Last Latexed: September 12, 2013 at 10:18 1

Physics 615, Lectures 1-3 Sept. 5–12, 2013

Introduction
c
Copyright 2005 by Joel A. Shapiro

1 Introduction
According to the catalogue, this course is “Overview of Quantum Field The-
ory”. Why “Overview”? When we teach Classical Mechanics or Quantum
Mechanics, we teach the groundwork of those subjects, at least up to some
point, and thus we call the courses “Introductions to” rather than “Overviews
of”. Giving an overview is an unusual way to start teaching one of the broad
conceptual frameworks to Physicists. We know that these theoretical frame-
works need to built on a firm conceptual basis. All physicists need that
understanding and so we don’t try to include a preview of time dependant
perturbation theory in freshman physics courses.
Quantum Field theory is also a broad conceptual framework underlying
much of how physicists understand their field. It is, except possibly for
the much more abstract string theory, the only framework for understanding
elementary particle physics at energies high enough for relativity to matter.
In ordinary Quantum Mechanics, the basic operators are the degrees of free-
dom, usually the positions and momenta of each particle. With sufficient
energy available, particles can be created, so the coordinates describing the
positions or spins of a list of individual particles cannot suffice to describe
the state of a system, and thus the quantum mechanical framework used in
nonrelativistic physics is inadequate.
Quantum field theory is also extensively used in condensed matter physics,
even though here there is not enough energy for electrons and positrons to
be created. In condensed matter physics one often describes the system
not in terms of fundamental particles or even nuclei or atoms, but in terms
of effective particles such as holes, phonons or excitons, and these are not
conserved.
Even in atomic physics, because intermediate states in perturbation the-
ory may not conserve energy, the possibility of particle creation has effects,
such as the Lamb shift in hydrogenic atoms, and larger effects in the inner
shells of heavy atoms, for which the potential energies may be comparable
to the rest masses of the electrons.
615: Introduction Last Latexed: September 12, 2013 at 10:18 2

So, if field theory is the language of so much of modern physics, why

an overview rather than a careful solid foundation-laying introduction? The
motivation is that right from the beginning, QFT raises a host of issues
which are quite complex. Unlike in classical mechanics or nonrelativistic
quantum mechanics, there are no nontrivial problems which can be solved
exactly. For example, even though we can write a “complete” theory of
electrodynamics, complete in the sense that it would be if there were not
other interactions, which is quite adequate for understanding a great deal of
observed physics, we still cannot do any exact calculations with that theory,
except if we assume there are no interactions at all. What we can do, very
successfully, are calculations in perturbation theory, treating the interaction
parameter, which is the charge of the electron, as a small parameter in terms
of which we can calculate scattering amplitudes as power series expansions.
But these calculations are both complicated and raise some very difficult
issues, with which you should be acquainted, even though you may not be
prepared to study these ideas in detail. Some examples:

• Renormalization is required to make sense of all but the lowest level

approximation to scattering amplitudes. The renormalization group is
important in understanding critical phenomena and also for the running
coupling constants one must use even in low order calculations of high
energy scattering experiments.

• Gauged symmetry groups are responsible for the standard model of

high energy physics and also provide models in CM, but a full discussion
of their quantization is very involved.

• Spontaneous symmetry breaking plays a crucial role in the standard

model, as well as in condensed matter (e.g. ferromagnetism), and is
probably responsible for the inflation that set up the initial conditions
for cosmology.

• The Higgs model is an elaboration on spontaneous symmetry breaking

in theories with gauged symmetry groups. Higgs models are essential in
both high energy (the standard model) and condensed matter physics.

The text, Peskin and Schroeder’s “Introduction to Quantum Field The-

ory”, does indeed give a good introduction laying out these ideas, but it is
far too much to teach in one semester. So this course will be a compromise,
615: Introduction Last Latexed: September 12, 2013 at 10:18 3

a fairly solid foundation to the beginning of quantum field theory, together

with a more qualitative or sketchy overview of the more advanced ideas that
I have just mentioned. We will learn how to use Feynman graphs to de-
scribe scattering amplitudes in perturbation theory, at least at the level that
experimentalists need for most purposes.
For today, we will have a brief review of Classical Mechanics in which we
make sure we learn to deal with fields. Then we will discuss the limitations
of particle quantum mechanics and begin to develop the quantum mechan-
ics of fields. This process has been called second quantization. In the
first quantization, in the 1920’s, we learned to replace the phase space vari-
ables, i.e. the positions and momenta of individual particles which describe
the classical mechanics of the system, by operators operating on the wave
function, an ordinary function of the positions. In Quantum mechanics it is
the wave function, rather than the coordinates, that describes the state of
the system. But in quantum field theory this wavefunction becomes a field
which is then quantized, and we will treat the degrees of freedom of this
field as operators.
I think this name, second quantization, is losing its appeal, and in a
sense it is misleading anyway. After all, historically the first thing to be
quantized was the electromagnetic field inside a black box. Here we were
used to the idea that light was described by fields, which could be considered
classical, while it was the particle nature of photons that was considered
quantum mechanical. For electrons, the particle nature was classical and
the wave function, a field ψ(~x, t) was considered quantum mechanical. But
the de Broglie relationship between momentum and wavelength is the same
for photons and electrons, and the Schrödinger Equation can be derived by
applying de Broglie to the Newtonian equations just as Maxwell’s equations
relates to electromagnetic waves.
Even in classical mechanics, relativity does not get along very well with
a particle description. There is no problem for free point particles, or for
describing collisions at a point. But for interesting particle mechanics we
need potential energies which are functions of the separation of the particles
~x1 − ~x2 , and not just delta functions δ 3 (~x1 − ~x2 ). But such a potential
describes action at a distance, the force on particle 1 at time t depending
on the position of particle 2 at the same time t, which is impossible for
particle 1 to know in relativity. The only relativistic interactions between
separated particles known to classical mechanics are electromagnetic (and,
in a sense, general relativistic gravity), and these interactions cannot be
615: Introduction Last Latexed: September 12, 2013 at 10:18 4

understood in general without the introduction of fields.

[I am not going to cover Chapter 1 of the book — read it if you like. I

will be covering, in the first four lectures, the material of Chapter 2, though
somewhat differently.]

2 Review of Classical Mechanics

Let us very briefly review the fundamentals of mechanics. In describing a
physical situation, the first thing one must do is decide what the coordinates
are. In classical mechanics of a discrete set of degrees of freedom, we will
have coordinates qj (t) which can depend on time. For the systems we wish
to deal with, the physics can be described by a Lagrangian

L(t) = L({qj (t)}, {q̇j (t)}, t)

which is a function of the coordinates and their first time derivatives at a

moment of time. The dynamics of the system tells how qj develops in time.
That is, it gives the path Γ : t ∈ [t1 , t2 ] 7→ qj (t) that the physical system
will actually take through coordinate space as a function of time. This is
determined by a functional called the action
Z t2
SΓ = L({qj (t)}, {q̇j (t)}, t) dt,
t1

where the integral is to be done with q and q̇ evaluated along an arbitrary

path Γ. Hamilton’s principle tells us the actual physical path will be one
which is stationary under small variations in the path, keeping the endpoints
qj (t1 ) and qj (t2 ) fixed. Considering an arbitrary infinitesimal variation of
the path, δqj (t), (with δqj (t1 ) = δqj (t2 ) = 0), and asking that δS = 0, gives
the Lagrange equations,
!
d ∂L ∂L
− = 0.
dt ∂ q̇j ∂qj

We should note here that the space in which q lives is not necessarily
ordinary three dimensional Euclidean space, or even a Euclidean space of
any number of dimensions, though for single particle motion it is ordinary
space. For example, for a pendulum swinging about a fixed axis, the sole
615: Introduction Last Latexed: September 12, 2013 at 10:18 5

coordinate may be θ, the angle with respect to the downward direction, which
lives on a circle (with θ = 0 and θ = 2π the same point). If it is swinging
from a fixed point rather than a fixed axis, its coordinates are the polar angle
θ and the azimuthal angle φ, which together live on a two-sphere S 2 , that is,
the surface of a three dimensional ball.
Classical mechanics of a conservative system can also be described in
terms of a Hamiltonian, which depends on momenta pj as well as the coor-
dinates, but does not depend on the time derivatives of the coordinates:

H({qj (t)}, {pj (t)}, t).

The canonical momentum pj conjugate to qj is given in the Lagrangian

formulation by
∂L
pj = . (1)
∂ q̇j
Eq. (1) is known as the constitutive equation. Note that these canonical
momenta, one for each degree of freedom, are to be distinguished from the
3 (or 4) components of the total momentum, except for a single particle
problem described by cartesian coordinates. Usually this expression for the
canonical momenta pi in terms of {qj } and {q̇j } can be inverted to give q̇j (t)
in terms of {qk (t)} and {pk (t)}, in which case one can define the Hamiltonian
from the Lagrangian by
X
H({qj (t)}, {pj (t)}, t) = q̇k (t)pk (t) − L({qj (t)}, {q̇j (t)}, t),
k

where it is understood that on the right hand side the q̇j ’s are to be replaced
by their values in terms of q and p. When Eq. (1) is not solvable for q̇, we
have a complicated situation related to gauge invariance. We will see that
this situation does arise for the electromagnetic field.
The equations of motion in Hamiltonian form are
∂H ∂H
q̇j = , −ṗj = ,
∂pj ∂qj
which determine the classical path the system takes through phase space,
(q(t), p(t)).
We can also relate any explicit time dependences of L and H by
∂L ∂H
− = .
∂t ∂t
615: Introduction Last Latexed: September 12, 2013 at 10:18 6

The Hamiltonian is a function on extended phase space, {qj , pj , t} while

the Lagrangian is not quite a function on coordinate space, because it de-
pends of q̇j as well as qj . The Hamiltonian equations of motion can be derived
from a variational principle in which one varies a path Γ through phase space
Γ : t 7→ (q(t), p(t)), varying q(t) and p(t) independently to extremize the ac-
tion  
Z X
S=  pj q̇j − H(q, p, t) dt,
j

where q̇ is evaluated from the path q(t) and is unrelated to p(t).

At the beginning of a course on classical or quantum mechanics we deal
with a system of discrete degrees of freedom, qj . For electromagnetism,
however, the degrees of freedom are not only the positions of the discrete
charged particles, but also the electric and magnetic fields at each point
in space, as a function of time. As we shall argue shortly, even for ordinary
particles we are going to need to deal with fields, where there is one (or
perhaps several) degree(s) of freedom for each point in space, at each time.
To make sure we understand formally how to deal with fields, let’s begin
with the simple example of the continuum limit of a one dimensional lattice
of masses on springs. Each lattice point xj = j a (for integer j, that is j ∈ Z)
has a mass m which may be displaced from its equilibrium position xj by an
amount ηj (t). Between neighboring points there are springs of equilibrium
length a and spring constant k, so the kinetic and potential energies of the
system are
1 X 2 1 X
T = m η̇j , V ({ηj }) = k (ηj+1 − ηj )2 .
2 j 2 j

In the continuum limit, it is more appropriate to use the linear mass density
µ = m/a, Young’s modulus Y and the Lagrangian density. Young’s modulus
is the force required to produce a unit extension per unit length, and as the
force required to stretch one spring, of length a, is F = k(ηj+1 − ηj ), we have
F = Y (ηj+1 − ηj )/a, so k = Y /a. Thus we may write the lagrangian as
 
1 2 
1 ηj+1 − ηj

η̇j2 − Y
X X
L=a Lj = a µ .
2 2 a 
j j

The Lagrangian density in the continuum limit is

!2
1 1 ∂η
L(x) = lim Lx/a = µη̇ 2 − Y .
a→0 2 2 ∂x
615: Introduction Last Latexed: September 12, 2013 at 10:18 7

R
and L = L(x)dx.
This Lagrangian, however, will not be of much use until we figure out what
is meant by varying it with respect to each dynamical degree of freedom or its
corresponding velocity. In the discrete case we have the canonical momenta
pj = ∂L/∂ η̇j , where the derivative requires holding all η̇j fixed, for j 6= i, as
well as all ηk fixed. This extracts one term from the sum 21 µ aη̇j2 , and this
P

would appear to vanish in the limit a → 0. Instead, we define the canonical

momentum as a density, pj → ap(x = ja), so

1 ∂ X
p(x = ja) = lim a L(η(x), η̇(x), x)|x=aj .
a→0 a ∂ η̇j
j

We may think of the last part of this limit,

X Z
lim a L(η(x), η̇(x), x)|x=aj = dx L(η(x), η̇(x), x),
a→0
j

if we also define a limiting operation

1 ∂ δ
lim → ,
a→0 a ∂ η̇j δ η̇(x)
1 ∂
and similarly for a ∂ηj
, which act on functionals of η(x) and η̇(x) by

δη(x1 ) δ η̇(x1 ) δη(x1 ) δ η̇(x1 )

= δ(x1 − x2 ), = = 0, = δ(x1 − x2 ).
δη(x2 ) δη(x2 ) δ η̇(x2 ) δ η̇(x2 )

Here δ(x′ − x) is the Dirac delta function, defined by its integral,

Z x2
f (x′ ) δ(x′ − x) dx′ = f (x)
x1

for any function f (x), provided x is contained in the interval (x1 , x2 ). Thus

δ ℓ 1 ℓ
Z Z
p(x) = dx′ µη̇ 2 (x′ , t) = dx′ µη̇(x′ , t)δ(x′ − x) = µη̇(x, t).
δ η̇(x) 0 2 0

We also need to evaluate

!2
δ δ ℓ
′ −Y ∂η
Z
L= dx .
δη(x) δη(x) 0 2 ∂x x=x′
615: Introduction Last Latexed: September 12, 2013 at 10:18 8

For this we need

δ ∂η(x′ ) ∂
′
= ′ δ(x′ − x) := δ ′ (x′ − x).
δη(x) ∂x ∂x

The generalized function δ ′ (x′ − x) is also defined by its integral,

x2 x2 ∂
Z Z
f (x′ )δ ′ (x′ − x)dx′ = f (x′ ) δ(x′ − x)dx′
x1 x1 ∂x′
′ ′ x
Z x2 ∂f
= f (x )δ(x − x)|x21 − dx′ ′
δ(x′ − x)
x1 ∂x
∂f
= − (x),
∂x
where after integration by parts the surface term is dropped because
δ(x − x′ ) = 0 for x 6= x′ , which it is for x′ = x1 , x2 if x ∈ (x1 , x2 ). Thus

δ ℓ ∂η ′ ′ ′ ∂2η
Z
L=− dx′ Y (x )δ (x − x) = Y 2 ,
δη(x) 0 ∂x ∂x

and Lagrange’s equations give the wave equation

∂2η
µη̈(x, t) − Y 2 = 0.
∂x

2.1 Higher dimensions

Now let us generalize to more than one spatial dimension, say D − 1, reserv-
ing D to represent the dimension of space-time. Of course we are usually
interested in D = 4, to describe the apparent real world, but other dimen-
sions will come up in condensed matter, in dimensional regularization, and
in string theory. We will use relativistic notation, with coordinates xµ , with
µ = 0, 1, . . . D − 1, and with x0 = ct = t, because we use units where the
speed of light is 1. Thus times and distances are both measured in meters.
We are distinguishing contravariant vectors xµ from covariant vectors xµ ,
related by a metric tensor

1 0 0 0
 
 0 −1 0 0 
gµν = g µν = ,

with xµ = g µν xν ,
0 0 −1 0 


0 0 0 −1
615: Introduction Last Latexed: September 12, 2013 at 10:18 9

and xµ = gµν xν . We are also using the “Einstein summation convention”,

that an index repeated up and down is automatically summed. The momenta
are naturally covariant, for in quantum mechanics
∂
pµ ∼ ih̄ ,
∂xµ
though if we write it in nonrelativistic notation,
 
~ .
pµ = (E, p~ ) ∼ i∂/∂t, −i∇

Note
1 0 0 0
 
∂ ν 0 1 0 0
x = δµν = 

.
∂xµ 0 0 1 0
0 0 0 1
The spatial part of a contravariant vector will be represented by ~x (boldface
in the book, whose notation we otherwise follow). The dot product of two D-
vectors is understood to use the Minkowski metric, p2 := g µν pµ pν = p20 −~p 2 =
E 2 − p~ 2 , which for a particle is m2 , classically at least.
Let us consider a scalar field φ(~x, t). As for the chain of masses on springs,
the Lagrangian will be a spatial integral of the Lagrangian density,
Z
L(t) = L dD−1x.

As for the chain, the Lagrangian density will depend on spatial as well as
temporal first derivatives of φ, and may also depend on φ itself:

L = L (φ, ∂µ φ, xµ ) ,

where we are switching to relativistic notation, with

∂ 
~ .

∂µ φ = φ = φ̇, ∇φ
∂xµ
The canonical momentum density we will1 call π(~x, t):

δL
π(~x, t) := ,
δ φ̇(~x, t)
1
Peskin defines π(~x, t) := ∂L/∂ φ̇(~x) which I agree with, but with some explanation.
615: Introduction Last Latexed: September 12, 2013 at 10:18 10

and the hamiltonian is

Z
H = H(~x, t) dD−1x
Z h  i
= dD−1 x π(~x, t)φ̇(~x, t) − L φ(~x, t), φ̇(~x, t), ∇φ(~x, t), ~x, t .

Before we give an example, let us note that we have not taken a very
relativistic attitude towards our Lagrangian density, for the momentum den-
sity π(~x, t) is its variation with respect to the time derivative of φ, but we
haven’t varied with respect to the spatial derivatives. We might consider the
canonical momentum π(~x) as one component of a four vector

δ δL(~x ′ , t)
Z
D−1 ′
L(t) = d x .
δ(∂µ φ(~x, t)) δ(∂µ φ(~x, t))

Our procedures do not fully treat δ/δ(∂µ φ(~x, t)) covariantly, however, for we
have been treating φ and φ̇ as completely independant variables, while φ and
∇φ are not. Thus the δ variation above is not well defined.
As for the mass chain, we have

δ ∂φ(~x ′ , t) ∂
′ j = ′ j δ D−1 (x′ − x),
δφ(~x, t) ∂x ∂x

where δ D−1 (~x ′ − ~x)Ris the D − 1 dimensional version of the Dirac delta, zero
unless ~x ′ = ~x with dD−1x′ f (~x ′ )δ D−1 (~x ′ − ~x) = f (~x). So we are not treating
∂j φ as independent of φ.
If we considered a different form of variation, δ̄, in which φ and ∇φ ~ are
considered independent and the Lagrangian density is L = L(φ, φ̇, ∇φ), then ~
our original δ variation can be expressed in terms of δ̄, with
δL δ
Z
= ~
dD−1 x′ L(φ, φ̇, ∇φ)(~
x ′)
δφ(~x) δφ(~x)
δ
Z
= dD−1 x′ ~ x ′)
L(φ, φ̇, ∇φ)(~
δφ(~x)
(
δ̄
Z
= dD−1 x′ ~
L(φ, φ̇, ∇φ)(~x ′) +
δ̄φ(~x)
" # )
δ̄ ~ ′ δ ′
L(φ, φ̇, ∇φ)(~
x) ∂j φ(~x )
δ̄∂j φ(~x ′ ) δφ(~x)
615: Introduction Last Latexed: September 12, 2013 at 10:18 11

~
(
∂L(φ, φ̇, ∇φ)
Z
D−1 ′
= d x (~x ′ )δ D−1 (~x ′ − ~x) +
∂φ
 
~
)
∂L(φ, φ̇, ∇φ) ′ ∂ D−1 ′
 (~x ) ′ j δ (~x − ~x),
∂(∂j φ) ∂x
 
~
∂L(φ, φ̇, ∇φ) ~
∂ ∂L(φ, φ̇, ∇φ)
= (~x) − j  (~x)
∂φ ∂x ∂(∂j φ)

Let us define
δ̄ Z
δ̄L(~x ′ , t) ∂L(φ, ∂µ φ)
π µ (~x, t) = L(t) = dD−1 x′ = (~x, t).
δ̄(∂µ φ(~x, t)) δ̄(∂µ φ(~x, t)) ∂(∂µ φ)

Thus we have
 
δL ~
∂L(φ, φ̇, ∇φ) ~
∂ ∂L(φ, φ̇, ∇φ)
= (~x) − j  (~x)
δφ(~x) ∂φ ∂x ∂(∂j φ)
∂L ∂
= − j π j (~x)
∂φ(~x) ∂x

Thus in writing out the Lagrange equations,

!
d δL δL d 0 ∂L ∂
− = π (~x) − + j π j (~x)
dt δ φ̇(x) δφ(x) dt ∂φ(x) ∂x
∂ µ ∂L
= π (~
x ) − = 0.
∂xµ ∂φ(x)

Thus we see that, despite a rather non-covariant formulation of the laws of

Lagrangian mechanics, the results are covariant equations of motion.
The covariant equations of motion are

∂ ∂L(φ, ∂ν φ) ∂L
− = 0.
∂xµ ∂(∂µ φ) ∂φ

Our first important example will be the Klein-Gordon field, in three spa-
tial dimensions, with a lagrangian density
1 1 1 1 1
L = φ̇2 − (∇φ)2 − m2 φ2 = (∂µ φ)(∂ µ φ) − m2 φ2
2 2 2 2 2
615: Introduction Last Latexed: September 12, 2013 at 10:18 12

This is the generalization of the mass chain to three dimensions, with an

added term that depends on φ itself, if the parameter m 6= 0. The choice
of setting the coefficient of φ̇2 to 12 may be viewed as setting the scale of φ,
while the coefficient of (∇φ)2 is then fixed if we are to have a relativistic
theory. Now
δL
π := = φ̇,
δ φ̇
and more generally
δ̄L
π µ := = ∂ µ φ,
δ̄∂µ φ
the equations of motion are

∂µ ∂ µ φ + m2 φ = 0.

and the Hamiltionian density is

1 1 1
H = π φ̇ − L = π 2 + (∇φ)2 + m2 φ2
2 2 2

We have several more things to discuss about classical field theory:

• Symmetries and conserved currents and charges (Noether’s theorem).

• Maxwell’s equations and electromagnetism, including L and H, and

four dimensional Fµν .

This will be postponed until the end of this chapter. First let’s take a step
into Quantum Mechanics.

3 Quantum Mechanics
In quantum mechanics the question of what path a system takes through
coordinate space is no longer meaningful, nor can a system be at a point in
phase space, for the Heisenburg uncertainty principle says the most localized
a state can be still requires a given volume in phase space. The possible
states of a system can be described in terms of wave functions on coordinate
space, and quantum mechanics (Schrödinger’s equation, for example) tells
us how those functions evolve with time. Thus we have a unitary operator,
615: Introduction Last Latexed: September 12, 2013 at 10:18 13

the time-evolution operator, which gives the amplitude for a state |ψ1 i at
one time to be in the state |ψ2 i a time t later, hψ2 | e−iHt/h̄ |ψ1 i, where we
have assumed the Hamiltonian is time-independent. In fact, this transition
amplitude can be understood in a functional integral formulation of quantum
mechanics as a sum over all possible paths Γ of eiSΓ /h̄ , for the system to go
from one configuration to another at a later time, rather than, as in classical
mechanics, choosing the one path that extremizes the action.
Actually, as we are going to be dealing only with quantum systems, it is
convenient to use units with h̄ = 1, thereby measuring energies and (ordi-
nary) momenta in units of inverse meters, as we have already set c = 1.
Let us look at a simple transition amplitude for a free nonrelativistic
particle, to move from a definite position ~x0 at time 0 to a point ~x at time t.
Here H = ~p 2 /2m, so the amplitude is
2
h~x| e−iHt |~x0 i = h~x| e−ip t/2m |~x0 i (2)
d3 p
Z
2
= 3
h~x| e−ip t/2m |~p i h~p |~x0 i (3)
(2π)
d3 p −i~p 2 t/2m+i~p·(~x−~x0 )
Z
= e (4)
(2π)3
m 3/2 im(~x−~x0 )2 /2t
 
= e . (5)
2πit
Here2 we are using states normalized so that h~x ′ | ~xi = δ 3 (~x ′ − ~x), with
momentum eigenstates normalized so that h~p ′ | p~i = (2π)3 δ 3 (~p ′ − ~p), and
completeness is
d3 p
Z
1= |~p i h~p | ,
(2π)3
and h~x| p~ i = eipx . This explains the second and third lines above. To get to
the last line, it is useful to know the Gaussian integral3
s
1 ~ ~
πD B · A−1 · B
Z
~ x
dD xe −~
x·A·~
x+B·~
= e4 ,
det A
2
Later we will change that normalization to be consistent with Peskin and Schroeder,
q p
p iPS = 2 p~ 2 + m2 |~p ihere .
|~

3
See http://en.wikipedia.org/wiki/Gaussian integral.
615: Introduction Last Latexed: September 12, 2013 at 10:18 14

if A is a diagonalizable matrix with all eigenvalues having a positive real part.

In our case we are at the boundary of this condition, but the expression still
holds.
Note there is a nonzero probability that the particle will be found at
any distance |~x − ~x0 | even after an arbitarily short time t. This is not sur-
prising, for an eigenstate of position ~x0 contains arbitrarily high momenta,
and for nonrelativistic mechanics this means arbitrarily high velocities. But
what might be surprising is that the q same thing happens even if we use the
relativistic form for the energy, E = p2 + m2 .
√
2 2
h~x| e−iHt |~x0 i = h~x| e−i p +m t |~x0 i (6)
Z
d3 p √
−i p2 +m2 t
= h~
x | e |~p i h~p |~x0 i (7)
(2π)3
1
Z ∞ √
2 −i p2 +m2 t
= p dp e (8)
(2π)3 0
Z π Z 2π
i|p||~
x−~
x0 | cos θ
× sin θ dθ e dφ
0 0
1
Z ∞ √ Z π
2 −i p2 +m2 t
= 2
dp p e sin θdθei|p||~x−~x0 | cos θ (9)
(2π) 0 0
1
Z ∞ √ Z 1
2 2
= 2
dp p2 e−i p +m t du ei|p||~x−~x0 |u (10)
(2π) 0 −1
1
Z ∞ √
−i p2 +m2 t
= dp p e sin(p|~x − ~x0 |) (11)
2π 2 |~x − ~x0 | 0

In terms of the dimensionless variables τ = mt and r = m|~x − ~x0 |, and a

dimensionless integration variable v = p/m, we have

m3 ∞ √
Z
−iτ 1+v2
h~x| e−iHt |~x0 i = dv v sin(vr)e (12)
2π 2 r 0
m3 d Z ∞ √
2
= − 2 dv cos(vr)e−iτ 1+v (13)
2π r dr 0
Gradshtein and Ryzhik 3.914 tells us that the last integral is on the border
iτ √
of the valid region and should be √ 2 2
K 1 ( r 2 − τ 2 ).
r −τ
It is pleasing to find that, except for the non-covariant normalization fac-
tor, the dependence is on the invariant m2 ((~x − ~x0 )2 − t2 ) but it is not so
615: Introduction Last Latexed: September 12, 2013 at 10:18 15

pleasant to see that the function does not vanish for positive, spacelike, argu-
ments. Evaluation
√ by steepest descent4 shows that for r ≫ τ the propagator
2 2
goes like e− r −τ , so it falls exponentially but is not identically zero, as rela-
tivity should have it. Thus we have a finite probability that the particle has
moved to a position it only could have gotten to by moving faster than the
speed of light!
We will see that the resolution of this contains
the creation of particle — antiparticle pairs, and ∆ x = ct
the probability of finding a particle at a distance t x
x > ct from where one particle was is due to cre-
ation of another particle from the vacuum, fol- x0
lowed by the antiparticle member of the pair later
annihilating our original particle.
So we see that the quantum mechanics of a single particle gives unphysical
results in a relativistic theory, and we need to consider rather quantizing the
field.

4 Field Quantization
Now consider again the Klein Gordon field described by the “coordinates”
φ(~x, t) and the canonical momenta π(~x, t), with the Hamiltonian density
1 1 1
H = π 2 + (∇φ)2 + m2 φ2 .
2 2 2
How do we quantize this? If we go back to the lattice, we would say
[Pi , qj ] = −iδij .
Our momenta π, however, are momentum densities, so the normalization we
should expect is d3 x[π(~x), φ(~x ′ )] = −i provided ~x ′ is in the integration
R

region, and 0 otherwise. This is what a Dirac delta was invented for, so
[π(~x), φ(~x ′ )] = −iδ 3 (~x − ~x ′ ).
Of course the coordinates commute with each other, as do the momenta:
[φ(~x), φ(~x ′ )] = 0, [π(~x), π(~x ′ )] = 0.
4
See, for example, Arfken, “Mathematical Methods for Physicists”, 2nd Ed., pp 373-
376.
615: Introduction Last Latexed: September 12, 2013 at 10:18 16

The classical mechanics of the wave equation is very simple, with solutions
corresponding to all 3-momenta ~k, with
q
φ(~x, t) ∝ ei~p·~x−iωt , with ω = ± p2 + m2 ,

so more generally the field can be expanded in terms of coefficients for each
momentum,
d3 p i~p·~x−iωt
Z
φ(~x, t) = e φ̃(~p).
(2π)3
Note that as φ(~x) is real (or, in quantum mechanics, hermitean), φ̃ is not
real but rather satisfies the condition φ† (~p) = φ(−~p). Thus we would do
better to write
Z
d3 p 1  i~
p·~
x−iωp~ t † −i~p·~
x+iωp~ t

φ(~x, t) = q ap~ e + ap~ e ,
(2π)3 2ωp~
√
which is automatically Hermitean. The factor 1/ 2ωp~ is introduced to set
the scale of a and a† conveniently, as we will explain in a minute. From π = φ̇
we gather that
d3 p ωp~ 
Z r 
π(~x, t) = −i ap~ ei~p·~x−iωp~ t − ap†~ e−i~p·~x+iωp~ t ,
(2π)3 2
At time t = 0 it may be more convenient, by changing the dummy variable
~p into −~p in the a† terms, to write these as
d3 p 1 
Z 
†
φ(~x, 0) = 3
q ap~ + a−~p ei~p·~x ,
(2π) 2ωp~

d3 p ωp~ 
Z r 
π(~x, 0) = −i ap~ − a†−~p ei~p·~x ,
(2π)3 2
Then
[φ(~x), φ(~x ′ )]
   
0
d3 p d 3 p′ 1
Z Z
′ ′
′ 
ei~p·~x+i~p ·~x
 3 ′ 
 [φ(~x), π(~x )]  =  iδ (~x − ~x )  =

q
(2π) 3 3
(2π) 2 ωp~ ω~p ′
[π(~x), π(~x ′ )] 0
[ap~ + a†−~p , a~p ′ + a†−~p ′ ]
 

×  −iω~p ′ [ap~ + a†−~p , a~p ′ − a†−~p ′ ] 

 

−ωp~ ω~p ′ [ap~ − a†−~p , a~p ′ − a†−~p ′ ]

615: Introduction Last Latexed: September 12, 2013 at 10:18 17

From the double Fourier transform of this, we see that

[ap~ + a†−~p, a~p ′ + a†−~p ′ ] = 0

[ap~ + a†−~p , a~p ′ − a†−~p ′ ] = −2(2π)3 δ 3 (~p + ~p ′ )
[ap~ − a†−~p , a~p ′ − a†−~p ′ ] = 0

or
[ap~ , a~p ′ ] = 0, [ap~ , a~p† ′ ] = (2π)3 δ 3 (~p − ~p ′ ), [ap†~ , a~p† ′ ] = 0.
Thus we have commuting operators for different ~p, and for each p~ we have a
set of ordinary harmonic oscillator raising and lowering operators, or at least
we would have if we were dealing with momenta discretized by working in a
finite box.
The Hamiltonian (at t = 0) can be written
1 1 1
Z  
H = d x π 2 (~x) + (∇φ)2 (~x) + m2 φ2 (~x)
3
2 2 2
( q
1
Z Z 3
dp
Z 3 ′
d p ωp~ ω~p ′  †

†

= d3 x − ap~ − a−~p a~
p ′ − a
−~p ′
2 (2π)3 (2π)3 2
2 ′ 
)
m − ~p · ~p †
 
†
 ′
+ q ap~ + a−~p a~p ′ + a−~p ′ ei(~p +~p)·~x
2 ωp~ ω~p ′
d3 p
(
1 ωp~ 
Z  
† †
= − a p
~ − a −~p a −~
p − a p
~
2 (2π)3 2
2 2 
)
m + p~ †

†

+ ap~ + a−~p a−~p + ap~
2ωp~
d3 p
( )
1
Z
† †
= ωp~ ap~ ap~ + ap~ ap~
2 (2π)3
d3 p 1
Z  
† †
= ωp~ ap~ ap~ + [ap~ , ap~ ]
(2π)3 2
3
dp 1 3
Z  
†
= ω p
~ ap a
~ p~ + δ (0)
(2π)3 2

Notice that the Hamiltonian separates, each spatial momentum component

~p decoupling from the others and entering as a simple harmonic oscillator.
However it is a bit disconcerting to have the zero point energy, 12 ωp~ , for all of
the infinite number of normal modes. As long as we avoid general relativity,
615: Introduction Last Latexed: September 12, 2013 at 10:18 18

and the coupling of this energy to gravitation, we can ignore this constant,
though infinite, contribution to the energy of every state in the system —
only energy differences have any effect. So we will drop this constant and
write Z
d3 p
H= ωp~ ap†~ ap~ .
(2π)3
We can find the possible states of the field theory by examining the com-
mution relations of H with ap~ and ap†~ . As

[a~p† ′ a~p ′ , ap~] = [a~p† ′ , ap~ ] a~p ′ = −(2π)3 δ 3 (~p ′ − p~ ) ap~,

and
[a~p† ′ a~p ′ , ap†~] = a~p† ′ [a~p ′ , ap†~ ] = (2π)3 δ 3 (~p ′ − p~ ) ap†~,
we have
d 3 p′
Z
[H, ap~ ] = 3
ω~p ′ [a~p† ′ a~p ′ , ap~ ] = −ωp~ ap~ ,
(2π)
Z
d 3 p′
[H, ap†~ ] = 3
ω~p ′ [a~p† ′ a~p ′ , a† ] = ωp~ ap†~ .
(2π)
If |ψi is a state with energy E, (H |ψi = E |ψi), then ap~ |ψi is a state
with energy E − ωp~ , unless ap~ |ψi = 0. Thus the lowest energy state of the
system must be the |ψi for which ap~ |ψi = 0 for all p~. This is called the
vacuum state, and has energy zero (after having dropped the constant term
from H). Thus we will describe this state as |0i, with H |0i = 0, ap~ |0i = 0
for all p~.
We can create other states from the vacuum state by applying raising
operators ap†~ , repeatedly, to the vacuum state. From their commutator with
H we see that the energy is

Hap†~1 ap†~2 |0i = [H, ap†~1 ]ap†~2 |0i + ap†~1 [H, ap†~2 ] |0i + ap†~1 ap†~2 H |0i
= ωp~1 ap†~1 ap†~2 |0i + ap†~1 ωp~2 ap†~2 ] |0i + 0
= (ωp~1 + ωp~2 )ap†~1 ap†~2 |0i

so the state ap†~1 ap†~2 |0i has energy (is an eigenstate of H with eigenvalue)
ωp~1 + ωp~2 . This is what a state of two noninteracting particles, each of mass
m, with momenta p~1 and p~2 respectively, should have.
615: Introduction Last Latexed: September 12, 2013 at 10:18 19

This is surely a hint that

N
!
ap†~i
Y
|0i
i=1

is a state of N particles of momenta p~1 , . . . , p~N , and we can get further ev-
idence by asking for the total momentum P~ of this state. The momentum
operator may be derived from translation invariance, as any continuous sym-
metry of a theory corresponds to a conserved quantity, which for translations
is the total momentum. We will discuss this relationship, called Noether’s
theorem, after we make a few more observations about the quantum mechan-
ics of a free scalar field. For the moment, let’s just accept that
~ R 3
~ x, t). Expanding that in terms of a and a† ,
P (t) = − d x φ̇(~x, t) ∇φ(~

d3 p′ ω~p ′ 
Z Z r 
P~ (0) = − d3 x(−i) a ~
p ′ − a †
−~p ′ ei~p ′ ·~
x
(2π)3 2
d3 p 1 
Z 
†
3
q ap~ + a−~p (i~p) ei~p·~x
(2π) 2ωp~
d3 p 1 
Z  
† †
= − a −~p − ap~ ap~ + a−~p ~p
(2π)3 2
Z
d3 p
= p~ a† ap~ ,
(2π)3 p~

where the integral over x has given us a (2π)3 δ 3 (~p +~p ′ ) which we have used to
do the d3 p′ integral, and in the last line we have dropped from the integrand
terms which are antisymmetric under p~ ↔ −~p (including d3 p ~pδ 3 (0).)
R

Notice that as ap~ |0i = 0 for all p~, P~ |0i = 0. Also, because

d 3 p′ ′ †
Z
[P~ , ap†~ ] = ~p a~p ′ (2π)3 δ 3 (~p − ~p ′ ) = p~ ap†~ ,
(2π)3

when P~ acts on the state ap†~1 ap†~2 |0i we get

P~ ap†~1 ap†~2 |0i = [P~ , ap†~1 ]ap†~2 |0i + ap†~1 [P~ , ap†~2 ] |0i + ap†~1 ap†~2 P~ |0i
= p~1 ap†~1 ap†~2 |0i + ap†~1 p~2 ap†~2 ] |0i + 0
= (~p1 + ~p2 )ap†~1 ap†~2 |0i
615: Introduction Last Latexed: September 12, 2013 at 10:18 20

so ap†~1 ap†~2 |0i is an eigenstate of P~ with total momentum p~1 + p~2 , and of
H with total energy ωp~1 + ωp~2 , just as one would expect for a state of two
noninteracting particles with momenta p~1 and p~2 respectively. Thus it is clear
that we can construct multiparticle states by applying a† ’s to the vacuum
state.
The normalization, however, is not what Peskin and Schroeder use, which
is instead
q q q
|~p i = 2ωp~ ap†~ |0i , |~p1 , ~p2 i = 2ωp~1 2ωp~2 ap†~1 ap†~2 |0i , . . .
for reasons they explain, which I will not repeat. (I will also switch from ωp~
to Ep~ to be consistent with them, though I don’t see why.) For 1-particle
states, this means
h~p |~q i = 2Ep~(2π)3 δ 3 (~p − ~q )
and
d3 p 1Z
1I1-part = |~p i h~p |
(2π)3 2Ep~
is the projection operator onto single particle states. This normalization is
Lorentz invariant, for the obviously (orthochronous5 ) Lorentz invariant
d4 p d3 p ∞
Z Z Z  
δ(p2 − m2 )Θ(p0 ) = dp 0
δ (p 0 2
) − Ep
2
~
(2π)3 (2π)3 0
d3 p d(p0 )2  0 2
Z Z ∞ 
2
= δ (p ) − E p
~
(2π)3 0 2p0
d3 p 1
Z
= .
(2π)3 2Ep~
Now that we understand ap~ and ap†~ as operators which annihilate or create
one particle of momentum ~p, we see that the field operator φ(~x) can
either create or annihilate a particle at ~x.
So far we have only considered these operators at a given time, as we
generally do in Hamiltonian mechanics, or the Schrödinger picture, where
the operators are fixed but the states transform with H ≡ i∂/∂t. In the
Heisenberg picture, the operators are time dependent, with
φ(xµ ) = φ(~x, t) = eiHt φ(~x, 0)e−iHt ,
5
Proper orthochronous Lorentz transformations are those which preserve the future
direction of time (orthochronous) and which do not reverse right and left hands (proper).
That is, Λ00 > 0 and det Λµν > 0.
615: Introduction Last Latexed: September 12, 2013 at 10:18 21

and similarly for all other operators O(t), with

∂
i O = [O, H].
∂t
This gives the equations of motion for the field operators
∂
φ(~x, t) = π(~x, t),
∂t
∂  
π(~x, t) = − −∇2 + m2 φ(~x, t),
∂t
or (∂µ ∂ µ + m2 ) φ(x) = 0.
There are two worthy things to note6 :
d3 p 1 µ µ 
Z  
φ(~x, t) = q ap~ e−ipµ x + ap†~ eipµ x  ,
(2π)3 2Ep~ p0 =Ep~

and ν ρ
φ(xµ ) = eiHt−iP ·~x φ(~0, 0)e−iHt+iP ·~x = eiPν x φ(~0, 0)e−iPρ x .
~ ~

Let us return to the question of causality which led us to question the

idea of using quantum mechanics for a fixed set of particles and argue that
field theory was necessary. To do things a little differently than the book,
let’s consider the two component free Klein-Gordon field, which, as problem
1 of homework 1 shows, is equivalent to a complex field φ satisfying (∂µ ∂ µ +
m2 )φ = 0. As there is no reality condition on φ, the coefficients of the
negative frequency modes are independent of the positive frequency ones,
and we write
Z
d3 p 1  −i(Et−~p·~x) † i(Et−~
p·~
x)

φ(~x, t) = q bp~ e + ap~ e ,
(2π)3 2Ep~

and φ† is, of course

d3 p 1 
Z 
† −i(Et−~
p·~
x) † i(Et−~
p·~
x)
φ (~x, t) = q ap~ e + bp~ e .
(2π)3 2Ep~

6
If ai and a†i satisfy [ai , a†j ] = δij , Mij is any c-number matrix, and O any function of
a and a† ,
† †
ea Ma O(a, a† )e−a Ma = O(e−M a, a† eM ).
615: Introduction Last Latexed: September 12, 2013 at 10:18 22

As φ† (~x, t) |0i is, up to normalization, a one-“b” particle state localized

at ~x at time t, we may ask what the probability of finding a one-“b”particle
state at ~x ′ at time t′ by evaluating
µ
D(x′ , xµ ) := h0| φ(~x ′ , t′ )φ† (~x, t) |0i (14)
Z
d 3 p′ 1 ′ ′ ′
= 3
q e−i(E~p ′ t − ~p · ~x )
(2π) 2E~p ′
d3 p 1
Z
× 3
q ei(Ep~ t − p~ · ~x) (2π)3 δ 3 (~p ′ − p~)
(2π) 2Ep~
d3 p 1 −iEp~ (t′ − t) i~p · (~x′ − ~x)
Z
= e e . (15)
(2π)3 2Ep~
As we saw earlier, the integration measure is invariant under proper or-
thochronous Lorentz transformations, and clearly it is invariant under
translations. So we may write D(x′ µ , xµ ) = D(x′ µ − xµ ) and recognize that
it is actually a function only of s2 = (t′ − t)2 − (~x ′ − ~x)2 and the sign of t′ − t
in the case that s2 > 0. If s2 > 0, with t′ > t, we may choose to work in a
frame with ~x ′ = ~x, t′ − t = s > 0,
d3 p 1 −iEp~ s
Z
′µ µ
D(x − x ) = e
(2π)3 2Ep~
Z ∞ 2
4π p dp −iEp~ s
= 3
e
(2π) 0 2Ep
1
Z ∞ √
= 2
dE E 2 − m2 e−iEs .
4π m
For large time, that is for large values of s, the integral is dominated7 by the
lowest value of E, and behaves like e−ims , as we might expect because the
lowest energy component of the state has energy m.
On the other hand, if x′ µ − xµ is spacelike, we can choose to work in a
frame with t′ = t, ~x ′ − ~x = rêz , in the z direction8 , and
d3 p 1 −ipz r
Z
′µ µ
D(x − x ) = e
(2π)3 2Ep~
7
Actually, for real s the integral doesn’t actually converge. See footnote 9 for a discus-
sion.
8
Notation: Although it is pµ rather than pµ whose spatial components are the usual
vector p~, when I use x, y and z as in pz , I will write them as subscripts as is customary in
Euclidean space. Thus pz = p3 . I hope this dubious notation doesn’t cause any difficulty.
615: Introduction Last Latexed: September 12, 2013 at 10:18 23

1 ∞ p2 π 2π
Z Z Z
= dp sin θdθe−ipr cos θ dφ
(2π)3 0 2Ep 0 0 0110
1 Z∞ p2 eipr − e−ipr 1010
= dp 1010
(2π)2 0 2Ep ipr 10
−i Z ∞
p 1010
= 2
dp √ 2 eipr 10
8π r −∞ p + m2 p

To evaluate this integral9 we note that for r > 0 we may add a semicircle in
the upper half complex p plane at |p| = ∞ to close the contour. Then we
may deform the contour to surround the cut along p = iy for
1
0
y ≥ m. Thus
0
1
0
1
−i
Z ∞ iy 0
1
0
1
µ
D(x′ − xµ ) = −i dy e−yr
0
1
q
8π 2 r m −i y2 − m2
0
1 p
−i ∞ iy 0
1
Z
+ i dy q e−yr 0
1
8π 2 r 2
i y −m
m 2
0
1
0
1
−i iy 0 m
1
Z ∞
= i dy q e−yr .
4π 2 r m 2
i y −m 2

While this does decrease exponentially (roughly as e−mr for large r), it is
certainly not zero (the integrand is positive everywhere), so we do not have
a zero probability of finding an “b” particle at a space-like separated point.
9
Again we see that the integral doesn’t quite converge. The oscillations which cause
this are thrown away in this argument at the ends of the arc at infinity. The justification
for treating these propagators as if they were well defined is to expect them to be convolved
with some wave packet which will smear out the oscillations. Another way to say this is
that the oscillations are an indication of delta functions which are irrelevant for nonzero
distance separations.
615: Introduction Last Latexed: September 12, 2013 at 10:18 24

We see that h0| φ(~x ′ , t′ )φ† (~x, t) |0i cannot repre-

sent the amplitude for creating a particle from the
vacuum at (~x, t) and have it propagate to (~x ′ , t′ ),
∆ x = ct
but rather something more complicated, including
virtual pair creation and annihilations between t t x
and t′ . In fact, we don’t even get zero if t′ is ear-
lier than t. As we will see later, causality in field x0
theory is related to the commutator of operators.
So we will now ask if the commutator of two fields
vanishes for spacelike separations.
For free fields it is clear that the commutator of two fields at any two times
is a c-number, and so we can evalate it by taking its vacuum-expectation value
(VEV):

[φ† (x), φ(y)] = h0| φ† (x)φ(y) |0i − h0| φ(y)φ†(x) |0i = D(x − y) − D(y − x),

were I have used the fact that the calculation of h0| φ(y)φ†(x) |0i is identical
to that for h0| φ† (y)φ(x) |0i, only replacing a and a† with b and b† . For
spacelike separations, however, x − y and y − x can be rotated into each
other, and as D is invariant under orthochronous Lorentz transformations,
which include rotations, the difference must vanish. Thus [φ† (x), φ(y)] = 0
for spacelike x − y. Of course for the complex field [φ(x), φ(y)] = 0 for all
x and y, as the a’s commute with both b and b† . We do, however, need to
be careful about [φ† (x), φ(y)] = 0 near xµ = y µ√ . For from the lagrangian
derived in your homework (with φ := (φ1 + iφ2 )/ 2)

L = ∂µ φ† ∂ µ φ − φ† φ,

we see that
δL
π(x) := = φ̇† (x),
δ φ̇(x)
and as we know that at equal times

[π(~x, t), φ(~y , t)] = −iδ 3 (~y − ~x)

we see that 
∂ † 
[φ (~x, t), φ(~y , t′ )] = −iδ 3 (~y − ~x).

∂t ′
t =t
So we see that we have singular behavior for xµ → y µ .
615: Introduction Last Latexed: September 12, 2013 at 10:18 25

More generally, for timelike separations we cannot map x−y to y −x with

an orthochronous Lorentz transformation, and in fact D(x − y) = D ∗ (y − x).
The commutator is
d3 p 1 µ µ 
Z  
†
h0| [φ (x), φ(y)] |0i = e−ip µ (x − y) −eip µ (x − y) 
(2π)3 2Ep~ 
p0 =Ep

d3 p 1
!
−ipµ (x − y)µ  −ipµ (x − y)µ 
Z  
= e − e (16)
(2π)3 2Ep~ 
p0 =Ep

p0 =−Ep

d4 p −ipµ (x − y)µ 2
Z
= 3
e δ(p − m2 )ǫ(p0 ) (17)
(2π)
I used the fact that the integration is symmetric under p~ to change the sign
of ~x − ~y in writing (16), and the property of the delta function that
X δ(x − xi )
δ(f (x)) =
xi ∋f (xi )=0
|f ′ (xi )|

to write (17). ǫ(x) = +1 for x > 0 and −1 for x < 0.

We can think of the p0 integral
Z
dp0 f (p0 )δ(p20 − E 2 )ǫ(p0 )

√ 1 f
(with E = p~ 2 + m2 ) as the integral around the poles of 2
dp0 .
2πi p0 − E 2
∞
If x0 > y 0 , the integral −∞
R
dp0 , distorted
as in contour Γ0 , may be closed in the lower p0
half plane with an infinite semicircle, and the −E p Γ0 +E p
delta function contributions can be considered
as residues of poles of (p2 − m2 )−1 . That is,
because
Γ+
1 1 1 1
Res 2 2
= , Res 2 2
=− ,
p0 =E p − m 2E p0 =−E p − m 2E
we can write the commutator as
d3 p dp0 −1 −ipµ (x − y)µ
Z Z
†
h0| [φ (x), φ(y)] |0i = e for x0 > y 0
(2π)3 Γ0 2πi p2 − m2
615: Introduction Last Latexed: September 12, 2013 at 10:18 26

The minus sign is because our contour runs clockwise rather than counter-
clockwise.
For x0 < y 0, we can close the contour Γ0 with a semicircle Γ− in the upper
half plane. This closed contour contains no poles or other singularities, so the
integral over Γ0 Γ− vanishes. Our function, integrated over Γ0 , is known as
S

the retarded Green’s function

DR (x − y) := θ(x0 − y 0) h0| [φ† (x), φ(y)] |0i
Z
d3 p Z dp0 −1 −ipµ (x − y)µ .
= e
(2π)3 Γ0 2πi p2 − m2
As implied by the “Green’s function” part of the name,

(∂x2 + m2 )DR (x − y) = −iδ 4 (x − y),

where ∂x2 = ∂µ ∂ µ involves differentiation with respect to x only, holding
y fixed. In the derivation in the book, one needs that the derivative of
the heaviside function Θ(t) is δ(t), and the derivative of δ(t) can only be
evaluated by differentiating whatever multiplies it,
δ ′ (x)f (x) ∼ −δ(x)f ′ (x).

DR (x − y) may also be written as the Fourier transform of D̃R (p),

d4 p −ipµ (x − y)µ , i
Z
DR (x−y) = D̃ R (p) e with D̃R (p) = , (18)
Γ0 (2π)4 p2 − m2
with the contour Γ0 indicating the contour for p0 for each fixed value of p~.
As with any Green’s function, there is an am- Γ−
biguity due to adding in solutions of the ho-
mogeneous equation. In terms of the Fourier
transforms, this ambiguity corresponds to p
0
−E p
choices of how the p0 contour deviates around
the poles. An important choice is called the Γ +E p
Feynman propagator, defined the same way
DR is in (18), except that the contour is de-
fined to pass below the pole at p0 = −Ep and
Γ+
above the one at p0 = −Ep , as Γ shown.
This choice is sometimes indicated by taking q
the Fourier transform to be
2 2
i/(p − m + iǫ), which has the poles at p0 = ± Ep~2 − iǫ = ±(Ep~ − iǫ′ ) for
615: Introduction Last Latexed: September 12, 2013 at 10:18 27

some infinitesimal but positive values of ǫ and ǫ′ . This places the poles just
above the negative real axis and just below the positive one, so dp0 can run
exactly on the real axis and pass to the sides of the poles Feynman requested.
Closing the contour with Γ+ in the lower half plane for x0 > y 0 or with Γ−
in the upper half plane for x0 < y 0 , we see that only the pole at p0 = Ep~
contributes for x0 > y 0, and that at p0 = −Ep~ for x0 < y 0 . The residues of
(p2 −m2 )−1 are +1/2Ep and −1/2Ep respectively, but the clockwise direction
of the contour Γ Γ+ gives an extra − sign in the x0 > y 0 case. Then we
S

have
DF (x − y) = θ(x0 − y 0)D(x − y) + θ(y 0 − x0 )D(y − x)
= θ(x0 − y 0) h0| φ† (x)φ(y) |0i + θ(y 0 − x0 ) h0| φ(y)φ†(x) |0i
=: h0| T φ† (x)φ(y) |0i . (19)
In the last expression, T is a kind of meta “time-ordering” operator, which
tells us to rewrite the operators following it in chronological sequence (right
to left increasing).
Thus far we have discussed only a free field — that is, as we have seen,
the solution to the equations of motion are all of the form that we have some
bunch of particles of various momenta, but each one has constant momentum,
and the number of particles is unchanged in time. This is elegant but not
very interesting — we need to include interactions to have interesting physics.
Unfortunately, interactions make the theory much more difficult to solve. The
simplest interaction our Klein-Gordon particles can have is with an external
classical source. In electromagnetism, this classical source is a 4-current j µ
as you considered for homework, which enters as an additional term −Aµ j µ
in the lagrangian. There are complications here, but for a real scalar field
the source is simpler, so that
1 1
L = (∂µ φ)2 − m2 φ2 + j(x)φ(x) =⇒ (∂ 2 + m2 )φ = j(x).
2 2
Thus the field φ(x) can be found using the retarded Green function
Z
φ(x) = φ0 + i d4 y DR (x − y)j(y).

The last section of Chapter 2 shows that the source creates particles of mo-
mentum p~ with a number expectation value given by the square of the fourier
transform q Z 
j̃( p~ 2 + m2 , ~p) = d4 y eip·y j(y) √

.
p0 = ~ 2 +m2
p
615: Introduction Last Latexed: September 12, 2013 at 10:18 28

You should read through this section, but I will not lecture on it.