propagator

We will seepbelow that the prescription we need in QFT is the one where the pole
at p0 p
= − p2 + m2 is shifted upwards (positive imaginary part), while the one at
p0 = p2 + m2 is shifted downwards (negative imaginary part), see fig. 2.1. This is
achieved by writing
d4 p ie−ip(x−y)
Z
DF (x − y) = . (2.69)
(2π)4 p2 − m2 + i
To see this, rewrite the expression as
d4 p ie−ip(x−y)
Z  
1 1
DF (x − y) = + .
(2π)4 2p0
p p
p0 + p2 + m2 − i0 p0 − p2 + m2 + i0
(2.70)
p
Here, 0 ≡ /(2 p2 + m2 ) is also infinitesimal. The quantity DF (x − y) is called the
Feynman propagator (hence the index F ). To carry out the complex contour integral
over p0 in eq. (2.70) explicitly, we close the integration contour in the complex plane by
a half-circle at infinity. For x0 > y 0 , the contour should be closed in the lower half plane,
for x0 < y 0 , it should be closed in the upper half plane for the exponential to go to zero
on the half circle. Using the residual theorem, we then obtain
d3 p
Z h i
−iEp (x0 −y 0 )+ip(x−y) 0 0 iEp (x0 −y 0 )+ip(x−y) 0 0
DF (x − y) = e θ(x − y ) + e θ(y − x )
(2π)3 2Ep
(2.71)
Z 3
d p h i
= e−ip(x−y) θ(x0 − y 0 ) + eip(x−y) θ(y 0 − x0 ) . (2.72)
(2π)3 2Ep
In the second term on the second line, we have substituted p → −p.

2.4.3 Relation to Correlation Functions

To find a physical interpretation for the Feynman propagator DF (x − y), we will relate
it to the 2-point correlation function
h0|φ(x)φ(y)|0i , (2.73)
which can be interpreted as the amplitude for a particle to propagate from y to x.
Indeed, since ap |0i = 0, only the a†p contributions are picked out of φ(y), i.e. φ(y)|0i
can be interpreted as creating a particle at y. Similarly, onlye the annihilation operators
contribute to h0|φ(x), so φ(x) here annihilates the particle at x. For the overall correlation
function to be nonzero, the particle must propagate from y to x. We can compute the
2-point correlation function for the free Klein–Gordon field:
D Z d3 p d3 p0
Z  E
−ipx+ip0 y † 
h0|φ(x)φ(y)|0i = 0 e a p p0 0
a (2.74)

p p
(2π)3 2Ep (2π)3 2Ep0

19
Chapter 2 The Klein–Gordon Field

d3 p
Z
= e−ip(x−y) . (2.75)
(2π)3 2Ep

Comparing to eq. (2.72), we can write

DF (x − y) = θ(x0 − y 0 )h0|φ(x)φ(y)|0i + θ(y 0 − x0 )h0|φ(y)φ(x)|0i . (2.76)

In other words, the Feynman propagator always describes the propagation of a particle
in the positive time direction. This motivates the particular choice for the shift of the
poles in eq. (2.69). We also see now why DF (x − y) deserves the name “propagator”.
A more compact way of writing DF (x−y) is obtained by introducing the time ordering
symbol T , which tells us to order any following field operators according to the zero
component of their argument in descending order:

DF (x − y) = h0|T φ(x)φ(y)|0i . (2.77)

20
 The Dirac Field
3
3.1 The Dirac Equation and its Solutions
3.1.1 The Equation and the Corresponding Lagrangian
Now that we have some understanding of how to quantize a scalar field, let us repeat the
same for fermions. Our starting point is the Dirac equation (written in covariant form)

(i∂/ − m)ψ(x) = 0 , (3.1)

where we use the notation ∂/ ≡ γ µ ∂µ , and γ µ are the Dirac matrices, which satisfy the
algebra

{γ µ , γ ν } = 2g µν . (3.2)

Here, g µν = (1, −1, −1, −1) is the Minkowski metric and {·, ·} is the anticommutator, i.e.

{γ µ , γ ν } ≡ γ µ γ ν + γ ν γ µ . (3.3)

One can get quite far without specifying a specific representation for the Dirac matrices
γ µ , but it is much easier to have one in mind. In the following, we will always use the
chiral representation, which reads

σi
   
0 0 12×2 i 0
γ = γ = . (3.4)
12×2 0 −σ i 0

Here, σ i are the Pauli matrices. Often, we will also encounter the 5-th gamma matrix

γ 5 ≡ iγ 0 γ 1 γ 2 γ 3 (3.5)
i
= − µνρσ γµ γν γρ γσ . (3.6)
4!

21
Chapter 3 The Dirac Field

In the chiral repesentation, it reads

 
5 −12×2 0
γ = . (3.7)
0 12×2

The matrix γ 5 has the property that it anticommutes with the other γ matrices, as can
be easily seen by using eq. (3.6) and the anticommutator (3.2).
Note that, by taken the Hermitian transpose of the Dirac equation and multiplying
by γ 0 from the right, we can immediately derive an equation for ψ̄ ≡ ψ † γ 0 . (We will see
shortly why it is useful to consider ψ̄ instead of simply ψ † .)

−i[∂µ ψ † (x)]γ µ† γ 0 − mψ̄ = 0 . (3.8)

This can be simplified using the identity

γ 0 γ µ† γ 0 = γ µ , (3.9)

which holds for the chiral representation (but not all representations) of the Dirac ma-
trices, as can be easily checked by direct computation. The complex conjugate Dirac
equation then becomes

−i[∂µ ψ̄(x)]γ µ − mψ̄ = 0 . (3.10)

The Lagrangian from which the Dirac equation and its complex conjugate are obtained
is

L = ψ̄(i∂/ − m)ψ . (3.11)

To check this, simply apply the Euler–Lagrange equations eq. (2.8), taking into account
that ψ(x) and ψ̄(x) are independent fields. (We could equivalently treat the real and
imaginary parts of ψ(x) as the independent degrees of freedom, but this would lead to
much more cumbersome equations.)

3.1.2 Solutions of the Dirac Equation

In relativistic quantum mechanics, the free solutions of the Dirac equation are given by

ψp,s,+ (x) = us (p)e−ipx , (positive energy solution) (3.12)

s ipx
ψp,s,− (x) = v (p)e , (negative energy solution) (3.13)

where p is an on-shell 4-momentum (i.e. p0 = Ep ) and the index s denotes the spin
orientation. The physical meaning of the negative energy solutions is a highly nontrivial
topic in relativistic quantum mechanics. Here, we will neglect it for the moment since
field theory will provide an elegant solution once we have proceeded to the quantized

22
 3.1 The Dirac Equation and its Solutions

Dirac field. The spinors us (p) and v s (p) must satisfy the relations (momentum space
Dirac equation)
s
/ − m)u (p) = 0 ,
(p
s
(p
/ + m)v (p) = 0 ,
(3.14)
ūs (p)(p
/ − m) = 0 ,
v̄ s (p)(p
/ + m) = 0
for ψp,s,+ (x) and ψp,s,− (x) to satisfy the Dirac equation eq. (3.1).
Explicitly, we have
√ √
p · σ ξs p · σ ξs
 
s s
u (p) = √ and v (p) = √ . (3.15)
p · σ̄ ξ s − p · σ̄ ξ s

We introduced here even more notation, namely we define the 4-vectors σ ≡ (12×2 , σ)
and σ̄ ≡ (12×2 , −σ). The 2-component vectors ξ 1 and ξ 2 distinguishing the two spin
orientations are simply two orthonormal basis vectors of R2 , for instance ξ 1 = (1, 0) and
ξ 2 = (1, 0). We can see that us (p) and v s (p) are indeed solutions of eq. (3.14) in the
following way:
 √
p · σ ξs
 
s 0 p·σ
p
/u (p) = p · σ̄ √
0 p · σ̄ ξ s
p 2 p · σ̄ ξ s

(p · σ)
= p
(p · σ̄)2 p · σ ξ s
p 2 (E + p · σ) ξ s

(E − p · σ)
= p
(E + p · σ)2 (E − p · σ) ξ s
p
(E 2 − 2E(p · σ) + p2 ) (E + p · σ) ξ s
 
= p 2
(E + 2E(p · σ) + p2 ) (E − p · σ) ξ s
p 3
E − 2E 2 (p · σ) + Ep2 + E 2 (p · σ) − 2Ep2 + p2 (p · σ) ξ s

= p
E 3 + 2E 2 (p · σ) + Ep2 − E 2 (p · σ) − 2Ep2 − p2 (p · σ) ξ s
 √
m √E − p · σ ξ s

=
m E + p · σ ξs
 √
m p · σ ξs

= √
m p · σ̄ ξ s
= mus (p) . (3.16)

Here, we have used multiple times the relation (p · σ)2 = pi pj σ i σ j = 12 pi pj {σ i , σ j } = p2 .

A similar derivation shows that also (p / + m)v s (p) = 0.
The spinors us (p) and v s (p) are normalized according to the Lorentz-invariant condi-
tion

ūs (p) ur (p) = −v̄ s (p) v r (p) = 2mδ rs , (3.17)

23
Chapter 3 The Dirac Field

where ūs ≡ (us )† γ 0 . Equation (3.17) can be easily checked by explicit calculation. It
is also easy to check that normalizing u† u rather than ūu would not yield a Lorentz-
invariant normalization condition. In fact,
us† (p) ur (p) = v s† (p) v r (p) = 2Ep δ rs . (3.18)
Therefore, in QFT, we almost always work with ψ̄ instead of ψ † . Note that the u and v
spinors are orthogonal:
ūs (p) v r (p) = −v̄ s (p) ur (p) = 0 . (3.19)
A similar relation does not hold for us† (p) v r (p) and v s† (p) ur (p) — these products are
in general nonzero. A relation that is, however, useful sometimes is
us† (p) v r (−p) = v s† (p) ur (−p) = 0 . (3.20)
(We write the argument of the u and v spinors as 3-vectors here to emphasize the sign
of the 3-momentum. It is of course implied that the 0-component of the momentum
4-vector is set by the relativistic energy–momentum relation.)

3.1.3 Spin Sums

We will often deal with systems in which the polarization of the incoming particles is
random, and the polarization of the outgoing particles is not measured. In these cases,
we need to sum over polarizations, which leads to spin sums of the form
X
us (p)ūs (p) (3.21)
s=1,2

Plugging in eq. (3.15) for us (p) and using s=1,2 ξ s ξ s† = 12×2 , we obtain
P

X X √p · σξ s  √ √
s s
u (p)ū (p) = √ s (ξ s† p · σ̄, ξ s† p · σ) (3.22)
p · σ̄ξ
s=1,2 s=1,2
√ √ √ √ 
p · σ p · σ̄ p·σ p·σ
= √ √ √ √ (3.23)
p · σ̄ p · σ̄ p · σ̄ p · σ
 
m p·σ
= (3.24)
p · σ̄ m
=p
/ + m. (3.25)
In the third equality, we have used
(p · σ)(p · σ̄) = (p0 − pi σ i )(p0 + pi σ i ) (3.26)
= (p0 )2 − (pi )2 (3.27)
2
=m . (3.28)
In analogy to eq. (3.25), we can also show that
X
v s (p)v̄ s (p) = p
/ − m. (3.29)
s=1,2

24
 3.2 Quantization of the Dirac Field

3.2 Quantization of the Dirac Field

3.2.1 How Not to Quantize the Dirac Field
The naı̈ve way of quantizing the Dirac field would be to proceed in analogy to section 2.3
and postulate the commutation relations

[ψa (x, t), ψb† (y, t)] = δ (3) (x − y)δab ,

(3.30)
[ψa (x, t), ψb (y, t)] = [ψa† (x, t), ψb† (y, t)] = 0

where a, b = 1 · · · 4 are spinor indices. Note that, again, we take the time coordinate
of the two fields to be the same. Let us try to quantize the Dirac field based on these
commutation relations and see what goes wrong.
We write the field ψ(x) as a superposition of all solutions of the free Dirac equation,
with operator-valued coefficients:

d3 p
Z X
s s −ipx s s ipx


ψ(x) = p a p u (p)e + bp v (p)e . (3.31)
(2π)3 2Ep s=1,2

Note that, unlike for the real Klein–Gordon field, the coefficients of eipx and e−ipx are
not related to each other for the Dirac field, which is complex. If the creation and
annihilation operators asp and bsp satisfy the commutation relations

[arp , as† r s† 3 (3) rs

q ] = [bp , bq ] = (2π) δ (p − q) δ , (3.32)

(and all other commutators being zero) we can show that ψ and ψ † satisfy the postulated
relations eq. (3.30):

d3 p d3 q
Z X
†
[ψ(x, t), ψ (y, t)] = 6
p p eipx−iqy
(2π) 2Ep 2Eq r,s
 
s†
× [arp , as†
q ]u r
(p)ūs
(q) + [br
,
−p −qb ]v r
(−p)v̄ s
(−q) γ0 (3.33)
d3 p
Z
eip(x−y) γ 0 Ep − γ i p + m + γ 0 Ep + γ i p − m γ 0


= 3
(2π) 2Ep
(3.34)
= δ (3) (x − y) 14×4 . (3.35)

Let us look at the quantized Hamilton operator, which is as usual obtained from a
Legendre transform of the Lagrangian:
Z  
3 δL
H= d x ∂0 ψ − L (3.36)
δ(∂0 ψ)
Z
= d3 x iψ̄γ 0 ∂0 ψ − ψ̄(i∂/ − m)ψ
 
(3.37)

25
Chapter 3 The Dirac Field

Z
d3 x ψ̄ − i(γ · ∇) + m ψ
 
= (3.38)

d3 p d3 q
Z Z Z Xh
−ipx
= d3 x as† s ipx
+ bs† s


3
p
3
p p ū (p)e p v̄ (p)e
(2π) 2Ep (2π) 2Eq s,r

i
· − i(γ · ∇) + m arq ur (q)e−iqx + brq v r (q)eiqx
 
(3.39)
d3 p
Z h
as† s

r r
= 3 p ū (p) γ · p + m ap u (p)
(2π) 2Ep
+ as† s

r r
p ū (p) γ · p + m b−p v (−p)
+ bs† s

r r
p v̄ (p) − γ · p + m a−p u (−p)

r r i
+ bs†
p v̄ s
(p) − γ · p + m bp u (p) (3.40)
d3 p X 
Z
Ep as† s s† s

= 3 p ap − Ep bp bp . (3.41)
(2π) s

In the fourth equality, we have plugged in eq. (3.31) and its complex conjugate. In the
sixth equality, we have used γ · p = −p / + Ep γ 0 and we have then invoked the momen-
tum space Dirac equation, eq. (3.14), as well as the orthogonality relations eqs. (3.18)
and (3.20).
We encounter a serious problem here: if we interpret the operators as† s s† s
p ap and bp bp as
particle number operators again, we are led to the conclusion that, the more particles we
create using the operator bs†
p , the lower the energy becomes. This seems like a terribly
unstable system.
A discussion of additional problems with the above procedure is given in [1].

3.2.2 Quantizing the Dirac Field with Anticommutators

The goal of avoiding negative energy states can lead us to the proper way of quantizing
the Dirac field.
A first attempt might be to replace bsp by bs†
p in eq. (3.31). This seems reasonable since
also for the Klein–Gordon field, the negative frequency term in the field operator came
with a creation operator rather than an annihilation operator. In this case, however,
the second term in eq. (3.34) would get a minus sign, and we would not reproduce the
desired commutation relations eq. (3.30). Also, the expression for H, eq. (3.41), would
change:

d3 p X 
Z
Ep as† s s s†

H= 3 p ap − Ep bp bp (3.42)
(2π) s
d3 p X 
Z
Ep as† s s† s 3 (0)

= 3 p ap − Ep bp bp − Ep (2π) δ (0) . (3.43)
(2π) s

26
 3.2 Quantization of the Dirac Field

Thus, H changes by an (infinite) constant from the commutator of bs† s

p and bp , but the
s† s
minus sign in the term proportional to bp bp would remain. The infinite constant is not
a problem because, again, experiments measure only energy differences, so H can be
renormalized to a finite expression by simply dropping the infinite constant.
Since the commutators eq. (3.30) didn’t lead us anywhere, let us give up on them.
Instead, note that the negative energy problem in the Hamiltonian is solved if we replace
bsp by bs†
p in eq. (3.31) and postulate in addition that the field operators anticommute.
This means we write
d3 p
Z X
s s −ipx s† s ipx


ψ(x) = p a p u (p)e + b p v (p)e , (3.44)
(2π)3 2Ep s=1,2
d3 p
Z X
s s −ipx s† s ipx


ψ̄(x) = p bp v̄ (p)e + a p ū (p)e (3.45)
(2π)3 2Ep s=1,2

and postulate

{arp , as† r s† 3 (3) rs

q } = {bp , bq } = (2π) δ (p − q) δ ,
{arp , asq } = {brp , bsq } = {arp , bsq } = 0 ,
(3.46)
{ar† s† r† s† r† s†
p , aq } = {bp , bq } = {ap , bq } = 0 ,
{arp , bs† r s†
q } = {bp , aq } = 0 .

Then, the Hamilton operator is

d3 p X 
Z
Ep as† s s s†

H= 3 p ap − Ep bp bp (3.47)
(2π) s
d3 p X 
Z
s† s s† s

= E p ap ap + Ep bp bp . (3.48)
(2π)3 s

(We have already dropped the infinite constant.) The field operators ψ(x, t) and ψ † (y, t)
now also satisfy anticommutation relations:
d3 p d3 q
Z X
† ipx−iqy
{ψ(x, t), ψ (y, t)} = p p e
(2π)6 2Ep 2Eq r,s
 
r†
× {arp , as†
q }u r
(p)ūs
(q) + {b ,
−p −q bs
}v r
(−p)v̄ s
(−q) γ 0 (3.49)
d3 p
Z
eip(x−y) γ 0 Ep − γ i p + m + γ 0 Ep + γ i p − m γ 0


= 3
(2π) 2Ep
(3.50)
= δ (3) (x − y) 14×4 . (3.51)

As usual, the vacuum is defined to be the state that satisfies

asp |0i = bsp |0i = 0 . (3.52)

27
Chapter 3 The Dirac Field

The interpretation of the operators asp and bsp is that both of them create (distinct)
particles with momentum p and positive energy Ep .

3.2.3 Physical Significance of the Quantized Dirac Field

Particles and Antiparticles
We see that fermions always come in pairs. We call the one created by as† p particle,
the one created by bs†
p antiparticle. (This is arbitrary—we could also interchange these
definitions.) Note that the Hamiltonian eq. (3.48) tells us that particles and antiparticles
have the same energy spectrum. In particular, they have identical mass.

The Pauli Exclusion Principle

The fact that the creation and annihilation operators anticommute allows us to derive
another important result, namely the Pauli exclusion principle, which states that no two
fermions can be in the same quantum state. In fact, consider a state
|p, si = 2Ep as†
p
p |0i . (3.53)
Assume we want to create another particle in the same state. Then, the operator that
does this must contain a†p , and we would obtain an amplitude proportional to (a†p )2 |0i
The anticommutation relation {a†p , a†q } = 0, however, tells us that (a†p )2 vanished. There-
fore, processes that create a particle in a state that is already occupied have zero ampli-
tude.

3.3 The Feynman Propagator for the Dirac Field

As for the Klein–Gordon field, we can derive Green’s functions for the Dirac operator,
i.e. functions that satisfy
 
∂
i(γ µ )ab µ − mδab Sbc (x − y) = iδ (4) (x − y)δac , (3.54)
∂x
where a, b and c are spinor indices, which we have written out explicitly here. We go to
Fourier space and write
d4 p −ip(x−y)
Z
Sac (x − y) = e S̃ac (p) . (3.55)
(2π)4
Plugging this into eq. (3.54), we find
 µ 
(γ )ab pµ − mδab S̃bc (p) = iδac . (3.56)
Thus we can formally write
i
S̃(p) = , (3.57)
p −
/ m

28
 3.3 The Feynman Propagator for the Dirac Field

where the 4 × 4 matrix (in spinor space) in the denominator means the inverse of that
matrix. If one wishes to avoid matrix inverses, one can equivalently write
i(p
/ + m)
S̃(p) = , (3.58)
p2 − m2
using that p
/p/ = p2 . With this S̃(p), eq. (3.55) is infinite, so we again need to introduce
a regularization scheme. As for the Klein–Gordon field, and by the same arguments, we
use the Feynman prescription for shifting the poles away from the real axis. This defines
the Feynman propagator for the Dirac field :
d4 p i(p
/ + m) −ip(x−y)
Z
SF (x − y) = e . (3.59)
(2π) p − m2 + i
4 2

Like the Klein–Gordon propagator, also the Feynman propagator can be related to
correlation functions. Indeed note that, by integrating over p0 , we can write
d3 p
Z h i
−ip(x−y) 0 0 ip(x−y) 0 0
SF (x − y) = e ( p
/ + m)θ(x − y ) + e (−p/ + m)θ(y − x ) .
(2π)3 2Ep
(3.60)
On the other hand, we also have
d3 p
Z X
h0|ψ(x)ψ̄(y)|0i = 3
us (p)ūs (p)e−ip(x−y) (3.61)
(2π) 2Ep s
d3 p
Z
−ip(x−y)
= (p
/ + m)e . (3.62)
(2π)3 2Ep
and
d3 p
Z X
h0|ψ̄(y)ψ(x)|0i = v s (p)v̄ s (p)eip(x−y) (3.63)
(2π)3 2Ep s
d3 p
Z
ip(x−y)
= / − m)e
(p . (3.64)
(2π)3 2Ep
Thus, we find again, similar to what we found for the Klein–Gordon field,
SF (x − y) = h0|ψ(x)ψ̄(y)|0i θ(x0 − y 0 ) − h0|ψ̄(y)ψ(x)|0i θ(y 0 − x0 ) (3.65)
= h0|T ψ(x)ψ̄(y)|0i . (3.66)
For the Klein–Gordon field, there was a plus sign between the two terms. The fact that
we now find a minus sign is a reflection of the anticommutating nature of fermion fields.
In the second line of eq. (3.66), we have extended the definition of the time ordering
symbol T to fermion fields. It is implied that any interchange of adjacent fermion field
operators that is necessary to bring the field order into time order contributes a minus
sign.
The physical interpretation of the Feynman propagator for Dirac fields is the following:
it describes the propagation of a particle from y to x (if x0 > y 0 ), or the propagation of
an antiparticle from x to y (if y 0 > x0 ).

29
Chapter 3 The Dirac Field

Figure 3.1: An active transformation (in this case a rotation) of a field configuration.
Figure taken from [1].

3.4 Symmetries of the Dirac Theory

Symmetries are one of the most fundamental concepts in particle physics: the Noether
theorem tells us that symmetries are related to conserved quantities, and conserved
quantities are extremely useful in solving the equations of motion for particles and fields.
More importantly still, symmetries dictate which terms are allowed in a Lagrangian and
which are not. This will be crucial when we discuss interacting field theories, where the
symmetry structure determines which interactions can exist.

3.4.1 Lorentz Invariance

Transformation Law for Spinor Fields
At the foundation of any relativistic theory should of course be Lorentz invariance, i.e.
the principle that the equations of motion must be the same in any inertial frame. Let
us consider a Lorentz transformation1

xµ → xµ0 = Λµν xν . (3.67)

We write the transformation of the spinor field ψ(x) under this transformation as

ψ(x) → ψ 0 (x) = S(Λ)ψ(Λ−1 x) , (3.68)

where S(Λ) is a linear transformation matrix, yet to be determined. (The linearity of

the transformation is for now an ansatz, but we will see shortly that it works.)
1
We work here with active transformations, i.e. we assume the field configuration to be rotated and/or
boosted, see fig. 3.1. For a passive transformation, where the coordinate system is redefined instead,
we would have to replace Λ by Λ−1 everywhere.

30
 3.4 Symmetries of the Dirac Theory

Lorentz invariance of the Dirac equation means that

∂ 0
iγ µ ψ (x) − mψ 0 (x) = 0 . (3.69)
∂xµ
Plugging in the transformation property of ψ(x), eq. (3.68), this becomes
∂
iγ µ S(Λ)ψ(Λ−1 x) − m S(Λ)ψ(Λ−1 x) = 0 . (3.70)
∂xµ
We proceed by making the substitution x → Λx, and obtain
ν ∂
iγ µ (Λ−1 )µ S(Λ)ψ(x) − m S(Λ)ψ(x) = 0 . (3.71)
∂xν
This can only be fulfilled if the γ-matrices transform according to

S −1 (Λ)γ µ S(Λ) = Λµν γ ν . (3.72)

This consistency condition basically says that the index µ on γ µ can indeed be treated
like any other Lorentz index.
Let us now construct S(Λ) explicitly. To do so, consider an infinitesimal Lorentz
transformation and write Λµν as

Λµν = g µν + ω µν ,
µ (3.73)
(Λ−1 ) ν = g µν − ω µν ,

where ω µν is infinitesimal. We know from the explicit form of Λµν , familiar from special
relativity, that ω µν is antisymmetric.
We can write S(Λ) and its inverse as
i
S(Λ) = 1 − σµν ω µν ,
4 (3.74)
−1 i
S (Λ) = 1 + σµν ω µν ,
4
with a yet-to-be-determined tensor σµν . Since ω µν is antisymmetric in µν, so must σµν
be as well. Plugging the infitesimal forms of Λµν and S(Λ), eqs. (3.73) and (3.74), into
the transformation law for γ µ , eq. (3.72), we find
i i
σαβ ω αβ γ µ − γ µ σαβ ω αβ = ω µν γ ν , (3.75)
4 4
or, equivalently,

ω αβ [σαβ , γ µ ] = −4iω αβ gα µ γβ = −2iω αβ (gα µ γβ − gβ µ γα ) . (3.76)

In the last step, we have used the antisymmetry of ω αβ . Since eq. (3.76) must be satisfied
for any ω αβ , it reduces to

[σαβ , γ µ ] = −2i(gα µ γβ − gβ µ γα ) . (3.77)

31
Chapter 3 The Dirac Field

We can show by explicit calculation that this is satisfied for

i
σ αβ ≡ [γ α , γ β ] . (3.78)
2
Together with eq. (3.74), this defines the action of an infinitesimal Lorentz transformation
on a spinor field. A finite transformation then has the form
µν
S(Λ) = e−(i/4)σµν ω , (3.79)

where ω µν is now finite.

The Generators of Lorentz Transformations

We recall from quantum mechanics that spatial rotations are generated by the angular
momentum operator L = x × (−i∇). This means that, for an infinitesimal rotation
described by a rotation matrix Rij = δ ij +rij , a wave function χ(x) transforms according
to

χ(x, t) → χ0 (x, t) = χ(R−1 x, t) (3.80)

ij j i
= χ(x, t) − r x ∇ χ(x, t) . (3.81)

If R describes a rotation around the z axis by an infinitesimal angle α,

 
0 α
r = −α 0  , (3.82)
0

this means for instance

χ(x, t) → χ0 (x, t) = χ(x, t) + α(x1 ∇2 − x2 ∇1 )χ(x, t) (3.83)

= (1 + iαL3 )χ(x, t) . (3.84)

A rotation about the z axis by a finite angle α is then given by

3
χ(x, t) → χ0 (x, t) = eiαL χ(x, t) . (3.85)

In an analogous way, also the action of a Lorentz transformation on a spinor field ψ(x),
given by eq. (3.68), can be expressed in terms of a generating operator Jµν . Consider
again an infinitesimal Lorentz transformation as in eq. (3.73) and write

ψ 0 (x) = S(Λ)ψ(Λ−1 x) (3.86)

 
i µν
= 1 − Jµν ω ψ(x) . (3.87)
2

32
 3.4 Symmetries of the Dirac Theory

(The factor i/2 is mere convention.) Plugging in our expressions for S(Λ) and Λ in the
infinitesimal case, eqs. (3.73) and (3.74), we obtain
   
i i
1 − σµν ω µν ψ(xµ − ω µν xν ) = 1 − Jµν ω µν ψ(x) . (3.88)
4 2
This leads to
i i
− σµν ω µν − ω µν xν ∂µ = − Jµν ω µν , (3.89)
4 2
or
1
Jµν = σµν + i(xµ ∂ν − xν ∂µ ) . (3.90)
2
In the last step, we have used the antisymmetry of ω µν . Note that the Jµν are block
diagonal in the chiral basis for the Dirac matrices. This follows from the definition of
σµν , eq. (3.78), together with the explicit form of the γ µ in the chiral basis, eq. (3.2). It
means that the spinor representation of the Lorentz group is reducible: if the upper or
lower two components in a 4-spinor are zero, they will remain zero under any Lorentz
transformation.

Spin
We know that angular momentum is the conserved quantity associated with symmetry
under spatial rotations. Therefore, we can now use the transformation properties of the
field operator under such rotations to derive the form of the angular momentum operator
for a fermion field. This will in particular help us better understand the internal angular
momentum (spin) of fermions.
Under an infinitesimal Lorentz transformation (of which spatial rotations are of course
a special case), a fermion field transforms as

ψ(x) → ψ(x) + δψ(x) (3.91)

with

δψ(x) ≡ S(Λ)ψ(Λ−1 x) − ψ(x) (3.92)

(see eq. (3.68)). Using the explicit form eq. (3.74) for S(Λ) and eq. (3.73) for Λ, this
becomes
i ∂
δψ(x) = − σµν ω µν ψ(x) − ω µν xν µ ψ(x) . (3.93)
4 ∂x
We now specialize to a rotation about the z axis (ω 12 = −ω21 = θ, all other ω µν = 0),
and use that
i
σ 12 = [γ 1 , γ 2 ] (3.94)
2

33
Chapter 3 The Dirac Field

i −σ 1 σ 2 + σ 2 σ 1
 
= (3.95)
2 −σ 1 σ 2 + σ 2 σ 1
 3 
σ
= (3.96)
σ3
≡ Σ3 . (3.97)

Here, we have used the relation [σ a , σ b ] = 2iabc σ c . We get

 
i 3 ∂ ∂
δψ(x) = − Σ θψ(x) − x2 1 − x1 2 θψ(x) ≡ θ ∆ψ . (3.98)
2 ∂x ∂x
The associated conserved quantity (integral of the time-component of the associared
Noether current) is
Z Z Z  
3 0 3 δL 3 0 1 ∂ 2 ∂ i 3
d xj = d x ∆ψ = −i d x ψ̄γ x −x + Σ ψ(x) .
δ(∂0 ψ) ∂x2 ∂x1 2
(3.99)

The right hand side is thus the third component of the angular momentum operator for a
fermion field. In exact analogy, also the other two components of the angular momentum
operator can be found, leading to
Z
J = d3 x ψ † x × (−i∇) + 21 Σ ψ .
 
(3.100)

The first term in square brackets is just the angular momentum operator in quantum
mechanics (dressed here with two field operators). For non-relativistic fermions, it can be
interpreted as giving the orbital angular momentum. The second term in square brackets
gives the internal angular momentum (spin). For relativistic fermions, this division into
spin and orbital angular momentum is not so trivial due to spin–orbit coupling.
Nevertheless, eq. (3.100) allows us to check that a Dirac fermion had indeed spin 1/2.
For this, it is sufficient to consider particles at rest (or very nearly so), so that the orbital
angular momentum term is negligible. In this case, we can plug the Fourier expansion
of ψ(x), eq. (3.44) into eq. (3.100) (omitting the orbital angular momentum term):

d3 p d3 p0
Z Z
0
j
J = d x 3
6
p p e−ip x eipx (3.101)
(2π) 2Ep 2Ep0
X  r0 † 0 0 0
 Σj  
· ap0 ur † (p0 ) + br−p0 v r † (−p0 ) arp ur (p) + br†
−p v r
(−p) . (3.102)
0
2
r,r

To apply this operator (or specifically its third component) to a one-particle state of zero
momentum, as† z
0 |0i, we use the fact that the vacuum has zero spin, i.e. J |0i = 0 and
therefore

J z as† z s†
0 |0i = [J , a0 ]|0i . (3.103)

34
 3.4 Symmetries of the Dirac Theory

We can then use the commutation relation [ar† r s† 3 (3) r† rs

p ap , a0 ] = (2π) δ (p) a0 δ , which is
easily derived using the canonical anticommutators, as well as the explicit form eq. (3.15)
of the u spinor, to find
1 s† Σ3 s
J z as†
0 |0i = u (0) u (0)as†0 |0i (3.104)
2m 2
σ3
= ξ s† ξ s as†
0 |0i . (3.105)
2
(All other commutators of creation and annihilation operators appearing in eq. (3.103)
vanish.) This shows that, for ξ s ∝ (1, 0), the particle has spin +1/2 along the z axis, for
ξ s ∝ (0, 1), it has spin −1/2 along the z axis.
We can do a similar derivation for an antiparticle state bs† 0 |0i. In this case, the only
r r† s† 3 (3) r† rs
relevant commutator is [bp bp , b0 ] = −(2π) δ (p) b0 δ . Note the extra minus sign!
We thus find
σ 3 s s†
J z bs†
0 |0i = −ξ
s†
ξ b0 |0i . (3.106)
2
Therefore, for antiparticles, the association between ξ s spinors and the orientation of the
physical angular momentum is reversed: for ξ s ∝ (1, 0), the particle has spin −1/2 along
the z axis, for ξ s ∝ (0, 1), it has spin +1/2 along the z axis.

Helicity, Chirality and Weyl Fermions

Let us consider an ultrarelativistic particle or antiparticle moving in the positive z di-
rection with momentum pz and spin up along the z axis, i.e. ξ s = (1, 0). (Note that a
particle moving along a line through the origin, such as a coordinate axis, has no orbital
angular momentum.) The u and v spinors for such a particle or antiparticle are
p

3 1
!

 E − p z σ 0 Em √ 0

   ' 2E
for spin up along the z axis
1

 pE + p σ 3 1


0

 z 0
u(p) = p

E − p σ 3 0
!
Em √


 z 1 0
1

 ' 2E for spin down along the z axis



0
 p

 E + pz σ 3 0 1
(3.107)
 p

E − pz σ 3 0
!

 1 Em √ 0
1

 ' 2E for spin up along the z axis



0
p
0
 − E + pz σ 3


1

v(p) =  p

E − pz σ 3 1
!
√

0

 0 Em

 ' 2E for spin down along the z axis




 p
1 − 1
 − E + p σ3

z 0 0

(3.108)

35
Chapter 3 The Dirac Field

We say that a particle whose spin is pointing in the direction of motion has a right-
handed (RH) helicity and a particle with its spin is pointing in the opposite direction
has a left-handed (LH) helicity.
We observe that, in the ultrarelativistic limit, for spinors for LH particles and RH
antiparticles, the lower components are zero, while for RH particles and LH antiparticles
the upper components vanish. This shows again that the upper and lower components of
a spinor are quite independent: they correspond to different spin orientations. We have
argued previously, below eq. (3.90), on more formal grounds that the upper and lower
components of a Dirac spinor are independent because the 4-dimensional spinor represen-
tation of the Lorentz group is reducible. It factorizes into two irreducible representations,
one describing the transformation properties of the upper two components of the Dirac
spinor, one describing the transformation properties of the lower two components.
It also makes sense that LH particles and RH antiparticles are in the same irreducible
representation of the Lorentz group. In the Dirac hole theory, an antiparticle corresponds
to the absence of a particle from the Dirac sea. If a left-handed particle is absent, the hole
it leaves back effectively has opposite spin and corresponds to a right-handed antiparticle.
To highlight the factorization of the Lorentz representation, one sometimes splits up
a Dirac spinor field into
 
ψL (x)
ψ(x) = (3.109)
ψR (x)

and works with the left-chiral Weyl fermion ψL (x) and the right-chiral Weyl fermion
ψR (x) separately. Note that, according to eqs. (3.107) and (3.108), in the ultrarelativistic
limit the Weyl fermions are helicity eigenstates, justifying the attribute “left” and “right”.
(Even though a left-chiral field corresponds to a LH particle, but a RH antiparticle.) In
terms of the Weyl fields the Dirac Lagrangian eq. (3.11) then becomes
†
L = ψL† (iσ̄ · ∂)ψL + ψR (iσ · ∂)ψL − mψ̄L ψR − mψ̄R ψL . (3.110)

We see that it is only the mass term that mixes the left- and right-chiral fields, thereby
destroying the one-to-one correspondence between chirality and helicity. This is a reflec-
tion of the fact that for massive fermions, helicity is a frame-dependent quantity. One
can always boost into a frame where the momentum of the particle is reversed.
When working with Weyl spinors, it is convenient to define the projection operators

1 − γ5 1 + γ5
PL ≡ PR ≡ , (3.111)
2 2
which project out the left-chiral and right-chiral components, respectively, of a Dirac
spinor. They satisfy PL/R2 = PL/R and PL PR = 0.
Weyl fermions play a crucial role in the theory of weak interactions, where the couplings
of left- and right-chiral fields are completely different.

36
 3.4 Symmetries of the Dirac Theory

3.4.2 Parity (P )
In addition to the continuous Lorentz transformations discussed above, there are several
discrete symmetries under which many quantum field theories are invariant. The first of
these is parity or space inversion which transforms a Lorentz 4-vector (t, x) into (t, −x).
A parity transformation thus reverses the 3-momentum of a particle, but leaves its spin
orientation unchanged. To see this, remember that orbital angular momentum L is a cross
product of x and p, and since both of these change sign under parity transformations,
L remains invariant. Spin and orbital angular momentum must transform the same
way if spin is to be interpreted as a form of angular momentum, therefore also spin
must remain invariant under parity. Consequently, the parity operator must transform
a quantum state asp |0i into as−p |0i. This implies that the parity operation P in Hilbert
space must act as
P asp P = ηa as−p and P bsp P = ηb bs−p , (3.112)
where ηa and ηb are phase factors. We are in principle free to choose these phase factors
arbitrarily—they are not restricted by the requirement that parity reverses space. They
are, however, constrained if we demand that applying the parity operator twice should
leave any physical observable unchanged. Since observables (such as the Hamiltonian)
are constructed from even numbers of fermion field operators, this implies ηa , ηb = ±1.
When we discussed continuous Lorentz transformations, we found that their action of
the field operator can be written as a multiplication by a matrix S(Λ) in spinor space, see
eq. (3.68). We will now find a similar transformation law also for parity transformations.
We write
d3 p
Z
ηa as−p us (p)e−ipx + ηb∗ bs†
X
s ipx


P ψ(x)P = 3
p −p v (p)e . (3.113)
(2π) 2Ep
We now substitute p = (p0 , p) → p0 ≡ (p0 , −p) in the integral. In applying this
substitution, we must express us (p) and v s (p) in terms of p0 . To do so, we use the explicit
form of these spinors (see eq. (3.15)) and observe that p · σ = p0 · σ̄ and p · σ̄ = p0 · σ.
This leads to
√
p · σξ s p · σ̄ξ s
 √ 0 
us (p) = √ = √ = γ 0 us (p0 ) , (3.114)
p · σ̄ξ s p0 · σξ s
√  √ 0
p · σξ s s

p · σ̄ξ
s
v (p) = √ = √ = −γ 0 v s (p0 ) . (3.115)
− p · σ̄ξ s − p0 · σξ s
With these relations, we arrive at
d3 p0
Z
s 0 s 0 −ip0 x0 ∗ s† 0 s 0 ip0 x0
X

P ψ(x)P = p η a ap0 γ u (p )e − ηb bp0 γ v (p )e . (3.116)
(2π)3 2Ep0
where we have defined x0 ≡ (x0 , −x). We now choose ηb∗ = −ηa . Then, we have the
transformation law
P ψ(x)P = ηa γ 0 ψ(x0 ) . (3.117)

37
Chapter 3 The Dirac Field

ψ̄ψ iψ̄γ 5 ψ ψ̄γ µ ψ ψ̄γ µ γ 5 ψ ψ̄σ µν ψ

P +1 −1 (−1)µ −(−1)µ (−1)µ (−1)ν
T +1 −1 (−1)µ (−1)µ −(−1)µ (−1)ν
C +1 +1 −1 +1 −1

Table 3.1: Transformation properties of Dirac field bilinearies under parity (P ), time
reversal (T ) and charge conjugation (C). The shorthand notation (−1)µ means +1 for
µ = 0 and −1 for µ = 1, 2, 3.

In computing physical observables and in dealing with interacting field theories, we

often encounter bilinearies of fermion fields, such as

ψ̄ψ , iψ̄γ 5 ψ , ψ̄γ µ ψ , ψ̄γ µ γ 5 ψ , ψ̄σ µν ψ . (3.118)

Let us discuss the transformation propertiers of these expressions under parity. Since
ψ̄(x) appears in all of them, we need the result that

P ψ̄(x)P = P ψ † (x)P γ 0 = ηa∗ ψ † (x)(γ 0 )2 = ηa∗ ψ̄(x)γ 0 . (3.119)

We then find for instance

P ψ̄(x)ψ(x)P = |ηa |2 ψ̄(x0 )ψ(x0 ) = ψ̄(x0 )ψ(x0 ) . (3.120)

Similarly, we can show that

(
µ 0 +ψ̄(x0 )γ µ ψ(x0 ) for µ = 0
0 µ 0 0
P ψ̄(x)γ ψ(x)P = ψ̄(x )γ γ γ ψ(x ) = . (3.121)
−ψ̄(x0 )γ µ ψ(x0 ) for µ = 1, 2, 3

In other words, ψ̄(x)γ µ ψ(x) transforms like a Lorentz vector. The transformation prop-
erties of the other bilinearies can be derived in a similar way. They are summarized in
table 3.1.

3.4.3 Time Reversal (T )

Next, we consider the time reversal operation that sends (t, x) to (−t, x). A momentum
3-vector (which is a derivative of a coordinate 3-vector with respect to time) transforms
under T as p → −p. Since coordinate 3-vectors are invariant under T while momentum
3-vectors change sign, also orbital angular momentum L = x × p, and thus also spin,
changes sign.
This means that T acts on asp and bsp according to

T asp T = a−s
−p and T bsp T = b−s
−p , (3.122)

In principle, we should again allow for arbitrary phase factors, but since they would not
affect the following derivations, we omit them here. This is, however, not the full story

38
 3.4 Symmetries of the Dirac Theory

yet: given some quantum mechanical transition amplitude hψ1 |ψ2 i, time reversal should
interchange the initial and final states, i.e. hT ψ1 |T ψ2 i = hψ2 |ψ1 i. (Such an operator is
called antiunitary.)
This implies that T not only acts on Hilbert space states, but also on complex numbers
by sending them to their complex conjugate. We thus have
d3 p
Z
−s† s∗
X
−s s∗ +ipx −ipx


T ψ(x)T = p a−p u (p)e + b−p v (p)e . (3.123)
(2π)3 2Ep
To write the right hand side as a linear transformation of ψ(x), we need a way of writing
us (p) in terms of u−s (p0 ), where p0 = (p0 , −p).
To do so, we first note again that p0 · σ = p · σ̄ and p0 · σ̄ = p · σ. Moreover, note that
a spin flip (sending s to −s) is achieved by the transformation

ξ s → ξ −s = −iσ 2 (ξ s )∗ . (3.124)

(We are not worrying about the inconsequential phase factors here.) To see this, assume
ξ s describes a spin along a unit vector n, i.e. ξ s is an eigenvector of the helicity operator
n · σ:

n · σξ s = +ξ s . (3.125)

Then, using σ i σ 2 = −σ 2 σ i∗ ,

(n · σ)[−iσ 2 (ξ s )∗ ] = −iσ 2 (−n · σ)∗ (ξ s )∗ (3.126)

2 s ∗
= −(−iσ (ξ ) ) . (3.127)

Therefore, we have
√
p · σ ξs

s √
u (p) = (3.128)
p · σ̄ ξ s
2 (ξ −s )∗ ]
√ 0 
p · σ̄ [−iσ
= √ 0 . (3.129)
p · σ [−iσ 2 (ξ −s )∗ ]

We would like to move the σ 2 matrices to the left of the square roots. To √ this end,
go to a reference frame where p is aligned along the z-axis. In this frame p0 · σ =
1−σ 3 1+σ 3
i 2 2 i∗
p p
Ep0 + |p0 |( 2 ) + Ep0 − |p0 |( 2 ). Therefore, using again σ σ = −σ σ , we find
p p
p0 · σ σ 2 = σ 2 p0 · σ̄ ∗ . (3.130)

Similarly,
p p
p0 · σ̄ σ 2 = σ 2 p0 · σ ∗ . (3.131)

This leads us to the relation

√
−iσ 2 √p0 · σ ∗ (ξ −s )∗
 
s
u (p) = (3.132)
−iσ 2 p0 · σ̄ ∗ (ξ −s )∗

39
Chapter 3 The Dirac Field

σ2
 
= −i [u−s (p0 )]∗ (3.133)
σ2
= −γ 1 γ 3 [u−s (p0 )]∗ . (3.134)

An analogous relation holds for v s (p):

v s (p) = −γ 1 γ 3 [v −s (p0 )]∗ . (3.135)

We are now ready to rewrite eq. (3.123):

d3 p0
Z
−ip0 x0 s† s ip0 x0
X
T ψ(x)T = −γ 1 γ 3 s s


p ap0 u (p)e + bp0 v (p)e (3.136)
(2π)3 2Ep0
= −γ 1 γ 3 ψ(x0 ) . (3.137)

Here, we have defined x0 = (−t, x), so that px = −p0 x0 .

As in section 3.4.2, we can again check the behavior of Dirac field bilinearies under
the T transformation. The results are summarized in table 3.1.

3.4.4 Charge Conjugation (C)

The final discrete symmetry we wish to consider here is charge conjugation C, under
which particles become antiparticles and vice-versa. In other words

Casp C = bsp and Cbsp C = asp . (3.138)

To write the action of C on the field ψ(x) as a linear operation as we did for the P and T
transforms, we will need a transformation that converts a u spinor into a v spinor. Here
it is:
√
p · σ ξs

s √
v (p) = (3.139)
− p · σ̄ ξ s
√
p · σ [−iσ 2 ξ −s ]∗

= √ (3.140)
− p · σ̄ [−iσ 2 ξ −s ]∗
√
−iσ 2 p · σ̄ ∗ (ξ −s )∗
 
= √ (3.141)
iσ 2 p · σ ∗ (ξ −s )∗
 √ ∗
0 −iσ 2 p · σ ξ −s

= √ (3.142)
iσ 2 0 p · σ̄ ξ −s
0 −iσ 2
 
= [us (p)]∗ (3.143)
iσ 2 0
= −iγ 2 [us (p)]∗ . (3.144)

In the second equality, we have used eq. (3.124). Note that in the last step we have us (p)
and not u−s (p) because we have seen in section 3.4.1 that the association between the

40
 3.4 Symmetries of the Dirac Theory

ξ s 2-spinors and the physical spin orientation is opposite for u and v spinors: ξ s = (1, 0)
corresponds to spin up along the z axis for a u spinor, but to spin down along the z axis
for a v spinor. Similarly, we also have

us (p) = −iγ 2 [v s (p)]∗ . (3.145)

(Note that to obtain eq. (3.145) directly from eq. (3.144), one has to take into account
that, in our conventions from eq. (3.124), ξ −(−s) = −ξ s .) Thus,

d3 p
Z X
bsp us (p)e−ipx + as† s ipx


Cψ(x)C = 3
p p v (p)e (3.146)
(2π) 2Ep
d3 p
Z X
2
bsp [v s (p)]∗ e−ipx + as† s ∗ ipx


= −iγ 3
p p [u (p)] e (3.147)
(2π) 2Ep
= −iγ 2 [ψ(x)]∗ (3.148)
0 2 T
= −i(ψ̄γ γ ) . (3.149)

The transformation properties of the various Dirac field bilinearies under this transfor-
mation are again summarized in table 3.1.

41
Chapter 3 The Dirac Field

42
 Interacting Fields and Feynman Diagrams
4
In the previous chapters, we have considered rather boring systems: non-interating scalar
and fermion fields. Of course, our ultimate goal is to study the interactions among fields,
and in particular to compute the transition amplitude from some initial state to some
final state as well as the associated cross section. We will now develop the tools to do
that. We will first develop a formalism for evaluating so-called correlation functions in
sections 4.1 to 4.3. Then, we will learn how to relate correlation functions to scattering
matrix elements in section 4.4. In section 4.5, we will finally employ our formalism to
evaluate scattering amplitudes. Along the way, we will get to know Feynman diagrams
as a pictorial way of representing the mathematical expressions for correlation functions
and scattering amplitudes. Up to this point, we will work in a theory containing only
a real scalar field φ(x), but no fermions. We will generalize the formalism to include
fermions in section 4.6.

4.1 Time-Dependent Perturbation Theory for Correlation

Functions
4.1.1 φ4 Theory
In this section and the following ones, we will consider an extension of the free real scalar
field theory called “φ4 theory”. It is perhaps the simplest interacting quantum field theory
and will serve here as a proxy for more realistic theories like quantum electrodynamics
(QED) and quantum chromodynamics (QCD) that we will consider later in this lecture.
φ4 theory itself describes for instance the Higgs boson and its self-interactions (but not
its interactions with other elemntary particles).
φ4 theory is obtained by adding a quartic interaction term to the Klein–Gordon La-

43
Chapter 4 Interacting Fields and Feynman Diagrams

grangian:
1 1 λ
L = (∂µ φ)2 − m2 φ2 − φ4 . (4.1)
2 2 4!
With the Fourier expansion of φ(x) in terms of creation and annihilation operators in
mind, we realize that the φ4 term describes for instance the scattering of two φ particles,
which can be viewed as the annihilation of two incoming particles, followed by the cre-
ation of two outgoing particles with possibly different momentum vectors. The quantity
λ in eq. (4.1) is a coupling constant that determines the strength of the interaction be-
tween φ particles, and the factor 4! in the denominator is included by convention. The
motivation for writing the interaction term in this way will become clear later in this
chapter.

4.1.2 The Vacuum State of the Interacting Theory

At this point, we come across a first important subtlety. The vacuum state |0i of the
free (non-interacting) theory is not the vacuum state of the interacting theory. This is
clear because the extra term Hint in the Hamiltonian should have some impact on the
energy eigenstates of theory, and in particular on the lowest one, which is by definition
the vacuum. We therefore have to be careful with our notation. We will henceforth
call the vacuum state of the free theory |0i and the vacuum state of the interacting
theory |Ωi. |0i has the property that ap |0i = 0 for all p, i.e. it contains no particles.
On the other hand, ap |Ωi 6= 0 in general. The physical interpretation of |Ωi is that it
is a complicated state containing vacuum fluctuations—short-lived particle–antiparticle
pairs that are created from the vacuum and annihilate again within the time window
allowed by the Heisenberg relation ∆E ∆t > 1/2. While they exist, they can interact.

4.1.3 Correlation Functions

As announced above, we cannot study transition matrix elements between multi-particle
initial and final states right away. This has to do with the difficulties associated with the
definition of appropriate initial and final
p states in the interacting theory. For instance,
†
the one particle states of the form 2Ep ap |0i will not be energy eigenstates of the
interacting theory, just as the vacuum states |0i and |Ωi were different. We will instead
first study correlation functions, i.e. expressions of the form

hΩ|φ(x1 ) · · · φ(xn )|Ωi . (4.2)

We have already encountered such a correlation function in the free Klein–Gordon theory:
the Feynman propagator was given by the two-point correlation function

d4 p ie−ip(x−y)
Z
DF (x − y) = h0|T φ(x)φ(y)|0i = , (free theory) (4.3)
(2π)4 p2 − m2 + i

44
 4.1 Time-Dependent Perturbation Theory for Correlation Functions

see eqs. (2.69) and (2.77). In the interacting theory, the expression will be more compli-
cated of course. In fact, it is impossible to write the propagator of the interacting theory
in closed form, but if λ is not too large, the interaction can be treated as a small per-
turbation, and we can use time-dependent perturbation theory to approximate it quite
well. This is what we are going to do now. After we have studied the two point corre-
lation function, we can easily generalize the formalism to more complicated correlation
functions and then to transition matrix elements.
We will make the connection between correlation functions and transition matrix ele-
ments in section 4.4 when we discuss the LSZ reduction formula.

4.1.4 Perturbation Theory

Splitting the Hamiltonian
In the spirit of perturbation theory, we write the Hamilton operator as

H = H0 + Hint , (4.4)

where
d3 p
Z
Ep a†p ap


H0 ≡ 3
(4.5)
(2π)

(cf. eq. (2.48)) and

Z
λ
Hint ≡ d3 x φ4 (x) . (4.6)
4!

(Remember that the Legendre transform that relates the Lagrangian and the Hamilto-
nian, eq. (2.13) acts nontrivially only on terms containing derivatives of the field, while
those without derivatives just get a minus sign.)

Rewriting the Field Operator φ(x, t)

Our goal is to expand the correlation function hΩ|T φ(x)φ(y)|Ωi in λ, assuming that
λ  1. Many, but not all, of the coupling constants appearing in nature fortunately are
indeed small. The main problem along the way is the fact that the time dependence of
the field operator φ(x) is non-trivial now, thank to the interaction. We therefore split
this time dependence into two pieces: the one induced by H0 and the one due to Hint .
At a fixed reference time t0 , we can still Fourier expand the field operator as usual (cf.
eqs. (2.34) and (2.39)):

d3 p
Z
ipx † −ipx


φ(x, t0 ) = p ap (t0 )e + a p (t0 )e . (4.7)
(2π)3 2Ep

45
Chapter 4 Interacting Fields and Feynman Diagrams

In quantizing the field, we impose as before that the creation and annihilation operators
satisfy the canonical commutation relations eq. (2.45).1 The full time-dependent field at
arbitrary time t is then formally given by

φ(x, t) = eiH(t−t0 ) φ(x, t0 )e−iH(t−t0 ) . (4.8)

We will define the interaction picture field operator φI (x, t), which contains the time-
dependence due to H0 , but not the one due to Hint . In other words, we define

φI (x, t) ≡ eiH0 (t−t0 ) φ(x, t0 )e−iH0 (t−t0 ) (4.9)

d3 p
Z 
−ipx † ipx 


= p ap e + ap e . (4.10)
(2π)3 2Ep  0
x =t−t0

This expression is of course (and not by chance) identical in form to the expression
for the time-dependent field operator in the free Klein–Gordon theory, eq. (2.61). In
the following, we will write the creation and annihilation operators at t = t0 simply
as a†p and ap , omitting the explicit mentioning of t0 . φI is an operator that we can
work with: it contains only c-numbers and creation and annihilation operators that obey
the canonical commutation relations. All the technology developed in chapter 2 can be
reused. Therefore, we would like to rewrite the two point correlations hΩ|T φ(x)φ(y)|Ωi
entirely in terms of φI . We have

φ(x, t) = eiH(t−t0 ) e−iH0 (t−t0 ) φI (x, t)eiH0 (t−t0 ) e−iH(t−t0 ) (4.11)

†
≡ U (t, t0 )φI (x, t)U (t, t0 ) . (4.12)

In the last step, we have defined the time evolution operator

U (t, t0 ) ≡ eiH0 (t−t0 ) e−iH(t−t0 ) , (4.13)

which describes all the non-trivial time-dependence of the theory. It satisfies the Schrödinger-
like equation
∂
U (t, t0 ) = eiH0 (t−t0 ) − H0 + H e−iH(t−t0 )


i (4.14)
∂t
= eiH0 (t−t0 ) − H0 + H e−iH0 (t−t0 ) U (t, t0 )


(4.15)
≡ Hint,I (t)U (t, t0 ) . (4.16)

Explicitly, the interaction Hamiltonian in the interaction picture Hint,I is given by

Z
λ
Hint,I (t) = d3 x eiH0 (t−t0 ) φ4 (x, t)e−iH0 (t−t0 ) (4.17)
4!
1
It is sufficient to impose the canonical commutation relations at the reference time t = t0 . They then
follow automatically also at other times because
[ap (t), a†p0 (t)] = eiH(t−t0 ) [ap (t0 ), a†p0 (t0 )]e−iH(t−t0 ) = (2π)3 δ (3) (p − p0 ) .

46
 4.1 Time-Dependent Perturbation Theory for Correlation Functions

Z
λ 4
= d3 x φ (x, t) . (4.18)
4! I

The solution of the above Schrödinger equation eq. (4.16), with the initial condition
U (t0 , t0 ) = 1, can be written as
Z t
U (t, t0 ) = 1 − i dt1 Hint,I (t1 )U (t1 , t0 ) (4.19)
t0

by simply integrating both sides. Plugging this expression into itself repeatedly, we
obtain the infinite series
Z t Z t Z t1
U (t, t0 ) = 1 + (−i) dt1 Hint,I (t1 ) + (−i)2 dt1 dt2 Hint,I (t1 )Hint,I (t2 )
t0 t0 t0
Z t Z tn−1
+ . . . + (−i)n dt1 · · · dtn Hint,I (t1 ) · · · Hint,I (tn ) + . . . . (4.20)
t0 t0

If λ, and thus Hint,I , is small, we can truncate the series after the first order or second
order term. Note that the integration boundaries are such that the integration vari-
ables t1 · · · tn satisfy t1 > t2 > · · · > tn . Therefore, using the time ordering symbol T
introduced in section 2.4.3, we can simplify the second order term in eq. (4.20) into
Z t Z t1 Z t Z t
2 1
(−i) dt1 dt2 Hint,I (t1 )Hint,I (t2 ) = (−i)2 dt1 dt2 T [Hint,I (t1 )Hint,I (t2 )] .
t0 t0 2 t0 t0
(4.21)

Here, we have used the fact that, with the time ordering symbol included, the integrans
is symmetric under t1 ↔ t2 , i.e. extending the domain of the t2 integral from [t0 , t1 ] to
[t0 , t] just double counts every point (see fig. 4.1. This is compensated by a prefactor
1/2.
The above arguments can be extended for the higher order terms in the perturbation
series:
Z t Z tn−1
n
(−i) dt1 · · · dtn Hint,I (t1 ) · · · Hint,I (tn )
t0 t0
Z t Z t
1
= (−i)n dt1 · · · dtn T [Hint,I (t1 ) · · · Hint,I (tn )] . (4.22)
n! t0 t0

Then, the perturbation series eq. (4.20) becomes an exponential series and we can write
  Z t 
0 0
U (t, t0 ) = T exp − i dt Hint,I (t ) , (4.23)
t0

where it is understood that the time ordering applies to each term in the exponential
series.

47
Chapter 4 Interacting Fields and Feynman Diagrams

Figure 4.1: Illustration of the arguments given below eq. (4.21), justifying why the
integration domain of the t2 integral in that equation can be extended from [t0 , t1 ] to
[t0 , t]. Figure taken from [1].

It is useful to generalize the definition of U (t, t0 ) to allow for both arguments to be

different from the reference time t0 at which φ(x, t) and φI (x, t) coincide by definition.
We define
  Z t 
0 00 00
U (t, t ) ≡ T exp − i dt Hint,I (t ) . (t ≥ t0 ) (4.24)
t0

It is easy to check that also U (t, t0 ) satisfies the Schrödinger-like eq. (4.16),
∂
i U (t, t0 ) = Hint,I (t)U (t, t0 ) . (4.25)
∂t
but now with the initial condition U (t, t0 ) = 1 at t = t0 . Another way of writing U (t, t0 )
is
0 0
U (t, t0 ) = eiH0 (t−t0 ) e−iH(t−t ) e−iH0 (t −t0 ) . (4.26)

This can be seen by checking that both eq. (4.24) and eq. (4.26) satisfy eq. (4.25) with
the same initial condition. Equation (4.26) shows in particular that U (t, t0 ) is unitary,
i.e. [U (t, t0 )]† U (t, t0 ) = 1 and that

[U (t, t0 )]† = U (t0 , t) . (4.27)

It also shows that U (t, t0 ) has the property that, for t1 ≥ t2 ≥ t3 ,

U (t1 , t2 ) U (t2 , t3 ) = U (t1 , t3 ) , (4.28)

U (t1 , t3 ) [U (t2 , t3 )]† = U (t1 , t2 ) . (4.29)

48
 4.1 Time-Dependent Perturbation Theory for Correlation Functions

Rewriting the Vacuum |Ωi

We now have the tools to rewrite φ(x, t) in terms of φI (x, t). To rewrite a correlation
function, however, we also need to rewrite the vacuum state |Ωi. In particular, we would
like to rewrite it in terms of the free vacuum |0i because we know how ap , a†p and thus
φI (x, t) act on |0i, while we have no idea how they act on |Ωi. We will now show that
|Ωi can be obtained by starting with |0i and evolving that state over a very long time
interval.
Consider the time evolution of |0i,
X
e−iHT |0i = e−iEn T |ni hn|0i , (4.30)
n

where n runs over a complete set of energy eigenstates of the interacting theory. The
interacting vacuum |Ωi is of course an energy eigenstate, with eigenvalue E0 , and can
therefore be separated from the sum:
X
e−iHT |0i = e−iE0 T |Ωi hΩ|0i + e−iEn T |ni hn|0i , (4.31)
n6=0

We would like to take the limit T → ∞, but for the oscillating exponentials, this limit
does not exist. We therefore have to regularize by taking the limit to a slightly imaginary
direction: T → ∞(1−i). Then, the term containing E0 is the slowest to die and therefore
dominates. We thus get rid of all states |ni except the desired vacuum state |Ωi. The
latter is thus given by
e−iHT |0i
|Ωi = lim . (4.32)
T →∞(1−i) e−iE0 T hΩ|0i

Since H0 |0i = 0 (see the explicit form eq. (4.5)), we can insert a factor eiH0 T before |0i.
Moreover, since T is very large, we can impose a small shift T → T + t0 without changing
the result. This leads to
e−iH(T +t0 ) eiH0 (T +t0 ) |0i
|Ωi = lim (4.33)
T →∞(1−i) e−iE0 (T +t0 ) hΩ|0i
U (t0 , −T )|0i
= lim −iE
. (4.34)
T →∞(1−i) e 0 (T +t0 ) hΩ|0i

Similarly, we can also derive

h0|U (T, t0 )
hΩ| = lim . (4.35)
T →∞(1−i) e−iE0 (T −t0 ) h0|Ωi
Keeping in mind our goal of rewriting everything in terms of φI (x, t), we see that the
only piece that is still problematic is hΩ|0i in the denominator. To simplify it, use that
1 = hΩ|Ωi (4.36)
h0|U (T, t0 )U (t0 , −T )|0i
= . (4.37)
e−2iE0 T | hΩ|0i |2

49
Chapter 4 Interacting Fields and Feynman Diagrams

Rewriting Correlation Functions

We have now all the tools to rewrite the two point correlation function hΩ|T φ(x)φ(y)|Ωi.
If we assume first x0 > y 0 > t0 , we obtain using φ(x) from eq. (4.12), the vacuum from
eqs. (4.34), (4.35) and (4.37), and the properties eqs. (4.28) and (4.29) of the operator
U (t, t0 ),
 −1
hΩ|T φ(x)φ(y)|Ωi = lim h0|U (T, t0 )U (t0 , −T )|0i
T →∞(1−i)

· 0U (T, t0 ) [U (x0 , t0 )]† φI (x, x0 )U (x0 , t0 )



· [U (y 0 , t0 )]† φI (y, y 0 )U (y 0 , t0 ) U (t0 , −T )0

 
(4.38)

h0|U (T, x0 )φI (x, x0 )U (x0 , y 0 )φI (y, y 0 )U (y 0 , −T )|0i

= lim .
T →∞(1−i) h0|U (T, −T )|0i
(4.39)
If instead y 0 > x0 > t0 , we obtain exactly the same expression, with x and y interchanged.
(We do not need to consider the case t0 > x0 or t0 > y 0 because we are free to choose
t0 such that it is smaller than both x0 and y 0 .) This allows us to write down the master
formula

 h RT
i  
0T φI (x, x0 )φI (y, y 0 ) exp − i −T dt Hint,I (t) 0
hΩ|T φ(x)φ(y)|Ωi = lim
 h RT
i   .
T →∞(1−i) 0T exp − i −T dt Hint,I (t) 0
(4.40)
This is the final result of this section. As promised, it is written entirely in terms of the
field operator φI (x, t) and the free vacuum |0i. The field operator is made up of creation
and annihilation operators whose action on each other and on the vacuum we know. The
generalization from the two point correlation function to correlation functions with more
than two field operator is straightforward: for each additional field operator φ(z) on the
left hand side, insert an extra factor of φI (z, z 0 ) in the numerator on the right hand side.
By expanding the exponential, we have a systematic way of approximately evaluating
correlation functions. The n-th term in the perturbation series for the two point corre-
lation function involves n spacetime integrals, n numerical factors of the form −iλ/4!,
and (in the numerator) a correlation function of the form
h0|T φI (x)φI (y)φI (z1 ) · · · φI (z4n )|0i . (4.41)
The correlation function in the denominator omits the factor φI (x)φI (y). We will now
develop a formalism that allows to compute such correlation functions in practice.

4.2 Wick’s Theorem

Our goal in this section is to reduce a correlation function of the form h0|T φI (x1 ) · · · φI (xn )|0i,
which still involves operators, to a purely algebraic expression. We already know that,

50
 4.2 Wick’s Theorem

for n = 2, we obtain just the Feynman propagator:

h0|T φI (x)φI (y)|0i = DF (x − y) . (4.42)

To generalize this two n > 2, it is useful to separate the field operator into a piece
containing only creation operators and a piece containing only annihilation operators:
−
φI (x) = φ+
I (x) + φI (x) , (4.43)

with
d3 p
Z
φ+
I (x) = p e−ipx ap , (4.44)
(2π)3 2Ep
d3 p
Z
φ−
I (x) = p e+ipx a†p . (4.45)
(2π)3 2Ep

(The superscripts + and − stand for positive and negative frequency.) This decomposi-
tion is useful because we know that

φ+
I (x)|0i = 0 and h0|φ−
I (x) = 0 . (4.46)

Decomposing the product of two fields yields

− − − −
φI (x)φI (y) = φ+ + + +
I (x)φI (y) + φI (x)φI (y) + φI (x)φI (y) + φI (x)φI (y) (4.47)
− − − −
= φ+ + + +
I (x)φI (y) + φI (y)φI (x) + φI (x)φI (y) + φI (x)φI (y)
−
+ [φ+
I (x), φI (y)] (4.48)

In the second line, we have rewritten the expression in such a way that all terms except
the commutator are in normal order, i.e. the creation and annihilation operators are
ordered such that all annihilation operators are on the right and all creation operator
are on the left. Normally ordered expression have the advantage that they vanish when
sandwiched between the vacuum |0i. There is even a special notation for normal ordering:
when operators are sandwiched between colons (: · :), it is implied that they should be
put into normal order. For example,

: ap a†k aq : ≡ a†k ap aq . (4.49)

−
The commutator [φ+ I (x), φI (y)] is also very useful because it is just a c-number (times the
identity operator). In fact, by virtue of the canonical commutation relations eq. (2.45),

d3 p d3 p0
Z
0
+ −
[φI (x), φI (y)] = 6
p p e−ipx+ip y [ap , a†p0 ] (4.50)
(2π) 2Ep 2Ep0
d3 p 1 −ip(x−y)
Z
= e . (4.51)
(2π)3 2Ep

51
Chapter 4 Interacting Fields and Feynman Diagrams

If we consider, instead of the simple product φI (x)φI (y) the time-ordered product
T φI (x)φI (y), we obtain for x0 > y 0 exactly the above expression eq. (4.48). For x0 < y 0 ,
−
the normal ordered terms are the same, but the commutator is [φ+ I (y), φI (x)]. There is
again a special symbol to simplify the notation in this situation. The contraction of two
fields is defined as
(
−
[φ+ 0
I (x), φI (y)] for x ≥ y
0
φI (x)φI (y) ≡ −
. (4.52)
[φ+ 0
I (y), φI (x)] for x < y
0

Writing out the commutators explicitly gives

d3 p 1 h −ip(x−y)
Z i
0 0 +ip(x−y) 0 0
φI (x)φI (y) = e θ(x − y ) + e θ(y − x ) . (4.53)
(2π)3 2Ep

Comparing to eq. (2.72), we see that the right hand side is just the Feynman propagator:

φI (x)φI (y) = DF (x − y) . (4.54)

This was to be expected because we know that h0|T φI (x)φI (y)|0i = DF (x − y) and we
have argued that, in decomposition like eq. (4.48), only the commutator but not the
normally ordered terms contribute.
To summarize the discussion so far, we have shown that

T φI (x)φI (y) = : φI (x)φI (y) : + φI (x)φI (y) . (4.55)

From here on, we will omit the subscript I on the field operators. Any field operator
appearing in our expressions in this section will always be meant to be written in the
interaction picture. In subsequent sections, it should be clear from the context where
φ(x) denotes the Heisenberg picture operator and where it denotes the interaction picture
operator.
Let us now generalize eq. (4.55) to time-ordered products of more than two fields. The
result we are going to prove is called Wick’s theorem and reads

T φ(x1 ) · · · φ(xn ) = : φ(x1 ) · · · φ(xn ) + all possible contractions : , (4.56)

where “all possible contractions” means one term for each way of contracting pairs of
fields. For instance, for n = 4, Wick’s theorem reads (with the shorthand notation
φa ≡ φ(xa ))

T φ1 φ2 φ3 φ4 = : φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4

+ φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4

+ φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4 : . (4.57)

52
 4.3 Feynman Diagrams

This expression means that all operators that are not contracted are in normal order,
while each pair of contracted operators gives a factor DF , irrespective of whether they
are adjacent or not. For instance:

: T φ1 φ2 φ3 φ4 : ≡ DF (x1 − x3 ) · : φ2 φ4 : . (4.58)
Note that, in vacuum expectation values of the form h0|T φ(x1 ) · · · φ(xn )|0i, only those
terms where all field are contracted contribute.
Let us now prove Wick’s theorem by induction. We have already proven it for n = 2.
Assume now that the theorem has been proven for n − 1 fields. Let us moreover assume
without loss of generality that x01 ≥ x02 ≥ · · · ≥ x0n . (If this condition is not satisfied, we
can simply relabel the xj to fulfill it.) Then,
−
T φ(x1 ) · · · φ(xn ) = (φ+

1 + φ1 ) · : φ(x2 ) · · · φ(xn )

+ [all contractions not involving φ1 ] : . (4.59)
− −
We would like to pull φ+ 1 and φ1 into the normally ordered expression. For φ1 , this is
easy because, according to the normal ordering prescription, it is anyway supposed to
stay on the left. The φ+ 1 operator, on the other hand, has to be commuted past all other
uncontracted fields. (Remember that fields that are contracted give just c-numbers that
commute with everything.) In a term with m uncontracted fields (which we take to be
φ2 , . . . , φm for definiteness), this leads to
φ+ + +
1 : φ2 · · · φm : = : φ2 · · · φm : φ1 + [φ1 , : φ2 · · · φm :]
+ − + −
= : φ+
1 φ2 · · · φm : + : [φ1 , φ2 ]φ3 · · · φm + φ2 [φ1 , φ3 ]φ4 · · · φm + · · · :

= : φ+
1 φ2 · · · φm : +φ1 φ2 φ3 · · · φm + φ1 φ2 φ3 · · · φm + · · · . (4.60)
For m = n, the first term on the right hand side gives the first term on the right hand
side of Wick’s theorem, eq. (4.56). The other terms for n = m give all possible terms
involving a contractions of φ1 with one of the other fields, and no other contractions. For
m < n, we obtain all terms involving an uncontracted φ1 together with contractions of
other pairs of fields, and all terms involving a contracted φ1 , together with contractions
of other pairs of fields. These are all the terms appearing on the right hand side of Wick’s
theorem, eq. (4.56), and the theorem is therefore proven.

4.3 Feynman Diagrams

The usefulness of Wick’s theorem comes from the fact that it allows us to turn any
correlation function h0|T φ(x1 ) · · · φ(xn )|0i into a sum of products of Feynman propaga-
tors. (Remember that all terms involving uncontracted fields vanish when sandwiched
between h0| · |0i thanks to the normal ordering.) The only difficulty is combinatorics,
i.e. not missing any contractions. This is where Feynman diagrams help by graphically
representing each combination of Feynman propagators.

53
Chapter 4 Interacting Fields and Feynman Diagrams

4.3.1 Basic Idea and Application to a Simple 4-Point Function

In a Feynman diagram, each field operator φ(x) is denoted by a point (called a vertex )
in the drawing plane. Each contraction of two fields is depicted by a line (called a
propagator ) joining the two corresponding spacetime points. For instance, let us consider
again the correlation function h0|T φ(x1 )φ(x2 )φ(x3 )φ(x4 )|0i. The time-ordered product
of field operators is rewritten according to eq. (4.57). Only the terms in the last line of
that expression give a nonzero contribution to the correlation function. The Feynman
diagram notation for these three terms is
1 2 1 2 1 2

h0|T φ1 φ2 φ3 φ4 |0i = + + . (4.61)

3 4 3 4 3 4
This diagrammatic notation also admits a straightforward physical interpretation: par-
ticles are created at two of the vertices, then each particle propagates to another vertex
and is annihilated there. There are three different ways of picking pairs of vertices, cor-
responding to the three Feynman diagrams. Each diagram corresponds to a transition
amplitude, and the amplitude for the overall process is the sum of the three Feynman
diagrams.

4.3.2 An Example in φ4 Theory

So far, we have assumed the spacetime points appearing in the correlation function to be
distinct. For correlation functions involving several field operators at the same spacetime
point, the corresponding vertices in the Feynman diagram coincide as well. In this case,
several propagators will connect to that vertex. Consider for instance the full propagator
of the interacting theory, hΩ|T φ(x)φ(y)|Ωi. According to our master formula, eq. (4.40),
we have to evaluate specifically the expression2

  Z T  
0T φ(x)φ(y) exp − i dt Hint (t) 0
−T
D   Z T   E
= 0T φ(x)φ(y) + T φ(x)φ(y) − i dt Hint (t) + · · · 0 , (4.62)
 
−T

where on the right hand side, we have expanded the perturbation series up to first order
in λ. The zeroth order term is simple: h0|T φ(x)φ(y)|0i = DF (x − y). The first order
term reads, after plugging in the expression for Hint
Z
λ
d4 z 0T φ(x)φ(y)φ(z)φ(z)φ(z)φ(z)0 .

  
−i (4.63)
4!
2
It is implicit here that the field operators φ(x), φ(y) appearing hΩ|T φ(x)φ(y)|Ωi are the full Heisenberg
fields, while those appearing in eq. (4.62) are the interaction picture operators which we denoted by
φI when we first introduced them.

54
 4.3 Feynman Diagrams

According to Wick’s theorem, we have to consider all possible pairwise contractions of

the 6 field operators, and combinatorics tells us that there are 15 such contractions. They
read
Z
λ  
−i d4 z 3 DF (x − y)[DF (z − z)]2 + 12 DF (x − z)DF (y − z)DF (z − z) .
4!
(4.64)

The diagrammatic representation of this expression is

x y z + x y (4.65)
z

Note that neither the prefactors nor the integral over d4 z is explicitly written out in
the diagrammatic notation. It is implied that each
R 4 internal vertex (i.e. each vertex that
comes from Hint ) comes with a factor −iλ/4! d z. There is, however, no direct way
of reading off the prefactors 3 and 12 from the diagrams. When translating Feynman
diagrams into algebraic expressions, we therefore have to be careful not to forget these
factors. This can sometimes be a headache.

4.3.3 A More Advanced Example

Let us now be more ambitious and apply our machinery to one of the contractions arising
from the 3rd order term in the perturbative expansion of hΩ|T φ(x)φ(y)|Ωi. In particular,
we consider

1 −iλ 3
  Z Z Z
d z φφφφ d w φφφφ d4 u φφφφ 0
4 4

  
0 T φ(x)φ(y)

3! 4!
 3 Z
1 −iλ
= d4 z d4 w d4 u DF (x − z) DF (y − w) DF (z − z)
3! 4!
· DF (z − w) [DF (w − u)]2 DF (u − u) . (4.66)

The corresponding Feynman diagram (“cactus diagram”)3 is

(4.67)
x z w y
3
It has become a sort of sport to name Feynman diagrams according to plants, animals and other
objects that they resemble. For instance: penguin diagram, lobster diagram, tadpole diagram, setting
sun diagram—just google for them.

55
Chapter 4 Interacting Fields and Feynman Diagrams

Of course, the contraction shown on the left hand side of eq. (4.66) is not the only
one leading to the expression on the right hand side. For instance, interchanging the
vertices z, w, u does not change the algebraic expression because the vertex coordinates
are all integrated over. There are 3! ways of interchanging the vertices. Moreover, the
contractions that involve φ(z) can connect to any of the four φ(z) factors. This leads to
a 4! = 24-fold ambiguity. Since, however, one contraction has its beginning and its end
on a φ(z) factor, the ambiguity is only 12-fold in our case. Similarly, the ambiguity in
how the contractions connect to the φ(w) fields is 24-fold and the one for the φ(u) fields
is 12-fold. Finally, note that ther are two contractions of the form DF (w − u):

φ(w)φ(w)φ(u)φ(u) . (4.68)

Interchanging the two φ(w) factors and the two φ(u) factors in this expression simul-
taneously does not change the contraction structure. Therefore, the multiplicity of the
diagram is reduced by a factor of 2. In total, the multiplicity factor is thus
4! 4! 1
3! × × 4! × × . (4.69)
|{z} 2 |{z} 2 2
interchange of |{z} interchange of |{z} |{z}
vertices interchange of φ(w) factors interchange of interchange of
φ(z) factors φ(u) factors the two
φ(w)–φ(u)
contractions

We see that the factor coming from the interchange of vertices cancels against the pref-
actor 1/3! from the Taylor series. This is true at any order in perturbation theory: the
n-th order term comes with a factor 1/n! from the series expansion, and interchanging
vertices leads to a multiplicity of n!. Also, each vertex has a prefactor 1/4! that cancels
against the factor 4! from the different ways in which contractions connect to the four
field operators in each vertex. (It now becomes clear why we have pulled the factor 1/4!
out of λ in defining the theory!) What remains is a factor 1/S = 1/8, called a symme-
try factor that accounts for permutations of vertices or fields that leave the contraction
structure invariant.

4.3.4 More Examples for Diagrams with Non-Trivial Symmetry Factors

In most practial calculations in elementary particle physics, symmetry factors do not even
appear. For instance, in more realistic theories like quantum electrodynamics, the vertex
connects three different field operators, so there is no ambiguity from permuting them.
Nevertheless, it is important to keep in mind that sometimes, especially in theories with
scalar fields, symmtry factors do show up. Therefore, we give here a few more examples:

S=2 (interchange of vertices in φ(z)φ(z)) (4.70)

x z

56
 4.3 Feynman Diagrams

(interchange of vertices within each

z S=8 contraction and interchange of con- (4.71)
tractions in φ(z)φ(z)φ(z)φ(z), )

(interchange of the 3 contractions

S = 3! between φ(x) and φ(w))
(4.72)
x z w y

w u
(symmetry factor of the “setting
S = 3! · 2 sun ” diagram times factor 2 from (4.73)
x z y interchanging u and w vertices.)

4.3.5 Position Space Feynman Rules

We can summarize the results of this section so far in a simple cooking recipe for evalu-
ating correlation function: to evaluate a correlation function

  Z T  
hΩ|T φ(x1 ) · · · φ(xn )|Ωi ∝ 0T φI (x1 ) · · · φI (xn ) exp − i dt Hint (t) 0 (4.74)
−T

up to a given order p in perturbation theory, draw all possible Feynman diagrams with
n external vertices (each of which connects to exactly one propagator) and p internal
vertices (each of which has four connection points). The algebraic expression for each
diagram is then composed of factors determined by the following Feynman rules

1. For each propagator x y = DF (x − y) (4.75)

Z
2. For each internal vertex z = −iλ d4 z (4.76)

3. For each external vertex x = 1 (4.77)

4. Figure out the symmetry factor and divide by it.

Let us briefly reiterate the physical interpretation of these rules: each propagator gives

57
Chapter 4 Interacting Fields and Feynman Diagrams

the amplitude for a particle to propagate between two spacetime points. At each external
vertex, one particle is created or annihilated. At each internal vertex, four propagators
meet. This describes processes where 4 particles are created from the vacuum, or 1
particle is annihilated and 3 are created, or 2 are annihilated and 2 are created, or 3 are
annihilated and 1 is cretaed, or 4 are annihilated. The amplitude for these interaction
processes is given by the factor −iλ. Finally, the integral over d4 z instructs us to integrate
over all spacetime points where an interaction can happen. This is a manifestation of
the superposition principle in quantum mechanics. If a process can happen in different
ways (in this case: if the interaction can happen at different spacetime points z), the
amplitudes for all these different ways should be added up.

4.3.6 Momentum Space Feynman Rules

The above cooking recipe was written entirely in coordinate space. In practice, it is
usually more convenient to work in momentum space. We can easily translate the Feyn-
man rules to momentum space by rearranging terms. In particular, remember that the
Feynman propagator is given by (cf. eq. (2.69))

d4 p ie−ip(x−y)
Z
DF (x − y) = , (4.78)
(2π)4 p2 − m2 + i

which is just the Fourier transform of

i
DF (p) ≡ . (4.79)
p2 − m2 + i
In momentum space, we will therefore associate this expression with the propagator. The
exponential factors will be associated with the vertices instead. To keep track of which
vertex gets the e−ipx factor and which one gets the e+ipy factor, we endow each propagator
with an arrow indicating the direction of momentum flow. The factor e+ipy is associated
with the vertex into which the momentum is flowing, the factor e−ipx is associated with
the vertex from which the momentum is flowing away. Since DF (x − y) = DF (y − x) we
can choose the direction of the arrow arbitrarily, it is only important that the momentum
on each propagator flows out of one vertex and into another one.
An external vertex, which gave just a factor 1 in coordinate space, will now contribute
a factor of e±ipx to the amplitude, where the + sign is relevant if the arrow of the attached
propagator points away from the vertex and the − sign is relevant if the arrow points
towards the vertex.
For an internal vertex, the exponential factors can be combined with the integral over
4
d z to yield a delta function. If we call the momenta of the propagators attached to a
given vertex p1 , p2 , p3 , p4 and assume for instance that p1 and p2 are incoming, while
p3 , p4 are outgoing, an internal vertex becomes

58
 4.3 Feynman Diagrams
p2 p4
Z
z = −iλ d4 ze−i(p1 +p2 +p3 +p4 )z = −iλ (2π)4 δ (4) (p1 + p2 − p3 − p4 ) .

p1 p3 (4.80)

If any of the arrows was reversed, the sign of the corresponding momentum would be
flipped on the right and side. The δ-function simply implies energy-momentum conser-
vation at the vertex: the 4-momentum carried into the vertex by the incoming particles
must be equal to the 4-momentum carried away by the outgoing particles.
The only pieces that are left are the 4-momentum integrals from the propagators.
Many of these can be carried out directly using the δ-functions from the vertices. The
remaining ones have to be included in the amplitude explicitly.
To summarize, the momentum space Feynman rules read

i
1. For each propagator = (4.81)
p2 − m2 + i

2. For each internal vertex = −iλ (4.82)

3. For each external vertex x = e±ipx (4.83)

4. Impose 4-momentum conservation at each vertex.

d4 p
Z
5. Integrate over momenta not determined by 4.: (4.84)
(2π)4

6. Figure out the symmetry factor and divide by it.

4.3.7 Disconnected Feynman Diagrams

Most Feynman diagrams we have considered so far were connected, i.e. each vertex in the
diagram could be reached from any other vertex by following propagator lines. However,
the Feynman rules and the underlying algebraic structures admit also diagrams were this
is not the case. Perhaps the simplest example for a disonnected diagram arises already
at first order in the perturbative expansion of the two-point function, see eq. (4.65):

59
Chapter 4 Interacting Fields and Feynman Diagrams

 
x y 
(4.85)
 z 

Other examples are

 
" #
,

 x y 


x1 x2
 
 
 

x3 x4

Each disconnected diagram is a product of several connected pieces. Some of these con-
tain external vertices, others contain only internal vertices. The latter type of diagrams
is called vacuum bubbles because they correspond to processes where particles are created
from the vacuum and annihilate back into the vacuum without coupling to an external
vertex.
Let us define the set {Vj } of all vacuum bubbles and the set {Ej } of diagrams in which
each internal vertex is connected to at least one external vertex. Note that the Ej do
not need to be connected—the only requirement is that each disconnected piece has to
contain at least one external vertex. In other words, there should be no vacuum bubbles.
Since every Feynman diagram is a product of one Ej and an arbitrary number of Vj ,
combinatorics tells us that the perturbation series can be reordered in the following way:
Z T i   X  X  Y 1 

 h 
ni
0 T φ(x1 ) · · · φ(xn ) exp − i
 dt Hint,I (t) 0 =
 Ej × (Vi ) .
−T ni !
j {ni } i
(4.86)

Here, the sum over {ni } runs over all ordered sets {n1 , n2 , · · · } of non-negative integers,
where each ni denotes how often the i-th vacuum bubble appears in the diagram. In any
fixed order Feynman diagram, only a finite number of the ni will be nonzero of course.
The factor 1/ni ! is a symmetry factor arising from the fact that interchanging identical
vacuum bubbles leaves the diagram and the underlying contraction structure unchanged.
We can simplify the contribution from the vacuum bubbles to eq. (4.86) further:
XY 1  YX ∞
ni 1
(Vi ) = (Vi )ni (4.87)
ni ! ni !
{ni } i i ni =0
Y

= exp Vi (4.88)
i

60
 4.4 The LSZ Reduction Formula

X 
= exp Vi . (4.89)
i

4.3.8 The Denominator of the Correlation Function

Up to this point were focused on computing the numerator in our master formula  for
evaluating correlation functions eq. (4.40). Of course, the denominator h0|T exp −
RT

i −T dt Hint,I (t) |0i can be obtained in complete analogy. The only difference compared
to the numerator is that the Feynman diagrams arising from the denominator have no
external vertices, i.e. they consist only of vacuum bubbles. The results from the previous
section (eqs. (4.86) and (4.89)) immediately tell us what the denominator looks like. It
is simply given by
Z T

 h  i  X 
0T exp − i dt Hint,I (t) 0 = exp Vi . (4.90)
−T i

This cancels exactly the corresponding exponential from the numerator!

In other words, a general correlation function can be evaluated by computing only
Feynman diagrams containing no vacuum bubbles
X
hΩ|T φ(x1 ) · · · φ(xn )|Ωi = Ej (4.91)
j
 
sum over all diagrams in which each connected
= .
piece contains at least one external vertex
(4.92)

4.4 The LSZ Reduction Formula

Now that we know how to evaluate a correlation function in an interacting quantum field
theory, we should proceed to discuss the connection to scattering amplitudes, which are
physically much more interesting.
A highly non-trivial problem in this context is to define appropriate initial and final
states (asymptotic state). We know that the creation operator a†p , when acting on the
vacuum of the free theory |0i, creates a one-particle state in the free theory. However,
this state will in general not be a momentum eigenstate of the interacting theory and
is therefore not suitable for constructing initial or final states for a scattering process.
Moreover, in the general Fourier expansion of the Heisenberg field φ(x),

d3 p
Z
ap (t)eipx + a†p (t)e−ipx ,


φ(x, t) = 3
p (4.93)
(2π) 2Ep

the creation and annihilation operators a†p (t) and ap (t) acquire a time dependence. In
section 4.1.4, we have imposed the canonical commutation relations at t = t0 to obtain
an object whose action on the free vacuum |0i we know. However, this does not help us

61
Chapter 4 Interacting Fields and Feynman Diagrams

define one-particle states in the interacting theory because we do not know how a†p (t0 )
acts on |Ωi.
We will assume that, in the interacting field theory, asymptotic one-particle states
with definite momentum are given by

|pi = 2Ep a†p (±∞)|Ωi ,

p
(4.94)

in analogy to the definition of one-particle momentum eigenstates in the free theory. We

use the creation operator at a†p (−∞) for initial states in scattering process and a†p (+∞)
for final states. The justification for this assumption is given for instance in Srednicki
[2], sec. 5. Since a one-particle state should be orthogonal to the zero-particle state—the
vacuum |Ωi—it follows from eq. (4.94) that

ap (±∞)|Ωi = 0 . (4.95)

We will assume that the one-particle states defined by eq. (4.94) are normalized in the
same way as in the free field theory:

p|p0 = (2π)3 2Ep δ (3) (p − p0 ) .


(4.96)

A further complication arises because the initial state of a scattering process involves
more than one particle. As soon as the wave functions of these particles overlap, they
will start to interact. If particles were described by plane wave states (momentum eigen-
states), their wave functions would overlap at all times. This would make it difficult to
define suitable initial and final states as combinations of one-particle states. In partic-
ular, even tough one particle states a†p1 (±∞)|Ωi and a†p2 (t = ±∞)|Ωi are momentum
eigenstates, the two particle state a†p2 a†p1 (±∞)|Ωi is not a momentum eigenstate because
of the interaction. The way out is to consider localized wave packets instead of plane
waves. Then, at t = ±∞, the wave packets will have negligible overlap. The different
particles in the initial and final states will know of each other only when they get close
to the origin, so their interaction does not hinder the definition of the asymptotic states.
We define the creation operator for a wave packet state centered around a momentum
pj as

d3 p
Z
ã†j (t) ≡ f (p; pj ) a†p (t) , (4.97)
(2π)3

where f (p; pj ) is the shape factor of the wave packet. Its exact form is unimportant,
but we could for instance choose a Gaussian:
(2π)3/4 (p − pj )2
 
f (p; pj ) = exp − . (4.98)
σ 3/2 4σ 2

(The normalization is chosen such that hΩ|ãj ã†j |Ωi = 1.) To be specific, let us consider
a 2 → 2 scattering process in which two initial state wave packets with central momenta

62
 4.4 The LSZ Reduction Formula

p1 and p2 scatter into two final state wave packets with central momenta p3 and p4 .
The initial and final states are thus

|ii = ã†1 (−∞) ã†2 (−∞) |Ωi (4.99)

and

hf | = hΩ| ã3 (+∞) ã4 (+∞) . (4.100)

We would ultimately like to relate scattering amplitudes hf |ii to correlation functions

hΩ|φ(x1 ) · · · φ(xn )|Ωi, therefore we should express the particle creation and annihilation
operators appearing in the definitions of |ii and |f i in terms of the field operators. The
relation that achieves this is
←
→
Z
i 
a†p (t → ±∞) = − p d3 x e−ipx ∂0 φ(x) , (4.101)

2Ep t→±∞

←→
where f (x) ∂0 g(x) ≡ f (x)(∂g(x)/∂x0 ) − (∂f (x)/∂x0 )g(x). The construction with the
derivative is necessary to extract only a†p (t) instead of a combination of a†p (t) and ap (t),
as can be easily shown by direct computation:
←→
Z
i 
−p d3 x e−ipx ∂0 φ(x)

2Ep t→±∞
3
−ipx ←
→
Z Z
i d q −iqx

† iqx 
= −p d3 x p e ∂0 aq e + aq e
(2π)3 2Eq

2Ep t→±∞
i −iEp t  
=− e − iEp ap e−iEp t + iEp a†p eiEp t + iEp ap e−iEp t + iEp a†p eiEp t 

2Ep t→±∞

= a†p (t → ±∞) . (4.102)

Note that here we have expanded the full Heisenberg field φ(x) in the same way as in
the free theory. This is not possible in general—the general time dependence of the
field operator is more complicated than that. However, by our assumption, at t → ±∞,
the operators a†p (t → ±∞) and ap (t → ±∞) create and annihilate energy-momentum
eigenstates of the full theory. This means that they commute with the Hamiltonian at
t → ±∞, and consequently, according to eq. (4.8) (or according to the derivation from
section 2.3.3) the time evolution of the field operator at t → ±∞ is indeed as in the free
theory.
To rewrite a scattering amplitude, it is moreover useful to be able to rewrite ã†j (−∞)
in terms of ã†j (+∞) and vice versa. Here is the relevant relation:
Z ∞
ã†j (+∞) − ã†j (−∞) = dt ∂0 ã†j (t)
−∞

63
Chapter 4 Interacting Fields and Feynman Diagrams

d3 p −i ←
→
Z Z  
= f (p; pj ) d4 x p ∂0 e−ipx ∂0 φ(x)
(2π)3 2Ep | {z }
←
→ ←
→
= −ip0 [e−ipx ∂0 φ(x)] + e−ipx ∂0 (∂0 φ(x))
= e−ipx [(p0 )2 + ∂02 ]φ(x)

d3 p −i −ipx  2
Z Z
d4 x p ∂0 + p2 + m2 φ(x)

= f (p; pj ) e
(2π)3 2Ep
d3 p −i −ipx  2 ← −
Z Z
f (p; pj ) d4 x p ∂0 − ∇2 + m2 φ(x)

= e
(2π)3 2Ep
d3 p −i −ipx  2
Z Z
f (p; pj ) d4 x p ∂ + m2 φ(x) .

= e (4.103)
(2π)3 2Ep

In the last step, we have integrated by parts twice to turn the left-acting derivative
←
−
operator ∇2 into a right-acting operator ∇2 . Note that (∂ 2 + m2 φ(x) would vanish in

the free theory due to the Klein–Gordon equation. In the interacting theory, however,
this expression is in general nonzero.
On the right hand side of eq. (4.103), the shape factor f (p, p0 ) of the wave packet no
longer plays an important (we only needed it to give meaning to the asymptotic states).
Therefore, we will from now on assume the momentum distribution q of the wave packets
3
to be very narrow, so that we can approximate f (p, pj ) ' (2π) 2Epj δ (3) (p−pj ). (The
normalization here is such that the states ã†j |Ωi are normalized in the same way as the
states |pj i, see eq. (4.96).
We can now go back to the scattering amplitude and write

hf |ii = hΩ|T ã4 (+∞) ã3 (+∞) ã†1 (−∞) ã†2 (−∞)|Ωi . (4.104)

On the right hand side, we have added a time ordering symbol T . We can do this because
the operators that follow it are in time order already. We now use eq. (4.103) and its
hermitian conjugate to turn all ãj (+∞) operators into ãj (−∞) operators and all ã†j (−∞)
into ã†j (+∞). The time ordering symbol then moves the annihilation operators to the
right, where they act on the vacuum to yield zero. What is left is only a term containing
field operators, but no ãj or ã†j operators any more. This term reads
Z
hf |ii = d4 x1 d4 x2 d4 x3 d4 x4 e−ip1 x1 (∂x21 + m2 ) · e−ip2 x2 (∂x22 + m2 )

· eip3 x3 (∂x23 + m2 ) eip4 x4 (∂x24 + m2 ) hΩ|T φ(x1 )φ(x2 )φ(x3 )φ(x4 )|Ωi . (4.105)

This is the desired relation between a scattering amplitude and a correlation function.
It is called the Lehmann-Symanzik-Zimmermann (LSZ) reduction formula. Generalizing
it from the 2 → 2 case to an arbitrary process with n particles in the initial state
(momenta p1 , . . . , pn ) and m particles in the final state (momenta pn+1 , . . . , pn+m ) is

64
 4.5 Computing S-Matrix Elements from Feynman Diagrams

straightforward:
Z
n+m
hf |ii = i d4 x1 · · · d4 xn+m

· e−ip1 x1 (∂x21 + m2 ) · · · e−ipn xn (∂x2n + m2 )

· eipn+1 xn+1 (∂x2n+1 + m2 ) · · · eipn+m xn+m (∂x2n+m + m2 )

· hΩ|T φ(x1 ) · · · φ(xn+m )|Ωi . (4.106)

Let us finally reiterate the main assumptions that went into the LSZ reduction formula.
The formula hinges on the fact that a†p (±∞)|Ωi is indeed a one-particle state. This can
only be the case if

hΩ|φ(x)|Ωi = 0 , (4.107)

i.e. if the vacuum expectation value of φ(x) vanishes. If eq. (4.107) was not satisfied, this
would mean that one or several of the a†p create a superposition of a one-particle state
and the vacuum.
The LSZ formula also relied on the normalization of the one-particle states, see eq. (4.96).
If the normalization was different, the numerical factors in the LSZ formula would change.
The normalization condition eq. (4.96) implies

hp|φ(x)|Ωi = eipx . (4.108)

A more detailed discussion of these conditions is given in Srednicki [2], sec. 5.

Equations (4.107) and (4.108) are crucial for our formalism to work, and we will go to
great lengths to make sure they are satisfied. In particular, we will have to renormalize
QFTs to remove violations of eqs. (4.107) and (4.108) that arise when computing loop
diagrams.

4.5 Computing S-Matrix Elements from Feynman Diagrams

We have come along way already: we have learned how to turn a general correlation func-
tion hΩ|φ(x1 ) · · · φ(xn )|Ωi, involving the unknown vacuum state |Ωi and the unknown
time-dependent field operator φ(x), into an expression we can compute perturbatively
using Feynman diagrams. The LSZ formula tells us how to turn the result of such a
computation into a scattering amplitude. However, the LSZ formula is a bit clumsy.
Therefore, we will now use it only to derive a new set of Feynman rules for scatter-
ing amplitudes. From then on, we can directly use these rules to evaluate scattering
amplitudes at our leisure.
Consider the momentum space Feynman rules derived in section 4.3.6. For each field
operator φ(xj ) appearing in the correlation function, they yield a factor eipxj for an in-
coming particle (arrow pointing into the diagram away from the external vertex) with

65
Chapter 4 Interacting Fields and Feynman Diagrams

4-momentum p and a factor e−ipxj for an outgoing particle (arrow pointing out of the
diagram towards the external vertex) with 4-momentum p. This factor is the only depen-
dence of the correlation function on xj . Therefore, applying the Klein–Gordon operator
∂x2j +m2 from the LSZ formula to these exponential factors yields simply (−p2 +m2 )e±ipxj .
The term in parentheses (together with one of the factors of i in the LSZ formula) cancels
exactly the i/(p2 − m2 + i) from the propagator connected to that external vertex.
The remaining exponential e±i(pj −p)xj , together with the integral over d4 xj , yields a
delta function (2π)4 δ (4) (pjR − p). These delta function can be used to evaluate some of
the momentum integrals d4 p/(2π)4 . Only one of them remains, enforcing overall 4-
momentum conservation among the external particles: (2π)4 δ (4) (±p1 ± p2 ± p3 ± · · · ),
where the plus signs stand for incoming particles and the minus signs for outgoing par-
ticles (or vice-versa). In practice, this delta function is usually not associated with the
matrix element, but is reintroduced in the formula for the cross section. We will follow
this convention here.
Beyond the overall delta function, neither external vertices nor propagators connected
to them contribute to the Feynman diagrams for the scattering amplitude. Propagator
factors need to be included only for internal propagators, i.e. propagators connecting
two internal vertices. To emphasize this point, incoming and outgoing particles in the
Feynman diagram for a scattering amplitude are depicted as lines with only one end
attached to a vertex.
These rules can be summarized as follows

i
1. For each (internal) propagator = (4.109)
p2 − m2 + i

2. For each internal vertex = −iλ (4.110)

3. For each incoming/outgoing particle = 1 (4.111)

4. Impose 4-momentum conservation at each vertex.

d4 p
Z
5. Integrate over momenta not determined by 4.: (4.112)
(2π)4

6. Figure out the symmetry factor and divide by it.

66
 4.5 Computing S-Matrix Elements from Feynman Diagrams

Here are two examples: first, the amplitude for the scattering of a φ particle off another
φ particle a lowest order in λ is simply
p2 p4

= −iλ (4.113)

p1 p3
Here pj are the momenta of the external particles. It is implied that p1 + p2 = p3 + p4 .
Going one order higher in λ, there are two diagrams:
p2 p4 p2 p4

p1 p3 p1 p3

d4 q
Z
2 i i
= (−iλ) · 2
(2π) (p1 + p2 + q) − m + i q − m2 + i
4 2 2

d4 q
Z
2 i i
+ (−iλ) · 2 . (4.114)
(2π) (p1 − p3 + q) − m + i q − m2 + i
4 2 2

We call diagrams like eq. (4.113), where no path taken through the diagrams hits a vertex
twice, a tree level diagram. Diagrams with closed loops, i.e. paths that return to the
vertex at which they originated are called loop diagrams. We will develop the technology
to evaluate loop diagrams, i.e. to simplify algebraic expressions like the right hand side
of eq. (4.114), in the second part of this course.
In principle, when considering the second order (in λ) corrections to the scattering rate
of two φ particles, one could also draw diagrams with loops connected to the external
legs, for instance

(4.115)

We will see later, when discussing renormalization, that such diagrams in which loops
are attached to one of the external legs, need not be considered in many cases. For
the moment, we simply accept this result and will therefore drop such contributions
henceforth.
A few comments are in order:
• Note that we have omitted the arrows on the lines. Since the only place where the
4-momenta appear is in the propagators, where they appear as p2 , the direction of
momentum flow is irrelevant.

67
Chapter 4 Interacting Fields and Feynman Diagrams

• In computing scattering amplitudes, we are only interested in connected diagrams.

We have already seen in section 4.3.8 that diagrams containing vacuum bubbles do
not contribute to correlation functions. The possibility remained, however, that a
diagram consists of several disconnected pieces, each of them containing external
vertices. However, such process are not scattering processes in the usual sense. For
instance, in the diagramss

+ + , (4.116)

contributing to the 4-point function, no momentum is exchanged.

• In a typical Feynman diagram, most momenta are determined by the fixed momenta
of the external particle, together with 4-momentum conservation at the vertices.
The undetermined momenta that we have to integrate over according to rule 5.
correspond to momentum flows inside the diagram along closed loops, i.e. beginning
and ending on the same vertex. For instance, in the diagram

(4.117)

there are two such loop momenta. To see this, note that there are three internal
propagators with per se undetermined momenta that should be integrated over.
Momentum conservation at the vertices imposes two constraints (in the form of
4d δ-functions). One of them becomes the overall energy-momentum-conserving
δ-factor that we have pulled out of the matrix element, the other eliminates the
integral over one of the internal momenta. We are left with two momentum inte-
grations, and we say that the diagram is a 2-loop diagram.

4.6 Feynman Rules for Fermions

Up to now, we have only considered interacting scalar field theories to keep things as
simple as possible. Unfortunately, most elementary particles are fermions. Therefore, let
us now generalize what we have learned to include fermions.

4.6.1 The Master Formula for Correlation Functions Involving Fermions

The derivation of our master formula eq. (4.40), which allows us to rewrite correlation
functions in the interacting theory in terms of correlation functions of the corresponding
free theory goes through also for fermions. There is only one small modification: in
several places along the way, we reordered the field operators in the correlation function

68
 4.6 Feynman Rules for Fermions

Rt Rt
to bring them into time order (see eq. (4.21), where we rewrote t0 dt1 t01 dt2 in terms
Rt Rt
of t0 dt1 t0 dt2 and the last step leading to the master formula eq. (4.40), where we
pulled the various U -factors together into one exponential). For fermions, reordering
field operators leads to extra minus signs due to their anticommuting nature. However,
these minus signs can be fully accounted for if we define the time ordering symbol T such
that any interchange of two fermion fields required to restore time ordering yields one
extra minus sign. For instance:

T ψ1 ψ̄2 ψ3 ψ4 = (−1)3 ψ3 ψ1 ψ4 ψ̄2 if x03 > x01 > x04 > x02 . (4.118)

– – ( ) ( )

Here, we have again used the shorthand notation ψ j ≡ ψ (xj ). For the special case of
only two fermion fields, we have already used this definition in section 3.3.

4.6.2 Wick’s Theorem for Fermions

For Wick’s theorem to hold for fermions, we should in a similar way also define the
normal ordering symbol : · : to yield an extra minus sign if an odd permutation of the
fermion fields is required to put them into normal order. For instance:

: ap bk a†q : = (−1)2 a†q ap bk = (−1)3 a†q bk ap . (4.119)

With this definition, we can easily generalize Wick’s theorem. Consider first again the
case of just two fermion fields (cf. eq. (4.48) et seqq.):

T ψ(x)ψ̄(y) = : ψ(x)ψ̄(y) : + ψ(x)ψ̄(y) (4.120)

Remember that the contraction term arises from those contributions that require an
interchange of fields in order to bring them into normal order. For scalar fields, this was
achieved by a commutator, for fermion fields we should use the anticommutator instead.
Since the time ordering symbol and the normal ordering symbols imply an extra minus
sign if y 0 > x0 , also the definition of the contraction has to include such an extra minus
sign:
(
{ψ + (x), ψ̄ − (y)} for x0 ≥ y 0
ψ(x)ψ̄(y) ≡ (4.121)
−{ψ̄ + (y), ψ − (x)} for x0 < y 0
= SF (x − y) . (4.122)

Here, ψ + (x) and ψ − (x) are again defined as the positive and negative frequency com-
ponents of ψ(x), respectively. The fact that the contraction of ψ(x) and ψ̄(y) equals
SF (x − y) is completely analogous to the proof of the corresponding identity for scalar
fields, eq. (4.54). Note that, unlike for the scalar propagator, for fermions SF (x − y) 6=
SF (y − x), so care must be taken to put ψ(x) and ψ̄(y) into the appropriate order before
evaluating the contraction.

69
Chapter 4 Interacting Fields and Feynman Diagrams

The time ordered product of two ψ fields or two ψ̄ fields is equal to the normally
ordered product of these fields because their anticommutator vanishes. In order for our
equations to have the same form in this case as in eq. (4.120), we define

ψ(x)ψ(y) = 0 (4.123)

ψ̄(x)ψ̄(y) = 0 . (4.124)

To carry out the induction step in the proof of Wick’s theorem for fermions, we need
to worry about one more source of minus signs. The analogue of eq. (4.60) for fermions
reads

ψ1+ : ψ2 · · · ψm : = − : ψ2 · · · ψm : ψ1+ + {ψ1+ , : ψ2 · · · ψm :}

= : ψ1+ ψ2 · · · ψm : + : {ψ1+ , ψ2− }ψ3 · · · ψm − ψ2 {ψ1+ , ψ3− }ψ4 · · · ψm + · · · :

= : ψ1+ ψ2 · · · ψm : + : ψ1 ψ2 ψ3 · · · ψm : + : ψ1 ψ2 ψ3 · · · ψm : + · · · .
(4.125)

(Replacing any of the ψ fields by ψ̄ fields would not change this result.) This implies
that a contraction of two non-adjacent field operators comes with a minus sign if an
odd number of anticommutations is necessary to move these field operators next to each
other. For example:

: ψ1 ψ2 ψ̄3 ψ̄4 : = −ψ1 ψ̄3 : ψ2 ψ̄4 : = −SF (x1 − x3 ) : ψ2 ψ̄4 : . (4.126)

To summarize, Wick’s theorem for fermions takes exactly the same form as for scalars:

T ψ(x1 ) · · · ψ(xn ) = : ψ(x1 ) · · · ψ(xn ) + all possible contractions : . (4.127)

For fermions, however, it is understood that an extra minus sign appears on the left hand
side when x1 , . . . , xn are such that bringing the field operators into time order requires
an odd permutation. Similarly, in the normally ordered terms on the right hand side,
any summand that requires an odd permutation to restore normal order gets a minus
sign. Finally in replacing contracted field operators by Feynman propagators, we should
first move the contracted fields next to each other, possible paying the price of another
minus sign if an odd number of anticommutations is required.

4.6.3 The LSZ Formula for Fermions

The LSZ reduction formula can also be generalized to theories with fermions. Here, we
only quote the result. The full proof is given in Srednicki, [2], sec. 41. For a scattering
process with n incoming fermions (4-momenta p1 , . . . , pn and spins s1 , . . . , sn ) and m
outgoing fermions (4-momenta pn+1 , . . . , pn+m and spins sn+1 , . . . , sn+m ) it reads:
Z
n+m
hf |ii = (−i) d4 x1 · · · d4 xn+m

70
 4.6 Feynman Rules for Fermions

· eipn+1 xn+1 ūsn+1 (pn+1 )(i∂/xn+1 − m) αn+1 · · · eipn+m xn+m ūsn+m (pn+m )(i∂/xn+m − m) αn+m
   

· hΩ|T ψαn+m (xn+m ) · · · ψαn+1 (xn+1 )ψ̄(x1 )α1 · · · ψ̄(xn )αn |Ωi
←− ←−
· (−i ∂/ x1 − m)us1 (p1 ) α1 e−ip1 x1 · · · (−i ∂/ xn − m)usn (pn ) αn e−ipn xn .
   

(4.128)

Here, α1 , . . . αn+m are spinor indices. If any of the incoming fermions is replaced by an
antifermion (created by ψ instead of ψ̄), the corresponding factor is replaced according
to
←−
−i(−i ∂/ xj − m)usj (pj )e−ipj xj → ie−ipj xj v̄ sj (pj )(i∂/xj − m) (4.129)

and is moved from the right of the correlation function to the left. Similarly, if an outgoing
fermion is replaced by an antifermion (annihilated by ψ̄ instead of ψ), the corresponding
factor is replaced according to
←−
−ieipj xj ūsj (pj )(i∂/xj − m) → i(−i ∂/ xj − m)v sj (pj )eipj xj (4.130)

and is moved from the left of the correlation function to the right.
There are two main differences between the LSZ formula for fermions compared to
the one for scalars: first, the replacement of the Klein–Gordon operator by the Dirac
operator comes about because the analogue of eq. (4.101) which allows us to write a
creation operator in terms of a field operator does not involve a derivative in the case of
fermions. Instead, one exploits the orthogonality relations of the u and v spinors, which
also explains the appearance of such spinors in the LSZ formula for fermions.

4.6.4 Yukawa Theory

To put the master formula and Wick’s theorem for fermions to good use, we need an
actual interacting field theory involving fermions. The simplest one of its kind is Yukawa
theory, a toy model involving one Dirac fermion ψ, one scalar field φ, and an interaction
between these fields. This theory was originally introduced by Hideki Yukawa (Nobel
Prize 1949) to describe the interaction between nucleons and mesons. In modern particle
physics, it also describes the interactions of the Higgs boson with quarks and electrons.
Finally, it can be considered a simplified toy model of quantum electrodynamics. Here
is the Lagrangian
1 1
LYukawa = ψ̄(i∂/ − mψ )ψ + (∂µ φ)(∂ µ φ) − m2φ φ2 − g ψ̄ψφ . (4.131)
2 2
Here, mψ and mφ are the fermion and scalar masses, respectively, and g is a dimensionless
coupling constant describing the strength of the Yukawa interactions. For perturbation
theory to work, g hadR better not be too large. The interaction Hamiltonian in Yukawa
theory is thus Hint = d3 x g ψ̄ψφ.

71
Chapter 4 Interacting Fields and Feynman Diagrams

Let us evaluate the fermion–fermion scattering rate in this model:

fermion(p) + fermion(k) → fermion(p0 ) + fermion(k 0 ) . (4.132)
The matrix element for this process can be obtained from the LSZ formula applied to the
correlation function hΩ|T ψ(x1 )ψ(x2 )ψ̄(x3 )ψ̄(x4 )|Ωi. This correlation function receives its
lowest order non-trivial contribution at order g 2 in perturbation theory:
Z Z
(−ig)2
 4
d4 y ψ̄ρ (y)ψρ (y)φ(y) 0 .
 
2! 0 T ψ (x )ψ (x )ψ̄ (x
α 1 β 2 γ 3 δ 4 ) ψ̄ (x ) d x ψ̄τ (x)ψτ (x)φ(x)
(4.133)
Here, α, β, γ, δ, τ , ρ are spinor indices. One contraction structure that is nonzero is the
following

Z Z
(−ig)2
 4
d4 y ψ̄ρ (y)ψρ (y)φ(y) 0 .
 
2! 0 T ψα (x1 )ψβ (x2 )ψ̄γ (x3 )ψ̄δ (x4 ) d x ψ̄τ (x)ψτ (x)φ(x)
(4.134)
Contracting ψα (x1 ) with ψ̄τ (x) and ψβ (x2 ) with ψ̄ρ (y) instead yields another nonzero
contribution. As usual, interchanging the two vertices x and y is always possible, and
the resulting factor 2! cancels the 1/2! prefactor. Therefore we can drop this prefactor
from now on. Note that, in Yukawa theory, the vertex couples three different fields, so
no extra factors arise from interchanging those. In other words, Yukawa theory does not
have symmetry factors!. Up to a possible minus sign, eq. (4.134) yields
Z Z
2
(−ig) d x d4 y [SF (x1 − y)]αρ [SF (x2 − x)]βτ [SF (y − x3 )]ργ [SF (x − x4 )]τ δ DF (x − y) .
4

(4.135)
We now apply the LSZ reduction formula to this expression. The terms for the incoming
fermions have the structure
←
−
Z
− i d4 x3 SF (y − x3 ) (−i ∂/ x3 − mψ )us3 (p)e−ipx3

d4 q i(/q + mψ ) −iq(y−x3 ) ←−
Z Z
= −i d x34
4 2 2 e (−i ∂/ x3 − mψ )us3 (p)e−ipx3
(2π) q − mψ + i
d4 q (/q + mψ )(/q − mψ ) −iq(y−x3 ) s3
Z Z
= d4 x3 e u (p)e−ipx3
(2π)4 q 2 − m2ψ + i
| {z }
=1
−ipy s3
=e u (p) . (4.136)
0
Similarly, for each outgoing fermion we obtain a factor of the form e+ip y ūs1 (p0 ) or
0
e+ik x ūs2 (k 0 ). The overall expression for the matrix element is thus, after plugging in
the Fourier expansion of DF (x − y),
iM · (2π)4 δ (4) (p + k − p0 + k 0 )

72
 4.6 Feynman Rules for Fermions

d4 q ie−iq(x−y)
Z
0 0
= (−ig)2 [ūs1 (p0 )us3 (p)][ūs2 (k 0 )us4 (k)]e−i(p−p )y e−i(k−k )x
(2π)4 q 2 − m2φ + i
i
= (−ig)2 [ūs1 (p0 )us3 (p)][ūs2 (k 0 )us4 (k)] · (2π)4 δ (4) (p + k − p0 + k 0 )
(p0 − p)2 − m2φ
(4.137)

As mentioned earlier, we pull the overall 4-momentum conserving delta function out of
the matrix element by convention. The procedure we have followed can obvious also be
applied to more general processes, and to higher order terms in the perturbation series.
What we have found can be summarized in the Feynman rules for Yukawa theory:

73
Chapter 4 Interacting Fields and Feynman Diagrams

1. Incoming fermion = us (p) (4.138)

p

p
−−→
2. Incoming antifermion = v̄ s (p) (4.139)

3. Outgoing fermion = ūs (p) (4.140)

p

p
−−→
4. Outgoing antifermion = v s (p) (4.141)

5. Incoming scalar = 1 (4.142)

p

6. Outgoing scalar = 1 (4.143)

p

i
7. Scalar propagator = (4.144)
p p2 − m2φ + i

i(p
/ + mψ )
8. Fermion propagator = (4.145)
p p2 − m2ψ + i

9. Vertex = −ig (4.146)

10. Impose 4-momentum conservation at each vertex.

d4 p
Z
11. Integrate over momenta not determined by 10.: (4.147)
(2π)4

12. Figure out the overall sign of the diagram.

74
 4.6 Feynman Rules for Fermions

As mentioned above, there are no symmetry factors in Yukawa theory. With these rules,
the scattering amplitude we just computed corresponds to the diagram

p p0
q
k k0
(4.148)

and the second diagram relevant for the process of fermion–fermion scattering is
p0

p
q

(4.149)
k
k0

A few notes on the Feynmana rules for fermions are in order

• By convention, the arrows on the fermion lines do not represent momentum flow,
but particle number flow: particle number flows into the diagram along an in-
coming fermion line and out of the diagram along an outgoing fermion line. For
antiparticles, the flow is reversed: for each incoming antifermion, one unit of par-
ticle number flows out of the diagram (or, equivalently, a unit of negative particle
number flows into the diagram) and for outgoing antifermions, particle number
flows into the diagram. On the external lines, the arrows remind us whether the
line corresponds to a barred spinor or an unbarred spinor. At the vertices, they
remind us of which propagator is contracted with the ψ̄ factor from the vertex and
which one is contracted with the ψ factor. The arrows thus also help us avoid
mistakes: if we end up with a vertex that has two arrows pointing towards it or
two arrows pointing away from it, that diagram is forbidden.

• Especially in the experimental literature, one sometimes draws an arrow of time

(typically pointing left to right) into a Feynman diagram and puts the initial parti-
cles on the left, the final state particles on the right of the diagram. We will refrain
from doing so and not associate any physical meaning with the orientation of the
Feynman diagram in the drawing plane.

• Spinor indices are always contracted along fermion lines. This can be understood
by noting that fermion lines are joint together at vertices, where the indices of
the ψ field corresponding to the line pointing towards the vertex and the ψ̄ field
corresponding to the line pointing away from the vertex are contracted. In practice,
when translating a Feynman diagram to an algebraic expression, we start at the end

75
Chapter 4 Interacting Fields and Feynman Diagrams

of a fermion lines (which gives ū or v̄ spinor) and then work our way backwards
along the lines, in the direction opposite to the arrows, applying the rules for
vertices and propagators as we encounter them.

• For fermion propagators, the direction of momentum flow matters. The above rule
holds if the momentum p flows along the arrow. (In case the momentum flows in
the opposite direction p should be replaced by −p.) This rule arise from the way the
oscillating exponential combine into 4-momentum conserving delta functions. To
see this explicitly, let us work out another example from first principles. Consider
the process of fermion–antifermion annihilation into two scalars:
p k
x→ −

q
p0 k0
−
→ →
−
y (4.150)

The relevant correlation function (with contractions already indicated) is

Z Z
0 T ψ̄(x1 )ψ(x2 )φ(x3 )φ(x4 )(−ig) d4 x ψ̄ψφ (−ig) d4 y ψ̄ψφ 0

  

Z Z
= (−ig)2 d4 x d4 y (−1)5 SF (x−x1 )SF (x2 −y)SF (y−x)DF (x−x4 )DF (y−x3 ) ,
(4.151)

where the coordinate x1 corresponds to the incoming fermion, x2 to the incoming

antifermion, x3 to the upper outgoing scalar, and x4 to the lower outgoing scalar.
Note the factor of (−1)5 coming from the interchange of fermion fields necessary
to put the operators contracted to SF (x − x1 ), SF (x2 − y) and SF (y − x) into the
right order (ψ before ψ̄). Applying the LSZ formula, we obtain

iM · (2π)4 δ (4) (p + p0 − k − k 0 )
d4 q 0 i(/
q + mψ )
Z Z Z  
0 0
= (−ig) 2 4
d x d y4
4
v̄(p ) 2 2 u(p) e−i(p−k−q)x e−i(p −k +q)y .
(2π) q − mψ
(4.152)

We see that the d4 x and d4 y integrals give delta functions that require the momen-
tum in the internal fermion propagator to flow from x to y, i.e. along the arrow.

• Perhaps the least useful Feynman rule is #12 (figure out the overall sign) because
it doesn’t really tell us what to do. In the above example of fermion–antifermion
annihilation, we have already seen what is needed: we have to count the number of

76
 4.6 Feynman Rules for Fermions

anticommutation operations necessary to put all pairs of contracted fermion fields

next to each other, with the ψ field to the right of the ψ̄ field. In eq. (4.151), 5 such
operations were required. In other words, we have to write down the correlation
function leading to the Feynman diagram in question, together with the contraction
structure leading to that diagram. Then, we need to figure out the number of
required contractions.
At this point, an ambiguity arises because the field operators corresponding to the
external fermions can be in arbirary order in the correlation function. Interchang-
ing two of them would lead to an extra minus sign. However, this minus sign would
be common to all diagrams contributing to a given scattering process, and since
physical observables depend only on the absolute value of the total matrix element
and on the relative signs of individual diagrams, such an overall minus sign is irrel-
evant. In other words, we are free to put the operators corresponding to incoming
and outgoing fields in arbitrary order, as long as we pick the same order for all
diagrams.
To illustrate this point, let us go back to the example of fermion–fermion scattering,

p0

p p0 p
+ (4.153)
k k0
k
k0
We repeat here for convenience the contractions from eq. (4.134), corresponding to
the first diagram:

Z Z
4
0 T ψ(x1 )ψ(x2 )ψ̄(x3 )ψ̄(x4 ) 2!1 (−ig) d4 y ψ̄ψφ 0 . (4.154)

  
d x ψ̄ψφ (−ig)

As it is written here, this contraction structure does not come with an extra minus
sign. The contractions corresponding to the second diagram, however, are

Z Z
4
0 T ψ(x1 )ψ(x2 )ψ̄(x3 )ψ̄(x4 ) 2!1 (−ig) d4 y ψ̄ψφ 0 . (4.155)

  
d x ψ̄ψφ (−ig)

Therefore, when applying the Feynman rules to write down the expression for the
amplitude of fermion–fermion scattering, the two diagrams should come with a
relative minus sign:

i
iM = (−ig) [ūs1 (p0 )us3 (p)] 0
2
[ūs2 (k 0 )us4 (k)]
(p − p)2 − m2φ

77
Chapter 4 Interacting Fields and Feynman Diagrams


s1 0 s4 i s2 0 s3
− [ū (p )u (k)] 0 [ū (k )u (p)] . (4.156)
(k − p)2 − m2φ

One special case is that of closed fermion loops. Consider for instance the leading
order diagram contributing to the scattering of two scalars:

∼ ψ̄ψ ψ̄ψ ψ̄ψ ψ̄ψ . (4.157)

We see that a closed fermion loop in a diagram always leads to an extra minus
sign.

4.6.5 The Yukawa Potential

We now have all the tools to compute scattering amplitudes in a straightfoward way,
it is perhaps time to finally compute some actual physical observable. We will consider
the scattering of non-relativistic fermions (e.g. nucleons) through the Yukawa interaction
(e.g. mediated by mesons) and compute the associated nucleon–nucleon potential V (r).
Our strategy is to compute the scattering matrix element first in QFT based on Feynman
diagrams, then in QM based on a Hamiltonian H = H0 + V (r). By comparing the two
expressions, we can determine V (r).
We assume one of the two fermions to be much heavier than the other so that its
4-momentum remains essentially unchanged in the scattering process and the approxi-
mation of a perfect central potential holds. The only way in which the mass difference of
the two fermions alters the process depicted in eq. (4.153) is by making the two particles
distinguishable, so that the second Feynman diagrams (with crossed outgoing legs) is
absent. To evaluate the first one (first line of eq. (4.156), we need to compute spinor
products of the form ū(p0 )u(p) in the non-relativistic limit. This is fairly easy because
√
in that limit, us (p) ' m(ξ s , ξ s ), i.e. the 3-momentum dependence in the spinors is
negligible. We therefore have
0 0 0
ūs (p0 )us (p) = 2mξ s† ξ s = 2mδ ss (4.158)

For the propagator, we also need

(p0 − p)2 ' −|p0 − p|2 + O(p4 ) . (4.159)

The matrix element is thus

0 0
2mψ1 2mψ2 ig 2 δ ss δ rr
iM = , (4.160)
|p0 − p|2 + m2φ

78
 4.6 Feynman Rules for Fermions

where, by assumption, the masses of the two fermions satisfy mψ2  mψ1 . The scattering
amplitude is as usual

hf |ii = iM (2π)4 δ (4) (p + k − p0 − k 0 ) . (4.161)

In non-relativistic quantum mechanics, the amplitude for scattering of a particle off a

potential V (r) is
RT 2
RT 2
0 −i
hf |ii = lim T hp |e −T dt [−∇ /(2mψ1 )+V (t,x)]
|pi−T · T hk0 |e−i −T dt [−∇ /(2mψ2 )]
|ki−T .
T →∞
(4.162)

The subscripts ±T on the bra and ket vectors indicate that the states are taken a
t = T → ±∞ and then evolved in time with the Hamiltonian H = H0 + V (t, x). In the
second scalar product in eq. (4.162), which comes from the very heavy fermion, we have
taken into account that this fermion is too heavy to be affected by the scattering process,
so the potential can be neglected in its Hamiltonian. Note also that, for very heavy mψ2 ,
the particle can absorb any momentum without its state changing. Therefore, we can
without loss of generality replace hk0 | by hk0 − k − p|. (We do this in order to bring the
final form of hf |ii to the same form as eq. (4.161), see below.)
The first factor on the right hand side of eq. (4.162) can be evaluated perturbatively
by expanding the exponential in V (t, x). The zeroth order term in this expansion is
uninteresting because no momentum is exchanged. We thus focus on the first order
term. Assuming that the potential is time-independent, it yields
Z RT 0 Rt 0
lim +T hp | − i dt e−i t dt H0 V (x)e−i −T dt H0 |pi−T
0
(4.163)
T →∞
Z Z
0
= lim −i dt d3 x e−i(Ep −Ep0 )t+i(p−p )x V (x) (4.164)
T →∞

= −iṼ (p0 − p) · 2π δ(Ep0 − Ep ) (4.165)

= −iṼ (p0 − p) · 2π δ(Ep + Ek − Ep0 − Ek0 ) . (4.166)

In the second line, we have used that |pi and |p0 i are eigenstates of the unperturbed
Hamiltonian H0 , and we have replaced the bra and ket vectors by the (time-dependent)
wave functions of the initial and final states. In the third line, we have introduced the
3-dimensional Fourier transform Ṽ (p0 − p) of the potential V (x). In the fourth line, we
have rewritten the δ-function to bring it to the form appearing in the QFT expression
eq. (4.161). In doing so, we have used that mψ2 is very large, so that Ek − Ek0 ' 0.
After a similar calculation for the second term in eq. (4.162), we obtain

hf |ii = −iṼ (p0 − p) · (2π)4 δ (4) (p + k − p0 − k 0 ) . (4.167)

Before directly comparing eq. (4.167) to the QFT expressions eqs. (4.160) and (4.161),
we need to remember that in QM a one-particle state |pi is usually normalized according

79
Chapter 4 Interacting Fields and Feynman Diagrams

to hp|p0 i = (2π)3 δ (3) (p − p0 ), whereas in QFT we have used the Lorentz-invariant

normalization hp|p0 i = 2Ep (2π)3 δ (3) (p − p0 ). Therefore, to make the comparison, we
should drop the factor 2mψ1 2mψ2 in eq. (4.160). We can then conclude that

−g 2
Ṽ (p0 − p) = . (4.168)
|p0 − p|2 + m2φ

All that’s left to do to obtain V (r) is to carry out an inverse Fourier transform:

d3 q −g 2
Z
V (x) = eiqx
(2π)3 p2 + m2φ
−g 2 ∞ iqr − e−iqr
Z
2 e 1
= dq q
2
4π 0 iqr q + m2φ
2

∞
−g 2 qeiqr
Z
= 2 dq 2 (4.169)
4π ir −∞ q + m2φ

This integral can be carried out using the residual theorem. The integration contour
can be closed in the complex q-plane by an infinite half-circle in the upper half-plane.
Thanks to the exponential, the integrand vanishes along that half-circle. We then pick
up the pole at q = +imφ and get

−g 2 imφ e−rmφ
V (x) = · 2πi
4π 2 ir 2imφ
g 2 1 −rmφ
=− e , (4.170)
4π r
where r ≡ |x|. The minus sign indicates that the Yukawa interaction leads to an attractive
force. The exponential limits the range of the force to distances . m−1
φ . For instance, for
nucleon–nucleon interactions mediated by pions, the range of the interaction is ' m−1π ∼
−1
(100 MeV) ∼ O(fm).

80
 Quantum Electrodynamics
5
After lots of formal developments, this section will be somewhat more practical. We will
abandon the toy models considered so far and move on to a theory of great practical
relevance: quantum electrodynamics (QED). In the context of QED, we will develop the
technology to efficiently calculate tree level Feynman diagrams and the associated cross
sections and decay rates.

5.1 The QED Lagrangian from Symmetry Arguments

Quantum electrodynamics is the simplest example of a gauge theory—a theory that is
invariant under a certain kind of symmetry called a gauge symmetry. In fact, one can
(and, in particle physics, usually does) argue that QED is defined through its gauge
symmetry. Let’s see how this argument goes.
We start from the theory of a free fermion, let’s called it the electron:


Lψ ≡ ψ̄ i∂/ − m ψ . (5.1)

We demand that the theory of electrons be invariant under U(1) gauge transformations
of the form

ψ(x) → ψ 0 (x) = e−iα(x) ψ(x) , (5.2)

where α(x) is an arbitrary function of x. The crucial point here is the x-dependence of α.
This is what makes the symmetry a gauge symmetry or local symmetry, as opposed to a
global symmetry, for which the transformation law is the same at each spacetime point.
The symmetry is called a U (1) gauge symmetry because, at fixed x, the set of all possible
transformations forms a mathematical group called U (1) (for “unitary transformations
in one dimension”).

81
Chapter 5 Quantum Electrodynamics

Obviously, the free Dirac Lagrangian eq. (5.1) is not invariant under U (1) gauge trans-
formations because of the derivative. Rather, Lψ transforms as

Lψ → L0ψ ≡ ψ̄ i∂/ − m ψ + ψ̄γ µ ψ(∂µ α) .

(5.3)

Gauge invariance could be achieved if the theory contained another field Aµ (judiciously
called the photon) that couples to ψ through a term

Lint ≡ −eψ̄γ µ ψAµ (5.4)

and has the gauge transformation property

1
Aµ → A0µ ≡ Aµ + (∂ µ α) . (5.5)
e
Gauge invariance immediately dictates many of the properties of the photon field. First,
it must be bosonic because α(x) is just a number, so if Aµ was fermionic, the gauge
transformation eq. (5.5) would destroy the canonical anticommutation relations. More-
over, the interaction term eq. (5.4) would not be Lorentz invariant. (Three spin 1/2 fields
can never couple to a scalar quantity.) More specifically, Aµ should be a real bosonic
field, otherwise the operator ψ̄γ µ ψAµ would not be Hermitian. Moreover, Aµ must be
a vector field, i.e. it carries a Lorentz index and thus has 4 components. Otherwise,
a transformation law like eq. (5.5) would not be Lorentz invariant. Finally, the photon
must be massless because mass term of the form m2 Aµ Aµ would not be gauge invariant.1
What gauge symmetry does not tell us is the value of the coupling constant e, which will
be interpreted as the electric unit charge.
The final missing ingredient is a kinetic term for the photon. The interaction term (5.4)
only describes photon interactions, but would not lead to a photon propagator. To find
the form of the photon kinetic term, we use our knowledge from classical electrodynamics,
where the Lagrangian describing photons is
1
Lγ ≡ − Fµν F µν . (5.6)
4
Here, the field strength tensor is defined as

F µν ≡ ∂ µ Aν − ∂ ν Aµ . (5.7)

It is easy to check that, with the Lagrangian (5.6), the Euler–Lagrange equations lead
to Maxwell’s equations:
δLγ
0 = ∂µ
δ(∂ µ Aν )
1
There are massive gauge bosons ins nature—the W and Z bosons mediating the weak interaction.
Understanding how they obtain their mass through the Higgs mechanism will be one of the climaxes
of the second part of this course.

82
 5.2 The Feynman Rules for QED

δ
= ∂µ
 ρ σ σ ρ
 
∂ A − ∂ A ∂ ρ A σ − ∂σ A ρ
δ(∂ µ Aν )
δ
= ∂µ
 ρ σ
µ ν
2(∂ A )(∂ρ Aσ ) − 2(∂ ρ Aσ )(∂σ Aρ ) (5.8)
δ(∂ A )
= ∂ µ 4∂µ Aν − 4∂ν Aµ
 

= 4 ∂ µ Fµν . (5.9)

This is just Maxwell’s first equation in covariant form: ∂ µ Fµν = 0.2 In summary, the
QED Lagrangian reads in total
1
LQED = ψ̄ i∂/ − m ψ − Fµν F µν − eψ̄γ µ ψAµ .


(5.10)
4
As a final remark related to considerations of gauge symmetry, let us apply Noether’s
theorem. The consdeved current is
δLQED
jµ = · ∆ψ ∝ ψ̄γ µ ψ . (5.11)
δ(∂µ ψ) |{z}
=−iαψ

5.2 The Feynman Rules for QED

The Feynman rules for QED can be derived in the same way as those of Yukawa theory.
The only exception is the photon propagator, a rigorous derivation of which we will
defer to a later part of this course, when we will have the path integral formalism at our
disposal.
The rules for incoming and outgoing (anti)fermions and for the fermion propagator are
completely identical to the ones in Yukawa theory. To guesstimate the rules for photons,
let us Fourier expand the photon field:

d3 p
Z X 
s s −ipx s† s∗ ipx
Aµ (x) = p ap µ (p)e + ap µ (p)e . (5.12)
(2π)3 2Ep s

We have here exploited the fact that Aµ is a real field. Otherwise, the coefficients of
the positive and negative frequency components would not be the complex conjugates of
each other. Since Aµ is bosonic, the creation and annihilation operators as† s
p and ap obey
canonical commutation relations of the form of eq. (2.45).
The numerical coefficients sµ account for the Lorentz structure of the field and will
be interpreted as photon polarization vectors. This is in full analogy to the u and v
spinors in the expansion of the Dirac field. The u and v spinors were solutions to the free
Dirac equation, and the field was expanded in these solutions. Here, the sµ are solutions
2
Note that the second of Maxwell’s equations, ∂ µ (µνρσ F ρσ ) = 0, is automatically satisfied because

∂ µ F ρσ + ∂ ρ F σµ + ∂ σ F µρ = 0 .

83
Chapter 5 Quantum Electrodynamics

to the free Maxwell exquations. The sum runs over a complete basis of such solutions.
This is actually where the complications with quantizing the photon field come into play:
we know that physical photons are always transversely polarized. Therefore, we must
restrict the polarization vectors of external photons to have the form µ
µ
√  = (0, ), with
 · p = 0. A suitable choice for p along the z axis is  = (0, 1, ±i, 0)/ 2, where the plus
sign corresponds to right-handed polarization and the minus sign corresponds to left-
handed polarizition. With this restriction, we can guess the Feynman rules for incoming
(outgoing) photons: they will contribute just a factor µ (µ∗ ), just as the appearance
of u and v spinors in the expansion of the Dirac field lead to the appearance of these
spinors in the Feynman rules.
The restriction of transversality does not necessarily need to apply to off-shell photons
(photons with p2 6= 0) which can propagate along an internal photon line in a Feynman
diagram. Nevertheless, even off-shell photons do not have 4 degrees of freedom because
of gauge invariance. The freedom to choose a gauge removes one degree of freedom.
This has to be carefully taken into account when deriving the expression for the photon
propagator. Here, we will settle for a naı̈ve guesstimate: the propagator should somehow
resemble that of the scalar field. However, since the propagator should again be just a
two point correlation function of the form h0|Aµ (x)Aν (y)|0i, it must have some Lorentz
structure. The simplest choice is a factor g µν . We will settle for that, but we will add
an extra minus sign and define the photon propagator as

d4 p −ig µν e−ip(x−y)
Z
µν
∆ (x − y) ≡ . (5.13)
(2π)4 p2 + i
The rationale for the extra minus sign in the numerator is as follows: For x → y and µ =
ν, the propagator is just the norm of Aµ (x), and that norm should be positive. Evaluating
the integral over p0 in eq. (5.13) using complex contour integration and assuming x0 > y 0 ,
we have
d3 p 1 −ip(x−y)
Z
µν x0 &y 0
∆ (x − y) = e · (−g µν )
(2π)3 2|p|
−g µν |p|2 ei|p||x−y| − e−i|p||x−y|
Z
= d|p|
4π 2 2|p| i|p||x − y|
−g µν
= . (5.14)
8π|x − y|
Thus, the spatial components of Aµ will have positive norm, while the 0-component has
negative form. This means that at least physical (transversely polarized) photons are
properly normalized. Regarding the unphysical time-like photons, we do not have the
tools yet to deal with them. We will see later that their contributions exactly cancel
against contributions from longitudinal photons (polarized along p), so that only the
physical polarization states remain.
The Feynman rule for the interaction vertex between photons and fermions can be
directly read off from the Lagrangian. As in the previously considered theories, all

84
 5.2 The Feynman Rules for QED

fields appearing in the interaction term end up contracted with other field operators,
yielding propagators, and the coupling constant (which here includes a γ µ matrix is
copied verbatim, but receives an extra factor
R of i from the fact that
R the master formula
eq. (4.40) involves an exponential of −i Hint (or equivalently +i Lint ), not just Lint .
We can thus summarize the QED Feynman rules as follows:

1. Incoming fermion = us (p) (5.15)

p
p
−−→
2. Incoming antifermion = v̄ s (p) (5.16)

3. Outgoing fermion = ūs (p) (5.17)

p
p
−−→
4. Outgoing antifermion = v s (p) (5.18)

5. Incoming photon = µ (5.19)

p

6. Outgoing photon = µ∗ (5.20)

p

−ig µν
7. Photon propagator = (5.21)
p p2 + i

i(p
/ + m)
8. Fermion propagator = (5.22)
p p2 − m2 + i

9. Vertex = −ieγ µ (5.23)

10. Impose 4-momentum conservation at each vertex.

d4 p
Z
11. Integrate over momenta not determined by 10.: (5.24)
(2π)4
12. Figure out the overall sign of the diagram.
Note that, as in Yukawa theory, QED does not have symmetry factors because the three
fields entering the vertex are all distinct.

85
Chapter 5 Quantum Electrodynamics

5.3 e+ e− → µ+ µ−
We now apply the QED Feynman rules to a simple process: the annihilation of an electron
and a positron into a muon and an antimuon, e+ e− → µ+ µ− . This was for instance
one of the most important processes at the Large Electron Positron (LEP) collider at
CERN. Similar processes (with the electron and the positron replaced by a quark and an
antiquark) are of utmost importance to the LHC. Our first goal is to compute the cross
section for e+ e− → µ+ µ− , developing along the way several important computational
techniques that are crucial to the efficient evaluation of Feynman diagrams.

5.3.1 Feynman Diagram and Squared Matrix Element

Let us first apply the Feynman rules, though. Compared to vanilla QED, we have two
fermion fields here—the electron and the muon—but their couplingsto the photon are
identical, therefore the generalization of the Feynman rules to the two-fermion case is
trivial. We obtain
e− µ−

p k  
0 −igµν 0
= v̄ s (p0 )(−ieγ µ )us (p) ūr (k)(−ieγ ν )v r (k 0 ) . (5.25)
q q2
→

→
−
−

e+ p0 k0 µ+
Since our goal is to compute a cross section, we ultimately need not the matrix element
iM, but its modulus square, |M|2 . We thus have to multiply eq. (5.25) by its complex
conjugate. Along the way, we encounter the complex conjugates of Dirac field bilinearies,
e.g.
0 0
(v̄ s (p0 )γ µ us (p))∗ = [us (p)]† (γ µ )† (γ 0 )† v s (p0 )
0
= ūs (p)γ µ v s (p0 ) . (5.26)

In the last step, we have used the identity

γ µ† = γ 0 γ µ γ 0 , (5.27)

which is most easily shown by direct calculation using the explicit form eq. (3.4) of the
gamma matrices. The squared matrix element is therefore

e 4  s0 0 µ s ν s0 0

r0 0 r0 0

|M|2 = v̄ (p )γ u (p)ūs
(p)γ v (p ) ūr
(k)γ µ v (k )v̄ (k )γ ν ur
(k)
q4
e4  0 0
  0 0

= 4 tr v s (p0 )v̄ s (p0 )γ µ us (p)ūs (p)γ ν tr ur (k)ūr (k)γµ v r (k 0 )v̄ r (k 0 )γν .
q
(5.28)

86
 5.3 e+ e− → µ+ µ−

The rearrangements in the second line will come in handy below. In practice, we often
have no control over the spin orientation of the initial state particles, and we do not
measure the spin orientation of the final state particles. For instance, in a particle collider,
it is very difficult to produce polarized beams—it requires that the electrons have some
polarizing interaction and to retain their polarization until they collide. One possibility
is the so-called Sokolov–Ternov effect: when electrons emit synchrotron radiation in a
magnetic field, their spin flips. Since the two spin states have different energies in the
magnetic field, transitions from the higher energy state to the lower energy one are
slightly more probable than interactions going the opposite way. After the electrons
have been in a storage ring for a long time, they can achieve a polarization up to 70%.
Measuring the spin of a final state particle can be done by observing the kinematics
(in particular angular correlations) between its decay products. In the case of muons,
however, this is difficult because they are too long-lived to decay inside a typical detector.
The upshot is that we are often interested in unpolarized cross sections, i.e. cross sec-
tions averaged over initial state spins and summed over final state spins.
P We have already
seen how toP evaluate spin sums in section 3.1.3. Using the relations s=1,2 us (p)ūs (p) =
/ + m and s=1,2 v s (p)v̄ s (p) = p
p / − m, we obtain
1 X 1 XXX
|M|2 ≡ |M|2
2 s 2 0 r 0 s r
e4  0 µ ν
  0 
= / − me )γ (p
tr (p / + me )γ tr (k/ + mµ )γµ (k/ − mµ )γν (5.29)
4q 4
Don’t let yourself get confused by the fact that µ appears both as a Lorentz index here
and as an index distinsguishing the electron mass me from the muon mass mµ .

5.3.2 Trace Technology

The reason an expression like eq. (5.29) is very convenient is that several identities exist
that make the evaluation of traces over gamma matrices very efficient. First, note that
the trace of a single gamma matrix vanishes:

tr γ µ = 0 . (5.30)

In fact the trace of any product of an odd number of gamma matrices vanishes because,
using (γ 5 )2 = 1 and the cyclic property of the trace, we have

tr γ µ1 · · · γ µ2n+1 = tr (γ 5 )2 γ µ1 · · · γ µ2n+1 (5.31)

5 µ1 µ2n+1 5
= −tr γ γ ···γ γ (5.32)
µ1 µ2n+1
= −tr γ ···γ . (5.33)

Thus, this trace must be zero.

For the product of two gamma matrices, we have

tr γ µ γ ν = tr 2g µν · 14×4 − γ ν γ µ
 

87
Chapter 5 Quantum Electrodynamics

= 8g µν − tr γ µ γ ν . (5.34)

Therefore,

tr γ µ γ ν = 4g µν . (5.35)

For any even number 2n of gamma matrices, we can use the anticommutator {γ µ , γ ν } =
2g µν to move the leftmost gamma matrix all the way to the right and then use the cyclic
property of the trace to bring it back to the left. This leads to the relation

tr γ µ1 · · · γ µ2n = tr 2g µ1 µ2 γ µ3 · · · γ µ2n − γ µ2 2g µ1 µ3 γ µ4 · · · γ µ2n



+ · · · + γ µ2 · · · γ µ2n−1 2g µ1 µ2n − γ µ2 · · · γ µ2n γ µ1 ,


(5.36)

which can be used to reduce the trace of 2n gamma matrices to a sum of traces of 2n − 2
gamma matrices. (In the last term on the right hand side, the cyclic property of the trace
needs to be exploited to bring it to the same form as the left hand side.) In particular,
for four gamma matrices, this leads to

tr γ µ γ ν γ ρ γ σ = tr g µν γ ρ γ σ − γ ν g µρ γ σ + γ ν γ ρ g µσ
 

= 4 g µν g ρσ − g µρ g νσ + g µσ g νρ .


(5.37)

Even though these are all the relations we need to evaluate the squared matrix element
for the process e+ e− → µ+ µ− , let us also mention some identities we will encounter later.
This concerns in particular traces involving γ 5 . Since γ 5 = iγ 0 γ 1 γ 2 γ 3 , it counts as an
even number of gamma matrices. This means in particular that any trace involving one or
several γ 5 and an odd number of γ 0 , . . . , γ 3 , vanishes. Also, the trace of γ 5 itself vanishes.
This is immediately obvious from the explicit form γ 5 = diag(−12×2 , 12×2 ) in the chiral
representation, but can also be shown more abstractly using only the representation
invariant properties of the gamma matrices:

tr γ 5 = tr γ 0 γ 0 γ 5 = −tr γ 0 γ 5 γ 0 = −tr γ 0 γ 0 γ 5 = −tr γ 5 . (5.38)

Similarly, for tr γ 5 γ µ γ ν , we can insert the square of a gamma matrix γ ρ with ρ 6= µ, ν

then anticommute one γ ρ factor all the way to the right to show that tr γ 5 γ µ γ ν is equal
to its negative and therefore must be zero. The same procedure works for a combination
of γ 5 with four other gamma matrices, except in the case where those four matrices
are all distinct, i.e. each γ 0 , . . . , γ 3 appears exactly once. Therefore, tr γ 5 γ µ γ ν γ ρ γ σ is
proportional to µνρσ . By using µνρσ = 0123 and calculating explicitly, we find the
prefactor to be −4i. Thus

tr γ 5 γ µ γ ν γ ρ γ σ = −4iµνρσ . (5.39)

88
 5.3 e+ e− → µ+ µ−

Figure 5.1: Kinematics of e+ e− → µ+ µ− . Figure taken from [1].

5.3.3 The Squared Matrix Element for e+ e− → µ+ µ− (Part II)

We are now ready to evaluate the spin-averaged matrix element (5.29). The first trace is
 0 µ ν
 0 µ ν 2 µν
/ − me )γ (p
tr (p / + me )γ = tr p /γ p /γ − 4me g
= 4 p0µ pν − (p0 · p)g µν + p0ν pµ − 4m2e g µν ,


(5.40)

and a similar expression holds for the second trace:

 0
tr (k/ − mµ )γν (k/ + mµ )γµ = 4 kν0 kµ − (k 0 · k)gµν + kµ0 kν − 4m2µ gµν ,


(5.41)

From here on, we will set me = 0 to keep our expressions shorter. At a typical high
energy collider, where kinetic energies are  GeV, the electron mass 0f 511 keV is indeed
negligible. The squared matrix element is then, after contracting Lorentz indices,

8e4 h 0 0 0 0 2 0
i
|M|2 = (p · k )(p · k) + (p · k)(p · k ) + mµ (p · p) . (5.42)
q4
At this point, it is useful to be a bit more explicit about the kinematics. Let us therefore
consider the collision of an e+ e− pair in the center of mass frame, where the two particles
are moving in opposite directions along the z axis. (This is the conventional coordinate
system chosen for processes at a high energy collider.) This kinematic setup is illustrated
in fig. 5.1. With the definitions given in the figure, we can compute the following dot
products:

q 2 = (p + p0 )2 = 4E 2 , p · p0 = 2E 2 ,
(5.43)
p · k = p0 · k 0 = E 2 − E|k| cos θ , p · k 0 = p0 · k = E 2 + E|k| cos θ .

With these replacements, the squared matrix element is

e4 h 2 2 2 2 2 2
i
|M|2 = E (E − |k| cos θ) + E (E + |k| cos θ) + 2m µ E
2E 4
m2µ m2µ
    
= e4 1 + 2 + 1 − 2 cos2 θ . (5.44)
E E

89
Chapter 5 Quantum Electrodynamics

We have written the last line such that the high energy limit E  mµ , relevant for
instance for the LEP collider, is easy to take. We have eliminated |k|2 in favor of E 2 and
m2µ .

5.3.4 The Cross Section — General Results

The procedure for obtaining the cross section from a squared matrix element is the same
in QFT as in QM. First, remember that, in our conventions for M, we had pulled out a
factor (2π)4 δ (4) (p + p0 − k − k 0 ). We should now put that factor back. When computing
the transition probability by squaring the matrix element (including the delta function),
we then encounter an awkward square of a delta function. To deal with it, rewrite one
of the δ-factors by using
Z T /2
2 ∆E T
2πδ(∆E) = lim dt ei ∆E t = lim sin . (5.45)
T →∞ −T /2 T →∞ ∆E 2
The physical interpretation here is that the process is happening in a finite time inter-
val [−T /2, T /2], so the time integral that leads to the delta function only leads to an
approximate delta function. The squared δ function is thus
2 ∆E T
(2π)2 δ 2 (∆E) = (2π)δ(∆E) lim lim sin
T →∞ ∆E→0 ∆E 2
= (2π)δ(∆E) lim T . (5.46)
T →∞

Similar expressions hold of course for the delta functions in the other components of
the 4-momentum vector. Here, the integration interval [−L/2, L/2] is interpreted as the
boundaries of a large box of volume V = L3 , to which the process is confined.
We also have to worry about the normalization of the external states: remember that
one-particle plane wave states are normalized according to hp|qi = (2π)3 2Ep δ (3) (p − q).
To derive a physical observable, we would rather like to have them normalized to one.
This can be achieved by adding a factor [2Ep ]−1 for each particle. For the initial state
particles, we should also divide by V (the height of the δ-peak in a finite volume). For
the final states, we instead integrate over the 3-momentum (the phase space integral).
Finally, in a collider environment we are not colliding individual electron with individ-
ual positrons, but rather beams containing many particles. We should therefore multiply
by the total number of electrons, N− , and positrons, N+ .
In total, we thus have for the transition probability:

d3 k d3 k 0
Z
1 1
P = lim T V N+ N−
T,L→∞ 2Ep V 2Ep0 V (2π)3 2Ek (2π)3 2Ek0

· (2π)4 δ (4) (p + p0 − k − k 0 ) · |M|2 . (5.47)

The total transition rate is this quantity divided by T , and the total cross section is
total transition rate
σ(e+ e− → µ+ µ− ) = , (5.48)
# of target e− × flux of incident e+

90
 5.3 e+ e− → µ+ µ−

where we have arbitrarily called the e− the “target particles” and the e+ the “incident
particles”. The number of target e− in the denominator cancels the factor N− from
eq. (5.47). The e+ flux is given by the e+ number density, N+ /V times the relative
velocity vrel of the e+ and the e− . Note that also the factors T and V cancel, so that
our final expression for the cross section carries no trace of the fact that we temporarily
worked with a finite spacetime volume. In other words, the limit for T, L → ∞ is trivial
to take and we obtain a master formula for 2 → 2 cross sections:
d3 k d3 k 0
Z
+ − + − 1
σ(e e → µ µ ) =
2Ep 2Ep0 vrel (2π)3 2Ek (2π)3 2Ek0

·(2π)4 δ (4) (p + p0 − k − k 0 ) · |M|2 . (5.49)

For processes with more than two R particles in the final state, one simply has to add more
final state phase space factors d3 kj /[(2π)3 Ekj ].
For the 2 → 2 case, the 4-dimensional delta function can be used to remove for instance
the integral over d3 k 0 and the integral over d|k|, leaving only the integral over the solid
angle of the first final state particle. If the masses of the two final state particles are m
and m0 , we find
d3 k d3 k 0
Z
(2π)4 δ (4) (p + p0 − k − k 0 )
(2π)3 2Ek (2π)3 2Ek0
|k|2
Z Z p p

= dΩ d|k| δ E + E 0 − |k| 2 + m2 − |k|2 + m02
p p
(2π)2 4Ek Ek0
|k|2 |k| −1
 
|k|
Z
= dΩ +
(2π)2 4Ek Ek0 Ek Ek0
|k|
Z
= dΩ 2
(5.50)
(2π) 4(Ek + Ek0 )
If we are not interested in the total cross section, but in a differential cross section (for
instance dσ/dΩ), we simply omit the integral over the corresponding kinematic variable
(for instance Ω) in the final state. For instance, for e+ e− → µ+ µ− , we have in the center
of mass frame,
s
dσ |M| 2 m2µ
= 1 − , (5.51)
dΩ 128π 2 E 2 vrel E2
where E is the common energy of the electrons and muons.

5.3.5 The Cross Section for e+ e− → µ+ µ−

Let us plug the squared matrix element eq. (5.44) into the cross section formula eq. (5.51):
s
2 m2µ m2µ m2µ
    
dσ α 2
= 1− 2 1 + 2 + 1 − 2 cos θ . (5.52)
dΩ 16E 2 E E E

91
Chapter 5 Quantum Electrodynamics

Figure 5.2: The total cross section for e+ e− → µ+ µ− as a function of the center of
2 to better expose
mass energy Ecm ≡ 2E. The vertical axis has been multiplied by Ecm
the asymptotic behavior for Ecm → ∞. Figure taken from [1].

Here we have introduced the electromagnetic fine structure constant α ≡ e2 /(4π). The
total cross section is
s
2 m2µ 1 m2µ
 
πα
σtotal = 1− 2 1+ . (5.53)
3E 2 E 2 E2

This expression is plotted in fig. 5.2

5.3.6 e+ e− → µ+ µ− : Summary
Let us briefly summarize how we obtained the cross section for the annihilation of an
electron–positron pair to muons:

1. Draw all relevant Feynman diagrams (here, there was only one)

2. Apply the Feynman rules to find the amplitude M

3. Square it and average over initial state spins/sum over P final state spins. Use
eqs. (3.25) and (3.29) to rewrite expressions of the form s ūs (p)us (p) as traces
over momenta, masses and Dirac matrices.

4. Use the relations derived in section 5.3.2 to evaluate the traces.

5. Pick a suitable reference frame and write the 4-momenta in terms of the measurable
kinematic variables in that frame.

92
 5.4 More Technology for Evaluating QED Feynman Diagrams

6. Integrate |M|2 over the final state phase space according to the master formula
eq. (5.49) to obtain the cross section.

5.4 More Technology for Evaluating QED Feynman Diagrams

Now that we have gone through one QED calculation in its full glory, we already have
most of the tools we need to evaluate any tree level diagram in QED. We will practice
these computational techniques thoroughly in the exercises.
A few important tricks are still missing, though.

5.4.1 Scattering of Polarized Particles

In section 5.3.1, we argued that it is difficult though not impossible to collide polarized
beams of particles and to analyze the spin of final state particles. It is therefore sometimes
important to compute scattering amplitudes between particles of definite spin. We could
of course use the explicit expressions for the u and v spinors directly after applying the
Feynman rules (i.e. in eq. (5.25)), but the resulting expressions can be quite horrible.
In the high energy limit, where all particles are ultrarelativistic, there is a much more
elegant approach, based on the observation that chirality and helicity coincide in that
limit (see section 3.4.1. Namely, we can simply restrict the amplitude to particular
helicities by inserting the chirality projection operators PL , PR (see eq. (3.111)) in the
fermion currents. For instance, to consider the annihilation of a left-handed electron and
a right-handed positron, we would replace
0 0
v̄ s (p0 )γ µ us (p) → [PL v s (p0 )]† γ 0 γ µ PL us (p)
0 1 + γ5 s
= v̄ s (p0 )γ µ u (p) . (5.54)
2
(Remember that a left-handed v-spinor annihilates a right-handed positron.)

5.4.2 External Photons

According to the Feynman rules, an incoming or outgoing photon in a scattering ampli-
tude contributes a polarization vector µ (p) or µ∗ (p). If the polarization of photons is
unknown, we have to sum over it when computing the squared matrix element, just as
we did for fermions in eq. (5.29). In this context an extremely useful identity is
X
∗µ (p)ν (p) ↔ −gµν . (5.55)
polarizations

Here, ↔ means that the left hand side and the right hand side are not identical, but
when computing squared matrix elements they can be used interchangeably. Here is the

93
Chapter 5 Quantum Electrodynamics

√
proof: consider the two physical transverse polarization states ±
µ (p) ≡ (0, 1, ±i, 0)/ 2,
assuming that p is aligned with the z axis. Then
 
0
X  1
∗µ (p)ν (p) = 


 1 
s=+,−
0

= −gµν + êtµ êtν − êzµ êzν , (5.56)

where êµt = (1, 0, 0, 0) and êµz = (0, 0, 0, 1). Next, note that

pµ − (êt · p)êµt
êµz = . (5.57)
êt · p
We now argue that the term proportional to pµ on the right hand side does not contribute
to a Feynman amplitude. The argument is gauge invariance: the matrix element of any
physical process involving an external photon field Aµ (x) is invariant under Aµ (x) →
Aµ (x) + 1e ∂ µ α(x). In momentum space, this gauge transformation rule reads

i
õ (p) → õ (p) − pµ α̃(p) . (5.58)
e

For the external photon, õ (p) ∝ µ (p). Thus, any contribution ∝ pµ to µ (p) must
vanish. Therefore,
X
∗µ (p)ν (p) ↔ −gµν + êtµ êtν − êtµ ẑtν = −gµν . (5.59)
s=+,−

94
 Path Integrals
6
Now that we have developed and applied the techniques required to compute physical
observables in quantum field theory, it is useful to pause a moment and think about
the physics hiding behind the algebra. In this chapter, we will therefore discuss an
approach to QFT different from the canonical quantization procedure introduced in
chapters 2 and 3: the path integral formalism. In that formalism, a scattering amplitude
is computed by considering all possible phase space trajectories that particle could take
to go from the initial state to the final state. The amplitudes for all these paths are then
integrated up, true to the superposition principle of quantum mechanics.

6.1 Path Integrals in Quantum Mechanics

To introduce path integrals, let us first take several steps back and consider good old non-
relativistic QM again. In particular, we consider the motion of a particle with momentum
p and coordinate q in one dimension. The Hamilton operator is1

p̂2
Ĥ(p, q) = + V (q̂) . (6.1)
2m

Here, V (q) is the potential. We wish to compute the amplitude hq 0 |e−iĤt |qi for the
particle to propagate from q at time 0 to q 0 at time t. We split the time interval [0, t]
into n + 1 small subintervals of length δt ≡ t/(n + 1). Between intervals, we insert a
complete set of coordinate eigenstates:
n
Z Y 
0 −iĤt
hq |e |qi = dqj hq 0 |e−iĤ δt |qn ihqn |e−iĤ δt |qn−1 i · · · hq1 |e−iĤ δt |qi . (6.2)
j=1

1
For the moment, we write operators with a hat, e.g. Ĥ and c-numbers without.

95
Chapter 6 Path Integrals

We would like to split each exponential factor into the kinetic part and the potential
part using the Campbell-Baker-Hausdorff formula
1
eT̂ +V̂ = eT̂ eV̂ e− 2 [T̂ ,V̂ ]+··· . (6.3)
In the limit n → ∞, i.e. δt → 0, the commutator term, which is of order (δt)2 and all
higher order terms can be neglected. Let us moreover insert a complete set of momentum
eigenstates. Then, each factor in eq. (6.2) takes the form
p̂2  ED 
Z
−iĤδt dpj D     E
hqj+1 |e |qj i ' qj+1  exp − i δt pj pj  exp − iV (q̂) δt qj
 
2π 2m
(6.4)
Z
pj2
dpj h i
' exp − i δt − iV (qj ) δt eipj (qj+1 −qj ) (6.5)
2π 2m
p2j
Z
dpj h i
' exp − i δt − iV (qj ) δt + ipj q̇j δt . (6.6)
2π 2m
In the second line, we have used that hq|pi = eipq , and in the last line, we have used that
(q2 − q1 )/δt ' q̇. To write the factors corresponding to first and the last time step in the
form of eq. (6.6), we define q0 ≡ q, qn+1 ≡ q 0 . Equation (6.2) now becomes
Z Y n Y n   n  2  
0 −iĤt dpk X pk
hq |e |qi = dqj exp − i + V (qk ) − pk q̇k δ t . (6.7)
2π 2m
j=1 k=0 k=0

We nowQtake the continuum limit nQ→ ∞ and introduce the abbreviations Dq ≡

limn→∞ nj=1 dqj and Dp ≡ limn→∞ nk=0 dp k
2π for the integration measures. The sum
in the exponent becomes an integral in the continuum limit:
p2
Z  Z  
0 −iĤt
hq |e |qi = Dq Dp exp i dt pq̇ − − V (q) . (6.8)
2m
Note that the integrand in the exponent is just the Legendre transform of the Hamilto-
nian, i.e. the Lagrangian. Thus,
Z  Z 
hq 0 |e−iĤt |qi = Dq Dp exp i dt L(q, p) . (6.9)

The physical interpretation of eq. (6.9) is the following: each set {q1 , . . . , qn , p0 , . . . pn }
corresponds to one possible phase space Rtrajectory leading from q to q 0 . The amplitude
for each path is given by the action exp(i dt L). The path integral (integral over Dq Dp)
sums up the amplitudes for all the paths. In that sense, any quantum mechanical tran-
sition process can be viewed as a continuum generalization of the two-slit experiment.
Already at this stage, we can glimpse why the path integral formalism may be useful
in QFT. Imagine we want to compute the expectation value of the particle’s position at
a time t1 , somewhere in the interval [0, t]. It is given by
hq 0 |q̂(t1 )|qi ≡ hq 0 |e−iH(t−t1 ) q̂e−iH(t1 −0) |qi

96
 6.2 The Path Integral for a Free Scalar Field

Z  Z 
= Dq Dp q(t1 ) exp i dt L(q, p) . (6.10)

So far, so good. Now, let’s insert one more q factor. If t1 > t2 , we have

hq 0 |q̂(t1 )q̂(t2 )|qi ≡ hq 0 |e−iH(t−t1 ) q̂e−iH(t1 −t2 ) q̂e−iH(t2 −0) |qi

Z  Z 
= Dq Dp q(t1 )q(t2 ) exp i dt L(q, p) , (6.11)

while for t1 < t2 , it is

hq 0 |q̂(t2 )q̂(t1 )|qi ≡ hq 0 |e−iH(t−t2 ) q̂e−iH(t2 −t1 ) q̂e−iH(t1 −0) |qi

Z  Z 
= Dq Dp q(t1 )q(t2 ) exp i dt L(q, p) . (6.12)

Thus, overall,
(
hq 0 |q̂(t1 )q̂(t2 )|qi
Z  Z  if t1 > t2
Dq Dp q(t1 )q(t2 ) exp i dt L(q, p) = (6.13)
hq 0 |q̂(t2 )q̂(t1 )|qi if t2 > t1
= hq 0 |T q̂(t1 )q̂(t2 )|qi . (6.14)

In other words, if we had an efficient way of evaluating path integrals, we’d have a new
way of computing time-ordered correlation functions. And we have seen in our discussion
of the LSZ reduction formula (section 4.4) that these are of paramount importance for
computing scattering amplitudes.
We will now introduce the path integral formalism in QFT, we will argue that it leads
to the same Feynman rules as our previous approach involving creation and annihilation
operators (though in an arguably more elegant way), and we will use path integrals to
proof the Feynman rule for the photon propagator.

6.2 The Path Integral for a Free Scalar Field

Generalizing the results of the previous section to a field theory is straightforward. First,
the initial and final states, which were coordinate points in QM, are now scalar field
configuration φ(x, t). Also the Hamilton operator changes and becomes the one for a
free scalar field (see eq. (2.46)):
Z  
3 1 2 1 2 1 2 2
Ĥ = d x [π̂(x, t)] + [∇φ̂(x, t)] + m [φ̂(x, t)] , (6.15)
2 2 2

with the canonical momentum π(x, t) = φ̇(x, t). Finally, instead of discretizing the
trajectory only in time, we discretize all four spacetime directions and integrate over the
field value at each spacetime point. The corresponding integration measure is
Y
Dφ ≡ dφ(x1i , x2j , x3k , tl ) . (6.16)
i,j,k,l

97
Chapter 6 Path Integrals

R
Similarly, also the path integral over the canonical momentum Dπ, is now understood
as
Y dπ(x1i , x2j , x3k , tl )
Dπ ≡ . (6.17)
2π
i,j,k,l

The transition amplitude between an initial field configuration φ(x, 0) and a final field
configuration φ0 (x, t) is then, in complete analogy to eq. (6.2),
Z
0 −iĤt
hφ (x, t)|e |φ(x, 0)i = Dφ hφ0 (x, t)|e−iĤδt |φ(x, tn )i

· hφ(x, tn )|e−iĤδt |φ(x, tn−1 )i · · · hφ(x, t1 )|e−iĤδt |φ(x, 0)i . (6.18)

Inserting conmplete sets of canonical momentum eigenstates, each factor in this product
becomes
hφ0 (x, tm+1 )|e−iĤt |φ(x, tm )i

dπ(x1i , x2j , x3k , tm ) D

Z Y Z
  1  E
' φ(x, tm+1 ) exp − i d3 x [π̂(x, t)]2 δt π(x, tm )
 
2π 2
i,j,k
Z n1
D   1 o  E
· π(x, tm ) exp − i d3 x [∇φ̂(x, t)]2 + m2 [φ̂(x, t)]2 δt φ(x, tm )
 
2 2
(6.19)
Z Y 1 2 3
dπ(xi , xj , xk , tm )
Z
h n 1 1
' exp − i d3 x [π(x, tm )]2 + [∇φ(x, tm )]2
2π 2 2
i,j,k
1 o i
+ m2 [φ(x, tm )]2 − π(x, tm )φ̇(x, tm ) δt . (6.20)
2
Overall, we thus find in anology to eq. (6.7),

hφ0 (x, t)|e−iĤt |φ(x, 0)i

Z  Z t Z  
3 1 2 1 2 1 2 2
= Dφ Dπ exp i dt d x π φ̇ − π − (∇φ) − m φ (6.21)
0 2 2 2
The path integral over Dπ can actually be evaluated directly. We have for each spacetime
point x = (x1i , x2j , x3k , tm ) an expression proportional to

dπ(x1i , x2j , x3k , tm )

Z  
4 i 2 4
exp iπ(x)φ̇(x) δ x − π (x) δ x . (6.22)
2π 2
This Gaussian integral can be evaluated after regularizing it by multiplying the exponent
by (1 − i), where  is an infinitesimal positive real number. The integral yields

2πi 1/2
   
1 4 2
exp i δ x φ̇ (x) . (6.23)
δ4x 2

98
 6.2 The Path Integral for a Free Scalar Field

The prefactor can be absorbed into a redefinition of the integration measure Dφ. Then,
the matrix element eq. (6.21) is
Z  Z t Z  
0 −iĤt 3 1 µ 1 2 2
hφ (x, t)|e |φ(x, 0)i = Dφ exp i dt d x (∂µ φ)(∂ φ) − m φ
0 2 2
Z  Z t Z 
= Dφ exp i dt d3 x L(φ̇, φ) . (6.24)
0

In QFT, an object of central importance will be the transition amplitude from the vacuum
at t = −∞ to the vacuum at t = +∞. It is called the partition function and is given by
Z  Z 
Z0 ≡ h0(t = +∞)|e−iĤt |0(t = −∞)i = Dφ exp i d4 x L(φ̇, φ) . (6.25)

Note the allusion to statistical mechanics implied by the term “partition function”. In
statistical mechanics, the partition function Z ≡ n hn|e−Ĥ/T |ni sums up the Boltzmann
P
factors for all microscopic states |ni of a system. In the path integral formalism of QFT,
the states become trajectories in configuration space, the temperature is set to 1, and
there is an extra factor of i in the exponent. Since a non-interacting system starting out
in its ground state will remain there forever, Z0 = 1 in the free theory. Defining the
partition function still makes sense because it will differ from unity as soon as we add
an interaction.
Note that, as in QM, the path integral formalism provides a way of writing time-
ordered correlation functions:
Z  Z 
4
h0|T φ(x1 ) · · · φ(xn )|0i = Dφ φ(x1 ) · · · φ(xn ) exp i d x L(φ̇, φ) . (6.26)

Before concluding this section, we will add a very particular interaction to the La-
grangian: a source term

Lsource ≡ φ · J . (6.27)

This term can be interpreted as a vertex that generates a particle out of nowhere, or
annihilates a particle into nothingness, without requiring the presence of another parti-
cle. The real coefficient function J(x) gives the strength of such processes and can be
interpreted as a classical (non-quantized) field. The reason for adding the source term
will become clear shortly. But first, let us define
Z  Z 
4


Z0 [J] ≡ Dφ exp i d x L(φ̇, φ) + φJ . (6.28)

Using the functional derivative δ/δf (x), defined via

δ
f (x2 ) ≡ δ (4) (x1 − x2 ) , (6.29)
δf (x1 )

99
Chapter 6 Path Integrals

we can then write time-ordered correlation functions very elegantly:

1 δ 
h0|φ(x1 )|0i = Z0 [J]J=0 , (6.30)
i δJ(x1 )
  
1 δ 1 δ 
h0|T φ(x1 )φ(x2 )|0i = Z0 [J]J=0 , (6.31)
i δJ(x1 ) i δJ(x2 )

and so on. Because of these relations, Z[J] is also called the generating functional of the
theory.

6.3 The Feynman Propagator from the Path Integral

Let us rewrite the action for a free real scalar field
Z  
4 1 µ 1 2 2
S0 = d x (∂µ φ)(∂ φ) − m φ + φ J (6.32)
2 2
in Fourier space, using
Z 4
d k −ikx
φ(x) ≡ e φ̃(k) . (6.33)
(2π)4
We obtain
d4 k
Z  
1 2 2


˜ ˜
S0 = φ̃(k) k − m φ̃(−k) + J(k)φ̃(−k) + J(−k)φ̃(k) . (6.34)
2 (2π)4

Note that we have split the Fourier transform of the source term into two (identical) pieces
and added a factor 1/2. Since the Fourier transform is a linear unitary transformation,
we can replace the path integral over Dφ in the definition of Z0 [J] by a path integral
over Dφ̃. The Jacobian corresponding to this substitution is 1.
Our goal is to simplify eq. (6.34). To this end, let us complete the square (in φ̃) in the
exponent by defining
˜
J(k)
χ̃(k) ≡ φ̃(k) + . (6.35)
k 2 − m2 + i
This substitution, being a constant shift, does not change the integration measure either,
i.e. we can simply replace Dφ̃ by Dχ̃. The action now becomes

1
Z 4 
d k ˜ J(−k)
J(k) ˜ 
2 2


S0 = χ̃(k) k − m χ̃(−k) − 2 (6.36)
2 (2π)4 k − m2 + i

and the partition function is consequently

Z  Z 4  
i d k 2 2


Z0 [J] = +∞ h0|0i−∞ = Dχ̃ exp χ̃(k) k − m χ̃(−k)
2 (2π)4

100
 6.4 Wick’s Theorem from the Path Integral


i
Z ˜ J(−k)
d4 k J(k) ˜ 
· exp − . (6.37)
2 (2π)4 k 2 − m2 + i

The first line is just Z0 [J = 0] = 1. The second line does not depend on χ̃ any more—we
have gotten rid of the path integral. We transform the second line back to coordinate
space using the obvious definition
Z
˜
J(k) ≡ d4 x eikx J(x) . (6.38)

Note that the k-dependent term obtained that way is just the Feynman propagator
DF (x − y). We thus obtain
 Z 
1 4 4 0 0 0
Z0 [J] = exp − d x d x J(x )DF (x − x )J(x) . (6.39)
2

6.4 Wick’s Theorem from the Path Integral

We have already seen in section 6.2 how time-ordered correlation functions can be con-
structed by repeatedly applying the functional derivative with respect to J(x) to the
partition function Z0 [J]. Let us now apply this procedure to the path integral-less form
of Z0 [J], eq. (6.39).
For the two-point function, this yields
  
1 δ 1 δ 
h0|T φ(x1 )φ(x2 )|0i = Z0 [J]J=0
i δJ(x1 ) i δJ(x2 )
 Z
1 1 δ
d4 x DF (x2 − x)J(x)Z0 [J]J=0

=−
i i δJ(x1 )
  
= DF (x2 − x1 ) + (terms proportional to J) · Z0 [J]J=0
= DF (x2 − x1 ) . (6.40)

We knew from section 2.4.3 that this had to be the outcome, but it is reassuring to see
that it can be obtained in the path integral formalism as well.
Consider next the 4-point function. By applying the functional derivative four times,
we obtain

h0|T φ(x1 )φ(x2 )φ(x3 )φ(x4 )|0i

= DF (x1 − x2 )DF (x3 − x4 ) + DF (x1 − x3 )DF (x2 − x4 ) + DF (x1 − x4 )DF (x2 − x3 )

1 2 1 2 1 2

= + + . (6.41)

3 4 3 4 3 4

101
Chapter 6 Path Integrals

This generalizes to arbitrary (even) numbers of functional derivative: each functional

derivative can act either on the factor Z0 [J], bringing an additional propagator (multi-
plied by J(x)) down, or it can act on one of the J(x) factors brought down by one of the
previous functional derivatives. In total, we obtain one term for each way of picking out
pairs of points from the set {x1 , . . . , xn } and connecting them with Feynman propagators
DF (xj − xk ). This is just Wick’s theorem, now proven in a completely different way than
in section 4.2.
Correlation functions of an odd number of fields will always vanish because there will
always be at least one factor of J in every term.

6.5 Interacting Field Theories in the Path Integral Formalism

The step from a free theory to an interacting theory in the path integral formalism is
quite straightforward. We split the Lagrangian into the free piece and the non-interacting
piece. For instance, in φ4 theory, we have
1 1
L0 = (∂µ φ)(∂ µ φ) − m2 φ2 (6.42)
2 2
and
λ
L1 = − φ4 . (6.43)
4!
We then define the partition function for the interacting theory,
Z  Z   Z 
4 4


Z1 [J] ∝ Dφ exp i d x L1 exp i d x L0 + φJ . (6.44)

By expanding the first exponential (but not the second), we recover a series of time-
ordered correlation function that coincides precisely with the perturbation series in the
numerator of our master formula, eq. (4.40). Wick’s theorem then tells us that, to
evaluate the correlation function, we have to consider all possible ways of connecting
the φ factors by propagators. We thus immediately recover the Feynman rules from
section 4.3.
We have thus rederived the formalism for evaluating correlation functions in terms of
Feynman diagrams without ever having to invoke creation and annihilation operators.
From here on, the remaining step towards the computation of physical observables are
the same as before: the LSZ formula relates correlation functions to matrix elements,
which are then integrated over phase space.
Note that all correlation functions we encountered in the path integral formalism so
far were correlation functions of the free theory, as they must be for our master formula
to be applicable. As we have seen in eq. (4.40), going from the free theory to the full
theory leads to an additional normalization factor, and this factor is the reason we have
written “∝” instead of “=” in eq. (6.44).

102
 6.6 Quantization of the Photon Field

6.6 Quantization of the Photon Field

With the path integral formalism at our disposal, we can now justify the expression for
the photon propagator in QED (eq. (5.13)),

d4 p −ig µν e−ip(x−y)
Z
µν
∆ (x − y) ≡ . (6.45)
(2π)4 p2 + i

Consider the path integral for the free photon field,

Z  Z  1 
4 µν µ
Z0 [J] = DA exp i d x − Fµν F + J Aµ . (6.46)
4
R R
Here, DA = DA0 DA1 DA2 DA3 implies the path integral over all four components
of Aµ . Note that also the source J µ is now a Lorentz vector. In other words, there is a
separate source for each component of Aµ . J µ must satisfy ∂ µ Jµ = 0 in order to preserve
gauge invariance. (To see this, consider how the term J µ Aµ transforms under a gauge
transformation, and integrate by parts.)
For the following discussion, it will be useful to rewrite
Z Z
1 4 µν 1
d4 x (∂ µ Aν )(∂µ Aν ) − (∂ µ Aν )(∂ν Aµ )
 
− d x Fµν F = −
4 2
Z
1
d4 x Aν ∂ µ ∂µ Aν − Aν ∂ µ ∂ν Aµ .
 
= (6.47)
2

In the second step, we have integrated by parts. As we did for the scalar field in eq. (6.34),
we now transform the path integral to Fourier space:
Z  Z 4 
i d k
Z0 [J] = Dà exp − õ (k)(k 2 g µν − k µ k ν )Ãν (−k)
2 (2π)4

˜µ ˜µ
− J (k)õ (−k) − J (−k)õ (k) . (6.48)

We have again split up the source term into two identical pieces and included a factor
1/2 to compensate for the implied double counting.
As in section 6.3, we would like to complete the square in the exponent to rewrite
Z0 [J] as Z0 [0] = 1, multiplied by a term not involving a path integral any more, but
containing J and the photon propagator. Doing so would require us to invert the matrix
k 2 g µν − k µ k ν ≡ k 2 P µν , so that we could define B̃ µ (k) ≡ õ (k) − [(k 2 P )−1 ]µν J˜µν (k).
Unfortunately, the matrix P is singular because P µν kν = 0.
The reason for the singular nature of P is gauge invariance. L is invariant under
A (x) → Aµ (x)+(1/e)∂ µ α(x), which in momentum space translates to õ (k) → õ (k)−
µ

(i/e)k µ α̃(k). This implies that the component of Aµ (k) proportional to k µ is unphysical,
but it also suggests a solution to our difficulties defining the photon propagator: we
simply drop the unphysical component of the photon field and redefine DÃ to mean

103
Chapter 6 Path Integrals

path integration only over components of à orthogonal to k, i.e. k µ õ (k) = 0. In other
words, we fix the gauge to the Lorenz gauge. Within this subspace, P µν is just the
identity matrix since

kµ kν kν k λ
  
µν λ µν λ
P (k)Pν (k) = g − 2 gν − 2
k k
µ
k k λ
= g µλ − 2
k
= P µλ (k) (6.49)

and

g µν Pµν (k) = 3 . (6.50)

The first relation implies that P µν (k) is a projection operator, i.e. its eigenvalues can
only be zero or one. The second relation gives just the trace of P µν and implies that three
of the eigenvalues are 1 and the fourth one is zero. We have already argued that the zero
eigenvalue corrresponds to the direction of k µ , which we have removed from the path
integral. Therefore, in the remaining subspace, all three eigenvalues are 1 and P µν (k)
must be the identity matrix. Therefore, in this subspace, (k 2 P µν )−1 = (1/k 2 )P µν . We
define

P µν (k) J˜µν (k)

B̃ µ (k) ≡ õ (k) − (6.51)
k2
and turn the partition function into
Z  Z 4  
i d k µ 2 ν ˜µ Pµν ˜ν
Z0 [J] = DB̃ exp − B̃ (k) k Pµν B̃ (−k) − J (k) 2 J (−k)
2 (2π)4 k
 Z 
1
= exp − d4 x d4 x0 J˜µ (x0 )∆µν (x − x0 )J˜ν (x) . (6.52)
2

In the second step, we have used that Z0 [0] = 1 to eliminate the path integral over DB̃.
We have also exploited the fact that k µ Jµ (x) = 0, so that replacing P µν = g µν − k µ k ν /k 2
by g µν leaves Z0 [J] invariant. Finally, we emphasize that the integration measure DB̃
implies integration only over the components of B̃ that are orthogonal to k µ .

6.7 Path Integrals for Fermions

We now generalize the path integral formalism to fermions. The problem is that we must
somehow reproduce the Pauli exclusion principle—intimately related to the anticommu-
tation relations of fermion fields—in a formalism that does not involve operators, but
only numbers. The solution is is to invoke so-called anticommuting number or Grassmann
numbers, which we will now introduce.

104
 6.7 Path Integrals for Fermions

6.7.1 Grassmann Numbers

The set G of (real) Grassmann numbers is defined by the property that, for θ, η ∈ G,

θη = −ηθ . (6.53)

For finite sets of Grassmann numbers, one can construct matrix representations of G.
The above definition is, however, more general. One immediate consequence of eq. (6.53)
is that

θ2 = 0 . (6.54)

Therefore, any real or complex valued function f (θ) can be written as

f (θ) = A + Bθ . (6.55)

In other words, the Taylor expansion of f (θ) terminates after the linear term. Note
that A is an ordinary number and B is a Grassmann number, so that Bθ is an ordinary
number.
R We will also want to integrate over Grassmann numbers, and we therefore define
dθ f (θ) by requiring the properties
Z Z
dθ c · f (θ) = c · dθ f (θ) , (6.56)
Z Z Z
 
dθ f (θ) + g(θ) = dθ f (θ) + dθ g(θ) , (6.57)
Z Z
dθ f (θ + η) = dθ f (θ) , (6.58)
Z
dθ 1 = 0 . (6.59)

for θ, η ∈ G and c ∈ C. These properties, inRparticular the invariance under shifts of the
integration variable, can be satisfied only if dθ f (θ) is a constant times B. We therefore
define
Z Z
dθ f (θ) = dθ (A + Bθ) ≡ B . (6.60)

For multidimensional Grassmann integrals, the ordering of the integrals and the Grass-
mann numbers in the integrand is important. We choose the convention
Z
dθ dη η θ = +1 . (6.61)

To take derivatives with respect to Grassmann numbers, we use the definition

d d
θη = − ηθ = −θ . (6.62)
dη dη

105
Chapter 6 Path Integrals

We can define a complex Grassmann number θ according to

1
θ ≡ √ (θr + iθi ) , (6.63)
2
√
where θr and θi are real Grassmann numbers. The prefactor 1/ 2 is convention. The
complex conjugate of θ is
1
θ∗ ≡ √ (θr − iθi ) . (6.64)
2
When taking the complex conjugate of a product of complex Grassmann numbers, we
again have to worry about their ordering. We define

(θη)∗ ≡ η ∗ θ∗ = −θ∗ η ∗ . (6.65)

When integrating over θ, we can either treat θr , θi as independent degrees of freedom,

or θ, θ∗ . We will choose the latter convention and thus have
Z
dθ∗ dθ θ θ∗ = 1 . (6.66)

Let us consider a few examples for working with Grassmann numbers:

1. A Gaussian integral: for complex Grassmann variables θ, θ∗ and a real number b,

we compute
Z Z
∗
dθ∗ dθ e−θ bθ = dθ∗ dθ 1 − θ∗ bθ = b .


(6.67)

∗
Compare this to the corresponding integral with only real numbers, dx dx∗ e−x bx =
R

2π/b.

2. Another Gaussian integral:

Z
∗
dθ∗ dθ θ θ∗ e−θ bθ = 1 . (6.68)

3. A multidimensional Gaussian integral. Let B be a Hermitian matrix with elements

Bij ∈ C. We would like to evaluate
Z
∗
I = dθ1∗ dθ1 · · · dθn∗ dθn e−θi Bij θj . (6.69)

We first note that, in the Taylor expansion of the exponential, Q the only term that
contributed is the one in which all θi , θi∗ appear exactly once: i θi∗ θi . The strategy
is to transform to an eigenbasis of B by applying a unitary transformation θi → θi0 =

106
 6.7 Path Integrals for Fermions

Uij θj . We will denote the eigenvalues of B by bk The product of all θi transforms

into
Y 1
θi0 = j1 j2 ··· θj0 2 θj0 2 · · · (6.70)
n!
i
1 j1 j2 ···
=  Uj1 k1 θk1 Uj2 k2 θk2 · · · (6.71)
n!
1 Y 
= j1 j2 ··· k1 k2 ··· Uj1 k1 Uj2 k2 · · · θi (6.72)
n!
i
Y 
= (det U ) θi (6.73)
i
Y
= θi . (6.74)
i

In the last step we have used the unitarity of U . We can now write
Z P 0∗ 0
I = dθ1∗ dθ1 · · · dθn∗ dθn e− k θk bk θk (6.75)
Z Y
= (−1)n dθ1∗ dθ1 · · · dθn∗ dθn bk θk∗ θk (6.76)
k
Y
= bk (6.77)
k
= det B . (6.78)
This should again be compared to the corresponding integral over ordinary num-
∗
bers, dx∗1 dx1 · · · dx∗n dxn e−xi Bij xj = (2π)n / det B.
R

6.7.2 Partition Function, Functional Derivative and Correlation Functions

for Fermions
Consider the free Dirac field with the Lagrangian
L0 = ψ̄(i∂/ − m)ψ + η̄ψ + ψ̄η . (6.79)
Note that we have again added source terms. The crucial point in the following will
be that we treat ψ(x), ψ̄(x), η(x) and η̄(x) as spinor-valued functions, with the entries
of the spinors being Grassmann numbers. We define the partition function for the free
fermion field as
Z h Z i
Z0 [η, η̄] ≡ Dψ Dψ̄ exp i d4 x L0 . (6.80)

Remember that for the free scalar field, we were able to rewrite the partition function in
a form without the path integral as
 Z 
1 4 4 0 0 0
Z0 [J] = exp − d x d x J(x )DF (x − x )J(x) , (6.81)
2

107
Chapter 6 Path Integrals

(see eq. (6.39)). To do the sameR for Z0 [η, η̄], we follow completely analogous arguments.
We transform the action S0 = d4 x L0 into Fourier space, using
Z 4
d k −ikx
ψ(x) ≡ e ψ̃(k) , (6.82)
(2π)4
Z 4
d k −ikx
η(x) ≡ e η̃(k) . (6.83)
(2π)4
This yields

d4 k h ˜
Z i
S0 = ψ̄ (k)(k/ − m) ψ̃(k) + ˜(k)ψ̃(k) + ψ̄˜(k)η̃(k) .
η̄ (6.84)
(2π)4

We then shift the field ψ̃(k) by defining

η̃(k)
χ̃(k) ≡ ψ̃(k) + . (6.85)
k/ − m + i
This leads to
Z 4 h
d k 1 i
S0 = ˜(k)(k/ − m)χ̃(−k) − η̄˜(k)
χ̄ η̃(k) (6.86)
(2π)4 k/ − m + i
0
d4 k eik(x −x)
Z Z
= d4 x χ̄(x)(i∂/ − m)χ(x) − d4 x d4 x0 η̄(x) η(x0 ) . (6.87)
(2π)4 k/ − m + i
from which it follows that
h Z i
Z0 [η, η̄] ≡ Z0 [0, 0] exp − d4 x d4 x0 η̄(x)SF (x − x0 )η(x0 ) . (6.88)

Here SF (x − x0 ) is the Feynman propagator for fermions,

0
d4 k ie−ik(x−x )
Z
SF (x − x0 ) = . (6.89)
(2π)4 k/ − m + i

Note that Z0 [0, 0] = 1 since a field that is initially in the vauum state will remain there
forever in the absence of external sources.
We can again obtain time-ordered correlation functions by taking functional derivatives
of Z0 [η, η 0 ]. For instance
  
1 δ 1 δ 
h0|T ψ(x1 )ψ̄(x2 )|0i = − Z0 [η, η̄]η=η̄=0 . (6.90)
i δ η̄(x1 ) i δη(x2 )

Note the extra minus sign in the functional derivative with respect to η(x2 ). Its origin
is the fact that, in the last term in the Lagrangian eq. (6.79), η appears to the right
of ψ̄, so that the derivative operator δ/δη(x2 ) needs to be anticommuted past ψ̄ before
evaluating it.

108
 6.8 The Quantum Equations of Motion: Schwinger–Dyson Equations

6.8 The Quantum Equations of Motion: Schwinger–Dyson

Equations
Consider an n-point correlation function in a scalar field theory,
Z R 4
hΩ|φ(x1 ) · · · φ(xn )|Ωi ∝ Dφ φ(x1 ) · · · φ(xn ) ei d x L[φ] . (6.91)

In quantum mechanics,
R we are used to deriving equations of motion by demanding that
the action S = dt L is invariant under infinitesimal variations of the system’s phase
space trajectory. Let us consider now infinitesimal variations of the field φ(x):

φ(x) → φ0 (x) ≡ φ(x) + (x) . (6.92)

Since the path integral means that the field value φ(x) is integrated over at each spacetime
point, this shift is nothing but a simple substitution of variables in the path integral.
At each spacetime point, eq. (6.92) is just a constant shift, which leaves the integration
measure Dφ invariant. Therefore,
Z Z
i d4 x L[φ] 0
R R 4
Dφ φ(x1 ) · · · φ(xn ) e = Dφ φ0 (x1 ) · · · φ0 (xn ) ei d x L[φ ] . (6.93)

Expanding in (x), this leads to

Z  Z Z 
i d4 x L[φ]
R δ 4 0 0
0 = Dφ e i d4 x (x)φ(x1 ) · · · φ(xn ) d x L[φ(x )]
δφ(x)
n
X 
+ φ(x1 ) · · · φ(xj−1 )(xj )φ(xj+1 ) · · · φ(xn ) (6.94)
j=1

The functional derivative in this expression is

Z 
δ 4 0 0 ∂L ∂L
d x L[φ(x )] = − ∂µ . (6.95)
δφ(x) ∂φ(x) ∂(∂µ φ(x))

This is just the expression that vanishes in the classical theory according to the Euler–
Lagrange equations of motion. In the terms in the second line of eq. (6.94), we can
write
Z
(xj ) = d4 x (x) δ (4) (x − xj ) . (6.96)

Noting now that eq. (6.94) has to hold for any (x), and writing the path integral in in
terms of correlation functions, we find
Z  E
D  δ 4 0 0
ΩT φ(x1 ) · · · φ(xn ) d x L[φ(x )] Ω
 
δφ(x)

109
Chapter 6 Path Integrals

n
X
ΩT φ(x1 ) · · · φ(xj−1 ) iδ (4) (x − xj ) φ(xj+1 ) · · · φ(xn )Ω .

  
= (6.97)
j=1

These equations are called the Schwinger–Dyson equations. The terms involving δ-
functions on the right hand side are called contact terms because they are non-zero
only when x equals one of the xj . The Schwinger–Dyson equations can be interpreted as
the equations of motion of the quantum theory: they tell us that the classical equations of
motion (which would dictate that eq. (6.95) vanishes) still hold inside a correlation func-
tion, provided the coordinate x does not equal any of the xj appearing in that correlation
function.

6.9 The Ward–Takahashi Identity

We now carry out a similar derivation as in the previous section for QED. Instead of the
general shifts considered above, we focus in particular on the transformation

ψ(x) → ψ 0 (x) ≡ [1 + ie α(x)]ψ(x) , (6.98)

with α(x) infinitesimal. This is just an infinitesimal gauge transformation on ψ(x). Note,
however, that we do not apply the corresponding gauge transformation to Aµ (x), i.e. the
Lagrangian is not invariant under eq. (6.98). Instead, it transforms as

LQED [ψ, ψ̄, A] → LQED [ψ 0 , ψ̄ 0 , A] ≡ LQED [ψ, ψ̄, A] − e [∂µ α(x)] ψ̄γ µ ψ . (6.99)

The crucial point is that the Lagrangian changes only by a term containing a deriva-
tive of α(x). The reason is that LQED is invariant under global (i.e. x-independent)
transformations ψ(x) → (1 + ieα)ψ(x).
Consider now the transformation of the path integral
Z R 4
Dψ̄ Dψ DA ei d x LQED [ψ,ψ̄,A] ψ(x1 )ψ̄(x2 )
Z
0 0
R 4
= Dψ̄ Dψ DA ei d x LQED [ψ ,ψ̄ ,A] ψ 0 (x1 )ψ̄ 0 (x2 ) . (6.100)

The identity follows from the fact that Dψ and Dψ̄ are invariant under eq. (6.98). Ex-
panding the right hand side of eq. (6.100) in α(x), and introducing the notation

j µ (x) ≡ e ψ̄(x)γ µ ψ(x) , (6.101)

we obtain
Z  Z
i d4 x LQED [ψ,ψ̄,A]
R
0= Dψ̄ Dψ DA e d4 x [∂µ α(x)] j µ (x) ψ(x1 )ψ̄(x2 )
−i

   
+ ieα(x1 )ψ(x1 ) ψ̄(x2 ) + ψ(x1 ) − ieα(x2 )ψ̄(x2 )

110
 6.9 The Ward–Takahashi Identity

Z Z 
i d4 x LQED [ψ,ψ̄,A]
R
= Dψ̄ Dψ DA e d x iα(x) [∂µ j µ (x)] ψ(x1 )ψ̄(x2 )
4


+ ieδ (4) (x − x1 )α(x1 )ψ(x1 )ψ̄(x2 ) − ieδ (4) (x − x2 )ψ(x1 )α(x2 )ψ̄(x2 ) .
(6.102)

In the last step, we have integrated by parts in the first term and pulled the last two
terms under the integral by introducing appropriate δ functions. SinceR eq. (6.102) has
to hold for any infinitesimal α(x), it must hold also when the integral d4 x is omitted.
Written in terms of correlation functions, this implies

i∂µ h0|T j µ (x)ψ(x1 )ψ̄(x2 )|0i = − ieδ (4) (x − x1 )h0|T ψ(x1 )ψ̄(x2 )|0i
+ ieδ (4) (x − x2 )h0|T ψ(x1 )ψ̄(x2 )|0i . (6.103)

This identity, which is called the Ward–Takahashi identity, can be expressed diagram-
matically as
x1 x1 x1
 
 
i∂µ · x
 (4) (4)

= δ (x − x1 )
ie  −δ (x − x2 ) .

 

x2 x2 x2 (6.104)

Note that the left hand side is just the full QED vertex (including all quantum correc-
tions), but with the photon field Aµ (x) replaced by a derivative ∂µ . The contact terms
on the right hand side involve, besides the δ-functions, the full propagator.
Let us elaborate more on the connection between the left hand side of the Ward–
Takahashi identity and the correlation function h0|T Aµ (x)ψ(x1 )ψ̄(x2 )|0i (the full QED
vertex), and also on its connection to a corresponding scattering amplitude hf |ii obtained
by applying the LSZ reduction formula to it. We use the equations of motion for the
photon field. Working in Lorenz gauge ∂µ Aµ = 0, they read

∂ ν ∂ν Aµ (x) = j µ (x) . (6.105)

We know from the previous section that these equations of motion still hold inside a
correlation function, up to contact terms. Therefore, eq. (6.103) is equivalent to

i∂µ ∂ 2 h0|T Aµ (x)ψ(x1 )ψ̄(x2 )|0i = contact terms . (6.106)

The double derivative ∂ 2 on the left hand side appears also in the LSZ reduction formula.
We now put in all the remaining pieces of that formula except the polarization vector
µ (pk). (Since Aµ (x) is already contracted with ∂µ , we would not have anything to

111
Chapter 6 Path Integrals

contract it with.) On the left-hand side of the LSZ formula, we write the scattering
amplitude as

hf |ii = µ (k) · iMµ (p, k, q) . (6.107)

The Feynman rules tell us that this is always possible. We obtain

Z Z Z
−ikµ M (p, k, q) = d x d x1 d4 x2 e−ikx ∂ 2 eiqx1 ū(q)(i∂/x1 − m)
µ 4 4
 

←−
× h0|T ∂µ Aµ (x)ψ(x1 )ψ̄(x2 )|0i (−i ∂/ x2 − m)u(p) e−ipx2 .
 

(6.108)
Z Z Z
d4 x d4 x1 d4 x2 e−ikx eiqx1 ū(q)(i∂/x1 − m)
 
= −i
←−
× (contact terms) (−i ∂/ x2 − m)u(p) e−ipx2 .
 
(6.109)

Let us first consider eq. (6.108). Integrating by parts over x1 , it turns into
Z
−ikµ Mµ (p, k, q) = lim (−i) d4 x1 eiqx1 ūs (q)(/q − m) . . . h0|T ψ(x1 ) . . . |0i
 
q 2 →m2

= lim (−i) ūs (q)(/q − m) . . . h0|T ψ̃(q) . . . |0i .

 
(6.110)
q 2 →m2

We have written out only the pieces corresponding to ψ(x1 ) here. The dots indicate the
other pieces. The limit is introduced to avoid the otherwise infinite expression 1/(q 2 −
m2 ), which would appear in the correlation function according to the Feynman rules. In
the
R 4 second line of eq. (6.110), we have introduced the Fourier-transformed field ψ̃(q) =
iqx
d x e ψ(x). Since the scattering matrix element hf |ii must be finite (up to the overall
energy-momentum conserving δ function), the correlation function h0|T ψ̃(q) . . . |0i must
include a pole proportional to 1/(/q − m) to cancel the factor /q − m from the LSZ formula,
and similar poles also for all other external particles. For the specific case of the full
QED vertex, the requisite pole structure is
1 1 1
. (6.111)
/ − m k2
/q − m p
When we discussed the Feynman rules for correlation functions, we saw that these pole
corresponded to the propagators attached to the external vertices. It is the coefficient of
all these poles that gives hf |ii. Terms that are lacking one or several of the poles do not
contribute since they go to zero in the limit q 2 → m2 , p2 → m2 , k 2 → 0.
Consider now the contact terms (right hand side of eq. (6.106) and last expression in
eq. (6.109)). The δ-function in a contact term eliminates one of the coordinate integrals
and simplifies the momentum dependence. For instance, consider the term containing
the delta function δ (4) (x − x1 ). After eliminating the x or x1 integral using this delta
function, the resulting term depends only on q−k, but not on the orthogonal combination

112
 6.9 The Ward–Takahashi Identity

q + k. Therefore, it cannot have separate 1/k 2 and 1/(/q − m) poles. It thus does not
have the structure of eq. (6.111) and therefore does not contribute to hf |ii.
These arguments can be generalized to arbitrary correlation functions and also to
theories other than QED. Therefore, we conclude that contact terms never contribute to
scattering matrix elements.
Coming back to the electron vertex function, we can now conclude from eqs. (6.108)
and (6.109) that

k µ Mµ = 0 . (6.112)

The implications of this become more clear when expressed diagrammatically:

q
k
→
−
kµ · = 0. (6.113)
p

In other words, when we take the matrix element describing the interaction of a photon
with two fermions and replace the photon polarization vector µ (k) by k µ , the result is
zero.
Of course, we could have carried out a fully analogous derivation also for QED diagrams
with more external vertices. The result would be the same: any QED scattering matrix
element µ (k)Mµ satisfies kµ Mµ = 0.
Remember that our starting point in this section was the invariance of LQED under
global gauge transformations (the statement that under a local gauge transformation
LQED changes only by a term containing a derivative of α(x)). The Ward–Takahashi
identity and the relation kµ Mµ = 0 are direct consequences of this. Another way of
phrasing eq. (6.112) is to say that the amplitude for the production of a (hypothetical)
longitudinal photon vanishes. It is often useful when computing complicated Feynman
diagrams to check the calculation by verifying that eq. (6.112) is satisfied.

113
Chapter 6 Path Integrals

114
 Weyl and Majorana Fermions
7
In section 3.4.1, we have already argued that the left-chiral and right-chiral components
of a 4-component spinor belong to different representations of the Lorentz group. (We
have argued that the transformation matrices are block diagonal, thus not mixing the
upper and lower components of a 4-component spinor.) In the Lagrangians we have
encountered so far, the only term that mixes the left-handed and right-handed pieces is
the mass term. For instance, consider the QED Lagrangian. Using

1 − γ5 1 + γ5
ψ(x) = ψ(x) + ψ(x)
2 2
= PL ψ(x) + PR ψ(x)
≡ ψL (x) + ψR (x) , (7.1)

it can be written as
1
LQED = ψ̄(i∂/ − m)ψ − eψ̄γ µ ψAµ − Fµν F µν
4
/ L − eψ̄L γ µ ψL Aµ + ψ̄R (i∂)ψ
= ψ̄L (i∂)ψ / R − eψ̄R γ µ ψR Aµ

1
− mψ̄L ψR + h.c. − Fµν F µν . (7.2)
4
We see that, indeed, apart from the mass term, this looks like the Lagrangian of two
independent fields, each coupled to the photon. Group theory allows one to classify
all possible representations of the Lorentz group SO+ (1, 3) and give them labels.1 By
convention, the representation corresponding to ψL is called ( 12 , 0), an the representation

1
When we write “Lorentz group” in the following, we consider for simplicity only the proper or-
thochronous Lorentz group, not the more general Lorentz group O(1, 3), which includes also time
reversal and parity transformations.

115
Chapter 7 Weyl and Majorana Fermions

corresponding to ψR is called (0, 12 ). Other repsentations include (0, 0) (Lorentz scalar)

and ( 21 , 12 ) (Lorentz vector).
In the following, we will exploit the possibility to separate left-handed from right-
handed spinors a little further. This is a preparation for a discussion of the Standard
Model of particle physics in the second part of this course, where the left-handed and
right-handed fermions in fact have different interactions. It is moreover useful when
studying massless or approximately massless fermions, for instance in collider physics
(where the quark masses are negligible compared to the typical collision energies) and in
neutrino physics (where we’re dealing with particles whos mass is  eV).

7.1 Spinor Indices

We first introduce some useful notation, recapitulating along the way the Lorentz trans-
formation properties of spinors.

7.1.1 Left-handed spinors

Consider a left-handed spinor Ψ(x). It can be understood as a 4-component object with
the lower two components set to zero, Ψ(x) = (ψ(x), 0)T . However, it is cumbersome to
always carry the zero components along, therefore we will in the following only consider
two-component spinors, in this case ψ(x). We will write out their spinor indices explicitly,
e.g. ψα (x). The position of the spinor index (subscript or superscript) will be meaningful,
as it is for Lorentz vector indices. We will see below how to raise or lower spinor indices.
Remember from eq. (3.68) that a (Dirac) spinor Ψ transforms under a Lorentz trans-
formation Λµν according to

Ψ(x) → Ψ0 (x) = S(Λ) Ψ(Λ−1 x) , (7.3)

where S(Λ) is the 4 × 4 block-diagonal matrix given in eq. (3.74) (for infinitesimal tran-
formations). We write
 
s(Λ)
S(Λ) = (7.4)
(s̄T )−1 (Λ)

(The transposition and the inverse in the second row are conventions that will turn out
to be useful later in this chapter. For now, this is just how we name the lower 2 × 2 block
of S(Λ).) The upper 2 × 2 block, which is relevant for the transformation of left-handed
spinors, has the form
i
s(Λ)αβ = δα β − (σµν )αβ ω µν , (7.5)
4
where in the 2-component world, σ µν is understood to mean
i µ ν
β
(σ µν )αβ = σ σ̄ − σ ν σ̄ µ α . (7.6)
2

116
 7.1 Spinor Indices

The tensor ω is, as before, defined via Λµν = g µν + ω µν .

Thus, in 2-component notation, the transformation property of a left-handed spinor is

ψα (x) → ψα0 (x) = s(Λ)αβ ψβ (Λ−1 x) , (7.7)

As we hinted at above, the position of the spinor indices used here is not arbitrary. The
convention is that an upper index is always contracted with a lower index and vice-versa.

7.1.2 Raising and lowering spinor indices

There is also a method for raising and lowering spinor indices, in analogy to the raising
and lowering of Lorentz vector indices using the metric g µν . To develop this method, let
us consider the antisymmetric tensor in two dimensions, αβ . Its diagonal components
vanish, while

12 = −21 = −1 . (7.8)

(The normalization is a mere convention.) The salient feature of αβ is its invariance
under Lorentz transformations. To see this, note first that det s(Λ) = 1, as befits a
symmetry transformation of a quantum field (after all the “O” in SO+ (1, 3) stands for
“orthogonal”). According to the definition of the determinant, this means in component
notation

1 = s(Λ)1γ s(Λ)2δ (−γδ ) , (7.9)

or, more generally,

αβ = s(Λ)αγ s(Λ)β δ γδ . (7.10)

This proves the invariance of αβ under Lorentz transformations, making it the equivalent
of gµν for spinor indices. We define αβ (with upper indices) as the inverse of αβ (with
lower indices). By defining

12 = 21 = 1 , 21 = 12 = −1 , (7.11)

we see that

αβ βγ = δα γ . (7.12)

Now, we can also define a spinor with an upper index:

ψ α ≡ αβ ψβ . (7.13)

Its transformation property is

ψ α (x) → ψ 0α (x) = αβ s(Λ)β γ ψγ (Λ−1 x) (7.14)

117
Chapter 7 Weyl and Majorana Fermions

This can be modified further by using the following relation, which follows from eq. (7.10):
η
θβ s(Λ)β δ αδ ακ = θβ s−1 (Λ)α ηβ ακ
θ
⇔ θβ s(Λ)β κ = s−1 (Λ)α ακ
α
⇔ αβ s(Λ)β γ = s−1 (Λ)β βγ . (7.15)

This relation leads to

α
ψ 0α (x) = ψ β (x) s−1 (Λ)β (7.16)

Lowering indices works the same way as raising them:

ψα = αβ ψ β = αβ βγ ψγ = δα γ ψγ . (7.17)

However, the antisymmetry of αβ introduces complications in the form of extra minus
signs sometimes:

ψ α = αβ ψβ = −ψβ βα . (7.18)

Also, when contracting several spinors, care must be taken:

ψ α χα = αβ ψβ χα = −ψβ βα χα = −ψβ χβ = χβ ψβ . (7.19)

The last equality follows from the anticommutation property of spinor fields. In the
following we will adopt the following convention for contractions of two spinors written
without spinor indices:

ψχ ≡ ψ α χα . (7.20)

Then, eq. (7.19) shows for instance that

ψχ = χψ . (7.21)

7.1.3 Right-handed spinors

Let us now consider a right-handed spinor ψ̄, i.e. a two-component object transforming
in (0, 12 ) of SO+ (1, 3). (For now, the bar in ψ̄ merely helps us to distinguish left-handed
from right-handed spinors. It has nothing to do yet with hermitian conjugation or so.
We will, however, make this connection shortly.) To distinguish left-handed from right-
handed spinors, we introduce the convention that spinor indices transforming in the
right-handed representation (0, 21 ) are dotted, e.g. ψ̄ α̇ , while those transforming in the
left-handed representation ( 21 , 0) are undotted as before, e.g. ψα . The transformation
property of ψ̄ α̇ is
α̇ α̇
ψ̄ α̇ (x) → ψ̄ 0α̇ (x) = (s̄T )−1 (Λ) β̇ ψ̄ β̇ (Λ−1 x) = ψ̄ β̇ (Λ−1 x) s̄−1 (Λ)β̇ , (7.22)

118
 7.1 Spinor Indices

where
α̇ i
s̄−1 (Λ)β̇ = δβ̇ α̇ − (σ̄µν )β̇ α̇ ω µν , (7.23)
4
with
i µ ν
α̇
(σ̄ µν )β̇ α̇ = σ̄ σ − σ̄ ν σ µ β̇ . (7.24)
2
Also in the (0, 21 ) representation, we can define an invariant tensor α̇β̇ and its inverse α̇β̇
by exactly the same arguments as for the ( 12 , 0) representation. The relations defining
these tensors are identical to eq. (7.11). Lowering and raising dotted indices works
in exactly the same way as for undotted indices, i.e. there are relations identical to
eqs. (7.12), (7.13) and (7.17) to (7.19), but with all undotted indices replaced by dotted
ones. The conventions for contractions without explicitly written indices is, however,
different: for right-handed spinors, we define

ψ̄ χ̄ ≡ ψ̄α̇ χ̄α̇ . (7.25)

7.1.4 Conjugate spinors

The next question to answer is what happens when we take the hermitian conjugate
of a spinor field. We know already from section 3.4.1 that, if ψ is a left-handed spinor
field, it describes left-handed particles and right-handed antiparticles. Therefore, ψ †
must describe right-handed particles and left-handed antiparticles, and should therefore
transform under (0, 12 ) of SO+ (1, 3). (An alternative way of seeing this is to note that
the transformation of a RH spinor, eq. (7.23), is just the Hermitian conjugate of the
transformation of a LH spinor, eq. (7.5).) Thus, ψ † should come with a dotted index:
ψα̇† . Note that this notation means (ψ † )α̇ , not (ψα̇ )† ! The fact that, for a left-handed
field ψ, its hermitian conjugate ψ † is always a right-handed field, justifies a posteriori
the notation ψ̄ that we have used above. When working with Weyl (2-component)
spinors, one typically writes all right-handed spinors as hermitian conjugates of left-
handed spinors.

7.1.5 Lorentz invariance of the Pauli matrices

There is one more object invariant under Lorentz transformations: the Pauli matrices,
provided we write them with one undotted and one dotted index: σαµα̇ . This follows
directly from eq. (3.72), which read

S −1 (Λ)γ µ S(Λ) = Λµν γ ν . (7.26)

Remembering that the γ matrices are composed of Pauli matrices, we can conclude
immediately that, in two-component notation

s−1 (Λ) σ µ (s̄T )−1 (Λ) = Λµν σ ν

119
Chapter 7 Weyl and Majorana Fermions

σ µ = Λµν s(Λ) σ ν s̄T (Λ) (7.27)

Since the first index on σ ν transforms with s(Λ), it must be an undotted index. Since the
second one transforms with s̄(Λ), it must be a dotted index. Putting in spinor indices,
eq. (7.27) can be rewritten as

(σ µ )αα̇ = Λµν s(Λ)αβ s̄(Λ)α̇β̇ (σ µ )β β̇ . (7.28)

Here, the index structure of s(Λ) can be read off from eq. (7.5). That of s̄T (Λ) must be
the same as that of (s̄T )−1 (Λ) from eqs. (7.22) and (7.23).
In the same way, we can also show that σ̄ µα̇α is invariant under Lorentz transforma-
tions. Alternatively, we can also show this by using that

σ̄ µα̇α = αβ α̇β̇ σβµβ̇ . (7.29)

7.1.6 One more example: the vector current

To get some practice with the formalism of dotted and undotted spinor indices, consider
a vector current:

ψ † σ̄ µ χ = ψα̇† σ̄ µα̇β χβ . (7.30)

Here, according to our conventions χ and ψ are left-handed and thus have undotted
indices, while ψ † ≡ ψ̄ is right handed and has a dotted index.
Since the spinor indices in eq. (7.30) are all contracted, the vector current transforms
like a Lorentz vector:

ψ † (x) σ̄ µ χ(x) → Λµν ψ † (Λ−1 x) σ̄ ν χ(Λ−1 x) . (7.31)

Let us compute the hermitian conjugate of this vector current. This is something we
need to do for instance when computing th squared matrix element |M|2 for a scattering
process. We find

[ψ † σ̄ µ χ]† = [ψα̇† σ̄ µα̇β χβ ]†

= χ†β̇ (σ̄ µ∗ )β̇α ψα

= χ†β̇ σ̄ µβ̇α ψα
= χ† σ̄ µ ψ . (7.32)

The third equality follows from the hermiticity of σ̄ µ . Remember that, for 4-component
spinors we had [ψ̄γ µ χ]† = χ̄γ µ ψ. We have now derived the 2-component version of this
identity.
Another useful relation is

ψ † σ̄ µ χ = ψα̇† σ̄ µα̇β χβ

120
 7.2 The QED Lagrangian in 2-Component Notation

= ψ †α̇ α̇β̇ σ̄ µβ̇γ γδ χδ

= −χδ σδµα̇ ψ †α̇
= −χσ µ ψ † , (7.33)

where in the third step we have used eq. (7.29) as well as the anticommutation property
of fermion fields.

7.2 The QED Lagrangian in 2-Component Notation

As a preparation for the discussion of the Standard Model of particle physics and its
extensions in the second part of this course, let us reqrite the QED Lagrangian using
2-component spinor fields. We write the 4-component electron field Ψ as
 
ψα
Ψ= . (7.34)
χ̄α̇

The QED Lagrangian is then

1
LQED = − Fµν F µν + Ψ̄(i∂/ − m)Ψ − eΨ̄γ µ Ψ Aµ
4
1
= − Fµν F µν + iψ̄σ̄ µ ∂µ ψ + iχσ µ ∂µ χ̄ − mψ̄ χ̄ − mχψ − eψ̄σ̄ µ ψ − eχσ µ χ̄
4
1
= − Fµν F µν + iψ̄σ̄ µ ∂µ ψ + iχ̄σ̄ µ ∂µ χ − mψ̄ χ̄ − mχψ − eψ̄σ̄ µ ψ + eχ̄σ̄ µ χ .
4
(7.35)

In the last line, we have used eq. (7.33) followed by an integration by parts on the kinetic
term of χ. We have also applied eq. (7.33) to the gauge interaction term of χ, where it
introduces a minus sign compared to the corresponding term for ψ. This makes sense:
remember that χ̄ is the right-handed component of Ψ, i.e. it has the same charge as ψ.
Consequently, χ must have opposite charge. As expected, we see that, apart from the
mass terms, ψ and χ are decoupled.
Note that eq. (7.35) is invariant under the transformation

ψα (x) → C −1 ψα (x)C ≡ χα (x) ,

χα (x) → C −1 χα (x)C ≡ ψα (x) , (7.36)
e → −e ,

which corresponds to charge conjugation. To see that it is equivalent to the C trans-

formation from section 3.4.4, note that eq. (7.36) is implemented on the 4-component
spinor Ψ = (ψα , χ̄†α̇ )T by the operation

Ψ(x) → Ψc (x) ≡ C Ψ̄T (x) (7.37)

121
Chapter 7 Weyl and Majorana Fermions

with the charge conjugation matrix

C ≡ −iγ 2 γ 0 . (7.38)

(This is identical to eq. (3.149).) In fact, when written in components, we have

σ2
  
1
C = −i (7.39)
σ̄ 2 1
−iσ 2
 
= (7.40)
−iσ̄ 2
 
0 −1
1 0 
= , (7.41)
 0 1
−1 0

and therefore

C Ψ̄T (x) = Cγ 0 Ψ∗ (x) (7.42)

!
−iσ 2 ψβ̇†
 
= (7.43)
−iσ̄ 2 χβ
!
ψβ̇†
 
αβ
= α̇β̇ (7.44)
 χβ
 
χα
= . (7.45)
ψ †α̇

One conclusion we can draw from this is that charge conjugation transforms a left-handed
4-spinor (ψα , 0)T into a right-handed 4-spinor (0, ψ †α̇ ).
Note that the matrix C has the following useful properties:

C T = C † = C −1 = −C and C∗ = C . (7.46)

This can be used for instance to show that applying charge conjugation twice is the
identity:

(Ψc )c = Cγ 0 (Ψc )∗ = Cγ 0 (Cγ 0 Ψ∗ )∗ = −C 2 (γ 0 )2 Ψ = Ψ . (7.47)

Other relations that are sometimes useful are

Ψc = (Cγ 0 Ψ∗ )† γ 0 = (γ 0 Cγ 0 Ψ)T = (−CΨ)T = CΨT (7.48)

and

Ψ̄Θc = Ψ† γ 0 Cγ 0 Θ∗

122
 7.3 Majorana Fermions

= −Θ† (γ 0 )T C T (γ 0 )T Ψ∗
= Θ̄Ψc . (7.49)

In the second equality, we have used the fact that all spinor indices are contracted, i.e.
the expression is just a number and we can without loss of generality replace it with its
transpose. We gave to introduce a minus sign, however, since the fermion fields Θ and Ψ
change places. (Write out the indices to see this!) In analogy to eq. (7.49), we also have

Ψc Θ = χc Ψ . (7.50)

7.3 Majorana Fermions

We could also have rewritten the Lagrangian eq. (7.35) by substituting

ψ= √1 (ξ1 + iξ2 ) , (7.51)

2
1
χ= √ (ξ1
2
− iξ2 ) . (7.52)

Plugging this into eq. (7.35) and using eqs. (7.21) and (7.32) leads to
1 1
LQED = − Fµν F µν + iξ¯1 σ̄ µ ∂µ ξ1 + iξ¯2 σ̄ µ ∂µ ξ2 − mξ1 ξ1 + mξ2 ξ2 + h.c.


4 2
− eξ¯1 σ̄ µ ξ2 − eξ¯2 σ̄ µ ξ1 . (7.53)

The kinetic and mass terms now describe two completely independent 2-component
spinor fields. The gauge interaction term is instead off-diagonal.
For uncharged particles (i.e. particles without electromagnetic gauge interaction terms,
for instance neutrinos), the decoupling of ξ1 and ξ2 is complete. Removing one of them
from the theory should therefore still lead to a valid model, with a Lagrangian of the
form

¯ µ ∂µ ξ − 1 mξξ − 1 mξ † ξ † .
LMajorana = iξσ̄ (7.54)
2 2
A field of this type is called a Majorana fermion. It is a massive fermion with only
two degrees of freedom. This is possible because the right-handed state of a Majorana
fermion is identical to the antiparticle of the left-handed state. To see this, define
 
ξ
Ψ(x) ≡ †αα̇ , (7.55)
ξ

which implies that Ψc = Ψ. We can also rewrite eq. (7.54) in 4-component notation:
1 1
LMajorana = Ψ̄γ µ i∂µ Ψ − mΨ̄Ψ . (7.56)
2 2

123
Chapter 7 Weyl and Majorana Fermions

7.4 Application: Majorana Neutrinos and the Seesaw

Mechanism
As alluded to before, neutrinos have all the properties required to be Majorana fermions—
in particular they are electrically neutral. In fact, we do not know whether they are Dirac
or Majorana particles. The interactions of the Standard Model (SM) couple only to left-
handed neutrinos (and their antiparticles, the right-handed anti-neutrinos). The only
interactions that couple right-handed neutrinos to left-handed neutrinos are the mass
terms, which are tiny for neutrinos. Thus, if right-handed neutrinos exist as distinct
particles (i.e. if neutrinos are Dirac), the only way to produce or detect them is through
interactions that are suppressed by the tiny neutrino masses. With present experiments,
it is impossible to detect such tiny interactions. Our best hope is the search for neutri-
noless double beta decay, a process in which two (virtual) neutrinos annihilate. This is
only possible if neutrinos are their own antiparticles, i.e. for Majorana neutrinos.

7.4.1 Neutrino mass terms

Let us now elaborate further on the physics of Majorana neutrinos. For phenomenological
purposes, we could simply write down a mass term
1
L ⊃ − mij νLi νLj + h.c. , (7.57)
2
where νLj is the 2-compoment neutrino spinor, and j is a flavor index. (There are three
species, or “flavors” of neutrinos: νe , νµ , ντ , just like there are three flavors of charged
leptons: the electron, the muon and the tau lepton.) Note that we have kept an index L
in eq. (7.57) to remind ourselves that this is a left-handed spinor. In the context of the
Standard Model, this index also implies that νLj couples to the weak interaction, which
affects only left-handed particles. The interaction vertices are
e ν

W Z

ν ν
where W and Z are the gauge bosons mediating weak interactions, and `j are the charged
lepton fields. But let’s come back to the mass term in eq. (7.57). One may ask the
question whether the “charge” of the neutrinos under the weak interaction does not
forbid this mass term, just like the electric charge forbids Majorana mass terms for the
other SM fermions. The truth is, eq. (7.57) is forbidden in the SM. In fact, all fermion
mass terms (Dirac and Majorana) are forbidden. Then, why are fermions obviously
massive? The solution is the Higgs mechanism, which we will discuss in detail in the
second part of this course, the essentials of which we can however already understand
now. The SM symmetries allow for the following Yukawa couplings:
L ⊃ y ij Li H̃N j + · · · + h.c. , (7.58)

124
 7.4 Application: Majorana Neutrinos and the Seesaw Mechanism

where Li = (νLi , `iL )T consists of the left-handed neutrino and charged lepton fields of
the i-th generating, arranged here in a 2-vector, and H̃ is the Higgs field, which is also
a 2-vector. (The reason for arranging these particles in 2-vectors is that they form a
2-dimensional representation of SU (2), one of the symmetry groups of the SM.) The
field N in eq. (7.58) describes the right-handed neutrinos, but as before, we write a right
handed field as the hermitian conjugate of a left-handed one, i.e. N are the left-handed
antiparticles of the RH neutrinos, and N̄ are the RH neutrinos themselves. Note that
in the original form of the SM, right-handed neutrinos are not included. Therefore,
the original SM predicted neutrinos to be exactly massless. With the discovery of neu-
trino oscillations, which prove that neutrinos have nonzero masses, the model had to be
amended. The dots in eq. (7.58) represent similar couplings for the charged leptons and
the quarks. The crucial point is now that hΩ|H̃|Ωi = (v, 0)T 6= 0, i.e. the Higgs field has
a non-vanishing vacuum expectation value. In other words, even when no Higgs particles
are present in a given process, the field is non-zero. Expanding H̃ around this vacuum
expectation value turns the Yukawa couplings in eq. (7.58) into

L ⊃ y ij v νLi N j + · · · + h.c. , (7.59)

|{z}
≡mij
D

a Dirac mass term for neutrinos. (The notation mij

D indicates that this parameter de-
scribes a Dirac mass.) The dots again indicate many terms we have omitted because
they are not relevant to our present discussion.

7.4.2 The seesaw mechanism

But the story does not end here: the right-handed singlet neutrinos N̄ j are uncharged
under all SM symmetries, i.e. they can have a Majorana mass term
1
L ⊃ − M ij N i N j + h.c. . (7.60)
2
The total Lagrangian describing neutrino masses (composed of eq. (7.59) and eq. (7.60))
can be written in matrix form:
  
1 0 mD νL
L ⊃ (νL , N ) + h.c. . (7.61)
2 mTD M N

Here, it is implied that νL = (νLe , νLµ , νLτ )T contains the three left-handed SM neutrino
flavors, and N = (N 1 · · · N n ) contains the singlet neutrinos. mD is thus a 3 × n matrix
and M is an n × n matrix.
We see that the neutrino mass matrix is off-diagonal, i.e. there is mixing between the
νL and N fields. For instance, an electron neutrino does not have a definite mass, but
is a superposition of several mass eigenstates. Similarly, a neutrino of definite mass is a
mixture of different flavor eigenstates. To transform the mass matrix in eq. (7.61) to the
mass basis, we diagonalize it by applying a unitary transformation to the neutrino state.

125
Chapter 7 Weyl and Majorana Fermions

To obtain the form of this transformation, let us consider for simplicity a simplified model
with only one νL flavor and one N flavor. The desired transformation can be written as
    
νmL cos θ sin θ νL
= . (7.62)
Nm − sin θ cos θ N

Here, the subscript m denotes mass eigenstates. By computing

   
cos θ sin θ 0 mD cos θ − sin θ
(7.63)
− sin θ cos θ mD M sin θ cos θ

and requiring that the off-diagonal elements vanish, we find for the mixing angle
2mD
tan 2θ = − . (7.64)
M
The eigenvalues of the mass matrix are
1
q
m1,2 = M ∓ M 2 + 4m2D ,


(7.65)
2
and the corresponding mass terms read
1 1
L2-flavor ⊃ − m1 νmL νmL − m2 Nm Nm + h.c. . (7.66)
2 2
There are theoretical reasons to believe that M  mD . Namely, while mD is determined
by the vacuum expectation value (vev) of the Higgs field v ∼ 174 GeV according to
eq. (7.59), the scale of M is completely free. (Note that y ij cannot be much larger than
O(1) in a consistent QFT, and values much smaller than O(1) are considered unnatural
by many theorists.) If M is determined by some underlying, more complete, theory, the
typical mass scales of that theory should be  v, otherwise we would have discovered
it already. If M  mD , we see from eq. (7.65) that one neutrino mass eigenstate is of
order M , while the other is

m2D
m1 ' . (7.67)
M
This provides a natural explanation of the smallness of neutrino masses. For instance, for
mD ∼ 100 GeV, M ∼ 1013 GeV, we obtain neutrino masses of order 0.1 eV, consistent
with experimental constraints.
The above mechanism is called the type-I seesaw mechanism because the fact that
the N have a are very large mass term reduces the effective mass of the light neutrinos,
which are mostly νL , with a small admixture of N .
Lifting the restrictions of the 2-flavor toy model and treating mD and M as matrices
again, the seesaw formua eq. (7.67) generalizes to

mν ' −mD M −1 mTD . (7.68)

126
 7.4 Application: Majorana Neutrinos and the Seesaw Mechanism

Note that the matrices mD and M can be off-diagonal. mD can in fact be an arbitrary
complex matrix, while M has to be complex symmetric. This follows from the form of
the N mass term, M ij N i N j + h.c. (see eq. (7.60)), which shows that any antisymmetric
contribution would cancel and thus be unphysical. The off-diagonal nature of the mass
matrices implies that also the SM neutrinos can mix among each other. This means, for
instance, that the electron neutrino does not have a definite mass, but is a mixture of
the three light mass eigenstate (plus a tiny, usually negligible admixture of order mD /M
of the heavy mass eigenstates). Conversely, the light neutrino mass eigenenstates are
mixtures of νe , νµ and ντ (and a tiny admixture of the heavy flavor eigenstates). This
mixing is at the core of the so-called neutrino oscillations.

7.4.3 Interlude: measuring neutrino masses

Direct neutrino mass measurements
Let us pause for a moment to discuss what we know experimentally about neutrino
masses. Obviously, decay processes involving neutrinos depend on their masses. The
most direct way of measuring neutrino masses is by studying the kinematics of β decay
processes like
3
H → 3 He + e− + ν̄e . (7.69)

The maximum energy of the electron is

Eemax = Q − min(mj ) , (7.70)

j

where Q = mH − mHe is the decay energy and mj are the neutrino mass eigenstates.
The spectrum of electron energies from β decay is given by

dNe (Ee ) p q
∝ F (Z, Ee ) Ee2 − m2e Ee (Q − Ee ) (Q − Ee )2 − m2ν̄e θ(Q − Ee − mν̄e ) .
dEe
(7.71)

Here, the Fermi function F (Z, Ee ) accounts for the interaction of the produced electron
with the Coulomb field of the nucleus. By θ(x) we denote the Heaviside function.
To make the dependence on the neutrino mass more visible, one defines the Kurie
functions
s
dNe (Ee )/dEe
K(Ee ) ≡ . (7.72)
F (Z, Ee )pe Ee

For mν̄e = 0, it should be a straight line ∝ Q − Ee , while for mν̄e 6= 0, it has a cutoff,
see fig. 7.1.

127
Chapter 7 Weyl and Majorana Fermions

�ν = �
�ν = �� ��
������ = ��� ��� �� ��

����� �������� [����]

����� ����� ����� ����� �����

�� [���]

Figure 7.1: The Kurie plot.

It is important that ν̄e does not have a definite mass, i.e. when we write mν̄e above, this
is actually ill-defined. We should therefore regard the process (7.69) as the combination
of the processes
3
H → 3 He + e− + ν̄1 , (7.73)
3 3 −
H → He + e + ν̄2 , (7.74)
3 3 −
H → He + e + ν̄3 .. (7.75)
Since we can only detect the electron, we do not know on an event-by-event basis which
of these processes has occured, hence we always have to take all three of them into
account. Thus, in particular, the maximum energy of the electron Eemax is given by
the maximum of the kinematic endpoints of the three processes. For neutrino mass
mj  maxk |mj − mk |, this subtlety is negligible.
For smaller neutrino masses, the Kurie function is
p
K(Ee ) = |Ue1 |2 [K1 (Ee )]2 + |Ue2 |2 [K2 (Ee )]2 + |Ue3 |2 [K3 (Ee )]2 , (7.76)
where Uej are elements of the leptonic mixing matrix that determine the admixture of
the mass eigenstate νj to the flavor eigenstate νe , and
r q
Kj (Ee ) ≡ C (Q − Ee ) (Q − Ee )2 − m2j θ(Q − Ee − mj ) . (7.77)

Here, C is a j-independent normalization constant. For Ee not too close to the endpoint,
we can expand
q m2j
(Q − Ee )2 − m2j ' Q − Ee − . (7.78)
2(Q − Ee )

128
 7.4 Application: Majorana Neutrinos and the Seesaw Mechanism

Plugging this into eq. (7.77) and (7.76), we get

v
m2j
u  X 
2|
u
K(Ee ) ' C (Q − Ee ) Q − Ee −
t |Uej θ(Q − Ee − mj ) . (7.79)
2(Q − Ee )
j

At energies below Q − maxj (mj ) (where all the θ functions are 1), this can be written as
r q
K(Ee ) ' C (Q − Ee ) (Q − Ee )2 − m2ν̄e , (7.80)

with the definition of the effective mass

sX
mν̄e ≡ 2 |m2 .
|Uej (7.81)
j
j

Very close to the endpoint, this description breaks down. Instead, several kinks are
expected in the spectrum.

Neutrinoless double beta decay

A second way of obtaining information on the masses of Majorana neutrinos is neutri-
noless double beta decay. To introduce this process, consider the level scheme in fig. 7.2,
taken from a table of isotopes [3]. Note that the isotopes in general follow two parabolas:
a higher energy one for the odd–odd nuclei (i.e. nuclei with an odd number of protons
and an odd number of neutrons), and a lower energy one for the even–even nuclei. This
leads to the situation that the energy of 136 Xe is lower than that of the isotope directly to
its right, 136 Cs, so that the direct β − decay of 136 Xe to 136 Cs is energetically forbidden.
On the other hand, two simultaneous β − decays are allowed:
136
Xe → 136 Ba + 2e− + 2ν̄e . (7.82)

The Feynman diagram for such a process of the form

(A, Z) → (A, Z − 2) + 2e− + 2ν̄e (7.83)

is
d u

W e−
ν̄e
ν̄e
W
e−
d u

129
Chapter 7 Weyl and Majorana Fermions

Sn (10100)

40000 A=136
NDS 52, 273(1987)
36000
NDS 71, 1(1994)(U) Sp (600)
Sn (11800) (3+) 3.7 s
EC
32000
(7+) 3.3 s
%p=0.09 (3.7 s) 136
%p=0.09 (3.3 s) EC 63Eu
p
28000 Sn (3500)
Sp 12010 Sn (9200)
Sp (3900) QEC(10400)

24000 0.82 s
136 0+ 47 s
51Sb β− %n=24.0
136
EC 62Sm
n+2n Sn (11070)
20000 Sp (2400)
Sn 4670
Qβ−(9300) Sp 8960 5(+),6– 107 s
EC
(2+) 47 s QEC(4500)
136
16000
Sn 8530 EC 61Pm
Sp 5540
0+ 17.5 s n
136
52Te β−
Sn 3780 QEC7850
1.1%
Sn 9940
12000 Sp 9926 Sp 4030
Qβ−5070
Sp 7197
(6–)
46.9 s –
640 β− Sn 7440
10000 (1 ) 0 Sn 8060 Sn 6828.1 Sn 9107.74 Sp 7130 0+ 50.65 m
83.4 s Sp 8594.0 136
136 Sp 5460 EC 60Nd
8000
53I β−
2+ 13.1 m
136
EC 59Pr
QEC2211
Qβ−6930
6000

5000 QEC5126

4000
IT

230+x
8– β− 114 ms
IT

0+x 1+ 0
3000 19 s 9.87 m
0+ 5+ 0
13.16 d 136 0+
EC 57La
>2.36×1021 y
136 136
55Cs β−
136
2000 54Xe 58Ce
IT

Qβ−β−2467 Qβ−2548.2 7– 2030.52

1000 QEC81 0.3084 s
QEC2870
500 Qβ−470

100

0
0+ 0 Evaluator: J.K. Tuli
136
56Ba

4632

Figure 7.2: Level scheme for A = 136 nuclei. Note that 136 Xe cannot undergo direct
β − decay to 136 Cs, but 0ν2β decay to 136 Ba is allowed. Similarly, 136 Ce cannot decay
via β + decay to 136 La, but 0ν2β + decay to 136 Ba is allowed.

130
 7.5 Twistors

Remember that replacing an outgoing antiparticle in a Feynman diagram by an ingoing

particle yields a valid Feynman diagram as well (“crossing symmetry”). Therefore, if the
neutrino is its own antiparticle (Majorana neutrinos), the two neutrino lines in the above
diagram can be connected. In other words, the neutrino emitted in the decay of the first
down quark can be absorbed by the second down quark. This leads to neutrinoless double
beta decay

(A, Z) → (A, Z − 2) + 2e− , (7.84)

the Feynman diagram for which is

d u

W
e−
M
e−
W

d u

This diagram is sometimes called “lobster diagram”. Since my artistic skills are not
sufficient to explain why, I have to appeal to the reader’s imagination . . .
The crucial point is that the rate of neutrinoless doble beta decay is proportional to
the neutrino masses. To see this, we must know that the W boson couples only to left-
handed particles and right-handed antiparticles. For instance, the upper vertex in te
lobster diagram can be viewed as emitting a LH electron and a RH antineutrino. The
lower vertex absorbs a LH neutrino and emits a LH electrons. This is only possible if the
neutrino experiences a chirality flip along the vertical propagator, i.e. if there is a mass
insertion (denoted by a little cross in the Feynman diagram).
While in two-neutrino double beta decay, part of the decay energy is carried away by
the neutrinos, in neutrinoless double beta decay it is all carried by the electrons and thus
visible to a detector. The telltale signature of neutrinoless double beta decay is thus a
monoergetic peak at the Q value of the decay.

7.5 Twistors
7.5.1 Unifying spinors and momentum 4-vectors
In high energy physics, one often deals with particles that are approximately massless.
For instance, at LHC energies of multiple TeV, the up, down and strange quark masses
of order MeV are completely negligible. We now introduce a formalism that allows
scattering amplitudes involving massless particles to be written in a very
P compact form.
s r r
Consider a massless spinor u (p), with s = ±. Using the spin sum r u (p)ū (p) = p /

131
Chapter 7 Weyl and Majorana Fermions

(see eq. (3.25)), we can write its outer product with itself as

1 + sγ 5  X r  1 − sγ 5 1 + sγ 5
us (p)ūs (p) = u (p)ūr (p) = p
/, (7.85)
2 r
2 2

where the projection operators (1 ± sγ 5 )/2 make sure that only the term with r = s
contributes in the spin sum. Similarly, for v spinors:

1 − sγ 5  X r  1 + sγ 5 1 − sγ 5
v s (p)v̄ s (p) = v (p)v̄ r (p) = p
/, (7.86)
2 r
2 2

Note that for massless spinors we have v s (p) = u−s (p). This follows from the explicit
expressions for the u and v spinors when taking the limit m → 0, see eqs. (3.107)
and (3.108). Therefore, in the following, it is sufficient to focus on u spinors only.
According to the Feynman rules, scattering amplitudes for massless particles depend
on spinors and on 4-momenta. We will now write the momenta in terms of spinor-like
objects as well. This will lead to some simplifications. We define

pαα̇ ≡ pµ σαµα̇ . (7.87)

We have already mentioned in eq. (7.29) that σ̄ µα̇α = αβ α̇β̇ σβµβ̇ . Therefore, we also have

pα̇α = αβ α̇β̇ pβ β̇ = pµ σ̄ µα̇α . (7.88)

Remembering that

0 σµ
   
µ 1 1 0
γ = , (1 − γ 5 ) = , (7.89)
σ̄ µ 0 2 0 0

we can write
 
− − 0 pαα̇
u (p) ū (p) = . (7.90)
0 0

To avoid all the zeros in this matrix, define

 
φα (p)
−
u (p) ≡ , and thus ū− (p) ≡ (φ†α̇ (p), 0) . (7.91)
0

Then, we have simply

pαα̇ = φα (p) φ†α̇ (p) . (7.92)

The 2-component object φα (p) is called a “twistor”. It carries a spinor index and trans-
forms accordingly under Lorentz transformation. Since twisters are not fields, but just
a numerical quantity, they are not anticommuting, though.

132
 7.5 Twistors

We can repeat the above arguments also for positive helicity spinors
 
+ 0
u (p) ≡ , and ū+ (p) ≡ (φαT (p), 0) , (7.93)
φ∗α̇ (p)

with φα (p) = αβ φβ (p). Note that the φ(p) appearing in u+ (p) is the same as the one
appearing in u− (p) (up to the raising and lowering of indices). This follows again from
the explicit expressions eq. (3.107). We can also write p in terms of u+ (p):
 
+ + 0 0
u (p) ū (p) = . (7.94)
pα̇α 0
or

pα̇α = φ†α̇ (p) φα (p) . (7.95)

We have thus seen that both the particles spinor and its momentum can be expressed
in terms of the twistor. It is therefore possible to express a Feynman amplitude for
massless particles entirely in terms of twistors.

7.5.2 Twistor notation

At this point, we again introduce some notation:

|p] ≡ u− (p) = v + (p) (7.96)

+ −
|pi ≡ u (p) = v (p) (7.97)
+ −
[p| ≡ ū (p) = v̄ (p) (7.98)
− +
hp| ≡ ū (p) = v̄ (p) (7.99)

Given two 4-momentum vectors p and k, and the corresponding twistors φα (p) and κα (k),
we thus have
[p k] ≡ φα (p) κα (k)
(7.100)
hp ki ≡ φ∗α̇ κ∗α̇ .

Since φα κα = αβ φβ κα = −βα κα φβ = κα φα , we find the identity

[p k] = −[k p] . (7.101)

Similarly,

hk pi = − hp ki (7.102)

Note that these identities appear to be at odds with eq. (7.21), which stated that ψχ =
χψ. Recall, however, that eq. (7.21) holds for anticommuting spinor fields, while we

133
Chapter 7 Weyl and Majorana Fermions

are now dealing with commuting twistors. Hence the extra minus sign. The definitions
eq. (7.100) imply

hp ki = [k p]∗ . (7.103)

Moreover, eqs. (7.91) and (7.93) show that

ū+ (p) u− (k) = [p k] (7.104)

− +
ū (k) u (p) = hp ki (7.105)
+ + − −
ū (p) u (k) = ū (p) u (k) = 0 . (7.106)

We can also rewrite the dot products of 4-momenta in Minkowski space in terms of the
twistor products hp ki and [p k] by using the identity

hp ki [p k] = (φ∗α̇ κ∗α̇ )(κα φα )

= −φα φ∗α̇ )(κ∗α̇ κα )
= −pαα̇ k α̇α
= −pµ kν σαµα̇ σ̄ ν α̇α
= −2pµ kµ . (7.107)

The last identity follows from tr(σ µ σ̄ ν ) = 2g µν .

When we did QED calculations in 4-component notation, we often encountered gamma
matrices sanwiched between spinors. In twistor notation, we write this as

ū+ (p)γ µ u− (k) = [p|γ µ |ki , (7.108)

+ µ + µ
ū (p)γ u (k) = [p|γ |ki , (7.109)

etc. If the gamma matrix is contracted with a momentum P , we abbreviate this by

ū+ (p)P
/ u+ (k) = [p|P |ki , (7.110)
ū− (p)P
/ u− (k) = hp|P |k] , (7.111)

If P is a light-like momentum (i.e. the on-shell momentum of a massless particles), we

can write P in terms of twistors, too. In particular, according to eqs. (7.92) and (7.95):

Pµ σαµα̇ = |P ]hP |
(7.112)
Pµ σ̄ µα̇α = |P i[P | ,

and thus
[p|P |ki = [p|α̇ Pα̇α |kiα = [pP ] hP ki ,
(7.113)
hp|P |k] = hp|α P αα̇ |k]α̇ = hpP i [P k] .

134
 7.5 Twistors

7.5.3 Examples
As a first example, consider the annihilation of two massless fermions into two scalars in
Yukawa theory:
e− , p1 φ, p3 e− , p1 φ, p3

e+ , p2 φ, p4 e+ , p2 φ, p4

/1 − p
i(p /3 ) /1 − p
i(p /4 ) s
 
2 s0
= g v̄ (p2 ) + u (p1 )
(p1 − p3 )2 (p1 − p4 )2
ip ip
 
0 /3 /4
= −g 2 v̄ s (p2 ) + us (p1 ) . (7.114)
2p1 p3 2p1 p4

/1 us (p1 ) = 0. If, for instance, s = s0 = +,

In the last line, we have used the Dirac eqiation p
we obtain
 
++ 2 i[23] h31i i[24] h41i
A ≡g +
h13i [13] h14i [14]
 
2 [23] [24]
= −ig + . (7.115)
[13] [14]

Note that, for brevity, we have replaced pj → j here. Moreover, we have used eq. (7.107)
to rewrite the contraction of the 4-momenta in the denominators in terms of twistors,
eq. (7.113) to rewrite the momenta contracted with gamma matrices, and eqs. (7.101)
and (7.102) to reorder the momenta inside twistor products.
The asmplitude for s = s0 = − is obtained in a similar way by replacing square brackets
by angle brackets and vice-versa. Using this observation, we can directly read it off from
eq. (7.115):
 
−− 2 h23i h24i
A = −ig + . (7.116)
h13i h14i

If, on th other hand, s = s0 , we would obtain mixed products of left-chiral and right-chiral
spinors like hpk], which vanish.

135
Chapter 7 Weyl and Majorana Fermions

136
 Acknowledgments

I would like to thank the students attending my lectures, in particular Moritz Breitbach,
for their critical reading of these lecture notes, and for their comments which helped to
improve them.

137
Chapter 7 Weyl and Majorana Fermions

138
 Bibliography

[1] M. E. Peskin and D. V. Schroeder, An Introduction to Quantum Field Theory.

Perseus Books, Cambridge, Massachusetts, 1995.

[2] M. Srednicki, Quantum Field Theory. Cambridge University Press, 2007.

[3] R. B. Firestone and V. S. Shirley, Table of isotopes. Wiley-Interscience, 8th ed.,

1998.

139