QFT Lecture Notes

ANDREAS WEILER, TUM
R E L AT I V I T Y , PA R T I C L E S , F I E L D S
Beachten Sie bitte, dass das Skript noch nicht gründlich Korrektur gelesen wurde und
es sicher noch Fehler gibt. Bitte schicken Sie sehr willkommene Vorschläge, Korrekturen,
Kommentare an:
andreas.weiler@tum.de
Thanks to Tobias Duswald, Nepomuk Ritz, Patrick Selle, Ennio Salvioni, Stefan Schuldt,
and Christian Schuster for helping to improve the script!
Der wackre Schwabe forcht’ sich nit,

ging seines Weges Schritt vor Schritt.1 1
Ludwig Uhland oft zitiert von
Albert Einstein.
c 2018 Andreas Weiler, TUM

Copyright
Sommersemester 2017, Version vom December 31, 2018

Contents
1 Foreword 9
1.1 Units and conventions 9
2 Introduction 11
2.1 Why quantum field theory? 11
2.1.1 Necessity of the field viewpoint: causality 12
2.1.2 General features of relativistic quantum field theories 13
2.2 String Theory 13
2.2.1 The discrete chain 13
2.2.2 The continuous string 14
2.2.3 Quantum string theory 16
2.2.4 Review of the simple harmonic oscillator 17
2.2.5 Fock space of the discrete chain 18
3 Relativity 23
3.1 Galileo and Newton 23
3.1.1 Maxwell vs. Newton 24
3.1.2 Derivation of the Lorentz transformations 24
3.2 Consequences of special relativity 25
3.2.1 Relativity of simultaneity 26
3.2.2 Time dilation 27
3.2.3 Lorentz contraction 27
3.2.4 Addition of velocities 28
3.3 Minkowski space 29
3.3.1 Fourvectors 30
3.3.2 Minkowski Metric 31
3.3.3 Discrete transformations 32
3.3.4 Motion of a free particle 32
4 andreas weiler, tum
3.4 Indices upstairs and downstairs 33

3.4.1 A convenient notation and derivatives 33
3.4.2 The appearance of anti-matter 34
3.5 Relativistic mechanics 34
3.5.1 E = mc2 36
3.5.2 Application: weak decay of a pion 36
4 Classical Field Theory 37

4.1 Dynamics of fields 37
4.1.1 Example: the electro-magnetic field 37
4.1.2 The Lagrangian density 37
4.1.3 Example 1: the Klein-Gordon equation 39
4.1.4 Example 2: first order Lagrangians 39
4.1.5 Example 3: the Maxwell equations 40
4.1.6 Locality 40
4.1.7 Lorentz invariance 41
4.2 Symmetries: Noether’s Theorem and conserved currents 42
4.2.1 Conserved currents 43
4.2.2 Example 1: internal symmetries U (1) 43
4.2.3 Example 2: an internal SO(N ) and a bit on Lie groups 44
4.2.4 Example 3: space-time translations 45
4.2.5 Summary of Noether’s theorem 47
4.3 Coulomb’s law and the propagator method 47
4.3.1 Fourier transformations 48
4.3.2 The Coulomb Potential 49
4.3.3 A non-linear example: the graviton Lagrangian 49
4.4 The Hamiltonian Formalism 51
4.4.1 Example: a real scalar field 52
5 Canonical Quantization 53
5.1 Free fields 53
5.1.1 The free scalar field 54
5.1.2 Infinities 55
5.1.3 Normal ordering 57
5.2 Particles 58
5.2.1 Multi-particle states and Bose statistics 59
relativity, particles, fields 5
5.3 Smeared operator valued distributions 60

5.3.1 Dimensional analysis and Coleman theorem 61
5.4 Some technicalities: relativistic normalization 62
5.4.1 Meaning of the state φ(x)|0i 64
5.5 Complex scalar fields 64
5.6 Time dependent operators: the Heisenberg picture 66
5.7 Causal quantum fields 68
5.7.1 Propagators in quantum field theory 69
5.8 The Stückelberg-Feynman propagator 71
5.8.1 Green’s Functions 73
5.8.2 Particle creation by a classical source 74
5.9 Non-relativistic fields 76
5.9.1 A special case: quantum mechanics 78
5.9.2 Non-relativistic interactions 80
6 Interacting quantum fields 81

6.1 Dimensional analysis of interactions 81
6.1.1 Some examples of weakly coupled theories 82
6.2 The interaction picture 84
6.2.1 Dyson’s formula 85
6.3 A first look at scattering 87
6.3.1 Scattering simplified 88
6.3.2 Example: Meson Decay φ → ψ̄ψ 90
6.4 Wick’s Theorem 90
6.4.1 Example: nucleon scattering ψψ → ψψ 92
6.5 Diagrammatic Perturbation theory: Feynman Diagrams 94
6.5.1 Feynman rules 95
6.5.2 Revisiting ψψ → ψψ scattering 98
6.6 More scattering processes in the ψ̄ψφ theory 98
6.6.1 Nucleon anti-nucleon scattering: ψ(p1 ) + ψ̄(p2 ) → ψ(q1 ) + ψ̄(q2 ) 98
6.6.2 Meson-meson scattering φφ → φφ 99
6.6.3 Symmetry factors 99
6.7 Potentials 102
6.7.1 The Yukawa Potential 102
6.8 How to calculate observables 104

6.8.1 Fermi’s Golden Rule 104
6.8.2 Decay Rates 106
6.8.3 Example: two-body phase space 108
6.8.4 Examples: particle decays 108
6.9 Cross Sections 109
6.10 Green’s Functions 110
6.10.1 The true vacuum 110
6.10.2 Connected diagrams and vacuum bubbles 112
6.10.3 Outlook to S-matrices from Green’s functions 114
7 Lorentz Representations 117

7.1 The Lorentz algebra 118
7.1.1 Representations 121
7.1.2 Spinor representations 121
7.1.3 The Dirac equation 124
7.1.4 Lorentz-invariant actions 127
7.1.5 The Dirac Lagrangian 129
7.1.6 The slash 130
7.1.7 Chiral spinors and the Weyl equation 130
7.1.8 γ5 131
7.2 Discrete symmetries 132
7.2.1 Parity 132
7.3 Solutions of the free Dirac equation 133
7.3.1 Negative energy solutions 135
7.3.2 Examples 135
7.3.3 Helicity 136
7.3.4 Useful results: inner and outer products 136
7.3.5 Outer products 137
7.4 Symmetries of the Dirac Lagrangian 138
7.4.1 Internal Symmetries 139
7.5 Quantizing the Dirac Field 140
7.5.1 The Hamiltonian 140
7.5.2 Quantization of fermions 140
7.5.3 Fermi-Dirac Statistics 142
7.5.4 Causal Fermi Fields 142
7.6 Perturbation Theory for Spinors 143

7.7 The Fermion Propagator 144
7.7.1 The Feynman propagator for Dirac fields 145
7.7.2 Nucleon-Nucleon scattering with spin 146
7.8 Feynman rules for Dirac Spinors 148
7.8.1 Example: Nucleon scattering ΨΨ → ΨΨ 149
7.8.2 Example: Nucleon-Anti-Nucleon scattering ΨΨ → ΨΨ 149
7.8.3 Example: Meson Scattering 149
7.8.4 Example: the Yukawa potential 150
8 Quantum Electro Dynamics 153

8.1 The Feynman Rules for Quantum Electrodynamics 153
8.2 Maxwell’s equations 154
8.3 Gauge Symmetry 155
8.3.1 Gauges 156
8.4 Quantizing the Electro-magnetic field 157
8.4.1 Coulomb gauge 157
8.4.2 Lorentz Gauge 159
8.4.3 The propagator 164
8.5 Coupling gauge fields to matter 164
8.5.1 Coupling photons to Dirac fermions 164
8.5.2 Coupling photons to charged scalars 165
8.6 The electric charge 166
8.7 Discrete symmetries of QED 166
8.7.1 Parity 167
8.7.2 Charge conjugation 167
8.7.3 Majorana neutrino 169
8.7.4 Furrys theorem 170
8.8 Time reversal 170
8.8.1 Quantum mechanics 170
8.8.2 Scalar Field 171
8.8.3 Spinor Field 172
8.9 Electron scattering and the Coulomb potential 172
8.9.1 Coulomb potential 173
9 Outlook 175
1
Foreword
These notes are not original but mostly a combination of results

found in books and lecture notes. My contribution is here foremost
in the selection and the perspective provided. I have made liberal
use of the following references The ? indicates the technical level of
the book. It is roughly proportional
• Peskin, Schroeder - An Introduction to Quantum Field Theory ?? to the time needed per page for a
full understanding. Note, that the
Zee book in particular is written in
• Schwartz - Quantum Field Theory and the SM ?? a conversational tone, but covers
more concepts than any of the other
• Mandl, Shaw - Quantum Field Theory ? books.
• Zee - Quantum Field Theory in a Nutshell ?
• Srednicki - Quantum Field Theory ? ? ?
• Weinberg - Quantum Field Theory I ? ? ??
• Ramond - Field Theory: a Modern Primer ? ? ?
• Itzykson, Zuber - Quantum Field Theory ? ? ?
and many more. I strongly suggest that you find a book (or books)
you like from the ones above and study it as a complement to these
lecture notes.
This course provides a hopefully gentle introduction into the
beautiful world of quantum field theory while it familiarizes you
with relativity. It is the first real theory course in that it is the
first course to tackle the theoretical framework that underlies all of
nature. Let’s go.
1.1 Units and conventions
There are three fundamental dimensionful constants in nature: the

speed of light c, Planck’s constant (divided by 2π) ~ , and Newton’s
constant GN . Their dimensions are
[c] = length × time−1 (1.1)

2 −1
[~] = length × mass × time (1.2)
[GN ] = length3 × mass−1 × time−2 (1.3)
The whole point of units is that you can choose whatever units are
most convenient! In particle physics and cosmology, we use “natural
units”
~=c=1 (1.4)
and so
length = time = mass−1 = energy−1 (1.5)
we can express all dimensionful quantities in terms of a single scale

which we choose to be mass or, equivalently, energy.1 I will tem- 1
Since E = mc2 has become E = m.
porarily reintroduce ~ and c whenever convenient, mostly to show
the classical ~ → 0 limit or when we start with relativity. To convert
the unit of energy back to units of length or time, we have to insert
the relevant powers of c and ~.
Energies will be given in units of eV (the electron volt)2 or more 2
The electron-volt is a unit of energy
often GeV = 109 eV or T eV = 1012 eV , since we are often dealing equal to approximately 1.6 × 10−19
joules (J). It is the amount of energy
with high energies. Newton’s constant defines a mass scale gained by the charge of a single
electron moving across an electric
GN = MP−2 , MP = 1.22 × 1019 GeV (1.6) potential difference of one volt.
this is the Planck scale. It corresponds to a length lP ≈ 10−33 cm.

Useful conversion factors are
(1 GeV)−1 (~c) = 0.1973 fm (1.7)

2 −24
(1 GeV)/c = 1.783 × 10 g (1.8)
Table 1.1: Overview over the relevant

Quantity Mass Length fundamental scales of Nature.
Cosmological constant Λ1/4 10−3 eV 10−4 m
Neutrino masses mν ≤ 0.12 eV ≥ 10−9 m

1
Atomic length (bohr a0 = me α ) 3.73 eV 5.29 · 10−11 m
Electron mass me− 0.511 keV 3 · 10−13 m
Muon mass mµ− 105.7 MeV 1.9 · 10−15 m
Proton mass mp 938.272 MeV 2.1 · 10−16 m
Z boson mass mZ 91.1876 GeV 2.16 · 10−18 m
Higgs mass mh 125.09 GeV 1.57 · 10−18 m
Top quark mass mPole

t 172.44 GeV 1.14 · 10−18 m
Planck Mass MP 2.4 · 1018 GeV 8 · 10−35 m

2
Introduction
2.1 Why quantum field theory?

It is the theory of the world. Here are the main arguments:
• All elementary particles look the same. An electron at the begin-

ning of the universe has the exact same properties as an electron
which is part of the atoms of your body. An excellent explanation
for this is, if we assume that there is one field per elementary
particle permeating the whole universe
Ψa = Ψa (x, t) (2.1)
Elementary particles are then local, quantized excitations of this

universal field. Each elementary particle corresponds to exactly
one quantum field.
• Once we try to combine special relativity (SR) with quantum

mechanics (QM), we find that particle number is not conserved
anymore
QM + SR = QF T
We will study Dirac’s argument for anti-particles in detail later.
Let it just suffice to say, that if we try to localize a particle in a
box smaller than This is the so-called Compton
wavelength.
~
λ= (2.2)
mc
then the energy uncertainty is of order of the mass of the particle.
To see this, start with the Heisenberg relation for the momentum
uncertainty of a particle in a box of size L
∆p ≥ ~/L
Relativistically, momentum and energy are on the same footing

∆E ≥ ~c/L.1 But if the uncertainty exceeds ∆E = 2mc2 then 1
E 2 = (pc)2 + (mc2 )2
we can create particle anti-particle pairs out of the vacuum.
The Compton wavelength is always smaller than the de Broglie
wavelength λdB = h/p. If you like, the de Broglie wavelength is
the distance at which the wavelike nature of particles becomes
apparent; the Compton wavelength is the distance at which
the concept of a single point like particle breaks down completely.
2.1.1 Necessity of the field viewpoint: causality

We will show now that causality makes a multi-particle theory
necessary. We start with the amplitude for a particle propagating
from x0 to x during the time t
U (t) = hx|e−iHt |x0 i (2.3)
The energy of a free particle in non-relativistic quantum mechanics is

E = p2 /(2m) and so2 2
Inserting the identity
d3 p
Z
2
U (t) = hx|e−ip̂ /(2m)t |x0 i (2.4) 1=
(2π)3
|pihp|
Z
d3 p 2 and
= hx|e−ip /(2m)t |pihp|x0 i (2.5) hx|pi = eip·x
(2π)3
Z
1 2
= 3
d3 p e−ip /(2m)t hx|pihp|x0 i (2.6)
(2π)
Z
1 2
= 3
d3 p e−ip /(2m)t eip·(x−x0 ) (2.7)
(2π)
m 3/2
im(x − x0 )2
= exp (2.8)
2πit 2t
where in the last step we have used a general gaussian integral3 with 3
A very useful formula:
A = it/m 1 and B = i(x − x0 ) This amplitude is non-zero for all x Z −1
n
P
Aij xi xj +
n
P
Bi xi
2
n i,j=1 i=1
and t, which implies that a particle can propagate between any two d xe
points in an arbitrarily short time. In a relativistic theory, no signal

Z
1 T T
= dn x e− 2 x Ax+B x
can propagate faster than light4 and this would mean that causality r
(2π)n

is violated! We can try to amend the above formula by using the 1 T −1
= exp B A B
det A 2
relativistic Pythagoras for the energy
4
More on that later.
2 2 2 2
E = (pc) + (mc ) (2.9)
and taking the positive root5 5

Why?
√ 2 2
U (t) = hx|e−i p̂ +m t |x0 i
Z √
d3 p −i p2 +m2 t
= hx|e |pihp|x0 i
(2π)3
Z √ 2 2
1
= 3
d3 p e−i p +m t eip·(x−x0 )
(2π)
Z ∞ Z 2π Z π √ 2 2
1
= 3
dpp 2
dφ dθ sin θ e−i p +m t eip|x−x0 | cos θ
(2π) 0 0 0
Z ∞ √ 2 2 Z 1
1
= dpp2 e−i p +m t d cos θeip|x−x0 | cos θ
(2π)2 0 −1
Z ∞ √
1 1 −i p2 +m2 t
= dpp e sin(p|x − x0 |)
2π 2 |x − x0 | 0
where we have chosen the z-axis along the x − x0 direction. We now The main idea of stationary phase
want an approximate solution for this integral for x2 t2 (far out- methods relies on the cancellation
of sinusoids with rapidly varying
side the light-cone). We will use the stationary phase approximation. phase. If many sinusoids have the
The phase function same phase and they are added
together, they will add constructively.
p If, however, these same sinusoids
φ(p) ≈ px − t p2 + m2 (2.10) have phases which change rapidly
as the frequency changes, they will
add incoherently, varying between
constructive and destructive addition
at different times.
https://en.wikipedia.org/wiki/
Stationary_phase_approximation
has a stationary point at

p

φ (p) = 0
0
p0 = imx/ x2 − t 2 (2.11)
p0
Substituting this value back for p, we find that up to a rational

function of x and t, that You can also solve this integral
√ exactly in terms of Bessel-functions
−m x2 −t2
U (t) ∼ eiφ(p0 ) ∼ e (2.12) and convince yourself of the veracity
of this statement.
We find a small, exponentially suppressed amplitude but still non-
zero outside the light-cone, causality is still violated.
We see from the exponential suppression that causality is violated
if we get to distances as small as x ∼ 1/m, for distances x 1/m
there is a negligible chance to find the particle outside the light-
cone6 , so at distances much greater than the Compton wavelength 6
Light-cone: x2 = c2 t2 , with
of a particle, the single-particle theory will not lead to measurable temporarily reintroduced c.
violations of causality. This is in accordance with our earlier ar-

guments based on the uncertainty principle: multi-particle effects
become important when you are working at distance scales of order
the Compton wavelength of a particle.
Quantum field theory finds a miraculous solution to the causality
problem: the propagation across a space-like interval is indistin-
guishable from the propagation of an anti-particle in the opposite
direction. We will find that the amplitudes for particle and anti-
particle propagation exactly cancel outside the light-cone – causality
is preserved!
2.1.2 General features of relativistic quantum field theories

• CPT
• Spin-statistics
• Very constraining for higher spins, only S ≤ 2 theories.
• Massless particles (m = 0): S = 1 gauge theory (like electro-

magnetism), S = 2 gravity For the massless spin-2 field, the
requisite gauge principle can be shown to be general covariance,
which leads to Einstein’s theory, if you want to couple S = 3/2 to
gravity: unique theory super-gravity
2.2 String Theory
2.2.1 The discrete chain

We end up with a quantum field by quantizing a classical field. A
simple example is the classical string. We will define our string
as a long-wavelength limit (= low energy limit) of a harmonically
coupled discrete chain. We take N + 2 masses with a quadratic
nearest-neighbor coupling
N
X +1 h i
m 2 κ
L(q, q̇) = T − V = q̇j − (qj − qj+1 )2 (2.13)
j=0
2 2
The coordinate qj (t) is the longitudinal displacement of the j-th

mass along the one-dimensional chain. The equilibrium distance a
between masses sets the rest location of the masses
... ...
xj ≡ j · a (2.14) qn+1
qn 1 qn
At the endpoints, the chain is fixed: Figure 2.1: One-dimensional discrete
chain.
q0 (t) = qN +1 (t) ≡ 0 (2.15)
The remaining N masses satisfy the equation of motion (EOM)
mq̈j − κ(qj+1 − 2qj + qj−1 ) = 0 (2.16)
for j = 1, . . . , N . The normal modes diagonalize the EOMs and have

the form
qj (t) = cos(ω t) sin(j p) (2.17)
The boundary-conditions in Eq. (2.15) require p to be
πn
pn = (2.18)
N +1
for n = 1, . . . , N . Substituting this into the EOMs, we get the
dispersion relations for the N independent normal frequencies ωn :
r
πn κ
ωn = ω0 sin , ω0 = 2 (2.19)
2(N + 1) m
The frequency ω0 is the cut-off frequency since the modes with
n > N are merely repeating the lower ones. For N = 4, there are 4 independent
modes. Above, n = 5 is trivial and
equivalent to n = 0, since sin(jp5 ) =
2.2.2 The continuous string sin(jπ) = sin(p0 j) = sin(0) = 0. And
n = 6 e.g. is the same as n = 4, and
The total length of the chain is so on.
R=Na (2.20)
We find the continuum limit by taking the mass distance a to zero

while keeping the the total length R fixed
a → 0, N → ∞, R = N a = const. (2.21)
In this limit the discrete chain approaches a continuous string and

the labeled coordinates qj (t) approach a classical field defined as
q(x, t) ≡ qj (t) (2.22)
We can take the limit also for the Lagrangian by replacing

2 2
qj − qj+1 ∂q(x, t)
(qj − qj+1 )2 = a2 → a2 (2.23)
a ∂x
X 1X Z R
1
= a → dx (2.24)
j
a j
a 0
Keeping the mass density ρ and string tension σ finite, we define

m
ρ= , σ = κa (2.25)
a
to obtain the Lagrangian
Z " 2 2 #
R
1 ∂q(x, t) ∂q(x, t)
L= dx ρ −σ (2.26)
2 0 ∂t ∂x
and the EOM

1 ∂ 2 q(x, t) ∂ 2 q(x, t)
− =0 (2.27)
c2 ∂t2 ∂x2
which is of course a wave-equation with a speed of sound of
σ
c2 = (2.28)
ρ
General solutions can be expanded in
q(x, t) = a · ei(kx−ωt) + b · e−i(kx−ωt) (2.29)
with a dispersion law

ω = ck (2.30)
As above, the boundary conditions of fixed ends
q(0, t) = q(R, t) = 0 (2.31)
give us a form of the normal modes
qn (x, t) = cos(ωn t) sin(kn x) (2.32)
with ωn = c kn and
πn
kn = (2.33)
R
The normal-modes reproduce the discrete chain as long as n/N
Figure 2.2: Normal modes of a

1.4 classical string ωn ∼ n and a
continuous string discrete chain for N = 8 which has
ωn ∼ sin(n/N ).
1.2
!1.0
0
0.8
discrete chain
0.6
0.4
0.2
0.0
0 2 4 6 8
1. We find however that the number of modes of the continuous

string is infinite! Only the first N modes correspond to those of the
discrete chain. Therefore, we see that there is a cut-off frequency
ωC which is
πc
ωC ≡ ωN = ∼ 1/a (2.34)
a
This is of the same parametric size as the maximum frequency of the

pκ
discrete chain ω0 = 2 m = 2 c/a, see Eq. (2.19). Our continuum
model q(x, t) is a good representation of the discrete system only for
ω ωC .
This is a first encounter with the important concept of renormalization. The cut-off frequency
ωC ∼ 1/a
is a theoretical necessity. The cut-off tells us until which scale, we can trust our continuum theory. A continuous
string at frequencies ω ωC is not a good description of the discrete chain: the minimal wavelength is set by the
atomic distance, below which there is nothing to oscillate. Without a cut-off the specific heat of the string would
diverge since at a temperature T each mode (equipartition theorem in statistical physics) contributes an amount
kT , where k is the Boltzmann constant.
Here, we cannot determine the value of the cut-off from the long-wavelength theory alone, because only the
combination
c ∼ a ωC
appears. We can absorb the cut-off in measurable parameters, as in Eq. (2.25). This is called renormalization. If
this can be done, a theory is called renormalizable. You will discuss this at length in the chapter of loop correc-
tions in quantum field theories.
If a theory is non-renormalizable, the behavior can be sensitive to short-distance physics, which in this case would
be the atomic motion. This behavior does appear random on macroscopic scales, as in the propagation of a crack
or the nucleation of a raindrop.
2.2.3 Quantum string theory

We quantize the classical continuum theory to obtain a first example
of a quantum field theory. The Hamiltonian of the discrete chain is
N
" #
X p2j κ 2
H(p, q) = T + V = + (qj − qj+1 ) (2.35)
j=1
2m 2
with pj = mq̇j . We can quantize the system, if we can replace pj and

qj with hermitian operators which satisfy the commutation In the following we will mostly set
~ = 1.
[p̂j , q̂k ] = −i~δjk (2.36)
We will switch to periodic boundary conditions now, Think of the system as being on a
torus. The spacing of the frequencies
q̂j+N (t) = q̂j (t) (2.37) is doubled but each frequency ap-
pears two-fold. We are not interested
since they allow us to use just the exponentials in the behaviour at the endpoints
and we will be at large N .
N/2
1 X
q̂j (t) = √ Q̂n ei2πnj/N (2.38)
N n=−N/2
N/2
1 X
p̂j (t) = √ P̂n ei2πnj/N (2.39)
N n=−N/2
where the operators P̂n and Q̂n satisfy
[P̂n† , Q̂m ] = −iδnm , (2.40)

P̂n† = P̂−n (2.41)
Q̂†n = Q̂−n (2.42)
Substituting back into the Hamilto-

The Hamiltonian is now simplified to a sum of independent nian, showing just the p2j -term.
harmonic oscillators N N N/2 N/2
X 1 X X X
p2j = P̂n ei2πnj/N P̂m ei2πm
N
X N j=1
1 1 j=1 n=−N/2 m=−N/2
H= P̂n† P̂n + mωn2 Q̂†n Q̂n (2.43)
n=1
2m 2 N/2
X N/2
X N
1 X i2π(n+m)j/N
= P̂n P̂m e
N j=1
n=−N/2 m=−N/2
with eigen-frequencies | {z }
δn,−m
4κ n N/2 N/2
ωn2 = sin2 π (2.44) =
X
P̂n P̂−n =
X
P̂n P̂n†
m N
n=−N/2 n=−N/2
As you remember from your QM course, we can write the energy- N
X
eigenvalues of the harmonic oscillator with frequency ω as = P̂n† P̂n
n=1

1 where in the last step we have used
n = ~ω +n , n = 0, 1, . . . (2.45) that P̂n† and P̂n commute and that
2
the product P̂n† P̂n therefore is
symmetric n → −n.
where n is the occupation number. We can therefore write the energy- The orthogonality relation follows
eigenvalues of H as a set of occupation numbers from the geometric series.
N
N/2
X X
ei2π(n+m)j/N =
1 − ei2π(n+m)j
=0
1 1 − ei2π(n+m)j/N
Eα = ωn + αn (2.46) j=1
2
n=−N/2
because we have ei2π(n+m)j = 1
since n, m ∈ N and ei2π(n+m)j/N 6= 1
with αn = 0, 1, 2, . . .. for n + m 6= 0 . The case n + m = 0 is
trivial.
Ex: Show the result for the qj
2.2.4 Review of the simple harmonic oscillator
substitution!
Consider the quantum mechanical Hamiltonian
1 2 1 2 2
H= p + ω q (2.47)
2 2
with the canonical commutation relations
[q, p] = i (2.48)
To find the spectrum we define the creation and annihilation opera-

tors (also known as raising/lowering operators, or sometimes ladder
operators)
r
ω i
a= q + √ p, (2.49)
2 2ω
r
ω i
a† = q− √ p (2.50)
2 2ω
which we can invert to
1
q = √ (a + a† ), (2.51)
2ω
r
ω
p = −i (a − a† ) (2.52)
2
Substituting the expressions we find
[a, a† ] = 1 (2.53)
while the Hamiltonian is

1
H= ω(aa† + a† a) (2.54)
2
1
H = ω(a† a + ) (2.55)
2
One can easily confirm that the commutators between the Hamilto-
nian and the creation and annihilation operators are given by
[H, a† ] = ωa† , (2.56)

[H, a] = −ωa (2.57)
Let |Ei be an eigenstate with energy E, so that
H|Ei = E|Ei (2.58)
Then we can construct more eigenstates by operating with a and a† ,
Ha† |Ei = (E + ω)a† |Ei, (2.59)

Ha|Ei = (E − ω)a|Ei (2.60)
So we find that we can generate a ladder of states for each k with

energies
. . . , E − ω, E, E + ω, E + 2ω, . . . (2.61)
The energy is bounded from below and there must be a ground state
|0i which satisfies
a|0i = 0 (2.62)
This has the ground-state energy (also called zero-point energy),
1
H|0i = ω|0i (2.63)
2
2.2.5 Fock space of the discrete chain

For this simple system, the Hilbert space can be identified with
the tensor product of the Hilbert spaces of the individual atoms
or masses. Thus, if we identify with |Ψin an arbitrary state in the
Hilbert space of the n-th atom, the states of the chain can then be
written as
|Ψi = |Ψi1 ⊕ |Ψi2 ⊕ . . . ⊕ |ΨiN (2.64)
This is a first example of the Fock space of multi-particle systems.
We can further diagonalize the Hamiltonian by making use of the

creation and annihilation operator formalism for the harmonic
oscillator. We will drop the ”ˆ” for operators
r from now on since it will be clear
mωk i from the context.
ak ≡ Qk + Pk (2.65)
2 mωk
r
mωk i
a†k ≡ Q−k − P−k (2.66)
2 mωk
With this definition we find that the ladder operators obey the
commutation relations.
[ak , a†k0 ] = δkk0 (2.67)

[ak , ak0 ] = [a†k , a†k0 ] =0 (2.68)
Proof: Plugging the definitions of ak and a†k into the commutator

and using the fact that the commutator is bilinear, we can calculate7 7
Thanks to Nepomuk Ritz!
h i r r
mωk i mωk0 i
ak , a†k0 = Qk + Pk , Q−k0 − P−k0
2 mωk 2 mωk0
 
√
m ωk ωk 0  i i 1
= [Qk , Q−k0 ] − [Qk , P−k0 ] + [Pk , Q−k0 ] + 2 [Pk , P−k0 ]
2 | {z } mωk0 mωk m ωk ωk0 | {z }
=0 =0
 
√ h i
m ωk ωk 0 
 i † i h † i† 

= P 0 , Qk − P , Q 0
2  mωk0 k k k 
| {z } mωk
=−iδkk0
√
m ωk ωk 0 δkk0 δkk0 mωk 2
= + = δkk0 = δkk0
2 mωk0 mωk 2 mωk
† †
where
h we
i have also used that P−k = Pk , Q−k = Qk and that
Pk†0 , Qk = −iδkk0 .
r r
mωk i mωk0 i
[ak , ak0 ] = Qk + Pk , Qk0 + Pk 0
2 mωk 2 mωk0
 
√
m ωk ωk0  i i 1
= [Qk , Qk0 ] + [Pk , Qk0 ] + [Qk , Pk0 ] − 2 [Pk , Pk0 ]
2 | {z } mωk mωk0 m ωk ωk0 | {z }
=0 =0
Since the commutator is antisymmetric, we have [Qk , Pk0 ] =

− [Pk0 , Qk ] and therefore, upon exchanging the indices k and k 0
for this term (which we can do, because [Pk0 , Qk ] = −iδk,−k0 ), the
remaining terms cancel and we have
[ak , ak0 ] = 0.
If we take the hermitian conjugate of this equation, we immediately

get h i h i
†
[ak , ak0 ] = a†k0 , a†k = − a†k , a†k0 = 0
and therefore also h i
a†k , a†k0 = 0.
which completes the proof.
The Hamiltonian becomes
N/2
X
1
H= ωk a†k ak + (2.69)
2
k=−N/2
Further we can show that
[H, ak ] = −ωk ak (2.70)

[H, a†k ] = ωk a†k (2.71)
and therefore these operators take us between energy eigenstates.

Let |Ei be an eigenstate with energy E, so that
H|Ei = E|Ei (2.72)
Then we can construct more eigenstates by operating with ak and

a†k ,
Ha†k = ωq (a†q aq + 1/2)a†k
X
Ha†k |Ei = (E + ωk )a†k |Ei, (2.73) q
ωq (a†q aq a†k + 1/2a†k )

X
Hak |Ei = (E − ωk )ak |Ei (2.74) =
q
ωq (a†q (a†k aq + δkq ) + 1/2a†k )

X
So we find that we can generate a ladder of states for each k with =
q
energies
= a†k (H + ωk )
. . . , E − ωk , E, E + ωk , E + 2ωk , . . . (2.75)
The energy is bounded from below and there must be a ground state
|0i which satisfies
ak |0i = 0 for all k (2.76)
This has the ground-state energy (also called zero-point energy),

1X
H|0i = ωk |0i (2.77)
2
k
Excited states then can be generated by repeated application of a†k ,
|n1 , n2 , . . . , nN i = (a†1 )n1 (a†2 )n2 · · · (a†N )nN |0i (2.78)
where we have not been careful about normalization, so hn|ni =

6 1. Ex: Derive the correct normalization
factor.
We now again move to the continuum limit with x = j · a to find
Z R " !#
1 2 σ ∂ φ̂(x, t)
Hcont = dx π̂ (x, t) + (2.79)
0 2ρ 2 ∂x
and the conjugate momentum operator
p̂j (t) ∂ q̂j (t) ∂ φ̂(x, t)

π̂(x, t) = =ρ =ρ (2.80)
a ∂t ∂t
The quantum field φ̂(x, t) and its conjugate momentum π̂(x, t)
satisfy the equal-time commutation relation This is the generalization of the
canonical commutation relations of
one-particle quantum mechanics,
[φ̂(x, t), π̂(x0 , t)] = i~ δ(x − x0 ) (2.81) [xa , pb ] = iδab .
where I have temporarily reintroduced ~ to show how to recover the

classical limit ~ → 0. Also for quantum string we have to introduce a
cut-off frequency ωC .
We have seen that in particular for distances smaller than x ∼

1/m or energies above E ∼ m, we need to extend the QM frame-
work because of causality violation and the possibility of particle
creation. These energies are however also energies where relativistic
effects become important. Therefore we will spend some time on the
basics of relativity in the next chapter.
We started with a discrete chain for pedagogical reasons. Of course we do not believe that the fields observed
in Nature, such as the electron field or the photon field, are actually constructed of point masses tied together
with springs. A modern picture which was first used by Landau-Ginzburg and fully understood by S. Wein-
berg, is that we start with the desired symmetry, say Lorentz invariance or a conserved U (1) charge (more on
both later) decide on the fields we want by specifying how they transform under the symmetry (in this case we
decided on a scalar field φ(x, t) as the local displacement) and then write down the action involving no more than
two time derivatives – because we don’t know how to quantize actions with more than two time derivatives (we
usually end up with ghosts). See also the discussion in the statistical physics script (in particular the chapter:
Landau-Ginzburg Theorie)
3
Relativity
3.1 Galileo and Newton

Two observers, one on a moving train (x, y, z) with velocity u and
one on the ground (x0 , y 0 , z 0 ), can be related by
t0 = t (3.1)
0
x =x+u·t (3.2)
0
y =y (3.3)
z0 = z (3.4)
which are known as the Galilean transformations. The differen-
tial form is
dt0 = dt (3.5)
0
dx = dx + u · dt (3.6)
We can add velocities trivially: e.g. a ball with 3 m/s on a train
with 20 m/s will be observed on the ground as 3 m/s + 20 m/s =
23 m/s. We see this also with the Galilean transformations
dx0 d dx
v0 = = 0 (x + u t) = +u=v+u (3.7)
dt0 dt dt
We now see that the invariance of Newtonian mechanics under
Galilean transformations follows merely because Newton’s law
involves 2nd derivatives, so

d2 x0 d dx d2 x
m 02 = m 0 +u =m 2 (3.8)
dt dt dt dt
We also see importantly where Galilean invariance in Newtonian
mechanics fails: if u changes in magnitude or direction with time
then we find1 1
All inertial reference frame is a
du frame in which Newton’s first axiom
F 0 = ma0 = ma + m (3.9) holds, i.e. every object on which no
dt force is acting either stays at rest or
we feel this as an additional force, e.g. as a centrifugal force in an moves uniformly.
accelerated frame of reference. Let us check that the Newtonian
action is invariant, too
Z 0 2 Z 2
01 dx 1 dx
S = dt m = dt m +u (3.10)
2 dt0 2 dt
Z 2 Z Z
1 dx dx 1
= dt m + u dtm + u2 dt m (3.11)
2 dt dt 2
for fixed initial and final conditions, the second term is just an
irrelevant additional constant, like the last term. We can generalize
this to a many particle system with interactions
 
Z X 1 2 X 
dxa
S = dt ma − V (xa − xb ) (3.12)
 2 dt 
a a6=b
Note, that it is necessary for the interaction potential to just de-

pend on the difference xa − xb . We also see that all observers have
to agree on the same mass m, there is no m0 otherwise Galilean
transformations would not work.
3.1.1 Maxwell vs. Newton

Maxwell’s equations are not invariant under Galilean transforma-
tions. In particular, the velocity of light is the same in all inertial Figure 3.1: Michelson and Morley’s
systems interferometric setup, mounted on a
stone slab that floats in an annular
c0 = c (3.13) trough of mercury.
contradicting Eq. (3.7). Various eminent physicists in the late 19th

century tried to reconcile this fact by postulating that light – like
sound – had to propagate in a medium, the aether. In 1887 (when
Einstein was eight years old) Michelson and Morley performed a
famous experiment to detect the aether and failed.
The Michelson–Morley experiment was performed over the spring and summer of 1887 by Albert A. Michel-
son and Edward W. Morley at what is now Case Western Reserve University in Cleveland, Ohio, and published
in November of the same year. It compared the speed of light in perpendicular directions, in an attempt to detect
the relative motion of matter through the stationary luminiferous aether (‘’aether wind”). The result was negative,
in that the expected difference between the speed of light in the direction of movement through the presumed
aether, and the speed at right angles, was found not to exist; this result is generally considered to be the first
strong evidence against the then-prevalent aether theory, and initiated a line of research that eventually led to
special relativity, which rules out a stationary aether.
Michelson–Morley type experiments have been repeated many times with steadily increasing sensitivity. These
include experiments from 1902 to 1905, and a series of experiments in the 1920s. More recent optical resonator
experiments confirmed the absence of any aether wind at the 10−17 level. Together with the Ives–Stilwell and
Kennedy–Thorndike experiments, Michelson–Morley type experiments form one of the fundamental tests of special
relativity theory. https://en.wikipedia.org/wiki/Michelson-Morley_experiment
3.1.2 Derivation of the Lorentz transformations

We consider two inertial systems and assume that they coincide for
t = 0 = t0 with a relative velocity v.
Let us know see how we can modify the Galilean transformations
so that a spherical light wave emanating from the origin
(ct)2 = x2 + y 2 + z 2 (3.14)
does not depend on the observer and is the same in the moving
inertial system
(ct0 )2 = x02 + y 02 + z 02 (3.15)

Let us assume that the origin is moving along the x-axis. We know Three useful hyperbolic function
then formulae
cosh2 η − sinh2 η = 1, (H1)
0 0 tanh η
y=y, z=z (3.16) sinh η = p , (H2)
1 − tanh2 η
1
The most general linear solution is, as can be verified by direct cosh η = p (H3)
1 − tanh2 η
substitution using (H1)
H2 and H3 can be easily shown by
! ! ! plugging in the definition of tanh η
0
x cosh η sinh η x and use of H1.
= (3.17)
ct sinh η cosh η ct0
This should remind you of an euclidean rotation which leaves x2 + y 2

invariant (instead of (ct)2 − x2 ).
To find the role of η in the physical setting, record the origin’s
progression, i.e. x0 = 0, x = vt. The equations become (using first
x0 = 0),
x = ct0 sinh η, (3.18)

0
ct = ct cosh η. (3.19)
Now divide :
x v
= tanh η = = β ⇒
ct c
v
c 1
sinhη = q , cosh η = q = γ,
v2 v2
1− c2 1− c2
where x = vt was used in the first step, (H2) and (H3) in the second,
which, when plugged back in (3.17), gives
x0 + vt0 t0 + v2 x0
x= q , t= q c , (3.20)
2 2
1 − vc2 1 − vc2
or, with the usual abbreviations,
x = γ(x0 + vt0 ), (3.21)

0 vx0
t=γ t + 2 , (3.22)
c
We see that the transformations for time and space are now symmet-
ric2 ! With 2
Very clear if you set c = 1.
Or,
1 x0 = γ(x − vt), (3.23)
γ=q (3.25)
v 2 vx
1−

c t0 = γ t − 2 . (3.24)
c
3.2 Consequences of special relativity
All reference frames in this section are inertial frames.

Figure 3.2: Light-cones and a world-

line in Minkowski-Space.
ct
world line
light-cone
future
space-like
x
past
time-like
3.2.1 Relativity of simultaneity

Two events happening in two different locations x1 and x2 that
occur simultaneously E.g. by sending light signals back
and forth. https://en.wikipedia.
t1 = t2 ∆t = 0 (3.26) org/wiki/Einstein_synchronisation
in the reference frame of one observer A, may occur non-simultaneously

in the reference frame of another observer B 0 . Let’s move in the
boosted frame B 0 traveling with v relative to A.
vx1 vx2
t01 = γ t − 2 , t02 = γ t − 2 (3.27)
c c
v
∆t0 = t01 − t02 = γ 2 (x2 − x1 ) 6= 0 (3.28)
c
This is counter-intuitive and seems to contradict our usual notion of
cause and effect. Can we change the order of two events? If t2 −t1 > 0,
we want that
v
0 < t02 − t01 = γ t2 − t1 − 2 (x2 − x1 ) (3.29)
c
We have to have that
v x2 − x1
t2 − t1 > (3.30)
c c
Since v < c, the order of the events is kept for
x2 − x1
t2 − t1 ≥ (3.31)
c
If two events are to be causally connected, the cause can only be
transmitted with finite speed v̄ ≤ c. This means
x2 − x1 x2 − x1
t2 − t1 = ≥ (3.32)
v̄ c
We find that cause and effect cannot be exchanged.

However, the order of causally disconnected events can very well
be changed.
Figure 3.3: Relativity of simultaneity

t t’ in a space-time diagram (c = 1). The
coordinate lines show unit-ticks and
unit-distances in the restframe (x, t)
and a moving inertial frame (x0 , t0 )
with v = 0.3. Two simultaneous
events A and B are happening at
different times in (x0 , t0 ).
tA tB = 0
A B x’
t0 > 0
3.2.2 Time dilation

Let us assume that we have a clock in A which ticks every ∆t sec-
onds. In B 0 we will find the ticks at
∆t
∆t0 = γ∆t = q ≥ ∆t (3.33)
v 2
1− c
The stationary clock ticks faster! Why is the situation not symmet-
ric? An observer in B 0 will have to use two clocks to compare to the
for him moving clock of A.
We can now define the important quantity of proper time ∆τ :
it is the time measured by a clock which is at a fixed place. We can
observe time dilation in (almost) everyday experiments like the decay
of unstable particles, like the muons generated in the atmosphere by
cosmic radiation. See Ex1!
3.2.3 Lorentz contraction

Suppose we measure a length as ∆x = l in A. How did we perform
the measurement? We put a standard ruler on the distance and
compare the endpoints simultaneously ∆t = 0. A measurement
of the length in B 0 moving with v relative to A is characterized by
∆t0 = 0. The positions of the endpoints in B 0 are
∆x0 = γ(∆x − v∆t) (3.34)

where ∆t = t2 − t1 =?. We have to measure simultaneously in B 0 :

v ! v
t1 − x1 = t 2 − 2 x2 (3.35)
c2 c
Therefore
v
t1 − t2 = (x1 − x2 ) (3.36)
c2
Plugging back

0 v2
l = x01 − x02 = γ x1 − x2 − 2 (x1 − x2 ) (3.37)
c
which means
r
0 v2
l =l 1− (3.38)
c2
A stickqof length l at rest in A will appear contracted in B 0 by a

2
factor 1 − vc2 < 1. Crucial for the length measurement is the
requirement of simultaneity.
Figure 3.4: Time dilation in a

t t’
space-time diagram (c = 1). The
with v = 0.3. We see that we need
two clocks in the rest frame to
measure the moving system’s clock.
time dilation
x’
t= t0 > t0
3.2.4 Addition of velocities

We can now derive the correct law of addition of velocities. Consider
again the simple case of an object moving in the x direction. The
Lorentz transformation reads
1
x0 = q (x + vt), (3.39)
v 2
1− c
1 vx
t0 = q t+ . (3.40)
v 2 c2
1− c
Figure 3.5: Length contraction in a

t t’ space-time diagram (c = 1). The
with v = 0.3. We see that when we
measure the length of the moving
measuring stick, we do so at the
same time t = 0, corresponding to
two different times t01 = 6 t02 in the
moving system.
x’
Lorentz
contraction
x
0 0
l=l/ <l
dx dx0
Let the velocity seen in A be u = dt and in B 0 be u0 = dt0 . Then
dividing dx0 by dt0 . We obtain
dx0 vdt + dx v + dx v+u

u0 = = = dt
dx = (3.41)
dt0 vdx
dt + c2 v
1 + cdt 1 + vu
c2
2
Instead of the Galilean u0 = u + v, the correct law contains a crucial

denominator
u+v
u0 = (3.42)
1 + vu
c2
This function is remarkable: it is symmetric under v ↔ u. If the

object is slowly moving u c, then u0 ≈ u + v, in accordance with
everyday intuition. But if a particle happens to be moving at the
c+v
speed of light (e.g. a photon), u = c, then u0 = 1+ vc = c.
c2
3.3 Minkowski space

We really, really set c = 1 from now
In Euclidean space, the invariance of the combination on.
dl2 = dx2 + dy 2 + dz 2 (3.43)
allows us to define dl as the distance between two points. Two ob-

servers whose coordinate systems are rotated to each other measure
the same distance.3 3
Indeed, the invariance of dl2 defines
With profound insight, Minkowski realized4 that the invariance of rotations and the Lie group SO(3).
4
In 1907, a mere 2 years after SR.
the combination
ds2 = dt2 − dx2 − dy 2 − dz 2 (3.44)

allows us to talk about distance in space-time. This invariance

determines the Lorentz transformation5 . Similar to the case of 5
Or the SO(1, 3) rotations
rotations, two observers in uniform motion relative to each other can
now agree on the space-time distance between two points.
We say that the separation between two nearby points in space-
time is
ds2 > 0 dt2 > dx2 + dy 2 + dz 2 timelike

2 2 2 2 2
ds = 0 dt = dx + dy + dz lightlike
ds2 < 0 dt2 < dx2 + dy 2 + dz 2 spacelike
The first two are causally connected. Light-like is also sometimes

called ’null’.
We see that we can identify ds with a proper-time intervall dτ
since it is the time measured by a resting particle ds2 = dt2 . Since
ds2 is Lorentz-invariant, we can determine it in any inertial system
by calculating ds2 = dt2 − (dx2 + dy 2 + dz 2 ), or

2 2 2 dx2 + dy 2 + dz 2
dτ = ds = dt 1 − (3.45)
dt2

= dt2 1 − v 2 (t) (3.46)
which is exactly the already discussed time dilation.

For an arbitrarily moving particle x(t) with a velocity v(t) =
dx(t)/dt, we can calculate
Z τB Z tB p
τB − τA = dτ = dt 1 − v 2 (t) ≤ tB − tA (3.47)
τA tA
3.3.1 Fourvectors
We define space-time points X using the components of a four-vector
 0  
x ct
x1   x 
   
xµ ≡  2  ≡   (3.48)
x   y 
x3 z
The components are defined relative to basis vectors eµ
3
X
X = xµ eµ = xµ eµ (3.49)
µ=0
where we have used the Einstein summation convention, when

an index variable appears twice in a single term, it implies summa-
tion of that term over all the values of the index. In the relativistic
context, we will always require one ’upstairs’ and one ’downstairs’
P3
index. In the following, we will therefore not show the ” µ=0 ”
anymore.
A Lorentz transformation can then be written as
x0µ = Λµ ν xν (3.50)
where we have moved one index to the right to make obvious, which
one is the first and which one is the second index of the matrix.
Since we have only changed the coordinates and the physical vectors
has to remain the same, we know
X = xµ eµ = x0µ e0µ = Λµ ν xν e0µ (3.51)
we find for the basis vectors
eν = Λµ ν e0µ (3.52)
and a Lorentz transformation along the x-direction would look like

 
γ γβ 0 0
γβ γ 0 0
 
(Λµ ν ) =   (3.53)
0 0 1 0
0 0 0 1
p
with γ = 1/ 1 − v 2 /c2 and β = v/c. This is called a boost. We
can of course also parametrize ordinary rotations within a Lorentz
transformation, say a rotation around the z-Axis
 
1 0 0 0
0 cos θ − sin θ 0
 
(Λµ ν ) =   (3.54)
0 sin θ cos θ 0
0 0 0 1
3.3.2 Minkowski Metric

Lorentz transformations leave the relativistic distance ds2 invariant.
This defines a metric of the Minkowski-space
ds2 = dt2 − dx2 − dy 2 − dz 2 = ηµν dxµ dxν (3.55)
with
 
1 0 0 0
0 −1 0 0
 
(ηµν ) =   (3.56)
0 0 −1 0
0 0 0 −1
The invariance of ds2 then implies
ds2 = ηµν dxµ dxν = ηµν dx0µ dx0ν = ηµν Λµ α Λν β dxα dxβ (3.57)
and we derive the condition
ηαβ = ηµν Λµ α Λν β (3.58)
which we will use extensively later. This defines the set of all
Lorentz transformations.6 6
To get the representations of the
Lorentz group SO(1, 3).
3.3.3 Discrete transformations

Space-time distances as measured by the Minkowski metric, or
Minkowski scalar products like
Vµ W µ (3.59)
are left invariant not only by transformations which can be continu-

ously connected to the the identity, but also by discrete transfor-
mations like parity Using
ηαβ = ηµν Λµ α Λν β
P : (t, x, y, z) −→ (t, −x, −y, −z) (3.60)
and taking the determinant, we find
or time reversal det Λ = ±1
T : (t, x, y, z) −→ (−t, x, y, z) (3.61)
Parity and time reversal are special because they cannot be written
as a product of rotations and boosts7 . We can write them as 7
Because rotations and boosts
reduce to 14×4 for v → 0 or θ → 0.
   
1 0 0 0 −1 0 0 0
0 −1 0 0  0 1 0 0
   
P = , T =  (3.62)
0 0 −1 0   0 0 1 0
0 0 0 −1 0 0 0 1
As you will learn in the discussion of the Standard Model of particle

physics, neither symmetry is conserved in nature!8 8
The weak interaction SU (2)L ×
U (1)Y are explicitly parity violating
because the gauge charges distinguish
3.3.4 Motion of a free particle between left-handed and right-
handed fermions. T is violated in
Newton enunciated that in the absence of external forces, a particle flavor changing transitions, since the
will maintain a constant velocity. Hence the equation of motion Yukawa couplings of the quarks have
a complex phase.
d2 X
=0 (3.63)
dt2
or in other words dX dt stays unchanged.
How does a free particle move in special relativity? Again, the
answer had to be enunciated (or guessed) by Einstein. He knew that
it must reduce to Newton’s equation for a slowly moving particle.
Also, all inertial frame observers have to agree that the particle is
freely moving. Both requirements are satisfied by
d2 X µ
=0 (3.64)
dτ 2
where X µ = X µ (τ ) denotes the location of a particle with τ being
the proper time. In another inertial frame X 0µ = Λµ ν X ν and so
d2 X 0µ 2 ν
µ d X
= Λ ν =0 (3.65)
dτ 2 dτ 2
and that dτ is the same for both observers. This motivates the
introduction of the four-velocity
dxµ
uµ ≡ (3.66)
dτ
which because of dt/dτ = γ(v) can be written in terms of the usual

three-velocity v as
!
µ dxµ dt c
u = = γ(v) (3.67)
dt dτ v
The (Lorentz) square of the four-velocity is9 9

See Ex for a four-acceleration
u2 = uµ uµ = γ(v)2 (c2 − v 2 ) = c2 (3.68)
3.4 Indices upstairs and downstairs

For now ηµν is the only object with lower indices. When we want to
sum over indices, the rule is that we multiply by ηµν and invoke the
Einstein summation convention, e.g. for two vectors pµ q ν
p · q ≡ ηµν pµ q ν (3.69)
To save ourselves from constantly writing the metric ηµν , we define,

when we are given a vector pµ a vector with lower index
pν ≡ ηµν pµ (3.70)
Or in other words,
pµ = (p0 , p), and pµ = (p0 , −p) (3.71)
and therefore
p · q = pµ q µ = p0 q 0 − p · q (3.72)
I hope you are not confused by this trivial act of notational sloth
and the fact that we have ”two kinds of vectors”, contravariant
vectors pµ and covariant vectors pµ . There is nothing profound
going on here. Just a convenient notation.10 10
See however box below!
Indeed, we can also define an inverse for ηµν , (ηµν )−1 = η µν
ηµν η νλ = δµλ (3.73)
As you can see it is numerically the same matrix11 and we will use 11
In general curved space-times the
the same symbol η with two covariant indices: ηµν . From this we can metric can be a function of xµ and
the inverse will in general not be the
see there is no such thing as δ µν or δµν , neither does ηνµ exist ! same matrix.
3.4.1 A convenient notation and derivatives

∂
It follows that the short-hand ∂µ for ∂xµ has to carry a lower index
because
∂xν
∂µ xν = = δµν (3.74)
∂xµ
We will use this fact repeatedly. You can also argue that a variation
of a Lorentz invariant scalar field φ(xµ ) should still be a scalar
∂φ µ
δφ = δx (3.75)
∂xµ
and therefore

∂ ∂ ∂ ∂ ∂
∂µ = = , , , (3.76)
∂xµ ∂t ∂x ∂y ∂z
and

∂ ∂ ∂ ∂ ∂
∂µ = = ,− ,− ,− (3.77)
∂xµ ∂t ∂x ∂y ∂z
A useful mnemonic is to think of xµ carrying an upper index but
since in ∂x∂ µ it appears in the denominator, it should carry a lower
index.
We will also use
∂µ Aµ = ∂0 A0 + ∂j Aj (3.78)
and
∂2
∂µ ∂ µ = − ∇2 = (3.79)
∂t2
Finally, we will make use of the completely anti-symmetric tensor
µναβ , known as the Levi-Civita tensor. It is defined as


1,
 if µναβ is an even permutation of (0, 1, 2, 3)
µναβ = −1, if µναβ is an odd permutation of (0, 1, 2, 3) (3.80)



0, if µναβ is not a permutation of (0, 1, 2, 3)
So e.g.0123 = −1023 = 1 and 0223 = 0.
3.4.2 The appearance of anti-matter

An actual physics process (event A), a proton turns into a neu-
tron by emitting a pion, the π + , which by conservation of charge
necessarily carries positive charge. In event B, the π + is absorbed
which turns a neighboring neutron into a proton. This generates an
attraction between the proton and the neutron and is an effective
description of the strong force.
The same process as seen by two observers. Since the pion prop-
agates over a spacelike intervall, it is possible to see a temporal
order in which A occurs after B. This something, by charge conser-
vation, has to carry negative charge which implies the existence of
anti-matter.
3.5 Relativistic mechanics

We define the four-momentum of a particle of mass m with
pµ ≡ muµ (3.84)
which we can write in an inertial system where the particle has the
three-velocity v(t)
! !
0
p c
pµ = = mγ(v) (3.85)
p v
p n Figure 3.6: The need for anti-

n p
matter. The same process as seen
by two observers. Since the pion
propagates over a spacelike intervall,
it is possible to see a temporal
order in which A occurs after B.
⇡+ B A ? This something ”?”, by charge
A conservation, has to carry negative
B charge
p
n p
n
The dual space: The Minkowski metric defines a (not positive definite) norm for four vectors
∆s2 = h∆X , ∆X i = ∆xµ ∆xν heµ , eν i (3.81)

| {z }
ηµν
and a Lorentz-invariant scalar product for arbitrary vectors A, B
hA, Bi = aµ bν ηµν (3.82)
Every vector space V has a dual space, often denoted V ∗ , which is simply the set of all linear maps from the
vector space into the real (or complex) numbers. We can find a basis eµ for the dual space with
eµ [eν ] = δνµ (3.83)
where eµ [·] takes a vector to produce a number.

The metric gives us a canonical isomorphism between the space of covariant vectors V and its dual space V ∗ ,
allowing us to be sloppy and conflate the two. More general objects with n upper indices and m lower indices can
be thought of as living in a tensor product of copies of V and V ∗ , i.e. V ⊗m ⊗ V ∗⊗n .
For us contra-variant and co-variant vectors are just a convenient notation but if you are a mathematician-want-to-
be you can read about this in more profound books, or here https://en.wikipedia.org/wiki/Dual_space.
The three-momentum is therefore
p = mγ(v) v (3.86)
This coincides with the usual definition of momentum if we introduce

a velocity dependent mass m(v) = γ(v)m. We call m = m(0) the
rest mass.
The relativistic energy12 E = cp0 is proportional to the 0- 12
See Ex for a derivation from the
component relativistic Lagrangian
Z
S = −mc ds
mc2 1 v2
E = cp0 = mγ(v)c2 = q ≈ mc2 1 + + . . . (3.87)
2
1 − vc2 2 c2 of a free point-particle.
where a new term, the rest energy, appeared. We find for the
square of the momentum
p2 = pµ pµ = (p0 )2 − p2 = m2 c2 (3.88)
where we have used uµ uµ = c2 .

3.5.1 E = mc2
The energy and the momentum of a particle of mass m form a four-
vector
!
µ E
p = (3.89)
p
whose square is
p2 = E 2 − p2 = m2 (3.90)
This is the relativistic Pythagoras which defines a Lorentz-

invariant relationship between the energy, the rest-mass and the
momentum of a particle:
E 2 = m2 + p2 (3.91)
µ+
⇡+
3.5.2 Application: weak decay of a pion
± ±
A charged π can decay to a muon µ and a muon anti-neutrino νµ .
The muon has about 3/4 the mass of a pion13 and the neutrino is
⌫¯µ
almost massless. We can calculate the energy and the velocity of the
decay products using the four-momentum conservation Figure 3.7: Feynman diagram of a π +
decay.
pµini = pµfinal
13
(3.92) mπ± = 139.57 MeV and mµ± =
105.658 MeV.
which is
! ! ! !
µ µ
Eini Eπ+ Eµ+ + Eν̄ Efinal
= = = (3.93)
pini pπ + pµ+ + pν pfinal
In the rest-system of the pion we derive from energy conservation
Eπ+ = mπ+ = Eµ+ + Eν (3.94)
The almost mass-less neutrino has Eν = |pν | and because the initial
three-momentum was zero in the rest system of the pion, we get
m2π+ − m2µ+
|pν | = |pµ+ | = (3.95)
2mπ+
For the decay to an electron and an electron anti-neutrino, the
electron mass is negligible: we find the decay products back-to-back
and the rest-mass has been converted entirely into kinetic energy
with |pν̄ | = |pµ+ | = mπ+ /2.
4
Classical Field Theory
We will first introduce various aspects of classical fields. We will

return to the subject at several later stages when we need new
concepts.
4.1 Dynamics of fields

In QM, observables are constructed out of q’s and p’s. Classical
fields are defined at each point in space-time.
φa (xµ ) (4.1)
Thus we are dealing with a system with infinitely many degrees of

freedom,1 at least one for each point x in space. Notice, that the 1
Compare the discrete chain to the
concept of position has been demoted from a dynamical variable in continous string!
particle mechanics to a mere label in field theory.
4.1.1 Example: the electro-magnetic field

You are very familiar with E(x, t) and B(x, t) from classical electro-
dynamics. In a more advanced treatment of electromagnetism, we
derive the spatial 3-vectors from a single 4-component field
!
µ µ φ
A (x ) = (4.2)
A
where the vector potential Aµ is a vector in space-time. The original

electric and magnetic fields are related by
∂A
E = −∇φ − , and B =∇×A (4.3)
∂t
which have two of Maxwell’s equations as an immediate consequence
∂B
∇ · B = 0, and = −∇ × E (4.4)
∂t
4.1.2 The Lagrangian density

The dynamics of fields is determined by a Lagrangian which is a
function of φ(x, t), φ̇(x, t), ∇φ(x, t). All the systems in this course
will have the form
Z
L(t) = d3 x L(φa , ∂µ φa ) (4.5)
where the L is the Lagrangian density2 . The action is 2

We will not be very careful in
Z t2 Z distinguishing the Lagrangian density
and the Lagrangian.
S= dt d3 x L (4.6)
t1
Remember that in classical mechanics, L depends on q and q̇. In Interactions with multiple derivatives,
field theory we analogously restrict to Lagrangians depending on e.g.
1
φ and φ̇ and not on φ̈. We could in principle write Lagrangian’s L= φ2 φ
M2
which depend on ∇φ, ∇2 φ, ∇3 φ and so on. But anticipating Lorentz-
may occur due to quantum effects in
invariance, we will only consider ∇φ and no higher derivatives. Also all but the simplest renormalizable
we will not include an explicit xµ dependence, all such dependence field theories. The are generic in all
effective field theories, as we may
comes only implicitly through φ and its derivatives. discuss at the end of the course, time
permitting.
In the following, we will use functional derivatives. A functional F [φ] for us is a map from a vector space of
δF [φ]
functions into the real numbers, like e.g. the action S[φ]. A functional derivative is denoted by δφ(x) and it is
defined as the linear term in an expansion
Z
δF [φ]
F [φ + η] = F [φ] + dx0 η(x0 ) + . . . (4.7)
δφ(x0 )
or
Z
δF F [φ + εη] − F [φ]
dx (x)η(x) = lim (4.8)
δφ ε→0 ε

d
= F [φ + η] , (4.9)
d =0
where η is an arbitrary function. In particular, for the functional F [φ] = φ(x), we find
δF [φ] δφ(x)
= = δ(x − x0 ) (4.10)
δφ δφ(x0 )
The quantity η is called the variation of φ. The differential (or variation or first variation) of the functional F [φ]
is
Z
δF
δη F (φ) = dx (x) η(x) (4.11)
δφ
This is similar in form to the total differential of a function. You should familiarize yourself with the mechanics of
functional derivatives on the next exercise sheet.
We can derive the equations of motion by the principle of least

action. We vary the field at each space-time point, keeping the
end-points fixed and requiring
δS = 0 (4.12)
with
Z
∂L ∂L
δS = d4 x δφa + δ(∂µ φa ) (4.13)
∂φa ∂(∂µ φa )
Z
∂L ∂L ∂L
= d4 x − ∂µ δφa + ∂µ δφa
∂φa ∂(∂µ φa ) ∂(∂µ φa )
The last term is a total derivative and therefore its integral only
depends on the field values at the spatial and temporal boundaries.
It vanishes for any δφa that decays at spatial infinity and obeys Note, we will always make the
assumption that we can drop such
δφa (x, t1 ) = δφa (x, t2 ) = 0 (4.14) total derivatives from the Lagrangian.
It let’s us integrate by parts without
consequence. We will therefore
equate
φ∂µ ψ = −(∂µ φ)ψ

in a Lagrangian.
Requiring δS = 0 then yields the Euler-Lagrange equations of

motions for the fields

∂L ∂L
− ∂µ =0 (4.15)
∂φa ∂(∂µ φa )
4.1.3 Example 1: the Klein-Gordon equation

Consider a real scalar field φ(xµ ) and the Lagrangian
1 µν 1
L= η ∂µ φ∂ν φ − m2 φ2 (4.16)
2 2
1 1 1
= (φ̇)2 − (∇φ)2 − m2 φ2 (4.17)
2 2 2
Comparison with a the usual structure L = T − V we see, that we
have a kinetic energy
Z
1
T = d3 x (φ̇)2 (4.18)
2
and the potential energy of the field
Z
3 1 2 1 2 2
V = d x (∇φ) + m φ (4.19)
2 2
where the first term is called gradient energy, while we usually call
the second term the ”potential”. Using
∂L ∂L
= −m2 φ, and = ∂µφ (4.20)
∂φ ∂(∂µ φ)
we get the Euler-Lagrange equation for the Klein-Gordon field
∂µ ∂ µ φ + m2 φ = 0 (4.21)
If we add an arbitrary potential V (φ) we can generalize the Klein-

Gordon equation to
∂V
φ + =0 (4.22)
∂φ
where we have used = ∂µ ∂ µ for the Laplacian in Minkowski space.
4.1.4 Example 2: first order Lagrangians

Alternatively we can write a Lagrangian which is linear in time
derivatives. Take a complex scalar field ψ whose dynamics is defined
by a real Lagrangian
i ∗ 1 mω 2 ∗
L= (ψ ∂t ψ − (∂t ψ ∗ )ψ) − ∇ψ ∗ · ∇ψ − ψ ψ (4.23)
2 2m 2
We derive the EOMs by treating the ψ and ψ ∗ as independent
variables
∂L i mω 2
∗
= ∂t ψ − ψ (4.24)
∂ψ 2 2
∂L i
=− ψ (4.25)
∂(∂t ψ ∗ ) 2
∂L 1
=− ∇ψ (4.26)
∂(∇ψ ∗ ) 2m
from which we obtain

∂ψ 1 2 mω 2
i =− ∇ ψ+ ψ (4.27)
∂t 2m 2
This looks like the Schrödinger equation except for the harmonic
potential, but its interpretation is very different. The field ψ is a
classical field and we have no probability interpretation attached to
it.
The plane which gives the initial conditions that determine the
future (and the past) uniquely is called a Cauchy surface. It differs
between the two examples above. If L ∝ (∂t φ)2 , we must specify
both φ and ∂t φ to fix the future evolution. For L ∝ ψ ∗ ∂t ψ, we only
need ψ and ψ ∗ .
4.1.5 Example 3: the Maxwell equations

The Maxwell equations without sources are given by the Lagrangian
1 1
L = − (∂µ Aν )(∂ µ Aν ) + (∂µ Aµ )2 (4.28)
2 2
Note that the minus signs are crucial to ensure positive kinetic terms
for the Ai
1 1 1
L= (∂t Ai )2 + (∇A0 )2 − (1 − δij )(∂i Aj )2 (4.29)
2 2 2
Observe, that there is no kinetic term for A0 (no L ∝ (∂t A0 )2 ),
which will have important consequences in QED. With
∂L
= −∂ µ Aν + (∂α Aα )η µν (4.30)
∂(∂µ Aν )
we get the Euler-Lagrange-equations
∂L
∂µ = −∂ 2 Aν + ∂ ν (∂α Aα ) = −∂µ (∂ µ Aν − ∂ ν Aµ ) (4.31)
∂(∂µ Aν )
≡ −∂µ F µν (4.32)
where the field strength is defined as3 3

You should check that this repro-
duces Maxwell’s equations in the
vacuum.
F µν = ∂ µ Aν − ∂ ν Aµ (4.33)
The Maxwell Lagrangian therefore can be written in the compact

form
1
L = − Fµν F µν (4.34)
4
where we’ve used an integration by parts.
4.1.6 Locality
All examples above are local Lagrangians. We have not written any
terms that look like
Z
L = d3 xd3 yφ(x)φ(y)
which would couple a field at x to a field at y. The closest we get is Again, effective field theories are
a coupling between φ(x) and φ(x + δx)through the gradient term special and can describe the correct
physics until some scale while being
(∇φ(x))2 . This property of locality is, as far as we know, a key non-local.
feature of all theories of Nature. Locality has a deep connection
to unitarity and one finds that non-local theories quickly violate
unitarity.
4.1.7 Lorentz invariance

An example for an active transfor-
Lorentz transformations have a representation on the fields. The mation: start with a temperature
simplest example is the scalar field which, under the Lorentz trans- field φ(x) which has a hotspot at
x = (1, 0, 0). After a rotation
formation xµ → Λµ ν xν , transforms as
x → Rx
φ(x) → φ0 (x) = φ(Λ−1 x) (4.35) The new field φ0 (x) will have the
hotspot at x = (0, 1, 0). If we want
There is an inverse Λ−1 in the argument because we are dealing with to express φ0 (x) in terms of the old
an active transformation in which we truly shift the field. field φ, we need to place ourselves at
x = (0, 1, 0) and ask ourselves what
We call a theory Lorentz invariant if φ(x) solves the EOMs then the old field looked like where we’ve
also φ(Λ−1 x) solves the EOMs. We can ensure that this holds by come from at
requiring the action to be Lorentz invariant. Let’s look at our three R−1 x = R−1 (0, 1, 0) = (1, 0, 0)
examples. See next comment on passive trans-
formations.
1. Klein-Gordon equation: For our real scalar field we have Continuing with the temperature
example: This R−1 is the reason for
φ(x) → φ0 (x) = φ(Λ−1 x) = φ(y). The contraction of deriva- the inverse transformation in the
tives is obviously invariant using Eq. (3.58). The potential terms formula.
If we were instead dealing with a
transform similarly, with φ2 (x) → φ2 (y). Putting this all to-
passive transformation in which we
gether, we find that the action is indeed invariant under Lorentz relabel our choice of coordinates, we
transformations, would have instead
Z Z φ(x) → φ0 (x) = φ(Λx)

S = d x L(x) = d4 x ∂µ φ(x)∂ µ φ(x) − m2 φ2 (x)
4
Z Z

−→ d4 y ∂µ0 φ(y)∂ 0µ φ(y) − m2 φ2 (y) = d4 y L(y) = S
∂
where we’ve denoted ∂µ0 = ∂y and used the fact that the Jacobian
is trivial
d4 x → det(Λ) d4 y = d4 y
with4 4
Ex: Show that
| det(Λ)| = 1
det(Λ) = 1
using Eq. (3.58).
for Lorentz transformations connected to the identity, which we
are using for now.
2. First order equation: The theory is not Lorentz invariant since

space and time are not on the same footing. In practice, it’s easy
to see if the action is Lorentz invariant: just make sure all the
Lorentz indices µ, ν, . . . are contracted with Lorentz invariant
objects, such as the metric ηµν .
3. Maxwell’s equations: Under a Lorentz transformation we

obtain
Aµ (x) → Λµ ν Aν (Λ−1 x)
and you can see that Maxwell’s Lagrangian is indeed invariant.

Historically of course, it was Maxwell’s Lagrangian which first
show the concept of Lorentz invariance.
4.2 Symmetries: Noether’s Theorem and conserved currents

Symmetries in field theory are probably even more important than
in particle mechanics. Every additional symmetry constraints and
reduces our freedom to pick terms and their coefficients in the
Lagrangian. There are Lorentz symmetries, internal symmetries (like
U (1) or SU (3) in QCD), gauge redundancies, supersymmetry and so
on.
They are the real point of the Lagrangian formalism: the La-
grangian provides a natural framework for the quantum mechanical
implementation of symmetry principles. Noether’s theorem is useful
because it allows you to make exact statements about the solutions
of a theory without solving it explicitly. Since in quantum field the-
ory we won’t be able to solve almost anything exactly, symmetry
arguments will be extremely important!
Noether’s theorem: Every continuous symmetry of the action
gives rise to a conserved current jµ (x) such that the equations of
motion imply
∂µ j µ (x) = 0 (4.36)
which in components is ∂t j 0 + ∇ · j = 0. We will now prove this.

Since we are assuming a continuous symmetry, we may prove it by
working infinitesimally. The transformation
δφa (x) = Xa (x) (4.37)
is a symmetry of the action. This means that the Lagrangian

changes at most by a total derivative
δXa (x) L = ∂µ B µ (4.38)
For an arbitrary variation δφa (x), we know that

∂L ∂L ∂L
δL = − ∂µ δφa + ∂µ δφa (4.39)
∂φa ∂(∂µ φa ) ∂(∂µ φa )
When the equations of motion are satisfied, the first term [. . .]

vanishes! We then have

∂L

δL EOM = ∂µ δφa (4.40)
∂(∂µ φa )
For a transformation which is a symmetry of the action, we know

however that δXa (x) L = ∂µ B µ . If we restrict ourselves to variations
that satisfy the EOMs, we obtain for the transformations which leave
the action invariant

∂L
δXa (x) LEOM = ∂µ Xa (x) = ∂µ B µ (4.41)
∂(∂µ φa )
or

∂L
∂µ Xa (x) − B µ =0 (4.42)
∂(∂µ φa )
This proves the theorem and we find Noether’s theorem was proved
for classical fields and one usually
extends it to quantum theory by
∂L replacing the fields by the corre-
∂µ j µ (x) = 0 with j µ (x) = Xa (x) − B µ (4.43) sponding quantized fields. This does
∂(∂µ φa ) however not always give a conserved
current in the quantum theory! A
non-zero divergence is called an
current anomaly. In the SM, e.g.
4.2.1 Conserved currents baryon and lepton number are global
symmetries which have a quantum
A conserved current implies a conserved charge Q with anomaly.
Z
Q ≡ d3 xj 0 (4.44)
where the integration is over spatial coordinates only. We can see

this by taking the time-derivative
Z 0 Z
dQ 3 ∂j
= d x = − d3 x ∇ · j = 0 (4.45)
dt ∂t
where we have assumed that j → 0 fast enough as x → ∞. Noether’s theorem does not apply
There is one such conserved current j µ and one constant of the to discrete symmetries, such as
φ → −φ of
motion Q for each independent infinitesimal symmetry transforma-
1
tion. L= φφ − m2 φ2 − λφ4
2
A conserved current is a stronger statement than a global con- with φ real. Discrete symmetries can
served charge (i.e. integrated over all of R3 ), since a conserved still lead to selection rules.
current implies a local conservation of charge! With the analysis

above for a finite volume V in R3
Z
QV = d3 x j 0 (4.46)
V
we find5 5
∂V is the boundary area of V .
Z Z
dQV
=− d3 x ∇ · j = − dS · j (4.47)
dt V ∂V
with Stokes theorem. Any charge leaving V per time interval must
be equal to the flow of three-current out of the volume. This holds in
any local field theory.
4.2.2 Example 1: internal symmetries U (1)

A global internal symmetry only involves a transformation of
the fields and acts the same at every point in space-time. A simple
example can be obtained using a complex scalar field ψ(x)
1
ψ(x) = √ (φ1 (x) + i φ2 (x))
2
with φi real fields. We assume the real Lagrangian
L = ∂µ ψ ∗ ∂ µ ψ − V (ψ ∗ ψ) (4.48)
where the potential is a general polynomial in |ψ|2 = ψ ∗ ψ. Instead

of expanding in φ1 , φ2 and deriving the equations of motion, we can
equivalently treat ψ ∗ , ψ as independent variables. We obtain the
EOM
∂V (ψ ∗ ψ)
∂µ ∂ µ ψ + =0 (4.49)
∂ψ ∗
after varying ψ ∗ . The Lagrangian is invariant under phase rotations
of ψ
ψ(x) → eiα ψ(x), or δψ = iα ψ (4.50)
This is called a global U (1) symmetry. We say the internal sym-

metry is G = U (1). This implies a conservation law, since the
Lagrangian remains invariant6 under this transformation δL = 0, 6
Not even a total derivative.
and so B µ = 0. We obtain for the conserved current7 7
We drop the constant α and we
perform the sum over internal indices
∂L ∂L ∂L 1, 2 by including ψ and ψ ∗ .
jµ = Xa (x) = δψ + δψ ∗ (4.51)
∂(∂µ φa ) ∂(∂µ ψ) ∂(∂µ ψ ∗ )
and find
j µ = i(∂ µ ψ ∗ )ψ − iψ ∗ (∂ µ ψ) (4.52)
Once we quantize this field theory, we will see that it can be in-
terpreted as electric charge and implies a conservation of particle
number.
4.2.3 Example 2: an internal SO(N ) and a bit on Lie groups

We can rewrite this Lagrangian using a basis of real fields8 8
A summation over repeated indices
is implied here, too.
L = ∂µ φa ∂ µ φa − V (φa φa ) (4.53)
We see that all the terms are polynomials of the form φa φa = φ21 + φ22 ,
similar to r2 = x2 + y 2 which is invariant under rotations. We
reproduce the complex transformation with
! ! !
φ1 (x) cos α sin α φ1 (x)
→ (4.54)
φ2 (x) − sin α cos α φ2 (x)
or
! ! !
δφ1 (x) 0 α φ1 (x)
→ (4.55)
δφ2 (x) −α 0 φ2 (x)
which we can write as Remember only greek indices like

µ, ν, α, . . . are Lorentz contracted
δφi (x) = αεij φj (x) (4.56) with ηµν , here we just sum over
repeated indices.
with εij = −εji and ε12 = 1. This is a two dimensional, real rotation
or G = O(2). If we restrict ourselves to transformations which can
be continuously connected to the identity, we would have G = SO(2)
which has the additional restriction that the group elements have
unit determinant or
det O = 1 (4.57)
This is called the special orthogonal group in 2 dimensions.9 9

We will always factor G into
Infinitesimal group elements can be written as a direct product of semi-simple
groups and discrete factors. For
the conservation laws, the discrete
O = 1 + αεij + . . . = 1 + iαT ij + . . . (4.58) factors are not important.
with O ∈ G = SO(2). We call the matrix T ij parameterizing

an infinitesimal rotation, a (or ”the” since there is only one here)
generator of the group. We could have written the transformation
of φa as
δφi (x) = iαT ij φj (x) (4.59)
and find the conserved current as
j µ = (∂ µ φ1 )φ2 − (∂ µ φ2 )φ1 (4.60)
or in terms of the generator
j µ = iT ij (∂ µ φi )φj (4.61)
Generally, we obtain one conserved current for each infinitesimal

transformation, or generator, that leaves the action invariant.
We can generalize this symmetry to a = 1, 2, . . . , N . This is a
non-Abelian symmetry group G = SO(N ). Similarly, we could
sum a complex field ψa over an index a and build a Lagrangian
with polynomials of ψa∗ ψa (and derivatives thereof) to obtain a
G = SU (N ) invariant theory. SU (N ) is complex non-abelian group
and contains all special unitary transformations in N dimensions.
These groups are called non-abelian, because their generators in
general do not commute, e.g. for SO(3)
[T a , T b ] = iεabc T c (4.62)
where we have suppressed the obvious matrix indices of the genera-

tors. We find 3 conserved currents:
jaµ = iTaij (∂ µ φi )φj (4.63)
We will return to currents and conserved charges once we have

introduced quantized fields.10 10
Anticipating the result, we state
that the non-abelian nature of the
algebra implies that it is not possible
4.2.4 Example 3: space-time translations to simultaneously measure more than
one of the SO(3) charges of a state:
Let us now study the case in which we transform space and time. the charges are non-commuting
Recall that in classical mechanics, invariance under spatial transla- observables.
tions gives rise to the conservation of momentum, while invariance

under time translations leads to the conservation of energy. Consider
an infinitesimal transformation11 11
Also called the Poincaré transfor-
mations.
x ν → x ν − ν (4.64)
φa (x) → φa (x) + ν ∂ν φa (x) (4.65)

Note the ”+” sign because we are performing an active transforma-

tion. The Lagrangian itself will transform as
L(x) → L(x) + ν ∂ν L(x) (4.66)
Noether’s theorem tells us that there are four conserved currents

(j µ )ν for each one of the four space-time translations ν . Together
with the total derivative (B µ )ν = δνµ µ L, we find
∂L
(j µ )ν = ∂ν φa − δνµ L ≡ T µ ν (4.67)
∂(∂µ φa )
This defines the energy-momentum-tensor T µ ν . The four con-

served quantities are the total energy E
Z
E = d3 xT 00 (4.68)
and
Z
Pi = d3 xT 0i (4.69)
where P i is the total physical momentum12 carried by the field. 12

Which is not to be confused with
the canonical momentum, which we
For the Klein-Gordon Lagrangian, we can e.g. compute13 will introduce below.
13
Ex: Verify that indeed
T µν = ∂ µ φ∂ ν φ − η µν L (4.70) ∂µ T µν = 0
using the equations of motion.
We find for E and P i
Z
3 1 2 1 2 1 2 2
E = d x (∂t φ) + (∇φ ) + m φ (4.71)
2 2 2
Z
P i = d3 x(∂t φ)∂ i φ (4.72)
or with ∂ i = −∂i = −∇i which we will use later:

Z
P = − d3 x(∂t φ)∇φ (4.73)
Note that the energy momentum tensor accidentally came out

symmetric here, such that T µν = T νµ . This might not always be the
case, but we can always find a symmetric form!14 The trick is that 14
In general relativity, we need a
we add an extra term which is anti-symmetric in the first two indices symmetric energy-momentum tensor
Γρµν = −Γµρν 8πG

Rµν − 12 R gµν = − Tµν ,
c4
since the metric and the Ricci tensor
Θµν = T µν + ∂ρ Γρµν are symmetric.
because the anti-symmetry of Γρµν guarantees that ∂µ ∂ρ Γρµν = 0,15 15

Thanks to Schwarz’s theorem or
which means that ∂µ Θµν = 0 still holds. the symmetry of second order partial
derivatives.
In general relativity, the metric gµν is a field and we can expand it
√
as gµν = ηµν + GN hµν + . . . . Once we enter this into the Einstein
action and expand, we will find for the linear terms in hµν a coupling
hµν Θµν .
4.2.5 Summary of Noether’s theorem

We summarize Noether’s theorem:
1. There is one conserved current j µ and one constant of motion Q

for each independent infinitesimal symmetry transformation.
2. The symmetry must be continuous, otherwise the δφa has no

meaning.
3. The current is conserved on-shell, that is when the equations of

motions are satisfied.
4. It applies to global symmetries, parametrized by numbers α, e.g.

φ → eiα φ.16 16
As we shall see later: also for
local symmetries parametrized by
Currents will appear as sources for vector fields, e.g. appearing in α = α(x)
the Lagrangian as
L = . . . − jµ (x)Aµ (x). (4.74)
This can be an explicit external current such as the charge current or

just a formal place-holder for the composite expression
j µ = i(∂ µ ψ ∗ )ψ − iψ ∗ (∂ µ ψ) (4.75)
We will discuss this issue in more detail once we tackle the issue of
gauge invariance.
4.3 Coulomb’s law and the propagator method

Let us do some calculations to better understand aspects of classical
field theory and introduce Fourier transformation techniques along
the way. We will derive Coulomb’s law from relativistic electro-
dynamics.
A charge e at the origin can be represented by the current j µ
! !
µ ρ(x) e δ 3 (x)
j = = (4.76)
v i (x) 0
Adding the source term to the Lagrangian, we get

1
L = − Fµν F µν − jµ (x)Aµ (x). (4.77)
4
The Euler-Lagrange equations are
∂L ∂L
0= ν
− ∂µ = −jµ + ∂ 2 Aν − ∂ ν (∂α Aα ) (4.78)
∂A ∂(∂µ Aν )
= −jµ + ∂µ F µν (4.79)
where we have used Eq. (4.31). As you may remember from your
electro-dynamics lectures, we can freely choose a gauge for Aµ and
we will use Lorenz gauge ∂α Aα = 0 to obtain
Aν = jµ (4.80)
which has the formal solution

1
Aν = jµ (4.81)

where 1/ is just the inverse of which we will define more pre-
cisely later. This is a very common procedure in quantum field
theory: we say that the field Aν is determined by the source j ν after
it propagates with the propagator
1
ΠA = (4.82)

We will discuss propagators in great detail. For our example above
we have to solve the following equations
Ai (x) = 0 (4.83)
e
A0 (x) = δ 3 (x) (4.84)

The homogeneous solutions Aν = 0 , which are the electro-
magnetic waves, we will ignore here since they have nothing to
do with our source.
4.3.1 Fourier transformations

We will determine the solution of the equation of motion by taking a
Fourier transformation. Generally, we have We will not be very careful in distin-
guishing the field from its Fourier
Z
d3 p ˜ transformation. It will be clear from
f (x) = f (p) eix·p (4.86) the variables what we mean. We’ll
(2π)3
use p or k for the transformed field,
e.g.
Recall, that the Fourier transformation of a δ-function (or δ tempered-
d3 p
Z
distribution) is just φ(x) = φ(p) eix·p (4.85)
(2π)3
Z
δ̃ 3 (p) = d3 x δ 3 (x) e−ix·p = 1 (4.87) Note the choice of normalization!
The D-dimensional momentum
integral is accompanied with a factor
and 1
(2π)D
Z
d3 p ix·p
δ 3 (x) = e (4.88)
(2π)3
We obtain for the derivatives of δ-functions

Z Z
d3 p d3 p
∆ δ 3 (x) = ∆ e ix·p
= (ip)2 eix·p (4.89)
(2π)3 (2π)3
This also works for four-vectors

Z
d4 k ixµ kµ
δ 4 (x) = e (4.90)
(2π)4
and so
Z
4 d4 k µ
δ (x) = (−k µ kµ ) eix kµ (4.91)
(2π)4
We will use this all the time: for you as a field theorist, means
”−k 2 ”.
4.3.2 The Coulomb Potential

Back to our problem: since δ 3 (x) is time-independent, we simplify
e 3 e
A0 (x) = δ (x) = − δ 3 (x) (4.92)
∆
We solve this using a Fourier transformation
Z
d3 p e
A0 (x) = (− ) eix·p
(2π)3 ∆
Z
d3 p 1 ix·p
=e e
(2π)3 p2
Z ∞ Z 1 Z 2π
e 2 1
= 3
dp p d cos θ dφ 2 eipr cos θ
(2π) 0 −1 0 p
Z ∞
e eipr − e−ipr
= dp
(2π)2 0 ipr
Z ∞
e 1 e − e−ipr
ipr
= 2
dp
8π ir −∞ p
Note that the integrand does not diverge at p → 0 since the numera-
tor is ∝ sin(pr) ≈ pr + . . .. So we can introduce a small shift in the
denominator i and simplify
Z ∞ Z ∞
eipr − e−ipr eipr − e−ipr
dp = lim dp
0 p →0 0 p + i
Assuming > 0, we see that the pole is on the negative imaginary
axis. We will now use the residue theorem to calculate the integral:
for eipr we need to close the contour in the upper half-plane to get
exponential suppression of the integrand ∼ e−|p|r , which misses the
pole. For e−ipr , we close the contour in the lower half plane and
obtain
Z ∞ −ipr
−e−ipr −e
dp = (−2πi) Resp→−i = (2πi) e−r (4.93)
−∞ p + i p + i
We therefore find
e 1 e −r
A0 (x) = 2
(2πi) e−r = e (4.94)
8π ir 4πr
and so for → 0, we obtain finally
e
A0 (x) = (4.95)
4πr
which is Coulomb’s law.
4.3.3 A non-linear example: the graviton Lagrangian

Propagators are a useful tool to solve similar types of equations. Let
us assume that our Lagrangian has a non-linear term A3 (or as we
will see: an interaction). Electro-magnetism does not have these
terms and the equations of motion are linear, but there are many
examples in nature, like the gluon or electro-weak theory. Another is
the graviton. A heuristic Lagrangian for the graviton h which is the
fluctuation around a flat background metric gµν (x) = ηµν +hµν (x)+. . .
can be written as See Ex!
1 1
L = − hh + λh3 + J · h (4.96)
2 3
where we have dropped the indices. This defines the non-linear
equations of motion. We solve perturbatively and we start with the
non-interacting theory with λ = 0
1
h0 = J (4.97)

We obtain the first oder correction h = h0 + h1 where h1 = O(λ)
(h0 + h1 ) − λ(h0 + h1 )2 − J = 0 (4.98)
and using the 0th order result above,
h1 = λh20 + O(λ2 ) (4.99)
This means that

1 2 1 1 1
h1 = λ h0 = λ ( J)( J) (4.100)

In summary, the first order solution is

1 1 1 1
h = h0 + h1 = J + λ ( J)( J) + O(λ2 ) (4.101)

This is the so-called Green’s function method. We will call the

1
Π=− (4.102)

the 2-point Green’s function or the propagator. If we define x =
∂ 2 with respect to xµ and
x Π(x, y) = −δ 4 (x − y) (4.103)
Note, Π(x, y) = Π(y, x). We can use Fourier transformations to show

Z
d4 p 1 i(x−y)·p
Π(x, y) = e (4.104)
(2π)4 p2
which we can check
Z
d4 p i(x−y)·p
x Π(x, y) = − e = −δ 4 (x − y) (4.105)
(2π)4
We ignore subtleties with boundary conditions here but we will come
back to this later on.
We can write the field as
Z Z
h(x) = d4 y δ 4 (x − y)h(y) = − d4 z (y Π(x, y))h(y) (4.106)
Z
= − d4 y Π(x, y)y h(y) (4.107)
where we have used integration by parts. We can solve the free field
with λ = 0 equation, by inserting y h0 (y) = J(y) to obtain
Z
h0 (x) = − d4 y Π(x, y)J(y) (4.108)
and the next order as
w h1 (w) = λh0 (w)2 (4.109)

Z Z
= λ d4 y Π(w, y)J(y) d4 z Π(w, z)J(z) (4.110)
Substituting again as above in Eq. (4.107) we find for the perturba-

tive solution
Z Z Z Z
h(x) = − d y Π(x, y)J(y) − λ d s d z d4 y Π(x, s)Π(s, y)Π(s, z) J(y)J(z) + O(λ2 )
4 4 4
which is the meaning of Eq. (4.101).

There is a nice pictorial representation of this solution in the form
of Feynman diagrams. This pictorial representation allows us to
J(z) Figure 4.1: Feynman diagrams in a

⇧(x, y) s classical field theory.
h(x) = J(y) + h(x) +...
J(y)
derive the most general λn expansion for a classical field with the
following classical Feynman rules:
1. Draw a line from a point x to a new point xj .
2. Truncate with a source J(xi ) or branch into two lines and multi-
ply with a factor λ.
3. Repeat the previous step.
4. The final result for h(x) is given by all graphs up to some order
λn with the ends truncated by currents J(xi ) and the lines re-
placed by propagators Π(xi , xj ). All internal points are integrated
over.
As we will see, in quantum field theory the Feynman rules are almost
identical except for ~ → 0 (classical limit), lines can not close
in on themselves.17 In classical general relativity, these diagrams 17
No loops!
describe the post-Newtonian effects of the Sun on Mercury! The
lines represent gravitons and the sources are in this case the Sun.
The first diagram is the well-known Newtonian potential while the
√
second order represents the self-interaction λ ∼ GN which is a
prediction of general relativity.
4.4 The Hamiltonian Formalism

In this course we will not discuss path integrals, and focus instead on
canonical quantization. For this we need the Hamiltonian formalism Figure 4.2: The perihelion precession
of field theory, since canonical commutators are defined between the of Mercury. https://en.wikipedia.org/wiki/
Two- body_problem_in_general_relativity
field and its canonically conjugate momentum.
We first define the conjugate momentum π a (x)
∂L
π a (x) = (4.111)
∂ φ̇a
The Hamilton density is obtained by a Legendre transformation The conjugate momentum π a (x)
should not be confused with the
H = π a (x)φ̇a (x) − L(x) (4.112) total momentum P i which is just
a number characterizing a field
configuration.
As in classical mechanics, we replace φ̇a (x) in favor of π a (x) every-
where. The Hamiltonian is simply
Z
H = d3 x H (4.113)
4.4.1 Example: a real scalar field

For the Lagrangian
1 1 1
L= (∂t φ)2 − (∇φ2 ) − m2 φ2
2 2 2
we obtain π = φ̇ and the Hamiltonian

1 2 1 1
H= π + (∇φ2 ) + m2 φ2
2 2 2
The advantage of the Lagrangian formalism is its explicit Lorentz
invariance. In contrast, the Hamiltonian formalism is not mani-
festly Lorentz invariant, since we have picked a preferred time. The
equations of motions do not look Lorentz invariant
∂H ∂H
∂t φ(x, t) = , ∂t π(x, t) = −
∂π(x, t) ∂φ(x, t)
The physics is of course unchanged (and still Lorentz invariant). In

the canonical formalism, we will have to check at various points that
that the invariance is indeed preserved.
5
Canonical Quantization
5.1 Free fields

You are very familiar with the transition from classical physics
to quantum mechanics by using canonical quantization: it is a
recipe based on the Hamiltonian formalism. We take the generalized
coordinates and their conjugate momenta qa , pa and promote them
to hermitian operators. This recipe is closest to classical mechanics,
when we formulate the dynamics using Poisson brackets {qb , pa } = δba
and morph them into commutation relations between operators
[qa , pb ] = i~ δba
[qa , qb ] = 0
[pa , q b ] = 0
As in the introductory example of the discrete chain, we will do the

same now for the field φa (x) and its momentum conjugate π b (x).
A quantum field operator valued function of space obeys the
commutation relations This is also often confusingly referred
to as second quantization. How-
ever, the fact that there are discrete
[φa (x), π b (y)] = i~ δba δ 3 (x − y) (5.1) modes is a classical phenomenon
(think of the eigenmodes of a string).
[φa (x), φb (y)] = 0 (5.2) The two steps are in fact (1) inter-
a b pret these modes as having energy
[π (x), π (y)] = 0 (5.3)
E = ~ω and (2) quantize each mode
as a harmonic oscillator. In that
sense we are only quantizing once.
We are working in the Schrödinger picture so that the operators
φa (x), π b (x) do not depend on time at all. Note, that we have lost
track of Lorentz invariance since we have separated space and time.
All time dependence in the Schrödinger picture is encapsulated in
the states |ψi which evolve according to the Schrödinger equation
d
i~ |ψi = H |ψi (5.4)
dt
This is exactly the same as in quantum mechanics, we are just
applying the formalism to the fields. Warning: the notation |ψi
is deceiving since the wave function in quantum field theory is a
functional of every possible configuration of the field φa .
In quantum mechanics we would determine the spectrum of the
Hamiltonian H and obtain the time evolution this way. This is too
hard in quantum field theories (most of the time). Clearly, dealing

with an infinite number of degrees of freedom1 complicates the 1
At least one for each point x in
space.
problem somewhat. However for free theories, we can solve the
dynamics by going to a basis where we can separate the dynamics of
the degrees of freedom.2 Typically, these will be Lagrangians which 2
Recall the eigen-modes of the
are quadratic in the fields. discrete chain.
The simplest relativistic field theory is the Klein-Gordon equation

for a real scalar field φ(x, t)
∂µ ∂ µ φ + m2 φ = 0
To decouple the degrees of freedom, we only need to take the Fourier
transformation
Z
d3 p
φ(x, t) = φ(p, t) eix·p (5.5)
(2π)3
Then

∂2 2 2
+ (p + m ) φ(p, t) = 0 (5.6)
∂t2
For each p, φ(p, t) solves the harmonic oscillator equation with a
frequency Reminder:
p 1 2 1
ωp = + p2 + m2 H= p + ω2 q2
2 2
The most general solution is a linear superposition of simple har- and the commutation relations
monic oscillators, each vibrating at a different frequency with a [q, p] = i

different amplitude. To quantize this theory we must quantize these Define
infinitely many harmonic oscillators. You should read Section 2.2.3
r
ω i
a= q + √ p,
again. 2 2ω
r
ω i
a† = q− √ p
2 2ω
5.1.1 The free scalar field
which we can invert to
We write the field and its conjugate momentum as a linear sum of 1
q = √ (a + a† ), (5.7)
infinitely many creation a†p and annihilation ap operators labeled by 2ω
r
ω
their 3-momentum p p = −i (a − a† ) (5.8)
2
Z Substituting the expressions we find
d3 p 1
φ(x) = 3
p ap eix·p + a†p e−ix·p (5.9) [a, a† ] = 1
(2π) 2ωp
and H = ω(a† a + 21 ).
and for the conjugate momentum3 3

Inspired by Eq. (2.51) or Eq. (5.8).
You can check that this will lead to
Z r the correct commutation relations for
d3 p ωp the fields consistent with the ladder
π(x) = 3
(−i) ap eix·p − a†p e−ix·p (5.10) operators.
(2π) 2
Later, when we will consider time-

As in the case of the quantized string, the commutation rela- dependent field operators (Heisen-
berg picture), you will see that this is
tions for φ(x) and π(x) of Eq. (5.1) are equivalent to commutation consistent with
relations for ap and a†p . Quantum field theory is just quantum me- ∂L
π= = ∂t φ
chanics with an infinite number of harmonic oscillators. ∂(∂t φ)
[ap , a†q ] = (2π)3 δ (3) (p − q) (5.11)

[ap , aq ] = [a†p , a†q ] =0 (5.12)
We will show this one way

 
Z 3 Z 3 r
d p d q −i ωq  
[φ(x), π(y)] = − [ap , a†q ] eix·p−iy·q + [a†p , aq ] e−ix·p+iy·q 
(2π)3 (2π)3 2 ωp | {z } | {z }
(2π)3 δ 3 (p−q) −(2π)3 δ 3 (p−q)
Z
d3 p −i h i(x−y)·p −i(x−y)·p
i
= −e − e
(2π)3 2
Z
d3 p i(x−y)·p
=i e
(2π)3
= i δ (3) (x − y)
We obtain for the Hamiltonian in terms of ap , a†q Ex: Assuming canonical com-
Z mutation relations for the fields,
1 show that the creation and
H= d3 x π(x)2 + (∇φ(x))2 + m2 φ(x)2
2 annihilation operators satisfy
Z r [ap , a†q ] = (2π)3 δ (3) (p − q).
1 3
3
d p d q 3
ωp ωq ix·p † −ix·p
ix·q † −ix·q

= d x 3 3
(−1) ap e − ap e aq e − aq e
2 (2π) (2π) 4
1
+p ip ap eix·p − ip a†p e−ix·p iq aq eix·q − iq a†q e−ix·q
4ωp ωq
1
+ m2 p ap eix·p + a†p e−ix·p aq eix·q + a†q e−ix·q
4ωp ωq
We will now collect the terms, integrate over d3 x and replace
e−ix·(p+p) → δ 3 (p − p) and e−ix·(p−p) → δ 3 (p + p) which allows us to
remove one d3 q integral.
Z
d3 p 1 h † †
i
H= ap a −p + a p a−p (−ωp2 + p2 + m2 )
(2π)3 4ωp
1
+ ap a†p + a†p ap (ωp2 + p2 + m2 )
4ωp
The first line vanishes because of ωp2 = p2 + m2 and we obtain
Z
1 d3 p
H= 3
ωp ap a†p + a†p ap (5.13)
2 (2π)
Z
d3 p † 1 3 (3)
= ω p a a
p p + (2π) δ (0) (5.14)
(2π)3 2
What is this!? We clearly have to work to interpret this properly.
There are two infinities: the delta-function has an infinite spike at 0
and the integral over ωp also diverges.4 4
The latter infinity is closely related
to the infinity of the specific heat
mentioned in the introduction.
5.1.2 Infinities
We will separate the infinities into two classes: infrared diver-
gencies which are caused by assuming infinitely large scales (or
arbitrarily low energies) and ultraviolet divergencies which come
from extrapolating to infinitely small scales (or arbitrarily high
energies).
We will better understand the problem by discussing the ground
state or the vacuum |0i. We define |0i as the state being annihi-
lated by all the ap
ap |0i ≡ 0 for all p (5.15)

We obtain for the ground-state energy

Z
d3 p † 1 3 (3)
H|0i = E0 |0i = ωp ap ap + (2π) δ (0) |0i (5.16)
(2π)3 2
Z 3
d p 1
= 3
ωp (2π)3 δ (3) (0)|0i (5.17)
(2π) 2
= ”∞2 ” |0i (5.18)
This is common for quantum field theories: infinities pop up left and
right. They tell us that we might be asking the wrong question or
are doing something wrong.
The infrared divergence or long-distance divergence can be taken
care of by putting the field in a box of length L or volume V = L3
and assuming periodic boundary conditions. We obtain
Z L Z L Z L
2 2 2
(2π)3 δ (3) (0) = lim dx dy dz eix·p =V (5.19)
L→∞ −L −L −L
2 2 2 p=0
This tells us: we should be calculating the energy density E0

instead of the total energy! We therefore divide by V to find
Z Z ∞ p
E d3 p 1 4π 1
E0 = = ωp = dp p2 m2 + p2 → ∞ (5.20)
V (2π)3 2 (2π)3 2 0
which is however still infinite.5 5

Although a little less infinite:
”∞” |0i...
The ultraviolet divergence or short-distance divergence can be
understood by recognizing this as the integral over all ground state
energies of the harmonic oscillators. In the limit that |p| → ∞ we
find E0 → ∞. This divergence is caused by us assuming that our
quantum field theory is valid down to the smallest distances (or
highest energies)6 . This is certainly not true. We need to cut-off the 6
Even above the Planck scale!
integral at high energy (or high momentum) to include the fact that Compare to the discrete chain,
where this would correspond to
we are ignorant above a certain scale. extrapolating to distances much
smaller than the average distance
We could introduce an explicit cut-off and stop integrating at some between the mass points
energy Λ since we do not want to extrapolate beyond this point.
Z Λ p
1
E0 (Λ) = dp p2 m2 + p2 ∝ aΛ4 + . . . (5.21)
4π 2 0
which is called regularization.7 7

Note, that this is not a Lorentz-
We can then renormalize the vaccuum energy to its observed invariant regulator: Lorentz invari-
ance is broken when we impose a
value by realizing that we could have added a constant V0 to the cut-off on the spatial momentum
Lagrangian only. To meaningfully evaluate the
zero-point energy density, you must
use a regularization scheme that
Lnew = L − V0 (5.22) does not break Lorentz invariance.
A possible choice is dimensional
This constant will not affect the equations of motion and we usually regularization which will be covered
in more advanced QFT courses.
just drop it. V0 is a bare parameter which we choose to cancel the
divergence of zero-point energies
V0 = −E0 (Λ) + χ (5.23)

this diverges as we take the cut-off to infinity V0 = −aΛ4 + . . .. But

now the observed energy-density is finite even for Λ → ∞ as
1
H |0i = (E0 (Λ) + V0 )|0i (5.24)
V
= (E0 (Λ) − E0 (Λ) + χ)|0i (5.25)
= Eobserved |0i (5.26)
the step is called renormalisation. Note that the finite piece

χ = Eobserved is completely arbitrary. It must be determined ex-
perimentally by measuring an observable – in this case the vacuum
energy.8 Since Λ only appears in the ”bare” Lagrangian and interme- 8
Vacuum energy is a meaningful
diate steps of the calculation, but not in physical observables, we can observable only in the presence of
gravity – see below.
remove it at the end.
5.1.3 Normal ordering

Alternatively, we can follow a more practical approach called nor-
mal ordering. Since we are only interested in energy-differences9 , 9
Again, as long as we ignore gravity,
the practical approach is good enough. We will just subtract off the which couples directly to the energy
momentum tensor.
infinity and redefine our Hamiltonian as
Z
d3 p
H0 = ωp a†p ap (5.27)
(2π)3
and now
H 0 |0i = 0 (5.28)
If you are a bit taken aback by the lack of rigor implied by an ar-
bitrary redefinition of our Hamilton, note the following: there is
an inherent ambiguity of ordering transitioning from classical to
quantum. If we had defined the classical Hamiltonian as
0 1 1 ω
Hclassical = (ωq − ip)(ωq + ip) = p2 + q 2 = Hclassical
2 2 2
0
which is classically equivalent Hclassical = Hclassical . We would
naturally get after quantization10 10
Compare to the definition
r
ω i
0
HQM = ω a† a a=
2
q + √ p,
2ω
r
ω i
a† = q− √ p=
This method of exploiting the order ambiguity is called normal 2 2ω
ordering: we define it as the ordering in which all the creation .
operators stand to the left of the annihilation operators.
(ap a†q )normal ordered = : ap a†q : = a†q ap

(a†p aq )normal ordered = : a†p aq : = a†p aq
In general, we define a string of operators as normal ordered
: φa (x1 )φb (x2 ) . . . φf (xn ) : (5.29)

The Cosmological Constant : As mentioned before, gravity can indeed measure energy differences. Gravity
sees everything! We would expect the sum of a all zero-point energies to contribute to the cosmological constant
Λ = E0 /V and appear on the right side of the Einstein-Hilbert equations
Rµν − 12 R gµν = −8πGTµν + Λ gµν ,
Cosmological observations tell us that ca. 70% of the energy density of the universe today is in a term that looks
like the cosmological constant with
Λ ∼ (10−3 eV)4
This is many orders of magnitude smaller than all the other scales in particle physics (the SM is valid to ∼ 200
GeV = 2 · 1011 eV). Why don’t zero-point energies contribute to Λ? What cancels them to such accuracy? We do
not know the answer. In fact, we don’t even know how to ask the right question.
if all the annihilation operators are placed to the right. Another

consequence of normal-ordering is that vacuum matrix elements
vanish
h0| : φa (x1 )φb (x2 ) . . . φf (xn ) : |0i
and the only normal-ordered expressions that do not vanish are

c-number functions11 which satisfy 11
The term c-number (or classical
number) is an old nomenclature
h0| : f (x) : |0i = f (x) used by Paul Dirac which refers to
real and complex numbers. It is
Our Hamiltonian then has the normal-ordered form used to distinguish from operators
Z (q-numbers or quantum numbers) in
d3 p quantum mechanics.
:H: = ωp a†p ap (5.30)
(2π)3
5.2 Particles
What are the excitations of the field with respect to the vacuum |0i?
We know that
[H, a†p ] = ωp a†p (5.31)

[H, ap ] = −ωp ap (5.32)
As in the case of the simple harmonic oscillator, we can construct Ex: Verify the relation
energy eigenstates by applying the creation operators a†p on the [H, a†p ] = ωp a†p
vacuum |0i. We define
1
a†p |0i = p |pi (5.33)
2Ep
p
The factor 2Ep constitutes a relativistic normalization – see
√
below.12 The energy of this state is 12
The 2 is for convenience only.
Z
d3 p p
: H : |qi = ωp a†p ap 2Eq a†q |0i
(2π)3
Z
p d3 p † † 3 (3)
= 2Eq ωp ap a q a p + (2π) δ (p − q) |0i
(2π)3
p
= ωq 2Eq a†q |0i
= ωq |qi
with
ωq2 = q 2 + m2 (5.34)
This is the relativistic dispersion relation (or relativistic Pythagoras)

for a particle of mass m and momentum q. We therefore interpret
this as a momentum eigenstate of a single particle of mass m.
Its energy is Eq2 = q 2 + m2 and we can interchange Eq and ωq .
What about the other quantum numbers? The total momen-
tum P of a classical field configuration which we have derived in
Eq. (4.73), can be turned into a (normal ordered) operator
Z
: P : = − d3 x : π(x)∇φ(x) :
Z r
d3 p d3 q ωp
= − d3 x 3 3
(−i) : ap eix·p − a†p e−ix·p ∇ aq eix·q + a†q e−ix·q :
(2π) (2π) 4ωq
Z r
3
d p d q 3
ωp
= − d3 x 3 3
(−i) : ap eix·p − a†p e−ix·p iq aq eix·q − a†q e−ix·q :
(2π) (2π) 4ωq
Z 3 h i
d p 1 † †
= (−1) p ap a−p + ap a −p + p a†p ap
(2π)3 2
The first term vanishes because the term in the brackets is sym-
metric under p → −p which together with p makes this part of the
integrand anti-symmetric and therefore vanishes after integrating
over a symmetric interval. We find
Z
d3 p
:P : = p a† ap (5.35)
(2π)3 p
In summary we find
Z
d3 p µ †
: Pµ : = p ap ap (5.36)
(2π)3
We can do the same for the classical expression of the angular

momentum13 to obtain the operator 13
See Ex and solution!
Z
J i = εijk d3 x(J 0 )jk (5.37)
Acting on a one-particle state with zero-momentum, we find
J i |p = 0i = 0
which we understand as telling us that the particle does not carry

internal angular momentum. Quantizing a scalar field gives rise to a
spin-0 particle!
5.2.1 Multi-particle states and Bose statistics

Similarly, we can see that the state
a†p1 a†p2 . . . a†pn |0i (5.38)

has momentum
p1 + p2 + . . . + pn (5.39)
We interpret this as an n-particle state. Note that these are not localized
in space; a†pn creates a momentum
|p1 , . . . , pn i = N a†p1 a†p2 . . . a†pn |0i (5.40) eigenstate.
where N is a normalization. Since the creation operators commute

[a†p1 , a†p2 ] = 0, these states are symmetric under exchange of any
two particles, e.g.
|p1 , p2 i = |p2 , p1 i (5.41)
and we know that these particles are bosons. We can construct the
Fock space as the Hilbert space spanned by all possible combina-
tions of the a†pi
|0i, a†p1 |0i, a†p1 a†p2 |0i, a†p1 a†p2 a†p3 |0i, . . . (5.42)
As we have seen in the introduction, this is simply the direct sum of

the n-particle Hilbert spaces
|Ψi = |Ψi1 ⊕ |Ψi2 ⊕ . . . (5.43)
We can count the number of particles in a given state in Fock space

using the number operator N defined as
Z
d3 p †
N= a ap (5.44)
(2π)3 p
which satisfies
N |p1 , . . . , pn i = n |p1 , . . . , pn i (5.45)
and it commutes with Hamiltonian,
[H, N ] = 0
which it should since in our free theory, particle number is conserved.

This will no longer hold once we include interactions (e.g. L ∼ φ3 ).
Interactions can destroy and annihilate particles, taking us between
different sectors in Fock space.
5.3 Smeared operator valued distributions

As already mentioned, |pi is not really a particle and not localized in
space but |pi is a momentum eigenstate.
Crucially, neither the operators φ(x) nor a†p are good operators
acting on the Fock space. 14 14
This is in direct analogy with the
position and momentum eigenstates
hp|pi = 2Ep h0|ap a†p |0i of quantum mechanics which are not
good elements of the Hilbert space
= 2Ep h0|a†p ap + (2π)3 δ (3) (p − p)|0i because they are not normalizable.
They normalize to delta-functions.
= (2π)3 2Ep δ (3) (0) h0|0i
| {z }
=1
and so Recall, ap |0i = 0 and therefore by

hermitian conjugation h0|a†p = 0.
hp|pi = 2Ep (2π)3 δ (3) (0) (5.46)
Similarly
h0|φ(x)φ(x)|0i = hx|xi = δ (3) (0) (5.47)
These are therefore operator valued distributions, rather than

functions. Even though we obtain a finite vacuum expectation value
h0|φ(x)|0i = 0 (5.48)
The quadratic fluctuations of the operator at a point are infinite

h0|φ(x)φ(x)|0i = ∞. We can obtain finite results by smearing the
distributions over space (or momentum) using test functions,
Z
φf (x) = d3 y φ(x)f (x − y)
with e.g. the (normalized) Gaussian smearing

1 x2
f (x) = 3 exp −
(a2 π) 2 a2
We can obtain h0|φf (x)φf (x)|0i by observing that of the four terms
after expanding φ(x)2 only the one proportional to h0|ap a†q |0i ∼
δ 3 (p − q) will be non-zero. We have
Z Z 2
1 d3 p 3 (x − y)2
h0|φf (x)φf (x)|0i = d y exp − + ip(x − y)
8(aπ)6 2Ep a2
The Gaussian integral can be easily calculated by completing the

squares15 and we get 15
Or by using Mathematica as the
author did.
Z 3
1 d p −p2 a2 /2
h0|φf (x)φf (x)|0i = 3
e
(2π) 2Ep
Z ∞
1 p2 2 2
= 2
dp p e−p a /2
(2π) 0 p2 + m2
For a 1/m we can ignore the m-dependence16 to obtain 16

pDefine q = ap and approximate
q 2 + (am)2 ≈ q
1 1
h0|φf (x)φf (x)|0i =
4π 2 a2
We see that the fluctuations diverge without bounds in the limit of
a → 0.
5.3.1 Dimensional analysis and Coleman theorem

We could have in fact guessed the result for the fluctuations easily.
In the limit a → 0, there is no scale17 other than a. Similarly, we 17
As a → 0, we can ignore the only
know by counting the dimensions in the expansion of h0|φ(x)φ(y)|0i other scale (m) since 1/a m.
and realizing again that the only scale is |x − y|, we have to have Up to factors of order one, like π etc.
1
h0|φ(x)φ(y)|0i ∼
|x − y|2
This dimensional analysis is equivalent to saying that the field has

dimensions of inverse length. We could have already realized that a
scalar field in 3 + 1 Dimensions has canonical mass dimension18 1, or 18
The mass dimension of a field ψ
[φ] = 1 by looking at the kinetic term of the action is defined to be the exponent
Z [ψ] = n ψ ∼ (mass)n
S = d4 x ∂µ φ(x)∂ µ φ(x) + . . . which means that ψ has dimension
(mass)n ∼ (energy)n ∼ (length)−n .
Since the action is dimensionless [S] = [~] = 0 and since [d4 x] = −4

and [∂µ ] = 1, we find
[φ(x)] = 1
So the growth of the fluctuations with |x − y|−2 is really just dimen-

sional analysis. We can obtain similar results in any dimension. In D
space-time dimensions
Z
S = dD x ∂µ φ(x)∂ µ φ(x) + . . .
from which we get

D−2
[φ(x)] =
2
and so in D−dimensions we obtain
1
h0|φ(x)φ(y)|0iD ∼
|x − y|D−2
We immediately see an interesting feature by going to lower dimen-

sions. The divergence of the fluctuations is less rapid. In particular
for D = 2, that is in 1 + 1 space-time dimensions, the power vanishes.
This is really a log and we get19 19
See Exercise 5.2.

1
h0|φ(x)φ(y)|0iD=2 ∼ ln
|x − y|
The fluctuations diverge logarithmically slow as |x − y| → 0. The

flip side of this is that also for large |x − y|, the fluctuations diverge
logarithmically.
So long-range fluctuations in D ≤ 2 are then actually very
important now and if you follow this through you find that these
fluctuations are large enough to prevent the formation of long-
range order.20 This is a form of the Coleman-Mermin-Wagner 20
which in particle physics language
theorem21 which holds in quantum field theory and in statistical is called spontaneous symmetry
breaking.
mechanics. They hold in both since we can get statistical mechanics 21
Continuous symmetries cannot be
from quantum field theory by analytic continuation – but this is for spontaneously broken in dimensions
D ≤ 2, https://en.wikipedia.org/wiki/
another lecture.
Mermin- Wagner_theorem
5.4 Some technicalities: relativistic normalization

The vacuum |0i is defined as normalized h0|0i = 1. Our one-particle
p
|pi = 2Ep a†p |0i states then have
p
hp|qi = (2π)3 4Ep Eq δ (3) (p − q)
This is a Lorentz invariant normalization. How do we show this? In

a quantum theory, we are always looking for unitary representa-
tions of our symmetries, such that
pµ → p̃µ = Λµ ν pν |pi → |p̃i = U (Λ)|pi
and therefore leave the normalizations unchanged
hp|pi = hp̃|p̃i
but since p is a three-vector, this is not at all clear. The problem is

that spatial volumes are not invariant under boosts V → V /γ! We
show that we have a Lorentz invariant normalization by proving that
another object is Lorentz invariant.
Claim: We claim that

Z
d3 p
(5.49)
2Ep
is the Lorentz invariant measure.
R
Proof: We show this by using that d4 p is obviously Lorentz
invariant and by using that
pµ pµ = m2 p20 = Ep2 = p2 + m2
is a Lorentz invariant statement. We choose the positive branch p0 =

+Ep , but Lorentz transformations (the ones which are continuously
connected to the 1) also leave this choice invariant. We conclude
that the following combination must be Lorentz invariant
Z Z 3
4 2 2 2
d p
d p δ(p0 − p − m ) =
p0 >0 2p0 p0 =Ep
which completes the proof. Now we know that the Lorentz-invariant Note, that
δ (3) -function for 3 vectors is: X δ(x − xi )
δ(g(x)) =
i
|g 0 (xi )|
2Ep δ (3) (p − q)
where the sum extends over all roots
of g(x) which are assumed to be
this holds because
simple. For example
Z
d3 p 1 h
2Ep δ (3) (p − q) = 1
i
δ x2 − α 2 =

δ (x + α)+δ (x − α) .
2Ep 2|α|
This finally explains our normalization for the momentum states,

p
|pi = 2Ep a†p |0i
These states now satisfy
hq|pi = (2π)3 2Ep δ (3) (p − q) (5.50)
The identity on one-particle states is therefore

Z
d3 p 1
1= |pihp|
(2π)3 2Ep
In the following, we will only use relativistically invariant normaliza-
tions.
5.4.1 Meaning of the state φ(x)|0i

Let us consider the meaning of φ(x)|0i. Using the expansion
Eq. (5.9) and Eq. (5.33), we obtain
Z
d3 p 1 −ip·x
φ(x)|0i = e |pi
(2π)3 2Ep
which is a linear superposition of single particle states with well-

defined momentum. This should remind you of the non-relativistic
expression for the eigen-state of position |xi, except for the factor
of 1/2Ep . We observe that for non-relativistic p the factor is nearly
a constant 1/2Ep ≈ 1/2m.
We will therefore put forward the same interpretation: the ope-
rator φ(x) acting on the vacuum creates a particle at position x.
Additional evidence comes from computing
Z h ip
d3 p0 1 ix·p0 † −ix·p0
h0|φ(x)|pi = h0| p a p 0 e + ap 0 e 2Ep a†p |0i
(2π)3 2Ep0
= eix·p (5.51)
This can be interpreted as the position-space representation of the

single-particle wavefunction of the state |pi, exactly like in non-
relativistic quantum mechanics hx|pi ∝ eix·p is the wavefunction of
the state |pi.
5.5 Complex scalar fields
Let us consider a complex scalar field ψ(x) with
L = ∂µ ψ ∗ ∂ µ ψ − m2 ψ ∗ ψ
with the equation of motion
∂µ ∂ µ ψ + m2 ψ = 0
and the same equation of motion for ψ ∗ . We expand the complex

field operator in the Schroedinger picture
Z
d3 p 1
ψ(x) = 3
p bp eix·p + c†p e−ix·p (5.52)
(2π) 2Ep
Z
d3 p 1 † −ix·p
ψ † (x) = 3
p bp e + cp eix·p (5.53)
(2π) 2Ep
The classical field is not real: the corresponding quantum field ψ

is not hermitian. This explains the appearance of the different
operators b and c† . The conjugate momentum is
∂L
π= = ∂t ψ ∗
∂∂t ψ
and22 22
Recall, that we define π(x) in
analogy with the result for the
harmonic oscillator and that it
satisfies the canonical commutation
relations.
Z r
d3 p Ep
π(x) = 3
(+i) b†p e−ix·p − cp eix·p (5.54)
(2π) 2
Z r
3
d p Ep
π † (x) = (−i) bp eix·p − c†p e−ix·p (5.55)
(2π)3 2
with the commutation relations
[ψ(x), π(y)] = iδ (3) (x − y)
all other commutators are vanishing
0 = [ψ(x), π(y)† ] = [ψ(x), ψ(y)] = [π(x), π(y)] = . . .
You can check these commutation relations imply
[bp , b†q ] = (2π)3 δ (3) (p − q) (5.56)

[cp , c†q ] 3 (3)
= (2π) δ (p − q) (5.57)
and the rest vanishes
0 = [bp , bq ] = [c†p , c†q ] = [bp , cq ] = [bp , c†q ]
naturally including their hermitian conjugates, e.g. [cp , cq ] = 0.
Interpretation: The quantization of a complex scalar field

gives rise to two types of creation operators b†q and c†q . These are
particles and anti-particles. Compare to a real scalar field.
The complex scalar field has an internal U (1) symmetry, which as There is only one type of particle: it
is its own anti-particle.
we saw in Sec. 4.2.2 gives rise to a conserved current
j µ = i(∂ µ ψ ∗ )ψ − iψ ∗ (∂ µ ψ) (5.58)
and therefore a conserved quantum number

Z Z
Q = i d x (ψ̇ ψ − ψ ψ̇) = i d3 x (πψ − ψ ∗ π ∗ )
3 ∗ ∗
(5.59)
The normal ordered quantum operator is

Z
: Q : = i d3 x : (πψ − ψ ∗ π ∗ ) :
Z Z Z s
d3
p d 3
q Eq
= − d3 x : bp eix·p + c†p e−ix·p b†q e−ix·q − cq eix·q : + h.c.
(2π)3 (2π)3 4Ep
Z
d3 p n † h i h io
=− 3
bp bp − c†p cp − b−p cp − b†−p c†p + b−p cp − b†−p c†p
(2π)
where we’ve used that hermitian conjugation of the first bracket

leaves it unchanged with normal-ordering. The second and third
bracket cancel. We finally obtain for the normal-ordered charge
operator
Z
d3 p †
:Q: = c cp − b†p bp = Nc − Nb (5.60)
(2π)3 p
which counts the number of anti-particles (created by c†p ) minus the

number of particles (created by b†p ). We know that23 23
Ex: Show this!
[H, Q] = 0
which tells us that Q is a conserved quantity. We saw that in the

free field theory this is trivially true with Eq. (5.44), since [N, H] =
0. Later, we will see that in interacting field theories, new particles
can be created, e.g. L ∼ φ3 can branch a particle into two new ones:
Nc and Nb will not be individually conserved. However, Q = Nc − Nb
survives as a conserved quantity.
5.6 Time dependent operators: the Heisenberg picture
We have been working in the Schroedinger picture, where the op-

erators depend on space but not on time. We will move to the
Heisenberg picture and make the operators time-dependent in
the usual way
φH (x) = φH (x, t) = eiHt φS (x) e−iHt
and In the Schroedinger picture, states

iHt −iHt evolve in time
π(x) = e π(x) e
d
i~ |p(t)iS = H |p(t)iS = Ep |p(t)iS
dt
In general
which means
OH (t) = eiHt OS e−iHt
|p(t)iS = e−iEp t |p(0)iS
and so States in the Heisenberg picture are

d related by the unitary transformation
OH (t) = i[H, OH (t)]
dt |ψ(t)iS = e−iHt |ψiH
In field theory we will drop the subscripts ”S ” or ”H ” and indicate and similarly operators
using the arguments if we are in the Schrödinger picture φ(x) or in OS = e−iHt OH (t) e+iHt
the Heisenberg picture φ(x, t) = φ(x). This means that the matrix-elements
We assume that say at t = 0 the two pictures coincide and we find are the same
that Eq. (5.1) become equal-time commutation relations H hψ|OH (t)|φiH = S hψ(t)|OS |φ(t)iS
We will assume that at t = 0 both

pictures coincide.
[φa (x, t), π b (y, t)] = i~ δba δ 3 (x − y) (5.61)
|ψ(0)iS = |ψiH
[φa (x, t), φb (y, t)] = 0 (5.62) OS = OH (0)
[π b (x, t), π b (y, t)] = 0 (5.63)
We can now discuss the time-evolution of the operator φ(x, t), e.g.24 24
We use
[A2 , B] = A[A, B] + [A, B]A = 2f A
∂t φ(y) = i[H(t), φ]
Z if the commutator is just a c-valued
i function
= d3 x π(x, t)2 + (∇φ(x, t))2 + m2 φ(x, t)2 , φ(y) [A, B] = f
2
Z e.g. f = δ(x − y).
= i d3 x π(x)(−i)δ (3) (x − y)
= π(y)
And
Z
i
∂t π(y) = i[H, π] = d3 x π(x)2 + (∇x φ(x))2 + m2 φ(x)2 , π(y)
2
Z
= i d3 x ∇x [φ(x), π(y)]∇x φ(x) + im2 φ(x)δ 3 (x − y)
Z
=− d x (∇x δ (x − y))∇x φ(x) − m2 φ(y)
3 3
= ∇2y φ(y) − m2 φ(y)
which when combined with ∂t φ(y) = π(y) gives the Klein-Gordon

equation
∂µ ∂ µ φ + m2 φ = 0
We find the important result: Field operators satisfy the same

equation as the classical field. This starts to resemble a relativistic
theory.
How does the Heisenberg field operator expansion look like?
We want to transform
φS (x) → φ(x) = eiHt φS (x) e−iHt
and therefore need to transform the creation and annihilation opera-

tors in the expansion
eiHt ap e−iHt = e−iEp t ap (5.64)

eiHt a†p e−iHt = e+iEp t a†p (5.65)
which we can prove using the Baker-Campbell-Hausdorff relation

X∞
1
eX Y e−X = [X, Y ]m (5.66)
m=0
m!
with the general commutator
[[X, Y ]m = [X, [X, Y ]m−1 ] and [X, Y ]0 = Y
We know that
[H, ap ] = −Ep ap , and so [iHt, ap ]m = (−iEp t)m ap
which is all we need to show

X∞
1
eiHt ap e−iHt = [iHt, ap ]m = e−iEp t ap
m=0
m!
An analogous calculation (or hermitian conjugation) allows us to

obtain the Heisenberg transformation of a†p .
Plugging this into Eq. (5.9) means that the field operator of the real
scalar in the Heisenberg picture φ(x, t) = eiHt φ(x) e−iHt has the
following expansion
Z
d3 p 1
φ(x) = 3
p ap e−ip·x + a†p eip·x (5.67)
(2π) 2Ep
where the exponent has the opposite sign, because we have used the
Lorentz-vectors
p · x = pµ xµ = Ep t − p · x
You can easily check that Eq. (5.67) indeed satisfies the Klein-
Gordon equation with Ep2 = p2 + m2 .
For completeness, we show the straight-forward result for the
complex scalar field of Eq. (5.52) in the Heisenberg picture
Z
d3 p 1
ψ(x) = 3
p bp e−ip·x + c†p eip·x (5.68)
(2π) 2Ep
Z
d3 p 1 † ip·x
ψ † (x) = 3
p bp e + cp e−ip·x (5.69)
(2π) 2Ep
5.7 Causal quantum fields

We already know from Eq. (5.61) that φ(x) and π(x) satisfy equal
time commutation relations, e.g.
[φ(x, t), π(y, t)] = i δ (3) (x − y) (5.70)
We would like to study now: how does the commutator look like at
arbitrary space-time distances? This brings us back to the discussion
of causality in the introduction. Recall that we must require that
space-like separated events (x − y)2 < 0 do not overlap. Or in other
words, we should be able to find a basis in which matrix elements of
operators can be brought into diagonal form (for x and y space-like).
This is equivalent to requiring that space-like separated operators
commute This is the very important causality
condition.
[O1 (x), O2 (y)] = 0 if (x − y)2 < 0 (5.71)
which ensures that a measurement at x will not affect a measure-

ment at y if they are not causally connected. Let us check if we
satisfy this property. We define
∆(x − y) ≡ [φ(x), φ(y)] (5.72)
By direct substitution we find that in the free theory the ∆(x − y) is

simply a c-number function with the integral expression25 25
Thanks to Nepomuk Ritz for
Z sending in the solution!
d3 p 1 h −i(x−y)·p i
∆(x − y) = 3
e − ei(x−y)·p
(2π) 2Ep
Plugging the expression for φ(x) from Eq. (5.67) into the definition of ∆(x − y)
yields
d3 p d3 q 1
Z Z h i
∆(x − y) ≡ [φ(x), φ(y)] = 3 3
p ap e−ip·x + a†p eip·x , aq e−iq·y + a†q eiq·y
(2π) (2π) 4Ep Eq
d3 p d3 q 1
Z Z n h i
−ip·x−iq·y †
= [a p , a q ] e + ap , a q e−ip·x+iq·y
(2π)3 (2π)3 4Ep Eq
p
h i h i o
+ a†p , aq eip·x−iq·y + a†p , a†q eip·x+iq·y .
Using the commutation relations

h i
[ap , aq ] = 0 ap , a†q = (2π)3 δ (3) (p − q)
h i h i
a†p , a†q = 0 a†p , aq = −(2π)3 δ (3) (q − p).
this expression simplifies to
d3 p 1 h −i(x−y)·p
Z i
∆(x − y) = 3
e − ei(x−y)·p
(2π) 2Ep
which is the desired result.
Properties of ∆(x − y):
1. ∆(x − y) is Lorentz-invariant. The exponentials are functions of

contracted four-vectors and we have just shown the integration
R
measure d3 p/(2Ep ) to be Lorentz-invariant.
2. For time-like distances ∆(x − y) does not vanish: if we take

z µ = xµ − y µ = (t, 0, 0, 0)µ , we obtain
Z
1
∆(z) = [φ(x, t), φ(x, 0)] ∼ d3 p sin(Ep t)
2Ep
which is non-vanishing.
3. For space-like distances ∆(x − y) vanishes. We first show it for

a space-like distance at equal times
(x − y)2 = (t − t)2 − (x − y)2 = −(x − y)2 < 0
We obtain in this case

Z h i
d3 p 1 i(x−y)·p −i(x−y)·p
[φ(x, t), φ(y, t)] = p e − e
(2π)3 2 p2 + m2
This vanishes. We can e.g. replace p → −p in the second term

since it is just an integration variable and we are integrating over
a symmetric interval, which changes the sign in the exponent and
causes the sum to identically vanish. Since ∆(x − y) is Lorentz-
invariant, it only depends on (x − y)2 and must therefore vanish for
all (x − y)2 < 0 (and not only for equal times).
We conclude that we have defined a theory of causal quantum

fields, which solves this problem of relativistic quantum mechanics
as described in the introduction. This property will continue to hold
in interacting theories. Even though [φ(x), φ(y)] will not be
a c-function anymore, but will be
operator valued.
5.7.1 Propagators in quantum field theory
Consider a purely space-like distance:
We have already seen the usefulness of the Green’s function method x0 − y 0 = 0 and x − y = r and repeat
in Sec. 4.3. We will encounter a similar object in quantum field steps after Eq. (4.92)
theories. Continuing with the causality problem, we could ask the d3 p 1 −ip·(x−y)
Z
D(x − y) = e
(2π)3 2Ep
question: what is the amplitude to create a particle at x and observe Z ∞
2π p2 eipr − e−ipr
it at y while leaving the vacuum |0i undisturbed? We calculate = dp p
(2π)3 0 2
2 p +m 2 ipr
Z ∞ ipr
−i pe
= dp p
2(2π)2 r −∞ p2 + m2
The integrand can be considered a
complex function with branch-cuts
on the imaginary axis (from the
√
. . .) starting at ±im. We evaluate
Z
d3 p d3 q 1
h0|φ(x)φ(y)|0i = p h0|ap a†q |0i e−ip·x+iq·y
(2π)3 (2π)3 4Ep Eq
Z
d3 p 1 −ip·(x−y)
= e
(2π)3 2Ep
≡ D(x − y) (5.74)
We call D(x − y) the propagator. For space-like separations (x −

y)2 < 0, we can show, see Eq. (5.73), that
D(x − y) ∼ e−m|x−y|
Although it decays exponentially quickly outside the light-cone, it

is non-vanishing! The quantum field leaks out of the light-cone, vio-
lating causality? How do we reconcile this with the fact, Eq. (5.71),
that all space-like separated measurements commute? We can write
the commutator Eq. (5.72) as branch-cut
[φ(x), φ(y)] = D(x − y) − D(y − x) = 0 if (x − y)2 < 0 (5.75)
cut
Iradial
Ibranch
Let us describe what this equation means: we know that for space-
like separation (x − y)2 < 0 of x and y, there is no Lorentz invariant
ordering of events: if a particle can travel from x −→ y, it can just +im
IR
as easily travel from y −→ x. In any measurement, the amplitudes
im
for the two processes cancel.
Let us consider this for a complex scalar-field, which has distinct
Figure 5.1: Since the integrand
particle and anti-particle excitations, see Eq. (5.52). When the has no poles or branch-cuts inside
complex scalar field ψ(x) is quantized, this contour, the integral has to
vanish. The integrand on the radial
• ψ(x) will create positive charged particles and destroy negatively boundary vanishes Iradial → 0. We
can therefore ’push the contour’ to
charged ones wrap around the branch cut. We
know IR + Ibranch−cut + Iradial = 0
• ψ † (x) will create negative charged particles and destroy positively and therefore IR = −Ibranch−cut
charged ones
The commutator [ψ(x), ψ † (y)] will be generally non-zero but most

delicately cancel outside of the light-cone to preserve causality.
The interpretation of the two terms in Eq. (5.75) now has charges
attached. Since
ψ(x)|0i ∼ (. . .) c†p |0i
produces an anti-particle at x and
ψ † (y)|0i ∼ (. . .) b†p |0i
produces a particle at y, we can interpret the commutator
[ψ(x), ψ † (y)] = D(x − y) − D(y − x)

= h0|ψ(x)ψ † (y)|0i − h0|ψ † (y)ψ(x)|0i
= ”particle y → x” − ”anti-particle x → y”
In oder for these two processes to cancel, both particles must exist
and they must have the same mass.
In quantum field theory, causality requires that every particle has

a corresponding anti-particle with the same mass and opposite
quantum numbers (here it would be electric charge). What is the
anti-particle of the real Klein-Gordon field? The particle of the real
field is its own anti-particle.26 26
And by the property of anti-
particles of having opposite quantum
numbers, we found an alternative
5.8 The Stückelberg-Feynman propagator way to show that real scalar fields
are neutral.
We will now discover one of the most important quantities in inter-
acting field theories, the Stückelberg-Feynman propagator. We
will define the product of fields so that an interpretation in terms of
the causal propagation of particles is always possible Its importance will become clearer
in the next chapter and we will call
 it only the Feynman propagator, as
 D(x − y) x0 > y 0 it is unfortunately common. One of
∆F (x − y) = h0| T φ(x)φ(y)|0i = the many cases of the Matthew effect
 D(y − x) y 0 > x0 in science, https://en.wikipedia.
org/wiki/Matthew_effect oder
”Der Teufel scheisst immer auf den
größten Haufen.”
(5.76)
where T is the time ordering, which places all operators evaluated

at later times to the left

 φ(x)φ(y) x0 > y 0
T φ(x)φ(y) = (5.77)
 φ(y)φ(x) y 0 > x0
Claim: We can express the Feynman propagator of Eq. (5.76) in

terms of a Lorentz-invariant expression
Z
d4 p i
∆F (x − y) = 4 2 2
e−ip·(x−y)
contour (2π) p − m
The meaning of the ”contour” will become clear soon. This is the
first integral over 4-momenta. Until now, we have integrated only
over 3-momenta and kept the p0 on the mass shell with p0 = Ep .
Here, we have no such condition. The integral for ε = 0 is ill-defined,
because for each p the denominator has a pole at
p
p2 − m2 = (p0 )2 − p2 − m2 p0 = ±Ep = ± p2 + m2
We need a prescription to avoid these singularities in the p0 integra-

tion. The Feynman propagator follows from the contour
Figure 5.2: The contour C for the

Ep Feynman propagator in the complex
p0 -plane.
+Ep
Proof: We first rewrite the numerator

1 1 1
2 2
= 0 2 2
= 0
p −m (p ) − Ep (p − Ep )(p0 + Ep )
The residue is Recall the residue for a simple pole is

1 1 Res [f (z)] = lim (z − c)f (z)
Res =± z=c z→c
p0 =±Ep (p0 0
− Ep )(p + Ep ) 2Ep and that for a positively oriented
simple closed curve C
For x0 > y 0 : we can close the contour in the lower half plane I
p0 = −i|ρ|, since here
X
f (z)dz = 2πi Res [f (z)]
C z=ci
i
−ip0 (x0 −y 0 ) −|ρ|(x0 −y 0 )
e =e →0 where the sum is over the residues
which are enclosed by C. Clockwise
for p0 → −i∞ and therefore the radial integration will not con- encircled residues contribute with the
opposite sign.
tribute. We pick up the residue at p0 = +Ep which gives
I
i 0 0 0 1 −iEp (x0 −y0 )
dp0 2 2
e−ip (x −y ) = −2πi e
C p −m 2Ep
where the minus sign is due to us taking a clockwise contour. There-
fore when x0 > y 0 , we get
Z
d3 p −2πi −iEp (x0 −y0 )+ip·(x−y)
∆F (x − y) = ie
(2π)4 2Ep
Z
d3 p 1 −ip·(x−y)
= e
(2π)3 2Ep
= D(x − y)
which is the Feynman propagator for x0 > y 0 .

For x0 > y 0 : we will close the contour in the upper-half plane. We
catch the residue at −Ep in an anti-clockwise direction and find
Z
d3 p 2πi 0 0
∆F (x − y) = ie+iEp (x −y )+ip·(x−y)
(2π)4 (−2Ep )
Z
d3 p 1 −iEp (y0 −x0 )−ip·(y−x)
= e
(2π)3 2Ep
Z
d3 p 1 −ip·(y−x)
= e
(2π)3 2Ep
= D(y − x)
where in the step from the 2nd to the 3rd line, we have flipped the
integration variable p → −p which is valid since the rest of the
expression (including the boundaries) is symmetric in p. Also in this
case we reproduce the Feynman propagator. QED.
Instead of explicitly showing the contour, we usually write for the

Feynman propagator the beautiful expression This is the most important result of
this chapter.
Z
d4 p i
∆F (x − y) = e−ip·(x−y) (5.78)
(2π)4 p2 − m2 + iε
with ε > 0 and infinitesimal. This shifts the poles slightly off the real
axis, such that the integral along p0 is equivalent to the contour C.
This form of the propagator is called the iε-prescription.
Figure 5.3: The iε-prescription for

=(p0 ) the Feynman contour. Note that the
Ep + i" ε is not exactly the same as in the
propagator, they are proportional to
each other since εprop = 2εplot Ep
<(p0 )
+Ep i"
5.8.1 Green’s Functions

We will now see that the objects we’ve been discussing above are
closely related to the Green’s functions which we have discussed in
Sec. 4.3. In fact, if we stay away from the singularities, we obtain
Z
d4 p i
2
( + m )∆F (x − y) = (−p2 + m2 ) 2 e−ip·(x−y)
(2π)4 p − m2
Z
d4 p −ip·(x−y)
= −i e
(2π)4
= −iδ (4) (x − y) (5.79)
Note, that we did not need the contour in the derivation. For other
applications it is useful to pick alternative contours, which also result
in Green’s functions.
We can define a retarded Green’s function ∆R (x − y) with

D(x − y) − D(y − x) x0 > y 0
∆R (x − y) =
0 y 0 > x0
which corresponds to the contour in Fig. 5.4 (left) This is useful if
retarded advanced
Ep +Ep
Ep +Ep
Figure 5.4: Contours for the retarded

and advanced Green’s function in the
we know the initial value of a field configuration and want to know
complex p0 -plane.
what it evolves into in the presence of a source, e.g.
φ + m2 φ = J(x)
Along the same lines, we can define an advanced Green’s function,

see contour in Fig. 5.4 (right)

0 x0 > y 0
∆A (x − y) =
−D(x − y) + D(y − x) y 0 > x0
which is helpful if we know the final field configuration and want to

figure out where it came from.
Using Heaviside θ-functions we can express the three types of

Green’s functions as
∆F (x − y) = h0| T φ(x)φ(y)|0i
= θ(x0 − y 0 ) h0| φ(x)φ(y)|0i + θ(y 0 − x0 ) h0| φ(y)φ(x)|0i
and
∆R (x − y) = θ(x0 − y 0 ) h0| [φ(x), φ(y)]|0i

∆A (x − y) = −θ(y 0 − x0 ) h0| [φ(x), φ(y)]|0i
We are still a long way from being able to do any real calculation,
since so far we have only talked about the free Klein-Gordon theory,
where the field equations are linear and there are no interactions.
On the other hand, the formalism here is extremely important since
the free theory forms the basis for doing perturbative calculations in
interacting theories.
5.8.2 Particle creation by a classical source

Even though we have no interactions yet, we can discuss what
happens if we disturb the field by a classical, external source J(x)
( + m2 )φ(x) = J(x) (5.80)
which follows from the Lagrangian

1 µ 1
L= ∂ φ∂µ φ − m2 φ2 + J · φ (5.81)
2 2
We will now turn on J(x) for a finite time. Before we turn on J(x),
the field has the form of a free field, see Eq. (5.67)
Z
d3 p 1
φ0 (x) = 3
p ap e−ip·x + a†p eip·x
(2π) 2Ep
without a source, this would be the solution for all time. We can con-
struct the solution with a source, by using the retarded Green’s
function
( + m2 )∆R (x − y) = −iδ (4) (x − y) (5.82)

0 0
∆R (x − y) = 0, (x < y ) (5.83)
The second requirement, that ∆R be the retarded Green function, is

required so that the boundary condition φ(x) → φ0 (x) as x0 → −∞
is satisfied. You can check that this is the correct
Z Ansatz, by plugging the first line into
4 Eq. (5.80) and using the properties
φ(x) = φ0 (x) + i d y ∆R (x − y)J(y)
of the retarded Green’s function
Z Z 3 h i Eq. (5.82)
d p 1
= φ0 (x) + i d4 y θ(x0 − y 0 ) e−ip·(x−y) − eip·(x−y) J(y)
3
(2π) 2Ep ( + m2 )φ(x)
Z Z
d3 p 1 h −ip·(x−y) i Z
x0 →+∞ 4 ip·(x−y) =( + m2 )φ0 (x) + i d4 y( + m2 )x ∆R (x − y)J(y)
= φ0 (x) + i d y e −e J(y)
(2π)3 2Ep
=0 + i(−i)J(x) = J(x)
We have used the fact that if we wait until all of J(y) is in the past, which clearly satisfies Eq. (5.80).
the the θ(x0 − y 0 ) equals 1 over the whole domain of integration
and may be dropped. Then the solution involves only the Fourier
transform of J(y)
Z
J(p) = d4 y eip·y J(y)
evaluated at pµ such that
p2 = m2 (on-shell) (5.84)
and with (J(p))∗ = J(−p) because J(y) is real, we obtain

Z
x0 →+∞ d3 p 1 −ip·x
φ(x) = φ0 (x) + i 3
e J(p) − eip·x (J(p))∗
(2π) 2Ep
We can now group positive frequency terms together with ap and

negative frequency terms together with a†p :
Z " ! ! #
d3 p 1 i i
φ(x) = p ap + p J(p) e−ip·x + a†p − p (J(p))∗ eip·x
(2π)3 2Ep 2Ep 2Ep
Since all observables are built out of the fields, we have solved the
theory. We can now guess (or derive) the Hamiltonian after J(x) has
been switched on and off. The free Hamiltonian (with J(x) ≡ 0) is
Z
d3 p
H= Ep a†p ap
(2π)3
and
h0|H|0i = 0
If we act with J(y) and study the system in the far future (x0 →
+∞), we can get HJ looking at the above discussion of the retarded
Green’s function by just replacing
i
ap → ap + p J(p)
2Ep
i
a†p → a†p − p (J(p))∗
2Ep
to obtain
Z ! !
d3 p i i
HJ = Ep a†p −p (J(p))∗ ap + p J(p)
(2π)3 2Ep 2Ep
The energy of the system in the far future after the source has been
switched off, is therefore
Z
d3 p |J(p)|2
h0|HJ |0i = Ep (5.85)
(2π)3 2Ep
Note that because we are in the Heisenberg representation, we

are still in the ground state of the free theory – the state has not
evolved.
How can we interpret Eq. (5.85)? In the far future, we are in the
free theory again and the spectrum of the Hamiltonian is just free
particles. This means that the expectation value of the total number
of particles created with momentum p is
|J(p)|2
dN (p) =
2Ep
and each Fourier component of J(p) produces particles27 with a 27

This can be made more explicit
probability density |J(p)|2 /(2Ep ) for creating a particle in the mo- once we discuss interacting fields
and have developed the relevant
mentum eigenstate p. The total number of particles produced formalism. Here you are asked to
is intuit this interpretation by starring
Z Z long enough at the form of the free
d3 p 1 Hamiltonian in the absence of a
dN = |J(p)|2 source.
(2π)3 2Ep
We see from the condition Eq. (5.84), that only the Fourier modes
of J(x) which are in resonance (p2 = m2 ) with the on-shell Klein-
Gordon waves are effective at creating particles. This is just the
classical phenomenon of resonance occurring in the quantum field
theory setting.
5.9 Non-relativistic fields

We will now derive the non-relativistic limit of the Klein-Gordon
equation to see if we can recover the Schrödinger equation. Let us
start with a classical complex scalar field which satisfies the Klein-
Gordon equation. We decompose
ψ(x, t) = e−imt ψ̃(x, t) (5.86)
Plugging this into the Klein-Gordon equation

0 = (∂t2 − ∇2 + m2 )ψ = e−imt ∂t2 − 2im∂t − ∇2 ψ̃(x, t) (5.87)
where the m2 term cancelled against the time derivatives. In the

non-relativistic limit
|p| m
and so after a Fourier transformation28 28

Recall
d3 p
Z
φ(p) e−ip·x

φ(x) =
∂t ψ(x, t) → −iEp ψ(p), ∂t ψ̃(x, t) → −i(Ep − m)ψ̃(p) (2π)3
with p2 = m2 , such that ( + m2 )φ =
in the non-relativistic limit, Ep − m m and we find 0
|∂t2 ψ̃(x, t)| m|∂t ψ̃(x, t)|.
We therefore drop the term with two time derivatives to obtain
∂ ψ̃ 1 2
i =− ∇ ψ̃ (5.88)
∂t 2m
This looks like the Schrödinger equation, except it does not have
a probabilistic interpretation – it is just a free classical field, evolving
according to an equation of first order in the time derivatives.
In Sec. 4.1.4, we discussed a Lagrangian which was first order
in the time derivatives. Once again, we can derive Eq. (4.23) from
the relativistic Klein-Gordon Lagrangian using ∂t ψ mψ and

ψ(x, t) = e−imt ψ̃(x, t)
L = ψ ∗ (− − m2 )ψ
= |ψ̇|2 − |∇ψ|2 − m2 |ψ|2
= | − imψ̃ + ∂t ψ̃|2 − |∇ψ̃|2 − m2 |ψ̃|2

≈ im ψ̃ ∗ (∂t ψ̃) − ψ̃ (∂t ψ̃ ∗ ) − |∇ψ̃|2
Once we divide by 1/2m, replace ψ̃ → ψ and partially integrate in

time, we recover Eq. (4.23) which also was first order in the time
derivatives
1
L = i ψ ∗ (∂t ψ) − (∇ψ ∗ )(∇ψ) (5.89)
2m
Like its relativistic origin, this Lagrangian has a U (1) symmetry
ψ → eiα ψ. The associated conserved current is Recall,
! ∂L ∂L
j µ (x) = δψ + δψ ∗
µ −ψ ∗ ψ ∂(∂µ ψ) ∂(∂µ ψ ∗ )
j (x) = i ∗ ∗
(5.90)
2m (ψ ∇ψ − ψ∇ψ )
Let us compute the Hamiltonian with the goal of quantizing the

theory. The conjugate momentum is
∂L
π= = iψ ∗
∂(∂t ψ)
Interestingly, the momentum does not depend on time derivatives,

which is consistent for a theory that only is first order in the time
derivatives.29 We compute the Hamiltonian H = π ψ̇ − L 29
The trajectory is determined
by ψ and ψ ∗ at a time t, no time-
1 derivatives on the initial slice re-
H= (∇ψ ∗ )(∇ψ) quired.
2m
where the time derivatives drop out. As before we quantize by
imposing in the Schrödinger picture the canonical commutation
relations
[ψ(x), π(y)] = iδ (3) (x − y)
or
[ψ(x), ψ † (y)] = δ (3) (x − y)
and all others vanishing
[ψ(x), ψ(y)] = [ψ † (x), ψ † (y)] = 0
We can Fourier expand the field operator ψ(x)

Z
d3 p
ψ(x) = ap eip·x
(2π)3
The commutation relations imply
[ap , a†q ] = (2π)3 δ (3) (p − q)

As before the vacuum (or ground-state) satisfies ap |0i = 0, the

multi-particle excitations are proportional to a†p1 . . . a†pn |0i. They are
eigenstates of the Hamiltonian
p2
H|pi = |pi
2m
The single particle states satisfy the non-relativistic dispersion
relation.
Comments
• Quantizing the first order Lagrangian above gives rise to non-

relativistic particles of mass m.
• We started with a complex field but found only a single type of

particle. The anti-particle is not part of the spectrum. 30 Only 30
which is present in the quantized
with relativity does the existence of anti-particles follow. relativistic Klein-Gordon theory.
• The conserved charge implied by Eq. (5.90) is equal to particle

number
Z
Q = d3 x : ψ † ψ :
Even if we introduce interactions ∆L = V (ψ ∗ ψ), particle number

will still be conserved. Only with relativity, anti-particles appear
and particle number can change.
• We cannot find a non-relativistic limit of a real scalar field. In

the relativistic theory the particles were their own anti-particles
in this case. There can be no way to construct a multi-particle
theory which conserves particle number!
5.9.1 A special case: quantum mechanics

How do we recover quantum mechanics? In quantum mechanics,
we describe physics in terms of the momentum operator P and the
position operator X. We already have the momentum operator
Z
d3 p
P = p a†p ap
(2π)3
In the non-relativistic limit we can also construct a position operator,

see discussion in Sec. 5.4.1. The operator
Z
† d3 p † −ip·x
ψ (x) = a e (5.91)
(2π)3 p
creates a particle at x and we write
|xi = ψ † (x)|0i
The best candidate for a position operator is then

Z
X = d3 x x ψ † (x)ψ(x)
so that
X |xi = x |xi
Let us now construct a state |ϕi which we would usually call the
Schrödinger wavefunction (in the position representation). We
obtain it by taking superpositions of one-particle states |xi
Z
|ϕi = d3 x ϕ(x) |xi
Let us check if it has the right properties. The position operator has
the right action, as we can see
Z
i
X |ϕi = d3 x xi ϕ(x) |xi (5.92)
The momentum operator acts as See Exercise!
Z
∂ϕ(x)
P i |ϕi = d3 x −i |xi (5.93)
∂xi
So we learn that when acting on one-particle states, P and X act as

position and momentum operators in quantum mechanics, with
[X i , P j ] |ϕi = iδ ij |ϕi
What about time-evolution? The wave-function ϕ(x, t) evolves in

time according to the Hamiltonian
Z Z
1 d3 p p2 †
H= d3 x ∇ψ ∗ ∇ψ = a ap
2m (2π)3 2m p
and we find
∂ϕ 1 2
i = Hϕ = − ∇ ϕ (5.94)
∂t 2m
This is the same equation obeyed by the original (full quantum) field.
Only this time, it is really the Schrödinger equation with the usual
probabilistic interpretation for the wavefunction ϕ. In particular, the
total probability which is conserved in quantum mechanics arises
as the conserved charge of the Noether current:
Z
Q= d3 x |ϕ(x)|2
It is useful to know that if we treat the one-particle Schrödinger

equation as the equation for a quantum field then it will give the
proper generalization to multi-particle quantum field theory. His-
torically, the fact that the equation for the classical field Eq. (5.88)
and the one-particle wavefunction Eq. (5.94) gave rise to consider-
able confusion and people thought that perhaps we are quantizing
the wavefunction itself and called it ”second quantization”. This is
clearly incorrect, we are only quantizing the classical field once!
5.9.2 Non-relativistic interactions

Often we are interested in some fixed background potential
V (x). This can be incorporated into field theory by working with a
Lagrangian with
1
L = i ψ ∗ (∂t ψ) − (∇ψ ∗ )(∇ψ) − V (x)ψ ∗ ψ (5.95)
2m
We have broken translational symmetry and there will not be an
associated, conserved energy-momentum tensor. Such Lagrangians
are useful in condensed-matter physics but we almost never
encounter them in high-energy physics, where everything is Poincaré
invariant, that is translational and Lorentz invariant.
We can also have interactions between particles, which require
n particle states with n ≥ 2. They can arise from Lagrangians of the
form
∆L = λ ψ ∗ (x)ψ ∗ (x)ψ(x)ψ(x)
which destroys two particles before creating two new ones. In the
following chapter we will explore interactions like there in detail for
relativistic theories.
6
Interacting quantum fields
We were able to completely solve the free scalar quantum field

theory, but nothing interesting happens in this theory. Most impor-
tantly, the particle excitations do not interact with each other. Now
we will discuss more complicated theories with interaction terms,
which in particular will lead to non-linear field equations.
6.1 Dimensional analysis of interactions

We can add the following terms to the free Lagrangian
1 1
L= ∂µ φ∂ µ φ − m2 φ2 − λ3 φ3 − λ4 φ4 − λ5 φ5 + . . .
2 2
We call the coefficients of the higher order terms λn coupling
constants. The equations of motion now contain non-linear terms
and they cannot be solve by Fourier analysis as the free Klein-
Gordon equation.
When are these additional terms small perturbations? Which
terms should we include? What about λ15 φ15 ? I have restricted the expansion
Recall our discussion in Sec. 5.3.1, where we derived the mass to powers of φ but we could have
equally added terms of the form
dimension of [φ] = 1 and of the action [S] = 0 from which we
1
immediately conclude [L] = 4. We obtain for the coupling constants (∂µ φ∂ µ φ)n φm
M 4−4n−m
with integer n, m. In particular, we
[λn ] = 4 − n, e.g. [λ3 ] = 1, [λ4 ] = 0, [λ5 ] = −1 could have added
1
So for all but λ4 we cannot just ask for λn 1, since this only ∆L = (∂µ φ∂ µ φ)2
M4
makes sense for dimensionless quantities. We will introduce mass Here you can see explicitly how the
scales Mi for the dimensionful couplings to write ’typical energy scale’ E appears if
you replace ∂µ ∼ pµ ∼ E. We obtain
1 1 λ̃5 5 for the typical size of the interaction
L= ∂µ φ∂ µ φ − m2 φ − λ̃3 M3 φ3 − λ̃4 φ4 − φ + ...
2 2 M5 E4
∼
M4
Now the statement of λ̃n 1 makes sense but it has be accom-
that it is clearly an irrelevant pertur-
panied by discussion of the typical energy E of the process under bation.
consideration vs. the mass scale Mi of the interaction.
We distinguish three classes of interactions Note here, that the meaning of
relevant, marginal and irrelevant
• Relevant perturbation: ∆L = −λ3 φ3 = −λ̃3 M3 φ3 and is tied to the number of space-time
dimensions. In D = 2 for example,
[λ3 ] = 1. The dimensionless parameter is
[φ]=0 and all the φn terms are
λ3 M3 relevant!
= λ̃3
E E
What is the meaning of the scale E? We typically take E to

be the energy scale of the process of interest.1 This means that 1
Think of the center of mass energy
of a scattering, or the mass of
when we test the theory a very high energies E λ3 or E
particle if you consider decays of
M3 ,2 λ3 φ3 is a small perturbation, suppressed by M3 /E. If we particles.
test at low energies E λ3 it is a large perturbation. We call 2
Always assuming that λ̃n = O(1).
terms like this relevant because they are most relevant at low
energies, which is where most of the physics we see takes place.
In relativistic theories, E > m, and we can make this a small
perturbation by taking λ3 m or M3 m.
• Marginal perturbation: ∆L = λ4 φ4 with [λ4 ] = 0. The

effects of terms with d = 4 are independent of energy, and these
interactions are called marginal. This term is small if λ4 1.
• Irrelevant perturbation: ∆L = λ5 φ5 with [λ5 ] = −1 (and

λn φn for n ≥ 5). The dimensionless parameter is
E
λ5 = λ̃5
M5
This interaction is small at low energies E M5 and large at
high energies. Such perturbations are called irrelevant.
Note, that his naive categorization of relevant, marginal and irrele-

vant is sometimes subject to change due to quantum corrections.3 3
One example are the pions of QCD,
Further, we see that irrelevant terms can be problematic in QFT they are scalars [φ] = 1 but emerge
as strongly coupled bound states of
since, as we have seen, we often have to integrate to very high two quarks [ψ̄ψ] = 3 at low energy.
energies. These terms lead to so-called ’non-renormalizable’ field Fermions in a Dirac spinor have
mass-dimension [ψ] = 3/2.
theories and which require some additional work to make sense of
the infinities at arbitrarily high energies.
6.1.1 Some examples of weakly coupled theories

Here we will only study theories where the interaction terms can be
considered small perturbations. You might discuss strongly interact-
ing theories in a QFT2 course.
• φ4 theory: This is a subset of the marginal and relevant interac-
tions above
1 1 λ
L= ∂µ φ∂ µ φ − m2 φ2 − φ4
2 2 4!
with small λ 1. Is there a reason why we should not have
a ∆L ∼ φ3 coupling? This could be explained by a discrete
symmetry φ(x) → −φ(x) which allows only even powers in the
field.
We can guess the effects of the extra term. If we expand out φ4 in
creation and annihilation operators, we find terms like
a†p1 a†p2 a†p3 a†p4 , a†p1 a†p2 a†p3 ap4 , ...
These terms create and destroy particles and particle number will
in general not be conserved and we could calculate [H, N ] 6= 0
(even though the discrete parity would be conserved.)
Why QFT is so simple

Typically we consider only relevant and marginal couplings, since irrelevant couplings become small at low energies.
This simplifies QFT a lot, since out of the infinite number of interaction terms we can write down, only a small
number is needed. In the case of a scalar field, we would have
1 1
L= ∂µ φ∂ µ φ − m2 φ − λ3 φ3 − λ4 φ4 + O(Λ/E)
2 2
Let us assume, we someday discover the true Lagrangian of the universe which is valid up to very high energies,
say Λ with [Λ] = 1. This could be the GUT scale (Λ ≈ 1016 GeV) or the Planck scale (Λ ≈ 1018 GeV). If we now
investigate nature at the small energies E accesible to us E Λ and if we assume she is described by a scalar field
φ (think of the Higgs). This scalar field will then have some complicated couplings λn φn which depend on the UV
completion of our amazing theory of everything at Λ. In terms of the dimensionless couplings λ̃n , we expect them
to scale like
λ̃n
λn =
Λn−4
Now for experiments at E Λ (think of E ≈ 104 GeV at the LHC), the interaction therms of φn with n > 4 will
be suppressed by very small powers of
n−4
E
Λ
This simple argument using dimensional analysis tells us that we can usually focus on the first few terms in the
interactions. It also tells us that it will be very difficult to figure out the high energy theory. These considerations
are a simplified version of what is called effective field theory and Wilson’s renormalization group. Due to the
unknown UV physics completing our QFT, we are generally required to write down all the allowed marginal and
relevant interactions to capture the IR behaviour of a theory with a scalar field.
This discussion is the modern viewpoint of what people previously discussed under the notion of ”renormalizable”
quantum field theories.
• Scalar Yukawa theory: We introduce a real scalar φ and a

complex scalar ψ with the Lagrangian We will use this theory in the sec-
tions below!
1 1
L = ∂µ ψ ∗ ∂ µ ψ − M 2 ψ ∗ ψ + ∂µ φ∂ µ φ − m2 φ2 − g ψ ∗ ψ φ
2 2
The coupling g has [g] = 1 and is a relevant coupling. For the per-
turbation to be small, we require g M, m. Particle number is
again not conserved anymore but since the Lagrangian is invariant
under a continuous phase symmetry (U (1))
ψ → eiα ψ
a conserved charge Q exists, with [Q, H] = 0. The number of

ψ-particles minus the number of ψ-anti-particles is conserved. We
will usually denote the antiparticle as ψ̄.
• Quantum electro-dynamics This theory involves Dirac 4-

spinors4 Ψ and the electro-magnetic vector potential Aµ . 4
Which we will discuss in the next
chapter
1
L = Ψ̄ (iγ µ ∂µ − m)Ψ − Fµν F µν − e Ψ̄ γ µ Ψ Aµ
4
You can easily determine the canonical dimensions of the fields to
find that [e] = 0. You know e = −|e| as the electron charge. In
our units |e| ≈ 0.3 and the interaction term is small.
• Fermion Yukawa theory This theory involves spinors Ψ and a

real scalar φ
1 1
L = Ψ̄ (iγ µ ∂µ − m)Ψ + ∂µ φ∂ µ φ − M 2 φ2 − g Ψ̄ Ψ φ
2 2
You can also determine the canonical dimensions of the fields
to find that [g] = 0. Yukawa invented this theory to describe
nucleons (Ψ ) and pions (φ). Nowadays we use it in the Standard
Model (SM), since this is the interaction between the Higgs
and the matter fermions (quarks and leptons). Most of the free
parameters in the SM are Yukawa coupling constants.
In our now interacting quantum field theory, we impose the equal-

time commutation relations
[φa (x, t), π b (y, t)] = iδba δ (3) (x − y)
which are unaffected by Lint .5 You can straight-forwardly check 5

Note, that if Lint contained a
then that the field operator in the Heisenberg picture satisfies the ∂µ φ, the definition of the conjugate
momentum π(x) would change.
equation of motion including interaction term.
6.2 The interaction picture
We have discussed the Schroedinger (states time-dependent, oper-

ators not) and the Heisenberg picture (operators time-dependent,
states not) in Sec. 5.6 and we will now introduce a hybrid called the
interaction picture. We write the Hamiltonian as the free theory
part H0 plus the interaction Hint
H = H0 + Hint (6.1)
for the φ4 -example that would be

Z
λ 4
H = HKlein-Gordon + d3 x φ (x) (6.2)
4!
We now define states and operators in the interaction picture as It is a hybrid: informally speaking it
is the Heisenberg picture for the free
H0 and the Schroedinger picture for
the Hint .
|ψ(t)iI = eiH0 t |ψ(t)iS (6.3)
iH0 t −iH0 t
OI (t) = e OS e (6.4)
when the interaction is small, e.g. λ 1, then the most important

time dependence generated by H0 is already taken care off. The
interaction Hamiltonian in the interaction picture is
HI = (Hint )I = eiH0 t (Hint ) e−iH0 t (6.5)
States in the interaction picture evolve according to (starting with

the Schroedinger picture result)
d
i |ψ(t)iS = HS |ψ(t)iS
dt
plugging in the definition of the interaction state |ψ(t)iI , we get
d −iH0 t
i (e |ψiI ) = (H0 + Hint )S e−iH0 t |ψiI (6.6)
dt
d
i |ψiI = HI |ψiI (6.7)
dt
where we have first canceled H0 on both sides, multiplied the equa-
tion by eiH0 t and used Eq. (6.5).
6.2.1 Dyson’s formula

No relation to the vacuum appliance
At a fixed time t0 , we can expand φ(x) as before in terms of ladder inventor.
operators
Z
d3 p 1
φ(x, t0 ) = 3
p ap eix·p + a†p e−ix·p
(2π) 2Ep
We now want a perturbative expansion that captures the full time-

dependence of the quantum field φ. In the Heisenberg picture, we
would get
φH (x, t) = eiH(t−t0 ) φH (x, t0 ) e−iH(t−t0 ) (6.8)
For λ = 0, we have H = H0 and this reduces with Eq. (6.4) to

φ(x, t)λ=0 = eiH0 (t−t0 ) φ(x, t0 ) e−iH0 (t−t0 ) = φI (x, t) (6.9)
When λ is small, this will give the most important part of the time-
dependence. Since H0 is diagonalized by the Fourier expansion, we
get
Z
d3 p 1 −ip·x

† ip·x
φI (x) = p ap e + ap e 0 (6.10)
(2π)3 2Ep x =t−t0
which is just the familiar result of Eq. (5.67). We now want to

express the Heisenberg picture field operator φ in terms of φI (x),
again as before to have the full time-dependence in the operators
(while keeping the states time-independent). We can write this as
φ(x, t) = eiH(t−t0 ) e−iH0 (t−t0 ) φI (x, t) eiH0 (t−t0 ) e−iH(t−t0 ) (6.11)

†
≡ U (t, t0 ) φI (x, t) U (t, t0 ) (6.12)
with the unitary operator
U (t, t0 ) = eiH0 (t−t0 ) e−iH(t−t0 ) (6.13)
which is also called the interaction picture propagator or time-

evolution operator. States in the interaction picture evolve as From the definition of a state in the
interaction picture
|ψ(t)iI = U (t, t0 )|ψ(t0 )iI (6.14) |ψ(t)iI = eiH0 t |ψ(t)iS
= eiH0 t e−iHt |ψ(0)iS = U (t, 0)|ψ(0)iS
We want to express U (t, t0 ) entirely in terms of φI for which we have
where we have use the time-evolution
an explicit expression in terms of ladder operators. We note, that in the Schroedinger picture.
U (t, t0 ) is the unique solution, with initial condition U (t0 , t0 ) = 1 of

a simple differential equation
∂
i U (t, t0 ) = eiH0 (t−t0 ) (H − H0 )e−iH(t−t0 )
∂t
= eiH0 (t−t0 ) (Hint )e−iH(t−t0 )
= eiH0 (t−t0 ) (Hint )e−iH0 (t−t0 ) eiH0 (t−t0 ) e−iH(t−t0 )
= HI (t) U (t, t0 )
with HI (t) as defined in Eq. (6.5).6 The solution of this equation 6

E.g.
Rt
should naively look something like U ∼ exp(−i dt0 HI (t0 )), but that λ 4
Z
HI (t) = d3 x φ (x)
ignores the fact that [HI (t), HI (t0 )] 6= 0 when t 6= t0 . 4! I
A more careful treatment shows that the actual solution is

Z t Z t Z t1
2
U (t, t0 ) = 1+(−i) dt1 HI (t1 ) + (−i) dt1 dt2 HI (t1 )HI (t2 )
t0 t0 t0
Z t Z t1 Z t2
+(−i)3 dt1 dt2 dt3 HI (t1 )HI (t2 )HI (t3 ) + . . .
t0 t0 t0
To verify, that this indeed the correct solution, take the derivative:
you find that each term gives the previous one ×(−i)HI (t). The For the first two terms:
initial condition is U (t0 , t0 ) = 1 is trivially satisfied. ∂
U (t, t0 )
We observe that that the factors of HI (ti ) stand in time order, ∂t
Z t
with the later operators to the left. We can now simplify using the = (−i) HI (t) + (−i)2 HI (t) dt2 HI (t2 ) + . . .
t0
time ordering T . Starting with HI2 Z t
= −iHI (t) 1 + (−i) dt1 HI (t1 ) + . . .
Z t Z t1 Z Z t
1 t t0
dt1 dt2 HI (t1 )HI (t2 ) = dt1 dt2 T {HI (t1 )HI (t2 )}
t0 t0 2 t0 t0
= −iHI (t)U (t, t0 )
The time-ordered double integral on the RHS just counts everything
twice, since in the t1 − t2 plane, the integrand T {HI (t1 )HI (t2 )} is
symmetric about the t1 = t2 line. Similarly, one can show
Z t Z tn−1 Z Z t
1 t
dt1 . . . dtn HI (t1 ) . . . HI (tn ) = dt1 . . . dtn T {HI (t1 ) . . . HI (tn )}
t0 t0 n! t0 t0
It’s not easy to draw this but you should be able to convince yourself
that this is in fact correct. We can now write the time evolution
operator in a very compact form:
Z t Z Z t
(−i)2 t
U (t, t0 ) = 1+(−i) dt1 HI (t1 ) + dt1 dt2 T {HI (t1 )HI (t2 )}
t0 2! t0 t0
Z Z t Z t
(−i)3 t
+ dt1 dt2 dt3 T {HI (t1 )HI (t2 )HI (t3 )} + . . .
3! t0 t0 t0
and we obtain Dyson’s formula
Z t
0 0
U (t, t0 ) = T exp −i dt HI (t ) (6.15)
t0
It is usually very hard (impossible) to compute time ordered expo-

nentials in practice. The power of Dyson’s formula is that just keep
the first couple of terms in the expansion when the interaction is a
small perturbation.
Z Z Figure 6.1: Dyson integration.

t2 t
dt1
t
dt2
t0 t0
t
Z t Z t1
dt1 dt2
t0 t0
t0
t1
t0 t
6.3 A first look at scattering

We will now apply what we have just learned to a the interaction in
the scalar Yukawa theory
Z
HI = g d3 x ψ † ψφ
where all the fields are in the interaction picture.7 This HI allows 7
Which means the field operators are
particles to morph into each other. To understand why that is, in the Heisenberg picture of the free
theory.
follow the time evolution of a state
|ψ(t)iI = U (t, t0 )|ψ(t0 )iI
where U (t, t0 ) is given by Dyson’s formula, which contains an expan-

sion in powers of HI . Now, HI contains annihilation and creation
operators for different types of particles. This is a simplified descrip-
tion of meson-nucleon interactions.8 We can have 8
Nucleons are neutron and the
proton, wich are composed of down
• ψ † ∼ b† + c : This operator creates ψ particles trough b† and and up quarks. We will now take the
historical perspective and consider
destroys anti-particles through c. Let us call these ψ-particles them elementary particles. In our
nucleons. toy model we will describe them as
scalars. In reality nucleons are spin
• ψ ∼ b + c† : This operator destroys ψ particles (nucleons) trough 1/2 particles, and therefore fermions.
b and creates anti-particles (anti-nucleons) through c† .
• φ ∼ a + a† We will call the φ particles mesons and φ can therefore

create and destroy mesons.
Due to the U (1) symmetry of ψ in L, we know that the charge

Q = Nc − Nb will be conserved even in the presence of LI = −HI .
The interaction ψ † ψφ contains a collection of creation and annihi-

lation operators, such as
c† b† a
which annihilates a φ particle (meson) and creates ψ particle and

antiparticle (nucleon and anti-nucleon): this corresponds to the
decay of the process φ → ψ̄ψ.
Another example
cc† a ψ+φ→ψ
which corresponds to the absorption of a meson by a nucleon.

At second order in perturbation theory, ∼ HI2 , which produces
more complicated processes like
ψ̄ + ψ → φ → ψ̄ + ψ
nucleon-antinucleon scattering through the creation of an intermedi-

ary meson.
6.3.1 Scattering simplified

We will now discuss scattering processes. They are particularly
convenient since in many cases the initial and final states look
like non-interacting particles. What does that mean? In a scat-
tering process, we start with an initial state |ii of isolated particles.
The particles are widely separated and do not feel the effects of the
interaction: they look like free plane wave states9 . Once the particles 9
Eigenstates of H0 and so eigen-
approach each other, they begin to feel the interaction (and the states also of N , even though in
general [H, N ] 6= 0
resulting potential) and they evolve in complicated non-linear way
as in Eq. (6.14). In this intermediate stage the system will look ex-
tremely complicated if expressed in terms of free particles. Particles
will be created and destroyed, since in general
[HI , N ] 6= 0
We will not just have two colliding protons but a very complicated
mess of protons, pions, photons, gluons, etc.
The result of the scattering process can have several outcomes.
Multiple initial particles could form a bound state.10 In this case, 10
e.g. p + p → 2 D, two protons
no matter how long we wait after the scattering, the final state will forming a deuterium nucleus.
never look like the eigenstate of the free Hamiltonian H0 , because

the interaction is responsible for the bound state.11 The formalism 11
Without the interaction the bound
here will not be useful in this case. state will fly apart.
We will now consider a process where no bound state is formed.

After some long time the system will just be a couple of widely
separated (non-interacting) particles. It will look simple again.
Before we go any further, you need to know that this is a bit of a
fake.12 Despite this, our quick and dirty scattering model will still 12
No matter how far in the future
work in the processes we consider here. We can see this by imagining we go, we never end up with a
collection of free particles. You
a theory like already know this from electro-
magnetism: the electron always
L = Lφ + Lψ − g f (t) ψ † ψφ (6.16) carries its electromagnetic field
with it (which corresponds to a
virtual cloud of photons around the
electron). Similarly, nucleons always
have a cloud of mesons around them.
Turning off our interaction, the states
will change, and our simple picture is
not quite right.
Figure 6.2: Turning on and off of

f (t) the function f (t) in Eq. (6.16). The
scattering process occurs near t = 0.
1
t
T
with f (t) like in Fig. 6.2. The function f (t) = 0 for large |t| and
f (t) = 1 around 0 where the scattering happens. A long time
after the scattering has occurred, we turn the interaction off very
slowly (adiabatically) over a period ∆: we expect that the simple
states in the real theory slowly turn into eigenstates of H0 with unit
probability.
There will be a one-to-one correspondence between asymptotic
simple eigenstates of the full Hamiltonian H and the eigenstates of
the free H0 . 13 We should recover the full theory if 13
Again, we can’t consider bound
states, which are not eigenstates of
T → ∞, ∆ → ∞, ∆/T → 0 the free Hamiltonian.
where the last limit is required to ensure no edge effects occur

(switching on and off of HI is adiabatic). We can justify this ap-
proach in a more rigorous manner but we do not have time for the
technical details in this course. Look for the LSZ reduction for-
mula in your favorite QFT textbook, named after the German
physicists Harry Lehmann, Kurt Symanzik, and Wolfhart Zimmer-
mann14 . 14
Wolfhart Zimmermann was hon-
After this long motivation, we finally get to the meat.15 We want orary professor of TUM and director
at the Heisenberg MPI in Freimann.
to solve Eq. (6.7) 15
Or the tofu if you are so inclined.
d
i |ψiI = HI |ψiI
dt
with the boundary condition I will drop the subscript ”I” on
states from now on, since we will be
|ψ(−∞)i = |ii (6.17) in the interaction picture from now
on if not otherwise stated.
and
|ψ(+∞)i = |f i (6.18)
The amplitude to go from |ii to |f i is
lim hf |U (t+ , t− )|ii ≡ hf |S|ii (6.19)
t± →±∞
where the unitary operator S is called the S-matrix. Alternatively,

we can with Eq. (6.15) write
Z
4
S = T exp −i d x HI (x) (6.20)
6.3.2 Example: Meson Decay φ → ψ̄ψ

We take the (relativistically) normalized initial φ and final states ψ̄ψ
p
|ii = |pφ i = 2Ep a†p |0i (6.21)
p
|f i = |q1 q2 i = 4Eq1 Eq2 b†q1 c†q2 |0i (6.22)
where q1,2 are the momenta of the decay products, a nucleon anti-
nucleon pair (ψ̄ψ). We can compute the amplitude for the decay
with Eq. (6.15) to leading order in g
Z ∞
hf |S|ii = hf |ii − ihf | dt HI (t)|ii + . . . (6.23)
Z −∞
= 0 − ighf | d4 x ψ † (x)ψ(x)φ(x) |ii (6.24)
We first expand φ ∼ a + a† . We will only need the a piece which The a† term turns |ii into a two
annihilates the initial meson state |ii into something ∼ |0i. We meson state with zero overlap with
the two nucleon state in |f i.
obtain,
Z ∞
hf |S|ii = ihf | dt HI (t) |ii + . . .
−∞
Z Z p
† d3 k 2Ep
4
= −ighf | d x ψ (x)ψ(x) 3
√ ak a†p e−ik·x |0i
(2π) 2Ek
Z
= −ighf | d4 x ψ † (x)ψ(x)e−ip·x |0i
where we have used the commutation relation of the ak . To get non-

zero overlap with hf |, only the b† and c† contribute, since we need to
create a nucleon and anti-nucleon final state from |0i. We have,
R 4 3 p
d xd k1 d3 k2 4Eq1 Eq2 −i(p−k2 −k1 )·x
hf |S|ii = −igh0| 6
p e bq1 cq2 b†k1 c†k2 |0i
(2π) 4Ek1 Ek2
= −ig (2π)4 δ (4) (p − q1 − q2 ) (6.25)
and we have found our first quantum field theory amplitude! Notice
the δ-function which encapsulates the 4-vector conservation. In
particular, the decay will only take place if mφ ≥ 2Mψ .16 16
Ex: Show this!
We post-pone the steps to turn this into a something observable,
like a life-time of the meson. You can already see that it might in-
volve some additional insights since once we calculated probabilities,
we will get the square of a δ-function.
6.4 Wick’s Theorem
We now want to find a systematic way to compute quantities like
hf |T {HI (x1 ) . . . HI (xn )}|ii
from Dyson’s formula. The time-ordering T fixes the order of the

operators, but we would have a much simpler calculation if we could
move all annihilation operators to the right, where they can remove
particles in |ii. What we want find is therefore: how to go from time
ordered products to normal order products of fields. This is called

Wick’s theorem.
We define the contraction of two fields
A(x)B(y) ≡ T {A(x)B(y)} − : A(x)B(y) : (6.26)
We can easily see that A(x)B(y) is a c-number. Let us consider the

example of a real scalar field
φ(x) = φ+ (x) + φ− (x)
decomposed as Note, that these are either free

fields in the Heisenberg picture, or
Z
d3 p 1 interaction picture fields φI (x) for
φ+ (x) = p ap e−ip·x H = H0 + HI .
(2π)3 2Ep
Z
d3 p 1
φ− (x) = 3
p a†p eip·x
(2π) 2Ep
where I would have used the opposite assignment of + and − but we

have Pauli and Heisenberg to blame.17 Let us consider x0 > y 0 , we 17
It is due to the sometimes used
obtain notion of positive energies in the
time evolution of φ+ ∼ e−iEp t and
the negative energies in the time
T {φ(x)φ(y)} = φ(x)φ(y) evolution of φ− ∼ e−i(−Ep )t .
= (φ+ (x) + φ− (x))(φ+ (y) + φ− (y))
= φ+ (x)φ+ (y) + φ− (x)φ+ (y) + φ+ (x)φ− (y) + φ− (x)φ− (y)
= φ+ (x)φ+ (y) + φ− (x)φ+ (y) + φ− (y)φ+ (x) + [φ+ (x), φ− (y)] + φ− (x)φ− (y)
= : φ(x)φ(y) : + [φ+ (x), φ− (y)]
= : φ(x)φ(y) : + D(x − y)
where we have achieved normal ordering at the price of an ex-

tra term [φ+ (x), φ− (y)] = D(x − y), which is the propagator of
Eq. (5.74).18 Similarly for y 0 > x0 , 18
Recall,
D(x − y) = h0|φ(x)φ(y)|0i
T {φ(x)φ(y)} = : φ(x)φ(y) : + [φ+ (y), φ− (x)]
= h0| : φ(x)φ(y) : + [φ+ (x), φ− (y)]|0i
= : φ(x)φ(y) : + D(y − x) = h0|[φ+ (x), φ− (y)]|0i
= [φ+ (x), φ− (y)]
Putting both together, we finally obtain
because in the last line, the commu-
tator is a c-number.
T {φ(x)φ(y)} = : φ(x)φ(y) : + ∆F (x − y) (6.27)
where ∆F (x − y) is the Feynman propagator! As derived in

Sec. 5.8, it has the integral representation
Z
d4 p i
∆F (x − y) = e−ip·(x−y)
(2π) p − m2 + iε
4 2
An alternative derivation proceeds as follows: we have shown above,

that the difference between normal ordering and time ordering
φ(x)φ(y) is just a c-number.19 So we can equally switch it between 19
The operator structure is lost once
we use the commutator for a and a† .
the vacuum
φ(x)φ(y) = h0|φ(x)φ(y)|0i
= h0|T {φ(x)φ(y)}|0i − h0| : φ(x)φ(y) : |0i
= h0|T {φ(x)φ(y)}|0i
= ∆F (x − y)
For charged fields, it is straight-forward to show that the propaga-

tor is
Z
† † d4 p i
ψ(x)ψ (y) = ψ (x)ψ(y) = e−ip·(x−y)
(2π)4 p2 − M 2 + iε
(6.28)
Note, that the only difference is the mass M of ψ. All other contrac-
tions vanish
ψ(x)ψ(y) = ψ † (x)ψ † (y) = 0
as they should. The last equation is easy to show since ψ ∼ c† + b

and ψ therefore only creates c-type particles and annihilates b-type
particles, ergo h0|T {ψ(x)ψ(y)|0i = 0.
We can now state Wick’s theorem. For any collection of fields
φ1 ≡ φa1 (x1 ), φ2 ≡ φa2 (x2 ), . . ., the T -product of the fields has the
expansion
T {φ1 . . . φn } = : φ1 . . . φn + all possible contractions :
For n = 2, it is identical to Eq. (6.26). The term ”all possible

contractions” means, there will be one term for each possible con-
traction of the n fields in pairs. The proof that shows this for all n is
by induction and so not very illuminating. With this theorem, we Proof: Suppose it is true for
obtain thus for n = 4, φ2 . . . φn and we now add φ1 . We
choose x01 > x0k for all 2 ≤ k ≤ n. We
move φ1 out to the left of the time
T {φ1 φ2 φ3 φ4 } = : φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4 ordered product
+φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4 T {φ1 φ2 . . . φn }
−
= (φ+
1 + φ1 )(: φ2 . . . φn + contractions :)
+φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4 + φ1 φ2 φ3 φ4 : The φ− †
1 ∼ a is already normal
ordered. To have the RHS as a
where a contraction across operators which are not adjacent, still normal ordered product, we need to
−
gives a factor of ∆F , e.g. commute φ+ 1 past the φk operators.
Each time φ+ 1 moves past one φ−
k , we
: φ1 φ2 φ3 φ4 : = ∆F (x1 − x3 ) : φ2 φ4 : pick up a factor
φ1 φk = ∆F (x1 − xk )
6.4.1 Example: nucleon scattering ψψ → ψψ Try it!
This proves Wick’s theorem by
We are scattering ψψ → ψψ and therefore have the in- and outgoing induction.
states
p p
|ii = |p1 p2 i = 2Ep1 2Ep2 b†p1 b†p2 |0i (6.29)
p p
|f i = |q1 q2 i = 2Eq1 2Eq2 b†q1 b†q2 |0i (6.30)
We now want to obtain the expansion of hf |S|ii. In fact, we are

interested in hf |S − 1|ii, since we do not care about amplitudes where
nothing happens.20 At second order in g, we get for hf |S − 1|ii 20
The two nucleons in this case
just continue to fly without any
Z
(−ig)2 interaction as free particles.
d4 x1 d4 x2 ψ † (x1 )ψ(x1 )φ(x1 )ψ † (x2 )ψ(x2 )φ(x2 ) (6.31)
2!
Wick’s theorem can relate this to a sum of normal-ordered products.
The interesting piece for us is
: ψ † (x1 )ψ(x1 )ψ † (x2 )ψ(x2 ) : φ(x1 )φ(x2 ) (6.32)
which contributes to ψψ → ψψ scattering, because the two ψ fields

annihilate the initial nucleons and the two ψ † ’s creates the outgoing
nucleons (or ψ-particles). Any other contraction will give a zero
result. We obtain with Eq. (5.68)
hq1 q2 | : ψ † (x1 )ψ(x1 )ψ † (x2 )ψ(x2 ) : |p1 p2 i

= hq1 q2 | ψ † (x1 )ψ † (x2 ) |0ih0| ψ(x1 )ψ(x2 ) |p1 p2 i

= eiq1 ·x1 +iq2 ·x2 + e+iq1 ·x2 +iq2 ·x1 e−ip1 ·x1 −ip2 ·x2 + e−ip1 ·x2 −ip2 ·x1
nh i o
= eix1 ·(q1 −p1 )+ix2 ·(q2 −p2 ) + eix1 ·(q2 −p1 )+ix2 ·(q1 −p2 ) + (x1 ↔ x2 )
Let us plug this into Eq. (6.31) to obtain

Z Z
(−ig)2 d4 k i
d4 x1 d4 x2 {. . . + (x1 ↔ x2 )} e−ik·(x1 −x2 )
2! (2π)4 k 2 − m2 + iε
where the {. . .} corresponds to the sum of exponentials above result-

ing from the matrix element of the normal-ordered operator. Since
the propagator is symmetric under x1 ↔ x2 we can drop this in
exchange for the 1/2 in front. The xi integrals result in δ-functions,
e.g. for the first term we get for the exponentials
Z
d4 x1 d4 x2 eix1 ·(q1 −p1 )+ix2 ·(q2 −p2 )−ik·(x1 −x2 )
= (2π)8 δ (4) (q1 − p1 − k)δ (4) (q2 − p2 + k)
and we find
Z h
2 d4 k (2π)8
i(−ig) δ (4) (q1 − p1 − k)δ (4) (q2 − p2 + k)
(2π)4 k 2 − m2 + iε
i
+ δ (4) (q1 − p2 + k)δ (4) (q2 − p1 − k)
We finally obtain

1 1
i(−ig)2 + (2π)4 δ (4) (p1 + p2 − q1 − q2 )
(p1 − q1 )2 − m2 (p1 − q2 )2 − m2
Notice that performing the final integral over δ-functions leaves us

with a factor of
(2π)4 δ (4) (p1 + p2 − q1 − q2 )
which just enforces 4-momentum conservation for the scattering

process. Since we usually have Poincaré invariant Lagrangians21 , it 21
Poincaré invariant = space-time
translation (and Lorentz) invariant.
For the δ (4) function the space-
time translation invariance is the
crucial part, since it guarantees total
4-momentum conservation.
is customary to define the invariant Feynman amplitude Af i by

22
22
We will get back to why we can
hf |S − 1|ii = i Af i (2π)4 δ (4) (pf − pi )
remove the δ-function later. We will
see it has to do with our assumption
where the factor i is by convention, since it reproduces the phase of free particles being in plane wave
states.
conventions for scattering in non-relativistic quantum mechanics.
P µ P µ
Further, pµf = µ
n pf,n and pi = n pi,n . We can ignore the iε,
because the denominator is never zero: in the center of mass frame,
we can write the momenta as23 23
Careful: p = |p| and not a 4-
vector!
p
p1 = ( p2 + M 2 , p, 0, 0) (6.33)
p
p2 = ( p2 + M 2 , −p, 0, 0) (6.34)
p
2 2
q1 = ( p + M , p cos θ, p sin θ, 0) (6.35)
p
q1 = ( p2 + M 2 , −p cos θ, −p sin θ, 0) (6.36)
which immediately gives24 24

We see that e.g. (p1 − q1 )2 < 0 and
so (p1 − q1 )2 6= m2 which justifies
(p1 − q1 )2 = −2p2 (1 − cos θ), (p1 − q2 )2 = −2p2 (1 + cos θ) dropping iε.
and so

2 1 1
A=g 2 2
+ 2 (6.37)
2p (1 − cos θ) + m 2p (1 + cos θ) + m2
Note that the two terms are required because of Bose statistics. Scat-
tering into two identical particles at an angle θ is indistinguishable
from scattering at an angle θ − π, and so the probability must be
symmetrical under the interchange of the two processes. Since these
particles are bosons, the amplitude must also be symmetric.
This was a bit tedious. There is a much simpler way which we
ψ†
will discuss in a little bit using Feynman diagrams.
φ
6.5 Diagrammatic Perturbation theory: Feynman Diagrams
ψ † ψφ
Our final result for ψψ → ψψ scattering in Eq. (6.37) was remark- ψ
ably simple, even though the intermediate steps were a bit messy.
This motivates a diagrammatic short-hand called Feynman dia-
grams. These are pictures of the fields and contractions which we
Figure 6.3: Interaction vertex. Each
need to evaluate to give the matrix element. At the nth order in per- field in HI is represented by a line
turbation theory HI will act n times and so a Feynman diagram will emanating from the vertex. The
ψ † -line creates a nucleon or destroys
contain n interaction vertices. For our toy model the interaction an anti-nucleon, the ψ-line creates an
vertex looks like in Fig. 6.3. anti-nucleon or destroys a nucleon,
We need to distinguish ψ from ψ † and we will therefore draw an and the φ line creates or annihilates
a meson.
arrow on the line.
Next, contractions are the lines joining different vertices. Every
time there is a contraction, we join the lines of the contracted fields.
The example of Eq. (6.32) corresponds to the diagram Fig. 6.4.
φ1 φ2
Arrows will line up because ψ(x)ψ(y) = ψ † (x)ψ † (y) = 0 and
the arrows indicate the flow of the U (1) charge. Now, any fields
Figure 6.4: Diagrammatic representa-

tion of Eq. (6.32).
which are left uncontracted must either annihilate particles from the
incoming state or create particles from the outgoing state. If there q1
p1
are different ways of doing this, they correspond to indistinguishable
processes, so we must add the corresponding amplitudes: we write k
down a separate Feynman diagram for each distinct labeling of the p2 q2
external legs. For ψψ → ψψ scattering, this results in two Feynman
diagrams
We will revisit them below. p1 q2
k
6.5.1 Feynman rules p2 q1
1. For each particle in |ii and |f i, draw an external line. Assign a
directed momentum pi for each line, and add an arrow to denote Figure 6.5: Feynman diagrams
contributing at order g 2 to ψψ → ψψ
its charge. scattering, see Eq. (6.32). We show
the explicit momentum assignment
2. Join the external lines together with vertices as in Fig. 6.3. for the external and internal lines.
There are two distinct options.
3. At each vertex, write a factor of
X
(−ig)(2π)4 δ (4) ( ki )
i
P
with i ki is the sum of all momenta flowing into the vertex.
4. For each internal φ line, we write a factor of

Z
d4 k i
(2π)4 k 2 − m2 + iε
and for each internal ψ line, we write

Z
d4 k i
(2π)4 k 2 − M 2 + iε
5. Divide the final result by the overall energy-momentum conserv-

ing δ-function, (2π)4 δ (4) (pf − pi ).
In fact, we can simplify things further and get rid of the trivial
delta-functions: we impose energy-momentum conservation on the
momenta flowing into each vertex and so we change
3.’ At each vertex, write a factor
(−ig)
4.’ Each contracted internal line represents a propagator (or M 2 for

internal ψ lines)
i
k2 − m2 + iε
k
This is ok for graphs like the ones we have been considering.
However, there are also diagrams with closed loops where energy-
momentum conservation at the vertices does not fully fix all the
internal momenta! For example the matrix element obtained from p p+k p
Figure 6.6: Diagrammatic repre-
sentation of Eq. (6.38) where the
blue arrows show the momentum
assignments.
hp| : ψ † (x1 )ψ(x2 )ψ(x1 )ψ † (x2 )φ(x1 )φ(x2 ) : |pi (6.38)
This means, we need to keep the factor

Z
d4 k
p
(2π)4
Similarly for the fully contracted term
h0| : ψ † (x1 )ψ(x2 )ψ(x1 )ψ † (x2 )φ(x1 )φ(x2 ) : |0i (6.39)

k
neither p nor k is constrained (not fixed by external momenta) and
we must integrate over both momenta. Thus we must add a rule that
R d4 k
we integrate with (2π) 4 over each internal line whose momentum
k is unconstrained . p+k
Figure 6.7: Diagrammatic repre-
sentation of Eq. (6.39) where the
blue arrows show the momentum
assignments.
Let us summarize our simplified and final Feynman rules to calcu-

late iAf i which we defined as hf |S − 1|ii = i Af i (2π)4 δ (4) (pf − pi )
1. For each particle in |ii and |f i, draw an external line. Assign a

directed momentum pi for each line, and add an arrow to denote
the flow of charge.
2. Join the external lines together with vertices as in Fig. 6.3.
3. At each vertex, write a factor of
(−ig)
and impose 4-momentum conservation.
4. For each internal φ line, we write a factor of

i
k2 − m2 + iε
and for each internal ψ line, we write
i
k2 − M 2 + iε
5. For each loop with momentum k unconstrained write a factor of

Z
d4 k
(2π)4
6.5.2 Revisiting ψψ → ψψ scattering

Let us apply the above Feynman rules to compute the amplitude of
ψψ → ψψ scattering at order g 2 . The two diagrams contributing to
this process at this order are and we obtain using the Feynman rules
Figure 6.8: Feynman diagrams

contributing at order g 2 to ψψ → ψψ
p1 q1 p q2 scattering. We show the explicit
1
momentum assignment for the
k k external and internal lines. There are
two distinct options.
p2 q 2 p2 q1

2 i i
iA = (−ig) 2 2
+ (6.40)
(p1 − q1 ) − m (p1 − q2 )2 − m2
The interpretation of these diagrams is the following: the nucleons ψ
exchange a meson φ which has momentum k = p1 − q1 = p2 − q2 . This
meson does not satisfy the usual energy momentum relation, since as
we saw above k 2 = −2p2 (1 − cos θ) < 0 and therefore in particular
k 2 6= m2 ! The meson φ is called a virtual particle and which is
off-shell.25 Heuristically we can say, it cannot live long enough for 25
k2 = m2 would be on-shell.
its energy to be measured to great accuracy.
6.6 More scattering processes in the ψ̄ψφ theory

Let us follow Feynman’s motto and let us ’shut up and calculate’ a
couple of sample processes.
6.6.1 Nucleon anti-nucleon scattering: ψ(p1 ) + ψ̄(p2 ) → ψ(q1 ) +

ψ̄(q2 )
There are two diagrams contributing to this process. Note the
opposite flow of the arrows on the Feynman-lines indicating the
difference between a nucleon and an anti-nucleon. We obtain,
p1 q1 Figure 6.9: Feynman diagrams

p1 q1 k contributing at order g 2 to ψ(p1 ) +
ψ̄(p2 ) → ψ(q1 ) + ψ̄(q2 ) scattering.
k We show the explicit momentum
assignment for the external and
p2 q2 internal lines. There are two distinct
p2 q2 options.

2 i i
iA = (−ig) + (6.41)
(p1 − q1 )2 − m2 (p1 + p2 )2 − m2
Note, since the diagrams are simply a shorthand for matrix elements
of operators in the Wick expansion, the orientation of the lines
inside the graphs have absolutely no significance. In the second
contribution, the virtual meson can go on-shell for certain values of
the external momenta. In the center of mass frame p1 = −p2 and

the 2nd denominator is
1 1
=
2
(p1 + p2 ) − m 2 4(M + p21 ) − m2
2
which can lead to a divergence if m > 2M for certain values of p1 .

This will be fixed if one adds higher order corrections to the Feyn-
man graph in the form of an imaginary contribution (the width) and
we will find a bump if we scan the momenta p1 in an experiment.
6.6.2 Meson-meson scattering φφ → φφ

For φφ → φφ, the simplest diagram appears only at g 4 . One of the
momenta remains unconstrained
Figure 6.10: Feynman diagram

p1 k q1 q1 contributing at order g 4 to φφ → φφ
scattering. We show the explicit
momentum assignment for the
k q 1 + p1 k external and internal lines.
p2 k + q2 q2
and we need to integrate over it,

Z
d4 k i i
iA =
(2π) k − M + iε (k − q1 ) − M 2 + iε
4 2 2 2
i i
×
(k − q1 + p1 ) − M + iε (k + q2 ) − M 2 + iε
2 2 2
This is a one-loop diagram and the integration of k µ can require

some additional insights as diagrams can be divergent. This was a
serious problem in the early days of quantum field theory but we
now understand it along the lines of the discussion in Sec. 5.1.2. In
the example at hand we are fine though, since the integration is
R
convergent, since iA ∼ d4 k/k 8 for very large k µ .
6.6.3 Symmetry factors

Are there combinatoric considerations in associating operators with
Feynman diagrams? Write the expansion of Eq. (6.20)
Z
4
S = T exp −i d x HI (x)
in powers of the interaction

∞
X
S= S (n) (6.42)
n=0
X∞ Z
(−i)n
= d4 x1 . . . d4 xn T {HI (x1 ) . . . HI (xn )} (6.43)
n=0
n!
Generally, we might expect the factor 1/n! to get canceled when

we carry out the Wick expansion, because we can permute the
vertices to find a different contraction that gives the same normal
ordered operator. But this counting can change if the diagrams have
symmetries, e.g. permutations of the vertices that do not give rise
to a new contraction!
Let us consider three Klein-Gordon scalar fields φ1 , φ2 , φ3 with
masses m1 , m2 , m3 and an interaction
HI = λ φ1 (x)φ2 (x)φ3 (x)
and discuss the diagram associated with the contraction
(φ1 φ2 φ3 )(φ1 φ2 φ3 )(φ1 φ2 φ3 ) (6.44)
There are 3! contractions associated with this diagram, and the 1/3!
1 Figure 6.11: A λ3 diagram in the

λ φ1 (x)φ2 (x)φ3 (x)-theory.
2
3
1
2
3
in front of the S (3) gets completely canceled.

We now need to add to our list of rules associating operators with
diagrams, we may need to add one more: Divide by the symme-
try factor of the graph. The symmetry factor is the number of
permutations of the vertices that leave the contraction unchanged. We will give a precise definition
In the λ φ1 (x)φ2 (x)φ3 (x)-theory, this consideration will give the below.
complete symmetry factor. In general, another contribution to the

symmetry factor of a graph occurs if there are identical fields at a
single vertex.
Let us illustrate this with in λ φ3 (x)-theory, where we consider a
single real KG-scalar field of mass m with the interaction
λ 3
HI = φ (x)
3!
The factor 1/3! is included since there are usually 3! different ways
of contracting the fields of a vertex with the fields of neighboring
vertices, and so the 1/3! gets canceled.
However, the 1/3! does not always get completely canceled be-
cause permutations of the lines might not give a new contraction.
We must therefore also include in the symmetry factor of a graph
the number of permutations of lines that leave the contraction
unchanged.
You can convince yourself, that we can put these considerations in a

simple formula for the symmetry factor S(G) of a graph G
n!(η)n
S(G) =
r
where n is the number of vertices, η comes from the coupling con-

λ 4 λ 3
stant and is η = 4! in 4! φ theory and η = 3! in 3! φ theory, and r is
the multiplicity of the diagram. The multiplicity r is determined by
labeling all vertices with three (or four) lines emerging. We assume
all these lines to be distinguishable. The multiplicity r is then the
total number of ways to connect external points and vertices to form
the diagram .26 26
If a diagram is direction sensitive
(charge flow), the we take this
To get the correct overall constant for a diagram, we divide by its
into account by only including the
symmetry factor. number of ways to draw the diagram
given the directional conditions of
λ λ the external points.
Examples: We will use the 3! φ3 (x) and 4! φ4 (x) theory
• No symmetry since all three external lines are distinct:
S(G) = 1
which comes about from

λ
hq1 q2 | : φ(x)φ(x)φ(x) : |p3 i
3!
λ
= 3!hq1 q2 | : φ(x)φ(x)φ(x) : |p3 i
3!
we can contract each of the three field operators with each of the
external fields. Let us check by calculating
1!(3!)1
S(G) = =1 (6.45)
3!
where η = 3! and r = 3! since there are 3! ways to connect the
external points to the 3 lines emerging from the vertex.
λ 3
• This is a 2nd order graph in 3! φ theory:
2!(3!)2
S(G) = =1
(3 · 2)2 2
λ 4
• This is a 2nd order graph in 4! φ theory.
2!(4!)2
S(G) = =2
8·3·4·3·2
with r = 8 · 3 · 4 · 3 · 2 because we can connect the first external
point to 8 different lines emerging from the two vertices, after
that we can use the remaining 3 lines and the 4 · 3 lines from the
other vertex. The internal propagator is then fixed apart from a
two-fold ambiguity.
λ 4
• This is a 2nd order graph in 4! φ with two loops
2!(4!)2
S(G) = =6
8·4·3·2
with r = 8 · 4 · 3 · 3 because there are 8 · 4 ways to connect to the
external points and once we choose one of the internal lines we
have first 3 options, then 2 to connect it to the other vertex.
λ 3
• A 4th order 3! φ graph:
4!(3!)4
S(G) = =1
12 · 2 · 9 · 2 · 6 · 2 · 3 · 2
Most people never need to evaluate a diagram with a symmetry
factor greater than 2, so you should not worry too much about this
technicality.
6.7 Potentials
Let us look at the non-relativistic limit of our scattering amplitude
with the goal to treat it as a potential in non-relativistic quantum
mechanics.
6.7.1 The Yukawa Potential

We first re-examine the problem of a classical Klein-Gordon field
in the presence a point source. We will just repeat the steps of
Sec. 4.3.2 where we now have to solve the static Klein-Gordon
equation
( + m2 )φ(x) = δ 3 (x)
we can invert this and simplify since δ 3 (x) is time-independent

1
φ(x) = δ 3 (x)
−∇2 + m2
We solve again using the Fourier transformation of the δ-function
Z
d3 p 1
φ(x) = eix·p
(2π) −∇ + m2
3 2
Z
d3 p 1
= eix·p
(2π)3 p2 + m2
Z
1 1 ∞ eipr − e−ipr
= 2
dp p
4π ir 0 p2 + m2
Z ∞
1 dp eipr
= p 2
2πr −∞ 2πi p + m2
where we have skipped in the 2nd and 3rd line the steps which we
have derived carefully in Sec. 4.3.2. We close the contour in the
upper half plane p → +i∞ where we pick up the pole at p = +im,
which finally gives
1 −m r
φ(x) = e (6.46)
4πr
If you compare this to the ∼ 1/r of Coulomb’s law Eq. (4.95) you see
that the field now dies off exponentially at distances > 1/m, which is
the Compton wavelength of the meson.
Can we now understand the profile of the φ field as a force be-
tween ψ particles? Recall, that we treat the electro-static potential
sourced by a δ-function as the potential energy for another charged
(test) particle moving in this background.
Is there a classical limit of the scalar Yukawa theory with the ψ
particles as δ-function sources for φ, creating the Yukawa potential?
The answer is yes, at least in the limit M m, that is, when the
nucleons are much heavier than the mesons and act as quasi static
sources.
Recall the Born approximation 27 in non-relativistic quantum me- 27
See e.g. Cohen-Tannoudji, Quan-
chanics: at first order in perturbation theory, we find the amplitude tum Mechanics, Vol. II, Chapter
VIII, especially section B.4
to scatter an incoming state with momentum p of a potential U (r)
into an outgoing state with momentum q
Z
AQM (p → q) = hq|U (r)|pi = −i d3 r U (r)e−i(q−p)·r
We will now compare the non-relativistic potential scattering with

that from the first diagram in Eq. (6.40) in the center of mass frame
i i
iA = −g 2 2 2
= g2 (6.47)
(p1 − q1 ) − m |p1 − q1 |2 + m2
where we have used the fact that in the center of mass frame the
energies of the incoming and scattered particles are the same, see
Eq. (6.33).
We need to account for the different normalizations of the rel-
ativistic and non-relativistic amplitudes and divide the result by
(2M )2 to account for the difference between relativistic and non-
relativistic normalization of states, see Sec. 5.4. We define the dimen-
sionless quantity
g
λ≡
2M
and obtain
Z
λ2
d3 r U (r)e−i(q1 −p1 )·r = − 2
|p1 − q1 | + m2
Inverting the Fourier-transform gives

Z
d3 p −ip·r 1
U (r) = −λ2 e
(2π)3 p2 + m2
This is exactly the integral, that we already calculated in the begin-

ning of the section and we find
λ2 −mr
U (r) = − e
4πr
This is the Yukawa potential, where the minus sign tells us that it
is attractive. Yukawa made this potential28 the basis for his theory 28
To be precise: he used the fermion-
of the nuclear force and worked backwards from the range of the scalar version of this interaction
which however gives the same poten-
force (about 1 fm) to predict the mass (about 200 MeV) of the tial.
required meson, the pion.
We can directly from the scattering amplitudes for ψψ → ψψ, Ex: Check this!
ψ ψ̄ → ψ ψ̄, and ψ̄ ψ̄ → ψ̄ ψ̄ that the sign of the spin-0 generated
Yukawa term is always positive – it leads to a universally attractive
potential. Compare this to the Coulomb potential which is mediated
by the exchange of a spin-1 boson, which can be either attractive
and repulsive. Gravity, which is mediated by a spin-2 field is again
universally attractive.
Notice that quantum field theory has given us an entirely new
view of forces between particles. Rather than being a fundamental
concept, the force arises from the virtual exchange of other particles,
in this case the meson.
6.8 How to calculate observables
We are now able to calculate amplitudes for a set of processes by

evaluating Feynman diagrams,
hf |S − 1|ii = iAf i (2π)4 δ (4) (pF − pI ) (6.48)
but we have yet to calculate a probability, which would be measur-

able. For this, we need to square the amplitudes and sum over all
observed final states. But it looks like we are in trouble here, since
h i2
|hf |S − 1|ii|2 = |Af i |2 (2π)4 δ (4) (pF − pI ) (6.49)
Squaring a δ-function makes no sense. What happened?

As we will see, the problem is that we are not working with
square-integrable states. Instead, our states are normalized to δ (3) -
functions, since they are plane waves which exist at every point in
space-time. Thus, the scattering process is occuring at all points in
space for all time. No suprise that we got divergent non-sense.
6.8.1 Fermi’s Golden Rule
Let us derive the familiar Fermi’s golden rule from Dyson’s formula
of Eq. (6.15). Consider two energy eigenstates |ni and |mi of H0 ,
with En 6= Em . We find in leading order in the perturbation Recall from Eq. (6.5), that the
interaction Hamiltonian in the
interaction picture is
HI = (Hint )I = eiH0 t (Hint ) e−iH0 t
Z t
hm|U (t)|ni = −ihm| dt0 HI (t0 )|ni
0
Z t
0
= −ihm|Hint |ni dt0 ei(Em −En )t
0
Z t
0
= −ihm|Hint |ni dt0 eiωt
0
eiωt − 1
= −hm|Hint |ni
ω
t2
with ω = Em − En . We obtain for the probability Pn→m (t) for the
sin2 (!t/2)
transition |mi → |ni after a time t, 4
!2

2 2 sin2 (ωt/2)
Pn→m (t) = |hm|U (t)|ni| = |hm|Hint |ni| 4
ω2
We show the function in brackets in Fig. 6.12. As you can see, after
a time t most of the transitions happen in a region between energy
eigenstates separated by !
⇡
2π t
∆E ≈
t sin2 (ωt/2)
h i
Figure 6.12: 4 ω2
vs ω.
As we take t → ∞, the function in brackets starts to approach a
δ-function. We get the normalization by
Z ∞
sin2 (ωt/2)
dω 4 = 2πt
−∞ ω2
and therefore
sin2 (ωt/2)
4 → 2πt δ(ω), for t → ∞
ω2
We consider now a transition to a cluster of states with density
ρ(E). In the limit t → ∞, we get
Z
sin2 (ωt/2)
Pn→m (t) = dEm ρ(Em )|hm|Hint |ni|2 4
ω2
→ 2πt |hm|Hint |ni|2 ρ(En )
We finally obtain a constant transition probability per unit time

for states around the same energy En ∼ Em = E. This is Fermi’s
golden rule for the transition rate Γn→m = Ṗn→m Note however, according to
Wikipedia: “Although named af-
ter Enrico Fermi, most of the work
Ṗn→m = 2π |hm|Hint |ni|2 ρ(E) (6.50) leading to the Golden Rule is due
to Paul Dirac who formulated 20
years earlier a virtually identical
equation, including the three com-
This was a useful exercise, since it showed us how to carefully ponents of a constant, the matrix
take the limit t → ∞. If we chose to compute the amplitude for the element of the perturbation and
an energy difference. It was given
state |ni at t → −∞ to transition to |mi at t → +∞ we would get
this name because, on account of its
the amplitude importance, Fermi dubbed it ’Golden
Z ∞ Rule No. 2.’ ”. This seems to show
how to avoid the Matthew principle,
−ihm| dt0 HI (t0 )|ni = −ihm|Hint |ni 2π δ(ω)
−∞
but you need to be yourself famous
enough. ’Golden Rule No. 1.’ states
Squaring this, we wold run into the same problem as for the Feynman- the second-order contributions to the
transition probability.
amplitude, since Pn→m (t) = |hm|Hint |ni|2 (2π)2 δ(ω)2 , a non-sensical
infinity. Comparing to the result above, we realize now that the

extra infinity is coming because Pn→m is the probability for the
transition to happen in infinite time t → ∞. We can write these
δ-functions as
(2π)2 δ(ω)2 = (2π) δ(ω) T (6.51)
where T signals t → ∞. We can now divide by T , to derive the

probability per unit time as We used a similar trick, when we
discussed the vacuum energy in
Sec. 5.1.2.
Ṗn→m = 2π |hm|Hint |ni|2 δ(ω)
which after integrating over the density of final states gives us

Fermi’s Golden rule again. We will now interpret the squares of the
δ (4) -functions in Eq. (6.49) as space-time volume factors.
6.8.2 Decay Rates

We now want to obtain the probability for a single particle |ii of
momentum pI to decay into some number of particles |f i with
P
momentum pF = l pfl . This probability is given by
|hf |S|ii|2
P =
hf |f ihi|ii
We have to be careful with the normalization since we work with

states obeying the relativistic normalization convention, see Sec. 5.4.
In particular, with Eq. (5.50), we obtain
hi|ii = (2π)3 2Ep δ (3) (0) = 2EpI V
and
Y
hf |f i = 2Epl V
final states
Let us place our initial particle at rest pI = 0 and EpI = mi , we

obtain for the probability for the decay
1
P = |hf |S|ii|2 × (6.52)
hf |f ihi|ii
1 Y 1
= |Af i |2 (2π)4 δ (4) (pI − pF ) V T ×
2mV 2Epl V
final states
where like in the derivation of the Golden Rule we have replaced

one of the δ (4) -functions with a space-time factor (2π)4 δ (4) = V T .
We have been already computing the amplitude Af i . We can now Recall the example of meson decay
get rid of T by calculating the transition rate, which we get by φ → ψ̄ψ of Eq. (6.25)
taking the T derivative (or just dividing by T ). One worry is still the hf |S|ii = −ig (2π)4 δ (4) (p − q1 − q2 )
sum over all final states. We proceed in two steps: first we put our = iAf i (2π)4 δ (4) (p − q1 − q2 )
theory in a box with side L assuming periodic boundary conditions and therefore
which allows us to count states. Then we integrate over all possible Af i = −g
momenta of all final states. in leading order.
As we discussed in the introduction, in a box of dimension L3

with periodic boundary conditions, the allowed momentum values
are of the form
2πni
ki = , with ni = 0, ±1, ±2, . . . , and i = x, y, z
L
which implies a density of states (L/(2π))3 in momentum space.
In the real world, no experimentalist can measure a cross-section
with exact momenta assignments. We can only measure all states
about some small region ∆k in momentum space. In a region of size
∆kx ∆ky ∆kz , there are
3
L V
∆kx ∆ky ∆kz × = ∆kx ∆ky ∆kz
2π (2π)3
states. The normalization of our momentum

The number of final one-particle states in an infinitesimal momen- states and the field expansion also
changes since for the discrete system
tum interval is
1 X −ik·x
δ (3) (x) = e
V V k
d3 k
(2π)3 which means
d3 p 1
Z h i
and we find for the density of final states that the factors of V φ(x) = 3
p ap e−ip·x + a†p eip·x
(2π) 2Ep
cancel
1 X 1 h i
→ √ p ap e−ip·x + a†p eip·x
V k 2Ep
Y d3 pl 1
dΠ = (2π)4 δ (4) (pI − pF ) (6.53) The reason, the √1 appears, is
V
(2π)3 2Epl because we have chosen a discrete
final states
normalization
[ak , a†p ] = δk,p

and find a Lorentz-invariant quantity, since the 1/(2Epl ) conspire
and φ therefore satisfies
to give us a Lorentz-invariant measure. This gives us our final
[φ(x, t), φ̇(y, t)] = iδ (3) (x − y)
expression for the decay probability per unit time Γ = Ṗ ,
We recover the continuum as in the
introduction with
X Z
1 L
Z
|Af i |2 dΠ
X
Γ= (6.54) → dk1
2m k1
2π
final states
V
X Z
→ 3
d3 k
k
(2π)
where the sum is over final states with different numbers and types 1/2
1
of particles.After a time t, the probability that the particle has not ak → ak
V
decayed is just e−Γt . We call Γ the width of the particle. It is the
inverse of the half-life τ = 1/Γ. If we consider the uncertainty
principle, we can see this: the particle exists for a time τ , any mea-
surement of its energy (or mass in the rest-frame) must be uncertain
by ∼ 1/τ = Γ. Therefore, a series of measurements of the particle
mass will have a characteristic spread of order Γ.
Since we will need it later, we will also derive a general expression

for the probability per unit time using Eq. (6.52)
Y 1
Ṗ = |Af i |2 V · · dΠ (6.55)
2Epi V
initial states
where we have used Eq. (6.53) for the phase space of the final-states.
6.8.3 Example: two-body phase space

For two particles, there are six integrals in d3 p1 d3 p2 but four of these
variables are constrained by 4-vector conservation δ (4) (p1 + p2 − pI ),
leaving only two variables to integrate over.
d3 p1 1 d3 p2 1
dΠ = (2π)4 δ (4) (p1 + p2 − pI )
(2π)3 2Ep1 (2π)3 2Ep2
In the center of mass frame, pI = 0 and EI ≡ ET which is total
energy in the process.
d3 p1 1 d3 p2 1
dΠ = (2π) δ(E1 + E2 − ET )(2π)3 δ (3) (p1 + p2 )
(2π)3 2Ep1 (2π)3 2Ep2
d3 p1 1
= (2π) δ(E1 + E2 − ET )
(2π)3 4Ep1 Ep2
1
= p2 dp1 dΩ1 (2π) δ(E1 + E2 − ET )
(2π)3 4Ep1 Ep2 1
We now need to convert the δ-function of energy into a constraint
on p1 . Using the notation of the margin note, we have g(p1 ) = We use again that
P p
E1 (p1 ) + E2 (p2 ) = i p2i + m2 and X δ(x − xi )
δ(g(x)) =
i
|g 0 (xi )|
2
X 1 1 p1 p1
g 0 (p1 ) = p (2p1 ) = + where the sum extends over all roots
i=1 p1 + m2 2
2 E1 E2 of g(x) which are assumed to be
simple. For example
p1 (E1 + E2 ) p1 ET
= = 1 h i
E1 E2 E2 E2 δ x2 − α 2 =

δ (x + α)+δ (x − α) .
2|α|
where we have used p1 = −p2 .
We obtain finally for the two-body phase space in the center of
mass frame,
1 p1
dΠ2 = dΩ (6.56)
16π 2 ET
Here we have assumed that the particles A and B in the final state
are distinguishable and we treated the final states |A(p1 ), B(p2 )i and
|A(p2 ), B(p1 )i as distinct. Note, that if the final states are instead
identical, we would have double-counted by 2!. In general, we need
to multiply an n-body phase space of n identical particles in the final
states by a factor of
1
dΠn → dΠn
n!
6.8.4 Examples: particle decays

Let us apply this to an example. We assume that m > 4M 2 and the
ψ†
pion to two nucleon decay
φ
φ → ψ̄ψ
is kinematically allowed. There is only one diagram at leading order ψ † ψφ

in perturbation theory, see Eq. (6.25), ψ
iA = −ig
Figure 6.13: Pion decay into two
nucleons. This is kinematically
allowed in our toy model if m > 4M 2 .
Note, in nature the pion is the
lightest QCD resonance and would
not be able to decay to two nucleons.
and the decay width of φ is therefore

Z
1
Γ(φ → ψ̄ψ) = dΠ |A|2
2m
Z
g 2 1 p1
= dΩ
2m 16π 2 ET
g 2 p1
=
2m 8πm
R
where we have used dΩ = 4π and ET = m. We adapt the result of
√
Sec. 3.5.2, and find p1 = 21 m2 − 4M 2 to obtain
r
g2 1 (2M )2
Γ(φ → ψ̄ψ) = 1− (6.57)
32π m m2
6.9 Cross Sections
In physical experiments, we often collide two beams of particles or a

beam hits a target. From time to time, they hit each other. We then
measure the resulting number of particles incident on a detector.
The fraction of the times (per unit flux) they collide is called
the cross section and we denote it as σ. The incoming flux F is
defined as the number of incoming particles per area per unit time.
The total number of scattering events N per unit time is then given
by
N =F ·σ (6.58)
We can use quantum field theory to calculate σ. In fact, we can

extract even more and we can calculate dσ, which is the differential
cross-section. This dσ is related the probability for a given scatter-
ing process into a solid angle (θ, φ). With E1 and E2 as the energies
of the two incoming particles, we obtain with Eq. (6.55)
differential probability
dσ =
unit time · unit flux
|Af i |2 1
= · · dΠ
2E1 2E2 V F
We now need an expression for the unit flux F . Let us assume for
simplicity that we are in the center of mass frame of the collision.
We have so far considered a single particle per spatial volume V such
that the flux is given in terms of the 3-velocities as This is easy to see: consider a beam
of particles perpendicular to a plane
|v1 − v2 | of area A moving with v. If the
F = density of particles is n, then after a
V
time t, the total number of particles
We therefore get passing through the plane is
N = |v|A · t · n
1 1
dσ = · |Af i |2 · dΠ (6.59) The flux is therefore N/(At) = |v| · n.
4E1 E2 |v1 − v2 |
With our normalization, there is one
particle in the box with volume V , so
You can now use Eq. (6.59) and take your favorite scattering n = 1/V and the flux is
amplitude to compute the probability for particles to end up at
F = |v|/V
various angles.
If we have two colliding beams, the
probability of finding either particle
is 1/V but since the collision can
occur anywhere in the box, the total
flux is
|v1 − v2 |/V 2 × V = |v1 − v2 |/V

Let us look in more detail at 2 → 2 scattering in the center of

mass frame. The 3-velocities are v1 = p1 /E1 and v2 = p2 /E2 =
−p1 /E2 and so

1 1 E1 + E2 p1 ET
|v1 − v2 | = |p1 | + = p1 =
E1 E2 E1 E2 E1 E2
which leads together with the two-body phase space of Eq. (6.56) to
1 1
dσ = · |Af i |2 · dΠ2
4E1 E2 |v1 − v2 |
1 E1 E2 1 pf
= |Af i |2 dΩ
4E1 E2 pi ET 16π 2 ET
1 1 pf |Af i |2
= dΩ
4pi ET 16π 2 ET
dσ 1 1 pf
= |Af i |2 (6.60)
dΩ 64π 2 ET2 pi
where pi and pf are the magnitudes of the three-momenta of the

incoming and outgoing particles.
6.10 Green’s Functions

Often times, we are not interested in questions that can be answered
by calculating scattering cross-sections. We might want to figure out
the non-Gaussianity of density perturbations arising in the CMB
from novel models of inflation or calculate the optical properties of
strange metals. These questions are best addressed with the help of
correlation functions, which we will now learn how to calculate
with Feynman diagrams.
6.10.1 The true vacuum

So far we have approximated the ground state with the ground state
of the free theory H0 |0i = 0. We will now define the true vacuum
of the interacting theory |Ωi. The energy is chosen such that for
H = H0 + Hint
H|Ωi = 0 (6.61)
and we normalize the vacuum to
hΩ|Ωi = 1
We define correlation functions as
G(n) (x1 , . . . , xn ) ≡ hΩ|T φH (x1 ) . . . φH (xn )|Ωi (6.62)
where φH is in the Heisenberg picture of the full theory. We also

call G(n) the Green’s functions of the full theory. How can we
compute these objects using Feynman Diagrams?
Claim: The n-point correlation function is given as
G(n) (x1 , . . . , xn ) = hΩ|T φH (x1 ) . . . φH (xn )|Ωi (6.63)

h0|T φI (x1 ) . . . φI (xn )S|0i
= (6.64)
h0|S|0i
where we the interaction picture operators on the right-hand side
are applied on the vacuum of the free theory |0i and the usual time
evolution operator
Z
4
S = T exp −i d x HI (x) = lim U (t+ , t− )
t+ →+∞
t− →−∞
Proof: The time evolution operator in the interaction picture

Z t Z
0 3 0
U (t, t0 ) = T exp −i dt d x HI (x, t )
t0
has the following properties
U † (t, t0 ) = U (t0 , t), U (t0 , t2 ) = U (t0 , t1 )U (t1 , t2 ), U (t0 , t0 ) = 1
The numerator of the RHS of Eq. (6.64) can be written with

φH (x) = U † (t, t0 ) φI (x, t) U (t, t0 ) (see Eq. (6.12)) as
h0|T φI (x1 ) . . . φI (xn )U (t+ , t− )|0i

= h0|T U (t0 , t1 )φH (x1 )U (t1 , t0 ) U (t0 , t2 )φH (x2 )U (t2 , t0 ) . . .
U (t0 , tn )φH (xn ) U (tn , t0 )U (t+ , t− )|0i
Since we are allowed to commute operators under time ordering T ,

using U (t0 , t1 )U (t1 , t0 ) = 1, we get
h0|T U (t+ , t0 )φH (x1 )φH (x2 ) . . . φH (xn )U (t0 , t− )|0i
Since t± → ±∞ the ordering of the remaining U is independent

of the time-ordering T and they will always appear to the left and
right of the string of operators. We therefore need to figure out the
meaning of the two boundary U operators, e.g.
lim U (t0 , t− )|0i

t− →∞
We consider an arbitrary state |Ψi,
hΨ|U (t0 , t− )|0i = hΨ|UH (t0 , t− )|0i
Where we have used that H0 |0i = 0 and UH is the time evolution

operator of the full theory. Let us insert a complete set of states, Recall from Eq. (6.13), that in the
which we take as energy eigenstates of the full Hamiltonian H = interaction picture
H0 + Hint , U (t, t0 ) = eiH0 (t−t0 ) e−iH(t−t0 )

  and the full time-evolution is
 X 
hΨ|UH (t, −∞)|0i = lim hΨ|UH (t0 , t− ) |ΩihΩ| + |nihn| |0i UH = e−iH(t−t0 )
t− →−∞  
n6=0
X
= lim hΨ| UH (t0 , t− )|ΩihΩ|0i + lim eiEn (t− −t) hΨ|nihn|0i
t− →−∞ | {z } t− →−∞
n6=0
|Ωi
X
= hΨ|ΩihΩ|0i + lim eiEn (t− −t) hΨ|nihn|0i
t− →−∞
n6=0
where we have used Eq. (6.61) in the first term. The second term
can also be simplified. In fact, it vanishes!
P R
The sum n is actually an integral d3 p . . . since the momentum
states form a continuum. We can now use the Riemann-Lebesgue Figure 6.14: The Riemann-Lebesgue
lemma. The function f (x) is multi-
lemma, which states that for any well-behaved function f (x), plied by a rapidly oscillating function.
Z β The product integrates to zero in the
limit of infinite oscillation frequency.
lim dx f (x) eiωx = 0
ω→∞ Proof:
α Z Z
f (x)e−iωx dx =
1 0 −iωx

We can alternatively derive this by shifting the integration slightly iω f (x)e dx

into the imaginary t → −∞(1 + iε), which exponentially suppresses all 1
Z
≤ |f 0 (x)| dx → 0 as ω → ±∞.
energy eigenstates except for the ground-state.29 The above result 29
See e.g. Peskin &|ω| Schroeder,
therefore gives, chapter 4.2
hΨ|UH (t, −∞)|0i = hΨ|ΩihΩ|0i (6.65)
In particular, we see that the vacuum of the free theory and the
vacuum of the full theory are proportional to each other Since the result is true for any state
hΨ|, we have
|0i = hΩ|0i · |Ωi UH (t, −∞)|0i = |ΩihΩ|0i
†
|0i = UH (t, −∞)|ΩihΩ|0i
which justifies the somewhat hand-waving argument in Sec. 6.3.1 to
|0i = |ΩihΩ|0i
use free states, or creation operators acting on the free vacuum |0i,
in scattering processes.
If we insert the result of Eq. (6.65) in the RHS of Eq. (6.64), we find With
h0|S|0i = lim h0|U (t+ , t− )|0i

t± →±∞
h0|T φI (x1 ) . . . φI (xn )S|0i = lim h0|U (t+ , t)U (t, t− )|0i
t± →±∞
h0|S|0i
= lim h0|UH (t+ , t)UH (t, t− )|0i
h0|ΩihΩ|T φH (x1 )φH (x2 ) . . . φH (xn )|ΩihΩ|0i t± →±∞
=
h0|ΩihΩ|ΩihΩ|0i = h0|ΩihΩ|ΩihΩ|0i
= hΩ|T φH (x1 )φH (x2 ) . . . φH (xn )|Ωi
where we have used hΩ|Ωi = 1. This completes the proof of this very
important result which we can write in slightly more explicit form
4
R
h0|T φI (x1 ) . . . φI (xn ) e−i d x HI (x) |0i
hΩ|T φH (x1 ) . . . φH (xn )|Ωi = R
4
h0| e−i d x HI (x) |0i
(6.66)
This expression for the n-point Green’s function G(n) (x1 , . . . , xn )

is ideally suited to doing perturbative calculations. We need only
keep as many terms as needed in the Taylor series expansions of the
exponentials.
6.10.2 Connected diagrams and vacuum bubbles

We will now sketch some further results before we move on to
greener pastures30 . With Dyson’s formula and Wick’s theorem, 30
Spin 1
2
representations!
we know how to calculate the expressions appearing in Eq. (6.66).
Feynman diagrams greatly simplify these calculations. But what is
the meaning of the factor (h0|S|0i)−1 ? What are we dividing the
Quantum Field Theory: Example Sheet 4
Dr David Tong, November 2007 relativity, particles, fields 113
1. A real scalar field with φ4 interaction has the Lagrangian

expression by? We will now sketch that the denominator removes
all vacuum bubbles and1 that we only1need to consider
λ connected
Feynman graphs, L
in = ∂µ φ∂ µ φ − m2 φ2 − φ4
words: (1)
2 2 4!
Use Dyson’s formula and Wick’s theorem Xto show that the leading order contribution
hΩ|T φH (x1 ) . . . φH (xn )|Ωi = connected Feynman graphs
to 3-particle → 3-particle scattering includes the amplitude
p1
(6.67)
where
p a connected
p1/
diagram 2
p / is defined by the
i that every
requirement
2 2 = (−iλ) (2)
line is joined to an external
p3/ leg. (p1 + p2 + p3 )2 − m2
Let
p3
us explain this in φ4 theory. The diagramatic expansion for
h0|S|0i is
Check that this result is consistent with the Feynman rules for the theory. What other
diagrams also contribute to this process?
2
2. Examine None⟨0|S|0⟩ todiagrams
of these in φ4totheory.
order λconnect Identifyand
external particles thewediﬀerent
there- diagrams with
forecontributions
call them vacuum 31
the diﬀerent arising from We
bubbles. an will not show this
application of here
Wick’s , theorem.
31
See Confirm
however Sheet 7, Ex 2!
but the
2 pre-factors of the diagrams consisting of combinatoric and
that to order λ , the combinatoric factors work out so that the the vacuum to vacuum
symmetry factors conspire such that the whole series can be expo-
amplitude is given by the exponential of the sum of distinct vacuum bubble types,
nentiated
⟨0| S |0⟩ = exp ( + + + ... ) (3)
and we find that the amplitude for the vacuum of the free theory to
evolve into itself (the numerator of the RHS of Eq. (6.66))
3. Consider the Lagrangian for 3 scalar fields φi , i = 1, 2, 3, given by
3 h0|S|0i = exp(all distinct vacuum
3 bubbles) 3
! 1 1 ! λ ! 2 2
L= (∂µ φi )(∂ µ φi ) − m2 ( φ2i ) − ( φi ) (4)
i=1
2
The same combinatoric 2
simplification takes place for 8
i=1
the numerator
i=1
evaluating of Eq. (6.66) generic correlation functions. One can show See e.g. Schwartz 7.2.5 or Peskin 4.4.
Compare also to the exponentiation
Show that the Feynman propagator for the free
Xfield theory (i.e. λ = 0) isofof
thethe form
cluster expansion in statistical
h0|T φI (x1 ) . . . φI (xn ) S |0i = h0|S|0i · connected diagrams physics, see script 3.4.1 of SS17 of
the TH4A lecture by yours truly.
⟨0|T φi(x)φj (y)|0⟩ = δij DF (x − y) (5)
where again connected means no vacuum bubbles, or that all the
lines are joined to an external leg. We can now conclude that the
where DF (x − y) is the usual scalar propagator. Write down the Feynman rules of the
division by h0|S|0i has a very cute interpretation in terms of Feyn-
theory. Compute the amplitude for the scattering φi φj → φkFeynman
man diagrams. We only need to calculate the connected
φl to lowest order in λ.
graphs and we do not have to care for the vacuum bubbles. We
therefore obtain as promised Eq. (6.67)
X1
hΩ|T φH (x1 ) . . . φH (xn )|Ωi = connected Feynman graphs
g 3
Here we show another example for HI (x) = 3! φ (x) theory for a
two-point function
4
R
h0|T φI (x1 )φI (x2 ) e−i d x HI (x) |0i =
Z
1 1 1
D12 − g 2 3
D1x Dxx Dyy Dy2 + D12 Dxx Dxy Dyy + D12 Dxy + ... =
4 8 12
D1x Dxy Dyx Dy2

D12
+
= +
where integration over repeated indices is implied and DF (x1 , x2 ) =

D12 . The sum over all graphs (disconnected and connected) in the
numerator is then the sum over all graphs with no bubbles multiply-
ing the sum over the bubbles. In the calculation of hΩ|T φH (x1 ) . . . φH (xn )|Ωi,
we therefore can ignore all bubble diagrams, they are automatically
included by using the real vacuum |Ωi instead of the free one |0i.
6.10.3 Outlook to S-matrices from Green’s functions

How can we related the Grenn’s functions back to S-matrix elements
relevant for scattering processes? The Feynman rules for the Green’s
functions include the propagators for the external legs, which we
have to amputate. Also, the 4-momenta assignment to the external
legs in G(n) is arbitrary and generally not on-shell. In order to
achieve both we can relate the Green’s function in momentum space
G̃(n) (p1 , . . . , pn ) to S-matrix elements
hq1 , . . . , qn |S − 1|p1 , . . . pn0 i

0
n
Y n
Y
n+n0 0
= (−i) (qi2 2
− m + iε) (p2j − m2 + iε) G̃(n+n ) (−q1 , . . . , −qn , p1 , . . . , pn0 )
i=1 j=1
where we take the limit p0 → Ep in all of the external legs. These

prefactors therefore vanish and we only get a non-zero answer for
diagrams contributing to Gn which have the same propagators for
each external leg.
The important step that we have taken here, is that this provides
a systematic approach to dealing with true particle states in an
interacting theory. This formula still applies once we take into
account virtual particles in asymptotic states. This is the formally
correct way to consider scattering and is known as the LSZ reduction
formula. It will be worked out and shown in more detail in your
quantum field theory lecture.
Figure 6.15: Self-energy-like di-

agrams representing 32 gauge-
invariant subsets contributing to the
muon g − 2 at the tenth order, which
I(a) I(b) I(c) I(d) I(e) leads to prediction for the anomalous
magnetic dipole moment a = g−2 2
as
aµ (SM) = 116 591 840 (59) × 10−11

I(f) I(g) I(h) I(i) I(j)
and compare to the measurement
aµ (exp) − aµ (SM) = 249 (87) × 10−11
II(a) II(b) II(c) II(d) II(e) which is a terrifically precise pre-

diction. From Phys.Rev.Lett. 109
(2012) 111808.
II(f) III(a) III(b) III(c) IV
V VI(a) VI(b) VI(c) VI(d) VI(e)
VI(f) VI(g) VI(h) VI(i) VI(j) VI(k)

7
Lorentz Representations
In non-relativistic quantum mechanics we learn that the spin ± 12

states of the electron along a specific direction, are conveniently
described by a complex doublet of states
+
ψ
↑
|ψi = (7.1)
ψ↓
The dynamics of the doublet ψ = hx|ψi is governed by the Schrödinger-

Pauli equation

1
i∂t ψ = (i∇ − eA)2 − eA0 12×2 + µB B · σ ψ (7.2)
2m
with the vector potential Aµ , B = ∇ × A, and the Bohr magneton
e
µB =
2me
describing the strength of the coupling between the B-field and the
electron’s magnetic moment.1 1
Recall the Stern-Gerlach experi-
The σ = (σ1 , σ2 , σ3 ) are the Pauli-matrices which we have ment.
combined into a vector:

! ! !
0 1 0 −i 1 0
σ1 = , σ2 = , σ3 = (7.3)
1 0 i 0 0 1
Now it would seem that we have a problem with rotational invari-
ance, since the σ-vector will not change under rotations contrary
to ψ and B. How do we make sure that the equation is rotationally
invariant? It works because
[σi , σj ] = 2i εijk σk
are the exact same algebraic relations that the generator of infinites-
imal rotations satisfy, see Eq. (7.8) and the discussion below. We
anticipate the result and conclude that the changes induced by
rotations cancel in
B ·σψ
or any other vector field, like ∂i which transforms also as a 3-vector.

So we could have written down a rotationally invariant equation for
ψ with
(1 ∂t − ∂i σi )ψ = 0 (7.4)
It turns out, this equation is Lorentz-invariant, too. We combine
σ µ = (1, σ1 , σ2 , σ3 ) (7.5)
to obtain This is the equation for a Weyl

2-spinor, which is a slightly differ-
σ µ ∂µ ψ = 0 (7.6) ent form of the equation which we
usually associate with the Dirac equa-
This is the Dirac equation! We could try to be bold and add a tion, which describes the dynamics of
a Dirac 4-spinor.
mass term with
(σ µ ∂µ + m)ψ = 0 (7.7)
but this is not a Lorentz invariant equation. Please forget Eq. (7.7)
immediately (and remember only Eq. (7.6)).
In conclusion, we need to understand the intricate Lorentz prop-
erties of spin- 12 particles, discuss their representations. The Dirac
equation and its non-relativistic limit, the Schrödinger-Pauli equa-
tion, will follow.
7.1 The Lorentz algebra

We will now be a bit more rigorous and introduce groups, algebras
and representations. A group consists of a set of elements {gi } and
a rule
gi × gj = gk
The rule is required to be associative and have an identity element.

A representation is a particular embedding of the gi into operators
that act in a vector space. For finite-dimensional representations this Think of e.g. matrices acting on
means an embedding of the gi into matrices. vectors. Technically, the matrix
embedding is the representation,
Every group has a trivial representation r : gi → 1. We dis- not the vectors, even though we are
tinguish this from a faithful representation, where each group not careful usually in making this
distinction.
element has its own matrix.
For the Lorentz-group, which we recall from Eq. (3.58) leaves
ηαβ = ηµν Λµ α Λν β
invariant, we have only discussed the fundamental representation

which acts on 4-vectors
xµ → Λµ ν xν
and if you go back to Sec. 3.3.1, you’ll find examples of this embed-
ding as the set of 4 × 4 rotation and boost matrices. Our goal is now
to find all the representations.
Our goal is now to extract the properties of the group inde-
pendent of one particular representation. We will now focus on
continuously connected groups, which can be described by a set
of coordinates which are real numbers. The easiest approach is to A non-continuous group, is e.g. the
consider infinitesimal transformations: for any group G, group reflection group Z2 , which has the
elements 1, −1
elements g ∈ G can be written as
g = exp(iαig λi )
where the αig are just numbers and the λi are group generators. Gen-
erators λi form an algebra, because we can add and multiply them,
whereas group elements can only be multiplied. They can be ex-
tracted from the group elements using infinitesimal transformations
g = exp(iαig λi ) = 1 + iαig λi + . . .
or through

1 dg
λi =
i dαig αg =0
i
Lie groups are a special class, with an infinite number of ele-

ments but with a finite number of generators. The generators of a Think of the rotation group in
Lie group form the Lie algebra. Lie groups are crucial for the SM 3D, which contains infinitely many
rotations matrices depending on
as it contains the SU (3) × SU (2) × U (1) special unitary groups. The the 3 continuous Euler rotation
Lorentz group is also a (non-compact) Lie group and is sometimes angles, but there are only 3 different
infinitesimal rotations in the x − y,
called O(1, 3), an orthogonal group that preserves vector products x − z, and y − z plane.
with a metric with the (1, 3) signature ηµν = (1, −1, −1, −1).
Let us start with a reminder of the rotation group O(3).2 The 2
We will denote the Lie group with
three generators Ji of the rotation group satisfy capital letters, like SO(3), and the
corresponding Lie algebra with so(3)
[Ji , Jj ] = iεijk Jk (7.8)
when we act on a space-time vector xµ they are represented as

     
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 i 0 0 −i 0
J1 = 
0
, J2 =  , J3 =  
0 0 −i 0 0 0 0 0 i 0 0
0 0 i 0 0 −i 0 0 0 0 0 0
and so e.g. for the group element G1 = exp(iθJ1 )

 
1 0 0 0
0 1 0 0 
G1 = 
0 0
 (7.9)
cos θ sin θ 
0 0 − sin θ cos θ
You can obtain J2 and J3 by straight-forward permutations. How Check that the commutator in
about the Lorentz-boosts? We recall Eq. (3.17) Eq. (7.8) is indeed satisfied!
! ! !
x cosh η sinh η x0
= (7.10)
t sinh η cosh η t0
which means infinitesimally
x → x+ηt
t → t+ηx
The generator of a Lorentzboost in the x direction

     
t t t
 x  x x
     
  =   + iK1  
y  y  y 
z z z
is therefore given by the hermitian matrix

 
0 1 0 0
1 0 0 0 
iK1 =  
0 0 0 0 
0 0 0 0
and analogously
   
0 0 1 0 0 0 0 1
0 0 0 0 0 0 0 0
iK2 = 
1
, iK3 =  
0 0 0 0 0 0 0
0 0 0 0 1 0 0 0
We can now check the commutator between the generators of rota-

tions Ji and the boost generators Ki . We find
[Ji , Kj ] = iεijk Kk (7.11)
What does that mean? This just reflects that the boost generators
Ki transform as a 3-vector under rotations, as expected since the
boost direction is a 3−vector characterized by the relative velocity v i .
We will now do one of the most important calculations in history. Under a rotation we would get for
Compute [K1 , K2 ]. What do we get? the boost group elements U =
exp(iη l Kl ) which are 4 × 4 matrices,
   
0 0 0 0 0 0 0 0 U → G−1 l
i U Gi = (1 − iJi )(1 + iη Kl )(1 − iJi ) + . . .
0 0 −1 0 0 0 0 0 = 1 + iη l Kl + η l [Kl , Ji ] + . . .
K1 K2 − K2 K1 = 
0 0
− 
0 0 0 −1 0 0
and so
0 0 0 0 0 0 0 0
  Kl → i[Ji , Kl ]
0 0 0 0
0 0 −1 0

=  = −i J3
0 1 0 0
0 0 0 0
Two boosts produce a rotation! Summarizing, we find the Lorentz-

algebra SO(1, 3)
[Ji , Jj ] = iεijk Jk (7.12)

[Ji , Kj ] = iεijk Kk (7.13)
[Ki , Kj ] = −iεijk Jk (7.14)
Please note the very important negative sign in the commutator of

two boosts!
Now use a crucial observation: the algebra falls apart in two
independent sets if complexify it using
1 1
Ji+ ≡ (Ji + iKi ), Ji− ≡ (Ji − iKi )
2 2
which satisfies
[Ji+ , Jj+ ] = iεijk Jk+ (7.15)

[Ji− , Jj− ] = iεijk Jk− (7.16)
[Ji+ , Jj− ] =0 (7.17)
This shows that the complexified Lie-algebra for the Lorentz-group

has two commuting sub-algebras. The algebra generated by J ±
is the 3D rotation algebra or so(3) = sl(2, R) = su(2). So we have
shown that
so(1, 3) = su(2) ⊕ su(2)
which means that the Lorentz group SO(1, 3) is locally isomorphic to

two independent SU (2) groups. As one can show, the fact that the
complexifications of the corresponding Lie algebras coincide implies
that the representation theory of both groups coincides.
7.1.1 Representations
We can now simply use our knowledge of SU (2) representations
to determine all representations of SO(1, 3). We know that the
representations are labeled by their spin
j = 0, 12 , 1, 32 , . . . (7.18)
and each SU (2) representation consists of (2j + 1) objects
ψm with m = −j, −j + 1, . . . , j − 1, j
which are rotated into each other by elements of SU (2). We there-

fore know that the irreducible representations of SO(1, 3) are
labeled by (j + , j − ) with j ± taking on the values of Eq. (7.18).
Each SO(1, 3) representation contains (2j + + 1)(2j − + 1) objects
Ψm+ ,m− with
Ψm+ ,m− with m± = −j ± , −j ± + 1, . . . , j ± − 1, j ±
Table 7.1: Decomposition of ir-

SU (2) ⊗ SU (2) (0, 0) ( 12 , 0) (0, 12 ) ( 12 , 12 ) (1, 0) (1, 1) reducible representations of the
Lorentz SU (2) × SU (2) into irre-
1 1
SO(3) 0 2 2 0⊕1 1 0⊕1⊕2 ducible representations of its SO(3)
subgroup describing spin.
We recognize the trivial, one-dimensional representation (0, 0)

as the Lorentz-scalar. Counting the size of the representation, we
conclude that the 4-dimensional representation ( 12 , 12 ) is the Lorentz
vector, which is the defining representation of the Lorentz group.
7.1.2 Spinor representations

How can we understand the representations ( 21 , 0)? We write The name ”left-handed” and ”right-
handed” from the way the spin
precesses as a massless fermion
ψL = ψα , with α = 1, 2 (7.19)
moves: we will see this in Sec. 7.3.4.
How about the notation ( 12 , 0) ? This says that Ji+ = 1

2 (Ji + iKi )
acting on ψα is represented by
1
σi
2
while Ji− = 12 (Ji − iKi ) is represented by 0. This means that the

representations of the generators (before complexification) are
1
Ji =
σi
2
1
iKi = σi
2
after solving the simple linear system. The equality here means
”represented by”.
We will similarly denote the ( 12 , 0) with the slightly exotic symbol We use dotted and undotted indices
to make sure we don’t accidentally
contract the α of ( 12 , 0) with an
ψR = χ̄α̇ , with α̇ = 1, 2 (7.20)
α̇ = 1, 2 of ( 12 , 0), which would not be
Lorentz invariant. This is called van
Do not be confused by the bar. For now it is just part of the typo- der Waerden notation. It will prove
graphic symbol. extremely useful in supersymmetry
and in superstring theory.
The two-component spinors ψα and χ̄α̇ are called Weyl spinors
and furnish perfectly good representations of the Lorentz group.
What about the Dirac spinor? If you have already encountered
it, you know that it has also 4 components. Why? The reason is
parity. We know that under parity, see XXXX, that
x → −x
t→t
and similarly for other 3-vectors like the 3-momentum or the boosts
p → −p
K → −K
Rotations are pseudo-vectors and they transform as Think of the angular momentum as
L = r × v or imagine a clock-wise
rotation in a mirrored coordinate
J →J
system.
This means for the complexified generators

1 1
Ji+ = (Ji + iKi ) → (Ji − iKi ) = Ji−
2 2
In other words J + ↔ J − and so under parity we exchange
( 12 , 0) ↔ (0, 12 )
In order to describe an electron in QED which conserves parity,

we need to use both 2-dimensional representations, which we can
describe mathematically
( 21 , 0) ⊕ (0, 12 )
This reducible representation is thus given by stacking two Weyl QED and QCD are parity invari-
spinors to form a Dirac spinor ant, which means the equations of
motions and the Lagrangian are
! ! invariant under P transformations. If
ψL ψα had chosen to write the theory using
Ψ= = just the ( 12 , 0) spinor, we would have
ψR χ̄α̇
explicitly violated P. This is in fact,
what happens in the electro-weak
sector: SU (2)L gauge bosons only
couple to ( 12 , 0) spinors and not to
the (0, 21 ) representation.
The spinor Ψ(p) is (like the scalar) a function of 4-momentum, or a

function of xµ if you Fourier-transform. We know its transformation
properties under rotations
!
1
σ 0
JDirac = 2 1
0 2σ
where the equality means ”represented by”. Similarly for boosts

!
1
2 σ 0
iKDirac =
0 − 12 σ
Note, the very important minus sign, which distinguishes the two
irreducible representations entering a Dirac spinor.
Parity forced us to us a 4-spinor but we know that a electron has
only two physical degrees of freedom. We must project out two of
four the components described by Ψ(p). Let us go to the rest-frame
(pµ )r ≡ (m, 0)
Anticipating the result, we will understand fully later, we write the

projection operator as You can already guess that γ 0 will
turn out to be one of the γ matrices.
1
P= (1 − γ 0 )
2
For now γ 0 is just a 4 × 4 matrix. A projection applied twice should
be the same as just applying it once
P2 = P (γ 0 )2 = 1
and the eigen-values of γ 0 are therefore ±1. Since we know that

under parity
ψL ↔ ψR
we cannot simply use the projection operator to set one of the fields
(e.g. χ̄α̇ ) to 0. Parity means that the two irreducibles should be on
the same footing. We choose the following representation of γ 0
!
0 0 12×2
γ =
12×2 0
or explicitly We could have used also a different

  basis choice.
0 0 1 0
0 0 0 1 
γ0 =  
1 0 0 0 
0 1 0 0
So in the rest-frame, we can write the projection to two degrees of

freedom as
P Ψ(pr ) = 0
which is the same as
ψL − ψR = 0 (7.21)
Since this is an equation equating the two-spinors, it removes two of

the four degrees of freedom contained in the Dirac spinor.
Eq. (7.21) does not look Lorentz invariant since it equates states
in two different representations. Note however, that we required a
specific frame (the rest frame) and therefore should not respect the
resulting equation to be covariant. Once we boost to a generic frame,
we will find that the projection equation is covariant.
7.1.3 The Dirac equation

We have ”derived” the Dirac equation above. Why? We know now
how to obtain the equation which is satisfied by Ψ(p) of a general pµ
by Lorentz-transforming the spinor! We boost
Ψ(p) = e−iη·K Ψ(pr )
which implies for the projection equation for a general pµ

e−iη·K γ 0 eiη·K − 1 Ψ(p)
we define
γ µ pµ
≡ e−iη·K γ 0 eiη·K
m
to obtain the Dirac-equation
(γ µ pµ − m)Ψ(p) = 0 (7.22) Figure 7.1: A memorial stone in

Westminster Abbey in the nave near
Newton’s monument. According to a
which we can write in coordinate space as physics legend, Dirac was staring into
a fire on evening in 1928 when the
equation occurred to him.
(iγ µ ∂µ − m)Ψ(x) = 0 (7.23)
which is a first-order differential equation for the Dirac 4-spinor.

We can expand on the last steps by noting that Ex: Show this!
!
0 exp(−η · σ)
e−iη·K γ 0 eiη·K =
exp(η · σ) 0
and with the unit vector η/|η| we get Ex: Show this, too! Use {σi , σj } =
σi σj + σj σi = 2δij
η
exp(η · σ) = cosh |η| + · σ sinh |η|
|η|
If we identify the relation between boost and the 3-momentum and
the relevant boost as Ex: Show this, too! You can imme-
diately convince yourself, that this is
η the correct result in 1 + 1 dimensions,
p=m sinh |η| (7.24)
|η| since
We can then easily see that pr = (m, 0) → (p0 , p1 )
! which requires a boost: p1 =

i
0 σ m sinh(η) + 0 · cosh(η) which is
γi = Eq. (7.24) in 1+1 space-time dimen-
−σ i 0
sions.
The Dirac matrices satisfy the anti-commutator relations {A, B} = AB + BA

{γ µ , γ ν } = 2g µν 14×4 (7.25)
The γ-matrices generate the Dirac algebra, which is a special

case of a so-called Clifford algebra. This particular basis is the the
Weyl-representation.
We additionally define
σ µ ≡ (1, σ), σ̄ µ ≡ (1, −σ) (7.26)
and we can write the Dirac matrices in the Weyl or chiral basis as
!
µ 0 σµ
γ = (7.27)
σ̄ µ 0
We have found a non-trivial representation of the Lorentz group.

Note however that the representation is non-unitary! The complexi-
fied algebra is unitary but it will come with imaginary coefficients.
You can see that most easily by looking at an example of a boost
in the z-direction for ( 12 , 0)-representation which generated by K3
!
σ3 1 −i 0
iK3 = K3 =
2 2 0 i
which means that the group g3 (η3 ) ∈ SO(1, 3) element is

η
!
e2 0
g3 (η3 ) = exp(iη3 K3 ) = −η
0 e 2
This is not a unitary matrix, since g3† g3 6= 1. This is expected since

the Lorentz group is not compact.3 and therefore there are no 3
The rotational subgroup is
finite dimensional unitary representations of the Lorentz group. We compact since it is parametrized by
angles θi which lie on a circle S 1
will have to make use of projective representations and work with a with 0 and 2π identified (there is
so-called little group, but this is for your next QFT course.4 more to this but since the topology
of SO(3) is a ball with antipodal
We can find a Lorentz-covariant form of the Lorentz-algebra in surface points identified). Boosts
Eq. (7.14) by generalizing the rotation group generators and writing however are parametrized by η = v/c
them as differential operators with 0 ≤ η < 1 where η = 0 and
η = 1 are not the same and the
group is therefore not compact.
J = x × p = x × (−i∇) 4
See e.g. S. Weinberg, QFT1, chap-
ter 2.
which we can also write as an anti-symmetric tensor with J 3 = J 12
and so on:
J ij = −i(xi ∇j − xj ∇i )
The generalization to 4 dimensions is naturally
J µν = i(xµ ∂ ν − xν ∂ µ ) (7.28)
which is an anti-symmetric 4 × 4 matrix, which contains N (N − 1)/2 =

3 · 4/2 = 6 elements. We can identify them with the previously defined
Ji and Ki by
J i → J ij and K i → J 0i
again with J 3 = J 12 and so on. A general Lorentz transformation is

parametrized by the anti-symmetric tensor
ωµν = −ωνµ
and
i µν
Λ = e− 2 ωµν J
The commutation rules of the generators can be determined simply

from the commutators of the differential operators5 5
See Ex Sheet 9!
[J µν , J ρσ ] = i(η νρ J µσ − η µρ J νσ − η νσ J µρ + η µσ J νρ ). (7.29)
You can see that the fundamental representation for J µν acting on

4-vectors is given by
(J µν )αβ = i(δαµ δβν − δβµ δαν ) (7.30)
Compare with the definition of Ki and Ji above to confirm. We can

understand it better by looking at an infinitesimal transformation of
xµ

i
α α µν α
x → δβ − ωµν (J ) β xβ
2
which becomes e.g. in the case of ω23 = −ω32 = θ
 
1
 1  ν
xα → 

x
1 −θ
θ 1
23
or in exponentiated form e−iω23 J , we reproduce Eq. (7.9)
 
1 0 0 0
0 1 0 0 
G1 = 
0 0
 (7.31)
cos θ sin θ 
0 0 − sin θ cos θ
and you can also verify that ω10 = −ω01 = η gives K1 .

The Lorentz-algebra for Dirac spinors can be represented by Or with the other common Dirac
combination
i µ ν
i σ µν = [γ , γ ]
S µν
= [γ µ , γ ν ] (7.32) 2
4 we trivially have
1 µν
S µν = σ
which can be written with Eq. (7.25) as 2

i 0 µ=ν
S µν = [γ µ , γ ν ] = (7.33)
4  i γµγν µ 6= ν
2
i i
= γ µ γ ν − η µν (7.34)
2 2
you can check that Eq. (7.32) is a representation of the Lorentz-
algebra satisfying Eq. (7.29)
[S µν , S ρσ ] = i(η νρ S µσ − η µρ S νσ − η νσ S µρ + η µσ S νρ ). (7.35)
by using Eq. (7.25) repeatedly. So for a given Lorentz-transformation

Λµ ν defined by the parameters ωµν , we have the Dirac-spinor repre-
sentation

i µν
S[Λ] = exp − ωµν S (7.36)
2
with the transformation
Ψ(x) → S[Λ]Ψ(Λ−1 x)
We can again see, that the Lorentz-transformations are not unitary,

since the the exponent is not hermitian6 because 6
For general unitary matrix
U = exp(iA)
i
(S µν )† = − [(γ ν )† , (γ µ )† ] in order to satisfy
4
i U †U = 1
= [(γ µ )† , (γ ν )† ]
4
we see
which would be hermitian if all the γ µ are hermitian (γ µ )† = γ µ , U † U = exp(−iA† ) exp(iA)
or all are anti-hermitian (γ µ )† = −γ µ . However this never happens = 1 − i(A† − A) + . . .
because of Eq. (7.25) we need
A = A†
(γ 0 )2 = 1 real eigenvalues
(γ i )2 = −1 imaginary eigenvalues
So we could pick γ 0 to be hermitian, but we can only chose γ i to be

anti-hermitian. Indeed, you can check that the Dirac matrices in the
Weyl or chiral basis of Eq. (7.27) exactly satisfy this
(γ 0 )† = γ 0 , (γ i )† = −γ i
We see that there is no way to pick the γ µ such that the exponent is
hermitian and the representation is therefore non-unitary.
7.1.4 Lorentz-invariant actions

We have now a new field that we can work with, the Dirac spinor Ψ.
We now want to construct a Lorentz invariant action, which will lead
to a Lorentz invariant equation of motion.
Let us try something naive, which will not work. We define the
adjoint spinor as
Ψ† (x) = (Ψ∗ )T (x)

P3
Can we build a Lorentz scalar by contracting Ψ† Ψ = ∗
l=0 (Ψ )l Ψl ?
How does this behave under Lorentz transformations
Ψ(x) → S[Λ]Ψ(Λ−1 x)
Ψ† (x) → Ψ† (Λ−1 x)S[Λ]†
Ψ† (x)Ψ(x) → Ψ† (Λ−1 x)S[Λ]† S[Λ]Ψ(Λ−1 x)

This would be invariant if only S[Λ] was unitary but we in general

have S[Λ]† S[Λ] 6= 1. This means Ψ† Ψ is not a Lorentz-invariant
scalar, and certainly cannot be used in an action.
But we can use our insight in why it fails to find a solution. We
pick a representation of the Clifford algebra which like the Weyl
basis satisfies (γ 0 )† = γ 0 and (γ i )† = −γ i , then we have
γ 0 γ µ γ 0 = (γ µ )†
which means that

i µ † ν †
(S µν )† = [(γ ) , (γ ) ] = γ 0 S µν γ 0
4
such that

i
S[Λ]† = exp ωµν (S µν )† = γ 0 (S[Λ]−1 )γ 0 (7.37)
2
This allows us to define the Dirac adjoint
Ψ(x) = Ψ† (x)γ 0 (7.38)
We now see that ΨΨ is a Lorentz scalar since with Eq. (7.37)
Ψ(x)Ψ(x) = Ψ† (x)γ 0 Ψ(x)

→ Ψ† (Λ−1 x) S[Λ]† γ 0 S[Λ] Ψ(Λ−1 x)
= Ψ† (Λ−1 x)γ 0 Ψ(Λ−1 x)
= Ψ(Λ−1 x)Ψ(Λ−1 x)
which is indeed the tranformation law for a Lorentz-scalar field, see

e.g. Eq. (4.35).
Let us show another important bi-linear: Ψγ µ Ψ is a Lorentz-
vector. We claim, it therefore transforms as
Ψ(x)γ µ Ψ(x) → Λµ ν Ψ(Λ−1 x)γ ν Ψ(Λ−1 x)
and we treat the µ = 0, 1, 2, 3 label on the γ matrices as a true

vector index. We can use this bi-linear to form Lorentz-scalars
by contracting it with other Lorentz-vectors. We suppress the x-
argument for notational simplicity.
Ψγ µ Ψ → Ψ S[Λ]−1 γ µ S[Λ] Ψ
To show that Ψγ µ Ψ transforms as a vector, we need
S[Λ]−1 γ µ S[Λ] = Λµ ν γ ν (7.39)
which we show by working infinitesimally

i
Λ = 1 − ωρσ J ρσ + . . .
2
i
S[Λ] = 1 − ωρσ S ρσ + . . .
2
From Eq. (7.39) we find that we have to show

[S ρσ , γ µ ] = −(J ρσ )µ ν γ ν (7.40)
which holds if we plug in the definition of J ρσ .We use
γ µ γ ν = −γ ν γ µ + {γ µ , γ ν }
= −γ ν γ µ + 2η µν
to obtain for ρ 6= σ with Eq. (7.33) for the LHS
i ρ σ µ
[S ρσ , γ µ ] = [γ γ , γ ]
2
i
= (γ ρ γ σ γ µ − γ µ γ ρ γ σ )
2
i
= (γ ρ {γ σ , γ µ } − γ ρ γ µ γ σ − {γ µ , γ ρ }γ σ + γ ρ γ µ γ σ )
2
= i(γ ρ η σµ − γ σ η µρ ) (7.41)
For the RHS we plug in Eq. (7.30) with (J ρσ )µ ν = i(η ρµ δνσ − η σµ δνρ )
to finally obtain
−(J ρσ )µ ν γ ν = −i(η ρµ γ σ − η σµ γ ρ )
which equals Eq. (7.41) and therefore proves Eq. (7.40).
Similarly, Ψγ µ γ ν Ψ transforms as a Lorentz-tensor. Ex: Show this!
7.1.5 The Dirac Lagrangian

We can now use the Lorentz-covariant bilinears ΨΨ, Ψγ µ Ψ, and
Ψγ µ γ ν Ψ to build Lorentz-invariant action. We choose
Z
S= d4 x Ψ(x)(iγ µ ∂µ − m)Ψ(x) (7.42)
This is the Dirac action. If we vary with respect to Ψ, we get

(iγ µ ∂µ − m)Ψ(x) = 0
which is the beautiful Dirac equation. It is first order in deriva-
tives but miraculously Lorentz invariant. If we tried a similar thing
with a scalar field it would look like v µ ∂µ φ = . . . which necessarily in-
volves a privileged vector v µ in space-time and would not be Lorentz
invariant.
The Dirac equation mixes the various components of Ψ through
the γ µ matrices. Note, each individual component itself solves the
Klein-Gordon equation. We write
(iγ ν ∂ν + m)(iγ µ ∂µ − m)Ψ(x) = (−γ ν γ µ ∂ν ∂µ + m2 )Ψ(x) = 0
with
1 ν µ
γ ν γ µ ∂ν ∂µ = {γ , γ } ∂ν ∂µ = ∂ν ∂ν
2
and so we get
−(∂ν ∂µ + m2 )Ψα (x) = 0
which is the Klein-Gordon equation. We show the spinor index α
explicitly to show that it applies to each component.
7.1.6 The slash

We will introduce some useful notation. Often we contract 4-vectors
with gamma matrices and we write
Aµ γ µ = A
/
and the Dirac equation reads
(i∂/ − m)Ψ = 0
7.1.7 Chiral spinors and the Weyl equation

We write our spinor in the Weyl basis as
!
ψL (x)
Ψ(x) =
ψR (x)
which is in the ( 12 , 0)⊕(0, 21 ) representation. From XXX, we know that

the Weyl spinors ψL (x) in ( 12 , 0) and ψR (x) in (0, 21 ) transform the
same way under rotations
!
exp( 2i θ · σ) 02
S[Λrot ] = (7.43)
02 exp( 2i θ · σ)
but differently under boosts

!
exp(+ 12 η · σ) 02
S[Λboost ] = (7.44)
02 exp(− 21 η · σ)
or in components for rotations

i
ψL/R → exp( θ · σ)ψL/R
2
and for boosts
1
ψL/R → exp(± η · σ)ψL/R
2
which decomposes the Dirac Lagrangian into
L = Ψ(iγ µ ∂µ − m)Ψ
† µ † µ † †
= iψL σ̄ ∂µ ψL + iψR σ ∂µ ψR − m(ψL ψR + ψR ψL ) (7.45)
We see that a massive fermion seems to require both ψL and ψR

since they couple trough the mass term.7 In the massless case the 7
You can also write a mass term
two spinors decouple and we get the the Weyl equations with only one of them called a
Majorana mass, see below.
iσ̄ µ ∂µ ψL = 0
iσ µ ∂µ ψR = 0
We have already seen in the ”derivation” of the Dirac equation in

Sec. ??, that the equation of motion projects out half the degrees of
freedom. The crucial aspect was, that the equation of motion was
first order rather than second order as the Klein-Gordon equation.
In particular the conjugate momentum to the Dirac Lagrangian is

given by
∂L
πΨ = = iΨ†
∂ Ψ̇
which is not proportional to the time-derivative of Ψ. Therefore, the
phase space of the spinor is parametrized by Ψ and Ψ† , while for a
scalar it is φ and π = φ̇. So the phase space is described by 8 real
dimensions and consequently the Dirac spinor has 4 real degrees of
freedom.
7.1.8 γ5
In our choice of basis for the γ µ matrices, the Lorentz-transformations
S[Λ] came out block-diagonal. What happens if we choose a different
representation of the Clifford algebra?
γ µ → U γ µ U −1 , and Ψ → UΨ
S[Λ] will now likely not be block-diagonal. Can we define the Weyl
spinors in an invariant way? We introduce a ”fifth” gamma-matrix8 8
You can convince yourself that we
can also write γ 5 as
γ 5 = iγ 0 γ 1 γ 2 γ 3 γ5 =
i
εµνσρ γ µ γ ν γ σ γ ρ
4!
which satisfies which in the Weyl basis looks like using the totally anti-symmetric four
index tensor with ε0123 = 1.
!
−1 0 γ 5 is called the ”fifth” gamma-
2 2
γ5 = (7.46) matrix, because with ΓA = γ A for
02 12 A = 0, 1, 2, 3 and Γ4 = iγ 5 it satisfies
the five-dimensional Clifford algebra
{ΓA , ΓB } = 2η AB
µ 5 5 2
{γ , γ } = 0, and (γ ) = +1
As you will show in the exercises
[Sµν , γ 5 ] = 0
which means that γ 5 is a scalar under rotations and boosts. This

is another manifestation of the fact, that the Dirac representation
is reducible, since eigenvectors of γ 5 whose eigenvalues are different
transform without mixing9 9
This criterium for reducibility is
We can therefore define Lorentz-invariant projection operators also known as Schur’s lemma.
1
PL/R = (1 ∓ γ 5 )
2
such that
PL2 = PL , PR2 = PR , PL PR = 04
with Eq. (7.46), we see that PL/R projects into the Weyl-basis
ψ L = PL Ψ
ψ R = PR Ψ
7.2 Discrete symmetries
7.2.1 Parity
We already saw that under parity P the left-handed and right-
handed spinors are exchanged
P : ψL ↔ ψR
Using this, we can write the action of parity on the Dirac spinor
itself as
P : Ψ(x, t) → γ 0 Ψ(−x, t)
Notice that the Dirac Lagrangian conserves parity as you can easily
see from Eq. (7.45). Let us now discuss how our interaction terms
change under parity. The ”mass” bi-linear transforms as
P : Ψ(x, t)Ψ(x, t) → Ψ(−x, t)Ψ(−x, t)
thanks to (γ 0 )2 = 1. This is the transformation of a scalar field. How

about the vector Ψγ µ Ψ? We split in temporal
P : Ψ(x, t)γ 0 Ψ(x, t) → Ψ(−x, t)Ψ(−x, t)
and spatial components
P : Ψ(x, t)γ i Ψ(x, t) → Ψ(−x, t)γ 0 γ i γ 0 Ψ(−x, t)

= −Ψ(−x, t)γ i Ψ(−x, t)
which shows that Ψγ µ Ψ is a vector. We can use γ 5 in the bilinear to

form another Lorentz scalar and vector
Ψγ 5 Ψ, and Ψγ 5 γ µ Ψ
We find with {γ 5 , γ µ } = 0 that
P : Ψ(x, t)γ 5 Ψ(x, t) → Ψ(−x, t)γ 0 γ 5 γ 0 Ψ(−x, t)

= −Ψ(−x, t)Ψ(−x, t)
it is a pseudo-scalar and for the Ψγ 5 γ µ Ψ bilinear


 − Ψ(−x, t)γ 5 γ 0 Ψ(−x, t)
P : Ψ(x, t)γ 5 γ µ Ψ(x, t) →
 Ψ(−x, t)γ 5 γ i Ψ(−x, t)
we find that it is a pseudo-vector or axial-vector. Like the rotation vector Ji or the

angular momentum L = x × p.
In summary we find the following bi-linears in Table 7.2.1 and their
properties with
Ψ(x) Γ Ψ(x)
for various Γ matrices. The total number of bi-linears is 16, which What about Γ = γ [µ γ ν γ σ] or
is exactly what we could hope for from a 4 × 4 component object. Γ = γ [µ γ ν γ σ γ ρ] ? The brackets
indicate anti-symmetrization, since
We have now a set of terms to build covariant Lagrangians with. we do not need to consider the
If we add γ 5 to our terms then these terms will typically break symmetric part which is ∼ η µν (anti-
symmetrization also explains why we
consider at most 4-tensors). One can
show that γ [µ γ ν γ σ] = −iεµνσρ γρ γ 5
and γ [µ γ ν γ σ γ ρ] = −iεµνσρ γ 5 and
so they are linearly related to the
simpler bilinears with Γ = γ ρ γ 5 and
Γ = γ5.
Γ type number of bi-linears
1 scalar 1
γµ vector 4
σ µν = 2i [γ µ , γ ν ] tensor 6
γµγ5 pseudo-vector 4
γ5 pseudo-scalar 1
16
parity. The electro-weak sector of the SM treats the ψL and ψR

fields differently and they have different quantum numbers. They
do not describe the same particle, e.g. the left-handed electron eL
which couples to W -bosons of the SU (2)L gauge interactions is
only distantly related to the right-handed electron eR which does
not. A theory which treats ψL and ψR on the same footing is called In fact they are only related through
a vector-like theory. A Lagrangian in which ψL and ψR appear a scalar coupling via the Higgs h
differently, is called a chiral theory. QCD and QED are vector-like, L ⊃ e†L heR
and the electro-weak sector including the Higgs are chiral theories.
7.3 Solutions of the free Dirac equation
Let us study the solutions of the free Dirac equation

!
−m σ µ pµ
/ − m)Ψ(p) =
(p Ψ(p) = 0 (7.47)
σ̄ µ pµ −m
We start with a simple ansatz
Ψ = u(p)e−ip·x , with p2 = m2
where the on-shell condition is motivated because the components

satisfy the Klein-Gordon equation. For now we focus on positive
frequencies, such that p0 > 0. The column vector u(p) needs to
satisfy the additional constraint
(γ µ pµ − m)u(p) = 0
We can easily analyze this equation in the rest frame, where pr =

(m, 0). We get the general solution by boosting with S[Λ]. In the
rest frame
!
0 −12 12
(mγ − m)u(pr ) = m u(pr ) = 0
12 −12
the solutions are

!
√ ξ
u(pr ) = m
ξ
for any two-component spinor ξ which is just a number. We conven-

tionally normalize
ξ†ξ = 1
√
and we added m for future convenience. The interpretation for the
ξ two spinor is given by the transformation under rotations, where
ξ transforms as an ordinary two-component spinor of the rotation
group, see Eq. (7.43). We have
ξ → exp( 2i θ · σ) ξ
and ξ therefore determines the spin orientation of the Dirac solution,

as usual. E.g. when
!
1
ξ=
0
then the particle has spin up (↑) along the z-direction. Notice again,
that after applying the Dirac equation we are free to choose only 2 of
the four components u(p).
Let us now apply a boost in the z-direction
! !
E m cosh η
=
p3 m sinh η
Now apply the same boost to u(p). With Eq. (7.44) we get Show this!
√ !
p·σξ
u(p) = √ (7.48)
p · σ̄ ξ
with p = (E, 0, 0, p3 ) and where the square-root of a matrix is

understood as taking the positive root of each eigenvalue. Let us
show that this satisfies the Dirac equation. With u(p)T = (u1 , u2 ) we
can write the Dirac equation Eq. (7.47)as
(p · σ)u2 = m u1 (7.49)
(p · σ̄)u1 = m u2 (7.50)
We can see that either of the two equations implies the other using
(p · σ)(p · σ̄) = p20 − pi pj σ i σ j = p20 − pi pj δ ij = pµ pµ = m2
Let us try the ansatz u1 = p · σξ 0 for some ξ 0 then Eq. (7.50) immedi-
ately gives
u2 = mξ 0
We find that any spinor of the form

!
(p · σ)ξ 0
u(p) = A
mξ 0
with constant A would be solution. To make this more symmetric,

√
we choose A = 1/m and ξ 0 = p · σ̄ ξ with constant ξ. Then
√ √
u1 = (p · σ) p · σ̄ = m p · σ̄ξ
7.3.1 Negative energy solutions

We get another solution to the Dirac equation with the ansatz
Ψ = v(p) e+ip·x
The u(p) solution oscillates in time as Ψ ∼ e−iEt and is called a

positive frequency solution. The v(p) solutions oscillates with
Ψ ∼ e+iEt and are called negative frequency solutions. Both are
solutions to the classical Dirac equation and note that both have
positive energy. The Dirac equation requires that the 4-component
spinor v(p) satisfies
!
m σ µ pµ
v(p) = 0
σ̄ µ pµ m
which with the above we see is solved by
√ !
p·σχ
v(p) = √ (7.51)
− p · σ̄ χ
with some normalized (χ† χ = 1) 2-component spinor, taken to be a

constant.
7.3.2 Examples
We take again the positive frequency solution with mass m at rest
p=0
!
√ ξ
u(pr ) = m
ξ
Consider ξ T = (1, 0) we boost to (E, 0, 0, p3 ) to get

 !  !
√ 1 p 1
 p·σ   E − p3 
 0   0 
u(p) = √
! =  !
 1  p
 1 
p · σ̄ E + p3
0 0
This expression in fact also makes sense for a massless particle with
E = p3 . For a massless spinor we therefore obtain We picked the normalization such
  that this limit would make sense.
0
√  0

u(p) = 2E  
1
0
Similarly, if we boost ξ T = (0, 1) (spin down), we get

 !  !  
√ 0 p 0 0
 p·σ   E + p 3 
 1   1  m→0 √  1
u(p) =  ! =  ! −→ 2E 



√ 0  p 0   0
   3 
p · σ̄ E−p
1 1 0
7.3.3 Helicity
We finally explain the meaning of ”left-handed” for ψL and ”right-
handed” for ψR . The massless solutions are eigenstates of the helic-
ity operator
!
1 σi 0
h ≡ p̂ · S = p̂i (7.52)
2 0 σi
where p̂i = p/|p|. In Ex9, you will discuss the Pauli-

A particle with h = +1/2 is called right-handed, while one with Lubanski polarization vector,
h = −1/2 is called left-handed. Wµ ≡ (1/2)µνσρ P ν J σρ

The helicity of a massive particle depends on the reference frame: which is a Lorentz covariant operator.
we can always boost to a frame in which its momentum is in the Using the commutation relations for
the generators of the Poincaré group,
opposite direction (but its spin stays unchanged). 10 one can show that P 2 = Pµ P µ
The Lorentz-invariance of helicity (for mass-less particles) is and W 2 = Wµ W µ , are Casimir
operators for the Poincaré algebra.
manifest in the representation of Weyl spinors ψL and ψR which live Since the Poincaré group has rank
in different representations of the Lorentz group. 2, P 2 and W 2 are its only Casimir
operators. Therefore a massive state
can be labelled by two numbers, its
7.3.4 Useful results: inner and outer products mass and spin. A massless state
can be labeled by only one number
It is convenient to introduce a basis ξ s and χs with s = 1, 2 for the (called helicity), since P µ and W µ for
two-component spinores such that massless particles are proportional to
each other.
10
For a massless particle moving
ξ r† ξ s = δ rs , and χr† χs = δ rs with v = c, we cannot perform such a
boost.
which e.g. is satisfied for
! !
1 0
ξ1 = , and ξ 2 =
0 1
and similarly for χs . The two independent solutions to the positive

frequency ansatz are
√ !
s p · σ ξs
u (p) = √
p · σ̄ ξ s
We can take the inner product in two ways, either u† · u or ū · u. Only

ū · u will be Lorentz-invariant of course, but we will need u† · u to
quantize the theory later. Let us determine the results.
√ !
r † s r† √ r† √ p · σ ξs
(u (p)) · u (p) = (ξ p · σ, ξ p · σ̄) √
p · σ̄ ξ s
= ξ r† p · σξ s + ξ r† p · σ̄ξ s
= 2ξ r† p0 ξ s
= 2E δ rs
and for the Lorentz-invariant inner product

! √ !
r† √ r† √ 0 1 p · σ ξs
ūr (p) · us (p) = (ξ p · σ, ξ p · σ̄) √ = 2m δ rs
1 0 p · σ̄ ξ s
Similarly for the negative frequency solutions we have

√ !
s p · σ χs
v (p) = √
− p · σ̄ χs
with
(v r (p))† · v s (p) = 2p0 δ rs

v̄ r (p) · v s (p) = −2mδ rs
What about the inner products of u and v ? We compute

! √ !
r s r† √ r† √ 0 1 p · σ χs
ū (p) · v (p) = (ξ p · σ, ξ p · σ̄) √
1 0 − p · σ̄ χs
√ √
= ξ r† (− p · σp · σ̄)χs + ξ r† p · σ̄p · σχs = 0
and similarly
v̄ r (p) · us (p) = 0
For the adjoint inner product, there is another combination which

has useful properties. With (p0 )µ = (p0 , −p), we have
√ 0 !
r † s 0 r† √ r† √ p · σ χs
(u (p)) · v (p ) = (ξ p · σ, ξ p · σ̄) √ (7.53)
− p0 · σ̄ χs
p p
= ξ r† (p · σ)(p0 · σ)χs − ξ r† (p · σ̄)(p0 · σ̄)χs
(7.54)
The terms under the square root are
(p · σ)(p0 · σ) = (p0 + pi σ i )(p0 − pi σ i ) = p20 − p2 = m2
and the same holds for (p · σ̄)(p0 · σ̄). This means the two terms in
Eq. (7.54) cancel and we get
(ur (p))† · v s (−p) = (v r (p))† · us (−p) = 0
7.3.5 Outer products

A final spinor identity will prove very useful before we turn to
quantizing everything. We claim
2
X
us (p)ūs (p) = p
/+m (7.55)
s=1
Note, the two spinors are not contracted, but placed back to back to
give a 4 × 4 matrix Similarly Think of |aihb| instead of hb|ai.
2
X
v s (p)v̄ s (p) = p
/−m
s=1
Now let us show this.

2 2 √ !
X X p · σ ξ s √ √
us (p)ūs (p) = √ s
(ξ s† p · σ̄, ξ s† p · σ) (7.56)
s=1 s=1
p · σ̄ ξ
Now
2
X
ξ s ξ s† = 12
s=1
which gives for Eq. (7.56)

2
!
X m p·σ
us (p)ūs (p) = =p
/+m (7.57)
s=1
p · σ̄ m
P2
You can now easily prove s=1 v s (p)v̄ s (p) = p
/ − m.
The Dirac Sea:

A historical detour: if we interpret Ψ as a single-particle wave function, the plane wave solutions to the Dirac
equation are energy eigenstates, with
∂Ψ
Ψ = u(p)e−ip·x i = +Ep Ψ
∂t
∂Ψ
Ψ = v(p)e+ip·x i = −Ep Ψ
∂t
which look like positive and negative energy solutions. The spectrum is unbounded below! This seems disas-
trous. Dirac invented the ’Dirac see’, where all negative energy states are filled up thanks to the Pauli-exclusion
principle. He postulated only charge differences are observable (similar to the energy difference argument mo-
tivating normal ordering). He postulated that if you excite a negative energy state, it leaves a hole with all the
properties of an electron, except that it carries positive charge! This is the prediction of anti-matter.
We now understand that the interpretation of the Dirac spinor as a single-particle wavefunction is not really
correct. The picture of a Dirac sea relies on the fact, that the particles are fermions – which breaks down for a
charged scalar, a boson, which also predicts anti-particles.
What we can still learn from Dirac’s analysis is that there is no consistent way to interpret the Dirac equation
as describing a single particle. We need to think of it as a classical field equation with only positive energy solu-
tions (the Hamiltonian is positive definite!). Once we quantize the theory, we encounter particle and anti-particle
excitations.
7.4 Symmetries of the Dirac Lagrangian
We can proceed as in the scalar Lagrangian and derive the space-

time symmetries
δΨ = εµ ∂µ Ψ
and use the standard formula Eq. (4.67) for the energy-momentum
tensor Recall
∂L
(j µ )ν = ∂ν φa − δνµ L ≡ T µ ν
T µν = iΨγ µ ∂ ν Ψ − L ∂(∂µ φa )
using the fact that L does not depend on ∂µ Ψ. Since the current
is only conserved on-shell, we can impose the equations of motions
directly on T µν and we won’t lose anything. Since the equations of
motions are first order, this affects the energy-momentum tensor. We
use (iγ µ ∂µ − m)Ψ(x) = 0 which means, we can set L = 0, to obtain
T µν = iΨγ µ ∂ ν Ψ
which e.g. leaves for the total energy

Z Z
E = d3 x T 00 = d3 x iΨγ 0 Ψ̇ (7.58)
Z
= d3 x iΨ† γ 0 (−iγ i ∂i + m)Ψ (7.59)
In the last equality we have again used the equations of motion.
7.4.1 Internal Symmetries

We can rotate the Dirac spinor by a phase
Ψ → eiα Ψ
which leaves the Dirac Lagrangian invariant. This gives rise to a

current
jVµ = Ψγ µ Ψ (7.60)
where we specify that it is a vector symmetry which transforms

ψL and ψR in the same way. We can easily obtain jVµ using the
Noether procedure or we can simply check that it is conserved on-
(
ψL → eiα ψL
Ψ → eiα Ψ
shell: ψR → eiα ψR
∂µ jVµ = (∂µ Ψ)γ µ Ψ + Ψγ µ (∂µ Ψ)

= imΨΨ − imΨΨ = 0
where we have used i∂Ψ/ = mΨ and the Dirac adjoint equation

←−
Ψi ∂/ = −mΨ, where the arrow points to the term we take the
derivative of. The conserved charge
Z Z
Q = d3 x Ψγ 0 Ψ = d3 x Ψ† Ψ (7.61)
has the interpretation of electric charge or particle number for

fermions.
For a massless Dirac spinor m = 0, the left-handed ψL and
right-handed ψR spinors decouple and we can rotate the left and
right-handed fermions in opposite directions
(
5 ψL → e−iα ψL
Ψ → eiα γ
5
(7.62) Ψ → eiαγ Ψ
ψR → e+iα ψR
which gives the conserved axial current
µ
jA = Ψγ 5 γ µ Ψ (7.63)
This is only conserved when m = 0, as we can see by taking the

divergence
µ
∂µ jA = (∂µ Ψ)γ µ γ 5 Ψ + Ψγ µ γ 5 (∂µ Ψ)
= (∂µ Ψ)γ µ γ 5 Ψ − Ψγ 5 γ µ (∂µ Ψ)
= 2imΨγ 5 Ψ
which vanishes only for m = 0. Once we couple an axial current to

a gauge field (similar to the photon field Aµ ), axial symmetries do
not survive the quantization process, they develop anomalies. An
Anomaly is a symmetry of the classical theory which is not preserved
in the quantum theory.11 11
Another symmetry which is
usually anomalous is scale symmetry
xµ → e−α xµ , see Ex4-2.
7.5 Quantizing the Dirac Field

If we were to use the same approach for the free Dirac Lagrangian
L = Ψ(x)(iγ µ ∂µ − m)Ψ(x)
as for the Klein-Gordon scalar theory, we would run into serious

trouble: the quantized free Hamiltonian would seem to be able
to lower the energy by creating particles (The Hamiltonian is not
bounded below). If we tried anti-commutation relations for a scalar
field, the cancellation between negative and positive frequencies to
guarantee causality in Eq. (5.71) and Eq. (5.72) would not work
anymore. This is the gist of the spin-statistics theorem. See Ex10-1.
7.5.1 The Hamiltonian

We will need the Hamiltonian. With the conjugate momentum
π = iΨ†
we obtain
H = π Ψ̇ − L = Ψ(−iγ i ∂i + m)Ψ (7.64)
which is the same as the result we obtained deriving the energy-

momentum tensor in Eq. (7.59).
We now want to turn this into an operator. We will expand Ψ(x)
into a basis of eigenfunctions of the free Dirac equation
2 Z
X d3 p −ip·x

ψ(x) = p bsp us (p)e+ip·x + cs† s
p v (p)e
s=1
2Ep
2 Z
X d3 p
ψ † (x) = p bs† s†
p u (p)e
−ip·x
+ csp v s† (p)e+ip·x
s=1
2Ep
where the operators bs†

p create particles associated with the spinor
state u (p), while the cs†
s s
p create anti-particles associated to v (p).
s s
Except for the u (p) and v (p), this has the exact same form as the
expansion of a complex scalar field, see Sec. 5.5.
7.5.2 Quantization of fermions

We need to account for the Fermi-Dirac-statistics for fermions,
meaning that the state picks up a minus sign upon the interchange
of any two particles. The spin-statistics theorem says that integer
spin must be quantized as bosons, while half-integer spin must be
quantized as fermions. If we try to do otherwise (as you should in
the exercise), we will find inconsistencies, such as unboundedness of
the free Hamiltonian.
How do we quantize fermionic fields? Recall for bosons
[a†p , a†q ] = 0 a†p a†q |0i ≡ |p, qi = |q, pi
If we want Fermi-Dirac statistics, we will need anti-commutation

relations {A, B} = AB + BA to hold.
Recall that for a scalar field φ, we could interpret a†p and ap as

creation an annihilation operators because of their commutation
relation with the number operator12 12
... or equivalently with the free
Z Hamiltonian
d3 k † Z
d3 k
Nφ = a ak H= Ek a†k ak
(2π)3 k (2π)3
Recall, that this worked because of the following identity for commu-
tators
[AB, C] = A[B, C] + [A, C]B
which immediately gives13 13

... and also
h i Z d3 k h † i [N, ak0 ] = −ak0
† †
N, ak0 = ak a k , ak 0 = a†k0
(2π)3 Recall that this means for eigenstates
of the number operator
Therefore a†k0 acting on a state raises the eigenvalue of N by one N |ni = n|ni
and the energy by Ek , as expected for a creation operator. However
and so
there is another useful relation for commutators
N a†k |ni = ([N, a†k ] + a†k N )|ni
[AB, C] = A{B, C} − {A, C}B (7.65) = (a†k + a†k N )|ni
= (1 + n)a†k |ni
with the anticommutator {A, B} = AB + BA. This means that if
we were to impose anticommutation relations on the a†k and ak , then and so
they could still have an interpretation as creation an annihilation a†k |ni ∝ |n + 1i

operators! up to normalization.
We will therefore ask the spinor field to satisfy
{Ψα (x), Ψβ (y)} = {Ψ†α (x), Ψ†β (y)} = 0 (7.66)
and
n o
Ψα (x), Ψ†β (y) = δαβ δ (3) (x − y) (7.67)
Using the expansion for Ψ and Ψ† gives
{brp , bsq } = {crp , csq } = {brp , cs†

q } = ... = 0
and the canonical non-zero relations
r s†
bp , bq = (2π)3 δ rs δ (3) (p − q) (7.68)
r s†
cp , cq = (2π)3 δ rs δ (3) (p − q) (7.69)
The calculation of the Hamiltonian is straight-forward but somewhat

tedious and we get Send me the LATEXand I’ll post it
Z here.
d3 p
H= 3
Ep bs† s s s†
p bp − cp cp
(2π)
Z
d3 p s† s s† s 3 (3)
= E p b b
p p + c c
p p − (2π) δ (0)
(2π)3
where the anti-commutation relation saved us from the indignity of
an unbounded Hamiltonian! Interestingly the contribution to the
cosmological constant now has the opposite sign compared to the
bosonic result.14 14
This anticipates a result of su-
persymmetry, which is a symmetry
between fermions and bosons. Exact
supersymmetry predicts H|0i =
((2π)3 δ (3) (0) − (2π)3 δ (3) (0))|0i = 0.
This holds after quantization and
even in the presence of interactions!
7.5.3 Fermi-Dirac Statistics

We define the (free) vacuum as usual
bsp |0i = csp |0i = 0
for all creation operators. The Hamiltonian still has nice commuta-
tion relations with the anti-commuting b and c operators. You can
easily see with Eq. (7.65) that
[H, bs† s†
p ] = Ep bp
[H, bsp ] = −Ep bsp
and exactly the same for b ↔ c. We can treat them therefore again
as ladder operators creating a tower of energy-eigenstates by acting
with bs† and cs† on the vacuum, just as in the bosonic case. The (not
relativistically normalized) one-particle states are
|p, ri = br†
p |0i
and the (not relativistically normalized) two-particle states statisfy
|p1 , r1 ; p2 , r2 i = (brp11 )† (brp22 )† |0i = −|p2 , r2 ; p1 , r1 i
exactly as we wanted, confirming that the particles do indeed obey

Fermi-Dirac statistics. In particular, we obtain the Pauli-Exclusion
principle
|p1 , r1 ; p1 , r1 i = 0
As for the scalar field (which had spin 0), we could act with the
angular momentum operator and confirm that a particle at rest
|p = 0, ri
does indeed carry intrinsic angular momentum 1/2 as expected.
7.5.4 Causal Fermi Fields

It is not clear that a theory with Fermi-Dirac statistics will lead to
causal predictions. For bosons, we said that
[φ(x), φ(y)] = 0, for (x − y)2 < 0.
For Fermi fields we now have the relations
{Ψα (x), Ψβ (y)} = 0
and
{Ψα (x), Ψ†β (y)} = 0, for (x − y)2 < 0
How do we see that this guarantees causality in the quantized the-

ory?
The reason is that observables are necessarily always bilinear in

the fields, e.g. the energy Eq. (??), momentum and fermion number
Eq. (7.61) (or charge) are given by
Z
E = d3 x iΨ† ∂t Ψ
Z
P = d3 x iΨ† ∂ i Ψ
i
Z
Q = d3 x Ψ† Ψ
This is not entirely surprising: we know that spinors form a double-

valued representation of the Lorentz group since they change
sign under 2π rotations. Observables, however, are unaffected by
a rotation by 2π and so must be composed of an even number of
spinor fields.
7.6 Perturbation Theory for Spinors
We will consider a simple nucleon-meson theory15 15

The nucleons are fermions now and
we are graduating from a toy model
1 µ2 2 to something closer to Nature, even
L = Ψ(iγ µ ∂µ − m)Ψ + (∂µ φ)2 − φ − λ ΨΓΨ φ (7.70) though it does not in fact describe
2 2 nucleon-meson interactions at low
energy correctly.
with Γ = 1 in case of φ being a scalar or Γ = iγ 5 for a pseudo-
scalar.16 16
We need the i for hermiticity of L.
Note, the dimensions of the fields, as before [φ] = 1, but the
kinetic term of the spinor requires [Ψ] = 3/2. This turns the in-
teraction to a marginal term [λ] = 0, compare to the dimensionful
(relevant) interaction in the case of a trilinear coupling between
scalars.
Dyson’s formula and Wick’s theorem can be used in almost
the same way for Fermi fields as for scalars. However, the an-
ticommutation relations lead to a crucial difference. Remem-
ber that for space-like distances (x − y)2 < 0, time-ordering is
not Lorentz-invariant17 Still, the T product of two scalar fields is 17
Since we can boost to a reference
Lorentz-invariant because the fields commute for space-like distances frame,where the temporal order
x0 > y 0 is reversed to x0 < y 0 .
(x − y)2 < 0, so
φ(x)φ(y) = φ(y)φ(x)
and the order here doesn’t matter. For fermions this no longer
holds! If (x − y)2 < 0, Fermi fields anticommute. For (x − y)2 < 0 and
if we are in a frame where x0 > y 0
T {Ψ(x)Ψ(y)} = Ψ(x)Ψ(y) (7.71)
in the Lorentz-transformed frame where y 0 > x0 , we have
T {Ψ(x)Ψ(y)} = Ψ(y)Ψ(x) = −Ψ(x)Ψ(y) (7.72)
We conclude that we must modify the definition of the T ordering of

Fermi-fields to make it Lorentz invariant.
The solution is straight-forward. We define the T -product to in-

clude a factor of (−1)n where n is the number of exchanges required
to time order the Fermi fields.
For two fields

Ψ(x)Ψ(y) for x0 > y 0
T {Ψ(x)Ψ(y)} = (7.73)
−Ψ(y)Ψ(x) for y 0 > x0
in particular we have
T {Ψ1 (x)Ψ2 (y)} = −T {Ψ2 (y)Ψ1 (x)}
The normal-ordering product is defined as before and we define

with
Ψ = Ψ(+) + Ψ(−)
where Ψ(+) multiplies an annihilation operator, and Ψ(−) multiplies

a creation operator. Remember the strange historic
convention stemming from the
(+) (+) (+) (−) (−) (+) (−) (−) positive/negative energy in the
: Ψ1 Ψ2 : = : Ψ1 Ψ2 + Ψ 1 Ψ2 + Ψ 1 Ψ2 + Ψ 1 Ψ2 :
exponent.
(+) (+) (−) (+) (−) (+) (−) (−)
= Ψ1 Ψ2 − Ψ2 Ψ1 + Ψ 1 Ψ2 + Ψ1 Ψ2
where the second picked up a (−1) because we had to interchange to

Fermi fields. Just as for the T product, we therefore have Recall that for scalar fields, their
order in T and N products was
unimportant.
: Ψ1 Ψ2 : = − : Ψ2 Ψ1 :
With these modifications, we can use Dyson’s formula and Wick’s

theorem as before. We just need to be careful with contractions in
Wick’s theorem, e.g.
: Ψ1 Ψ2 Ψ3 Ψ4 : = − : Ψ1 Ψ3 Ψ2 Ψ4 : = − Ψ1 Ψ3 : Ψ2 Ψ4 :
7.7 The Fermion Propagator
We will now move to the Heisenberg picture and define spinor

operators at every point in spacetime Ψ(x, t) such that they evolve
according to
∂Ψ
= i[H, Ψ]
∂t
which is the usual operator time evolution. As before, we can solve
this by expanding (replacing ix · p → −ix · p in the exponent)
2 Z
X d3 p
Ψ(x) = p bsp us (p)e−ip·x + cs† s
p v (p)e
ip·x
(7.74)
s=1
2Ep
2 Z
X
† d3 p
Ψ (x) = p bs† s†
p u (p)e
ip·x
+ csp v s† (p)e−ip·x (7.75)
s=1
2Ep
We define the fermionic propagator to be
iSαβ ≡ {Ψα , Ψβ } (7.76)
Inserting the expansion Eq. (7.74) we obtain We will in the following often drop
the indices, but you should remember
Z h
d3 pd3 q 1 −i(p·x−q·y)
that S(x − y) = {Ψ(x), Ψ(y)} is a
iSαβ (x − y) = p {bsp , br† s r
q } uα (p)ūβ (q)e 4 × 4 matrix.
(2π)6 4Ep Eq
i
+{cs† r s r
p , cq } vα (p)v̄β (q)e
+i(p·x−q·y)
We use Eq. (7.55) and the anticommutation relations for b and c of

Eq. (7.68) to obtain
Z
d3 p 1 h s i
iSαβ (x − y) = 3
uα (p)ūsβ (p)e−ip·(x−y) + vαs (p)v̄βs (p)e+ip·(x−y)
(2π) 2Ep
Z
d3 p 1 h −ip·(x−y) +ip·(x−y)
i
= (p/ + m) αβ e + ( p
/ − m) αβ e
(2π)3 2Ep
We can therefore write
iS(x − y) = (i∂/x + m)(D(x − y) − D(y − x))
in terms of the propagator for the real scalar field D(x − y). If we stay Recall, we can write
away from singularities, the propagator satisfies Z
d3 p 1 −ip·(x−y)
D(x − y) = e
(2π)3 2Ep
(i∂/x − m)S(x − y) = 0 with pµ = (Ep , p) or p2 = m2 .
which directly follows because
(i∂/x − m)S(x − y) = (i∂/x − m)(i∂/x + m)(D(x − y) − D(y − x))

= −(x + m2 )(D(x − y) − D(y − x)) = 0
using the properties of the scalar propagator for p2 = m2 away from

the singularities.
7.7.1 The Feynman propagator for Dirac fields

Similarly, we obtain first for the vacuum expectation value
Z
d3 p 1 −ip·(x−y)
h0|Ψα (x)Ψβ (y)|0i = (p
/ + m)αβ e
(2π)3 2Ep
= (i∂/x + m)D(x − y)
and
Z
d3 p 1 +ip·(x−y)
h0|Ψβ (y)Ψα (x)|0i = / − m)αβ e
(p
(2π)3 2Ep
= (−i∂/x − m)D(y − x)
from which we define the Feynman propagator Note, it is also a 4 × 4 matrix.
SF (x − y) = Ψ(x)Ψ(y)
or

h0|Ψ(x)Ψ(y)|0i x0 > y 0
SF (x − y) = h0|T Ψ(x)Ψ(y)|0i ≡
h0| − Ψ(y)Ψ(x)|0i y 0 > x0
Note, the minus sign! As we have argued around Eq. (7.72) it is

necessary for Lorentz-invariance. Let us relate the spinor Feynman
propagator to the scalar Feynman propagator
Ψ(x)Ψ(y) = θ(x0 − y0 )(i∂/x + m)D(x − y) + θ(y0 − x0 )(i∂/x + m)D(y − x)

= (i∂/x + m) [ θ(x0 − y0 )D(x − y) + θ(y0 − x0 )D(x − y) ]
= (i∂/x + m)φ(x)φ(y)
where the Feynman propagator for scalar fields is If you are worried about the time-
Z derivative of the θ function, you can
d4 p −ip·(x−y) i see that
φ(x)φ(y) = ∆F (x − y) = e
(2π)4 p2 − m2 + iε ∂t θ(x0 − y0 )D(x − y) = δ(x0 − y0 )D(x − y) = 0
The integral representation of the Feynman propagator for Dirac using the properties of the scalar
propagator.
spinors is therefore We can write Eq. (7.77) also as
d4 p e−ip·(x−y)
Z
Z SF (x − y) = i
d4 p −ip·(x−y) p
/+m (2π)4 p
/ − m + iε
SF (x − y) = i 4
e (7.77)
(2π) p − m2 + iε
2
which is a commonly used short
hand. We see that like for the scalar
field, the Feynman propagator is just
which satisfies the inverse of the Dirac equation
(with suitable ε pole structure).
(i∂/x − m)SF (x − y) = iδ (4) (x − y)
thanks to −(x + m2 )φ(x)φ(y) = iδ(x − y) derived in Eq. (5.79).

We conclude that SF (x − y) is a Green’s function of the Dirac
operator.
7.7.2 Nucleon-Nucleon scattering with spin

Let us study ΨΨ → ΨΨ scattering again. We have already discussed
this process in Sec. 6.4.1, except that the nucleons now have spin.
The initial state is
p p
|ii = |p1 , s; p2 , ri = 2Ep1 2Ep2 bs† r†
p1 bp2 |0i (7.78)
and the final state is

p p 0 0
|f i = |q1 , s0 ; q2 , r0 i = 2Eq1 2Eq2 bsq1† brq2† |0i (7.79)
We need to be careful with signs, since the b† now anti-commute.

For the adjoint we therefore have
p 0 0
hf | = 4Eq1 Eq2 h0|brq2 bsq1 (7.80)
We are interested in hf |S − 1|ii at order λ2 ,

Z
(−iλ)2
d4 x1 d4 x2 T Ψ(x1 )Ψ(x1 )φ(x1 )Ψ(x2 )Ψ(x2 )φ(x2 ) (7.81)
2
where as usual, all field operators are in the interaction picture.

As in the pure-scalar calculation, the contribution to Nucleon-
Nucleon scattering is due to the contraction
: Ψ(x1 )Ψ(x1 )Ψ(x2 )Ψ(x2 ) : φ(x1 )φ(x2 )

We now have to be careful about how the spinor indices are con-
tracted. Let us start with the Ψ and deal with Ψ later. Since we will
only find no non-vanishing contributions for the c† at this order, we
can focus on
: Ψ(x1 )Ψ(x1 )Ψ(x2 )Ψ(x2 ) : bs† r†
p1 bp2 |0i
Z 3
d k1 d3 k2 m n e−ik1 ·x1 −ik2 ·x2 m n s† r†
=− (Ψ(x 1 ) · u (k 1 ))(Ψ(x 2 ) · u (k2 )) p bk1 bk2 bp1 bp2 |0i
(2π)6 4Ek1 Ek2
where the − sign is from commuting Ψ past the Ψ. Let us now
anticommute the b’s past the b† ’s 18 to obtain 18
Using {bi , b†j } = δij and bj |0i = 0,
we find
1
= −p (Ψ(x1 ) · ur (p2 ))(Ψ(x2 ) · us (p1 ))e−ip2 ·x1 −ip1 ·x2 b1 b2 b†3 b†4 |0i =(−b1 b†3 b2 b†4 + b1 δ23 b†4 )|0i
4Ep1 Ep2
=(−b1 b†3 δ24 + δ23 δ14 )|0i
−(Ψ(x1 ) · us (p1 ))(Ψ(x2 ) · ur (p2 ))e−ip1 ·x1 −ip2 ·x2 |0i
=(−δ13 δ24 + δ23 δ14 )|0i.
Note again the relative minus sign between the two terms. What
happens when we contract this with hf |? Let’s see the first term
0 0
h0|brq2 bsq1 (Ψ(x1 ) · ur (p2 ))(Ψ(x2 ) · us (p1 )) |0i =
e+iq2 ·x1 +iq1 ·x2 s0 0
p (ū (q1 ) · ur (p2 ))(ūr (q2 ) · us (p1 ))
2 Eq1 Eq2
e+iq1 ·x1 +iq2 ·x2 r0 0
− p (ū (q2 ) · ur (p2 ))(ūs (q1 ) · us (p1 ))
2 Eq1 Eq2
the second term contributes the same and cancels the 1/2 in front
√
of Eq. (7.81). The factors of 1/ E cancel the relativistic state
normalization. Combining everything, we obtain
Z 4 "
d x1 d4 x2 d4 k iek·(x1 −x2 ) 0 0
−(−iλ) 2
4 2 2
(ūs (q1 ) · ur (p2 ))(ūr (q2 ) · us (p1 ))e+i(q2 −p1 )·x1 +i(q1 −p2 ·x2
(2π) k − µ + iε
#
0 0
−(ūr (q2 ) · ur (p2 ))(ūs (q1 ) · us (p1 ))e+i(q1 −p1 )·x1 +i(q2 −p2 ·x2
where have used the φ-propagator. Integrating over x1 and x2 , we

obtain
Z 4 "
d k i(−iλ)2 (2π)4 0 0
− (ūs (q1 ) · ur (p2 ))(ūr (q2 ) · us (p1 ))δ (4) (p1 − q1 + k)δ (4) (p2 − q2 + k)
k 2 − µ2 + iε
#
r0 r s0 s (4) (4)
−(ū (q2 ) · u (p2 ))(ū (q1 ) · u (p1 ))δ (p1 − q2 + k)δ (p2 − q1 + k)
Almost done. We use the usual normalization

hf |S − 1|ii = iA (2π)4 δ (4) (p1 + p2 − q1 − q2 )
we find
" 0 0 0 0
#
(ūr (q2 ) · ur (p2 ))(ūs (q1 ) · us (p1 )) (ūs (q1 ) · ur (p2 ))(ūr (q2 ) · us (p1 ))
2
iA = (−iλ ) −
(p2 − q1 )2 − µ2 + iε (p1 − q1 )2 − µ2 + iε
which is the final result.
7.8 Feynman rules for Dirac Spinors

The calculation above is really cumbersome and you will hopefully
never have to redo it this way ever again. Luckily introducing Feyn-
man rules simplifies the calculation significantly and we will see that
we will be able to derive the final result in a couple of easy steps.
1. Propagators:
i
φ(x)φ(y) = =
p2 − µ2 + iε
i(p
/ + m)
Ψ(x)Ψ(y) = =
p2 − m2 + iε
2. Vertices:
−iλ =
3. External legs incoming
p
= us (p)
p
= v̄ s (p)
4. External legs outgoing
p
= ūs (p)
p
= v s (p)
5. Impose momentum conservation at each vertex.

R d4 k
6. Integrate over each undetermined momentum k with (2π)4 .
7. Figure out the overall sign of the diagram.
8. A closed fermion loop always gives a factor of −1 and the trace of

a product of Dirac matrices.19 19
We will see an example of this
below.
We draw arrows on fermion lines to represent the direction of
particle-number flow. The momentum assigned to a fermion
propagator flows in the direction of this arrow. For external antipar-
ticles, however, the momentum flows opposite to the arrow. We show
this explicitly by drawing a second arrow next to the line.
7.8.1 Example: Nucleon scattering ΨΨ → ΨΨ

The two diagrams are
q q0
q
q0
+ (7.82)
p p0 0
p p
"
i
iA = (−ig)2 ū(q 0 ) · u(q) ū(p0 ) · u(p)
(q − q 0 )2 − µ2
#
0 i 0
−ū(p ) · u(q) 0 ū(q ) · u(p) (7.83)
(p − q)2 − µ2
Where we have suppressed the spinor spin labels. The minus sign
between the diagrams is a reflection of Fermi-Dirac statistics.
7.8.2 Example: Nucleon-Anti-Nucleon scattering ΨΨ → ΨΨ

The two diagrams are
q q0
q q0
+ p p0 (7.84)
p p0
with the Feynman rules we have

2 ū(p0 ) · u(p) v̄(q) · v(q 0 ) v̄(q) · u(p) ū(p0 ) · v(q 0 )
iA = (−iλ) − +
(p − p0 )2 − µ2 (p + q)2 − µ2
There is a relative minus sign between the two diagrams, which is a

bit subtle to figure out. You can convince yourself that it is correct,
by repeating the relevant steps using the explicit Wick technology
above.
7.8.3 Example: Meson Scattering

Finally, in φφ → φφ scattering, the leading contribution appears first
at 1-loop. The amplitude for the diagram is
p0 q0
p q
Z 0 0 0
4 d4 k k/ + m k/ + p
/ +m k/ + p
/ − /q + m k/ − p
/+m
iA = −(−iλ) Tr 2
(2π)4 k − m2 (k + p0 )2 − m2 (k + p0 − q0 )2 − m2 (k + p)2 − m2
where we have suppressed the +iε piece in the denominator. Note,
that the integral diverges logarithmically20 In the k → ∞ limit, the 20
Compare to the pure scalar case
integral behaves like which was convergent.
Z 4
d k
iA ∼ ∼ ln Λ
k4
You will have to come to the QFT course next semester to learn how
to make sense of results like these. There is an overall minus sign Like for the cosmological constant in
sitting in front the of the amplitude. Sec. ?? we will have to regularize and
renormalize the result in a systematic
In complicated diagrams, we can often find the minus signs, by way.
noting that the product (ΨΨ) or any other pair of fermions is a
commuting object. So,
. . . (ΨΨ)x (ΨΨ)y (ΨΨ)z . . . = . . . (+1)(ΨΨ)x (ΨΨ)z (ΨΨ)y . . .

. . . SF (x − z)SF (z − y) . . .
In a closed loop of n fermion propagators we have
which involves the contraction
Ψα (x)Ψα (x)Ψβ (y)Ψβ (y) = − Ψβ (y)Ψα (x)Ψα (x)Ψβ (y)

= −Tr[SF (y − x)SF (x − y)]
This is a general result: a closed fermion loop results in a factor of

(−1) and a trace of a string of γ matrices.
7.8.4 Example: the Yukawa potential

We saw in Sec. ?? that the exchange of a real scalar between to
charged scalars gives rise to always attractive Yukawa potential
between two spin zero particles. Do we still find the same for the
scalar force between two spin 1/2 particles?
We will again take the non-relativistic limit p m which
implies for the spinors
√ ! !
s p · σ ξs √ ξs
u (p) = √ → m s
p · σ̄ ξ s ξ
and
√ ! !
s p · σ ξs √ ξs
v (p) = √ → m
− p · σ̄ ξ s −ξ s
If the two interacting particles are distinguishable only the first one
in Eq. (7.82) contributes. In this limit, the spinor contractions in the
ψψ → ψψ scattering are
0
ūs (q 0 ) · us (q) = 2mδss0
and the amplitude becomes

δs0 s δr0 r
iA = −i(−iλ)2 (2m)
(p − p0 )2 + µ2
We see that the spin is conserved at each vertex in the non-relativistic

limit (due to the δ factors) and that the momentum dependence is
exactly the same as in the bosonic case Eq. (6.47), telling us that the
particles feel an attractive Yukawa potential,
λ2 e−µr
U (r) = −
4πr
If we repeat the calculation for ΨΨ → ΨΨ, two minus signs can-
cel and the Yukawa interaction once again leads to an attractive
potential.
8
Quantum Electro Dynamics
8.1 The Feynman Rules for Quantum Electrodynamics

We replace the scalar particle φ with a vector particle Aµ and re-
place the Yukawa interaction Hamiltonian with
Z
Hint = d3 x e Ψγ µ ΨAµ
What are the Feynman rules? We can easily guess, even though the
theory behind them is quite involved.
1. Interaction vertex:
−ieγ µ =
2. Propagators:
−iηµν
Aµ (x)Aν (y) = =
q 2 + iε
3. External photon lines:

Outgoing
p
= ∗µ (p)
Incoming
p
= µ (p)
We draw photons as wavy lines. The symbol µ (p) stands for

the polarization vectors of the initial or final state photons. The
polarization vectors are always of the form
!
0
µ = , with p · = 0 (8.1)

If p is along the z-axis, the right- and left-handed polarization

vectors are
 
0
1 1


µ = √  
2 ±i
0
8.2 Maxwell’s equations

The Maxwell-Lagrangian in the absence of any sources is simply
1
L = − Fµν F µν (8.2)
4
with the field strength
Fµν = ∂µ Aν − ∂ν Aµ (8.3)
and the equations of motions are

∂L
∂µ = −∂µ F µν = 0
∂(∂µ Aν )
Additionally, the field strength satisfies the Bianchi identities
∂λ F µν + ∂µ F νλ + ∂ν F λν = 0
The version of Maxwell’s equation you learn in pre-school is recov-

ered atopwithdelims
!
µ φ
A =
A
with the electric and magnetic fields E and B
∂A
E = −∇φ −
∂t
and
B =∇×A
which in terms of the field strength is

 
0 Ex Ey Ez
−E 0 −B By 
 x z 
Fµν =  
−Ey Bz 0 −Bx 
−Ez −By Bx 0
and the Bianchi identity results in two Maxwell equations which are
unchanged even in the presence of sources
∇·B =0
∂B
= −∇ × E
∂t
The equations of motions give the two remaining Maxwell equations
∇·E =0
∂E
=∇×B
∂t
Note, the interesting electro-magnetic
duality E → B and B → −E!
8.3 Gauge Symmetry

We have used Aµ to describe a massless vector field, which would
seem to imply that since the gauge field has 4 components, we
naively count 4 degrees of freedom. Yet we know has only two E.g. from counting states and
degrees of freedom, which are the polarization states we’ve used calculating the thermodynamics of a
blackbody in thermodynamics.
in the Feynman rules. How can this be resolved? There are two
mechanisms at work which ensure this.
• The field A0 is not dynamical: It has no kinetic term Ȧ0 in the
Lagrangian. If we fix Ai and Ȧi at initial some time t0 , then A0 is
fully determined by the equation of motion ∇ · E = 0 which we can
expand out to obtain
∂A
∇ 2 A0 + ∇ · =0
∂t
which has the solution gauge orbits
Z 0
∇x0 · ∂A(x )
A0 (x) = d3 x0 ∂t
(8.4)
4π|x − x0 |
gauge fixing
which you can convince yourself is correct using the arguments
around Eq. (4.95) and the Green’s function method. So A0 is
clearly dependent and we do not get to specify it fixing the initial
conditions. This reduces our the degrees of freedom from 4 to 3.
field space
• The Maxwell Lagrangian has a very large redundancy,
Figure 8.1: Schematic picture of
the gauge redundancy. The gauge
Aµ (x) → Aµ (x) + ∂µ α(x) (8.5) orbits are families of gauge-equivalent
configurations. The physically
independent states are in a gauge
for any function α(x). We will require that α(x) vanishes fast fixed slice that intersects each orbit
once.
enough at x → ∞. This is a gauge symmetry. The field
strength is invariant under it
Fµν → ∂µ (Aν + ∂ν α) − ∂ν (Aµ + ∂µ α) = Fµν
We seem to have a theory with an infinite number of symmetries,

each for each local choice of function α(x). All our other internal
symmetries where global and acted the same at every point in
spacetime, e.g.
ψ → eiβ ψ
with β = const.. These symmetries lead to selection rules or

conservation laws, thanks to Noether’s theorem. Do we now have
infinitely many conserved currents?
No! A symmetry takes a physical state to different physical

state with the same properties, the gauge symmetry however is a
redundancy in our description. The two states connected by a
gauge symmetry have to be identified, they are the same physical
state. You can see this also in the Maxwell’s equation which are
not sufficient to specify the time evolution of Aµ
(ηµν (∂ λ ∂λ ) − ∂µ ∂ν )Aν = 0
Now, the differential operator (ηµν (∂ λ ∂λ ) − ∂µ ∂ν ) is not invert-

ible, since it has a zero eigenvector for functions of the form ∂µ λ.
Fixing initial data, we can not uniquely determine Aµ since there
no way to distinguish between
Aµ , and Aµ + ∂µ λ
If we are willing to identify Aµ and Aµ + ∂µ λ as the same

physical state, then this is not a problem anymore.
How about formulating the theory solely in E and B? Lorentz-

invariance would not be manifest, but we might be able to deal
with this. However, this is not advisable given the relevance of Aµ
for quantum mechanical effects like the Aharonov-Bohm effect. Figure 8.2: The the Aharonov-
Further, we would like to describe charged gauge fields, which also Bohm effect is a quantum mechan-
ical phenomenon in which an elec-
requires generalizations of Aµ .
trically charged particle is affected
We discover an enlarged phase space, foliated by gauge orbits. by an electromagnetic potential Aµ ,
To remove the redundancy, we need to pick one representative from despite being confined to a region in
which both the magnetic field B and
each equivalent gauge orbit. We call these different configuration electric field E are zero. The underly-
of a physical state, different gauges. We can pick gauges as we ing mechanism is the coupling of the
electromagnetic potential with the
please, but as for coordinate systems, there are better and less suited
complex phase of a charged particle’s
choices depending on the problem. wave function,
q
Z
ϕ= A · dx
8.3.1 Gauges ~ P
and the Aharonov-Bohm effect is ac-
We start with the Lorentz gauge1 cordingly illustrated by interference
experiments. https://en.wikipedia.
org/wiki/Aharonov-Bohm_effect.
∂ µ Aµ = 0 (8.6) 1
Other popular choices are
A0 = 0 (temporal gauge)
This does not yet pick a unique representation from the gauge orbit, 3
A =0 (axial gauge)
since we are free to make further gauge transformations which are
functions that satisfy
∂µ ∂ µ λ(x) = 0
which certainly has non-trivial solutions. The advantage of the

Lorentz gauge condition is clear from its name: it is manifestly
Lorentz invariant.
Another useful gauge, is the Coulomb gauge
∇·A=0 (8.7)
there is still options to choose harmonic functions which satisfy

∇2 λ = 0. Since Eq. (8.4) fixes A0 , we find that now2 2
This will not hold in the presence
of sources, as you know from electro-
A0 = 0 statics, since in this case, e.g. a
charge induces a potential for A0 = φ
Coulomb gauge is not manifestly Lorentz invariant, but it clearly

shows the degrees of freedom of the physical system: The 3 degrees
of freedom satisfy a single constraint ∇ · A = 0, leaving us with just 2
independent degrees of freedom.
8.4 Quantizing the Electro-magnetic field

We will now try to quantize the electric field in Coulomb and in
Lorentz gauge. A common subtlety to both methods is due to the
absence of a canonical momentum for A0 :
∂L
π0 = =0 (8.8)
∂(∂t A0 )
∂L
πi = = −F 0i = E i (8.9)
∂(∂t Ai )
We again see that A0 is not a dynamical field. The Hamiltonian is
Z
H = d3 x(π i Ai − L)
Z
1 1
= d3 x( E · E + B · B − A0 (∇ · E))
2 2
We see that A0 acts as a Lagrange multiplier, which has the equation
of motion
∇·E =0
which is just enforcing Gauss’ law. What do the different gauge

fixing conditions mean for the system?
8.4.1 Coulomb gauge

In Coulomb gauge, the equation of motion for A is
∂µ ∂ ν A = 0
which we solve as usual by a Fourier-ansatz

Z
d3 p
A(x) = ξ(p)eip·x
(2π)3
with p2 = 0 or p20 = |p|2 . The Coulomb gauge condition ∇ · A = 0

requires
ξ(p) · p = 0
which means that ξ(p) is orthogonal to the direction of motion of

the gauge field p. We can choose it to be a linear combination of two
ortho-normal vectors r with r = 1, 2 which satisfy
r (p) · p = 0
and
r (p) · s (p) = δrs
These are the two polarization states of the photon. Now we

attempt to quantize and we turn the classical Poisson brackets into
commutator for operator valued fields. Naively (and wrongly, we try
for the equal-time commutator
[Ai (x), E j (y)] = iδij δ (3) (x − y) (wrong)

and
[Ai (x), Aj (y)] = [E i (x), E j (y)] = 0
This is not correct! Why? It does not work with the gauge and
Gauss constraints. We need to have imposed on the operators
∇·A=∇·E =0
but if we evaluate the operator relations above, we find
[∇x · A(x), ∇y · E(y)] = i∇2 δij δ (3) (x − y) 6= 0
You can see this problem already in the classical Poisson bracket
structure.3 The correct Poisson bracket structure implies different 3
See e.g. P. Ramond - QFT, or
commutation relations in Coulomb gauge Dirac ”Lectures on Quantum me-
chanics”.

∂i ∂j
[Ai (x), E j (y)] = i δij − 2 δ (3) (x − y) (8.10)
∇
Let us check if this is consistent with the constraints. We write the δ

function in momentum space
Z
j d3 p j pi pj
[Ai (x), E (y)] = i δi − 2 eip·(x−y)
(2π)3 p
Applying e.g. the Coulomb gauge fixing
Z
d3 p j pi pj
[∂ix Ai (x), E j (y)] = i ipi δ i − eip·(x−y)
(2π)3 p2
Z
d3 p p2 pj
=i i p j − eip·(x−y) = 0
(2π)3 p2
We expand in modes again as usual
2 Z
X d3 p 1
A(x) = 3
p r (p) arp eip·x + arp † e−ip·x
r=1
(2π) 2|p|
2 Z p
X d3 p (−i) |p|
E(x) = 3
√ r (p) arp eip·x − arp † e−ip·x
r=1
(2π) 2
with the before define polarization vectors.
r (p) · p = 0 and r (p) · s (p) = δrs
You should convince yourself that the commutation relations imply

for the ladder operators
[arp , asq † ] = (2π)3 δ rs δ (3) (p − q)
and
[arp , asq ] = [arp † , asq † ] = 0

where you will have to use the outer product (or completeness
relation) for the polarization vectors
2
X pi pj
ir (p)jr (p) = δ ij −
i=1
p2
We can confirm this equation by acting on both sides with a com-

plete basis (1 (p), 2 (p), p).
We substitute the expansion into the Hamiltonian and obtain
after normal ordering
Z X2
d3 p
H= |p| arp † arp
(2π)3 r=1
Quantizing in the Coulomb gauge gave us direct insight into

the physical degrees of freedom. Lorentz invariance is however not
manifest anymore. We can see this looking at the propagator in the
Heisenberg picture
tr
Dij (x − y) ≡ h0|T Ai (x)Aj (y)|0i
Z
d4 p i j pi pj
= δ − 2 e−ix·p
(2π)4 p2 + iε i p
Where explicitly denote with tr, the transverse part of the photon.
Let us therefore explore the manifestly invariant Lorentz gauge.
8.4.2 Lorentz Gauge

The Lorentz gauge condition is
∂ µ Aµ = 0
which implies for the equations of motion
∂ µ ∂µ Aν = 0
We will depart from the procedure enforcing Coloumb gauge.4 and 4

This can be motivated much more
will modify the Lagrangian such that the equations of motion are cleanly using the path integral
formalism and a trick by Fadeev and
directly in Lorentz gauge. With Popov.
1 1
L = − Fµν F µν − (∂µ Aµ )2
4 2
we obtain
∂µ F µν + ∂ ν (∂µ Aµ ) = ∂ µ ∂µ Aν = 0
We can be more general and choose

1 1
L = − Fµν F µν − (∂µ Aµ )2
4 2α
with arbitrary and constant α. We confusingly call α = 1 the
Feynman gauge and α = 0 the Landau gauge.
We will now firstly quantize the theory and only secondly enforce
the gauge constraint. Thankfully, both conjugate momenta are
non-vanishing
∂L
π0 = = −∂µ Aµ (8.11)
∂(∂t A0 )
∂L
πi = = ∂ i A0 − ∂ 0 Ai (8.12)
∂(∂t Ai )
We impose the equal-time commutation relations in Lorentz-gauge
[Aµ (x), πν (y)] = i ηµν δ (3) (x − y) (8.13)
and
[Aµ (x), Aν (y)] = [πµ (x), πν (y)] = 0
Expanding again
3 Z
X d3 p 1
Aµ (x) = 3
p λµ (p) aλp eip·x + aλp † e−ip·x
(2π) 2|p|
λ=0
3 Z p
X d3 p +i |p| λ
πµ (x) = 3
√ µ (p) aλp eip·x − aλp † e−ip·x
(2π) 2
λ=0
Note that we here have (+i) in the momentum, rather than the
usual (−i). The reason is that the conjugate momentum is
π µ = −Ȧµ + . . .
We now have to deal with four polarization vectors! This is much

worse than the two physical polarizations in Coloumb gauge. We
choose 0 time-like and i to be space-like. We choose the normaliza-
tion
0 0
λ · λ = η λλ
which implies
0
(µ )λ (ν )λ ηλλ0 = ηµν
We choose two of the space-like polarizations to be transverse to the

momentum pµ = (|p|, p)
1 · p = 2 · p = 0 (8.14)
The third vector is a longitudinal polarization. If pµ(z) ∝ (1, 0, 0, 1),

we would have
       
1 0 0 0
0 1 0 0
       
0 =   , 1 =   , 2 =   , 3 =   ,
0 0 1 0
0 0 0 1
We define for a general four-momentum pµ ∝ (|p|, p),
     
1 0 0
0 1 p1  r 
     
0 (p) =   , 3 (p) =   , r (p) =  1r  ,
0 |p| p2  1 
0 p3 r1
with r (p) · p = r (p) · p = 0, r = 1, 2
For later use, we note for the time-like and longitudinal polarizations
that
0 (p) · p = |p| 3 (p) · p = −|p| (8.15)
We derive the ladder operator commutation relations using the field

expansion to obtain
0 0
[aλp , aλq † ] = −η λλ (2π)3 δ (3) (p − q)
with all other commutators vanishing as usual. The minus sign

is problematic! For space-like λ, λ0 = 1, 2, 3 we are in well-know
territory
0 0
[aλp , aλq † ] = (2π)3 δ λλ δ (3) (p − q), λ, λ0 = 1, 2, 3
but for time-like ladder operators, we have
[a0p , a0q † ] = −(2π)3 δ (3) (p − q)
This means trouble. Let us see what it implies e.g. for one-particle
states. With the vacuum
aλp |0i = 0
we can define (yet to be normalized) one-particle states as usual
|p, λi = aλq † |0i
For space-like λ = 1, 2, 3 we are in the green, but for the time-like

polarization λ = 0, the state |p, λ = 0i has negative norm
hp, 0|q, 0i = h0|a0p a0q † |0i = −(2π)3 δ (3) (p − q)
This strange. Negative norm would imply negative probabilities! We

put ourselves in danger already from the start as we can see from the
kinetic term for A0 which has the wrong sign
1 2 1
L= Ȧ − Ȧ20 + . . .
2 2
How can we makes sense of this?
We have yet to impose the gauge constraint ∂µ Aµ = 0 on the theory.
It will be responsible to remove the time-like negative norm states
and remove all but two polarizations. How can we implement the
constraint in a quantum theory?
• Let us start with the strongest possible condition. We impose

∂µ Aµ = 0 as a operator equation. This does not work, because
it is not compatible with the commutation relations Eq. (8.13),
since the conjugate momentum Eq. (8.11)
∂L
π0 = = −∂µ Aµ
∂(∂t A0 )
would vanish universally.

• A weaker option would be to impose it on a subset of the Hilbert

space instead of on the operators. We could imagine splitting
the Hilbert space into ”good” states |ψi and a ”bad” states which
somehow are not part of physical observables. How can we define
”good” states. We could impose
∂µ Aµ |ψi = 0 (8.16)
on all the ”good” states. However this does not work either. It is
−
still too strong, since we can e.g. decompose Aµ = A+
µ + Aµ with
3 Z
X d3 p 1
A+
µ (x) = p λ (p) aλp e−ip·x
(2π)3 2|p| µ
λ=0
3 Z
X d3 p 1
A−
µ (x) = p λ (p) aλp † e+ip·x
(2π)3 2|p| µ
λ=0
The vacuum satisfies
A+
µ |0i = 0
by definition. But we get
∂ µ A−
µ |0i =
6 0
which shows that the vacuum as a physical state isn’t even one of
the good states.
• The solution: We want the vacuum as one of the good physical

states. We therefore ask physical states |ψi to only satisfy the
Gupta-Bleuler condition5 5
The problem with the Gupta-Bleuer
condition is that is not suited for
non-abelian gauge theories due to
∂ µ A+
µ |ψi = 0 (8.17) the non-linearity of the free field
equations (they contain terms
∼ A3 , A4 ). The modern approach
replacing this QED-only ’hack’ is
and the adjoint equation called BRST symmetry (Becchi,
Rouet, Stora and Tyutin) and will
hψ|∂ µ A−
µ =0 be developed in the QFT lectures.
One identifies a nilpotent symmetry
With this we make sure that generator which commutes with the
Hamiltonian H which allows to split
the eigenstates of H into unphysical
hψ 0 |∂ µ A− µ + 0 µ
µ + ∂ Aµ |ψi = hψ |∂ Aµ |ψi = 0 and physical sub-sectors which have
zero inner product with each other.
holds. Since the constraint is linear, the physical states span the
physical Hilbert space Hphys . How does the physical Hilbert space
look like?
We consider a basis for the Fock space. We decompose any ele-

ment of this basis into
|ψi = |ϕT i|ϕL−0 i
where |ϕT i contains the transverse photons (created by a1p † and a2p † )
and |ϕL−0 i contains time-like and longitudinal photons (created by
a0p † and a3p † ).
The Gupta-Bleuer condition Eq. (8.17) therefore reads with

the polarization condition for the transverse states Eq. (8.14) and
Eq. (8.15) as
(a0p − a3p ) |ϕL−0 i = 0 (8.18)
What does this mean? Condition Eq. (8.18) requires all physical
states to contain pairs of timelike and longitudinal photons: for each
timelike photon with p, it must also contain a longitudinal photon
(n)
with the same p. We can write |ϕL−0 i as a sum over states |ϕL−0 i
containing n pairs of timelike and longitudinal photons
∞
X (n)
|ϕL−0 i = cn |ϕL−0 i (8.19)
n=0
(0)
where |ϕL−0 i = |0i is just the vacuum.
One can show that Eq. (8.18) indeed decouples the negative norm
states and the all the remaining timelike and longitudinal photons
have zero norm
(n) (n)
hϕL−0 |ϕL−0 i = δn,0 δm,0
We have therefore defined an inner product on the physical Hilbert

space which is positive semi-definite. This is clearly a step in the
right direction. We will now have to find a way to deal with the
zero-norm states.
We understand the zero-norm states if we treat them as gauge-
equivalent to the vacuum. If two states that differ only by pairs
(n)
of timelike and longitudinal photons, |ϕL−0 i with n ≥ 1, they
describe the same physical state.
We need to check then that they give the same result for all
physical observables. Let us start with the Hamiltonian, which you
can easily compute
Z " 3 #
d3 p X
i† i 0† 0
H= |p| ap ap − ap ap
(2π)3 i=1
The Gupta-Bleuer condition Eq. (8.18) guarantees that
hψ|a3p † a3p |ψi = hψ|a0p † a0p |ψi
and the contributions of of timelike and longitudinal photons cancel

each other in the Hamiltonian. Further, the Hamiltonian evaluated
on physical states satisfying the Gupta-Bleuer condition is posi-
tive definite and contains only the contributions from transverse
photons, as it should be.
One can show that expectation values of gauge-invariant operators
evaluted on physical states are independent of the coefficients cn of
Eq. (8.19).
8.4.3 The propagator

The propagator for the photon in Lorentz gauge is given by
Z
d4 p −iηµν −ip·(x−y)
h0|T Aµ (x)Aν (x)|0i = Aµ (x)Aν (y) = e
(2π)4 q 2 + iε
Alternatively, you can show that for general gauges with arbitrary α
in Eq. (??), we would obtain in momentum space Keeping the α dependence and
testing that it cancels out in the final
−i pµ pν result is a useful check.
ηµν + (α − 1)
q 2 + iε p2
8.5 Coupling gauge fields to matter

We now want to couple the gauge field Aµ to other fields, like scalars
or fermions (or more poetically, coupling light to matter). Let us see
how we can do this. We know that the simplest coupling has to have
the form
1
L = − Fµν F µν − Aµ Rµ (8.20)
4
What are the conditions on Rµ ? We know that we need the gauge
redundancy to find the correct number of degrees of freedom. The
equations of motions are
∂µ F µν = Rν
Let us apply ∂ν on both sides
∂ν ∂µ F µν = 0 = ∂ν Rν
In other words Rν must be a conserved current! We write Rν = j ν

and recongnize this as the usual conservation equation
∂µ j µ = 0
We have already encountered a large number of these conserved

currents j µ . Which are the Noether currents related to the gauge
symmetry for the photon field?
8.5.1 Coupling photons to Dirac fermions

Let us try with the U (1) global, internal symmetry of the Dirac
Lagrangian
Ψ → eiα Ψ
which with Eq. (7.60) gives rise to a current
j µ = Ψγ µ Ψ (8.21)
Now let us couple this to the gauge field L ⊃ Aµ j µ to obtain to-

gether with the kinetic term of the Dirac spinor
1
L = − Fµν F µν + iΨ∂Ψ
/ − eΨγ µ Ψ Aµ (8.22)
4
where I have introduced a yet to be specified constant e. We can

rewrite this as
1
L = − Fµν F µν + iΨDΨ
/ (8.23)
4
with the covariant derivative The ”covariant” is with respect to
it’s behaviour under gauge transfor-
mations as we will see below.
Dµ = ∂µ + ieAµ (8.24)
Under a gauge transformation Aµ (x) → Aµ (x) + ∂µ λ(x) we find that

the Lagrangian transforms as
0
L → L − e Ψ γ µ (∂µ λ(x))Ψ0
we see that we need to cancel the transformation of Aµ with a

transformation of Ψ → Ψ0 . Let us try a local generalization of the
global U (1) transformation
Ψ → e−ie λ(x) Ψ
The covariant derivative transforms as
Dµ Ψ = [∂µ + ieAµ ]Ψ → [∂µ + ieA0µ ]Ψ0

= [∂µ + ie(Aµ + ∂µ λ(x))]e−ie λ(x) Ψ
= e−ie λ(x) [∂µ − ie∂µ λ(x) + ie(Aµ + ∂µ λ(x))]Ψ
= e−ie λ(x) Dµ Ψ
which shows that the covariant derivative has the convenient prop-
erty that it only picks up a factor exp(−ieλ(x)), with the derivative
canceling the transformation of the gauge field. The whole La-
grangian Eq. (8.26) is then gauge invariant, since
Ψ → Ψe+ie λ(x)
Let us summarize now the gauge symmetry of QED
Aµ (x) → Aµ (x) + ∂µ λ(x), Ψ → e−ie λ(x) Ψ (8.25)
/ invariant.
which leaves the Lagrangian and in particular iΨDΨ
8.5.2 Coupling photons to charged scalars

We will write ϕ for the charged scalar field and not ψ as previously
in order no to confuse it with the spinors.
How would we couple the charged scalar to a gauge boson? The
linear coupling to the U (1) current would not work6 but we can use 6
Try it!
the covariant derivative Dµ
Dµ = ∂µ + ieAµ
and replace the ∂µ everywhere. This gives

1
L = − Fµν F µν + (Dµ ϕ)∗ Dµ ϕ − m2 ϕ∗ ϕ (8.26)
4
a simultaneous gauge transformation
Aµ (x) → Aµ (x) + ∂µ λ(x), ϕ(x) → e−ie λ(x) ϕ(x) (8.27)
This tricks works for any theory. Replacing the derivatives by co-
variant derivatives in suitable representations works in almost any
theory. This procedure is called minimal coupling.
8.6 The electric charge

Back to QED and the gauge-spinor coupling: he factor e has the
interpretation of the coupling strength between the charged field and
the photon. We know from electro-magnetism that j 0 is the charge
density, as can be seen from the equations of motion, ∂µ F µν = j ν or
XXX. The total charge Q is then given by
Z
Q = e d3 x Ψγ 0 Ψ
Inserting the Heisenberg expansion of the free spinor, we get a

similar expression as for the charged scalar field Eq. (5.60)
Z 2
d3 q X s† s
Q=e b b − cs† s
q cq
(2π)3 r=1 q q
We find again that this is the number of particles minus the num-
ber of antiparticles. The value of the coupling constant in QED is
usually written in terms of the fine-structure constant There is a subtlety here, in that
these couplings are scale dependent
2
e 1
α= ≈ (8.28) α = α(µ)
4π~c 137
√ Once you include loop corrections,
which in natural units means for e = 4πα ≈ 0.3. the value of the fine-structure con-
stant grows logarithmically as
the energy scale is increased. The
8.7 Discrete symmetries of QED value here is associated with the
energy scale of the electron mass, so
In addition to the Lorentz transformations which are continuously α(µ = me ) ≈ 1/137, however e.g. at
the mass of the Z boson
connected to the 1, there are two other space-time operations which
can be symmetries of the Lagrangian: parity P and time-reversal α(µ = MZ ) ≈ 1/129
T . Additionally, there can be a particle-antiparticle symmetry called The scale dependence is encoded in
the renormalization group equation
charge conjugation C. We have already encountered parity as which you will discuss in the QFT
lecture
P : (t, x) → (t, −x)
∂e(µ) e(µ)3
≡ β(e) = + ...
which reverses handedness. Time reversal acts as ∂ ln µ 12π 2
where just show the leading order
T : (t, x) → (−t, x) dependence. The running of the
coupling is a loop effect as you see
both leave the Minkowski interval x2 = t2 − x2 invariant. Formally, from its e(µ)3 dependence.
we can say that the full Lorentz group breaks into four disconnected
subsets. We call the continuous Lorentztransformation the proper,
orthochronous Lorentz group L↑+
P
L↑+ L↑−
T T
P
L↓+ L↓−
Although a relativistic field theory must be invariant under L↑+ , it

is not necessarily invariant under C, P, or T . The gravitational,
electromagnetic and the strong force respect all three, but weak
interactions violate C and P separately, and certain suppressed pro-
cesses violate CP and T violation. The combination CPT is observed
in all processes so far in nature. There is also the famous CPT the-
orem which states that any Lorentz invariant local quantum field
theory with a Hermitian Hamiltonian must have CPT symmetry.7 7
It is very difficult to write a con-
In all the quantum field theories that you will encounter in this lec- sistent quantum field theory which
would lead to observable CPT viola-
ture (and in QFT1/2) you can therefore assume that CPT is good tion.
symmetry and as a consequence that CP = T , which means that
the measured violation of CP symmetry implies that nature is not
time-reversal symmetric.
8.7.1 Parity
We have already discussed parity extensively. Let us just add the
result for the gauge field and any four-vector
P : Aµ (x) = (A0 , A) → (A0 , −A)
the Dirac spinor in the Weyl basis exchanges ψL and ψR , whose

representation is
P : Ψ(x, t) → γ 0 Ψ(−x, t)
8.7.2 Charge conjugation

We now know what the meaning of charge is in an interaction and
hence we can talk about charge conjugation. We will try to flip the
charge e in
iγ µ (∂µ + ie Aµ )Ψ = 0
Start with the complex conjugation of the equation of motion
−iγ µ∗ (∂µ − ie Aµ )Ψ = 0
If we complex conjucate the Clifford algebra {γ µ , γ ν } = 2g µν 1,

we see that (−γ µ∗ ) also satisfies it. Therefore (−γ µ∗ ) must be the
gamma matrices expressed in a different basis, which means there
exists a matrix Cγ 0 such that Including γ 0 in the definition is
convention.
−γ µ∗ = (Cγ 0 )−1 γ µ Cγ 0 (8.29)
We plug this into the conjugate equation of motion to find
iγ µ (∂µ − ie Aµ )ΨC = 0
with
ΨC ≡ Cγ 0 Ψ∗
Let us find the specific form of the charge conjugation matrix C.

We can write Eq. (8.29) with (γ 0 )2 = 1 and so (γ 0 )−1 = γ 0 as
Cγ 0 γ µ∗ γ 0 C −1 = −γ µ
We know that the hermitian conjugate is (γ µ )† = γ 0 γ µ γ 0 and

taking the complex conjugate gives (γ µ )T = γ 0 (γ µ∗ )γ 0 if γ 0 is real.
Therefore we have
(γ µ )T = −C −1 γ µ C
which explains why C is defined with a γ 0 attached. In both bases

(Weyl and Dirac from the Ex-sheet), γ 2 is the only imaginary matrix,
so γ µ∗ = γ µ for µ 6= 2 and γ 2∗ = −γ 2 . Then the defining equation Recall, that
Eq. (8.29) just says, 2

0 σ2

γWeyl =
−σ 2 0
{Cγ 0 , γ µ } = 0, for µ 6= 2, with
0 2
[Cγ , γ ] = 0

0 −i
σ2 =
i 0
which means
C = −iγ 2 γ 0
up to an arbitrary phase8 . So we find the simple relation for the 8

We choose −i for convenience.
charge conjugate spinor where we multiply by (−i) for conve-
nience
ΨC = −iγ 2 Ψ∗ (8.30)
What does it mean in terms of the Weyl spinors? We find

! !
∗ ∗
2 ψL −iσ 2 ψR
ΨC = −iγ ∗
= ∗
ψR iσ 2 ψL
This is interesting! So the charge conjugate of a left-handed field is

right-handed and vice versa. In terms of the Lorentz representations, Experimentally, we know that the
we have found an important relation, if neutrino is left-handed. Thus, we
are now able to predict that the
∗ anti-neutrino is right-handed!
ψL : ( 12 , 0), then iσ 2 ψL : (0, 21 )
and vice versa. We can explicitly check that ΨC transforms as a This is due to the fact the represen-
spinor with the known Lorentz transformation, Eq. (7.36), Ψ → tations of SU (2) are pseudo-real.
i µν A representation and its complex
e− 4 ωµν σ Ψ we have for the complex conjugate conjugate are related to each other
µν ∗
by a simple transformation S, e.g.
i
Ψ∗ → e+ 4 ωµν (σ )
Ψ∗ for the complex doublet 2
2∗ = S −1 2S
hence
2 it is just
in this case with εij = iσij
2 + 4i ωµν (σ µν )∗ ∗
ΨC → −iγ e Ψ 2∗i = εij 2j
− 4i ωµν (σ µν )
=e ΨC Since for µ, ν 6= 2
∗
i
where we have used the fact, that [γ 2 , γ µ γ ν ] = 0 for µ, ν 6= 2 (σµν )∗ = [γµ , γν ]
2
and {γ 2 , γ 2 γ ν } = 0 for ν =
6 2. This shows that ΨC transforms i ∗ ∗
= − [γµ , γν ]
like a regular Dirac spinor and confirms our statement about the 2
i
representations of ψL and ψR being truly exchanging roles. = − [γµ , γν ]
2
= −σµν
since these γ matrices are real (only
γ 2 is purely imaginary). If one of the
indices µ = 2 or ν = 2, then e.g.
i
(σ2ν )∗ = − [(−)γ2 , (±)γν ]
2
8.7.3 Majorana neutrino

Because ΨC transforms as a spinor, Majorana understood that
/ = mΨ but
Lorentz invariance no only allows the Dirac equation i∂Ψ
also the Majorana equation
/ = mΨC
i∂Ψ (8.31)
This type of mass term is called a Majorana mass. We can obtain

it from the Lagrangian
1 T

L = Ψi∂Ψ/ − m ΨT CΨ + ΨCΨ
2
Since Ψ and ΨC carry opposite charge, we can use the Majorana
equation, unlike the Dirac equation, only for electrically neutral
fields. You can also see this from the Lagrangian: the mass term
breaks the U (1) symmetry Ψ → eiα Ψ. We can therefore also not
couple it to gauge fields. How does it look in components? The mass
contains terms like
ΨT CΨ = Cαβ Ψα Ψβ
but since
!
−iσ2 0
C = −iγ2 γ0 =
0 iσ2
and the anti-symmetric matrix tensor for i, j = 1, 2
i(σ2 )ij = εij
we have Cαβ = −Cαβ this seems to indicate that the term has
to vanish if the spinor fields Ψα commute. In your QFT lecture,
you will learn that spinors Ψ have to be treated as anticommuting
Grassmannian numbers,
Ψ1 Ψ2 = −Ψ2 Ψ1
which we have already (secretly) realised when defining time-ordered

or normal ordered products of Fermion fields.
In terms of the Weyl two-spinors Ψ = (ψL , ψR ) the mass term
looks like
1 1
mΨT CΨ + h.c. = − m(ψL T T
iσ2 ψL − mψR iσ2 ψR + h.c.)
2 2
We see that we could have been more economical and started with
just one Weyl fermion, ψL since the ψL and ψR fields are now not
mixing through the mass term.9 A minimal real Majorana field 9
As we know, they are irreducible
representations of the Lorentz group
would be
and will be separately Lorentz
covariant.
L = iψ̄L γ µ ψL + m(ψL
T
iσ2 ψL + h.c.) (8.32)
which brings us back to the initial mystery of how to give a single

Weyl fermion a Lorentz-invariant mass term. All that we needed to
save the naive guess in Eq. (7.7), was an anti-symmetric matrix.
†
iσ̄ µ ∂µ ψL − mψL iσ2 = 0
(e) D =0 (f) D= 1
What happens to spinor(g) bilinears under charge conjugation?

D =0 You
can easily check that
Figure 10.2. The seven QED µamplitudes µwhose superficial degree of di-
C : CΨγ ΨC → −Ψγ Ψ
vergence (D) is > 0. (Each circle represents the sum of all possible QED
diagrams.)
This means As that
explained in thej µtext,
the current amplitude
of QED changes (a) is irrelevant to scattering
sign.
processes, while amplitudes (b) and (d) vanish because of symmetries. Am-
plitude
8.7.4(e) Furrys
is nonzero, but its divergent parts cancel due to the Ward identity.
theorem
The remaining amplitudes ( c, f, and g) are all logarithmically divergent, even
We can
though D > start with
0 for (c) the
andphoton
(f). one-point function. We note that the
external photon must be attached to a QED vertex. Neglecting the
photonexternal
propagator, this the
propagator, amplitude
amplitudeisistherefore
hence
- q
=
Z
-ie J
4
x e-iq·x
d−iq·x
= −ie d4 xe hΩ|T jµ(OJ T Jµ(x)(8.33)
(x)|Ωi JO), (10.5)
-
where jl' = Ψγisµ Ψ
= j'ljryl''lj;
where µ the electromagnetic
is the electro-magneticcurrent operator.The
current operator. But the vacuum
expectation value of jl' must vanish
µ by Lorentz invariance,
vacuum expection value of j has to vanish because of Lorentz- since otherwise it
would be a preferred
invariance: 4-vector.
otherwise j µ would be a preferred direction in Minkowski
Thespace.
photon one-point function also vanishes for a second reason: charge-
conjugation
Theinvariance.
photon one-point Recall that C
function hasistoa vanish
symmetry ofother
also for QED, rea-so C JO) = JO).
But jl'(x)
sons,changes
namely charge sign under charge
conjugation Since CCjl'(x)Ct
conjugation,
invariance. is symmetry = of-jl'(x), so its
vacuumQED,
expectation value must vanish:
we know that the vacuum |0i and the vacuum of the interact-
ing theory
C|Ωi = |Ωi
The same argument applies to any vacuum expectation value of an odd num-
are C invariant. A currents.
ber of electromagnetic diagram with n external photons
In particular, (and nothree-point
the photon external function,
fermions) is proportional to
Fig. 10.2d, vanishes. (This result is known as Furry's theorem.) It is not hard
to check explicitly that the hΩ|Tphoton
jµ1 (x1 ) . .one- and
. jµn (x three-point functions vanish in
n )|Ωi
the leading order of perturbation theory (see Problem 10.1).
TheIf remaining
n is odd thenamplitudes in Fig. 10.2 are all nonzero, so we must analyze
their structures in more detail. Consider, for example, the electron self-energy
hΩ|T j (x ) . . . j (x )|Ωi = hΩ|T CCj (x )CC . . . CCj (x )CC|Ωi
µ1 1 µn n µ1 1 µn n
= hΩ|T (Cjµ1 (x1 )C)(C . . . C)(Cjµn (xn )C)|Ωi

Figure 8.3: Example of a vanishing
= (−1)n hΩ|T jµ1 (x1 ) . . . jµn (xn )|Ωi 3-point function.
This means that the a pure photon n-point function vanishes, for n
odd. This is known as Furry’s theorem.
8.8 Time reversal
8.8.1 Quantum mechanics

We will now complete our discussion of the discrete symmetries of
QED and discuss time reversal invariance. Time reversal symmetry
is a bit subtle as you might already know from your quantum me-
chanics course. Let us investigate this first using the Schrödinger
equation of QM We are suppressing the trivial x
dependence in this section.
∂
i ψ(t) = Hψ(t) (8.34)
∂t
1
with for concreteness think H = − 2m ∇2 − V (x). We now consider
the time reversal transformation
t → t0 = −t
Our goal is now to find ψ 0 (t0 ) which satisfies
∂
iψ(t0 ) = Hψ 0 (t0 ) (8.35)
∂t0
We write for the transformed field
ψ 0 (t0 ) = T ψ(t)
where T is an operator which we will determine in the following. Let Again, we can fix it up to an arbi-
us plug this into Eq. (8.35), we obtain trary phase.
∂
i T ψ(t) = HT ψ(t)
∂(−t)
∂
T −1 i T ψ(t) = T −1 HT ψ(t)
∂(−t)
Since H is time-independent by assumption, the time-inversion
operator has no effect and
T −1 H = H T −1 ,
which results in
∂
T −1 (−i)T ψ(t) = Hψ(t)
∂t
Comparing to Eq. (8.34), we are no forced to conclude
T −1 (−i)T = i (8.36)
We can define
T ≡UK
where K complex conjugates everything to the right. Applied to For real φ1 , φ2 we have
Eq. (8.36) we get with T −1 = KU −1 K(φ1 + iφ2 ) = φ1 − iφ2
KU −1 (−i)U K = U −1∗ (+i)U ∗ K 2 = U −1∗ (+i)U ∗

We see that it does the job, if U −1 iU = i, that is, if U −1 is just an
ordinary (unitary) operator that leaves i be.10 The contribution of 10
We will determine U below
K to T makes it antiunitary.
8.8.2 Scalar Field

Let us check how T acts on a scalar state solving the free Klein-
Gordon equation φ(t) = e−ip·x = ei(k·x−Et) . We have
φ0 (t0 ) = T φ(t) = U Kφ(t) = U φ∗ (t) = U e−i(k·x−Et)
Since φ has just one component, U is a trivial phase-factor.11 We 11
Since we require |φ|2 = |φ0 |2 , we
can rewrite this as choose the phase factor here to be
U = 1.
φ0 (t) = e−i(k·x+Et) = ei(−k·x−Et)
from which we see that φ0 describes a wave moving in the opposite
direction. The energy is still positive since ψ ∝ e−iEt . We see that
two time-reversals are equal to the identity
T 2 = U KU K = U U ∗ K 2 = +1
8.8.3 Spinor Field

Now we need to consider how T acts on spinors. The Dirac equa-
tion is after multiplying by γ 0 from the left
∂
i Ψ(t) = HΨ(t)
∂t
with H = −iγ 0 γ i ∂i + γ 0 m. As above, we want
∂ 0 0
i Ψ (t ) = HΨ(t0 )
∂t0
with
Ψ0 (t0 ) = T Ψ(t)
Following our discussion above, we can make similar arguments if

T −1 HT = H or
KU −1 HU K = H
which means we require
KU −1 γ 0 U K = γ 0
and
KU −1 (iγ 0 γ i )U K = iγ 0 γ i
If we move K to the RHS, we see that we can determine U with
U −1 γ 0 U = γ 0∗ , and U −1 γ i U = −γ i∗
Restricting ourselves to the Weyl (and Dirac) basis, we know that γ 2

is the only imaginary matrix. So we need to find a U which flips γ 1
and γ 3 but not γ 0 and γ 2 ? This choice works12 12
Again, with an arbitrary phase
which we set to 1.
U = γ1γ3
and so
 
0 1
−1 0 
 
Ψ(t, x) → γ 1 γ 3 Ψ(−t, x) =   Ψ(−t, x)
 0 1
−1 0
Thus T flips the spin of particles.
8.9 Electron scattering and the Coulomb potential

We describe electron scattering e− e− → e− e− with the two tree-level
Feynman diagrams as
q q0
q
q0
+ (8.37)
p p0
p p0
Figure 8.4: Summary of C, P, T .
"
(−iηµν )
2
iA = (−ig) ū(q 0 )γ µ u(q) ū(p0 )γ ν u(p)
(q − q 0 )2 − µ2
#
0 µ (−iηµν ) 0 ν
− ū(p )γ u(q) 0 ū(q )γ u(p) (8.38)
(p − q)2 − µ2
Compared to the scalar mediator we seem to find an overall minus
sign, but remember that the the expression is really positive for
µ, ν = 1, 2, 3.
8.9.1 Coulomb potential

We take the first diagram of e− e− → e− e− and calculate the non-
relativistic limit. The steps proceed as in Sec. 6.7.1 and Sec. 7.8.4
for the Yukawa potential. The only difference is that we now need to
evaluate ūγ µ u in Eq. (8.38) using the non-relativistic spinor
!
√ ξ
u(p) → m
ξ
We see that
ūr (p)γ 0 us (q) → 2mδ rs
ūr (p)γ i us (q) → 0
Comparing the scattering amplitude in the non-relativistic limit
to the quantum mechanical expression again, we find an effective
potential between two electrons
Z
d3 p eip·r e2
U (r) = +e2 = +
(2π)3 |r|2 4πr
Which is the familiar repulsive Coulomb potential between two
electrons. What about e− e+ → e− e+ scattering? The relevant
diagram in the non-relativistic limit is
q q0
p p0
which gives
" #
2 0 µ (−iηµν ) 0 ν
iA = −(−ig) v̄(q )γ v(q) ū(p )γ u(p) (8.39)
(q − q 0 )2
The overall + sign comes from treating the fermions correctly: we

saw the same minus sign when studying scattering in Yukawa theory.
The non-relativistic limit is now difference however, since v̄γ 0 v → 2m
which gives us Compare to the Yukawa potential
where the limit is v̄v → −2m.
Z
d3 p eip·r e2
U (r) = −e2 = −
(2π)3 |r|2 4πr
As expected, we find an attractive force between an electron and a

positron. We can see the same effect without all the spinor-ology in
scalar QED. The interaction is (Dµ φ)∗ Dµ φ and so the relevant term
for the scalar-photon interaction is
−ieAµ (φ∗ ∂ µ φ − φ∂µ φ∗ ) + e2 Aµ Aµ φ∗ φ
the two vertices are

q
p
= −ie(p + q)µ , = +2ie2 ηµν
(8.40)
The momentum dependence comes from the ∂ µ term and the factor
of 2 is because of the identical particles Aµ Aµ appearing in the
Lagrangian Similar to the cancellation of the 1/4!
We can see that the difference in sign comes from the A0 piece in the φ4 vertex.
of the propagator −iηµν /p2 , we have in the non-relativistic limit of

scalar e− e− → e− e− scattering
(p + p0 )µ (q + q 0 )ν (2m)2
= −iηµν (−ie)2 0
→ −i(−ie)2
(p − p) 2 −(p − p0 )2
where the numerator selects the A0 piece since (p + p0 )µ (q + q 0 )µ ≈

(p + p0 )0 (q + q 0 )0 ≈ (2m)2 . The Coulomb potential for spin 0 particles
is again repulsive. If switch to scalar e + e− → e + e− scattering,
one of the arrows on the legs changes direction and the amplitude
picks up an additional minus sign because since the charge arrows
are correlated with momentum arrows. This leads as expected to an
attractive potential
9
Outlook
If you found this interesting you should continue and attend the
QFT course in the next semester. You have yet to encounter many
of the truly awesome and deep insights that quantum field theory
contains. It is the language in which the laws of Nature are written.
As Sidney Coleman said: ”Not only God knows, I know, and by the
end of the semester, you will know.”

QFT Lecture Notes

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

QFT Lecture Notes

Hochgeladen von

Copyright:

Verfügbare Formate

ANDREAS WEILER, TUM

Der wackre Schwabe forcht’ sich nit,

c 2018 Andreas Weiler, TUM

Sommersemester 2017, Version vom December 31, 2018

3.4 Indices upstairs and downstairs 33

4 Classical Field Theory 37

5.3 Smeared operator valued distributions 60

6 Interacting quantum fields 81

6.8 How to calculate observables 104

7 Lorentz Representations 117

7.6 Perturbation Theory for Spinors 143

8 Quantum Electro Dynamics 153

These notes are not original but mostly a combination of results

• Zee - Quantum Field Theory in a Nutshell ?

• Srednicki - Quantum Field Theory ? ? ?

• Weinberg - Quantum Field Theory I ? ? ??

• Ramond - Field Theory: a Modern Primer ? ? ?

• Itzykson, Zuber - Quantum Field Theory ? ? ?

1.1 Units and conventions

There are three fundamental dimensionful constants in nature: the

[c] = length × time−1 (1.1)

length = time = mass−1 = energy−1 (1.5)

we can express all dimensionful quantities in terms of a single scale

this is the Planck scale. It corresponds to a length lP ≈ 10−33 cm.

(1 GeV)−1 (~c) = 0.1973 fm (1.7)

Table 1.1: Overview over the relevant

Cosmological constant Λ1/4 10−3 eV 10−4 m

Neutrino masses mν ≤ 0.12 eV ≥ 10−9 m

Electron mass me− 0.511 keV 3 · 10−13 m

Muon mass mµ− 105.7 MeV 1.9 · 10−15 m

Proton mass mp 938.272 MeV 2.1 · 10−16 m

Z boson mass mZ 91.1876 GeV 2.16 · 10−18 m

Higgs mass mh 125.09 GeV 1.57 · 10−18 m

Top quark mass mPole

Planck Mass MP 2.4 · 1018 GeV 8 · 10−35 m

2.1 Why quantum field theory?

• All elementary particles look the same. An electron at the begin-

Elementary particles are then local, quantized excitations of this

• Once we try to combine special relativity (SR) with quantum

Relativistically, momentum and energy are on the same footing

2.1.1 Necessity of the field viewpoint: causality

U (t) = hx|e−iHt |x0 i (2.3)

The energy of a free particle in non-relativistic quantum mechanics is

points in an arbitrarily short time. In a relativistic theory, no signal

and taking the positive root5 5

has a stationary point at

Substituting this value back for p, we find that up to a rational

violations of causality. This is in accordance with our earlier ar-

2.1.2 General features of relativistic quantum field theories

• Very constraining for higher spins, only S ≤ 2 theories.

• Massless particles (m = 0): S = 1 gauge theory (like electro-

2.2 String Theory

2.2.1 The discrete chain

The coordinate qj (t) is the longitudinal displacement of the j-th

The remaining N masses satisfy the equation of motion (EOM)

mq̈j − κ(qj+1 − 2qj + qj−1 ) = 0 (2.16)

for j = 1, . . . , N . The normal modes diagonalize the EOMs and have

We find the continuum limit by taking the mass distance a to zero

In this limit the discrete chain approaches a continuous string and

q(x, t) ≡ qj (t) (2.22)

We can take the limit also for the Lagrangian by replacing

Keeping the mass density ρ and string tension σ finite, we define

to obtain the Lagrangian

and the EOM