Quantum Field Theory I

Babis Anastasiou
Institute for Theoretical Physics,
ETH Zurich,
8093 Zurich, Switzerland
E-mail: babis@phys.ethz.ch
January 23, 2010
Contents
1 Quantum Field Theory. Why? 7
2 Theory of Classical Fields 9
2.1 The principle of least action . . . . . . . . . . . . . . . . . . . 9
2.2 Fields from a discretized space (lattice) . . . . . . . . . . . . . 10
2.3 Euler-Lagrange equations for a classical eld from a Lagrangian
density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Field Hamiltonian Density . . . . . . . . . . . . . . . . . . . . 16
2.5 An example: acoustic waves . . . . . . . . . . . . . . . . . . . 17
2.6 Noethers theorem . . . . . . . . . . . . . . . . . . . . . . . . 19
2.6.1 Field symmetry transformations . . . . . . . . . . . . . 19
2.6.2 Space-Time symmetry transformations . . . . . . . . . 20
3 Quantization of the Schrodinger eld 30
3.1 The Schrodinger equation from a Lagrangian density . . . . . 30
3.2 Quantization of Fields . . . . . . . . . . . . . . . . . . . . . . 32
3.3 Quantized Schrodinger eld . . . . . . . . . . . . . . . . . . . 33
3.4 Particle states from quantized elds . . . . . . . . . . . . . . . 35
3.5 What is the wave-function in the eld quantization formalism? 39
4 The Klein-Gordon Field 42
4.1 Real Klein-Gordon eld . . . . . . . . . . . . . . . . . . . . . 42
4.1.1 Real solution of the Klein-Gordon equation . . . . . . . 43
4.1.2 Quantization of the real Klein-Gordon eld . . . . . . . 45
4.1.3 Particle states for the real Klein-Gordon eld . . . . . 46
4.1.4 Energy of particles and normal ordering . . . . . . . 47
4.1.5 Field momentum conservation . . . . . . . . . . . . . . 49
4.1.6 Labels of particle states? . . . . . . . . . . . . . . . . . 50
1
4.2 Two real Klein-Gordon elds . . . . . . . . . . . . . . . . . . . 50
4.2.1 Two equal-mass real Klein-Gordon elds . . . . . . . . 52
4.2.2 Two real Klein-Gordon elds = One complex Klein-
Gordon eld . . . . . . . . . . . . . . . . . . . . . . . . 55
4.3 Conserved Charges as generators of symmetry transformations 57
4.4 Casimir eect: the energy of the vacuum . . . . . . . . . . . . 58
4.5 Can the Klein-Gordon eld be an one-particle wave-function? 62
5 The Dirac Equation 64
5.1 Mathematical interlude . . . . . . . . . . . . . . . . . . . . . . 65
5.1.1 Pauli matrices and their properties . . . . . . . . . . . 65
5.1.2 Kronecker product of 2 2 matrices . . . . . . . . . . 66
5.2 Dirac representation of -matrices . . . . . . . . . . . . . . . . 67
5.3 Traces of matrices . . . . . . . . . . . . . . . . . . . . . . 69
5.4 matrices as a basis of 4 4 matrices . . . . . . . . . . . . . 69
5.5 Lagrangian for the Dirac eld . . . . . . . . . . . . . . . . . . 71
6 Lorentz symmetry and free Fields 73
6.1 Field transformations and representations of the Lorentz group 75
6.2 Scalar representation M() = 1 . . . . . . . . . . . . . . . . . 76
6.3 Vector representation M() = . . . . . . . . . . . . . . . . . 77
6.4 How to nd representations? . . . . . . . . . . . . . . . . . . . 77
6.4.1 Generators of the scalar representation . . . . . . . . . 80
6.4.2 Generators of the vector representation . . . . . . . . . 81
6.5 Spinor representation . . . . . . . . . . . . . . . . . . . . . . . 82
6.6 Lorentz Invariance of the Dirac Lagrangian . . . . . . . . . . . 83
6.7 General representations of the Lorentz group . . . . . . . . . . 85
6.8 Weyl spinors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.8.1 The Dirac equation with Weyl spinors . . . . . . . . . 89
7 Classical solutions of the Dirac equation 90
7.1 Solution in the rest frame . . . . . . . . . . . . . . . . . . . . 91
7.2 Lorentz boost of rest frame Dirac spinor along the z-axis . . . 92
7.3 Solution for an arbitrary vector . . . . . . . . . . . . . . . . . 95
7.4 A general solution . . . . . . . . . . . . . . . . . . . . . . . . . 95
2
8 Quantization of the Dirac Field 97
8.1 One-particle states . . . . . . . . . . . . . . . . . . . . . . . . 99
8.1.1 Particles and anti-particles . . . . . . . . . . . . . . . . 100
8.1.2 Particles and anti-particles of spin-
1
2
. . . . . . . . . . 102
8.2 Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
8.3 Lorentz transformation of the quantized spinor eld . . . . . . 105
8.3.1 Transformation of the ladder operators . . . . . . . . . 105
8.3.2 Transformation of the quantized Dirac eld . . . . . . 108
8.4 Parity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
8.5 Other discrete symmetries . . . . . . . . . . . . . . . . . . . . 112
9 Quantization of the free electromagnetic eld 114
9.1 Maxwell Equations and Lagrangian formulation . . . . . . . . 114
9.2 Quantization of the Electromagnetic Field . . . . . . . . . . . 118
10 Propagation of free particles 123
10.1 Transition amplitude for the Schrodinger eld . . . . . . . . . 123
10.2 Transition amplitude for the real Klein-Gordon eld . . . . . . 124
10.3 Time Ordering and the Feynman-St uckelberg propagator for
the real Klein-Gordon eld . . . . . . . . . . . . . . . . . . . . 128
10.4 Feynman propagator for the complex Klein-Grodon eld . . . 130
10.5 Feynman propagator for the Dirac eld . . . . . . . . . . . . . 131
10.6 Feynman propagator for the photon eld . . . . . . . . . . . . 132
11 Scattering Theory (S-matrix) 133
11.1 Propagation in a general eld theory . . . . . . . . . . . . . . 134
11.1.1 A special case: free scalar eld theory . . . . . . . . . . 138
11.1.2 Typical interacting scalar eld theory . . . . . . . . 139
11.2 Spectral assumptions in scattering theory . . . . . . . . . . . . 141
11.3 In and Out states . . . . . . . . . . . . . . . . . . . . . . 142
11.4 Scattering Matrix-Elements . . . . . . . . . . . . . . . . . . . 144
11.5 S-matrix and Greens functions . . . . . . . . . . . . . . . . . 145
11.6 The LSZ reduction formula . . . . . . . . . . . . . . . . . . . . 147
11.7 Truncated Greens functions . . . . . . . . . . . . . . . . . . . 150
11.8 Cross-sections

. . . . . . . . . . . . . . . . . . . . . . . . . . 151
3
12 Perturbation Theory and Feynman Diagrams 152
12.1 Time evolution operator in the interaction picture . . . . . . . 154
12.2 Field operators in the interacting and free theory . . . . . . . 156
12.3 The ground state of the interacting and the free theory . . . . 158
12.4 Wicks theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 160
12.5 Feynman Diagrams for
4
theory . . . . . . . . . . . . . . . . 164
12.6 Feynman rules in momentum space . . . . . . . . . . . . . . . 167
12.7 Truncated Greens functions in perturbation theory . . . . . . 171
13 Loop Integrals 174
13.1 The simplest loop integral. Wick rotation . . . . . . . . . . . . 174
13.2 Dimensional Regularization . . . . . . . . . . . . . . . . . . . 177
13.2.1 Angular Integrations . . . . . . . . . . . . . . . . . . . 178
13.2.2 Properties of the Gamma function . . . . . . . . . . . . 180
13.2.3 Radial Integrations . . . . . . . . . . . . . . . . . . . . 181
13.3 Feynman Parameters . . . . . . . . . . . . . . . . . . . . . . . 182
14 Quantum Electrodynamics 186
14.1 Gauge invariance . . . . . . . . . . . . . . . . . . . . . . . . . 186
14.2 Perturbative QED . . . . . . . . . . . . . . . . . . . . . . . . 189
14.3 Dimensional regularization for QED . . . . . . . . . . . . . . . 192
14.3.1 Gamma-matrices in dimensional regularization . . . . . 193
14.3.2 Tensor loop-integrals . . . . . . . . . . . . . . . . . . . 194
14.4 Electron propagator at one-loop . . . . . . . . . . . . . . . . . 198
14.5 Photon propagator at one-loop . . . . . . . . . . . . . . . . . . 201
15 One-loop renormalization of QED 202
4
Bibliography
[1] The Quantum Theory of Fields, Volume I Foundations, Steven Wein-
berg, Cambridge University Press.
[2] An introduction to Quantum Field Theory, M. Peskin and D. Schroeder,
Addison-Wesley
[3] Quantum Field Theory in a nutshell, A. Zee, Princeton University Press.
[4] Quantum Field Theory, Mark Srednicki, Cambridge University Press.
[5] An introduction to Quantum Field Theory, George Sterman, Cambridge
University Press.
[6] Classical Mechanics, Goldstein, Poole and Safko, Addison-Wesley
[7] Lectures On Qed And Qcd: Practical Calculation And Renormalization
Of One- And Multi-loop Feynman Diagrams, Andrea Grozin, World
Scientic
5
Empty!
6
Chapter 1
Quantum Field Theory. Why?
The goal of this lecture series is to introduce a beautiful synthesis of quan-
tum mechanics and special relativity into a unied theory, the theory of
quantized elds. We have already seen in the course of Quantum Mechanics
II that quantized elds can be used to describe systems of many identical par-
ticles (fermions or bosons). In this formalism, a quantum-mechanical state
of particles at xed positions is generated from the vacuum state, containing
no particles, after we act on it with eld operators,
[ x
1
, x
2
, . . . , x
n
) f

( x
1
, t)f

( x
2
, t) . . . f

( x
n
, t) [0) . (1.1)
We can also prove (QMII) that there is a wave-function for non-interacting
identical particles which satises the Schrodinger equation, if the eld oper-
ator itself f(x, t) satises the Schrodinger equation. The power of the eld
theory formulation is that particle-exchange symmetry for bosons and anti-
symmetry for fermions are automatically satised, if we impose commutation
quantization conditions for boson elds and anti-commutation quantization
conditions for fermion elds:
[f( x
1
, t), f( x
2
, t)]

=
_
f

( x
1
, t), f

( x
2
, t)

= 0
_
f( x
1
, t), f

( x
2
, t)

=
(3)
( x
1
x
2
), (1.2)
where the commutator (- for bosons) and anticommutator (+ for fermions)
are dened as,
[A, B]

= AB BA. (1.3)
For example, we can easily verify that the Pauli exclusion principle for iden-
tical fermions is automatically respected. A state of two identical fermions
7
in the same position is, in quantum eld theory,
[x, x) f

(x, t)f

(x, t) [0) =
1
2
_
f

(x, t), f

(x, t)

+
[0) = 0 [0) = 0, (1.4)
which vanishes.
In non-relativistic quantum mechanics, quantized elds oer an elegant
framework which is convenient to describe a system of many identical par-
ticles. However, an equivalent description in terms of wave-functions yields
the same physical results and it is always possible. Traditional quantum me-
chanics fails when we must merge it with special relativity. A number of new
phenomena manifest themselves at small distances (quantum eects) and
relativistic velocities; these are routinely studied at modern collider experi-
ments. An example physical processes, exhaustivelly studied at the Large-
Electron-Positron (LEP) collider in Geneva, is the production of a muon and
its anti-particle out of the annihilation of an electron and a positron:
e

+ e
+

+
+
. (1.5)
Such a phenomenon cannot be explained with a wave-function description.
We can start with a wave-function for the electron/positron system a long
time before they collide. A long time after the collision, we have a new
wave-function for two dierent particles, the muon and the anti-muon. The
Schrodinger equation, and the corresponding time evolution, does not predict
wave-functions which are destroyed and replaced by others.
Quantized elds allow for the creation and annihilation of particles. From
our example reaction, we conclude that there exist an electron eld and a
muon eld. As we shall see later, the quantization of these elds introduces
creation and annihilation operators for the electron the muon and their anti-
particles. In Quantum Field Theory, it is also possible to describe how elds
interact, and to compute the probability for a state with a certain particle
content to mutate to another state with a dierent particle content after
some time. Field quantization is an essential part of quantum relativistic
laws, and not an optional formalism.
8
Chapter 2
Theory of Classical Fields
We start by introducing a Lagrangian formalism for elds at the classical
level.
2.1 The principle of least action
The Lagrangian formalism has been introduced in classical mechanics for
systems with a nite number of degrees of freedom. We review here the salient
features of the formalism. We consider a physical system with
i
(t), i =
1 . . . N generalized coordinates. For any time interval t
2
t
1
, the action is
given by,
S =
_
t
2
t
1
dtL[
1
(t), . . .
N
(t),
1
(t), . . .
N
(t)] , (2.1)
where L[
i
,
i
] is the Lagrangian function of the system. According to the
principle of the least action, the physical generalized coordinates yield a
stationary value for the action:
S
_

i
=
phys
i
_
= S
_

i
=
phys
i
+ a
i
_
, (2.2)
where
i
(t
1
) =
i
(t
2
) = 0, and a is a small parameter deforming slightly the
generalized coordinate functions from their physical solution
phys
i
(t).
Eq. 2.2 determines the dynamical behaviour of the physical coordinates
9

phys
i
(t). Performing a Taylor expansion in a through O(a), we obtain,
_
t
2
t
1
dt
N

j=1
_
L
_

phys
,
phys

phys
j

j
+
L
_

phys
,
phys

phys
j

j
_
= 0
;
_
t
2
t
1
dt
N

j=1
_
L
_

phys
,
phys

phys
j

d
dt
_
L
_

phys
,
phys

phys
j
__

j
+
_
t
2
t
1
dt
d
dt
_
N

j=1
L
_

phys
,
phys

phys
j

j
(t)
_
= 0
;
_
t
2
t
1
dt
N

j=1
_
L
_

phys
,
phys

phys
j

d
dt
_
L
_

phys
,
phys

phys
j
__

j
= 0,
(2.3)
where the integral over the total derivative is zero due to
j
(t
1
) =
j
(t
2
) = 0.
The functions
j
(t) are arbitrary smooth functions for t ,= t
1
, t
2
; for the
above integral to vanish for arbitrary choices of
j
, the physical generalized
coordiates must obey the Euler-Lagrange equations
L
_

phys
,
phys

phys
j

d
dt
_
L
_

phys
,
phys

phys
j
_
= 0. (2.4)
2.2 Fields from a discretized space (lattice)
A classical eld is a continuous function dened at every point in space-time.
For example, the amplitude of a mechanical wave or an electromagnetic wave
are elds. We may derive classical equations of motion for elds using the
principle of least action, which we have postulated for a discrete number of
generalized coordinates in the last section. An extension of the formalism to
elds is made, if we rst assume that a eld function takes non-zero values
only at discrete points of a lattice which spans all space, and then take the
continuous limit of a zero distance between the lattice points.
As a simple example, we shall study the vibration motion in an one-
dimensional elastic rod. The rod has a mass density . For an elastic rod,
the force applied on it is proportional to the elongation or the compression
per unit length () of the rod,
F = Y , (2.5)
10
where Y is a constant called Youngs modulus.
A non-vibrational rod can be seen as an ininite number of equally spaced
particles at rest, all having the same mass m, with a relative distance a. For
a small spacing a, we have
=
dm
dx
= lim
a0
m
a
. (2.6)
A vibration is created when these particles are displaced from their positions
at rest. We will assume that each particle in the lattice can interact with its
immediate neighbours only. When two neighboring particles are displaced to
a larger relative distance, an attractive force tends to bring them back to their
resting position, while a repulsive force is developed when they are found at
a shorter relative distance. Such a force can be approximated (elastic rod) by
considering an elastic spring between the particles, with a Hookes constant
. The force required to elongate one spring is,
F = (y
i+1
y
i
) = (a)
y
i+1
y
i
a
= (a) . (2.7)
We can then relate the microspopic Hookes constant to the macroscopic
Youngs modulus:
Y = lim
a0
(a) . (2.8)
The potential energy in the rod is equal to the sum of the potential
energies in all springs,
V =

i
1
2
y
2
i
=

i
1
2
(y
i+1
y
i
)
2
, (2.9)
where y
i
= the expansion or contraction of the spring in between particle
i and particle i + 1, and y
i
is the displacement of particle i from its position
at rest. The kinetic energy of all particles is,
T =

i
1
2
m y
2
i
. (2.10)
The Lagrangian of the system is simply given by,
L = T V
; L =

i
_
1
2
m y
2
i

1
2
(y
i+1
y
i
)
2
_
. (2.11)
11
We can easily nd the equations of motion for each particle j in the discretized
rod, from the Euler-Lagrange equations 2.4,
d
dt
_
L
y
k
_

L
y
k
= 0
;
d
dt
_
T
y
k
_
+
V
y
k
= 0
;
d
dt
_

i
1
2
m y
2
i
y
k
_
+

i
1
2
(y
i+1
y
i
)
2
y
k
= 0
;
d
dt
_

1
2
m y
2
k
y
k
_
+

_
1
2
(y
k+1
y
k
)
2
+
1
2
(y
k
y
k1
)
2

y
k
= 0
; m y
k
+ (y
k
y
k1
) (y
k+1
y
k
) = 0. (2.12)
Let us now consider the limit that the spacing in the discretized rod tends
to zero. The vibration amplitude y
k
(t), denes a continuous function at each
position x in the one-dimensional rod.
y
k
(t) y(x, t)
y
k+1
(t) y(x + a, t)
y
k1
(t) y(x a, t)
y
k+2
(t) y(x + 2a, t)
y
k2
(t) y(x 2a, t)
. . .
The equations of motion are written as,
m y(x, t) [(y(x + a, t) y(x, t)) (y(x, t) y(x a, t))] = 0
;
m
a
y(x, t)
_
y(x + a, t) y(x, t)
a

y(x, t) y(x a, t)
a
_
= 0
;
m
a
y(x, t) (a)
y(x+a,t)y(x,t)
a

y(x,t)y(xa,t)
a
a
= 0. (2.13)
Taking the limit a 0, we obtain the equation of motion for the vibration
eld y(x, t),

2
y(x, t)
t
2
Y

2
y(x, t)
x
2
= 0. (2.14)
12
It is interesting to take the continous limit in the expression of Eq. 2.11
for the Lagrangian as well. We have,
L =

i
a
_
1
2
m
a
y
2
i

1
2
(a)
_
y
i+1
y
i
a
_
2
_
. (2.15)
In the continuoum limit, the summation turns into an integration over the
position variable x, leading to
L =
_
dx
_
1
2

_
y(x, t)
t
_
2

1
2
Y
_
y(x, t)
x
_
2
_
. (2.16)
The action is then written as an integral in time and space dimensions,
S =
_
dtdx/, (2.17)
over a Lagrangian density,
/ =
1
2

_
y(x, t)
t
_
2

1
2
Y
_
y(x, t)
x
_
2
. (2.18)
Can we obtain the equations of motion (Eq. 2.14) from the action integral
over the Lagrangian density of Eq. 2.17 using the principle of least action?
If so, we can avoid the cumbersome discretization derivation and work using
a direct formalism for continuoum systems. We require that

_
dtdx/
_
y,
y
t
,
y
x
_
= 0, (2.19)
if we vary the eld y(x, t) from its physical solution,
y(x, t) y(x, t) + y(x, t). (2.20)
We also require that the variation of the eld vanishes at the boundaries of
the integrations,
y(x, t
1
) = y(x, t
2
) = 0, y(x
1
, t) = y(x
2
, t) = 0. (2.21)
13
We then have,

_
dtdx/
_
y,
y
t
,
y
x
_
= 0
;
_
dtdx
_
/
y
y +
/
(
t
y)
(
t
y) +
/
(
x
y)
(
x
y)
_
= 0 (2.22)
The eld variation commutes with space and time derivatives, and we can
rewrite the above equation as,
_
dtdx
_
/
y
y +
/
(
t
y)

t
(y) +
/
(
x
y)

x
(y)
_
= 0
;
_
dtdx
__
/
y

t
/
(
t
y)

x
/
(
x
y)
_
y
+
t
_
/
(
t
y)
y
_
+
x
_
/
(
x
y)
y
__
= 0. (2.23)
Since the eld variation y is generic and it vanishes at the boundaries, we
obtain an Euler-Lagrange dierential equation for y(x, t),
/
y

t
/
(
t
y)

x
/
(
x
y)
= 0. (2.24)
Substituting into this equation the Lagrangian density for the elastic rod, we
nd the following equation of motion:

2
y(x, t)
t
2
Y

2
y(x, t)
x
2
= 0, (2.25)
which is the same as Eq. 2.12
2.3 Euler-Lagrange equations for a classical
eld from a Lagrangian density
We generalize readily the Lagrangian formalism developed for the one-dimensional
elastic rod example, to elds which are dened in four space-time dimen-
sions x

= (x
0
= ct, x
1
= x, x
2
= y, x
3
= z). In this lecture series, we
use natural units, setting h = c = 1, and the space-time metric is dened as
14
g

= diag(1, 1, 1, 1). We consider a system of N elds

(i)
, i = 1, . . . , N
with a Lagrangian density which depends only on the elds and their rst
(for simplicity) space-time derivatives
/ = /
_

(i)
,

(i)

. (2.26)
The action is given by the integral,
S
_
t
2
t
1
dx
0
_
V
d
3
x/
_

(i)
,

(i)

. (2.27)
and we consider small variations
(i)
(x, t)
(i)
(x, t) +
(i)
(x, t) of the
elds from their physical solutions which are zero at the times t
1
, t
2
or at the
boundary surface S(V ) of the volume V, but arbitrary otherwise. Applying
the variational principle, we obtain
S = 0
;
_
d
4
x/ = 0
;
_
d
4
x
_
/

(i)

(i)
+
/
(

(i)
)

(i)
_
_
= 0
;
_
d
4
x
_
/

(i)

/
(

(i)
)
_

(i)
+
_
d
4
x

_
/
(

(i)
)

(i)
_
= 0
;
_
d
4
x
_
/

(i)

/
(

(i)
)
_

(i)
= 0, (2.28)
where we have used the conditions for vanishing eld variations at the bounar-
ies. Given that the functional form of the eld variation
(i)
is arbitrary,
we obtain the Euler-Lagrange dierential equations,
/

(i)

/
(

(i)
)
= 0. (2.29)
From the above derivation we can easily see that two Lagrangian densities,
/ and /+

(
(i)
,
(i)
), which dier by a total divergence yield identical
Euler-Lagrange identities. Indeed the variation of the total divergence term
in the action integral is zero,

_
V,T
d
4
x

=
_
S(V,T)
d

=
_
S(V,T)
d

= 0. (2.30)
In the above we have used Stokes theorem and that the variation of a smooth
function F of the elds and their derivatives vanishes at the boundaries of
the action integral.
15
2.4 Field Hamiltonian Density
For a system with a nite number of degrees of freedom, we nd the conjugate
momentum of a generalized coordinate by dierentiating the Lagrangian with
respect to the time derivative of the coordinate:
p
j
=
L
q
j
(2.31)
The Hamiltonian of the system is then given by,
H =

j
p
j
q
j
L. (2.32)
Let us now start from the Lagrangian of a eld (in one dimension for
simplicity),
L =
_
dx/, (2.33)
which after discretization becomes
L =

i
aL
i
, (2.34)
where L
i
is the value of the Lagrangian density at the point i of the lattice
(discretized line in our case). The eld q(x, t) has a value q
i
(t), at the same
point. The corresponding conjugate mometum is,
p
j
=
L
q
j
=

i
a
L
i
q
j
. (2.35)
The Hamiltonian of discretized system is then,
H =

i
a
_

j
q
j
L
i
q
j
L
i
_
. (2.36)
We now assume that the discretized Lagrangian at any point i, contains
the time derivative of only one degree of freedom q
i
, and no other (e.g.
q
i+1
, q
i1
, . . .). This was indeed the case for the example of the elastic rod, and
it is the case for all physical systems that we will study. It essentially means
that each point in the lattice has its own self-determined kinetic energy. On
16
the other hand, the potential term in L
i
can depend on the coordinates of
its neighbours. The Hamiltonian now becomes,
H =

i
a
_
q
i
L
i
q
i
L
i
_
. (2.37)
Finally, by taking the continuoum limit a 0, we obtain a volume integral
over a it Hamiltonian density,
H =
_
dxH (2.38)
with
H = q(x, t)(x, t) /, (2.39)
with the eld conjugate momentum dened as,
(x, t) =
/
q(x, t)
. (2.40)
We notice that in the derivation of Euler-Lagrange equations (Eq. 2.29),
derivatives with respect to space and time coordinates are treated on the
same footing. In fact, the Lagrangian formulation is manifestly covariant.
However, in computing the Hamiltonian (Eq. 2.36), dierentiation with time
maintains a special role and covariance is not manifest.
2.5 An example: acoustic waves
As an example, we use an elastic medium (e.g. air) for acoustic waves. The
Lagrangian density is a generalization in three space dimensions of the elastic
rod Lagrangian that we have already studied.
/ =

2
_
y(r, t)
t
_
2

v
2
sound
2
_

y(r, t)
_
2
, (2.41)
with v
sound
the speed of sound. Setting v
sound
= 1, we write
/ =

2
(

y) (

y)

2
_
(
x
0
y)
2
(
x
1
y)
2
(
x
2
y)
2
(
x
3
y)
2

. (2.42)
17
We rst nd the corresponding Euler-Lagrange equations, where we need the
derivatives
/
y
= 0, (2.43)
/
(

y)
=

y,

/
(

y)
=

y
2
y. (2.44)
The equations of motion are

/
(

y)
=
/
y
;
2
y = 0. (2.45)
A solution of the equation of motion is a plane-wave
y = e
ikx

e
ikx
, with k

= k
2
= 0. (2.46)
The conjugate momentum of the eld y is,

/
(
0
y)
=
0
y
;
0
y =

. (2.47)
The Hamiltonian density is
H
0
y /
; H =

2
2
+
1
2

y
_
2
. (2.48)
In the theory of elds, we make a reference to the Lagrangian L =
_
d
3
x/
and the Hamiltonian H =
_
d
3
xHrarely. In practice, we always work directly
with the corresponding densities / and H. Thus, it has prevailed that we
call H the Hamiltonian and / the Lagrangian assuming imlicitly that
we refer to their densities. We shall adopt the same terminology from now
on.
18
2.6 Noethers theorem
We can prove a very powerful theorem for Lagrangian systems wich exhibit
symmetries, which is known as the Noether theorem. The theorem states
that for every symmetry of the system, there exists a conserved current and
a corresponding conserved charge.
2.6.1 Field symmetry transformations
Consider a Lagrangian density
/
_

(i)
,

(i)

, (2.49)
which we assume to be invariant under innitesimal transformations of the
elds

(i)
(x)
(i)
(x) +
(i)
(x),

(i)
(x)

(i)
(x) +

(i)
(x). (2.50)
The change in the Lagrangian density can be computed as,
/ =

i
_
/

(i)

(i)
+
/
(

(i)
)

(i)
_
=

i
_

(i)
_
/

(i)

/
(

(i)
)
_
+

_
/
(

(i)
)

(i)
__
;/ =

_
/
(

(i)
)

(i)
_
. (2.51)
Invariance of the Lagrangian under the symmetry transformation of Eq. 2.50
requires that
/ =

_
/
(

(i)
)

(i)
_
= 0. (2.52)
Therefore the currents
J

=
/
(

(i)
)

(i)
, (2.53)
are conserved:

= 0. (2.54)
The conserved currents J

are as many as the number of independent gen-

erators for the transformations
(i)
.
19
It may happen that a eld symmetry transformation does not leave the
Lagrangian invariant, but it only changes it by a total derivative,
/ =

. (2.55)
In that case, we still obtain the same physical equations of motion since the
action integral remains invariant,
_
d
4
x/ =
_
d
4
x

= 0. (2.56)
From Eq. 2.55 and our general result of Eq. 2.51, we nd that
J

=
/
(

(i)
)

(i)

, (2.57)
are conserved currents

= 0.
To each conserved current, corresponds a conserved charge, i.e. a physical
quantity which maintains the same value at all times. Indeed,

= 0
;
J
0
t
+
J
i
x
i
= 0
;
J
0
t
+

J = 0
;
_
d
3
x
J
0
t
+
_
d
3
x

J = 0
;
Q
t
= 0 with Q =
_
d
3
xJ
0
(2.58)
2.6.2 Space-Time symmetry transformations
We now consider a more complicated version of the theorem, which deals
with space-time symmetries. We consider a classical system of i = 1 . . . N
physical elds
(i)
, and we assume that the Lagrangian density does not
depend explicitly on space-time coordinates. This is a reasonble assumption
if we require that physical laws which are valid in the present should also be
valid after we perform a time translation, i.e. the same physical will hold in
the future and they held in the past. Also, this assumption anticipates that
experiments in two dierent positions discover the same physics. However,
20
the Lagrangian depends on the space-time coordinates implicitly, through
the elds
(i)
(x

).
/ = /
_

(i)
,

(i)

. (2.59)
We now require that if we perform a space-time symmetry transformation,
such as a translation or a rotation or a boost
x

, (2.60)
the action remains always the same:
S =
_
V
d
4
x/
_

(i)
(x),

(i)
(x)

=
_
V

d
4
x

/
_

(i)
(x

),

(i)
(x

)
_
. (2.61)
In this way, since the action is invariant under the transformation, the vari-
ational principle S = 0 is automatically satised after the transformation
if it was satised before the transformation. In general, the elds
(i)
also
transform under general space-time tranformations. They may be scalar,
vectors, etc. We write,

(i)
(x

) =
(i)
(x) +
(i)
(x) (2.62)

(i)
(x

) =

(i)
(x) +
_

(i)
(x)
_
(2.63)
(2.64)
The action after the transformation becomes,
S =
_
V

d
4
x

/
_

(i)
(x

),

(i)
(x

)
_
=
_
V
d
4
xdet
_
x

_
/
_

(i)
(x) +
(i)
(x),

(i)
(x) +
_

(i)
(x)
_
.
(2.65)
The variation of the Lagrangian density due to the symmetry transfor-
mation can be computed as,
/ = /
_

(i)
(x) +
(i)
(x),

(i)
(x) +
_

(i)
(x)
_
/
_

(i)
(x),

(i)
(x)

;/ =

i
_
/

(i)

(i)
+
/
(

(i)
)

(i)
_
_
+O
_
_

(i)
_
2
_
(2.66)
where the term inside the square brackets vanishes by requiring that
(i)
are
physical elds satisfying the Euler-Lagrange equations.
21
We consider continuous transformations for which there exists a choice
of the transformation parameters resulting to a unit transformation, i.e. no
transformation. An example is a Lorentz boost with some velocity v, where
for v = 0 we have x

= x

. There are examples of symmetry trans-

formations where this does not occur. For example a parity transformation
does not have this property, and the Noether theorem is not applicable then.
For continuous symmetry transformations, we can consider that they are
innitesimally dierent than the unit transformation,
x

= x

+ x

(2.67)
with x

very small. The Jacobian of such a transformation is,

det
_
x

_
= det
_
[x

+ x

]
x

_
= det
_

_
. (2.68)
We now need the determinant of a four by four matrix (, = 0, 1, 2, 3). We
can verify with an explicit calculation that through order O(x):
det
_
x

_
= 1 +

(x

) +O((x)
2
). (2.69)
[exercise: Verify the above explicitly]
Substituting the expansions of Eq. 2.66 and Eq. 2.69 in the action integral
of Eq. 2.65, and keeping terms through O
_

(i)
_
, we nd
S = S +
_
d
4
x[/ +/

]
;0 =
_
d
4
x
_

i
_
/

(i)

(i)
+
/
(

(i)
)

(i)
_
_
+/

_
(2.70)
An untransformed eld evaluated at a transformed coordinate is expanded
as

(i)
(x

) =
(i)
(x + x)
;
(i)
(x

) =
(i)
(x) + x

(i)
+O(x
2
)
;
(i)
(x) = x

(i)
(2.71)
22
Similarly,

(i)
(x

) =

(i)
(x + x)
;

(i)
(x

) =

(i)
(x) + x

(i)
+O(x
2
)
;

(i)
(x) = x

(i)
(2.72)
The dierence of a transformed eld evaluated at a transformed coordinate
with respect to an untransformed eld evaluated at a non-tranformed coor-
dinate is:

(i)
=

(i)
(x

)
(i)
(x)
;
(i)
=

(i)
(x

)
(i)
(x

) +
(i)
(x

)
(i)
(x)
;
(i)
=

(i)
(x

)
(i)
(x

) +
(i)
(x)
;
(i)
=

(i)
(x) +
(i)
(x)
;
(i)
=

(i)
(x) + x

(i)
, (2.73)
where we separate explicitly the variation of the eld at a point

(i)
(x) =

(i)
(x

)
(i)
(x

), (2.74)
from the variation due to changing the space-time position. Similarly, for
the variation of the eld derivative, we have

(i)
=

(i)
(x

(i)
(x)
;

(i)
=

(i)
(x

(i)
(x

) +

(i)
(x

(i)
(x)
;

(i)
=

(i)
(x

)
(i)
(x

)
_
+

(i)
(x + x)

(i)
(x)
;

(i)
=

(i)
(x) + x

(i)
(x). (2.75)
Substituting Eq 2.73 and Eq 2.75 into Eq. 2.70, we obtain:
0 =
_
d
4
x

i
_
/

(i)

(i)
+
/
(

(i)
)

(i)
_
_
+
_
d
4
x
_
/

+ x

i
_
/

(i)

(i)
+
/
(

(i)
)

(i)
_
_
(2.76)
23
Applying the chain rule, we see that the term in the curly brackets is a total
divergence,

(/x

) = /

+ x

i
_
/

(i)

(i)
+
/
(

(i)
)

(i)
_
. (2.77)
The term in the square brackets of Eq. 2.76 can also be written as a to-
tal divergence, applying integration by parts and using the Euler-Lagrange
equations,
/

(i)

(i)
+
/
(

(i)
)

(i)
_
=
_
/

(i)

/
(

(i)
)
_

(i)
+

_
/
(

(i)
)

(i)
_
=

_
/
(

(i)
)

(i)
_
(2.78)
Eq. 2.76, with Eq. 2.78 and Eq. 2.78, becomes
0 =
_
d
4
x
_

_
/
(

(i)
)

(i)
_
+

(/x

)
_
; 0 =
_
d
4
x

i
/
(

(i)
)

(i)
+/x

_
. (2.79)
This is essentially the proof of Noethers theorem. For simplicity, we re-
strict ourselves to elds
(i)
which transform as scalars under a space-time
transformation, i.e. they do not transform:

(i)
(x

) =
(i)
(x) ;
(i)
= 0. (2.80)
For this to happen, the variation of a eld at a point

(i)
must compensate
the variation
(i)
due to changing the space-time position:

(i)
= 0 ;

(i)
+
(i)
= 0
;

(i)
=
(i)
;

(i)
= x

(i)
. (2.81)
Substituting Eq. 2.80 into Eq 2.79 we obtain
_
V
d
4
x

(T

) = 0. (2.82)
24
where we have dened:
T

i
/
(

(i)
)
_

(i)
_
/g

(2.83)
The tensor T

is known as the energy-momentum tensor for reasons that will

become obvious in a while. The name stress-energy tensor is also encountered
very often.
The volume of integration in equation Eq. 2.71 is considered to have
boundaries which are arbitrary. In addition, the space-time variation x

depends on small but otherwise arbitrary parameters, as many of them as

the generators of the symmetry transformation. Then, in order for Eq. 2.71
to hold, the integrand must vanish. We have therefore proven, that if the
system possesses a space-time symmetry transformation, there are as many
conserved currents as the generators of the symmetry transformation.

= 0, J

= x

. (2.84)
We emphasize that we can write as many currents J

as the number of
the independent x

space-time symmetry transformations which are obeyed

by the physical system.
Translation symmetry transformations
As a rst application of the Noether theorem, we nd the conserved currents
and the corresponding charges for physical systems which are symmetric
under space-time translations. An innitesimal translation transformation is
x

= x

;x

, (2.85)
where

is a small constant four-vector.

The corresponding conserved current is
J

= T

, (2.86)
satisfying the continuity equation

= 0
;

= 0. (2.87)
25
The vector

is small but otherwise arbitrary. Then, we must have that the

above equation is satised in general if

= 0, (2.88)
for every value of the index = 0, 1, 2, 3 seperately.
To the four conserved currents correspond four conserved charges (Eq. 2.58),
P

=
_
d
3
xT
0
. (2.89)
Specically, the time-component of P

is
P
0
=
_
d
3
xT
00
=
_
d
3
x
_

i
/
(
0

(i)
)
_

(i)
_
g
00
/
_
. (2.90)
We recall that the conjugate momentum for the eld
(i)
is

(i)
=
/
(
0

(i)
)
, (2.91)
and thus
P
0
=
_
d
3
x
_

(i)

(i)
/
_
. (2.92)
In the square bracket, we recognize the Hamiltonian density (Eq. 2.39),
P
0
=
_
d
3
xH. (2.93)
concluding that the charge P
0
is the energy of the system, and it is con-
served. Energy conservation is thus a pure consequence of time translation
symmetry. Similarly, we can identify the charges P
i
, i = 1, 2, 3 as the mo-
mentum components of the system in the three space directions, and they
are of course also conserved.
26
Lorentz symmetry transformations
Lorentz transformations,
x

, (2.94)
preserve the distance
ds
2
= g

dx

dx

= g

dx

dx

. (2.95)
This leads to
g

dx

dx

= g

dx

dx

= g

dx

dx

;g

= g

. (2.96)
Considering an innitesimal transformation,

, +O(
2
) (2.97)
we obtain
g

= g

;g

= g

+ . . .
_
(

+ . . .)
;g

= g

+O(
2
)
;

. (2.98)
The parameters

for generating Lorentz transformations are therefore

antisymmetric.
A transformed space-time coordinate is given by
x

= (g

+ . . .) x

= x

+O(
2
)
;x

. (2.99)
If Lorentz transformations are a symmetry of a physical system, then the
corresponding conserved currents are
J

= T

= T

=
1
2
T

]
=
1
2
[T

] =
1
2
[T

]
;J

=
1
2

with

. (2.100)
27
The continuity equations for the conserved currents J

are,

= 0 ;

= 0, (2.101)
and the correspoding conserved charges are:
M

=
_
d
3
xJ
0
=
_
d
3
x
_
T
0
x

T
0
x

. (2.102)
Poincare symmetry transformations
We now require that a physical system possesses both space-time translation
and Lorentz transformation symmetry,
x

. (2.103)
The symmetry group of both translations and Lorentz transformations is
called the Poincare group. The following two charge conservation equations
should hold:

= 0, due to translation symmetry, (2.104)

= 0, due to Lorentz symmetry. (2.105)

This is possible if the energy-momentum tensor T

is symmetric. Indeed,
0 =

[T

]
= x

+ T

= x

0 + T

0 T

= T

; T

= T

. (2.106)
It is not obvious that the energy-momentum tensor is symmetric. In fact,
the denition of the tensor
T

i
/
(

(i)
)
_

(i)
_
/g

(2.107)
does not exhibit an explicit symmetry in the indices and . However, the
energy momentum tensor is not uniquely dened. Let us consider a general
tensor of the form

,
28
which is antisymmetric in the indices , :
f

= f

.
We can show that such a tensor is a conserved current,

) =
1
2
[

] (antisymmetry)
=
1
2
[

] (relabeling)
=
1
2
[

] = 0. (commuting derivatives) (2.108)

However it produces a null conserved charge, as we can verify easily:
Q =
_
d
3
x

f
0
=
_
d
3
x
_

0
f
00
+
i
f
i0
_
=
_
d
3
x
_

0
0 +
i
f
i0
_
=
_
d
3
x
i
f
i0
(using Stokes theorem)
= 0. (2.109)
We can therefore add such a tensor to the energy-momentum tensor,
T

= T

(2.110)
without modifying the corresponding conserved charges (energy and mo-
menta). The function f

can be chosen so that

T

is manifestly symmetric
in the indices , .
29
Chapter 3
Quantization of the
Schrodinger eld
We start our study of eld quantization, by rst reproducing the known
results of Quantum Mechanics, for a system of many identical non-interacting
particles.
3.1 The Schrodinger equation from a Lagrangian
density
As a rst step, we introduce a Lagrangian that yields the Schrodinger equa-
tion with the Euler-Lagrange formalism. We can verify that the following
density,
/ = i

t

_

__

_
2m
(3.1)
does exactly this. and

turn out to be complex conjugate when they take

their physical solutions, but this is not an assumption that we have to do
now ourselves. It will be a natural consequence of solving the Euler-Lagrange
30
equations, which are:
0 =

/
(

)

/

=
0
/
(
0
)
+
i
/
(
i
)

/

= i

t
+

i

2m
+ 0
;0 =
_
i

t


2
2m
_

, (3.2)
and
0 =

/
(

)

/

=
0
/
(
0

)
+
i
/
(
i

)

/

= 0 +

i

2m
i

t
;0 =
_
i

t
+

2
2m
_
. (3.3)
Indeed, we obtain the Schrodinger equation for and its complex conjugate
for

. A general solution of the above equations is a superposition of plane-

wave solutions,
(x, t) =
_
d
3

k
(2)
3
a(

k)e
i(
k
t

kx)
, with
k
=

k
2
2m
. (3.4)
It is also useful to compute the conjugate momenta for and

as well as
the Hamiltonian density corresponding to the Schrodinger Lagrangian. We
nd
=
/
(
0
)
= i

, (3.5)
and

=
/
(
0

)
= 0. (3.6)
The Hamiltonian density is
H =
0
+

/
;H =
(
i

) (
i
)
2m
=
_

__

_
2m
. (3.7)
31
Notice that this Hamiltonian density diers from

2m
=
_

__

_
2m

_
2m
, (3.8)
by a total divergence which does not contribute to the Hamiltonian integral
H =
_
d
3
xH.
Substituting the general solution of Eq. 3.4 in the expression for the
Hamiltonian density, Eq. 3.7, we obtain
H =
_
d
3

k
1
(2)
3
d
3

k
2
(2)
3

k
1

k
2
2m
e
i[(
k
2

k
1
)t(

k
2

k
1)x]
a

(k
2
)a(k
1
) (3.9)
The Hamiltonian of the Schrodinger eld is,
H =
_
d
3
xH, (3.10)
and we can perform the volume integration easily using
_
d
3
xe
i

kx
= (2)
3

(3)
(

k).
We nd the following expression for the Hamiltonian,
H =
_
d
3

k
(2)
3

k
a

(k)a(k). (3.11)
3.2 Quantization of Fields
For a system with a nite number of degrees of freedom L[q
i
q
i
], the quan-
tization procedure is the following. For each generalized coordinate we nd
rst the corresponding conjugate momentum
p
j
=
L
q
j
. (3.12)
Then we promote all q
i
, p
i
to operators, which satisfy commutation relations,
[q
i
, p
j
] = i
ij
, (3.13)
and
[p
i
, p
j
] = [q
i
, q
j
] = 0. (3.14)
32
We can generalize this quantization procedure to systems of elds, with an
inntite number of degrees of freedom. Viewing elds as the continuum limit
of a discrete spectrum of degrees of freedom is particularly useful. In this
way, we can make the discrete to continuum correspondece:
i x,
q
i
(t) (x, t)
p
i
(t) (x, t). (3.15)
The corresponding quantization conditions for elds are:
[(x
1
, t), (x
2
, t)] = i
(3)
(x
1
x
2
), (3.16)
and
[(x
1
, t), (x
2
, t)] = [(x
1
t), (x
2
, t)] = 0. (3.17)
3.3 Quantized Schrodinger eld
We can apply the eld quantization conditions of the previous section (Eqs 3.16-
3.17) to the Schrodinger eld (x, t) and its conjugate momentum (x, t) =
i

(x, t). We obtain

[(x
1
, t), (x
2
, t)] = i
(3)
(x
1
x
2
)
;[(x
1
, t),

(x
2
, t)] =
(3)
(x
1
x
2
), (3.18)
and
[(x
1
, t), (x
2
, t)] = [(x
1
, t), (x
2
, t)] = 0
;[(x
1
, t), (x
2
, t)] = [

(x
1
, t),

(x
2
, t)] = 0. (3.19)
We now run into a problem if we impose a quantization condition for

and

, since we have found

= 0 in Eq. 3.6. The requirement,

(3)
(x
1
, x
2
) = [

(x
1
, t),

x
2
, t] = [

(x
1
, t), 0] = 0, (3.20)
is clearly inconsistent. However, we can write a dierent Lagrangian for
the Schrodinger eld than the one in Eq. 3.1, which is classically equivalent
yielding identical equations of motion,
/
new
=
i
2

t

i
2

t

_

__

_
2m
. (3.21)
33
Exercise: Verify that this Lagrangian yields the Scrodinger equa-
tion as Euler-Lagrange equations
This Lagrangian diers from the one of Eq. 3.1 by a time derivative
i
2

t
(

).
With /
new
both and

are non-vanishing, and all commutation relations

that we can write lead to one of the Eqs 3.18-3.19. This is a good example
showing that not all Lagrangian densities which are equivalent at the clas-
sical level, i.e. yield the same equations of motion, can be subjected to a
self-consistent quantization. In our case, /
new
can be quantized while / not.
The quantization conditions using the conjugate momenta and

as de-
rived from the Lagrangian density /
new
give the following three commutation
relations:
[(x
1
, t), (x
2
, t)] = [

(x
1
, t),

(x
2
, t)] = 0,
[(x
1
, t),

(x
2
, t)] =
(3)
(x
2
x
1
) . (3.22)
From these quantization conditions for the elds, it is easy to derive the cor-
responding quantization conditions for the operators a(

k) and a

k). First
we observe, from Eq. 3.4 that the operators a(

k) and a

k) are Fourier trans-

forms of and

respectively,
a(

k) =
_
d
3
x(x, t)e
i[
k
t

kx]
. (3.23)
Exercise: Verify explicitly the above by substituting in the expres-
sion of Eq. 3.4 for the eld .
Then, we have
_
a(

k
1
), a

k
2
)
_
=
_
d
3
x
1
d
3
x
2
e
i[
k
1
t

k
1
x
1]
e
i[
k
2
t

k
2
x
2]
[(x
1
, t),

(x
2
, t)]
=
_
d
3
x
1
d
3
x
2
e
i[
k
1
t

k
1
x
1]
e
i[
k
2
t

k
2
x
2]

(3)
(x
2
x
1
)
=
_
d
3
xe
i[(
k
1

k
2
)t(

k
1

k
2)x]
;
_
a(

k
1
), a

k
2
)
_
= (2)
3

(3)
_

k
1

k
2
_
. (3.24)
Similarly (exercise), we can prove that
_
a(

k
1
), a(

k
2
)
_
=
_
a

k
1
), a

k
2
)
_
= 0. (3.25)
34
To emphasize that a

and

are now operators, we write

a

.
3.4 Particle states from quantized elds
Let us now look at the so called number density operator
N(k) a

(k)a(k). (3.26)
We can prove that
[N(k), a(p)] =
_
a

(k)a(k), a(p)

= a

(k) [a(k), a(p)] +

_
a

(k), a(p)

a(k)
; [N(k), a(p)] = (2)
3

(3)
(

k p)a(p). (3.27)
Also, in the same manner we nd that
_
N(k), a

(p)

= (2)
3

(3)
(

k p)a

(p). (3.28)
It is also easy to prove ((exercise)) that two number density operators com-
mute with each other,
[N(k), N(p)] =
_
N(k), a

(p)a(p)

=
_
N(k), a

(p)

a(p) + a

(p) [N(k), a(p)]

= (2)
3

(3)
(

k p)a

(p)a(p) (2)
3

(3)
(

k p)a

(p)a(p)
; [N(k), N(p)] = 0. (3.29)
We recall that the Hamiltonian is (Eq. 3.11),
H =
_
d
3

k
(2)
3

k
a

(k)a(k) ;H =
_
d
3

k
(2)
3

k
N(k). (3.30)
Obviously, it commutes with the number density operator,
[N(p), H] =
_
N(p),
_
d
3

k
(2)
3

k
N(k)
_
=
_
d
3

k
(2)
3

k
[N(p), N(k)] = 0.
(3.31)
35
As a consequence, the eigenstates of the Hamiltonian can be found as eigen-
states of the number density operator, or the particle number operator,
dened as

N
_
d
3

k
(2)
3
N(k) =
_
d
3

k
(2)
3
a

(k)a(k), (3.32)
and which also obviously commutes with the Hamiltonian.
From Eqs 3.27-3.28, we derive that
_

N, a

(p)
_
= +a

(p), (3.33)
and
_

N, a(p)
_
= a(p). (3.34)
If a state [n) is eigenstate of the number operator

N with eigenvalue n,

N [n) = n[n) , (3.35)

then the states a

(p) [n) and a(p) [n) are also eigenstates with eigenvalues
n + 1 and n 1 correspondingly. Indeed,

N
_
a

(p) [n)
_
=
__

N, a

(p)
_
+ a

(p)

N
_
[n)
;

N
_
a

(p) [n)
_
=
_
+a

(p) + a

(p)n
_
[n)
;

N
_
a

(p) [n)
_
= (n + 1)
_
a

(p) [n)
_
, (3.36)
and

N (a(p) [n)) =
__

N, a(p)
_
+ a(p)

N
_
[n)
;

N (a(p) [n)) = a(p) + a(p)n [n)
;

N (a(p) [n)) = (n 1) (a(p) [n)) . (3.37)
We can prove that the eigenvalues of the number operator must always
be positive, if the correspondinge eigestates are states of a Hilbert space with
a positive norm. Consider the state [) = a(p) [n),
0 [[ [) [[
2
= [ ) = n[ a

(p)a(p) [n) = n[ N(p) [n)

; 0 n[
__
d
3
p
(2)
3
N(p)
_
[n) = n[

N [n) = nn[ n) = n[[ [n) [[
2
; n 0. (3.38)
36
This condition is in an apparent contradiction with Eq. 3.34, since a repeated
application of the operator a(p) on a positive eigenvalue will eventually yield
an eigenstate with a negative eigenvalue. This is true , if all eigenstates are
not annihilated by the a(p) operator. Assume, however, that there is a state
[0) which is annihilated by a(p),
a(p) [0) = 0. (3.39)
This state is an eigenstate of the number operator with a zero eigenvalue:
a(p) [0) = 0
;a

(p)a(p) [0) = 0 [0) ;

__
d
3
p
(2)
3
a

(p)a(p)
_
[0) = 0 [0)
;

N [0) = 0 [0)
(3.40)
Then, Eq. 3.39 only permits to act on the vacuum state with a

operators,
producing eigenstates of the number operator with positive integer eigenval-
ues. For example, for a general state
[1
particle
)
_
d
3
p
1
(2)
3
f(p
1
)a

(p
1
) [0) , (3.41)
we have (exercise):

N [1
particle
) = (+1) [1
particle
) . (3.42)
Similarly, for a general state:
[2
particles
)
_
d
3
p
1
(2)
3
d
3
p
2
(2)
3
f(p
1
, p
2
)a

(p
1
)a

(p
2
) [0) , (3.43)
we have (exercise):

N [2
particles
) = (+2) [2
particles
) , (3.44)
and so on:

N [m
particles
) = (+m) [m
particles
) , (3.45)
37
with
[m
particles
)
_
m

i=1
_
d
3
p
i
(2)
3
a

(p
i
)
_
[0) f(p
1
, . . . , p
m
). (3.46)
Suggestively, we have labeled the eigenstates of the number operator

N,
as states of a discrete number of particles. This becomes better justied
when we nd common eigenstates with the Hamiltonian H. The states of
Eq. 3.46 with f(p
1
, . . . , p
m
) =

i
(2)
3

(3)
(p
i
k
i
), i = 1 . . . m:
[m
particles
, E)
_
m

i=1
a

(k
i
)
_
[0) (3.47)
diagonalize the Hamiltonian,
H[m
particles
, E) = E [m
particles
, E) , with E =
m

i=1

k
i
(3.48)
Exercise: Show the above for m=1 and m=2. Get convinced that
it is valid for arbitrary m.
The Hamiltonian H is, according to Noethers theorem, a conserved quan-
tity due to the invariance of the action under time-translations. Symmetry
under translations in the three space-time dimensions predicts the existence
of a conserved eld momentum

P, with
P
i
=
_
d
3
xT
0i
. (3.49)
Explicitly for the Schr udinger eld we nd the eld,

P =
_
d
3

k
(2)
3
N(k)

k (3.50)
Exercise: Prove the above
It is easy to see that a state [m
particles
, E) of Eq. 3.47 is also an eigenstate of
the eld momentum operator:

P [m
particles
, E) = p [m
particles
, E) , with p =
m

i=1

k
i
(3.51)
38
Exercise: Show the above for m=1 and m=2. Get convinced that
it is valid for arbitrary m.
To summarize, we have proven that the states of Eq. 3.47 are simultaneous
eigenstates of the number-operator, the eld Hamiltonian operator, and
the eld momentum operator. Their eigenvalues are the number of particles,
the sum of their energies, and the number of their momenta respectively. We
therefore see that the quantization of the Schrodinger eld is a very elegant
formalism to describe multi-particle states.
It turns out that the number-operator is also a conserved quantity.
The Lagrangian of the Schrodinger eld has an additional symmetry which
is responsible for this. The Lagrangian is invariant under the symmetry
transformation:
(x, t) e
ia
(x, t)

(x, t) e
ia

(x, t). (3.52)

You can prove that the corresponding charge to this symmetry transforma-
tion is

N =
_
d
3
xJ
0
, with J
0
=

(x, t) (x, t). (3.53)

Thus, we have found that multi-particle systems where we have measured
their number of particles at one moment, will always maintain the same
number of particles in the future.
3.5 What is the wave-function in the eld
quantization formalism?
In quantum mechanics, the wave function y(x, t)
1
corrsponding to a single
particle state is the probability amplitude to nd a particle at a position x,
y(x, t) = 1
particle
[ x) . (3.54)
For a system of N particles the wave-function is the probability amplitude
to nd the N particles at positions x
1
, x
2
, . . . x
N
.
y(x
1
, . . . x
N
, t) = x
1
, . . . , x
N
[ N
particles
) . (3.55)
1
In this section, we use the letter y to denote the wave-function, and the letter for
the Schrodinger eld operator.
39
The wave-function satises the Schrodinger equation, which for free particles
is:
_
i

t
+
N

i=1

2
x
i
2m
_
y(x
1
, . . . x
N
, t) = 0. (3.56)
Let us now look at a generic N-particle state of the previous section,
Eq. 3.46,
[N
particles
) =
_
_
N

i=1
d
3
p
i
(2)
3
a

(p
i
)
_
[0) f(p
1
, . . . , p
N
), (3.57)
where we can express the creation operators a

(p
i
) as Fourier transforms of
the eld operators

(x, t). Substituting Eq. 3.23 into Eq. 3.57, we obtain

[N
particles
) =
_
_
N

i=1
d
3
x
i
_
[S(x
1
, . . . , x
N
))

f(x
1
, . . . , x
N
, t), (3.58)
with,
[S(x
1
, . . . , x
N
)) =
_
N

i=1

(x
i
)
_
[0) , (3.59)
and

f(x
1
, . . . , x
N
, t) =
_ _
d
3
p
i
(2)
3
e
i(p
i
t p
i
x
i)
_
f(p
1
, . . . , p
N
). (3.60)
By an explicit calculation (exercise), we can verify that
_
i

t
+
N

i=1

2
x
i
2m
_

f(x
1
, . . . x
N
, t) = 0, (3.61)
i.e. it satises the Schrodinger equation. Therefore the function

f, is identi-
ed as the wave-function y(x
1
, . . . x
N
, t),
y(x
1
, . . . x
N
, t) = y(x
1
, . . . x
N
, t) = x
1
, . . . , x
N
[ N
particles
) . (3.62)
40
Using this observation, we write Eq.3.58 as
[N
particles
) =
_
_
N

i=1
d
3
x
i
_
[S(x
1
, . . . , x
N
)) x
1
, . . . , x
N
[ N
particles
) ,
;[N
particles
) =
_
_
_
N

i=1
d
3
x
i
_
[S(x
1
, . . . , x
N
)) x
1
, . . . , x
N
[
_
[N
particles
)
;1 = [N
particles
)
_
_
_
N

i=1
d
3
x
i
_
[S(x
1
, . . . , x
N
)) x
1
, . . . , x
N
[
_
[N
particles
)
(3.63)
Inside the square bracket we have a unit operator, which leads us to the
conclusion that
[S(x
1
, . . . , x
N
)) = [x
1
, . . . , x
N
) . (3.64)
We summarize the main result of this section, which connect the eld opera-
tor formalism with the traditional wave-function description of non-relativistic
Quantum Mechanics.
A quantum mechanical state of N particles at positions x
1
, . . . x
N
is
given by the action of eld operators on the vacuum state:
[x
1
, . . . x
N
) =
_
N

i=1

(x
i
)
_
[0) . (3.65)
A general Nparticle state is a superposition,
[N
particles
) =
_
d
3
x
1
. . . d
3
x
1

f(x
1
, . . . , x
N
) [x
1
, . . . x
N
) . (3.66)
The kernel of the superposition integral is the wave-function, satisfying
the Schrodinger equation
_
i

t
+
N

i=1

2
x
i
2m
_

f(x
1
, . . . x
N
, t) = 0. (3.67)
41
Chapter 4
The Klein-Gordon Field
In the study of the Schrodinger eld, we demonstrated how Quantum Field
theory can be made to construct quantum states of particles, with certain
momentum and an energy momentum relation:
E =
p
2
2m
. (4.1)
This is the correct energy-momentum relation in the non-relativistic limit.
Would it be possible to modify this procedure in such a way so that we obtain
the relativistic energy-momentum relation
E
2
= p
2
+ m
2
? (4.2)
Such an energy-momentum relation is not a novelty of special relativity.
In our classical mechanics example of acoustic waves, the Euler-Lagrange
equations gave wave solutions of the form:
(x, t) e
i(Et

kx)
, (4.3)
with
E
2
=

k
2
v
2
sound
. (4.4)
4.1 Real Klein-Gordon eld
If we were to quantize acoustic waves, we would obtain particles which would
always travel with the speed of sound. It is to be demonstrated that such
42
a quantization can be made consistently, but let us assume that this is pos-
sible. Then particles from the quantization of the acoustic eld (usually
called phonons) have the same energy-momentum relation as massless rela-
tivistic particles, if we were to identify the speed of sound as the speed of
light (v
sound
= c). It is easy to modify the Lagrangian for acoustic waves
so that the solutions for the eld (Eq. 4.3), encode the correct relativistic
energy-momentum relation of Eq. 4.2 for massive particles. The correspond-
ing Lagrangian is
/ =
1
2
(

) (

)
1
2
m
2

2
. (4.5)
We will then attempt for the rst time to write a quantum relativistic theory
of free particles, by quantizing the eld whose dynamics is described by
this Lagrangian. In the case of acoustic waves, the eld corresponds to the
amplitude of an oscillation, and it is therefore real.
4.1.1 Real solution of the Klein-Gordon equation
The Euler-Lagrangian equation for the eld of Eq. 4.5 is,

L
(

)

/

= 0
;
_

2
+ m
2

(x) = 0. (4.6)
This is known as the Klein-Gordon equation.
We shall nd a general solution of the Klein-Gordon equation for a real
eld . We start with the ansatz:
(x) =
_
d
4
k
(2)
4
_
f(k)e
ikx
+ f

(k)e
+ikx

, (4.7)
which is manifestly real. In addition, it is invariant under Lorentz transfor-
mations, as it is anticipated for a scalar eld. Substituting into the Klein-
Gordon equation we nd,
_
d
4
k
(2)
4
_
m
2
k
2
_ _
f(k)e
ikx
+ f

(k)e
+ikx

= 0. (4.8)
The above is satised if we have,
f(k) = (2)(k
2
m
2
)c(k), f

(k) = (2)(k
2
m
2
)c

(k). (4.9)
43
A general real solution of the Klein-Gordon equation is:
(x

) =
_
d
4
k
(2)
3

_
m
2
k
2
_ _
c(k)e
ikx
+ c

(k)e
+ikx

. (4.10)
The delta function requires that
k
2
= k

= m
2
;k
0
=
k
, with
k
=
_

k
2
+ m
2
. (4.11)
The delta function can be written as,
(k
2
m
2
) = (k
2
0

2
k
) =
(k
0

k
) + (k
0
+
k
)
2
k
. (4.12)
We can now perform the k
0
integration in Eq. 4.10, obtaining
(x

) =
_
d
3

k
(2)
3
2
k
_
c(
k
,

k)e
i(
k
t

kx)
+ c(
k
,

k)e
+i(
k
t+

kx)
+c

(
k
,

k)e
+i(
k
t

kx)
+ c(
k
,

k)e
i(
k
t+

kx)
_
(4.13)
We perform a change of integration

k

k to the integrals corresponding

to the second and fourth term of the sum inside the square bracket. Then,
the four terms can be grouped together into a sum of only two exponentials,
and we can write the general solution as:
(x

) =
_
d
3

k
(2)
3
2
k
_
a(

k)e
i(
k
t

k x)
+ a

k)e
i(
k
t

kx)
_
, (4.14)
where a(

k) = c(
k
,

k) + c

(
k
,

k).
We remark that we often write the integration measure in an explicitly
Lorentz invariant form,
_
d
3

k
(2)
3
2
k
=
_
d
4
k
(2)
3
(k
2
m
2
)(k
0
> 0)
_
d
4
k
(2)
3

(+)
(k
2
m
2
), (4.15)
with

(+)
(k
2
m
2
) (k
2
m
2
)(k
0
> 0). (4.16)
The measure is invariant under orthochronous transformations which do not
change the sign of the time-like component of the integration variable.
44
4.1.2 Quantization of the real Klein-Gordon eld
The conjugate momentum of the Klein-Gordon eld, (x, t), is
=
/
(
0
)
=
0
. (4.17)
The Hamiltonian of the Klein-Gordon eld is,
H =
_
d
3
x
(
0
)
2
+
_

_
2
+ m
2

2
2
. (4.18)
Substituting into the Hamiltonian integral the solution of Eq. 4.14, and per-
forming the integration over the space volume, we nd:
H =
_
d
3

k
(2)
3
2
k

k
2
_
a

(k)a(k) + a(k)a

(k)

, (4.19)
where we have promoted the eld into an operator. We now impose commu-
tation relations for the eld and its conjugate momentum,
[(x
1
, t), (x
2
, t)] = [(x
1
, t), (x
2
, t)] = 0, (4.20)
and
[(x
1
, t), (x
2
, t)] = i
(3)
(x
2
x
1
) . (4.21)
With a straightforward calculation, we can verify (exercise) that the oper-
ators a and a

are written in terms of the and operators as,

a(k) = i
_
d
3
xe
i(
k
t

kx)
_
(

x, t) i
k
(x, t)

, (4.22)
and
a

(k) = i
_
d
3
xe
i(
k
t

kx)
_
(

x, t) + i
k
(x, t)

. (4.23)
Then we can compute the commutators that can be formed with a and a

.
We nd (exercise),
[a(k
1
), a(k
2
)] =
_
a

(k
1
), a

(k
2
)

= 0, (4.24)
and
_
a(k
1
), a

(k
2
)

= (2)
3
(2
k
1
)
(3)
_

k
1

k
2
_
. (4.25)
45
4.1.3 Particle states for the real Klein-Gordon eld
We can now construct the particle number operator,

N =
_
d
3

k
(2)
3
2
k
a

(k)a(k), (4.26)
for which we can prove that
_

N, a(p)
_
= a(p), (4.27)
and
_

N, a

(p)
_
= +a

(p). (4.28)
Therefore, the operators a(p) and a

are ladder operators. In particular, if

a state [S) is an eigenstate of the number operator

N with eigenvalue n,
then the states a(p) [S) and a

(p) [S) are also eigenstates with eigenvalues

n 1 and n + 1 respectively. The construction of particle states proceeds
identically, as in our study of the quantization for the Schrodinger eld, after
we observe that the eld Hamiltonian and the number operator commute
(exercise):
_

N, H
_
= 0. (4.29)
The above equations lead to the existence of a vacuum state [0), which is
annihilated by the operator a(k),
a(k) [0) = 0. (4.30)
The vacuum state is an eigenstate of the number operator with a zero eigen-
value,

N [0) = 0 [0) . (4.31)

All other states can be produced from the vacuum state, by acting on it with
creation operators a

(p
i
). A generic state
[
m
)
_
m

i=1
a

(p
i
)
_
[0) , (4.32)
contains m particles,

N [
m
) = m[
m
) . (4.33)
46
4.1.4 Energy of particles and normal ordering
We consider the simplest case for [)
m
state,
[p) = a

(p) [0) , (4.34)

which is an one-particle state, satisfying

N [p) = 1 [p) . (4.35)

We can compute the energy level of the state by acting on it with the eld
Hamiltonian operator,
H[p) =
_
d
3

k
(2)
3
2
k

k
2
_
a

(k)a(k) + a(k)a

(k)
_
[p)
=
_
d
3

k
(2)
3
2
k

k
2
_
2a

(k)a(k) +
_
a(k), a

(k)
_
[p)
=
_
d
3

k
(2)
3
2
k

k
2
_
2a

(k)a(k) + (2)
3
2
k

(3)
(0)
_
[p)
;H[p) =
_

p
+
(3)
(0)
__
d
3

k
2
__
[p) . (4.36)
To our surprise, we nd an innite eigenvalue! This is at rst sight very
embarrassing.
We can make sense of this innity, if we look at the energy of the vacuum
state, which contains no particles.
H[0) =
_
d
3

k
(2)
3
2
k

k
2
_
a

(k)a(k) + a(k)a

(k)
_
[0)
=
_
d
3

k
(2)
3
2
k

k
2
_
2a

(k)a(k) +
_
a(k), a

(k)
_
[0)
=
_
d
3

k
(2)
3
2
k

k
2
_
2a

(k)a(k) + (2)
3
2
k

(3)
(0)
_
[0)
; H[0) =
(3)
(0)
__
d
3

k
2
_
[0)
;
(3)
(0)
__
d
3

k
2
_
= 0[ H[0) = E
vacuum
. (4.37)
47
We nd that the eld Hamiltonian gives an energy eigenvalue for the vacuum
state in Eq. 4.37 which is equal to the same innite constant that appeared
in the energy eigenvalue (Eq. 4.36) of the one-particle state. We can then
re-write Eq. 4.36 in an apparently innocent manner,
H[p) =
_
_
p
2
+ m
2
+ E
vacuum
_
[p) . (4.38)
Using the property (exercise)
_
H, a

(p
i
)

= +(p
i
)a

(p
i
), (4.39)
we can prove that states with more particles, have an energy,
H[
m
) =
_
E
vacuum
+

i
_
( p
2
i
+ m
2
)
_
[
m
) . (4.40)
All Hamiltonian eigenvalues contain and identical innite constant which
is equal to the vacuum energy. However, measurements of absolute energy
levels are not possible. The vacuum energy is harmless if we want to measure
the energy dierence of the states [
m
) from the vacuum. We can then
recalibrate our energy levels (by an innite constant) removing from the
Hamiltonian operator the energy of the vacuum:
H : H : H E
vacuum
= H 0[ H [0) . (4.41)
Explicitly, the Hamiltonian operator after we subtract the vacuum constant
can be written as
:
_
d
3

k
(2)
3
2
k

k
2
_
a

(k)a(k) + a(k)a

(k)

: =
_
d
3

k
(2)
3
2
k

k
_
a

(k)a(k)

.
(4.42)
A practical trick to remove from a conserved operator, such as the Hamil-
tonian, its vacuum expectation value:
: O : = O 0[ O[0) (4.43)
is to perform what is known as normal ordering. As you can see in Eq. 4.42,
this is simply achieved by putting to creation operators to the left of anni-
hilation operators when these refer to the same momentum

k. With this
practical rule,
:
_
a

(k)a(k) + a(k)a

(k)

: = a

(k)a(k) + a

(k)a(k) = 2a

(k)a(k). (4.44)
48
This is equivalent to using the commutation relation [a(k), a

(k)] = (2)
3
2
k

(3)
(0)
in order to put the creation operators to the right and dropping innities

(3)
(0). These leaves annihilation operators to the left, and guarantees that
the vacuum is automatically annihilated by the modied operator.
4.1.5 Field momentum conservation
Noethers theorem predicts that eld momentum is a conserved operator,
due to the symmetry of the Klein-Gordon action under space translations.
The momentum operator is given by,

P =
_
d
3

k
(2)
3
2
k

k
2
_
a

(k)a(k) + a(k)a

(k)

=
_
d
3

k
(2)
3
2
k

k
_
a

(k)a(k)

+

(3)
(0)
2
_
d
3

k
;

P =
_
d
3

k
(2)
3
2
k

k
_
a

(k)a(k)

. (4.45)
Here, the innity
(3)
(0) cancels because the momentum integral multiplying
it is vanishing due to the antisymmetry of the integrand under the transfor-
mation

k

k.

P commutes with both the Hamiltonian, with which it has
common eigenstates.
The momentum of the vacuum state is zero,

P [0) = 0. (4.46)
The

P eigenvalues of multi-particle states can be easily found by using,
_

P, a

(p)
_
= + pa

(p). (4.47)
For example, its eigenvalue for the one-particle state [p) = a

(p) [0) is

P [p) =
_

P, a

(p)
_
[0) = pa

(p) [0) = p [p) , (4.48)

etc.
49
4.1.6 Labels of particle states?
The main outcome of quantizing the real Klein-Gordon eld, is a spectrum of
states which can be identied as states of many particles with denite energy
and momentum. A general state,
[
n
)
_
n

i=1
a

(p
i
)
_
[0) , (4.49)
describes m identical particles,

N [
n
) = n[
n
) , (4.50)
each of them carrying momentum p
i
,

P [
n
) =
_
n

i=1
p
i
_
[
n
) , (4.51)
and relativistic energy +
_
p
2
i
+ m
2
,
H[
n
) =
_
n

i=1
_
p
2
i
+ m
2
_
[
n
) . (4.52)
We have therefore arrived to a consistent combination of quantum mechanics
and special relativity, where eld quanta give rise to particles with the correct
properties as anticipated from relativity.
4.2 Two real Klein-Gordon elds
Our task is to describe all known particles and their interactions. It is then
interesting to study the quantization of a system with more than one elds.
We can start simply, describing a system of two Klein-Gordon elds which
they dier only in their mass parameter,
/ =
1
2
(

1
) (

1
)
m
2
1
2

2
1
+
1
2
(

2
) (

2
)
m
2
2
2

2
2
. (4.53)
50
This Lagrangian yields two Klein-Gordon equations as equations of motions
for the elds
1
and
2
.,
_

2
+ m
2
1
_

1
= 0,
_

2
+ m
2
2
_

2
= 0, (4.54)
with solutions

1
(x

) =
_
d
3

k
(2)
3
2
1
_
a
1
(

k)e
ikx
+ a

1
(

k)e
ikx
_
, (4.55)

2
(x

) =
_
d
3

k
(2)
3
2
2
_
a
2
(

k)e
ikx
+ a

2
(

k)e
ikx
_
, (4.56)
and

1
=
_

k
2
+ m
2
1
,
2
=
_

k
2
+ m
2
2
. (4.57)
The Hamiltonian H, eld momentum

P and number operator

N can be
written as

N =

N
1
+

N
2

N
i
=
_
d
3

k
(2)
3
2
i
a

i
(

k)a
i
(

k), (4.58)

P =

P
1
+

P
2

P
i
=
_
d
3

k
(2)
3
2
i

k
2
_
a

i
, a
i
_
, (4.59)
H = H
1
+ H
2
H
i
=
_
d
3

k
(2)
3
2
i

i
2
_
a

i
, a
i
_
. (4.60)
We can construct particle states in the same fashion as with the La-
grangian of just a single Klein-Gordon eld. Products of a

1
operators acting
on the vacuum state create relativistic particles with mass m
1
, while a

2
op-
erators create particles with mass m
2
. For example, the states
[S
1
) a

1
(p) [0) , [S
1
) a

1
(p) [0) (4.61)
have

N [S
i
) = +1 [S
i
) , (4.62)

P [S
i
) = + p [S
i
) , (4.63)
H[S
1
) =
_
p
2
+ m
2
1
[S
1
) and H[S
2
) =
_
p
2
+ m
2
2
[S
2
) . (4.64)
51
They are degenerate in that, they are single particle states with the same
momentum p, however they can be distinguished by measuring the energy of
the particles as long as the masses m
1
and m
2
.
The possibility of m
1
= m
2
is interesting; states created with the a

1
or
a

2
operators are degenerate in both the energy and momentum.
4.2.1 Two equal-mass real Klein-Gordon elds
The special case of m
1
= m
2
= m worthies a detailed study, because a new
rotation symmetry emerges in the space of elds
1
and
2
. This leads,
according to Noethers theorem, to a conserved quantity. We rewrite the
Lagrangian in a form where the symmetry is manifest:
/ =
1
2
_

__

T
_
, (4.65)
where we have dened the eld vector

_

1

2
_
with its transpose

T
=
_

1

2
_
. (4.66)
Obviously, the Lagrangian is invariant (exercise) if we perform an orthogo-
nal transformation,

= R

, with R
T
= R
1
. (4.67)
To compute the conserved current we need to nd the change
i
of the
elds under an innitesimal symmetry transformation. We then expand,
R
ij
=
ij
+
ij
+O(
2
). (4.68)
Orthogonality of R implies that the matrix
ij
is antisymmetric,
R
T
ij
= R
1
ij
;
ji
+
ji
=
ij

ij
;
ij
=
ji
. (4.69)
The variation of the elds is:

1
= R
1j

j
= (
1j
+
1j
)
j
=
1
+
11

1
+
12

2
=
1
+
12

2
;
1
=
12

2
, (4.70)
52
and, similarly,

2
=
21

1
=
12

1
. (4.71)
The Noether theorem current is,
J

=
/
(

1
)

1
+
/
(

2
)

2
;J

=
12
[(

1
)
2
(

2
)
1
] (4.72)
and the charge
Q =
_
d
3
x[(

1
)
2
(

2
)
1
] , (4.73)
is conserved. Substituting in the above expressions the physical solutions for
the elds
i
, and performing the x integration we obtain (exercise):
Q = i
_
d
3

k
(2)
3
2
k
_
a
1
(k)a

2
(k) a
2
(k)a

1
(k)
_
. (4.74)
A Noether charge is dened uniquely up to a constant multiplicative fac-
tor. Obviously, if Q is a time-independent (conserved) quantity, then also
Q, where is a number constant, is also conserved. The normalization in
Eq. 4.74 is convenient yielding a Hermitian Q (exercise):
Q

= Q. (4.75)
We now proceed to nd the eigenstates of the Q charge operator. This
becomes easy, if we construct ladder operators for Q. We compute rst the
commutators of the charge with a
1
and a
2
(exercise). We nd
[Q, a
1
(p)] = ia
2
(p), (4.76)
and
[Q, a
2
(p)] = +ia
1
(p). (4.77)
Dene the linear combinations,
a(p)
a
1
(p) + ia
2
(p)

2
, (4.78)
and
b(p)
a
1
(p) ia
2
(p)

2
. (4.79)
53
These are indeed ladder operators (exercise), satisfying the following com-
mutation relations with the charge Q:
[Q, a(p)] = a(p), [Q, b(p)] = +b(p). (4.80)
By taking the hermitian conjugate of the above equations, and using Q

= Q,
we nd that
_
Q, a

(p)

= +a

(p),
_
Q, b

(p)

= b

(p). (4.81)
From a state [S) with charge q,
Q[S) = q [S) , (4.82)
we can obtain states with charges q 1 by applying the ladder operators
a

(p) and b

(p) (exercise),
Q
_
a

(p) [S)
_
= (q + 1)
_
a

(p) [S)
_
, (4.83)
Q
_
b

(p) [S)
_
= (q 1)
_
b

(p) [S)
_
. (4.84)
We remark that the operators a

, b

are also ladder operators for the Hamil-

tonian H and the eld momentum operator

P. This is a consequence of the
fact that a

and b

are linear combinations of a

1
and a

2
which yield degenerate
eigenstates for H and P.
To nd all the common eigenstates of the charge Q , it suces to start
from one common eigenstate and use the operators a

and b

to nd all
the others. The vacuum state is also an eigenstate of the charge operator
(exercise):
Q[0) = 0. (4.85)
Repeated application of the a

operator on the vacuum, builds states of

positive charge:

(+)
n
_

_
n

i=1
a

(p
i
)
_
, (4.86)
54
with

(+)
n
_
= n

(+)
n
_
(4.87)

(+)
n
_
=
_
n

i=1
p
i
_

(+)
n
_
(4.88)
H

(+)
n
_
=
_
n

i=1
_
p
i
2
+ m
2
_

(+)
n
_
(4.89)
Q

(+)
n
_
= +n

(+)
n
_
(4.90)
Repeated application of the b

operator on the vacuum, builds states of

positive charge:

()
n
_

_
n

i=1
b

(p
i
)
_
, (4.91)
with

()
n
_
= n

()
n
_
(4.92)

()
n
_
=
_
n

i=1
p
i
_

()
n
_
(4.93)
H

()
n
_
=
_
n

i=1
_
p
i
2
+ m
2
_

()
n
_
(4.94)
Q

()
n
_
= n

()
n
_
. (4.95)
To summarize the main results of this section, mass degeneracy resulted
to a new O(2) symmetry of the Lagrangian. This gives rise to a new con-
served quantity, the charge Q. A particle state, is now characterized by its
momentum, the mass of the particle (or equivalently the energy), and its
charge which can be either positive or negative. We have just demonstrated
a method to describe physical systems of particles which are accompanied by
their anti-particles.
4.2.2 Two real Klein-Gordon elds = One complex
Klein-Gordon eld
We can recast the original expressions for the operators whose eigenvalues
characterize particle (

N,

P, H, Q) in terms of the charge ladder operators a, b,
55
rather than a
1
, a
2
. Substituting,
a
1
=
a + b

2
, a
2
=
a b

2
, (4.96)
we nd

N =
_
d
3

k
(2)
3
2
k
_
a

k)a(

k) + b

k)b(

k)
_
, (4.97)

P =
_
d
3

k
(2)
3
2
k

k
2
__
a

, a
_
+
_
b

, b
_
(4.98)
H =
_
d
3

k
(2)
3
2
k

k
2
__
a

, a
_
+
_
b

, b
_
. (4.99)
We can now re-write the eld operators in terms of a and b. We nd that

1
=
+

2
,
2
=
+

2i
, (4.100)
with
(x) =
_
d
3

k
(2)
3
2
k
_
a(k)e
ikc
+ b

(k)e
ikc

, k
0
=
k
=
_

k
2
+ m
2
.
(4.101)
We can cast the Lagrangian representing the elds in terms of and

,
rather than
1
and
2
. We nd,
/ =
1
2
_

_
T
_

1
2
m
2

T
= (

)
_

_
m
2

, (4.102)
where

=
_

1

2
_
, and =

1
+ i
2

2
,

=

1
i
2

2
. (4.103)
The Lagrangian in the complex eld representation, is more suggestive to
the fact that it yields particle and anti-particle states. We could compute the
Hamiltonian and momentum operators directly in terms of and

arriving
to the same expressions as in the representation with two real elds. To
compute the charge Q, we would need to identify what is the symmetry of
this new Lagrangian. It is perhaps already obvious that the Lagrangian is
invariant under a eld phase-redenition,
e
i
,

e
+i

. (4.104)
56
This is the equivalent of the rotation symmetry transformation that we have
found earlier, in the complex eld representation. Let us verify this, by
performing a rotation on the vector

,
_

1

2
_

_
cos sin
sin cos
__

1

2
_
;
_

1
i
2
_

_
cos i sin
i sin cos
__

1
i
2
_
;
1
+ i
2
e
i
(
1
+ i
2
)
; to e
i
. (4.105)
There is an ambiguity worth noting, when applying Noethers theorem
to nd the conserved charged under the transformation e
i
and

e
i

. The corresponding eld variations are,

= i,

= i

. (4.106)
Noethers theorem states that the quantity,
Q =
_
d
3
x
_
/
(
0
)
+
/
(
0

_
;Q =
_
d
3
x
_
+

(4.107)
is conserved. Obviously, if Q is conserved, then also every other operator
Q+c, with lambda, c constant numbers is also conserved. The expressions
for Q are therefore unique up to a multiplicative and an an additive constant.
The ambiguity on the additive constant is removed when we remove the
contribution of the vacuum to the charge of particle states (as we have done
for the energy). The operator
: Q : = Q0[ Q[0) , (4.108)
is ambiguous only up to a multiplicative factor, which essentially denotes the
units in which we measure the charge of a state.
4.3 Conserved Charges as generators of sym-
metry transformations
It may occur that we know of a conserved charge, without knowing the corre-
sponding symmetry transformation. For example, in our previous discussion,
57
we found the charge from the Klein-Gordon Lagrangian in the representation
of two-real elds before we discovered the U(1) symmetry of the Lagrangian
in the complex eld representation. A conserved charge is a generator of the
symmetry transformation. Let us assume that a Lagrangian is symmetric
under a eld transformation,

i

i
+
i
(4.109)
leading to a conserved charge
Q =
_
d
3
x

i
/
(
0

i
)

i
=
_
d
3
x

i
. (4.110)
The commutator of the charge operator Q with a eld operator
j
is,
[Q,
j
(x)] =
_
d
3
y

i
[
i
(y)
i
(y),
j
(x)]
=
_
d
3
y

i
([
i
(y),
j
(x)]
i
(y) +
i
(y) [
i
(y),
j
(x)])
=
_
d
3
y

i
_
i
ij

(3)
(y x)
j
(x) + 0
_
;[Q,
j
(x)] = i
j
(x). (4.111)
Exercise: Verify this explicitly computing the commutators [Q, ]
and
_
Q,

for the complex Klein-Gordon eld.

4.4 Casimir eect: the energy of the vacuum
A remarkable consequence of the quantization of the Klein-Gordon eld is
that a state with no particles, the vacuum [0), has energy. For the real
Klein-Gordon eld, for example, we nd that the vacuum energy is:
E
vacuum
=
(3)
(0)
_
d
3

k
2
. (4.112)
We can associate the innity
3
(0), with the volume of our physical system.
We have considered that the classical solutions for the elds extend to an
58
innite volume, and we have performed all space integrations from to
. If we were to consider a nite volume, this delta function would be
replaced by the volume of integration. Recall that,
_
d
3
xe
i

kx
= (2)
3

(3)
(

k). (4.113)
For

k = 0, we have
V =
_
d
3
x = (2)
3

(3)
(

0), (4.114)
and the density of the vacuum energy can be cast in the physically more
appealing form,
=
E
vacuum
V
=
_
d
3

k
(2)
3
_
m
2
+

k
2
2
. (4.115)
A second source of innity in the expression for the vacuum energy is the
integration over all frequency modes
k
=
_
m
2
+

k
2
.
The value of the energy of the vacuum, does not enter physical predictions
and we may not worry that it comes out innite. But we should be able to
see that the vacuum energy is dierent if, for a reason, the elds vanish in
some region of the space-volume if some frequencies
k
do not contribute
to the vacuum. This can be achieved, if the eld is forces to satisfy some
boundary conditions. Let us assume, that the Klein-Gordon eld is forced
to vanish on the planes x = 0 and x = L,
(x = 0, y, z, t) = (x = L, y, z, t) = 0. (4.116)
We should also assume that there is no mass term, m = 0, for convenience.
The boundary conditions lter only the wave-lengths with
k
x
=
n
L
. (4.117)
One must make a careful promotion of the eld into an operator. The general
eld solution which in addition satises the boundary conditions is:
(x

) =

n=0
sin
_
nx
L
_
_
d
2

(2)
2
2
k
_
a(k)e
ikx

+ a

(k)e
ikx

, (4.118)
with

k
=
_

k
2

+
n
2

2
L
2
. (4.119)
59
Notice that we only integrate over the perpendicular directions k
y
, k
z
, since
k
x
is discretized. Consequently, the volume integral
(3)
(0) will be replaced
by the surface of the planes
(3)
(0)
(2)
(0) = S/(2)
2
.
The vacuum energy in the new conguration with the two eld-free
planes is,
E
vacuum
S
=

n=0
_
d
2

(2)
2
_

2
n
2
L
2
+

k
2

2
, (4.120)
where we have dened

k

= (k
y
, k
z
). S is now the surface of each planes. In
polar coordinates,
E
vacuum
S
=

n=0
_
dk

(2)
_

2
n
2
L
2
+ k
2

2
. (4.121)
This integral appears to be divergent in the limit k

. Let us now work

with a slightly modied integrand,
E
vacuum
S
=

n=0
_
dk

k
1

(2)
_

2
n
2
L
2
+ k
2

2
. (4.122)
There exists a value of for which the integral is well dened. We shall
perform our calculation for such a value, and then we shall try to analytically
continue the result to = 0. We change variables once again,
k

= l

n
L
, (4.123)
and we obtain,
E
vacuum
S
=
1
4
_

L
_
3
_

n=1
n
3
_
_

0
dl

l
1

_
1 + l
2

; =
1
8
_

L
_
3
_

n=1
1
n
3+
_
_

0
dl
2

(l
2

2
_
1 + l
2

(4.124)
We recognize the sum as the -function,

n=1
1
n
3+
= ( 3). (4.125)
60
Performing the change of variables,
l
2

x
1 x
, (4.126)
we nd that the integral is,
_

0
dl
2

(l
2

2
_
1 + l
2

=
_
1
0
dxx

2
(1 x)
5
2
= B(1 /2, 3/2 + /2).
(4.127)
where we have recognized the beta function,
B(x, y) =
(x)(y)
(x + y)
. (4.128)
The nal result reads,
E
vacuum
S
=
1
8
_

L
_
3
B(1 /2, 3/2 + /2)( 3). (4.129)
Amazingly
1
, the limit 0 exists and it is nite. We nd,
E
vacuum
S
=

2
1440L
3
. (4.130)
The vacuum energy depends on the distance between the two planes, on
which the eld vanishes. Can we realize this in an experiment?
The electromagnetic eld is zero inside a conductor. If we place two
innite conducting sheets parallel to each other at a distance L, then we
can reproduce the boundary conditions of the setup that we have just studied.
The quantization of the electromagnetic eld is rather more complicated than
the Klein-Gordon eld, but as far as the vacuum energy is concerned, we shall
see that they are very similar.
Our analysis, leads to an amazing prediction. Two electrically neutral
conductors attract each other. This is known as the Casimir eect. Notice,
that the energy of the vacuum E
vacuum
gets smaller when the conducting
plates are closer. Therefore, there is an attractive force between them. This
is an eect that it has been veried experimentally.
The force per unit area (pressure or rather anti-pressure) between the two
conductors is:
T =

Evacuum
S
L
=

2
480L
4
. (4.131)
1
Many subtleties have been swept under the carpet in this calculation. They will
become clear when we introduce properly the method of dimensional regularization.
61
4.5 Can the Klein-Gordon eld be an one-
particle wave-function?
In the previous chapter, we performed the quantization of the Schrodinger
eld. We found out that eld quantization, although an elegant formalism,
it was not indispensable. We were able to connect this formalism with a tra-
ditional description where the Schrodinger eld was taken to be directly, a
wave-function of a single particle. It is a question worth answering, whether
eld quantization is truly necessary for a relativistic eld or it is just a su-
peruous formalism.
An one-particle wave-function, as a probability amplitude, is in general
complex. We will then assume that the complex Klein-Gordon eld is one.
Recall that the Lagrangian is written as,
/
KG
= (

) (

) m
2

. (4.132)
We denote the eld with in order to emphasize our intention to consider it
a wave-function. As such,

is its complex conjugate and not a hermitian

conjugate operator.
The complex Klein-Gordon eld Lagrangian and the Schrodinger eld
Lagrangian,
/
S
=

i
t
+
[[
2
2m
, (4.133)
are both symmetric under a phase transformation, as we have discussed in
detail. In the Schrodinger case, this lead to a conserved charge, which we
identied it with the probability for the particle to be anywhere in space,
Total Probability =
_
d
3
x[(x, t)[
2
= constant. (4.134)
Therefore, we could also interpret the quantity,
[(x, t)[
2
, (4.135)
as a probability density.
An attempt to repeat the same steps for the case of the Klein-Gordon
eld fails badly. The total probability integral is now the charge,
Q =
_
d
3
x
_
(x, t)

(x, t)

(x, t)

(x, t)
_
, (4.136)
62
with a density
(x, t)

(x, t)

(x, t)

(x, t), (4.137)

which is not positive denite, and thus physically unacceptable.
A second embarrassment from the Klein-Gordon equation as a relativistic
wave-function equation, is that it predicts particles with a negative energy.
This may be accepted for a free particle, if we assume that for whatever rea-
son we can only see particles with positive energy. However, if we couple
these particles to photons as it is required for all particles with electromag-
netic interactions, spontaneous emission of photons will turn all particles
into negative energy particles. Clearly, a single particle wave-function inter-
pretation of the Klein-Gordon eld is wrong and it cannot describe physical
particles. Quantum Field Theory is the only consistent approach.
63
Chapter 5
The Dirac Equation
Dirac identied the source of the problems in interpreting the Klein-Gordon
equation as a wave -function equation the fact that it was a second order
dierential equation in time. The correspondence,
i
t
E,
resulted to a second order polynomial equation for the energy of a free-
particle obeying, which of course has both a positive and a negative solution.
Diracs idea was then to nd a linear equation, whose dierential operator
represented a sort of a square-root of the dierential operator in the Klein-
Gordon equation. Explicitly, Diracs linear equation is,
(i

m) = 0. (5.1)
Multiplying the Dirac equation with the dierential operator (i

m)
from the left,
; (i

m) (i

m) = 0
;
_

+ m
2
_
= 0
;
_
1
2

+
1
2

+ m
2
_
= 0
;
_
1
2

+
1
2

+ m
2
_
= 0
;
_
1
2

+ m
2
_
= 0, (5.2)
64
which for

= 2g

, (5.3)
gives the Klein-Gordon equation.
In other words, if we can construct objects

which satisfy the algebra

of Eq. 5.3, known as Cliord algebra, then every solution (x, t) of the Dirac
equation, will also be a solution of the Klein-Gordon equation.
The objects

are anti-commuting and they cannot be just numbers.

They have to be matrices. The rst dimensionality for which we can nd
four non-commuting matrices
0
,
1
,
2
,
3
is four. In the following we shall
construct the so called gamma-matrices in four dimensions.
5.1 Mathematical interlude
It is useful to review here the properties of the 2 2 Pauli matrices, which
we shall use as building blocks for constructing -matrices by taking their
Kronecker product.
5.1.1 Pauli matrices and their properties
The Pauli matrices are,

1
=
_
0 1
1 0
_
,
2
= i
_
0 1
1 0
_
,
3
=
_
1 0
0 1
_
. (5.4)
The Pauli matrices satisfy the following commutation and anti-commutation
relations,
[
i
,
j
] = 2i
ijk

k
(
123
= 1), (5.5)
and

i
,
j
= 2
ij
I
22
. (5.6)
Adding the two equations together, we nd that the product of two Pauli
matrices is,

j
=
ij
I
22
+ i
ijk

k
. (5.7)
For example,

2
i
= I
22
,
1

2
= i
3
,
3

2
= i
1
, . . . (5.8)
The determinant and trace of Pauli matrices are,
det (
i
) = 1, (5.9)
65
and
Tr (
i
) = 0. (5.10)
The Pauli matrices, together with the unit matrix I
22
constitute a basis
of general 2 2 matrices M
22
. We write,
M
22
= a
0
I
22
+
3

i=1
a
i

i
. (5.11)
From the above decomposition we have,
Tr (M) = a
0
(M) +
3

i=1
a
i
Tr (
i
) = 2a
0
, (5.12)
and (exercise),
Tr (
k
M) = 2a
k
. (5.13)
We can therefore write,
M
22
=
Tr (M)
2
I
22
+
3

i=1
Tr (M
i
)
2

i
. (5.14)
5.1.2 Kronecker product of 2 2 matrices
Let us dene the Kronecker product of two 2 2 matrices
A =
_
a
11
a
12
a
21
a
22
_
, B =
_
b
11
b
12
b
21
b
22
_
, (5.15)
as
AB
_
b
11
A b
12
A
b
21
A b
22
A
_
=
_
_
_
_
b
11
a
11
b
11
a
12
b
12
a
11
b
12
a
12
b
11
a
21
b
11
a
22
b
12
a
21
b
12
a
22
b
21
a
11
b
21
a
12
b
22
a
11
b
22
a
12
b
21
a
21
b
21
a
22
b
22
a
21
b
22
a
22
_
_
_
_
. (5.16)
The following properties can be proved (exercise) explicitly,
det (A B) = (det A)
2
(det B)
2
, (5.17)
Tr (AB) = Tr (A) Tr (B) , (5.18)
66
(A B)

= A

, (5.19)
(A+ C) B = A B + C B, (5.20)
A(B + C) = AB + AC, (5.21)
and
(A
1
B
1
) (A
2
B
2
) = A
1
A
2
B
1
B
2
. (5.22)
5.2 Dirac representation of -matrices
The matrices

0
= I
22

3
, (5.23)
and

j
=
j
(i
2
) , (j = 1, 2, 3) (5.24)
satisfy the Cliord algebra,

= 2g

I
44
. (5.25)
We nd,
_

0
_
2
= (I
22

3
) (I
22

3
)
= I
2
22

2
3
= I
22
I
22
= I
44
;
_

0
,
0
_
= 2g
00
I
44
. (5.26)
Also, for the anti-commutator
_

j
,
k
_
with j, k = 1, 2, 3 we have
_

j
,
k
_
=
j

k
+
k

j
= (
j
(i
2
)) (
k
(i
2
)) + (
k
(i
2
)) (
j
(i
2
))
=
j

k
(i
2
)
2
+
k

j
(i
2
)
2
=
j
,
k
(I
22
) = (2
jk
I
22
) (I
22
)
;
_

j
,
k
_
= 2
jk
I
44
;
_

j
,
k
_
= 2g
jk
I
44
. (5.27)
67
Finally, the anti-commutator
0
,
j
with j = 1, 2, 3 is
_

0
,
j
_
=
0

j
+
j

0
= (I
22

3
) (
j
(i
2
)) + (
j
(i
2
)) (I
22

3
)
=
j
(i
2

3
) +
j
(i
3

2
)
=
j
(i
2
,
3
)
=
j

_
i2
23
I
22
_
=
j
0
22
= 0
44
;
_

0
,
j
_
= 2g
0j
I
44
. (5.28)
We have therefore found a set of - matrices satisfying the Cliord alge-
bra. This set is not unique. Let us dene new matrices,

= M
1
44

M
44
, (5.29)
with M
44
any 4 4 matrix with an inverse. Then,
_

_
= M
1
44

M
44
= 2g

M
1
44
I
44
M
44
= 2g

I
44
. (5.30)
The matrices

satisfy the same algebra, furnishing a dierent representa-

tion.
Let us now compute the hermitian conjugate of the matrices. We nd

=
0

0
, (5.31)
which can also be written as,

0
=
0
_

,
0
_

0
_
2

= 2g
0

. (5.32)
Explicitly, -matrices are hermitian for = 0,

=
0
, (5.33)
while they are anti-hermitian for = j = 1, 2, 3

=
j
. (5.34)
It is easy to verify this in the Dirac representation. We have

= (I
22

3
)

= I

22

3
= I
22

3
=
0
, (5.35)
and

= (
j
(i
2
))

j
(i

2
) =
j
(i
2
) =
j
. (5.36)
68
5.3 Traces of matrices
We can prove that matrices are traceless in every representation. Let us
start with
Tr (

) = Tr
_

)
2
_
= Tr (

) = g

Tr (

) (5.37)
(No summation over the index )
This is also equal to
Tr (

) = Tr (

) Tr
_
(

)
2

_
= 2g

Tr (

) g

Tr (

) .
(5.38)
For ,= , we then have
g

Tr (

) = g

Tr (

) ;Tr(

) = 0. (5.39)
It is left as an exercise to prove a number of useful identities for the traces
of gamma matrices (series 4).
5.4 matrices as a basis of 4 4 matrices
The Cliord algebra restricts the number of independent matrices that we
can construct by taking products or combinations of products of matrices
in a given representation. First, it imposes that the square of a matrix is
proportional to the unit 4 4 matrix,
(

)
2
=
1
2

= g

I
44
=
_
+I
44
, = 0
I
44
, = j = 1, 2, 3
(5.40)
This limits the number of terms in independent products of matrices to
at most four.
The only independent product of four matrices is,

5
= i
0

3
. (5.41)
Exercise: Show that you can write the matrix
5
as

5
=
i
4!

, (
0123
= 1). (5.42)
69
There are four independent products of three matrices,
0

2
,

3
,
2

1
and
3

0
. These can be written as a linear com-
bination of

, (5.43)
and products of two only matrices, by using the Cliord algebra.
Due to the Cliord algebra, only the six antisymmetric combinations
of products of two matrices are independent. These are the combi-
nations
01
,
02
,
03
,
12
,
13
,
23
with

=
i
2
[

] . (5.44)
In the Dirac representation
5
takes the form,

5
= i
0

3
= i (I
22

3
) (
1
(i
2
)) (
2
(i
2
)) (
3
(i
2
))
= i (
1

3
) (i
3

2
)
= i
_
i
2
3
_
(
1
) = I
22

1
;
5
=
_
0 I
22
I
22
0
_
. (5.45)
Exercise: Show that in the Dirac representation,

0j
= i
_
0
22

j

j
0
22
_
, and
jk
= i
jkl
_

l
0
22
0
22

l
_
(5.46)
with j, k = 1, 2, 3.
A general 4 4 matrix has 16 independent components. From a repre-
sentation of matrices we can construct 15 independent 4 4 matrices.
Together with the unit matrix I
44
, they constitute a basis. In other words,
a general 4 4 matrix M can be written as,
M
44
= a
0
I
44
+ a

+ a
5

+ a

. (5.47)
Exercise: Find the coecients a
0
, a
5
, a

for a given 4 4 matrix

M.
70
5.5 Lagrangian for the Dirac eld
After this survey of the properties of matrices, we return to the Dirac
equation,
(i, mI
44
) = 0. (5.48)
We have introduced the slash notation:
,a

, ,

. (5.49)
The Dirac dierential operator i, m is a 4 4 matrix. The object is a
four dimensional Dirac spinor and it has its own Lorentz transformation,
which we shall derive in detail later.
(x, t)
a
(x, t) =
_
_
_
_

1
(x, t)

2
(x, t)

3
(x, t)

4
(x, t)
_
_
_
_
. (5.50)
It is very important not to confuse a spinor with a four dimensional vector,
which has truly dierent Lorentz transformations.
We can now take the Hermitian conjugate of the Dirac equation,
(i, mI
44
) = 0
; i (

= 0
; i
_

= 0
; i
_

0
m

= 0 (5.51)
Multiplying this equation with
0
from the left, we obtain
i
_

(
0
)
2
m

0
= 0
; i
_

0
_

m(

0
) = 0. (5.52)
We now dene,

0
. (5.53)
Then, the Hermitian conjugate of the Dirac equation reads
i

= 0. (5.54)
71
We can obtain the Dirac equation and its conjugate from a Lagrangian
density,
/ =

(i, m) . (5.55)
Exercise: Derive the Euler-Lagrange equations for the spinor eld
components
a
and

a
.
This is a very important piece of information if we want to quantize the Dirac
eld . We will postpone the quantization of the Dirac eld until after we
gain a deeper understanding of the origin of the Lagrangian, and the Lorentz
transformation properties of spinor elds.
72
Chapter 6
Lorentz symmetry and free
Fields
A fundamental requirement for physical laws is that they must be valid for
all relativistic observers. Let us consider elds
i
(x

) describing a physical
system, in a certain coordinate reference frame. Consider now a dierent
relativistic observer. Space-time coordinates in the frame of the two observers
are related with a Lorentz transformation,
x

. (6.1)
The new observer will have to transform appropriately both elds and coor-
dinates when computing the Lagrangian,

i
(x)

i
(x

), /(x) /

(x

). (6.2)
We will require that the action is the same for both observers. In this way, if
it is an extremum for the physical values of
i
(x) in the rst reference frame,
it will also be an extremum for the physical values of the eld

i
(x

) in the
second relativistic frame. For proper Lorentz transformations (det (

) = 1)
we have
d
4
x

= d
4
xdet
_
x

_
= d
4
xdet (

) = d
4
x. (6.3)
Then
S = S

;
_
d
4
x/(x) =
_
d
4
x/

(x

) ;/(x) = /

(x

). (6.4)
73
We have just found that we can write relativistically invariant actions, and
therefore physical laws for all relativistic observers, if the Lagrangian density
is a scalar, i.e. it does not transform under a Lorentz transformation:
/

(x

) = /(x), x

. (6.5)
How many such Lagrangian densities can we nd? Obviously, Lorentz
symmetry is not sucient on its own to determine all physical laws. We can
restrict the problem to Lagrangians which describe free relativistic particles,
which do not interact with each other. The Lagrangians that we are after
may contain up to one only parameter, which is the mass m of the free
particles. The equations of motion and eld quantization should lead to the
correct relativistic energy-momentum relation,
p
2
= m
2
. (6.6)
A Lagrangian of free particles must contain derivatives of the elds

i
.
They are needed to generate the momentum

i
p

in the energy-
momentum relation. We include a generic label i, allowing for the possibility
of a particle with a mass m and multiple degrees of freedom, such as charge,
spin, etc.
Our question has now become more specic: How can we combine the
elds
i
and their derivatives

i
into a scalar Lagrangian with one only
mass parameter? We need two ingredients:
The transformation of derivatives

under x

,
The elds transformations
i
(x)

i
(x

) under x

.
It is straightforward to nd the transformation of derivatives,

=

x

=
x

=
_

1
_

. (6.7)
Similarly,

=
_

1
_

. (6.8)
74
6.1 Field transformations and representations
of the Lorentz group
In this section, we will examine the allowed possibilities for linear eld trans-
formations
1
,

i
(x)

i
(x

) = M()
ij

j
(x) (6.9)
under a space-time Lorentz transformation
x

. (6.10)
Lorentz transformations are a group, i.e if we perform two successive
Lorentz transformations,
x x

= x,
and
x

,
this is equivalent to performing a direct Lorentz transformation
x x

x,
with

. (6.11)
How about the matrices M()
ij
? Consider again the transformation x
x

x, where the eld is transformed as,

(x)

(x

) = M(

)(x). (6.12)
We can arrive to the frame x

by performing successive Lorentz transforma-

tions, x x

= x x

x. Then we have that,

(x

) = M(

(x

) = M(

)M()(x). (6.13)
For the last two equations to be consistent with each other, eld transforma-
tions M() must obey:
M(

) = M(

)M(). (6.14)
In group theory terminology, this means that the matrices M() furnish
a representation of the Lorentz group. Field Lorentz transformations are
therefore not random, but they can be found if we nd all (nite dimension)
representations of the Lorentz group.
1
Non-linear transformations can be made linear, and are not necessary to consider
(eg BRST symmetry in QFTII)
75
6.2 Scalar representation M() = 1
The simplest representation M() is the scalar representation, where a eld
(x) does not change under Lorentz transformations,
(x)

(x

) = (x). (6.15)
A Lagrangian of a scalar eld and its derivatives must have the transfor-
mations of the derivatives canceled among themselves. A single derivative
term,

is not a scalar. The minimum number of derivatives that are needed for a
scalar is two,
(

) (

) ,
2
.
The second term is actually equivalent to the rst up to a total divergence.
Let us assume that a Lagrangian contains this term,
/ = (

) (

) + . . . (6.16)
Then we can nd the mass dimension of the eld , by requiring that the
action is dimensionless (in natural units where c = h = 1 ).
[mass]
0
= [S] =
__
d
4
x/
_
= [x]
4
[/] = [mass]
4
[/] (6.17)
which means that all Lagrangian densities must have mass dimensionality 4,
[/] = [mass]
4
. (6.18)
From this we nd that a scalar eld has a mass dimensionality one,
[(

) (

)] = [mass]
4
; [

]
2
[]
2
= [mass]
4
; [mass]
2
[]
2
= [mass]
4
; [] = [mass]
1
. (6.19)
A relativistically invariant Lagrangian may include terms with no deriva-
tives, of the type

n
.
Notice that the Klein-Gordon equation
/
KleinGordon
=
1
2
(

) (

)
m
2
2

2
, (6.20)
is, indeed, invariant under Lorentz transformations.
76
6.3 Vector representation M() =
Another representation of the Lorentz group which is easy to gure out is
the vector representation, corresponding to
M(

) =

. (6.21)
A eld in the vector representation transforms as,
A

(x) A

(x

) =

(x), x

. (6.22)
Terms where all Lorentz indices are contracted are invariant under Lorentz
transformations, Exercise: Show that the following terms are invari-
ant under Lorentz transformations
(

) , (

) (

) , (

) (

) . (6.23)
Lorentz invariance is not the only symmetry that is required for free elds.
For, example, the Lagrangian for the electromagnetic eld is (as we shall see
in detail)
/
EM
=
1
2
[(

) (

) (

) (

)] , (6.24)
where only a certain combination out of two dimension four operators with
two derivatives is present in the Lagrangian. This is because of an addi-
tional symmetry under gauge transformations which constrains further the
Lagrangian form. Invariance of the Lagrangian under Lorentz transforma-
tions is a denite requirement for relativistic theory, but this is not sucient
in order to determine its exact form.
6.4 How to nd representations?
Are the scalar and the vector representation the only representations that
we can nd? How do we nd the rest of them? This question simplies
enormously if we look at small Lorentz transformations in arbitrary repre-
sentations and the corresponding generators.
A small Lorentz transformation is,

, (6.25)
77
and any n n matrix representation of this transformation,
M
ab
(

) = M
ab
(

) , a, b = 1 . . . n (6.26)
can be expanded in the small parameters

,
M
ab
(

) = M
ab
(

) +

M
ab
()

=0
+O(
2
). (6.27)
The ndimensional representation M
ab
(

) of the unit transformation is the

n n unit matrix,
M
ab
(

) = I
nn
. (6.28)
The derivatives of the representation matrices M with respect to the small
transformation parameters dene the so called generators of the representa-
tion,
(J

M
)
ab

1
i
M
ab
(

=0
. (6.29)
The generators form a basis, such that the representation of every small
Lorentz transformation can be written as a linear combination of the gen-
erators plus the unit,
M
ab
() = I
nn
+ i

M
. (6.30)
Finite Lorentz transformations are also generated by exponentiating lin-
ear combinations of the generators. If

is not expandable in small param-

eters, we can use the representation property,
M() = M(
1
) . . . M(
N
), with =
1

2
. . .
N
(6.31)
and construct it from the innite product of innitesimal Lorentz transfor-
mations with small parameters
ij
/N,
M() = lim
N
_
I
nn
+ i

N
J

M
_
N
= e
iJ

M
. (6.32)
The generators of a group representation is all we ever need to determine
completely the representation.
The full set of commutators of the generators dene what is known as
the Lie algebra of the group. Consider a Lorentz transformation made up
from the following sequence of small transformations,
=
1

1
1

1
2
. (6.33)
78
The representation of is then
M() = M(
1
)M(
2
)M(
1
1
)M(
1
2
)
=
_
I
nn
+ i
(1)

M
_ _
I
nn
+ i
(2)

M
_ _
I
nn
i
(1)

M
_ _
I
nn
i
(2)

M
_
;M() = I
nn
+
(1)

(2)

[J

M
, J

M
] + . . . (6.34)
Now we take seriously the statement that representations M() of any small
transformation can be written as a linear combination of the generators
of the representation plus the unit matrix. There must exist parameters
X

such that
M() =
_
I
nn
+ i
(X)

M
_
+ . . . (6.35)
From the last equations we have that

(1)

(2)

[J

M
, J

M
] = i
(X)

M
. (6.36)
Notice that the number of generators J
M
is nite and always the same for
every representation of the group. Specically for the Lorentz group we can
write only six independent derivatives lim
0
M

for a given representa-

tion M (recalling the antisymmetry of

).
Eq. 6.36 tells us that the commutator of two generators is a linear com-
bination of generators,
[J

M
, J

M
] = iC

M
. (6.37)
The coecients C

as known as structure constants. Powerful theorems

from group theory state the following:
1. The structure constants are indeed constants. The same ones irrespec-
tive of the choices of group parameters in Eq. 6.36.
2. The structure constants are characteristic of the group and they are
the same for all representations M().
We now have a better strategy in order to nd representations of the
Lorentz group:
First, we will use any of the representations that we already know, the
scalar or vector representation, in order to nd the structure constants
C

of the Lorentz group.

79
Then we shall attempt to nd matrices J

nn
for each nite dimension
n which satisfy the same commutation relations (known as Lie algebra)
of Eq. 6.37. If we succeed, then we have found an nn representation
of the generators, and by exponentiation, an n n representation of
the group.
6.4.1 Generators of the scalar representation
We shall use the known scalar representation in order to nd the structure
constants of the Lorentz group. In this representation, under a Lorentz
transformation,
x

(6.38)
a scalar eld remains invariant

(x

) = (x

) ;

) = (x

). (6.39)
Setting x
1
x in the above equation we have,

(x

) = (
1

), (6.40)
which for small Lorentz transformations becomes

(x

) = (x

)
= (x

(x

) +O(
2
)
= (x

)
1
2
(

) x

(x

) +O(
2
)
= (x

) +

1
2
(x

) (x

) +O(
2
)
;

(x) =
_
1
i
2

scalar
_
(x) (6.41)
with the generators of the Lorentz group in the scalar representation being:
J

scalar
= i (x

) . (6.42)
. In an exponentiated form we have,

(x) = e

i
2
J

scalar
(x). (6.43)
80
With one representation of the group generators at hand we can nd the
structure constants. An explicit calculation (exercise) yields the identity,
[J

2
scalar
, J

4
scalar
] =
i (g

3
J

4
scalar
+ g

4
J

3
scalar
g

4
J

3
scalar
g

3
J

4
scalar
) . (6.44)
This is the Lie algebra of the Lorentz group, and we expect it to be satised
by every other representation of the generators.
6.4.2 Generators of the vector representation
We can verify that the Lie algebra is universal to all generator representations
by repeating the same analysis for the generators of the vector representation.
A vector eld A

(x

), transforms as
A

(x

) A

(x

) =

(x

)
=
_

_
A

(x

) +O(
2
)
= A

(x

) +

(x

) +O(
2
)
= A

(x

) +
1
2
_

_
A

(x

) +O(
2
)
= A

(x

) +

2
_

_
A

(x

) +O(
2
)
; = A

(x

) =
_

i
2

(J

vector
)

_
A

(x

), (6.45)
with
(J

vector
)

= i
_

_
. (6.46)
A Lorentz transformation on a vector eld changes its orientation by mix-
ing the indices of A

. The generators of the transformation for the vector

orientation are given in Eq. 6.46. In an exponentiated form, we have
A

(x

) = e
i(J

vector
)

(x). (6.47)
Exercise: Prove that a Lorentz transformation of a vector eld at
a point is:
A

(x) = e
i
h
(J

vector
)

scalar
i
A

(x). (6.48)
81
We can now compute explicitly (exercise) the commutator,
[J

2
vector
, J

4
vector
]

=
i
_
g

3
(J

4
vector
)

+ g

4
(J

3
vector
)

4
(J

3
vector
)

3
(J

4
vector
)

_
.
(6.49)
We nd that the commutators of the vector representation satisfy the same
Lie algebra as in the scalar representation (Eq. 6.44), which is of course what
we expect from the theory of Lie groups.
6.5 Spinor representation
We can nd all possible generators for the representations of the Lorentz
with a nite dimension in a systematic manner. Of particular interest to us
is the Dirac or spinor representation of the Lorenz group, which has the same
dimensionality 4 4 s the vector representation. We can verify, after some
algebra, that the 4 4 matrices,
S

=
i
4
[

] , (6.50)
satisfy the same commutation relation as in Eq. 6.44. Explicitly,
[S

2
, S

4
] =
i (g

3
S

4
+ g

4
S

3
g

4
S

3
g

3
S

4
) . (6.51)
Therefore, they are generators of a new representation for the Lorentz group,
which is neither the scalar nor the vector representation. We call this new
representation spinor representation. A spinor eld
(x) =
_
_
_
_

1
(x)

2
(x)

3
(x)

4
(x)
_
_
_
_
, (6.52)
transforms under as,

(x

) = 1
2
(x), (6.53)
82
with a nite spinor transformation being
1
2
= e

i
2
S

. (6.54)
We nd that (exercise) the commutator of the generator S

with a
matrix is
[

, S

] = i
_

_
, (6.55)
or, in the more suggestive form,
[

, S

] = (J

vector
)

. (6.56)
The above equation can be exponentiated into a general result,

1
1
2

1
2
=

. (6.57)
Exercise: Prove Eq. 6.57 in the case of very small Lorentz transfor-
mations, by expanding in the transformation parameters

both
sides of the equation and using Eq 6.56.
In the same manner, we verify that:
1
2

1
1
2
= (
1
)

. (6.58)
6.6 Lorentz Invariance of the Dirac Lagrangian
It is easy to verify that both terms,

and

,, in the Dirac Lagrangian are
invariant under Lorentz transformations. As in many previous instances, we
shall use the mathematical properties of Lie groups, which allow us to nd
transformation rules for representations for small Lorentz transformations,
and simply exponentiate the results at the end in order to write down the
general transformations.
Starting from the transformation of a spinor eld, we have:
1
2

1
2

_
1
i
2

_
1 +
i
2

_
(6.59)
83
We also have,
S

=
_
i
4
[

]
_

=
i
4
_

=
i
4
_

0
,
0

;S

=
0
S

0
. (6.60)
Thus, we have for the transformation of the conjugate spinor eld,

0
_
1 +
i
2

0
, (6.61)
and for nite Lorentz transformations,

1
1
2

0
. (6.62)
We observe that

does not really transform with the inverse transformation

of . However,

does. Multiplying from the left with
0
both sides of the
above transformation rule, we obtain

1
1
2
_

0
_
2
;

1
1
2
. (6.63)
It is now obvious that the mass term in the Dirac Lagrangian is Lorentz
invariant,
m

1
1
2
1
2
= m

. (6.64)
The derivative term of the Dirac equation is also Lorentz invariant,

1
1
2

1
_

1
2

1
1
2

1
2
_
_

1
_

1
_

=

,. (6.65)
We see that the Dirac Lagrangian contains only the two simplest scalar
terms that may be formed out of elds transforming in the spinor represen-
tation of the Lorentz group. The relative sign between these two terms is
easy to determine by requiring that the equations of motion will also satisfy
the Klein-Gordon equation, which gives the relativistically correct energy
momentum relations, p
2
= m
2
.
84
6.7 General representations of the Lorentz
group
The scalar, vector, and spinor representations are the simplest representa-
tions of the Lorentz group. However, there is an innite number of n n
representations. All of them can be found easily by observing that the Lie
algebra of the Lorentz group is the same as for an SU(2) SU(2) group. Let
us consider the generators in an arbitrary representation, which satisfy the
commutation relationship,
[J

2
, J

4
] =
i (g

3
J

4
+ g

4
J

3
g

4
J

3
g

3
J

4
) .
(6.66)
We now dene the linear combinations,
K
i
J
0i
, (6.67)
and
L
i

1
2

ijk
J
jk
. (6.68)
We can change the basis of generators comprising from the six
2
generators
J

with the linear combinations L

i
, K
i
. The generators L
i
are generators
of rotations around the axisi, and the generators K
i
are generators of boosts
in the directioni (exercise: demonstrate this for the generators in
the vectorial representation). Rotations are a subgroup of the Lorentz
group, which satisfy an SU(2) algebra. We nd the commutation relations,
[L
i
, L
j
] = i
ijk
L
k
. (6.69)
However, boosts do not form a subgroup of the Lorentz group. Two boosts
in two dierent directions are not a boost in a third direction, but rather a
rotation. We nd that the commutators of boost generators are,
[K
i
, K
j
] = i
ijk
L
k
. (6.70)
The commutator of a boost and a rotation generator is,
[L
i
, K
j
] = i
ijk
K
k
. (6.71)
2
they are antisymmetric
85
Dene now the linear combinations,
J

i

L
i
iK
i
2
. (6.72)
Then the Lie algebra breaks into two independent SU(2) algebras,
_
J
+
i
, J
+
j

= i
ijk
J
+
k
, (6.73)
_
J

i
, J

= i
ijk
J

k
, (6.74)
and
_
J
+
i
, J

= 0. (6.75)
We have already learned about the representations of the SU(2) group in
Quantum Mechanics. These can be labeled according to a spin number,
j = 0,
1
2
, 1,
3
2
, . . .
Each representation acts on (2j + 1) objects
m
with
m = j, j + 1, . . . j 1, j, (6.76)
transforming them into each other under SU(2).
A general representation of the Lorentz group acts on (2j
+
+1)(2j

+1)
objects
m
+
,m
, and is labeled by the numbers (j
+
, j

), e.g.
(0, 0), (
1
2
, 0), (0,
1
2
), (0, 1), (
1
2
,
1
2
), . . . . (6.77)
The massive Dirac eld transforms in the mixed representation,
(
1
2
, 0) (0,
1
2
).
6.8 Weyl spinors
As we have already noted, matrices are not unique, and there are many
equivalent representations. We introduce here a representation which shows
better the decomposition of the Lorentz group into two SU(2) subgroups.
This is called the chiral or Weyl representation, dened as

0
= I
22

1
,
i
=
i
(i
2
). (6.78)
86
As an exercise, you can prove that indeed this set of gamma-matrices
satises the Cliord algebra,

= 2g

I
44
. (6.79)
In the chiral representation, the generators of the spinorial Lorentz group
representation
S

=
i
4
[

] , (6.80)
take a block-diagonal form,
S
0j
=
i
2

3
=
i
2
_

j
0
22
0
22

j
_
, (6.81)
and
S
jk
=
1
2

jki

l
I
22
=
1
2

jki
_

l
0
22
0
22

l
_
. (6.82)
A Lorentz transformation for a spinor is
1
2
= e

i
2
S

, (6.83)
and the exponent takes the explicit form,
i
i
2

=
i
2

0j
S
0j

i
2

j0
S
j0

i
2

jk
S
jk
=

2

3
_
i

2
I
22
_
, (6.84)
with

j
=

0j
2
and
l
=
1
2

jkl

jk
. (6.85)
Under a small Lorentz transformation a spinor transforms as,

= e

i
2
S

_
1 i
i
2

=
_
_
_
_
_
_
_
1
i
2
_

_

_
_

1

2
_
_
1
i
2
_

+ i

_

_
_

3

4
_
_
_
_
_
_
_
(6.86)
87
The two-dimensional objects,

L

_

1

2
_
and
R

_

3

4
_
, (6.87)
are called Weyl spinors and transform independently under a Lorentz
transformation,

L
=
_
1
i
2
_

_

_

L
, (6.88)
and

R
=
_
1
i
2
_

+ i

_

_

R
. (6.89)
In an exponential form,

L
=
L

L
and
L

R
=
R

R
, (6.90)
with

L
= e

i
2
(

)
, (6.91)
and

R
= e

i
2
(

+i

)
. (6.92)
The two-transformations are connected via complex conjugation.
Exercise: First we prove the identity,

=
2
. (6.93)
The spinor

c
L

2

L
, (6.94)
transforms as,

L

2
(1
i
2
_

_
)

L
=
2
(1 +
i
2
_

+ i

L
= (1
i
2
_

+ i

_
)
2

L
;
c
L

R

c
L
(6.95)
Similarly,

c
R

L

c
R
. (6.96)
88
6.8.1 The Dirac equation with Weyl spinors
We now write the Dirac equation in terms of Weyl spinors
L
and
R
,
0 = (i, m)
=
_
i
0

0
+ i
j

j
m1
44
_

=
_
i(I
22

1
)
0
i ( (i
2
))

m1
22
1
22
_

;
_
_
m i
_

0
+

_
i
_

_
m
_
_
_

L

R
_
= 0. (6.97)
The mass m mixes the two Weyl spinors
L
,
R
. However, for a massless
particle the Dirac equation is diagonalized into two independent equations
for the Weyl spinors,

R
=

R
, (6.98)
and

L
=

L
. (6.99)
These are known as Weyl equations.
89
Chapter 7
Classical solutions of the Dirac
equation
In this chapter, we shall solve classically the equations of motion for the Dirac
eld, i.e. the classical Dirac equation. As we have already demonstrated, a
eld which satises the Dirac equation it also satises the Klein-Gordon
equation. So, it must admit a plane-wave solution. We consider rst the
general case
(x) = u(p)e
ipx
, (7.1)
with
p
2
= m
2
. (7.2)
For now, we shall restrict ourselves to the cases with positive energy solu-
tions, p
0
> 0. Substituting into the Dirac equation, we obtain additional
restrictions for the spinor u(p),
(i, m) u(p)e
ipx
= 0
; (,p m) u(p) = 0. (7.3)
Using the Weyl representation for matrices,
0
= I
22

1
,
J
=
j
(i
2
),
we obtain
_
mI
22
p
0
I
22
p
p
0
I
22
+ p mI
22
__

_
=
_
0
0
_
, (7.4)
90
where we write the four-component spinor u(p) in terms of two Weyl-spinors
, with two components each:
u(p)
_

_
. (7.5)
It is of course possible to nd solutions of the above equation directly.
However, we may also want to use some cleverness, and exploit the fact that
the eld transforms as a spinor. Consequently,

(x

) = 1
2
(x), (7.6)
and equivalently
u

(p

)e
ip

= 1
2
u(p)e
ipx
;u

(p)e
i(p)(x)
= 1
2
u(p)e
ipx
;u

(p)e
ipx
= 1
2
u(p)e
ipx
u

(p) = 1
2
u(p). (7.7)
and u(p) also transforms as a spinor, under a Lorentz transformation,
p

. (7.8)
The strategy for nding a general solution u(p) is then simple. We rst solve
the Dirac equation for an especially simple value of the momentum p

, and
then we perform a Lorentz transformation to obtain the result for a general
value of the momentum.
7.1 Solution in the rest frame
The Dirac equation (Eq. 7.4) assumes its simplest form in the rest frame,
p

= m(1,

0). (7.9)
yielding,
_
mI
22
mI
22
mI
22
mI
22
__

_
=
_
0
0
_
, (7.10)
91
and, equivalently,
= . (7.11)
The general solution for the spinor u in the rest frame is,
u
_
p

= (m,

0)
_
=

m
_

_
. (7.12)
The factor

m is a convenient normalization for future purposes, and the
Weyl spinor is arbitrary. We can choose it to be a linear combination,
=

s=+,
c
s

s
, (7.13)
with

+
=
_
1
0
_
, and

=
_
0
1
_
. (7.14)
Notice that this basis diagonalizes
3
, with

. (7.15)
From this solution of the Dirac equation in the rest frame, we can go to
any other momentum value with a boost.
7.2 Lorentz boost of rest frame Dirac spinor
along the z-axis
First, we boost any vector along the z-axis, with a very small Lorentz trans-
formation,
p

, (7.16)
where,

= 0, except
03
=
30
= Y. (7.17)
Notice that,

0
3
= g
00

03
= Y,
3
0
= g
33

30
= (Y ) = Y. (7.18)
The transformed momenta are,
p
0

= p
0
+ Y p
3
, p
1

= p
1
, p
2

= p
2
, p
3

= p
3
+ Y p
0
, (7.19)
92
which is conveniently written as,
_
p
0

p
3

_
= (I
22
+ Y
1
)
_
p
0
p
3
_
. (7.20)
For a nite boost of a vector along the z-direction, we have
_
p
0

p
3

_
= e
Y
1
_
p
0
p
3
_
. (7.21)
We can compute the exponential easily, noting that
2
1
= I
22
,
e
Y
1
=

n=0
Y
n
n!

n
1
=

k=0
Y
2k
(2k)!

2k
1
+

k=0
Y
2k+1
(2k + 1)!

2k+1
1
= I
22

k=0
Y
2k
(2k)!
+
1

k=0
Y
2k+1
(2k + 1)!
;e
Y
1
= I
22
cosh Y +
1
sinh Y. (7.22)
We apply this transformation for the momentum vector at rest,
_
m
0
_

_
E
p
3
_
= (I
22
cosh Y +
1
sinh Y )
_
m
0
_
;
_
E
p
3
_
=
_
mcosh Y
msinh Y
_
. (7.23)
We can then invert the above equations and compute the parameter Y which
generates the Lorentz boost, known as rapidity, in terms of the components
of the boosted momentum. We nd (exercise:)
e
Y
=
E + p
3
m
, e
Y
=
E p
3
m
, Y =
1
2
ln
E + p
3
E p
3
. (7.24)
Let us now nd the transformation of a spinor, under the same Lorentz
93
boost. The transformation matrix is,
1
2
= e

i
2
S

= e

i
2

03
S
03

i
2

30
S
30
= e

Y
2

3

3
=

n=0
_

Y
2
_
n
n!
(
3

3
)
n
=

n=0
_

Y
2
_
n
n!

n
3

n
3
=

n=odd
_

Y
2
_
n
n!

n
3

n
3
+

n=even
_

Y
2
_
n
n!

n
3

n
3
=

n=odd
_

Y
2
_
n
n!

3

3
+

n=even
_

Y
2
_
n
n!
I
22
I
22
;1
2
= cosh
_
Y
2
_
I
22
I
22
sinh
_
Y
2
_

3
. (7.25)
Applying this spin boost transformation to our solution of the Dirac equation
at the rest frame, we obtain
u(p) = 1
2

m
_

s

s
_
;u
s
(p) =
_
_
_
_
E + p
3
(1
3
)
2
+
_
E p
3
(1+
3
)
2
_

s
_
_
E + p
3
(1+
3
)
2
+
_
E p
3
(1
3
)
2
_

s
_
_
(7.26)
where we have used the expression for the rapidity, derived in Eq. 7.24. We
observe that,
_
1
3
2
_
2
=
1
3
2
, (7.27)
or, equivalently,
1
3
2
=
_
1
3
2
. (7.28)
We can then collect all terms in Eq. 7.26 under a common square root,
u
s
(p) =
_
_
_
(E + p
3
)
(1
3
)
2
+ (E p
3
)
(1+
3
)
2

s
_
(E + p
3
)
(1+
3
)
2
+ (E p
3
)
(1
3
)
2

s
_
_
(7.29)
94
which simplies to
u
s
(p) =
_ _
E p
3

s
_
E + p
3

s
_
(7.30)
This is the general solution for the spinor u(p), when p

= (E, 0, 0, p
3
).
7.3 Solution for an arbitrary vector
We can obtain the spinor solution u(p) of the Dirac equation for an arbitrary
vector p

, by performing a rotation of the solution that we found for p

=
(E, 0, 0, p
3
). In fact, things are simpler now and there is not much work to
do, if we can write down the previous result in a covariant form.
We dene,

(1, ) , (7.31)
and

(1, ) . (7.32)
Then the expression for the spinor u
s
(p), Eq. 7.30, is written as
u
s
(p) =
_

p
s

p
s
_
(7.33)
You can verify (exercise) that this is actually a solution of the Dirac equation
for an arbitrary vector p

. For this, you will need the generally useful identity,

( p)( p) = p
2
= m
2
, (7.34)
and to rewrite the Dirac dierential operator as
,p m =
_
m p
p m
_
(7.35)
in Eq. 7.4.
7.4 A general solution
We have found a general solution for the Dirac equation, of the form
(x) = u
s
(p)e
ipx
, with p
0
> 0. (7.36)
95
Another possibility is,
(x) = v(p)e
+ipx
, with p
0
> 0. (7.37)
Substituting into the Dirac equation we nd that the spinor v
s
(p) satises
(,p + m) v(p) = 0. (7.38)
Repeating the same steps as for u(p), we nd this equation is satised by
spinors
v
s
(p) =
_

p
s

p
s
_
. (7.39)
A general solution for the classical Dirac equation is,
(x) =

s
_
d
3
p
(2)
2
2
p
_
a
s
(p)u
s
(p)e
ipx
+ b

s
(p)v
s
(p)e
+ipx

, (7.40)
where the a
s
(p) and b
s
(p) are arbitrary coecients, and

p
= p
0
=
_
p
2
+ m
2
. (7.41)
We have not discussed explicitly solutions with p
0
< 0. At the classical
level, these are included in the above integral, as you can easily verify by
repeating here the analysis of Section 4.1.1. Dirac interpreted these negative
energy solutions with an ingenious theory (Dirac sea), which lead to the
discovery of antimatter. We do not need to bother with this interpretation,
except out of historic interest. As with the Klein-Gordon equation, eld
quantization gives consistently rise to physical states which have a positive
energy.
96
Chapter 8
Quantization of the Dirac Field
To quantize the Dirac eld, we promote the eld into an operator,
(x, t) =
_
d
3
p
(2)
3
2
p

s
_
a
s
(p)u
s
(p)e
ipx
+ b

s
(p)v
s
(p)e
ipx

. (8.1)
The conjugate momentum of the eld is,
(x, t) =
/

= i

(x, t), (8.2)

where the Lagrangian of the Dirac eld is,
/ =

(i, m) . (8.3)
The conjugate eld is given by,

(x, t) =
_
d
3
p
(2)
3
2
p

s
_
a

s
(p)u

s
(p)e
+ipx
+ b
s
(p)v

s
(p)e
ipx

. (8.4)
We also nd,

(x, t)

(x, t)
0
=
_
d
3
p
(2)
3
2
p

s
_
a

s
(p) u
s
(p)e
+ipx
+ b
s
(p) v

s
(p)e
ipx

.
(8.5)
In terms of and

, the operators a
a
, b
s
are written as,
a
s
(p) =
_
d
3
x u
s
(p)e
ipx

0
(x), (8.6)
97
and
b
s
(p) =
_
d
3
x

(x)
0
e
ipx
v
s
(p). (8.7)
To prove the above you will need the spinor summations,

s
u
s
(p) u
s
(p) = ,p + m, and

s
v
s
(p) v
s
(p) = ,p m, (8.8)
which can be derived easily from the explicit expressions for u
s
(p), v
s
(p) of
the last chapter.
Using Noethers theorem, we can compute the Hamiltonian of the Dirac
eld, which is
H =
_
d
3

k
(2)
3
(2
k
)

s
_
a

s
(k)a
s
(k) b
s
(k)b

s
(k)

. (8.9)
In the above, we have been careful to preserve the order of operators as they
appear.
We can now proceed with the quantization procedure. It is left as an
exercise to show that imposing commutation relations, leads to particle states
b

s
(p) [0) which have a negative energy. This is due to the minus sign in
the second term of the Hamiltonian integral in Eq. 8.9. To obtain positive
energy eigenvalues for such states, we are forced to quantize by requiring
anti-commutation relations,
(x
1
, t), (x
2
, t) = i
(3)
(x
1
x
2
), (8.10)
and
(x
1
, t), (x
2
, t) = (x
1
, t), (x
2
, t) = 0. (8.11)
For the operators a
s
, b
s
this results to the anticommutation relations,
_
a
s
( p
1
), a

( p
2
)
_
=
_
b
s
( p
1
), b

( p
2
)
_
=
s

s
(2)
3
(2
p
1
)
(3)
( p
2
p
1
), (8.12)
a
s
( p
1
), a
s
( p
2
) = b
s
( p
1
), b
s
( p
2
) = 0, (8.13)
_
a

s
( p
1
), a

( p
2
)
_
=
_
b

s
( p
1
), b

( p
2
)
_
= 0. (8.14)
Using the anticommutation relation of Eq. 8.12, we cast the Hamiltonian of
Eq. 8.9 as,
H =
_
d
3

k
(2)
3
(2
k
)

s
_
a

s
(k)a
s
(k) + b

s
(k)b
s
(k)

+0[ H[0) , (8.15)

98
with
0[ H[0) =
(3)
(0)
_
d
3

k
k
. (8.16)
Exercise: Notice that the sign of the vacuum energy is now nega-
tive.
1. What would be the Casimir force in a gedanken experiment
where we could impose that the Dirac eld vanishes on two-
parallel plates?
2. In supersymmetric theories, for each fermion eld exists a
symmetric boson eld, with the same mass. What is the
Casimir force?
As in the Klein-Gordon eld, we can redene the eld Hamiltonian to
have zero vacuum expectation value,
: H : = H 0[ H[0) . (8.17)
The subtraction of the vacuum energy is implemented at the operator level
with normal ordering, which in the case of anticommuting elds requires an
additional (-) sign,
: b
s
(k)b

s
(k) : = b

s
(k)b
s
(k). (8.18)
Again with the use of Nothers theorem, we nd that the momentum is
conserved and takes the form,

P =
_
d
3

k
(2)
3
(2
k
)

s
_
a

s
(k)a
s
(k) + b

s
(k)b
s
(k)

. (8.19)
8.1 One-particle states
It is left as an exercise to repeat the steps that we followed in the case of the
Schrodinger and the Klein-Gordon elds, and to to prove that
there exists a vacuum state with
a
s
(p) [0) = b
s
(p) [0) = 0, (8.20)
the operators a
s
(p), b
s
(p), a

s
(p), b

s
(p) ladder operators.
99
We can build two types of one-particle states, using either the a

s
creation
operator
[p, s)
a
a

s
(p) [0) , (8.21)
or the b

s
creation operator
[p, s)
b
b

s
(p) [0) . (8.22)
These states are degenerate eigenstates of the Hamiltonian and momentum
operators. We dene the eld momentum four-vector,
P

(: H : ,

P). (8.23)
Then
P

[p, s)
a
= p

[p, s)
a
, (8.24)
and
P

[p, s)
b
= p

[p, s)
b
, (8.25)
with
p

=
_
_
p
2
+ m
2
, p
_
. (8.26)
These are states of positive energy, and the correct relativistic energy-momentum
relation,
p
2
= m
2
. (8.27)
8.1.1 Particles and anti-particles
The Dirac Lagrangian is symmetric under the U(1) transformation,
(x)

(x) = e
i
(x),

(x)

(x) = e
i

(x). (8.28)
For small values of , we nd
=

= i, (8.29)
and

= i

, (8.30)
or, equivalently,

= i

. (8.31)
100
Using Noethers theorem, we nd that the current,
J

/
(

)
+
/
(

)
,
J

, (8.32)
is conserved,

= 0. The corresponding conserved charge is,

Q =
_
d
3
xJ
0
=
_
d
3
x

0
=
_
d
3
x

. (8.33)
We can express Q in terms of creation and annihilation operators,
Q =
_
d
3

k
(2)
3
2
k

s
_
a

s
(k)a
s
(k) + b
s
(k)b

s
(k)

. (8.34)
We calibrate the charge of the vacuum state to be zero with normal ordering,
: Q : = Q0[ Q[0)
=
_
d
3

k
(2)
3
2
k

s
_
: a

s
(k)a
s
(k) : + : b
s
(k)b

s
(k) :

: Q : =
_
d
3

k
(2)
3
2
k

s
_
a

s
(k)a
s
(k) b

s
(k)b
s
(k)

. (8.35)
The states and [p, s)
a
and [p, s)
b
are non-degenerate eigenstates of the
charge operator. We nd,
: Q : [p, s)
a
= (+1) [p, s)
a
, (8.36)
and
: Q : [p, s)
b
= (1) [p, s)
b
. (8.37)
Therefore the operators a

s
(p) give rise to particles with momentum p

and
charge (+1), while the operators b

s
(p) give rise to anti-particles with the
same momentum p

and opposite charge (1). To emphasize this, we shall

display explicitly the charge of one-particle states,
[p, s, +) [p, s)
a
= a

s
(p) [0) , and [p, s, ) [p, s)
b
= b

s
(p) [0) . (8.38)
101
8.1.2 Particles and anti-particles of spin-
1
2
We have demonstrated in previous chapters that the Dirac Lagrangian is
invariant under Lorentz transformations. The spinor eld transforms at a
point as,
(x)

(x) = 1
2
(
1
x). (8.39)
The variation of the eld at a point is

(x) (x)
= 1
2
(
1
x) (x)
=
_
1
i
2

_
(x

) (x

) +O(
2
)
= (x

) (x

)
i
2

(x

) +O(
2
).
(8.40)
We can expand,
(x

) = (x

(x) i

2
J

scalar
(x

), (8.41)
where we remind that,
J

scalar
= i (x

) . (8.42)
Eq. 8.40 becomes,

=
i
2

(J

scalar
+ S

) . (8.43)
Noethers theorem, Eq. 2.79, leads to the result that the quantity
J

=
_
d
3
x
/
(
0
)
(J

scalar
+ S

) , (8.44)
is conserved. Explicitly,
J

=
_
d
3
x

0
(J

scalar
+ S

) . (8.45)
and, equivalently,
J

=
_
d
3
x

(J

scalar
+ S

) . (8.46)
102
This is very close to what we nd for the total angular momentum of a scalar
eld. But not exactly. The S

is novel, and it arises here for the rst time.

This term is responsible to giving spin, an intrinsic angular momentum, to
Dirac particle states.
How can we gure out if a particle has spin? We can do the following
test. We put the particle at rest, and measure its total angular momentum.
If we nd it zero, then the particle has an angular momentum only when it
moves and therefore has no intrinsic spin. If it is not zero besides having a
zero momentum, then it is all due to the intrinsic spin of the particle.
First, we would like to use normal ordering, as with every conserved
quantity, to make sure that the vacuum has a zero total angular momentum:
: J

: [0) = 0. (8.47)
Let us consider a state of a positively charged Dirac particle which has a zero
momentum.
[0, s, +) = a

s
(0) [0) . (8.48)
Does it have any angular momentum? The action of the total-angular mo-
mentum on the state is,
: J

: [0, s, +) =
_
: J

:, a

s
(0)

[0)+a

s
(0) : J

: [0) =
_
: J

:, a

s
(0)

[0) .
(8.49)
We now specialize on a rotation around the zaxis,

12
=
21
= , otherwise

= 0. (8.50)
After some algebra, we nd that
J
3
[0, s, +) : J

: [0, s, +) =
_
: J

:, a

s
(0)

[0)
=

r
_

3
2

r
_
[0, r, +)
=
s

3
2

(1)
[0, (1), +) +
s

3
2

(2)
[0, (2), +) . (8.51)
where, we choose the basis

(1)
=
_
1
0
_
, and
(2)
=
_
0
1
_
. (8.52)
which diagonalizes
3
,

(1)
= +
(1)
,
3

(2)
=
(2)
. (8.53)
103
Then we have,
J
3
[0, s, +) =
_

s,(1)
1
2

s,(2)
1
2
_
[0, s, +) . (8.54)
We now have a physical interpretation of the quantum number s. A particle
at rest with charge + and s = (1) has total angular momentum +
1
2
in the
zdirection,
J
3
[0, s, +) =
_

s,(1)
1
2

s,(2)
1
2
_
[0, s, +) . (8.55)
A particle at rest with charge + and s = (2) has total angular momentum

1
2
in the zdirection. We have then discovered that a general superposition
of

s
c
s
[p, s, +) describes a particle with total spin-
1
2
. Repeating the same
steps, we also conclude that a general anti-particle state

s
c
s
[p, s, ) is also
a state with spin-
1
2
.
In summary, the quantization of the Dirac eld yields quantum states
with spin
1
2
. Spin emerged simply as a component of the total angular mo-
mentum, due to the Dirac elds transforming in the spinorial representation
of the Lorentz group. This theory predicts also the existence of both matter
and anti-matter. Of course, the theory is highly successful. The electron
and positron are states of the quantized Dirac eld. The same is valid for
other known spin-
1
2
particles, the muon and tau leptons, the quarks and their
antiparticles.
8.2 Fermions
Particles of the Dirac eld are fermions. A state with two positively charged
particles is
[p
1
, s
1
, +; p
2
, s
2
, +) a

s
1
(p
1
)a

s
2
(p
2
) [0)
;[p
1
, s
1
, +; p
2
, s
2
, +) =
1
2
_
a

s
1
(p
1
)a

s
2
(p
2
) a

s
1
(p
1
)a

s
2
(p
2
)

[0) using Eq. 8.14

(8.56)
The state where all quantum numbers of the two particles are identical
p
1
= p
2
and s
1
= s
2
is forbidden,
[p, s, +; p, s, +) =
1
2
_
a

s
(p)a

s
(p) a

s
(p)a

s
(p)

[0) = 0. (8.57)
104
which is Paulis exclusion principle for fermions. Using the anti-commutation
of creation operators we can write a state of N identical particles or anti-
particles as,
[p
1
, s
1
, ; p
2
, s
2
, ; . . . ; p
N
, s
N
, ;) =
1
n!

inSn
sgn()a

s
(1)
a

s
(2)
. . . a

s
(N)
[0) ,
(8.58)
which is manifestly anti-symmetric.
The quantization of the Dirac eld explains another mystery of quantum
mechanics., i.e. that elementary particles with spin
1
2
are fermions.
8.3 Lorentz transformation of the quantized
spinor eld
A spinor eld transforms classically, as
(x)

(x

) = 1
2
(x). (8.59)
What is the transformation of the corresponding quantum eld? It does not
have to be the same, and in fact it is not, as for the classical eld. Relativistic
invariance at the classical level is cast as the requirement that the Lagrangian
is a scalar,
/

(x

) = /(x).
This is sucient to guarantee that observers in dierent reference frames nd
the same minimum value for the action. However, this is not the same crite-
rion that we must introduce for Lorentz invariance of the quantum system.
8.3.1 Transformation of the ladder operators
We consider a quantum one-particle state, e.g.
[p, s, +) .
Under a Lorentz transformation , a state transforms
[p, s, +) U() [p, s, +) . (8.60)
105
What is the new state, i.e. what is the transformation representation U()?
It is logical to require that the action of U() yields a particle state with a
momentum p

,
[p, s, +) U() [p, s, +) = N(, p) [p, s, +) . (8.61)
Here we encounter an ambiguity. Consider a four vector p

= (m, 0).
An SO(3) rotation leaves the vector p

invariant. In general, for any vector

k

there are Lorentz transformations W, forming what is called the little

group, which leave it invariant,
W

= p

. (8.62)
Therefore the representation U() is not unique, since, for example, a dif-
ferent one U()U(W) could also be used for transforming a state with a
momentum p to a state with a momentum p. It is beyond our scope to
discuss how this ambiguity is removed. We will assume here for simplic-
ity that there exists a method to choose uniquely one of the little group
transformations. The vacuum state has zero momentum and all Lorentz
transformations belong to the little group. We then have that
U() [0) = [0) . (8.63)
Requiring that the normalization of a state is Lorentz invariant, we obtain
[[p, s, +)[
2
= [U() [p, s, +)[
2
;U

= U
1
. (8.64)
We express Eq. 8.61 in terms of creation operators,
U() [p, s, +) = N(, p) [p, s, +)
;U()a

s
(p) [0) = N(, p)a

s
(p)U() [0)
;U()a

s
(p)U
1
() = N(, p)a

s
(p), (8.65)
where N(, p) is a normalization factor that we need to determine. Conju-
gating Eq. 8.65 we nd that,
U()a
s
(p)U
1
() = N(, p)

a
s
(p). (8.66)
106
Particle states are orthogonal,
p
1
, s
1
, +[ p
2
, s
2
, +) = 0[ a
s
1
(p
1
)a

s
2
(p
2
) [0)
= 0[
_
a
s
1
(p
1
), a

s
2
(p
2
)

[0) 0[ a

s
2
(p
2
)a
s
1
(p
1
) [0)
= (2)
3
2
p
1

(3)
(p
2
p
1
)
s
1
s
2
0[ 0) 0
;p
1
, s
1
, +[ p
2
, s
2
, +) = (2)
3
2
p
1

(3)
(p
2
p
1
)
s
1
s
2
. (8.67)
Two dierent particles in a reference frame should remain distinct in any
other reference frame. If we perform a Lorentz transformation, Eq. 8.61, we
have
p
1
, s
1
, +[ U
1
U [p
2
, s
2
, +) = 0[ Ua
s
1
(p
1
)U
1
Ua

s
2
(p
2
)U
1
[0)
= 0[ Ua
s
1
(p
1
)U
1
Ua

s
2
(p
2
)U [0)
= [N(, p
1
)[
2
(2)
3
2
p
1

(3)
(p
2
p
1
)
s
1
s
2
(8.68)
We can now prove that
2
p
1

(3)
(p
1
p
2
) = 2
p
1

(3)
(p
1
p
2
). (8.69)
Proof: Consider a scalar function f( p). Then,
f( p

) =
_
d
3
p
2
p
2
p

(3)
( p p

). (8.70)
The measure
_
d
3
p
2
p
=
_
d
4
p(p
2
m
2
)
_
p
0
> 0
_
is invariant under Lorentz transformations. Given that f is a scalar, we must
have that

(3)
( p p

)
is also a scalar, which proves Eq. 8.69. The orthogonality of the states is
therefore preserved under Lorentz transformations. Eq. 8.68 becomes
p
1
, s
1
, +[ U
1
U [p
2
, s
2
, +) = [N(, p
1
)[
2
(2)
3
2
p
1

(3)
(p
2
p
1
)
s
1
s
2
,
(8.71)
which for N(, p) = 1, guarantees that
p
1
, s
1
, +[ U
1
U [p
2
, s
2
, +) = p
1
, s
1
, +[ p
2
, s
2
, +) . (8.72)
107
8.3.2 Transformation of the quantized Dirac eld
Quantum eld theories provide us with states and operators which are all
made out of creation and annihilation operators. The latter transform under
Lorentz transformations in a representation U() of the Lorentz group, such
that (explicitly written for the Dirac eld ladder operators)
U()a

s
(p)U
1
() = a

s
(p), (8.73)
U()a
s
(p)U
1
() = a
s
(p), (8.74)
U()b

s
(p)U
1
() = b

s
(p), (8.75)
U()b
s
(p)U
1
() = b
s
(p). (8.76)
Measurable observables are obtained from matrix-elements of the type,
STATE
1
[

O[STATE
2
) . (8.77)
We measure rst one such matrix-element,
p, s, +[ (x) [p, s, +) , (8.78)
in a certain reference system. Then we Lorentz transform the state,
[p, s, +) [p, s, +)

= U() [p, s, +) , (8.79)

and the quantum eld,
(x)

(x

). (8.80)
We should not hurry into claiming that

(x

) = 1
2
(x). This is the trans-
formation of the classical eld. It does not have to transform in the same
representation as the quantum eld, which is now an operator. What is then
the quantum

(x

)? It should be such that the two measurements in the

two frames are the same,
p, s, +[ U
1
()

(x

)U() [p, s, +) = p, s, +[ (x) [p, s, +)

;U
1
()

(x

)U() = (x)
;

(x

) = U()(x)U
1
(). (8.81)
108
We can compute the right hand side, knowing the transformation of the
creation and annihilation operators from Eq. 8.73,
U(x)U
1
=
_
d
3
k
(2)
3
2
k

s
U
_
a
s
(k)u(k)e
ikx
+ b

s
(k)v(k)e
+ikx

U
1
=
_
d
3
k
(2)
3
2
k

s
_
a
s
(k)u(k)e
ikx
+ b

s
(k)v(k)e
+ikx

.
(8.82)
Now we use that,
d
3
k
(2)
3
2
k
=
d
3
k
(2)
3
2
k
, k x = k x, (8.83)
and
u(k) =
1
1
2
u(k), and v(k) =
1
1
2
v(k), (8.84)
which yields, setting p = k,
U(x)U
1
=
1
1
2
_
d
3
p
(2)
3
2
p

s
_
a
s
(p)u(p)e
ipx
+ b

s
(p)v(p)e
+ipx

=
1
1
2
(x). (8.85)
We have found the transformation of the quantum Dirac eld. It reads,
(x)

(x

) = U()(x)U
1
() =
1
1
2
(x). (8.86)
8.4 Parity
We now derive the transformation properties of a spinor eld under parity.
This is a discrete transformation,
(t, x) (t, x). (8.87)
Under parity, the momentum of a particle transforms as,
p p (8.88)
109
The representation of the parity transformation on an one-particle state
should be, for example,
U
P
[p, s, +) = [ p, s, +) , (8.89)
where is such that
1 = p, s, +[ U

P
U
P
[ p, s, +) = [[
2
p, s, +[ p, s, +) ;[[
2
= 1. (8.90)
This yield the following transformation for the creation and annihilation
operators,
U
P
a
s
( p)U
1
P
=
a
a
s
( p), (8.91)
and
U
P
b
s
( p)U
1
P
=
b
b
s
( p). (8.92)
Under a parity transformation, a Dirac quantum eld transforms as
U
P
(x, t)U
1
P
=
_
d
3

k
(2)
3
2
k

s
U
P
_
a
s
(

k)u(

k)e
ikx
+ b

s
(

k)v(

k)e
+ikx
_
U
1
P
=
_
d
3

k
(2)
3
2
k

s
_

a
a
s
(

k)u(k)e
ikx
+

b
b

s
(

k)v(

k)e
+ikx
_
(8.93)
Now we use that the measure is invariant under a parity transformation,
d
3

k
(2)
3
2
k
=
d
3
(

k)
(2)
3
2
k
, (8.94)
and we rewrite,
k x = Et

kx = Et (

k)(x). (8.95)
Eq. 8.93 becomes
U
P
(x, t)U
1
P
=
_
d
3
p
(2)
3
2
p

s
_

a
a
s
( p)u( p)e
i(Et p(x))
+

b
b

s
( p)v( p)e
+i(Et p(x))

, (8.96)
where we set p =

k. Recall now that, in the Weyl representation,

u
s
(p) =
_
E p

E + p
_
and
0
=
_
0 I
22
I
22
0
_
. (8.97)
110
We then have

0
u
s
( p) = u
s
( p), (8.98)
and similarly,

0
v
s
( p) = v
s
( p). (8.99)
We can now x the arbitrary normalization phases
a
,
b
to

a
=

b
= 1. (8.100)
Then, Eq. 8.96 yields a simple parity transformation for the eld,
U
P
(x, t)U
1
P
=
0
(x, t). (8.101)
It is easy to show that we can take U
P
to be its own inverse. We have,
U
P
(x, t)U
1
P
=
0
(x, t)
;(x, t) =
0
U
1
P
(x, t)U
P
;
0
(x, t) = (
0
)
2
U
1
P
(x, t)U
P
;U
P
(x, t)U
1
P
= U
1
P
(x, t)U
P
, (8.102)
which is satised for
U
P
= U
1
P
. (8.103)
We can study the transformation of composite operators made up from
two Dirac elds. The combination

which transforms as a scalar under
Lorentz transformations, transforms under parity as a scalar too.

(t, x)(t, x)

(t, x)

(t, x) = U
P

(t, x)U
P
U
P
(t, x)U
P
=

(t, x)
0

0
(t, x, t)
=

(t, x)(t, x). (8.104)
However, there is a combination

5
which transforms as a scalar under
Lorentz transformations (exercise: prove that it is indeed so), but trans-
forms dierently under parity.

(t, x)
5
(t, x)

(t, x)
5

(t, x) = U
P

(t, x)U
P

5
U
P
(t, x)U
P
=

(t, x)
0

0
(t, x, t)
=

(t, x)
5

0
(t, x, t)
=

(t, x)
5
(t, x). (8.105)
111
This is called a pseudoscalar.
The combination

which transforms as a vector under Lorentz trans-

formations (exercise), transforms also as a vector under parity.

(t, x)

(t, x)

(t, x)

(t, x) = U
P

(t, x)U
P

U
P
(t, x)U
P
=

(t, x)
0

0
(t, x, t)
=

(t, x)
_
2g
0

(
0
)
2
_
(t, x)
=
_
2g
0
1
_

(t, x)

(t, x) (8.106)
For = 0,

(t, x)
0
(t, x)

(t, x)
0
(t, x). (8.107)
For = i = 1, 2, 3,

(t, x)
i
(t, x)

(t, x)
i
(t, x). (8.108)
This is exactly how a vector x

= (t, x) (t, x) transforms under parity.

The combination

5
which transforms as a scalar under Lorentz
transformations (exercise: prove that it is indeed so), transforms oppo-
sitely under parity.

(t, x)
m
u
5
(t, x)

(t, x)

(t, x) = U
P

(t, x)U
P

5
U
P
(t, x)U
P
=

(t, x)
0

0
(t, x, t)
=

(t, x)
0

5
(t, x, t)
=

(t, x)
_
2g
0

(
0
)
2
_

5
(t, x)
=
_
2g
0
1
_

(t, x)

5
(t, x) (8.109)
This is called a pseudo-vector.
The Dirac Lagrangian,
i

, m

,
is composed from true scalar terms, and it transforms as a scalar under parity.
Exercise: Prove that the classical Dirac eld has the same parity
transformation as the quantum Dirac eld.
8.5 Other discrete symmetries
Exercise: Prove the following,
112
1. Under time-reversal (t, x) (t, x),
U
P
a
s
( p)U
P
= a
s
( p),
s
= i
2
(
s
)

. (8.110)
and
U
P
(t, x)U
P
= (t, x). (8.111)
2. Under charge conjugation, exchanging a particle with its an-
tiparticle,
U
C
a
s
( p)U
C
= b
s
( p), U
C
b
s
( p)U
C
= a
s
( p), (8.112)
and
U
C
(t, x)U
C
= i(

(t, x)
0

2
)
T
. (8.113)
113
Chapter 9
Quantization of the free
electromagnetic eld
Light is travelling with the maximum possible relativistic velocity. The quan-
tization of the electromagnetic elds and the realization that light is pack-
aged in photons, are one of the greatest advances ever made in the history
of physics.
9.1 Maxwell Equations and Lagrangian for-
mulation
We rst review the equations of Maxwell,

B = 0, (9.1)

E +

B
t
= 0, (9.2)

E = , (9.3)

E
t
=

j. (9.4)
We now introduce the four-vector,
A

_
,

A
_
, (9.5)
114
where the components are chosen such as,

B =

A, (9.6)
and

E =

A
t


. (9.7)
Then, the rst two Maxwell equations (Eq. 9.1, Eq. 9.2) are automatically
satised,

B =

A
_
= 0, (9.8)
and

A
t

_
+

t

A =

_
=
ijk

j
= 0. (9.9)
We now dene the tensor,
F

. (9.10)
The components of F

are,
F
0i
=
0
A
i

i
A
0
=
_

A
t
+

_
(i)
= E
i
, (9.11)
and
F
ij
=
i
A
j

j
A
i
=
ijk
B
k
. (9.12)
All other components are zero, since F

is anti-symmetric. In a matrix form

we have,
F

=
_
_
_
_
0 E
1
E
2
E
3
E
1
0 B
3
B
2
E
2
B
3
0 B
1
E
3
B
2
B
1
0
_
_
_
_
(9.13)
This is called the Electromagnetic eld tensor.
Dene now the current four-vector,
j

_
,

j
_
. (9.14)
115
We can easily verify that the remaining Maxwell equations are given by the
compact equation,

= j

. (9.15)
For = 0,

F
0
=
;
0
F
00
+
3

i=1

i
F
i0
=
;

E = . (9.16)
For = 1,

F
1
= j
1
;
0
F
01
+
3

i=1

i
F
i1
= j
1
;
E
1
t
+
B
3
x
2
+
B
2
x
3
= j
1
, (9.17)
which is the rst component of the vector equation

E
t
+

B =

j. (9.18)
Maxwell equations, expressed in terms of the vector potential A

, take
the form,

= 0
;

) = 0
;
2
A

) = 0. (9.19)
Classical gauge Invariance and gauge-xing
Maxwell equations, Eq. 9.15, are invariant under gauge transformations,
where the vector potential transforms as
A

= A

, (9.20)
116
or, in components,

= + (9.21)

A

A

=

A

. (9.22)
is an arbitrary function of space-time coordinates. Indeed, the eld tensor
F

is invariant under this transformation,

F

(A

(A

)
= F

. (9.23)
We can often derive simplify our calculations if we exploit gauge invari-
ance. Suppose that we are given a eld A

satisfying Maxwell equations

Eq. 9.19. We can x the gauge, i.e. perform a special gauge transforma-
tion, A

, such that the second term in Eq. 9.19 disappears,

0 =

+
2
;
2
=

. (9.24)
This particular choice of gauge is known as the Lorentz gauge. In the
Lorentz gauge, Maxwell equations take the very simple form (dropping the
prime)

2
A

= j

. (9.25)
In the vacuum, j

= 0, we have

2
A

= 0. (9.26)
This equation is particularly easy to solve as plane-waves, and we write the
general real classical eld solution as an integral
A

(x) =
_
d
3

k
(2)
3
2
k

=0

()
(k)
_
a

(k)e
ikx
+ a

(k)e
ikx

, (9.27)
with k

=
_

k
,

k
_
and
k
=

. We have introduced a basis of four-vectors

which determine the direction of the electromagnetic eld (polarization),

(0)
=
_
_
_
_
1
0
0
0
_
_
_
_
,
(1)
=
_
_
_
_
0
1
0
0
_
_
_
_
,
(2)
=
_
_
_
_
0
0
1
0
_
_
_
_
,
(3)
=
_
_
_
_
1
0
0
1
_
_
_
_
. (9.28)
117
The polarization vectors are normalized as,

()

(

)
= g

. (9.29)
The Lorentz gauge condition,

= 0, imposes in addition that,

_
d
3

k
(2)
3
2
k

=0
_
k
()
(k)
_ _
a

(k)e
ikx
a

(k)e
ikx

= 0. (9.30)
Lagrangian of the electromagnetic eld
The following Lagrangian,
/ =
1
4
F

, (9.31)
gives rise to the Maxwell equations as classical equations of motion. Explic-
itly,
/ =
1
2
[(

) (

) (

) (

)] . (9.32)
It is easy to verify that the Euler-Lagrange equations are

/
(

)

/
A

= 0
;
2
A

) = 0. (9.33)
9.2 Quantization of the Electromagnetic Field
The conjugate momentum for the eld components A

are,

=
/
(
0
A

)
= F
0
. (9.34)
One would like to impose the bosonic quantization condition,
[A

(x),

(x

)] = g

(3)
(x x

) . (9.35)
Notice the presence of g

, which is needed in order to maintain covariance on

the indices , of the quantization condition. We immediately come across
a problem. For = 0,

0
= F
00
= 0,
118
we encounter an inconsistency
0 =
_
A
0
(x),
0
(x

=
(3)
(x x

) ,= 0. (9.36)
Our rst attempt to quantizing the classical Lagrangian of the electromag-
netic eld has failed. However, it is possible to quantize a dierent La-
grangian which nevertheless gives rise to the same physics as the Lagrangian
of Eq. 9.31. We can exploit the fact that classical physics is the same in
every gauge for A

.
Consider a modied Lagrangian,
/ =
1
4
F

2
(

)
2
. (9.37)
This yields the following Euler-Lagrange equations, which is not invariant
under gauge transformations,

2
A

(1 )

) = 0. (9.38)
which are generally dierent than Maxwell equations. Notice however, that
for a special value = 1, the equations of motion are identical to Maxwell
equations,
2
A

= 0, in the Lorentz gauge. For the new Lagrangian and

with = 1, we nd

0
=
/

A
0
=

, (9.39)
which is not zero, as long as we do not impose the Lorentz condition

=
0. This is at rst irreconcilable with classical physics, however there is a
way out. We may think the classical electromagnetic eld as the expectation
value of a quantum eld in a physical quantum state [), i.e.
A

classic
[ A

operator
[) . (9.40)
We therefore only need to guarantee that the expectation value of

on
physical states [),
[

[) = 0, (9.41)
which may be done in a way such that

,= 0 as an operator identity.
We can now carry on with the quantization of the Lagrangian of Eq. 9.37.
The quantization conditions of Eq. 9.35, substituting in the expressions of
the conjugate momenta

=
L
(
0
A)
, yield
_

A

(x, t) , A

(x

, t)
_
= ig

(3)
(x x

) . (9.42)
119
We now substitute the solution of Eq. 9.27 into Eq. 9.43. We obtain,
_
a

(k), a

(k

)
_
= g

2k
0
(2)
3

(3)
_

_
. (9.43)
We can safely interpret the a
i
(k) and a

i
(k) for i = 1, 2, 3 as annihilation and
creation operators correspondingly, of photons with space-like polarizations

(i)
. But this is problematic for the operators a
0
(k), a

0
(k), which satisfy
_
a

(k), a

0
(k

)
_
= 2k
0
(2)
3

(3)
_

_
. (9.44)
Consider a general one-photon state with a time-like polarization,
[1) =
_
d
3

k
(2)
3
(2k
0
)
f(k)a

0
(k) [0) . (9.45)
The norm of the state is,
1[ 1) =
_
d
3

k
(2)
3
(2k
0
)
[f(k)[
2
0[ 0) , (9.46)
which is negative. In general, a state with n
t
photons with a time-like polar-
ization has a norm
n
t
[ n
t
) = (1)
nt
. (9.47)
This is a serious problem. Let us calculate the energy of such states. After a
standard, by now, calculation, we nd that the Hamiltonian is given by the
expression:
H =
_
d
3

k
(2)
3
2k
0
k
0
_
3

=1
a

(k)a

(k) a

0
(k)a
0
(k)
_
(9.48)
with k
0
=

> 0. The energy of the state [1) is

E
1
=
1[ H[1)
1[ 1)
, (9.49)
which yields (exercise),
E
1
=
_
d
3
k
(2)
3
(2k
0
)
k
0
[f(k)[
2
_
d
3
k
(2)
3
(2k
0
)
[f(k)[
2
< 0. (9.50)
120
We have found that this state has a negative energy. This problem is removed
if we implement properly a quantum Lorentz gauge condition.
We need Eq. 9.41 to hold for any physical state. We will now describe
how this can be achieved following the prescription of Gupta and Bleuler. We
separate the electromagnetic eld into a positive and a negative frequency
component,
A

(x) = A
(+)

(x) + A
()

(x), (9.51)
with
A
(+)

=
_
d
3

k
(2)
3
2k
0

(k)e
ikx
, (9.52)
and
A
()

=
_
d
3

k
(2)
3
2k
0

(k)e
+ikx
. (9.53)
Physical states [) are required that they satisfy

A
(+)

[) = 0. (9.54)
This is satised automatically by the vacuum state, [) = [0), since A
(+)

contains only annihilation operators. We also nd that,

[

[)
= [

A
(+)

[) +[

A
()

[)
= [

A
(+)

[) +[

A
(+)

[)

= 2Re [

A
(+)

[)
= 0. (9.55)
The Bleuler-Gupta condition takes the explicit form

A
(+)

[) =
_
d
3

k
(2)
3
2k
0
e
ikx
3

=0
k

(k)a

(k) [) = 0. (9.56)
It is easy to verify that the above is not satised for photon-states with a
purely longitudinal polarization, [) = [1), as in Eq. 9.45.
What is a physical state for a single photon? Consider the general one-
photon state with a momentum p

= (1, 0, 0, 1), choosing its momentum

along the z-axis. This state can be written as,
[) =
3

=0
c

(p) [0) , (9.57)

121
where c

are arbitrary coecients determining the mixture of polarizations

in the photon physical state. We must have,
0 =

A
(+)

[)
;0 =
_

_
= 0. (9.58)
From the expressions for the vector polarizations in Eq. 9.28 we derive that,
p
1
= p
2
= 0 (9.59)
and
p
0
= p
3
. (9.60)
Therefore, for a physical photon state we need that c
1
= c
3
. In other
words, a component of the photon state with a time-like polarization must
be accompanied with an opposite component with a longitudinal polariza-
tion. The contributions of these components cancel each other in all physical
quantities.
122
Chapter 10
Propagation of free particles
In previous chapters we studied the particle states from the quantization
of free elds, such as the Schrodinger, the Klein-Gordon, the spinor, and
the electromagnetic eld. We are now ready to study the simplest transition
amplitude, for the transition of a free particle from a space-time position y

(y
0
, y) to a position x

(x
0
, x), with x
0
> y
0
. The transition amplitude is
given by,
M
xy
= x[ y)[
x
0
>y
0
. (10.1)
where [x) , [y) are states of a single particle at positions x, y correspondingly.
10.1 Transition amplitude for the Schrodinger
eld
We shall compute the transition amplitude for the case of the Scrodinger
eld. This will give us an opportunity to compare with other theories, which
are relativistic, and appreciate the merits of the latter. We recall that the
Schrodinger eld is given by,
(x
0
, x) =
_
d
3

k
(2)
3
a(k)e
i(E
k
x
0

kx)
, (10.2)
where
E
k
=

k
2
2m
, (10.3)
123
and the operator a(k) satises the quantization condition,
_
a(k), a

(k

= (2)
3

(3)
(

). (10.4)
A state of a particle at a position x is given by the action of a eld on
the vacuum (see Section 3.5),
[x) =

(x
0
, x) [0) , [y) =

(y
0
, y) [0) . (10.5)
The transition amplitude of Eq. 10.1 is then,
M
yx
= 0[ (x
0
, x)

(y
0
, y) [0)
=
_
d
3

k
(2)
3
e
i(E
k
t

kx)
, (10.6)
with
t = x
0
y
0
, x = x y. (10.7)
We can compute the above integral exactly. We rst change to spherical
coordinates,
d
3

k = dkk
2
dcos d, (10.8)
and write

kx = k x cos . (10.9)
Then it is easy to perform successively the integrations in , cos and k.
The result reads,
M
yx
=
_
m
2it
_3
2
e
it
h
1
2
m(
x
t
)
2
i
. (10.10)
This is an oscillatory function which is dened for arbitrary values of the par-
ticle speed
x
t
even when its value is not smaller than the speed of light. Par-
ticles which are quanta of the Schrodinger eld, can propagate at space-time
intervals anywhere outside the light-cone, which is in blatant disagreement
with special relativity.
10.2 Transition amplitude for the real Klein-
Gordon eld
We write the real Klein-Gordon eld as a sum of a positive frequency and
a negative frequency term,
(x) =
(+)
(x) +
()
(x), (10.11)
124
with

(+)
(x) =
_
d
4
k
(2)
3
(k
2
m
2
)(k
0
)e
ikx
a(k), (10.12)
and

()
(x) =
_
d
4
k
(2)
3
(k
2
m
2
)(k
0
)e
+ikx
a

(k). (10.13)
We notice that,

(+)
(x) [0) = 0, (10.14)
and
0[
()
(x) = 0. (10.15)
The operator a(k) satises the quantization condition,
_
a(k), a

(k

= (2)
3
2
k

(3)
(

), (10.16)
and

k
=
_

k
2
+ m
2
. (10.17)
The amplitude for the transition y

, with x
0
> y
0
, is then
M
yx
= 0[

(x)(y) [0)
= 0[ (x)(y) [0)
= 0[
_

(+)
(x) +
()
(x)
_ _

(+)
(y) +
()
(y)
_
[0)
= 0[
(+)
(x)
(+)
(y) +
(+)
(x)
()
(y) +
()
(x)
(+)
(y) +
()
(x)
()
(y) [0)
= 0[
(+)
(x)
()
(y) [0)
= 0[
_

(+)
(x),
()
(y)

+
()
(x)
(+)
(y) [0)
= 0[
_

(+)
(x),
()
(y)

[0) . (10.18)
The commutator gives,
_

(+)
(x),
()
(y)

=
_
d
4
k
(2)
3
(k
0
)e
ik(xy)
(k
2
m
2
), (10.19)
which is a cnumber. We then have for the transition amplitude,
M
yx
= 0[
_

(+)
(x),
()
(y)

[0) =
_

(+)
(x),
()
(y)

0[ 0) =
_

(+)
(x),
()
(y)

,
(10.20)
125
and explicitly,
M
yx
=
_
d
4
k
(2)
3
(k
0
)e
ik(xy)
(k
2
m
2
). (10.21)
We would like to investigate whether the above expression for the tran-
sition amplitude respects the expectation from special relativity, which re-
stricts particles to transitions within the light-cone. It is now a bit more
tedious to evaluate the integral of Eq. 10.21. For simplicity, we shall evalu-
ate it for two special transitions.
Transition A: where we take x = x y = 0 and t = x
0
y
0
> 0,
which is a time-like transition (within the light-cone)
Transition B: where we take [x[ = [x y[ > 0 and t = x
0
y
0
= 0,
which is a space-like transition (within the light-cone).
Time-like transition A
The transition amplitude becomes,
M
yx
=
_
d
4
k
(2)
3
(k
0
)e
ik
0
t
(k
2
m
2
)
=
_
d
3

k
(2)
3
2
_

k
2
+ m
2
e
it

k
2
+m
2
=
1
(2)
2
_

0
dk
k
2

k
2
+ m
2
e
it

k
2
+m
2
(10.22)
Setting E =

k
2
+ m
2
, we nd that
M
yx
=
1
(2)
2
_

m
dE

E
2
m
2
e
iEt
(10.23)
For large times, t infty, we can prove (exercise) that
lim
t
e
imt
. (10.24)
The transition amplitude for a particle to remain within the light-cone, such
as in our exemplary case A, is then nite.
126
Space-like transition B
It is interesting to nd out whether transitions outside the light-cone, such as
in case B, are forbiden in a relativistic eld theory. In this case, the transition
amplitude becomes
M
yx
=
_
d
4
k
(2)
3
(k
0
)e
+i

kx
(k
2
m
2
)
=
_
dk
(2)
2
dcos
k
2
e
+ik(x) cos

k
2
+ m
2
=
1
(2)
2
_

0
dk
k
2
2

k
2
+ m
2
e
ik(x)
e
ik(x)
ik(x)
=
i
2(2)
2
(x)
_
+

dk
ke
ik(x)

k
2
+ m
2
. (10.25)
The above integrals has branch cuts at k = im, as in Figure 10.1. For
Figure 10.1: Branch cuts of the transition amplitude integral for a space-like
transition
(x) > 0 we can wrap the contour of integration around the upper branch
cut. The integral is then becoming, after setting k = i,
M
yx
=
1
4
2
(x)
_

m
d
e
(x)
_

2
m
2
. (10.26)
127
Asymptotically, for a large space-like interval x , the transition inte-
gral becomes
lim
(x)
M
yx
e
m(x)
. (10.27)
We nd that the transition amplitude vanishes exponentially the further
away we move from the light-cone (x = 0). Is this a contradiction with
special relativity? Actually it is not when we remember that our theory is in
addition a quantum theory. Essentially, a particle can escape from the light-
cone by a distance x which is not observable to us due to the uncertainty
principle. The transition amplitude is only signicant for
m(x) 1, (10.28)
which is the uncertainty we expect for determining its position for a momen-
tum uncertainty p m. Essentially, to observe that a particle of a certain
mass has escaped the light-cone we must rst measure the momentum of the
particle with an accuracy as good as its mass (in order to identify the particle
from its mass). Then, the uncertainty in its position will be as large as the
distance that we expect a nite probability for the particle to escape away
from the light-cone.
10.3 Time Ordering and the Feynman-St uckelberg
propagator for the real Klein-Gordon eld
In order for the correlation function,
0[ (x)(y) [0) ,
to be a transition amplitude where a particle is created at point y

and it
is destroeyd at a point x

, the time x
0
must occur after y
0
, i.e. x
0
> y
0
.
However,
0[ (x)(y) [0) = 0[ [(x), (y)] [0) +0[ (y)(x) [0)
= 0[ 0 [0) +0[ (y)(x) [0)
= 0[ (y)(x) [0) . (10.29)
We can write the same correlation function with the space-time points x
and y interchanged. How can we interpret this? If we insist on a causal
128
interpretation, where a particle is rst created at a point x and it is destroyed
at a point y, we must have y
0
> x
0
.
The Feynman-St uckelberg propagator is dened to account for both cases,
x
0
> y
0
and x
0
< y
0
, for arbitrary x, y. This is written as,
0[ T(x)(y) [0) = (x
0
y
0
) 0[ T(x)(y) [0)
+(y
0
x
0
) 0[ T(y)(x) [0) . (10.30)
The time-ordering symbol T is dened to order the operators following it
from the later to the earlier times.
The Feynman-St uckelberg propagator has a very nice integral represen-
tation. Explicitly, from the above denition and the result of Eq. 10.21, we
nd
0[ T(x)(y) [0) =
_
d
3

k
(2)
3
1
2
_

k
2
+ m
2
_
(x
0
y
0
)e
ik(xy)
+ (y
0
x
0
)e
ik(yx)

= lim
0
_
+

d
4
k
(2)
4
i
k
2
m
2
+ i
e
ik(xy)
. (10.31)
We can verify that the four-dimensional integral of the last line is equal to
the line before explicitly, by using Cauchys theorem. In Fig. 10.2 we plot
the poles of the integrand on the energy k
0
complex-plane. These are located
Figure 10.2: Poles of the integrand for the Feynman-St uckelberg propagator
129
at k
0
= + i and k
0
= i, with =
_

k
2
+ m
2
. At k
0
= i, the
exponential in the integrand behaves as,
e
ik(xy)
e
+i

k(x y)
e
(x
0
y
0
)
. (10.32)
If x
0
> y
0
we can close the contour of integration to the lower complex haplf-
plane, guaranteeing that the integrand vanishes at i. If x
0
< y
0
we must
close the contour to the upper half-plane. In each case, we pick up one of the
residue at k
0
= , recovering the two terms of the rst line in Eq. 10.31.
The expression for the Feynman propagator,
0[ T(x)(y) [0) = lim
0
_
+

d
4
k
(2)
4
i
k
2
m
2
+ i
e
ik(xy)
, (10.33)
is of paramount importance for the computation of generic transition ampli-
tudes in quantum eld theory. An important property is that it is a Greens
function of the Klein-Gordon equation. We can act on it with the dierential
operator of the Klein-Gordon equation,
_

2
+ m
2
_
0[ T(x)(y) [0) =
_

2
+ m
2
_
_
+

d
4
k
(2)
4
i
k
2
m
2
+ i
e
ik(xy)
= i
_
+

d
4
k
(2)
4
e
ik(xy)
;
_

2
+ m
2
_
0[ T(x)(y) [0) = i
4
(x y). (10.34)
10.4 Feynman propagator for the complex Klein-
Grodon eld
We will now compute the transition amplitude, or the Feynman propagator,
for the complex Klein-Gordon eld,
(x, t) =
_
d
3

k
(2)
3
2
k
_
a(k)e
ikx
+ b

(k)e
ikx

. (10.35)
Notice that the hermitian conjugate of the eld acting on the vacuum creates
a particle with charge +1 at a certain position x,
[particle at position x) =

(x, t) [0) , (10.36)

130
while a eld acting on the vacuum creates an anti-particle with charge -1,
[anti-particle at position x) = (x, t) [0) . (10.37)
Consider the correlation function,
0[ (y)

(x) [0) (10.38)

which we can interprete it as a causal sequence where a particle is rst
created at x

(x
0
, x) and it is destroyed at y

(y
0
, y). We must then
have y
0
> x
0
.
It is easy to prove that
0[ (y)

(x) [0) = 0[

(x)(y) [0) (10.39)

The inerpretation of the rhs is that an anti-particle is rst created at y

(y
0
, y) and it is destroyed at x

(x
0
, x). We must then have x
0
> y
0
.
We dene the Feynman propagator to account for the two equivalent
possibilities for the forward in time propagation of either a particle or an
anti-particle.
0[ T(x)

(y) [0)
= (x
0
y
0
) 0[ (x)

(y) [0) + (y
0
x
0
) 0[

(y)(x) [0) (10.40)

We can repeat the same computation as for the Feynamn propagator of the
real Klein-Gordon eld. The result is identical:
0[ T(x)

(y) [0) = lim

0
_
+

d
4
k
(2)
4
i
k
2
m
2
+ i
e
ik(xy)
. (10.41)
10.5 Feynman propagator for the Dirac eld
Let us now compute the correlation function of two Dirac spinor elds,
0[
a
(x)

b
(y) [0) =
_
d
3
p
(2)
3
2
p

s
u
s
a
(p) u
s
b
(p)e
ip(xy)
=
_
d
3
p
(2)
3
2
p
(,p + m)
ab
e
ip(xy)
;0[
a
(x)

b
(y) [0) =
_
i

+ m1
_
ab
_
d
3
p
(2)
3
2
p
e
ip(xy)
(10.42)
131
Similarly, we nd (exercise)
0[

a
(y)
b
(x) [0) =
_
i

+ m1
_
ab
_
d
3
p
(2)
3
2
p
e
ip(yx)
(10.43)
Notice the overall (-) sign for the transition amplitude of an anti-particle
with respect to the corresponding amplitude in the case of a particle.
We can combine the two forward in time propagations by including this
relative minus sign into the denition of the time ordering for fermions.
0[ T
a
(x)

b
(y) [0) = (x
0
y
0
) 0[
a
(x)

b
(y) [0)(y
0
x
0
) 0[

a
(y)
b
(x) [0)
(10.44)
From the above, we nd for:
0[ T
a
(x)

b
(y) [0) =
_
i

+ m1
_
ab
_
d
3
p
(2)
3
2
p
_
(y
0
x
0
)e
ip(yx)
+(x
0
y
0
)e
ip(xy)

=
_
i

+ m1
_
ab
_
+

d
4
k
(2)
4
i
k
2
m
2
+ i
e
ik(xy)
, (10.45)
which yields for the fermion Feynman propagator the integral representation,
0[ T(x)

(y) [0) =
_
+

d
4
k
(2)
4
i (,k + m)
k
2
m
2
+ i
e
ik(xy)
. (10.46)
10.6 Feynman propagator for the photon eld
The photon Feynman propagator is dened as,
0[ TA

(x)A

(y) [0) = 0[ TA

(x)A

(y) [0) (x
0
y
0
)+0[ TA

(x)A

(x) [0) (y
0
x
0
)
(10.47)
This is worked out in the exercise tutorials. We nd that
0[ TA

(x)A

(y) [0) =
_
d
4
p
(2)
4
ig

p
2
+ i
e
ip(xy)
. (10.48)
132
Chapter 11
Scattering Theory (S-matrix)
We have studied the quantum eld theories of free elementary particles with
spin-0 (Klein-Gordon eld), spin-
1
2
(Dirac spinor eld), and spin-1 (Electro-
magnetic photon eld). In all these cases, we could identify a eld four mo-
mentum P

=
_
H,

P
_
which was a conserved quantity. It was also relatively
easy to nd a complete set of eigenstates for P

. In all these free eld theo-

ries the Hamiltonian eigenstates happened to be, in addition, eigenstates of
a time-independent number operator. States with denite energy and mo-
mentum were, therefore, also states of a conserved number of indestructible
particles
1
.
In reality, however, particles interact and may also be destroyed or cre-
ated in a scattering process. Realistic eld theories contain additional terms
in their Lagrangian which do not make it possible to nd a constant particle-
number operator. In a free eld theory, conservation of energy and momen-
tum is a consequence of the fact that the theory is invariant under time and
space translations. This is a property which will also hold for interacting eld
theories. Conservation of the particle number was, on the other hand, only a
consequence of an accident that a eld free theory was rather simple. This
accident does not occur anymore in an interacting eld theory, where we
expect it to describe successfully particle creation and annihilation.
In this chapter we will consider interacting eld theories in general, with-
out specifying the details of their Lagrangian. We will only require that they
yield physical laws which are invariant under space-time translations, and
therefore their is a conserved eld momentum operator. We will also make
1
In mathematical parlance, the Hilbert space of states had a Fock-state representation.
133
one further assumption, which we will explain later in its detail, that par-
ticle interactions happen at short distances, and that particles when are at
far distances they behave as if they were free. These two considerations will
lead us to general expressions for the probability amplitude of a scattering
process.
11.1 Propagation in a general eld theory
We start our analysis by studying the propagator of elds in a general theory,
as we have specied it above. We shall also assume that there is a ground
state, which we denote by
[) .
For simplicity we consider here the propagator of a scalar (interacting) eld
(x), which we write as
[ T (x)(y) [) = [ T (x)1(y) [) . (11.1)
Let us form a unit operator 1 out of a complete set of states. Space-time
translation invariance dictates that P

(H,

P), the eld four momentum
of the Lagrangian system is a conserved quantity. Notice that we do not
need the explicit expression for P

; Noethers theorem guarantees that such

an operator does exist. We denote with [p) the corresponding eigenstates of
the momentum-operator,
P

[p) = p

[p) , (11.2)
or, in components,
H[p) = E
p
[p) ,

P [p) = p [p) , (11.3)
with
p

= (E
p
, p) . (11.4)
As we have remarked, the states [p), are not , necessarily, states with a
denite number of particles. To simplify the discussion, we also assume here
that all momentum eigenvalues are not light-like, having p
2
,= 0. Then,
for each state [p) with momentum p

, we can perform a Lorentz boost which

renders p = 0. Conversely, all states [p) can be produced by applying Lorentz
transformations to states [
0
) with

P [
0
) = 0. (11.5)
134
Let [
p
) a boosted state from [
0
). Then we can write,
1 = [) [ +

_
d
3
p
(2)
3
2E
p
()
[
p
)
p
[ . (11.6)
The sum in runs over all possible states with a zero space momentum [
0
).
Let us now assume that x
0
> y
0
. The propagator is then written as,
[ T (x)(y) [)[
x
0
>y
0
=
[ (x)
_
[) [ +

_
d
3
p
(2)
3
2E
p
()
[
p
)
p
[
_
(y) [)
= [ (x) [) [ (y) [)
+

_
d
3
p
(2)
3
2E
p
()
[ (x) [
p
)
p
[ (y) [) . (11.7)
Consider now the matrix-element
[ (x) [
p
)
, and recall that the system is invariant under space-time translations x

. The eld four-momentum is the generator of space-time translations.

If we know the value of the quantum eld at a point y

, then we can nd its

value at a dierent space-time point by applying,
(x

) = e
iP(xy)
(y)e
iP(xy)
. (11.8)
It is very convenient to express the eld in the matrix-element in terms of
the value of the quantum eld at the origin,
[ (x) [
p
) = [ e
iPx
(0)e
iPx
[
p
) . (11.9)
The ground state is invariant under space-time translations,
e
iPx
[) = [) . (11.10)
Also, a state [
p
) has a denite momentum p

= (E
p
(), p). Thus,
e
iPx
[
p
) = [
p
) e
ipx

p
0
=Ep()
. (11.11)
135
We can then write,
[ (x) [
p
) = [ (0) [
p
) e
ipx

p
0
=Ep()
. (11.12)
We now recall that the state [
p
) is produced from a state with a zero space-
momentum with a Lorentz boost,
[
p
) = U (
p
) [
0
) . (11.13)
We can also exploit that the vacuum state is invariant under Lorentz trans-
formations,
U (
p
) [) = U
1
(
p
) [) = [) , (11.14)
and cast Eq. 11.12 as
[ (x) [
p
) = [ U
1
(
p
) (0)U (
p
) [
0
) e
ipx

p
0
=Ep()
. (11.15)
For a scalar eld,
U (
p
) (x

)U
1
(
p
) = (

) , (11.16)
which for a position at the origin x

= 0 yields
U (
p
) (0)U
1
(
p
) = U
1
(
p
) (0)U (
p
) = (0). (11.17)
We then nd that the space-time dependence of the matrix-element in Eq. 11.15
is a simple exponential,
[ (x) [
p
) = [ (0) [
0
) e
ipx

p
0
=Ep()
. (11.18)
Similarly,

p
[ (y) [) =
0
[ (0) [) e
+ipy

p
0
=Ep()
. (11.19)
Finally,
[ (y) [) = [ e
iPy
(0)e
iPy
[) = [ (0) [) . (11.20)
Using Eqs 11.18-11.20, we nd that the propagator in Eq. 11.7 becomes,
[ T (x)(y) [)[
x
0
>y
0
= [ (0) [)
2
+

[[ (0) [
0
)[
2
_
d
3
p
(2)
3
2E
p
()
e
ip(xy)

p
0
=Ep()
. (11.21)
136
In the same fashion, we obtain for the second time-ordering possibility
[ T (x)(y) [)[
y
0
>x
0
= [ (0) [)
2
+

[[ (0) [
0
)[
2
_
d
3
p
(2)
3
2E
p
()
e
ip(yx)

p
0
=Ep()
. (11.22)
The propagator is,
[ T (x)(y) [) = [ T (x)(y) [)[
x
0
>y
0
(x
0
y
0
)
+ [ T (x)(y) [)[
x
0
>y
0
(x
0
y
0
), (11.23)
which yields,
[ T (x)(y) [) = [ (0) [)
2
+

[[ (0) [
0
)[
2

_
d
3
p
(2)
3
2E
p
()
_
e
ip(xy)
(x
0
y
0
) + e
ip(yx)
(y
0
x
0
)

p
0
=Ep()
.
(11.24)
We recognize that the above integral is the Feynman-St uckelberg propagator
of Eq. 10.31, of the free scalar eld theory. We denote by m
2

the eigen-value
of the squared eld-momentum P
2
corresponding to the state [
0
),
P
2
[
0
) = m
2

[
0
) . (11.25)
Then, for the boosted state [
p
), we have
P
2
[
p
) =
_
H
2

P
2
_
[
p
) =
_
E
p
()
2
p
2
_
[
p
) . (11.26)
Given that [
p
) = U(
p
), and P
2
is invariant under a Lorentz transformation,
we obtain that
E
p
()
2
p
2
= m
2

. (11.27)
Then, using the result of Eq. 10.31, we write
[ T (x)(y) [) = [ (0) [)
2
+

[[ (0) [
0
)[
2
_
d
4
p
(2)
4
i
p
2
m
2

+ i
e
ip(xy)
(11.28)
137
This is a general result for the propagator of any scalar eld in any quantum
eld theory for which is symmetric under Poincare (Lorentz and translation)
transformations.
Let us now assume that there exists a density of states [
0
), with P
2
[
0
) =
m
2

[
0
). We can replace the sum in the above expression with an integral,
[ T (x)(y) [) = [ (0) [)
2
+
_
dM
2
(M
2
m
2

) [[ (0) [
0
)[
2
_
d
4
p
(2)
4
i
p
2
M
2
+ i
e
ip(xy)
(11.29)
This is the KallenLehmann representation for the scalar eld propagator. It
is an amazing result for its simplicity, given its generality. Up to constants
([ (0) [)
2
, [[ (0) [
0
)[
2
), the propagator is determined fully by the
density of the energy eigenstates with zero space-momentum.
11.1.1 A special case: free scalar eld theory
Our result for the Feynman-St uckelberg can be now rederived as a special
case of the Kallen-Lehmann propagator of Eq. 11.29 when considering the
free Klein-Gordon eld. In this case, we are able to nd a simple expression
for the quantum eld (x),
(x) =
_
d
3
p
(2)
2
2
p
_
a(p)e
ipx
+ a

(p)e
+ipx

, (11.30)
with
p
=
_
p
2
+ m
2
. The vacuum state is,
[) = [0) , (11.31)
and it is annihilated by the ladder operator a(p),
a(p) [0) = 0[ a

(p) = 0. (11.32)
We can easily see that
0[ (0) [0) = 0. (11.33)
We also need to compute the density of states [
0
), with H[
0
) = m

[
0
)
and

P [
0
) = 0. In the case of the free Klein-Gordon eld we can identify
a number of particles for each eigenstate of the P

operator. We can build

138
states of higher and higher energy with p = 0, by applying repeatedly the
creation operator a

(0) on the vacuum. We have

H
_
a

(0) [0)
_
= m
_
a

(0) [0)
_
,
H
_
a

(0)
2
[0)
_
= 2m
_
a

(0)
2
[0)
_
,
. . .
H
_
a

(0)
n
[0)
_
= nm
_
a

(0)
n
[0)
_
,
. . .
The density is then
(M
2
m
2

) =

n=1
(M
2
n
2
m
2
). (11.34)
Finally, we require the constants
0[ (0) [
0
) = 0[ (0)
_
a

(0)
n
[0)
_
=
n,1
. (11.35)
Exercise: prove the above.
We now substitute Eqs 11.33, 11.34 and Eq. 11.35 into Eq. 11.29. We
recover
0[ T (x)(y) [0) =
_
d
4
p
(2)
4
i
p
2
m
2
+ i
e
ip(xy)
, (11.36)
which is the familiar Feynman-St uckelberg propagator for the free Klein-
Gordon eld.
11.1.2 Typical interacting scalar eld theory
It is not possible to compute the density (M
2
m
2

) of zero-momentum p = 0
states [
0
) in realistic eld theories where the Lagrangian contains additional
potential terms to the ones of the free scalar Klein-Gordon eld. We can
make, however, some reasonable assumptions which agree with observations
on the systems that we would like to describe.
We have emphasized several times that in interacting theories it is not, in
general, possible to identify a number of particles corresponding to a certain
139
energy-momentum eigenstate. Nevertheless, we do observe particles which
act as if they were free and isolated. Such a state is characterized by the mass
m of the particle. If we now consider a state of two approximately free
particles the energy of the state when the combined momentum of the two
particles is p = 0 is also approximately twice the mass of one-particle 2 m.
However, their interaction potential will not allow for this exact value, and
in general, their energy will be larger. States other than one-particle states
form a continuum of masses M
2
which is larger than the minimum energy
(corresponding to free particles) required to get two particles in the same
state. An exception to this occurs when the two-particles form a bound state,
where their binding energy is negative and their eective mass is generally
smaller than 2 m.
Although a very important issue, we will not analyze in this course bound
state problems. Then, a typical density is,
(M
2
m
2

) = (M
2
m
2
1particle
) + (M
2
4m
2
)
cont
(M
2
m
2

). (11.37)
where
cont
a continuous function. Substituting this density functional form
into the general expression of Eq. 11.29 for the scalar propagator we nd,
[ T (x)(y) [) =

[ (0)

1
particle
0
_

2
_
d
4
p
(2)
4
i
p
2
m
2
1
particle
+ i
e
ip(xy)
+
_
dM
2

cont
(M
2
m
2

) [[ (0) [
0
)[
2
_
d
4
p
(2)
4
i
p
2
M
2
+ i
e
ip(xy)
+[ (0) [)
2
(11.38)
The above result is very important, stating that one-particle states are re-
sponsible for poles in the propagator of an interacting eld. If a theory has
states which behave as isolated then we anticipate to nd poles in the prop-
agator for squared momenta p
2
= m
1
particle
which correspond to the physical
mass of the particles. Suggestive to the above, it is then worth to remember
in practice that the scalar propagator is
[ T (x)(y) [) =
_
d
4
p
(2)
4
i

e
ip(xy)
p
2
m
2
phys
+ i
+ continuum, (11.39)
140
where

[ (0)

1
particle
0
_

2
, (11.40)
a constant, and m
phys
the physical mass of a particle.
11.2 Spectral assumptions in scattering the-
ory
We are now ready to study using quantum eld theory the scattering of el-
ementary particles. Scattering processes occur in Lagrangian systems with
interaction potentials, i.e. additional terms in the Lagrangian other than the
ones that we have found in the quantization of free elds. It has never been
possible to nd an exact solution for the Hamiltonian of any eld theory
which describes realistic particle interactions in four space-time dimensions.
However, it is possible under some reasonable assumptions to develop a the-
ory for scattering probability amplitudes. For simplicity, we shall carry out
our analysis for the case of interacting scalar elds. This is sucient to
demonstrate the salient features of the scattering theory.
Our basic set of assumptions is the following,
The theory has a ground state [)
There are rst excited states [p) with
P
2
[p) = m
2
phys
[p) , (11.41)
corresponding intuitively to single-particle states. We also assume
that m
2
phys
0, which means that there are no tachyons (particles
which travel faster than the light)
All remaining eigenstates [q) of the eld momentum operator have
P
2
[q) = m
2
phys
[q) , (11.42)
with
M
2
4m
2
phys
. (11.43)
These states form a continuum, and they correspond, intuitively, to
multi-particle excitations.
141
The expectation value of the eld operator in the ground state is zero,
[ (x) [) = 0. (11.44)
Roughly, (x) [) should correspond to an one-particle excited state. A single
particle should not disappear spontaneously into a state with no particles [).
The corresponding probability amplitude should therefore vanish,
[ 1
particle
) [ (x) [) = 0.
11.3 In and Out states
As we have discussed earlier, states in interacting eld theories do not longer
describe systems of particles with a necessarily constant number of them.
Free eld theories have a Hilbert-space of states with states of denite particle
number, but Hamiltonian eigenstates in interacting theories are presumably.
Nevertheless, we do observe systems of free electrons and photons when they
are suciently separated. By the uncertainty principle, such states describing
particles which are isolated at certain times must be wave-packets.
Two classes of wave-packets are interesting in order to describe the scat-
tering of freely moving particles.
Wave-packets which are isolated at a time very far in the past, t = ,
but may be overlapping at some nite time t
Wave-packets which are isolated at in the far future, t = +, but may
be overlapping at some nite time t.
We should remember that in our development of the eld theory formal-
ism, we are using the Heisenberg picture for time evolution. In this pic-
ture, eld operators depend on time, while states are time independent
2
.
Time-independent states are perhaps counter-intuitive, and can mislead us
to believe that they describe non-evolving systems. But this is of course not
2
For example, in the free scalar eld case the eld has a solution which is explicitly
time-dependent,
(x) =
_
d
3
p
(2)
2
2
p
_
a(p)e
ipx
+ a

(p)e
+ipx

,
while states, [0) , [p) = a

(p) [0) , . . ., are manifestly time-independent.

142
true. Measurements of expectation values of eld operators on a state are
time-dependent, and in this sense, states encapsulate the time-evolution of
the physical system that they describe. We dene a state as an [in) state,
if it describes a system of freely moving in the very far past, which may or
may not be interacting at a later time. We dene a state as an [out) state,
if it describes a system of particles which may or may not be interacting at
a nite time but they are freely moving in the far future.
Consider an [in) state which describes a single particle with momentum p
in the far past. We will assume that free particles are indestructible, stable,
in the absence of other particles. Then this state will also describe a free
single particle in the far future, and it will also be an state [in):
[in, single particle, p) = [out, single particle, p) . (11.45)
More complicated in and out states containing many free particles in
the very far past or future respectively, allow for particle interactions at a
nite time. What are the possible in and out states? We will make the
following assumption:
Every possible state is a linear combination of either [in) or [out) states.
In physical terms, this means that every physical system will behave as a
system of freely moving particles which are getting increasingly isolated if we
wait long enough. Conversely, if we look further back in the far past, every
physical system originates from a system of freely moving isolated particles.
The above assumption is in accordance with the particle interactions oc-
cur when particles are brought very close together. Short range interactions
cannot keep particles together perpetually. There is a striking important ex-
ception to this assumption: elementary particles may form bound states.
In practice, we know from experimentation which are the bound states that
may be formed (hadrons, atoms, molecules, etc ). These are stable, and we
can account for them by including them in a generalized denition of [in)
and [out) states.
We adopt the following notation. We denote by
[p
i
, in)
an in state where the i
th
particle has momentum p
i
. We denote by
[k
j
, out)
143
an out state where the j
th
particle has momentum k
j
. These states are
complete. Every state can be written as a linear combination of either of
them, and we have:

[p
i
, in) in, p
i
[ =

[k
j
, out) out, k
j
[ = 1 (11.46)
11.4 Scattering Matrix-Elements
Obviously, an [p
i
, in) state can be written as a superposition of [k
j
, out)
states. This physically means that a system of originally free particles will
evolve to a system of ultimately free particles. If these particles never inter-
act, the in state is the same as the out state. If they do, then we may
have isolated particles after the interaction with dierent momenta or even
a dierent number of them. We write,
[p
i
, in) =

{k
j
}
c (k
j
, p
i
) [k
j
, out) . (11.47)
The probability amplitude for the transition of a system of i = 1 . . . N in-
coming particles with momenta p
i
to a system of j = 1 . . . M outgoing
particles with momenta k
j
is known as the scattering matrix-element,
S
ji
c (k
j
, p
i
) [k
j
) = out, k
j
[ p
i
, in) (11.48)
The scattering probability is
P (p
i
k
j
) = [S
ji
[
2
= [out, k
j
[ p
i
, in)[
2
. (11.49)
The scattering matrix-elements constitute the S-matrix, which is a matrix
acting on the space of in states and transforms them into out states,
S S
ij
, [, out) = S

[, in) . (11.50)
The S-matrix is unitary,
_
SS

, in[ out, ) , out[ , in)

= , in[
_

[, out) out, [
_
[, in) =

. (11.51)
144
We are usually not interested in computing the probability of not having
a scattering. The matrix-elements of the S-matrix which are relevant for a
true scattering are the non-diagonal. We then dene the so called transition-
matrix T, which is given by
S 1 +iT. (11.52)
In the rest of this chapter we shall prove a fundamental relation of S-
matrix elements to expectation values of eld operators in the ground state
of the interacting theory.
11.5 S-matrix and Greens functions
In the last section, we established very few properties of the S-matrix which,
at a rst sight, oer little help in computing it explicitly. It is possible,
however, to extract S-matrix elements from Greens functions, which are
dened as
G(x
1
, x
2
, . . . , x
N
) [ T (x
1
)(x
2
) . . . (x
N
) [) . (11.53)
In the simplest case of one space-time point,
[ (x) [) = 0, (11.54)
we have assumed that the corresponding Greens function vanishes, as part
of our spectral assumptions for an interacting eld theory.
The two-point Greens function is our familiar propagator, for which we
have found the Kallen-Lehmann representation
G(x, y) =
_
d
4
p
(2)
4
i

e
ip(xy)
p
2
m
2
phys
+ i
+ continuum. (11.55)
Consider the explicit case with x
0
> > > y
0
with , for two
space-time points where one is very far in the future and the second very far
145
in the past. Then,
G(x, y)[
x
0
+
y
0

= [ T (x)(y) [)
= [ (x)(y) [)
= [ (x)1 1 (y) [)
= [ (x)
_

out
[out) out[
_ _

in
[in) in[
_
(y) [)
; G(x, y)[
x
0
+
y
0

in

out
[ (x) [out) out[ in) in[ (y) [) (11.56)
We have dened an [out) as any state which behaves as a state of any number
of free isolated particles in the far future. The the sum over all out states
corresponds to

out
[ (x) [out) out[ [ (x) [) [
+
_
d
3
p
1
(2)
3
2
p
1
[ (x) [p
1
; out) p
1
; out[
+
_
d
3
p
1
(2)
3
2
p
1
d
3
p
2
(2)
3
2
p
2
[ (x) [p
1
, p
2
; out) p
1
, p
2
; out[
+ . . . (11.57)
We now make use the property of out states behaving as states of free
particles at large times x
0
> , and compute [ (x) [out) in free eld theory.
In free eld theory, the state (x) corresponds to a state with exactly one-
particle, and it has no overlap with states of a dierent particle multiplicity.
We then have,
[ (x) [) = [ (x) [p
1
, p
2
; out) = [ (x) [p
1
, p
2
, p
3
; out) = . . . = 0,
(11.58)
and only
[ (x) [p
1
; out) , = 0. (11.59)
We have already computed this non-vanishing already up to a constant. From
Eq. 11.18 and Eq. 11.40 we have
[ (x) [p; out) =
_

e
ipx
(11.60)
146
We compute the sum over in states in Eq. 11.56 in an identical fashion.
Then we arrive to a simple result
G(x, y)[
x
0
+
y
0

=

Z

_
d
3
p
(2)
3
2
p
d
3
q
(2)
3
2
q
e
i(pxqy)
p; out[ q; in) . (11.61)
Eq. 11.61 relates the Greens function function for two well separated in
time points to the transition amplitude from an in single-particle state to
an out single-particle state. How about more complicated Greens func-
tions and transition amplitudes? We can repeat the same procedure for a
general Greens function
G(x
1
, . . . , y
1
. . .)[
x
0
i
+
y
0
i

= [ T (x
1
) . . . (y
y
) . . . [)
x
0
i
+
y
0
i

. (11.62)
with N
out
points at far future times x
0
i
and N
in
points y
j
at far past times.
We nd that
G(x
1
, . . . , y
1
. . .)[
x
0
i
+
y
0
i

=
_
_
Nout

i=1
d
3
p
i
e
ip
i
x
i
(2)
3
2
p
i
__
N
in

j=1
d
3
q
j
e
+iq
j
y
j
(2)
3
2
q
j
_

Z
N
in
+N
out
2

p
1
, . . . , p
Nout
; out[ q
1
, . . . q
N
in
; in) (11.63)
We arrive to a very useful result. Greens functions with N
in
+ N
out
, with
N
in
of them chosen in the far past and N
out
in the far future are, up to an
overall constant, some sort of a Fourier transform of S-matrix elements for
the scattering of N
in
particles to N
out
particles. In the next section we shall
use Eq. 11.63 to compute the matrix-elements.
11.6 The LSZ reduction formula
We will compute scattering matrix elements by inverting Eq. 11.63. This
would require a type of a Fourier transformation on the space-time coordi-
nates x

i
and y

j
. The integration over the space components, x
i
and y
j
, can
be unrestricted (from to +), but we need to be careful concerning the
time integrations, since we have made assumptions about x
0
i
being far future
and y
0
j
far past times.
147
Let us try to integrate x

i
and y

j
over the maximum space-time region
where we do not contradict our time-ordering assumptions. Consider the
integral,
I [k
i
, l
j
]
_

i
_

dx
0
i
_
d
3
x
i
e
ik
i
x
i
__

j
_

dy
0
j
_
d
3
y
j
e
il
j
y
j
_
G(x
1
, . . . , y
1
, . . .). (11.64)
We now substitute the expression for the Greens function that we have found
in Eq. 11.63, and perform the space integrations using the identity
_
d
3
xe
i

kx
= (2)
3

(3)
(

k). (11.65)
The integral of Eq. 11.64 becomes,
I [k
i
, l
j
]
_

i
_

dx
0
i
e
i(k
0
i
(

k
i
))x
0
i
2(

k
i
)
__

j
_

dy
0
j
e
i(l
0
j
(

l
j
))y
0
j
2(

l
j
)
_

Z
N
in
+N
out
2

k
1
, . . . , k
Nout
; out[ l
1
, . . . l
N
in
; in) (11.66)
The integrals over the time variables are also easy to perform,
_

dte
it(l
0
(

l))
=
i
l
0
(

l)
_
lim
T
e
iT(l
0
(

l))
e
i(l
0
(

l))
_
(11.67)
The above is problematic due to the term which requires a not well dened
limit at innity. Here we pay a price for not having considered proper lo-
calized wave-packets for in and out states. This undened limit can be
evaded in a rigorous proof. An easy way out is to consider an alternative
integration over a time which is slightly complex
3
,
lim
0
_

dte
it(1+i)(l
0
(

l))
=
i
l
0
(

l)
e
i(l
0
(

l))
(11.68)
3
Certainly, our analysis here is very sloppy with exchanging limits, as well as with
not having considered proper wave-packets for in and out states. Careful and also
lengthier proofs can be found in the original literature and the book of Weinberg
148
Given the above caveats, we can write
I [k
i
, l
j
]
_

i
i
k
0
(

k)
e
i(k
0
(

k))
__

j
i
l
0
j
(

l
j
)
e
i(l
0
j
(

l
j
))
_

Z
N
in
+N
out
2

k
1
, . . . , k
Nout
; out[ l
1
, . . . l
N
in
; in) (11.69)
We have now achieved to relate Smatrix elements for the probability of a
transition of a system with N
in
particles with momenta l

j
to a system with
N
out
particles with momenta k
j
to a Fourier-type integrals over Greens
functions.
Let us analyze Eq. 11.69 further. We observe that a Fourier transformed
Greens function on the lhs has poles at
l
0
i
= (

l
i
) =
_

l
2
i
+ m
2
phys
,
due to every particle in the in and out states. We also notice that
the poles are independent of the time , since the dependent exponential
becomes one at the pole position l
0
= . Indeed , we can expand
e
i(l
0
)
2(l
0
)
=
1
2(l
0
)
+
i
2
+O((l
0
)
2
) (11.70)
This means that had we decided to put no restrictions in the time integra-
tions for I [k
i
, l
j
] the coecient of the poles (their residue) would be
unchanged. We can then write,
_
Regular at l
0
i
, k
0
j
(

l
i
), (

l
i
)
_
+
_
_

Z
1
2

2(

l
i
)(l
0
(

l
i
))
_
_
_
_

Z
1
2

2(

k
j
)(k
0
(

k
j
))
_
_
k
1
, . . . , k
Nout
; out[ l
1
, . . . l
N
in
; in)
=

G(k
1
, . . . , k
Nout
; l
1
, . . . l
N
in
) , (11.71)
where the Greens function

G in momentum space is dened as a normal
Fourier transform of a Greens function in xspace,

G(k
1
, . . . , k
Nout
; l
1
, . . . l
N
in
) =
_
+

d
4
x
1
. . . d
4
x
Nout
d
4
y
1
. . . d
4
y
N
in
e
i[
P
i
x
i
k
i

P
j
y
j
l
j]
[ T (x
1
) . . . (x
Nout
)(y
1
) . . . (y
N
in
) [)
(11.72)
149
A comment is in order concerning the time integrations in the above
integral. The exponential factors must be slightly deformed in the complex
plane,
e
ikx
e
ik
0
x
0
(1+i)i

kx
. (11.73)
Alternatively, we can leave intact the exponentials and perform a real time-
integration in Eq. 11.72, but then we have to evaluate the kernel Greens
function at deformed space-time points,

G(k
1
, . . . , k
Nout
; l
1
, . . . l
N
in
) =
_
+

d
4
x
1
. . . d
4
x
Nout
d
4
y
1
. . . d
4
y
N
in
e
i[
P
i
x
i
k
i

P
j
y
j
l
j]
[ T ( x
1
) . . . ( x
Nout
)( y
1
) . . . ( y
N
in
) [)
(11.74)
where
x

= (x
0
(1 i), x). (11.75)
We now multiply Eq. 11.71 with a factor
l
2
m
2
phys
i
_

for each in state and out state particle, and take the limit l
2
= (l
0

)(l
0
+ ) m
2
phys
. The regular term vanishes in this limit from the lhs of
the equation. We are left with the LSZ reduction formula,
k
1
, . . . , k
Nout
; out[ l
1
, . . . l
N
in
; in) = lim
l
2
i
=m
2
phys
k
2
j
=m
2
phys
_
_

i
l
2
i
m
2
phys
i
_

_
_
_
_

j
k
2
j
m
2
phys
i
_

_
_

G(k
1
, . . . , k
Nout
; l
1
, . . . l
N
in
) . (11.76)
which is due to Lehmann, Symanzin and Zimmermann.
11.7 Truncated Greens functions
Let us recall here the Kahlen-Lehman representation of the propagator of
Eq. 11.39, which we can invert by applying a Fourier transformation. We
nd

G(p) =
i

p
2
m
2
phys
+
_
Regular at p
2
m
2
phys
_
(11.77)
150
where we have dened the propagator in momentum space as

G(p)
_
d
4
xe
ipx
[ T (x)(0) [) . (11.78)
We can then cast the S-matrix from the LSZ formula as
k
1
, . . . , k
Nout
; out[ l
1
, . . . l
N
in
; in) =

Z
N
in
+N
out
2

G(k
1
, . . . , k
Nout
; l
1
, . . . l
N
in
)
_

G(k
i
)
__

G(l
j
)
_

k
2
i
m
2
phys
l
2
j
m
2
phys
(11.79)
A Greens function which is divided by a propagator for each external
particle in the in and out states is called truncated:

G
trunc
(k
1
, . . . , k
Nout
; l
1
, . . . l
N
in
)

G(k
1
, . . . , k
Nout
; l
1
, . . . l
N
in
)
_

G(k
i
)
__

G(l
j
)
_ (11.80)
The LSZ formula takes the form
k
1
, . . . , k
Nout
; out[ l
1
, . . . l
N
in
; in) =

Z
N
in
+N
out
2

G
trunc
(k
1
, . . . , k
Nout
; l
1
, . . . l
N
in
)

k
2
i
m
2
phys
l
2
j
m
2
phys
(11.81)
We now have a relation of probability amplitudes for the scattering of
free particles in Qunatum Field theory, in terms of truncated Greens func-
tions. As we have discussed already, there is no realistic eld theory in four
dimensions where all such probabilities can be computed exactly. In the
next chapter, we shall resort to perturbation theory, which turns to be an
amazingly powerful tool.
11.8 Cross-sections

151
Chapter 12
Perturbation Theory and
Feynman Diagrams
We have been able to solve exactly Hamiltonian systems for the simplest
quantum eld theories which describe free particles. For these, we could
nd all eigenstates and eigenvalues of the Hamiltonian. However, in general,
when interactions among particles are present, we must resort to perturbation
theory.
As a concrete example, we consider a simple new term in the Hamiltonian
of a real Klein-Gordon eld,
/ =
1
2
(

)(

)
1
2
m
2

4!

4
. (12.1)
The Hamiltonian density is,
H =

/, =
/

, (12.2)
The Hamiltonian is then found to be
H = H
0
+ H

, (12.3)
where
H
0
=
_
d
3
x

2
+
_

_
2
+ m
2

2
2
, (12.4)
is the Hamiltonian a free real Klein-Gordon eld, and
H

=
_
d
3
x

4!

4
, (12.5)
152
is a new term.
For a generic ,= 0, it is not possible to nd an exact expression for the
quantum eld . In this chapter, we will learn how to construct a solution
using perturbation theory, by expanding around the known result for = 0.
It is only necessary to know the eld (x, t) at a specic time value t = .
We can use symmetry under time translations in order to determine the eld
at any other moment t. The generator of space-time translations x

is the eld momentum P

=
_
H,

P
_
. Specically for time translations, the
generator is the Hamiltonian H. Fields at two dierent times are related via,
(x, t) = e
+iH(t)
(x, )e
iH(t)
(12.6)
(this is an exponentiated form of an equation such as Eq. 4.111, with Q =
P

, x
i
= 0). As you may have noticed eld operators evolve with time,
while states do not. This is usually termed as the Heisenberg picture.
Let us now compute the time evolution of the eld (x, ) in the hypo-
thetical case that there is no interaction ( = 0). We nd that

I
(x, t) = e
+iH
0
(t)
(x, )e
iH
0
(t)
, (12.7)
and solving for (x, ), we nd
(x, ) = e
iH
0
(t)

I
(x, t)e
+iH
0
(t)
. (12.8)

I
(x, ) is the eld in the interaction picture. The elds in the interaction
picture and the Heisenberg picture are identical if H = H
0
.
Combining Eq 12.6 and Eq. 12.8 we nd the identity
(x, t) = U
1
(t, )
I
(x, t)U(t, ) (12.9)
where we have dened the time-evolution operator in the interaction picture
U(t, ) e
+iH
0
(t)
e
iH(t)
. (12.10)
We now make a very important assumption. We require that there is a
time, for example in the far past, for which the eld of the full Hamiltonian H
is a solution of the free Hamiltonian H
0
. This is a reasonable approximation
if we can identify a time that particles are far from each other and they do
not feel their interaction can be neglected. We can formally implement this
by requiring for example that the interaction switches on at a certain time.
Given the existence of such a special time, the interaction eld
I
(x, t)
will continue to be a solution of the free Hamiltonian at any time t. Then,
Eq. 12.9 is a transformation of the eld
I
in the free theory (H
0
) to the eld
in the full interacting theory (H).
153
12.1 Time evolution operator in the interac-
tion picture
The calculation of the eld operator in the full theory proceeds through the
evaluation of the time-evolution operator U(t, ) given that we can determine
the eld operator in the free theory.
We shall rst derive a dierential equation for U(t, ). Dierentiating
with respect to time, we have:
i
U(t, )
t
= i
_
iH
0
e
iH
0
(t)
e
iH(t)
ie
iH
0
(t)
He
iH(t)

= e
iH
0
(t)
(H H
0
)e
iH(t)
= e
iH
0
(t)
(H H
0
)e
iH
0
(t)
e
iH
0
(t)
e
iH(t)
(12.11)
We write the above in the form,
i
U(t, )
t
= V
I
(t )U(t, ), (12.12)
with the interaction potential dened as,
V
I
(t ) e
iH
0
(t)
(H H
0
)e
iH
0
(t)
. (12.13)
We can cast the general solution of the dierential equation Eq. 12.12
as a time ordered exponential (as with an ordinary Schrodinger equation).
Integrating both sides of the equation with respect to time, we obtain
U(t, ) = U(, ) +
1
i
_
t

dt
1
V
I
(t
1
)U(t
1
, ). (12.14)
Let us rewrite the above equation, replacing t with t
1
and t
1
with t
2
. We
have,
U(t
1
, ) = U(, ) +
1
i
_
t

dt
2
V
I
(t
2
)U(t
2
, ). (12.15)
We can substitute Eq. 12.15 into Eq. 12.14, obtaining
U(t, ) =
_
1 +
1
i
_
t

dt
1
V
I
(t
1
)
_
U(, )
+
_
1
i
_
2
_
t

dt
1
_
t
1

dt
2
V
I
(t
1
)V
I
(t
2
)U(t
2
, ).
(12.16)
154
Obviously, we can repeat inserting Eq. 12.14 to itself as many times as we
wish. After an innite number of iterations we obtain
U(t, ) =
_
1 +

n=1
_
1
i
_
n
_
t

dt
1
_
t
1

dt
2
. . .
_
t
n1

dt
n
V
I
(t
1
)V
I
(t
2
) . . . V
I
(t
n
)
_
U(, ). (12.17)
Time-Ordering
Consider one of the simplest integrals in the above series,
I[t, ] =
_
t

dt
1
_
t
1

dt
2
V
I
(t
1
)V
I
(t
2
). (12.18)
It can be written as
I[t, ] =
_
t

dt
1
_
t

dt
2
V
I
(t
1
)V
I
(t
2
)(t
1
t
2
). (12.19)
By changing integration variables, t
1
t
2
and t
2
t
1
, we obtain
I[t, ] =
_
t

dt
1
_
t

dt
2
V
I
(t
2
)V
I
(t
1
)(t
2
t
1
). (12.20)
Adding Eq. 12.19 and Eq. 12.20, we obtain
_
t

dt
1
_
t
1

dt
2
V
I
(t
1
)V
I
(t
2
) =
1
2!
_
t

dt
1
dt
2
T V
I
(t
1
)V
I
(t
2
) ,
(12.21)
where the symbol T. . . is our familiar time-ordering symbol, ordering op-
erators from the largest to the smallest times,
T V
I
(t
1
)V
I
(t
2
) = V
I
(t
1
)V
I
(t
2
)(t
1
t
2
)
+V
I
(t
2
)V
I
(t
1
)(t
2
t
1
). (12.22)
It is not hard to convince ourselves that in general,
_
t

dt
1
V
I
(t
1
)
_
t
1

dt
2
V
I
(t
2
) . . .
_
t
n1

dt
n
V
I
(t
n
)
=
1
n!
_
t

dt
1
. . . dt
n
T V
I
(t
1
) . . . V
I
(t
n
) . (12.23)
155
Time-Ordered Exponentiated Integral
We dene a time-ordered exponentiated integral as the time-ordering of its
Taylor series expansion,
Te
R
t

dt

A(t

)
T

n=0
_
_
t

dt

A(t

)
_
n
n!
= 1 +
1
n!

n=1
_
t

dt
1
. . . dt
n
T A(t
1
) . . . A(t
n
) . (12.24)
For times t > t
a
> t
o
we have the identity,
Te
R
t
ta
dt

A(t

)
Te
R
ta

dt

A(t

)
= T
_
e
R
t
ta
dt

A(t

)
e
R
ta

dt

A(t

)
_
= Te
R
t

dt

A(t

)
, (12.25)
where we can put the two exponentials under a common time-ordering symbol
given that the times of the operators in the rst exponent are always larger
than the times of the operators in the second exponent.
We can now write a compact expression for the time-evolution operator of
Eq. 12.17 as a time-ordered exponential, by using Eq. 12.23 and the denition
of Eq. 12.25. We obtain that,
U(t, ) = Te
i
R
t

dt

V (t

)
U(, ). (12.26)
12.2 Field operators in the interacting and
free theory
Let us summarize here what we have achieved.
The eld in the full theory (x, t) is related to the eld in the free
theory
I
(x, t) at any time t, via a simple relation,
(x, t) = U(t, )
1

I
(x, t)U(t, ). (12.27)
The time evolution operator, U(t, ), is a time ordered exponential
given by Eq. 12.26.
156
To determine the operator U(t, ) at an arbitrary time, we require a
boundary value U(, ) at a time . We can select to be any time for which
the eld in the full theory is essentially equal to the eld in the free theory,
(x, ) =
I
(x, ). (12.28)
Such as special time is not hard to nd, if the assumptions that we have
made for the Smatrix of the theory are correct. Essentially, elds are most
of the time free, except for the short duration of particle interactions. We can
then select to be a time in the far past or the far future of the scattering
event that we would like to describe. For this time, we can set
U(, ) = 1. (12.29)
A time-ordered product of elds in the full theory can be written as,
T (x
1
)(x
2
) . . . (x
n
) =
T
_
U
1
(x
0
1
, )
I
(x
1
)

U(x
0
1
, x
0
2
)
I
(x
2
)

U(x
0
2
, x
0
3
) . . .

U(x
0
n1
, x
0
n
)
I
(x
n
)U(x
0
n
, )
_
,
(12.30)
where we dene the operator

U(t
2
, t
1
) U(t
2
, )U
1
(t
1
, ). (12.31)
The above operator is independent of the reference time . It is given by
(exercise):

U(t
2
, t
1
) = Te
i
R
t
2
t
1
dt

V
I
(t

)
, (12.32)
where t
2
> t
1
. It is easy to prove that

U(t
3
, t
2
)U(t
2
, t
1
) = U(t
3
, t
1
), (12.33)
with t
3
> t
1
. Notice that, for t
2
> x
0
> t
1
, we have
T
_
. . .

U(t
2
, t
1
)
I
(x) . . .
_
= T
_
. . .

U(t
2
, x
0
)
I
(x)

U(x
0
, t
1
) . . .
_
. (12.34)
157
12.3 The ground state of the interacting and
the free theory
Our goal is to develop a formalism for the evaluation of Greens functions
[ T (x
i
) . . . (x
n
) [) ,
in the interacting theory. As we have seen, their Fourier transforms will give
us, with the LSZ reduction formula, the probability amplitudes for physical
scattering processes. We remind that the points x

i
in the above need to have
time components which are slightly complex (Eq. 11.75).
From the discussion of the previous section we have that
[ T (x
i
) . . . (x
n
) [) =
[ T
_
U
1
(x
0
1
, )

U(x
0
1
, x
0
n
)
I
(x
i
) . . .
I
(x
n
)U(x
0
n
, )
_
[) (12.35)
In general, we anticipate that the ground state [) of the interacting
theory H = H
0
+H

is a dierent state from the vacuum state [0) of the free

theory H
0
. Let us assume that the Hamiltonian H has a spectrum [
n
) with
H[
n
) = E
n
[
n
) , (12.36)
and [
0
) [). Eigenstates of the full Hamiltonian form a complete set,
1 = [) [ +

n=0
[
n
)
n
[ . (12.37)
Let us now act with an operator to the vacuum state [0) of the free theory
H
0
,
e
iH(+T)(1i)
e
+iH
0
(+T)(1i)
[0) ,
choosing T a large time in the future and a very small dumbing parameter
0
+
. This is in anticipation of needing to evaluate Greens functions with
slightly complex times. We have,
e
iH(+T)(1i)
e
+iH
0
(+T)(1i)
[0) = e
iH(+T)(1i)
e
+i0(+T)(1i)
[0)
= e
iH(+T)(1i)
[0) (12.38)
158
We now use the completeness of the Hamiltonian eigenstates for the full
theory.
e
iH(+T)(1i)
e
+iH
0
(+T)(1i)
[0) = e
iH(+T)(1i)
1 [0)
= e
iH(+T)(1i)
_
[) [ +

n=0
[
n
)
n
[
_
[0)
= [) e
iE
0
(+T)E
0
(+T)
[ 0) +

n=0
[
n
) e
iEn(+T)En(+T)

n
[ 0) .
(12.39)
Assuming a hierarchy E
n
> E
0
for the rst energy level of the full Hamilto-
nian with respect to the ground energy, the factor e
En(+T)
vanishes faster
than e
E
0
(+T)
, as 0
+
. In this limit, we can write a simple relation be-
tween the ground state in the full theory and the vacuum in the free theory,
e
iH(+T)(1i)
e
+iH
0
(+T)(1i)
[0) = [) e
iE
0
(1i)(+T)
[ 0) . (12.40)
We now observe that the operator on the left side of the above equation
is nothing else than U
1
(T(1 i), (1 i)). We then have
[) = ^U
1
(T(1 i), (1 i)) [0) , (12.41)
[ = ^

0[ U (T(1 i), (1 i)) , (12.42)

where ^ is a normalization constant, which we determine from the normal-
ization condition [ ) = 1,
[^[
2
= [

U (T(1 i), T(1 i)) [) (12.43)
Substituting into Eq. 12.35, we have that
[ T (x
i
) . . . (x
n
) [) = [^[
2
0[ U(T, )
T
_
U
1
(x
0
1
, )

U(x
0
1
, x
0
n
)
I
(x
i
) . . .
I
(x
n
)U(x
0
n
, )
_
U
1
(T, ) [0) (12.44)
159
Given that the times T > x
0
i
> T, we can put all operators under the time-
ordering symbol. We then obtain, the nal result for the Greens function
[ T (x
i
) . . . (x
n
) [) = lim
0
T
0[ T
I
(x
1
) . . .
I
(x
n
)e
i
R
T
T
dt

V
I
(t

)
[0)
0[ Te
i
R
T
T
dt

V
I
(t

)
[0)
,
(12.45)
where the points x

i
are computed in slightly complex times x
0
i
(1 i).
Eq. 12.45 is an exact result, but it can serve as our basis for perturbation
theory. The exponential can be written as a function of the perturbation
Lagrangian with free elds. For our example interaction
e
i
R

dt

V
I
(t

)
= e
i
R
d
4
x

4!

I
(x)
4
, (12.46)
and it can be expanded as a Taylor series in .
12.4 Wicks theorem
In the previous section, we found that Greens functions can be cast in a form
suitable for applying the method of perturbation theory. For the example
Hamiltonian of Eq. 12.5, we can write the result,
[ T (x
1
) . . . (x
n
) [) = lim
0
T
0[ T
I
(x
1
) . . .
I
(x
n
)

n=0
_
i
4!
_
d
4
x
I
(x)
4
_
n
[0)
0[

n=0
_
i
4!
_
d
4
x
I
(x)
4
_
n
[0)
,
(12.47)
To evaluate the terms of the right hand side of Eq. 12.47, we need to be able to
compute expectation values in the free-eld theory of time-ordered products
of eld operators. This can be achieved by means of Wicks theorem.
We decompose the free scalar eld
I
into a term with a creation operator
and a term with an annihilation operator,

I
(x) =
+
(x) +

(x), (12.48)
160
with

(x) =
_
d
3

k
(2)
3
2
k
e
ikx
a(k), (12.49)
and

+
(x) =
_
d
3

k
(2)
3
2
k
e
+ikx
a

(k), (12.50)
The time-ordered product of two scalar elds is,
T
I
(x)
I
(y) = (x
0
y
0
)
I
(x)
I
(y) + (x

)
= (x
0
y
0
)
_

+
(x)
+
(y) +

(x)

(y) +
+
(x)

(y)
+

(x)
+
(y)
_
+ (x

)
= (x
0
y
0
)
_

+
(x)
+
(y) +

(x)

(y) +
+
(x)

(y)
+
+
(y)

(x) + [
+
(x),

(y)]
_
+ (x

)
=
_
(x
0
y
0
) + (y
0
x
0
)

+
(x)
+
(y) +

(x)

(y)
+
+
(x)

(y) +
+
(y)

(x)
_
+[

(x),
+
(y)] (x
0
y
0
) + [

(y),
+
(x)] (y
0
x
0
) (12.51)
The sum of the theta functions in the rst term is equal to one. We observe
that in the curly bracket all eld products appear with creation operators
preceding annihilation operators. The curly bracket is then just the normal
ordering of the product of the two eld operators
_
. . .
_
=:
I
(x)
I
(y) : (12.52)
The sum of the two terms in the last line is a known object, the Feynman-
St uckelberg propagator, which is not an operator but a cnumber. Using
161
that the relation
[a(k), a

(k

)] = (2)
3
2
k

(3)
(

), (12.53)
we nd that
[

(x),
+
(y)] =
_
d
3
k
(2)
3
2
k
e
ik(xy)
. (12.54)
Combining the two terms together we have
[

(x),
+
(y)]
_
x
0
y
0
_
+ [

(y),
+
(x)]
_
y
0
x
0
_
=
_
d
3
k
(2)
3
2
k
_
e
ik(xy)
(x
0
y
0
) + e
ik(yx)
(y
0
x
0
)

= 0[ T
I
(x)
I
(y) [0) . (12.55)
We can then write that,
T
I
(x)
I
(y) =:
I
(x)
I
(y) : +0[ T
I
(x)
I
(y) [0) . (12.56)
We have proved that the time-ordered product of two free-eld operators
is the normal-ordering of the same product plus a cfunction which is the
propagator of the two-elds.
This result generalizes easily to the time-ordered product of an arbitrary
number of eld operators. For three elds we have
T
I
(x
1
)
I
(x
2
)
I
(x
3
) = :
I
(x
1
)
I
(x
2
)(x
3
) :
+ 0[ T
I
(x
1
)
I
(x
2
) [0)
I
(x
3
)
+ 0[ T
I
(x
1
)
I
(x
3
) [0)
I
(x
2
)
+ 0[ T
I
(x
2
)
I
(x
3
) [0)
I
(x
1
) (12.57)
For four elds we have
T
I
(x
1
)
I
(x
2
)
I
(x
3
)(x
4
) = :
I
(x
1
)
I
(x
2
)(x
3
)
I
(x
4
) :
+ 0[ T
I
(x
1
)
I
(x
2
) [0) :
I
(x
3
)(x
4
) :
+ 0[ T
I
(x
1
)
I
(x
3
) [0) :
I
(x
2
)(x
4
) :
+ 0[ T
I
(x
1
)
I
(x
4
) [0) :
I
(x
2
)(x
3
) :
+ 0[ T
I
(x
1
)
I
(x
2
) [0) 0[ T
I
(x
3
)(x
4
) [0)
+ 0[ T
I
(x
1
)
I
(x
3
) [0) 0[ T
I
(x
2
)(x
4
) [0)
+ 0[ T
I
(x
1
)
I
(x
4
) [0) 0[ T
I
(x
2
)(x
3
) [0) .
(12.58)
162
In general, Wicks theorem states that
T
I
(x
1
) . . . (x
n
) =
= :
I
(x
1
) . . . (x
n
) + all contractions : (12.59)
where a contraction means to replace one or more pairs of elds with their
propagator. The theorem can be proved easily by induction.
Notice that the normal ordered products in the expressions produced with
Wicks theorem are vanishing when bracketed with the vacuum,
0[ :

I
(x
j
) : [0) = 0, (12.60)
given that creation operators are placed before annihilation operators in the
normal ordering and
0[ a

(k) = a(k) [0) = 0.

From Eqs 12.56-12.58 we derive the tautology
0[ T
I
(x)
I
(y) [0) = 0[ T
I
(x)
I
(y) [0) , (12.61)
and the more informative equations,
0[ T
I
(x
1
)
I
(x
2
)
I
(x
3
) [0) = 0, (12.62)
and
0[ T
I
(x
1
)
I
(x
2
)
I
(x
3
)(x
4
) [0) = 0[ T
I
(x
1
)
I
(x
2
) [0) 0[ T
I
(x
3
)(x
4
) [0)
+ 0[ T
I
(x
1
)
I
(x
3
) [0) 0[ T
I
(x
2
)(x
4
) [0)
+ 0[ T
I
(x
1
)
I
(x
4
) [0) 0[ T
I
(x
2
)(x
3
) [0) .
(12.63)
Such equations admit a graphical representation. Let us represent the
two-point correlation function with a straight line
0[ T
I
(x
1
)
I
(x
2
) [0) =
1
2
(12.64)
163
The four point function is represented as
0[ T
I
(x
1
)
I
(x
2
)
I
(x
3
)(x
4
) [0) =
1 4
3 2
contracted
=
+
1 4
3 2
+
1
1
4 4
2
2
3
3
(12.65)
It is then very easy to apply Wicks theorem pictorially, simply by drawing
all possible pairing of the points which appear in 0[ T
I
(x
1
)
I
(x
2
) . . . [0).
12.5 Feynman Diagrams for
4
theory
The pictorial application of Wicks theorem yields to the representation of
Greens functions in the full theory as a perturbative expansion in terms of
Feynman diagrams. As a concrete example, we shall consider the two-point
function in the full theory, through order O() in the coupling parameter .
From Eq. 12.47 we have
[ T (x
1
)(x
2
) [)
=
0[ T
I
(x
1
)
I
(x
2
)
_
1 +
(i)
4!
_
d
4
x
I
(x)
4
_
[0) +O(
2
)
0[ T
_
1 +
(i)
4!
_
d
4
x
I
(x)
4
_
[0) +O(
2
)
= 0[ T
I
(x
1
)
I
(x
2
) [0) +
(i)
4!
_
d
4
x0[ T
_

I
(x
1
)
I
(x
2
)
I
(x)
4
_
[0)

(i)
4!
0[ T
I
(x
1
)
I
(x
2
) [0)
_
d
4
x0[
I
(x)
4
[0) +O(
2
) (12.66)
Let us compute pictorially the second term in the expansion,
(i)
4!
_
d
4
x0[ T
_

I
(x
1
)
I
(x
2
)
I
(x)
4
_
[0)
164
x
1
2
Figure 12.1: The term 0[ T
I
(x
1
)
I
(x
2
)
I
(x)
4
[0) as the sum of all con-
tractions of the graph above
The corresponding Greens function is equal to the contractions of the graph
in Fig 12.5. We nd two dierent types of contractions which lead to two
dierent Feynman diagrams.
Diagram I: The two elds dened at the external points x
1
and x
2
get contracted with each other. The four elds dened at the internal
point x contract then among themselves.
1 2
x
This Feynman diagram is classied as disconnected, meaning that
the internal point x is not connected to any of the external points x
1
and x
2
.
This conguration occurs in more than one ways. The external elds
at x
1
and x
2
can be contracted in 1 way. Take now one of the four
elds at the internal point x. This eld can be contracted with any
of the remaining internal elds with 3 ways. Finally, the remaining
two internal elds can be contracted in 1 way only. Therefore, this
Feynman diagram appears
1 3 1 times.
Diagram II: The two external elds dened at x
1
and x
2
contract
with internal elds at x.
1 2
x
165
This is a connected diagram, meaning that internal points are con-
nected to external points by following the lines of the graph.
We can now compute the multiplicity of the Feynman diagram. There
are 4 ways that we can contract the eld at the rst external point with
one of the elds at the internal point x. The second external point can
then be contracted to another internal eld in three ways. Finally, two
remaining internal elds can be contracted with each other in one way.
Therefore, this Feynman diagram occurs
4 3 1 =
4!
2
times.
Now we examine the last term

(i)
4!
0[ T
I
(x
1
)
I
(x
2
) [0)
_
d
4
x0[
I
(x)
4
[0)
in Eq. 12.66, which is the contribution of the denominator in Eq. 12.47 to
the Taylor expansion. It is easy to convince ourselves that this term yields
the same Feynman diagram as the disconnected Diagram I. The two ex-
ternal elds are contracted together. The four internal elds are also con-
tracted among themselves. However, this terms carries an overall minus sign.
As a result, the disconnected Feynman diagram drops out from the result.
This observation holds at all orders in perturbation theory. Disconnected
Feynman diagrams do not contribute to the perturbative expansion of Greens
functions
1
. These factorize in the numerator of Eq. 12.47 and cancel exactly
against the denominator.
In summary, the perturbative expansion of the two-point Greens func-
tion through order O() can be represented very simply with the following
Feynman diagrams.
[ T (x
1
)(x
2
) [) =
1 2
+
1
2
1
x
2 +
...
(12.67)
We can obtain a concrete mathematical expression from the above diagrams
by following very simple rules, pioneered by Feynman:
1
The general proof of this statement is easier with the path integral formalism, and we
postpone it for QFTII.
166
1. We associate a propagator to each line
1
2
0[ T
I
(x
1
)
I
(x
2
) [0) (12.68)
2. We associate a factor of (i) and a space-time integration to each
vertex
x
(i)
_
d
4
x (12.69)
These rules give us a simple pictorial representation in terms of Feynman
diagrams of the perturbative series for any Greens function G(x
1
, x
2
, . . . , x
N
).
We draw all possible graphs with N external and with no more vertices than
the maximum order in the perturbative expansion
n
that we require. Then
we translate the Feynman diagrams into mathematical expressions using the
rules above. One diculty in this procedure is to determine the combinato-
rial rational factor that multiplies the diagram, as for example the factor 1/2
in the second diagram of Eq. 12.67. This number can always be obtained
following the procedure in Eq. 12.67, which is a rather brute force method.
For everything practical, this method is sucient and for more complicated
cased we can easily program it in a computer code. The combinatorial fac-
tor, also known as symmetry factor, can be determined cleverly as the inverse
of the independent exchange symmetry operations of the Feynman diagram
(exchanging vertices and propagators) which leave it intact. This is a nice
method for everyone who is condent in spotting all symmetries without
double counting equivalent ones. It is, however, prone to human errors.
12.6 Feynman rules in momentum space
The LSZ reduction formula expresses scattering amplitudes as truncated
Greens functions in momentum space. We can develop simple Feynman
rules for computing them directly.
We start with the Fourier transform of the propagator from the origin to
a point x.

G(p) =
_
d
4
xe
ipx
[ T (x)(0) [) (12.70)
167
At leading order in perturbation theory we have,

G(p) =
_
d
4
xe
ipx
0[ T
I
(x)
I
(0) [0) +O()
=
_
d
4
xe
ipx
_
d
4
k
(2)
4
i
k
2
m
2
+ i
e
ikx
+O()
=
_
d
4
k
(2)
4
i
k
2
m
2
+ i
_
d
4
xe
i(pk)x
+O()
=
_
d
4
k
(2)
4
i
k
2
m
2
+ i
(2)
4

(4)
(p k) +O()
;

G(p) =
i
p
2
m
2
+ i
+O() (12.71)
It is interesting to look at the transition amplitude for a particle with
momentum p to a particle with momentum p

. The corresponding Greens

function is

G(p; p

) =
_
d
4
ye
+ip

y
_
d
4
xe
ipx
[ T (x)(y) [) , (12.72)
At leading order in perturbation theory gives (exercise)

G(p; p

) = (2)
4

(4)
(p p

)
i
p
2
m
2
+ i
+O(). (12.73)
Notice the delta function, which simply imposes momentum conservation.
Execrise: What is the corresponding truncated Greens function?
A more complicated case is the scattering of four particles. The Greens
function we require is
G(x
1
, x
2
, x
3
, x
4
) =
+
3
4
x
2
1
3
+
4
2
1
4
2
1
3
4
3 2
1
+
1
2
3
4
Ga
Gb
Gc
Gd Ge
x y
+
1
2
+. . . (12.74)
168
Let us now compute the Greens function in momentum space,

G(p
3
, p
4
; p
1
, p
2
) =
_
d
4
x
1
e
i(x
1
p
1
+x
2
p
2
x
3
p
3
x
4
p
4
)
G(x
1
, x
2
, x
3
, x
4
), (12.75)
where we consider p
1
, p
2
in the initial state and p
3
, p
4
in the nal state. We
write,

G(p
3
, p
4
; p
1
, p
2
) = G
a
+ G
b
+ G
c
+ G
d
+ G
e
+ . . . , (12.76)
where G
a,b,c
are the contributions of the rst three diagrams and G
d
, G
e
are
the contributions of the fourth and fth diagram at order O() and O(
2
).
The rst three diagrams give,
G
a
+ G
b
+ G
c
= (2)
4

4
(p
1
p
4
)
i
p
2
1
m
2
+ i
(2)
4

4
(p
2
p
3
)
i
p
2
2
m
2
+ i
+ (2)
4

4
(p
1
p
3
)
i
p
2
1
m
2
+ i
(2)
4

4
(p
2
p
4
)
i
p
2
2
m
2
+ i
+ (2)
4

4
(p
1
+ p
2
)
i
p
2
1
m
2
+ i
(2)
4

4
(p
3
+ p
4
)
i
p
2
3
m
2
+ i
+ (12.77)
The fourth diagram is converted into a mathematical expression following
again the Feynman rules of the previous section. We have
G
d
=
_
d
4
x
1
d
4
x
2
d
4
x
3
d
4
x
4
e
i(x
1
p
1
+x
2
p
2
x
3
p
3
x
4
p
4
)
(i)
_
d
4
x ( vertex )
0[ T
I
(x
1
)
I
(x) [0) ( contraction of x
1
and x )
0[ T
I
(x
2
)
I
(x) [0) ( contraction of x
2
and x )
0[ T
I
(x
3
)
I
(x) [0) ( contraction of x
3
and x )
0[ T
I
(x
4
)
I
(x) [0) ( contraction of x
4
and x ).(12.78)
Substituting the expression for 0[ T
I
(x
3
)
I
(x) [0) Performing the x and
x
i
integrations we nd,
G
d
= (i)
i
p
2
1
m
2
+ i
i
p
2
2
m
2
+ i
i
p
2
3
m
2
+ i
i
p
2
4
m
2
+ i
(2)
4

4
(p
1
+p
2
p
3
p
4
).
(12.79)
169
The diagram G
e
is a loop diagram. Let us write the corresponding
expression, using the Feynman rules in position space. We have
G
e
=
_
d
4
x
1
d
4
x
2
d
4
x
3
d
4
x
4
e
i(x
1
p
1
+x
2
p
2
x
3
p
3
x
4
p
4
)
(i)
_
d
4
xd
4
y ( vertices )
0[ T
I
(x
1
)
I
(x) [0) ( contraction of x
1
and x )
0[ T
I
(x
2
)
I
(x) [0) ( contraction of x
2
and x )
0[ T
I
(x
3
)
I
(y) [0) ( contraction of x
3
and y )
0[ T
I
(x
4
)
I
(y) [0) ( contraction of x
4
and y )
0[ T
I
(x)
I
(y) [0)
2
( two contractions of x and y )
(12.80)
Performing all xspace integrations we nd,
G
e
=
i
p
2
1
m
2
+ i
i
p
2
2
m
2
+ i
i
p
2
3
m
2
+ i
i
p
2
4
m
2
+ i
(2)
4

4
(p
1
+ p
2
p
3
p
4
)
(i)
2

_
+

d
4
k
(2)
4
i
k
2
m
2
+ i
i
(k + p
1
+ p
2
)
2
m
2
+ i
.
(12.81)
Observe our nal expressions for Ga, b, c, Gd and Ge. We can make easy
rules (Feynman rules) to produce them from the Feynman diagrams.
In every vertex we pick up a factor
(i),
With each propagator comes a factor
i
p
2
m
2
+ i
.
At each vertex, we have delta functions guaranteeing that momentum
is conserved. As a result, for all particles that are connected with each
170
other there is an overal factor
(2)
4

(4)
_

in
p
in

out
p
out
_
,
which forces the sum of incoming momenta in a connected diagram to
be equal to the sum of the outcoming momenta.
The loop in diagram G
e
introduces an integration over a loop momen-
tum and a measure
_
+

d
4
k
(2)
4
.
12.7 Truncated Greens functions in pertur-
bation theory
Figure 12.2: Feynman diagrams which are correction to external lines and
do not contribute to physical scattering amplitudes.
We are interested in computing amplitudes for physical scattering pro-
cesses, as given by the LSZ formula. For the scattering of two particles
with momenta p
1
, p
2
to two particles with momenta p
3
, p
4
we then need a
171
truncated Greens function
p
3
, p
4
[ p
1
, p
2
) =

Z
2

G
trunc
(p
3
, p
4
; p
1
, p
2
)

p
2
i
=m
2
phys
(12.82)
compute for squared external momenta equal to the physical mass of the
particles. The truncated Greens function is dened as

G
trunc
(p
3
, p
4
; p
1
, p
2
) =

G(p
3
, p
4
; p
1
, p
2
)

G(p
1
)

G(p
2
)

G(p
3
)

G(p
4
)
, (12.83)
where the Greens function in momentum space is divided with the propa-
gator of each of the particles in the initial and nal states.
The Kahlen-Lehmann representation for the pole of the propagator is

G(p) =
i

Z
p
2
m
2
phys
+ regular terms (12.84)
In the previous section we computed the perturbative result,

G(p) =
i
p
2
m
2
+O(). (12.85)
At leading order in perturbation theory, we can then identify the physical
mass of a particle m
phys
with the mass parameter m of our Lagrangian, and
determine the normalization

Z

= 1.
We then nd that the diagrams G
a
, G
b
, G
c
which do not connect all eter-
nal particles with each other do not contribute to the scattering amplitude
p
3
, p
4
[ p
1
, p
2
). We nd for example that, at leading order in ,
G
a

G(p
1
)

G(p
2
)

G(p
3
)

G(p
4
)
(p
2
3
m
2
)(p
2
4
m
2
)(p
4
p
1
)(p
3
p
2
)

p
2
i
=m
2
= 0.
(12.86)
The diagrams G
d
and G
e
connect all external particles and have poles
i
p
2
i
m
2
for each one of them. These contribute to the scattering amplitude.
We can perform the division with the external propagators as it is de-
manded by the LSZ formula we need three more rules in our set of Feynman
rules.
172
Lines starting from an external point contribute a factor 1 (and not
i/(p
2
m
2
)).
We must be, though, a bit more careful. Eq. 12.85 has perturbative cor-
rections of order . We need to divide our Feynman diagrams with the full
perturbative expansion of the propagator and not just the leading order con-
tribution. We can account for this division entirely with one more clever
trick.
Through away all Feynman diagrams which correct external lines.
For example, the diagrams of Fig. 12.2 should all be disregarded when com-
puting truncated Greens functions for physical scattering amplitudes.
173
Chapter 13
Loop Integrals
In order to compute the perturbative expansion of any Greens function,
we must perform unrestricted integrations over the momenta of particles
circulating in loops of Feynman diagrams. The exact evaluation of loop
integrals is a dicult and often prohibitive task. Many loop integrals are
divergent! This is a very unpleasant surprise if we aim towards a realistic
description of physical phenomena with nite probabilities. The problem
of divergences is a very serious one, and it jeopardizes the mathematical
foundation of the perturbative method.
13.1 The simplest loop integral. Wick rota-
tion
Let us consider the simplest loop-integral in eld theory,
I
_
d
4
k
(2)
4
1
k
2
m
2
+ i
(13.1)
This is a formidable integral, given that it requires four integrations for each
space-time dimension. We notice that only the combination k
2
= k
2
0

k
2
appears in the integrand. The integral would be easily solvable had we had
a Euclidean metric for k
2
E
= k
2
0
+

k
2
, by using spherical coordinates in four
dimensions. With a Minkowski metric, we ought to treat the time integration
specially. We write
I = lim
0
+
_
+

d
3

k
(2)
4
_
+

dk
0
1
(k
0

k
+ i) (k
0
+
k
i)
, (13.2)
174
with

k
=
_

k
2
+ m
2
. (13.3)
The integrand has two poles, at k
0
=
k
i and k
0
=
k
+ i, which
are found in the lower-right and upper-left quarter planes in the (k
0
, k
0
)
plane We can perform the so called Wick rotation, rotating the integration
Figure 13.1: Wicks rotation: The integrand has no poles on the upper-right
and lower-left k
0
complex plane. We can rotate the integration axis by 90
o
,
and integrate along the imaginary axis k
o
[i, +i].
axis by 90 degrees and integrating over the imaginary axis,
I =
_
+

d
3

k
(2)
4
_
+i
i
dk
0
1
k
2
0

k
2
m
2
+ i
. (13.4)
Setting
k
0
= ik

0
,
the integral becomes
I = i
_
+

d
3

k
(2)
4
_
+i
i
dk

0
1
(k

0
)
2

k
2
m
2
+ i
. (13.5)
175
Notice that after Wicks rotation we can consider the four-vector k
E
(k

0
,

k)
in Euclidean space, and express k
E
in four-dimensional spherical coordinates:
dk
E
= d[k[ [k[
3
d
4
. (13.6)
The integral then becomes
I = i
4
_

0
d[k[
[k[
3
[k[
2
+ m
2
i
. (13.7)
Notice that the integral diverges as [k[ , which is an embarrassment.
Innities were historically the biggest puzzle in the development of quan-
tum eld theory. But, we now know what to do with them for many La-
grangian systems which are realistic descriptions of nature. We will learn
later that we can sweep innities under the carpet, exploiting certain free-
doms that we have. We are allowed to absorb innities in the the normal-
ization constants Z

of the LSZ formula, or in the denition of physical mass

and coupling parameters in terms of the corresponding parameters appearing
in the Lagrangian. These are very few freedoms, and it is amazing that
they suce to solve the problem.
The method of sweeping innities under the carpet is called renormal-
ization. A prerequisite to it is to quantify the innities with the help of a
regulator. We shall introduce a parameter which renders the integrations
well dened and nite except at a certain limit. For example, we can restrict
the integration in Eq. 13.7 below a certain cut-o value for [k[,
I() i
4
_

0
d[k[
[k[
3
[k[
2
+ m
2
i
= i
4
_

2
2
+ . . .
_
(13.8)
The regularized integral I() is equal to the original integral I in the limit
of an innite ,
lim

I() = I. (13.9)
What is the benet of introducing a regularization method? It will give us
the possibility to compute the divergent parts of all loop-integrals that enter
the evaluation of Smatrix elements for values of the regulator where they
are all well-dened. Then, we can perform renormalization and express
the parameters of the Lagrangian in terms of physical mass and coupling
parameters. Then we shall take the limit of the regulator which corresponds
to the original loop integrals. It will turn out that for many eld theories the
nal renormalized S-matrix elements are nite.
176
13.2 Dimensional Regularization
A method to regularize loop integrals is not unique. A well chosen regular-
ization procedure can be very benecial for practical computations, since the
diculty in carrying out the integrations may vary signicantly. What is
more important, some regularization methods may violate some of the sym-
metries of the theory. For example, a cut-o regularization violates Lorentz
invariance. When symmetries are broken due to the regularization method it
may be required to pay additional eorts in order to relate S-matrix elements
to physical observables.
The most elegant regularization method known to date was developed
by tHooft and Veltman in the 70s. It is called dimensional regularization.
It has revolutionized the evaluation of loop integrals for its simplicity. It is
also known to preserve most of the symmetries which are found in physical
Lagrangian systems.
In dimensional regularization, the regulator is the number of space-time
dimensions d. We shall explain how this work, by revisiting the simplest
example of a loop integral in Eq. 13.7. Let us now compute this integral in
an arbitrary number of dimensions,
I
d
=
_
d
d
k
(2)
d
1
k
2
m
2
+ i
, (13.10)
which after Wicks rotation becomes
I
d
= i
_
d
d
k
(2)
d
1
k
2
m
2
+ i
, (13.11)
and
k
2
= k
2
0
+

k
2
. (13.12)
In spherical coordinates,
I
d
=
i
d
(2)
d
_

0
dk
k
d1
k
2
m
2
+ i
. (13.13)
For innitely large loop momenta it behaves as,
I
d

i
d
(2)
d
_

dkk
d3
. (13.14)
177
We observe that the integral has a nite ultraviolet limit (k ) for val-
ues of the dimension d < 3. In dimensional regularization we compute the
integral for a generic value of the dimension d which is allowed for it in in
order to be well-dened. At the end of our computation, we shall perform
an analytic continuation to a physical dimension value d = 4.
13.2.1 Angular Integrations
For d = 2, the solid angle is
d
2
= d,
for d = 3, we have
d
3
= sin dd,
and so on. We are used to dening solid angles for an integer number of
dimensions. What is the value of the solid angle
d
for an arbitrary real
valued dimension d? Our generalization of the dimension from integer to
real values for the solid angle is mathematically similar to the generalizing
of the factorial to the complex-valued function,
n! (z),
and relies on nding a dening recurrence relation.
Let us take a ddimensional vector

k
(d)
=
_

k
(d1)
, k
d
_
, (13.15)
where k
d
is the component corresponding to the d
th
dimension. We shall
assume that we know how to express the

k
(d1)
in terms of (d 1) polar
coordinates, i.e. a radial r
d1
coordinate and (d 2) angular coordinates.
Then
_
+

dk
1
. . . dk
d
=
_
+

dk
d
_

0
dr
(d1)
r
d2
(d1)
_
d
d1
. (13.16)
The ddimensional vector is now expressed in cylindrical coordinates,

k
(d)
=
_
k
d1
u
(d1)
(
1
,
2
, . . .
d2
) , k
d
_
, (13.17)
where u
(d1)
is a unit vector. Now we change variables to polar coordinates
in ddimensions by performing the transformation,
r
(d1)
= l sin
d1
k
d
= l cos
d1
. (13.18)
178
This gives for the integration measure
_
d
d
k,
_
+

dk
d
_

0
dr
(d1)
r
d2
(d1)
_
d
d1
=
_

0
dll
d1
_

0
d
d1
sin
d2

d1
_
d
d1
.
(13.19)
We have then arrived to the recurrence relation,
_
d
d
=
_

0
d
d1
sin
d2

d1
_
d
d1
. (13.20)
Recall that for d = 2,
_
d
2
=
_
2
0
d
1
. (13.21)
Using this as a boundary condition for Eq. 13.20, we obtain the solid-angle
in arbitrary integer dimensions,
_
d
d
=
_

0
d
d1
sin
d2

d1
. . .
_

0
d
2
sin
2
_
2
0
d
1
. (13.22)
All integrals can be performed easily using that,
_

0
d sin
x
=
_
1
1
d(cos )(1 cos )
x1
2
=
_
1
0
daa
1
2
1
(1 a)
x+1
2
1
(a = cos
2
)
;
_

0
d sin
x
= B
_
1
2
,
1 + x
2
_
=

_
1
2
_

_
x+1
2
_

_
x+2
2
_ . (13.23)
The the solid angle in ddimensions is

d
= 2

d
2

_
d
2
_. (13.24)
On the rhs we have a function which is dened, as we shall discuss imme-
diately, to arbitrary complex values of the dimension parameter d, with the
exception of non-positive integers. We thus have derived a generalization of
the solid angle to an arbitrary such value of d.
179
A quicker derivation
1
of Eq. 13.24 proceeds as follows. Consider the
exponential integral,
_

dxe
x
2
=

. (13.25)
For d such integrals we have,

d
2
=
_
d
d
xe

P
d
i=1
x
2
i
=
_
d
d
_

0
dxx
d1
e
x
2
. (13.26)
Setting t = x
2
, we nd

d
2
=
1
2

d
_

0
dtt
d
2
1
e
t
=
1
2

_
d
2
_
. (13.27)
13.2.2 Properties of the Gamma function
In the above, the Beta function is dened as
B(x, y) =
_
2
0
1dxi
x1
(1 )
y1
. (13.28)
The function is dened through the integral representation,
(z)
_

0
dxx
z1
e
x
. (13.29)
By using integration by parts, we can easily show that
(z + 1) = z(z), (13.30)
which is the same recursion relation as for the factorial of integers. For z = n
a positive integer, we have
(n) = (n 1)! (13.31)
The function (z) is analytic except for z = 0, 1, 2, . . .. From the
integral representation we conclude that the function is convergent for
Re z > 0. But, the recursion relation of Eq. 13.30 denes it for all complex
values of z, with the exception of non-positive integers.
1
shown in class
180
A useful value is

_
1
2
_
=

. (13.32)
We can also prove the identity,
(x)
_
1
2
_
= 2
x1

_
x
2
_

_
x + 1
2
_
. (13.33)
We shall often need to expand a Gamma function around an integer value
for its argument. We have
(1 +) =
_

0
x

e
x
=
_

0
e
lnx
e
x
=
_

0
_
1 + ln x +

2
2
ln
2
x + . . .
_
e
x
=
;(1 +) = 1 +
2
_

2
2
+

2
12
_
+O(
3
). (13.34)
The Euler-gamma constant is dened as

_

0
dxlog(x) 0.544 . . . (13.35)
13.2.3 Radial Integrations
We now continue with the remaining integration of Eq. 13.14 over the mag-
nitude of the Euclidean four-momentum. We have
I
d
=
i
2

d
(2)
d
_
dk
k
d2
k
2
+ m
2
i
. (13.36)
We perform the change of variables
k
2

x
1 x
_
m
2
i
_
, (13.37)
which yields
I
d
=
i
2

d
(2)
d
_
m
2
i
_d
2
1
_
1
0
dxx
d
2
1
(1 x)

d
2
(13.38)
181
which yields the nal result,
_
d
d
k
i
d
2
1
k
2
m
2
+ i
=
_
1
d
2
_
_
m
2
i
_d
2
1
(13.39)
Notice that the integral is divergent for d = 2, 3, 4, . . ., but it is well-dened
for any other value of d.
Exercise: Prove that
_
d
d
k
i
d
2
1
[k
2
m
2
+ i]

= (1)

_

d
2
_
()
_
m
2
i
_d
2

. (13.40)
An interesting case is when m
2
= 0. In dimensional regularization, where
the dimension parameter is considered non-integer, we have that
lim
m0
m
d
= 0
d
= 0. (13.41)
We then have that integrals with no mass scales vanish, and in our examplary
case,
_
d
d
k
i
d
2
1
[k
2
+ i]

= 0. (13.42)
13.3 Feynman Parameters
Consider a more complicated integral. For example,
I
2
=
_
d
d
k
i
d
2
1
[k
2
m
2
+ i] [(k + p)
2
m
2
+ i]
. (13.43)
The integrand has now four poles for
k
0
=
__

k
2
+ m
2
i
_
,
_
_
(

k + p)
2
+ m
2
i
_
. (13.44)
It is now more dicult to perform the integration over k
0
, as well as the
angular integrations since the denominator depends explicilty on angles,
(k + p)
2
m
2
= k
2
+ p
2
m
2
+ 2k
0
p
0
2[

k[[ p[ cos . (13.45)

The method of Feynman parameters allows to integrate out the loop
momentum k easily, applying Wicks rotation and integrating over angles in
182
exactly the same manner as for the simplest integral of Eq. 13.40. The price
that we have to pay is to introduce new integration variables over Feynman
parameters. It is easy to prove that,
1
AB
=
_
1
0
dx
1
[xA + (1 x)B]
2
. (13.46)
We can use such identity in order to combine denominators in loop intergrals
under one term. For example, the integral of Eq. 13.43 can be written as
I
2
=
_
d
d
k
i
d
2
1
[x((k + p)
2
m
2
) + (1 x) (k
2
m
2
) + i]
2
=
_
d
d
k
i
d
2
1
[k
2
m
2
+ 2xk p + xp
2
+ i]
2
=
_
d
d
k
i
d
2
1
[(k + px)
2
m
2
+ x(1 x)p
2
+ i]
2
(13.47)
We perform a shift in the integration variable k, usually called loop mo-
mentum, dening
K

= k

+ xp

. (13.48)
We then have,
I
2
=
_
1
0
dx
_
d
d
k
i
d
2
1
[K
2
(m
2
x(1 x)p
2
i)]
2
(13.49)
The integral over K

can be solved using the general result of Eq. 13.40,

I
2
=
_
2
d
2
__
1
0
dx
_
m
2
x(1 x)p
2
i
_d
2
2
(13.50)
The above integral is only one-dimensional. We still have the task of
performing the task of integrating over Feynman parameters. Feynman pa-
rameter integrals are generalized hypergeometric functions, which are only
partially understood in the mathematical science. Said dierently, the math-
ematics of hypergeometric functions is far from sucient in order to tackle
the Feynman integrals that appear in the perturbative expansion of Greens
functions. This limits computations to simple cases. Nevertheless, even with
the mathematical understanding of loop integrations being in its infancy, we
183
can still derive rather precise theoretical predictions for S-matrix elements of
interesting scattering processes.
For physics predictions, we are interested in the value of loop integrals as
a Laurent expansions around the physical value of the dimension parameter,
d = 4. It is customary to write,
d = 4 2, (13.51)
and expand around = 0. The () sign is because we usually think of
the dimension parameter as being smaller than four, in order to make an
integral convergent in the UV limit [k[ . The factor of 2 in front of
is convenient due to that typically the combination
d
2
emerges as result of
loop-integrations. The integral of Eq. 13.50 becomes
I
2
= ()
_
1
0
dx
_
m
2
x(1 x)p
2
i
_

+
_
1
0
dxlog
_
m
2
x(1 x)p
2
i

+O() . (13.52)
Exercise: Solve the above integral, assuming that p
2
< 4m
2
. For p
2
> 4m
2
the integral develops an imaginary part. Compute it, by performing an
analytic coninuation of the result for p
2
< 4m
2
.
The procedure of combining denominators in a single term using Feyn-
man parameters and making complete squares of the loop momenta can be
performed in general. For an arbitrary number of denominators in a loop-
integral we can use the formula,
1
A

1
1
. . . A
n
n
=
(
1
+ . . .
n
)

(
i
)
_
1
0
dx
1
. . . dx
n

_
1

i
x
i
_
x

1
1
1
. . . x
n1
n
[

x
i
A
i
]
P

i
(13.53)
Proof: Let us perform the change of variables
x
i
=
a
i
1 + a
n
(13.54)
in the rhs of Eq. 13.53. The delta function becomes,

_
1
n

i=1
a
i
1 + a
n
_
=
_
1

n1
i=1
a
i
1 + a
n
_
= (1+a
n
)
_
1
n1

i=1
a
i
_
. (13.55)
184
The parameters a
i
are constrained to be in the interval [0, 1], due to the
function for i = 1 . . . n 1, while the parameter a
n
ranges in between
a
n
= 0(x
n
= 0) and a
n
= (x
n
= ). The Jacobian of the transformation
is
dx
i
=
da
i
1 + a
n
, i ,= n, (13.56)
and
dx
n
=
da
n
(1 + a
n
)
2
. (13.57)
All factors conspire in Eq. 13.53, so that the integrand is almost the same
as the original, replacing x
i
with a
i
. The only dierence is that the delta
fucntion does not contain a
n
anymore, and that the range of this variable is
from 0 to innity. Explicitly,
rhs of Eq. (13.53) =
(
1
+ . . .
n
)

(
i
)

_

0
da
n
_
1
0
da
1
. . . da
n1

_
1
n1

i=1
a
i
_
a

1
1
1
. . . a
n1
n
[

a
i
A
i
]
P

i
(13.58)
We now change once more variables to
a
n
=
x
1 x

n1
i=1
a
i
A
i
A
n
, (13.59)
and perform the integration over x, which ranges from 0 to 1. The result is
rhs of Eq. (13.53) =
1
A
n
n

(
1
+ . . .
n1
)
(
1
) . . . (
n1
)
_
1
0
dx
1
. . . dx
n

_
1

i
x
i
_
x

1
1
1
. . . x
n1
n
[

x
i
A
i
]
P

i
(13.60)
We have now factorized the term 1/A
n
n
out of an integral, which is now
cast in the same form as the original but with n 1 Feynman parameters.
Obviously, we can repeat the same procedure as many times as needed in
order to factorize all
1
A

i
i
terms, as in the lhs of Eq. 13.53.
185
Chapter 14
Quantum Electrodynamics
One of the most amazing successes of Quantum Field theory, is the preci-
sion in which it describes the interaction of radiation and matter. The rules
governing these interactions are very simple, and are given by the theory of
Quantum Electrodynamics (QED). What is also astonishing, emerges natu-
rally by combining two very simple principles: symmetry and localilty. The
QED Lagrangian is symmetric under phase-redenitions (U(1) transforma-
tions) of the electron eld which are local, i.e. we can choose the symmetry
trnsformation parameters dierently at each space-time point.
Local symmetry transformations govern all quantum eld theories, QED,
QCD and the Standard Model of electroweak interactions, which describe the
three forces of nature other than gravity. These theories describe physical
phenomena extremelly well within the experimentally accessible accuracy.
14.1 Gauge invariance
We consider a free electron eld:
/ =

(x) (i, m) (x). (14.1)
As we have discussed already, this Lagrangian is invariant under a U(1)
transformation,
(x)

(x) = exp(ie)(x), (14.2)

if the phase is global, i.e. it is chosen to be the same at avery point in
space-time
_

x
= 0
_
.
186
Let us now take a bold step and ask whether we can have a a dierent
phase at every space-time point x

,
U(x) = exp (ie(x)) . (14.3)
The Lagrangian is no longer invariant,

,

, +

e
ie
_
,e
ie

. (14.4)
The problem is that the space-time derivative does not transform simply
under the local U(1) transformation any more,
, e
ie
, + something else .
If we insist on locality, we shall need to modify the derivative so that it
transforms more conveniently. We will look for a new derivative which, under
a local U(1) transformation transforms as:
,D U(x),D. (14.5)
The simplest modication we can think of, is to add to the space-time deriva-
tive a space-time function,

+ function

(14.6)
which transforms in the same way as the space-time derivative under Lorentz
transformations, i.e. it transforms as vector. A function of space-time, in
eld theory, is nothing else but a eld. We are then proposing to modify the
denition of a derivative by adding to it a vector eld. We write a covarian
derivative,
D

ieA

(x), (14.7)
where the ie factor is conventional.
The vector eld A

(x) must have a very specic trasformation which we

can nd, in order for the modied derivative to transform simply. We require
that
D

(x) D

= U(x)D

;
_

ieA

_
(U(x)) = U(x) (

ieA

)
; U(x)

+ [

U(x)] ieA

U(x) = U(x)

ieA

U(x)
; A

= A

i
g
U
1
(x)

U(x) (14.8)
187
This is an astonishing result. We nd that the vector eld transforms as,
A

= A

+ g

, (14.9)
which is a gauge transformation, the transformation of the photon eld. In
other words, if we would like that a Lagrangian of electrons is invariant
under local U(1) transformations, then the same Lagrangian must describe
a photon eld.
The covariant derivative transforms as:
D

ieA

ie
_
A

i
g
U
1
U
_
=

ieA

U
1
(

U)
=

ieA

+ U
_

U
1
_
= U(x) (

ieA

) U
1
(x) (14.10)
Therefore:
D

= U(x)D

U
1
(x) (14.11)
We can now replace the free Lagrangian of the spin-1/2 eld with a new
Lagrangian which is symmetric locally,
/ =

[i,D m]

U
1
U [i,D m] U
1
U

[i,D m] .
If A

is a physical eld, we need to introduce a kinetic term in the La-

grangian for it. We will insist on constructing a fully gauge invariant La-
grangian. To this purpose, we can use the covariant derivative as a building
block. Consider the gauge transformation of the product of two covariant
derivatives:
D

= UD

U
1
UD

U
1
= UD

U
1
.
This is not a gauge invariant object. Now look at the commutator:
[D

, D

]
_
D

, D

= U [D

, D

] U
1
(14.12)
188
This is gauge invariant. To convince ourselves we write the commutator
explicitly:
[D

, D

] = (

ieA

) (

ieA

) [ ]
=

ie (

) ieA

ieA

+ (ie)
2
A

[ ]
= ie [

] . (14.13)
Inserting Eq. 14.13 into Eq. 14.12, we nd that the commutator of covariant
derivatives (in the abelian case) is gauge invariant. We have also found that
it is proportional to the eld strength tensor of the gauge (photon) eld:
F

=
i
e
[D

, D

] =

(14.14)
We now have invariant terms for a Lagrangian with an electron and a
photon eld. The classical Lagrangian for QED reads
/ =

(i,D m)
1
4
F

. (14.15)
This Lagrangian accounts for the majority of the phenomena that we expe-
rience in nature. It is a beautiful synthesis of locality and symmetry!
Exercise: Find the Noether current and conserved charge due to the
invariance under the U(1) gauge transformation of the QED Lagrangian.
14.2 Perturbative QED
Let us take the QED Lagrangian of Eq. 14.15 and substitute the expression
for the covariant derivative. We have,
/ =

[i, m] (free electron eld)
+
1
4
F

( free photon eld)

+ e

,A (electron-photon interaction)
(14.16)
The last term, which couples the electron and fermion elds, dierentiates
the QED Lagrangian from the Lagrangian of two free electron and photon
elds. Can we solve the energy eigenstates of the QED Hamiltonian? No,
189
unless we resort to perturbation theory. The strength of the photon-electron
interaction is very small. We can consider the last term a small perturbation,
and expand probability amplitudes around e = 0.
For e = 0, we should recover the results from the quantization of the free
photon and free electron elds. We then encounter the same problems as
when we quantized the free electromagnetic eld, which required to x the
gauge for the eld A

. We then add to the QED Lagrangian a term,

/
gaugefix
=
1
2
(

)
2
. (14.17)
In this way, we can at least be sure that the theory is correctly dened for
e = 0. It is not obvious that this modication will solve the problem of
obtaining a consistently quantized theory at higher orders in perturbation
theory. It actually does! But the justication of this statement will be a
very important topic of study in QFTII. As a primitive diagnostic, we shall
consider an arbitrary parameter in our calculations. We will then check
that higher order corrections do not modify it.
We then have,
/
QED
= /
e
+/

+/
int
, (14.18)
with
/
e
=

[i, m] , (14.19)
/

=
1
4
F

2
(

)
2
, (14.20)
and
/
int
= e

,A. (14.21)
We can develop a similar formalism for perturbation theory as in the case
of a scalar eld theory. This leads to the derivation of Feynman rules for
Smatrix elements for QED transition amplitudes. Essentially, we can pic-
ture QED processes as photons and electrons propagating in between interac-
tions, where a photon is absobed or emmitted by electrons. For every photon
absorption or emission we need a factor of e, which is a number represent-
ing the strenght of the interaction. We then have to include all possibilities
for a scattering process to happen, with so many interaction vertices as the
maximum order in our perturbative series. The rules are:
190
For the propagation of a photon with momentum p, assign the factor
i
p
2
+ i
_
g

+ (1 )
p

p
2
+ i
_
. (14.22)
For the propagation of an electron, assign the factor
i
,p m + i
. (14.23)
For photon absorption or emission, assign a factor
ie

(14.24)
In addition,
For each loop in a Feynman diagram, we perform an integration over
the loop momentum
_
d
d
k
(2)
d
, (14.25)
where d is the dimension.
For each fermion loop we multiply with a factor
(1) (14.26)
We divide each Feynman diagram with a symmetry factor, account-
ing for the equivalent eld permutations.
The LSZ reduction formula is slightly modied for fermions and photons,
where the constants

Z
1/2
are replaced by

Z
1/2
times a spin factor which is
dierent from one. For truncated external lines, we have the rules
multiply with a factor
u(p),
for each outgoing fermion,
191
a factor
v(p),
for each outgoing anti-fermion,
a factor
u(p),
for each incoming fermion,
a factor
v(p),
for each incoming anti-fermion,
and a factor

(p)
for each external photon.
14.3 Dimensional regularization for QED
QED is plagued by divergent loop integrals. We shall use dimensional reg-
ularization as our regularization method. In this Section, we discuss some
new features that arise in dimensional regularization in gauge theories.
First, we consider the QED action integral in arbitrary d dimensions.
S =
_
d
d
x/
QED
. (14.27)
The action has no mass dimensions,
[S] =
_
d
d
x/
QED

= [mass]
0
. (14.28)
Since the space-time coordinates have an inverse mass dimension we nd that
all terms in the Lagrangian must have mass dimensionality d,
[/
QED
] = [mass]
d
. (14.29)
From the kinetic terms of the photon and fermion elds we can deternime
that their mass dimensionalities ought to be,
_
(

)
2

= [mass]
d
;[A

] = [mass]
d
2
1
, (14.30)
192
and
_

= [mass]
d
;[] = [mass]
d1
2
. (14.31)
From the gauge xing term, we nd that the parameter
[] = [mass]
0
, (14.32)
is dimensionless. Finally, the interaction term gives,
_
e

,A

= [mass]
d
;[e] = [mass]
4d
2
. (14.33)
The coupling constant e is a dimensionful parameter in dimensions other
than four. In dimensional regularization, we take
d = 4 2,
which translates into
[e] = [mass]

. (14.34)
We can then substitute in the Lagrangian and the Feynman rule for photon
absorption or emission,
e e

, (14.35)
introducing an arbitrary mass scale in the Lagrangian. Naively, given that
we shall take the 0 after renormalization, the arbitrary scale seems
innocuous. In practice, we cannot get rid of easily, unless we are able to
compute the perturbative series at all orders. Loop integrals will produce
poles in of the form,

=
1

+ log() +O(). (14.36)

The procedure of renormalization will eliminate the residues of the poles in
in the expressions for scattering matrix-elements. However the log() terms
can only eliminated by means of a resummation of all perturbative orders.
14.3.1 Gamma-matrices in dimensional regularization
How do we treat -matrices in dimensional regularization? The prescription
that is followed in conventional dimensional regularization for the Cliord
algebra is

= 2g

1
44
, (14.37)
193
where the metric is taken in D = 4 2 dimensions,
g

= 4 2, (14.38)
and the dimensionality of the gamma matrices is four by four,
tr1
44
= 4. (14.39)
Let us do some simple calculations with this algebra.

=
1
2

= g

1
44
= (4 2)1
44
. (14.40)
Also,

= 2g

(4 2)

= 2(1 )

. (14.41)
and so on.
A diculty arises in dening the matrix
5
in 4 2 dimensions. In four
dimensions this is dened as,

5
= i
0

3
. (14.42)
It is often just ne to use a prescription of an anti-commuting
5
,

,
5
= 0,
in performing the gamma-matrix algebra for QED amplitudes. But a discus-
sion of this issue lies beyond the scope of this course.
14.3.2 Tensor loop-integrals
In computing QED scattering amplitude, we need to calculate loop integrals
where tensors of the loop-momentum appear in the numerator. These arise
due to the Feynman rule for the fermion propagator, which is written as
i
,k m
=
i (

+ m)
k
2
m
2
. (14.43)
194
Particles propagating in loops, called virtual, generate loop-momenta in the
numerators of loop integrands.
At one-loop we can always express tensor integrals in terms of scalar
integrals. The technique is known as Passarino-Veltman technique and we
can illustrate it with a couple of examples. Consider the following tensor
one-loop integral,
I [k

]
_
d
d
k
k

k
2
(k + p)
2
, (14.44)
where the +i term in the denominator is implicit and we take for simplicity
m = 0. The integrand is a rank one vector in the loop momentum. Due
to Lorentz convariance, the result of the integration will also be a rank one
tensor. We can then write,
I [k

] = Ap

, (14.45)
where on the rhs we have written the most general rank-one tensor that
we can construct from the momenta in the integral other than the loop
momentum. The coecient A is as yet undetermined. We can express it in
terms of integrals which have scalar numerators. We multiply Eq. 14.45 with
p

and rearrange
A =
1
p
2
I [k p] . (14.46)
Notice that the scalar product in the numerator of the integral above can be
expressed in terms of the denominators of the integral,
k p =
1
2
(k + p)
2

k
2
2

p
2
2
. (14.47)
We then have that,
A =
1
2
_

_
d
d
k
1
k
2
(k + p)
2
+
1
p
2
_
d
d
k
1
k
2

1
p
2
_
d
d
k
1
(k + p)
2
_
. (14.48)
Notice, that we only nd integrals with a constant numerator, which is inde-
pendent of the loop momentum. The last two integrals are identical, as we
can see by performing the shift k k +p in one of them. In this particular
example, they cancel against each other, and we have
_
d
d
k
k

k
2
(k + p)
2
=
p

2
_
d
d
k
1
k
2
(k + p)
2
. (14.49)
195
Consider now a more complicated example,
I [k

]
_
d
d
k
k

k
2
(k + p)
2
. (14.50)
The most general rank two tensor that we can write using the external mo-
menta of the integral is,
I [k

] = A
1
g

+ A
2
p

p
2
. (14.51)
Contracting with g

and
pp
p
2
, we obtain the system of equations,
_
g

p
2
g

pp
p
2
p

p
2
pp
p
2
_
_
A
1
A
2
_
=
_
I [k
2
]
I[(kp)
2
]
(p
2
)
2
_
. (14.52)
In arbitrary dimensions,
g

= d = 4 2,
and the above system of equations becomes,
_
4 2 1
1 1
__
A
1
A
2
_
=
_
I [k
2
]
I[(kp)
2
]
(p
2
)
2
_
. (14.53)
with a solution,
_
A
1
A
2
_
=
1
3 2
_
1 1
1 4 2
_
_
I [k
2
]
I[(kp)
2
]
(p
2
)
2
_
. (14.54)
The two integrals on the rhs can be expressed in terms of two scalar master
integrals,
I
a

_
d
d
k
1
k
2
, (14.55)
and
I
b

_
d
d
k
1
k
2
(k + p)
2
. (14.56)
Specically, we have
I
_
k
2

=
_
d
d
k
k
2
k
2
(k + p)
2
=
_
d
d
k
1
(k + p)
2
=
_
d
d
(k + p)
1
(k + p)
2
=
_
d
d
k
1
k
2
= I
a
. (14.57)
196
I
a
is a scaleless integral and, as we have discussed, it is zero in dimensional
regularization. We let it as an exercise to prove that
I
_
(k p)
2

=
p
2
2
I
a
+
(p
2
)
2
4
I
b
. (14.58)
We can express all one-loop tensor integrals in terms of four master inte-
grals,
M
A

_
d
d
k
i
d
2
1
k
2
m
2
1
+ i
, (14.59)
M
B

_
d
d
k
i
d
2
1
(k
2
m
2
1
+ i) [(k + p)
2
m
2
2
+ i]
, (14.60)
M
C

_
d
d
k
i
d
2
1
(k
2
m
2
1
+ i) [(k + p
1
)
2
m
2
2
+ i] [(k + p
2
)
2
m
2
3
+ i]
,
(14.61)
and
M
D

_
d
d
k
i
d
2

1
(k
2
m
2
1
+ i) [(k + p
1
)
2
m
2
2
+ i] [(k + p
2
)
2
m
2
3
+ i] [(k + p
2
)
2
m
2
4
+ i]
.
(14.62)
Exercise: Express the following tensor integrals in terms of master integrals,
_
d
d
k
k

k
2
m
2
_
d
d
k
k

k
2
m
2
_
d
d
k
k

(k
2
m
2
1
) [(k + p)
2
m
2
2
]
_
d
d
k
k

(k
2
m
2
1
) [(k + p)
2
m
2
2
]
(14.63)
The above master integrals are known analytically since many years.
Their computation is involved for the general case of arbitrary masses m
i
and momenta p
i
. We shall need the following two special cases,
_
d
d
k
i
d
2
1
k
2
m
2
= (1 + )
_
m
2
_
1
, (14.64)
197
and
_
d
d
k
i
d
2
1
k
2
(k + p)
2
=
c

(1 2)
_
p
2
_

, (14.65)
where
c

=
(1 +)(1 )
2
(1 2)
. (14.66)
Exercise: Prove the above using the method of Feynman parame-
ters
14.4 Electron propagator at one-loop
As an application of the above, we can compute the electron propagator
through one-loop order. Applying the QED Feynman rules we have,
_
d
4
xe
ipx
[ T(x)

(0) [) =
i
,p m
+
i
,p m
(i
2
(p))
i
,p m
+ . . . (14.67)
with
i
2
=
_
d
d
k
(2)
d
(ie

)
i
,k +,p m
(ie

)
i
k
2
_
g

+ (1 )
k

k
2
_
,
(14.68)
where the i terms in the denominators are implicit. As we have discussed,
in d = 4 2 dimensions, we need to introduce a mass scale factor

for
every coupling factor e.
There are many ways that one can decide to compute this integral. Per-
haps, the easiest is with applying directly Feynman parameters. I choose
a bit of a longer method here, the details are not important, but the out-
come is, in the fact that it contains ultraviolet divergences. We rst make
all denominators quadratic in the loop momentum by writing,
1
,k +,p m
=
,k +,p + m
(,k +,p m) (,k +,p + m)
=
,k +,p + m
(k + p)
2
m
2
, (14.69)
198
and then contract the Lorentz indices and use the d-dimensional Cliord
algebra. We obtain
i
2
= e
2

2
_
d
d
k
(2)
d
(2 d)(,k +,p) + md
k
2
[(k + p)
2
m
2
]
+ e
2

2
(1 )
_
d
d
k
(2)
d
,k(,k +,p + m),k
(k
2
)
2
[(k + p)
2
m
2
]
(14.70)
We can simplify the second integral if we use the identity
,k
1
,k +,p m
,k = (,k , p + m) + (,p m)
1
,k +,p m
(,p m) , (14.71)
which is proved easily by writing ,k = (,k + ,p m) (,p m). The rst
term in the rhs of the identity leads to a scaleless integral which is zero in
dimensional regularization. We then have
i
2
= e
2

2
_
d
d
k
(2)
d
(2 d)(,k +,p) + md
k
2
[(k + p)
2
m
2
]
+ e
2

2
(1 ) (,p m)
_
d
d
k
(2)
d
(,k +,p + m)
(k
2
)
2
[(k + p)
2
m
2
]
(,p m) ,
(14.72)
and by changing variables to k k p, we write
i
2
= e
2

2
_
d
d
k
(2)
d
(2 d),k + md
(k p)
2
[k
2
m
2
]
+ e
2

2
(1 ) (,p m)
_
d
d
k
(2)
d
(,k + m)
[(k p)
2
]
2
(k
2
m
2
)
(,p m) .
(14.73)
Now we can write
,k =

,
and reduce the emerging tensor integrals to master integrals (exercise).
However, there is an easier method. First, we simplify further the gamma
matrices in the numerators and we write
i
2
= e
2

2
_
d
d
k
(2)
d
(2 d),k + md
(k p)
2
[k
2
m
2
]
e
2

2
(1 )
_
d
d
k
(2)
d
, k(p
2
m
2
) + 2,p(p k m
2
) + m
_
(k
2
m
2
) (k p)
2

[(k p)
2
]
2
(k
2
m
2
)
.
(14.74)
199
We anticipate the result to be written in terms of the vector p

, contracted
with gamma matrices. We can write the general ansatz,
i
2
=

n=0
(,p)
n
f
n
(p
2
, m
2
). (14.75)
Given that ,p
2
= p
2
, ,p
3
= p
2
,p etc, the series terminates after the second term,
and we cast the general functional form of the one-loop electron propagator
as,
i
2
= i
2,V
(p
2
, m
2
),p im
2,S
(p
2
, m
2
)1. (14.76)
We can nd expressions for the functions
2,S
and
2,V
by taking the traces
im
2,S
(p
2
, m
2
) =
1
4
tr (i
2
) , (14.77)
and
i
2,V
(p
2
, m
2
)p
2
=
1
4
tr (i
2
,p) . (14.78)
Explicitly,

2,S
(p
2
, m
2
) = ie
2

2
[d (1 )]
_
d
d
k
(2)
d
1
(k p)
2
[k
2
m
2
]
, (14.79)
and

2,V
(p
2
, m
2
)p
2
= ie
2

2
_
d
d
k
(2)
d
(2 d)k p
(k p)
2
[k
2
m
2
]
ie
2

2
(1 )
_
d
d
k
(2)
d
k p(p
2
+ m
2
) 2p
2
m
2
[(k p)
2
]
2
(k
2
m
2
)
.
(14.80)
Both functions
2,S
and
2,V
are divergent due to the ultraviolet limit k
. In this limit, we can ignore the electron mass in the above expressions.
We then have

2,S
(p
2
, 0) = ie
2

2
[d (1 )]
_
d
d
k
(2)
d
1
(k p)
2
[k
2
]
, (14.81)
which has a single pole in ,

2,S
(p
2
, m
2
) =
2,S
(p
2
, 0)

UVpole
+O(
0
) =
e
2
4
2
(4
2
)

(3+)
_
1

+O(
0
)
_
(14.82)
200
with d = 4 2.
Similarly, the UV poles of
2,V
can be computed from

2,V
(p
2
, 0)p
2
= ie
2

2
_
d
d
k
(2)
d
(2 d)k p
(k p)
2
k
2
ie
2

2
(1 )p
2
_
d
d
k
(2)
d
k p
[(k p)
2
]
2
k
2
.
(14.83)
Writing
k p =
1
2
_
k
2
+ p
2
(k p)
2
_
,
and setting to zero scale-less integrals we have

2,V
(p
2
, 0)p
2
= (14.84)
14.5 Photon propagator at one-loop
201
Chapter 15
One-loop renormalization of
QED
202