Advanced Quantum Mechanics
A. W. Stetz
January 22, 2006
2
Contents
1 Introduction and Review 7
2 The Path Integral Formulation 13
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 The Feyman Path Integral . . . . . . . . . . . . . . . . . . . . 19
2.3 The free particle propagator . . . . . . . . . . . . . . . . . . . 20
2.4 Causality and the propagator . . . . . . . . . . . . . . . . . . 21
2.5 Generalizing the oneparticle propagator . . . . . . . . . . . . 24
2.6 Correlation functions . . . . . . . . . . . . . . . . . . . . . . . 25
3 Mathematical Appendix 29
3.1 Functional Diﬀerentiation . . . . . . . . . . . . . . . . . . . . 29
3.2 Functional Taylor series . . . . . . . . . . . . . . . . . . . . . 31
4 Introduction to Field Theory 33
4.1 Introduction to Continuum Mechanics . . . . . . . . . . . . . 35
4.2 Correlation functions . . . . . . . . . . . . . . . . . . . . . . . 40
4.3 The Feyman Propagator . . . . . . . . . . . . . . . . . . . . . 44
4.4 Interacting Fields . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.5 Momentum Space . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.6 Virtual Particles . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.7 A Word About Renormalization . . . . . . . . . . . . . . . . 56
5 Identical Particles and ManyParticle States 61
5.1 The Quantum Mechanics of Identical Particles . . . . . . . . 61
5.2 Boson States . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.3 More General Operators . . . . . . . . . . . . . . . . . . . . . 66
5.4 Field Operators . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.5 Field Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.6 Momentum Representation . . . . . . . . . . . . . . . . . . . 72
3
4 CONTENTS
5.7 Superﬂuids . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.7.1 Bogoliubov Theory . . . . . . . . . . . . . . . . . . . . 75
5.7.2 The Bogoliubov Transformation . . . . . . . . . . . . 77
5.8 Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.8.1 Inclusion of Spin . . . . . . . . . . . . . . . . . . . . . 84
5.8.2 The Fermi Sphere and the Debye Cutoﬀ . . . . . . . . 84
5.9 Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.9.1 Eﬀective ElectronElectron Interaction . . . . . . . . . 87
6 Second Quantization and Relativistic Fields 95
6.1 Introduction to Field Theory . . . . . . . . . . . . . . . . . . 95
6.2 Introduction to Continuum Mechanics . . . . . . . . . . . . . 97
6.3 Introduction to Second Quantization . . . . . . . . . . . . . . 101
6.4 Field Theory and the KleinGordon Equation . . . . . . . . . 110
6.5 The Propagator . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7 The Interaction Picture and the SMatrix 115
7.1 The Interaction Picture . . . . . . . . . . . . . . . . . . . . . 116
7.2 The S Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
7.3 The LSZ Reduction Scheme . . . . . . . . . . . . . . . . . . . 122
7.4 Correlation Functions . . . . . . . . . . . . . . . . . . . . . . 127
7.5 The Wick Expansion . . . . . . . . . . . . . . . . . . . . . . . 130
7.6 An Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
8 Cross Sections and Decay Rates 137
8.1 Classical Scattering . . . . . . . . . . . . . . . . . . . . . . . . 137
8.2 Quantum Scattering . . . . . . . . . . . . . . . . . . . . . . . 139
8.3 Phase Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
8.4 TwoParticle Scattering . . . . . . . . . . . . . . . . . . . . . 147
8.5 The General Case . . . . . . . . . . . . . . . . . . . . . . . . . 148
9 The Dirac Equation 151
9.1 The Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
9.2 Plane Wave Solutions . . . . . . . . . . . . . . . . . . . . . . 154
9.3 Charge Conjugation and Antiparticles . . . . . . . . . . . . . 156
9.4 Quantizing the Field . . . . . . . . . . . . . . . . . . . . . . . 161
9.5 The Lorentz Group . . . . . . . . . . . . . . . . . . . . . . . . 165
9.6 Spinor Representations . . . . . . . . . . . . . . . . . . . . . . 169
9.7 The Dirac Propagator . . . . . . . . . . . . . . . . . . . . . . 172
CONTENTS 5
10 The Photon Field 175
10.1 Maxwell’s Equations . . . . . . . . . . . . . . . . . . . . . . . 175
10.2 Quantization in the Coulomb Gauge . . . . . . . . . . . . . . 178
11 Quantum Electrodynamics 183
11.1 Gauge Invariance . . . . . . . . . . . . . . . . . . . . . . . . . 183
11.2 Noether’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . 185
11.3 Feynman’s Rules for QED . . . . . . . . . . . . . . . . . . . . 187
11.4 The Reaction e
−
+e
+
→µ
−
+µ
+
. . . . . . . . . . . . . . . 193
11.4.1 Trace Technology . . . . . . . . . . . . . . . . . . . . . 195
11.4.2 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . 195
11.5 Introduction to Renormalization . . . . . . . . . . . . . . . . 196
12 Weak Interactions 205
12.1 Parity Violation . . . . . . . . . . . . . . . . . . . . . . . . . . 208
12.2 Neutrinos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
12.3 Chirality, Helicity and Electrons . . . . . . . . . . . . . . . . 213
12.4 Lepton Number . . . . . . . . . . . . . . . . . . . . . . . . . . 215
13 The Electroweak Theory 219
13.1 Weak interaction of quarks . . . . . . . . . . . . . . . . . . . 219
13.2 Weak Isospin . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
13.3 The Intermediate Vector Bosons . . . . . . . . . . . . . . . . 222
13.4 The Feynman Rules . . . . . . . . . . . . . . . . . . . . . . . 223
14 Symmetry Breaking 225
14.1 A Simple Example . . . . . . . . . . . . . . . . . . . . . . . . 226
14.2 Goldstone Bosons . . . . . . . . . . . . . . . . . . . . . . . . . 228
14.3 A More Abstract Look . . . . . . . . . . . . . . . . . . . . . . 230
14.4 NonRelativistic Field Theory . . . . . . . . . . . . . . . . . . 233
14.5 Spontaneous Breaking of Local Symmetry . . . . . . . . . . . 237
15 LNAGS and the Standard Model 243
15.1 Spontaneously broken global SU(2) symmetry . . . . . . . . . 244
15.2 Quarks and Leptons . . . . . . . . . . . . . . . . . . . . . . . 247
6 CONTENTS
Chapter 1
Introduction and Review
The business of mechanics, both classical and quantum, is to predict the
future, i.e. given a system with some set of initial conditions, describe its
subsequent behavior. Quantum mechanics gives us a deceptively simple
looking solution to this problem; it’s called Schrodingers’ equation.
ψ(t) = e
−i
ˆ
H(t−t
)
ψ(t
) (1.1)
The equation is not very useful because
ˆ
H =
ˆ
T +
ˆ
V is the sum of two non
commuting operators.
1
There is the trivial case in which the state is an
eigenstate of the complete Hamiltonian. If
ˆ
Hψ
n
(x, t) = E
n
ψ(x, t) (1.2)
then
ψ
n
(t) = e
−iE
n
(t−t
)
ψ
n
(t
) (1.3)
These are called “stationary stattes” because there is no time dependence
except for the unobservable phase factor. If everything were that simple
there would be nothing to live for. The fact that such states exist, how
ever, suggests a way of calculating time dependence. We simply expand the
wave function in a series of energy eigenstates, each one of winch has trivial
time dependence. The following calculation is done in every introductory
quantum text. We deﬁne energy eigenstates (with no time dependence) as
ˆ
Hφ
n
(x) = E
n
φ(x) (1.4)
1
I will use “hats” on symbols when I want to call attention to the fact that they are
operators.
7
8 CHAPTER 1. INTRODUCTION AND REVIEW
Now for some t
< t,
ψ(x, t) =
n
c
n
φ(x)e
−iE
n
t
(1.5)
_
dx
φ
∗
m
(x
) ψ(x
, t
) = c
m
e
−iE
m
t
(1.6)
I have assumed that
_
dxφ
∗
m
(x)φ
n
(x) = δ
mn
(1.7)
Now reinsert (6) into (7)
ψ(x, t) =
_
dx
_
n
φ
∗
n
(x
)φ
n
(x)e
−iE
n
(t−t
)
_
ψ(x
, t
) (1.8)
The quantity in brackets is called the propagator.
K(x, t; x
, t
) ≡
n
φ
∗
n
(x
)φ
n
(x)e
−iE
n
(t−t
)
(1.9)
ψ(x, t) =
_
dx
K(x, t; x
, t
)ψ(x
, t
) (1.10)
Obviously, the propagator has the ability to “move the wave function along
in time.” Equation (9) is a complete inprinciple solution to the problem of
time development, but it’s still not very useful. For one thing, there are
only a few potintials for which we can ﬁnd exact eigenfunctions, and these
are usually exotic transcendental functions. The worst part of it is that
there is no way to make systematic approximations. Nontheless, the notion
of a propagator as deﬁned implicitly by (10) is very powerful. Everything
we will do in the next few chapters has to do ket in some way with ﬁnding
propagators.
In order to keep the discussion on familiar ground (for a little while), I
will redo the calculation above with a more abstract notation. This would
be a good time to review the diﬀerence between the Schrodinger picture and
the Heisenberg picture in elementary quantum mechanics. When we write
an equation like,
[ψ(t))
S
= e
−iHt
[ψ(0)) (1.11)
we are implicitly working the Schrodinger picture. The state vectors evolve
in time, whereas the operators do not. I have emphasized this by putting
the subscript S on the ket on the left. (There is no subscript on the other ket
9
for reasons that will soon be clear.) Operators in the Schrodinger picture
usually don’t depend on time, but of course, their expectation values do.
¸Q(t)) = ¸ψ
S
(t)[
ˆ
Q
S
[ψ
S
(t)) = ¸ψ(0)[e
i
ˆ
Ht
ˆ
Q
S
e
−i
ˆ
Ht
[ψ(0)) (1.12)
Deﬁne
ˆ
Q
H
(t) = e
i
ˆ
Ht
ˆ
Q
S
e
−i
ˆ
Ht
, (1.13)
so that
¸Q(t)) = ¸ψ(0)[
ˆ
Q
H
(t)[ψ(0)) (1.14)
All time dependence is now gathered into the operator. Since the two terms
on the right of (12) are identical for t = 0, we make no distinction between
Schrodinger states and Heisenberg states at this time. For this reason we
often write [ψ(0)) = [ψ). The question of exactly what time it was when
t = 0 (last week, last year?) just doesn’t come up.
If the Hamiltonian itself doesn’t depend on time, then the Heisenberg
operators satisfy their own version of Schrodinger’s equation.
d
ˆ
Q
H
(t)
dt
= i
_
ˆ
H,
ˆ
Q
H
(t)
_
(1.15)
Now look at an eigenstate of
ˆ
Q.
ˆ
Q
S
[q)
S
= q[q)
S
(1.16)
e
i
ˆ
Ht
ˆ
Q
S
e
−i
ˆ
Ht
e
i
ˆ
Ht
[q)
S
= qe
i
ˆ
Ht
[q)
S
(1.17)
Deﬁne
e
i
ˆ
Ht
[q)
S
≡ [q, t)
H
(1.18)
so that
ˆ
Q
H
[q, t)
H
= q[q, t) (1.19)
These are the “Heisenberg eigenstates.” There is something paradoxical
about them. We are used to saying the Heisenberg states do not change in
time, but in fact, the eigenstates of Heisenberg operators transform “back
wards” in time.
The Creed of Quantum Mechanics says, among other things, that to
every observable quantity there corresponds a Hermitian operator. The
eigenvalue spectrum of this operator corresponds to all possible results of
measurements of the observable, and the eigenstates themselves constitute
a complete set of states in terms of which any arbitrary state vector can be
expanded. This is not something we can prove mathematically. We believe
10 CHAPTER 1. INTRODUCTION AND REVIEW
in creeds, because life wouldn’t make sense otherwise. Certainly the position
of a particle is something that can be measured, so there must be a position
operator with eigenstates.
ˆ
X[x) = x[x) (1.20)
The eigenvalue spectrum is continuous, so we can take advantage of com
pleteness by writing
ˆ
I =
_
dx[x)¸x[ (1.21)
Students often ﬁnd this troubling, and for good reason. The variable x, in
addition to being an observable quantity, is also, along with time, one of the
independent variables in Schrodinger’s equation. And yet – we never think
of time as an observable, and the notion of a “time eigenstate” is bizarre!
The fact that position has this ambiguous dual role is one of the drawbacks
of Schrodinger theory. The problem largely disappears in the Heisenberg
picture, however, and for this and other reasons having to do with the role
of space and time, we will use the Heisenberg picture almost exclusively in
what follows.
Position eigenstates can be thought of as a notational device for switching
back and forth between the elementary formulation of quantum mechanics
in terms of wave functions and the more abstract formulation in terms of
bras and kets. For example
ψ(x, t) = ¸x, t[ψ)
H
= [x)ψ(t))
S
(1.22)
Notice that wave functions are really scalar products and as such do not
change between the Heisenberg and Schrodinger pictures. We can reverse
the procedure as follows:
[ψ(t))
S
=
_
dxψ(x, t)[x)
S
(1.23)
Now we can get back to the business of calculating propagators. Deﬁne
energy eigenstates as
ˆ
H[n) = E
n
[n) (1.24)
The symbol n stands for all the discrete quantum numbers required to spec
ify the eigenstate. We’ll assume that the states are complete and normalized
so that
ˆ
I =
n
[n)¸n[ (1.25)
e
−iH(t−t
)
= e
−iH(t−t
)
n
[n)¸n[ =
e
−iE
n
(t−t
)
[n)¸n[ (1.26)
11
This is sometimes called the “spectral representation” of the operator e
−i
ˆ
H(t−t
)
.
This gives us an alternative derivation of (8).
ψ(x, t) = ¸x[ψ(t)) = ¸x[e
−iH(t−t
)
[ψ(t
)) =
n
e
−iE
n
(t−t
)
¸x[n)¸n[ψ(t
))
=
_
dx
n
e
−iE
n
(t−t
)
¸x[n)¸n[x
)ψ(x
, t
) (1.27)
This is identical to (8) if we remember that ¸x[n) = φ
n
(x).
In order to take the next step toward the path integral formulation of
quantum mechanics, it will be necessary to calculate the propagator without
using energy eigenstates. So one last time
¸x[ψ(t)) =
_
dx
¸x[e
−iH(t−t
)
[x
)¸x
[ψ(t
)) (1.28)
Compare this with (10). We see the propagator lurking inside the integral.
K(x, t; x
t
) = ¸x[e
−iH(t−t
)
[x
) =
_
¸x[e
−iHt
)
_
_
e
iHt
[x
)
_
= ¸x, t[x
, t
)
(1.29)
This is the starting point for the path integral formulation in the next sec
tion. Before going on, we need one allimportant result.
ˆ
I =
_
dx[x)¸x[ =
_
dx e
iHt
[x)¸x[e
−iHt
=
_
dx¸x, t[x, t) (1.30)
Insert this in (29).
K(x, t; x
, t
) =
_
dx
¸x, t[x
, t
)¸x
, t
[x
, t
) =
_
dx
K(x, t; x
, t
)K(x
, t
; x
, t
)
(1.31)
This is the “composition property” of propagators. It implies that we can
always insert a complete set of Heisenberg position eigenstates using (30).
Physically it means that the probability for a particle to propagate from
(x
, t
) to (x, t) is equal to the probability of propagating from (x
, t
) to
(x
, t
) time the probability of propagating from (x
, t
) to (x, t) integrated
over all x
.
12 CHAPTER 1. INTRODUCTION AND REVIEW
Chapter 2
The Path Integral
Formulation
2.1 Introduction
In the usual formulation of quantum mechanics, one tries to calculate the
time dependence of the wave functions directly using Schrodinger’s equation.
The path integral formulation seeks to calculate the propagator K(x
f
, t
f
; x
◦
, t
◦
).
The procedure follows:
1. Draw all causal paths in the xt plane connecting (x
◦
, t
◦
) with (x
f
, t
f
).
By “causal” I mean that the paths must not loop back in time. There
are no other restrictions. The paths can be wildly unphysical.
2. Find the classical action S[x(t)] for each path x(t).
3. Perform the following sum.
K(x
f
, t
f
; x
◦
, t
◦
) = A
paths
e
iS[x(t)]/¯h
(2.1)
The constant A is a normalization factor, more about this later. I have
put in the ¯ h to illustrate the classical limit. The real question is how to do
the sum over paths. This will require learning some new math, functional
integration. First let’s do a brief review of classical mechanics.
The classical equations of motion can be derived from Hamilton’s prin
ciple. We start with the Lagrangian L = T − V , the kinetic energy minus
13
14 CHAPTER 2. THE PATH INTEGRAL FORMULATION
the potential energy. Calculate the action.
S[q, ˙ q] =
_
t
f
t
◦
L(q
1
q
n
, ˙ q
1
˙ q
n
)dt (2.2)
The q
i
’s are the generalized coordinates. They are functions of time, and
these functions q
i
= q
i
(t) collectively describe the path taken by the system.
Each choice of functions will yield a numerical value for the action. We say
that it is a functional of the coordinates and their derivatives. Hamilton’s
principle says that the actual path followed by the system is the one that
makes the action an extremum, i.e. a maximum or a minimum. Your favorite
classical mechanics text will now prove that this statement is equivalent to
a set of n EulerLagrange equations,
d
dt
∂L
∂ ˙ q
i
−
∂L
∂q
i
= 0 (2.3)
We will be much concerned with the propagation of free particles. Let’s
see how that works classically in one dimension We know the answer ahead
of time
x(t) = x
◦
+
(x
f
−x
◦
)
(t
f
−t
◦
)
(t −t
◦
) (2.4)
S =
_
t
f
t
◦
1
2
m˙ x
2
dt =
m
2
_
t
f
t
◦
_
x
f
−x
◦
t
f
−t
◦
_
2
dt =
m
2
(x
f
−x
◦
)
2
(t
f
−t
◦
)
(2.5)
In classical mechanics, the particle follows just this one path. In quan
tum mechanics, the particle somehow follows all paths (that don’t involve
propagation backwards in time), but as can be seen from (1), those with
action much diﬀerent from the classical action will tend to cancel one an
other through destructive interference. Terms close to the classical path add
coherently. Let’s put in some numbers. Suppose a classical particle travels
1 cm in 1 second. If it starts at the origin and follows the “right” path x = t
in the appropriate units
S =
_
1
0
1
2
m(1)
2
dt =
1
2
m. (2.6)
However, if the particle follows an unphysical path, say x = t
2
then
S =
_
1
0
1
2
m(2t)
2
dt =
2
3
m. (2.7)
The units of S must be ergseconds. If the mass were one gram, the “right”
and “wrong” action would diﬀer by only 1/6 ergsec., but remember that in
2.1. INTRODUCTION 15
(1) the action is divided by ¯ h = 10
−27
ergsec. Consequently the wrong path
is out of phase by 10
26
radians! On the other hand, if we are talking about
an electron with a mass of 9.11 10
−28
grams, then the wrong path is out
of phase by only 8.7 degrees. Clearly the classical particle must follow the
straight and narrow way, and how narrow the path must be is determined
by the extremely small size of ¯ h. To put it in less biblical terms, S ¸ ¯h
gives the classical limit.
The postulate (1) is usually attributed to Feynman
1
who claimed that
he discovered it in a moment if inspiration during a high school physics
class. What Feynman could intuit, the rest of us can prove, though the
proof will take many pages. The key idea is to discretize time. We will
divide the interval between t
◦
and t
f
into n + 1 equal slices bounded by
t
◦
< t
1
< t
2
< < t
n−1
< t
f
. We represent the path of the particle as a
set of n + 1 straightline segments connecting the spacetime points, i.e.
(x
◦
, t
◦
) →(x
1
, t
1
) →(x
2
, t
2
) → →(x
n
, t
n
) →(x
f
, t
f
)
We now use the composition property (1.31) n + 1 times.
¸x
f
, t
f
[x
◦
, t
◦
) =
_
dx
n
dx
n−1
dx
1
¸x
f
, t
f
[x
n
, t
n
)¸x
n
, t
n
[x
n−1
, t
n−1
) ¸x
1
, t
1
[x
◦
, t
◦
)
(2.8)
We will use the more compact notation,
¸x
f
, t
f
[x
◦
, t
◦
) =
_
n
i=1
dx
i
n
j=0
¸x
j+1
, t
j+1
[x
j
, t
j
), (2.9)
but remember that x
0
= x
◦
and x
n+1
= x
f
. We will eventually take the
limit n → ∞ to represent all smooth paths, but this is a strange limit in
which there are an inﬁnite number of integrations! We now work on the
shorttime propagator,
¸x
j+1
, t
j+1
[x
j
, t
j
) = ¸x
j+1
[e
−iHτ
[x
j
) ≈ ¸x
j+1
[(1 −iHτ[x
j
) (2.10)
The ﬁrst step makes use of (1.29). The second step takes advantage of the
fact that τ = t
j+1
− t
j
is very small. At this point I would like to insert
a complete set of momentum eigenstates, but there are a few technicalities
regarding normalization and boundary conditions that will repay a brief
degression.
1
Actually, most of these ideas were published earlier by Dirac. I don’t know if Feynman
was aware of his work.
16 CHAPTER 2. THE PATH INTEGRAL FORMULATION
The momentum eigenvalue equation in the Schrodinger picture is
ˆ
Pξ
p
(x) = p ξ(x) (2.11)
ξ
p
∝ e
ipx
We often make the eigenvalue spectrum discrete by imposing periodic bound
ary conditions, ξ(x) = ξ(x +L). in this case
ξ
n
=
1
√
L
e
ip
n
x
p
n
=
2πn
L
(2.12)
The normalization is chosen so that
_
L
0
dx ξ
∗
n
ξ
n
= δ
nn
(2.13)
If we don’t use periodic boundary conditions, the eigenvalue spectrum
is continuous. You can go back and forth between these two representations
with the following “cookbook” recipe.
ξ
n
=
e
ip
n
x
√
L
↔ξ
p
= e
ipx
n
↔
1
2π
_
∞
−∞
dp (2.14)
In the language of bras and kets
ˆ
P[p) = p[p) (2.15)
but we have to pay some attention to the normalization of [p). Let’s choose
¸x[p) = ξ
p
(x) ¸x[p
n
) = ξ
n
(x) (2.16)
We would also like ¸x
[x) = δ(x−x
). In terms of discrete states this requires
¸x
[x) =
n
¸x
[p
n
)¸p
n
[x) =
n
ξ
n
(x
)ξ
∗
n
(x) = δ(x −x
) (2.17)
So far, so good, When we make the transition to continuum states according
to (14) this becomes
¸x
[x) =
1
2π
_
dp¸x
[p)¸p[x) = δ(x −x
). (2.18)
The point is that whereas
_
dx[x)¸x[ =
ˆ
I,
2.1. INTRODUCTION 17
to be consistent we must have
1
2π
_
dp[p)¸p[ =
ˆ
I (2.19)
It is this last expression that we need for the shorttime propagator.
The Hamiltonian in (10) can be written
ˆ
H =
ˆ
P
2
2m
+
ˆ
V (x) (2.20)
Our strategy will be to use position and momentum eigenstates to replace
the momentum and position operators with their eigenvalues. When we
have done that, there will no longer be operators in our formalism. They
will have been replaced by classical or cnumber variables. Let’s do the
kinetic energy term ﬁrst.
¸x
j+1
[
ˆ
P
2
2m
[x
j
) =
1
2π
_
dp
j
¸x
j+1
[
p
2
j
2m
[p
j
)¸p
j
[x
j
) (2.21)
=
1
2π
_
dp
j
e
ip
j
(x
j+1
−x
j
)
p
2
j
2m
The potential energy term is simpler.
¸x
j+1
[V (x)[x
j
) = V (x
j
)¸x
j+1
[x
j
) = V (x
j
)
1
2π
_
dp
j
e
ip
j
(x
j+1
−x
j
)
(2.22)
Combine (10), (21) and (22).
¸x
j+1
, t
j+1
[x
j
, t
j
) =
1
2π
_
dp
j
e
ip
j
(x
j+1
−x
j
)
[1 −iτH(p
j
, x
j
)] (2.23)
Two points to notice:
• The Hamiltonian contains no operators. It is the classical Hamilto
nian.
• The symbol p
j
does not imply discrete momentum values. The sub
script means that [p
j
) is being used to help do the integrals associated
with the t
j
time slice.
Use the fact that τ is small so that [1 −iτH] ≈ e
−iτH
.
¸x
j+1
, t
j+1
[x
j
, t
j
) =
1
2π
_
dp
j
exp ¦i [p
j
(x
j+1
−x
j
) −τH(p
j
, x
j
)]¦ (2.24)
Substitute (24) into (9).
18 CHAPTER 2. THE PATH INTEGRAL FORMULATION
¸x
f
, t
f
[x
◦
, t
◦
) =
_
_
n
i=1
dx
i
_
_
_
n
j=0
1
2π
_
dp
j
exp ¦i [p
l
(x
j+1
−x
j
) −τH(p
j
, x
j
)]¦
_
_
=
_
_
n
i=1
dx
i
_
_
_
n
j=0
1
2π
_
dp
j
_
_
exp
_
i
n
l=0
[p
l
(x
l+1
−x
l
) −τH(p
l
, x
l
)]
_
(2.25)
The notation in the last line is potentially confusing. I have used three
independent indices, i, j, and l, since there are three indexed operations,
two products and a sum. The intent is to describe an equation with n
factors, each consisting of an integral over one of the x’s and one of the p’s.
In order to calculate a propagator, one evaluates (25) term by term.
The ﬁnal result will be an expression that depends on n. One then takes
the limit n → ∞ in the hope that the limit is welldeﬁned. Unfortunately,
there are only a few simple potentials for which this agenda can be carried
through exactly, so we are usually working with approximations. To this end,
note that something remarkable happens when we take the limit n → ∞.
Concentrate on the sum inside the exponential.
lim
n→∞
n
l=0
τ
_
p
l
_
x
l+1
−x
l
t
l+1
−t
l
_
−H(p
l
, x
l
)
_
=
_
t
f
t
◦
dt(p ˙ x −H) =
_
t
f
t
◦
L = S
(2.26)
Yes! L is the classical Lagrangian and S is the classical action. Equation
(25) can now be collapsed as follows:
lim
n→∞
n
j=0
1
2π
_
∞
−∞
dp
j
≡
_
D[p] (2.27)
lim
n→∞
n
i=1
_
dx
i
≡
_
D[x] (2.28)
¸x
f
, t
f
[x
◦
, t
◦
) =
_
D[x]D[p]e
iS
(2.29)
Equations (25) through (29) not only prove Feynman’s conjecture (1), they
also tell us what it means to sum over paths. They are the most general
formulas for evaluating path integrals.
2.2. THE FEYMAN PATH INTEGRAL 19
The mathematics before the limit n → ∞ is conventional math. Equa
tions like (25) can be done with ordinary calculus without much regard to
rigor. After the limit, the equations are in the new arena of functional in
tegration. Eventually we will do functional integrals just as we do ordinary
integrals – by following a set of rules we have memorized. In order to derive
the rules we must go back to (25), do what needs to be done, and then retake
the limit. Our ﬁrst example of this is the Feynman propagator, which we
will do in the next section.
2.2 The Feyman Path Integral
Most nonrelativistic Hamiltonians consist of a kinetic energy term, which
is some simple quadratic function of the momenta, and a potential energy
term, which is only a function of the coordinates. In the simplest case,
H = p
2
/2m + V (x). When this is true the momentum integrations in (25)
or (27) can be done once and for all. The resulting expression is called the
Feynman propagator.
Let’s go back to the integral in (24). In order to simplify the notation,
I will drop all the subscripts.
_
dp exp
_
i
_
p∆x −τ
p
2
2m
−τV
__
(2.30)
The gaussian integral comes in many diﬀerent forms. This is one of them.
The formula we need here is
_
∞
−∞
dp exp
_
i(−ap
2
+bp +c)
¸
=
_
π
ia
exp
_
ib
2
/4a +ic
¸
(2.31)
We use this to evaluate (30) and substitute the result back into (25),
¸x
f
, t
f
[x
i
, t
i
) =
_
_
n
i=1
dx
i
_
_
m
2πiτ
_n+1
2
exp
_
iτ
n
l=0
_
m
2
_
∆x
l
τ
_
2
−V (x
l
)
__
,
(2.32)
where ∆x
l
= x
l+1
−x
l
. In the limit that n →∞the multiplicative constant
is
N = lim
n→∞
_
m
2πiτ
_n+1
2
(2.33)
This diverges like ∞
∞
, but no matter, you have to get used to this sort of
thing. As you will see when we calculate things that are physically relevant,
this constant always factor out or disappears in some miraculous way. All
20 CHAPTER 2. THE PATH INTEGRAL FORMULATION
this is to say the the limit n →∞ is tricky. If you are trying to be rigorous,
the limit should be the last step in the calculation after all the troublesome
factors have eaten each other.
There is another bit of math that we have ignored so far. The integral
(31) doesn’t really converge to anything, because of the oscillating expo
nential. This, as the engineers would say, is not a bug but a feature. We
could make it converge unambiguously either by deforming the contour on
integration slightly into the complex plane or by adding a small convergence
factor in the exponential. The way we choose to do this will determine
whether the particle propagates forwards in time or backwards in time or
some combination of the two We will return to this issue presently.
So with all these caveats, let’s take the limit n → ∞ with our ﬁngers
ﬁrmly crossed behind our backs.
¸x
f
, t
f
[x
i
, t
i
) = N
_
D[x]e
iS
(2.34)
This is the Feynman path integral.
2.3 The free particle propagator
So far as I know, there are only two problems for which this last integral can
be done exactly, the free particle and the harmonic oscillator. Everything
else is approximation, but the approximations are all based on these two
cases. We need to study them carefully for this reason. Let’s start with the
free particle. Go back to (32) and set V = 0. The crux of the matter is this
integral,
_
dx
1
dx
n
exp
_
im
2τ
_
(x
f
−x
n
)
2
+ (x
n
−x
n−1
)
2
+ + (x
1
−x
i
)
2
¸
_
(2.35)
This is another example of the famous gaussian integral. Watch what hap
pens when we integrate over dx
1
using (31).
_
dx
1
exp
_
im
2τ
_
(x
2
−x
1
)
2
+ (x
1
−x
i
)
2
¸
_
=
_
iπτ
m
exp
_
im
4τ
(x
2
−x
i
)
2
_
(2.36)
After you have done n of these integral you will be left with
¸x
f
, t
f
[x
i
, t
i
) =
_
m
2πi(t
f
−t
i
)
exp
_
im(x
f
−x
i
)
2
2(t
f
−t
i
)
_
(2.37)
2.4. CAUSALITY AND THE PROPAGATOR 21
Exercise: Do the calculation and conﬁrm this result.
There are several points here to contemplate.
• We never had to take the limit n → ∞, because by the time we had
done all the integrals, there were no n’s left.
• By the same token, there are no divergent factors.
• Look at our example (5). The argument of the exponential is just the
classical action, so
¸x
f
, t
f
[x
i
, T
i
) = Ce
iS
(2.38)
• We could have gotten the same result directly from (1.6). The point is,
I guess, that for free particles, energy eigenstates are also momentum
eigenstates, so the sum in (1.6) it trivial.
• The propagator (37) seems to allow propagation backwards in time.
This is not surprising, since we could have replaced t → −t in any of
our derivations and gotten more of less the same results. As hinted
previously, this ambiguity in the direction of time is related to the
ambiguity associated with the convergence of integrals with oscillat
ing phase factors. This phenomenon is not peculiar to path integrals
or even to quantum mechanics. One encounters the same problem
calculating Greens’ functions for the propagation of electromagnetic
waves. The free particle propagator is just simple enough to allow us
to investigate these issues carefully in the next section.
2.4 Causality and the propagator
Equation (1.9) gives a simple formula for the freeparticle propagator in
terms of the energy eigenfunctions. In our current notation,
¸x
f
, t
f
[x
i
, t
i
) =
n
φ
∗
n
(x
i
)φ
n
(x
f
)e
−iE
n
(t
f
−t
i
)
(2.39)
In the case of a free particle, energy eigenfunctions are also momentum
eigenfunctions, since
ˆ
Hφ
n
=
p
2
n
2m
φ
n
. (2.40)
We can use the momentum eigenfunctions deﬁned in (12)and (14) for our
φ’s,
φ
n
(x) =
1
√
L
e
ip
n
x
→e
ipx
, (2.41)
22 CHAPTER 2. THE PATH INTEGRAL FORMULATION
where the arrow represents the continuum limit. Replace the sum in (39)
with an integral as in (14)
¸x
f
, t
f
[x
i
, t
i
) =
1
2π
_
dp e
i(x
f
−x
i
)
e
−i
p
2
2m
(t
f
−t
i
)
(2.42)
This can be integrated immediately using (31) to get (37), but for the pur
poses of discussing causality, it is better to get rid of the time and space
variables by means of a Fourier transform.
Propagators are closely related to Green’s functions. For example, we
can deﬁne a “retarded” Green’s function as
G
R
(x
f
, t
f
; x
i
, t
i
) ≡ −iθ(t
f
−t
i
)¸x
f
, t
f
[x
i
, t
i
) (2.43)
= −
i
2π
_
dp θ(t
f
−t
i
) exp
_
i
_
−
p
2
2m
(t
f
−t
i
) +p(x
f
−x
i
)
__
I put in the step function “by hand” so that the Green’s function is zero for
all t
i
> t
f
. The −i factor is a convention, but there is a good reason for it
as you will see eventually. In order to save ink, I will set x
i
and t
i
equal to
zero and abbreviate
G
R
(x
f
, t
f
; 0, 0) ≡ G(x, t) (2.44)
Eliminate the x variable with a Fourier transform.
_
dx e
−ip
x
G(x, t) = −iθ(t)e
−iE
p
t
≡ G(p
, t) (2.45)
Finally, take the Fourier transform with respect to time.
_
dt e
iωt
G(p, t) = −i
_
∞
0
e
−i(E−ω)t
dt ≡ G(p, ω) (2.46)
The last integral is not well deﬁned. There are several strategies for dealing
with it. The simplest is to add a small negative imaginary part to the energy,
E →E −iδ. Then
G(p, ω) =
1
ω −E +iδ
(2.47)
To see the signiﬁcance of the “fudge factor,” Fourier transform back into
conﬁguration space.
G(x, t) =
1
(2π)
2
_
dp
_
dω
e
i(px−ωt)
ω −E
p
+iδ
(2.48)
2.4. CAUSALITY AND THE PROPAGATOR 23
The ω integration can be done as a complex contour integral. For t < 0,
the integration contour closes in the upper half ω plane and the integral is
zero. When t > 0, the contour closes in the lower half plane and the integral
is just −2πi times the residue of the pole at ω = E − iδ. This displacing
the pole slightly below the real axis gives the retarded Green’s function, i.e.
the one that vanishes when t < 0. Obviously, we could get the advanced
Green’s function by displacing the pole upward.
Here is an assortment of points that you should absorb from the preced
ing derivation:
• Compare the formula for the propagator in conﬁguration space, equa
tion (37), with the Fourier transform of the Green’s function, (47).
Not only is (47) much simpler mathematically, its causality properties
are unambiguous and obvious. (The −i factor in (43) was inserted to
make (47) as simple as possible.) This is a recurring theme in ﬁeld
theory. Equations of motion and path integrals always start out in
conﬁguration space, but the results are simpler in momentum space.
• Integrals over momentum like (42) tend to be mathematically am
biguous because the integrand oscillates inﬁnitely rapidly as p → ∞.
They are physically ambiguous because they could describe propaga
tion backward in time just as well as propagation forward in time. We
can remove this ambiguity in several ways. In (43) we simply put in
a theta function. This makes causality explicit, but is clumsy to work
with. A better scheme is to give energies a small negative imaginary
part, or alternatively, just remember that the propagator in momen
tum space always has the form (47). This apparently ad hoc scheme
removes the physical ambiguity and the mathematical ambiguity si
multaneously, because they are really two manifestations of the same
problem.
• In nonrelativistic quantum mechanics based on Schrodinger’s equa
tion, cause and eﬀect always propagate “forward” in time. The point
is that the equation is second order in space and ﬁrst order in time, so
that if you replace t →−t in Schrodinger’s equation, you get a diﬀerent
equation. I emphasize this because relativistic theories are profoundly
diﬀerent; space and time coordinates (except for that damnable minus
sign) must be handled in the same way. The KlineGordon equation,
for example, is second order in space and time. The Dirac equation is
ﬁrst order in both variables. Such theories do not have a welldeﬁned
sense of past and future. The propagators must therefore account for
24 CHAPTER 2. THE PATH INTEGRAL FORMULATION
both forward and backward propagation. I’ll have much more to say
about this later.
• If you replace t → −it in Schrodinger’s equation, you get something
that looks like the equation for heat ﬂow,
∇
2
φ −a
∂φ
∂t
= 0 (2.49)
This is signiﬁcant for two reasons. First, we all know that heat ﬂows
from somewhere hot to somewhere colder. The second law of ther
modynamics deﬁnes an unambiguous direction of time. Schrodinger’s
equation has this kind of causality built into it. Second, the trans
formation t → −it is called a Wick rotation. It is a profound and
puzzling fact that the Wick rotation always turns quantum mechanics
into thermodynamics. You have just seen one example of this. There
are more to follow.
• You doubtless learned in Middle School that the Green’s function is
the ﬁeld due to a unit impulse function. You should convince yourself
that G(x, t) is no diﬀerent, i.e.
_
ˆ
H −i
∂
∂t
_
G(x, t) = −δ(x)δ(t) (2.50)
2.5 Generalizing the oneparticle propagator
We have been working on the propagator
¸x
f
, t
f
[x
i
, t
i
) = ¸x
f
[e
−iH(t
f
−t
i
)
[x
i
) (2.51)
Remember that the states on the left are timedependent position
eigenstates in the Heisenberg picture. The states on the right are
timeindependent Schrodinger states in which there is one particle that
happens to be at x
i
or x
f
. Let’s think through a series of generaliza
tions. First, we replace x with a generalized coordinate q. We can
measure this coordinate, so by the Central Dogma of Quantum Me
chanics, there must be some corresponding operator, eigenvalues and
eigenstates.
ˆ
Q[q) = q[q) (2.52)
2.6. CORRELATION FUNCTIONS 25
We could repeat all our derivations with q replacing x and get
¸q
f
, t
f
[q
i
, t
i
) = ¸q
f
[e
−iH(t
f
−t
i
)
[q
i
) (2.53)
Now that we have introduced one generalized coordinate, there is no
reason why we can’t have N of them. The classical Hamiltonian is
H = K(p
1
p
N
) +V (q
1
q
N
) (2.54)
where
p
j
=
∂L
∂ ˙ q
j
(2.55)
Again, repeat all the derivations. This time replace the single integrals
with multiple integrals
_
dp
j
→
_
_
dp
j1
dp
jN
(2.56)
_
dq
j
→
_
_
dq
j1
dq
jN
This does not introduce any additional problems.
In many respects we seem to be doing classical Langrangian mechanics
with N degrees of freedom. In order to do ﬁeld theory, however, we
must pass to the limit of continuum mechanics, which is to say, me
chanics with an inﬁnite number of degrees of freedom. In this case the
coordinates q
j
are replaced with displacement ﬁelds φ(x). The moti
vations for doing this and the mathematics involved are the subjects
of the next chapter.
2.6 Correlation functions
It turns out that many of the quantities that will be of interest to us have
the form
¸F[T
_
ˆ
Q(t
1
)
ˆ
Q(t
2
)
ˆ
Q(t
n
)
_
[I) (2.57)
The
ˆ
Q operators were explained in the previous section. The states [I) and
¸F[ are shorthand notation for the initial and ﬁnal states. The notation
T[ ] introduces the timeordered product. It means that the operators in
the brackets must be arranged so that operators evaluated at earlier time
26 CHAPTER 2. THE PATH INTEGRAL FORMULATION
always stand to the right of operators evaluated at later times. for example,
in the case of a single particle,
¸F[T[
ˆ
Q(t
1
)
ˆ
Q(t
2
)][I) = ¸q
f
, t
f
[
ˆ
Q(t
1
)
ˆ
Q(t
2
)[q
i
, t
i
)θ(t
1
−t
2
) (2.58)
±¸q
f
, t
f
[
ˆ
Q(t
2
)
ˆ
Q(t
1
)[q
i
, t
i
)θ(t
2
−t
1
)
The ±sign anticipates some arcane math that will be necessary when we deal
with fermion operators, which anticommute. For the time being, this will
be a plus sign. As a mnemonic for remembering (58), recall that quantum
mechanics reads like Hebrew, from right to left, so that t
f
> t
2
≥ t
1
> t
i
.
I will anticipate the next section by saying that we will usually evaluate
(57) in the limits, t
i
→−∞and t
f
→+∞. This has the eﬀect of projecting
out the ground state or lowestenergy state of the system, so that we are
really concerned with
¸Ω[T
_
ˆ
Q(t
1
)
ˆ
Q(t
2
)
ˆ
Q(t
n
)
_
[Ω), (2.59)
where [Ω) and ¸Ω[ are the initial and ﬁnal ground states. I would prefer to
do this one step at a time. First I will show how to evaluate (57) , and in a
later section, I’ll show the signiﬁcance of the limits t
i
→−∞and t
f
→+∞.
We start by evaluating (57) with a single operator. Transforming it into
the Schrodinger picture gives
¸q
f
, t
f
[
ˆ
Q(t)[q
i
, t
i
) = ¸q
f
[e
−iH(t
f
−t)
ˆ
Qe
−iH(t−t
i
)
[q
i
). (2.60)
Now insert a complete set of states just to the right of
ˆ
Q and take advantage
of the fact that
ˆ
Q[q) = q[q).
=
_
dq q ¸q
f
[e
−iH(t
f
−t)
[q)¸q[e
−iH(t−t
i
)
[q
i
) =
_
dq q(t) ¸q
f
, t
f
[q, t)¸q, t[q
i
, t
i
)
In this last expression we have two ordinary propagators ¸q
f
, t
f
[q, t) and
¸q, t[q
i
, t
i
). Each one could be written as in (32) or (34). This is just as
if we were calculating ¸q
f
, t
f
[q
i
, t
i
) and stuck in one “extra” time slice at t
between t
i
and t
f
. Repeating the steps that led to (34) we have
¸q
f
, t
f
[
ˆ
Q(t)[q
i
, t
i
) =
_
D[q]q(t)e
iS
(2.61)
Exercise: Verify this. Pay careful attention to the normalization factor.
2.6. CORRELATION FUNCTIONS 27
This procedure can be repeated as many times as we like, so
¸q
f
, t
f
[T
_
ˆ
Q(t
1
)
ˆ
Q(t
2
)
ˆ
Q(t
n
)
_
[q
i
, t
i
) =
_
D[q]q(t
1
)q(t
2
) q(t
n
)e
iS
(2.62)
Notice that there is a timeordering operator on the left side of the equation
and none on the right. One way to look at this is that the q’s commute,
so the order in which they are written makes no diﬀerence. A better way
to look at it is that when we insert sets of complete sets as in (60), we
will always put them in in such a way that t
f
> t
n
> > t
2
> t
1
> t
i
(in order to prevent backwards propagation in time). Thus the q(t
i
)’s are
automatically time ordered.
Path integrals are notoriously diﬃcult to do, so what hope do we have of
evaluating an integral like (62) with all those additional q(t)’s? The answer
is that if we can evaluate it exactly without the q’s, then we can do it with
the q’s almost trivially! If we can get an approximate answer without the
q’s, then to the same degree of approximation, we can evaluate it with the
q’s. In order to understand how this trick works, look at the simpler integral
_
∞
−∞
dq q
n
e
−
1
2
aq
2
We know that
_
∞
−∞
dq e
−
1
2
aq
2
+Jq
=
_
2π
a
_1
2
e
J
2
/2a
Diﬀerentiating this n times with respect to J,
_
∞
−∞
dq q
n
e
−
1
2
aq
2
+Jq
=
d
n
dJ
n
_
∞
−∞
dq e
−
1
2
aq
2
+Jq
=
d
n
dJ
n
_
2π
a
_1
2
e
J
2
/2a
Finally
_
∞
−∞
dq q
n
e
−
1
2
aq
2
= lim
J→0
d
n
dJ
n
_
2π
a
_1
2
e
J
2
/2a
(2.63)
You see the point? Integration is hard. Diﬀerentiation is easy. Using the “J
trick,” we can do diﬃcult integrals by diﬀerentiating!
Now go back to (???) and put in the J.
z[J] ≡ ¸q
f
, t
f
[q
i
, t
i
)
J
≡
_
D[q] exp
_
i
_
t
f
t
i
dt [L(t) +J(t)q(t)]
_
(2.64)
I can always get the propagator by taking the limit J → 0, but in the
meantime, I can put as many q’s as I like in the integrand by diﬀerentiating
28 CHAPTER 2. THE PATH INTEGRAL FORMULATION
(64) with respect to J. z[J] is called a generating function for this reason.
This is more than just a trick as you will see later on In some theories J has
a clear physical interpretation as a source of particles.
Before we can use (64) we must dispose of a technical diﬃculty. In (63),
J is a parameter, In (64) it’s a function of t buried inside a deﬁnite integral.
This requires a new technique called functional diﬀerentiation, I have put
this material in an appendix labelled Chapter 3, to which I will be referring
from time to time. This would be a good time to study it, We can use (3.5)
to get our ﬁnal result:
¸q
f
, t
f
[T
_
ˆ
Q(t
1
)
ˆ
Q(t
2
)
ˆ
Q(t
n
)
_
[q
i
, t
i
)
=
_
1
i
_
n
δ
n
z[J]
δJ(t
1
)δJ(t
2
) δJ(t
n
)
¸
¸
¸
¸
J=0
(2.65)
Chapter 3
Mathematical Appendix
3.1 Functional Diﬀerentiation
Consider the expression
F(J(τ))
By this we understand that τ is the independent variable and F is a function
of a function of τ. The familiar rules of diﬀerentiation apply, so for example,
dF =
∂F
∂J
dJ
dτ
dτ
These rules fail when there is buried inside F some deﬁnite integral in which
τ appears as the variable of integration. This is the case in equation (2.64),
which I repeat here for reference.
¸q
f
, t
f
[q
i
, t
i
)
J
≡
_
D[q] exp
_
i
_
t
f
t
i
dt [L(t) +J(t)q(t)]
_
(3.1)
We call a expression of this sort a functional of J and write it F[J]. We
can’t diﬀerentiate this with respect to J, but we can do something similar
called functional diﬀerentiation.
Suppose, for example that the function f = f(J(τ)) is a function of J,
which itself is a function of τ. The function f is turn is buried inside an
integral, so
F[J] ≡
_
f(J(τ))dτ (3.2)
Deﬁne the variation of F as follows.
1
Let η(τ) be any old wellbehaved
1
This deﬁnition is taken from Goldstein, Classical Mechanics, 3’rd ed. Section 2.3
29
30 CHAPTER 3. MATHEMATICAL APPENDIX
function of τ, and deﬁne
J(τ, α) = J(τ) +αη(τ)
∂[F(α)]
∂α
=
_
∂f
∂J
∂J
∂α
dτ =
_
∂f
∂J
ηdτ (3.3)
The variation is then
δF ≡
_
∂F
∂α
_
δα = F[J +δη] −F[J] (3.4)
Where δα is some very small value of α, and δη = ηδα. Combine (3.3) and
(3.4)
δF =
_
∂f
∂J
δη dτ (3.5)
This is usually written
F[J +δη] −F[J] =
_
δF[J]
δJ(τ)
δη dτ (3.6)
Equation (3.6) implicitly deﬁnes the functional derivative
δF[J]
δJ(τ)
and also
gives us an algorithm for calculating it. As an example, take a simpliﬁed
form of (2.54)
F[J] = exp
_
i
_
τ
f
τ
i
dτ[L +J(τ)q(τ)]
_
(3.7)
F[J+δη] = exp
_
i
_
τ
f
τ
i
dτ[L +J(τ)q(τ) +δη(τ)q(τ)]
_
= F[J]e
i
R τ
f
τ
i
δη(τ)q(τ)dτ
δF = iF[J]
_
τ
f
τ
i
q(τ)δη(τ)dτ
δF[J]
δJ(τ)
= iq(τ)F[J] (3.8)
This is a good time to point out that the functional derivative of a functional
is a function.
Now that you have the idea, you can take the shortcut. (This is some
times used as a deﬁnition of the functional derivative.)
δF[J(τ)]
δJ(τ
)
= lim
→0
F[J(τ) +δ(τ −τ
)] −F[J(τ)]
(3.9)
The notation F[J(τ)] means that τ is the variable of integration.
3.2. FUNCTIONAL TAYLOR SERIES 31
We can use the shortcut to evaluate a more diﬃcult and important case
from a later chapter. The following expression occurs naturally when we
calculate correlation functions using the path integral formalism.
F[J] = −
i
2
_
d
4
x d
4
yJ(x)D(x −y)J(y) (3.10)
All you need to know about this is that x and y are 4vectors and D(x) =
D(−x). We use functional diﬀerentiation to pull D(x − y) out of the func
tional. To do this, we need to evaluate the expression,
δ
2
F[J]
δJ(x
)δJ(y
)
(3.11)
The ﬁrst derivative is found as follows:
δF
δJ(x
)
= lim
→0
_
1
__
−
i
2
_
d
4
xd
4
y[J(x) +δ
(4)
(x −x
)]D(x −y)
[J(y) +δ
(4)
(y −x
)] +
i
2
_
d
4
xd
4
yJ(x)D(x −y)J(y)
_
= −i
_
d
4
xD(x −x
)J(x)
Notice that both J’s are incremented by the term. The last step is easier.
δ
2
F
δJ(x
)δJ(y
)
= −iD(x
−y
)
The primes have done their job of distinguishing the independent variables
from the variables of integration. We don’t need them anymore, so let’s
write it
δ
2
F
δJ(x)δJ(y)
= −iD(x −y) (3.12)
We diﬀerentiated with respect to J(x) and J(y), therefore the resulting
function must be a function of x and y.
3.2 Functional Taylor series
“When in doubt, expand in a Taylor series.” This is good advice with
functionals as well as ordinary functions This is particularly true, since most
path integrals cannot be evaluated exactly, and we need some way to make
systematic approximations.
32 CHAPTER 3. MATHEMATICAL APPENDIX
As in (3.2) we can write
F[J(α)] =
_
f[J(τ) +αη(τ)]dτ (3.13)
This time we will expand F[J(α)] in an ordinary power series in α and set
α = 1 at the end of the calculation.
F[J(α)] = F[J(0)] +
d
dα
F[J(α)][
α=0
+
1
2
d
2
dα
2
F[J(α)][
α=0
+ (3.14)
Repeating the steps from (3.3) – (3.6)
d
dα
F[J(α)][
α=0
=
_
∂f
∂J
ηdτ =
_
δF[J]
δJ(τ)
η(τ)dτ (3.15)
d
2
dα
2
F[J(α)] =
_
dτ
δ
J(τ)
η(τ
)
_
dτ
δF[J]
δJ(τ
)
η(τ
) (3.16)
Collecting these terms and setting α = 1 gives
F[J +η] = F[J] +
_
δF[J]
δJ(τ)
η(τ)dτ +
1
2
_ _
dτdτ
η(τ)η(τ
)
δ
2
F[J]
δJ(τ)δJ(τ
)
+
(3.17)
Chapter 4
Introduction to Field Theory
The formalism I have employed so far tries to look as much as possible like
classical Lagrangian mechanics with a ﬁnite number of degrees of freedom.
Our strategy has been to get rid of operators and replace them with classical,
cnumber variables. These variables commute with one another, which as a
great simplifying feature. They are embedded in the path integral formalism,
which is anything but simple, but most of the time this formalism “lurks
in the background.” We can construct Lagrangians, calculate interactions,
and discuss symmetries without actually doing path integrals. So long as
we work with a ﬁnite number of degrees of freedom, however, we can never
accommodate the creation and annihilation of particle. There is no way to
change the number of q’s in our Lagrangian. To do this, we need to replace
the mechanics of discrete masses with the mechanics of continuous media.
The motivation for this is explained nicely in Zee’s book in the section about
the mattress, which I will paraphrase.
Imagine that space is like a rubber sheet. If I put a bowling ball on
the sheet, it will create a depression, and nearby objects will roll into it.
This is an imperfect analogy for an attractive potential. We could describe
the attraction in one of two ways: we could say that there is an attractive
potential between any pair of pointlike masses, or we could introduce a
continuous variable, φ(x, y) which describes the displacement of the sheet
as a function of position. Such a continuous displacement variable is a ﬁeld in
the strict mathematical sense: it assigns a numerical value (or set of values)
to each point in space. The quantum mechanics of such ﬁelds is called
quantum ﬁeld theory. Now suppose that instead of using a bowling ball I
jump up and down on the sheet. The sheet will oscillate in response. My
activity becomes a source of energy, which propagates outward in the form
33
34 CHAPTER 4. INTRODUCTION TO FIELD THEORY
of waves. This is the rubbersheet analogy to the propagation of particles.
This analogy can easily be misleading. For one thing, I don’t want you
to think we are doing general relativity. The rubber sheet is not intended as
an analogy for ordinary spacetime as it is often used in explaining general
relativity. The ﬁeld φ(x, y) describes a displacement, and I know you want
to ask, “Displacement of what?”
The same question comes up in classical electromagnetic theory. When
an electromagnet wave is propagating through space, what is waving? Folks
in the 19’th century thought it must be some sort of mechanical medium,
which they called the ether. According to the textbooks, Michaelson and
Morley proved that wrong with their famous interferometer. But just saying
that the ether does’t exist doesn’t answer the question, it just makes it
impossible to answer! Let’s bite the bullet and agree for the purposes of this
course that space is pervaded by a medium, which for lack of a better name,
we will call the ether. Well, actually the ethers. Each species of particle
corresponds to a set of vibrations in it’s own speciﬁc ether. Electrons are
all vibrations in the electron ether, etc. Spacetime points in the ether can
be labelled with Lorentz fourvectors or (x, t) as usual, and these points
obey the usual rules for Lorentz transformations. This much is required
by the MM experiment. Ordinary bulk media have elastic properties that
are described by two parameters, the density and Young’s modulus. These
parameters are not themselves relevant to our formalism, but their ratio
gives the velocity of propagation, which is what we really care about.
I am fond of saying, “When correctly viewed, everything is a harmonic
oscillator.” Now you see that this is profoundly true. Each point on the
rubber sheet or ether acts like a harmonic oscillator! Quantum ﬁeld theory
is a theory about harmonic oscillators.
Well – I have to modify that slightly. If each point on the sheet be
haved like a simple harmonic oscillator with a quadratic potential, the waves
propagating on the sheet would never interact. The principle of linear su
perposition would hold everywhere. This is a theory of free particles. If
our theory is to describe interactions, then we must modify the potential
so that it becomes anharmonic. Unfortunately, the anharmonic oscillator
cannot be solve exactly in quantum mechanics. (If you think of a way to
do it, tell me and I’ll be famous.) We have to resort to approximations,
and here is where the path integral formalism really becomes a help rather
than a complication. It makes possible a systematic and understandable
perturbation theory that, with enough patience, can be carried out to any
degree of accuracy desired.
There is an alternative way of dealing with the creation and annihilation
4.1. INTRODUCTION TO CONTINUUM MECHANICS 35
of particles. It is the older way, sometimes called canonical quantization
or second quantization. The path integral formalism, seeks to banish all
operators from the theory. Second quantization goes in the other direc
tion. It turns the wave functions themselves into operators by imbedding
creation and annihilation operators into them; but they are the raising and
lowering operators of the harmonic oscillator! The universe, according to
second quantization, is an inﬁnity of harmonic oscillators. This approach is
complementary to path integrals in other ways as well. We need to master
both.
Continuum mechanics is not covered in most graduate mechanics classes.
There is a good discussion in the last chapter of Goldstein, but we never
make it that far. What follows is a brief introduction.
4.1 Introduction to Continuum Mechanics
The rules of continuum mechanics are derived by starting with a system
with a ﬁnite number of degrees of freedom and then passing to the limit in
which the number becomes inﬁnite. Let’s do this with the simplest possible
system, a long chain of masses connected by springs. It’s a onedimensional
problem. The masses can only oscillate along the chain. We will use ϕ
i
,
the displacement of the ith particle from its equilibrium position, as the
generalized coordinate. The Lagrangian is constructed in the obvious way.
T =
1
2
i
m¨ ϕ
2
i
(4.1)
V =
1
2
i
k(ϕ
i+1
−ϕ
i
)
2
(4.2)
L = T −V =
1
2
i
a
_
m
a
˙ ϕ
2
i
−ka
_
ϕ
i+1
−ϕ
i
a
_
2
_
(4.3)
The equilibrium separation between masses is a. The spring constant is k.
The EulerLagrange equations of motion are obtained from
d
dt
∂L
∂ ˙ ϕ
i
−
∂L
∂ϕ
i
= 0 (4.4)
If there are N masses, then there are N coupled equation of this sort. They
look like
m
a
¨ ϕ
i
−ka
_
ϕ
i+1
−ϕ
i
a
2
_
+ka
_
ϕ
i
−ϕ
i−1
a
2
_
= 0 (4.5)
36 CHAPTER 4. INTRODUCTION TO FIELD THEORY
We need diﬀerent parameters to describe the continuum limit:
m/a →µ mass per unit length
ka →Y Young’s modulus
The index i points to the ith mass, and ϕ
i
gives its displacement. In the
continuum limit, the index is replaced by the coordinate x. In elementary
mechanics, x would be the displacement of a particle. Here ϕ(x) is the
displacement of the string at the point x. In the continuum limit
ϕ
i+1
−ϕ
i
a
→
ϕ(x +a) −ϕ(x)
a
→
dϕ
dx
L →
1
2
_
dx
_
µ ˙ ϕ
2
−Y
_
dϕ
dx
_
2
_
≡
_
dxL(ϕ, ˙ ϕ) (4.6)
The last integral implicitly deﬁnes the Lagrangian density . The continuum
version of the EulerLagrange equation is
1
d
dt
_
_
∂L
∂
_
dϕ
dt
_
_
_
+
d
dx
_
_
δL
∂
_
dϕ
dx
_
_
_
−
∂L
∂ϕ
= 0 (4.7)
Use the Lagrangian density from (6) in (7)
∂
2
ϕ
∂x
2
=
_
µ
Y
_
d
2
ϕ
dt
2
(4.8)
(4) and (5) represent a set of N coupled equations for N degrees of freedom.
(7) is one equation for an inﬁnite number of degrees of freedom. In this
sense, continuum mechanics is much easier that discrete mechanics.
(8) should remind you of the equation for the propagation of electromag
netic waves.
_
∂
2
ϕ
∂x
2
_
+
_
∂
2
ϕ
∂y
2
_
+
_
∂
2
ϕ
∂z
2
_
=
1
c
2
_
∂
2
ϕ
∂t
2
_
As you know, photons are massless particles. Notice that a string of massive
particles yields a wave equation that when quantized describes the propa
gation of massless particles. (With a diﬀerent velocity, of course.) This is
worth a brief digression.
1
See Goldstein for a derivation of this important equation.
4.1. INTRODUCTION TO CONTINUUM MECHANICS 37
What does it mean to say that a wave function describes the propagation
of a particle of a particular mass? The wave function ψ = e
i(kx−ωt)
might
describe a wave in classical E&M, or a massive particle in nonrelativistic
or relativistic quantum mechanics. The question is, what is the relation
between k and ω? The relationship between the two is called a dispersion
relation. It contains a great deal of information. In the case of EM waves
in vacuum, k = ω/c. Frequency and wave number are simply proportional.
This is the hallmark of a massless ﬁeld. The velocity is the constant of
proportionality, so there can only be one velocity. In Schrodinger theory
¯h
2
k
2
2m
= ¯hω (4.9)
The relationship is quadratic. The relativistic wave equation for a spinzero
particle is called the KlineGordon equation.
_
∇
2
−
1
c
2
∂
2
∂t
2
_
ϕ −
m
2
c
2
¯h
2
ϕ = 0 (4.10)
The dispersion relation is
(c¯hk)
2
+m
2
c
4
= (¯ hω)
2
, (4.11)
or in other words, p
2
c
2
+ m
2
c
4
= E
2
. All these equations can be obtained
from (7) with the appropriate Lagrangian density. They are all three
dimensional variations of our “waves on a rubber sheet” model. What does
this have to do with the particle’s mass? It’s useful to plot (9) and (11), i.e.
plot ω versus k for small values of k. In both cases the curves are parabolas.
This means that in the limit of small k, the group velocity,
v
group
=
dω
dk
≈
¯hk
m
(4.12)
In other words, the group velocity is equal to the classical velocity for a
massive particle v = p/m. All the wave equations I know of fall in one of
these two categories; either ω is proportional to k, in which case the particle
is massless and its velocity v = ω/k, or the relationship is quadratic, in
which case
m = lim
k→0
_
¯hk
dk
dω
_
. (4.13)
So long as we are talking about waveparticle duality, this is what mass
means.
38 CHAPTER 4. INTRODUCTION TO FIELD THEORY
One of the advantages of using Lagrangians rather than Hamiltonians
is that Lagrangians have simple transformation properties under Lorentz
transformations. To see this, let’s rewrite (7) in relativistic notation. Con
struct the contravariant and covariant fourvectors
x
µ
≡ (x
0
, x
1
, x
2
, x
3
) = (ct, x, y, z) (4.14)
x
µ
= (x
0
, x
1
, x
2
, x
3
) = (ct, −x, −y, −z) (4.15)
and the corresponding contravariant and covariant derivatives
∂
µ
≡
∂
∂x
µ
∂
µ
≡
∂
∂x
µ
. (4.16)
This puts the EulerLagrange equation in tidy form
∂
µ
_
∂L
∂(∂
µ
ϕ)
_
−
∂L
∂ϕ
= 0 (4.17)
This is slightly amazing. Equation (7) was derived without reference to
Lorentz transformations, and yet (17) has the correct form for a scalar wave
equation. We get relativity for free! If we can manage to make L a Lorentz
scalar, then (17) will have the same form in all Lorentz frames. Better yet,
the action
S =
_
dt L =
_
dt
_
d
3
x L =
1
c
_
d
4
x L (4.18)
is also a Lorentz scalar. We can do relativity with path integrals without
having to “ﬁx” anything.
Here’s an example. Rewrite (6) in 3d
L =
1
2
_
µ
_
∂ϕ
∂t
_
2
−Y
_
_
∂ϕ
∂x
_
2
+
_
∂ϕ
∂y
_
2
+
_
∂ϕ
∂z
_
2
__
(4.19)
This would be the Lagrangian density for oscillations in a huge block of
rubber. Take
µ
Y
=
1
c
2
. (4.20)
Obviously L can be multiplied by any constant without changing the equa
tions of motion. Rescale it so that it becomes
L =
1
2
_
_
∂ϕ
∂t
_
2
−c
2
_
_
∂ϕ
∂x
_
2
+
_
∂ϕ
∂y
_
2
+
_
∂ϕ
∂z
_
2
__
(4.21)
4.1. INTRODUCTION TO CONTINUUM MECHANICS 39
Substituting (21) into (17) yields the usual equation for EM waves, 2ϕ = 0.
Notice how the Lagrangian for oscillations a block of rubber (19) turns
into the Lagrangian for oscillations in the ether (21). We don’t have to worry
about the mechanical properties of the ether, because µ and Y are scaled
away. Despite what you may have been told, the MichelsonMorley experi
ment proves the existence of the ether. When correctly viewed, everything
is a bunch of harmonic oscillators, even the vacuum!
Using Einstein’s neat notation, we can collapse (21) into one term
L =
1
2
(∂
µ
ϕ)(∂
µ
ϕ) ≡
1
2
(∂ϕ)
2
(4.22)
The last piece of notation (∂ϕ)
2
, is used to save ink. The fact that we can
write L like this is proof that it is a Lorentz scalar. This is an important
point; we can deduce the symmetry properties of a theory by glancing at L.
Now you can make up your own ﬁeld theories. All you have to do is add
scalar terms to (22). Try it. Name the theory after yourself. Here’s a theory
that already has a name. It’s the KlineGordon theory.
L =
1
2
_
(∂ϕ)
2
−m
2
φ
2
¸
(4.23)
(I have set c = 1 and ¯ h = 1.) Using our new notation, the equation of
motion is
(∂
µ
∂
µ
+m
2
)ϕ = 0 (4.24)
If we assume that ϕ(x) (x is a 4vector in this notation.) is a onecomponent
Lorentz scalar, then this describes a spinless particle with mass m propagat
ing without interactions. Spin can be included by adding more components
to ϕ. More about this later.
It is plausible that we can adopt all the path integral results from the
previous sections. Simply replace the displacement q
i
with the displacement
ﬁeld ϕ(x) and the momentum p
i
with the canonical momentum π(x) =
∂L
∂ ˙ ϕ
.
Let’s look at this more closely. We have been calculating the propagator,
¸q
f
, t
f
[q
i
, t
i
)
H
= ¸q
f
[e
−iH(t
f
−t
i
)
[q
f
)
S
=
_
D[x]D[p]e
iS
. (4.25)
(Equation (29) in Chapter 2) The natural generalization of this is
¸ϕ
f
(x)[e
−iH(t
f
−t
i
)
[ϕ
i
(x)) =
_
D[ϕ]D[π] exp
_
i
_
t
f
t
i
d
4
xL(ϕ)
_
(4.26)
40 CHAPTER 4. INTRODUCTION TO FIELD THEORY
Where the functions ϕ(x) over which we integrate are constrained to the
speciﬁc conﬁgurations ϕ
i
(x) at x
0
= t
i
, and ϕ
f
(x) at t
0
= t
f
.
In the previous chapter we were able to derive (2.29) directly from Feyn
man’s sum over paths hypothesis. Here here that derivation breaks down.
The trouble is that the ﬁeld operators ˆ ϕ do not have complete sets of eign
states. The derivation can be completed using the notion of coherent states,
2
but I propose a diﬀerent approach. To make the notation a bit more compact
I will write the propagator,
¸ϕ
f
, t
f
[ϕ
i
, t
i
) = U(ϕ
i
, ϕ
f
; T), (4.27)
where T = t
f
−t
i
. Now it is possible to show
3
without using the completeness
hypothesis that U satisﬁes Schrodinger’s equation
i¯h
∂
∂T
U = HU (4.28)
The Hamiltonian in this equation is obtained from the Lagrangian in (26)
by the usual procedure. We argue that (26) is correct, even if we are unable
to derive it from ﬁrst principles. Let’s just take it as a deﬁnition of U.
Since the exponent in (26) is quadratic in π, we can complete the square
and evaluate the D[π] integral to obtain
¸ϕ
f
(x)[e
−iH(t
f
−t
i
)
[ϕ
i
(x)) =
_
D[ϕ] exp
_
i
_
t
f
t
i
d
4
x L
_
(4.29)
You will recall from (2.32) and (2.33) that this integration produces a badly
behaved normalization constant. Back then, I simply called it N and re
assured you that it would go away. Here, this is true a fortiori. I have no
idea what the constant is and only a clue what it means. I do know that all
physically interesting quantities are quotients of two terms, each containing
the same constant. Since it always cancels, I have not even bothered to
write it in (29).
4.2 Correlation functions
The propagator in (25) is just the ticket for ordinary quantum mechanics.
After all, the q’s are measurable quantities. Quantum ﬁelds, on the other
hand, are not measurable, so U as it stands is not of much use. It turns out
2
This is worked out in Prof. Wasserman’s lecture notes.
3
See Peskin and Schroeder, Section 9.1
4.2. CORRELATION FUNCTIONS 41
that the quantities of interest are the correlation functions deﬁned in Section
2.6, equation (59), but with ﬁelds replacing the
ˆ
Q operators. In other words
we are looking for
G(x
1
, x
2
, , x
n
) ≡ ¸Ω[T [ ˆ ϕ(x
1
) ˆ ϕ(x
2
) ˆ ϕ(x
n
)] [Ω) (4.30)
This is sometimes called the npoint function. The most important instance
of this is the twopoint correlation function, which we will now investigate.
Consider the following object:
4
¸ϕ
b
[e
−iHT
T[ ˆ ϕ(x
1
) ˆ ϕ(x
2
)]e
−iHT
[ϕ
a
) (4.31)
The ˆ ϕ’s are ﬁeld operators in the Heisenberg picture. The times are arranged
so that both x
0
1
and x
0
2
fall in the interval between −T and T. ϕ
a
represents
the ﬁeld evaluated at time −T, ϕ
a
= ϕ(−T, x) and ϕ
b
= ϕ(T, x). The
construction T[ ] refers to the time ordering operator, as in Section 2.6.
Let’s assume to start with that x
0
2
> x
0
1
. We can convert the Heisenberg
ﬁelds to Schrodinger ﬁelds, e.g.
ˆ ϕ(x
2
) = e
iHx
0
2
ˆ ϕ
S
(x
2
)e
−iHx
0
2
Now (31) can be written
¸ϕ
b
[e
−iH(T−x
0
2
)
ˆ ϕ
S
(x
2
)e
−iH(x
0
2
−x
0
1
)
ˆ ϕ
S
(x
1
)e
−iH(x
0
1
+T)
[ϕ
a
) (4.32)
At this point in Section 2.6 we inserted complete sets of position eigen
states and replaced the operators with their eigenvalues. As mentioned
previously, this step is not immediately justiﬁed, because the ﬁeld operators
don’t have eigenstates. I don’t want to get into the technicalities of coherent
states, so let me just reassure you that we won’t make any mistakes, if we
assume that we can replace ﬁeld operators with cnumber ﬁelds
ˆ ϕ
S
(x
1
)[ϕ
1
) = ϕ
1
(x
1
)[ϕ
1
),
and that the completeness relation
_
D[ϕ
1
] [ϕ
1
)¸ϕ
1
[ =
ˆ
1,
4
I am using the notation from Section 9.2 of Peskin and Schroeder. I hope no one will
be so symbolminded as to confuse the T in the exponentials (meaning some ﬁxed time)
and the T in front of the ﬁelds (the time ordering operator).
42 CHAPTER 4. INTRODUCTION TO FIELD THEORY
makes sense. Now insert
ˆ
1 after each of the ﬁelds in (32). The result is
_
D[ϕ
1
(x)]
_
D[ϕ
2
(x)] ϕ
1
(x
1
)ϕ
2
(x
2
)¸ϕ
b
[e
−iH(T−x
0
2
)
[ϕ
2
) (4.33)
¸ϕ
2
[e
−iH(x
0
2
−x
0
1
)
[ϕ
1
)¸ϕ
1
[e
−iH(x
0
1
+T)
[ϕ
a
)
This is a product of three propagators that jolly along the ﬁeld from −T →
x
0
1
→x
0
2
→T. For example,
¸ϕ
b
[e
−iH(T−x
0
2
)
[ϕ
2
) =
_
D[ϕ] exp
_
i
_
T
x
0
2
d
4
x L(ϕ)
_
combining these three factors yields the ﬁnal result
¸ϕ
b
[e
−iHT
T[ ˆ ϕ(x
1
) ˆ ϕ(x
2
)]e
−iHT
[ϕ
a
) (4.34)
=
_
D[ϕ]ϕ(x
1
)ϕ(x
2
) exp
_
i
_
T
−T
d
4
x L(ϕ)
_
where the boundary conditions on the path integral are ϕ(−T, x) = ϕ
a
(x)
and ϕ(T, x) = ϕ
b
(x) for some ϕ
a
and ϕ
b
. Notice that if we had started with
the assumption that x
0
1
> x
0
2
, we would have arrived at the same result. We
get the time ordering for free.
I have set up the correlation function with the e
−iHT
’s in (31) because I
want to discuss the limit T → ∞. This has the eﬀect of projecting out the
ground state, but exactly what does this mean and what is the ground state
anyhow? We are accustomed to dividing the Hamiltonian into two parts, the
freeparticle and the potential or interaction term, H = H
0
+H
I
. If we set
H
I
= 0, the particle has energy eigenstates and eigenenergies, and clearly,
the minimum energy is E
0
= 0. This makes sense mathematically, but not
physically: nature doesn’t come equipped with switches to do this. This
is a highly nontrivial problem, because a single particle can interact with
itself via intermediate particles. Worse yet, even the vacuum can interact
with itself through the spontaneous creation of particleantiparticle pairs.
We call this minimumenergy vacuum state [Ω).
Deﬁne the eigenstates of H as follows:
H[n) = E
n
[n)
Imagine starting with [ϕ
a
), and evolving through time with H.
e
−iHT
[ϕ
a
) =
n
e
−iE
n
T
[n)¸n[ϕ
a
)
4.2. CORRELATION FUNCTIONS 43
= e
−iE
0
T
[Ω)¸Ω[ϕ
a
) +
n=0
e
−iE
n
T
[n)¸n[ϕ
a
)
Give a small negative imaginary part to T and look at what happens when
T → ∞(1 − i). Each term tends toward zero, but since E
0
< E
n
for all
n ,= 0, the ﬁrst term approaches zero more slowly. Put it another way: if we
divide the equation through by e
−iE
0
T
¸Ω[ϕ
a
), the term containing the sum
will vanish.
[Ω) = lim
T→∞(1−i)
(e
−iE
0
T
¸Ω[ϕ
a
))
−1
e
−iHT
[ϕ
a
). (4.35)
Insert (35) into (34).
¸Ω[T[
ˆ
ϕ(x
1
)
ˆ
ϕ(x
2
)][Ω) = lim
T→∞(1−i)
_
D[ϕ]ϕ(x
1
)ϕ(x
2
) exp
_
i
_
T
−T
d
4
xL
_
_
D[ϕ] exp
_
i
_
T
−T
d
4
xL
_
(4.36)
I remarked previously that the measure
_
D[ϕ] produced weird normaliza
tion constants. Here as always, these constants cancel between the numera
tor and denominator. Often in doing perturbation theory, we are concerned
with the correlation function for two free particles. In this case we replace
[Ω) with [0). The denominator is then proportional to ¸0[0) = 1, so we can
ignore it altogether.
Back at the end of Chapter 2, I introduced the notion of a generating
function. The key equations here are (2.64) and (2.65), which I repeat here
for reference.
z[J] ≡ ¸q
f
, t
f
[q
i
, t
i
)
J
≡
_
D[q] exp
_
i
_
t
f
t
i
dt [L(t) +J(t)q(t)]
_
(4.37)
¸q
f
, t
f
[T
_
ˆ
Q(t
1
)
ˆ
Q(t
2
)
ˆ
Q(t
n
)
_
[q
i
, t
i
) =
_
1
i
_
n
δ
n
z[J]
δJ(t
1
)δJ(t
2
) δJ(t
n
)
¸
¸
¸
¸
J=0
(4.38)
We take advantage of recent developments by upgrading these formulas in
two ways: (1) replacing the Q
i
’s with the ﬁelds, ϕ(x), and (2) taking the
limit T → ∞(1 − i). It is customary to use the capital Z for this new
generating functional.
Z[J] ≡
_
D[ϕ] exp
_
i
_
∞
−∞
d
4
x[L(ϕ) +J(x)ϕ(x)]
_
(4.39)
We can use this technology to rewrite (36) as follows:
¸Ω[T[
ˆ
ϕ(x
1
)
ˆ
ϕ(x
2
)][Ω) =
1
Z[0]
_
1
i
_
2
δ
2
Z[J]
δJ(x
1
)δJ(x
2
)
¸
¸
¸
¸
J=0
(4.40)
44 CHAPTER 4. INTRODUCTION TO FIELD THEORY
The generalization to higherorder correlation functions is obvious. We
sometimes use the notation
¸Ω[T[
ˆ
ϕ(x
1
)
ˆ
ϕ(x
2
) ˆ ϕ(x
n
)][Ω) = G(x
1
, x
2
, , x
n
) (4.41)
The function G(x
1
, x
2
, , x
n
) is called the “npoint function.”
4.3 The Feyman Propagator
I will now show how this formalism works with the free KleinGordon ﬁeld.
Use the Lagrangian (23) in (39)
Z[J] =
_
D[ϕ] exp
_
i
_
d
4
x
_
1
2
_
(∂ϕ)
2
−m
2
ϕ
2
+iϕ
2
_
+Jϕ
__
(4.42)
I have inserted the iϕ
2
to ensure that the exponential integrals converge
as explained previously. The key to doing the
_
d
4
x integral is to Fourier
transform the integrand to momentum space.
ϕ(x) =
_
d
4
p
(2π)
2
e
−ipx
˜ ϕ(p) (4.43)
Substituting (43) into (42) gives the following:
_
d
4
x
1
2
_
(∂ϕ)
2
−m
2
ϕ
2
+iϕ
2
¸
(4.44)
=
_
d
4
p
1
d
4
p
2
2(2π)
4
=
_
−p
1
p
2
−m
2
+i
¸
˜ ϕ(p
1
) ˜ ϕ(p
2
)
_
d
4
xe
−i(p
1
+p
2
)
=
1
2
_
d
4
p ˜ ϕ(−p) ˜ ϕ(p)(p
2
−m
2
+i)
On the other hand
_
d
4
xJ(x)ϕ(x) =
_
d
4
p
˜
J(−p) ˜ ϕ(p) or
_
d
4
p
˜
J(p) ˜ ϕ(−p) (4.45)
or (in fact)
1
2
_
d
4
p
_
˜
ϕ(P)
˜
J(−p) + ˜ ϕ(−p)
˜
J(p)
_
The reason for the two terms will be clear soon. In order to make the
notation a bit more compact, I will deﬁne
∆
−1
≡ p
2
−m
2
+i (4.46)
4.3. THE FEYMAN PROPAGATOR 45
The complete action becomes
S =
_
d
4
x (L +J ˆ ϕ)
=
1
2
_
d
4
p
_
˜ ϕ(−p) ˜ ϕ(p)∆
−1
+
˜
J(p) ˜ ϕ(−p) +
˜
J(−p) ˜ ϕ(p)
_
(4.47)
Deﬁne a new ﬁeld
˜ ϕ(±p) = ˜ ϕ
0
(±p) −
˜
J(±p)∆ (4.48)
S =
1
2
_
d
4
p
_
˜ ϕ
0
(−p) ˜ ϕ
0
(p)∆
−1
−
˜
J(p)
˜
J(−p)∆
_
(4.49)
You see the point of (49)? It break up the action into two pieces, one
depending only on the ﬁelds and one depending only on the sources. The
peculiar sum in (45) was necessary to make this happen. Substituting into
(42) yields
Z[J] =
_
D[ϕ
0
] exp
_
−
i
2
_
d
4
p ˜ ϕ
0
(p) ˜ ϕ
0
(−p)∆
−1
_
(4.50)
exp
_
i
2
_
d
4
p
˜
J(p)
˜
J(−p)∆
_
= Z[0]e
iW[J]
(4.51)
where
W[J] ≡ −
1
2
_
d
4
p
˜
J(p)
˜
J(−p)∆(p) (4.52)
Most path integrals are impossible to do, we just don’t have to do them.
We don’t have a snowball’s chance of calculating Z[0], but (40) guarantees
that it will always cancel out!
At this point it is useful to transform back to conﬁguration space.
˜
J(±p) =
_
d
4
x
(2π)
2
e
±ipx
J(x) (4.53)
∆(p) =
_
d
4
x e
ipx
D(x)
Our ﬁnal result is
W = −
1
2
_
d
4
x d
4
y J(x)J(y)D(x −y) (4.54)
46 CHAPTER 4. INTRODUCTION TO FIELD THEORY
where
D(x −y) ≡
_
d
4
p
(2π)
4
e
ip(x−y)
p
2
−m
2
+i
(4.55)
This is the Feynman propagator.
5
It is the relativistic generalization of the
oneparticle Green’s function (2.48). As such it represents the amplitude
for a particle at the space time point x to propagate to y. The boundary
conditions are diﬀerent, however, since D(x − y) = D(y − x). Relativity
forces this on us, as I mentioned before.
The derivation of Z[J] assumed that sources were turned on and oﬀ at
ﬁnite times before and after the limit T →∞. Let’s see how this idea plays
out in (50). First expand Z[J]/Z[0] in an exponential series. (See equation
(51).)
Z[J]
Z[0]
= 1 +iW +
i
2
2
W
2
+ (4.56)
The second term is easy to understand.
iW = −
i
2
_ _
d
4
x d
4
y J(x)J(y)D(x −y) (4.57)
Suppose our source consists of two local centers of activity, so that J(x) =
J
0
[δ(x −x
1
) +δ(x −x
2
)]. Then
iW = −iJ
2
0
[D(x
1
−x
2
) +D(0)] (4.58)
The D(0) is meaningless. It’s an artifact of our delta function sources. The
other term is simply the probability of a particle to be created by the source
at x
1
(when the source was turned on) and reabsorbed at x
2
(when the
source was turned oﬀ) and vice versa.
Now calculate the twopoint function corresponding to iW without spec
ifying the nature of the sources. The following calculations use equations
5
To be consistent I suppose, we should use
˜
D(p) instead of ∆(p), but this is not
customary notation.
4.4. INTERACTING FIELDS 47
(40), (41), (56) and (57).
G(x
1
, x
2
) = ¸Ω[T[ ˆ ϕ(x
1
) ˆ ϕ(x
2
)][Ω)
=
1
Z[0]
_
1
i
_
2
δ
2
Z[J]
δJ(x
1
)δJ(x
2
)
¸
¸
¸
¸
J=0
=
_
1
i
_
2
δ
2
iW[J]
δJ(x
1
)δJ(x
2
)
¸
¸
¸
¸
J=0
=
_
1
i
_
2
δ
2
δJ(x
1
)δJ(x
2
)
_
−
i
2
_ _
d
4
x d
4
y J(x)J(y)D(x −y)
_
J=0
(4.59)
This functional derivative was worked out as an example in Chapter 3.
The relevant equations are (3.10) and (3.12). The eﬀect of the functional
derivatives has been to “peel oﬀ” the sources. The ﬁnal result is
G(x
1
, x
2
) = iD(x
1
−x
2
). (4.60)
This is in fact the exact amplitude for a free particle to propagate from
x
1
to x
2
. The higher order terms in (56) vanish when we set J = 0. It
is easy to see that the nth term of the expansion (56) contains 2n factors
of J and that all the npoint functions vanish except for G(x
1
, x
2
, , x
2n
).
Your should work out the 4point function as an exercise. It consists or a
sum of three terms corresponding to all possible ways in which two particles
can propagate between four points. Now that you have the idea, you can
represent any npoint function as a set of (2n1)!! diagrams. Just draw
n points labelled x
1
, x
2
, , x
2n
, and draw lines connecting them in pairs,
These are Feynman diagrams.
The theory so far has been trivial in the sense that the particles don’t
interact with one another. We started with the Lagrangian based on the
simple harmonic oscillator, i.e. with a quadratic potential, and in such a
model all the normal modes are linearly independent. This is well known
from continuum mechanics, and it’s true in quantum mechanics as well.
Interacting ﬁeld theories have additional terms in the Lagrangian. We’ll
look at a simple example in the next section.
4.4 Interacting Fields
In order to make out ﬁelds interact, we must add some additional term to the
Lagrangian. Many texts use the interaction term −(λ/4!)ϕ
4
(x), which is as
48 CHAPTER 4. INTRODUCTION TO FIELD THEORY
good as any. As you will see, this describes identical particles interacting in
groups of four. An example is the scattering process in which two particles
come together at a point and two particles emerge from that point. Another
is the decay process in which one particle decays into three clones of itself.
The parameter λ is called a “coupling constant.” It determines the strength
of the interaction. You probably don’t know any particles that act like this,
but it turns out to be an important ingredient in the nonAbelian gauge
theories we will study much later in the course. More familiar theories all
involve two or more diﬀerent particle species and spin, but “Suﬃcient unto
the day is the evil thereof.”
The simplest nontrivial process involving these hypothetical particles is
the twobody scattering process described above, It turns out that every
thing we can know about this process is contained in the fourpoint Green’s
function. We need to calculate the fourﬁeld version of (40),
1
Z[0]
_
1
i
_
4
δ
4
Z[J]
δJ(x
1
)δJ(x
2
)δJ(x
3
)δJ(x
4
)
¸
¸
¸
¸
J=0
= G(x
1
, x
2
, x
3
, x
4
) (4.61)
with our new Z[J],
Z[J] =
_
D[ϕ] exp
_
i
_
d
4
x
_
1
2
_
(∂ϕ)
2
−m
2
ϕ
2
+iϕ
2
_
+Jϕ −(λ/4!)ϕ
4
__
(4.62)
The tricks we used in arriving at (50) no longer work, and we are forced
to make approximations. As I am fond of saying, “When in doubt, expand
in a power series.” Feynman diagrams are a kind of perturbation theory in
which the coupling constant is assumed to be “small” so that perturbation
series is really an expansion in powers of the coupling constant, in this case
λ. Let’s start with the ﬁrstorder term.
Z
1
[J] =
−iλ
4!
_
dω
_
D[ϕ] exp
_
i
_
d
4
x
_
−
1
2
ϕ(∂
2
+m
2
−i)ϕ +Jϕ
__
ϕ
4
(ω)
(4.63)
You will notice that I replaced (∂ϕ)
2
in the exponential with −ϕ∂
2
ϕ. This
4.4. INTERACTING FIELDS 49
is just integration by parts. Now substitute (4.63) into (4.61)
G(x
1
, x
2
, x
3
, x
4
) =
1
Z[0]
_
1
i
_
4
δ
4
Z
1
[J]
δJ(x
1
)δJ(x
2
)δJ(x
3
)δJ(x
4
)
¸
¸
¸
¸
J=0
=
−iλ
4!
1
Z[0]
_
d
4
ω
_
D[ϕ] exp
_
i
_
d
4
x
_
−
1
2
ϕ(∂
2
+m
2
−i)ϕ +Jϕ
__¸
¸
¸
¸
J=0
ϕ
4
(ω)ϕ(x
1
)ϕ(x
2
)ϕ(x
3
)ϕ(x
4
)
(4.64)
You see that long string of ϕ’s on the last line of (4.64)? That is the
paradigmatic integral. If you can do that, you can calculate any Green’s
function to any order in λ. In fact, there is a simple, intuitive ”cookbook”
procedure for evaluating all such expressions eﬀortlessly. The trick is yet
another application of our basic gaussian integral.
6
_
∞
−∞
dx
1
dx
2
dx
N
e
−
i
2
x·A·x+iJ·x
=
¸
(2πi)
N
det [iA]
e
i
2
J·A
−1
·J
(4.65)
The notation x A x means
ij
x
i
A
ij
x
j
, etc. The matrix A is assumed to
be real and symmetric. Let’s deﬁne
¸x
i
x
j
x
k
x
l
) ≡
_
∞
−∞
_
∞
−∞
dx
1
dx
2
dx
N
e
−
i
2
x·A·x
x
i
x
j
x
k
x
l
_
∞
−∞
_
∞
−∞
dx
1
dx
2
dx
N
e
−
i
2
x·A·x
(4.66)
By repeated diﬀerentiation of (4.65) we can derive
¸x
i
x
j
x
k
x
l
) =
Wick
(A
−1
/i)
ab
(A
−1
/i)
cd
(4.67)
The subscripts ab and cd in (4.67) stand for any pair of the indices i, j, k, l.
A sum over “Wick contractions” consists of all possible ways of connecting
the indices in pairs. If there is only one pair,
¸x
i
x
k
) = (A
−1
/i)
ik
(4.68)
There is a useful analogy between (4.65) and (4.66) and the last two lines
of (4.64). Clearly J
i
is analogous to J(x), and x
i
is analogous to ϕ(x). We’ve
discussed this before. So far so good. The term
i
2
J A
−1
J is reminiscent
6
There is a brief proof of this in Appendix 1 of Zee’s book. I will ask you to derive it
in more detail for homework.
50 CHAPTER 4. INTRODUCTION TO FIELD THEORY
of (54), so it’s a good guess that A
−1
is analogous to −D(x −y), but if this
is true, A must be analogous to (∂
2
+ m
2
− i). Quel horreur! Worse yet,
what does it mean to take the determinant of it?
The answer to the last question is clear: we don’t care, since the determi
nant cancels in equation (4.66). In fact, come to think of it, the denominator
in (4.66) must be analogous to Z[0]. The other point is almost clear, since
(∂
2
+m
2
−i)D(x −y) = −δ
(4)
(x −y) (4.69)
The proof of (4.69) is easily established using the deﬁnition (55). D(x −y)
is, after all, a Green’s function, and Green’s functions are by deﬁnition,
inverses of diﬀerential operators. That’s what they do for a living. In
summary then, I claim that we can use (4.65) and (4.66) for ﬁelds rather
than discrete variables by making the following replacements:
x
i
→ ϕ(x
i
)
J
i
→ J(x
i
)
A
−1
→ −D(x −y)
i
2
J A
−1
J → −
i
2
_
d
4
xd
4
yJ(y)D(x −y)J(y)
A → (∂
2
+m
2
−i)
−
i
2
x A x → −
i
2
_
d
4
xϕ(x)(∂
2
+m
2
−i)ϕ(x)
_
∞
−∞
dx
1
dx
2
dx
n
→
_
D[ϕ]
The proof that my claim is valid is sketched out in Ryder,
7
but in fact,
we have implicitly proved it, at least for expressions of the form (4.64), as
you will see. Anyhow, if my claim is true then the correct generalization of
(4.66) is
¸ϕ(x
i
)ϕ(x
j
) ϕ(x
k
)ϕ(x
l
))
=
_
D[ϕ] exp
_
i
_
d
4
x
_
−
1
2
ϕ(∂
2
+m
2
−i)ϕ
¸_
ϕ(x
i
)ϕ(x
j
) ϕ(x
k
)ϕ(x
l
)
_
D[ϕ] exp
_
i
_
d
4
x
_
−
1
2
ϕ(∂
2
+m
2
−i)ϕ
¸_
(4.70)
The denominator is just Z
0
[0], ie. equation (4.62) with J = λ = 0.
7
Lewis Ryder, Quantum Field Theory, Section 6.2
4.4. INTERACTING FIELDS 51
To see how this works, let’s evaluate (4.70) in the case of two ﬁelds.
G(x
1
, x
2
) =
1
Z
0
[0]
_
1
i
_
2
δ
2
Z
0
[J]
δJ(x
1
)δJ(x
2
)
¸
¸
¸
¸
J=0
=
_
D[ϕ] exp
_
i
_
d
4
x
_
−
1
2
ϕ(∂
2
+m
2
−i)ϕ
¸_
ϕ(x
1
)ϕ(x
2
)
_
D[ϕ] exp
_
i
_
d
4
x
_
−
1
2
ϕ(∂
2
+m
2
−i)ϕ
¸_
(4.71)
The ﬁrst line follows from (59), and the second line uses (4.62) with λ = 0.
A glance at (4.70) reveals that what we have just calculated is ¸ϕ(x
1
)ϕ(x
2
)).
Let’s pause for a moment to see how all the pieces ﬁt together. ¸ϕ(x
1
)ϕ(x
2
))
as deﬁned by (4.70) equals G(x
1
, x
2
). That proof required nothing more than
functional diﬀerentiation. Reasoning by analogy from (4.68) led us to hope
that ¸ϕ(x
1
)ϕ(x
2
)) would be equal to −D(x
1
−x
2
)/i. But we already know
from (60) that G(x
1
, x
2
) = iD(x
1
−x
2
), so our cookbook analogy has given
us the right answer for the Feynman propagator without any of the clever
calculations in Section 4.3. Put it another way: we have proved that the
analogy works correctly in the case of (71). The analogy to (4.67) is
¸ϕ(x
i
)ϕ(x
j
) ϕ(x
k
)ϕ(x
l
)) =
Wick
G(x
a
, x
b
) G(x
c
, x
d
) (4.72)
So here’s the procedure: make a list of all possible ways of pairing up the
indices i, j, , k, l. Each entry in the list corresponds to one term in the
sum. That’s all there is to it. I’m willing to believe this without further
proof.
Now you can evaluate (4.64) using (4.70).
G(x
i
, x
2
, x
3
, x
4
) = −
iλ
4!
_
d
4
ω¸ϕ(x
1
)ϕ(x
2
)ϕ(x
3
)ϕ(x
4
)ϕ
4
(ω)) (4.73)
The Wick sum contains in principle (N − 1)!! = 7!! = 105 terms. It is
probably the case, however, that most of these terms are of no interest.
Draw some pictures to ﬁnd the interesting ones. Label ﬁve points 1,2, 3,
4, and ω. Each term in the expansion corresponds to a way of connecting
these points. Here are the rules:
• There are eight ﬁelds in (4.73), therefore there must be four lines
connecting points. Each line corresponds to one of the G(x
a
, x
b
)’s in
(4.72).
• The points 1, 2, 3, and 4, must be attached to one and only one line.
Since there are four factors of ϕ(ω) in (4.73), there must be four “line
ends” at the ω point, ie. it can be connected to itself by one or two
lines.
52 CHAPTER 4. INTRODUCTION TO FIELD THEORY
The points 1, 2, 3, and 4 are connected to external sources, or to put it an
other way, J(x
1
), J(x
2
), J(x
3
), and J(x
4
) were used to deﬁne G(x
1
, x
2
, x
3
, x
4
).
The ω point is not. If you want all four of these particles to interact, they
must each be contracted with one of the ϕ(ω)’s. There are 4! terms like this
in the Wick sum. The ﬁnal result is
G(x
1
, x
2
, x
3
, x
4
) = −iλ
_
d
4
ωG(x
1
, ω)G(x
2
, ω)G(x
3
, ω)G(x
4
, ω) (4.74)
Forgive me if I gloat a bit. A few paragraphs ago we had some algebra,
which, if done correctly, would have yielded 105 terms. Now we draw a few
pictures, and the answer is obvious by inspection. These pictures are called
Feynman diagrams.
You have noticed that we have ignored 1054!=81 terms, These are all
pathological in the sense that the ω point is connected to itself by at least
one line. It is a general rule that all Feynman diagrams with loops give rise
to divergent integrals. This is ultimately related to the fact that there are
parameters such as the mass and coupling constant (or charge) of a particle
that are fundamentally mysterious. If these integrals converged, then we
might be able to calculate them. These divergences are nature’s way of
slamming the door in our face. That’s the big bad news. The little bit of
good news is that we can go on doing ﬁeld theory calculating everything else,
if we treat these divergences very carefully. This comes under the general
heading of renormalization, to which we will return later.
4.5 Momentum Space
Our derivations have for the most part used conﬁguration space. Momentum
space was only used occasionally as a tool for evaluating certain integrals.
Momentum space is more natural for many quantum processes, however.
Scattering is a good example. The particles in the beam have a welldeﬁned
momentum. The target particle is at rest. The momenta of the ﬁnal state
particles are determined by our apparatus. The rules for Feynman diagrams
discussed brieﬂy above also have a simpler form in momentum space.
Deﬁne the momentum space Green’s functions in the obvious way.
˜
G(p
1
, p
2
, , p
N
) =
_
d
4
x
1
d
4
x
2
d
4
x
N
e
i(±p
1
x
1
±p
2
x
2
±···±p
N
x
N
)
G(x
1
, x
2
, , x
N
)
(4.75)
I call your attention to the ± signs in the exponential. To some extent the
sign doesn’t matter, since G(x) = G(−x) and ∆(p) = ∆(−p). Moreover,
4.5. MOMENTUM SPACE 53
when x or p appear as integration variables, one is always free to replace
x → −x or p → −p. In one respect, however, the signs are allimportant.
Take the Fourier transform of (4.74) using (4.75) and (4.53).
˜
G(p
1
, p
2
, p
3
, p
4
) = −iλ(2π)
4
δ
(4)
(±p
1
±p
2
±p
3
±p
4
)∆(p
1
)∆(p
2
)∆(p
3
)∆(p
4
)
(4.76)
The δfunction is there to enforce momentum conservation. For example, if
the scattering process is one in which p
1
and p
2
are the incident momentum
and p
3
and p
4
are the outgoing momentum, then the δfunction must be
δ
(4)
(p
1
+p
2
−p
3
−p
4
) or δ
(4)
(p
3
+p
4
−p
1
−p
2
). Let’s agree on a convention:
outgoing momenta are positive and incoming momenta are negative. That
way we will always get the right signs in (4.75).
Equation (4.76) illustrates some of the rules associated with Feynman
diagrams. The momentum space Green’s function corresponding to any one
diagram consists of a product of the following factors.
• A factor −iλ for each vertex.
• A factor i∆(p
i
) for each external line.
• An energymomentum conserving δfunction (2π)
4
δ
(4)
(±p
1
±p
2
±p
3
±
p
4
)
There is one more important item that we have not yet encountered.
What happens to internal lines, ie. those that do not terminate on external
sources. To see that, let’s look at one of the pathological diagrams in which
one of the lines connected to the ω loops back on itself. One such term in
the Wick expansion is
8
G(x
1
, x
2
, x
3
, x
4
) = −
iλ
4!
_
d
4
ω G(x
1
, x
2
)G(x
3
, ω)G(x
4
, ω)G(ω, ω) (4.77)
The momentum space Green’s function is
G(p
1
, p
2
, p
3
, p
4
) = −
iλ
4!
(2π)
4
δ
(4)
(p
1
−p
2
)(2π)
4
δ
(4)
(p
3
−p
4
)
∆(p
1
)∆(p
3
)∆(p
4
)
_
d
4
p
4
(2π)
4
∆(p
4
) (4.78)
In this case, the particle that starts out with momentum p
1
does not interact
with the one with momentum p
3
, therefore their momenta are conserved
8
We have not yet come to terms with indistinguishable particles. For the moment,
assume that the particles have little numbers on them like billiard balls.
54 CHAPTER 4. INTRODUCTION TO FIELD THEORY
separately. There are two δfunction The new feature here is an internal
line, this one labelled by p
4
. Since this momentum can have any value, there
is nothing to constrain the
_
d
4
p
4
integral. Such integrals always diverge.
To see that this is so, look at
_
d
4
p
4
p
2
4
−m
2
+i
(4.79)
The numerator contains four powers of the momentum, whereas the denom
inator contains only two. There is no oscillating phase factor, so the integral
diverges quadratically. Such integrals can be dealt with rigorously with a
variety of diﬀerent techniques. For example, one can put a cutoﬀ on the
range on integration, and then investigate analytically how the result de
pends on the cutoﬀ. For a certain privileged class of theories, one can show
that the only cutoﬀdependent quantities are the particle’s mass and cou
pling constant and the normalization of the wave function. Since we know
these things already, we can simply ignore the cutoﬀdependent parts. Such
theories are said to be renormalizable. The fact that we have to do this is
a sign that our knowledge and theory are incomplete. The fact that we can
do it is also signiﬁcant. All theories we know that describe the real world,
at least approximately, are renormalizable. If you give me a theory that is
not renormalizable, I know that it’s wrong!
4.6 Virtual Particles
We have encountered the freeparticle Feynman propagator, ﬁrst in mo
mentum variables (46) and later in spacetime variables (55). When we
evaluate out Green’s functions they appear attached to the external sources
as well as to points inside the region where the particles interact with them
selves. In both cases they have a peculiar feature: in some sense they repre
sent particles that are not “on the mass shell.” For free particles, after all,
p
2
= p
µ
p
µ
= m
2
, so it seems that the denominators of (46) and (55) should
vanish. They don’t of course, because p
µ
is not a physical momentum, but
a variable of integration. There is more to it than that, however. We can
not specify space and momentum coordinates with arbitrary accuracy. The
propagator D(x −y) describes a situation in which x and y are microscopi
cally close together, so there is some “fuzziness” in the momentum as well.
Since we cannot specify momentum and energy with arbitrary precision, we
can’t insist that they are perfectly conserved. The propagator takes this into
account in a way that is consistent with relativity. p
2
is a Lorentz scalar,
4.6. VIRTUAL PARTICLES 55
it’s just not equal to m
2
. We say that the propagator describes “virtual
particles.”
The propagator ∆(p
4
) in (4.78) describes particles that are virtual be
cause they are emitted and absorbed in a very short distance. The other
propagators in (4.77) and (4.78) describe particles that are virtual for a
diﬀerent reason: the sources J(x
i
) that we built into the theory are also
quantum mechanical objects. They are not external sources like xray tubes
or cyclotrons. That should be evident from the discussion in Section 4.2.
The sources were turned on sometime after −T and turned oﬀ again before
T, but the limits −T →−∞ and +T →+∞ that were used to project out
the ground states refer to the macroscopic past and future where we live
and move and have our being. Put it another way, we live in the ground
state. The action takes place after the sources turn on , and it stops when
the sources turn oﬀ.
We often use the momentum space Green’s functions to calculate scatter
ing reactions. Indeed, that is their principal use. But particles in scattering
experiments are produced by cyclotrons and other macroscopic sources. We
can use our Green’s functions for this by the simple expedient of removing
the ∆’s from the external legs. This is sometimes called “amputating the
legs of the diagram.”
9
This procedure then becomes part of the rules for
Feynman diagrams.
Leaving aside the very technical issue of renormalization, you now have
everything you need to do perturbation theory calculations with scalar ﬁeld
theory to any order. There is a complete description of the rules in Zee’s
book, pages 53 and 54. I will restate them here for reference.
To calculate a momentum space Green’s function corresponding to a
Feynman diagram, do the following:
• Draw the diagram and label each line with a momentum.
• Associate each line carrying momentum p with a propagator i/(p
2
−
m
2
+i).
• Associate with each interaction vertex the coupling −iλ and (2π)
4
δ
(4)
(
i
p
i
−
j
p
j
), forcing the sum of momenta
i
p
i
going out of the vertex to
be equal to the sum of momenta
j
p
j
going out of the vertex.
• Momenta associated with internal lines are to be integrated over with
the measure d
4
p/(2π)
4
.
• Remove those propagators associated with external lines.
9
Don’t let PETA ﬁnd out about this.
56 CHAPTER 4. INTRODUCTION TO FIELD THEORY
4.7 A Word About Renormalization
I have said repeatedly that all diagrams with complete loops lead to diver
gent integrals. That’s bad news enough, but what is worse, these divergent
integrals contribute real measurable eﬀects to the physical processes de
scribed by the Green’s functions. The challenge is to divide these integrals
into two parts; one part that is inﬁnite and one part that is ﬁnite and con
tains all measurable eﬀects. This is really a huge subject, but I would like to
give you a few of the key ideas. I will do this in the context of the simplest
possible divergent integral. It arises when one tries to do calculations in the
socalled “phithird theory.”
L
int
= −i
λ
3!
ϕ
3
(4.80)
When one calculates the twopoint function G(x
1
, x
2
) to second order in
perturbation theory, one comes up against the integral
G
2
(x
1
, x
2
) = (−iλ)
2
_
d
4
w d
4
z G(x, w)G(w, z)G(w, z)G(z, x
2
) (4.81)
This will diverge because of the loop. It represents the secondorder correc
tion to the propagator.
First take the Fourier transform
G
2
(p
1
, p
2
) =
_
d
4
x
1
d
4
x
2
d
4
w d
4
z e
i(p
2
x
2
−p
1
x
1
)
_
d
4
p
1
d
4
p
4
(2π)
16
e
i(x
1
−w)
e
i(w−z)
e
i(w−z)
e
i(z−x
2
)
i∆(p
1
)i∆(p
2
)i∆(p
3
)i∆(p
4
) (4.82)
Do the integrals over
_
d
4
x
1
d
4
x
2
d
4
w d
4
z, then do three momentum inte
grals
_
d
4
p
1
d
4
p
3
d
4
p
4
. The result when all is said and done is
G
2
(p
1
, p
2
) = (−iλ)
2
(2π)
4
δ
(4)
(p
1
−p
2
)∆(p
1
)∆(p
2
)
_
d
4
k
(2π)
4
∆(k)∆(p
1
−k)
(4.83)
The integration variable p
2
has be renamed k, since there are no other primed
variables left.
That last integral is going to give us trouble. Let’s concentrate on it by
deﬁning
F(p) ≡
_
d
4
k
(2π)
4
∆(k)∆(p −k) (4.84)
The integral is not only diﬃcult, it’s divergent. We will deal with these
two problems with two amazing tricks; the ﬁrst due to Feynman. Feynman
4.7. A WORD ABOUT RENORMALIZATION 57
argued that a fourfold integral is hard to do, so let’s make it a ﬁvefold
integral! The basic formula is this:
1
A
1
A
2
=
_
1
0
dz[A
1
z +A
2
(1 −z)]
−2
≡
_
1
0
dz
D(z)
2
(4.85)
The key idea is that the product A
1
A
2
is replaced by the single factor D
2
.
This, as it turns out, is enough to make the d
4
k integrable doable. In our
case
D
2
= B
2
+Q
2
−k
2
−i = C
2
−k
2
−i (4.86)
where
Q = p(1 −z) B
2
= m
2
−p
2
(1 −z) k = k
+Q c
2
= B
2
+Q
2
Dropping the prime, we have
F(p) =
_
1
0
dz
_
d
4
k
(2π)
4
1
[C
2
−k
2
−i]
2
(4.87)
A bit of power counting shows that there are four powers of k in the nu
merator and four in the denominator. The integral diverges logarithmically,
and here is where the next great idea comes in. Note that the integral
_
∞
1
dx
x
diverges, but
_
∞
1
dx
x
1+
=
1
ie. increasing the exponent in the denominator by an inﬁnitesimal amount
makes the integral convergent. We would like to integrate over 1/D
2+
but
this makes no sense whatsoever (D is a four vector). Suppose we could
integrate over slightly less that four dimensions! In the context of the trivial
example above, suppose we could integrate
_
∞
1
d
(1−)
x
x
Here’s the idea
_
d
d
k
(2π)
d
1
(C
2
−k
2
−i)
2
=
i
(4π)
d/2
Γ(2 −d/2)
Γ(2)
_
1
C
2
_
2−d/2
(4.88)
58 CHAPTER 4. INTRODUCTION TO FIELD THEORY
This was derived for integer d (don’t ask me how), but the Γ functions and
everything else on the right side of this equation are perfectly welldeﬁned
continuous functions of d. We use (4.88) as a deﬁnition of what it means
to integrate over a noninteger number of dimensions. We will use (4.88) to
integrate (4.87) and then see what happens in the limit d →4. Incidentally,
Γ(x) =
_
∞
0
dt t
x−1
e
−t
, (4.89)
Also Γ(n) = (n −1)! for integer n, and Γ(x + 1) = xΓ(x) in general.
After some algebra we ﬁnd
F(p) =
i
(4π)
d/2
2Γ(1 +/2)
_
1
0
dz
[m
2
−p
2
z(1 −z)]
/2
(4.90)
where ≡ 4 − d. Well Γ(1) = 1, so the integral still diverges in the limit
→0. We expected that, but I have one more trick up my sleeve. Look at
the expansion
A
−/2
= 1 −
2
ln A+
1
8
2
(lnA)
2
+ (4.91)
Insert (4.91) into (4.90).
lim
→0
F(p) = ∞−
i
(2π)
2
_
1
0
dz ln [m
2
−p
2
z(1 −z)] (4.92)
There are two points to notice here.
• The remaining integral is straightforward, since it involves only one
real variable of integration.
• The term “∞” in (4.92) is really
lim
→0
2i
(4π)
2
1
(4.93)
The point is that this is a pure number; It does not depend on p! We
have cleanly separated the measurable physics in the second term from
the meaningless constant in (4.93). To used an old cliche, we don’t
have to worry about throwing the baby(s) out with the bath water,
because we have gotten all the kids out of the tub!
Well, that’s not quite true. Recall that F(p) is multiplied by λ
2
in (4.81).
It’s as if we were working with an inﬁnite coupling constant. Rather than
4.7. A WORD ABOUT RENORMALIZATION 59
“throwing out an inﬁnite term,” let’s say that the λ that appears in (4.80)
is not physically measurable. It’s sometimes called the “bare” coupling con
stant λ
0
. What we are doing is replacing λ
0
∞→λ. We have “renormalized”
the coupling constant. If we were to calculate G(p
1
, p
2
) to higher orders of
perturbation theory, we would ﬁnd that these divergent integrals also have
the eﬀect of renormalizing the mass. We use the same argument; the “bare”
mass m
0
is not measurable, but we can always replace m
0
∞→m, where m
is the physical mass. It is remarkable fact that all the inﬁnities that occur to
all orders of perturbation theory can be disposed of using these two maneu
vers. A theory in which all inﬁnities can be sequestered in a ﬁnite number
of physical parameters is said to be “renormalizible.” We have learned from
decades of experience that all theories that are not renormalizible are wrong
in the sense that they make predictions that are falsiﬁed by experiments.
The bad news is that we can’t calculate m and λ. The good news is that we
can calculate everything else. This raises a profound and unanswered ques
tion. Are the masses and coupling constants of elementary particles really
fundamental constants, or can they be calculated on the basis of something
else, and if so, what is that something else?
60 CHAPTER 4. INTRODUCTION TO FIELD THEORY
Chapter 5
Identical Particles and
ManyParticle States
Ordinary matter consists of vast numbers of identical particles at low energy.
Quantum mechanics requires that the wave functions for such ensembles be
totally symmetric or totally antisymmetric, and this in turn is responsible
for many important properties, including the fact that atoms don’t simply
collapse into their nuclei! Ordinary quantum mechanics is capable of de
scribing such systems so long as the number of particles remains constant.
In a sense this is always true; there is never enough energy in a semiconduc
tor, for example, to create a single electron; but it is convenient to regard
holes in the FermiDirac sea as real particles. Thus the formation of a hole
entails the creation of two particles, a positive “particle,” the hole, and a
negative particle, the excited electron. There are equivalent processes in
Boson condensates as we shall see. Second quantization is ideally suited to
describe such processes, but ﬁrst we need to develop the ordinary quantum
mechanics of manyparticle systems.
5.1 The Quantum Mechanics of Identical Particles
Let’s write a manyparticle wave function as follows:
ψ = ψ(1, 2, . . . , N) (5.1)
The notation 1, for example, means x
1
, σ
1
. Suppose the particles had num
bers on them like billiard balls. They don’t of course, that’s the whole point,
but suppose they did. Then ψ(1, 2) means that the particle numbered 1 was
at the point x
1
and it’s zcomponent of spin was σ
1
. The wave function
61
62CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
ψ(2, 1) means that the numberone ball has components x
2
and σ
2
. Our
job is to construct a theory in which the sequence of numbers in ψ has
no observable consequences. That is what we mean by indistinguishable
particles.
Deﬁne the permutation operator P
ij
P
ij
ψ(. . . , i, . . . , j, . . .) = ψ(. . . , j, . . . , i, . . .) (5.2)
Here are a few simple theorems about permutations. The reader is invited
to supply the simple proofs.
1. For every symmetric operator S(1, . . . , N) we have
[P, S] = 0 (5.3)
An important example of a symmetric operator is the Hamiltonian.
This should be obvious, since a lowenergy Hamiltonian will have the
form
H =
i
p
2
i
2m
+
i,j
V
ij
where V
ij
is the interaction potential between particles i and j. From
Newton’s third law, we expect V
ij
= V
ji
.
2.
¸Pϕ[Pψ) = ¸ϕ[ψ), (5.4)
which follows by renaming the integration variables.
3. Permutations have no eﬀect on the expectation values of symmetric
operators.
¸Pϕ[S[Pψ) = ¸ϕ[P
†
SP[ψ) = ¸ϕ[S[ψ) (5.5)
4. If ¸ϕ[O[ψ) = ¸Pϕ[O[Pψ) for all ϕ and ψ, then O must be symmet
ric. We believe that all physical observables must be invariant under
all permutations of the particles, so all operators corresponding to
observable quantities must be symmetric.
It is well known that wave functions describing identical bosons must
be symmetric with respect to the exchange of any pair of particles. Func
tions describing identical fermions must be antisymmetric in the same sense.
There is a vast amount of experimental evidence to corroborate this. There
is also a deep result known as the spinstatistics theorem, which shows that
5.2. BOSON STATES 63
it is virtually impossible to construct a covariant theory that does not have
this property.
One way to make completely symmetric or antisymmetric states is simply
to multiply singleparticle states in all possible combinations. We’ll call the
basic ingredient [i)
α
. By this I mean that the ball that wears the number
α is in the quantum state given by i. We assume these are orthonormal,
¸i[j) = δ
ij
. We can write an Nparticle state
[i
1
, i
2
, . . . , i
N
) = [1)
1
[2)
2
[i
N
)
N
(5.6)
The symmetrized and antisymmetrized basis states are can be written
S
±
[i
1
, i
2
, . . . , i
N
) ≡
1
√
N!
P
(±)
P
P[i
1
, i
2
, . . . , i
N
) (5.7)
The sum goes over all of the N! possible permutations of N objects. Equa
tion (5.6) deﬁnes the symmetric and antisymmetricmaking operators S
±
.
The symbol (±)
P
requires further explanation. All the N! permutations of
N objects can be obtained by exchanging successive pairs of particles. The
P in (±)
P
stands for the number of such exchanges required to achieve the
corresponding permutation. Of course, the upper sign refers to bosons and
the lower, to fermions. It is a fact that no permutation can be reached by
an even and an odd number of exchanges. Put it another way, even and odd
permutations constitute two disjoint sets.
5.2 Boson States
We must allow for the possibility that there are several particles in one
quantum state. If, for example, there are n
i
particles in the ith state, there
will be n
i
! permutations that leave the Nparticle state unchanged. In this
case (5.6) will not be normalized to unity. A properly normalized state can
be constructed as follows:
[n
1
, n
2
, . . . ) = S
+
[i
1
, i
2
, . . . , i
N
)
1
√
n
1
!n
2
!
(5.8)
The numbers n
1
, n
2
, etc. are called occupation numbers. The sum of all
occupation numbers must equal to the total number of particles:
i
n
i
= N (5.9)
64CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
All this assumes that there are exactly N particles. We are interested
in precisely the situation in which the total number of particles is not ﬁxed.
We allow for this by taking the basis states to be the direct sum of the space
with no particles, the space with one particle, the space with two particles,
etc. A typical basis element is written [n
1
, n
2
, . . .). There in no constraint
on the sum of the n
i
. The normalization is
¸n
1
, n
2
, . . . [n
1
, n
2
, . . .) = δ
n
1
,n
1
δ
n
2
,n
2
(5.10)
and the completeness relation
n
1
,n
2
,...
[n
1
, n
2
, . . .)¸n
1
, n
2
, . . . [ = 1 (5.11)
Since there are physical processes that change the number of particles in
a system, it is necessary to deﬁne operators that have this action. The basic
operators for so doing are the creation and annihilation operators. As you
will see, they are formally equivalent to the raising and lowering operators
associated with the harmonic oscillator. For example, suppose a state has
n
i
particles in the i’th eigenstate. We introduce the creation operator a
†
i
by
a
†
i
[ . . . , n
i
, . . .) =
√
n
i
+ 1[ . . . , n
i
+ 1, . . .), (5.12)
ie. a
†
i
increases by one the number of particles in the i’th eigenstate. The
adjoint of this operator reduces by one the number of particles. This can
be seen as follows: Take the adjoint of (5.12) and multiply on the right by
[ . . . , n
i
+ 1, . . .).
¸. . . , n
i
, . . . [a
i
[ . . . , n
i
+ 1, . . .)
=
√
n
i
+ 1¸. . . , n
i
+ 1, . . . [ . . . , n
i
+ 1, . . .) =
√
n
i
+ 1
Now replace n
i
everywhere by n
i
−1.
¸. . . , n
i
−1, . . . [a
i
[ . . . , n
i
, . . .)
=
√
n
i
¸. . . , n
i
, . . . [ . . . , n
i
, . . .) =
√
n
i
(5.13)
The eﬀect of a
i
on the state [ . . . , n
i
, . . .) has been to produce a state in
which the number of particles in the i’th state has been reduced by one.
Eqn. (5.13) also tells us what the normalization must be. In short
a
i
[ . . . , n
i
, . . .) =
√
n
i
[ . . . , n
i
−1, . . .) for n
i
≥ 1 (5.14)
5.2. BOSON STATES 65
Of course if n
i
= 0, the result is identically zero.
a
i
[ . . . , n
i
= 0, . . .) = 0
The commutation relations are important. Clearly all the a
i
’s commute
among one another, since it makes no diﬀerence in what order the diﬀerent
states are depopulated, and by the same argument, the a
†
i
’s commute as
well. a
i
and a
†
j
also commute if i ,= j, since
a
i
a
†
j
[ . . . , n
i
, . . . , n
j
, . . .)
=
√
n
1
_
n
j
+ 1[ . . . , n
i
−1, . . . , n
j
+ 1, . . .)
= a
†
j
a
i
[ . . . , n
i
, . . . , n
j
, . . .)
Finally
_
a
i
a
†
i
−a
i
†a
i
_
[ . . . , n
i
, . . . , n
j
, . . .)
=
_√
n
i
+ 1
√
n
i
+ 1 −
√
n
i
√
n
i
_
[ . . . , n
i
, . . . , n
j
, . . .)
In summary
[a
i
, a
j
] = [a
†
i
, a
†
j
] = 0, [a
i
, a
†
j
] = δ
ij
(5.15)
If it were not for the i and j subscripts, (5.15) would be the commutation
relations for the harmonic oscillator, [a, a] = [a
†
, a
†
] = 0, [a, a
†
] = 1. In this
context they are called ladder operators or raising and lowering operators.
This is the essence of second quantization. Try to imagine a quantum system
as an inﬁnite forest of ladders, each one corresponding to one quantum state
labelled by an index i. The rungs of the i’th ladder are labelled by the integer
n
i
. The entire state of the system is uniquely speciﬁed by these occupation
numbers. The eﬀect of a
i
and a
†
i
is to bump the system down or up one
rung of the i’th ladder. There are several important results from harmonic
oscillator theory that can be taken over to second quantization. One is
that we can build up manyparticle states using the a
†
i
’s. Starting with the
vacuum state [0) with no particles, we can construct singleparticle states,
a
†
i
[0), twoparticle states
1
√
2!
_
a
†
i
_
2
[0) or a
†
i
a
†
j
[0),
or in general, manyparticles states.
[n
1
, n
2
, . . .) =
1
√
n
1
!n
2
!
_
a
†
1
_
n
1
_
a
†
2
_
n
2
[0) (5.16)
66CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
Another familiar result is that the number operator
N
i
= a
†
i
a
i
(5.17)
is a Hermitian operator whose eigenvalue is the number of particles in the
i’th quantum state.
N
i
[ . . . , n
i
, . . . ) = n
i
[ . . . , n
i
, . . . ) (5.18)
Here is a useful result that you can prove by brute force or induction.
_
a
i
,
_
a
†
i
_
n
_
= n
_
a
†
i
_
n−1
Use this to do the following exercises.
• Show that (5.16) has the normalization required by (5.10).
• Prove (5.18).
• Show that the mysterious factor of
√
n
i
+ 1 in (5.12) is in fact required
by (5.16).
5.3 More General Operators
The matrix element of a singleparticle operator
1
in a oneparticle state is
just
O
ij
= ¸i[
ˆ
O[j). (5.19)
If the [i)’s represent a complete set of states, we can represent the operator
as
ˆ
O =
i,j
O
ij
[i)¸j[ (5.20)
In a manyparticle system, we must distinguish between operators that work
on speciﬁc individual particles and those that operate on all the particles
in the system. One way to do this is to simply sum over oneparticle states
and generalize (5.20) as follows:
ˆ
O =
i,j
O
ij
N
α=1
[i)
α
¸j[
α
(5.21)
1
I will sometimes use “hats” to emphasize that some particular symbol represents
an operator. I will do this only occasionally, however, and I reserve the right to be
inconsistent.
5.4. FIELD OPERATORS 67
The index α points to the individual particles, but the sum over α is awk
ward; we would like to express it in terms of creation and annihilation oper
ators. Note that when
[i)
α
¸j[
α
operates on an arbitrary Nparticle state,
it produces n
j
terms in which [j) is replaced by [i), one for each particle in
the state, but these n
j
terms are still normalized as if they had n
j
particles
in the j’th state and n
i
particles in the i’th state. Taking this into account
gives
n
α=1
[i)
α
¸j[
α
[ . . . , n
i
, . . . , n
j
, . . .)
= n
j
[ . . . , n
i
+ 1, . . . , n
j
−1, . . .)
√
n
i
+ 1
√
n
j
=
√
n
j
√
n
i
+ 1[ . . . , n
i
+ 1, . . . , n
j
−1, . . .)
= a
†
i
a
j
[ . . . , n
i
, . . . , n
j
, . . .)
≡ N
ij
[ . . . , n
i
, . . . , n
j
, . . .) (5.22)
You see that N
ij
= a
†
i
a
j
is a generalization of (5.17). Evidentally, N
ii
as
deﬁned by (5.22) is the same as N
i
deﬁned by (5.17). Finally
ˆ
O =
ij
O
ij
N
ij
(5.23)
From here it’s only a hop, skip, and jump to deﬁning twoparticle operators.
ˆ
O
(2)
=
1
2
i,j,k,m
¸i, j[
ˆ
O
(2)
[k, m)a
†
i
a
†
j
a
k
a
m
, (5.24)
where
¸i, j[
ˆ
O
(2)
[k, m) =
_
dx
_
dy ϕ
∗
i
(x)ϕ
∗
j
(y)
ˆ
O
(2)
(x, y)ϕ
k
(x)ϕ
m
(y). (5.25)
The factor
1
2
in (5.25) insures that each interaction is counted only once,
since for identical particles
ˆ
O
(2)
(x, y) =
ˆ
O
(2)
(y, x).
5.4 Field Operators
I have used the symbol [i) to indicate a one particle “quantum state.” For
example, if one were dealing with hydrogen, i would stand for the discrete
eigenvalues of some complete set of commuting operators, in this case n,
68CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
l, m, and m
s
. The creation operators a
†
i
create particles in such a state.
Whatever these operators might be, however, none of them is the position
operator. An important question is what the creation operator formalism
looks like in terms of the position operator. First consider two basis systems
based on two alternative sets of commuting observables. Use i to index one
set and λ to index the other. Then
[λ) =
i
[i)¸i[λ). (5.26)
Since this is true for states, it must also be true of creation operators.
a
†
λ
=
i
¸i[λ)a
†
i
(5.27)
Now remember that what we call a wave function in elementary quantum
mechanics is really a scalar product on Hilbert space of the corresponding
state and eigenstates of the position operator, i.e.
¸x[i) = ϕ
i
(x). (5.28)
We assume that the ϕ
i
are a complete set of orthonormal functions, so that
_
d
3
x ϕ
∗
i
(x)ϕ
j
(x) = δ
ij
(5.29)
and
i
ϕ
∗
i
(x)ϕ
i
(x
) = δ
(3)
(x −x
) (5.30)
Combining (5.27) and (5.28) gives
ψ
†
(x) =
i
ϕ
∗
i
(x)a
†
i
(5.31)
and its adjoint
ψ(x) =
i
ϕ
i
(x)a
i
ψ
†
(x) is the creation operator in the basis deﬁned by the position opera
tor. It creates a particle at x. ψ
†
and ψ are called ﬁeld operators. Their
commutation relations are important.
[ψ(x), ψ(x
)]
±
= [ψ
†
(x), ψ
†
(x
)]
±
= 0 (5.32)
[ψ(x), ψ
†
(x
)]
±
= δ
(3)
(x −x
)
5.4. FIELD OPERATORS 69
I have used the symbol [ , ]
±
to allow for both boson (commutation)
and fermion (anticommutation) rules. The ﬁrst line of (5.32) is more or less
obvious. The second line follows from (5.30)
Quantum mechanics according to Heisenberg was obtained by replacing
classical quantities like P and L with noncommuting operators. Our for
malism takes the further step of making the wave functions themselves into
operators via (5.32). This, I suppose, is the reason why it’s called “second
quantization.” Many texts in fact, start with (5.32) and use it to infer the
existence of the creation and annihilation operators. Let’s see what ordi
nary quantum mechanics looks like in this formalism and then see what new
aspects appear. Start with the kinetic energy. Eqn. (5.19) gives the matrix
element
T
ij
=
_
d
3
x ϕ
∗
i
(x)
_
−
¯h
2
2m
∇
2
_
ϕ
j
(x) =
¯h
2
2m
_
d
3
x ∇ϕ
∗
i
(x) ∇ϕ
j
(x). (5.33)
The last step involves an integration by parts. Eqn.(5.23) becomes
ˆ
T =
¯h
2
2m
ij
_
d
3
x a
†
i
∇ϕ
∗
i
(x) ∇ϕ
j
a
j
(x) =
¯h
2
2m
_
d
3
x ∇ψ
†
(x) ∇ψ(x)
(5.34)
The potential operator is similar.
ˆ
U =
_
d
3
x ψ
†
(x)U(x)ψ(x) (5.35)
Equations (5.24) and (5.25) give us to form of any twoparticle operator,
ˆ
V =
1
2
i,j,k,m
_
d
3
x
_
d
3
y a
†
i
a
†
j
ϕ
∗
i
(x)ϕ
∗
j
(y)V (x, y)ϕ
k
(x)ϕ
m
(y)a
k
a
m
(5.36)
=
1
2
_
d
3
x
_
d
3
y ψ
†
(x)ψ
†
(y)V (x, y)ψ(y)ψ(x)
The total Hamiltonian operator is then
ˆ
H =
ˆ
T +
ˆ
U +
ˆ
V (5.37)
From (5.17) N =
i
a
†
i
a
i
is the total number of particles operator. It
follows that ρ(x, t) = ψ
†
(x, t)ψ(x, t) is the spatial density operator. This
can be seen as follows. Inverting (5.31) gives
a
†
i
(t) =
_
d
3
x ψ
†
(x, t)ϕ
∗
i
(x)
70CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
a
i
(t) =
_
d
3
x ϕ
i
(x)ψ(x)
N =
i
a
†
i
a
i
=
i
__
d
3
x ψ
†
(x, t)ϕ
∗
i
(x, t)
_ __
d
3
x
ϕ
i
(x
)ψ(x
, t)
_
=
_
d
3
x
_
d
3
x
ψ
†
(x, t)ψ(x
, t)δ
(3)
(x −x
)
=
_
d
3
x ψ
†
(x, t)ψ(x, t) =
_
d
3
x ρ(x, t)
Notice that
ρ(x
, t)ψ
†
(x, t)[0) = ψ
†
(x
, t)ψ(x
, t)ψ
†
(x, t)[0)
= ψ
†
(x
, t)
_
ψ
†
(x, t)ψ(x
, t) +δ
(3)
(x −x
)
_
[0)
= δ
(3)
(x −x
)ψ
†
(x
, t)[0) = δ
(3)
(x −x
)ψ
†
(x, t)[0)
The point of this is that ψ
†
(x, t)[0) is an eigenstate of ρ with eigenvalue
δ
(3)
(x−x
). This is consistent both with our interpretation of ψ
†
as a creation
operator and with our interpretation of ρ as a particle density operator.
ψ
†
(x, t) does indeed create a pointlike particle at x and time t.
5.5 Field Equations
The ﬁeld operators so far have had no time dependence. We have been
working in the Schrodinger picture. Field theory calculations are typically
done in the Heisenberg picture, however, so we must derive an equation of
motion. The basic relation is
ψ(x, t) = e
iHt/¯h
ψ(x, 0)e
−iHt/¯h
(5.38)
Where H is the operator given by (5.37). The Heisenberg equation of motion
is obtained by diﬀerentiating (5.38),
i¯h
∂
∂t
ψ((x), t) = −[H, ψ(x, t)] = −e
iHt/¯h
[H, ψ(x, 0)]e
−iHt/¯h
(5.39)
The commutator is something of a challenge. We’ll do it one term at a time.
[T, ψ(x)] =
_
−
¯h
2
2m
_
_
_
ij
_
d
3
x
ϕ
∗
i
(x
)∇
2
ϕ
j
(x
)a
†
i
a
j
,
k
ϕ
k
(x)a
k
_
_
5.5. FIELD EQUATIONS 71
I have used (5.33) for T and (5.31) for ψ. At this point we use the com
mutation relations, either (5.15) for bosons or (5.93) for fermions, to reduce
the product of three a
i
’s.
[a
†
i
a
j
, a
k
] = −a
j
δ
ki
[T, ψ(x)] =
_
¯h
2
2m
__
d
3
x
i
ϕ
i
(x)ϕ
∗
i
(x
)∇
2
j
ϕ
j
(x
)a
j
=
_
¯h
2
2m
__
d
3
x
δ
(3)
(x −x
)∇
2
ψ(x
) =
_
¯h
2
2m
_
∇
2
ψ(x) (5.40)
The potential energy is rather easier.
[U, ψ(x)] =
_
d
3
xU(x
)[ψ
†
(x
)ψ(x
), ψ(x)] = −U(x)ψ(x) (5.41)
Finally, the interaction term. A straightforward application of (5.32) and
(5.36) gives
[V, ψ(x)] =
1
2
_
d
3
x
d
3
x
[ψ
†
(x
)ψ
†
(x
)V (x
, x
)ψ(x
)ψ(x
), ψ(x)]
= −
_
d
3
x
ψ(x
)V (x, x
)ψ(x
)ψ(x) (5.42)
Incidentally, these results are valid for both boson and fermion ﬁelds so long
as one keeps the ﬁelds ordered in exactly the way shown above. (The reader
is invited to work through these derivations to see how this comes about.
Finally combining (5.37), (5.39), (5.40), (5.41), and (5.42) gives the equation
of motion
i¯h
∂
∂t
ψ(x, t) =
_
−
¯h
2
2m
∇
2
+U(x)
_
ψ(x, t)+
_
d
3
x
ψ
†
(x
, t)V (x, x
)ψ(x
, t)ψ(x, t) (5.43)
The ﬁrst line looks like the conventional Schrodinger equation but with ﬁeld
operators rather than the usual wave functions. The last term is nonlinear
and has no counterpart in singleparticle systems.
72CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
5.6 Momentum Representation
As we have seen before, it’s usually easier to formulate a theory in position
space and easier to interpret it in momentum space. In this case we work
exclusively in a ﬁnite volume with discrete momentum eigenvalues. The
basic eigenfunctions are
ϕ
k
=
1
√
V
e
ik·x
(5.44)
We assume the usual periodic boundary conditions force the momentum
eigenvalues to be
k = 2π
_
n
x
L
x
,
n
y
L
y
,
n
z
L
z
_
(5.45)
Where each of the n’s can take the values 0, ±1, ±2, independently. With
this normalization, the eigenfunctions are orthonormal,
_
V
d
3
x ϕ
∗
k
(x)ϕ
k
(x) = δ
k,k
(5.46)
We can now calculate matrix elements of the Hamiltonian. Eqn. (5.33)
gives
¸k[
ˆ
T[k
) =
_
d
3
x ϕ
∗
k
(x)
_
−
¯h
2
2m
∇
2
_
ϕ
k
(x) =
¯h
2
k
2
2m
δ
k,k
(5.47)
For the potential energy
¸k[
ˆ
U[k
) =
_
d
3
x ϕ
∗
k
(x)U(x)ϕ
k
(x) =
1
V
_
d
3
x e
i(k−k
)·x
U(x) (5.48)
≡
1
V
U
k−k
This is the Fourier transform of U(x) with respect to the momentum k −k
.
The twoparticle potential in (5.36) should only be a function of the diﬀer
ence of the two coordinates, ie. V (x, y) = V ([x −y[). As with U(x) we can
take the Fourier transform
2
V
q
≡
_
d
3
x e
−iq·x
V (x) (5.49)
V (x) =
1
V
q
V
q
e
iq·x
2
Anyone who confuses the volume V with the potential V will be taken out and tied
to a tree.
5.7. SUPERFLUIDS 73
The matrix element of V is then
¸p
, k
[V (k −k
)[p, k) =
1
V
q
V
q
δ
q,p
−p
δ
q,k−k
(5.50)
The complete Hamiltonian is then
H =
k
(¯ hk)
2
2m
a
†
k
a
k
+
1
V
k
,k
U
k
−k
a
†
k
a
k
+
1
2V
∆
V
k−k
a
†
k
a
†
p
a
k
a
p
(5.51)
The meaning of the sum over ∆ is that one sums over all values of k, p, k
, p
such that k +p = k
+p
. This is easy to interpret. Two particles with mo
menta k and p interact in such a way that there is a momentumk −k
= p −p
transferred between them. After this scattering process their new momenta
are k
and p
.
5.7 Superﬂuids
4
He atoms have zero angular moment and as such are, to a good approx
imation, Bose “particles”. Helium is unique in that it does not crystalize
even in the limit of zero temperature (except under extreme high pressure).
The reason is that it is so light that its endpoint oscillations have suﬃcient
energy to prevent the formation of a lattice. Helium does liquify at T ≈ 4K
and at the lambda point T
λ
= 2.18K it enters the superﬂuid state with
some remarkable properties. One of the most striking is the ability to ﬂow
through narrow channels without friction. Another is the existence of quan
tized vortices. The occurrence of frictionless ﬂow can be described in a rough
qualitative way by the twoﬂuid model. The two ﬂuids – the normal and the
superﬂuid components – are interpenetrating, and their densities depend on
temperature. At very low temperatures the density of the normal compo
nent vanishes, while the density of the superﬂuid component approaches the
total density of the liquid. Near the transition temperature to the normal
state, the situations is reversed: the superﬂuid density tends towards zero
while the normal density approaches the density of the liquid.
At a more microscopic level, superﬂuidity is explained in terms of ele
mentary excitations. In an ideal gas an elementary excitation corresponds to
the addition of a single quasiparticle in a momentum eigenstate. For small
momenta the excitations in liquid
4
He are sound waves or phonons. The
phonons have a linear dispersion relation E ∝ k, which is to say that they
are massless particles. This point is crucial to the phenomena of superﬂu
idity as can be seen from the following argument. Imagine helium ﬂowing
74CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
Coordinate System S S
Ground State E = E
0
E
= E
0
+Mv
2
/2
P = 0 P
= −Mv
GS + One Quasiparticle E = E
0
+(p) E
= E
0
+(p) −p v +Mv
2
/2
P = p P
= p −Mv
Excitation Energy ∆E = (p) ∆E
= (p) −p v
Table 5.1: Kinematics for Quasiparticle Formation
through a tube at constant velocity. We can describe this in terms of two
diﬀerent coordinate systems. In the system S the helium is at rest and the
tube is moving with velocity v, and in system S
the tube is at rest and the
helium has velocity −v. Evidentally S
is moving relative to S with velocity
v. Momentum and energy are related by the Galilean transformation,
P
= P−Mv (5.52)
E
= E −P v +Mv
2
/2
If the ﬂuid is ordinary, ie. not a superﬂuid, there will be friction between
the ﬂuid and the walls of the tube. In S it would appear that the tube was
dragging a thin layer of ﬂuid along with the it. In S
the tube is decelerating
this same layer. In the limit T = 0, this deceleration comes about through
the creation of quasiparticles, and this is only kinematically possible if it
lowers the total energy of the ﬂuid. Let’s try to create our ﬁrst quasiparticle
in S with energy (p) and momentum p. Call the ground state energy of
the ﬂuid E
0
and its momentum, P. Table 1 shows the relevant kinematics.
Only when ∆E
< 0 does the ﬂowing ﬂuid lose energy. This means there is
a minimum velocity
v
min
=
p
. (5.53)
This is called the Landau criterion. If the ﬂow velocity is smaller that this,
no quasiparticles are excited and the ﬂuid ﬂows unimpeded and lossfree
through the tube. If the quasiparticles had ﬁnite mass so that (v) = p
2
/2m,
then v
min
= 0 in the limit p = 0, and there could be no superﬂuidity. It is
essential that /p pass through the origin with a ﬁnite positive slope.
It is possible to calculate the dispersion relation for lowenergy quasipar
ticles using the ﬁeld theory formalism we have just developed. This was done
originally by N.N. Bogoliubov,
3
and is one of the crown jewels of manybody
3
N. N. Bogoliubov, J. Phys. U.S.S.R. 11, 23 (1947)
5.7. SUPERFLUIDS 75
theory. Though this was not realized in 1947, this calculation introduced
into physics some profound ideas about symmetry breaking that are impor
tant in theories of fundamental interactions. I will take you through the
Bogoliubov calculation now and come back to the fundamental issues later.
5.7.1 Bogoliubov Theory
We have derived the relevant Hamiltonians, (5.33) through (5.37) in coor
dinate space and (5.51) in momentum space. Since we are not concerned
with the eﬀect of external potentials, we set U = 0. Notice that (5.37) in
invariant under the simple gauge transformation,
ψ →ψ
= e
−iα
ψ. (5.54)
Moreover (5.51) obeys a simple conservation law; is conserves the number
of particles in the system. Each term has the same number of creation
and annihilation operators. For every particle created, one is annihilated
and vice versa. Though it’s by no means obvious, these two properties are
closely related. For every continuous symmetry transformation that leaves a
Hamiltonian invariant there is a corresponding conservation law. This deep
result is called Noether’s theorem. I will have much more to say about it
later.
We know from statistical mechanics, that if the particles do not interact
with one another, then in the limit of zero temperature, they are all in
the ground state. If we introduce a weak twoparticle interaction via V in
(5.51), then it is plausible that most of the particles remain in the ground
state. The crucial approximation consists in ignoring interactions between
excited particles and keeping only those terms corresponding to particles
in the ground state interacting with one another and terms in which one
excited particle interacts with one particle in the ground state. That means
keeping in (5.51) only those terms that have at least two of the ground state
creation and annihilation operators a
i
and a
†
i
. We are left with
H ≈
k
(¯ hk)
2
2m
a
†
k
a
k
+
1
2V
V
0
a
†
0
a
†
0
a
0
a
0
+
1
V
k =0
(V
0
+V
k
)a
†
0
a
0
a
†
k
a
k
+
1
2V
k=0
V
k
(a
†
k
a
†
−k
a
0
a
0
+a
†
0
a
†
0
a
k
a
−k
) (5.55)
We have ignored those terms with three a
0
’s, because they would not con
serve momentum. Bogoliubov argued that the a
0
and a
†
0
operators in (5.55)
76CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
can be replaced with cnumbers, a
0
→
√
N
0
, a
†
0
→
√
N
0
where N
0
is the
number of particles in the ground state. For one thing, the system is in an
eigenstate of a
†
0
a
0
(with eigenvalue N
0
) and of a
0
a
†
0
(with eigenvalue N
0
+1).
It ﬁgures that since N
0
is a number on the order of 10
20
or 10
22
, we can
neglect the 1 in N
0
+ 1. To this extent a
0
and a
†
0
commute! Another point
the that the a
0
’s and a
†
0
’s are all divided by the volume V .
1
V
[a
0
, a
†
0
][N
0
) =
1
V
[N
0
) (5.56)
The left side of (5.56) consist of two terms, both equal to the particle density.
This right side goes to zero as V → ∞. With that replacement (5.55)
becomes
H ≈
k
(¯ hk)
2
2m
a
†
k
a
k
+
1
2V
N
2
0
V
0
+
N
0
V
k=0
_
(V
0
+V
k
)a
†
k
a
k
+
1
2
V
k
(a
†
k
a
†
−k
+a
k
a
−k
)
_
(5.57)
The a
†
k
a
k
operator has three contributions. The ﬁrst is the kinetic en
ergy. The second, N
0
V
0
/V , is called the Hartree energy , which comes from
the direct interaction of a particle in the state k with the N
0
atoms in the
zeromomentum state. The third is the exchange , or Fock term, in which
an atom in the state k is scattered into the zeromomentum state, while a
second atom is simultaneously scattered from the condensate to the state k.
We can get some insight about the approximations made so far by notic
ing the following. We obtained (5.51) from (5.36) by replacing
ϕ(x) =
1
√
V
k
a
k
e
ik·x
. (5.58)
If we had made the replacement
ϕ(x) →
√
ρ
0
+
1
√
V
k=0
a
k
e
ik·x
(5.59)
where ρ
0
= N
0
/V , we would have gotten (5.57) directly. You see that
(5.59) diﬀers from (5.58) in two related ways: the classical density ρ
0
and
the k = 0 exclusion in the summation. The meaning is clear. We are
treating the ground state like a classical ﬂuid and all the other particles as
a dilute quantum ﬂuid. I should mention a related matter. The original
Hamiltonian conserved particle number: for every creation operator there
5.7. SUPERFLUIDS 77
was a corresponding annihilation operator. Eqn. (5.57) no longer does. The
particles simply appear out of or disappear into the ground state where they
are “invisible,” since we are treating the ground state classically. We will
have much more to say about this later on in the course when we discuss
spontaneous symmetry breaking.
Eqn. (5.57) refers to N
0
, the number of particles in the ground state,
but this is not something we know ahead of time. We do know the total
number of particles, that’s something that can be measured. The two can
be related as follows:
N = N
0
+
k=0
a
†
k
a
k
(5.60)
You see the similarity between (5.60) and (5.59). N
0
is a cnumber, but
the number of excited particles is obtained via an operator operating on the
state function. Substituting (5.60) into (5.57) gives
H ≈
k=0
E
k
a
†
k
a
k
+
N
2
2V
V
0
+
N
2V
k=0
V
k
(a
†
k
a
†
−k
+a
k
a
−k
) (5.61)
E
k
=
k
2
2m
+
N
V
V
k
(5.62)
From (5.49), V
k
= V
−k
, so that (5.61) can be written in a more symmetric
fashion,
H ≈
N
2
2V
V
0
+
k>0
_
E
k
(a
†
k
a
k
+a
†
−k
a
−k
) +
N
V
V
k
(a
†
k
a
†
−k
+a
k
a
−k
)
_
(5.63)
5.7.2 The Bogoliubov Transformation
Physically realizable states must be eigenstates of the Hamiltonian, but
systems with a deﬁnite number of particles in the various energy levels are
not eigenstates of (5.63), since it does not conserve particle number. In
order to study the physical states, we must diagonalize H. The procedure
for doing this was originally invented by Bogoliubov in this context, and
the technique as since been applied to many diﬀerent problems. In order
to work this out in a generic fashion (and also to keep the algebra under
control), I will pose the problem like this.
4
Each term in the sum in (5.63)
has the form
H =
0
(a
†
a +b
†
b) +
1
(a
†
b
†
+ba) (5.64)
4
I am following the development in BoseEinstein Condensation in Dilute Gasses, C.
J. Pethick & H. Smith, Cambridge (2002)
78CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
In this particular instance, a = a
k
and b = a
−k
, but the ﬁnal result will hold
with the generality of (5.64).
0
and
1
are arbitrary numbers with units of
energy, or since ¯ h = 1, of frequency. Of course
[a, a
†
] = [b, b
†
] = 1, and [a, b] = [a, b
†
] = [a
†
, b] = [a
†
, b
†
] = 0 (5.65)
We would like to ﬁnd linear combinations of these
α = ua +vb
†
, and β = ub +va
†
(5.66)
such that
•
[α, α
†
] = [β, β
†
] = 1, and [α, β
†
] = [β, α
†
] = 0 (5.67)
• The only operators appearing in the Hamiltonian are α
†
α and β
†
β.
In other words, we will diagonalize the Hamiltonian, and resulting states,
whatever they may be, will be the observable, physical states. Insert (5.66)
into (5.67) and use (5.65). Assuming u and v to be real we get,
u
2
−v
2
= 1. (5.68)
The inverse of (5.66) is
a = uα −vβ
†
, b = uβ −vα
†
. (5.69)
Now substitute (5.69) into (5.64). After a bit of algebra we get,
H = 2v
2
0
−2uv
1
+ [
0
(u
2
+v
2
) −2uv
1
](α
†
α +β
†
β)
+[
1
(u
2
+v
2
) −2uv
0
](αβ +β
†
α
†
) (5.70)
We would like the last term to vanish, i.e.
1
(u
2
+v
2
) −2uv
0
= 0 (5.71)
Combining (5.68) and (5.71) gives
u
2
=
1
2
_
0
+ 1
_
and v
2
=
1
2
_
0
−1
_
, (5.72)
where
=
_
2
0
−
2
1
(5.73)
5.7. SUPERFLUIDS 79
It is necessary to choose the positive branch of the square root so that u
and v are real. Substituting these into (5.70) gives
H = (α
†
α +β
†
β) + −
0
(5.74)
It’s time to make contact with our original Hamiltonian, (5.61) or (5.63).
The parameter we have been calling is best thought of as a frequency.
Where convenient I will use n ≡ N/V .
0
→
k
2
2m
+nV
k
1
→nV
k
→ω
k
≡
¸
_
k
2
2m
_
2
+
nk
2
V
k
m
(5.75)
H ≈
N
2
2V
V
0
+
k>0
_
ω
k
(α
†
k
α
k
+β
†
k
β
k
) +ω
k
−
_
k
2
2m
+nV
k
__
=
N
2
2V
V
0
−
1
2
k=0
_
k
2
2m
+nV
k
−ω
k
_
+
k=0
ω
k
α
†
k
α
k
(5.76)
The Hamiltonian consists of the groundstate energy and a sum of oscillators
of energy ω
k
. Our new creation and destruction operators α
†
k
and α
k
refer
to the quasiparticles discussed earlier. The ground state of the system is
that in which there are no quasiparticles.
α
k
[Ω) = 0 for all k. (5.77)
This new ground state plays the same role for the α
k
’s as the quantum
vacuum [0) plays for the a
k
’s. It is in some sense, the new vacuum. I’m sure
this was not realized back in 1947 when this work was originally published,
but we are standing on the threshold of a profound new idea. The point
is that in making the transition from the Hamiltonian (5.51) and the ﬁeld
operator (5.31) to The Hamiltonian (5.63) and ﬁeld (5.59), we automatically
make a transition to a new vacuum with some unusual properties. I will get
back to that in a minute.
This solution makes an interesting point about perturbation theory.
Let’s calculate the number of real particles, helium atoms say, that are
excited out of the ground state.
N
ex
= ¸Ω[
k=0
a
†
k
a
k
[Ω) = ¸Ω[
k=0
v
2
k
α
k
α
†
k
[Ω) =
k=0
v
2
k
(5.78)
80CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
To make my point, I will take the simplest possible potential, V (x) = λδ(x),
so that V
k
= λ. If we were doing ordinary perturbation theory, we would
pray that λ was “small” and expand the solution in powers of nλ. In this
case we can get a solution which is “exact” in the sense that we don’t need
this approximation. We can replace the sum in (5.78) with an integral. After
a horrendous integration we get the simple result
n
ex
≡
N
ex
V
=
m
3/2
3π
2
(λn)
3/2
(5.79)
Now λ
3/2
considered as a complex function of λ is nonanalytic at λ = 0.
It has a branch point. You can’t make a power series expansion around
a branch point; it just doesn’t make sense. In this problem perturbation
theory, in the usual sense, would always get the wrong answer. It’s a sobering
thought that in scattering theory we always use perturbation theory. We
aren’t smart enough to do anything else. And it could all be wrong.
It’s clear that ω
k
from Eqn. (5.75) is the energy of a single quasiparticle.
In the limit of small k we have
ω
k
= vk with v =
_
nV
0
m
(5.80)
The linear dispersion relation is the hallmark of massless particles. If there
were no potential or the potential somehow vanished at low energy, the
dispersion relation would be
ω
k
=
k
2
2m
.
This is the relation for an ordinary particle of mass m, i.e. the potential
creates massless particles. We could say that (5.75) interpolates between
phononlike behavior at small k and massive particlelike behavior at large
k.
Now let’s get back to the ground state. Since the α
k
’s and a
k
’s are
related by a canonical transformation, there must exist an unitary operator
U such that
α
k
= Ua
k
U
−1
a
k
= U
−1
α
k
U. (5.81)
It is possible to ﬁnd an exact form for this operator,
5
but for our purposes it
is enough to know that it exists. Since a
k
[0) = 0, it must be that α
k
U[0) = 0.
Look at (5.77). It must be that
[Ω) = U[0). (5.82)
5
This is assigned as Problem 17.4 in Gauge Theories in Particle Physics, Vol. II, I.J.R.
Aitchison and A. J. G. Hey, Institute of Physics Publishing, 2004
5.7. SUPERFLUIDS 81
The exact form of U shows that [Ω) is a linear superposition of states with
diﬀerent numbers of particles, therefor it is not an eigenstate of the num
ber operator (5.17). We had an intimation of this, since the Hamiltonian
(5.57) does not conserve particle number. It will turn out that the conser
vation of particle number is related to gauge transformations. Our original
Hamiltonian is invariant under the simple U(1) symmetry transformation,
ϕ(x) → e
−iα
ϕ(x), whereas the modiﬁed Hamiltonian is not. It turns out
that the breaking of this symmetry is directly responsible for the existence
of the massless quasiparticles. I will prove this assertion when we get around
to the Goldstone theorem.
The expectation value of the ﬁeld ϕ(x) from (5.58) for any state with a
deﬁnite number of particles will always vanish. Let [N) be any state with
N noninteracting particles. Then
¸N[ϕ(x)[N) = 0,
but this is not true of our new ﬁeld (5.59).
¸N[ ˜ ϕ(x)[N) = ρ
1/2
0
. (5.83)
We can rewrite (5.59) in terms of the α
k
’s. Using (5.69) and (5.75) we get
˜ ϕ(x) = ρ
1/2
0
+
1
√
V
k
(u
k
α
k
−v
k
α
−k
)e
ik·x
Then from (5.77),
¸Ω[ ˜ ϕ(x)[Ω) = ρ
1/2
0
(5.84)
All this was obtained as the end result of a number of approximations.
Suppose we could do the theory exactly (whatever that means), and ﬁnd
the “true” ﬁeld and vacuum state. We postulate that something of (5.84)
would survive, that in fact
¸Ω[ ˜ ϕ(x)[Ω) , = 0. (5.85)
While we are speculating, let’s guess that because we are not making
any approximations, the Hamiltonian would preserve the U(1) symmetry
discussed in connection with (5.54). Then we should not change any physical
results by replacing
˜ ϕ(x) → ˜ ϕ
(x) = e
−iα
˜ ϕ(x) (5.86)
This transformation can be brought about by a unitary operator U
α
, ˜ ϕ
=
U
α
˜ ϕU
−1
α
. Then
¸Ω[U
α
˜ ϕU
−1
α
[Ω) = e
−iα
¸Ω[ ˜ ϕ[Ω) (5.87)
82CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
I choose to interpret (5.87) as follows. I regard U
−1
α
[Ω) ≡ [Ω, α) as an
alternative ground state. Then (5.87) becomes
¸Ω, α[ ˜ ϕ[Ω, α) = e
−iα
¸Ω[ ˜ ϕ[Ω) (5.88)
Obviously there are an inﬁnite number of these ground states, (5.84) corre
sponds to the special case α = 0, but the choice of α can have no observable
physical consequences.
It’s time to summarize the plot. It is postulated that our exact Hamil
tonian is invariant under the U(1) transformation (5.54). The Hamiltonian
gives rise to an inﬁnite number of alternative ground states, however, and
in choosing one of them we break the symmetry. This is called “spontaneous
symmetry breaking,” and I will have much more to say about it later. In the
meantime, there is another example of spontaneous symmetry breaking that
you’re all familiar with, the ferromagnet. In this case the broken symmetry
is rotation invariance. Surely the Hamiltonian for the interactions of atoms
and electrons is invariant under rotation, and so an inﬁnite mass of molten
iron is also, but when the iron is cooled below the Curie point, the iron
spontaneously (spontaneously – get it?) “chooses” one direction to line up
all the magnetic moments, resulting in a ground state that has a preferred
direction. Thus rotation invariance is broken. It is widely speculated that
the “Hamiltonian of the universe” has many symmetries of which we are un
aware, but we are living in or close to the ground state of this Hamiltonian,
and much of the physical world is the end result of this sort of symmetry
breaking.
5.8 Fermions
The fact that fermion wave functions are antisymmetric introduces a few
small complications regarding the creation and annihilation operators. They
are easy to explain looking at twoparticle states. When I write [i
1
, i
2
), I
mean that particle 1 is in state i
1
, which is to say that the leftmost entry
in the ket refers to particle 1, the second entry on the left refers to particle
2, etc. Antisymmetry then decrees that [i
1
, i
2
) = −[i
2
, i
1
). If both particles
were in the same state [i
1
, i
1
) = −[i
1
, i
1
), so double occupancy is impossible.
If I describe this state in terms of occupation numbers [n
1
, n
2
), the leftmost
entry refers to the ﬁrst quantum state (rather than the ﬁrst particle), but
which state is the ﬁrst state? You have to decide on some convention for
ordering states and then be consistent about it.
5.8. FERMIONS 83
These states are constructed with creation and annihilation operators
as in the case of bosons, but now we must be more careful about ordering.
Consider the following two expressions.
a
†
i
1
a
†
i
2
[0) = [i
1
, i
2
)
a
†
i
2
a
†
i
1
[0) = [i
2
, i
1
) = −[i
1
, i
2
)
I have decreed, and again this is a convention, that the ﬁrst operator on the
left creates particle 1, etc. Obviously
a
†
i
1
a
†
i
2
+a
†
i
2
a
†
i
1
= 0 (5.89)
We say that fermion operators anticommute. The usual notation is
[A, B]
+
= ¦A, B¦ = AB +BA (5.90)
Of course, fermions don’t have numbers painted on them any more than
bosons do, so we must use occupation numbers. Here the convention is
[n
1
, n
2
, ) =
_
a
†
i
_
n
1
_
a
†
2
_
n
2
[0) n
i
= 0, 1 (5.91)
The eﬀect of the operator a
†
i
must be
a
†
i
[ . . . , n
i
, . . .) = η[ . . . , n
i
+ 1, . . .), (5.92)
where η = 0 of n
i
= 1 and otherwise η = +/−, depending on the number
of anticommutations necessary to bring the a
†
i
to the position i. The com
mutation relations of the a
i
’s and a
†
i
’s are obtained by arguments similar to
those leading to (5.15). The results are
¦a
i
, a
j
¦ = ¦a
†
i
, a
†
j
¦ = 0 ¦a
i
, a
†
j
¦ = δ
ij
(5.93)
One can construct one and twoparticle operators just as we did for bosons.
Formulas (5.23) and (5.24) are correct as they stand, but one must be careful
to keep the creation and annihilation operators in the exact order in which
they appear in these equations.
6
6
Strictly speaking, I should prove this. If you want to see all the excruciating details,
consult F. Schabel Advanced Quantum Mechanics, Section 1.4.2
84CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
5.8.1 Inclusion of Spin
Spin is just “tacked on” to nonrelativistic quantum mechanics. The wave
function for a spin1/2 particle is just
ψ
σ
(r, t) = ψ(r, t)u
σ
(t). (5.94)
The spin wave function u
σ
is a twocomponent column matrix. The subscript
σ is an index that takes on the values ±1/2 corresponding to spin “up” or
“down” (with respect to some axis of quantization). The space wave function
is ψ(r, t). The point is that spin and space degrees of freedom are completely
decoupled. (This is not true in relativistic theory.) Sometime we need to
specify that a particle is in a speciﬁc spin state. This can be done in cartoon
fashion by writing ψ
↑
or ψ
↓
.
It’s easy to incorporate spin into Fockstate formulation. For example,
the operator a
†
kσ
creates a particle in a momentum and spin eigenstate with
momentum k and zprojection of spin σ. If we say that there are n
i
particles
in the i’th eigenstate, we must remember that a complete speciﬁcation of
the i’th eigenstate must include the spin projection σ
i
.
5.8.2 The Fermi Sphere and the Debye Cutoﬀ
The ground state of a system of bosons is one in which all the particles
have zero energy. Because of the exclusion principle, this is not possible
with fermions. In the case of fermions the ground state is that in which all
the energies are as small as possible. Let’s try to make that more precise.
Since the kinetic energy is a function of k
2
, T = ¯ h
2
k
2
/2m, there will be a
certain momentum associated with the highest occupied energy level. This
is called the Fermi momentum ¯ hk
F
. The ground state can be constructed
in momentum space as
[Ω) =
k
k<k
F
σ
a
†
kσ
[0) (5.95)
The set of all wave numbers [k[ < k
F
is called the Fermi sphere. Suppose
the system is contained in a cube of length L on a side. If we impose periodic
boundary conditions along x axis for example, the wave numbers are limited
by the condition e
ik
x
x
= e
ik
x
(x+L)
, or k
x
L = 2πn
x
, where n
x
is an integer.
dn
x
=
L
2π
dk
x
, (5.96)
5.8. FERMIONS 85
where dn
x
is the number of allowed momentum states in the interval between
k
x
and k
x
+dk
x
. The number of states in the Fermi sphere is then
N = 2V
_
k
f
0
d
3
k
(2π)
3
=
V k
3
F
3π
2
(5.97)
The factor of 2 in front of the integral arises because there are two allowed
spin states for every one momentum state. It follows that
k
3
F
= 3π
2
n, (5.98)
where n = N/V is the average particle density. The Fermi energy is deﬁned
by
F
= (¯ hk
F
)
2
/(2m). Typically it is on the order of a few eV ’s.
At absolute zero temperature all the electrons would be inside the Fermi
sphere. The simplest excitation of this system is obtained by promoting
an electron from inside the sphere to an energy level outside. In terms of
creation and annihilation operators, such a state is written
[k
2
σ
2
, k
1
σ
1
) = a
†
k
2
σ
2
a
k
1
σ
1
[Ω) (5.99)
One can think of this as the creation of an electronhole pair. We can deﬁne
hole creation and annihilation operators as b
kσ
≡ a
†
−k,−σ
and b
†
kσ
≡ a
−k,−σ
.
Then (5.98) becomes
[k
2
σ
2
, k
1
σ
1
) = a
†
k
2
σ
2
b
†
−k
1
,−σ
1
[Ω) (5.100)
Bosons do not obey the exclusion principle, so there is nothing exactly
like the Fermi sphere for integerspin particles. There is something simi
lar for lowenergy phonons in crystalline solids, however, called the Debye
cutoﬀ. Phonons in crystals are sound waves in which the crystal planes os
cillate longitudinally along the direction of propagation. The fact that the
oscillators are discrete imposes limitations on the wave numbers that can
exist. Consider a line of length L consisting of N + 1 atoms spaced at a
distance a. Assume that the ﬁrst and last atoms don’t move because they
are at the surface of the crystal. The normal modes will be standing waves
in which there are an integer number of halfwavelengths in the length L.
Consequently, the allowed values are
k =
π
L
,
2π
L
,
3π
L
, ,
(N −1)π
L
(5.101)
Notice that if k = Nπ/L, there would be a node at the site of each atom, so
that no atom could move! It doesn’t make sense to have halfwavelengths
86CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
smaller than the interatomic separation, so the number of allowed modes
is one less that the number of atoms – just those given in (5.101). We can
now calculate the number of modes with wave number less than k. The
calculation is identical to (5.97) (without the factor of 2),
N = V
_
k
0
d
3
k
(2π)
3
=
V k
3
6π
2
. (5.102)
We know that the number of modes is just equal to the number of atoms
that are oscillating. If we call that number N also, (5.102) gives us a formula
for the maximum allowed k.
k
D
= (6π
2
N/V )
1/3
(5.103)
In the longwavelength limit, the energy and frequency are simply propor
tional.
ω = vk (5.104)
where v is the velocity of sound. This allows us to deﬁne a cutoﬀ frequency
7
ω
D
= (6π
2
v
3
N/V )
1/3
. (5.105)
It will be relevant to our discussion of superconductors that the Debye energy
¯hω
D
is usually several orders of magnitude smaller than the Fermi energy.
5.9 Superconductors
Some metals when cooled to a temperature close to absolute zero enter a dis
tinct thermodynamic state in which current ﬂows without resistance. Metals
in this state have many other remarkable properties as well. For example,
they are perfectly diamagnetic. Magnetic ﬁelds are excluded except for a
small penetration depth at the surface. This is called the Meissner eﬀect.
This rich phenomenology will be covered in the spring quarter of your solid
state physics class. Here I want to concentrate on the basic physics of the
superconducting state.
Our basic ideas about superconductors come from a classic paper by
Bardeen, Cooper, and Schrieﬀer.
8
They were awarded the Noble Prize for
7
Unlike (5.103), (5.105) is not an exact result. The point is that (5.104) holds in the
longwavelength limit, whereas the Debye cutoﬀ argument is based on shortwavelength
behavior. Nonetheless, these ideas are at the heart of several useful results for solids at
low temperature. See C. Kittel, Introduction to Solid State Physics.
8
J. Bardeen, L. N. Cooper, J. R. Schrieﬀer, Phys. Rev. 106, 162 (1957); 108, 1175
(1957)
5.9. SUPERCONDUCTORS 87
this, and the body of theory has come to be called “BCS theory.” Electrons
of course repel one another, and even with the screening eﬀect of the lattice,
the net coulomb force is still repulsive. They can interact in another way,
however, through the exchange of phonons. It was L. N. Cooper who ﬁrst
realized that this interaction produces a weak attractive force over a narrow
range of electron energies. This force is suﬃcient under some circumstances
to produce a weakly bound spinzero state of two electrons called a “Cooper
pair.” Since the energy of an electron in a pair is less than the Fermi energy,
the ground state becomes unstable, and electrons rush out of the Fermi sea
to make pairs. The pairs act as bosons, so they can “condense” much like
the helium atoms in a superﬂuid. The crucial diﬀerence is that their wave
functions overlap, so there is a strong coherence among the pairs. For this
reason they are able to spontaneously lock into phase with one another with
a macroscopic coherence length. The pairs move, not as individual particles,
but as a ﬂuid with a rigid phase. It’s not easy to scatter one of the pairs,
because it’s locked “in step” with all the others. It is this that accounts for
the lack of electrical resistance.
This is the barest outline of an enormously rich and elegant body of the
ory. I will concentrate on the central ideas and make numerous idealizations
and approximations along the way.
5.9.1 Eﬀective ElectronElectron Interaction
Phonons in this context are acoustic waves in crystals. They arise because
the crystal lattice can “ﬂex” slightly. The phonon waves are purely longitudi
nal, they have no polarization. In this respect they are a threedimensional
generalization of compression waves on a rubber band. Since we will be
dealing with phonons in the longwavelength approximation, the classical
analogy is apt. In both cases, the frequency and energy are proportional.
Phonons are massless particles. If we think of phonons as particles with
deﬁnite momentum, we must allow for them to scatter from electrons. The
Hamiltonian for this interaction is derived in every advanced text on solid
state physics.
H
=
k,q
D
q
c
†
k+q
c
k
(a
q
+a
†
−q
) (5.106)
Here c
k
is the annihilation operator for electrons, and a
q
is the correspond
ing operator for phonons. D
q
is the coupling “constant.” It actually has
some momentum dependence, which we will ignore. The derivation of this
constant involves some assumptions about the mechanical properties of the
lattice. This is really a topic for a solid state class, so I will just take (5.106)
88CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
as “given.” The signiﬁcance of the terms is clear enough. This ﬁrst term cor
responds to an electron of momentum k absorbing a phonon of momentum
q and going on its way with momentum q +k. The second term gives the
emission of a phonon of momentum q leaving an electron with momentum
k −q. This is the Hamiltonian we would guess, even if we had no theory
to ﬁnd D
q
. In addition to the interaction term, there is the freeparticle
Hamiltonian,
H
0
=
q
¯hω
q
a
†
q
a
q
+
k
k
c
†
k
c
k
. (5.107)
The exchange of a phonon between two electrons is at least a twostep
process, so we need to do something like secondorder perturbation theory to
combine two factors of (5.106) to get an eﬀective electronelectron Hamilto
nian. There is an elegant trick for doing this without any special formalism.
Suppose you have a Hamiltonian of the form
H = H
0
+λH
. (5.108)
The parameter λ is supposed to be “small.” We will eventually set it equal
to one. It’s really just a bookkeeping device. Our strategy is to ﬁnd a
transformed Hamiltonian
˜
H ≡ e
−S
He
S
= H + [H, S] +
1
2
[[H, S], S] + , (5.109)
such that all ﬁrstorder terms in λ are transformed away.
9
This can be
accomplished by
λH
+ [H
0
, S] = 0 (5.110)
Substitute (5.110) into (5.109) bearing in mind that S is itself ﬁrst order in
λ.
˜
H = H
0
+
1
2
[λH
, S] +O(λ
3
) (5.111)
That commutator will be our secondorder interaction Hamiltonian. We can
make an educated guess at the form of S (with λ = 1).
S =
k,q
(Aa
†
−q
+Ba
q
)D
q
c
†
k+q
c
k
(5.112)
9
The solid state community calls this the SchrieﬀerWolﬀ transformation.
5.9. SUPERCONDUCTORS 89
Here A and B are functions that we are about to determine. A long tedious
calculation gives
[H
0
, S] =
k,q
D
q
a
†
−q
c
†
k+q
c
k
(¯ hω
q
+
k+q
−
k
)A
+
k,q
D
q
a
q
c
†
k+q
c
k
(
k+q
−
k
−¯hω
q
)B (5.113)
I have assumed that ω
q
= ω
−q
. Now all this is supposed to equal −H
from
(5.106). It must be that
(
k
−
k+q
−¯hω
q
)A = 1
(
k
−
k+q
+ ¯hω
q
)B = 1 (5.114)
Use the notation ∆
±
(k, q) =
k
−
k+q
±¯hω
q
. Our formula for S is then.
S =
k,q
_
a
†
−q
∆
−
(k, q)
+
a
q
∆
+
(k, q)
_
D
q
c
†
k+q
c
k
(5.115)
Some comments about (5.115) are in order.
• It’s possible to show that S is antihermitian, i.e. S
†
= −S. This is
necessary so that (5.109) is a unitary transformation. In some sense
we have “diagonalized” the Hamiltonian. I’ll explain what I mean by
that below.
• The energy denominators are typical for nonrelativistic perturbation
theory. They vanish on the energy shell. For ∆
+
this corresponds to
the absorption of a phonon of momentum q, and ∆
−
to the emission
of a phonon of momentum −q.
• Our new secondorder interaction Hamiltonian from (5.111) is [H
, S]/2.
In order to calculate this we will need commutators like
[(a
†
−q
+a
q
)c
†
k+q
c
k
, a
†
−q
c
†
k
+q
c
k
] = δ
q,−q
c
†
k+q
c
k
c
k
+q
c
k
+ (5.116)
The in (5.116) refers to terms with one pair of boson operators and
one pair of electron operators. These are second order all right, but
they are the product of two singleparticle interactions and as such do
not interest us. Now you see the sense in which (5.109) diagonalizes
the Hamiltonian. Our new Hamiltonian connects twoelectron states
with twoelectron states, and states with one electron and one boson
with similar states with one boson and one electron.
90CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
It’s instructive to combine the two terms of (5.115) over a common denomi
nator. That gives the ﬁnal form for our eﬀective, secondorder, two electron
Hamiltonian as
1
2
[H
, S] =
1
2
k,k
,q
D
2
q
c
†
k+q
c
†
k
−q
c
k
c
k
¯hω
(
k
−
k
−q
)
2
−(¯hω)
2
(5.117)
The electron energies
k
, as I pointed out in Section 8.2, are much larger than
phonon frequencies, ω
q
, but there is a narrow window around [
k
−
k
−q
[ ≈
¯hω
D
in which the potential changes sign. Outside the window the potential is
repulsive (just like a positive Coulomb potential), but inside it is attractive.
This is the famous “energy gap” upon which all superconducting phenomena
depend. Electrons with just this range of energy form Cooper pairs.
At this point it is customary to make several approximations based partly
on mathematical convenience and partly on an understanding of the physics
of superconductors. First it is known that most of the phonons in a low
temperature solid have energies close to the Debye limit. Since the spread
in energies is so small, we do not sum over q in (5.117) but gather all the
numerical factors into one phenomonological (positive) coupling constant
V . Second, for a variety of reasons, electron pairs tend to have zero total
momentum and zero total spin. Thus if one electron of a pair has momentum
k and s
z
= 1/2, the other will have −k and s
z
= −1/2. We therefore
adopt the convention that a state written explicitly as k has spin ↑, and one
written as −k has spin ↓. The complete electron Hamiltonian with these
assumptions is
H
BCS
=
k
k
(c
†
k
c
k
+c
†
−k
c
−k
) −V
k,k
c
†
k
c
†
−k
c
−k
c
k
(5.118)
This is known as the BCS reduced Hamiltonian. It operates only on the
subspace consisting of the Cooper pairs that satisfy the above assumptions.
Notice that the interaction part of this Hamiltonian cannot be written
as a product of number operators. Put it another way, a state described
in terms of occupation numbers is not an eigenstate of H
BCS
. We have not
completely succeeded in our goal of diagonalizing it. You will recall that
this was the situation with the superﬂuid Hamiltonian. In fact, (5.118)
apart from the double sum is formally equivalent to to (5.63). It won’t be a
surprise then that the Bogoliubov transformation is the key to diagonalizing
it.
In 1956 Cooper
10
proved a profound result that was the prelude to the
full BCS theory. He showed that two electrons directly below the Fermi
10
L. N. Cooper, Phys. Rev. 104, 1189 (1956)
5.9. SUPERCONDUCTORS 91
surface can lower their energy by being excited into a Cooper pair with
momentum (k, −k) just above the Fermi surface, provided that an attrac
tive interaction of the form (5.118) exists. This is known as the Cooper
instability.
11
He also showed that a small center of mass momentum has
the eﬀect of much reducing the binding energy. This is one reason why
we only consider (k, −k) pairs. It is also the reason why superconductivity
only exists at low temperature. Now suppose we could start with a free
electron gas close to T = 0, so that the Fermi sphere is completely ﬁlled
up to a radius
F
. As we “turn on” the interaction (5.118), electrons in a
thin shell,
F
− ω
D
< <
F
+ ω
D
form Cooper pairs. The oneelectron
states in kspace in this shell are used in forming the BCS ground state, i.e.
the minimumenergy of the Hamiltonian (5.118). Part of our job will be to
construct this state explicitly. For the time being I will simply call it [Ω).
We’ll start by deriving equations of motion for the c
k
’s.
i ˙ c
k
=
k
c
k
−c
†
−k
V
k
c
−k
c
k
i ˙ c
†
−k
= −
k
c
†
−k
−c
k
V
k
c
†
k
c
†
−k
(5.119)
The key to solving the superﬂuid problem was to linearize the equations of
motion by replacing the product of operators a
†
k
a
k
with their expectation
value in the ground state, [Ω). Here the “particles” we are concerned with
are pairs, so the corresponding approximation is
V
k
c
−k
c
k
→V ¸Ω[
k
c
−k
c
k
[Ω) ≡ ∆
k
V
k
c
†
k
c
†
−k
→V ¸Ω[
k
c
†
k
c
†
−k
[Ω) ≡ ∆
∗
k
(5.120)
Substituting (5.120) back into (5.119) puts us back in the realm of familiar
mathematics. We look for solutions of the form a
k
(t) = a
k
(0)e
−iω
k
t
.
ω
k
c
k
(0) =
k
c
k
(0) −∆
k
c
†
−k
(0)
ω
k
c
†
−k
(0) = −
k
c
†
−k
(0) −∆
∗
k
c
k
(0) (5.121)
These equations are consistent if
ω
k
−
k
∆
k
∆
∗
k
ω
k
+
k
= 0 (5.122)
11
See, P. Philips, Advanced Solid State Physics for an accessible proof.
92CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
Or
ω
k
=
_
2
k
+ ∆
2
k
(5.123)
This is an important result. It means that all excited states are separated
from
k
by a ﬁnite amount, the gap parameter ∆.
The eigenvectors of (5.121) are in fact the Bogoliubov transformations
from Section 7.2. The analog to (5.66) is
β
k
= u
k
c
k
−v
k
c
†
−k
β
−k
= u
k
c
−k
−v
k
c
†
k
(5.124)
with inverses
c
k
= u
k
β
k
+v
k
β
†
−k
c
−k
= u
k
β
−k
−v
k
β
†
k
(5.125)
Both u
k
and v
k
are real numbers; u
k
is an even function of k, v
k
is odd. We
would like the β’s to satisfy fermion anticommutation relations.
¦β
k
, β
k
¦ = u
k
u
k
¦c
k
, c
k
¦ = v
k
v
k
¦c
†
−k
, c
−k
¦ = δ
kk
(u
2
k
+v
2
k
)
This is the ﬁrst condition on u
k
and v
k
.
u
2
k
+v
2
k
= 1 (5.126)
The natural parameterization is
u
k
= cos θ
k
v
k
= sinθ
k
(5.127)
Substituting (5.125) into the ﬁrst of (5.121) gives
ω
k
u
k
=
k
u
k
+ ∆u
k
v
k
. (5.128)
Square and substitute (5.123)
∆
2
(u
2
k
−v
2
k
) = 2
k
∆u
k
v
k
(5.129)
from which we conclude
tan 2θ
k
= ∆/
k
cos
2
θ
k
=
1
2
_
1 +
k
ω
k
_
sin
2
θ
k
=
1
2
_
1 −
k
ω
k
_
(5.130)
So far I have said nothing about the phase of ∆. There is an important
issue lurking here. The BCS Hamiltonian (5.118) is invariant under the
global gauge transformation c
k
→ c
k
= e
−iα
c
k
for all k. (It is clear that
if H
BCS
did not observe this symmetry, it would also not conserve fermion
5.9. SUPERCONDUCTORS 93
number.) It is speculated that the ground state observes this symmetry as
well so that any choice of the phase of ∆ is equally valid. We take ∆ real.
The magnitude of ∆ can be estimated as follows. Substitute (5.125) into
(5.120) and use the fermion anticommutation relations.
∆ = V ¸Ω[
k
(cos θ
k
β
−k
+ sin θ
k
β
†
k
)(cos θ
k
β
k
+ sin θ
k
β
†
−k
)[Ω)
= V ¸Ω[
k
cos θ
k
sinθ
k
β
−k
β
†
−k
[Ω)
= V
k
∆
2
_
2
k
+ ∆
2
¸
1/2
(5.131)
I have used (5.130) in the last step. This can be converted into an integral
equation,
1 =
V N
F
2
_
ω
D
−ω
D
d
[
2
+ ∆
2
]
1/2
= V N
F
sinh
−1
(ω
D
/∆) (5.132)
The limits of the integral are chosen in light of the fact that the Hamiltonian
only works on the states
k
where
F
−ω
D
<
k
<
F
+ω
k
. N
F
is the density
of states at the Fermi level. Eqn. (5.132) can be inverted to yield
∆ =
ω
D
sinh(1/V N
F
)
(5.133)
This was one of the original discoveries of the BCS collaborators. I under
stand that it is in reasonable agreement with experiment. Remember that
ordinary perturbation theory requires that the interaction be in some sense
“small.” Here the magnitude of the interaction is set by the quantity V N
F
.
In the limit that this is small, the denominator looks like the hyperbolic
sine of inﬁnity! This is not something you will ever get from a perturbation
theory.
Finally as promised, we construct the explicit form of the BCS ground
state wave function. I claim that apart from normalization
[Ω) =
k
β
−k
β
k
[0) (5.134)
Substituting (5.124) we get,
[Ω) =
k
(−v
k
)(u
k
+v
k
c
†
k
c
†
−k
) (5.135)
94CHAPTER 5. IDENTICAL PARTICLES ANDMANYPARTICLE STATES
It turns out that the right normalization is obtained by dropping the factors
of (−v
k
). This can be seen as follows:
¸Ω[Ω) = ¸0[
k
(u
k
+v
k
c
−k
c
k
)(u
k
+v
k
c
†
k
c
†
−k
[0))
=
k
(u
2
k
+v
2
k
)¸0[0) = 1 (5.136)
Thus the normalized ground state is
[Ω) =
k
(u
k
+v
k
c
†
k
c
†
−k
) (5.137)
Just to make sure, let’s verify that β[Ω) = 0.
β
k
[Ω) = β
k
β
−k
β
k
k
β
−k
β
k
[0) = 0
The prime on the product indicates that the k = k
term has been factored
out. The expression is zero because β
k
β
k
= 0 for a fermion operator.
Eqn. (5.137) represents a coherent superposition of correlated pairs, with
no restraint on the particle number.
Chapter 6
Second Quantization and
Relativistic Fields
In Chapter 5 I introduced the concept of the ﬁeld operators ψ(x) and ψ
†
(x)
(5.31). I explained that ψ
†
(x) is the operator that creates a particle at the
position x. In this context it seemed like a technical device for introducing
interactions in which particles are exchanged so that the number of particles
is not a constant of the motion. All this was done to “soup up” Schrodinger’s
equation. We didn’t worry too much about the physical interpretation.
There are other ways of regarding quantum ﬁeld operators, however, that
are important when we are exploring other wave equations, particularly
those arising in relativistic theories.
1
6.1 Introduction to Field Theory
Imagine that space is like a rubber sheet. If I put a bowling ball on the
sheet, it will create a depression, and nearby objects will roll into it. This
is an imperfect analogy for an attractive potential. We could describe the
attraction in one of two ways: we could say that there is an attractive
potential between any pair of pointlike masses, or we could introduce a
continuous variable, φ(x, y) which describes the displacement of the sheet
as a function of position. Such a continuous displacement variable is a ﬁeld in
the strict mathematical sense: it assigns a numerical value (or set of values)
to each point in space. The quantum mechanics of such ﬁelds is called
quantum ﬁeld theory. Now suppose that instead of using a bowling ball I
1
This material also appears at the beginning of Chapter 4. It is included here to make
the next few chapters selfcontained.
95
96CHAPTER 6. SECONDQUANTIZATIONANDRELATIVISTIC FIELDS
jump up and down on the sheet. The sheet will oscillate in response. My
activity becomes a source of energy, which propagates outward in the form
of waves. This is the rubbersheet analogy to the propagation of particles.
This analogy can easily be misleading. For one thing, I don’t want you
to think we are doing general relativity. The rubber sheet is not intended as
an analogy for ordinary spacetime as it is often used in explaining general
relativity. The ﬁeld φ(x, y) describes a displacement, and I know you want
to ask, “Displacement of what?”
The same question comes up in classical electromagnetic theory. When
an electromagnet wave is propagating through space, what is waving? Folks
in the 19’th century thought it must be some sort of mechanical medium,
which they called the ether. According to the textbooks, Michaelson and
Morley proved that wrong with their famous interferometer. But just saying
that the ether does’t exist doesn’t answer the question, it just makes it
impossible to answer! Let’s bite the bullet and agree for the purposes of this
course that space is pervaded by a medium, which for lack of a better name,
we will call the ether. Well, actually the ethers. Each species of particle
corresponds to a set of vibrations in it’s own speciﬁc ether. Electrons are
all vibrations in the electron ether, etc. Spacetime points in the ether can
be labelled with Lorentz fourvectors or (x, t) as usual, and these points
obey the usual rules for Lorentz transformations. This much is required
by the MM experiment. Ordinary bulk media have elastic properties that
are described by two parameters, the density and Young’s modulus. These
parameters are not themselves relevant to our formalism, but their ratio
gives the velocity of propagation, which is what we really care about.
I am fond of saying, “When correctly viewed, everything is a harmonic
oscillator.” Now you see that this is profoundly true. Each point on the
rubber sheet or ether acts like a harmonic oscillator! Quantum ﬁeld theory
is a theory about harmonic oscillators.
Well – I have to modify that slightly. If each point on the sheet behaved
like a simple harmonic oscillator with a quadratic potential, the waves prop
agating on the sheet would never interact. The principle of linear superposi
tion would hold everywhere. This is a theory of free particles. If our theory
is to describe interactions, then we must modify the potential so that it
becomes anharmonic. Unfortunately, the anharmonic oscillator cannot be
solve exactly in quantum mechanics. (If you think of a way to do it, tell
me and I’ll be famous.) We have to resort to approximations. The generic
name for these approximations is perturbation theory. The path integral
formalism in Chapters 14 comes equipped with its own natural way of do
ing perturbation theory, which you understand by now if you have plowed
6.2. INTRODUCTION TO CONTINUUM MECHANICS 97
through this material.
There is an alternative way of dealing with interaction involving the
creation and annihilation of particles. It is the older way, sometimes called
canonical quantization or second quantization. The path integral formalism,
seeks to banish all operators from the theory. Second quantization goes in
the other direction. It turns the wave functions themselves into operators
by imbedding creation and annihilation operators into them; but they are
the raising and lowering operators of the harmonic oscillator! The universe,
according to second quantization, is an inﬁnity of harmonic oscillators. This
approach is complementary to path integrals in other ways as well. We need
to master both.
Continuum mechanics is not covered in most graduate mechanics classes.
There is a good discussion in the last chapter of Goldstein, but we never
make it that far. What follows is a brief introduction.
6.2 Introduction to Continuum Mechanics
The rules of continuum mechanics are derived by starting with a system
with a ﬁnite number of degrees of freedom and then passing to the limit in
which the number becomes inﬁnite. Let’s do this with the simplest possible
system, a long chain of masses connected by springs. It’s a onedimensional
problem. The masses can only oscillate along the chain. We will use ϕ
i
,
the displacement of the ith particle from its equilibrium position, as the
generalized coordinate. The Lagrangian is constructed in the obvious way.
T =
1
2
i
m¨ ϕ
2
i
(6.1)
V =
1
2
i
k(ϕ
i+1
−ϕ
i
)
2
(6.2)
L = T −V =
1
2
i
a
_
m
a
˙ ϕ
2
i
−ka
_
ϕ
i+1
−ϕ
i
a
_
2
_
(6.3)
The equilibrium separation between masses is a. The spring constant is k.
The EulerLagrange equations of motion are obtained from
d
dt
∂L
∂ ˙ ϕ
i
−
∂L
∂ϕ
i
= 0 (6.4)
98CHAPTER 6. SECONDQUANTIZATIONANDRELATIVISTIC FIELDS
If there are N masses, then there are N coupled equation of this sort. They
look like
m
a
¨ ϕ
i
−ka
_
ϕ
i+1
−ϕ
i
a
2
_
+ka
_
ϕ
i
−ϕ
i−1
a
2
_
= 0 (6.5)
We need diﬀerent parameters to describe the continuum limit:
m/a →µ mass per unit length
ka →Y Young’s modulus
The index i points to the ith mass, and ϕ
i
gives its displacement. In the
continuum limit, the index is replaced by the coordinate x. In elementary
mechanics, x would be the displacement of a particle. Here ϕ(x) is the
displacement of the string at the point x. In the continuum limit
ϕ
i+1
−ϕ
i
a
→
ϕ(x +a) −ϕ(x)
a
→
dϕ
dx
L →
1
2
_
dx
_
µ ˙ ϕ
2
−Y
_
dϕ
dx
_
2
_
≡
_
dxL(ϕ, ˙ ϕ) (6.6)
The last integral implicitly deﬁnes the Lagrangian density . The continuum
version of the EulerLagrange equation is
2
d
dt
_
_
∂L
∂
_
dϕ
dt
_
_
_
+
d
dx
_
_
δL
∂
_
dϕ
dx
_
_
_
−
∂L
∂ϕ
= 0 (6.7)
Use the Lagrangian density from (6.6) in (6.7).
∂
2
ϕ
∂x
2
=
_
µ
Y
_
d
2
ϕ
dt
2
(6.8)
(6.4) and (6.5) represent a set of N coupled equations for N degrees of
freedom. (6.7) is one equation for an inﬁnite number of degrees of freedom.
In this sense, continuum mechanics is much easier that discrete mechanics.
Equation (6.8) should remind you of the equation for the propagation of
electromagnetic waves.
_
∂
2
ϕ
∂x
2
_
+
_
∂
2
ϕ
∂y
2
_
+
_
∂
2
ϕ
∂z
2
_
=
1
c
2
_
∂
2
ϕ
∂t
2
_
2
See Goldstein for a derivation of this important equation.
6.2. INTRODUCTION TO CONTINUUM MECHANICS 99
As you know, photons are massless particles. Notice that a string of massive
particles yields a wave equation that when quantized describes the propa
gation of massless particles. (With a diﬀerent velocity, of course.) This is
worth a brief digression.
What does it mean to say that a wave function describes the propagation
of a particle of a particular mass? The wave function ψ = e
i(kx−ωt)
might
describe a wave in classical E&M, or a massive particle in nonrelativistic
or relativistic quantum mechanics. The question is, what is the relation
between k and ω? The relationship between the two is called a dispersion
relation. It contains a great deal of information. In the case of EM waves
in vacuum, k = ω/c. Frequency and wave number are simply proportional.
This is the hallmark of a massless ﬁeld. The velocity is the constant of
proportionality, so there can only be one velocity. In Schrodinger theory
¯h
2
k
2
2m
= ¯hω (6.9)
The relationship is quadratic. The relativistic wave equation for a spinzero
particle is called the KleinGordon equation.
_
∇
2
−
1
c
2
∂
2
∂t
2
_
ϕ −
m
2
c
2
¯h
2
ϕ = 0 (6.10)
The dispersion relation is
(c¯hk)
2
+m
2
c
4
= (¯ hω)
2
, (6.11)
or in other words, p
2
c
2
+ m
2
c
4
= E
2
. All these equations can be obtained
from (6.7) with the appropriate Lagrangian density. They are all three
dimensional variations of our “waves on a rubber sheet” model. What does
this have to do with the particle’s mass? It’s useful to plot (6.9) and (6.11),
i.e. plot ω versus k for small values of k. In both cases the curves are
parabolas. This means that in the limit of small k, the group velocity,
v
group
=
dω
dk
≈
¯hk
m
(6.12)
In other words, the group velocity is equal to the classical velocity for a
massive particle v = p/m. All the wave equations I know of fall in one of
these two categories; either ω is proportional to k, in which case the particle
is massless and its velocity v = ω/k, or the relationship is quadratic, in
which case
m = lim
k→0
_
¯hk
dk
dω
_
. (6.13)
100CHAPTER 6. SECONDQUANTIZATIONANDRELATIVISTIC FIELDS
So long as we are talking about waveparticle duality, this is what mass
means.
One of the advantages of using Lagrangians rather than Hamiltonians
is that Lagrangians have simple transformation properties under Lorentz
transformations. To see this, let’s rewrite (6.7) in relativistic notation. Con
struct the contravariant and covariant fourvectors
x
µ
≡ (x
0
, x
1
, x
2
, x
3
) = (ct, x, y, z) (6.14)
x
µ
= (x
0
, x
1
, x
2
, x
3
) = (ct, −x, −y, −z) (6.15)
and the corresponding contravariant and covariant derivatives
∂
µ
≡
∂
∂x
µ
∂
µ
≡
∂
∂x
µ
. (6.16)
This puts the EulerLagrange equation in tidy form
∂
µ
_
∂L
∂(∂
µ
ϕ)
_
−
∂L
∂ϕ
= 0 (6.17)
This is slightly amazing. Equation (6.7) was derived without reference to
Lorentz transformations, and yet (6.17) has the correct form for a scalar
wave equation. We get relativity for free! If we can manage to make L a
Lorentz scalar, then (6.17)) will have the same form in all Lorentz frames.
Better yet, the action
S =
_
dt L =
_
dt
_
d
3
x L =
1
c
_
d
4
x L (6.18)
is also a Lorentz scalar. We can do relativistic quantum mechanics using
the canonical formalism of classical mechanics.
Here’s an example. Rewrite (6.6) in 3d
L =
1
2
_
µ
_
∂ϕ
∂t
_
2
−Y
_
_
∂ϕ
∂x
_
2
+
_
∂ϕ
∂y
_
2
+
_
∂ϕ
∂z
_
2
__
(6.19)
This would be the Lagrangian density for oscillations in a huge block of
rubber. Take
µ
Y
=
1
c
2
. (6.20)
Obviously L can be multiplied by any constant without changing the equa
tions of motion. Rescale it so that it becomes
L =
1
2
_
_
∂ϕ
∂t
_
2
−c
2
_
_
∂ϕ
∂x
_
2
+
_
∂ϕ
∂y
_
2
+
_
∂ϕ
∂z
_
2
__
(6.21)
6.3. INTRODUCTION TO SECOND QUANTIZATION 101
Substituting (6.21) into (6.17) yields the usual equation for EM waves, 2ϕ =
0.
Notice how the Lagrangian for oscillations a block of rubber (6.19) turns
into the Lagrangian for oscillations in the ether (6.21). We don’t have to
worry about the mechanical properties of the ether, because µ and Y are
scaled away. Despite what you may have been told, the MichelsonMorley
experiment proves the existence of the ether. When correctly viewed, ev
erything is a bunch of harmonic oscillators, even the vacuum!
Using Einstein’s neat notation, we can collapse (6.21) into one term
L =
1
2
(∂
µ
ϕ)(∂
µ
ϕ) ≡
1
2
(∂ϕ)
2
(6.22)
The last piece of notation (∂ϕ)
2
, is used to save ink. The fact that we can
write L like this is proof that it is a Lorentz scalar. This is an important
point; we can deduce the symmetry properties of a theory by glancing at L.
Now you can make up your own ﬁeld theories. All you have to do is
add scalar terms to (6.22). Try it. Name the theory after yourself. Here’s a
theory that already has a name. It’s the KleinGordon theory.
L =
1
2
_
(∂ϕ)
2
−m
2
φ
2
¸
(6.23)
(I have set c = 1 and ¯ h = 1.) Using our new notation, the equation of
motion is
(∂
µ
∂
µ
+m
2
)ϕ = 0 (6.24)
If we assume that ϕ(x) (x is a 4vector in this notation.) is a onecomponent
Lorentz scalar, then this describes a spinless particle with mass m propagat
ing without interactions. Spin can be included by adding more components
to ϕ. More about this later.
6.3 Introduction to Second Quantization
So far this has all been “classical ﬁeld theory,” ie. the ﬁelds are ordinary
functions, cnumbers as they are called, and not operators. We have not yet
introduced second quantization. In the previous chapter I introduced the
ﬁeld operator
ˆ
ψ(x) =
1
√
V
k
e
ik·x
ˆ a
k
(6.25)
I said that ˆ a
†
k
and ˆ a
k
were creation and annihilation operators and that all
this was necessary to treat systems in which the number of particles was
102CHAPTER 6. SECONDQUANTIZATIONANDRELATIVISTIC FIELDS
not constant. In this section I would like to examine the motivations behind
(6.25) more carefully and also investigate the many subtleties that arise
when we apply these ideas to relativistic wave equations. We will eventually
derive a completely relativistic generalization of (6.25), which will be our
starting point for doing relativistic ﬁeld theory.
We have encountered so far three quantum mechanical wave equations,
the Schrodinger equation, the KleinGordon equation, and the equation for
massless scalar particles, which is just the KG equation with m = 0. I will
write the freeparticle versions of them in a peculiar way to emphasize the
complementary roles of time and energy.
i
∂
∂t
ψ(x, t) =
ˆ
k
2
2m
ψ(x, t) (6.26)
_
i
∂
∂t
_
2
ψ(x, t) = (
ˆ
k
2
+m
2
)ψ(x, t) (6.27)
_
i
∂
∂t
_
2
ψ(x, t) =
ˆ
k
2
ψ(x, t) (6.28)
I have used
ˆ
k = −i∇
x
and ¯ h = c = 1. The operator on the right side of (6.26)
is the kinetic energy. Einstein’s equation E
2
= k
2
+ m
2
suggests that the
operators on the right side of (6.27) and (6.28) are the total energy squared.
Suppose that the ψ’s are eigenfunctions of these operators with eigenvalue
ω(k). (Each of these three equations will deﬁne a diﬀerent functional relation
between k and ω, of course.) The equations become
i
∂
∂t
ψ
ω
(x, t) = ω(k)ψ
ω
(x, t) (6.29)
_
i
∂
∂t
_
2
ψ
ω
(x, t) = ω
2
(k)ψ
ω
(x, t) (6.30)
_
i
∂
∂t
_
2
ψ
ω
(x, t) = ω
2
(k)ψ
ω
(x, t) (6.31)
Although we don’t usually use this language, we could think of i∂/∂t as
a kind of energy operator whose eigengvalues are the total energy of the
particle. Suppose now that the ψ
ω
’s are also momentum eigenstates so that
ˆ
kψ
ω
= kψ
ω
. The simplest solutions of (6.26) and (6.29) with ω = k
2
/2m
are
ψ
k
(x, t) =
1
√
V
e
i(±k·x−ωt)
(6.32)
6.3. INTRODUCTION TO SECOND QUANTIZATION 103
whereas the simplest solutions of (6.27) and (6.30) with ω
2
= k
2
+ m
2
or
(6.28) and (6.31) with ω
2
= k
2
are
ψ
k
(x, t) =
1
√
V
e
i(±k·x∓ωt)
(6.33)
(The 1/
√
V is a normalization factor put in for later convenience.) Eviden
tally the solutions to (6.30) and (6.31) comprise a larger family than those
of (6.29), and it is this larger family that I want to investigate.
To avoid ambiguity, I will assume that the symbol ω refers to a positive
quantity. Then
i
∂
∂t
e
∓iωt
= ±ωe
∓iωt
(6.34)
Since ¯ h = 1, ω has the units of energy. Schrodinger’s equation does not
have negative energy solutions. This is the clue that the upper sign in
(6.33) and (6.34) gives positiveenergy solutions and the lower sign gives
negativeenergy solutions whatever that may mean! What about the other
sign ambiguity? Think about the solution e
i(k·x−ωt)
. Pick out some point
on the wave where the phase of the exponential is φ. As time goes by, this
point will move so that k x −ωt = φ, or
k x = ωt +φ.
This point is moving in the general direction of k. We call this a positive
frequency solution. If ω and k x have opposite signs, the solution has
positive frequency (in the sense above). If the signs are the same, one gets
the negativefrequency solution.
Now take an arbitrary timeindependent wave function and expand it in a
Fourier series. Assume periodic boundary conditions so that k is discretized
as in (5.45).
ψ(x) =
1
√
V
k
e
ik·x
a
k
(6.35)
At this point a
k
is a Fourier coeﬃcient and nothing more. We can make ψ
time dependent by building the time dependence into the a
k
’s, a
k
→a
k
(t).
In order that (6.30) and (6.31) be satisﬁed, the a
k
’s should satisfy
¨ a
k
+ω
2
k
a
k
= 0 (6.36)
This is the diﬀerential equation for the harmonic oscillator, except for two
peculiar features. First, the a
k
’s are complex functions, and second, the
frequency (and hence the “spring constant”) is a function of k. In some
104CHAPTER 6. SECONDQUANTIZATIONANDRELATIVISTIC FIELDS
sense, each term in (6.35) has a harmonic oscillator associated with it. We
can tie into the usual harmonic oscillator formalism and avoid the complex
coordinates at the same time by deﬁning the real generalized coordinate,
q
k
(t) =
1
√
2ω
k
[a
k
(t) +a
∗
k
(t)] . (6.37)
The conjugate momentum is given by p(t) = ˙ q(t), but before we take the
derivative, we must decide on whether we are dealing with the positive or
negativeenergy solution. In order that each term in (6.35) has the form
(6.33), ˙ a
k
(t) = ∓iωa
k
. For the time being take positive energy (upper sign)
p
k
(t) = −i
_
ω
k
2
[a
k
(t) −a
∗
k
(t)] (6.38)
These are real variables oscillating with frequency ω. We know that the
Hamiltonian for simple harmonic motion is
H
k
=
1
2
_
p
2
k
+ω
2
k
q
2
k
¸
. (6.39)
You can verify with the deﬁnitions (6.37) and (6.38) that H
k
is time
independent, that p
k
is canonically conjugate to q
k
, and that Hamilton’s
equations of motion are satisﬁed. We can turn (6.39) into a quantum
mechanical Hamiltonian by simply making a
k
, a
∗
k
, q
k
and p
k
into opera
tors. We know from our work in Chapter 5 that ˆ a
k
must be an annihilation
operator with the commutation relations (5.15). The operators in (5.15),
however, are timeindependent, Schrodingerpicture operators as is the ﬁeld
operator (6.25). We will want to work in the Heisenberg representation, so
we must be careful about the time dependence. The natural assumption is
a
k
(t) →ˆ a
k
e
−iωt
a
∗
k
→ˆ a
†
k
e
+iωt
(6.40)
In (6.40) ˆ a
k
and ˆ a
†
k
are timeindependent, Schrodingerpicture operators.
I’ll argue presently that these is a consistent and reasonable assumption.
The commutation relations are then,
[ˆ a
k
, ˆ a
†
k
] = δ
k,k
[ˆ a
†
k
, ˆ a
†
k
] = [ˆ a
k
, ˆ a
k
] = 0 (6.41)
Since ˆ p
k
and ˆ q
k
don’t have this simple time dependence, the commutation
relations must be taken at equal times.
[ˆ q
k
(t), ˆ p
k
(t)] = iδ
k,k
[ˆ q
k
(t), ˆ q
k
(t)] = [ ˆ p
k
(t), ˆ p
k
(t)] = 0 (6.42)
6.3. INTRODUCTION TO SECOND QUANTIZATION 105
With this substitution (6.39) becomes
ˆ
H
k
=
1
2
ω
k
_
ˆ a
†
k
ˆ a
k
+ ˆ a
k
ˆ a
†
k
_
= ω
k
_
ˆ a
†
k
ˆ a
k
+
1
2
_
(6.43)
The same replacement turns (6.35) into (6.25).
The last factor of 1/2 in (6.43) presents something of a dilemma. This
ˆ
H
k
is just the Hamiltonian for a single k value. The complete Hamiltonian
is a sum over all values.
ˆ
H =
k
ˆ
H
k
(6.44)
An inﬁnite number of 1/2’s is still inﬁnity. It is customary to discard the
constant with some weak argument to the eﬀect that in deﬁning energy,
additive constants are meaningless. Since this problem will appear again
and again in diﬀerent contexts, it is useful to have some formal procedure
for sweeping it under the rug. To this end we introduce the concept of
“normal ordering.” We will say that an operator has been normal ordered
if all creation operators are placed to the left of all annihilation operators.
The usual symbol to indicate that an operator has been normal ordered is
to place it between colons, so for example,
:
ˆ
H
k
:= ω
k
ˆ a
†
k
ˆ a
k
(6.45)
To put it another way, (6.45) was obtained from (6.43) by commuting the ˆ a
k
past the ˆ a
†
k
in the second term and discarding the commutator. Whenever
we use a Hamiltonian in a practical calculation, we will assume that it has
been normal ordered.
We can check that this Hamiltonian is consistent with the time depen
dence assumed in (6.40) First note that [ˆ a
k
,
ˆ
H] = ω
k
ˆ a
k
, so
ˆ
Hˆ a
k
= ˆ a
k
(
ˆ
H −ω
k
) (6.46)
hence
ˆ
H
n
ˆ a
k
= ˆ a
k
(
ˆ
H −ω
k
)
n
(6.47)
as a consequence
ˆ a
k
(t) = e
i
ˆ
Ht
ˆ a
k
e
−i
ˆ
Ht
= ˆ a
k
e
−iω
k
t
ˆ a
†
k
(t) = e
i
ˆ
Ht
ˆ a
†
k
e
−i
ˆ
Ht
= ˆ a
†
k
e
iω
k
t
(6.48)
The deeper question is why this quantization procedure makes any sense
at all. The justiﬁcation lies in the canonical quantization procedure from
106CHAPTER 6. SECONDQUANTIZATIONANDRELATIVISTIC FIELDS
elementary quantum mechanics. It uses the Poisson bracket formalism of
classical mechanics and then replaces the Poisson brackets with commutator
brackets to get the corresponding quantum expression A Poisson bracket is
deﬁned as
¦F, G¦ ≡
N
k=1
_
∂F
∂q
k
∂G
∂p
k
−
∂F
∂p
k
∂G
∂q
k
_
(6.49)
where q
k
and p
k
are any pair of conjugate variables, and F and G are any
two arbitrary functions of q
k
and p
k
. The sum runs over the complete set
of generalized coordinates. Obviously
¦q
n
, p
m
¦ = δ
mn
¦q
n
, q
m
¦ = ¦p
n
, p
m
¦ = 0
(6.50)
This looks like the uncertainty relation in Quantum Mechanics, [x, p] =
i¯h. We get the quantum expression from the classical expression by the
replacement
¦F, G¦ →[
ˆ
F,
ˆ
G]/i¯h, (6.51)
where
ˆ
F and
ˆ
G are the quantum mechanical operators corresponding to the
classical quantities F and G, and [
ˆ
F,
ˆ
G] =
ˆ
F
ˆ
G−
ˆ
G
ˆ
F. In the case where
ˆ
F
and
ˆ
G depend on time, the commutator is taken at equal times. This seems
like a leap of faith, but it is valid for all the familiar operators in quantum
mechanics.
3
Now inverting (6.38) and (6.39) gives
a
k
=
ip
k
+ω
k
q
k
√
2ω
k
a
∗
k
=
−ip
k
+ω
k
q
k
√
2ω
k
(6.52)
Substituting (6.52) into (6.49) gives ¦a
k
, a
∗
k
¦ = −iδ
k,k
and ¦a
k
, a
k
¦ =
¦a
†
k
, a
∗
k
¦ = 0, so that
[ˆ a
k
, ˆ a
†
k
] = δ
k,k
[ˆ a
k
, ˆ a
k
] = [ˆ a
†
k
, ˆ a
†
k
] = 0
(6.53)
(with ¯ h = 1).
The value of the Poisson bracket ¦F, G¦ is independent of the choice
of canonical variables. That is a fundamental theorem. Since (6.37) and
(6.38) are together a canonical transformation, (6.42) and (6.53) are identi
cal. Any choice of variables will do so long as they are related to q
k
and p
k
3
Much of the formal structure of quantum mechanics appears as a close copy of the
Poisson bracket formulation of classical mechanics. See Goldstein, Poole and Safko, Clas
sical Mechanics Third Ed., Sec. 9.7
6.3. INTRODUCTION TO SECOND QUANTIZATION 107
by a canonical transformation. We simply chose q
k
so that it was real and
had convenient units. The rest followed automatically. The fact that the
resultant Hamiltonian is that of harmonic oscillators is simply a consequence
of the fact that we choose to expand ψ(x, t) in a Fourier series.
I can now write Equation (6.25) as
ˆ
ψ
(+)
(x, t) =
1
√
V
k
e
i(k·x−ω(k)t)
ˆ a
k
(6.54)
The superscript (+) means the positive energy solution. The functional
form of ω(k) is determined by the wave equation it represents, but I want
to concentrate on solutions of the KleinGordon equation. Suppose we had
chosen the negative energy solution.
ˆ
ψ
(−)
(x, t) =
1
√
V
k
e
i(k·x+ω(k)t)
ˆ c
k
(6.55)
(This automatically becomes a negative frequency solution as well.) The op
erators ˆ c
k
and ˆ c
†
k
annihilate and create these new negative energy particles.
Everything goes through as before except that ˆ p
k
(t) =
ˆ
˙ q
k
(t) = +iωˆ q
k
(t)
changes sign, so that (6.38) becomes
ˆ p
k
= i
_
ω(k)
2
_
ˆ c
k
− ˆ c
†
k
_
. (6.56)
The counterpart of (6.41) is
[ˆ c
†
k
, ˆ c
k
] = δ
k,k
(6.57)
It seems that the new creation operator c
†
k
stands in the place of the old
annihilation operator ˆ a
k
. This is not just a mathematical accident. It
points to an important piece of physics. To see this we deﬁne another pair
of creation and annihilation operators.
ˆ
d
k
= ˆ c
†
−k
ˆ
d
†
k
= ˆ c
−k
(6.58)
Substituting this in (6.55) and changing the sign of the summation variable
from k to −k gives
ˆ
ψ
(−)
(x, t) =
1
√
V
k
e
−i(k·x−ω(k)t)
ˆ
d
†
k
(6.59)
108CHAPTER 6. SECONDQUANTIZATIONANDRELATIVISTIC FIELDS
What is it that the
ˆ
d’s are creating and destroying, and what is the
signiﬁcance of the change in the sign of the momentum? In Chapter 5 we
considered the Fermi sea to be a set of lowlying energy levels all occupied by
fermions. Removing one particle from the sea leaves a “hole,” which behaves
in some ways like a real particle. If the hole moves in the positive k direction,
a real particle must move in the −k direction to backﬁll the hole. Dirac
proposed this mechanism to explain the negative energy states that appear
in relativistic electron theory. The correspondence between holes moving
forward and real particles moving backward is a good way of visualizing the
signiﬁcance of (6.58). Unfortunately, the KleinGordon equation describes
bosons, so there is no Fermi sea.
4
Nowadays, we regard these negative
energy solutions as representing real positiveenergy antiparticles. There
are two lines of evidence for this. For one thing, the states created by
ˆ
d
k
have momentum k (rather than −k). This can be proved formally, but it is
almost obvious from (6.59), which is a positivefrequency solution. Later on
when we discuss the interaction of the electromagnetic ﬁeld with bosons, we
will show that
ˆ
d
†
k
creates a particle of the opposite charge to that created
by ˆ a
†
k
. The complete operatorvalued wave function is the sum of (6.54) and
(6.59).
ˆ
ψ(x, t) =
1
√
V
k
_
e
i(k·x−ωt)
ˆ a
k
+e
−i(k·x−ωt)
ˆ
d
†
k
_
(6.60)
There are several profound reasons why the positive and negativeenergy
solutions must be added in just this way. These will emerge as we go along.
Let’s note in passing that there are several neutral spinzero particles
such as the π
0
that have no nonzero additive quantum numbers. Such
particles are thereby identical to their antiparticles. If ˆ a
†
k
creates a π
0
, then
ˆ
d
k
destroys the same particle. In this case there is no point in distinguishing
between ˆ a
k
and
ˆ
d
k
. We write (6.60)
ˆ
ψ(x, t) =
1
√
V
k
_
e
i(k·x−ωt)
a
k
+e
−i(k·x−ωt)
ˆ a
†
k
_
(6.61)
Fields corresponding to neutral particles are Hermitian. Those correspond
ing to charged particles are not.
In some ways (6.60) and (6.61) are relics of our nonrelativistic ﬁelds
from Chapter 5. Because they are based on discrete k values and periodic
boundary conditions they behave under Lorentz transformations in a most
4
The idea is only relevant to lowtemperature conductors anyway.
6.3. INTRODUCTION TO SECOND QUANTIZATION 109
awkward fashion. We are accustomed to passing to the continuum limit
through the replacement
1
V
k
→
_
d
3
k
(2π)
3
,
but this may be a bad idea for relativistic ﬁelds. The trouble is that the
integration measure d
3
k does not transform like a scalar under Lorentz trans
formations. A better choice might be
1
V
k
→
_
d
4
k
(2π)
3
δ(k
2
−m
2
), (6.62)
which is clearly invariant. (The symbol k here refers to the usual fourvector
k
µ
→(k
0
, k).) The dk
0
integration is done as follows.
_
dk
0
δ(k
2
−m
2
) =
_
d(k
0
)
2
_
dk
0
d(k
0
)
2
_
δ((k
0
)
2
−ω
2
k
)
=
_
d(k
0
)
2
2k
0
δ((k
0
)
2
−ω
2
k
) =
1
2ω
k
Equation (6.62) becomes
1
V
k
→
_
d
3
k
(2π)
3
2ω
k
(6.63)
Although this convention is sometimes used, it is somewhat simpler to use
the following
ˆ ϕ(x, t) =
_
d
3
k
_
(2π)
3
2ω
k
_
ˆ a(k)e
−ikx
+
ˆ
d
†
(k)e
ikx
_
(6.64)
where kx = ωt −k x. The point is that we must also consider the transfor
mation properties of the creation and annihilation operators. The natural
generalization of (6.41) is
[ˆ a(k), ˆ a
†
(k
)] = δ
(3)
(k −k
) [ˆ a(k), ˆ a(k
)] = [ˆ a
†
(k), ˆ a
†
(k
)] = 0 (6.65)
and and similarly for the
ˆ
d’s. Although δ
(3)
(k −k
) by itself is not a Lorentz
scalar, the ﬁeld deﬁnition (6.64) together with (6.65) does have the right
transformation properties. This will be apparent once we have calculated
the propagator.
110CHAPTER 6. SECONDQUANTIZATIONANDRELATIVISTIC FIELDS
6.4 Field Theory and the KleinGordon Equation
The KleinGordon equation and its associated Lagrangian were discussed
brieﬂy in Section 6.2. The Lagrangian density (6.54) is
5
L =
1
2
_
(∂ϕ)
2
−m
2
ϕ
2
¸
(6.66)
Suppose we regard ϕ as a generalized “coordinate” perhaps referring to the
displacement of some hypothetical continuous medium. The conjugate ﬁeld
is
π(x) =
δL
δ ˙ ϕ
= ∂
0
ϕ(x) (6.67)
We can use the Poisson bracket approach to quantize these ﬁelds just as we
quantized the a
k
’s in the previous section. Classically,
6
¦ϕ(x, t), π(x
, t)¦ = δ
(3)
(x −x
) (6.68)
Our quantized version of this is
[ ˆ ϕ(x, t), ˆ π(x
, t)] = iδ
(3)
(x −x
) (6.69)
It is easy to verify that (6.64) satisﬁes (6.69) so long as the creation oper
ators are normalized according to (6.65). It’s crucial that the two ﬁelds in
(6.69) are evaluated at the same time. As such they are called equal time
commutation relations. Many texts in fact consider (6.69) to be the funda
mental postulate of ﬁeld theory and use it to derive the properties of the
creation and annihilation operators.
Before we leave the subject of commutation relations, there is an issue
that has been hanging ﬁre since (6.60). In that equation the positive and
negativeenergy solutions appear with equal magnitude. Nothing I have
said so far requires the presence of the negativeenergy term, let alone that
it have the same normalization as the positiveenergy part. In fact, we have
no choice in the matter. The argument goes like this. Consider two space
like separated points (x, t) and (x
, t
). There will always be a Lorentz frame
such that t = t
. If x and x
are distinct points, then a signal can propagate
from one to the other only by travelling at inﬁnite velocity. We believe this
to be impossible, so our theory must not allow it, even in principle. We
5
I am using ϕ in this section rather than ψ to refer speciﬁcally to solutions of the
KleinGordon equation.
6
See J.V. Jose and E. J. Saletan, Classical dynamics: a contemporary approach, Sec
9.3.1 for a derivation of this rather nontrivial result.
6.5. THE PROPAGATOR 111
call this the requirement of causality. It can be argued that a necessary
requirement is that the ﬁelds ˆ ϕ(x) and ˆ ϕ
†
(x
) also commute at equal times.
7
Let’s rewrite (6.64) with an additional parameter α that we can tweak at
our pleasure.
ˆ ϕ(x, t) =
_
d
3
k
_
(2π)
3
2ω
k
_
ˆ a(k)e
−ikx
+αˆ a
†
(k)e
ikx
_
(6.70)
A simple calculation now gives
[ ˆ ϕ(x, t), ˆ ϕ
†
(x
, t)] =
_
d
3
k
2(2π)
3
ω
k
(1 −[α[
2
)e
ik·(x−x
)
(6.71)
This is not zero (because of the ω in the denominator) unless [α[ = 1.
Relativity and causality together require an equal admixture of negative
and positiveenergy states. This argument takes on additional signiﬁcance
when spin is taken into account. It can be shown that the requirement
proved above only holds for integer spin. In the case of halfodd integer
spin, the result only holds if the corresponding creation and annihilation
operators anticommute.
6.5 The Propagator
At this point I must anticipate some developments from the next few chap
ters. It will turn out that one of the key ingredients of any perturbation
calculation is the Feynman propagator deﬁned by
G(x, y) = iD(x −y) = ¸0[T[
ˆ
φ(x),
ˆ
φ(y)][0) (6.72)
Where T[ , ] is the “timeordered product” deﬁned by
T[ ˆ ϕ(x) ˆ ϕ(y)] = θ(x
0
−y
0
) ˆ ϕ(x) ˆ ϕ(y) +θ(y
0
−x
0
) ˆ ϕ(y) ˆ ϕ(x) (6.73)
In a timeordered product the timedependent operators are ordered so that
later times stand to the left of earlier times. Timeordered products read like
Hebrew, right to left. There are several ways of interpreting D(x−y). From
a mathematical point of view, it is the Green’s function of the KleinGordon
equation, i.e.
(∂
µ
∂
µ
−m
2
)D(x −y) = δ
(4)
(x −y) (6.74)
7
See Paul Teller, An Interpretive Introduction to Quantum Field Theory, Chapter 4,
for a careful discussion of this point.
112CHAPTER 6. SECONDQUANTIZATIONANDRELATIVISTIC FIELDS
From a physical point of view, it is the probability amplitude for a particle
to propagate from y to x. I need to prove the central result that
iD(x −y) = i
_
d
4
k
(2π)
4
e
−ik(x−y)
k
2
−m
2
+i
(6.75)
Each kind of particle has its own propagator, but the +i term in the denom
inator is ubiquitous. The stands for an inﬁnitesimal positive quantity. It’s
job is to get the boundary conditions right as you will see in the derivation.
We know that this will be a function of x−y, so we can make the algebra
a bit simpler by setting y = 0. Just remember to replace x → x − y at the
end of the calculation. Substituting the ﬁelds from (6.70) into (6.72) and
taking the vacuum expectation value gives
iD(x) = ¸0[T[ ˆ ϕ(x, t) ˆ ϕ(0, 0)][0)
=
_
d
3
k
(2π)
3
2ω
k
_
θ(t)e
−i(ω
k
t−k·x)
+θ(−t)e
i(ω
k
t−k·x)
_
(6.76)
Equations (6.75) and (6.76) are really the same result, though this is far
from obvious. In principle, we could derive either form from the other, but
it’s probably easier to start from (6.75).
iD(x) = i
_
d
3
k
(2π)
4
e
ik·x
_
dk
0
e
−ik
0
t
(k
0
−ω
k
+i)(k
0
+ω
k
−i)
(6.77)
Notice how the denominator is factored. Multiplying the two factors and
making the replacements, 2iω
k
→i and
2
→0, gives the same denomina
tor as (6.75). The dk
0
integration is now performed as a contour integration
in the complex k
0
plane. For t < 0 the contour is completed in the upper
halfpane enclosing the point k
0
= −ω
k
+ i, and for t > 0 the contour is
completed in the lower halfplane enclosing the point k
0
= ω−i. The result
is identical to (6.76). You see how the i in the denominator displaces the
poles so as to pick up the right integrand depending on whether t is positive
or negative. Notice ﬁnally that (6.73) is a Lorentz scalar since kx, k
2
and
d
4
k are all scalar quantities. You will see how D(x−y) becomes a key player
in perturbation theory via the interaction picture in the next chapter.
6.5. THE PROPAGATOR 113
Im k
0
Re k
0
t<0
t>0
k
0
=ω
0
iε
k
0
=ω
0
+iε
Figure 6.1: The complex k
0
plane
114CHAPTER 6. SECONDQUANTIZATIONANDRELATIVISTIC FIELDS
Chapter 7
The Interaction Picture and
the SMatrix
Most of what we know about subatomic physics comes from two kinds of
experiments: decay experiments and scattering experiments. In a decay
experiment, one starts with some system such as an atom or nucleus or
“elementary” particle and observes the spontaneous transitions that it un
dergoes. One can determine the lifetime of the system, the identity of the
decay products, the relative frequency of the various decay modes, and the
distribution of momentum and energy among the resulting particles. In a
scattering experiment, one starts with a stable system and bombards it with
another particle. One measures the distribution of momenta among the var
ious particles produced by the reaction and determines the probability that
the scattering will lead to a particular ﬁnal state. One common feature of
both these experiments is that the particles are detected when they are not
interacting with one another. They are detected on a scale of distance that
is many orders of magnitude larger than the characteristic de Broglie wave
length and at times that are vastly longer than the time of interaction. In
nonrelativistic quantum mechanics these processes are calculated with ﬁrst
order, time dependent perturbation theory using a bit of hocus pocus called
“Fermi’s golden rule.” This is not altogether wrong, but it is inadequate
for several reasons: it can’t accommodate the creation of new particles, it’s
hopelessly nonrelativistic, and it only works to ﬁrst order.
Real scattering theory is diﬃcult. There are many subtle issues involved.
Much of the material in advanced quantum books relates to scattering in one
way or another. I say this because it’s easy to lose sight of the goal amidst all
the technical diﬃculties. Roughly speaking, there are two basic issues: how
115
116CHAPTER 7. THE INTERACTIONPICTURE ANDTHE SMATRIX
do the quantum ﬁelds ϕ(x) evolve in time, and given this information, how
can we calculate the results of these experiments in terms of the momenta
of particles measured in the asymptotic regime as explained above? The
ﬁrst question is answered neatly by the interaction picture formalism ﬁrst
developed by Freeman Dyson. The second question is much more diﬃcult
and will require the remainder of this chapter.
7.1 The Interaction Picture
Path integrals use the Lagrangian; the interaction picture uses the Hamilto
nian; so I should say a few words about Hamiltonians in general. In classical
mechanics, the two are related by
H = p ˙ q −L (7.1)
In our variety of continuum mechanics the Hamiltonian density becomes
H = π ˙ ϕ −L (7.2)
For the free KleinGordon ﬁeld
L =
1
2
_
(∂
µ
ϕ)(∂
µ
ϕ) −m
2
ϕ
2
¸
=
1
2
_
˙ ϕ
2
−(∇ϕ)
2
−m
2
ϕ
2
¸
(7.3)
H =
1
2
_
˙ ϕ
2
+ (∇ϕ)
2
+m
2
ϕ
2
¸
(7.4)
We get the Hamiltonian by integrating (7.4) over all space.
H =
_
d
3
xH (7.5)
We assume that the Hamiltonian can be split up into two pieces: H =
H
0
+H
int
. H
0
is the Hamiltonian of the free ﬁeld, and H
int
is everything left
over. We assume the H
int
is “small” in some sense; so that the perturbation
series converges. We also have to assume that H
int
is a polynomial of the
ﬁelds and their derivatives. The reason for this technical assumption will
appear later.
You will recall the relation between the Schrodinger and Heisenberg pic
tures is
[ψ(t))
S
= e
−iHt
[ψ)
H
(7.6)
Q
H
(t) = e
iHt
Q
S
e
−iHt
7.1. THE INTERACTION PICTURE 117
This is the usual convention at any rate. [ψ)
H
, which does not depend on
time, is a “snapshot” of [ψ(t))
S
taken at t = 0. This is not necessarily the
most convenient time to take the picture as we will see.
We deﬁne the interaction state by
[ψ(t))
I
≡ e
iH
0
t
[ψ(t))
S
(7.7)
Q
I
(t) = e
iH
0
t
Q
S
e
−iH
0
t
Some comments:
• It’s clear from (7.6) the Hamiltonian H is the same in the Schrodinger
and Heisenberg pictures, and equally clear from (7.7) that H
0
is the
same in the interaction picture as well. This is not true of the inter
action piece; since in general, [H
int
, H
0
] ,= 0. I will write
H
I
int
(t) ≡ e
iH
0
t
H
S
int
e
−iH
0
t
(7.8)
Naturally, H
S
int
is the interaction Hamiltonian in the Schrodinger pic
ture. Notice that H
I
int
(t) depends on time.
• If there were no interactions, [ψ(t))
I
= [ψ)
H
, so the interaction pic
ture state would be stationary. Now think of a scattering experi
ment. When the particles are far apart, there is no interaction between
them. In this regime, the interaction picture and Heisenberg pictures
are identical. As the particles begin to interact, H
int
“turns on” and
[ψ(t))
I
begins to change. Eventually, after the particles are separated
it settles down to a new Heisenberg state.
Actually, this business of “turning on” and “ turning oﬀ” skirts a
profound and diﬃcult problem. The trouble is that H
int
never really
turns oﬀ. Even when the particles are not interacting with one another,
they are still interacting with themselves by emitting and reabsorbing
virtual particles. Worse yet, even the vacuum interacts with itself by
creating particleantiparticle pairs. This is a diﬃcult problem, but one
that can be dealt with rigorously.
Deﬁne the time evolution operator in the interaction picture, U(t, t
0
), by
[ψ(t))
I
= U(t, t
0
)[ψ(t
0
))
I
. (7.9)
Since H is Hermitian and the norm of [ψ)
I
is conserved, we feel entitled by
the rules of quantum mechanics to the following basic relations:
U
†
(t, t
0
)U(t, t
0
) = 1 (7.10)
118CHAPTER 7. THE INTERACTIONPICTURE ANDTHE SMATRIX
U(t, t) = 1 (7.11)
U(t, t
0
) = U
−1
(t
0
, t) (7.12)
U(t
2
, t
0
) = U(t
2
, t
1
)U(t
1
, t
0
) (7.13)
We can derive an equation of motion for U by starting with Schrodinger’s
equation
i
∂
∂t
[ψ(t))
S
= H[ψ(t))
S
. (7.14)
A quick calculation with (7.7) and (7.9) yields
i
∂
∂t
U(t, t
0
) = H
I
int
(t)U(t, t
0
). (7.15)
We know how to solve equations like this.
U(t, t
0
) = 1 −i
_
t
t
0
dt
H
I
int
(t
)U(t
, t
0
) (7.16)
Well – maybe solve is too strong a word, since U appears on both sides of
(7.16). We would like to claim that
U(t, t
0
) = (?) exp
_
−i
_
t
t
0
dt
H
I
int
(t
)
_
This would be the obvious solution if U and H were not operators. The ﬂaw
in the reasoning here is that factors of H
int
don’t commute at diﬀerent times,
[H
int
(t), H
int
(t
)] ,= 0. We can come up with a valid solution by iterating
(7.16) paying careful attention to the time dependence.
U(t, t
0
) = 1 −i
_
t
t
0
dt
1
H
I
int
(t
1
)
+(−i)
2
_
t
t
0
dt
1
_
t
1
t
0
dt
2
H
I
int
(t
1
)H
I
int
(t
2
) +
(7.17)
The entire series can be summarized by
U(t, t
0
) =
∞
n=0
(−i)
n
_
t
t
0
dt
1
_
t
1
t
0
dt
2
_
t
n−1
t
0
dt
n
H
I
int
(t
1
)H
I
int
(t
2
) H
I
int
(t
n
)
(7.18)
The series (7.18) is more diﬃcult than it looks. Since the Hamiltonians
don’t commute, we must be meticulously careful to keep later times to
7.2. THE S MATRIX 119
the left of earlier times. This is called time ordering. We need some
machinery to do this for us. Deﬁne the time ordering operator,
T[H(t
1
)H(t
2
)] ≡ H(t
1
)H(t
2
)θ(t
1
−t
2
) +H(t
2
)H(t
1
)θ(t
2
−t
1
) (7.19)
The generalization to three or more Hamiltonians is obvious. You should
convince yourself that the following things are true:
_
t
t
0
dt
1
_
t
1
t
0
dt
2
H(t
1
)H(t
2
) =
_
t
t
0
dt
2
_
t
2
t
0
dt
1
H(t
2
)H(t
1
)
=
1
2
_
t
t
0
_
t
t
0
dt
1
dt
2
T[H(t
1
)H(t
2
)],
and in general
_
t
t
0
dt
1
_
t
n−1
t
0
dt
n
H(t
1
) H(t
n
) =
1
n!
_
t
t
0
dt
1
_
t
t
0
dt
n
T[H(t
1
) H(t
n
)]
So our ﬁnal result for the U operator is
U(t, t
0
) =
∞
n=0
(−i)
n
n!
_
t
t
0
_
t
t
0
dt
1
dt
2
dt
n
T[H
I
int
(t
1
)H
I
int
(t
2
) H
I
int
(t
n
)],
(7.20)
which can be written in shorthand notation as
U(t, t
0
) = T exp
_
−i
_
t
t
0
dt
H
I
int
(t
)
_
(7.21)
The reason that this is a workable approximation is that the ﬁelds that make
up H
I
int
are interactionpicture ﬁelds, which according to (7.8), transform as
free ﬁelds. Equation (7.20) is a “recipe.” You just insert the formulas for
the free ﬁelds and do the algebra.
7.2 The S Matrix
We can imagine quantum scattering as taking place in three phases. In the
limit t → −∞ the particles are separated and not interacting with one an
other. As the particles approach one another, the interaction between them
“turns on” as they exchange virtual particles. At some later time t →+∞,
this interaction turns oﬀ and they again become free particles. This is diﬃ
cult to treat theoretically, since the interaction Hamiltonian itself in no way
120CHAPTER 7. THE INTERACTIONPICTURE ANDTHE SMATRIX
turns on or oﬀ. So far as theoretical diﬃculties are concerned, this is only
the tip of the iceberg. As soon as we allow particles to interact with other
particles, they inevitably interact with themselves. When this happens, the
notions of free particles, charge, and vacuum become profoundly subtle.
From the point of view of theory, a free particle is a boiling stream of
selfinteractions. From the point of view of experiment, every electron in
the universe is an unremarkable little thing with a welldeﬁned mass, spin,
charge, and magnetic moment. These two views can exist simultaneously
because the selfinteractions are virtual processes, and as such they are con
tained within a region roughly the size of the Compton wavelength. So long
as one does not probe inside this region, the electron has it’s usual phys
ical properties, and probing inside the region means doing a highenergy
scattering experiment.
The same sort of ambiguities also complicate our understanding of the
vacuum state. According to perturbation theory, virtual particleantiparticle
pairs are spontaneously emitted from and reabsorbed into empty space. This
fact is established experimentally (to some extent) by an experiment measur
ing the socalled Casimir eﬀect,
1
but apart from this, these vacuum ﬂuctua
tions have very little consequence, and at any rate, are not well understood.
2
Finally consider the charge of an electron (or more generally, the cou
pling constant for any sort of interaction). The charge is the measure of
the strength with which photons couple to electrons, but this can only be
measured by doing a scattering experiment, and the experiment must be
interpreted in light of the virtual processes involved.
We believe that all these problems can be addressed, at least for quantum
electrodynamics, but this is a long story, which will occupy us for the next
few chapters. For the time being, let us say that the theory recognizes two
kinds of mass and two kinds of charge. There are the physical or “dressed”
masses and coupling constants – those that are measured in the lab – and
the “bare” masses and coupling constants, which are the parameters that go
into the interaction Lagrangians and Hamiltonians. Somehow the eﬀect of all
virtual processes formulated in terms of the bare particles is to produce the
physical particles. What then are the numerical values of these bare masses
and charges? Alas, we do not know. There is no way to measure them
directly. They could be calculated in principle, but every such calculation
yields inﬁnity! Despite this, we are able to calculate everything that can be
1
Two charged plates suspended parallel to one another in vacuum experience a very
small repulsive force because of virtual electronpositron pairs created between them.
2
One can calculate the gravitational mass of all the virtual particleantiparticle pairs
in the universe. The result come out too large by many orders of magnitude.
7.2. THE S MATRIX 121
measured (except mass and charge, of course) using only the physical mass
and charge. The fact that we have learned to skirt inﬁnities so violent as to
give sciﬁ fans nightmares, is one of the ironic triumphs of modern physics.
In view of this, the interaction picture formulation in the previous section
is a bit too naive. What masses are we to put in H
int
, the bare or the physical
mass? It seems that either way we are stuck. We need to back up and look
at the basic description of scattering processes.
Consider the state of a scattering experiment before the particles have
interacted. For mathematical purposes we say this corresponds to a time
“t → −∞” (The actual time for a relativistic particle could be as small
as t → −10
−23
seconds.) At this time the particles are free in the sense
explained above. It is customary to call such states “in states,” and write
them [α, in). The symbol α stands for all the quantum numbers required
to completely specify the state. We will be particularly concerned with
the momentum. If there are n particles in the initial state, we can write
[k
1
, k
2
, . . . , k
n
, in). The ﬁeld operators introduced in the previous chapter
can be written
ϕ
in
(x) =
_
d
3
x
_
2E
k
(2π)
3
_
a
in
(k)e
−ikx
+a
†
in
(k)e
ikx
_
(7.22)
It is understood that in calculating with this expression, one uses the phys
ical mass of the particle. In the same way, one deﬁnes “out states” ¸β, out[,
“out ﬁelds” ϕ
out
(x), and “out” creation operators a
†
out
(k) corresponding to
the limit t → ∞. A scattering experiment can be idealized as follows. The
initial target and beam particles are prepared so that they are in a unique
eigenstate of all the relevant operators. We call this state [α, in). The detec
tors of the scattered particles are so reﬁned that they can specify that the
ﬁnal state was in the exact eigenstate ¸β, out[. The object of the experiment
is to ﬁnd what fraction of particles in [α, in) make the transition to ¸β, out[.
In other words, we are measuring the probability
P
βα
= [S
βα
[
2
(7.23)
where the socalled S matrix is deﬁned by
S
βα
= ¸β, out[α, in) (7.24)
If we assume that the in states and out states are both complete sets, there
must be some operator S that transforms in states into out states.
¸β, out[ = ¸β, in[S (7.25)
122CHAPTER 7. THE INTERACTIONPICTURE ANDTHE SMATRIX
S
βα
= ¸β, in[S[α, in) (7.26)
The goal of scattering theory is to calculate the S matrix and the corre
sponding S operator.
Here are some commonsense properties of S.
1. Nothing in – nothing out, i.e. [S
00
[
2
= 1. Put it another way
¸0, in[S = ¸0, out[ = e
iθ
¸0, in[ (7.27)
The vacuum is stable, but we must allow the possibility that phase of
the vacuum state evolves with time.
2. Oneparticle states are also stable.
¸p, out[p, in) = ¸p, in[S[p, in) = ¸p, in[p, in) = 1 (7.28)
3.
ϕ
in
= Sϕ
out
S
−1
(7.29)
Proof:
¸β, out[ϕ
out
[α, in) = ¸β, in[Sϕ
out
[α, in)
But ¸β, in[ϕ
out
is itself an in state, so we can write
¸β, in[ϕ
out
[α, in) = ¸β, out[ϕ
in
S[α, in)
Comparing these two equations gives
Sϕ
out
= ϕ
in
S
4. S is unitary.
Proof:
¸β, in[S = ¸β, out[ S
†
[α, in) = [α, out)
¸β, in[SS
†
[α, in) = ¸α, out[β, out) = S
βα
7.3 The LSZ Reduction Scheme
What is the relation between the fullyinteracting ﬁeld ϕ and ϕ
in
? The
natural guess is
lim
t→−∞
ϕ(x) = ϕ
in
(x)
7.3. THE LSZ REDUCTION SCHEME 123
lim
t→∞
ϕ(x) = ϕ
out
(x)
On second thought though, this can’t be right. The complete ﬁeld ϕ(x)
is capable of creating states out of the vacuum with two or more virtual
particles. A more careful analysis suggests
lim
t→−∞
ϕ(x) =
√
Z ϕ
in
lim
t→∞
ϕ(x) =
√
Z ϕ
out
(7.30)
Z is understood as the probability for ϕ(x) to produce a singleparticle state
out of the vacuum. Even this is not quite correct as an operator equation for
some subtle mathematical reasons that are discussed in the original article.
3
It is correct in terms of matrix elements, for example
lim
t→−∞
¸α[ϕ(x)[β) =
√
Z¸α[ϕ
in
(x)[β)
where ¸α[ and [β) are arbitrary states. Since we are only interested in
matrix elements in the long run, we will assume (7.30) without any further
comment. Furthermore, the processes in which ϕ(x) produces multiparticle
states out of the vacuum only occur in higher orders of perturbation theory.
When we are working to lowest nontrivial order we can (and will) set Z = 1.
Now we have the apparatus at hand for deﬁning and studying the S
matrix deﬁned in (7.24). We start from an initial state of a system with n
noninteracting physical particles, denoted by
[k
1
k
n
in) = [α in), (7.31)
and a ﬁnal state in which m particles emerge denoted by
¸p
1
p
m
out[ = ¸β out[ (7.32)
Our agenda is to convert (7.24) into an expression involving vacuum expec
tation values of interacting ﬁelds. This is the LSZ reduction technique.
Let us write the inﬁeld in a way analogous to (6.64)
ϕ
in
(x) =
_
d
3
k
_
f
k
(x)a
in
(k) +f
∗
k
(x)a
†
in
(k)
_
(7.33)
where
f
k
(x) =
e
−ikx
_
(2π)
3
2ω
k
(7.34)
3
See also Quantum Theory of Point Particles and Strings, Brian Hatﬁeld, Chapter 7.
124CHAPTER 7. THE INTERACTIONPICTURE ANDTHE SMATRIX
Equation (7.33) can be inverted to yield
a
in
(k) = i
_
d
3
x f
∗
k
(x)
↔
∂
0
ϕ
in
(x)
a
†
in
(k) = −i
_
d
3
x f
k
(x)
↔
∂
0
ϕ
in
(x) (7.35)
The notation a
↔
∂
0
b means a∂
0
b − (∂
0
a)b. Since the instate in (??) [α in)
is made up with a product of creation operators as in (??), we can write
S
βα
= ¸β out[a
†
in
(k)[α −k in) (7.36)
where [α −k in) represents the initial assemblage of particles represented
by α with one particle of momentum k removed. The procedure is now to
use (7.35) to replace a
†
in
(k) with an integral over ϕ
in
(x). The procedure is
then repeated until all the particles have been removed from the in and
out states leaving the vacuum expectation value of the ﬁelds.
S
βα
= ¸β out[a
†
out
(k)[α −k in) (7.37)
+¸β out[
_
a
†
in
(k) −a
†
out
(k)
_
[α −k in)
= ¸β−k out[α−k in)−i¸β out[
__
d
3
x f
k
(x)
↔
∂
0
(ϕ
in
(x) −ϕ
out
(x))
_
[α −k in)
I have added and subtracted a
†
out
(k) and used (7.35) to eliminate a
†
in
and
a
†
out
. The ﬁrst term in the last line vanishes unless the initial and ﬁnal
states are identical. This is the unscattered wave, which we will neglect for
the time being. Now use (7.30) to replace the in and out ﬁelds with the
complete interacting ﬁeld. The scattered part can be written as follows:
S
βα
∼
i
√
Z
_
lim
x
0
→∞
− lim
x
0
→−∞
__
d
3
x f
k
(x)
↔
∂
0
¸β out[ϕ(x)[α −k in) (7.38)
(The symbol ∼ means that I have temporarily neglected the forward scat
tering term.) The limits have a nasty noncovariant look about them. I will
clean up the expression with a series of elegant mathematical tricks. For
any two functions such as f
k
(x) and ϕ(x),
_
lim
x
0
→∞
− lim
x
0
→−∞
__
d
3
x f
k
↔
∂
0
ϕ =
_
∞
−∞
d
4
x ∂
0
_
f
k
↔
∂
0
ϕ
_
7.3. THE LSZ REDUCTION SCHEME 125
=
_
∞
−∞
d
4
x
_
f
k
∂
2
∂x
2
0
ϕ −ϕ
∂
2
∂x
2
0
f
k
_
(7.39)
The fact that f
k
satisﬁes the KleinGordon equation allows us to write
ϕ
∂
2
∂x
2
0
f = ϕ(∇
2
−m
2
)f →f(∇
2
−m
2
)ϕ (7.40)
The sense of the right arrow is that the subsequent expression is obtained
by integrating by parts twice and discarding the surface terms. Finally
substituting (7.39) and (7.40) into (7.38) gives the ﬁnal result
S
βα
= ¸β out[in α) = ¸β −k out[α −k in) (7.41)
+
i
√
Z
_
d
4
x f
k
(x)(2+m
2
)¸β out[ϕ(x)[α −k in)
You see the pattern? We have “reduced” an asymptotic particle from the in
state and replaced it with an interacting ﬁeld ϕ(x). The price we pay for this
is an integral, a wave function f
k
(x), and a diﬀerential operator (2 + m
2
).
We will eventually Fourier transform this expression, whereupon all these
complications will disappear leaving only the inverse of the momentum space
propagator ∆(k).
As an exercise (I am getting tired of typing) you should take the next
step and reduce a particle of momentum p out of the outstate. Convince
yourself that the result is
S
βα
= ¸β −p out[α −k in) (7.42)
∼
_
i
√
z
_
2
_
d
4
x
_
d
4
y f
k
(x)f
†
p
(y)(2
x
+m
2
)(2
y
+m
2
)
¸β −p, out[T[ϕ(x)ϕ(y)][α −k, in)
The ∼ sign again means that we have dropped the forward scattering terms.
The new thing here is the time ordered product. You should do the calcu
lation carefully to see how this comes about. Now suppose there are m
particles in the initial state and n particles in the ﬁnal state. Just repeat
this procedure n +m times to get the following result:
S
βα
=¸p
1
p
n
out[k
1
k
m
in)
∼
_
i
√
Z
_
m+n m
i=1
_
d
4
x
i
f
k
i
(x
i
)(2
x
i
+m
2
)
n
j=1
d
4
y
j
f
∗
p
j
(y
j
)(2
y
j
+m
2
)
¸0[T[ϕ(y
1
) ϕ(y
n
)ϕ(x
1
) ϕ(x
m
)][0)
(7.43)
126CHAPTER 7. THE INTERACTIONPICTURE ANDTHE SMATRIX
Equation (7.43), to quote Bjorken and Dell, “serves as a cornerstone for all
calculation of scattering amplitudes in modern quantum ﬁeld theory.” (This
was before the days of path integrals.)
Feynman rules are usually expressed in momentum space, so introduce
the Fourier transform,
G(x
1
x
m
y
1
y
n
) =
m
i=1
_
d
4
k
i
e
ik
i
x
i
(2π)
4
n
j=1
_
d
4
p
j
e
−ip
j
y
j
(2π)
4
˜
G(k
1
k
m
p
1
p
n
)
(7.44)
There is a possible confusion about this equation. Since
G(x
1
x
m
y
1
y
n
) = ¸0[T[ϕ(y
1
) ϕ(y
n
)ϕ(x
1
) ϕ(x
m
)][0) (7.45)
There is no distinction between x’s and y’s, but in (7.44), x’s are clearly
associated with incoming particles, and y’s are associated with outgoing
particles. The point is that G(x
1
x
m
y
1
y
n
) is a “general purpose”
tool that can be used in many situations including those on a quantum
mechanical time scale in which there is no clearcut distinction between
past and future. In scattering experiments, on the other hand, past and
future are unambiguous. We therefore associate some coordinates with in
states and some with outstates. We do this by assigning them to k’s (for
incoming particles) or with p’s (for outgoing particles). In this simple theory
with selfinteracting scalar ﬁelds, the only diﬀerence is the + or − signs in
the exponentials in (7.44), but this is necessary to get the conservation
of momentum right. When there are several diﬀerent kinds of particles
involved, this choice becomes even more signiﬁcant. For example, if we were
looking at the process e
+
+ e
−
→ γ + γ, (positron annihilation) we would
need to assign x’s to the electron ﬁelds and y’s to the photon ﬁelds. After
making this assignment, substitute (7.44) into (7.43) and do the necessary
integrals.
S
βα
=¸p
1
p
n
out[k
1
k
m
in)
∼
_
i
√
Z
_
m+n m
i=1
1
_
(2π)
3
2E
k
i
(k
2
i
−m
2
)
n
j=1
1
_
(2π)
3
2E
p
j
(p
2
j
−m
2
)
˜
G(k
1
k
m
p
1
p
n
)
(7.46)
7.4. CORRELATION FUNCTIONS 127
7.4 Correlation Functions
The heart of (7.43) is the vacuum correlation function
G(x
1
x
n
) = ¸0[T[ϕ(x
1
)ϕ(x
2
) ϕ(x
n
)][0) (7.47)
The ﬁelds ϕ(x) are the fully interacting ﬁelds in the Heisenberg picture.
If we are to calculate anything in this formalism, we must somehow relate
them to the “in” ﬁelds deﬁned in (7.30). Let’s hope there exists an operator
U(t) such that
ϕ(x) = U
−1
(t)ϕ
in
(x)U(t) with t = x
0
(7.48)
I happen to know that the necessary operator is given by
U(t) = lim
t
0
→−∞
U(t, t
0
) (7.49)
where U(t, t
0
) is time timedevelopment operator given by (7.21). This is
plausible, but it need to be proved.
Both ϕ and ϕ
in
satisfy Heisenberg equations of motion.
∂
∂t
ϕ(t) = i[H(ϕ), ϕ(t)] (7.50)
∂
∂t
ϕ
in
(t) = i[H
0
(ϕ
in
), ϕ
in
(t)] (7.51)
I have written H(ϕ) and H
0
(ϕ
in
) to emphasize that the Hamiltonians are
functionals of their respective ﬁelds. I have also written ϕ(t). I realize the
ϕ is really a function of the four vector x
µ
, but I am speciﬁcally concerned
with the time development here, and x
0
= t. Note that
d
dt
_
UU
−1
¸
=
dU
dt
U
−1
+U
dU
−1
dt
= 0
In a condensed notation this is
U
˙
U
−1
= −
˙
UU
−1
(7.52)
Now substitute (7.48) into (7.51) and use (7.50) to eliminate ˙ ϕ.
∂
∂t
ϕ
in
=
∂
∂t
_
UϕU
−1
¸
=
˙
UϕU
−1
+U ˙ ϕU
−1
+Uϕ
˙
U
−1
=
˙
U(U
−1
ϕ
in
U)U
−1
+Ui [H(ϕ), ϕ] U
−1
+U(U
−1
ϕ
in
U)
˙
U
−1
128CHAPTER 7. THE INTERACTIONPICTURE ANDTHE SMATRIX
=
˙
UU
−1
ϕ
in
−ϕ
in
˙
UU
−1
+i
_
(UH(ϕ)U
−1
), UϕU
−1
¸
=
_
˙
UU
−1
, ϕ
in
_
+i [H(ϕ
in
), ϕ
in
] = i [H
0
(ϕ
in
), ϕ
in
] (7.53)
I have used (7.52) to simplify the second line of (7.53). I have also used the
fact that Hamiltonians are polynomials of ϕ (and possibly its derivatives),
so for example, UH(ϕ)U
−1
= H(ϕ
in
). Therefore
__
˙
UU
−1
+iH
int
(ϕ
in
)
_
, ϕ
in
_
= 0
The operator in curly brackets commutes with all ϕ
in
. It must therefore
be a c number. We could think of this constant as a number added to the
Hamiltonian. We have had several occasions to discard such numbers before
and will do so now. We conclude that (to within a meaningless additive
constant)
i
dU(t)
dt
= H
int
(ϕ
in
)U(t) (7.54)
This is equation (7.15) with the important clariﬁcation that H
int
is con
structed with in ﬁelds, i.e. ﬁelds using the physical masses of the particles
involved. We can take U(t, t
0
) to be given by (7.21) and U(t) by (7.49).
Using (7.12) and (7.13) we can argue that
U(t
2
, t
1
) = lim
t→−∞
U(t
2
, t
0
)U(t
0
, t
1
)
= lim
t→−∞
U(t
2
, t
0
)U
−1
(t
1
, t
0
) = U(t
2
)U
−1
(t
1
) (7.55)
Now consider a set of timeordered points x
1
, , x
n
satisfying x
1
0
>
x
0
2
> > x
0
n
. Using (7.55) we can write
G(x
1
, , x
n
) = ¸0[ϕ(x
1
) ϕ(x
n
)[0)
= ¸0[U(t
0
, t
1
)ϕ
in
(x
1
)U(t
1
, t
2
)ϕ
in
(x
2
) U(t
n−1
, t
n
)ϕ
in
(x
n
)U(t
n
, t
0
)[0)
(7.56)
The sequence of times reading from left to right runs from t
0
→ −∞ to t
1
and back again to −∞. We can deform the contour by writing U(t
0
, t
1
) =
U(t
0
, −t
0
)U(−t
0
, t
1
). An operator of the form U(−t
0
, t
0
) runs from −∞ to
+∞. It has the eﬀect of transforming asymptotic initial states to asymptotic
ﬁnal states. This is just the role of the S operator deﬁned in (7.50). We will
simply call U(t
0
, −t
0
) = S
−1
. With this change, our correlation function
becomes
G(x
1
, , x
n
)
= ¸0[S
−1
U(−t
0
, t
1
)ϕ
in
(x
1
)U(t
1
, t
2
) U(t
n−1
, t
n
)ϕ
in
(x
n
)U(t
n
, t
0
)[0)
(7.57)
7.4. CORRELATION FUNCTIONS 129
According to (7.27), the vacuum state is an eigenstate of S with a complex
phase. We can write ¸0[S
−1
= e
−iθ
¸0[ and ﬁnd the phase as follows.
¸0[S[0)¸0[S
−1
= ¸0[
The point is that since the vacuum is stable, [0)¸0[ constitutes a complete
set of states. Consequently
¸0[S
−1
=
¸0[
¸0[S[0)
(7.58)
Equation (7.57) becomes
G(x
1
, , x
n
) (7.59)
=
¸0[U(−t
0
, t
1
)ϕ
in
(x
1
)U(t
1
, t
2
) U(t
n−1
, t
n
)ϕ
in
(x
n
)U(t
n
, t
0
)[0)
¸0[S[0)
Consider an operator like U(t
i−1
, t
i
). It contains an inﬁnite sum of prod
ucts of ﬁelds evaluated at various times, but every term in the sum contains
ﬁelds that are timeordered between t
i−1
and t
i
. The entire numerator of
(7.59) consists of a vast array of ﬁelds, but they are all time ordered. Of
course, we started out with the assumption that the sequence of coordi
nates x
1
, x
2
, x
n
was already time ordered, so this makes the “scaﬀolding”
around which the U’s are time ordered. The original deﬁnition of G, equa
tion (7.47) already contains the time ordering operator, so we need not have
made that assumption. We just need to write it
G(x
1
, , x
n
) (7.60)
=
¸0[T [U(−t
0
, t
1
)ϕ
in
(x
1
)U(t
1
, t
2
) U(t
n−1
, t
n
)ϕ
in
(x
n
)U(t
n
, t
0
)] [0)
¸0[S[0)
With the timeordering operator in place, we are free to move the factors
around in the numerator to suit our convenience. In particular, we can
gather all the U’s together.
U(−t
0
, t
1
)U(t
1
, t
2
) U(t
n
, t
0
) = U(−t
0
, t
0
) = S
Our ﬁnal result for the correlation function function is
G(x
1
, , x
n
) =
¸0[T [ϕ
in
(x
1
)ϕ
in
(x
2
) ϕ
in
(x
n
)S] [0)
¸0[S[0)
(7.61)
130CHAPTER 7. THE INTERACTIONPICTURE ANDTHE SMATRIX
Itzykson and Zuber
4
in a masterstroke of French understatement, call
this derivation “heuristic.” The fact is that we just don’t know how to
do rigorous mathematics with such badlybehaved operator expansions as
(7.20), and there are many subtle physical questions as well. You might call
this “experimental mathematics.” We use this formula and it works. It is
also true that equivalent formulas can be derived using the path integral for
malism. This also goes way beyond the limits of conventional mathematics,
but the same result is obtained in the end.
7.5 The Wick Expansion
If we can calculate S
βα
in (7.46), we can calculate all possible decay rates
and cross sections. In order to do this we need the Fourier transformed
Green’s function given (implicitly) by (7.44).
˜
G(k
1
k
m
p
1
p
n
) =
m
i=1
_
d
4
x
i
e
−ik
i
x
i
n
j=1
_
d
4
y
j
e
ip
j
y
j
G(x
1
x
m
y
1
y
n
)
(7.62)
Now we need the conﬁguration space Green’s function G(x
1
x
m
y
1
y
n
).
The formula for this is (7.61)
G(x
1
, x
2
, , x
n
) = ¸0[T[ϕ(x
1
)ϕ(x
2
) ϕ(x
n
)][0)
=
¸0[T[ϕ
I
(t
1
)ϕ
I
(t
2
) ϕ
I
(t
n
)S][0)
¸0[S[0)
,
(7.63)
By S in (7.63) I mean the operator U(−t
0
, t
0
) in the limit t
0
→−∞. Then
the equation for S is (7.20) with inﬁnite limits on the integrals.
S =
∞
n=0
(−i)
n
n!
_
_
d
4
x
1
d
4
x
2
d
4
x
n
T[H
I
int
(x
1
)H
I
int
(x
2
) H
I
int
(x
n
)],
(7.64)
where it is understood that the limits on the time integrations are −∞ <
x
0
< ∞ and that H
I
int
is constructed with ϕ
in
given by (7.33).
This seems (and is) an excruciatingly complicated set of equations to
wade through. The fact is that when you ﬁnally get the results in momentum
space, they are often embarrassingly simple. This led Feynman to formulate
a set of rules that give the right answer in momentum space without having
4
Quantum Field Theory, C. Itzykson and J.B. Zuber, McGrawHill 1980
7.5. THE WICK EXPANSION 131
to calculate anything. I could just tell you the rules now, but they would
be devoid of any intellectual content. Let’s derive them step by step.
Take a look at the derivation of the propagator at the end of Chapter 6.
There we were dealing with free ﬁelds so we should rewrite (6.72)
¸0[T[ϕ
I
(x
1
)ϕ
I
(x
2
)][0) = iD(x
1
−x
2
) (7.65)
I am going to describe the steps that led from (6.72) to (6.75) in rather
formal terms that make it easy to generalize to situations in which there are
more than two ﬁelds. In order to keep the notation under control, I will write
ϕ(x) = ϕ
+
(x) +ϕ
−
(x), where ϕ
+
is the positiveenergy piece containing an
annihilation operator, and ϕ
−
is the negativeenergy term with the creation
operator.
• Multiply the ﬁelds together with the timeordering θ functions. This
gives eight terms, each one containing a product of two of the creation
and/or annihilation operators.
T[ϕ(x
1
)ϕ(x
2
)] = (ϕ
+
(x
1
) +ϕ
−
(x
1
))(ϕ
+
(x
2
) +ϕ
−
(x
2
))θ(t
1
−t
2
)
+ (ϕ
+
(x
2
) +ϕ
−
(x
2
))(ϕ
+
(x
1
) +ϕ
−
(x
1
))θ(t
2
−t
1
)
• Use the commutation relations (6.65) to rewrite each term (if neces
sary) in such a way that all creation operators stand to the left
of all annihilation operators. This is called normal ordering. The
usual notation for this is to enclose the expression between colons, i.e.
: ϕ(x
1
)ϕ(x
2
) : means that the terms between the colons have already
been placed in normal order. In the course of normal ordering there
will appear terms without operators since, for example
ϕ
+
(x
1
)ϕ
−
(x
2
) =: ϕ
−
(x
2
)ϕ
+
(x
1
) : +
_
d
3
k
(2π)
3
2ω
e
−ik(x
1
−x
2
)
The ﬁrst term on the right is normal ordered. The second term is a
cnumber.
• Take advantage of the fact that the vacuum expectation value of any
normal ordered product is zero. As a consequence
¸0[T[ϕ(x
1
)ϕ(x
2
)][0) = ¸0[ : ϕ(x
1
)ϕ(x
2
) : [0) +iD(x
1
−x
2
) (7.66)
Of course, the ﬁrst term on the right is zero. We say that iD(x
1
−x
2
) is
the contraction of ϕ(x
1
) and ϕ(x
2
). I will use the overbrace to indicate
contraction. In this notation (7.66) is
T[ϕ(x
1
)ϕ(x
2
)] =: ϕ(x
1
)ϕ(x
2
) : +
¸ .. ¸
ϕ(x
1
)ϕ(x
2
) (7.67)
132CHAPTER 7. THE INTERACTIONPICTURE ANDTHE SMATRIX
I am now in a position to state Wick’s theorem. The time ordered
product of n ﬁelds T[ϕ(x
1
)ϕ(x
2
) ϕ(x
n
)] can be written as a sum of terms
as follows:
• A single term in which all n ﬁelds are normalordered.
• A sum of terms, each one of which consists of one pair of contracted
ϕ’s. The n−2 remaining ﬁelds are normalordered. There is one term
in the sum for each possible pairing of two of the ﬁelds.
• A sum of terms, each one of which consists of two pairs of contracted
ϕ’s. The remaining n −4 remaining ﬁelds are normalordered. There
is one term in the sum for each way in which two distinct pairs can be
made out of n ﬁelds.
• You see the pattern? One term with no contractions plus all possible
terms with one contraction plus all possible terms with two contrac
tions plus all possible terms with three contractions and so forth. If
n is even, repeat this procedure n/2 times. The last group of terms
will consist of n/2 contractions. There will be one term in this group
for each way of choosing n/2 pairs out of n objects. If n is odd, re
peat this procedure (n − 1)/2 times. The last terms will have one
normalordered ﬁeld left over
This is all much easier to understand that to explain. The following
example with n = 4 should make this clear.
T[ϕ(x
1
)ϕ(x
2
)ϕ(x
3
)ϕ(x
4
)] =: ϕ(x
1
)ϕ(x
2
)ϕ(x
3
)ϕ(x
4
) :
+
¸ .. ¸
ϕ(x
1
)ϕ(x
2
) : ϕ(x
3
)ϕ(x
4
) : +
¸ .. ¸
ϕ(x
1
)ϕ(x
3
) : ϕ(x
2
)ϕ(x
4
) :
+
¸ .. ¸
ϕ(x
1
)ϕ(x
4
) : ϕ(x
2
)ϕ(x
3
) : +
¸ .. ¸
ϕ(x
2
)ϕ(x
3
) : ϕ(x
1
)ϕ(x
4
) :
+
¸ .. ¸
ϕ(x
2
)ϕ(x
4
) : ϕ(x
1
)ϕ(x
3
) : +
¸ .. ¸
ϕ(x
3
)ϕ(x
4
) : ϕ(x
1
)ϕ(x
2
) :
+
¸ .. ¸
ϕ(x
1
)ϕ(x
2
)
¸ .. ¸
ϕ(x
3
)ϕ(x
4
) +
¸ .. ¸
ϕ(x
1
)ϕ(x
3
)
¸ .. ¸
ϕ(x
2
)ϕ(x
4
)
+
¸ .. ¸
ϕ(x
1
)ϕ(x
4
)
¸ .. ¸
ϕ(x
2
)ϕ(x
3
)
(7.68)
You should try working out one of these time ordered products by brute force
starting with (6.64) and (6.65). You will be convinced that the theorem is
true. The proof is more work that it’s worth.
5
5
If you want a proof see J. D. Bjorken and S. D. Drell, Relativistic quantum Fields,
Section 17.4
7.6. AN EXAMPLE 133
This would be a good time to go back to the free ﬁeld Hamiltonian, Eq.
(6.43).
H =
1
2
n=∞
n=−∞
¯hω
n
(a
n
a
†
n
+a
†
n
a
n
) =
n=∞
n=−∞
¯hω
n
_
a
†
n
a
n
+
1
2
_
We had to discard the ¯ hω/2 term to avoid an inﬁnite sum. This is a common
problem whenever we make Hamiltonians our of ﬁeld operators. In order
to insure that this never happens, we will agree that all Hamiltonians come
already normally ordered. Accordingly (7.64) should be written
S =
∞
n=0
(−i)
n
n!
_
_
d
4
x
1
d
4
x
2
d
4
x
n
T[: H
I
int
(x
1
) :: H
I
int
(x
2
) : : H
I
int
(x
n
) :].
(7.69)
When we use Wick’s theorem on this expression, we only make contractions
of pairs of ﬁelds from diﬀerent Hamiltonians.
7.6 An Example
Let’s take the interaction Hamiltonian H
int
= (λ/4!)ϕ
4
(x) that we stud
ied in Sections 4.3 and 4.4, and calculate the fourpoint Green’s function
G(x
1
, x
2
, x
3
, x
4
) to ﬁrst order in λ. The Soperator (7.64) to ﬁrst order is
S = 1 −
iλ
4!
_
d
4
x : ϕ
4
(x) : + (7.70)
Note that to ﬁrst order ¸0[S[0) = 1, so (7.63) becomes
G(x
1
, x
2
, x
3
, x
4
) = ¸0[T[ϕ
I
(x
1
)ϕ
I
(x
2
)ϕ
I
(x
3
)ϕ
I
(x
4
)][0)
−
iλ
4!
¸0[T
_
ϕ
I
(x
1
)ϕ
I
(x
2
)ϕ
I
(x
3
)ϕ
I
(x
4
)
_
d
4
x : ϕ
4
I
(x) :
_
[0) +
(7.71)
The zerothorder term can be read oﬀ of (7.68)
G(x
1
, x
2
)G(x
3
, x
4
) +G(x
1
, x
3
)G(x
2
, x
4
) +G(x
1
, x
4
)G(x
2
, x
3
) (7.72)
This is the sum of all possible ways two particles can propagate between two
pairs of points. As you would expect in this zerothorder term, the particles
do not interact .
The ﬁrstorder term has eight ﬁelds each with two terms. If you were to
simply multiply them out there would be 2
8
= 256 terms, so here’s where
134CHAPTER 7. THE INTERACTIONPICTURE ANDTHE SMATRIX
Mr. G. C. Wick earns his keep. First notice that the vacuum expectation
value of any normalordered product of ﬁelds vanishes. We need only be
concerned with terms in which all eight ﬁelds are contracted. Because of
the normalordered Hamiltonian, we don’t contract the four ϕ(x)’s with one
another. The only surviving terms are those in which each of ϕ(x
1
), ϕ(x
2
),
ϕ(x
3
), and ϕ(x
4
) is contracted with one of the ϕ(x)’s. There are, of course,
4! ways of doing this all leading to the same expression for the ﬁrstorder
contribution to G(x
1
, x
2
, x
3
, x
4
).
G(x
1
, x
2
, x
3
, x
4
) = −iλ
_
d
4
xG(x
1
, x)G(x
2
, x)G(x
3
, x)G(x
4
, x), (7.73)
which is exactly (4.74).
Now suppose you want to calculate the S matrix element for the scat
tering of two particles k
1
+ k
2
→ p
1
+ p
2
. There are two ways of doing
this.
• First do the d
4
x integration in (7.73). Then substitute G(x
1
, x
2
, x
3
, x
4
)
into (7.62) and ﬁnd the Fourier transform
˜
G(k
1
, k
2
, p
1
, p
2
). Finally
substitute
˜
G into (7.46) to get S
βα
.
• G is a Green’s function after all, so
(2
x
+m
2
)G(x, x
) = i(2
x
+m
2
)D(x −x
) = −iδ
(4)
(x −x
) (7.74)
You see, all the (2
x
i
+ m
2
) operators in (7.43) produce factors of
−iδ
(4)
(x
i
− x) when they operate on G(x
i
, x). Now all the integrals
are trivial.
S
βα
=¸p
1
p
2
out[k
1
k
2
in)
=
_
i
√
Z
_
4
(−iλ)(2π)
4
δ
(4)
(p
1
+p
2
−k
1
−k
2
)
2
i=1
1
_
(2π)
3
2E
k
i
2
j=1
1
_
(2π)
3
2E
p
j
(7.75)
This second technique, though less obvious, saves a factor of two in scratch
paper.
This simple calculation illustrates most of the rules for calculating more
complicated Smatrix elements in the ϕ
4
theory. To start with draw some
diagrams with the incoming and outgoing particles labelled with their mo
menta. The diagrams will consist of straight lines and vertices where four
7.6. AN EXAMPLE 135
lines meet. The number of vertices must be less than or equal to the corre
sponding order of perturbation theory, i.e. if you working at the nth order
approximation, you must draw all possible diagrams in which there are n
or fewer vertices. In the example, there was only one vertex corresponding
to ﬁrstorder perturbation theory. There will be many diagrams in which
one or more of the particles don’t interact with the others. One way to say
this is that these diagrams can be separated into two or more pieces without
cutting any lines. These are called “disconnected” diagrams. These will
fall into two categories: “vacuum ﬂuctuation” diagrams in which there is a
piece that is disconnected from all the external particles and “self interac
tion” diagrams in which one of the external particles emits and reabsorbs
virtual particles without interacting with the other external particles. Dis
card all these. It can be shown that the vacuum ﬂuctuations only contribute
to the ¸0[S[0) term, which cancelled in the formula for the Smatrix. The
self interaction diagrams represents particles that don’t scatter, and we are
presumably not interested in the probability of things not happening. Each
of the remaining diagrams contributes a term to the perturbation expansion.
Each term will have the following factors.
• A factor 1/
_
(2π)
3
2E
k
for each external particle with momentum k.
• A factor (−iλ) for each vertex.
• A factor (2π)
4
δ
(4)
(k
1
+k
2
−p
1
−p
2
− ) to enforce allover momentum
conservation.
These factors are all evident in (7.75). There are several new features that
are only present in higherorder terms.
• There will be internal lines, i.e. lines that begin and end in the dia
gram. Each one of these gets a propagator
i
p
2
−m
2
+i
where p is the momentum of the internal particle.
• Impose momentum conservation at each vertex.
• Integrate over each undetermined loop momentum:
_
d
4
p
(2π)
4
136CHAPTER 7. THE INTERACTIONPICTURE ANDTHE SMATRIX
• Divide by the symmetry factor.
This last issue, the symmetry factor, has to do with redundant diagrams.
It’s too complicated to explain in the text. I’ll explain it in class when I can
draw lots of diagrams on the board.
There’s still one more step on the road. How do you turn an Smatrix
into a cross section. That is the subject of the next chapter.
Chapter 8
Cross Sections and Decay
Rates
The last step in the road from Lagrangians to cross sections is this: take
the Smatrix computed in Chapter 7 and convert it to a scattering cross
section or decay rate. There are two easy ways of doing this and one right
way. One easy way is to assume that the initial and ﬁnal state particles can
be described by plane waves. Plane waves occupy inﬁnite volume, unfortu
nately, so several key integrals diverge. This problem can be ducked with
some judicious hand waving. The other easy way is to quantize the wave
functions in a box of ﬁnite size. This avoids the inﬁnite integrals, but of
course, real experiments aren’t done in a box. It is best to acknowledge that
the beam and target particles should be represented by wave packets, but
this makes for a long tedious derivation. We’ll do it the ssecond easy way
(at least in this edition).
8.1 Classical Scattering
From the point of view of classical mechanics, the cross section of a particle
is just its projected area. If the particle is a sphere of radius R, the cross
section σ
T
= πR
2
. This is usually called the “total” cross section. If the
particle were very small, we might contrive to measure its cross section by
means of a scattering experiment. First we prepare a target with n
t
of
these particles per unit volume, and bombard this target with N
i
incident
pointlike particles. We assume that the target is so thin and the target
particles so small that no one of them gets in front of another. We then
count the number N of incident particles that are deﬂected as a result of a
137
138 CHAPTER 8. CROSS SECTIONS AND DECAY RATES
collision. The fraction of particles that are scattered, N/N
i
must be equal
to the fraction of the target area covered by the projected area of the target
particles, n
t
Lσ
T
, where L is the target thickness. The cross section is then
given by
σ
T
=
N
N
i
n
t
L
The idea works even if σ
T
is arbitrarily small. We just have to make N
i
and
n
t
large enough.
We can also determine the shape with a scattering experiment. Imagine
that the target particles have concentric circles painted on them like an
archery target. If one of the incident particles hits a particular area dσ it
will scatter into a solid angle dΩ. OK – so your’re not as good with a bow
and arrow as Legolas. Just ﬁre a gazillion arrows at random an observe the
distribution of scattering angles. Your favorite theorist should then be able
to infer the shape of the target particle from the distribution. The relevant
equation is
dσ
dΩ
=
1
N
i
n
t
L
dN
dΩ
(8.1)
If you are an experimentalist, you measure dN/dΩ in your lab, compute
dσ/dΩ, and pass it along to the theorist. This computed quantity is called
the “diﬀerential cross section.”
The quantities N
i
, n
t
, and L in (8.1) are directly measurable, but for
purposes of comparing with theory, it is more convenient to express them in
terms of ﬂux. If the beam is moving at velocity v toward a stationary target,
then the number of particles in the beam N
i
is equal to the density of the
beam n
i
times the volume. If the beam is a pulse that is turned on for T
seconds, then the volume of the beam is vTA, where A is the crosssectional
area of the beam (assumed smaller than the target.) Therefore N
i
= n
i
vTA.
The cross section can be written as:
dσ
dΩ
=
1
(n
i
vTA)n
t
L
dN
dΩ
=
1
I
dR
dΩ
,
(8.2)
where the transition rate R is the number of scattering events per unit time
per unit target volume, V = AL.
R = N/V T, (8.3)
and the ﬂux I is
I = n
i
n
t
v (8.4)
8.2. QUANTUM SCATTERING 139
8.2 Quantum Scattering
In the quantum mechanics regime, quantities like size and shape don’t have
any direct meaning, since any putative size is bound to be smaller that the
corresponding deBroglie wavelength; but the experiment can still be done
exactly as I have described it, because all the measurements are macro
scopic and asymptotic. The diﬀerential cross section (DCS for short) can’t
be interpreted directly in terms of shape, but it does contain most of the
information we are entitled to have about particles and other subatomic
systems.
The description of scattering in terms of archery targets is appropri
ate to lowenergy potential scattering, but at higher energies new particles
can be created and the total momentum shared among them. It requires
more than a scattering angle to describe the ﬁnal state. Rather we have
to specify the momenta of each of the particles: p
1
, p
2
, . . . , p
n
. Not all val
ues of the momenta are allowed of course, because of the conservation of
momentum and energy. We can imagine the momenta as occupying some
complicatedshaped region in a 3ndimensional space whose coordinates are
the components of the individual momenta. This construct is called “phase
space.”
1
A typical multiparticle scattering experiment measures the transi
tion probability into some region of phase space. We still use the diﬀerential
cross section notation dσ, but now dσ might be an 3nfold diﬀerential, de
pending on the details of the experiment. I will also use the notation ∆Ω to
indicate some small region of phase space (not just a region of solid angle).
Now let’s put on our theorist’s hats. We know how to calculate the S
matrix elements. How do we now ﬁnd the cross sections? You will recall
from the end of Chapter 7 that the ﬁrst term of the perturbation theory
expansion of the Smatrix describes particles that don’t scatter. Feynman’s
rules decree that all other terms contain the factor (2π)
4
δ
(4)
(P
f
−P
i
), where
P
i
and P
f
are the total momentum fourvectors of the initial and ﬁnal state
particles. It is customary to deﬁne a transition matrix T
fi
as follows:
S
fi
= δ
fi
−i(2π)
4
δ
(4)
(P
f
−P
i
)T
fi
(8.5)
The δ
fi
just represents the particles that don’t scatter. T is the nontrivial
part of the Smatrix with the delta function “peeled oﬀ.” If should be clear
from (6.25) and (6.35) that the S operator is unitary. The Smatrix in
the current notation is S
fi
= ¸f[S[i), where [i) and ¸f[ represent complete
1
Note that phase space in thermodynamics is 6ndimensional, since it includes space
as well as momentum coordinates.
140 CHAPTER 8. CROSS SECTIONS AND DECAY RATES
sets of states. As such we expect the Smatrix to be unitary as well. Very
schematically, this means that
f
S
∗
fi
S
fk
= δ
ik
(8.6)
I say “very schematically” because it’s not at all clear what it means to sum
over one of these indices. I’ll get back to that.
Since S
fi
is a probability amplitude, we should get the corresponding
probability by squaring it. So for example,
P
fi
= [S
fi
[
2
(8.7)
is the probability that the initial state [i) will scatter into the ﬁnal state [f).
Unfortunately this means that we have to square the delta function in (8.5),
and this is bad news, since δ(P
f
−P
i
)δ(P
f
−P
i
) = δ(P
f
−P
i
)δ(0) and δ(0)
is inﬁnite.
I’ll deal with this last problem ﬁrst. Squaring (8.5) gives
P
fi
= (2π)
4
δ
(4)
(0)(2π)
4
δ
(4)
(P
f
−P
i
)[T
fi
[
2
(8.8)
(I have ignored the nonscattering term.) Here’s the kludge. The factor
(2π)
4
δ
(4)
(0) is interpreted as a volume of spacetime, since in the limit P →0
(2π)
4
δ
(4)
(P) =
_
d
4
x e
iPx
→
_
d
4
x = V T. (8.9)
But the time and volume of what? We can use the derivation of (8.2) as
a model. Let V be a small arbitrary volume inside the interaction region.
Assume that it is small enough to ﬁt inside the beam and target but very
large compared the typical quantummechanical dimensions. T is the time
required for a beam particle to pass through V . The analogy to (8.3) is
R
fi
= transition rate per unit volume =
P
fi
V T
=
(2π)
4
V Tδ
(4)
(P
f
−P
i
)[T
fi
[
2
V T
=(2π)
4
δ
(4)
(P
f
−P
i
)[T
fi
[
2
(8.10)
This is the ﬁrst of several instances in which the arbitrary V ’s and T’s cancel.
Now to the question of counting and normalizing states. This is not
explained clearly in any book I know of, so I hope you will indulge me while
8.2. QUANTUM SCATTERING 141
I explain this boring technical detail in excruciating detail. Someone has to
do it.
First of all, KleinGordon wave functions have a conserved norm, but it
isn’t
_
dV [ϕ[
2
. In order to have the right behavior under Lorentz transforma
tions, the norm must transform like the timelike component of a conserved
current. That is, there must be a probability density ρ and a probability
current j such that
∂ρ
∂t
+∇· j = 0 (8.11)
The reason is that the rate of decrease of particles in a given volume is equal
to the total ﬂux of particles out of that volume, that is
−
∂
∂t
_
V
ρ dV =
_
S
j ˆ n dS =
_
V
∇ j dV
The procedure for ﬁnding this current should be familiar from nonrelativistic
quantum mechanics. Write the KleinGordon equation as
−
∂
2
ϕ
∂t
2
+∇
2
ϕ = m
2
ϕ
Multiply this equation by −iϕ
∗
and the complex conjugate equation by −iϕ
and subtract.
∂
∂t
_
i
_
ϕ
∗
∂ϕ
∂t
−ϕ
∂ϕ
∗
∂t
__
. ¸¸ .
ρ
+∇ [−i(ϕ
∗
∇ϕ −ϕ∇ϕ
∗
)]
. ¸¸ .
j
= 0 (8.12)
As a consequence the norm is
ρ = i
_
d
3
x ϕ
∗
(x)
←→
∂
0
ϕ(x) (8.13)
By ϕ(x) I mean the wave function, not the quantized ﬁeld, and
ϕ
∗
←→
∂
0
ϕ ≡ ϕ
∗
∂
0
ϕ −(∂
0
ϕ
∗
)ϕ.
We have been using the quantized ﬁeld operator
ˆ ϕ =
_
d
3
k
_
(2π)
3
2E
k
_
ˆ a
k
e
−ikx
+
ˆ
d
†
k
e
ikx
_
142 CHAPTER 8. CROSS SECTIONS AND DECAY RATES
from (6.64). We obtain the classical wave function ϕ from the quantized
ﬁeld operator ˆ ϕ as follows
ϕ(x) = ¸0[ ˆ ϕ(x)[p) =
_
d
3
k
_
(2π)
3
2E
k
e
−ikx
¸0[ˆ a
k
ˆ a
†
p
[0)
=
e
−ipx
_
(2π)
3
2E
p
(8.14)
This result depends on three normalization conventions that are to some ex
tent arbitrary. First is the factor 1/
_
(2π)
3
2E in the ﬁeld operator (6.64).
Second is the commutation relations for the creation and annihilation op
erators (6.65). Finally, we have agreed to construct singleparticle states
as [p) = ˆ a
†
p
[0). Only the equaltime commutation relations (6.69) are non
negotiable. We will ﬁnally arrive at formulas for the cross sections and decay
rates that must be independent of all conventions. This will happen, but no
two books do it the same way!
Our choice of normalization for the ﬁeld operator has the advantage of
making (6.65) as simple as possible, but it results in a rather oddlooking
norm.
ρ = i(ϕ
∗
←→
∂
0
ϕ) = 2Eϕ
∗
ϕ = (2π)
−3
_
V
ρ dV = (2π)
−3
V
(8.15)
Yes that’s right – there are (2π)
−3
particles per unit volume!
Equation (8.10) refers to scattering into a single ﬁnal state [f). What
does this mean? Since we already are committed to the imaginary box,
we can impose periodic boundary condition to discretize the momenta. For
a single particle, quantum theory restricts the number of ﬁnal states in a
volume V with momenta in element d
3
p to be V d
3
p/(2π)
3
. Our V has
(2π)
−3
V particles, however, so
No. of ﬁnal states/particle =
V d
3
p
(2π)
3
(2π)
−3
V
= d
3
p (8.16)
In the general case in which two particles go into N particles with initial
and ﬁnal states given by
[i) = [k
2
, k
2
)
[f) = [p
1
, p
2
, , p
N
)
(8.17)
8.3. PHASE SPACE 143
the “density of states” is
No. of available states =
N
j=1
d
3
p
j
(8.18)
Now turn to the incident ﬂux (8.4). With our norm n
i
= n
t
= (2π)
−3
,
and v is the relative velocity of the colliding particles.
v =
¸
¸
¸
¸
k
1
E
1
−
k
2
E
2
¸
¸
¸
¸
=
1
E
1
E
2
_
(k
1
k
2
)
2
−(m
1
m
2
)
2
(8.19)
The generalization of (8.4) is
dσ =
R
I
density of states
Using (8.10), (8.18) and (8.7) we get
dσ =
(2π)
4
δ(P
f
−P
i
)[T
fi
[
2
(2π)
−6
v
N
j=1
d
3
p
j
=
1
v
(2π)
10
δ
(4)
(P
f
−P
i
)[T
fi
[
2
N
j=1
d
3
p
j
.
(8.20)
Finally, we use this formalism to compute the probability of the decay
of a single particle. The diﬀerential decay rate dΓ of an unstable particle is
deﬁned by
dΓ =
R
n
t
density of states
= (2π)
7
δ
(4)
(P
f
−P
i
)[T
fi
[
2
N
j=1
d
3
p
j
(8.21)
In this case n
t
is the number of decaying particles per unit volume, which
as usual is equal to (2π)
−3
.
8.3 Phase Space
The outcome of a scattering experiment as described by (8.20) depends,
generally speaking, on three kinds of things. First there are the fundamental
dynamics, which are encoded in the Hamiltonian of the system. Second,
144 CHAPTER 8. CROSS SECTIONS AND DECAY RATES
there are the details of the experiment itself, e.g. the number of particles
in the beam and target, the construction of the detectors, etc. There is a
third category of inﬂuences that arise because momentum and energy are
conserved and because the experiment must be describable in a relativistic
invariant way. These last considerations are incorporated in the notion of
phase space. Phase space in this context is the set of momentum states
available to the particles after the interaction. Imagine leaving the parking
lot in front of a movie theater after the show is over. If there is only one
narrow road leading out of the lot, then it will take you a long time to
escape. If there are freeways leading out in all directions, the lot will empty
quickly. This is an analogy based on coordinate space. Try to imagine
something like this in momentum space and you will have a feeling for what
phase space means. If there are many momentum states available to the
ﬁnalstate particles, the scattering cross section will be large or the decay
will happen quickly. I will explain this again in the next paragraph using
a semiquantitative argument based on the uncertainty principle and ﬁnally
present a version of the argument that is correct relativistically.
In classical statistic mechanics the state of motion of a particle is repre
sented as a point in a sixdimensional manifold called phase space consisting
of three position and three momentum coordinates, i.e. (x, y, z, p
x
, p
y
, p
z
).
Classical mechanics places no restrictions on the density of these points,
since x and p are supposedly measurable with complete precision. Quan
tum mechanics on the other hand limits the precision by the uncertainty
principle,
∆x
j
∆p
j
≥ 2π¯h.
Phase space can be thought of as being subdivided into elementary cells of
size (2π¯h)
3
. The particle’s coordinates cannot be localized to any smaller
region. The number of states available to one particle will therefore be equal
to the total volume of phase space divided by the size of the elementary
cell. Assuming that the particle is contained in a volume V , the number of
available states is
N =
1
(2π¯h)
3
_
d
3
x d
3
p =
V
(2π¯h)
3
_
d
3
p.
The limits of integration come about from the conservation of mass and
energy. Since p
µ
p
µ
= m
2
, the momentum integration is bounded by a sphere
of radius [p[
2
= E
2
− m
2
, so that N depends on m and E. The factor
V/(2π¯h)
2
is part of the normalization discussed in the previous section. I
am concerned here with the integral over d
3
p.
8.3. PHASE SPACE 145
Now suppose there are N such particles in the ﬁnal state of a scattering
experiment.
N
N
=
_
N−1
j=1
d
3
p
j
(8.22)
There are only N − 1 integrals in (8.22), because the N p
j
’s are not all
independent,
N
j=1
p
j
≡ P
f
= P
i
.
We can incorporate this constraint in a more ﬂexible way by introducing a
δfunction.
_
d
3
p
N
δ
(3)
_
p
N
+
N−1
k=1
p
k
−P
i
_
= 1
N
N
=
_
N−1
j=1
d
3
p
j
=
_
N
j=1
d
3
p
j
δ
(3)
_
N
k=1
p
k
−P
i
_
Energy is also conserved, and this constraint can be included with a similar
trick.
_
dE
i
δ(E
f
−E
i
) = 1
N
N
=
_
dE
i
N
j=1
d
3
p
j
δ
(3)
_
N
k=1
p
k
−P
i
_
δ
_
N
k=1
E
k
−E
i
_
dN
N
dE
i
=
_
N
j=1
d
3
p
j
δ
(4)
(P
f
−P
i
) (8.23)
These are just the kinematic factors appearing in (8.20). It appears from
this heuristic argument, that the reaction rate is proportional to the density
of states dN
N
/dE
i
.
2
This is not the whole story, because (8.23) as it stands
is not Lorentz invariant, but according to the Feynman rules from Section
2
You will recall from elementary quantum mechanics that the transition probability
rate for a system to make a transition from an energy E
l
to E
k
is given by “Fermi’s golden
rule,”
w
lk
=
2π
¯h
g(E
k
)H
kl

2
where g(E
k
) is the density of states whose energy is close to E
k
. See R. L. Liboﬀ, Intro
ductory Quantum Mechanics.
146 CHAPTER 8. CROSS SECTIONS AND DECAY RATES
7.5, the square of the scattering amplitude, [¸f[T [i)[
2
in (8.20), contains a
factor 1/2E
i
(2π)
3
for each particle in the ﬁnal state. This is good, since
_
d
4
p δ(p
2
−m
2
) =
_
d
3
p
2E
is clearly a Lorentz scalar. The point is that the factors
_
_
N
j=1
_
d
3
p
j
_
_
(2π)
4
δ
(4)
(P
i
−P
f
)
from (8.20) together with the factors 1/2E
i
from [¸f[T [i)[
2
can be combined
into a Lorentzinvariant quantity
T =
_
N
j=1
d
3
p
j
2E
j
δ
(4)
(P
f
−P
i
)
=
_
N
j=1
d
4
p
j
δ(p
2
j
−m
2
) δ
(4)
(P
f
−P
i
)
(8.24)
The quantity T is called the Lorentzinvariant phase space. In general we
are more interested in integrating some of the momenta in (8.24) over some
portion of their allowed kinematic range, in which case
dT =
_
p∈∆
N
j=1
d
4
p
j
δ(p
2
j
−m
2
)δ
(4)
(P
f
−P
i
). (8.25)
You might say that (8.24) and (8.25) lie on the interface between theory and
experiment. For example, if you were using (8.20) to predict the results of an
experiment, the limits on the integrals would be set to represent the details
of the experimental apparatus. If the apparatus did not detect particle j (it
might be a neutral particle), then the corresponding momentum p
j
would
be integrated out. If the momentum of the k’th particle was measured, then
the corresponding integral would not be done, and T would be left as a
function of p
k
.
For manyparticle reactions, T will have a complicated dependence on
the momentum variables.
3
It is often the case that the data from an ex
periment are determined mostly by T regardless of the dynamics. Such
an experiment is a failure almost by deﬁnition. The general purpose of an
experiment is always to ﬁnd some deviation from phase space.
3
There are entire books written on the subject, e.g. Kinematics and Multiparticle
Systems, M. Nikolic, Gordon and Breach (1968)
8.4. TWOPARTICLE SCATTERING 147
8.4 TwoParticle Scattering
To make this more concrete, let’s calculate the cross section for the elastic
scattering of two particles k
1
+k
2
→p
1
+p
2
from the example in the previous
chapter. Comparing (7.75) (with Z = 1) with (8.5) gives
T
fi
=
λ
(2π)
6
_
2E
k
1
2E
k
2
2E
p
1
2E
p
2
(8.26)
Then (8.20) gives
dσ =
λ
2
4(2π)
2
vE
k
1
E
k
2
_
d
3
p
1
2E
p
1
d
3
p
2
2E
p
2
δ
(4)
(P
f
−P
i
)
=
λ
2
4(2π)
2
vE
k
1
E
k
2
_
dT
(8.27)
The transition matrix T
fi
is especially simple in this example. This is why
there is nothing inside the integral except dT. In general there will be some
complicated function that we may or may not want to integrate.
I will complete the calculation in the center of mass system assuming
equalmass particles. In this case [k
1
[ = [k
2
[ = [p
1
[ = [p
2
[, k
1
= −k
2
,
p
1
= −p
2
, E
p
2
= E
p
1
≡ E
1
, and E
k
1
+E
k
2
≡ E
i
. For the remainder of this
section only I will use the notation [p
1
[ = p
1
. Integrating out d
3
p
2
gives
T =
_
1
4E
2
1
p
2
1
dp
1
δ(2E
p
1
−E
i
)dΩ
1
It’s convenient to change the variable of integration from p
1
to E
1
.
p
1
dp
1
= E
1
dE
1
δ(2E
1
−E
i
) =
1
2
δ(E
1
−E
i
/2)
T =
_
p
1
8E
1
dE
1
δ(E
1
−E
i
/2)dΩ
1
There is still one δfunction, which will enforce energy conservation and
hence ﬁx the value of p
1
.
T =
_
p
1
8E
1
dΩ
1
=
πp
1
4E
1
. (8.28)
Notice that symmetry together with the conservation of momentum and
energy largely predetermine the outcome of such a scattering experiment.
If the particles are spinless there can be no dependence on the azimuthal
148 CHAPTER 8. CROSS SECTIONS AND DECAY RATES
angle. Only the scattering angle θ
1
is unconstrained, and the only way
the underlying dynamics could manifest itself is by aﬀecting the diﬀerential
cross section as a function of scattering angle. Our example is trivial in the
sense that there is no θ dependence, but let’s go ahead and calculate the
diﬀerential cross section anyhow. For this purpose it is best to leave (8.28)
in the form of a diﬀerential
dT =
p
1
8E
1
dΩ
1
(8.29)
We also need v from (8.19), which in this simple case gives
v =
2p
1
E
1
(8.30)
Equation (8.27) becomes
dσ =
λ
2
4(2π)
2
E
2
1
E
1
2p
1
p
1
8E
1
dΩ
1
dσ
dΩ
1
=
λ
2
(16πE
1
)
2
(8.31)
As anticipated, the scattering is isotropic. There is no θ
1
dependence.
8.5 The General Case
Here is a cookbook procedure for calculating scattering cross sections and
decay rates for the general case in which there are N particles in the ﬁnal
state.
• Use Feynman’s rules from Section 7.5 to calculate the S matrix.
• Find the transition matrix T
fi
from (8.5).
• Find the relative velocity v from (8.19).
• The diﬀerential cross sections and decay rates are calculated from
(8.20) and (8.21) integrated over the appropriate regions of phase
space.
dσ =
(2π)
10
v
_
p∈∆Ω
N
j=1
d
3
p
j
δ
(4)
(P
f
−P
i
)[T
fi
[
2
(8.32)
8.5. THE GENERAL CASE 149
dΓ = (2π)
7
_
p∈∆Ω
N
j=1
d
3
p
j
δ
(4)
(P
f
−P
i
)[T
fi
[
2
(8.33)
When there are only two particles in the ﬁnal state, one integrates
over all the momentum components that are constrained by the delta
function and leaves the remaining two angles in the form of a diﬀer
ential. When there are more than two particles in the ﬁnal state, the
decision regarding which variables to integrate and which to leave in
the diﬀerential depends on the experiment that the theory is describ
ing. That is the meaning of the notation p ∈ ∆Ω that speciﬁes the
limits of the integrals
Because of our normalization conventions, [T
fi
[
2
will contain one factor of
1/(2π)
3
2E for each external particle in the reaction. The factors of 2π as well
as the energies of the initialstate particles are constants that are gathered
together outside the integrals. The energies of the ﬁnalstate particles are
functions of the variables of integration. It is best to think of them as part of
the phase space deﬁned by (8.24). Needless to say, these rules are tailored
to the Feynman’s rules from Section 7.5. If you combine formulas from
diﬀerent books, you are likely to get the wrong answer.
The rules so far only hold for spinless particles. The Dirac wave functions
used for spin
1
2
scattering are normalized diﬀerently. I’ll return to that point
at the end of Chapter 10.
150 CHAPTER 8. CROSS SECTIONS AND DECAY RATES
Chapter 9
The Dirac Equation
There is a story to the eﬀect that Neils Bohr once asked Dirac what he was
working on. Dirac, who was famous for not saying much, replied, “I’m try
ing to take the square root of the KleinGordon equation.” The background
to the story is this: the KleinGordon equation is invariant under Lorentz
transformations, but it was known to have a strange pathology, it has nega
tive energy solutions. This is inevitable with a secondorder equation. The
Schrodinger equation, on the other hand, is hopeless so far as relativity is
concerned, but because it has only a ﬁrst derivative with respect to time,
it has only positiveenergy solutions. (I’m talking about freeparticle equa
tions here. Bound state problems usually have negative total energy.) Dirac
thought that he could combine the two equations in a way that would pre
serve the best features of each. The result is called the Dirac equation.
Ironically, it has negativeenergy solutions anyway, but it does provide a
formalism for treating spin 1/2 particles in a way that is consistent with
relativity.
9.1 The Equation
Let’s follow in the footsteps of our taciturn master and see what we can
come up with. Start with the Schrodinger equation.
i
∂ψ
∂t
= Hψ
Can we come up with a Hamiltonian that (1) is ﬁrst order in space deriva
tives, (2) is Hermitian, and (3) leads to a covariant equation of motion? The
151
152 CHAPTER 9. THE DIRAC EQUATION
answer is yes if ψ is a matrix. Try
i
∂ψ
∂t
= Hψ = (α p +βm)ψ (9.1)
here α and β are Hermitian matrices and p = −i∇. We will also use the
replacement E = i∂/∂t. If E is the total energy of special relativity, it must
satisfy E
2
= p
2
+m
2
. Combine this with equation (9.1)
E
2
ψ =
_
i
∂
∂t
_
2
ψ = (α· p +βm)
_
i
∂ψ
∂t
_
= (α· p +βm)
2
ψ
We would like this to equal (p
2
+m
2
)ψ. This will be possible if the matrices
have some special properties.
(α
i
p
i
+βm)
2
= β
2
m
2
+ (α
i
)
2
(p
i
)
2
+¦α
i
, β¦ mp
i
+
1
2
¦α
i
, α
j
¦
i=j
p
i
p
j
(9.2)
I have used the convention that repeated indices are summed. The curly
brackets are endemic to Diracology. They represent the anticommutator.
¦A, B¦ ≡ AB +BA
We would like (9.2) to boil down to p
i
p
i
+m
2
. This will be possible if these
matrices obey the following constraints:
β
2
= (α
i
)
2
= 1 for each i (9.3)
¦α
i
, β¦ = ¦α
i
, α
j
¦
i=j
= 0 (9.4)
We are trying to construct a theory of spin 1/2 particles, and we know
that in nonrelativistic QM Pauli spinors have two components. We expect
therefore that α and β will be 2 2 matrices. Alas, it is not to be. They
must be at least 4 4. Here is the argument:
1. βα
i
+α
i
β = 0 so βα
i
β = −α
i
β
2
= −α
i
. Take the trace of this equation.
Tr[βα
i
β] = Tr[α
i
β
2
] = Tr[α
i
] = −Tr[α
i
]. So α
i
is traceless. In the
same way we could prove that Tr[β] = 0.
2. We want α and β to be Hermitian. This together with (9.3) is suﬃcient
to prove that their eigenvalues are ±1
3. Arguments 1. and 2. lead us to conclude that the matrices must have
even dimension.
9.1. THE EQUATION 153
4. There are only three independent traceless Hermitian 2 2 matrices,
and you know what they are: the Pauli spin matrices.
σ
1
=
_
0 1
1 0
_
σ
2
=
_
0 −i
i 0
_
σ
1
=
_
1 0
0 −1
_
(9.5)
Unfortunately, we need four such matrices, so 22 won’t do. We must
have 4 4. The traditional choice is
β =
_
I 0
0 −I
_
α
i
=
_
0 σ
i
σ
i
0
_
(9.6)
All this has a nasty, nonrelativistic look about it. We therefore abandon
Dirac’s original notation for the following:
1
γ
0
= β γ
i
= βα
i
=
_
0 σ
i
−σ
i
0
_
(9.7)
Now (9.4) becomes
¦γ
µ
, γ
ν
¦ = 2g
µν
(9.8)
and (9.1) is
(iγ
µ
∂
µ
−m)ψ = 0 (9.9)
This is Dirac’s equation. Of course ψ is a fourcomponent column matrix,
the socalled Dirac spinor. g
µν
is the usual metric tensor. There is one other
convenient bit of notation.
γ
µ
B
µ
= γ
µ
B
µ
≡,B (9.10)
Dirac’s equation ﬁnally is (i,∂ − m)ψ = 0. It looks simple. It’s not. It is
however, covariant. This is far from obvious, but I will not prove it at this
time.
Wave functions are supposed to have something to do with probability.
This is true of ψ as well, but to see the connection, we need to derive
the formula for a conserved current. Starting with (9.1) and taking the
Hermitian conjugate, we have
−i
∂ψ
†
∂t
= i
∂ψ
†
∂x
i
α
i
+ψ
†
βm (9.11)
1
Some authors use diﬀerent conventions. This is consistent with Zee, Gross, and many
other texts.
154 CHAPTER 9. THE DIRAC EQUATION
Multiply on the right by β
2
= 1 and deﬁne
ψ ≡ ψ
†
β = ψ
†
γ
0
(9.12)
This object ψ is called the Dirac conjugate. It often appears in bilinear
expressions in place of the usual Hermitian conjugate. Anyhow, the resulting
equation is
−i
∂ψ
∂t
γ
0
= −i
∂ψ
∂x
i
γ
i
+ψm
or
ψ(i
←
,∂ +m) = 0 (9.13)
The arrow indicates that the derivative acts to the left. To ﬁnd the current,
multiply (9.9) on the left by ψ and (9.13) on the right by ψ and add the
two. The result is
∂
µ
ψγ
µ
ψ +ψγ
µ
∂
µ
ψ = 0 = ∂
µ
(ψγ
µ
ψ) (9.14)
So the conserved current is
j
µ
= ψγ
µ
ψ (9.15)
The current density j
0
= ψ
†
ψ is positive and has the same form [ψ[
2
as the
probability density in nonrelativistic QM.
9.2 Plane Wave Solutions
We know that electrons have two spin components. Why then are there four
components to the Dirac spinor? The answer is that despite Dirac’s best
intentions, there are still negative energy solutions. This in turn is related
to the existence of antiparticles. We will get to the particle interpretation
later. For the time being we will just reconcile ourselves to negative energies.
We start with
ψ
±
= e
∓ipx
_
χ
η
_
(9.16)
As a convention, the symbol E = p
0
will always refer to a positive number.
χ and η are twocomponent spinors. They are not independent; the Dirac
equation imposes a relationship between the two. Our job is to ﬁnd it. Start
with the upper sign (positive energy) solution. Substitute (9.16) into (9.1).
_
E −m −σ p
−σ p E +m
__
χ
η
_
= 0
9.2. PLANE WAVE SOLUTIONS 155
This will only have a solution if the determinant is equal to zero, i.e. E
2
−
m
2
−(σ p)
2
= 0. Remember that
σ
2
i
= I ¦σ
i
, σ
j
¦
i=j
= 0 (9.17)
so that
(σ p)
2
= p
2
, (9.18)
which is a useful identity to have at your disposal. We conclude that E
2
=
p
2
+ m
2
, which we knew anyhow. Unfortunately, this means that E =
±
_
p
2
+m
2
, so we are stuck with negative energy solutions. At any rate
η =
_
σ p
E +m
_
χ
We can make χ into a Pauli spinor that satisﬁes χ
†
χ = 1.
χ
(1)
=
_
1
0
_
χ
(2)
=
_
0
1
_
(9.19)
The complete positive energy solution to the Dirac equation is written
ψ
+
(x, s) = u(p, s)e
−ipx
(9.20)
where
u(p, s) = C
_
¸
¸
_
χ
(s)
σ·p
E+m
χ
(s)
_
¸
¸
_
s = 1, 2 (9.21)
Notice that in the nonrelativistic limit, the third and fourth components
become negligible.
The normalization constant C has to be chosen according to some con
vention. We will use u(p, s)u(p, s) = 1. In order to appreciate this, it is
necessary to use one of these group theory results that I have not yet proved.
It is a fact that combinations like ψψ and uu are Lorentz scalars, so that
if we choose the normalization in the rest frame of the particle, it will have
the same value in all other frames. A quick calculation shows that
u(p, s)u(p, s) = C
2
2m
E +m
,
so our normalized spinor is
u(p, s) =
_
E +m
2m
_
¸
¸
_
χ
(s)
σ·p
E+m
χ
(s)
_
¸
¸
_
s = 1, 2 (9.22)
156 CHAPTER 9. THE DIRAC EQUATION
so that in the rest frame
u(0, 1) =
_
¸
¸
_
1
0
0
0
_
¸
¸
_
u(0, 2) =
_
¸
¸
_
0
1
0
0
_
¸
¸
_
(9.23)
Now back to the negative energy solutions, which I will provisionally call
u(p, 3) and u(p, 4). Repeating the above arguments with the lower sign in
(9.16) brings us to
χ = −
_
σ p
E +m
_
η
and
u(p, s) =
_
E +m
2m
_
¸
¸
_
−
σ·p
E+m
χ
(s−2)
χ
(s−2)
_
¸
¸
_
s = 3, 4 (9.24)
It is easy to see that in the rest frame (and so everywhere else) (9.24) is
normalized so that u(p, s)u(p, s) = −1. The minus sign is forced on us. The
complete negativeenergy solution is
ψ
−
(x, s) = u(p, s)e
ipx
s = 3, 4 (9.25)
9.3 Charge Conjugation and Antiparticles
The appearance of negativeenergy states was both an embarrassment and a
disappointment; a disappointment because this wasn’t supposed to happen
with a linear wave equation and an embarrassment because of its physical
implications. Let us consider an atomic electron. An excited electron will
normally lose energy by emitting photons until it drops down to the lowest
energy or ground state. In the original version of Dirac theory there is no
lowestenergy state since there exists an continuum of negativeenergy states
from −mc
2
to −∞. The electron would fall into a bottomless pit emitting
an inﬁnite amount of energy in the process! To sidestep this diﬃculty, Dirac
proposed that under normal circumstances, all negativeenergy states are
completely ﬁlled. This agglomeration of ﬁlled states is called the Dirac sea.
Like all of our theories I suppose, it’s a half truth. Unfortunately, the other
half is very confusing. For one thing, it depends on the Pauli exclusion
principle. The states are full because no two Fermions can occupy a single
state. This is not true for integer spin particles, and the KleinGordon
9.3. CHARGE CONJUGATION AND ANTIPARTICLES 157
equation also has negative energy states. Obviously the Dirac sea can’t be
the whole answer, but it contains a surprising amount of physical insight.
2
Suppose one of the negativeenergy electrons were to absorb a photon of
energy > 2mc
2
and become an E > 0 state. As a result a “hole” is created
in the Dirac sea. The observable energy of the Dirac sea is now the energy
of the vacuum minus the negative energy of the vacated state, hence a pos
itive quantity. In this way we expect that the absence of a negativeenergy
electron appears as the presence of a positiveenergy particle. The same
argument works with charge. Let Q
0
represent the charge of the vacuum
and e the negative charge of an electron in the sea, then the total charge of
the Dirac sea is
Q = Q
0
−e = Q
0
−(−[e[) = Q
0
+[e[ (9.26)
The observable charge is then Q
obs
= Q−Q
0
= [e[. This is interpreted as a
particle with positive energy and charge. This process actually happens. In
the laboratory it looks like γ → e
−
+ e
+
. (It can’t happen in empty space
because of energy and momentum conservation. It does happen whenever
high energy photons pass through matter.) When Dirac ﬁrst proposed this
in 1930, however, no such particle was known. For this reason it was as
sumed that the theory couldn’t be right. The positron was discovered soon
thereafter.
According to this theory, the total energy of the vacuum as well as the
total charge is negative inﬁnity. We have to take it on faith that these
inﬁnities are not observable. What is observable are the deviations from
inﬁnity corresponding to the energy and charge of a single electron! This is
an unsatisfactory aspect of the theory of course, but it will continue to devil
us long after we have abandoned the Dirac sea.
The argument made above regarding charge and energy also works for
momentum and spin. The absence of momentum p in the Dirac sea appears
as the presence of momentum −p. Similarly the absence of a spinup E < 0
electron is to be interpreted as the presence of a spindown E > 0 positron.
This is all well and good, but we must face the fact that the Dirac
sea is really just a metaphor. We need to describe antiparticles in a way
that does not invoke negative energies. I once heard Edward Teller explain
antiparticles like this: “Take a particle in one hand and an antiparticle in
the other. Put them together, and what you have is – nothing”! Put less
cryptically, all the additive quantum numbers sum to zero. What are the
2
The best discussion of this entire subject is still the classic, Advanced Quantum Me
chanics, by J. J. Sakurai, AddisonWesley 1967, Sections 3.93.10
158 CHAPTER 9. THE DIRAC EQUATION
additive quantum numbers? Well, charge, baryon number, three kinds of
lepton number, and strangeness for starters. Electrons have zero baryon
number and strangeness anyhow. I will come to the issue of lepton number
presently. Let’s concentrate on the matter of charge. We can deﬁne a
transformation called “charge conjugation.” It has two eﬀects. First, it
replaces all particles with their antiparticles, and second, it changes the
wave function in such a way that it correctly describes the new situation.
It’s easy to modify the Dirac equation to include the electromagnetic
potential. The procedure is called “minimal substitution.” The rule is
everywhere you see the derivative ∂
µ
simply replace it with ∂
µ
→ D
µ
=
∂
µ
+ iqA
µ
where q is the charge of the particle. This rule is based on the
idea of gauge invariance, which we will discuss in the next chapter. For the
time being – just believe it. With this replacement the Dirac equation for
an electron with charge −e (e > 0) becomes
(i,∂ +e ,A−m)ψ = 0 (9.27)
Charge conjugation must change the charge of the electron so that Teller’s
dictum comes true. The transformed equation is
(i,∂ −e ,A−m)ψ
c
= 0. (9.28)
Here ψ
c
is the new “charge conjugated” wave function. Take the complex
conjugate of (9.28) and multiply by C
0
, a 4 4 matrix that operates on the
space of the γ’s.
C
0
(iγ
µ∗
∂
µ
+eγ
µ∗
A
µ
+m)ψ
∗
c
=
_
(C
0
γ
µ∗
C
−1
0
)(i∂
µ
+eA
µ
) +m
¸
C
0
ψ
∗
c
= 0
(9.29)
At this point we don’t know what C
0
is. The challenge is to get (9.28)
looking like (9.27). For this we will need C
0
γ
µ∗
C
−1
0
= −γ
µ
and C
0
ψ
∗
c
= ψ.
It doesn’t take much trial and error to realize that all the γ’s are pure real
except γ
2
. Try C
0
= iγ
2
. (The i is necessary to make C
0
Hermitian.)
3
You
can check that C
2
0
= I and C
0
γ
µ∗
C
−1
0
= −γ
µ
as promised. The bottom line
is
ψ
c
= C
0
ψ
∗
= iγ
2
ψ
∗
(9.30)
Let us see what this operation does to our prefab spinors. The following
identities are useful.
σ · p =
_
p
z
p
−
p
+
−p
z
_
(9.31)
3
There are several diﬀerent representations of the gamma matrices that are useful in
other contexts. Equation (9.30) is valid only in the the standard or DiracPauli represen
tation we are using.
9.3. CHARGE CONJUGATION AND ANTIPARTICLES 159
C
0
=
_
_
_
_
0 0 0 −1
0 0 1 0
0 1 0 0
−1 0 0 0
_
_
_
_
(9.32)
ψ
+
(x, 1) = u(p, 1)e
−ipx
=
_
E +m
2m
_
¸
¸
_
1
0
p
z
/(E +m)
p
+
/(E +m)
_
¸
¸
_
e
−ipx
(9.33)
ψ
+
(x, 2) = u(p, 2)e
−ipx
=
_
E +m
2m
_
¸
¸
_
0
1
p
−
/(E +m)
−p
z
/(E +m)
_
¸
¸
_
e
−ipx
(9.34)
ψ
−
(x, 3) = u(p, 3)e
ipx
=
_
E +m
2m
_
¸
¸
_
−p
z
/(+m)
−p
+
/(E +m)
1
0
_
¸
¸
_
e
+ipx
(9.35)
ψ
−
(x, 4) = u(p, 4)e
ipx
=
_
E +m
2m
_
¸
¸
_
−p
−
/(E +m)
p
z
/(E +m)
0
1
_
¸
¸
_
e
+ipx
(9.36)
Let’s ﬁnd the chargeconjugated version of ψ
+
(x, 1). Using (9.30), (9.32),
and (9.33) gives
ψ
+
c
(x, 1) =
_
_
_
_
0 0 0 −1
0 0 1 0
0 1 0 0
−1 0 0 0
_
_
_
_
_
E +m
2m
_
¸
¸
_
1
0
p
z
/(E +m)
p
+
/(E +m)
_
¸
¸
_
∗
e
+ipx
=
_
E +m
2m
_
¸
¸
_
−p
−
/(E +m)
p
z
/(E +m)
0
−1
_
¸
¸
_
e
+ipx
= −u(−p, 4)e
ipx
(9.37)
In the same way one can prove that
ψ
+
c
(x, 2) = u(−p, 3)e
ipx
(9.38)
Thus the chargeconjugate wave function obtained from the positiveenergy
planewave solution is the wave function for a negativeenergy plane wave
160 CHAPTER 9. THE DIRAC EQUATION
with the same [E[ and opposite momentum. The spin direction is also
reversed since the spinup spinor with s = 1 is changed into the spindown
spinor with s = 4. According to Dirac’s hypothesis, the observable behavior
of the chargeconjugate wave function should be a positron with E > 0 and
the same momentum and spin. This is consequence of the absence of an
electron with negative charge and energy and reversed momentum and spin.
If you buy into this bizarre notion of the Dirac sea, the rest is logically
consistent, but expectation values of the chargeconjugated wave functions
do not give the results you would like in terms of antiparticles. Try for
example, to calculate the expectation value of the charge of a positron. The
electromagnetic current should be proportional to the zeroth component of
the probability current (9.15).
Q = e
_
j
0
d
3
x = e
_
ψ
†
ψd
3
x (9.39)
The chargeconjugated wave functions give the same value for Q, since
Q
c
= e
_
ψ
†
c
ψ
c
d
3
x = e
_
(C
0
ψ
∗
)
†
(C
0
ψ
∗
)d
3
x = e
_
(ψ
†
ψ)
∗
d
3
x = Q (9.40)
where I have used the fact that C
0
is Hermitian, C
2
0
= I, and currents
are real functions. It’s trivial to show that the expectation values of the
momentum and spin do change sign. The reason is that ψ
c
is not the positron
wave function, it’s the wave function that is interpreted as the positron
wave function. Under the skin, it’s still the wave function of a ﬁctitious
negative energy particle.
4
In order to get sensible expectation values we
need to introduce second quantization with creation operators that create
real positrons. In order to do that, it is customary to deﬁne spinors for real
positrons as follows.
v(p, 1) ≡ −u(−p, 4)
v(p, 2) ≡ +u(−p, 3)
(9.41)
With this deﬁnition
u
c
(p, s) = C
0
u(p, s)
∗
= v(p, s) s = 1, 2
v
c
(p, s) = C
0
v(p, s)
∗
= u(p, s) s = 1, 2
(9.42)
u(p, 3) and u(p, 4) are properly called negativeenergy spinors, while v(p, 1)
and v(p, 2) are positron spinors.
4
Again – thanks to Sakurai for ﬁnally making me understand this point.
9.4. QUANTIZING THE FIELD 161
Here is a summary of the important equations involving u and v:
ψ
+
(x) = u(p, s)e
−ipx
(9.43)
ψ
−
(x) = v(p, s)e
+ipx
(9.44)
u(p, s)u(p, s
) = −v(p, s)v(p, s
) = δ
ss
(9.45)
u
†
(p, s)u(p, s
) = v
†
(p, s)v(p, s
) =
E
m
δ
ss
(9.46)
u(p, s)v(p, s
) = v(p, s)u(p, s
) = 0 (9.47)
u
†
(−p, s)v(p, s
) = v
†
(p, s)u(−p, s
) = 0 (9.48)
s
u(p, s)u(p, s) =
_
,p +m
2m
_
(9.49)
s
v(p, s)v(p, s) =
_
,p −m
2m
_
(9.50)
These equations will all be useful later on. You can verify them by substi
tuting the representations (9.22), (9.24), and (9.7).
9.4 Quantizing the Field
We repeat the procedure from Chapter 5 used to quantize the KlineGordon
ﬁeld. First we’ll need a Lagrangian, which can be taken as
L = ψ(x)(i ,∂ −m)ψ(x) (9.51)
The conjugate momentum is
π =
∂L
∂
˙
ψ
= iψ
†
(9.52)
and the Hamiltonian is
H =
_
d
3
x (π
˙
ψ −L) =
_
d
3
x ψiγ
0
∂
0
ψ (9.53)
Next we will need the plane wave solutions of the Dirac equation. These are
u(p, s)e
−ipx
and v(p, s)e
ipx
. We then expand ψ(x) and π(x) in a Fourier
series in which creation and annihilation operators appear as Fourier coeﬃ
cients. This should look like (6.60) with the following diﬀerences:
162 CHAPTER 9. THE DIRAC EQUATION
1. ψ is a fourcomponent spinor rather than a scalar like ϕ.
2. u(p, x) and v(p, s) depend on spin, so in addition to integrating over all
momentum components, we must also sum over the spin components.
3. We need two kinds of creation and annihilation operators, one set for
electrons and one for positrons. In the usual notation
b(p, s) Annihilates a positive energy electron
b
†
(p, s) Creates a positive energy electron
d(p, s) Annihilates a positive energy positron
d
†
(
p, s) Creates a positive energy positron
(9.54)
The analogs to (6.60) are
ψ(x) =
_
_
m
E
d
3
p
_
(2π)
3
s
[b(p, s)u(p, s)e
−ipx
+d
†
(p, s)v(p, s)e
ipx
]
ψ(x) =
_
_
m
E
d
3
p
_
(2π)
3
s
[b
†
(p, s)u(p, s)e
ipx
+d(p, s)v(p, s)e
−ipx
]
(9.55)
In analogy with (6.69) we would expect the equaltime commutation rela
tions to be
[ψ
α
(x, t), ψ
†
β
(y, t)] = δ
αβ
δ
3
(x −y)
[ψ
α
(x, t), ψ
β
(y, t)] = [ψ
†
α
(x, t), ψ
†
β
(y, t)] = 0.
Unfortunately, this leads to some wildly unphysical consequences. The ori
gin of the problem is that electrons are fermions, i.e. particles with halfodd
integer spin. Such particles obey the Fermi exclusion principle, no two par
ticles can occupy the same quantum state. Mathematically, this means that
creation and annihilation operators anticommute, and so as a consequence,
must the ﬁelds themselves. We must have
_
ψ
α
(x, t), ψ
†
β
(y, t)
_
= δ
αβ
δ
3
(x −y)
¦ψ
α
(x, t), ψ
β
(y, t)¦ =
_
ψ
†
α
(x, t), ψ
†
β
(y, t)
_
= 0.
(9.56)
Pay careful attention to the notation. ψψ
†
is a 4 4 matrix,
whereas ψ
†
ψ is a 11 matrix, so [ψ, ψ
†
] doesn’t make any sense.
The commutators in (9.56) refer to the αth component of one
ψ matrix with the βth component of another. Put it another
way, each commutator is a 4 4 matrix with rows and columns
indexed by α and β. Be alert to this ambiguity in other contexts.
9.4. QUANTIZING THE FIELD 163
You should verify that (9.55) is consistent with (9.56) assuming the following
anticommutation relations
_
b(p, s), b
†
(p
, s
)
_
= δ
ss
δ
3
(p −p
)
_
d(p, s), d
†
(p
s
)
_
= δ
ss
δ
3
(p −p
) (9.57)
All other combinations of b and d anticommute. The factor of
_
m/E in
(9.55) was chosen so that (9.56) would come out right.
There are at least three places where we have to make a some
what arbitrary choice of normalization. One is the normalization
of u and v, equations (9.22) and (9.24). The second is the above
mentioned factor in the Fourier expansion of ψ. The third is the
normalization of the creation and annihilation operators, (9.57).
Only (9.56) is nonnegotiable. It is only a slight exaggeration to
say that no two books do it the same way. At least Zee, Michio
Kaku (Quantum Field Theory, A Modern Introduction), and I
are consistent.
It’s easy to generalize charge conjugation (9.30) for the quantized ﬁelds.
Deﬁne the charge conjugation operator
ˆ
C by
ˆ
ψ
c
≡
ˆ
C
ˆ
ψ
ˆ
C
−1
= C
0
ˆ
ψ
†T
= iγ
2
ˆ
ψ
†T
(9.58)
The point is that complex conjugation is not deﬁned for the creation and
annihilation operators. The combination
†T
has the same eﬀect as
∗
on
matrices and the same eﬀect as
†
on
ˆ
b and
ˆ
d. (I have put the hats back on
the operators for emphasis. They are implied in the other operator equations
in this section.) Now using (9.29), (9.55) becomes
ψ(x) =
_
_
m
E
d
3
p
_
(2π)
3
s
[d(p, s)u(p, s)e
−ipx
+b
†
(p, s)v(p, s)e
ipx
]
ψ(x) =
_
_
m
E
d
3
p
_
(2π)
3
s
[d
†
(p, s)u(p, s)e
ipx
+b(p, s)v(p, s)e
−ipx
]
(9.59)
The only change is that particle operators have been replaced by antiparti
cles operators and vice versa, b(p, s) ↔d(p, s) and b
†
(p, s) ↔d
†
(p, s) with
no change in spin. This point will be central importance in the case of the
weak interactions.
164 CHAPTER 9. THE DIRAC EQUATION
Now insert the Fourier decomposition back into the expression for the
Hamiltonian, equation (9.53). Again – as an exercise in Diracology, you
should verify that the following is obtained.
H =
_
d
3
pE
s
[b
†
(p, s)b(p, s) −d(p, s)d
†
(p, s)] (9.60)
Here is the ﬁrst of many places where commutation relations lead to un
physical results. If d and d
†
commuted, the last term could be written
−d
†
(p, s)d(p, s), so by creating many particles with d
†
we could create states
of arbitrarily large negative energy. With the anticommutation rules we have
: H :=
_
d
3
pE
s
[b
†
(p, s)b(p, s) +d
†
(p, s)d(p, s)] (9.61)
As in (6.45), the Hamiltonian must be normal ordered to avoid the inﬁnite
constant. When dealing with fermions, it is necessary to include the minus
sign arising from the anticommutators. For example, if a, b, c, and d are
fermion annihilation operators,
: (a +b
†
)(c +d
†
) := ac −d
†
a +b
†
c +b
†
d
†
(9.62)
Take another look at the total charge calculated in (9.39)
Q = e
_
d
3
x ψ
†
(x)ψ(x)
where e is the electron charge, a negative number. The usual Dirac business
gives us
Q = e
_
d
3
p
s
[b
†
(p, s)b(p, s) +d(p, s)d
†
(p, s)] (9.63)
This operator also has to be normal ordered.
: Q := e
_
d
3
p
s
[N
−
(p, s) −N
+
(p, s)] (9.64)
Where N
+
and N
−
are the number operators for positive and negative
energy electrons respectively.
9.5. THE LORENTZ GROUP 165
9.5 The Lorentz Group
We can’t get much further with the Dirac equation without using some group
theory. Here is a brief review of the Lorentz group.
The basic equation of Lorentz transformation is
x
µ
= Λ
µ
ν
x
ν
(9.65)
Λ
µ
ν
is any 4 4 real matrix that satisﬁes
g
µν
= Λ
σ
µ
g
σγ
Λ
γ
ν
, (9.66)
or in matrix notation
g = Λ
T
gΛ. (9.67)
It is easy to show that this implies det Λ = ±1. The identity transformation
is given by Λ = I. Any transformation that can be obtained from this by
a continuous variation of the transformation parameters (β for example) is
called a proper transformation. There are three ways that a transformation
might be not proper.
1. Try Λ
0
0
= −1, Λ
i
i
= 1, and all oﬀdiagonal terms equal to zero. This
simply makes time run backwards, and so it is called the time reversal
transformation or T for short. It is easy to show from (9.66) that this
is improper.
2. Λ
0
0
= 1, Λ
i
i
= −1, and all oﬀdiagonal terms equal to zero. This
reverses the spatial coordinates. We call it a parity transformation
and refer to it as P.
3. We could have a product of the two, TP=PT.
These are examples of discrete transformations, and T
2
= P
2
= I. The most
general Lorentz transformation consists of a product of a proper transfor
mation and P, T, or PT.
5
All this is preliminary to analyzing the proper Lorentz group and its
associated Lie algebra.
6
Consider the basic Lorentz transformation
x
= γ(x −vt) y
= y t
= γ(t −vx/c
2
) (9.68)
5
Later on we will use P and T to represent more general operators that have the eﬀect
of reversing x or t.
6
I am following a particularly ﬁne exposition in Quantum Field Theory by Michio
Kaku.
166 CHAPTER 9. THE DIRAC EQUATION
If we make the standard replacement
γ =
1
_
1 −v
2
/c
2
= cosh ζ
1
βγ = sinh ζ
1
β = v/c (9.69)
then this transformation can be written as:
_
_
_
_
x
0
x
1
x
2
x
3
_
_
_
_
=
_
_
_
_
cosh ζ
1
−sinh ζ
1
0 0
−sinh ζ
1
cosh ζ
1
0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
_
_
_
x
0
x
1
x
2
x
3
_
_
_
_
(9.70)
Transformations of this sort are called boosts and the angle ζ
1
that parame
terizes the boost in (9.69) and (9.70) is called the rapidity. The 44 matrix
above represents the tensor Λ
µ
ν
in (9.65). In matrix notation x
= Λx.
According to the doctrine of Lie groups, we can calculate the group gen
erators by diﬀerentiating Λ with respect to its parameters and then setting
the parameters equal to zero.
K
1
≡
∂
∂ζ
1
Λ(ζ
1
)
¸
¸
¸
¸
ζ
1
=0
(provisional deﬁnition)
The K’s obtained in this way are constant matrices called the “generators”
of the transformation. The group can then be “reconstituted” by exponen
tiating the generators.
Λ(ζ
1
) = e
K
1
ζ
1
At this point physicists often muddy the water by inserting i’s in ways that
appal the mathematicians. We usually write the above as
K
1
≡ −i
∂
∂ζ
1
Λ(ζ
1
)
¸
¸
¸
¸
ζ
1
=0
(9.71)
and
Λ(ζ
1
) = e
iK
1
ζ
1
(9.72)
The reason for putting in the i and then taking it out will appear presently.
With this convention
K
1
= −i
_
_
_
_
0 −1 0 0
−1 0 0 0
0 0 0 0
0 0 0 0
_
_
_
_
(9.73)
9.5. THE LORENTZ GROUP 167
The generators of boosts in the y and z directions are
K
2
= i
_
_
_
_
0 0 1 0
0 0 0 0
1 0 0 0
0 0 0 0
_
_
_
_
K
3
= i
_
_
_
_
0 0 0 1
0 0 0 0
0 0 0 0
1 0 0 0
_
_
_
_
(9.74)
Rotations are also Lorentz transformations, or to put it another way, the
rotation group is a subgroup of the Lorentz group. The usual parameters
are a set of three angles. For example, a rotation through an angle θ
1
around
the 1 or x axis would yield
Λ(θ
1
) =
_
_
_
_
1 0 0 0
0 1 0 0
0 0 cos θ
1
sin θ
1
0 0 −sin θ
1
cos θ
1
_
_
_
_
(9.75)
The corresponding generators are
J
1
= i
_
_
_
_
0 0 0 0
0 0 0 0
0 0 0 1
0 0 −1 0
_
_
_
_
J
2
= i
_
_
_
_
0 0 0 0
0 0 0 1
0 0 0 0
0 −1 0 0
_
_
_
_
(9.76)
J
3
= i
_
_
_
_
0 0 0 0
0 0 −1 0
0 1 0 0
0 0 0 0
_
_
_
_
(9.77)
Notice that the J
i
’s are antisymmetric and because of the i’s, Hermitian.
The corresponding rotation matrices
Γ(θ) = e
iJ·θ
(9.78)
are unitary. This is why physicists (who are fond of unitarity) put in the i’s
and mathematicians (who would like to keep the structure constants real)
do not. The boost generators by contrast, are symmetric and boost matrices
are not unitary. The generators have the following commutation relations:
7
[K
i
, K
j
] = −i
ijk
J
k
[J
i
, J
j
] = i
ijk
J
k
[J
i
, K
j
] = i
ijk
K
k
(9.79)
7
This is also worked out in Jackson, Sec. 11.7
168 CHAPTER 9. THE DIRAC EQUATION
The matrices above constitute the deﬁning representations of the group
and the algebra. That is to say, they operate on fourvectors. We are
looking for a representation that transforms Dirac spinors, however, and
for this purpose the commutation relations (9.79) must be put in a more
convenient form. To this end we start by proving a simple theorem about
the symmetry properties of the generators.
Consider inﬁnitesimal Lorentz transformations of the form
Λ
µ
ν
= δ
µ
ν
+M
µ
ν
(9.80)
with M
µ
ν
a constant matrix, i.e. a member of the algebra, and inﬁnitesi
mal. Substitute (9.80) into (9.66) and discard the secondorder terms.
M
µν
+M
νµ
= 0 (9.81)
This is equivalent in matrix notation to the statement that M (with both
indices down) is an antisymmetric matrix. It is easy to check that the
maximum number of independent 4 4 antisymmetric matrices is six and
that the commutator of two antisymmetric matrices is itself antisymmetric.
Thus this set of matrices constitutes a closed Lie algebra, the algebra of the
Lorentz group. A particularly useful way of choosing such matrices is
(M
µν
)
αβ
= i(δ
µ
α
δ
ν
β
−δ
µ
β
δ
ν
α
) (9.82)
Just so there’s no confusion, µν is the name of the matrix and α and β index
it’s elements. It’s easy to remember. The matrix named M
µν
contains all
zeros except for two elements. The element that sits on the µ’th row and
ν’th column is i and the element at the ν’th row and µ’th column is −i.
One can derive the following commutation relations:
[M
µν
, M
ρσ
] = i(g
νρ
M
µσ
−g
µρ
M
νσ
−g
νσ
M
µρ
+g
µσ
M
νρ
) (9.83)
The M’s deﬁned by (9.82) constitute a matrix representation, but any set
of objects that satisfy the commutation relations (9.83) also constitutes a
valid representation of the Lie algebra. There are two other representations
that are important. One is the diﬀerential operator
M
µν
= i(x
µ
∂
ν
−x
ν
∂
µ
). (9.84)
The other is the spinor representation, which I will come to shortly.
Referring again to (9.82), there are twelve nonzero 4 4 antisymmetric
matrices related in a simple way to the K’s and J’s deﬁned previously. It’s
9.6. SPINOR REPRESENTATIONS 169
easy to verify that gM
01
= K
1
, gM
12
= J
3
, etc. Inﬁnitesimal transforma
tions are written
Λ
α
β
(ω) = δ
α
β
−
i
2
ω
µν
(M
µν
)
α
β
, (9.85)
where ω
µν
is an inﬁnitesimal antisymmetric tensor. We can iterate (9.85) to
build up ﬁnite transformations as usual.
Λ(ω) = exp
_
−
i
2
ω
µν
M
µν
_
(9.86)
9.6 Spinor Representations
The anticommutation relations (9.8) deﬁne what is called a Cliﬀord algebra.
(This works in any even dimension.) These algebras can always be used to
deﬁne a new sort of representation. Deﬁne
σ
µν
≡
i
2
[γ
µ
, γ
ν
] (9.87)
It is straightforward to show using (9.8) that the object
1
2
σ
µν
satisﬁes the
commutation relations (9.83) and hence constitute a representation of the
Lorentz group. The operator
S(ω) ≡ exp
_
−
i
4
ω
µν
σ
µν
_
(9.88)
is exactly analogous to (9.86). This can also be written S(Λ) to remind our
selves that given the parameters ω
µν
, we can calculate the deﬁning represen
tation Λ
µ
ν
as well as the spinor representation S. It is also straightforward
to show that
[γ
µ
,
1
2
σ
ρσ
] = (M
ρσ
)
µ
ν
γ
ν
(9.89)
where M
ρσ
is deﬁned by (9.82).
Before we apply all this to the Dirac equation, we should think about
what it means to transform a vector ﬁeld. The Dirac spinor ψ(x), like the
EM vector potential A
µ
(x), describes some multicomponent property (of
the ether, if you like) at the spacetime point x. I take the passive point of
view, that a Lorentz transformed ﬁeld describes the same phenomena at the
same spacetime point, but both the point and the components of the ﬁeld
are referred to a diﬀerent set of axis, so
S(Λ)ψ(x) = ψ
(x
). (9.90)
170 CHAPTER 9. THE DIRAC EQUATION
x
= Λx as usual, and the new ﬁeld components ψ
will be determined by
S(Λ). If the Dirac equation is to make any sense, it must remain form
invariant under this transformation, so that we must have both
(γ
µ
p
µ
−m)ψ(x) = 0 (9.91)
and
(γ
µ
p
µ
−m)ψ
(x
) = 0 (9.92)
(Here p
µ
= −i∂
µ
.) This requirement leads to an important equation for
S(Λ). Substitute (9.90) and p
µ
= Λ
ν
µ
p
ν
into (9.92).
S
−1
(Λ)(γ
µ
Λ
ν
µ
p
ν
−m)S(Λ)ψ(x) = 0
This will be identical to (9.91) if
S
−1
(Λ)γ
µ
S(Λ) = Λ
µ
ν
γ
ν
(9.93)
This is interesting. The left side is a similarity transformation. This is how
one transforms a matrix like γ
µ
. The right side treats γ
µ
like a fourvector.
So is γ
µ
a matrix like S(Λ) or a fourvector like p
µ
? The answer of course,
is yes!
Or at least we hope so. The covariance of the Dirac equation stands or
falls depending on (9.93). Here its proof: The ﬁrst step is to prove that
(9.93) is true for inﬁnitesimal transformations
(1 +
i
4
ω
ρσ
σ
ρσ
)γ
µ
(1 −
i
4
ω
ρσ
σ
ρσ
) = (γ
µ
−
i
2
ω
ρσ
[γ
µ
,
1
2
σ
ρσ
])
=(1 −
i
2
ω
ρσ
M
ρσ
)
µ
ν
γ
ν
(9.94)
The last line makes use of (9.89). The left side of (9.94) is the inﬁnitesimal
form of (9.93) by construction. A glance at (9.86) shows that the last line
of (9.94) is also the inﬁnitesimal form of the right side of (9.93). Thus
(9.93) is valid for inﬁnitesimal transformations. It can be proved for ﬁnite
transformations by iterating inﬁnitesimal transformations. It’s not a new
argument, but it’s very tedious to write out. I’ll leave it as an exercise.
We are now in a position to prove the claim made earlier that ψψ is
a Lorentz scalar. First take the Hermitian conjugate of (9.90) and right
multiply by γ
0
.
ψ
†
(x
)γ
0
= ψ
†
(x)γ
0
γ
0
S
†
(Λ)γ
0
Remember that (γ
0
)
2
= 1 and (γ
0
)
†
= γ
0
.
ψ
(x
) = ψ(x)γ
0
S(Λ)
†
γ
0
9.6. SPINOR REPRESENTATIONS 171
= ψ(x)S
−1
(Λ) (9.95)
The last line follows from (9.87) and (9.88). Evidentally, ψ
(x
)ψ
(x
) =
ψ(x)ψ(x). By the same token, ψγ
µ
ψ is a genuine vector.
ψ
(x
)γ
µ
ψ
(x
) = ψ(x)S
−1
γ
µ
Sψ = ψ(x)Λ
µ
ν
γ
ν
ψ(x) (9.96)
The last step used (9.93).
Expressions like ψψ and ψγ
µ
ψ are called Dirac bilinear forms. Notice
that unlike ψ and γ
µ
, they are not matrices but ordinary tensors. We
can also show that ψσ
µν
ψ is a secondrank antisymmetric tensor. Since
these forms are essential in describing interactions, we should enumerate
the various possibilities. We need to deﬁne a new gamma matrix:
γ
5
= γ
5
= iγ
0
γ
1
γ
2
γ
3
(9.97)
The following properties can be veriﬁed with the help of the anticommuta
tion relations.
(γ
5
)
†
= γ
5
(9.98)
(γ
5
)
2
= 1 (9.99)
_
γ
5
, γ
µ
_
= 0 (9.100)
With our conventions
γ
5
=
_
0 1
1 0
_
(9.101)
There are sixteen numbers in a 4 4 matrix, so there should be six
teen linearly independent 4 4 matrices with welldeﬁned transformation
properties. Here they are,
1 scalar 1
γ
µ
vector 4
σ
µν
tensor 6
γ
µ
γ
5
pseudovector 4
γ
5
pseudoscalar 1
for a total of sixteen. The terms pseudotensor and pseudoscalar are
“pseudo” in the sense that they change sign under parity conjugation. I
will return to this important subject later. One can show that these sixteen
matrices are indeed linearly independent. Any product of gamma matrices
can be reduced to a linear combination of these terms.
172 CHAPTER 9. THE DIRAC EQUATION
9.7 The Dirac Propagator
Consider a physical process in which an electron is created out of the vacuum
at the point x and is subsequently reabsorbed at x
. This emission and
reabsorption might take place because of sources, which we discussed in
connection with path integrals, or it might come about because the electron
interacted with other particles at x and x
. In any case, the amplitude
for this process is called the propagator. It is the basic ingredient in all
perturbation theory calculations. I have argued before that the ﬁeld ψ(x)
creates an electron at x and ψ destroys it, so the product ¸0[ψ(x
)ψ(x)[0)
must have something to do with the propagator. This can’t be the whole
story, however, because we cannot annihilate the electron before it has been
created. True, simultaneity is relative, but it is logically impossible in any
reference frame to make something disappear that doesn’t exist.
8
Therefore,
we should write out propagator as ¸0[ψ(x
)ψ(x)[0)θ(t
−t). This looks non
covariant, because or the θ(t
−t). The argument can be made in a completely
covariant way, but this is just pedantry. We get a covariant result in the
end anyway.
This is still not the complete propagator, because there is a distinct pro
cess that is physically equivalent. ψ doesn’t just create electrons, it reduces
the charge. So ¸0[ψ(x
)ψ(x)[0) also include the process in which negative
energy electrons travel backwards in time from x to x
, or equivalently,
positrons travel from x
to x. We must still insist that the positron be
created before it is destroyed, so ﬁnally
iS
F
(x
, x) = ¸0[ψ(x
)ψ(x)[0)θ(t
−t) −¸0[ψ(x)ψ(x
)[0)θ(t −t
) (9.102)
So why subtract the two terms rather than adding them? The glib answer
is that the spinor ﬁelds anticommute. We will get more insight into this
presently. Just as we did in the case of the KlineGordon equation, eq.
(5.33), we introduce the time ordering operator
T[ψ(x
)ψ(x)] =
_
ψ(x
)ψ(x) t
> t
−ψ(x)ψ(x
) t > t
(9.103)
Finally
i(S
F
)
αβ
(x
, x) = ¸0[T[ψ
α
(x
)ψ
β
(x)][0) (9.104)
S
F
is a 44 matrix indexed with α and β. It’s easy to calculate. Substitute
(9.55) into (9.102) and use (9.49) and (9.50) to do the spin sums. Zee
8
“Yesterday upon a stair, / I met a man who wasn’t there. / He wasn’t there again
today. / I wish that man would go away.”
9.7. THE DIRAC PROPAGATOR 173
goes through this calculation in detail on pages 108109. He arrives at the
following:
iS(x) = i
_
d
4
p
(2π)
4
e
−ipx
,p +m
p
2
−m
2
+i
(9.105)
It turns out that (,p −m)(,p +m) = p
2
−m
2
. We could say that the factors
of (,p + m) in the numerator and denominator “cancel” leaving the rather
whimsical equation,
9
iS(x) =
_
d
4
p
(2π)
4
e
−ipx
i
,p −m+i
. (9.106)
In momentum space this is
iS(p) =
i
,p −m+i
(9.107)
Causality requires that the propagator vanishes outside the light cone.
You can show that (9.105) does have that property. It also reproduces the
minus sign in (9.102). It can also be shown
10
that without that minus sign,
the propagator violates causality. This is an illustration of a deep result,
called the spinstatistics theorem. It says that in order to preserve causality,
Fermion ﬁelds must anticommute, and boson ﬁelds must commute.
At this point in our treatment of the scalar ﬁeld, we added some in
teractions and did perturbation theory, but electrons don’t interact with
themselves like our hypothetical scalar particle. They do interact with the
electromagnetic vector potential and with the potential that carries the weak
interactions. (The’re the same ﬁeld actually, but that’s a story for another
time.) Our next order of business is the electromagnetic ﬁeld, to which we
turn in the next chapter.
9
Your instructor will become homicidal if you divide by a matrix under any other
circumstances.
10
Peskin and Schroeder, An Introduction to Quantum Field Theory page 56
174 CHAPTER 9. THE DIRAC EQUATION
Chapter 10
The Photon Field
10.1 Maxwell’s Equations
In some sense Maxwell discovered both quantum electrodynamics and rela
tivity. I mean by this odd remark that Maxwell’s equations are consistent
with both. Nothing about them needs to be changed. Here they are:
∇ E = ρ
∇B −
∂E
∂t
= j
∇ B = 0
∇E +
∂B
∂t
= 0
(10.1)
You may never have seen them in this simple form. It arises because: (1) We
use HeavisideLorentz units. (See the appendix in Jackson regarding units.)
In this system, all the factors of 4π are absent from Maxwell’s equation.
(They retaliate by showing up in Coulomb’s law, but we don’t care about
that.) There are no dielectrics or magnetic materials, so = µ = 1. (3) We
set c = 1 as usual. The electromagnetic current is conserved.
∂ρ
∂t
+∇ j = 0 (10.2)
The vector ﬁelds can be derived from potentials.
E = −∇A
0
−
∂A
∂t
B = ∇A (10.3)
175
176 CHAPTER 10. THE PHOTON FIELD
This can all be put in covariant notation by deﬁning the following tensors:
A
µ
= (A
0
, A)
j
µ
= (ρ, j)
F
µν
= ∂
µ
A
ν
−∂
ν
A
µ
(10.4)
Then current conservation is just
∂
µ
j
µ
= 0 (10.5)
and Maxwell’s equations can be summarized as:
∂
µ
F
µν
=j
ν
∂
µ
F
νρ
+∂
ν
F
ρµ
+∂
ρ
F
µν
= 0
(10.6)
The fact that these can be written in covariant form indicates that relativity
is “built in” to Maxwell’s equations.
We need a Lagrangian to do ﬁeld theory, Jackson has some obscure things
to say about Lagrangians in general, but if you know the answer ahead of
time, it’s really very easy.
L = −
1
4
F
µν
F
µν
−j
µ
A
µ
(10.7)
To show that this does indeed generate the right equations of motion, sim
plify the expression as follows:
−
1
4
F
µν
F
µν
= −
1
4
(∂
µ
A
ν
−∂
ν
A
µ
)(∂
µ
A
ν
−∂
ν
A
µ
)
= −
1
2
[(∂
µ
A
ν
)(∂
µ
A
ν
) −(∂
µ
A
ν
)(∂
ν
A
µ
)]
∂L
∂(∂
µ
A
ν
)
= −∂
µ
A
ν
+∂
ν
A
µ
= −F
µν
(10.8)
The EulerLagrange equations are
∂
µ
_
∂L
∂(∂
µ
A
ν
)
_
−
∂L
∂A
ν
= −∂
µ
F
µν
+j
ν
= 0,
which is exactly (10.6).
The potential contains some unphysical information in the sense that
Maxwell’s equations are invariant under gauge transformations. The familiar
form for gauge transformation is
A →A+∇Λ
10.1. MAXWELL’S EQUATIONS 177
φ →φ −
∂Λ
∂t
where Λ(x) is any arbitrary function of space and time. We can write this
in covariant form as follows.
A
µ
(x) →A
µ
(x) −∂
µ
Λ(x) (10.9)
This leaves the ﬁeld tensor invariant F
µν
(x) →F
µν
(x), and hence preserves
Maxwell’s equations. We can use this symmetry to make A
µ
satisfy certain
relations.
1
For example, we can choose Λ so that A
0
(x) = 0. Imposing
constraints like this is called “gauge ﬁxing.” This particular constraint de
ﬁnes what is called temporal gauge. It is possible to impose an additional
constraint
∇ A = 0 (10.10)
This is known as Coulomb gauge or radiation gauge. Another choice is
Lorentz gauge.
∂
µ
A
µ
= 0 (10.11)
This has the advantage of being covariant, but it does not fully ﬁx the
gauge. This poses a dilemma: we can’t fully ﬁx the gauge without making
the equations noncovariant. We can’t make the theory covariant without
building into the theory unphysical degrees of freedom known as Fadeev
Popov ghosts.
2
The subject is so diﬃcult and convoluted that one distin
guished text
3
simply refuses to discuss it!
Components of A
µ
that can be set equal to zero by a gauge transfor
mation cannot have any physical signiﬁcance. It’s a wellknown result from
classical ﬁeld theory that any vector can be decomposed into a longitudinal
and a transverse part.
A = A
T
+A
L
where by deﬁnition
∇A
L
= 0 ∇ A
T
= 0
Furthermore, the decomposition is unique up to an additive constant. Since
F
ij
is basically the curl of A, it cannot depend on the longitudinal compo
nents. Only the transverse components are meaningful. There are only two
1
This subject is discussed extensively in Jackson and our Ph632 class. Here I am
quoting the main results without proof.
2
Where is Buﬀy now that we need her?
3
Peskin and Schroder
178 CHAPTER 10. THE PHOTON FIELD
physical degrees of freedom, which correspond to the two polarization states
of an electromagnetic wave.
Perhaps the easiest way out of this mess is to quantize in Coulomb gauge,
though even this is not straightforward. This is the approach taken in the
previous generation of texts.
4
Modern texts do this with path integrals. It’s
ironic that the photon propagator, which is all you need to do perturbation
theory in QED, is very simple. This is my excuse for spending minimal
eﬀort on this vexatious subject.
10.2 Quantization in the Coulomb Gauge
We have two goals.
5
First, to ﬁnd an expansion of the photon ﬁeld in
terms of creation and annihilation operators, and second, ﬁnd the photon
propagator. The Lagrangian density for the free ﬁelds is
L(x) = −
1
4
F
µν
F
µν
=
1
2
(E
2
−B
2
). (10.12)
The conjugate momenta to A
µ
are,
π
0
(x) =
∂L
∂(∂
0
A
0
)
= 0
π
i
(x) =
∂L
∂(∂
0
A
i
)
= −E
i
(10.13)
and the Hamiltonian density is
H = π
µ
˙
A
µ
−L =
1
2
(E
2
+B
2
) +E ∇A
0
(10.14)
The next step is to impose the equaltime commutation relations. (Compare
with (73) in the previous chapter.)
[A
α
(x, t), π
β
(y, t)] = iδ
αβ
δ(x −y) (10.15)
[A
α
(x, t), A
β
(y, t)] = [π
α
(x, t), π
β
(y, t)] = 0
We are missing one relation, since π
0
= 0. This is an invitation to duck into
coulomb gauge and set A
0
= ∇ A = 0. We are not out of the woods yet,
4
J. D. Bjorken and S. D. Drell, Relativistic Quantum Mechanics and Relativistic Quan
tum Fields is the Gold Standard.
5
I am following B. Hatﬁeld, Quantum Field Theory of Point Particles and Strings.
10.2. QUANTIZATION IN THE COULOMB GAUGE 179
however. Eq. (10.3) gives ∇ E = 0. This is Gauss’s law; so far so good. If
we take the divergence of (10.15)
[A
i
(x, t), E
j
(y, t)] = −iδ
ij
δ(x −y), (10.16)
we get the odd result
0 = −i∂
i
δ(x −y)
This is time for some creative thinking. Suppose we modify (10.15) slightly
so that we avoid this contradiction. Lets replace the δfunction with a
function f(x−y) such that ∂
i
f(x−y) = 0. The usual choice is the transverse
δfunction,
δ
tr
ij
(x −y) ≡
_
d
4
k
(2π)
3
e
ik·(x−y)
_
δ
ij
−
k
i
k
j
k
2
_
. (10.17)
δ
tr
ij
(x) has the desired property: ∂
i
δ
tr
ij
(x) = 0. We modify (16) to read
[A
i
(x, t), E
j
(y, t)] = −iδ
tr
ij
δ(x −y), (10.18)
The following comments are in order:
• δ
tr
ij
(x) ,= 0 so A
i
and E
i
don’t commute at spacelike separations. This
in itself is OK since A is not measurable.
• Electric and magnetic ﬁelds calculated using (10.18) do commute at
spacelike separations, and in fact, we get the same commutators for E
and B regardless whether we use the transverse or ordinary δfunction
in (10.18).
• The underlying idea here is that there are only two degrees of freedom,
but (10.16) tries to quantize three. The transverse δfunction remedies
this problem.
Now we are home safe. It remains only to expand A in plane waves and
calculate the propagator.
A(x) =
_
d
3
k
_
(2π)
3
2k
0
2
λ=1
ε
λ
(k)
_
a
λ
(k)e
−ikx
+a
λ†
(k)e
ikx
_
(10.19)
Since there is no antiphoton (or more correctly, the photon is its own antipar
ticle) there is only one set of creation and destruction operators. Since we
are working in transverse gauge, A must be perpendicular to the direction
180 CHAPTER 10. THE PHOTON FIELD
of propagation. This is accomplished by deﬁning the photon polarization
vectors ε
λ
such that
ε
λ
k = 0 (10.20)
We also require that they be orthonormal.
ε
λ
(k) ε
λ
(k) = δ
λλ
(10.21)
Equation (10.18) will be satisﬁed if
[a
λ
(k), a
λ†
(k
)] = δ
λλ
δ
(3)
(k −k
) (10.22)
Our last mission is to calculate the propagator. The process should be
quite familiar by now.
iD
F
(x
−x)
µν
= ¸0[T[A
µ
(x
)A
ν
(x)][0)
= i
_
d
4
k
(2π)
4
e
−ik(x
−x)
k
2
+i
2
λ=1
ε
λ
µ
(k)ε
λ
ν
(k) (10.23)
The polarization tensors are not in a convenient form, and since they refer
to a particular frame of reference, they are not covariant. That is the price
of working in Coulomb gauge. Here is an argument that addresses both
issues. We construct a set of orthonormal vectors based on the coordinate
system in which the polarization vectors are deﬁned. In this frame they
are perpendicular to k, and since they have no time component ε
µ
k
µ
= 0
(regardless of the time component of k). We introduce a timelike unit vector
η = ( 1 0 0 0 ), η
µ
ε
µ
= 0, but η
µ
k
µ
,= 0. We complete our set of four
unit vectors by projecting out the part of k
µ
along η
µ
.
ˆ
k
µ
=
k
µ
−(k η)η
µ
_
(k η)
2
−k
2
(10.24)
The four unit vectors then are η, ε
1
, ε
2
, and
ˆ
k. If you think of them as
column matrices,
ηη
†
−ε
1
ε
1†
−ε
2
ε
2†
−
ˆ
k
ˆ
k
†
= g,
where g is the usual metric tensor. Writing this as a tensor equation gives,
2
λ=1
ε
λ
µ
(k)ε
λ
ν
(k) = −g
µν
+η
µ
η
ν
−
ˆ
k
µ
ˆ
k
ν
. (10.25)
The usual fourvector argument goes like this: equation (10.25) is valid in
one particular reference frame. g
µν
and
ˆ
k
µ
ˆ
k
ν
are legitimate fourtensors. If
10.2. QUANTIZATION IN THE COULOMB GAUGE 181
η
µ
η
ν
were also, then (24) would be valid in all reference frames. Unfortu
nately, it isn’t, so it’s not. Let’s go ahead and substitute (10.24) into (10.25)
anyhow.
2
λ=1
ε
λ
µ
(k)ε
λ
ν
(k) = −g
µν
−
k
µ
k
ν
(k η)
2
−k
2
+
(k η)(k
µ
η
ν
+k
ν
η
µ
)
(k η)
2
−k
2
−
k
2
η
µ
η
ν
(k η)
2
−k
2
(10.26)
This is not covariant, but when the photon propagator is used in a actual
calculation it is always coupled to a conserved current. Eq. (10.5) written
in momentum space is k
µ
j
µ
= 0. All the terms above with uncontracted k’s
vanish! The last term is something of an embarrassment. When substituted
into (10.23) it gives the term
−η
µ
η
ν
δ(t −t
)
4π[x −x
[
This is the instantaneous Coulomb interaction. It is a well known artifact
of Coulomb gauge in classical E&M.
6
It can’t be right, of course. Even the
Coulomb interaction doesn’t propagate instantaneously through space. In
this case it is cancelled by another spurious term in the Hamiltonian.
7
The
remaining term in (10.26) gives the correct propagator,
D
F
(x
−x)
µν
= g
µν
_
d
4
k
(2π)
4
e
−ik(x
−x)
k
2
+i
. (10.27)
In momentum space, this is simply,
D(k) =
g
µν
k
2
+i
(10.28)
This argument guarantees that so long as we use (10.27) or (10.28) in our
calculations, the resulting S matrix elements and Green’s functions will have
the correct covariant behavior, even though the quantization was done in
Coulomb gauge.
6
See Jackson, Sec. 6.5 and references therein.
7
See M. Kaku, Quantum Field Theory, Section 4.3
182 CHAPTER 10. THE PHOTON FIELD
Chapter 11
Quantum Electrodynamics
11.1 Gauge Invariance
Before we can do any realistic calculations, we must ﬁgure out how the
electromagnetic ﬁeld couples to electrons. We could guess this as follows:
we know that the classical electromagnetic Lagrangian has the term −j
µ
A
µ
,
so we assume that the interaction Hamiltonian must be of the form j
µ
A
µ
.
Furthermore, the electron probability current is j
µ
=
¯
ψγ
µ
ψ, so the charge
current must be e
¯
ψγ
µ
ψ. We conclude that H
int
= e
¯
ψγ
µ
ψA
µ
. This is true,
but there is a profound principle here that we should linger over.
The freeparticle Dirac Lagrangian is L =
¯
ψ(i ,∂ − m)ψ. If we make
the transformation ψ → e
iqθ
ψ, then L → L. This is called a “global gauge
transformation.” It’s not what we usually think of as a gauge transforma
tion in E&M, but they are related as you will see. Try an inﬁnitesimal
transformation δψ = iqδθψ. If the Lagrangian is to remain invariant under
this transformation we must have
δL = 0 =
δL
δψ
δψ +
δL
δ(∂
µ
ψ)
δ(∂
µ
ψ) (11.1)
= iqδθ
_
δL
δψ
ψ +
δL
δ(∂
µ
ψ)
∂
µ
ψ
_
Here θ is an inﬁnitesimal and q is something like a charge. The Euler
Lagrange equation of motion is
δL
δψ
= ∂
µ
δL
δ(∂
µ
ψ)
(11.2)
183
184 CHAPTER 11. QUANTUM ELECTRODYNAMICS
Substituting (11.2) into (11.1) gives
iqδθ∂
µ
_
δL
δ(∂
µ
ψ)
ψ
_
= 0 (11.3)
Evidentally, the quantity in brackets is a conserved current.
j
µ
∝
δL
δ(∂
µ
ψ)
ψ (11.4)
and ∂
µ
j
µ
= 0. Applying this formula to the Dirac Lagrangian gives j
µ
=
¯
ψγ
µ
ψ, which we knew before.
There is a stronger statement to be made if we make a diﬀerent trans
formation at each point in space, i.e. if we make a local transformation.
ψ
= e
iqθ(x)
ψ. (11.5)
The function θ(x) is arbitrary. Now
∂
µ
ψ
= e
iqθ
∂
µ
ψ +iq(∂
µ
θ)e
iqθ
ψ
Now L is not invariant by itself, but it can be made invariant by incor
porating an additional ﬁeld with special transformation properties. Let’s
insist that we replace ∂
µ
ψ everywhere with (∂
µ
−ieqA
µ
)ψ. If the following
statement is true, the electron Lagrangian will be invariant:
(∂
µ
−ieqA
µ
)ψ
= e
iqθ
(∂
µ
−ieqA
µ
)ψ,
but
(∂
µ
−ieqA
µ
)ψ
= e
iqθ
∂
µ
ψ +iq(∂
µ
θ)e
iqθ
ψ −ieqA
µ
e
iqθ
ψ
We conclude that
A
µ
=
1
e
∂
µ
θ +A
µ
. (11.6)
In other words, the electron Lagrangian will be invariant under the trans
formation (11.5) if our new ﬁelds transform according to (11.6), but these
new ﬁelds must also appear in our Lagrangian, and this Lagrangian must
be invariant under the transformation
A
µ
→A
µ
+
1
e
∂
µ
θ(x). (11.7)
Equation (11.7) is of course, the usual statement of gauge invariance in clas
sical E&M, but now we see it in a larger context; quantum gauge invariance,
(11.5), requires the existence on an additional vector ﬁeld.
11.2. NOETHER’S THEOREM 185
We know that the Lagrangian for classical E&M (excluding sources) is
L =
1
4
F
µν
F
µν
where
F
µν
= ∂
µ
A
ν
−∂
ν
A
µ
Transformation (11.7) does leave this invariant, but if this were the La
grangian for a massive ﬁeld, it would contain an additional term
m
2
A
†µ
A
µ
which would not be invariant. We conclude that local gauge invariance
requires a massless vector ﬁeld.
You will notice that the “charge” q cancels in the above calculation. As a
consequence the sign and magnitude of e are undetermined. As a convention
we usually write
L =
¯
ψ(iγ
µ
∂
µ
−eγ
µ
A
µ
−m)ψ (11.8)
so that
H
int
= −L
int
= e
¯
ψγ
µ
ψ = j
µ
A
µ
(11.9)
11.2 Noether’s Theorem
Zee calls Noether’s theorem, ”one of the most profound observations in theo
retical physics.” The theorem simply states that every continuous symmetry
transformation is connected with a conserved current and hence, with some
sort of conserved “charge.” The converse is also true, for every conserved
current, there is a corresponding symmetry group. The proof is a simple
generalization to that given above. Given the Lagrangian density composed
of N ﬁelds ϕ
i
and their derivatives, L = L(ϕ
i
, ∂
µ
ϕ
i
). Consider an inﬁnitesi
mal change, δϕ
i
, brought about by some small parameter or parameters δ
j
,
and calculate the action.
δS =
N
i=1
_
d
4
x
_
δL
δϕ
i
δϕ
i
+
δL
δ(∂
µ
ϕ
i
)
δ(∂
µ
ϕ
i
)
_
=
N
i=1
_
d
4
x
_
δL
δϕ
i
δϕ
i
+
δL
δ(∂
µ
ϕ
i
)
∂
µ
δϕ
i
_
(11.10)
186 CHAPTER 11. QUANTUM ELECTRODYNAMICS
(The repeated µ indices are summed.) We can use the N EL equations of
motion
0 = ∂
µ
_
δL
δ(∂
µ
ϕ
i
)
_
−
δL
δϕ
i
to combine the two terms in (11.10).
δS =
N
i=1
_
d
4
x∂
µ
_
δL
δ(∂
µ
ϕ
i
)
∂ϕ
i
∂
j
δ
j
_
(11.11)
If the action is unchanged by this variation, we can deﬁne a current
J
µ
j
=
δL
δ(∂
µ
ϕ
i
)
∂ϕ
i
∂
j
(11.12)
Equation (11.11) says that ∂
µ
J
µ
j
= 0. From this conserved current, we can
also create a conserved charge by integrating over the time component of
the current:
Q
j
≡
_
d
3
x J
0
j
(11.13)
Now integrate the conservation equation:
0 =
_
d
3
x ∂
µ
J
µ
j
=
_
d
3
x ∂
0
J
0
j
+
_
d
3
x ∇ J
j
=
d
dt
Q
j
+
_
s
dS J
j
In the last term I have used the divergence theorem to convert the volume
integral into a surface integral. Assume that the ﬁelds vanish at inﬁnity so
the last term can be neglected. Then:
d
dt
Q
j
(t) = 0 (11.14)
In summary: a continuous symmetry implies a conserved current, which
in turn implies a conserved charge. It’s important to understand that the
“charge” so deﬁned usually has no connection with electrical charge. Most
of the conserved quantities in particle physics, such as strangeness, charm,
lepton number, etc. come about in this way. Many theories in particle
physics came about because the conserved quantities were ﬁrst discovered
experimentally. Theorists then reasoned “backwards” from this discovery to
construct the underlying conserved currents.
11.3. FEYNMAN’S RULES FOR QED 187
11.3 Feynman’s Rules for QED
The derivation of Feynman’s rules for electrons and photons proceeds along
the same lines as the derivation for scalar particles in Chapter 7. Because
fermion ﬁelds anticommute, there are some additional minus signs to keep
track of. The timeordered product was deﬁned in (9.103)
T[ψ
α
(x
1
)ψ
β
(x
2
)] = ψ
α
(x
1
)ψ
β
(x
2
)θ(x
0
1
−x
0
2
) −ψ(x
2
)
β
ψ(x
1
)
α
θ(x
0
2
−x
0
1
)
(11.15)
Decompose the ﬁeld into positive and negativeenergy parts. ψ = ψ
+
+ψ
−
.
Remember that ψ
+
contains the electron annihilation operator
ˆ
b
k
and ψ
+
contains the positron creation operator
ˆ
d
†
k
. Rewrite (11.15) in such a way
that all creation operators stand to the right of all annihilation operators.
T[ψ
α
(x
1
)ψ
β
(x
2
)]
= ¦ψ
+
α
(x
1
), ψ
+
β
(x
2
)¦θ(x
0
1
−x
0
2
) −¦ψ
−
β
(x
2
), ψ
−
α
(x
1
)¦θ(x
0
2
−x
0
1
)
−ψ
+
β
(x
2
)ψ
+
α
(x
2
) +ψ
+
α
(x
1
)ψ
−
β
(x
2
) +ψ
−
α
(x
1
)ψ
+
β
(x
2
) +ψ
−
α
(x
1
)ψ
−
β
(x
2
)
(11.16)
You see that the second line of this equation is a c number and the third, a q
number. We call the cnumber part the “contraction” of the two ﬁelds. The
qnumber part is “normal ordered” in the sense that all creation operators
stand to the left of all annihilation operators, but because of the additional
minus sign we must amend the deﬁnition of normal ordering somewhat.
Deﬁnition: The ﬁelds : ψ(x
1
)ψ(x
2
) ψ(x
n
) : are normal ordered if in
each term, all creation operators stand to the left of all annihilation opera
tors. Each term is positive unless it required on odd number of permutations
to bring the various creation and annihilation operators into that order.
In a way analogous to (7.67), we introduce the notion of the contraction
of two ﬁelds.
¸ .. ¸
ψ
α
(x
1
)ψ
β
(x
2
) ≡ T[ψ
α
(x
1
)ψ
β
(x
2
)]− : ψ
α
(x
1
)ψ
β
(x
2
) : (11.17)
Since the vacuum expectation value of normalordered ﬁelds vanishes, we
have
¸ .. ¸
ψ
α
(x
1
)ψ
β
(x
2
) = ¸0[T[ψ
α
(x
1
)ψ
β
(x
2
)][0) = i(S
F
)
αβ
(x
1
−x
2
), (11.18)
where I have borrowed the deﬁnition of the propagator from (9.104). It’s
clear that
¸ .. ¸
ψ(x
1
)ψ(x
2
) =
¸ .. ¸
ψ(x
1
)ψ(x
2
) = 0 (11.19)
188 CHAPTER 11. QUANTUM ELECTRODYNAMICS
An additional complication arises when there are more than two ﬁelds
to time order. In this case we must include a minus sign for each operator
interchange. For example, (compare with (7.68))
T[ψ(x
1
)ψ(x
2
)ψ(x
3
)ψ(x
4
)] =: ψ(x
1
)ψ(x
2
)ψ(x
3
)ψ(x
4
) :
−
¸ .. ¸
ψ(x
1
)ψ(x
3
) : ψ(x
2
)ψ(x
4
) : +
¸ .. ¸
ψ(x
1
)ψ(x
4
) : ψ(x
2
)ψ(x
3
) :
+
¸ .. ¸
ψ(x
2
)ψ(x
3
) : ψ(x
1
)ψ(x
4
) : −
¸ .. ¸
ψ(x
2
)ψ(x
4
) : ψ(x
1
)ψ(x
3
) :
−
¸ .. ¸
ψ(x
1
)ψ(x
3
)
¸ .. ¸
ψ(x
2
)ψ(x
4
) +
¸ .. ¸
ψ(x
1
)ψ(x
4
)
¸ .. ¸
ψ(x
2
)ψ(x
3
)
(11.20)
Wick’s theorem enables us to calculate the vacuum expectation value of
any product of boson and fermion ﬁelds. The next step toward calculating
the S matrix requires a fermion version of the LSZ reduction formulas de
veloped in Section 7.3. The derivation is similar to that leading to (7.43).
There are no new ideas, just a lot of niggling details regarding minus signs.
I will simply state the results and refer the reader to the classic exposition
in Bjorken and Drell.
1
In analogy with (7.30) we deﬁne the fermion inﬁeld
lim
t→−∞
ψ(x) =
_
Z
2
ψ
in
(x) (11.21)
The constant Z
2
is called the electron wave function renormalization con
stant. There are three such constants in QED. One for the electron wave
function, one for the photon, and one for the electric charge. If it were not for
the fact that all Feynman diagrams with closed loops give divergent results,
we would be able to calculate them. As it is, we must be sure that they do
not appear in any ﬁnal results. As it turns out, the photon renormalization
constant can be set equal to unity and the other two can be gathered into
the deﬁnition of the physical mass and charge of the electron. This is to
say that QED is renormalizable. One of the key ideas to come out of this
is that renormalizability is a very special property possessed by only a few
theories, and nonrenormalizable theories are wrong!
Deﬁne the planewave solutions of Dirac’s equation as follows.
U
ks
(x) =
_
m
(2π)
3
E
k
u(k, s)e
−ikx
V
ks
(x) =
_
m
(2π)
3
E
k
v(k, s)e
ikx
(11.22)
1
Relativistic Quantum Fields, J. D. Bjorken and S. D. Drell, McGrawHill 1965, Section
16.9
11.3. FEYNMAN’S RULES FOR QED 189
In this notation the secondquantized Dirac ﬁeld is
ψ
in
(x) =
_
d
3
k
s
[b
in
(k, s)U
ks
(x) +d
†
in
(k, s)V
ks
] (11.23)
There are several formulas analogous to (7.41) depending on whether we
remove a particle or antiparticle from the instate or outstate. If we remove
a particle from the instate (refer to Section 7.3 for the signiﬁcation of the
notation)
S
βα
= ¸β out[α in)
∼ −
i
√
Z
2
_
d
4
x ¸β out[ψ(x)[α −k in)
←−−−−−−−
(−i,∂ −m)U
ks
(x)
(11.24)
Removing an antiparticle from the instate leads to
∼
i
√
Z
2
_
d
4
x V
ks
−−−−−−→
(i,∂ −m)¸β out[ψ(x)[α −k in) (11.25)
Removing a particle from the outstate leads to
∼ −
i
√
Z
2
_
d
4
x U
ks
−−−−−−→
(i,∂ −m)¸β −k out[ψ(x)[α in)U
ks
(x) (11.26)
Removing an antiparticle from the outstate gives
∼
i
√
Z
2
_
d
4
x ¸β −k out[ψ(x)[α in)
←−−−−−−−
(−i,∂ −m)V
ks
(x) (11.27)
After two or more particles have been “reduced,” the ﬁelds inside the braket
combination are time ordered.
Finally, we need to be able to reduce photon ﬁelds. In this case the
asymptotic condition is
lim
t→−∞
A(x, t) =
_
Z
3
A
in
(x, t) (11.28)
The constant Z
3
will turn out to renormalize the electron charge. The wave
function is
A
kλ
=
1
_
(2π)
3
2ω
e
−ikx
(k, λ) (11.29)
so that the Fourier expansion of A
in
(x) is
A
in
=
_
d
3
k
λ
[a
in
(k, λ)A
kλ
(x) +a
†
in
(k, λ)A
∗
kλ
(x)] (11.30)
190 CHAPTER 11. QUANTUM ELECTRODYNAMICS
The reduction formulas are virtually identical to (7.43) (with m = 0 of
course) except for one minus sign.
S
βα
= ¸β out[α in)
∼
−i
√
Z
3
_
d
4
x ¸β out[A
µ
(x)[α −(k, λ))
←−
2A
µ∗
kλ
(x)
(11.31)
The additional minus sign in (11.31) comes from the spacelike nature of the
polarization unit vector
µ
µ
= − · = −1
This would be a good point to summarize the results so far. According
to (8.5) the Smatrix can be rewritten
S
fi
= δ
fi
−i(2π)
4
δ
(4)
(P
f
−P −i)T
fi
(11.32)
The δ
fi
stands for all those terms appearing in the reduction procedure in
which the two initial particles don’t scatter.
2
The (2π)
4
δ(P
f
−P
i
) appears
when we do the last integral over coordinate space. The interesting part is
T
fi
, which is the crucial ingredient in (8.20) and (8.21) to calculate cross
sections and decay rates. To calculate the Smatrix, we start with the
basic QED Hamiltonian (11.9) and substitute it into the formula for the
Soperator (7.69). This is an inﬁnite series. We keep only those terms
corresponding to the desired degree of approximation. These terms are
substituted into (??)
S
βα
= ¸β out[S[α in). (11.33)
In this context, α stands for the speciﬁc initial state you have set up in your
experiment, and β represents the particular ﬁnal state whose probability
you wish to calculate. We then use (11.24), (11.25), (11.26), (11.27), and
(11.31) to reduce out the various electron and photon ﬁelds. Finally, do all
the implied diﬀerentiation and integration.
This is a “cookbook” procedure. No problemsolving ability is required.
You should be able to do it while in a coma. You have to have patience
though, even a modest calculation can involve, implicitly at least, tens of
thousands of terms. After you have done a few such calculations, you will
realize that you are doing the same few things over and over. These things
can be boiled down to a few simple rules known as Feynman’s rules.
2
In a real experiment, the vast majority of incident beam particles don’t scatter. All
this business about calculating cross sections has to do with the inﬁnitesimal fraction that
does.
11.3. FEYNMAN’S RULES FOR QED 191
1. Draw all possible connected, topologically distinct diagrams, including
loops, with n external legs. Ignore vacuumtovacuum graphs. Each
vertex must have a continuous electron or positron line and a photon
line terminating on the fermion.
2. There is a factor given by the following rules for each external line.
Incoming fermion:
p
=
u(p, s)
_
Z
2
(2π)
3
2E
p
Incoming antifermion:
p
=
v(p, s)
_
Z
2
(2π)
3
2E
p
Outgoing fermion:
p
=
u(p, s)
_
Z
2
(2π)
3
2E
p
Outgoing antifermion:
p
=
v(p, s)
_
Z
2
(2π)
3
2E
p
Incoming photon:
p
=
µ
(p)
_
Z
3
(2π)
3
2E
p
192 CHAPTER 11. QUANTUM ELECTRODYNAMICS
Outgoing photon:
p
=
µ∗
(p)
_
Z
3
(2π)
3
2E
p
3. For each internal line there is a propagator
Electron propagator:
p
=
i(,p +m)
p
2
−m
2
+i
Photon propagator:
p
µ ν
=
−ig
µν
p
2
+i
4. At each vertex, place a factor of −ieγ
µ
.
Electronphoton vertex:
= −ieγ
µ
5. Insert an additional factor of 1 for each closed fermion loop.
6. For each internal loop, integrate over:
_
d
4
q
(2π)
4
(11.34)
7. A relative factor 1 appears between graphs that diﬀer from each other
by an interchange of two identical external fermion lines.
8. Internal fermion lines appear with arrows in both clockwise and coun
terclockwise directions. However. diagrams that are topologically
equivalent are counted only once.
Remember that the electronphoton coupling has the form eψγ
µ
ψA
µ
. As
a consequence, the spinors associated with any vertex must appear in the
order uγ
µ
u (with v replacing u as needed).
With this in mind, let’s do some examples.
11.4. THE REACTION E
−
+E
+
→µ
−
+µ
+
193
11.4 The Reaction e
−
+e
+
→µ
−
+µ
+
Consider the reaction e
−
+ e
+
→µ
−
+µ
+
. In terms of momentum, that is
p + p
→ k + k
, and in terms of spin indices we have s and s
→ r and r
.
The momentum of the exchanged photon is q = p +p
= k +k
. The muons
are almost identical to electrons, i.e. they have the same electromagnetic
interactions, but they are heavier by a factor of 200. This simpliﬁes the
problem in several ways. For one thing, we are usually entitled to neglect
the mass of the electron. For another, there are no identical particles in this
example. Feyman’s rules give
iM =
ie
2
q
2
_
v
e
(p
)γ
µ
u
e
(p)
_ _
u
µ
(k)γ
µ
v
µ
(k
)
_
(11.35)
I have added subscripts e and µ to the spinors to indicate that they contain
diﬀerent masses. The spin indices are implicit. I will get back to spins in a
minute.
In order to obtain a diﬀerential cross section from M we must do three
diﬃcult things: square it, sum over ﬁnal spins, and average over initial spins.
In order to make the notation more compact, I will write (11.35) as
iM =
ie
2
q
2
a
µ
b
µ
(11.36)
In order to calculate [M[
2
, we will need to ﬁgure out
[ab[
2
= a
µ
a
ν†
b
µ
b
†
ν
(11.37)
Notice that a and b are 1 1 matrices in spin space and fourvectors in
Minkowski space. There are two useful consequences; they all commute and
complex conjugation and Hermitian conjugation have the same eﬀect. This
allows us to write
a
ν†
= u
†
γ
ν†
γ
0†
v = u
†
γ
0
γ
ν
v = uγ
ν
v (11.38)
With that insight (11.37) becomes
[M[
2
=
e
4
q
4
_
v(p
)γ
µ
u(p)u(p)γ
ν
v(p
)
_ _
u(k)γ
µ
v(k
)v(k
)γ
ν
u(k)
_
(11.39)
Now make use of the spin information. The diﬀerential cross section will be
proportional to
1
2
s
1
2
s
r
r
[M(s, s
, r, r
)[
2
(11.40)
194 CHAPTER 11. QUANTUM ELECTRODYNAMICS
Remember from a previous chapter
s
u(p, s)u(p, s) =
,p +m
2m
(11.41)
s
v(p, s)v(p, s) =
,p −m
2m
Let’s do the spin sums on the ﬁrst term of (11.39)
s,s
v(p
, s
)γ
µ
u(p, s)u(p, s)γ
ν
v(p
, s
) (11.42)
=
1
2m
s
v
a
(p
, s
)γ
µ
ab
(,p +m)
bc
γ
ν
cd
v
d
(p
, s
)
I apologize for the notational overload. An object like γ
µ
is a 4 4 matrix
in spin space. I have turned it in to a tensor by adding the matrix indices
a and b. Tensors commute as you know, so (11.42) can be rewritten
=
1
2m
s
v
d
(p
, s
)v
a
(p
, s
)γ
µ
ab
(,p +m)
bc
γ
ν
cd
Now v and v are in the right relation to one another so that we can use
(11.41)
=
_
1
2m
_
2
(,p
−m)
da
γ
µ
ab
(,p +m)
bc
γ
ν
cd
Now this is recognizable as the trace of a product of matrices.
=
_
1
2m
_
2
Tr
_
(,p
−m)γ
µ
(,p +m)γ
ν
¸
Finally combining (11.39), (11.40), and (11.42), we get
1
4
spins
[M[
2
=
1
4
_
e
2mq
_
4
Tr
_
(,p
−m
e
)γ
µ
(,p +m
e
)γ
ν
¸
Tr
_
(,p
−m
µ
)γ
µ
(,p +m
µ
)γ
ν
¸
(11.43)
There you have it. QED calculations eventually come down to taking traces.
11.4. THE REACTION E
−
+E
+
→µ
−
+µ
+
195
11.4.1 Trace Technology
The things we are taking traces of are all 4 4 matrices, even when they
don’t look like that. This results in some paradoxical formulas like Tr 1 = 4
and Tr m = 4m. You can check from the representation of the gamma
matrices that Trγ
µ
= 0. Traces of an even number of γ’s are generally not
zero. Here is a simple result.
Tr(,a ,b) = a
µ
b
ν
Tr(γ
µ
γ
ν
) =
1
2
a
µ
b
ν
Tr¦γ
µ
, γ
ν
¦ = a
µ
b
ν
Tr(g
µν
) = 4a b
I have used the important fact that the trace of a product of matrices such as
Tr(abc z) is invariant under any cyclic permutation of the matrices. Thus
in the equation above Tr(γ
µ
γ
ν
) = Tr(γ
ν
γ
µ
). Our result can be summarized
Tr(γ
µ
γ
ν
) = 4g
µν
(11.44)
The trace of an odd number of γ’s always vanishes. The proof requires
a trick. Remember the γ
5
matrix. It has the properties
Tr(γ
5
) = 0 (γ
5
)
2
= 1 ¦γ
5
, γ
µ
¦ = 0
Now the line of argument goes
Tr(,a
1
,a
2
,a
n
) = Tr(,a
1
,a
2
,a
n
γ
5
γ
5
) = Tr(γ
5
,a
1
,a
2
,a
n
γ
5
)
= (−1)
n
Tr(,a
1
,a
2
,a
n
γ
5
γ
5
) = (−1)
n
Tr(,a
1
,a
2
,a
n
)
The trace is zero for n odd.
There are many other identities that are easily proved. In a moment I
will use
Tr(γ
µ
γ
ν
γ
ρ
γ
σ
) = 4(g
µν
g
ρσ
−g
µρ
g
νσ
+g
µσ
g
νρ
) (11.45)
which is obtained from repeated application of (11.44). Others can be found
in any standard text. At the moment I am looking at Peskin and Schroder,
Introduction to Quantum Field Theory Appendix A3.
11.4.2 Kinematics
A short calculation using (11.45) yields
Tr
_
(,p
−m)γ
µ
(,p +m)γ
ν
¸
= 4[p
µ
p
ν
+p
ν
p
µ
−g
µν
(p p
+m
2
e
)] (11.46)
We can set m
2
e
= 0. Then (11.43) can be evaluated by using (11.46) twice.
1
4
spins
[M[
2
=
8e
4
q
4
[(p k)(p
k
) + (p k
)(p
k) +m
2
µ
(p p
)] (11.47)
196 CHAPTER 11. QUANTUM ELECTRODYNAMICS
Equation (11.47) is valid in any Lorentz frame. This is ﬁne as far as it
goes, but in order to compare the result with an experiment, we need to
specialize to a deﬁnite frame of reference. Such an experiment is typically
done with a colliding beam machine for which the laboratory is the CM
frame. Let us assume then that k = (E, k), k
= (E, −k), p = (E, Eˆ z) and
p
= (E, −Eˆ z). The angle between k and the z axis is θ. The following
kinematic identities follow with m
e
= 0.
q
2
= (p +p
)
2
= 4E
2
p p
= 2E
2
p k = p
k
= E
2
−E[k[ cos θ p
k = p k
= E
2
+E[k[ cos θ
Eqn. (11.47) can be rewritten
1
4
spins
[M[
2
= e
4
__
1 +
m
2
µ
E
2
_
+
_
1 −
m
2
µ
E
2
_
cos
2
θ
_
(11.48)
Back in Chapter 8 I derived the relationship between M and the diﬀer
ential cross section. Since then we have introduced the Dirac spinors, which
must be normalized diﬀerently from the KlineGordon ﬁelds. I will therefore
state (without proof) the appropriate version of ?? is
_
dσ
dΩ
_
CM
=
1
2E
A
E
B
[v
A
−v
B
[
[p
1
[
(2π)
2
4E
CM
[M(p
A
+p
B
→p
1
+p
2
)[
2
(11.49)
This holds in any coordinate system in which generic particles A and B
scatter into a ﬁnal state consisting of two particles with momenta p
1
and
p
2
. In our case particles A and B are collinear and assumed massless. In
this case
[v
A
−v
B
[ =
¸
¸
¸
¸
k
A
E
A
−
k
B
E
B
¸
¸
¸
¸
= 2 (11.50)
In our example the M in (11.49) is replaced by the spinaveraged M of
(11.48).
11.5 Introduction to Renormalization
Every Feynman diagram with one or more closed loops yields a divergent in
tegral.
3
One of the most important discoveries of twentiethcentury physics
3
There are some exceptions in the case of spinzero particles, but it’s universally true
in qed.
11.5. INTRODUCTION TO RENORMALIZATION 197
is that these divergent terms contain real physical content that is part of
the “right answer” to any higherorder calculation. The program of extract
ing useful information from meaningless inﬁnities is called renormalization.
This is an extremely technical and complicated subject, but I would like
to expose you to some of the basic ideas. To do this, I will concentrate
on a subset of divergent diagrams that go under the heading of electron
selfenergy.
Consider the quantum process in which a single photon of momentum
k is emitted and reabsorbed by a virtual electron with initial and ﬁnal mo
mentum p. In the interim between emitting and reabsorbing the photon the
electron has momentum q = p − k. According to Feynman’s rules, Section
11.3, the amplitude for this process is
iS
F
(p)
_
d
4
k
(2π)
4
(−ieγ
µ
)
_
−ig
µν
k
2
+i
_
i(,p− ,k +m)
(p −k)
2
−m
2
+i
(−ieγ
ν
)iS
F
(p)
(11.51)
= iS
F
(p)
_
−ie
2
_
d
4
k
(2π)
4
γ
µ
(m+ ,p− ,k)γ
µ
(−k
2
−i)(m
2
−(p −k)
2
−i)
_
S
F
(p)
≡ iS
F
(p)Σ(p)S
F
(p)
The last equality implicitly deﬁnes the divergent quantity Σ, know as a
“selfenergy insertion.” It is clearly divergent, since there are ﬁve powers of
k in the numerator and only four in the denominator. We will eventually
evaluate it, but for the time being I would like to investigate some of its
general properties. Such as term would arise in perturbation theory as a
secondorder correction to the electron propagator. We could deﬁne a “more
nearly exact” propagator
iS
F
(p) ≈ iS
F
(p) +iS
F
(p)Σ(p)S
F
(p)
I say “more nearly exact” meaning that iS
F
(p) contains all the second
order corrections that would arise in the usual perturbation expansion of
the propagator. We could make an even better approximation by including
another selfenergy insertion.
iS
F
≈ iS
F
+iS
F
ΣS
F
+iS
F
ΣS
F
ΣS
F
This is not quite correct, even to fourth order, since I have ignored the dia
gram in which the electron emits another photon before reabsorbing the ﬁrst.
Diagrams of this sort are called “overlapping divergences.” Such diagrams
present a huge additional complication, but “suﬃcient unto the day is the
198 CHAPTER 11. QUANTUM ELECTRODYNAMICS
evil thereof.” I will ignore them. My conclusions will still be qualitatively
correct.
I can go on adding diagrams like this ad inﬁnitum. The sum would look
like this.
iS[1 + ΣS + ΣSΣS + (ΣS)
3
+ ] (11.52)
(I have suppressed the F subscript and well as the (p) to make the equations
easier to read.) This looks like the familiar geometric series (1 − x)
−1
=
1 +x
2
+x
3
+ , so I am tempted to write
iS
= iS
1
1 −ΣS
(11.53)
The question is, what does it mean to divide by a 4 4 matrix? You can
regard (11.53) as notational shorthand for (11.52). Here are some further
observations.
• It’s clear from (11.52) that Σ has the form
Σ = mA(p
2
)+ ,pB(p
2
) (11.54)
It’s true that A and B are inﬁnite, but let’s ignore that for the time
being. Furthermore
S =
,p +m
p
2
−m
2
(I will ignore the i’s for the time being also. I don’t want to distract
you with details. I promise I haven’t forgotten them and will put
them back when necessary.) It follows that [S, Σ] = 0. Therefore the
following forms are equivalent.
S
= S
1
1 −ΣS
=
1
1 −ΣS
S =
S
1 −ΣS
(11.55)
• Since ,p ,p = p
2
it makes sense to write in the same spirit
S =
,p +m
p
2
−m
2
=
,p +m
(,p −m)(,p +m)
=
1
,p −m
, (11.56)
and ﬁnally
S
−1
=,p −m. (11.57)
The propagator can now be rewritten
iS
=
iS
1 −ΣS
S
−1
S
−1
=
i
,p −m−Σ
(11.58)
11.5. INTRODUCTION TO RENORMALIZATION 199
Leaving aside the fact that Σ is inﬁnite, there is still a strange pathology
in (11.57). Every propagator we have encountered so far has had a pole at
the mass of the particle. It is almost axiomatic that a propagator is the
probability amplitude for the particle to propagate from x to y such that
p
2
= m
2
on the mass shell. It must be that the series of selfenergy terms we
have summed has modiﬁed the mass of the particle. We have to confess that
at least in the context of perturbation theory, the m that appears in the
Hamiltonian is not the physical mass of the particle. We call the m in the
Hamiltonian the “bare” mass and the mass that marks the pole of the exact
propagator, the “dressed” or physical mass, which I will call m. It seems
that given the bare mass, we should be able to calculate the physical mass
or vice versa. The fact that Σ is inﬁnite makes this permanently impossible.
We have not faced up to this problem so far, because we have always worked
at the lowest nontrivial order of perturbation theory. To that order we can
assume the the bare and physical masses are identical, and everything works
out OK. As soon as we include the selfenergy terms, we must keep track
of both m and m. Worse yet, the same sort of thing happens when we
calculate higherorder corrections to the electronphoton vertex. There we
ﬁnd that the charge e put into the Hamiltonian is not the physical charge of
the electron but some “bare” charge, and again we must keep track of the
e’s and e’s.
The agenda of keeping track of these corrections goes by the name “renor
malization.” It is a remarkable fact that all the inﬁnite corrections brought
about by all loop diagrams to all orders of perturbation theory can swept un
der the rug of these two constants. A theory for which this is possible is said
to be “renormalizable.” It is an article of faith that no nonrenormalizable
theory can be completely right, however useful it might be phenomenologi
cally. Fermi’s theory of weak interactions, which we will study in the next
chapter, is a good example. It explains nuclear beta decay quite nicely, but
it would require an inﬁnite number of inﬁnite constants to take care of all
the loops! It was eventually modiﬁed by the addition of the intermediate
vector bosons. The resulting theory is called the “standard model.” It is
renormalizable and right! It is also possible to do quantum ﬁeld theory with
gravitons. The theory at least predicts Newton’s law of gravitation. To this
extent it is correct. It is nonrenormalizable, however, and we have no idea
what the true theory might be.
It is necessary to rewrite (11.58) in such a way that the physical mass
appears in the denominator. To this end we expand the self energy term as
200 CHAPTER 11. QUANTUM ELECTRODYNAMICS
follows.
Σ(,p) = Σ(m) + (,p −m)Σ
(m) + (,p −m)
2
R(p
2
) (11.59)
This is the key equation, and it deserves some comment.
• It looks like a power series expansion about the point ,p = m, but in
fact, ,p can never equal m, since m is diagonal and ,p never is. Rather,
(11.59) deﬁnes implicitly what is meant by R(p
2
). I will sometimes
write it
Σ(,p) = Σ(m) + (,p −m)Σ
(m) + Σ
R
(,p) (11.60)
In which case it deﬁnes Σ
R
. By construction Σ
R
(m) = 0.
• It appears that in the term Σ
we have diﬀerentiated with respect to a
matrix! That is not so strange as it seems. The point is that p
2
=,p
2
.
For example, the A(p
2
) term in (11.54) can be diﬀerentiated as follows.
A
(m) =
dA(p
2
)
d ,p
¸
¸
¸
¸
p=m
=
dA(p
2
)
dp
2
d ,p
2
d ,p
¸
¸
¸
¸
p=m
= 2m
dA(p
2
)
dp
2
¸
¸
¸
¸
p
2
=m
2
(11.61)
Strictly speaking, Σ(m) and Σ
(m) are both inﬁnite, so the operations
in (11.60) and (11.61) are not well deﬁned. Our strategy for giving them a
precise meaning works as follows. I will introduce an additional parameter
into (11.51) in such a way the the integrals converge for > 0. It will
turn out that A(p
2
) and B(p
2
) in (11.54) can each be split into two parts.
One part will depend on . It becomes inﬁnite as → 0, but for ,= 0 it is
a wellbehaved mathematical expression. The other part will depend on p
2
but not on . It is ﬁnite, calculable, and important. I am about to show that
all the epsilondependent terms can be lumped together with the bare mass
to give the physical mass m and a new epsilondependent term Z
2
called
the wave function renormalization factor, which cancels out in any complete
calculation of a Feynman diagram. The point is that all these calculations
are done with ﬁnite (though epsilondependent) terms, and when the time
comes to calculate some physical quantity such as a cross section or reaction
rate, there are no epsilondependent terms remaining.
Now substitute (11.60) into (11.58)
iS
=
−i
m+ Σ(m)− ,p + (,p −m)Σ
(m) + Σ
R
(11.62)
11.5. INTRODUCTION TO RENORMALIZATION 201
I looks like the “real” mass is m = m+Σ(m). We need one further constant.
4
Z
−1
2
≡ 1 −Σ
(m) (11.63)
Equation (11.62) can be rewritten
iS
=
i(,p +m)
p
2
−m
2
Z
2
1 + (m− ,p)Z
2
R(p
2
)
(11.64)
The looks like the “bare” propagator iS (with m replaced by m) multiplied
by a factor, which, if Z
2
were not epsilondependent, would be ﬁnite and
calculable. Before we set = 0, however, Z
2
is ﬁnite, and we can show why it
does not appear in the ﬁnal result of any complete calculation of a Feynman
diagram. First consider the Z
2
in the denominator of (11.64). You can see
from (11.51) that Σ is proportional to e
2
. We must confess again the the e
that appears in the Hamiltonian is not the measured charge of the electron,
but rather a “bare” charge, which is modiﬁed by all the loop diagrams to all
orders of perturbation theory. Part of this modiﬁcation is due to Z
2
. Let’s
tentatively deﬁne a “dressed” charge, e =
√
Z
2
e. Then Z
2
disappears from
the denominator because it is absorbed into Σ. The Z
2
disappears from the
numerator when we realize that our virtual electron will not be observable
until it couples to a photon at each end. (Draw some Feynman diagrams
to convince yourself this is true.) Each electronphoton vertex comes with
a factor of e, so the entire diagram is proportional to e
2
Z
2
= e
2
.
Now I will evaluate the very diﬃcult integral (11.51). Notice that the
denominator is the product of two factors that can be labelled
A = m
2
−(p −k)
2
−i B = −k
2
−i
Feyman realized that a fourdimensional integral can be simpliﬁed by making
it ﬁvedimensional. The trick is
1
AB
=
_
1
0
dz
[Az +B(1 −z)]
2
≡
_
1
0
dz
D
2
(11.65)
The term D can be simpliﬁed by completing the square. Deﬁne k
µ
≡ k
µ
+
zp
µ
. Then
D = C
2
−k
2
where C
2
≡ z[m
2
−p
2
(1 −z)]
4
This is standard notation. There are other Z’s, but since we are only dealing with
the electron selfenergy part, these other constants don’t arise.
202 CHAPTER 11. QUANTUM ELECTRODYNAMICS
With these substitutions, the numerator of (11.51) becomes
N(p, z) = γ
µ
[m+ ,p(1 −z)− ,k
]γ
µ
I can delete the ,k
term using the familiar argument that an odd function
integrated over even limits yields zero. Using the fact that γ
µ
γ
µ
= 4, gives
us the ﬁnal form of the numerator function.
N = 2[2m− ,p(1 −z)] (11.66)
With all these deﬁnitions the selfenergy term becomes
Σ(p) = −ie
2
_
1
0
dzN(p, z)
_
d
4
k
(2π)
4
1
(C
2
−k
2
−i)
2
(11.67)
The integral can be done for integer dimension d using the remarkable for
mula
5
_
d
d
k
(2π)
d
1
(C
2
−k
2
−i)
n
=
i
(4π)
d/2
Γ(n −d/2)
Γ(n)
_
1
C
2
_
n−d/2
(11.68)
Here Γ is the Euler gamma function deﬁned by
Γ(α) ≡
_
∞
0
dt t
α−1
e
−t
(11.69)
If α is an integer, say n, Γ(n) = (n−1)!, and in general Γ(α) = (α−1)Γ(α−1).
It’s useful to get Maple to plot this function for you. You will see that Γ
interpolates smoothly between the integer values of α. You will also see that
Γ(0) = ∞. We expected that. The integral (11.67) has to diverge. Now
of course, the integral on the left side of (11.68) only makes sense when d
is an integer, but the expression on the right interpolates smoothly between
integer values of d. Even if d is only inﬁnitesimally less than four, it is still
ﬁnite. Now deﬁne ≡ 4 −d. For our purposes, n = 2.
Γ(n −d/2)
Γ(n)
=
Γ(3 −d/2)
2 −d/2
=
2
Σ(p) =
2e
2
(4π)
2
_
1
0
dzN
C
−
(11.70)
Insert the expansion
C
−
= 1 − ln C +O(
2
)
5
This is proved in Gross, Relativistic Quantum Mechanics and Field Theory, page 344
11.5. INTRODUCTION TO RENORMALIZATION 203
Σ(p) =
2e
2
(4π)
2
_
1
0
dzN(,p, z)
_
1
−ln C(p
2
, z)
_
+O() (11.71)
The integrals in (11.71) are easy enough to do, although the resulting ex
pressions are not very illuminating. The important point is that Σ(,p) can
be divided up as follows.
Σ(,p) = (mB
1
+ ,pB
2
)/ +
˜
Σ(,p) (11.72)
where
˜
Σ(,p) does not contain and so remains ﬁnite in the limit →0. B
1
and B
2
are simple constants. When we expand Σ(,p) using (11.60), Σ
R
(,p)
is as promised, ﬁnite. Thus we can absorb all the inﬁnite terms into the
two constants m and Z
2
, and calculate Σ
R
(,p)and R(p
2
) which are, again as
promised, ﬁnite and important.
204 CHAPTER 11. QUANTUM ELECTRODYNAMICS
Chapter 12
Weak Interactions
The ﬁeld of nuclear physics began with the study of radioactive decay. There
were three identiﬁable processes called α β and γdecay corresponding
to the emission of alpha particles, electrons, and photons respectively. In
retrospect, they manifest the strong, weak, and electromagnetic forces, the
three forces that operate at the level of nuclei and particles. Gravity is too
weak to have any eﬀect, and there has never been any convincing evidence
of a “ﬁfth force.”
Consider some generic nucleus that undergoes βdecay, say
N
X
Z
. By
1930 the process was believed to be
1 N
X
Z
→
N
X
Z+1
+ e
−
. Unfortunately,
the energy spectrum of the emitted electron looks something like the sketch
in Fig. 1. If there were only two particles in the ﬁnal state, all the elec
trons should have the energy marked E
max
. Several wild hypotheses were
put forward to explain this including the suggestion that energy and/or mo
mentum were not conserved in these reactions. In 1930 W. Pauli suggested
that βdecay was in fact a threebody reaction,
N
X
Z
→
N
X
Z+1
+ e
−
+ ¯ ν.
The ¯ ν was a hitherto unknown particle, which Pauli called the “neutron.”
The neutron as we know it is a constituent of nuclei, so the name was later
changed to “neutrino.” The particle would have to have zero charge and
spin ¯ h/2 to conserve charge and angular momentum. So far as one could
tell from the endpoint energy, it had to have a very small mass. Since it
has proved very diﬃcult to detect, it must interact weakly with ordinary
matter.
2
Now we know that there are actually three kinds of neutrinos all
1
There is a similar decay in which a positron is emitted. This is harder to recognize,
however, because the positron immediately captures an electron forming positronium.
This decays quickly into two or three photons.
2
I have written ¯ ν rather than ν, because in this case, the particle is actually an an
tineutrino. The distinction between neutrinos and antineutrinos is a subtle matter, to
205
206 CHAPTER 12. WEAK INTERACTIONS
Electron Energy
Events E
max
Figure 12.1: The βdecay energy spectrum
with very small masses, but for the time being I will consider only one kind,
the socalled electron neutrino and make the very good approximation that
it is massless.
3
It is ironic that whereas the radioactive decay of nuclei was the ﬁrst evi
dence of the operation of nuclear forces, it is also so very complicated that no
ﬁrstprinciples calculation could ever reproduce the lifetimes and branching
ratios of these decays. Nuclei are very complicated objects, and the strong,
electromagnetic, and weak forces are inextricably entangled in all of these
processes. The “scientiﬁc method” dictates that we should understand the
simplest instances ﬁrst. These turn out to be the purely leptonic decays
such as µ → e + ν + ¯ ν where the strong interactions play no role. I would
like to postpone that reaction for a bit and follow a more historical line of
development. We know that once outside of a nucleus, a neutron will decay,
n → p + e
−
+ ¯ ν in about 13 minutes. This process is complicated slightly
by the eﬀect of the strong interactions, but no nuclear physics is involved,
and this was the ﬁrst decay to be studied quantitatively.
By the mid 1950’s it was believed that neutron decay as well as all
the other weak interactions known at the time could be described by the
interaction Hamiltonian
H
int
=
G
√
2
J
α
(x)J
†
α
(x) (12.1)
which I will return later.
3
“Suﬃcient unto the day is the evil thereof.”
207
e

e
+
µ
−
µ
+
γ
Figure 12.2: e
+
+e
−
→µ
+
+µ
−
via one photon exchange
where
G = 1.166 10
−5
GeV
−2
(12.2)
is the universal weak coupling constant or Fermi constant. This is modelled
after the electromagnetic interactions that we have already studied. For
example the process e
+
+ e
−
→µ
+
+ µ
−
shown in Fig. 2 can be described
by the matrix element
/
int
= −ie(¯ v
e
γ
α
u
e
)D
αβ
(¯ u
µ
γ
β
v
µ
) = −ieJ
α
e
D
αβ
J
β
µ
(12.3)
where D
αβ
is the photon propagator. Fermi suggested that protons and
neutrons (collectively called nucleons) and electrons and neutrinos (collec
tively called leptons) could also constitute currents just as the electrons and
muons do in (12.3). The weak current J
α
consists of a leptonic part l
α
and
a hadronic part h
α
,
J
α
(x) = l
α
(x) +h
α
(x). (12.4)
The are three diﬀerent leptons, the e, µ and τ, so l
α
consists of a sum
of three separate currents. Likewise there are many hadrons that undergo
weak decay, so in principle h
α
should be a sum of of all the corresponding
currents. In this way (12.1) describes many diﬀerent processes. There will
be purely leptonic process such as muon decay, semileptonic processes like
neutron decay, and many purely hadronic processes such as Λ → p + π
and K → π + π. It is now known that the weak interactions are mediated
by the exchange of vector bosons just as the electromagnetic interactions
are mediated by the exchange of the photon. The bosons, however, are
208 CHAPTER 12. WEAK INTERACTIONS
ν
n
p
e

Figure 12.3: Feynman diagram for neutron decay
extremely heavy, so for ordinary nuclear processes, the propagator collapses
into a single constant that is absorbed into G. Weak decays therefore are
essentially pointlike interactions. Neutron decay, for example, is described
by the Feynman diagram Fig 13.3.
Equation (12.1), the socalled “currentcurrent interaction,” had its suc
cesses, but there is much more to the story. For one thing, (12.1) does not
allow the nucleon’s spin to ﬂip, at least not in the nonrelativistic limit. The
spin frequently does ﬂip in nuclear βdecay, so other interaction terms must
be considered. It is also nonrenormalizable so it is at best a lowenergy
approximation.
12.1 Parity Violation
By far the most profound modiﬁcation of (12.1), however, has to do with
parity. Until 1956 it was assumed that parity was conserved in all interac
tions. Roughly speaking, this means that the mirror image of any allowed
process is also an allowed process with all the same reaction rates. In more
mathematical terms, there is an operation called parity conjugation, which
has two eﬀects: ﬁrst it replaces x
µ
with x
µ
= (ct, −x), and second, it
replaces every wave function or quantum ﬁeld ψ(x) with a new function
ψ
(x
) that describes the particle or system in the mirrorreﬂected world.
The statement that parity is conserved is equivalent to saying that this
transformation leaves all the equations of motion form invariant. In 1956
12.1. PARITY VIOLATION 209
Lee and Yang pointed out that there was no evidence that this was true for
the weak interactions. It was shown shortly thereafter that weak interac
tions violate this principle maximally.
4
This was done by polarizing
60
Co
nuclei in a strong magnetic ﬁeld.
60
Co βdecays emitting an electron. It
was found that the intensity distribution of the emitted electrons could be
described by the formula
I(θ) = 1 −¸J) p/E = 1 −v cos θ (12.5)
Here J is the nuclear spin and p the momentum of the electron. The impor
tant point about (12.5) is that the distribution is not symmetric between the
forward (with respect to the nuclear spin or magnetic ﬁeld) and backward
direction. This in itself is proof that parity is not conserved. To see how
this follows, imagine viewing the experiment in a mirror held perpendicular
to the nuclear spin. Since angular momentum has the form L = x ×p, it
does not change sign under parity conjugation. Its mirror image is identical
to the real thing. It is an example of a pseudo or axial vector. The elec
tron distribution, however, does change. In this world it is peaked in the
direction opposite to J. In the mirror world it is peaked along J. Ergo,
parity is not conserved. Parenthetically,
60
Co also undergoes γdecay, and
the photon distribution is symmetric between the forward and backward
directions. This is because electromagnetic interactions do conserve parity.
That was a simple argument, but incorporating parity into the theory
of relativistic electrons requires a some formalism. Our postulate of form
invariance requires that the Dirac equation in the mirror reﬂected world
should be
(iγ
µ
∂
µ
−m)ψ
(x
) = 0 (12.6)
Multiply the ordinary Dirac equation on the left by γ
0
.
γ
0
(iγ
µ
∂
µ
−m)ψ(x) = (iγ
µ
∂
µ
−m)γ
0
ψ(x) = 0 (12.7)
In order for (12.6) to be true, it must be that
ψ
(x
) = γ
0
ψ(x) (12.8)
In Chapter 9.4 we touched brieﬂy on the ﬁve bilinear forms that can be
constructed with Dirac spinors. I showed that
¯
ψψ transforms like a scalar
under Lorentz transformations and that
¯
ψγ
µ
ψ transforms like a vector. I
also introduced the γ
5
matrix and made some vague remarks to the eﬀect
4
I will explain eventually what I mean by this
210 CHAPTER 12. WEAK INTERACTIONS
that it had something to do with parity. For easy reference, here are its
deﬁnition and properties.
γ
5
= iγ
0
γ
1
γ
2
γ
3
(12.9)
(γ
5
)
†
= γ
5
(12.10)
(γ
5
)
2
= 1 (12.11)
¦γ
5
, γ
5
¦ = 0 (12.12)
Let’s take a look at
¯
ψγ
5
ψ. Under parity conjugation,
¯
ψ(x)γ
5
ψ(x) →
¯
ψ
(x
)γ
5
ψ
(x
) =
¯
ψ(x)γ
0
γ
5
γ
0
ψ(x) = −
¯
ψ(x)γ
5
ψ(x) (12.13)
It is, as claimed, a pseudo scalar. In the same way it can be shown that
¯
ψγ
µ
γ
5
ψ is a pseudo or axial vector.
How are we to modify (12.1) to accommodate parity violation? At ﬁrst
sight one could replace γ
α
by γ
α
γ
5
making the currents into pseudovectors.
But since the Hamiltonian is a product of two such currents, the minus
signs cancel, and parity conjugation has no eﬀect. In principle one could
use all ﬁve of the bilinear forms in arbitrary linear combinations, and in the
case of some nuclear βdecay, that possibility must be taken seriously. In
order to understand the fundamental nature of weak interactions, however,
it is useful to look at reactions in which there are no strong interactions
to complicate things. There are a few purely weak decay processes, and
the most accessible is ordinary muon decay, µ → e + ν + ¯ ν. After much
experimental work it is clear that the way to construct purely leptonic weak
currents is as follows.
l
µ
= ¯ u
e
γ
µ 1
2
(1 −γ
5
)v
ν
(12.14)
This is called the VA or vector minus axial vector interaction. This precise
form is tied up with the properties of the neutrino to which we now turn.
12.2 Neutrinos
We know that there are three kinds of neutrinos called the electron neutrino,
the muon neutrino, and the tau neutrino. There is considerable indirect
evidence that they all have some small mass, which from a theoretical point
of view is both a complication and an embarrassment. The neutrino that
appears in (12.14) is the electron neutrino whose mass is a tiny fraction of an
electron volt. Except for those experiments involving neutrinos from the sun
or from outer space, this mass is completely negligible, and it will simplify
our work considerably if we ignore it. In this limit, half of the neutrino
12.2. NEUTRINOS 211
spinor is “missing,” and the role of the
1
2
(1 −γ
5
) in (12.14) is to “delete” it.
In order to see what this means, it is useful to reconstruct the Dirac spinors
according to a diﬀerent plan.
The gamma matrices are deﬁned by their anticommutation relations,
but there are several diﬀerent matrix representations that have the same
relations. The choice given in Equations 9.6 and 9.7 is called the Dirac or
standard representation. When spinors are constructed with this choice, the
third and fourth components are proportional to p/(E + m) and so vanish
in the lowenergy limit. There is another important representation called
the Weyl or chiral representation.
5
γ
0
=
_
0 1
1 0
_
γ
i
=
_
0 σ
i
−σ
i
0
_
γ
5
=
_
−1 0
0 1
_
(12.15)
The important diﬀerence is that γ
0
is diagonal in the standard representa
tion, and γ
5
is diagonal in the chiral representation.
The Dirac equation in the limit m = 0 and in momentum space is simply
,pu = 0. Using the representation (12.15) this is
_
0 E −σ · p
E +σ · p 0
_ _
φ
χ
_
= 0. (12.16)
The determinant of the coeﬃcient matrix must vanish, and so E = ±[p[.
Concentrate on the positiveenergy solution for the time being. We are left
with
χ = σ ˆ p χ (12.17)
φ = −σ ˆ p φ
In this limit the upper and lower components decouple. The operator σ p
represents the component of spin along the particle’s direction of motion.
Its eigenvalues are called helicity. According to (12.17), massless particles
are helicity eigenstates. The spin is either aligned parallel to the direction
of motion (positive helicity, righthanded particles) or antiparallel (negative
helicity, lefthanded particles).
Let us reﬁne our notation somewhat and use λ = ±1 to represent the
helicity. Introduce the helicity eigenstates deﬁned by
σ · p φ
λ
= λ[p[φ
λ
σ · p χ
λ
= λ[p[χ
λ
(12.18)
5
Beware. There are several diﬀerent sign conventions used by other authors.
212 CHAPTER 12. WEAK INTERACTIONS
Massless positiveenergy spinors are then
u(p, λ = 1) =
_
0
φ
+
_
u(p, λ = −1) =
_
φ
−
0
_
(12.19)
The matrix γ
5
is called the chirality operator.
6
In the chiral representa
tion the operator
1
2
(1 ±γ
5
) has a simple structure.
1
2
(1 +γ
5
)u =
_
0 0
0 1
_ _
φ
χ
_
=
_
0
χ
_
(12.20)
1
2
(1 −γ
5
)u =
_
1 0
0 0
_ _
φ
χ
_
=
_
φ
0
_
Clearly the
1
2
(1 ± γ
5
)’s are acting like projection operators. Let’s be brief
and write P
±
≡
1
2
(1 ± γ
5
). These operate like any Godfearing projection
operators. For example, P
2
+
= P
+
, P
+
P
−
= 0, and P
+
+ P
−
= 1. The fact
that u
ν
in (12.14) is always multiplied by P
−
means that only lefthanded
neutrinos interact.
P
−
u
ν
(λ = 1) =
_
0
0
_
P
+
u
ν
(λ = −1) =
_
φ
−
0
_
(12.21)
So far as one can tell from the form of the interaction, it might be that
neutrinos come in two helicity states, but somehow only the lefthanded
states are allowed to participate. In fact, neutrinos only interact weakly and
always with the
1
2
(1 −γ
5
) in front of the spinor, so whether or not there are
positive helicity states is something of a metaphysical question.
7
It is these
“missing” or noninteracting helicity states that are directly responsible for
parity violation. If we take the neutrino spinor to be
P
−
u
ν
=
_
φ
−
0
_
(12.22)
then the parity conjugation operation (12.8) forces us to conclude that
P
−
u
ν
= P
−
γ
0
u
ν
=
_
0
0
_
(12.23)
6
The word “chirality” comes from the Greek meaning “hand.” Nonetheless, the terms
right and lefthanded refer to helicity not chirality.
7
The fact that the lefthanded rather than righthanded states interact was established
in a classic experiment by Goldhaber et al (1958) described in Perkins, Introduction to
High Energy Physics, AddisonWesley 1987
12.3. CHIRALITY, HELICITY AND ELECTRONS 213
When I said that parity was violated maximally, I meant that the right side
of (12.23) is zero without any remaining small components.
8
12.3 Chirality, Helicity and Electrons
This simple association of chirality with helicity in (12.19) does not hold for
massive particles. The Dirac equation in our new basis including the mass
term is
_
−m E −σ · p
E +σ · p −m
_ _
φ
χ
_
= 0 (12.24)
(E +σ · p)φ = mχ
(E −σ · p)χ = mφ
It’s convenient to take φ and χ to be helicity eigenstates. Equations (12.24)
become
1
m
(E +λ[p[)φ
λ
= χ
λ
(12.25)
The complete spinors are
u(λ) = C
_
¸
¸
_
φ
λ
E+λp
m
φ
λ
_
¸
¸
_
=
1
√
2m
_
¸
¸
_
_
E −λ[p[ φ
λ
_
E +λ[p[ φ
λ
_
¸
¸
_
(12.26)
The normalization constant C is chosen according to our convention that
¯ uu = 1. Now look at the relative sizes of the upper and lower components.
_
E −[p[
_
E +[p[
=
_
E
2
−[p[
2
E +[p[
=
m
E +[p[
≈
m
2E
(12.27)
which vanishes in the limit m → 0 or equivalently the limit [p[ ¸ m.
9
In
this limit
P
−
u(λ = 1) =
1
√
2m
_
0
0
_
(12.28)
P
−
u(λ = −1) =
1
√
2m
_ √
2E φ
−
0
_
8
Now you see why the neutrino mass is such a problem. The statement is not quite
correct.
9
It is still meaningful to take the limit m →0 despite the factor of 1/
√
2m. The point is
that all physical results depend on bilinear products like ¯ uu. Because of our normalization
condition, all the m’s cancel.
214 CHAPTER 12. WEAK INTERACTIONS
The point is that for [p[ ¸ m, the P
−
in the interaction suppresses the
positive helicity components by the factor m/2E. It is easy to show that if
u
1
and u
2
are any two spinors,
¯ u
1
γ
µ
P
−
u
2
= P
−
¯ u
1
γ
µ
u
2
= P
−
¯ u
1
γ
µ
P
−
u
2
(12.29)
so that in this limit, only the lefthanded components of all fermions partic
ipate in the interaction.
One can construct the spinors for antiparticles by repeating the pro
cedure with E replaced by −E in (12.16) and (12.24). Equation (12.17)
becomes
φ = −σ ˆ p φ (12.30)
χ = σ ˆ p χ
The signs have changed. Massless righthanded antiparticles have negative
chirality. Only righthanded antineutrinos participate in weak interactions.
Equation (12.26) becomes
v(λ) = C
_
¸
¸
_
−
1
m
(E +λ[p[)χ
λ
χ
λ
_
¸
¸
_
=
1
√
2m
_
¸
¸
_
−
_
E +λ[p[χ
λ
_
E −λ[p[χ
λ
_
¸
¸
_
. (12.31)
(Remember that ¯ vv = −1.) For [p[ ¸m, the negative helicity components
are suppressed by m/2E.
The missing helicity state in the neutrino wave function also causes
charge conjugation symmetry to be violated. According to (12.20), neutri
nos have negative helicity. Equation (9.59) shows that the chargeconjugate
state must also have negative helicity, but we have just showed that the
helicities of massless antiparticles are opposite to those of particles. Nega
tive helicity antineutrinos don’t exist, or if they do, they don’t participate.
The combination of parity conjugation and charge conjugation is a diﬀerent
story. The helicity operator σ p is odd under parity conjugation, so the
helicity of a parityconjugated state is opposite to that of the original state.
Two minuses make a plus, right? Weak interactions are invariant under CP.
Well – almost. There is a “superweak” component to the weak inter
actions that does violate CP. It is so weak that it only manifests itself in
the decays of the neutral K and B mesons, and then only because of a
complicated set of fortuitus circumstances. CP violation is one of the most
profound discoveries of particle physics in recent years. It is believed to be
the origin of the preponderance of matter over antimatter in the universe.
12.4. LEPTON NUMBER 215
12.4 Lepton Number
I mentioned (without explanation) that ordinary muon decay µ
−
→ e
−
+
ν + ¯ ν, requires a neutrino and an antineutrino in the ﬁnal state. I also
stated (again without explanation) (12.14) that the lepton current should
be l
µ
= ¯ u
e
γ
µ 1
2
(1−γ
5
)v
ν
rather than say l
µ
= ¯ u
e
γ
µ 1
2
(1−γ
5
)u
µ
, both of which
are allowed by Feynman’s rules. How is one to know about these things?
The following reactions and decays all conserve energy and charge, but
none have ever been observed.
π
0
,→µ
+
+e
−
µ
−
,→e
−
+γ
µ
−
,→e
+
+e
−
+e
−
µ
−
+p ,→p +e
−
¯ ν +n ,→p +e
−
(12.32)
The last reaction seems impossible to realize, but there’s a trick. Park a
swimming pool full of cleaning ﬂuid in front of a nuclear reactor. Reactors
produce mostly antineutrinos, so one looks for the reaction
¯ ν +
37
Cl →
37
A+e
−
This can be identiﬁed since
37
Ar is a radioactive gas that can be separated
from the liquid and its radioactivity measured. This is a heroic experiment,
but it’s been done, and the reaction just doesn’t occur. What is “wrong”
with these reactions? The (almost) obvious guess is that these leptons carry
some sort of additive quantum numbers that in (12.32) just don’t add up.
First of all, muons can’t decay into electrons, so there must be some measure
of “muonness” and “electronness” that must be conserved. As a convention,
we say that negative electrons have an electron number = +1, and positrons
have electron number = −1. Similarly, negative muons have muon number
= +1, and positive muons have muon number = −1. (There is also a tau
lepton, so it must carry a tau number assigned in the same way.) Reactions
like e
+
+e
−
→γ +γ are allowed, since γ’s are not leptons, and so all their
lepton numbers are zero. What are we to make of µ
−
→ e
−
+ ¯ ν + ν? We
have to acknowledge that neutrinos also carry lepton number and that all
neutrinos are not created identical. There must be electron neutrinos that
carry electron number just as electrons do, as well as muon neutrinos and
tau neutrinos. The complete set of quantum numbers is given in Table 13.1.
The correct way to specify muon decay is µ
−
→e
−
+ ¯ ν
e
+ν
µ
. Now it’s clear
216 CHAPTER 12. WEAK INTERACTIONS
e
−
ν
e
µ
−
ν
µ
τ
−
ν
τ
n
e
1 1 0 0 0 0
n
µ
0 0 1 1 0 0
n
τ
0 0 0 0 1 1
Table 12.1: Standard lepton number assignments. Antiparticle lepton num
bers have the opposite sign. All other particles have zero lepton number.
µ
−
ν
µ
e

ν
e
Figure 12.4: Muon decay µ
−
→e
−
+ ν
µ
+ ¯ ν
e
. The solid lines represent the
ﬂow of muon number current. The dotted lines represent electron number
current.
that both sides of the reaction have n
µ
= 1 and n
e
= 0.
Lepton numbers also resolve the question raised above regarding the
construction of lepton currents. You will recall that the conservation of
charge was a key requirement in constructing currents for electromagnetic
currents. Take for example, the term j
α
e
= ¯ v
e
γ
α
u
e
in (12.5) illustrated in
Figure 13.2. It represents an electron current that “ﬂows” up the electron
leg of the diagram and down the positron leg. Conservation of charge means
that you can trace the current along this path following the arrows without
lifting your pencil from the page. In the same way lepton number represents
a kind of “charge” that must be conserved. In Figure 13.4 I have illustrated
muon decay using solid lines to represent the ﬂow of muon number and
dotted lines to represent electron number current. The complete matrix
12.4. LEPTON NUMBER 217
element for muon decay is then
/=
4G
√
2
¯ u
ν
γ
α 1
2
(1 −γ
5
)u
µ
¯ u
e
γ
α
1
2
(1 −γ
5
)u
µ
(12.33)
218 CHAPTER 12. WEAK INTERACTIONS
Chapter 13
The Electroweak Theory
13.1 Weak interaction of quarks
From the point of view of quarks, ordinary neutron decay, n →p +e
−
+ ¯ ν,
amounts to a process in which a down quark beta decays into an up quark
and a lepton pair, d →u+e
−
+¯ ν. When allowance is made for the diﬀerences
in mass and phase space, the strength of this reaction is almost the same
as the purely leptonic process µ → e + ν + ¯ ν. We can construct weak VA
currents for these processes as follows,
j
µ
(lepton) ∝ ¯ u(ν)γ
µ 1
2
(1 −γ
5
)u(e) (13.1)
∝ ¯ ν
L
γ
µ
e
L
j
µ
(quark) ∝ ¯ u(u)γ
µ 1
2
(1 −γ
5
)u(d) (13.2)
∝ ¯ u
L
γ
µ
d
L
Notice the condensed notation: ¯ u
L
is the lefthanded projection of the up
quark. At this point I am not concerned with whether this is a utype or a
vtype spinor. Of course that is important when doing an exact calculation,
but you have Feynman’s rules to ﬁgure that out. Here I am trying to present
an idea with a minimum of notational complications. By the same token,
∝ means “proportional to.” The constant of proportionality will be all
important, but not now. Finally, notice that the projection operator,
1
2
(1 −
γ
5
) projects the lefthanded components of both spinors as shown in class.
The currents in (13.1) and (13.2) are both “charged currents,” in the
sense that the currents change charge in the process of the interaction. Put
it another way, they couple to charged intermediate vector bosons. Until
the 1970’s it was believed that there were no neutral current interactions,
219
220 CHAPTER 13. THE ELECTROWEAK THEORY
since processes like K
0
→ µ
+
µ
−
seem to be forbidden. The K
0
carries
one unit of strangeness, however, so this is an example of a strangeness
changing interaction, and these are forbidden. With the advent of high
energy neutrino beams it became possible to see reactions such as ν
µ
+N →
ν
µ
+ X, where N is some heavy nucleus, and X stands for some hadronic
”stuﬀ.” In terms of quarks, these are examples of the reaction ν +q →ν +q.
(q stands for any quark.) There is no reason to believe that these interactions
have the pure VA form, so the correspondinq currents are
j
µ
(leptonic neutral current) ∝ ¯ νγ
µ 1
2
(1 −γ
5
)ν (13.3)
j
µ
(neutral hadronic current) ∝ ¯ qγ
µ 1
2
(c
V
−c
A
γ
5
)q (13.4)
The constants c
V
and c
A
have been determined experimentally with inelastic
neutrino scattering. As a consequence of (13.4), weak neutral currents have
both right and lefthanded components.
13.2 Weak Isospin
The electroweak theory per se does not involve quarks, nonetheless, the
following facts established in the previous section are foundational.
• Weak leptonic currents and weak charged hadronic currents have only
lefthanded components.
• The weak, neutral, hadronic currents that they couple to have both
right and lefthanded components.
Electrons, of course, also have both right and lefthanded components, but
only the lefthanded components participate in the weak interactions. Or
so it seems. The underlying idea of the electroweak theory is that the weak
and electromagnetic interactions are closely related, and the relationship is
cemented by the marriage of the righthanded piece of the electron current
and the righthanded piece of the weak, neutral, hadronic current.
To bring this marriage about I will deﬁne
J
+
µ
≡ ¯ ν
L
γ
µ
e
L
(13.5)
J
−
µ
= (J
+
µ
)
†
= ¯ e
L
γ
µ
ν
L
(13.6)
The notation J
+
(J
−
) means that the initial lepton gains (loses) one unit of
charge in the course of the interaction. It does this by coupling to a negative
13.2. WEAK ISOSPIN 221
(positive) intermediate vector boson. It is customary to use the formalism
of spin or isospin to combine these various currents. To this end, deﬁne the
“spinor,”
χ
L
=
_
ν
e
−
_
L
, (13.7)
and raising and lowering operators,
σ
±
=
1
2
(σ
1
±σ
2
) (13.8)
σ
+
=
_
0 1
0 0
_
σ
−
=
_
0 0
1 0
_
The σ
i
’s are the usual Pauli spin matrices. Then
J
i
µ
≡ ¯ χ
L
γ
µ
1
2
σ
i
χ
L
J
±
µ
≡ ¯ χ
L
γ
µ
σ
±
χ
L
(13.9)
This is familiar angular momentum physics in an unfamiliar context. Think
of χ
L
as a Pauli spinor. We can combine two spin1/2 doublets to make a
spin1 triplet and a spin0 singlet. In this case, the underlying quantity is
not spin but something conventionally called ‘weak isospin.” The currents
J
±
µ
are two components of the triplet. The third member is
J
3
µ
≡ ¯ χ
L
γ
µ
1
2
σ
3
χ
L
=
1
2
¯ ν
L
γ
µ
ν
L
−
1
2
¯ e
L
γ
µ
e
L
(13.10)
This J
3
µ
is a weak neutral current, but it can’t be the whole story, because
the complete neutral weak current has righthanded components. The key
idea is to split up the electron current into right and lefthanded pieces.
j
em
µ
= −¯ eγ
µ
e = −¯ eγ
µ
1
2
(1 −γ
5
)e − ¯ eγ
µ
1
2
(1 +γ
5
)e (13.11)
= −¯ e
R
γ
µ
e
R
− ¯ eLγ
µ
e
L
Despite the fact that it contains electron spinors, j
em
µ
is a neutral current
for reasons explained previously. Notice in this particular deﬁnition I have
not included the electron charge. In (13.11) e is just the electron spinor.
Suppose the “missing” righthanded components of the weak neutral current
actually come from the righthanded components of the electromagnetic
current. This is the idea, but it plays out in ways that are rather subtle.
We assume that in addition to charge and the third component of isotopic
spin, there is a third “good quantum number” called hypercharge or Y . The
three are related by
Q = T
3
+Y/2 (13.12)
222 CHAPTER 13. THE ELECTROWEAK THEORY
lepton T T
3
Q Y
ν
e
1
2
1
2
0 1
e
−
L
1
2
−
1
2
1 1
e
−
R
0 0 1 2
Table 13.1: Weak quantum number assignments
According to Noether’s theorem, for every conserved charge there is a con
served current. The currents in this case must be related by
j
em
µ
= J
3
µ
+
1
2
j
Y
µ
(13.13)
The new current j
Y
µ
is called the weak hypercharge current.
1
The quantum
numbers are given in Table 12.1. Similar assignments can be made for the
quarks.
13.3 The Intermediate Vector Bosons
In conventional QED, the interactions are mediated by the photon ﬁeld A
µ
and the basic interaction vertex is −ieJ
µ
. In electroweak theory there are
four currents, the three components of J
i
µ
and the hypercharge current j
Y
µ
.
In the conventional notation they couple to vector boson ﬁelds as follows
−ig
_
J
i
_
µ
W
i
µ
−i
g
2
_
j
Y
_
µ
B
µ
(13.14)
The two new coupling constants g and g
will be related to the Fermi coupling
constant and the electric charge later on. The ﬁelds W
i
and B are assumed
to be massless. They acquire mass through a mechanism called spontaneous
symmetry breaking, which will be explained in a subsequent chapter. The
socalled Weinberg angle θ
W
, is another constant that must be determined
from experiment. The particles that are detected in the laboratory are the
photon which is massless, two heavy charged bosons, the W
±
’s, and the
even heavier neutral vector boson, the Z
0
. Their ﬁelds are related to those
in (13.15) as follows
W
±
µ
=
_
1
2
(W
1
µ
∓iW
2
µ
) (13.15)
A
µ
= B
µ
cos θ
W
+W
3
µ
sinθ
W
(13.16)
Z
µ
= −B
µ
sinθ
W
+W
3
µ
cos θ
W
1
There is a strong hypercharge also.
13.4. THE FEYNMAN RULES 223
The neutral current piece of (13.14) is
−ig
_
J
3
_
µ
W
3
µ
−i
g
2
_
j
Y
_
µ
B
µ
(13.17)
= −i(g sin θ
W
J
3
µ
+g
cos θ
W
j
Y
µ
/2)A
µ
−i(g cos θ
W
J
3
µ
−g
sin θ
W
j
Y
µ
/2)Z
µ
Since A
µ
is the actual photon ﬁeld, the current coupled to it must be j
em
µ
from (13.13). This can only be so if
g
cos θ
W
= g sinθ
W
= e (13.18)
Whatever it is that couples to the Z
0
must be the complete neutral weak
current, which I will call J
NC
µ
. The last term in (13.17) can be rewritten
−ig
_
cos θ
W
J
3
µ
−
sin
2
θ
W
cos
2
θ
W
(j
em
µ
−J
3
µ
)
_
Z
µ
= −i
g
cos θ
W
J
NC
µ
Z
µ
(13.19)
where
J
NC
µ
≡ J
3
µ
−sin
2
θ
W
j
em
µ
(13.20)
As promised J
NC
µ
inherits its righthanded components from j
em
µ
and its left
handed components form J
3
µ
and j
em
µ
.
13.4 The Feynman Rules
Now that the smoke has cleared, we are left with the following interactions.
• Two charged IVB’s, the W
±
’s, coupling to the weak charged current.
The interaction term in (13.14) can be rewritten using (13.6), (13.9),
and (13.15) as
−ig
_
(J
1
)
µ
W
1
µ
+ (J
2
)
µ
W
2
µ
¸
= −i
g
√
2
(J
µ
W
+
µ
+J
µ†
W
−
µ
) (13.21)
• One neutral IVB, the Z
0
, coupled to J
NC
with coupling constant
g/ cos θ
W
. Typically we write the vertex function
−i
g
cos θ
W
γ
µ 1
2
(c
V
−c
A
γ
5
) (13.22)
The constants c
V
and c
A
can be determined from based on the charge
and isotopic spin of the particles that make up J
NC
.
224 CHAPTER 13. THE ELECTROWEAK THEORY
• The photon coupling to the electromagnetic current with the usual
constant e.
The Feynman rules from Section 11.3 require the following additions.
1. At each vertex at which a W
±
couples to a charged current, insert the
vertex factor
−i
g
√
2
γ
µ 1
2
(1 −γ
5
) (13.23)
At each vertex at which a Z
0
couples to a neutral current, insert the
vertex factor
−i
g
cos θ
W
γ
µ 1
2
(c
V
−c
A
γ
5
) (13.24)
2. For each virtual W or Z insert the propagator
i(−g
µν
+p
µ
p
ν
/M
2
)
p
2
−M
2
(13.25)
3. For each free W or Z insert the spin polarization tensor ε
µ
(λ). When
summing over the polarization state λ use the completeness relation
λ
ε
(λ)∗
µ
(p)ε
(λ)
ν
(p) = −g
µν
+
p
µ
p
ν
M
2
(13.26)
The other rules are unchanged.
Chapter 14
Symmetry Breaking
Consider an atom with a spherically symmetric potential. Because of the
symmetry, the energy eigenstates are independent of the m and m
s
quantum
numbers. Put it another way, the SU(2) symmetry of the Hamiltonian
causes these states to be degenerate. If we put the atom in an external
magnetic ﬁeld, the degeneracy is removed and the energies become functions
of both m and m
s
. The system is still symmetric about the direction of the
ﬁeld, so the imposition of the external ﬁeld has broken the symmetry from
SU(2) down to U(1). This sort of symmetry breaking is called the Wigner
Mode.
There are several other ways to break symmetry that are more relevant
to this chapter. Try standing a straight ﬂexible rod upright and press down
on the top of it with a force F. Let’s assume that the rod remains straight for
some minimal force, but beyond some threshold force F
c
the rod “buckles,”
and assumes a curved shape. The symmetry is broken, the rod is no longer
straight, but there are an inﬁnite number of equivalent states corresponding
to the possible orientations of the rod. We can summarize this as follows:
1. A parameter (in this case F) assumes some critical value. Beyond that
value,
2. the symmetric situation becomes unstable, and
3. the ground state is degenerate.
The second example is the ferromagnet. The atoms in a ferromagnet
interact through a spinspin interaction
H = −
i,j
J
ij
S
i
S
j
225
226 CHAPTER 14. SYMMETRY BREAKING
which is a scalar and therefore invariant under rotations. The ground state
however, is one in which all the spins are aligned, and this is clearly not ro
tationally invariant. The direction of spontaneous magnetisation is random,
and all the degenerate ground states may be reached from a given one by
rotation. The spontaneous magnetisation disappears at high temperature,
when the ground state becomes symmetric.
It is clear that the general situation here is the same as in our ﬁrst exam
ple, the relevant parameter here being T. These two examples exhibit what
is known as “spontaneous symmetry breaking. In both cases the system pos
sesses a symmetry (rotation symmetry) but the ground state in not invariant
under that symmetry; rather, it changes into one of the other (degenerate)
ground states.
14.1 A Simple Example
Consider a classical selfinteracting real scalar ﬁeld φ with a Lagrangian
density
L =
1
2
(∂
µ
φ)(∂
µ
φ) −
1
2
µ
2
φ
2
−
1
4
λφ
4
(14.1)
We assume λ > 0 so that there is a ﬁnite minimum energy. This Lagrangian
is invariant under the discrete transformation
φ →−φ. (14.2)
There are two quite diﬀerent cases depending on the sign of µ
2
. The po
tential for µ
2
> 0 has a single minimum, V (φ) = 0 at φ = 0. This point
corresponds to the ground state, so in quantum mechanical terms
¸φ)
0
= ¸0[φ[0) = 0
Expanding the Lagrangian about ¸φ)
0
to second order gives
L ≈
1
2
(∂
µ
φ)(∂
µ
φ) −
1
2
µ
2
φ
2
(14.3)
which is the Lagrangian density of a free scalar ﬁeld of mass µ. The pertur
bations induced by the φ
4
term can be thought of as oscillations about the
origin.
Now consider the case µ
2
< 0. Now the potential has minima at
¸φ)
0
= ±
_
−µ
2
λ
≡ ±v. (14.4)
14.1. A SIMPLE EXAMPLE 227
Now there are two degenerate vacuum states. The minima at ¸φ)
0
= ±v are
equivalent and either may be chosen as the classical ground (vacuum) state
of the system. Once this choice has been made, the vacuum is no longer
invariant under the transformation (2). This is a new form of symmetry
breaking: the Lagrangian is invariant but the vacuum state is not. This
is called the Goldstone mode, spontaneous symmetry breaking, or hidden
symmetry.
Let’s choose
¸φ)
0
= +v. (14.5)
We can no longer do perturbation theory by expanding the Lagrangian about
¸φ)
0
= 0, since the slightest perturbation about this point would send the
system plunging down into one of the two minima. It makes more sense to
shift the ﬁeld by deﬁning
ξ(x) ≡ φ(x) −¸φ)
0
= φ(x) −v. (14.6)
In terms of this new variable the vacuum state is ¸ξ)
0
= 0, and the La
grangian density is (neglecting constant terms)
L = L
0
−λvξ
3
−
1
4
λξ
4
(14.7)
We can think of the ξ
3
and ξ
4
terms as perturbations of the freeﬁeld La
grangian
L
0
=
1
2
(∂ξ)
2
−λv
2
ξ
2
(14.8)
which is the Lagrangian for a free scalar ﬁeld with mass m
ξ
=
_
−2µ
2
.
(Remember that µ
2
< 0.)
This is a toy model, of course, but it illustrates many of the features of
spontaneous symmetry breaking as it manifests itself in quantum mechan
ics.
1
1. There is a nonzero expectation value of some ﬁeld in the vacuum state.
2. The resulting classical theory has a degenerate vacuum, with the choice
among the equivalent vacuum states completely arbitrary.
3. The transition from a symmetric vacuum to a degenerate vacuum typ
ically occurs as a phase transition as some order parameter (µ
2
in the
example) is varied.
1
I am copying almost verbatim from Mike Guidry, Gauge Field Theories John Wiley
& Sons (1991)
228 CHAPTER 14. SYMMETRY BREAKING
4. The chosen vacuum state does not possess the same symmetry as the
Lagrangian.
5. After we have expanded the Lagrangian around the chosen vacuum,
the original symmetry is no longer apparent. The degenerate vacua
are related to each other by symmetry operations, which tells us that
the symmetry is still there, but it is not manifest; it is hidden.
6. The masses of the particles appearing in the theory with and without
the spontaneous symmetry breaking may diﬀer substantially. We say
that the masses have been acquired spontaneously in the latter case.
7. Once the theory develops degenerate vacua the origin becomes an un
stable point. Thus the symmetry may be “broken spontaneously” in
the absence of external intervention.
14.2 Goldstone Bosons
The transformation (2) was discrete. The next generalization makes use of
a continuous global gauge transformation. For this we will need a complex
or nonHermitian ﬁeld. In previous chapters we used Hermitian scalar ﬁelds
for spin zero particles. In terms of second quantization, this means that the
particle is its own antiparticle. If the particle caries charge or some other
quantum number that is diﬀerent for particles and antiparticles, then the
ﬁeld must be nonHermitian (as is the case with electron ﬁelds). Take the
Lagrangian to be
L = (∂
µ
)
†
(∂
µ
φ) −µ
2
φ
†
φ −λ(φ
†
φ)
2
(14.9)
The λ terms represents self interaction. Under normal circumstances, we
would regard µ as a mass. Here µ
2
is just a parameter, since we are going to
make µ
2
< 0. This Lagrangian is invariant under the group of global gauge
transformations,
φ(x) →φ
(x) = e
iΛ
φ(x). (14.10)
Using polar coordinates ρ
2
= φ
†
φ, we may identify a potential
V (ρ) = µ
2
ρ
2
+λρ
4
. (14.11)
As before, set µ
2
< 0. The minima now occur in the complex φ plane on a
circle of radius
[φ[
2
=
−µ
2
2λ
≡ a
2
(14.12)
14.2. GOLDSTONE BOSONS 229
The shape of the potential has been likened to a Mexican hat or perhaps the
bottom of a champagne bottle. There is now an inﬁnity of degenerate ground
states, corresponding ro diﬀerent positions on the ring of minima in the
complex φ plane, and the symmetry operation (10) relates one to another.
In the quantum theory, where φ becomes an operator, this condition refers
to the vacuum expectation value of φ.
[¸0[φ[0)[
2
= a
2
. (14.13)
We could choose any point in the minimum and expand the Lagrangian
about that point. For simplicity, we choose the real axis Re(φ) = a, and
expand around it to investigate the spectrum. It is convenient to work in
polar “coordinates”
φ(x) = ρ(x)e
iθ(x)
, (14.14)
so the complex ﬁeld φ is expressed in terms of two real scalar ﬁelds ρ and θ.
In quantum language, we choose the vacuum state
¸0[φ[0) = a (14.15)
where a is real; then
¸0[ρ[0) = a, ¸0[θ[0) = 0 (14.16)
Now let us put
φ(x) = [ρ
(x) +a]e
iθ(x)
(14.17)
We regard ρ
and θ as “small” displacements from the point φ = a. Substi
tute (17) into (9)and keeping only the quadratic terms gives
L = (∂
µ
ρ
)(∂
µ
ρ
)+(ρ
+a)
2
(∂
µ
θ)(∂
µ
θ)−4λa
2
ρ
2
+O(ρ
3
)+ constants (14.18)
The coeﬃcient of the square of the ﬁeld in a freeﬁeld Lagrangian is the
mass squared of the corresponding particle. In this case the ρ
ﬁeld has a
mass given by
m
2
ρ
= 4λa
2
(14.19)
There is no term in θ
2
, so θ is a massless ﬁeld. As a result of spontaneous
symmetry breaking, what would otherwise be two massive ﬁelds (the real
and imaginary parts of φ), become one massive and one massless ﬁeld. The
diﬀerence between the two coordinates is easy to visualize. It requires no
energy to displace a particle in the θ direction, but it does cost energy to dis
place ρ
against the restoring force of the potential. The θ particle is known
230 CHAPTER 14. SYMMETRY BREAKING
as a Goldstone boson. This phenomenon is general: spontaneous breaking
of a continuous symmetry entails the existence of a massless particle. This
is the Goldstone theorem.
Try the same thing in cartesian coordinates. Introduce two real ﬁelds φ
1
and φ
2
.
φ = a +
φ
1
+iφ
2
√
2
, (14.20)
so that the vacuum expectation values of both ﬁelds are zero. Then
L =
1
2
(∂
µ
φ
1
)
2
+
1
2
(∂
µ
φ
2
)
2
−2λa
2
φ
2
1
−
√
2λφ
1
(φ
2
1
+φ
2
2
)−
λ
4
(φ
2
1
+φ
2
2
)
2
(14.21)
The φ
2
ﬁeld is massless, but the φ
1
has a mass squared of 4λa
2
, the same
as (19).
There is an analogy here to spin waves in a ferromagnet. Spin waves are
periodic variations in M. Because the forces in a ferromagnet are of short
range, it costs very little energy excite such a wave. Such a wave could carry
zero energy in the limit of inﬁnite wavelength, i.e. it’s a massless excitation.
14.3 A More Abstract Look
There is a profound result regarding symmetries known as Noether’s theo
rem, to the eﬀect that there is a conserved current associated with each gen
erator of a continuous symmetry. The proof is simple: consider a Lagrangian
composed of several ﬁelds, which we call φ
a
. Our symmetry transformation
changes each ﬁeld ay an inﬁnitesimal δφ
a
. since L does not change, we have
0 = δL =
δL
δφ
a
δφ
a
+
δL
δ(∂
µ
φ
a
)
δ(∂
µ
φ
a
)
The equations of motion are
δL
δφ
a
= ∂
µ
δL
δ(∂
µ
φ
a
)
so we can combine the two and get
0 = ∂
µ
_
δL
δ(∂
µ
φ
a
)
δφ
a
_
(14.22)
If we deﬁne
J
µ
≡
δL
δ(∂
µ
φ
a
)
δφ
a
(14.23)
14.3. A MORE ABSTRACT LOOK 231
Then (22) says that ∂
µ
J
µ
= 0. (We are also assuming that repeated a’s are
summed.) (We have already used this argument to get the electron current
in Dirac theory.)
To make this more concrete, imagine that the N ﬁelds make up a mul
tiplet,
φ ≡
_
¸
¸
¸
_
φ
1
φ
2
.
.
.
φ
N
_
¸
¸
¸
_
,
and the Lagrangian contains terms like (∂φ)
2
= (∂
µ
φ
a
)(∂
µ
φ
a
), and φ
2
=
φ
a
φ
a
, arranged in the form
L =
1
2
_
(∂φ)
2
−m
2
φ
2
¸
−c
2
(φ
2
)
2
−c
3
(φ
2
)
3
− (14.24)
(Unless otherwise stated, repeated indices are summed.) This Lagrangian
will be invariant under a “rotation,” φ
a
→R
ab
φ
b
, where R is a member of the
rotation group SO(N). (This is sometimes called an “internal symmetry.”)
We can write R = e
iθ·X
where the X
i
are the group generators. Under a
inﬁnitesimal transformation φ
a
→R
ab
φ
b
≈ (1+θ
i
X
i
)
ab
φ
b
, or in other words,
we have the inﬁnitesimal change δφ
a
= θ
i
X
i
ab
φ
b
. Equation (23) assures us
that the N(N −1)/2 currents
J
i
µ
= (∂
µ
φ
a
)X
i
ab
φ
b
(14.25)
are conserved. So there is one conserved current, J
i
, for each of the N
generators of the symmetry group.
For every conserved current there is a conserved “charge”
Q ≡
_
d
3
xJ
0
=
_
d
3
x
δL
δ(∂
0
φ
a
)
δφ
a
(14.26)
We notice that δL/δ(∂
0
φ
a
) = π
a
, the canonically conjugate ﬁeld. The equal
time commutation relation then gives the elegant result
i[Q, φ
a
] = δφ
a
(14.27)
We are now in a position to prove the Goldstone theorem. The charges
deﬁned by (26) are conserved, they are constant in time. Therefore,
[H, Q] = 0 (14.28)
232 CHAPTER 14. SYMMETRY BREAKING
Let the vacuum (or ground state in quantum mechanics) be denoted by [0).
We can always write H[0) = 0. (The vacuum must be an eigenstate of the
Hamiltonian. If it should happen that H[0) = c, we could always redeﬁne
the Hamiltonian H →H −c.) Normally the vacuum is invariant under the
symmetry transformation, e
iθQ
[0) = [0), or in other words Q[0) = 0.
In the previous examples, we had to choose a speciﬁc vacuum that was
not invariant under the symmetry transformation; i.e. Q[0) ,= 0. So what is
this state, Q[0)? Clearly, it’s a zeroenergy eigenstate of the Hamiltonian.
HQ[0) = [H, Q][0) = 0 (14.29)
I claim it corresponds to a massless particle. Consider the state
[ψ(k)) =
_
d
3
xe
ik·x
J
0
(x, t)[0) (14.30)
We make use of the fact that the momentum operator P satisﬁes
[P, φ(x)] = −i∇φ(x), (14.31)
where φ(x) could be any ﬁeld operator including J(x). Then
P[ψ(k)) =
_
d
3
xe
ik·x
PJ
0
(x, t)[0)
=
_
d
3
xe
ik·x
[P, J
0
(x, t)][0) = −i
_
d
3
xe
ik·x
∇J
0
(x, t)[0)
Integration by parts then gives
P[ψ(k)) = k[ψ(k)) (14.32)
Evidentally [ψ(k)) describes a particle with momentum k. In the limit that
k goes to zero, (30) becomes
lim
k→0
[ψ(k)) = Q[0) (14.33)
We know from (29) that Q[0) is a state of zero energy. This relationship
between momentum and energy is the signature of a massless particle.
In general there will be as many conserved charges as there are generators
in the symmetry group. For each one of these charges that does not leave
the vacuum invariant, i.e. those for which Q
i
[0) ,= 0, there will be one of
these massless Goldstone bosons.
14.4. NONRELATIVISTIC FIELD THEORY 233
14.4 NonRelativistic Field Theory
Field theory is necessary whenever particles are created and destroyed. This
is bound to happen at relativistic energies, but it can happen at very low
energy, if the “particles” represent some collective phenomena. Spontaneous
symmetry breaking can occur in solids and liquids. In fact, superconductiv
ity and superﬂuidity are outstanding examples of the mechanism. In order
to study these examples, we will have to have a lowenergy version of the φ
4
theory discussed previously. Start with the Lagrangian
L = (∂Φ)
†
(∂Φ) −m
2
Φ
†
Φ −λ(Φ
†
Φ)
2
(14.34)
When λ = 0 this gives the KlineGordon equation
(∂
2
+m
2
)Φ = 0. (14.35)
We can turn this into Schrodinger’s equation as follows. Deﬁne
Φ = e
−iωt
ϕ(x, t) (14.36)
∂Φ
∂t
= e
−iωt
_
−im+
∂
∂t
_
ϕ
∂
2
Φ
∂t
2
= e
−iωt
_
−im+
∂
∂t
_
2
ϕ
The lowenergy approximation consists in setting ∂
2
ϕ/∂t
2
= 0. Then (35)
becomes
−
∇
2
2m
ϕ = i
∂ϕ
∂t
, (14.37)
which is Schrodinger’s equation.
The same technique works on the Lagrangian, but we must be careful
about the normalization.
Φ =
1
√
2m
e
−imt
ϕ(x, t) (14.38)
First note that
∂Φ
†
∂t
∂Φ
∂t
−m
2
Φ
†
Φ =
1
2m
_
(im+
∂
∂t
)ϕ
†
(−im+
∂
∂t
)ϕ −m
2
ϕ
†
ϕ
_
(14.39)
≈
i
2
_
ϕ
†
∂ϕ
∂t
−
∂ϕ
†
∂t
ϕ
_
→iϕ
†
∂
0
ϕ
234 CHAPTER 14. SYMMETRY BREAKING
The last step required an integration by parts. The ﬁnal result is
L = iϕ
†
∂
0
ϕ −
1
2m
∂
i
ϕ
†
∂
i
ϕ −g
2
(ϕ
†
ϕ)
2
(14.40)
where g
2
= λ/4m
2
. (Relativistic is beautiful. Nonrelativistic is ugly.)
It’s often useful to write the ﬁelds in terms of polar coordinates.
ϕ =
√
ρe
iθ
(14.41)
The Lagrangian is
L =
i
2
∂
0
ρ −ρ∂
0
θ −
1
2m
_
ρ(∂
i
θ)
2
+
1
4ρ
(∂
i
ρ)
2
_
−g
2
ρ
2
(14.42)
ρ and θ should be thought of as independent ﬁelds, just like ϕ and ϕ
†
. In
this case the canonical momentum is
δL
δ(∂
0
θ)
= −ρ, (14.43)
so the equaltime commutation relation is
[ρ(x, t), θ(x
, t)] = iδ(x −x
) (14.44)
We can normalize the wave function so that
N =
_
ρ(x, t) dx = number of bosons (14.45)
then
[N, θ] = i (14.46)
In other words, number is conjugate to phase angle.
What sort of interactions do these Lagrangians describe? There is a hint
in (42). The interaction Hamiltonian is
H
int
= −L
int
= g
2
ρ
2
Since ρ represents the density of bosons, we can see that packing bosons
together costs energy, i.e. bosons repel regardless of the sign of λ. We
can learn more about this with a complicated argument from Zee I.3. Let’s
use the generating functional formalism from Fall quarter to calculate the
propagator for a Lagrangian such as (34). Look at
Z =
_
D[Φ]e
i
R
d
4
xL(Φ)
(14.47)
14.4. NONRELATIVISTIC FIELD THEORY 235
Last quarter you proved that
i
_
dx
i
e
i
2
x·A·x+iJ·x
=
_
(2πi)
N
det[A]
_
1/2
e
−
i
2
J·A
−1
·J
(14.48)
We learned to generalize equations such as this by replacing the coordinates
with ﬁelds.
_
D[ϕ]e
i
R
d
4
x(
1
2
ϕ·A·ϕ+J·ϕ)
∝ e
−
i
2
R
d
4
x J·A
−1
·J
(14.49)
Our strategy will be to introduce a new “auxiliary” ﬁeld called σ which has
the eﬀect of transmitting the repulsive between the bosons. To see how this
works, make the following substitutions in (49):
J →2ΦΦ
†
ϕ →σ A →2/λ
Then
1
2
ϕ A ϕ +J ϕ →σ
2
/λ + 2σΦΦ
†
−
1
2
J A
−1
J →2σΦΦ
†
The identity (49) becomes
_
D[σ]e
i
R
d
4
x[2σΦΦ
†
+σ
2
/λ]
= e
−i
R
d
4
xλ(ΦΦ
†
)
2
(14.50)
The right side of (50) contains the boson interaction term λΦΦ
†
, the left
side represents a σ “particle” coupling to the Φ. Using this in (47) gives
Z =
_
D[φ]D[σ] exp
_
i
_
d
4
x[(∂Φ)
2
−m
2
Φ
2
+ 2σΦΦ
†
+σ
2
/λ]
_
(14.51)
To put it another way, we have a new Lagrangian
2
L = (∂Φ)
2
−m
2
Φ
2
+ 2σΦΦ
†
+σ
2
/λ (14.52)
It contains a scalar Φ particle obeying the KleinGordon equation. We know
that the exchange of such a particle will produce a force of range ∼ m
−1
. It
is coupled to a σ particle via the 2σΦΦ
†
term, so presumably the exchange
of a σ will produce a force between two Φ’s, although it’s not clear at this
2
Zee calls the transition from (47) to (52) the HubbardStrantonovic transformation.
I will have to take his word for it.
236 CHAPTER 14. SYMMETRY BREAKING
point what the range and sign may be. To determine these things I will ﬁrst
review Zee’s argument from Section I.34 regarding the Φ.
Introduce the Fourier transform
Φ(x) =
_
d
4
p
(2π)
2
e
−ipx
˜
Φ(p). (14.53)
Then
_
d
4
x [(∂Φ)
2
−m
2
Φ
2
] =
_
d
4
p
˜
Φ
2
(p)(p
2
−m
2
) (14.54)
We have learned to extract the propagator from this integral. The argument
brieﬂy is this:
Z[J] = Z[0]e
iW[J]
= ¸0[e
−iH(T−T
)
[0) (14.55)
W[J] = −
1
2
_
d
4
xd
4
yJ(x)J(y)D(x −y) (14.56)
D(x −y) =
_
d
4
p
(2π)
4
e
ip(x−y)
p
2
−m
2
+i
(14.57)
Let’s put two delta function sources in (56) and see how they aﬀect one
another.
J(x) = δ
(3)
(x −x
1
) +δ
(3)
(x −x
2
)
W[J] will contain terms with δ
(3)
(x−x
1
)δ
(3)
(y−x
2
) and δ
(3)
(x−x
2
)δ
(3)
(y−
x
1
) representing the interaction between the two sources. Substitute them
into (56)
W = −
_
dx
0
_
dy
0
_
dp
0
2π
e
ip
0
(x−y)
0
_
d
3
p
(2π)
3
e
ik·(x
1
−x
2
)
k
2
−m
2
+i
Integrate over y
0
to get δ(p
0
), then integrate over p
0
.
W =
_
T/2
−T/2
dx
0
_
d
3
k
(2π)
3
e
ik·(x
2
−x
1
)
k
2
+m
2
(14.58)
Have another look at (55). It seems that ¸0[e
−iHT
[0) ought to be some
thing like e
−iET
where E is the energy due to the presence of the two sources
acting on one another. Since this is just a qualitative argument that should
do for now. In this spirit then −iRT = iW. (58) gives us
E = −
_
d
3
k
(2π)
3
e
ik·(x
2
−x
1
)
k
2
+m
2
(14.59)
14.5. SPONTANEOUS BREAKING OF LOCAL SYMMETRY 237
This contains two great truths: (1) a negative potential means an attractive
force, and (2) the range of the force is ∼ 1/m. The integral can be done
easily. The result is
E = −
1
4πr
e
−mr
. (14.60)
Now go back to the Lagrangian (52). We have just analyzed the free Φ
ﬁeld, and the interaction term speaks for itself, but what are we to make of
the free σ ﬁeld? There is nothing in the Lagrangian but σ
2
/λ. Let’s repeat
the steps from (52) to (57)
σ(x) =
_
d
4
p
(2π)
2
e
−ipx
˜ σ(p) (14.61)
_
d
4
x
1
λ
σ
2
(x) =
1
λ
_
d
4
p˜ σ
2
(p) (14.62)
Compare this with (54). The momentum space propagator is simply +λ.
The Fourier transform of this is the delta function. The force is repulsive
and zerorange.
14.5 Spontaneous Breaking of Local Symmetry
The situation is quite diﬀerent when the broken symmetry is a local or gauge
symmetry. Let’s revisit the model from equations (9) and (10), this time
with the transformation
φ →e
iΛ(x)
φ. (14.63)
We discussed this in the chapter on quantum electrodynamics. The La
grangian will remain invariant only if there is an additional massless ﬁeld
A
µ
. Moreover, this ﬁeld must appear in the Lagrangian through the replace
ment
∂
µ
→∂
µ
+ieA
µ
, (14.64)
and in the kinetic energy term −
1
4
F
µν
F
µν
, where F
µν
= ∂
µ
A
ν
− ∂
ν
A
µ
.
Substituting this into (9) gives
L = [(∂
µ
+ieA
µ
)φ]
†
[(∂
µ
−ieA
µ
)φ] −µ
2
φ
∗
φ −λ(φ
∗
φ)
2
−
1
4
F
µν
F
µν
(14.65)
As before, we treat φ as a complex classical ﬁeld and µ
2
as a parameter that
can be positive or negative.
238 CHAPTER 14. SYMMETRY BREAKING
The equations of motion for the A
µ
ﬁeld can be found from the Euler
Lagrange equations
∂
µ
δL
δ(∂
µ
A
ν
)
−
δL
δA
ν
= 0 (14.66)
The resulting equation is
2A
ν
−∂
ν
(∂
µ
A
µ
) = j
ν
(14.67)
where
j
ν
= iq
_
φ
†
∂
ν
φ −(∂
ν
φ
†
)φ
_
−2q
2
φ
†
φA
ν
(14.68)
For the time being, I will use the Lorentz gauge condition ∂
µ
A
µ
= 0. If we
take φ to be real with constant value
φ =
√
¯ ρ (14.69)
a simple result emerges.
2A
µ
= j
µ
= −2q
2
¯ ρA
µ
, (14.70)
which is the equation for a free vector particle of mass M =
√
2¯ ρq. Consider
for a moment the static case. (70) becomes
∇
2
B = M
2
B (14.71)
In one dimension this would give
B = B
0
e
−Mx
. (14.72)
In other words, the magnetic ﬁeld is screened out by the currents. We
might say that the magnetic ﬁeld has developed a “mass.” This is the origin
of the Meisner eﬀect in superconductors. The boson ﬁeld φ is a coherent
manybody wave function consisting of Cooper pairs.
How could this be relevant to particle physics? What is there in free
space that is like the BoseEinstein condensate in a superconductor? Per
haps an analogy will help. Maxwell noticed that in some sense, something
like current ﬂows between the plates of a capacitor when it is charging or
discharging. He therefore postulated a displacement current
J
d
=
∂E
∂t
.
14.5. SPONTANEOUS BREAKING OF LOCAL SYMMETRY 239
Is there really a current ﬂowing between the plates? It depends on what you
mean by “really.” In the same sense we will postulate vacuum screening cur
rents. They come about through the mechanism of spontaneous symmetry
breaking.
In the secondquantized version of quantum ﬁeld theory, we represent
the ﬁeld operators in terms of creation and annihilation operators.
ˆ
φ =
_
d
3
k
_
(2π)
3
2ω
k
[e
−ikx
ˆ a(k) +e
ikx
ˆ a
†
(k)] (14.73)
with the understanding that
¸0[ˆ a(k)[0) = ¸0[ˆ a
†
[0) = 0 (14.74)
i.e. the vacuum is the state in which there are no particles. (I am temporar
ily using “hats” to emphasize that we are dealing with operators rather than
classical ﬁelds.) In the path integral formulation, the vacuum was regarded
as the lowest energy state, cf. Section 2.4. The two are not necessarily the
same thing. In superconductors, for example, the ground state is one in
which all the Cooper pairs are in their lowest energy state. It would not
make sense to represent this state with a ﬁeld operator of the form (73). It
is at least plausible under these circumstances that ¸0[
ˆ
φ[0) ,= 0. The generic
classical φ
4
Lagrangian (65) we have been considering is a phenomonolog
ical model for such systems. If −µ
2
is taken as a positive parameter, the
minimum energy state appears at [φ[ = a, where
a =
_
−µ
2
2λ
(14.75)
In polar form, the ground state wave function is
ˆ
φ = [
ˆ
φ[e
i
ˆ
θ(x)
. (14.76)
Any phase is possible, the Lagrangian is still symmetric, but the system
will choose some vacuum, and this will break or hide the symmetry. The
most familiar example of this symmetry breaking is the ground state of a
ferromagnet below the transition temperature. The spinspin interaction is
invariant under rotation, but once the spins have “decided” to align them
selves in one particular direction, the symmetry is apparently broken.
Let us assume then along with the highenergy physics community, that
there is some ﬁeld
ˆ
φ called the Higgs ﬁeld that ﬁlls space and has a nonzero
240 CHAPTER 14. SYMMETRY BREAKING
vacuum expectation value. We choose the vacuum phase θ = 0. In the
vicinity of this point we can write
ˆ
φ = a +
ˆ
h (14.77)
The implication is that ¸0[
ˆ
φ[0) = a is a constant given by (75), and
ˆ
h is a
quantum ﬁeld. Inserting (77) into (67) (with ∂
µ
ˆ
A
µ
= 0) gives
2
ˆ
A
µ
= −2q
2
a
2
ˆ
A
µ
+ terms depending on h (14.78)
Compare with (70), the photon has acquired a mass M =
_
2q
2
a
2
.
This argument seems to depend on the gauge condition ∂
µ
ˆ
A
µ
= 0, so
we have lost gauge invariance. Can we have our mass and gauge invariance
too? To answer this question, ﬁrst combine (76) and (77) so that
ˆ
φ = ae
i
ˆ
θ(x)
+ terms depending on
ˆ
h (14.79)
and substitute into the operator forms of (67) and (68)(again ignoring
ˆ
h).
2
ˆ
A
µ
−∂
ν
(∂
µ
ˆ
A
µ
) = −2q
2
a
2
(
ˆ
A
ν
+q
−1
∂
ν
ˆ
θ) (14.80)
Let’s deﬁne
ˆ
A
ν
→
ˆ
A
ν
=
ˆ
A
ν
+q
−1
∂
ν
ˆ
θ (14.81)
This is just a gauge transformation after all, and we know that the left side
of (80) is gauge invariant, so
(2−M)
ˆ
A
ν
−∂
ν
(∂
µ
ˆ
A
µ
) = 0. (14.82)
Diﬀerentiating with respect to ∂
ν
reveals
∂
ν
ˆ
A
ν
= 0, (14.83)
so we can have our proverbial cake and eat it too.
(2+M
2
)
ˆ
A
ν
= 0 (14.84)
Gauge transformations are customarily written
ˆ
A
µ
→
ˆ
A
µ
=
ˆ
A
µ
−∂
µ
ˆ χ
µ
, (14.85)
so we have simply chosen the gauge ˆ χ = −
ˆ
θ/q.
Conclusions:
14.5. SPONTANEOUS BREAKING OF LOCAL SYMMETRY 241
• Setting θ = 0 in (76) breaks the rotational symmetry of the theory
but it does not break gauge invariance, because any choice of θ can be
compensated by an equivalent gauge transformation of A
µ
.
• By allowing the photon to interact with a scalar ﬁeld with a nonzero
vacuum expectation value, we enable the photon to acquire a mass.
So far, we have only looked at the freeparticle component of the theory.
We can get an idea of the interactions by substituting the complete form of
(77) into (68).
(2+M
2
)
ˆ
A
ν
= −4q
2
a
ˆ
h
ˆ
A
ν
−2q
2
ˆ
h
2
ˆ
A
ν
(14.86)
There is no such thing as a relativistic superconductor, of course, this ex
ample is only intended to illustrate the ideas behind the Higgs mechanism.
Equation (86) is intended to illustrate one ﬁnal point. There is a new par
ticle, the
ˆ
h or Higgs boson, and it couples to the photon ﬁeld in a speciﬁc
way. The mechanism not only produces the mass, it decides the form of the
interaction.
Suppose we did not set
ˆ
θ = 0 but kept it around as an additional particle?
This would change the equations of motion and hence the propagator. This
is not necessarily a bad thing, because the renormalization properties of
the theory are more transparent with other choices of the gauge function
ˆ
θ.
This is the key to the remarkable fact that gauge theories are renormalizable.
This a technical issue that will not be covered in these lectures.
In the absence of the gauge ﬁeld, the vacuum is at
[φ[ = a =
_
−µ
2
2λ
_
(14.87)
As in (20) we introduce two real ﬁelds φ
1
and φ
2
and substitute into (36).
Keeping only the quadratic terms, we have
L = −
1
4
F
µν
F
µν
+e
2
a
2
A
µ
A
µ
+
1
2
(∂
µ
φ
1
)
2
+
1
2
(∂
µ
φ
2
)
2
−2λa
2
φ
2
1
+
√
2eaA
µ
∂
µ
φ
2
+ (14.88)
Now it appears from the second term that the photon has developed a mass
m
2
= e
2
a
2
. The φ
1
ﬁeld has a mass, and φ
2
, the Goldstone boson, is
massless, but it couples to the “photon” with a derivative coupling. This
suggests that the photon could simply turn itself into a φ
2
. Let’s count the
degrees of freedom. Originally we had
242 CHAPTER 14. SYMMETRY BREAKING
2 (complex scalar ﬁeld)
+2 (transverse ﬁeld for massless photon)
4
After the spontaneous symmetry breaking we have
1 (φ
1
, a massive scalar ﬁeld)
1 (φ
2
a massless scalar ﬁeld)
+3 (massive vector ﬁeld A
µ
)
5
Something is wrong. A degree of freedom seems to have been gained in the
spontaneous symmetry breaking. This is an illusion, as can be make obvious
by an appropriate change of gauge. The φ
2
ﬁeld, however, can be removed
by means of a gauge transformation. Putting an inﬁnitesimal Λ in (34) gives
φ
1
= φ
1
−Λφ
2
φ
2
= φ
2
+ Λφ
1
+
√
2Λa (14.89)
We can always choose Λ so that φ
2
= 0. This will aﬀect the higher
order terms, of course, but for the moment we are only concerned with
the quadratic terms. Simply set φ
2
= 0 in (38)
L = −
1
4
F
µν
F
µν
+e
2
a
2
A
µ
A
µ
+
1
2
(∂
µ
φ
1
)
2
−2λa
2
φ
2
1
+ (14.90)
There are now only two ﬁelds. A massive spinone “photon” and a massive
scalar ﬁeld, the φ
1
. The Goldstone boson has disappeared. This is called
the Higgs mechanism.
Chapter 15
LNAGS and the Standard
Model
These notes are intended to summarize the last few lectures on local non
abelian gauge ﬁelds (LNAGS) and the standard model. They will not be a
substitute for your own careful notes.
We have been considering the consequences of the transformation
ψ →ψ
= e
igτ·α/2
ψ (15.1)
Here τ represent the three Pauli spin matrices, here interpreted as isospin
operators, and ψ represents some generic twocomponent spinor. The La
grangians we have used all have the “kinetic energy” term
(∂
µ
ψ)
†
(∂
µ
ψ) (15.2)
In order to keep (2) invariant under the transformation (1), desperate mea
sures are called for. We need to introduce a triplet of massless spinone
ﬁelds
W
µ
= (W
µ
1
, W
µ
2
, W
µ
3
), (15.3)
and construct with them the “covariant derivative”
D
µ
= ∂
µ
+igτ W
µ
/2 (15.4)
These ﬁelds must satisfy additional constraints so that
D
µ
ψ
= e
igτ·α(x)/2
D
µ
ψ (15.5)
One consequence of these constraints is that the W
µ
ﬁelds interact with
themselves. As a practical measure, we can derive the correct Lagrangian
243
244 CHAPTER 15. LNAGS AND THE STANDARD MODEL
and equations of motion by simply replacing ∂
u
→D
µ
wherever it appears.
The Lagrangian also has to be modiﬁed by the addition of the “freeparticle”
energy
1
F
µν
= ∂
µ
W
ν
−∂
ν
W
µ
−gW
µ
W
ν
(15.6)
15.1 Spontaneously broken global SU(2) symme
try
The W particles will eventually acquire a mass via the Higgs mechanism.
Let us ﬁrst see how the massless precursors of the W’s arise as goldstone
bosons from spontaneously broken local SU(2) symmetry. We consider an
SU(2) doublet, not of fermions this time, but of bosons.
φ =
_
φ
+
φ
0
_
=
_
(φ
1
+iφ
2
)/
√
2
(φ
3
+iφ
4
)/
√
2
_
(15.7)
If these are interpreted as particle creation and annihilation operators, φ
+
destroys a positive particle or creates a negative antiparticle, whereas φ
−
destroys a neutral particle or creates a neutral antiparticle.
The Lagrangian we choose is the generic “phi to the fourth” interaction
L = (∂
µ
φ
†
)(∂
µ
φ) +µ
2
φ
†
φ −
λ
4
(φ
†
φ)
2
(15.8)
This has a classical minimum at
(φ
†
φ)
min
= 2µ
2
/λ ≡ v
2
/2, (15.9)
which we interpret as a condition on the vacuum expectation value of φ
†
φ
¸0[φ
†
φ[0) = v
2
/2 (15.10)
Notice that (8) is invariant under the transformation (1) and also under a
separate global U(1) transformation
φ →φ
= e
−iyα
φ, (15.11)
where α is separate from α = (α
1
, α
2
, α
3
). The constant y is included for
later use. It will stand for the weak hypercharge. The full symmetry is then
referred to as SU(2) U(1).
1
“Freeparticle” is in quotes because the particles are still interacting with themselves.
15.1. SPONTANEOUSLY BROKEN GLOBAL SU(2) SYMMETRY 245
There are inﬁnitely many ways of implementing (10), all summarized by
the equation,
¸0[φ
†
φ[0) = ¸0[(φ
2
1
+φ
2
2
+φ
2
3
+φ
2
4
)[0) = v
2
. (15.12)
There is an important subtlety here. We have to choose ¸0[φ
i
[0) ,= 0 for
at least one of the ﬁelds φ
i
. Since there are four independent symmetry
transformation that leave the Lagrangian invariant, we expect four massless
ﬁelds. It is inevitable, however, that any choice we make will be invari
ant under one of these four transformations, and hence the corresponding
particle will remain massive. We choose
¸0[φ[0) =
_
0
v/
√
2
_
. (15.13)
In this case the vacuum remains invariant under the combined transforma
tion of the third component of the SU(2) transformation (1) plus the U(1)
transformation (11)
(1/2 +t
3
)¸0[φ[0) = 0 (15.14)
and hence
¸0[φ[0) →¸0[φ[0)
= exp[iα(1/2 +t
3
)]¸0[φ[0) = ¸0[φ[0) (15.15)
As usual t
3
= τ
3
/2. I am using lower case t for weak isospin. The point is
that the vacuum is invariant under (15) so the corresponding ﬁeld will not
acquire a mass.
We now need to consider oscillations about (10) in order to see the
physical particle spectrum. We parameterize these as
φ = exp(−iθ(x) τ/2v)
_
0
(v +H(x))/
√
2
_
(15.16)
Now φ is invariant under the SU(2) gauge transformations, and clearly
setting θ = 0 is such a transformation, so consider it done and write
φ =
_
0
(v +H(x))/
√
2
_
(15.17)
As a consequence of the gauge symmetry (11), the Lagrangian will con
tain an additional massless vector ﬁeld, which we call B
µ
. Our Lagrangian
now is
L = (D
µ
φ)
†
(D
µ
φ) +µ
2
φ
†
φ −
λ
4
(φ
†
φ)
2
−
1
4
F
µν
F
µν
−
1
4
G
µν
G
µν
(15.18)
246 CHAPTER 15. LNAGS AND THE STANDARD MODEL
where
D
µ
= ∂
µ
+igτ W
µ
/2 +ig
yB
µ
/2 (15.19)
G
µν
= ∂
µ
B
ν
−∂
ν
B
µ
(15.20)
and F
µν
is given by (6). Substitute (19), (6), and (20) into (18) and keep
only terms that are quadratic in the ﬁelds. We ﬁnd
L =
1
2
(∂
µ
H)(∂
µ
H) −µ
2
H
2
−
1
4
(∂
µ
W
1ν
−∂
ν
W
1µ
)(∂
µ
W
ν
1
−∂
ν
W
µ
1
) +
1
8
g
2
v
2
W
1µ
W
µ
1
−
1
4
(∂
µ
W
2ν
−∂
ν
W
2µ
)(∂
µ
W
ν
2
−∂
ν
W
µ
2
) +
1
8
g
2
v
2
W
2µ
W
µ
2
−
1
4
(∂
µ
W
3ν
−∂
ν
W
3µ
)(∂
µ
W
ν
3
−∂
ν
W
µ
3
) −
1
4
G
µν
G
µν
+
1
8
v
2
(gW
3µ
−g
B
µ
)(gW
µ
3
−g
B
µ
) (15.21)
The ﬁelds W
3
and B are mixed in a way that suggests that neither of
them are real physical ﬁelds. They can be unmixed by introducing
Z
µ
= cos θ
W
W
µ
3
−sin θ
W
B
µ
(15.22)
and
A
µ
= sinθ
W
W
µ
3
+ cos θ
W
B
µ
(15.23)
The resulting Langrangian is
L =
1
2
∂
µ
H∂
µ
H −µ
2
H
2
−
1
4
(∂
µ
W
1ν
−∂
ν
W
1µ
)(∂
µ
W
ν
1
−∂
ν
W
µ
1
) +
1
8
g
2
v
2
W
1µ
W
µ
1
−
1
4
(∂
µ
W
2ν
−∂
ν
W
2µ
)(∂
µ
W
ν
2
−∂
ν
W
µ
2
) +
1
8
g
2
v
2
W
2µ
W
µ
2
−
1
4
(∂
µ
Z
ν
−∂
ν
Z
µ
)(∂
µ
Z
ν
−∂
ν
Z
µ
) +
1
8
v
2
(g
2
+g
2
)Z
µ
Z
µ
−
1
4
F
µν
F
µν
(15.24)
where
F
µν
= ∂
µ
A
ν
−∂
ν
A
µ
(15.25)
and
tan θ
W
= g
/g (15.26)
15.2. QUARKS AND LEPTONS 247
This is the “freeparticle” Lagrangian for the vector ﬁelds. Now that
the smoke has cleared, we see the consequences of our various maneuvers.
The three W particles that were initially massless have acquired masses,
although W
3
has done so by mixing with the B particle. We are left with
a single massless particle, the A, which we identify with the photon. The
masses and coupling constants can be read oﬀ the Lagrangian. This is left
as an exercise.
Anyhow, these are the “messenger particles.” But whose messages do
they carry? The answer, of course, is the leptons and quarks. How they ﬁt
in is the subject of the next section.
15.2 Quarks and Leptons
The Dirac Langrangian is
L =
¯
ψ(i∂
µ
γ
µ
−m)ψ (15.27)
Because of our gauge hypothesis we must replace ∂
µ
→ D
µ
where D
µ
is
given by (19), but now ψ is a two component object containing the rele
vant fermion ﬁelds. We have discussed weak interaction processes in which
leptons scattered from leptons and those in which leptons scattered from
quarks. Our new Lagrangian should include both processes, so we must
construct ψ’s with both quarks and leptons. We have in hand several clues
to help us do this.
• The ψ’s should transform as doublets under the weak isospin group
SU(2)
W
.
• Leptonic transitions associated with charged currents change leptons
into neutrinos and vice versa. For example, ν
e
↔ e or ν
µ
↔ µ. This
suggests that these should be doublets of the group.
• We also talked about the quark doublet,
q =
_
u
d
c
_
=
_
u
d cos θ
c
+s sin θ
c
_
, (15.28)
where θ
c
is the Cabibo angle.
• Consider a process like e
+
+d →ν
e
+u resulting from the exchange of a
W
+
. The current that raises the quark charge from down (q = −1/3)
248 CHAPTER 15. LNAGS AND THE STANDARD MODEL
to up (q = 2/3) is
J
µ
=
g
√
2
¯ qγ
µ
_
τ
+
2
_
1
2
(1 −γ
5
)q (15.29)
=
g
√
2
¯ u
L
d
L
cos θ
c
+
g
√
2
¯ u
L
s
L
sin θ
c
The
1
2
(1 − γ
5
) projects out the lefthanded parts of the quark wave
functions
1
2
(1 −γ
5
)q ≡ q
L
(15.30)
So far as we know, there is no process that connects right and left
handed states, so we can write
J
+µ
=
g
√
2
¯ q
L
γ
µ
_
τ
+
2
_
q
L
(15.31)
We can combine charge raising, charge lowering and neutralcurrent
terms by deﬁning the operators
τ
±
2
=
σ
1
±iσ
2
2
τ
0
2
=
σ
3
2
(15.32)
With this, (31) can be written more compactly as
J
αµ
=
g
√
2
¯ q
L
γ
µ
τ
α
2
q
L
, (15.33)
which also includes the neutralcurrent contribution.
Here then is the family portrait.
leptons
_
ν
e
e
−
_
L
_
ν
µ
µ
−
_
L
_
ν
τ
τ
−
_
L
(15.34)
quarks
_
u
d
c
_
L
_
c
s
c
_
L
_
t
b
c
_
L
(15.35)
The subscript “c” on the d, s, and b remind us that these are not pure
quark states, but rather they are mixed according to a scheme that we
haven’t time to discuss. The symbols that appear in these arrays stand for
the Dirac spinors u and v with the lefthanded parts projected out. For
example “e
−
” stands for
1
2
(1 −γ
5
)u
e
(k, λ). (λ is the spin index.)
15.2. QUARKS AND LEPTONS 249
The couplings of the gauge ﬁelds to these lefthanded doublets is given
simply by
L
int
= i
¯
ψ
D
µ
γ
µ
ψ (15.36)
where ψ and ψ
stand for any of the doublets in (34) and (35), and D
µ
is
given by (19). There is a technical point regarding the W’s however. W
1
and
W
2
are real ﬁelds, and as such they don’t represent the physical particles.
The real charged particles are given by
W
±
=
W
1
±iW
2
√
2
(15.37)
The
√
2 in the denominator is an indirect consequence of the normalization
we chose in (7). As a consequence, the vertex factor that appears whenever
a W couples to a quark or lepton is
−i
g
√
2
γ
µ
1 −γ
5
2
(15.38)
Still unaccounted for are the righthanded components of the fermion
ﬁelds. There is at present no evidence for any weak interactions coupling
the the righthanded ﬁeld components and it is therefore natural that all ‘R’
components are singlets under the weak isospin group. The ‘R’ components
do interact via the U(1) ﬁeld B
µ
: it is this that allows electromagnetism
to emerge free of parity violating γ
5
terms. Just for the record then, the
right handed ﬁelds are e
R
, µ
R
, τ
R
, u
R
, d
R
, s
L
, c
R
, t
R
, and b
R
, all with weak
isospin, t = 0. We will not have time this quarter to consider any of the
interactions of these ﬁelds.
2
Contents
1 Introduction and Review 2 The 2.1 2.2 2.3 2.4 2.5 2.6 Path Integral Formulation Introduction . . . . . . . . . . . . . . . . The Feyman Path Integral . . . . . . . . The free particle propagator . . . . . . . Causality and the propagator . . . . . . Generalizing the oneparticle propagator Correlation functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 13 13 19 20 21 24 25
3 Mathematical Appendix 29 3.1 Functional Diﬀerentiation . . . . . . . . . . . . . . . . . . . . 29 3.2 Functional Taylor series . . . . . . . . . . . . . . . . . . . . . 31 4 Introduction to Field Theory 4.1 Introduction to Continuum Mechanics 4.2 Correlation functions . . . . . . . . . . 4.3 The Feyman Propagator . . . . . . . . 4.4 Interacting Fields . . . . . . . . . . . . 4.5 Momentum Space . . . . . . . . . . . . 4.6 Virtual Particles . . . . . . . . . . . . 4.7 A Word About Renormalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 35 40 44 47 52 54 56 61 61 63 66 67 70 72
5 Identical Particles and ManyParticle States 5.1 The Quantum Mechanics of Identical Particles 5.2 Boson States . . . . . . . . . . . . . . . . . . . 5.3 More General Operators . . . . . . . . . . . . . 5.4 Field Operators . . . . . . . . . . . . . . . . . . 5.5 Field Equations . . . . . . . . . . . . . . . . . . 5.6 Momentum Representation . . . . . . . . . . . 3
4 5.7 Superﬂuids . . . . . . . . . . . . . . . . . . . . 5.7.1 Bogoliubov Theory . . . . . . . . . . . . 5.7.2 The Bogoliubov Transformation . . . . Fermions . . . . . . . . . . . . . . . . . . . . . . 5.8.1 Inclusion of Spin . . . . . . . . . . . . . 5.8.2 The Fermi Sphere and the Debye Cutoﬀ Superconductors . . . . . . . . . . . . . . . . . 5.9.1 Eﬀective ElectronElectron Interaction . . . . . . . . . . . . . . . . . . . . . . . . .
CONTENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 75 77 82 84 84 86 87 95 95 97 101 110 111 115 116 119 122 127 130 133
5.8
5.9
6 Second Quantization and Relativistic Fields 6.1 Introduction to Field Theory . . . . . . . . . 6.2 Introduction to Continuum Mechanics . . . . 6.3 Introduction to Second Quantization . . . . . 6.4 Field Theory and the KleinGordon Equation 6.5 The Propagator . . . . . . . . . . . . . . . . . 7 The 7.1 7.2 7.3 7.4 7.5 7.6 Interaction Picture and the The Interaction Picture . . . . The S Matrix . . . . . . . . . . The LSZ Reduction Scheme . . Correlation Functions . . . . . The Wick Expansion . . . . . . An Example . . . . . . . . . . . SMatrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
8 Cross Sections and Decay Rates 8.1 Classical Scattering . . . . . . . 8.2 Quantum Scattering . . . . . . 8.3 Phase Space . . . . . . . . . . . 8.4 TwoParticle Scattering . . . . 8.5 The General Case . . . . . . . . 9 The 9.1 9.2 9.3 9.4 9.5 9.6 9.7 Dirac Equation The Equation . . . . . . Plane Wave Solutions . Charge Conjugation and Quantizing the Field . . The Lorentz Group . . . Spinor Representations . The Dirac Propagator .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
137 . 137 . 139 . 143 . 147 . 148 151 . 151 . 154 . 156 . 161 . 165 . 169 . 172
. . . . . . . . . . . . . . . . Antiparticles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . . . . . . . . 14. . Helicity 12. . . . .4 Lepton Number . . . .4 NonRelativistic Field Theory .4 . . . . . 14. . . . . . . . .2 13. . . . . . . . . . . . . . . . . . . . .4. 13 The 13. . . . . . . . . . . . . . . . . . . . . . . 11. . .3 A More Abstract Look . .2 Quarks and Leptons . . . . . . . . . . . . . . . . 11. . . . . . . . . . . . . .1 Gauge Invariance . . . . . . . . . . . . . . . . 178 11 Quantum Electrodynamics 11. . . . . . . . . 228 . . . . . 233 . . . . . . . . . . . . . .1 Trace Technology . . . Weak Isospin . . . . . . . . . . . 183 183 185 187 193 195 195 196 205 208 210 213 215 219 219 220 222 223 Electroweak Theory Weak interaction of quarks . . .1 Parity Violation . . . . . . . . . . . . . . . . . . . 12. . . . . . . . . . . .1 Maxwell’s Equations . . 11. . . . . . . . . . . . . . . .3 Chirality. . . . . . . . . . . . .4 The Reaction e− + e+ → µ− + µ+ 11. . . . . . . . . . . . . . . . . . . . 226 . . . . . .1 Spontaneously broken global SU (2) symmetry . .CONTENTS 5 10 The Photon Field 175 10. . . . . . . . . . . . 237 15 LNAGS and the Standard Model 243 15. . . . . . 12 Weak Interactions 12. . . . . . . . . 230 . . . . . . . . . . . . . . . . .1 13. . 247 . . . .2 Kinematics . 244 15. . . . .1 A Simple Example . . . . . . . . . . . . . . . . . . . . . 175 10. . . . . 14. . . . . . . . 14 Symmetry Breaking 14. . The Intermediate Vector Bosons The Feynman Rules . . . . . .2 Neutrinos . . . . . .2 Quantization in the Coulomb Gauge . . .2 Goldstone Bosons . . . . . . . . . . . . . . . . . . . . . . .5 Introduction to Renormalization . . . . . . . . . . . . . . . . . . . . . . 11. . . . . . . .3 Feynman’s Rules for QED . . . . . . . . . . . . . .5 Spontaneous Breaking of Local Symmetry 225 . . . . . . . 14. . . . . . . . . . . . . . .3 13. . . . . . . . . . . . . . . . . . . . . . 12. and Electrons . . 11. . . . . . . . .4. . . . . . . . . . . . . . .2 Noether’s Theorem . . . . . . . . .
6 CONTENTS .
given a system with some set of initial conditions. however.3) These are called “stationary stattes” because there is no time dependence except for the unobservable phase factor. Quantum mechanics gives us a deceptively simplelooking solution to this problem. We simply expand the wave function in a series of energy eigenstates. suggests a way of calculating time dependence. We deﬁne energy eigenstates (with no time dependence) as ˆ Hφn (x) = En φ(x) (1. i. eigenstate of the complete Hamiltonian. each one of winch has trivial time dependence. it’s called Schrodingers’ equation.1) ˆ ˆ ˆ The equation is not very useful because H = T + V is the sum of two non1 There is the trivial case in which the state is an commuting operators. If everything were that simple there would be nothing to live for. If ˆ Hψn (x. t) = En ψ(x. describe its subsequent behavior. The following calculation is done in every introductory quantum text.4) (1. The fact that such states exist. both classical and quantum.Chapter 1 Introduction and Review The business of mechanics.e. t) then ψn (t) = e−iEn (t−t ) ψn (t ) (1. ψ(t) = e−iH(t−t ) ψ(t ) ˆ (1. 7 . is to predict the future.2) 1 I will use “hats” on symbols when I want to call attention to the fact that they are operators.
t) = n cn φ(x)e−iEn t (1.10) Obviously. x . t. t) = (1. CHAPTER 1. t ) = cm e−iEm t m I have assumed that dxφ∗ (x)φn (x) = δmn m Now reinsert (6) into (7) ψ(x.11) we are implicitly working the Schrodinger picture. K(x. This would be a good time to review the diﬀerence between the Schrodinger picture and the Heisenberg picture in elementary quantum mechanics.7) φ∗ (x )φn (x)e−iEn (t−t ) ψ(x . (There is no subscript on the other ket .9) ψ(x. I have emphasized this by putting the subscript S on the ket on the left. and these are usually exotic transcendental functions. When we write an equation like. t ) n (1. I will redo the calculation above with a more abstract notation. INTRODUCTION AND REVIEW ψ(x.6) (1. whereas the operators do not. For one thing.” Equation (9) is a complete inprinciple solution to the problem of time development. there are only a few potintials for which we can ﬁnd exact eigenfunctions.8) The quantity in brackets is called the propagator. the notion of a propagator as deﬁned implicitly by (10) is very powerful. The worst part of it is that there is no way to make systematic approximations. t) = dx n (1. Nontheless.8 Now for some t < t. t. x .5) dx φ∗ (x ) ψ(x . t ) ≡ n φ∗ (x )φn (x)e−iEn (t−t ) n dx K(x. Everything we will do in the next few chapters has to do ket in some way with ﬁnding propagators. the propagator has the ability to “move the wave function along in time. but it’s still not very useful. In order to keep the discussion on familiar ground (for a little while). t )ψ(x . The state vectors evolve in time. ψ(t) S = e−iHt ψ(0) (1. t ) (1.
14) Deﬁne so that ˆ ˆ ˆ ˆ QH (t) = eiHt QS e−iHt . We believe . ˆ ˆ ˆ ˆ Q(t) = ψS (t)QS ψS (t) = ψ(0)eiHt QS e−iHt ψ(0) (1. t H H = qq.9 for reasons that will soon be clear. ˆ QS q ˆ ˆ S ˆ (1. QH (t) dt ˆ Now look at an eigenstate of Q. but in fact. This is not something we can prove mathematically. then the Heisenberg operators satisfy their own version of Schrodinger’s equation.” There is something paradoxical about them. and the eigenstates themselves constitute a complete set of states in terms of which any arbitrary state vector can be expanded.19) ˆ eiHt QS e−iHt eiHt q Deﬁne so that eiHt q ˆ QH q. We are used to saying the Heisenberg states do not change in time. ˆ dQH (t) ˆ ˆ = i H. but of course. t These are the “Heisenberg eigenstates. The eigenvalue spectrum of this operator corresponds to all possible results of measurements of the observable.12) (1. among other things. that to every observable quantity there corresponds a Hermitian operator. t ˆ S = qeiHt q ≡ q.18) (1.16) (1. For this reason we often write ψ(0) = ψ . The Creed of Quantum Mechanics says.) Operators in the Schrodinger picture usually don’t depend on time. last year?) just doesn’t come up. the eigenstates of Heisenberg operators transform “backwards” in time. ˆ Q(t) = ψ(0)QH (t)ψ(0) All time dependence is now gathered into the operator.15) = qq S S ˆ S (1.17) (1.13) (1. If the Hamiltonian itself doesn’t depend on time. The question of exactly what time it was when t = 0 (last week. we make no distinction between Schrodinger states and Heisenberg states at this time. their expectation values do. Since the two terms on the right of (12) are identical for t = 0.
is also. so we can take advantage of completeness by writing ˆ I = dxx x (1. ˆ Xx = xx (1. INTRODUCTION AND REVIEW in creeds. and for this and other reasons having to do with the role of space and time.26) . For example ψ(x.10 CHAPTER 1. we will use the Heisenberg picture almost exclusively in what follows. however. one of the independent variables in Schrodinger’s equation.21) Students often ﬁnd this troubling. The problem largely disappears in the Heisenberg picture.22) Notice that wave functions are really scalar products and as such do not change between the Heisenberg and Schrodinger pictures. and the notion of a “time eigenstate” is bizarre! The fact that position has this ambiguous dual role is one of the drawbacks of Schrodinger theory.20) The eigenvalue spectrum is continuous. The variable x. t)x S (1. Certainly the position of a particle is something that can be measured. Position eigenstates can be thought of as a notational device for switching back and forth between the elementary formulation of quantum mechanics in terms of wave functions and the more abstract formulation in terms of bras and kets. because life wouldn’t make sense otherwise. t) = x. in addition to being an observable quantity. so there must be a position operator with eigenstates. We’ll assume that the states are complete and normalized so that ˆ I= n n (1. tψ H = x ψ(t) S (1. And yet – we never think of time as an observable.23) Now we can get back to the business of calculating propagators.24) The symbol n stands for all the discrete quantum numbers required to specify the eigenstate. and for good reason. along with time. We can reverse the procedure as follows: ψ(t) S = dxψ(x.25) n e −iH(t−t ) =e −iH(t−t ) n n n = e−iEn (t−t ) n n (1. Deﬁne energy eigenstates as ˆ Hn = En n (1.
t ) to (x . t. t.29) This is the starting point for the path integral formulation in the next section. t = dx K(x. This gives us an alternative derivation of (8). t. t x . t . t ) = dx x. t) = xψ(t) = xe−iH(t−t ) ψ(t ) = n ˆ e−iEn (t−t ) xn nψ(t ) = dx n e−iEn (t−t ) xn nx ψ(x . = x. tx . K(x.31) This is the “composition property” of propagators. x t ) = xe−iH(t−t ) x = xe−iHt eiHt x Insert this in (29).27) This is identical to (8) if we remember that xn = φn (x). t) integrated over all x . So one last time · · · xψ(t) = dx xe−iH(t−t ) x x ψ(t ) (1.30) K(x. t ) to (x. t ) time the probability of propagating from (x . ψ(x. x . Before going on. ˆ I= dxx x = dx eiHt x xe−iHt = dx x. . t )K(x . We see the propagator lurking inside the integral. x . It implies that we can always insert a complete set of Heisenberg position eigenstates using (30). tx.28) Compare this with (10). t) is equal to the probability of propagating from (x . t x . it will be necessary to calculate the propagator without using energy eigenstates. t ) to (x. t ) (1. Physically it means that the probability for a particle to propagate from (x . In order to take the next step toward the path integral formulation of quantum mechanics.11 This is sometimes called the “spectral representation” of the operator e−iH(t−t ) . t (1. t ) (1. we need one allimportant result. tx . x . t (1.
12 CHAPTER 1. INTRODUCTION AND REVIEW .
The paths can be wildly unphysical. Perform the following sum.Chapter 2 The Path Integral Formulation 2. x◦ . The real question is how to do ¯ the sum over paths. more about this later. tf ). t◦ ) = A paths The constant A is a normalization factor. one tries to calculate the time dependence of the wave functions directly using Schrodinger’s equation.1 Introduction In the usual formulation of quantum mechanics. the kinetic energy minus 13 h eiS[x(t)]/¯ (2. tf . tf . The path integral formulation seeks to calculate the propagator K(xf . K(xf . This will require learning some new math. There are no other restrictions. The procedure follows: 1. We start with the Lagrangian L = T − V . I have put in the h to illustrate the classical limit. t◦ ). Find the classical action S[x(t)] for each path x(t). functional integration. 3. By “causal” I mean that the paths must not loop back in time. x◦ . First let’s do a brief review of classical mechanics. The classical equations of motion can be derived from Hamilton’s principle. Draw all causal paths in the xt plane connecting (x◦ .1) . 2. t◦ ) with (xf .
Let’s see how that works classically in one dimension We know the answer ahead of time (xf − x◦ ) x(t) = x◦ + (t − t◦ ) (2.3) We will be much concerned with the propagation of free particles. q1 · · · qn )dt ˙ ˙ (2. Your favorite classical mechanics text will now prove that this statement is equivalent to a set of n EulerLagrange equations. the “right” and “wrong” action would diﬀer by only 1/6 ergsec. THE PATH INTEGRAL FORMULATION the potential energy. the particle somehow follows all paths (that don’t involve propagation backwards in time). say x = t2 then 1 S= 0 1 2 m(2t)2 dt = m.7) The units of S must be ergseconds. those with action much diﬀerent from the classical action will tend to cancel one another through destructive interference. if the particle follows an unphysical path. Suppose a classical particle travels 1 cm in 1 second.. q] = ˙ tf t◦ L(q1 · · · qn . but remember that in . In quantum mechanics. Let’s put in some numbers. 2 2 (2. Hamilton’s principle says that the actual path followed by the system is the one that makes the action an extremum. the particle follows just this one path. i. Calculate the action. but as can be seen from (1). Terms close to the classical path add coherently.14 CHAPTER 2. They are functions of time.4) (tf − t◦ ) S= tf t◦ m 1 mx2 dt = ˙ 2 2 tf t◦ xf − x◦ tf − t◦ 2 dt = m (xf − x◦ )2 2 (tf − t◦ ) (2. and these functions qi = qi (t) collectively describe the path taken by the system. 2 3 (2. a maximum or a minimum. Each choice of functions will yield a numerical value for the action.5) In classical mechanics. We say that it is a functional of the coordinates and their derivatives. ∂L d ∂L − =0 dt ∂ qi ∂qi ˙ (2. If the mass were one gram. S[q.6) However. If it starts at the origin and follows the “right” path x = t in the appropriate units 1 S= 0 1 1 m(1)2 dt = m.e.2) The qi ’s are the generalized coordinates.
(2. tj .9) but remember that x0 = x◦ and xn+1 = xf . To put it in less biblical terms. tn xn . Consequently the wrong path ¯ 26 radians! On the other hand. but there are a few technicalities regarding normalization and boundary conditions that will repay a brief degression. tf ) We now use the composition property (1. The postulate (1) is usually attributed to Feynman1 who claimed that he discovered it in a moment if inspiration during a high school physics class. t◦ (2. Clearly the classical particle must follow the straight and narrow way.11 × 10−28 grams. tf xn . tj+1 xj .10) The ﬁrst step makes use of (1. but this is a strange limit in which there are an inﬁnite number of integrations! We now work on the shorttime propagator.2. xj+1 .7 degrees. INTRODUCTION 15 (1) the action is divided by h = 10−27 ergsec. t◦ = dxn dxn−1 · · · dx1 xf . S ¯ h ¯ gives the classical limit. tf x◦ . We will divide the interval between t◦ and tf into n + 1 equal slices bounded by t◦ < t1 < t2 < · · · < tn−1 < tf . The key idea is to discretize time. tn−1 · · · x1 . We will eventually take the limit n → ∞ to represent all smooth paths. At this point I would like to insert a complete set of momentum eigenstates. tf x◦ . most of these ideas were published earlier by Dirac. We represent the path of the particle as a set of n + 1 straightline segments connecting the spacetime points. the rest of us can prove.8) We will use the more compact notation. xf .31) n + 1 times. tj = xj+1 e−iHτ xj ≈ xj+1 (1 − iHτ xj (2. I don’t know if Feynman was aware of his work. . tj+1 xj .1.29). t1 ) → (x2 .e. t1 x◦ . t2 ) → · · · → (xn . (x◦ . 1 Actually. though the proof will take many pages. if we are talking about is out of phase by 10 an electron with a mass of 9. i. t◦ ) → (x1 . t◦ = i=1 dxi j=0 xj+1 . tn ) → (xf . then the wrong path is out of phase by only 8. What Feynman could intuit. n n xf . tn xn−1 . The second step takes advantage of the fact that τ = tj+1 − tj is very small. and how narrow the path must be is determined by the extremely small size of h.
Let’s choose xp = ξp (x) xpn = ξn (x) (2. When we make the transition to continuum states according to (14) this becomes x x = The point is that whereas ˆ dxx x = I.12) (2. THE PATH INTEGRAL FORMULATION The momentum eigenvalue equation in the Schrodinger picture is ˆ P ξp (x) = p ξ(x) ξp ∝ eipx (2. You can go back and forth between these two representations with the following “cookbook” recipe. eipn x ξn = √ ↔ ξp = eipx L In the language of bras and kets ˆ P p = pp (2.17) So far. in this case 1 ξn = √ eipn x L The normalization is chosen so that L 0 ∗ dx ξn ξn = δnn pn = 2πn L (2. In terms of discrete states this requires x x = n x pn pn x = n ∗ ξn (x )ξn (x) = δ(x − x ) (2.14) but we have to pay some attention to the normalization of p .15) ↔ n 1 2π ∞ dp −∞ (2.18) . ξ(x) = ξ(x + L).13) If we don’t use periodic boundary conditions.16 CHAPTER 2. the eigenvalue spectrum is continuous.16) We would also like x x = δ(x−x ). 1 2π dp x p px = δ(x − x ).11) We often make the eigenvalue spectrum discrete by imposing periodic boundary conditions. so good. (2.
tj = Two points to notice: • The Hamiltonian contains no operators. . xj+1 V (x)xj = V (xj ) xj+1 xj = V (xj ) Combine (10). Use the fact that τ is small so that [1 − iτ H] ≈ e−iτ H . tj+1 xj . there will no longer be operators in our formalism.24) 1 2π dpj eipj (xj+1 −xj ) [1 − iτ H(pj .23) 1 2π dpj eipj (xj+1 −xj ) (2. xj+1 .21) p2 1 j dpj eipj (xj+1 −xj ) 2π 2m The potential energy term is simpler. xj+1 .2. (21) and (22). xj )] (2.19) It is this last expression that we need for the shorttime propagator.22) Substitute (24) into (9).1. When we have done that. It is the classical Hamiltonian. tj = 1 2π dpj exp {i [pj (xj+1 − xj ) − τ H(pj . xj )]} (2. • The symbol pj does not imply discrete momentum values. They will have been replaced by classical or cnumber variables. tj+1 xj . The subscript means that pj is being used to help do the integrals associated with the tj time slice.20) Our strategy will be to use position and momentum eigenstates to replace the momentum and position operators with their eigenvalues. INTRODUCTION to be consistent we must have 1 2π ˆ dpp p = I 17 (2. Let’s do the kinetic energy term ﬁrst. The Hamiltonian in (10) can be written ˆ P2 ˆ ˆ H= + V (x) 2m (2. xj+1  ˆ 1 P2 xj = 2m 2π = dpj xj+1  p2 j pj pj xj 2m (2.
j.29) Equations (25) through (29) not only prove Feynman’s conjecture (1). one evaluates (25) term by term. two products and a sum. Concentrate on the sum inside the exponential. and l. each consisting of an integral over one of the x’s and one of the p’s. To this end. The ﬁnal result will be an expression that depends on n. tf x◦ .25) The notation in the last line is potentially confusing. xl )] (2.18 CHAPTER 2. t◦ = n dxi i=1 n j=0 = i=1 j=0 1 2π dpj exp i l=0 [pl (xl+1 − xl ) − τ H(pl . xj )]} n xf .28) xf . n n→∞ lim τ pl l=0 xl+1 − xl tl+1 − tl − H(pl . Equation (25) can now be collapsed as follows: n n→∞ lim j=0 1 2π n ∞ −∞ dpj ≡ D[p] (2. I have used three independent indices. so we are usually working with approximations.26) Yes! L is the classical Lagrangian and S is the classical action. i. . t◦ = D[x]D[p]eiS (2. since there are three indexed operations. In order to calculate a propagator. tf x◦ . One then takes the limit n → ∞ in the hope that the limit is welldeﬁned.27) n→∞ lim dxi ≡ i=1 D[x] (2. The intent is to describe an equation with n factors. there are only a few simple potentials for which this agenda can be carried through exactly. xl ) = tf t◦ dt(px − H) = ˙ tf t◦ L=S (2. Unfortunately. THE PATH INTEGRAL FORMULATION n dxi n 1 2π dpj exp {i [pl (xj+1 − xj ) − τ H(pj . They are the most general formulas for evaluating path integrals. they also tell us what it means to sum over paths. note that something remarkable happens when we take the limit n → ∞.
you have to get used to this sort of thing. In the limit that n → ∞ the multiplicative constant is m n+1 2 N = lim (2. which we will do in the next section. This is one of them.33) n→∞ 2πiτ This diverges like ∞∞ . In the simplest case. Eventually we will do functional integrals just as we do ordinary integrals – by following a set of rules we have memorized. n xf . Let’s go back to the integral in (24). THE FEYMAN PATH INTEGRAL 19 The mathematics before the limit n → ∞ is conventional math. H = p2 /2m + V (x). which is some simple quadratic function of the momenta. 2.32) where ∆xl = xl+1 − xl . this constant always factor out or disappears in some miraculous way.30) The gaussian integral comes in many diﬀerent forms. tf xi . do what needs to be done. All .2.2. and then retake the limit. which is only a function of the coordinates. In order to simplify the notation. In order to derive the rules we must go back to (25). The formula we need here is ∞ −∞ dp exp i(−ap2 + bp + c) = π exp ib2 /4a + ic ia (2. I will drop all the subscripts. (2. Our ﬁrst example of this is the Feynman propagator. the equations are in the new arena of functional integration. After the limit. As you will see when we calculate things that are physically relevant. dp exp i p∆x − τ p2 − τV 2m (2. The resulting expression is called the Feynman propagator. ti = i=1 dxi m 2πiτ n+1 2 n exp iτ l=0 m 2 ∆xl τ 2 − V (xl ) . Equations like (25) can be done with ordinary calculus without much regard to rigor. but no matter.2 The Feyman Path Integral Most nonrelativistic Hamiltonians consist of a kinetic energy term. and a potential energy term.31) We use this to evaluate (30) and substitute the result back into (25). When this is true the momentum integrations in (25) or (27) can be done once and for all.
dx1 · · · dxn exp iπτ im exp (x2 − xi )2 m 4τ (2. Everything else is approximation. So with all these caveats. because of the oscillating exponential. ti = N This is the Feynman path integral. tf xi . Let’s start with the free particle. the limit should be the last step in the calculation after all the troublesome factors have eaten each other. but the approximations are all based on these two cases. The crux of the matter is this integral. im (xf − xn )2 + (xn − xn−1 )2 + · · · + (x1 − xi )2 2τ (2. the free particle and the harmonic oscillator.20 CHAPTER 2. Watch what happens when we integrate over dx1 using (31). Go back to (32) and set V = 0. is not a bug but a feature. D[x]eiS (2. xf .37) im (x2 − x1 )2 + (x1 − xi )2 2τ . let’s take the limit n → ∞ with our ﬁngers ﬁrmly crossed behind our backs. We need to study them carefully for this reason. This. We could make it converge unambiguously either by deforming the contour on integration slightly into the complex plane or by adding a small convergence factor in the exponential.3 The free particle propagator So far as I know.36) After you have done n of these integral you will be left with dx1 exp = xf . as the engineers would say. If you are trying to be rigorous. ti = im(xf − xi )2 m exp 2πi(tf − ti ) 2(tf − ti ) (2.34) 2. THE PATH INTEGRAL FORMULATION this is to say the the limit n → ∞ is tricky. tf xi . there are only two problems for which this last integral can be done exactly. The way we choose to do this will determine whether the particle propagates forwards in time or backwards in time or some combination of the two We will return to this issue presently. The integral (31) doesn’t really converge to anything. There is another bit of math that we have ignored so far.35) This is another example of the famous gaussian integral.
2.4. CAUSALITY AND THE PROPAGATOR Exercise: Do the calculation and conﬁrm this result. There are several points here to contemplate.
21
• We never had to take the limit n → ∞, because by the time we had done all the integrals, there were no n’s left. • By the same token, there are no divergent factors. • Look at our example (5). The argument of the exponential is just the classical action, so xf , tf xi , Ti = CeiS (2.38) • We could have gotten the same result directly from (1.6). The point is, I guess, that for free particles, energy eigenstates are also momentum eigenstates, so the sum in (1.6) it trivial. • The propagator (37) seems to allow propagation backwards in time. This is not surprising, since we could have replaced t → −t in any of our derivations and gotten more of less the same results. As hinted previously, this ambiguity in the direction of time is related to the ambiguity associated with the convergence of integrals with oscillating phase factors. This phenomenon is not peculiar to path integrals or even to quantum mechanics. One encounters the same problem calculating Greens’ functions for the propagation of electromagnetic waves. The free particle propagator is just simple enough to allow us to investigate these issues carefully in the next section.
2.4
Causality and the propagator
Equation (1.9) gives a simple formula for the freeparticle propagator in terms of the energy eigenfunctions. In our current notation, xf , tf xi , ti =
n
φ∗ (xi )φn (xf )e−iEn (tf −ti ) n
(2.39)
In the case of a free particle, energy eigenfunctions are also momentum eigenfunctions, since p2 ˆ Hφn = n φn . (2.40) 2m We can use the momentum eigenfunctions deﬁned in (12)and (14) for our φ’s, 1 φn (x) = √ eipn x → eipx , (2.41) L
22
CHAPTER 2. THE PATH INTEGRAL FORMULATION
where the arrow represents the continuum limit. Replace the sum in (39) with an integral as in (14) xf , tf xi , ti = 1 2π dp ei(xf −xi ) e−i 2m (tf −ti )
p2
(2.42)
This can be integrated immediately using (31) to get (37), but for the purposes of discussing causality, it is better to get rid of the time and space variables by means of a Fourier transform. Propagators are closely related to Green’s functions. For example, we can deﬁne a “retarded” Green’s function as GR (xf , tf ; xi , ti ) ≡ −iθ(tf − ti ) xf , tf xi , ti =− i 2π dp θ(tf − ti ) exp i − p2 (tf − ti ) + p(xf − xi ) 2m (2.43)
I put in the step function “by hand” so that the Green’s function is zero for all ti > tf . The −i factor is a convention, but there is a good reason for it as you will see eventually. In order to save ink, I will set xi and ti equal to zero and abbreviate GR (xf , tf ; 0, 0) ≡ G(x, t) (2.44) Eliminate the x variable with a Fourier transform. dx e−ip x G(x, t) = −iθ(t)e−iEp t ≡ G(p , t) Finally, take the Fourier transform with respect to time. dt eiωt G(p, t) = −i
0 ∞
(2.45)
e−i(E−ω)t dt ≡ G(p, ω)
(2.46)
The last integral is not well deﬁned. There are several strategies for dealing with it. The simplest is to add a small negative imaginary part to the energy, E → E − iδ. Then 1 (2.47) G(p, ω) = ω − E + iδ To see the signiﬁcance of the “fudge factor,” Fourier transform back into conﬁguration space. G(x, t) = 1 (2π)2 dp dω ei(px−ωt) ω − Ep + iδ (2.48)
2.4. CAUSALITY AND THE PROPAGATOR
23
The ω integration can be done as a complex contour integral. For t < 0, the integration contour closes in the upper half ω plane and the integral is zero. When t > 0, the contour closes in the lower half plane and the integral is just −2πi times the residue of the pole at ω = E − iδ. This displacing the pole slightly below the real axis gives the retarded Green’s function, i.e. the one that vanishes when t < 0. Obviously, we could get the advanced Green’s function by displacing the pole upward. Here is an assortment of points that you should absorb from the preceding derivation: • Compare the formula for the propagator in conﬁguration space, equation (37), with the Fourier transform of the Green’s function, (47). Not only is (47) much simpler mathematically, its causality properties are unambiguous and obvious. (The −i factor in (43) was inserted to make (47) as simple as possible.) This is a recurring theme in ﬁeld theory. Equations of motion and path integrals always start out in conﬁguration space, but the results are simpler in momentum space. • Integrals over momentum like (42) tend to be mathematically ambiguous because the integrand oscillates inﬁnitely rapidly as p → ∞. They are physically ambiguous because they could describe propagation backward in time just as well as propagation forward in time. We can remove this ambiguity in several ways. In (43) we simply put in a theta function. This makes causality explicit, but is clumsy to work with. A better scheme is to give energies a small negative imaginary part, or alternatively, just remember that the propagator in momentum space always has the form (47). This apparently ad hoc scheme removes the physical ambiguity and the mathematical ambiguity simultaneously, because they are really two manifestations of the same problem. • In nonrelativistic quantum mechanics based on Schrodinger’s equation, cause and eﬀect always propagate “forward” in time. The point is that the equation is second order in space and ﬁrst order in time, so that if you replace t → −t in Schrodinger’s equation, you get a diﬀerent equation. I emphasize this because relativistic theories are profoundly diﬀerent; space and time coordinates (except for that damnable minus sign) must be handled in the same way. The KlineGordon equation, for example, is second order in space and time. The Dirac equation is ﬁrst order in both variables. Such theories do not have a welldeﬁned sense of past and future. The propagators must therefore account for
24
CHAPTER 2. THE PATH INTEGRAL FORMULATION both forward and backward propagation. I’ll have much more to say about this later. • If you replace t → −it in Schrodinger’s equation, you get something that looks like the equation for heat ﬂow,
2
φ−a
∂φ =0 ∂t
(2.49)
This is signiﬁcant for two reasons. First, we all know that heat ﬂows from somewhere hot to somewhere colder. The second law of thermodynamics deﬁnes an unambiguous direction of time. Schrodinger’s equation has this kind of causality built into it. Second, the transformation t → −it is called a Wick rotation. It is a profound and puzzling fact that the Wick rotation always turns quantum mechanics into thermodynamics. You have just seen one example of this. There are more to follow. • You doubtless learned in Middle School that the Green’s function is the ﬁeld due to a unit impulse function. You should convince yourself that G(x, t) is no diﬀerent, i.e. ∂ ˆ H −i ∂t G(x, t) = −δ(x)δ(t) (2.50)
2.5
Generalizing the oneparticle propagator
We have been working on the propagator xf , tf xi , ti = xf e−iH(tf −ti ) xi (2.51)
Remember that the states on the left are timedependent position eigenstates in the Heisenberg picture. The states on the right are timeindependent Schrodinger states in which there is one particle that happens to be at xi or xf . Let’s think through a series of generalizations. First, we replace x with a generalized coordinate q. We can measure this coordinate, so by the Central Dogma of Quantum Mechanics, there must be some corresponding operator, eigenvalues and eigenstates. ˆ Qq = qq (2.52)
2.6. CORRELATION FUNCTIONS We could repeat all our derivations with q replacing x and get qf , tf qi , ti = qf e−iH(tf −ti ) qi
25
(2.53)
Now that we have introduced one generalized coordinate, there is no reason why we can’t have N of them. The classical Hamiltonian is H = K(p1 · · · pN ) + V (q1 · · · qN ) where pj = ∂L ∂ qj ˙ (2.54)
(2.55)
Again, repeat all the derivations. This time replace the single integrals with multiple integrals dpj → dqj → ··· ··· dpj1 · · · dpjN dqj1 · · · dqjN (2.56)
This does not introduce any additional problems. In many respects we seem to be doing classical Langrangian mechanics with N degrees of freedom. In order to do ﬁeld theory, however, we must pass to the limit of continuum mechanics, which is to say, mechanics with an inﬁnite number of degrees of freedom. In this case the coordinates qj are replaced with displacement ﬁelds φ(x). The motivations for doing this and the mathematics involved are the subjects of the next chapter.
2.6
Correlation functions
It turns out that many of the quantities that will be of interest to us have the form ˆ ˆ ˆ F T Q(t1 )Q(t2 ) · · · Q(tn ) I (2.57) ˆ The Q operators were explained in the previous section. The states I and F  are shorthand notation for the initial and ﬁnal states. The notation T [· · · ] introduces the timeordered product. It means that the operators in the brackets must be arranged so that operators evaluated at earlier time
26
CHAPTER 2. THE PATH INTEGRAL FORMULATION
always stand to the right of operators evaluated at later times. for example, in the case of a single particle, ˆ ˆ ˆ ˆ F T [Q(t1 )Q(t2 )]I = qf , tf Q(t1 )Q(t2 )qi , ti θ(t1 − t2 ) ˆ ˆ ± qf , tf Q(t2 )Q(t1 )qi , ti θ(t2 − t1 ) The ± sign anticipates some arcane math that will be necessary when we deal with fermion operators, which anticommute. For the time being, this will be a plus sign. As a mnemonic for remembering (58), recall that quantum mechanics reads like Hebrew, from right to left, so that tf > t2 ≥ t1 > ti . I will anticipate the next section by saying that we will usually evaluate (57) in the limits, ti → −∞ and tf → +∞. This has the eﬀect of projecting out the ground state or lowestenergy state of the system, so that we are really concerned with ˆ ˆ ˆ ΩT Q(t1 )Q(t2 ) · · · Q(tn ) Ω , (2.59) (2.58)
where Ω and Ω are the initial and ﬁnal ground states. I would prefer to do this one step at a time. First I will show how to evaluate (57) , and in a later section, I’ll show the signiﬁcance of the limits ti → −∞ and tf → +∞. We start by evaluating (57) with a single operator. Transforming it into the Schrodinger picture gives ˆ ˆ qf , tf Q(t)qi , ti = qf e−iH(tf −t) Qe−iH(t−ti ) qi . (2.60)
ˆ Now insert a complete set of states just to the right of Q and take advantage ˆ = qq . of the fact that Qq = dq q qf e−iH(tf −t) q qe−iH(t−ti ) qi = dq q(t) qf , tf q, t q, tqi , ti
In this last expression we have two ordinary propagators qf , tf q, t and q, tqi , ti . Each one could be written as in (32) or (34). This is just as if we were calculating qf , tf qi , ti and stuck in one “extra” time slice at t between ti and tf . Repeating the steps that led to (34) we have ˆ qf , tf Q(t)qi , ti = D[q]q(t)eiS (2.61)
Exercise: Verify this. Pay careful attention to the normalization factor.
2.6. CORRELATION FUNCTIONS This procedure can be repeated as many times as we like, so ˆ ˆ ˆ qf , tf T Q(t1 )Q(t2 ) · · · Q(tn ) qi , ti = D[q]q(t1 )q(t2 ) · · · q(tn )eiS
27
(2.62) Notice that there is a timeordering operator on the left side of the equation and none on the right. One way to look at this is that the q’s commute, so the order in which they are written makes no diﬀerence. A better way to look at it is that when we insert sets of complete sets as in (60), we will always put them in in such a way that tf > tn > · · · > t2 > t1 > ti (in order to prevent backwards propagation in time). Thus the q(ti )’s are automatically time ordered. Path integrals are notoriously diﬃcult to do, so what hope do we have of evaluating an integral like (62) with all those additional q(t)’s? The answer is that if we can evaluate it exactly without the q’s, then we can do it with the q’s almost trivially! If we can get an approximate answer without the q’s, then to the same degree of approximation, we can evaluate it with the q’s. In order to understand how this trick works, look at the simpler integral
∞ −∞
dq q n e− 2 aq
1
2
We know that
∞ −∞
dq e
− 1 aq 2 +Jq 2
=
2π a
1 2
eJ
2 /2a
Diﬀerentiating this n times with respect to J,
∞ −∞
dq q e Finally
n − 1 aq 2 +Jq 2
dn = dJ n
∞ −∞
dq e
− 1 aq 2 +Jq 2
dn = dJ n
1 2
2π a
1 2
eJ
2 /2a
∞ −∞
dq q e
n − 1 aq 2 2
dn = lim J→0 dJ n
2π a
eJ
2 /2a
(2.63)
You see the point? Integration is hard. Diﬀerentiation is easy. Using the “J trick,” we can do diﬃcult integrals by diﬀerentiating! Now go back to (???) and put in the J. z[J] ≡ qf , tf qi , ti
J
≡
D[q] exp i
tf ti
dt [L(t) + J(t)q(t)]
(2.64)
I can always get the propagator by taking the limit J → 0, but in the meantime, I can put as many q’s as I like in the integrand by diﬀerentiating
28
CHAPTER 2. THE PATH INTEGRAL FORMULATION
(64) with respect to J. z[J] is called a generating function for this reason. This is more than just a trick as you will see later on In some theories J has a clear physical interpretation as a source of particles. Before we can use (64) we must dispose of a technical diﬃculty. In (63), J is a parameter, In (64) it’s a function of t buried inside a deﬁnite integral. This requires a new technique called functional diﬀerentiation, I have put this material in an appendix labelled Chapter 3, to which I will be referring from time to time. This would be a good time to study it, We can use (3.5) to get our ﬁnal result: ˆ ˆ ˆ qf , tf T Q(t1 )Q(t2 ) · · · Q(tn ) qi , ti = 1 i
n
δ n z[J] δJ(t1 )δJ(t2 ) · · · δJ(tn )
(2.65)
J=0
Suppose. which I repeat here for reference.1 Functional Diﬀerentiation F (J(τ )) By this we understand that τ is the independent variable and F is a function of a function of τ . dF = ∂F dJ dτ ∂J dτ Consider the expression These rules fail when there is buried inside F some deﬁnite integral in which τ appears as the variable of integration.1 Let η(τ ) be any old wellbehaved 1 This deﬁnition is taken from Goldstein. We can’t diﬀerentiate this with respect to J. tf qi . for example that the function f = f (J(τ )) is a function of J. The function f is turn is buried inside an integral. qf . 3’rd ed. This is the case in equation (2.64).Chapter 3 Mathematical Appendix 3.2) Deﬁne the variation of F as follows.1) We call a expression of this sort a functional of J and write it F [J]. ti J ≡ D[q] exp i tf ti dt [L(t) + J(t)q(t)] (3. which itself is a function of τ . Classical Mechanics. but we can do something similar called functional diﬀerentiation. Section 2. The familiar rules of diﬀerentiation apply. so for example. so F [J] ≡ f (J(τ ))dτ (3.3 29 .
4) ∂f ∂J dτ = ∂J ∂α ∂f ηdτ ∂J (3.6) δF [J] Equation (3. and deﬁne CHAPTER 3.9) .8) δF [J] = iq(τ )F [J] δJ(τ ) This is a good time to point out that the functional derivative of a functional is a function. As an example.54) F [J] = exp i F [J+δη] = exp i τf τi τf dτ [L + J(τ )q(τ )] = F [J]ei R τf τi (3. and δη = ηδα. α) = J(τ ) + αη(τ ) ∂[F (α)] = ∂α The variation is then δF ≡ ∂F ∂α δα = F [J + δη] − F [J] (3. (This is sometimes used as a deﬁnition of the functional derivative.30 function of τ . you can take the shortcut.6) implicitly deﬁnes the functional derivative δJ(τ ) and also gives us an algorithm for calculating it. (3.7) δη(τ )q(τ )dτ τi dτ [L + J(τ )q(τ ) + δη(τ )q(τ )] δF = iF [J] τf τi q(τ )δη(τ )dτ (3. Combine (3.3) and (3.) δF [J(τ )] F [J(τ ) + δ(τ − τ )] − F [J(τ )] = lim →0 δJ(τ ) The notation F [J(τ )] means that τ is the variable of integration.3) Where δα is some very small value of α. Now that you have the idea.5) ∂J This is usually written F [J + δη] − F [J] = δF [J] δη dτ δJ(τ ) (3.4) ∂f δF = δη dτ (3. take a simpliﬁed form of (2. MATHEMATICAL APPENDIX J(τ.
12) δJ(x)δJ(y) We diﬀerentiated with respect to J(x) and J(y). . therefore the resulting function must be a function of x and y. δ 2 F [J] δJ(x )δJ(y ) The ﬁrst derivative is found as follows: δF = lim →0 δJ(x ) 1 − i 2 d4 xd4 y[J(x) + δ (4) (x − x )]D(x − y) i 2 d4 xd4 yJ(x)D(x − y)J(y) (3. F [J] = − i 2 d4 x d4 yJ(x)D(x − y)J(y) (3. FUNCTIONAL TAYLOR SERIES 31 We can use the shortcut to evaluate a more diﬃcult and important case from a later chapter.” This is good advice with functionals as well as ordinary functions This is particularly true. We don’t need them anymore. we need to evaluate the expression. To do this.2. expand in a Taylor series.2 Functional Taylor series “When in doubt. and we need some way to make systematic approximations.11) ×[J(y) + δ (4) (y − x )] + = −i d4 xD(x − x )J(x) term. since most path integrals cannot be evaluated exactly. We use functional diﬀerentiation to pull D(x − y) out of the functional. so let’s write it δ2F = −iD(x − y) (3.10) All you need to know about this is that x and y are 4vectors and D(x) = D(−x). The following expression occurs naturally when we calculate correlation functions using the path integral formalism. 3. Notice that both J’s are incremented by the δ2F = −iD(x − y ) δJ(x )δJ(y ) The primes have done their job of distinguishing the independent variables from the variables of integration.3. The last step is easier.
14) Repeating the steps from (3.3) – (3.6) d F [J(α)]α=0 = dα d2 F [J(α)] = dα2 dτ ∂f ηdτ = ∂J δ η(τ ) J(τ ) δF [J] η(τ )dτ δJ(τ ) dτ δF [J] η(τ ) δJ(τ ) (3.16) Collecting these terms and setting α = 1 gives F [J + η] = F [J] + 1 δF [J] η(τ )dτ + δJ(τ ) 2 dτ dτ η(τ )η(τ ) δ 2 F [J] +· · · δJ(τ )δJ(τ ) (3.32 CHAPTER 3. F [J(α)] = F [J(0)] + d 1 d2 F [J(α)]α=0 + · · · F [J(α)]α=0 + dα 2 dα2 (3.2) we can write F [J(α)] = f [J(τ ) + αη(τ )]dτ (3.15) (3. MATHEMATICAL APPENDIX As in (3.13) This time we will expand F [J(α)] in an ordinary power series in α and set α = 1 at the end of the calculation.17) .
however. y) which describes the displacement of the sheet as a function of position. There is no way to change the number of q’s in our Lagrangian. The motivation for this is explained nicely in Zee’s book in the section about the mattress. Such a continuous displacement variable is a ﬁeld in the strict mathematical sense: it assigns a numerical value (or set of values) to each point in space. This is an imperfect analogy for an attractive potential. but most of the time this formalism “lurks in the background. The sheet will oscillate in response. calculate interactions. which is anything but simple. The quantum mechanics of such ﬁelds is called quantum ﬁeld theory. φ(x. Now suppose that instead of using a bowling ball I jump up and down on the sheet. or we could introduce a continuous variable. Our strategy has been to get rid of operators and replace them with classical. and discuss symmetries without actually doing path integrals. We could describe the attraction in one of two ways: we could say that there is an attractive potential between any pair of pointlike masses. and nearby objects will roll into it.” We can construct Lagrangians. So long as we work with a ﬁnite number of degrees of freedom. which as a great simplifying feature.Chapter 4 Introduction to Field Theory The formalism I have employed so far tries to look as much as possible like classical Lagrangian mechanics with a ﬁnite number of degrees of freedom. which propagates outward in the form 33 . If I put a bowling ball on the sheet. cnumber variables. My activity becomes a source of energy. They are embedded in the path integral formalism. we can never accommodate the creation and annihilation of particle. which I will paraphrase. we need to replace the mechanics of discrete masses with the mechanics of continuous media. Imagine that space is like a rubber sheet. To do this. it will create a depression. These variables commute with one another.
This much is required by the MM experiment. it just makes it impossible to answer! Let’s bite the bullet and agree for the purposes of this course that space is pervaded by a medium. y) describes a displacement. Well – I have to modify that slightly. which for lack of a better name. can be carried out to any degree of accuracy desired. INTRODUCTION TO FIELD THEORY of waves. I don’t want you to think we are doing general relativity. It makes possible a systematic and understandable perturbation theory that. and these points obey the usual rules for Lorentz transformations. etc. When an electromagnet wave is propagating through space. Spacetime points in the ether can be labelled with Lorentz fourvectors or (x. But just saying that the ether does’t exist doesn’t answer the question. and here is where the path integral formalism really becomes a help rather than a complication. Each species of particle corresponds to a set of vibrations in it’s own speciﬁc ether. what is waving? Folks in the 19’th century thought it must be some sort of mechanical medium.) We have to resort to approximations. I am fond of saying. According to the textbooks. The principle of linear superposition would hold everywhere. the anharmonic oscillator cannot be solve exactly in quantum mechanics. Well. Michaelson and Morley proved that wrong with their famous interferometer. “When correctly viewed. t) as usual.” Now you see that this is profoundly true. The rubber sheet is not intended as an analogy for ordinary spacetime as it is often used in explaining general relativity.34 CHAPTER 4. everything is a harmonic oscillator. which they called the ether. (If you think of a way to do it. “Displacement of what?” The same question comes up in classical electromagnetic theory. Ordinary bulk media have elastic properties that are described by two parameters. actually the ethers. we will call the ether. There is an alternative way of dealing with the creation and annihilation . the waves propagating on the sheet would never interact. tell me and I’ll be famous. Electrons are all vibrations in the electron ether. which is what we really care about. This analogy can easily be misleading. If each point on the sheet behaved like a simple harmonic oscillator with a quadratic potential. The ﬁeld φ(x. These parameters are not themselves relevant to our formalism. Each point on the rubber sheet or ether acts like a harmonic oscillator! Quantum ﬁeld theory is a theory about harmonic oscillators. If our theory is to describe interactions. but their ratio gives the velocity of propagation. and I know you want to ask. Unfortunately. This is the rubbersheet analogy to the propagation of particles. with enough patience. This is a theory of free particles. then we must modify the potential so that it becomes anharmonic. the density and Young’s modulus. For one thing.
according to second quantization. then there are N coupled equation of this sort. We need to master both. The Lagrangian is constructed in the obvious way.4) If there are N masses. 4. T = 1 2 1 2 mϕ2 ¨i i (4.3) The equilibrium separation between masses is a.2) 2 L=T −V = a i m 2 ϕ − ka ˙ a i ϕi+1 − ϕi a (4. The path integral formalism. but we never make it that far. The EulerLagrange equations of motion are obtained from d ∂L ∂L − =0 dt ∂ ϕi ∂ϕi ˙ (4. This approach is complementary to path integrals in other ways as well. They look like m ϕi+1 − ϕi ϕi − ϕi−1 ϕi − ka ¨ + ka =0 (4.4. It turns the wave functions themselves into operators by imbedding creation and annihilation operators into them. the displacement of the ith particle from its equilibrium position. Second quantization goes in the other direction. a long chain of masses connected by springs. There is a good discussion in the last chapter of Goldstein. is an inﬁnity of harmonic oscillators.1. The masses can only oscillate along the chain. It’s a onedimensional problem. as the generalized coordinate. It is the older way. The spring constant is k.1) V = 1 2 k(ϕi+1 − ϕi )2 i (4. We will use ϕi . but they are the raising and lowering operators of the harmonic oscillator! The universe. What follows is a brief introduction. Let’s do this with the simplest possible system. sometimes called canonical quantization or second quantization.1 Introduction to Continuum Mechanics The rules of continuum mechanics are derived by starting with a system with a ﬁnite number of degrees of freedom and then passing to the limit in which the number becomes inﬁnite. Continuum mechanics is not covered in most graduate mechanics classes.5) 2 a a a2 . seeks to banish all operators from the theory. INTRODUCTION TO CONTINUUM MECHANICS 35 of particles.
In the continuum limit ϕ(x + a) − ϕ(x) dϕ ϕi+1 − ϕi → → a a dx L→ 1 2 dx µϕ2 − Y ˙ dϕ dx 2 ≡ dxL(ϕ. x would be the displacement of a particle. (7) is one equation for an inﬁnite number of degrees of freedom. 1 See Goldstein for a derivation of this important equation. In the continuum limit. Notice that a string of massive particles yields a wave equation that when quantized describes the propagation of massless particles. (8) should remind you of the equation for the propagation of electromagnetic waves. In this sense. In elementary mechanics. of course. INTRODUCTION TO FIELD THEORY We need diﬀerent parameters to describe the continuum limit: m/a → µ ka → Y mass per unit length Young’s modulus The index i points to the ith mass. continuum mechanics is much easier that discrete mechanics. and ϕi gives its displacement.8) (4) and (5) represent a set of N coupled equations for N degrees of freedom.) This is worth a brief digression.7) + − dt ∂ dϕ dx ∂ dϕ ∂ϕ dt dx Use the Lagrangian density from (6) in (7) ∂2ϕ µ = 2 ∂x Y d2 ϕ dt2 (4.36 CHAPTER 4. photons are massless particles. ϕ) ˙ (4. Here ϕ(x) is the displacement of the string at the point x.6) The last integral implicitly deﬁnes the Lagrangian density . ∂2ϕ ∂x2 + ∂2ϕ ∂y 2 + ∂2ϕ ∂z 2 = 1 c2 ∂2ϕ ∂t2 As you know. . (With a diﬀerent velocity. the index is replaced by the coordinate x. The continuum version of the EulerLagrange equation is1 d ∂L d δL ∂L =0 (4.
or the relationship is quadratic. in which case the particle is massless and its velocity v = ω/k. h h (4. the group velocity. i. INTRODUCTION TO CONTINUUM MECHANICS 37 What does it mean to say that a wave function describes the propagation of a particle of a particular mass? The wave function ψ = ei(kx−ωt) might describe a wave in classical E&M.13) m = lim ¯ k h k→0 dω So long as we are talking about waveparticle duality.9) The relationship is quadratic. The relativistic wave equation for a spinzero particle is called the KlineGordon equation. The question is.1. All these equations can be obtained from (7) with the appropriate Lagrangian density. The velocity is the constant of proportionality. plot ω versus k for small values of k. . This is the hallmark of a massless ﬁeld. so there can only be one velocity. They are all threedimensional variations of our “waves on a rubber sheet” model. what is the relation between k and ω? The relationship between the two is called a dispersion relation. in which case dk . vgroup = h ¯k dω ≈ dk m (4. either ω is proportional to k. All the wave equations I know of fall in one of these two categories. It contains a great deal of information.e.10) The dispersion relation is (c¯ k)2 + m2 c4 = (¯ ω)2 . (4. In both cases the curves are parabolas. k = ω/c. this is what mass means. Frequency and wave number are simply proportional. p2 c2 + m2 c4 = E 2 . or a massive particle in nonrelativistic or relativistic quantum mechanics. This means that in the limit of small k.4. In Schrodinger theory h ¯ 2 k2 = ¯ω h 2m (4. 2 − 1 ∂2 c2 ∂t2 ϕ− m2 c2 ϕ=0 h ¯2 (4. the group velocity is equal to the classical velocity for a massive particle v = p/m. What does this have to do with the particle’s mass? It’s useful to plot (9) and (11).11) or in other words.12) In other words. In the case of EM waves in vacuum.
x. −x. (4. −y.15) This puts the EulerLagrange equation in tidy form ∂µ ∂L ∂(∂ µ ϕ) − ∂L =0 ∂ϕ (4.16) (4. ∂xµ (4.20) Y c Obviously L can be multiplied by any constant without changing the equations of motion. z) xµ = (x0 . x1 .18) S = dt L = dt d3 x L = c is also a Lorentz scalar. x3 ) = (ct. Equation (7) was derived without reference to Lorentz transformations. INTRODUCTION TO FIELD THEORY One of the advantages of using Lagrangians rather than Hamiltonians is that Lagrangians have simple transformation properties under Lorentz transformations. Construct the contravariant and covariant fourvectors xµ ≡ (x0 . We get relativity for free! If we can manage to make L a Lorentz scalar. x3 ) = (ct. Better yet. Rescale it so that it becomes L= 1 2 ∂ϕ ∂t 2 − c2 ∂ϕ ∂x 2 + ∂ϕ ∂y 2 + ∂ϕ ∂z 2 (4. Take µ 1 = 2.38 CHAPTER 4.14) (4. Rewrite (6) in 3d 1 L= 2 µ ∂ϕ ∂t 2 −Y ∂ϕ ∂x 2 + ∂ϕ ∂y 2 + ∂ϕ ∂z 2 (4. the action 1 d4 x L (4. To see this. then (17) will have the same form in all Lorentz frames. Here’s an example. −z) and the corresponding contravariant and covariant derivatives ∂µ ≡ ∂ ∂xµ ∂µ ≡ ∂ . We can do relativity with path integrals without having to “ﬁx” anything.19) This would be the Lagrangian density for oscillations in a huge block of rubber. and yet (17) has the correct form for a scalar wave equation. x2 . x2 . let’s rewrite (7) in relativistic notation.21) .17) This is slightly amazing. y. x1 .
The fact that we can write L like this is proof that it is a Lorentz scalar. We don’t have to worry about the mechanical properties of the ether. Notice how the Lagrangian for oscillations a block of rubber (19) turns into the Lagrangian for oscillations in the ether (21).) is a onecomponent Lorentz scalar. even the vacuum! Using Einstein’s neat notation. INTRODUCTION TO CONTINUUM MECHANICS 39 Substituting (21) into (17) yields the usual equation for EM waves. More about this later. Try it. ti H = qf e−iH(tf −ti ) qf S = D[x]D[p]eiS .23) (I have set c = 1 and h = 1. It’s the KlineGordon theory. Now you can make up your own ﬁeld theories. tf qi .1. Simply replace the displacement qi with the displacement ﬁeld ϕ(x) and the momentum pi with the canonical momentum π(x) = ∂L . then this describes a spinless particle with mass m propagating without interactions. everything is a bunch of harmonic oscillators.22) The last piece of notation (∂ϕ)2 .26) . is used to save ink. This is an important point. the equation of ¯ motion is (∂µ ∂ µ + m2 )ϕ = 0 (4. When correctly viewed. Name the theory after yourself. qf . because µ and Y are scaled away. L= 1 (∂ϕ)2 − m2 φ2 2 (4. We have been calculating the propagator.24) If we assume that ϕ(x) (x is a 4vector in this notation. we can collapse (21) into one term 1 1 L = (∂µ ϕ)(∂ µ ϕ) ≡ (∂ϕ)2 2 2 (4.) Using our new notation. ∂ϕ ˙ Let’s look at this more closely. Here’s a theory that already has a name. we can deduce the symmetry properties of a theory by glancing at L. Despite what you may have been told. Spin can be included by adding more components to ϕ. 2ϕ = 0.4. All you have to do is add scalar terms to (22).25) (Equation (29) in Chapter 2) The natural generalization of this is ϕf (x)e−iH(tf −ti ) ϕi (x) = D[ϕ]D[π] exp i tf ti d4 xL(ϕ) (4. It is plausible that we can adopt all the path integral results from the previous sections. the MichelsonMorley experiment proves the existence of the ether. (4.
The derivation can be completed using the notion of coherent states.2 but I propose a diﬀerent approach. We argue that (26) is correct. I simply called it N and reassured you that it would go away. are not measurable. I have no idea what the constant is and only a clue what it means. Since it always cancels. we can complete the square and evaluate the D[π] integral to obtain ϕf (x)e−iH(tf −ti ) ϕi (x) = D[ϕ] exp i tf ti d4 x L (4. the q’s are measurable quantities. T ). ϕf .29) directly from Feynman’s sum over paths hypothesis.32) and (2. Wasserman’s lecture notes. To make the notation a bit more compact I will write the propagator. Section 9. Now it is possible to show3 without using the completeness hypothesis that U satisﬁes Schrodinger’s equation i¯ h ∂ U = HU ∂T (4. tf ϕi .28) The Hamiltonian in this equation is obtained from the Lagrangian in (26) by the usual procedure. ϕf . ti = U (ϕi . See Peskin and Schroeder. I do know that all physically interesting quantities are quotients of two terms.33) that this integration produces a badly behaved normalization constant. so U as it stands is not of much use.27) where T = tf −ti . on the other hand.2 Correlation functions The propagator in (25) is just the ticket for ordinary quantum mechanics. Quantum ﬁelds. In the previous chapter we were able to derive (2. I have not even bothered to write it in (29). After all. INTRODUCTION TO FIELD THEORY Where the functions ϕ(x) over which we integrate are constrained to the speciﬁc conﬁgurations ϕi (x) at x0 = ti .40 CHAPTER 4. The trouble is that the ﬁeld operators ϕ do not have complete sets of eignˆ states. even if we are unable to derive it from ﬁrst principles. Since the exponent in (26) is quadratic in π. this is true a fortiori.29) You will recall from (2. Let’s just take it as a deﬁnition of U . Back then. It turns out 2 3 This is worked out in Prof. (4.1 . and ϕf (x) at t0 = tf . each containing the same constant. Here. Here here that derivation breaks down. 4.
because the ﬁeld operators don’t have eigenstates. CORRELATION FUNCTIONS 41 that the quantities of interest are the correlation functions deﬁned in Section ˆ 2.31) The ϕ’s are ﬁeld operators in the Heisenberg picture. as in Section 2. ϕa represents 1 2 the ﬁeld evaluated at time −T .32) At this point in Section 2.6. x2 . I am using the notation from Section 9. but with ﬁelds replacing the Q operators. xn ) ≡ ΩT [ϕ(x1 )ϕ(x2 ) · · · ϕ(xn )] Ω ˆ ˆ ˆ (4. ϕa = ϕ(−T. this step is not immediately justiﬁed. x) and ϕb = ϕ(T. As mentioned previously.6. x). We can convert the Heisenberg 2 1 ﬁelds to Schrodinger ﬁelds. equation (59). Consider the following object:4 ϕb e−iHT T [ϕ(x1 )ϕ(x2 )]e−iHT ϕa ˆ ˆ (4. ˆ and that the completeness relation D[ϕ1 ] ϕ1 ϕ1  = ˆ 1.4. I hope no one will be so symbolminded as to confuse the T in the exponentials (meaning some ﬁxed time) and the T in front of the ﬁelds (the time ordering operator). The construction T [· · · ] refers to the time ordering operator. 4 .g.6 we inserted complete sets of position eigenstates and replaced the operators with their eigenvalues. In other words we are looking for G(x1 . e. ϕ(x2 ) = eiHx2 ϕS (x2 )e−iHx2 ˆ ˆ Now (31) can be written ˆ ˆ ϕb e−iH(T −x2 ) ϕS (x2 )e−iH(x2 −x1 ) ϕS (x1 )e−iH(x1 +T ) ϕa 0 0 0 0 0 0 (4. Let’s assume to start with that x0 > x0 .30) This is sometimes called the npoint function. The times are arranged ˆ so that both x0 and x0 fall in the interval between −T and T . The most important instance of this is the twopoint correlation function. so let me just reassure you that we won’t make any mistakes. which we will now investigate.2 of Peskin and Schroeder. I don’t want to get into the technicalities of coherent states. if we assume that we can replace ﬁeld operators with cnumber ﬁelds ϕS (x1 )ϕ1 = ϕ1 (x1 )ϕ1 .2. · · · .
This has the eﬀect of projecting out the ground state. and clearly. INTRODUCTION TO FIELD THEORY makes sense. H = H0 + HI . because a single particle can interact with itself via intermediate particles. even the vacuum can interact with itself through the spontaneous creation of particleantiparticle pairs. 1 2 ϕb e−iH(T −x2 ) ϕ2 = 0 T D[ϕ] exp i x0 2 d4 x L(ϕ) combining these three factors yields the ﬁnal result ϕb e−iHT T [ϕ(x1 )ϕ(x2 )]e−iHT ϕa ˆ ˆ T (4. the freeparticle and the potential or interaction term. I have set up the correlation function with the e−iHT ’s in (31) because I want to discuss the limit T → ∞. Deﬁne the eigenstates of H as follows: Hn = En n Imagine starting with ϕa . The result is 1 D[ϕ1 (x)] D[ϕ2 (x)] ϕ1 (x1 )ϕ2 (x2 ) ϕb e−iH(T −x2 ) ϕ2 0 0 0 0 (4. This is a highly nontrivial problem.33) × ϕ2 e−iH(x2 −x1 ) ϕ1 ϕ1 e−iH(x1 +T ) ϕa This is a product of three propagators that jolly along the ﬁeld from −T → x0 → x0 → T . the particle has energy eigenstates and eigenenergies.42 CHAPTER 4. If we set HI = 0. Now insert ˆ after each of the ﬁelds in (32). e−iHT ϕa = n e−iEn T n nϕa . This makes sense mathematically.34) = D[ϕ]ϕ(x1 )ϕ(x2 ) exp i d4 x L(ϕ) −T where the boundary conditions on the path integral are ϕ(−T. we would have arrived at the same result. and evolving through time with H. x) = ϕa (x) and ϕ(T. Notice that if we had started with the assumption that x0 > x0 . but not physically: nature doesn’t come equipped with switches to do this. but exactly what does this mean and what is the ground state anyhow? We are accustomed to dividing the Hamiltonian into two parts. the minimum energy is E0 = 0. Worse yet. We call this minimumenergy vacuum state Ω . x) = ϕb (x) for some ϕa and ϕb . We 1 2 get the time ordering for free. For example.
CORRELATION FUNCTIONS = e−iE0 T Ω Ωϕa + n=0 43 e−iEn T n nϕa Give a small negative imaginary part to T and look at what happens when T → ∞(1 − i ). Each term tends toward zero. Back at the end of Chapter 2. and (2) taking the limit T → ∞(1 − i ).38) We take advantage of recent developments by upgrading these formulas in two ways: (1) replacing the Qi ’s with the ﬁelds. ˆ ˆ ˆ qf . the ﬁrst term approaches zero more slowly. these constants cancel between the numerator and denominator. Often in doing perturbation theory. which I repeat here for reference. I introduced the notion of a generating function.39) We can use this technology to rewrite (36) as follows: ˆ ˆ ΩT [ϕ(x1 )ϕ(x2 )]Ω = 1 Z[0] 1 i 2 δ 2 Z[J] δJ(x1 )δJ(x2 ) (4. In this case we replace Ω with 0 . so we can ignore it altogether. It is customary to use the capital Z for this new generating functional.64) and (2. ti = 1 i ∞ Z[J] ≡ D[ϕ] exp i −∞ d4 x [L(ϕ) + J(x)ϕ(x)] (4.40) J=0 .4.65). ˆ ˆ ΩT [ϕ(x1 )ϕ(x2 )]Ω = lim D[ϕ]ϕ(x1 )ϕ(x2 ) exp i D[ϕ] exp i T −T T −T T →∞(1−i ) lim (e−iE0 T Ωϕa )−1 e−iHT ϕa . (4.35) d4 xL T →∞(1−i ) d4 xL (4. z[J] ≡ qf . the term containing the sum will vanish. tf qi .2.37) δ n z[J] δJ(t1 )δJ(t2 ) · · · δJ(tn ) J=0 (4. The key equations here are (2. ϕ(x). but since E0 < En for all n = 0. Here as always. tf T Q(t1 )Q(t2 ) · · · Q(tn ) qi . The denominator is then proportional to 00 = 1. Put it another way: if we divide the equation through by e−iE0 T Ωϕa .36) I remarked previously that the measure D[ϕ] produced weird normalization constants. ti J ≡ D[q] exp i tf ti dt [L(t) + J(t)q(t)] n (4. Ω = Insert (35) into (34). we are concerned with the correlation function for two free particles.
41) 4. INTRODUCTION TO FIELD THEORY The generalization to higherorder correlation functions is obvious.44 CHAPTER 4. x2 .3 The Feyman Propagator I will now show how this formalism works with the free KleinGordon ﬁeld.45) ˜ ˜ ˜ d4 p ϕ(P )J(−p) + ϕ(−p)J(p) ˜ The reason for the two terms will be clear soon.” (4. We sometimes use the notation ˆ ˆ ΩT [ϕ(x1 )ϕ(x2 ) · · · ϕ(xn )]Ω = G(x1 .46) . Use the Lagrangian (23) in (39) Z[J] = D[ϕ] exp i d4 x 1 (∂ϕ)2 − m2 ϕ2 + i ϕ2 + Jϕ 2 (4. I will deﬁne ∆−1 ≡ p2 − m2 + i (4. xn ) is called the “npoint function. In order to make the notation a bit more compact. ϕ(x) = d4 p −ipx e ϕ(p) ˜ (2π)2 (4.42) I have inserted the i ϕ2 to ensure that the exponential integrals converge as explained previously. · · · . The key to doing the d4 x integral is to Fourier transform the integrand to momentum space. · · · .44) d4 p1 d4 p2 = −p1 p2 − m2 + i ϕ(p1 )ϕ(p2 ) ˜ ˜ 2(2π)4 = 1 2 d4 p ϕ(−p)ϕ(p)(p2 − m2 + i ) ˜ ˜ On the other hand d4 xJ(x)ϕ(x) = or (in fact) ˜ d4 pJ(−p)ϕ(p) ˜ 1 2 or ˜ ˜ d4 pJ(p)ϕ(−p) (4.43) Substituting (43) into (42) gives the following: d4 x = 1 (∂ϕ)2 − m2 ϕ2 + i ϕ2 2 d4 xe−i(p1 +p2 ) (4. xn ) ˆ The function G(x1 . x2 .
We don’t have a snowball’s chance of calculating Z[0]. we just don’t have to do them.53) .54) d4 x ±ipx e J(x) (2π)2 d4 x eipx D(x) (4.47) Deﬁne a new ﬁeld ˜ ϕ(±p) = ϕ0 (±p) − J(±p)∆ ˜ ˜ S= 1 2 ˜ ˜ d4 p ϕ0 (−p)ϕ0 (p)∆−1 − J(p)J(−p)∆ ˜ ˜ (4.48) (4.4. THE FEYMAN PROPAGATOR The complete action becomes S= 1 = 2 4 45 d4 x (L + J ϕ) ˆ −1 d p ϕ(−p)ϕ(p)∆ ˜ ˜ ˜ ˜ ˜ + J(p)ϕ(−p) + J(−p)ϕ(p) ˜ (4.51) (4. one depending only on the ﬁelds and one depending only on the sources.49) You see the point of (49)? It break up the action into two pieces.50) × exp ˜ ˜ d4 p J(p)J(−p)∆ (4. Substituting into (42) yields Z[J] = D[ϕ0 ] exp − i 2 i 2 d4 p ϕ0 (p)ϕ0 (−p)∆−1 ˜ ˜ (4. The peculiar sum in (45) was necessary to make this happen. but (40) guarantees that it will always cancel out! At this point it is useful to transform back to conﬁguration space.3.52) = Z[0]eiW [J] where W [J] ≡ − 1 2 ˜ ˜ d4 p J(p)J(−p)∆(p) Most path integrals are impossible to do. ˜ J(±p) = ∆(p) = Our ﬁnal result is W =− 1 2 d4 x d4 y J(x)J(y)D(x − y) (4.
we should use D(p) instead of ∆(p). The boundary conditions are diﬀerent. The derivation of Z[J] assumed that sources were turned on and oﬀ at ﬁnite times before and after the limit T → ∞. As such it represents the amplitude for a particle at the space time point x to propagate to y. so that J(x) = J0 [δ(x − x1 ) + δ(x − x2 )].56) iW = − (4. The other term is simply the probability of a particle to be created by the source at x1 (when the source was turned on) and reabsorbed at x2 (when the source was turned oﬀ) and vice versa. Now calculate the twopoint function corresponding to iW without specifying the nature of the sources. It’s an artifact of our delta function sources. . i 2 d4 x d4 y J(x)J(y)D(x − y) (4. (See equation (51).5 It is the relativistic generalization of the oneparticle Green’s function (2.48).55) This is the Feynman propagator.46 where CHAPTER 4. Relativity forces this on us. but this is not customary notation. Let’s see how this idea plays out in (50). however. since D(x − y) = D(y − x).57) Suppose our source consists of two local centers of activity. Then 2 iW = −iJ0 [D(x1 − x2 ) + D(0)] (4. INTRODUCTION TO FIELD THEORY D(x − y) ≡ d4 p eip(x−y) (2π)4 p2 − m2 + i (4. The following calculations use equations 5 ˜ To be consistent I suppose.) i2 Z[J] = 1 + iW + W 2 + · · · Z[0] 2 The second term is easy to understand. as I mentioned before.58) The D(0) is meaningless. First expand Z[J]/Z[0] in an exponential series.
4. x2n . These are Feynman diagrams. INTERACTING FIELDS (40). i. with a quadratic potential. It is easy to see that the nth term of the expansion (56) contains 2n factors of J and that all the npoint functions vanish except for G(x1 . x2 ) = iD(x1 − x2 ). (56) and (57). x2 ) = ΩT [ϕ(x1 )ϕ(x2 )]Ω ˆ ˆ = = = 1 Z[0] 1 i 1 i 2 47 1 i 2 δ 2 Z[J] δJ(x1 )δJ(x2 ) J=0 J=0 δ 2 iW [J] δJ(x1 )δJ(x2 ) 2 δ2 i − δJ(x1 )δJ(x2 ) 2 d4 x d4 y J(x)J(y)D(x − y) (4. G(x1 . ·. and draw lines connecting them in pairs. 4. Just draw n points labelled x1 . x2 .10) and (3. It consists or a sum of three terms corresponding to all possible ways in which two particles can propagate between four points. x2 . The relevant equations are (3. The ﬁnal result is G(x1 . Your should work out the 4point function as an exercise. we must add some additional term to the Lagrangian. Many texts use the interaction term −(λ/4!)ϕ4 (x). The theory so far has been trivial in the sense that the particles don’t interact with one another.59) J=0 This functional derivative was worked out as an example in Chapter 3. The higher order terms in (56) vanish when we set J = 0. We’ll look at a simple example in the next section. and it’s true in quantum mechanics as well. · · · . (41). We started with the Lagrangian based on the simple harmonic oscillator.60) This is in fact the exact amplitude for a free particle to propagate from x1 to x2 . The eﬀect of the functional derivatives has been to “peel oﬀ” the sources. Interacting ﬁeld theories have additional terms in the Lagrangian. and in such a model all the normal modes are linearly independent.12). This is well known from continuum mechanics. you can represent any npoint function as a set of (2n1)!! diagrams. (4. which is as .4.4 Interacting Fields In order to make out ﬁelds interact. Now that you have the idea. x2n ).e.
48 CHAPTER 4.” Feynman diagrams are a kind of perturbation theory in which the coupling constant is assumed to be “small” so that perturbation series is really an expansion in powers of the coupling constant. More familiar theories all involve two or more diﬀerent particle species and spin. Z[J] = D[ϕ] exp i d4 x Z1 [J] = −iλ 4! dω D[ϕ] exp i 1 d4 x − ϕ(∂ 2 + m2 − i )ϕ + Jϕ 2 ϕ4 (ω) (4. and we are forced to make approximations. but “Suﬃcient unto the day is the evil thereof. It turns out that everything we can know about this process is contained in the fourpoint Green’s function.” It determines the strength of the interaction.63) You will notice that I replaced (∂ϕ)2 in the exponential with −ϕ∂ 2 ϕ. x3 .62) The tricks we used in arriving at (50) no longer work. x2 . this describes identical particles interacting in groups of four. 1 (∂ϕ)2 − m2 ϕ2 + i ϕ2 + Jϕ − (λ/4!)ϕ4 2 (4. Let’s start with the ﬁrstorder term. An example is the scattering process in which two particles come together at a point and two particles emerge from that point. As I am fond of saying. “When in doubt. in this case λ. As you will see. INTRODUCTION TO FIELD THEORY good as any. but it turns out to be an important ingredient in the nonAbelian gauge theories we will study much later in the course. You probably don’t know any particles that act like this. 1 Z[0] 1 i 4 δ 4 Z[J] δJ(x1 )δJ(x2 )δJ(x3 )δJ(x4 ) = G(x1 .61) with our new Z[J].” The simplest nontrivial process involving these hypothetical particles is the twobody scattering process described above. Another is the decay process in which one particle decays into three clones of itself. This . We need to calculate the fourﬁeld version of (40). The parameter λ is called a “coupling constant. x4 ) J=0 (4. expand in a power series.
. The term 2 J · A−1 · J is reminiscent 6 There is a brief proof of this in Appendix 1 of Zee’s book. x4 ) = = −iλ 1 4! Z[0] d4 ω 1 Z[0] δ 4 Z1 [J] δJ(x1 )δJ(x2 )δJ(x3 )δJ(x4 ) J=0 1 D[ϕ] exp i d4 x − ϕ(∂ 2 + m2 − i )ϕ + Jϕ 2 1 i 4 49 J=0 × ϕ4 (ω)ϕ(x1 )ϕ(x2 )ϕ(x3 )ϕ(x4 ) (4. The matrix A is assumed to be real and symmetric. We’ve i discussed this before.4. So far so good. j. The trick is yet another application of our basic gaussian integral. etc.65) The notation x · A · x means ij xi Aij xj . Let’s deﬁne xi xj · · · xk xl ≡ i ∞ ∞ − 2 x·A·x xi xj · · · xk xl −∞ · · · −∞ dx1 dx2 · · · dxN e i ∞ ∞ − 2 x·A·x −∞ · · · −∞ dx1 dx2 · · · dxN e (4. · · · k.66) and the last two lines of (4. intuitive ”cookbook” procedure for evaluating all such expressions eﬀortlessly. x2 . and xi is analogous to ϕ(x). you can calculate any Green’s function to any order in λ.64)? That is the paradigmatic integral. I will ask you to derive it in more detail for homework.64) You see that long string of ϕ’s on the last line of (4. there is a simple.4.68) There is a useful analogy between (4.67) stand for any pair of the indices i.65) and (4. Now substitute (4. x3 . xi xk = (A−1 /i)ik (4. A sum over “Wick contractions” consists of all possible ways of connecting the indices in pairs.67) The subscripts ab and cd in (4.63) into (4.66) By repeated diﬀerentiation of (4. If you can do that. INTERACTING FIELDS is just integration by parts. l.6 ∞ −∞ dx1 dx2 · · · dxN e− 2 x·A·x+iJ·x = i (2πi)N i J·A−1 ·J e2 det [iA] (4. If there is only one pair.61) G(x1 .65) we can derive xi xj · · · xk xl = (A−1 /i)ab · · · (A−1 /i)cd Wick (4. In fact. Clearly Ji is analogous to J(x).64).
so it’s a good guess that A−1 is analogous to −D(x − y). In fact. if my claim is true then the correct generalization of (4. what does it mean to take the determinant of it? The answer to the last question is clear: we don’t care.66) must be analogous to Z[0]. That’s what they do for a living. we have implicitly proved it. Anyhow.69) is easily established using the deﬁnition (55). Section 6. inverses of diﬀerential operators.62) with J = λ = 0. The other point is almost clear. Quel horreur! Worse yet. 7 Lewis Ryder. as you will see. In summary then. Quantum Field Theory. the denominator in (4.2 .66). since (∂ 2 + m2 − i )D(x − y) = −δ (4) (x − y) (4. after all.64). I claim that we can use (4. A must be analogous to (∂ 2 + m2 − i ). come to think of it. but if this is true.65) and (4.66) for ﬁelds rather than discrete variables by making the following replacements: xi → ϕ(xi ) → −D(x − y) i i J · A−1 · J → − d4 xd4 yJ(y)D(x − y)J(y) 2 2 A → (∂ 2 + m2 − i ) i i − x·A·x → − d4 xϕ(x)(∂ 2 + m2 − i )ϕ(x) 2 2 ∞ −∞ A Ji −1 → J(xi ) dx1 dx2 · · · dxn → D[ϕ] The proof that my claim is valid is sketched out in Ryder. INTRODUCTION TO FIELD THEORY of (54). equation (4. a Green’s function.50 CHAPTER 4. at least for expressions of the form (4. ie. and Green’s functions are by deﬁnition. D(x − y) is.70) The denominator is just Z0 [0].7 but in fact.69) The proof of (4.66) is ϕ(xi )ϕ(xj ) · · · ϕ(xk )ϕ(xl ) = D[ϕ] exp i 1 d4 x − 2 ϕ(∂ 2 + m2 − i )ϕ ϕ(xi )ϕ(xj ) · · · ϕ(xk )ϕ(xl ) D[ϕ] exp i d4 x − 1 ϕ(∂ 2 + m2 − i )ϕ 2 (4. since the determinant cancels in equation (4.
l. That proof required nothing more than functional diﬀerentiation. and ω. must be attached to one and only one line.2.72) Wick So here’s the procedure: make a list of all possible ways of pairing up the indices i. x2 ) = iD(x1 − x2 ). That’s all there is to it. so our cookbook analogy has given us the right answer for the Feynman propagator without any of the clever calculations in Section 4.64) using (4. x3 . xd ) (4. x2 ) = 1 Z0 [0] 1 i 2 51 δ 2 Z0 [J] δJ(x1 )δJ(x2 ) J=0 D[ϕ] exp i d4 x − 1 ϕ(∂ 2 + m2 − i )ϕ ϕ(x1 )ϕ(x2 ) 2 = 1 D[ϕ] exp i d4 x − 2 ϕ(∂ 2 + m2 − i )ϕ (4. therefore there must be four lines connecting points. ϕ(x1 )ϕ(x2 ) as deﬁned by (4. But we already know from (60) that G(x1 . Since there are four factors of ϕ(ω) in (4.71) The ﬁrst line follows from (59).73) The Wick sum contains in principle (N − 1)!! = 7!! = 105 terms.70) reveals that what we have just calculated is ϕ(x1 )ϕ(x2 ) .4. 2. INTERACTING FIELDS To see how this works. 3. and the second line uses (4. x2 ). G(xi . k.62) with λ = 0. The analogy to (4. Now you can evaluate (4. 4. ie. Label ﬁve points 1. j. x2 . it can be connected to itself by one or two lines.4.68) led us to hope that ϕ(x1 )ϕ(x2 ) would be equal to −D(x1 − x2 )/i. Put it another way: we have proved that the analogy works correctly in the case of (71). Draw some pictures to ﬁnd the interesting ones.70). however. let’s evaluate (4. xb )’s in (4. Each term in the expansion corresponds to a way of connecting these points. 3.67) is ϕ(xi )ϕ(xj ) · · · ϕ(xk )ϕ(xl ) = G(xa .73). x4 ) = − iλ 4! d4 ω ϕ(x1 )ϕ(x2 )ϕ(x3 )ϕ(x4 )ϕ4 (ω) (4. that most of these terms are of no interest. xb ) · · · G(xc . I’m willing to believe this without further proof. It is probably the case. there must be four “line ends” at the ω point. .70) in the case of two ﬁelds. Here are the rules: • There are eight ﬁelds in (4. • The points 1. and 4. G(x1 . Reasoning by analogy from (4. Each line corresponds to one of the G(xa . A glance at (4.70) equals G(x1 . Let’s pause for a moment to see how all the pieces ﬁt together.73).72). Each entry in the list corresponds to one term in the sum.3. · · · .
p2 . or to put it another way. Scattering is a good example. This is ultimately related to the fact that there are parameters such as the mass and coupling constant (or charge) of a particle that are fundamentally mysterious. x3 . 4. The rules for Feynman diagrams discussed brieﬂy above also have a simpler form in momentum space. pN ) = d4 x1 d4 x2 · · · d4 xN ei(±p1 x1 ±p2 x2 ±···±pN xN ) G(x1 . ω)G(x3 . 3. ω) (4. There are 4! terms like this in the Wick sum. . x2 . J(x3 ). Momentum space was only used occasionally as a tool for evaluating certain integrals.75) I call your attention to the ± signs in the exponential. To some extent the sign doesn’t matter. J(x2 ). Now we draw a few pictures. Deﬁne the momentum space Green’s functions in the obvious way. if we treat these divergences very carefully. If these integrals converged. If you want all four of these particles to interact. The ﬁnal result is G(x1 . x4 ). and 4 are connected to external sources. The ω point is not. 2. ˜ G(p1 . and J(x4 ) were used to deﬁne G(x1 . INTRODUCTION TO FIELD THEORY The points 1. It is a general rule that all Feynman diagrams with loops give rise to divergent integrals. These are all pathological in the sense that the ω point is connected to itself by at least one line. The momenta of the ﬁnal state particles are determined by our apparatus. · · · . You have noticed that we have ignored 1054!=81 terms. x4 ) = −iλ d4 ωG(x1 . and the answer is obvious by inspection. however. they must each be contracted with one of the ϕ(ω)’s. xN ) (4. J(x1 ). since G(x) = G(−x) and ∆(p) = ∆(−p). ω)G(x4 . if done correctly. would have yielded 105 terms. Moreover. then we might be able to calculate them. These divergences are nature’s way of slamming the door in our face. to which we will return later. x2 . · · · . x2 . The little bit of good news is that we can go on doing ﬁeld theory calculating everything else. A few paragraphs ago we had some algebra. x3 .52 CHAPTER 4. Momentum space is more natural for many quantum processes. These pictures are called Feynman diagrams. The particles in the beam have a welldeﬁned momentum. The target particle is at rest. which. That’s the big bad news. This comes under the general heading of renormalization. ω)G(x2 .5 Momentum Space Our derivations have for the most part used conﬁguration space.74) Forgive me if I gloat a bit.
p3 . x2 )G(x3 .76) The δfunction is there to enforce momentum conservation.75).5. ˜ G(p1 . Let’s agree on a convention: outgoing momenta are positive and incoming momenta are negative.75) and (4. the particle that starts out with momentum p1 does not interact with the one with momentum p3 . That way we will always get the right signs in (4.4. • A factor i∆(pi ) for each external line. ω)G(ω. • An energymomentum conserving δfunction (2π)4 δ (4) (±p1 ± p2 ± p3 ± p4 ) There is one more important item that we have not yet encountered. x3 . one is always free to replace x → −x or p → −p. One such term in the Wick expansion is8 G(x1 .76) illustrates some of the rules associated with Feynman diagrams. For the moment. MOMENTUM SPACE 53 when x or p appear as integration variables. In one respect. . p4 ) = −iλ(2π)4 δ (4) (±p1 ± p2 ± p3 ± p4 )∆(p1 )∆(p2 )∆(p3 )∆(p4 ) (4. then the δfunction must be δ (4) (p1 + p2 − p3 − p4 ) or δ (4) (p3 + p4 − p1 − p2 ). x4 ) = − iλ 4! d4 ω G(x1 . • A factor −iλ for each vertex. Equation (4. let’s look at one of the pathological diagrams in which one of the lines connected to the ω loops back on itself. x2 . if the scattering process is one in which p1 and p2 are the incident momentum and p3 and p4 are the outgoing momentum. ie. p4 ) = − iλ (2π)4 δ (4) (p1 − p2 )(2π)4 δ (4) (p3 − p4 ) 4! d4 p4 × ∆(p1 )∆(p3 )∆(p4 ) ∆(p4 ) (4.53). ω)G(x4 . p2 . For example. therefore their momenta are conserved 8 We have not yet come to terms with indistinguishable particles. the signs are allimportant. however.77) The momentum space Green’s function is G(p1 . ω) (4. What happens to internal lines.78) (2π)4 In this case. assume that the particles have little numbers on them like billiard balls. p2 . those that do not terminate on external sources. To see that. Take the Fourier transform of (4. p3 .74) using (4. The momentum space Green’s function corresponding to any one diagram consists of a product of the following factors.
Since this momentum can have any value. p2 = pµ pµ = m2 . Since we know these things already. Such integrals always diverge. To see that this is so. . The propagator D(x − y) describes a situation in which x and y are microscopically close together. so the integral diverges quadratically. and then investigate analytically how the result depends on the cutoﬀ. Such theories are said to be renormalizable. there is nothing to constrain the d4 p4 integral. we can simply ignore the cutoﬀdependent parts. When we evaluate out Green’s functions they appear attached to the external sources as well as to points inside the region where the particles interact with themselves. Such integrals can be dealt with rigorously with a variety of diﬀerent techniques. I know that it’s wrong! 4. after all. at least approximately. The propagator takes this into account in a way that is consistent with relativity. so it seems that the denominators of (46) and (55) should vanish. because pµ is not a physical momentum.” For free particles. There are two δfunction The new feature here is an internal line. so there is some “fuzziness” in the momentum as well. whereas the denominator contains only two. They don’t of course. All theories we know that describe the real world.54 CHAPTER 4. p2 is a Lorentz scalar. For a certain privileged class of theories. ﬁrst in momentum variables (46) and later in spacetime variables (55). we can’t insist that they are perfectly conserved. one can show that the only cutoﬀdependent quantities are the particle’s mass and coupling constant and the normalization of the wave function.6 Virtual Particles We have encountered the freeparticle Feynman propagator. Since we cannot specify momentum and energy with arbitrary precision. We cannot specify space and momentum coordinates with arbitrary accuracy. For example. but a variable of integration.79) The numerator contains four powers of the momentum. In both cases they have a peculiar feature: in some sense they represent particles that are not “on the mass shell. There is more to it than that. one can put a cutoﬀ on the range on integration. however. The fact that we can do it is also signiﬁcant. The fact that we have to do this is a sign that our knowledge and theory are incomplete. If you give me a theory that is not renormalizable. INTRODUCTION TO FIELD THEORY separately. are renormalizable. There is no oscillating phase factor. look at d4 p4 p42 − m2 + i (4. this one labelled by p4 .
9 i pi − Don’t let PETA ﬁnd out about this. The sources were turned on sometime after −T and turned oﬀ again before T . Put it another way. Leaving aside the very technical issue of renormalization.2. Indeed.78) describes particles that are virtual because they are emitted and absorbed in a very short distance.4. . We often use the momentum space Green’s functions to calculate scattering reactions. but the limits −T → −∞ and +T → +∞ that were used to project out the ground states refer to the macroscopic past and future where we live and move and have our being. The other propagators in (4. They are not external sources like xray tubes or cyclotrons.77) and (4. forcing the sum of momenta i pi going out of the vertex to be equal to the sum of momenta j pj going out of the vertex. That should be evident from the discussion in Section 4. To calculate a momentum space Green’s function corresponding to a Feynman diagram. We say that the propagator describes “virtual particles. I will restate them here for reference.”9 This procedure then becomes part of the rules for Feynman diagrams.” The propagator ∆(p4 ) in (4. pages 53 and 54. We can use our Green’s functions for this by the simple expedient of removing the ∆’s from the external legs. There is a complete description of the rules in Zee’s book. • Remove those propagators associated with external lines. that is their principal use. But particles in scattering experiments are produced by cyclotrons and other macroscopic sources. The action takes place after the sources turn on . • Associate each line carrying momentum p with a propagator i/(p2 − m2 + i ). and it stops when the sources turn oﬀ.6. • Momenta associated with internal lines are to be integrated over with the measure d4 p/(2π)4 .78) describe particles that are virtual for a diﬀerent reason: the sources J(xi ) that we built into the theory are also quantum mechanical objects. you now have everything you need to do perturbation theory calculations with scalar ﬁeld theory to any order. we live in the ground state. • Associate with each interaction vertex the coupling −iλ and (2π)4 δ (4) ( j pj ). VIRTUAL PARTICLES 55 it’s just not equal to m2 . do the following: • Draw the diagram and label each line with a momentum. This is sometimes called “amputating the legs of the diagram.
but what is worse. x2 ) = (−iλ)2 d4 w d4 z G(x. First take the Fourier transform G2 (p1 . It represents the secondorder correction to the propagator. p2 ) = d4 x1 d4 x2 d4 w d4 z ei(p2 x2 −p1 x1 ) d4 p1 · · · d4 p4 (2π)16 × ei(x1 −w) ei(w−z) ei(w−z) ei(z−x2 ) i∆(p1 )i∆(p2 )i∆(p3 )i∆(p4 ) (4. x2 ) to second order in perturbation theory. the ﬁrst due to Feynman. I will do this in the context of the simplest possible divergent integral. z)G(z. The challenge is to divide these integrals into two parts. Feynman . x2 ) (4. This is really a huge subject.” λ Lint = −i ϕ3 3! (4. one part that is inﬁnite and one part that is ﬁnite and contains all measurable eﬀects. these divergent integrals contribute real measurable eﬀects to the physical processes described by the Green’s functions. w)G(w. p2 ) = (−iλ)2 (2π)4 δ (4) (p1 − p2 )∆(p1 )∆(p2 ) The integral is not only diﬃcult. That last integral is going to give us trouble.84) (2π)4 G2 (p1 . it’s divergent. z)G(w.81) This will diverge because of the loop. The result when all is said and done is d4 k ∆(k)∆(p1 − k) (2π)4 (4. Let’s concentrate on it by deﬁning d4 k F (p) ≡ ∆(k)∆(p − k) (4.82) Do the integrals over d4 x1 d4 x2 d4 w d4 z.80) When one calculates the twopoint function G(x1 . one comes up against the integral G2 (x1 . then do three momentum integrals d4 p1 d4 p3 d4 p4 .7 A Word About Renormalization I have said repeatedly that all diagrams with complete loops lead to divergent integrals. since there are no other primed variables left. It arises when one tries to do calculations in the socalled “phithird theory. We will deal with these two problems with two amazing tricks.83) The integration variable p2 has be renamed k. INTRODUCTION TO FIELD THEORY 4. That’s bad news enough.56 CHAPTER 4. but I would like to give you a few of the key ideas.
7.4. so let’s make it a ﬁvefold integral! The basic formula is this: 1 = A1 A2 1 0 dz[A1 z + A2 (1 − z)]−2 ≡ 1 0 dz D(z)2 (4. Note that the integral ∞ 1 dx x diverges.85) The key idea is that the product A1 A2 is replaced by the single factor D2 . and here is where the next great idea comes in. In our case D2 = B 2 + Q2 − k 2 − i = C 2 − k 2 − i (4.86) where Q = p(1 − z) B 2 = m2 − p2 (1 − z) k =k +Q c2 = B 2 + Q2 Dropping the prime. as it turns out. A WORD ABOUT RENORMALIZATION 57 argued that a fourfold integral is hard to do. The integral diverges logarithmically.87) A bit of power counting shows that there are four powers of k in the numerator and four in the denominator. Suppose we could integrate over slightly less that four dimensions! In the context of the trivial example above. We would like to integrate over 1/D2+ but this makes no sense whatsoever (D is a four vector). is enough to make the d4 k integrable doable. we have 1 F (p) = 0 dz d4 k 1 (2π)4 [C 2 − k 2 − i ]2 (4.88) . but 1 ∞ dx 1 = 1+ x ie. This. suppose we could integrate ∞ 1 d(1− ) x x Here’s the idea dd k 1 i Γ(2 − d/2) = d (C 2 − k 2 − i )2 d/2 (2π) Γ(2) (4π) 1 C2 2−d/2 (4. increasing the exponent in the denominator by an inﬁnitesimal amount makes the integral convergent.
so the integral still diverges in the limit → 0.90) where ≡ 4 − d. Recall that F (p) is multiplied by λ2 in (4. Rather than .88) to integrate (4. but I have one more trick up my sleeve.92) There are two points to notice here. • The remaining integral is straightforward. After some algebra we ﬁnd F (p) = i 2Γ(1 + /2) d/2 (4π) 1 0 dz [m2 − p2 z(1 − z)] /2 (4. and Γ(x + 1) = xΓ(x) in general. It does not depend on p! We have cleanly separated the measurable physics in the second term from the meaningless constant in (4. (4.81). that’s not quite true.91) into (4. We use (4.91) A− /2 = 1 − ln A + 2 (ln A)2 + · · · 2 8 Insert (4. INTRODUCTION TO FIELD THEORY This was derived for integer d (don’t ask me how). • The term “∞” in (4. It’s as if we were working with an inﬁnite coupling constant.89) Also Γ(n) = (n − 1)! for integer n. To used an old cliche.93) The point is that this is a pure number.88) as a deﬁnition of what it means to integrate over a noninteger number of dimensions. We expected that. because we have gotten all the kids out of the tub! Well. Look at the expansion 1 (4. since it involves only one real variable of integration. We will use (4. Well Γ(1) = 1.58 CHAPTER 4.90). lim F (p) = ∞ − →0 i (2π)2 1 0 dz ln [m2 − p2 z(1 − z)] (4.87) and then see what happens in the limit d → 4. we don’t have to worry about throwing the baby(s) out with the bath water.92) is really lim 2i 1 →0 (4π)2 (4.93). ∞ Γ(x) = 0 dt tx−1 e−t . Incidentally. but the Γ functions and everything else on the right side of this equation are perfectly welldeﬁned continuous functions of d.
4.” let’s say that the λ that appears in (4. What we are doing is replacing λ0 ∞ → λ. and if so. We have “renormalized” the coupling constant.” We have learned from decades of experience that all theories that are not renormalizible are wrong in the sense that they make predictions that are falsiﬁed by experiments. If we were to calculate G(p1 . where m is the physical mass. Are the masses and coupling constants of elementary particles really fundamental constants. We use the same argument. A WORD ABOUT RENORMALIZATION 59 “throwing out an inﬁnite term. but we can always replace m0 ∞ → m. The good news is that we can calculate everything else. the “bare” mass m0 is not measurable.80) is not physically measurable. or can they be calculated on the basis of something else. we would ﬁnd that these divergent integrals also have the eﬀect of renormalizing the mass. This raises a profound and unanswered question. what is that something else? . A theory in which all inﬁnities can be sequestered in a ﬁnite number of physical parameters is said to be “renormalizible. It is remarkable fact that all the inﬁnities that occur to all orders of perturbation theory can be disposed of using these two maneuvers. It’s sometimes called the “bare” coupling constant λ0 . p2 ) to higher orders of perturbation theory. The bad news is that we can’t calculate m and λ.7.
INTRODUCTION TO FIELD THEORY .60 CHAPTER 4.
and a negative particle. a positive “particle. . and this in turn is responsible for many important properties. In a sense this is always true. Thus the formation of a hole entails the creation of two particles. but suppose they did. Then ψ(1. there is never enough energy in a semiconductor. 2) means that the particle numbered 1 was at the point x1 and it’s zcomponent of spin was σ1 . N ) (5.1) Let’s write a manyparticle wave function as follows: The notation 1. means x1 . to create a single electron. Suppose the particles had numbers on them like billiard balls. including the fact that atoms don’t simply collapse into their nuclei! Ordinary quantum mechanics is capable of describing such systems so long as the number of particles remains constant. for example. Second quantization is ideally suited to describe such processes. but ﬁrst we need to develop the ordinary quantum mechanics of manyparticle systems. σ1 . The wave function 61 . . but it is convenient to regard holes in the FermiDirac sea as real particles. Quantum mechanics requires that the wave functions for such ensembles be totally symmetric or totally antisymmetric. They don’t of course. for example. that’s the whole point. . 2.” the hole.Chapter 5 Identical Particles and ManyParticle States Ordinary matter consists of vast numbers of identical particles at low energy. the excited electron. .1 The Quantum Mechanics of Identical Particles ψ = ψ(1. 5. There are equivalent processes in Boson condensates as we shall see.
we expect Vij = Vji . That is what we mean by indistinguishable particles. . 3. Deﬁne the permutation operator Pij Pij ψ(. . . so all operators corresponding to observable quantities must be symmetric. From Newton’s third law. 1.j where Vij is the interaction potential between particles i and j. then O must be symmetric.) (5. . i.2) Here are a few simple theorems about permutations. j. There is a vast amount of experimental evidence to corroborate this. which follows by renaming the integration variables. We believe that all physical observables must be invariant under all permutations of the particles. The reader is invited to supply the simple proofs. . P ϕP ψ = ϕψ . . . since a lowenergy Hamiltonian will have the form p2 i + Vij H= 2m i i. It is well known that wave functions describing identical bosons must be symmetric with respect to the exchange of any pair of particles. . Our job is to construct a theory in which the sequence of numbers in ψ has no observable consequences. . IDENTICAL PARTICLES AND MANYPARTICLE STATES ψ(2. P ϕSP ψ = ϕP † SP ψ = ϕSψ (5. j. . . . . S] = 0 (5. . . . . This should be obvious. . There is also a deep result known as the spinstatistics theorem.62CHAPTER 5. . which shows that (5. 2. i. . Permutations have no eﬀect on the expectation values of symmetric operators.3) An important example of a symmetric operator is the Hamiltonian.5) 4.) = ψ(. For every symmetric operator S(1. . Functions describing identical fermions must be antisymmetric in the same sense. . . . If ϕOψ = P ϕOP ψ for all ϕ and ψ.4) . 1) means that the numberone ball has components x2 and σ2 . N ) we have [P.
Of course. . All the N ! permutations of N objects can be obtained by exchanging successive pairs of particles. iN ≡ √ N! (±)P P i1 . n2 . i2 . . We assume these are orthonormal. The symbol (±)P requires further explanation. . We’ll call the basic ingredient i α . i2 . . .9) . Put it another way. . . . A properly normalized state can be constructed as follows: 1 n1 . 5. . In this case (5. . . . .6) deﬁnes the symmetric. iN √ n1 !n2 ! · · · (5. there are ni particles in the ith state. . even and odd permutations constitute two disjoint sets. etc. The sum of all occupation numbers must equal to the total number of particles: ni = N i (5. i2 .6) The symmetrized and antisymmetrized basis states are can be written 1 S± i1 . The P in (±)P stands for the number of such exchanges required to achieve the corresponding permutation. .8) The numbers n1 . iN = 1 1 2 2 · · · iN N (5. i2 .5. to fermions. By this I mean that the ball that wears the number α is in the quantum state given by i.7) The sum goes over all of the N ! possible permutations of N objects. We can write an N particle state i1 . BOSON STATES 63 it is virtually impossible to construct a covariant theory that does not have this property.and antisymmetricmaking operators S± . . ij = δij .6) will not be normalized to unity. . there will be ni ! permutations that leave the N particle state unchanged. are called occupation numbers. .2 Boson States We must allow for the possibility that there are several particles in one quantum state. for example. n2 . . iN P (5. One way to make completely symmetric or antisymmetric states is simply to multiply singleparticle states in all possible combinations. It is a fact that no permutation can be reached by an even and an odd number of exchanges. Equation (5. If. the upper sign refers to bosons and the lower. = S+ i1 .2.
. . A typical basis element is written n1 . the space with two particles. ni + 1. the space with one particle.  . . . . n1 .13) The eﬀect of ai on the state  . . . . . As you will see. (5. . This can be seen as follows: Take the adjoint of (5. it is necessary to deﬁne operators that have this action. . ni . . . .  = 1 n1 . . ..n2 · · · and the completeness relation n1 . . . . . . . We allow for this by taking the basis states to be the direct sum of the space with no particles. .. ni . The basic operators for so doing are the creation and annihilation operators. ni . . . . . .. We are interested in precisely the situation in which the total number of particles is not ﬁxed. . . . . . . . n2 . . . . ni . . The i adjoint of this operator reduces by one the number of particles.n1 δn2 .10) (5. n2 .64CHAPTER 5. . . suppose a state has ni particles in the i’th eigenstate. .12) ie. . . . . . . . ai  . etc. ni − 1. . ni . .13) also tells us what the normalization must be. ni + 1. . has been to produce a state in which the number of particles in the i’th state has been reduced by one. . = ni + 1 Now replace ni everywhere by ni − 1. ni . . . they are formally equivalent to the raising and lowering operators associated with the harmonic oscillator. . . .  . In short ai  . . . . . n1 . √ √ = ni + 1 . n2 . . . . Eqn. for ni ≥ 1 (5. . . . a† increases by one the number of particles in the i’th eigenstate. . . = √ √ ni . (5. . There in no constraint on the sum of the ni . n2 . . . .11) Since there are physical processes that change the number of particles in a system. . . . . = ni (5. . . . . . . . = δn1 . . . . . . ni + 1. ni − 1. .12) and multiply on the right by  . .14) . = √ ni  . . . = i √ ni + 1 . For example. n2 . IDENTICAL PARTICLES AND MANYPARTICLE STATES All this assumes that there are exactly N particles. . ni + 1.n2 . . ni . . . . . (5. ai  . . ni + 1. . . . We introduce the creation operator a† by i a†  . The normalization is n1 .
. . = √ a† n1 !n2 ! · · · 1 n1 a† 2 n2 · · · 0 (5. . . . . nj . a† ] = δij j (5.15) If it were not for the i and j subscripts. twoparticle states i 1 √ a† i 2! 2 0 or a† a† 0 .2. . . . n2 . . . a† 0 . . . In this context they are called ladder operators or raising and lowering operators. . The eﬀect of ai and a† is to bump the system down or up one i rung of the i’th ladder. . . . There are several important results from harmonic oscillator theory that can be taken over to second quantization. . . . . aj ] = [a† . The entire state of the system is uniquely speciﬁed by these occupation numbers. = 0 65 The commutation relations are important. The rungs of the i’th ladder are labelled by the integer ni . . nj + 1. each one corresponding to one quantum state labelled by an index i. ni . . . .5. . . . (5. a† ] = 0. . . . . we can construct singleparticle states. j Finally ai a† − ai †ai  . ai  . . a† ] = 0. . . manyparticles states. the a† ’s commute as i well. ni − 1. Starting with the i vacuum state 0 with no particles. a] = [a† .16) . . ni = 0. the result is identically zero. . nj . . . . . BOSON STATES Of course if ni = 0.15) would be the commutation relations for the harmonic oscillator. . . One is that we can build up manyparticle states using the a† ’s. . . [a. 1 n1 . a† ] = 1. ai and a† also commute if i = j. . This is the essence of second quantization. and by the same argument. . . since it makes no diﬀerence in what order the diﬀerent states are depopulated. . i j or in general. nj . = In summary [ai . i √ √ √ √ ni + 1 ni + 1 − ni ni  . . . . ni . . . j √ = n1 nj + 1 . = a† ai  . Clearly all the ai ’s commute among one another. ni . . ni . . nj . i j [ai . . [a. Try to imagine a quantum system as an inﬁnite forest of ladders. since j ai a†  . . .
16) has the normalization required by (5. (5. • Show that the mysterious factor of by (5.3 More General Operators The matrix element of a singleparticle operator1 in a oneparticle state is just ˆ Oij = iOj . ni .20) as follows: N ˆ O= i. .16). . and I reserve the right to be inconsistent. however. . . we must distinguish between operators that work on speciﬁc individual particles and those that operate on all the particles in the system. • Show that (5. • Prove (5.10).j In a manyparticle system. = ni  .j 1 Oij α=1 i α jα (5.20) i.12) is in fact required 5. One way to do this is to simply sum over oneparticle states and generalize (5. . IDENTICAL PARTICLES AND MANYPARTICLE STATES Another familiar result is that the number operator Ni = a† ai i (5.21) I will sometimes use “hats” to emphasize that some particular symbol represents an operator. . ni . √ ni + 1 in (5.19) If the i ’s represent a complete set of states.18).17) is a Hermitian operator whose eigenvalue is the number of particles in the i’th quantum state. . . a† i n = n a† i n−1 Use this to do the following exercises. . . Ni  . I will do this only occasionally.18) Here is a useful result that you can prove by brute force or induction. we can represent the operator as ˆ O= Oij i j (5. . . .66CHAPTER 5. (5. ai .
i ≡ Nij  . . . ni . y)ϕk (x)ϕm (y). . . . . Taking this into account gives n i α=1 α jα  .25) insures that each interaction is counted only once. in this case n. . . ni .22) is the same as Ni deﬁned by (5.25) ˆ i. Nii as i deﬁned by (5. i j i.22) You see that Nij = a† aj is a generalization of (5. jO(2) k.4. ni + 1. ni . . . m = dx ˆ dy ϕ∗ (x)ϕ∗ (y)O(2) (x. . but these nj terms are still normalized as if they had nj particles in the j’th state and ni particles in the i’th state. . Finally ˆ O= ij Oij Nij (5. skip. . nj − 1. if one were dealing with hydrogen. nj .17). . .k.24) The factor 1 in (5. . m a† a† ak am . . . . . . . . . . .4 Field Operators I have used the symbol i to indicate a one particle “quantum state. . 5.17). . one for each particle in the state. . . . . . jO(2) k. ni + 1.23) From here it’s only a hop. (5. . Evidentally. . x). 1 ˆ O(2) = 2 where ˆ i. . . . i j (5. . i would stand for the discrete eigenvalues of some complete set of commuting operators. and jump to deﬁning twoparticle operators. √ ni + 1 = nj  . FIELD OPERATORS 67 The index α points to the individual particles. we would like to express it in terms of creation and annihilation operators.5. nj . but the sum over α is awkward. . . y) = O(2) (y.” For example. . .m (5. √ nj √ √ = nj ni + 1 . it produces nj terms in which j is replaced by i . = a† aj  . . . 2 ˆ ˆ since for identical particles O(2) (x. . . Note that when i α jα operates on an arbitrary N particle state. . nj − 1. .j. . nj . .
[ψ(x). (5.30) Combining (5.32) ψ † (x) . and ms . Then λ = i i iλ . (5. Use i to index one set and λ to index the other. The creation operators a† create particles in such a state.29) (5. none of them is the position operator. xi = ϕi (x). so that d3 x ϕ∗ (x)ϕj (x) = δij i and ϕ∗ (x)ϕi (x ) = δ (3) (x − x ) i i (5.68CHAPTER 5. First consider two basis systems based on two alternative sets of commuting observables. ψ † (x )]± = 0 [ψ(x). i Whatever these operators might be. it must also be true of creation operators. a† = λ i iλ a† i (5.27) Now remember that what we call a wave function in elementary quantum mechanics is really a scalar product on Hilbert space of the corresponding state and eigenstates of the position operator. An important question is what the creation operator formalism looks like in terms of the position operator.31) and its adjoint ψ(x) = i ϕi (x)ai is the creation operator in the basis deﬁned by the position operator. ψ † and ψ are called ﬁeld operators. IDENTICAL PARTICLES AND MANYPARTICLE STATES l.27) and (5. ψ † (x )]± = δ (3) (x − x ) (5.e. ψ(x )]± = [ψ † (x). i. Their commutation relations are important.28) We assume that the ϕi are a complete set of orthonormal functions. however.26) Since this is true for states.28) gives ψ † (x) = i ϕ∗ (x)a† i i (5. It creates a particle at x. m.
start with (5.5.32) is more or less obvious.36) j i j i 1 d3 x d3 y ψ † (x)ψ † (y)V (x. This can be seen as follows. t)ψ(x. · · · ]± to allow for both boson (commutation) and fermion (anticommutation) rules.” Many texts in fact. It i follows that ρ(x. is the reason why it’s called “second quantization.m d3 y a† a† ϕ∗ (x)ϕ∗ (y)V (x. FIELD OPERATORS 69 I have used the symbol [· · · .31) gives a† (t) = i d3 x ψ † (x. t)ϕ∗ (x) i .37) From (5.19) gives the matrix element Tij = d3 x ϕ∗ (x) − i h ¯2 2m 2 ϕj (x) = h ¯2 2m d3 x ϕ∗ (x) · ϕj (x). Inverting (5.33) i The last step involves an integration by parts. Start with the kinetic energy.4. 1 ˆ V = 2 d3 x i.32). y)ψ(y)ψ(x) 2 The total Hamiltonian operator is then = ˆ ˆ ˆ ˆ H =T +U +V (5.k.17) N = i a† ai is the total number of particles operator.24) and (5.32) and use it to infer the existence of the creation and annihilation operators. This. (5.30) Quantum mechanics according to Heisenberg was obtained by replacing classical quantities like P and L with noncommuting operators. Eqn.j. Eqn.34) The potential operator is similar. t) = ψ † (x. The ﬁrst line of (5. The second line follows from (5. y)ϕk (x)ϕm (y)ak am (5. t) is the spatial density operator. I suppose.35) Equations (5.23) becomes h ¯2 ˆ T = 2m d3 x a† ϕ∗ (x) · i i ij ϕj aj (x) = h ¯2 2m d3 x ψ † (x) · ψ(x) (5.25) give us to form of any twoparticle operator. ˆ U= d3 x ψ † (x)U (x)ψ(x) (5. Our formalism takes the further step of making the wave functions themselves into operators via (5.(5. (5. Let’s see what ordinary quantum mechanics looks like in this formalism and then see what new aspects appear.
t)] = −eiHt/¯ [H. t)ψ(x .37). t) i d3 x = = Notice that d3 x ψ † (x.39) The commutator is something of a challenge. We’ll do it one term at a time.70CHAPTER 5. 0)e−iHt/¯ (5. t)ψ † (x. t)0 = δ (3) (x − x )ψ † (x. t) = −[H. t)ψ(x. 0)]e−iHt/¯ ∂t (5. t)ψ † (x. The basic relation is h h ψ(x. ψ(x)] = − d3 x ϕ∗ (x ) 2 ϕj (x )a† aj . Field theory calculations are typically done in the Heisenberg picture. t) a† ai = i i d3 x ψ † (x. This is consistent both with our interpretation of ψ † as a creation operator and with our interpretation of ρ as a particle density operator. ϕk (x)ak i i 2m ij k . The Heisenberg equation of motion is obtained by diﬀerentiating (5. ψ(x. t)0 = ψ † (x . 2 h ¯ [T. t) = ρ(x . t) d3 x ψ † (x. t)0 The point of this is that ψ † (x. however. IDENTICAL PARTICLES AND MANYPARTICLE STATES ai (t) = N= i d3 x ϕi (x)ψ(x) d3 x ϕi (x )ψ(x . t)ψ(x . t) + δ (3) (x − x ) 0 = δ (3) (x − x )ψ † (x . t)ψ(x . ψ † (x. We have been working in the Schrodinger picture.38) Where H is the operator given by (5. t)0 is an eigenstate of ρ with eigenvalue δ (3) (x−x ). t) does indeed create a pointlike particle at x and time t. t) = eiHt/¯ ψ(x. 5. t)δ (3) (x − x ) d3 x ρ(x.38). i¯ h ∂ h h ψ((x). ψ(x. t)ϕ∗ (x. so we must derive an equation of motion. t)0 = ψ † (x . t) ψ † (x.5 Field Equations The ﬁeld operators so far have had no time dependence.
42) gives the equation of motion i¯ h ∂ ψ(x.33) for T and (5.43) The ﬁrst line looks like the conventional Schrodinger equation but with ﬁeld operators rather than the usual wave functions.37). ψ(x)] = d3 xU (x )[ψ † (x )ψ(x ).5. x )ψ(x )ψ(x ).93) for fermions.40) The potential energy is rather easier. The last term is nonlinear and has no counterpart in singleparticle systems.42) Incidentally.15) for bosons or (5.36) gives [V. to reduce the product of three ai ’s. (5. (The reader is invited to work through these derivations to see how this comes about. and (5. t) (5. ψ(x)] =− d3 x ψ(x )V (x. either (5. ψ(x)] = 1 2 d3 x d3 x [ψ † (x )ψ † (x )V (x .41). x )ψ(x )ψ(x) (5.5. t) = ∂t − h ¯2 2m 2 + U (x) ψ(x.39). the interaction term. t)V (x. A straightforward application of (5. [U.40). At this point we use the commutation relations. x )ψ(x . ψ(x)] = −U (x)ψ(x) (5. ak ] = −aj δki i [T. t)ψ(x.31) for ψ.41) Finally. t)+ d3 x ψ † (x . [a† aj . (5. FIELD EQUATIONS 71 I have used (5.32) and (5. Finally combining (5. ψ(x)] = h ¯2 2m d3 x i ϕi (x)ϕ∗ (x ) i h ¯2 2m 2 j ϕj (x )aj = h ¯2 2m d3 x δ (3) (x − x ) 2 ψ(x ) = 2 ψ(x) (5. (5. these results are valid for both boson and fermion ﬁelds so long as one keeps the ﬁelds ordered in exactly the way shown above. .
k k (5. d3 x ϕ∗ (x)ϕk (x) = δk.36) should only be a function of the diﬀerence of the two coordinates.6 Momentum Representation As we have seen before. ±1.72CHAPTER 5.44) V We assume the usual periodic boundary conditions force the momentum eigenvalues to be nx ny nz k = 2π . IDENTICAL PARTICLES AND MANYPARTICLE STATES 5. Eqn.47) For the potential energy ˆ kU k = d3 x ϕ∗ (x)U (x)ϕk (x) = k ≡ 1 V d3 x ei(k−k )·x U (x) (5.46) V We can now calculate matrix elements of the Hamiltonian. . In this case we work exclusively in a ﬁnite volume with discrete momentum eigenvalues. With this normalization. (5. it’s usually easier to formulate a theory in position space and easier to interpret it in momentum space.k (5.49) V (x) = 2 Anyone who confuses the volume V with the potential V will be taken out and tied to a tree. The twoparticle potential in (5. V (x.33) gives ˆ kT k = d3 x ϕ∗ (x) − k h ¯2 2m 2 ϕk (x) = h ¯ 2 k2 δ 2m k. y) = V (x − y). . · · · independently. As with U (x) we can take the Fourier transform2 Vq ≡ d3 x e−iq·x V (x) 1 V Vq eiq·x q (5. (5. the eigenfunctions are orthonormal.45) Lx Ly Lz Where each of the n’s can take the values 0. ie.48) 1 Uk−k V This is the Fourier transform of U (x) with respect to the momentum k − k . ±2. The basic eigenfunctions are 1 ϕk = √ eik·x (5.
to a good approximation. The two ﬂuids – the normal and the superﬂuid components – are interpenetrating.18K it enters the superﬂuid state with some remarkable properties. k V (k − k )p.51) k p ∆ The meaning of the sum over ∆ is that one sums over all values of k. In an ideal gas an elementary excitation corresponds to the addition of a single quasiparticle in a momentum eigenstate. Bose “particles”. At very low temperatures the density of the normal component vanishes. After this scattering process their new momenta are k and p .p −p δq.k 1 2V Vk−k a† a† ak ap (5. and their densities depend on temperature. Helium does liquify at T ≈ 4K and at the lambda point Tλ = 2. Another is the existence of quantized vortices.7 4 He Superﬂuids atoms have zero angular moment and as such are. SUPERFLUIDS The matrix element of V is then p .50) (¯ k)2 † h 1 ak ak + 2m V Uk −k a† ak + k k . p. the situations is reversed: the superﬂuid density tends towards zero while the normal density approaches the density of the liquid.7. Near the transition temperature to the normal state. k . The phonons have a linear dispersion relation E ∝ k. This point is crucial to the phenomena of superﬂuidity as can be seen from the following argument. Two particles with momenta k and p interact in such a way that there is a momentum k − k = p − p transferred between them. while the density of the superﬂuid component approaches the total density of the liquid.5. One of the most striking is the ability to ﬂow through narrow channels without friction. p such that k + p = k + p . superﬂuidity is explained in terms of elementary excitations. At a more microscopic level. Imagine helium ﬂowing . Helium is unique in that it does not crystalize even in the limit of zero temperature (except under extreme high pressure). For small momenta the excitations in liquid 4 He are sound waves or phonons. This is easy to interpret. The occurrence of frictionless ﬂow can be described in a rough qualitative way by the twoﬂuid model. 5. k = The complete Hamiltonian is then H= k 73 1 V Vq δq.k−k q (5. which is to say that they are massless particles. The reason is that it is so light that its endpoint oscillations have suﬃcient energy to prevent the formation of a lattice.
no quasiparticles are excited and the ﬂuid ﬂows unimpeded and lossfree through the tube. Call the ground state energy of the ﬂuid E0 and its momentum. and this is only kinematically possible if it lowers the total energy of the ﬂuid. IDENTICAL PARTICLES AND MANYPARTICLE STATES Coordinate System Ground State GS + One Quasiparticle Excitation Energy S E = E0 P=0 E = E0 + (p) P=p ∆E = (p) S E = E0 + M v2 /2 P = −M v E = E0 + (p) − p · v + M v2 /2 P = p − Mv ∆E = (p) − p · v Table 5. ie. 11. In S it would appear that the tube was dragging a thin layer of ﬂuid along with the it. In S the tube is decelerating this same layer. If the quasiparticles had ﬁnite mass so that (v) = p2 /2m. This means there is a minimum velocity (5. Bogoliubov. N. If the ﬂow velocity is smaller that this. this deceleration comes about through the creation of quasiparticles. Bogoliubov.S. J. This was done originally by N.N. then vmin = 0 in the limit p = 0. and there could be no superﬂuidity. We can describe this in terms of two diﬀerent coordinate systems.74CHAPTER 5. and in system S the tube is at rest and the helium has velocity −v.53) vmin = .52) N. P = P − Mv E = E − P · v + M v2 /2 If the ﬂuid is ordinary. 23 (1947) . Table 1 shows the relevant kinematics. p This is called the Landau criterion. U. Only when ∆E < 0 does the ﬂowing ﬂuid lose energy.R. Momentum and energy are related by the Galilean transformation. In the limit T = 0. Evidentally S is moving relative to S with velocity v. It is essential that /p pass through the origin with a ﬁnite positive slope. Let’s try to create our ﬁrst quasiparticle in S with energy (p) and momentum p. In the system S the helium is at rest and the tube is moving with velocity v. Phys. there will be friction between the ﬂuid and the walls of the tube. not a superﬂuid.3 and is one of the crown jewels of manybody 3 (5.S. It is possible to calculate the dispersion relation for lowenergy quasiparticles using the ﬁeld theory formalism we have just developed. P.1: Kinematics for Quasiparticle Formation through a tube at constant velocity.
51). because they would not conserve momentum. then in the limit of zero temperature. This deep result is called Noether’s theorem. Bogoliubov argued that the a0 and a† operators in (5. The crucial approximation consists in ignoring interactions between excited particles and keeping only those terms corresponding to particles in the ground state interacting with one another and terms in which one excited particle interacts with one particle in the ground state. We are left with i H≈ k 1 1 (¯ k)2 † h ak ak + V0 a† a† a0 a0 + 0 0 2m 2V V + 1 2V (V0 + Vk )a† a0 a† ak 0 k k=0 Vk (a† a† a0 a0 + a† a† ak a−k ) 0 0 k −k k=0 (5.5. Each term has the same number of creation and annihilation operators. that if the particles do not interact with one another.1 Bogoliubov Theory We have derived the relevant Hamiltonians.51) only those terms that have at least two of the ground state creation and annihilation operators ai and a† .55) We have ignored those terms with three a0 ’s. one is annihilated and vice versa. ψ → ψ = e−iα ψ. is conserves the number of particles in the system. I will take you through the Bogoliubov calculation now and come back to the fundamental issues later.54) Moreover (5.7. If we introduce a weak twoparticle interaction via V in (5. this calculation introduced into physics some profound ideas about symmetry breaking that are important in theories of fundamental interactions.37) in coordinate space and (5. (5. Since we are not concerned with the eﬀect of external potentials. 5.37) in invariant under the simple gauge transformation. That means keeping in (5. then it is plausible that most of the particles remain in the ground state.7.51) obeys a simple conservation law. these two properties are closely related. SUPERFLUIDS 75 theory. they are all in the ground state. Though it’s by no means obvious. Notice that (5. (5. we set U = 0.55) 0 . For every particle created.51) in momentum space.33) through (5. I will have much more to say about it later. For every continuous symmetry transformation that leaves a Hamiltonian invariant there is a corresponding conservation law. Though this was not realized in 1947. We know from statistical mechanics.
The third is the exchange . or Fock term. We can get some insight about the approximations made so far by noticing the following.58) in two related ways: the classical density ρ0 and the k = 0 exclusion in the summation. we can neglect the 1 in N0 + 1. We are treating the ground state like a classical ﬂuid and all the other particles as a dilute quantum ﬂuid. we would have gotten (5. both equal to the particle density. the system is in an eigenstate of a† a0 (with eigenvalue N0 ) and of a0 a† (with eigenvalue N0 +1). The original Hamiltonian conserved particle number: for every creation operator there . The ﬁrst is the kinetic enk ergy. The second. IDENTICAL PARTICLES AND MANYPARTICLE STATES √ √ can be replaced with cnumbers. For one thing. This right side goes to zero as V → ∞. is called the Hartree energy . N0 V0 /V .55) becomes (¯ k)2 † h 1 2 H≈ a ak + N V0 2m k 2V 0 k + N0 V k=0 1 (V0 + Vk )a† ak + Vk (a† a† + ak a−k ) k k −k 2 (5. 0 1 1 [a0 .56) consist of two terms. a† ]N0 = N0 0 V V (5. With that replacement (5. in which an atom in the state k is scattered into the zeromomentum state. 0 0 It ﬁgures that since N0 is a number on the order of 1020 or 1022 .57) The a† ak operator has three contributions. The meaning is clear. To this extent a0 and a† commute! Another point 0 the that the a0 ’s and a† ’s are all divided by the volume V . while a second atom is simultaneously scattered from the condensate to the state k. I should mention a related matter.59) diﬀers from (5. We obtained (5. k (5. which comes from the direct interaction of a particle in the state k with the N0 atoms in the zeromomentum state.58) (5. You see that (5.59) where ρ0 = N0 /V . a0 → N0 .51) from (5.56) The left side of (5.57) directly. a† → N0 where N0 is the 0 number of particles in the ground state.36) by replacing 1 ϕ(x) = √ V If we had made the replacement ϕ(x) → 1 √ ρ0 + √ V ak eik·x k=0 ak eik·x .76CHAPTER 5.
since it does not conserve particle number. The procedure for doing this was originally invented by Bogoliubov in this context. We will have much more to say about this later on in the course when we discuss spontaneous symmetry breaking.” since we are treating the ground state classically. Vk = V−k .61) can be written in a more symmetric fashion. In order to study the physical states.57) gives H≈ k=0 Ek a† ak + k N2 N V0 + 2V 2V Ek = Vk (a† a† + ak a−k ) k −k k=0 (5. that’s something that can be measured. Eqn. The two can be related as follows: N = N0 + a† ak (5. (5.57) no longer does.62) 2m V From (5.60) into (5.7. J. Smith.64) 4 I am following the development in BoseEinstein Condensation in Dilute Gasses. but this is not something we know ahead of time.63) 5. we must diagonalize H.5. The particles simply appear out of or disappear into the ground state where they are “invisible. the number of particles in the ground state.7. but the number of excited particles is obtained via an operator operating on the state function.60) k k=0 You see the similarity between (5. (5.57) refers to N0 .63). In order to work this out in a generic fashion (and also to keep the algebra under control). SUPERFLUIDS 77 was a corresponding annihilation operator.49).59). Pethick & H.61) N k2 + Vk (5. so that (5. but systems with a deﬁnite number of particles in the various energy levels are not eigenstates of (5. Eqn. H≈ N2 V0 + 2V Ek (a† ak + a† a−k ) + k −k k>0 N Vk (a† a† + ak a−k ) k −k V (5. We do know the total number of particles. and the technique as since been applied to many diﬀerent problems.2 The Bogoliubov Transformation Physically realizable states must be eigenstates of the Hamiltonian. C. Cambridge (2002) . N0 is a cnumber.60) and (5.63) has the form H = 0 (a† a + b† b) + 1 (a† b† + ba) (5.4 Each term in the sum in (5. I will pose the problem like this. Substituting (5.
72) − 2 1 (5.78CHAPTER 5. or since h = 1. whatever they may be.68) (5. Of course ¯ [a. 0 and 1 are arbitrary numbers with units of energy. IDENTICAL PARTICLES AND MANYPARTICLE STATES In this particular instance.69) into (5. and resulting states. α† ] = 0 (5.64). we will diagonalize the Hamiltonian. The inverse of (5.67) and use (5. H = 2v 2 0 − 2uv 1 + [ 0 (u2 + v 2 ) − 2uv 1 ](α† α + β † β) (5.66) is a = uα − vβ † . b† ] = 0 We would like to ﬁnd linear combinations of these α = ua + vb† . u2 − v 2 = 1.64).66) (5. and [a. After a bit of algebra we get. b† ] = [a† . of frequency. (5. physical states. In other words.67) • The only operators appearing in the Hamiltonian are α† α and β † β. b† ] = 1. β † ] = [β. 2 1 (u + v 2 ) − 2uv 0 =0 (5. α† ] = [β. but the ﬁnal result will hold with the generality of (5.68) and (5.66) into (5. (5. b] = [a† . i. will be the observable.69) (5.e. Insert (5.70) +[ 1 (u2 + v 2 ) − 2uv 0 ](αβ + β † α† ) We would like the last term to vanish. and β = ub + va† such that • [α. b = uβ − vα† . β † ] = 1.73) .65) Now substitute (5.71) gives u2 = where = 2 0 1 2 0 +1 and v 2 = 1 2 0 −1 . a† ] = [b. Assuming u and v to be real we get. and [α. a = ak and b = a−k .65).71) Combining (5. b] = [a.
77) This new ground state plays the same role for the αk ’s as the quantum vacuum 0 plays for the ak ’s. that are excited out of the ground state.5.59). I’m sure this was not realized back in 1947 when this work was originally published. the new vacuum. helium atoms say. Where convenient I will use n ≡ N/V .70) gives H = (α† α + β † β) + − 0 (5.78) .61) or (5. but we are standing on the threshold of a profound new idea. SUPERFLUIDS 79 It is necessary to choose the positive branch of the square root so that u and v are real. (5. 0 → k2 + nVk 2m k2 2m 2 1 → nVk nk2 Vk m k2 + nVk 2m † ωk αk αk k=0 → ωk ≡ H≈ N2 V0 + 2V N2 2V + (5. (5.31) to The Hamiltonian (5. αk Ω = 0 for all k.63). Our new creation and destruction operators αk and αk refer to the quasiparticles discussed earlier.51) and the ﬁeld operator (5. Substituting these into (5. This solution makes an interesting point about perturbation theory.75) † † ωk (αk αk + βk βk ) + ωk − k>0 = V0 − 1 2 k=0 k2 + nVk − ωk 2m + (5. The ground state of the system is that in which there are no quasiparticles. Let’s calculate the number of real particles. I will get back to that in a minute. we automatically make a transition to a new vacuum with some unusual properties. The point is that in making the transition from the Hamiltonian (5.74) It’s time to make contact with our original Hamiltonian. Nex = Ω k=0 a† ak Ω = Ω k k=0 † 2 vk αk αk Ω = k=0 2 vk (5.76) The Hamiltonian consists of the groundstate energy and a sum of oscillators † of energy ωk .63) and ﬁeld (5. The parameter we have been calling is best thought of as a frequency. It is in some sense.7.
81) ωk = It is possible to ﬁnd an exact form for this operator. it must be that αk U 0 = 0. Look at (5. II. Since ak 0 = 0. After a horrendous integration we get the simple result nex ≡ Nex m3/2 (λn)3/2 = V 3π 2 (5. (5. i.J. It’s a sobering thought that in scattering theory we always use perturbation theory. It has a branch point.78) with an integral. there must exist an unitary operator U such that αk = U ak U −1 ak = U −1 αk U.80) m The linear dispersion relation is the hallmark of massless particles. You can’t make a power series expansion around a branch point. V (x) = λδ(x).82) 5 This is assigned as Problem 17. Vol. If we were doing ordinary perturbation theory.e. In this problem perturbation theory. so that Vk = λ.80CHAPTER 5. Aitchison and A. (5.79) Now λ3/2 considered as a complex function of λ is nonanalytic at λ = 0. We could say that (5. (5.5 but for our purposes it is enough to know that it exists. If there were no potential or the potential somehow vanished at low energy. It’s clear that ωk from Eqn. we would pray that λ was “small” and expand the solution in powers of nλ. Now let’s get back to the ground state. In the limit of small k we have nV0 (5. We aren’t smart enough to do anything else. I will take the simplest possible potential.75) is the energy of a single quasiparticle. G. would always get the wrong answer.75) interpolates between phononlike behavior at small k and massive particlelike behavior at large k.R. Since the αk ’s and ak ’s are related by a canonical transformation. We can replace the sum in (5.77). Institute of Physics Publishing. 2004 . IDENTICAL PARTICLES AND MANYPARTICLE STATES To make my point. the potential creates massless particles. the dispersion relation would be ωk = vk with v = k2 . It must be that Ω = U 0 . I. in the usual sense. J. In this case we can get a solution which is “exact” in the sense that we don’t need this approximation. And it could all be wrong. it just doesn’t make sense. 2m This is the relation for an ordinary particle of mass m. Hey.4 in Gauge Theories in Particle Physics.
let’s guess that because we are not making any approximations. We had an intimation of this.86) This transformation can be brought about by a unitary operator Uα .84) would survive. Suppose we could do the theory exactly (whatever that means).77).75) we get ϕ(x) = ρ0 ˜ Then from (5. Ωϕ(x)Ω = ρ0 ˜ 1/2 1/2 (uk αk − vk α−k )eik·x k (5. We postulate that something of (5.17).58) for any state with a deﬁnite number of particles will always vanish.54).5. the Hamiltonian would preserve the U (1) symmetry discussed in connection with (5.69) and (5. therefor it is not an eigenstate of the number operator (5.83) We can rewrite (5.85) While we are speculating. The expectation value of the ﬁeld ϕ(x) from (5.59). SUPERFLUIDS 81 The exact form of U shows that Ω is a linear superposition of states with diﬀerent numbers of particles. Our original Hamiltonian is invariant under the simple U (1) symmetry transformation. ϕ = ˜ Uα ϕUα . but this is not true of our new ﬁeld (5. ˜ (5. ϕ(x) → e−iα ϕ(x).7. Then ˜ −1 ΩUα ϕUα Ω = e−iα ΩϕΩ ˜ −1 ˜ (5. since the Hamiltonian (5. It turns out that the breaking of this symmetry is directly responsible for the existence of the massless quasiparticles. that in fact Ωϕ(x)Ω = 0. Let N be any state with N noninteracting particles. It will turn out that the conservation of particle number is related to gauge transformations. ˜ 1 +√ V 1/2 (5.87) . Using (5. Then N ϕ(x)N = 0. Then we should not change any physical results by replacing ϕ(x) → ϕ (x) = e−iα ϕ(x) ˜ ˜ ˜ (5.57) does not conserve particle number. and ﬁnd the “true” ﬁeld and vacuum state.59) in terms of the αk ’s.84) All this was obtained as the end result of a number of approximations. I will prove this assertion when we get around to the Goldstone theorem. N ϕ(x)N = ρ0 . whereas the modiﬁed Hamiltonian is not.
the second entry on the left refers to particle 2.87) becomes Ω.8 Fermions The fact that fermion wave functions are antisymmetric introduces a few small complications regarding the creation and annihilation operators. 5. and much of the physical world is the end result of this sort of symmetry breaking. IDENTICAL PARTICLES AND MANYPARTICLE STATES −1 I choose to interpret (5. which is to say that the leftmost entry in the ket refers to particle 1. When I write i1 . In this case the broken symmetry is rotation invariance. but when the iron is cooled below the Curie point. αϕΩ. so double occupancy is impossible. i1 . i2 . I mean that particle 1 is in state i1 . the ferromagnet. The Hamiltonian gives rise to an inﬁnite number of alternative ground states.84) corresponds to the special case α = 0.82CHAPTER 5. If I describe this state in terms of occupation numbers n1 . i2 = −i2 . there is another example of spontaneous symmetry breaking that you’re all familiar with. Then (5. Thus rotation invariance is broken.54). . n2 . i1 = −i1 . Surely the Hamiltonian for the interactions of atoms and electrons is invariant under rotation. This is called “spontaneous symmetry breaking. and so an inﬁnite mass of molten iron is also. It is postulated that our exact Hamiltonian is invariant under the U (1) transformation (5. It is widely speculated that the “Hamiltonian of the universe” has many symmetries of which we are unaware. and in choosing one of them we break the symmetry. etc. but we are living in or close to the ground state of this Hamiltonian. It’s time to summarize the plot. resulting in a ground state that has a preferred direction. but which state is the ﬁrst state? You have to decide on some convention for ordering states and then be consistent about it. α = e−iα ΩϕΩ ˜ ˜ (5.” and I will have much more to say about it later. Antisymmetry then decrees that i1 . the leftmost entry refers to the ﬁrst quantum state (rather than the ﬁrst particle). i1 . but the choice of α can have no observable physical consequences. In the meantime. α as an alternative ground state. If both particles were in the same state i1 . They are easy to explain looking at twoparticle states. however. the iron spontaneously (spontaneously – get it?) “chooses” one direction to line up all the magnetic moments. (5.87) as follows.88) Obviously there are an inﬁnite number of these ground states. I regard Uα Ω ≡ Ω.
. Consider the following two expressions. . . consult F. .8. If you want to see all the excruciating details.and twoparticle operators just as we did for bosons. .91) The eﬀect of the operator a† must be i a†  . i (5. i2 i i a†2 a†1 0 = i2 . a†1 a†2 0 = i1 . Schabel Advanced Quantum Mechanics. fermions don’t have numbers painted on them any more than bosons do.90) (5. .6 6 Strictly speaking.5. Obviously a†1 a†2 + a†2 a†1 = 0 i i i i We say that fermion operators anticommute. . i1 = −i1 . Here the convention is n1 . . . The usual notation is [A. = η . Section 1. a† } = δij j (5. and again this is a convention. The results are {ai . .93) One can construct one. . i2 i i I have decreed. The comi mutation relations of the ai ’s and a† ’s are obtained by arguments similar to i those leading to (5. FERMIONS 83 These states are constructed with creation and annihilation operators as in the case of bosons.92) where η = 0 of ni = 1 and otherwise η = +/−.15).4. I should prove this. ni + 1.23) and (5.24) are correct as they stand. · · · = a† i n1 a† 2 n2 · · · 0 ni = 0. ni . . . n2 . but one must be careful to keep the creation and annihilation operators in the exact order in which they appear in these equations. a† } = 0 i j {ai .89) Of course. B} = AB + BA (5. Formulas (5. 1 (5. etc. that the ﬁrst operator on the left creates particle 1.2 . but now we must be more careful about ordering. so we must use occupation numbers. depending on the number of anticommutations necessary to bring the a† to the position i. aj } = {a† . B]+ = {A.
(This is not true in relativistic theory. IDENTICAL PARTICLES AND MANYPARTICLE STATES 5.) Sometime we need to specify that a particle is in a speciﬁc spin state. where nx is an integer. dnx = L dkx . If we say that there are ni particles in the i’th eigenstate. 2π (5.8. The point is that spin and space degrees of freedom are completely decoupled. This can be done in cartoon fashion by writing ψ↑ or ψ↓ . 5. Let’s try to make that more precise. For example. (5. If we impose periodic boundary conditions along x axis for example. Suppose the system is contained in a cube of length L on a side.2 The Fermi Sphere and the Debye Cutoﬀ The ground state of a system of bosons is one in which all the particles have zero energy. t)uσ (t). Because of the exclusion principle.96) . T = h2 k 2 /2m. t) = ψ(r. It’s easy to incorporate spin into Fockstate formulation. we must remember that a complete speciﬁcation of the i’th eigenstate must include the spin projection σi . Since the kinetic energy is a function of k 2 .8.84CHAPTER 5. t). or kx L = 2πnx . the operator a† creates a particle in a momentum and spin eigenstate with kσ momentum k and zprojection of spin σ. This is called the Fermi momentum hkF . The wave function for a spin1/2 particle is just ψσ (r. The ground state can be constructed ¯ in momentum space as Ω = k k<kF a† 0 kσ σ (5. In the case of fermions the ground state is that in which all the energies are as small as possible.94) The spin wave function uσ is a twocomponent column matrix.1 Inclusion of Spin Spin is just “tacked on” to nonrelativistic quantum mechanics. this is not possible with fermions. The space wave function is ψ(r. The subscript σ is an index that takes on the values ±1/2 corresponding to spin “up” or “down” (with respect to some axis of quantization).95) The set of all wave numbers k < kF is called the Fermi sphere. there will be a ¯ certain momentum associated with the highest occupied energy level. the wave numbers are limited by the condition eikx x = eikx (x+L) .
It follows that 3 kF = 3π 2 n. the allowed values are k= π 2π 3π (N − 1)π .98) where n = N/V is the average particle density.−σ and bkσ ≡ a−k. so that no atom could move! It doesn’t make sense to have halfwavelengths . k1 σ1 = a† 2 σ2 ak1 σ1 Ω k (5.8. L L L L (5. We can deﬁne † hole creation and annihilation operators as bkσ ≡ a† −k.100) Bosons do not obey the exclusion principle.99) One can think of this as the creation of an electronhole pair. The number of states in the Fermi sphere is then N = 2V 0 kf 3 V kF d3 k = (2π)3 3π 2 (5. The normal modes will be standing waves in which there are an integer number of halfwavelengths in the length L. Typically it is on the order of a few eV ’s. The fact that the oscillators are discrete imposes limitations on the wave numbers that can exist. called the Debye cutoﬀ. k1 σ1 = a† 2 σ2 b† 1 . Consequently. In terms of creation and annihilation operators. Consider a line of length L consisting of N + 1 atoms spaced at a distance a.−σ . The simplest excitation of this system is obtained by promoting an electron from inside the sphere to an energy level outside.−σ1 Ω k −k (5. such a state is written k2 σ2 . Phonons in crystals are sound waves in which the crystal planes oscillate longitudinally along the direction of propagation. (5. h At absolute zero temperature all the electrons would be inside the Fermi sphere. FERMIONS 85 where dnx is the number of allowed momentum states in the interval between kx and kx + dkx . there would be a node at the site of each atom. however. . Then (5.5. The Fermi energy is deﬁned by F = (¯ kF )2 /(2m). Assume that the ﬁrst and last atoms don’t move because they are at the surface of the crystal. so there is nothing exactly like the Fermi sphere for integerspin particles.101) Notice that if k = N π/L.97) The factor of 2 in front of the integral arises because there are two allowed spin states for every one momentum state. ··· . There is something similar for lowenergy phonons in crystalline solids. .98) becomes k2 σ2 .
Metals in this state have many other remarkable properties as well. The calculation is identical to (5. 5. The point is that (5. Schrieﬀer. 1175 (1957) . so the number of allowed modes is one less that the number of atoms – just those given in (5.101). Kittel. they are perfectly diamagnetic. Bardeen. 162 (1957). (5. Rev.86CHAPTER 5. This is called the Meissner eﬀect. 106.103). Cooper.97) (without the factor of 2). IDENTICAL PARTICLES AND MANYPARTICLE STATES smaller than the interatomic separation. See C. For example. J. Our basic ideas about superconductors come from a classic paper by Bardeen. If we call that number N also. This allows us to deﬁne a cutoﬀ frequency7 ωD = (6π 2 v 3 N/V )1/3 . Phys. Here I want to concentrate on the basic physics of the superconducting state. We can now calculate the number of modes with wave number less than k.105) It will be relevant to our discussion of superconductors that the Debye energy h ¯ ωD is usually several orders of magnitude smaller than the Fermi energy. 8 J. ω = vk (5.102) We know that the number of modes is just equal to the number of atoms that are oscillating. (5. 108. Nonetheless. k N =V 0 V k3 d3 k = .103) In the longwavelength limit. Introduction to Solid State Physics. N. This rich phenomenology will be covered in the spring quarter of your solid state physics class.104) holds in the longwavelength limit. (2π)3 6π 2 (5. L.105) is not an exact result. the energy and frequency are simply proportional.8 They were awarded the Noble Prize for 7 Unlike (5. Magnetic ﬁelds are excluded except for a small penetration depth at the surface.104) where v is the velocity of sound. and Schrieﬀer. whereas the Debye cutoﬀ argument is based on shortwavelength behavior. Cooper.9 Superconductors Some metals when cooled to a temperature close to absolute zero enter a distinct thermodynamic state in which current ﬂows without resistance. R. (5. kD = (6π 2 N/V )1/3 (5. these ideas are at the heart of several useful results for solids at low temperature.102) gives us a formula for the maximum allowed k.
The phonon waves are purely longitudinal. so there is a strong coherence among the pairs. N. The crucial diﬀerence is that their wave functions overlap.5. Dq is the coupling “constant. and electrons rush out of the Fermi sea to make pairs. It was L. we must allow for them to scatter from electrons. through the exchange of phonons. and the body of theory has come to be called “BCS theory. The pairs act as bosons. 5. the net coulomb force is still repulsive. In this respect they are a threedimensional generalization of compression waves on a rubber band.1 Eﬀective ElectronElectron Interaction Phonons in this context are acoustic waves in crystals. They arise because the crystal lattice can “ﬂex” slightly. so I will just take (5.” Electrons of course repel one another. however. The derivation of this constant involves some assumptions about the mechanical properties of the lattice. they have no polarization. This is really a topic for a solid state class. Phonons are massless particles. but as a ﬂuid with a rigid phase. so they can “condense” much like the helium atoms in a superﬂuid. the classical analogy is apt. The pairs move. It’s not easy to scatter one of the pairs. which we will ignore. and even with the screening eﬀect of the lattice. It is this that accounts for the lack of electrical resistance. SUPERCONDUCTORS 87 this. because it’s locked “in step” with all the others. They can interact in another way.q Here ck is the annihilation operator for electrons. the frequency and energy are proportional. I will concentrate on the central ideas and make numerous idealizations and approximations along the way. If we think of phonons as particles with deﬁnite momentum. Cooper who ﬁrst realized that this interaction produces a weak attractive force over a narrow range of electron energies.” Since the energy of an electron in a pair is less than the Fermi energy.” It actually has some momentum dependence. H = Dq c† ck (aq + a† ) (5. The Hamiltonian for this interaction is derived in every advanced text on solid state physics.106) −q k+q k.106) .9. This is the barest outline of an enormously rich and elegant body of theory. For this reason they are able to spontaneously lock into phase with one another with a macroscopic coherence length. In both cases. the ground state becomes unstable.9. Since we will be dealing with phonons in the longwavelength approximation. not as individual particles. This force is suﬃcient under some circumstances to produce a weakly bound spinzero state of two electrons called a “Cooper pair. and aq is the corresponding operator for phonons.
S] + · · · . there is the freeparticle Hamiltonian. Suppose you have a Hamiltonian of the form H = H0 + λH .108) The parameter λ is supposed to be “small. H0 = q h ¯ ωq a† aq + q k † k ck ck . S] + [[H.107) The exchange of a phonon between two electrons is at least a twostep process. IDENTICAL PARTICLES AND MANYPARTICLE STATES as “given. 2 (5.106) to get an eﬀective electronelectron Hamiltonian. S] = 0 (5. even if we had no theory to ﬁnd Dq . S= k. .109) bearing in mind that S is itself ﬁrst order in λ. This is the Hamiltonian we would guess.” The signiﬁcance of the terms is clear enough. S]. The second term gives the emission of a phonon of momentum q leaving an electron with momentum k − q. We can make an educated guess at the form of S (with λ = 1).110) Substitute (5. (5.” We will eventually set it equal to one.110) into (5. It’s really just a bookkeeping device.q 9 (Aa† + Baq )Dq c† ck −q k+q (5. (5. There is an elegant trick for doing this without any special formalism. so we need to do something like secondorder perturbation theory to combine two factors of (5. This ﬁrst term corresponds to an electron of momentum k absorbing a phonon of momentum q and going on its way with momentum q + k.111) 2 That commutator will be our secondorder interaction Hamiltonian. In addition to the interaction term.109) such that all ﬁrstorder terms in λ are transformed away. 1 ˜ H = H0 + [λH .112) The solid state community calls this the SchrieﬀerWolﬀ transformation.88CHAPTER 5.9 This can be accomplished by λH + [H0 . Our strategy is to ﬁnd a transformed Hamiltonian 1 ˜ H ≡ e−S HeS = H + [H. S] + O(λ3 ) (5.
i. and states with one electron and one boson with similar states with one boson and one electron. • Our new secondorder interaction Hamiltonian from (5.q − k+q ± ¯ ωq .109) is a unitary transformation. • It’s possible to show that S is antihermitian. In order to calculate this we will need commutators like [(a† +aq )c† ck . Our formula for S is then. This is necessary so that (5.113) h I have assumed that ωq = ω−q . It must be that ( ( k k − − k k+q k+q − ¯ ωq )A = 1 h + ¯ ωq )B = 1 h (5.−q c† ck ck +q ck +· · · (5. but they are the product of two singleparticle interactions and as such do not interest us.111) is [H .114) Use the notation ∆± (k.116) refers to terms with one pair of boson operators and one pair of electron operators. S]/2. These are second order all right. h (5. S] = k. . S † = −S. I’ll explain what I mean by that below.115) a† aq −q + Dq c† ck k+q ∆− (k. They vanish on the energy shell.q k+q − k )A Dq aq c† ck ( k+q k+q − k − ¯ ωq )B (5. a† c† +q ck ] = δq. Our new Hamiltonian connects twoelectron states with twoelectron states.109) diagonalizes the Hamiltonian. In some sense we have “diagonalized” the Hamiltonian.5. q) Some comments about (5. Now all this is supposed to equal −H from (5.116) −q k+q k+q −q k The · · · in (5. A long tedious calculation gives [H0 .115) are in order.q Dq a† c† ck (¯ ωq + h −q k+q + k. SUPERCONDUCTORS 89 Here A and B are functions that we are about to determine.9. Now you see the sense in which (5. q) ∆+ (k. For ∆+ this corresponds to the absorption of a phonon of momentum q.106). and ∆− to the emission of a phonon of momentum −q.e. • The energy denominators are typical for nonrelativistic perturbation theory. q) = S= k.
Second. Rev. This is the famous “energy gap” upon which all superconducting phenomena depend. N. but inside it is attractive. Outside the window the potential is repulsive (just like a positive Coulomb potential). electron pairs tend to have zero total momentum and zero total spin.117) but gather all the numerical factors into one phenomonological (positive) coupling constant V . a state described in terms of occupation numbers is not an eigenstate of HBCS . for a variety of reasons.90CHAPTER 5.63).2. but there is a narrow window around  k − k −q  ≈ h ¯ ωD in which the potential changes sign.k . are much larger than phonon frequencies. The complete electron Hamiltonian with these assumptions is HBCS = k † k (ck ck + c† c−k ) − V −k k. secondorder. as I pointed out in Section 8. Thus if one electron of a pair has momentum k and sz = 1/2. At this point it is customary to make several approximations based partly on mathematical convenience and partly on an understanding of the physics of superconductors. He showed that two electrons directly below the Fermi 10 L. 104. First it is known that most of the phonons in a lowtemperature solid have energies close to the Debye limit.117) k+q k 2 2 ( k − k −q )2 − (¯ ω)2 h k. Electrons with just this range of energy form Cooper pairs.118) apart from the double sum is formally equivalent to to (5. Phys.118) This is known as the BCS reduced Hamiltonian. Notice that the interaction part of this Hamiltonian cannot be written as a product of number operators. two electron Hamiltonian as 1 1 h ¯ω 2 [H . and one written as −k has spin ↓.115) over a common denominator.q The electron energies k . In 1956 Cooper10 proved a profound result that was the prelude to the full BCS theory. Cooper.k c† c† c−k ck k −k (5. we do not sum over q in (5. IDENTICAL PARTICLES AND MANYPARTICLE STATES It’s instructive to combine the two terms of (5. You will recall that this was the situation with the superﬂuid Hamiltonian. It won’t be a surprise then that the Bogoliubov transformation is the key to diagonalizing it. That gives the ﬁnal form for our eﬀective. ωq . 1189 (1956) . S] = Dq c† c† −q ck ck (5. the other will have −k and sz = −1/2. We therefore adopt the convention that a state written explicitly as k has spin ↑. Put it another way. We have not completely succeeded in our goal of diagonalizing it. It operates only on the subspace consisting of the Cooper pairs that satisfy the above assumptions. Since the spread in energies is so small. In fact. (5.
e. ick = ˙ k ck − c† V −k k c−k ck c† c† k −k k ic† = − k c† − ck V ˙−k −k (5. Philips. i.11 He also showed that a small center of mass momentum has the eﬀect of much reducing the binding energy. so the corresponding approximation is V k c−k ck → V Ω k c−k ck Ω ≡ ∆k c† c† Ω ≡ ∆∗ k k −k k V k c† c† → V Ω k −k (5. We look for solutions of the form ak (t) = ak (0)e−iωk t . This is one reason why we only consider (k. It is also the reason why superconductivity only exists at low temperature.118) exists.120) back into (5. Ω . ωk ck (0) = k ck (0) − ∆k c† (0) −k (5. We’ll start by deriving equations of motion for the ck ’s. Part of our job will be to construct this state explicitly. The oneelectron states in kspace in this shell are used in forming the BCS ground state. For the time being I will simply call it Ω .118).119) puts us back in the realm of familiar mathematics.122) See.9.119) The key to solving the superﬂuid problem was to linearize the equations of motion by replacing the product of operators a† ak with their expectation k value in the ground state. As we “turn on” the interaction (5.120) Substituting (5. electrons in a thin shell. −k) just above the Fermi surface. Advanced Solid State Physics for an accessible proof. the minimumenergy of the Hamiltonian (5. F − ωD < < F + ωD form Cooper pairs. Now suppose we could start with a free electron gas close to T = 0. provided that an attractive interaction of the form (5. so that the Fermi sphere is completely ﬁlled up to a radius F . SUPERCONDUCTORS 91 surface can lower their energy by being excited into a Cooper pair with momentum (k. . −k) pairs. P. This is known as the Cooper instability.5. Here the “particles” we are concerned with are pairs.121) ωk c† (0) = − k c† (0) − ∆∗ ck (0) k −k −k These equations are consistent if ωk − ∆∗ k 11 k ∆k ωk + k =0 (5.118).
125) Both uk and vk are real numbers.129) from which we conclude tan 2θk = ∆/ cos2 θk = 1 2 1+ k k ωk sin2 θk = 1 2 1− k ωk (5. 2 u2 + vk = 1 k (5. The eigenvectors of (5.123) 2 ∆2 (u2 − vk ) = 2 k ∆uk vk k k uk + ∆uk vk . it would also not conserve fermion .118) is invariant under the global gauge transformation ck → ck = e−iα ck for all k.124) (5. It means that all excited states are separated from k by a ﬁnite amount.128) (5.125) into the ﬁrst of (5.66) is βk = uk ck − vk c† −k with inverses † ck = uk βk + vk β−k † c−k = uk β−k − vk βk β−k = uk c−k − vk c† k (5. The analog to (5.123) This is an important result. ck } = vk vk {c† . vk is odd.92CHAPTER 5. the gap parameter ∆.121) are in fact the Bogoliubov transformations from Section 7. (It is clear that if HBCS did not observe this symmetry. IDENTICAL PARTICLES AND MANYPARTICLE STATES Or ωk = 2 k + ∆2 k (5. (5. The BCS Hamiltonian (5. βk } = uk uk {ck . We would like the β’s to satisfy fermion anticommutation relations. 2 {βk .130) So far I have said nothing about the phase of ∆. uk is an even function of k.126) The natural parameterization is uk = cos θk vk = sin θk (5.127) Substituting (5. There is an important issue lurking here.121) gives ωk uk = Square and substitute (5. c−k } = δkk (u2 + vk ) k −k This is the ﬁrst condition on uk and vk .2.
Finally as promised. We take ∆ real.9.124) we get.134) Substituting (5. I understand that it is in reasonable agreement with experiment. The magnitude of ∆ can be estimated as follows.” Here the magnitude of the interaction is set by the quantity V NF .133) This was one of the original discoveries of the BCS collaborators.120) and use the fermion anticommutation relations.131) I have used (5.135) . Substitute (5.5.132) can be inverted to yield ∆= ωD sinh(1/V NF ) (5. SUPERCONDUCTORS 93 number.132) The limits of the integral are chosen in light of the fact that the Hamiltonian only works on the states k where F −ωD < k < F +ωk . 1= V NF 2 ωD −ωD d [ 2 + ∆2 ]1/2 = V NF sinh−1 (ωD /∆) (5. NF is the density of states at the Fermi level.130) in the last step. Remember that ordinary perturbation theory requires that the interaction be in some sense “small. This can be converted into an integral equation. ∆ = V Ω k † † (cos θk β−k + sin θk βk )(cos θk βk + sin θk β−k )Ω = V Ω k † cos θk sin θk β−k β−k Ω =V k ∆ 2 2 k + ∆2 1/2 (5. I claim that apart from normalization Ω = k β−k βk 0 (5. we construct the explicit form of the BCS groundstate wave function. In the limit that this is small. (5. the denominator looks like the hyperbolic sine of inﬁnity! This is not something you will ever get from a perturbation theory.125) into (5. Ω = k (−vk )(uk + vk c† c† ) k −k (5. Eqn.) It is speculated that the ground state observes this symmetry as well so that any choice of the phase of ∆ is equally valid.
94CHAPTER 5. βk Ω = βk β−k βk k β−k βk 0 = 0 The prime on the product indicates that the k = k term has been factored out.137) represents a coherent superposition of correlated pairs. . Eqn.136) Thus the normalized ground state is Ω = k (uk + vk c† c† ) k −k (5. let’s verify that βΩ = 0. The expression is zero because βk βk = 0 for a fermion operator. This can be seen as follows: ΩΩ = 0 k (uk + vk c−k ck )(uk + vk c† c† 0 ) k −k = k 2 (u2 + vk ) 00 = 1 k (5. IDENTICAL PARTICLES AND MANYPARTICLE STATES It turns out that the right normalization is obtained by dropping the factors of (−vk ). with no restraint on the particle number. (5.137) Just to make sure.
φ(x.Chapter 6 Second Quantization and Relativistic Fields In Chapter 5 I introduced the concept of the ﬁeld operators ψ(x) and ψ † (x) (5. There are other ways of regarding quantum ﬁeld operators. or we could introduce a continuous variable.1 Introduction to Field Theory Imagine that space is like a rubber sheet. and nearby objects will roll into it. however.1 6. that are important when we are exploring other wave equations. In this context it seemed like a technical device for introducing interactions in which particles are exchanged so that the number of particles is not a constant of the motion. We could describe the attraction in one of two ways: we could say that there is an attractive potential between any pair of pointlike masses. y) which describes the displacement of the sheet as a function of position. I explained that ψ † (x) is the operator that creates a particle at the position x. The quantum mechanics of such ﬁelds is called quantum ﬁeld theory. it will create a depression. Such a continuous displacement variable is a ﬁeld in the strict mathematical sense: it assigns a numerical value (or set of values) to each point in space. All this was done to “soup up” Schrodinger’s equation.31). We didn’t worry too much about the physical interpretation. particularly those arising in relativistic theories. If I put a bowling ball on the sheet. 95 . Now suppose that instead of using a bowling ball I 1 This material also appears at the beginning of Chapter 4. This is an imperfect analogy for an attractive potential. It is included here to make the next few chapters selfcontained.
which they called the ether. If each point on the sheet behaved like a simple harmonic oscillator with a quadratic potential. the waves propagating on the sheet would never interact. When an electromagnet wave is propagating through space. t) as usual. This analogy can easily be misleading. (If you think of a way to do it. Ordinary bulk media have elastic properties that are described by two parameters. Spacetime points in the ether can be labelled with Lorentz fourvectors or (x. which you understand by now if you have plowed .96CHAPTER 6. which is what we really care about. The sheet will oscillate in response. I am fond of saying. The rubber sheet is not intended as an analogy for ordinary spacetime as it is often used in explaining general relativity. which propagates outward in the form of waves. For one thing. actually the ethers. Unfortunately. The ﬁeld φ(x. SECOND QUANTIZATION AND RELATIVISTIC FIELDS jump up and down on the sheet. Each point on the rubber sheet or ether acts like a harmonic oscillator! Quantum ﬁeld theory is a theory about harmonic oscillators. and I know you want to ask. we will call the ether. tell me and I’ll be famous. Well. This is the rubbersheet analogy to the propagation of particles. According to the textbooks. the anharmonic oscillator cannot be solve exactly in quantum mechanics. Michaelson and Morley proved that wrong with their famous interferometer. The generic name for these approximations is perturbation theory. This is a theory of free particles. but their ratio gives the velocity of propagation.” Now you see that this is profoundly true. etc. My activity becomes a source of energy. everything is a harmonic oscillator. Each species of particle corresponds to a set of vibrations in it’s own speciﬁc ether. which for lack of a better name. I don’t want you to think we are doing general relativity. the density and Young’s modulus. The path integral formalism in Chapters 14 comes equipped with its own natural way of doing perturbation theory. The principle of linear superposition would hold everywhere. If our theory is to describe interactions. This much is required by the MM experiment. But just saying that the ether does’t exist doesn’t answer the question.) We have to resort to approximations. y) describes a displacement. “When correctly viewed. “Displacement of what?” The same question comes up in classical electromagnetic theory. it just makes it impossible to answer! Let’s bite the bullet and agree for the purposes of this course that space is pervaded by a medium. Electrons are all vibrations in the electron ether. Well – I have to modify that slightly. and these points obey the usual rules for Lorentz transformations. then we must modify the potential so that it becomes anharmonic. These parameters are not themselves relevant to our formalism. what is waving? Folks in the 19’th century thought it must be some sort of mechanical medium.
but they are the raising and lowering operators of the harmonic oscillator! The universe. 6. There is a good discussion in the last chapter of Goldstein. but we never make it that far. seeks to banish all operators from the theory.1) V = k(ϕi+1 − ϕi )2 i (6. T = 1 2 1 2 mϕ2 ¨i i (6. The Lagrangian is constructed in the obvious way. Second quantization goes in the other direction. a long chain of masses connected by springs. We will use ϕi . Continuum mechanics is not covered in most graduate mechanics classes.2. It’s a onedimensional problem. sometimes called canonical quantization or second quantization. INTRODUCTION TO CONTINUUM MECHANICS 97 through this material. What follows is a brief introduction.3) The equilibrium separation between masses is a.2 Introduction to Continuum Mechanics The rules of continuum mechanics are derived by starting with a system with a ﬁnite number of degrees of freedom and then passing to the limit in which the number becomes inﬁnite. It is the older way. the displacement of the ith particle from its equilibrium position. Let’s do this with the simplest possible system. It turns the wave functions themselves into operators by imbedding creation and annihilation operators into them. We need to master both.6. This approach is complementary to path integrals in other ways as well. as the generalized coordinate. The masses can only oscillate along the chain. according to second quantization.2) 2 L=T −V = 1 2 a i m 2 ϕ − ka ˙ a i ϕi+1 − ϕi a (6.4) . The spring constant is k. There is an alternative way of dealing with interaction involving the creation and annihilation of particles. The EulerLagrange equations of motion are obtained from d ∂L ∂L − =0 dt ∂ ϕi ∂ϕi ˙ (6. is an inﬁnity of harmonic oscillators. The path integral formalism.
In this sense.5) represent a set of N coupled equations for N degrees of freedom. SECOND QUANTIZATION AND RELATIVISTIC FIELDS If there are N masses.6) The last integral implicitly deﬁnes the Lagrangian density . and ϕi gives its displacement.98CHAPTER 6. (6. In elementary mechanics.8) (6. the index is replaced by the coordinate x.7). ∂2ϕ ∂x2 2 + ∂2ϕ ∂y 2 + ∂2ϕ ∂z 2 = 1 c2 ∂2ϕ ∂t2 See Goldstein for a derivation of this important equation. ϕ) ˙ (6.8) should remind you of the equation for the propagation of electromagnetic waves.7) dt ∂ dϕ dx ∂ dϕ ∂ϕ dt dx Use the Lagrangian density from (6. ∂2ϕ µ = 2 ∂x Y d2 ϕ dt2 (6. continuum mechanics is much easier that discrete mechanics. x would be the displacement of a particle. In the continuum limit ϕi+1 − ϕi ϕ(x + a) − ϕ(x) dϕ → → a a dx L→ 1 2 dx µϕ2 − Y ˙ dϕ dx 2 ≡ dxL(ϕ.5) 2 a a a2 We need diﬀerent parameters to describe the continuum limit: m/a → µ ka → Y mass per unit length Young’s modulus The index i points to the ith mass. The continuum version of the EulerLagrange equation is2 d δL ∂L d ∂L + − =0 (6. Here ϕ(x) is the displacement of the string at the point x.7) is one equation for an inﬁnite number of degrees of freedom. . Equation (6.6) in (6. In the continuum limit.4) and (6. then there are N coupled equation of this sort. They look like m ϕi+1 − ϕi ϕi − ϕi−1 ϕi − ka ¨ + ka =0 (6.
2.e. All these equations can be obtained from (6. or the relationship is quadratic. The question is. of course.11).9) and (6. so there can only be one velocity.) This is worth a brief digression. (6. In both cases the curves are parabolas. h h (6. INTRODUCTION TO CONTINUUM MECHANICS 99 As you know. All the wave equations I know of fall in one of these two categories. vgroup = h ¯k dω ≈ dk m (6. They are all threedimensional variations of our “waves on a rubber sheet” model. in which case dk m = lim ¯ k h .9) The relationship is quadratic. k = ω/c. This is the hallmark of a massless ﬁeld. p2 c2 + m2 c4 = E 2 . i. the group velocity. It contains a great deal of information. what is the relation between k and ω? The relationship between the two is called a dispersion relation. in which case the particle is massless and its velocity v = ω/k. the group velocity is equal to the classical velocity for a massive particle v = p/m.11) or in other words. either ω is proportional to k. plot ω versus k for small values of k. This means that in the limit of small k. Notice that a string of massive particles yields a wave equation that when quantized describes the propagation of massless particles. In the case of EM waves in vacuum.7) with the appropriate Lagrangian density.10) The dispersion relation is (c¯ k)2 + m2 c4 = (¯ ω)2 . (With a diﬀerent velocity. What does it mean to say that a wave function describes the propagation of a particle of a particular mass? The wave function ψ = ei(kx−ωt) might describe a wave in classical E&M. photons are massless particles.12) In other words. In Schrodinger theory h ¯ 2 k2 = ¯ω h 2m (6. The relativistic wave equation for a spinzero particle is called the KleinGordon equation.13) k→0 dω . The velocity is the constant of proportionality. 2 − 1 ∂2 c2 ∂t2 ϕ− m2 c2 ϕ=0 h ¯2 (6. Frequency and wave number are simply proportional.6. What does this have to do with the particle’s mass? It’s useful to plot (6. or a massive particle in nonrelativistic or relativistic quantum mechanics.
To see this. x2 . Rewrite (6.21) . then (6. x3 ) = (ct. We can do relativistic quantum mechanics using the canonical formalism of classical mechanics. Construct the contravariant and covariant fourvectors xµ ≡ (x0 .14) (6. x2 .6) in 3d L= 1 2 µ ∂ϕ ∂t 2 −Y ∂ϕ ∂x 2 + ∂ϕ ∂y 2 + ∂ϕ ∂z 2 (6.20) Y c Obviously L can be multiplied by any constant without changing the equations of motion. x1 . x3 ) = (ct.100CHAPTER 6. z) xµ = (x0 . Here’s an example.15) This puts the EulerLagrange equation in tidy form ∂µ − ∂L =0 ∂ϕ (6. and yet (6. this is what mass means.7) was derived without reference to Lorentz transformations. We get relativity for free! If we can manage to make L a Lorentz scalar.17)) will have the same form in all Lorentz frames. Better yet. let’s rewrite (6.17) has the correct form for a scalar wave equation. Rescale it so that it becomes L= 1 2 ∂ϕ ∂t 2 − c2 ∂ϕ ∂x 2 + ∂ϕ ∂y 2 + ∂ϕ ∂z 2 (6.17) This is slightly amazing. ∂xµ (6. −x. (6.18) is also a Lorentz scalar. x.19) This would be the Lagrangian density for oscillations in a huge block of rubber. y.7) in relativistic notation. the action S= dt L = dt d3 x L = 1 c d4 x L (6. One of the advantages of using Lagrangians rather than Hamiltonians is that Lagrangians have simple transformation properties under Lorentz transformations. Equation (6. −y. Take µ 1 = 2. SECOND QUANTIZATION AND RELATIVISTIC FIELDS So long as we are talking about waveparticle duality. x1 .16) (6. −z) and the corresponding contravariant and covariant derivatives ∂µ ≡ ∂ ∂xµ ∂L ∂(∂ µ ϕ) ∂µ ≡ ∂ .
In the previous chapter I introduced the ﬁeld operator 1 ˆ ψ(x) = √ eik·x ak ˆ (6.3 Introduction to Second Quantization So far this has all been “classical ﬁeld theory.17) yields the usual equation for EM waves. we can collapse (6. Spin can be included by adding more components to ϕ. the equation of ¯ motion is (∂µ ∂ µ + m2 )ϕ = 0 (6. All you have to do is add scalar terms to (6.” ie. More about this later. the MichelsonMorley experiment proves the existence of the ether. and not operators.22) The last piece of notation (∂ϕ)2 . When correctly viewed. is used to save ink.6. even the vacuum! Using Einstein’s neat notation. INTRODUCTION TO SECOND QUANTIZATION 101 Substituting (6. We don’t have to worry about the mechanical properties of the ether.21). This is an important point. Name the theory after yourself.21) into (6. Try it.) is a onecomponent Lorentz scalar. L= 1 (∂ϕ)2 − m2 φ2 2 (6.19) turns into the Lagrangian for oscillations in the ether (6. 2ϕ = 0. we can deduce the symmetry properties of a theory by glancing at L. We have not yet introduced second quantization.21) into one term 1 1 L = (∂µ ϕ)(∂ µ ϕ) ≡ (∂ϕ)2 2 2 (6. the ﬁelds are ordinary functions.24) If we assume that ϕ(x) (x is a 4vector in this notation.25) V k I said that a† and ak were creation and annihilation operators and that all ˆk ˆ this was necessary to treat systems in which the number of particles was . because µ and Y are scaled away. cnumbers as they are called.22). Despite what you may have been told. The fact that we can write L like this is proof that it is a Lorentz scalar. everything is a bunch of harmonic oscillators.) Using our new notation.3. 6. then this describes a spinless particle with mass m propagating without interactions. Here’s a theory that already has a name. It’s the KleinGordon theory. Now you can make up your own ﬁeld theories. Notice how the Lagrangian for oscillations a block of rubber (6.23) (I have set c = 1 and h = 1.
Einstein’s equation E 2 = k2 + m2 suggests that the operators on the right side of (6.26) ¯ is the kinetic energy. the Schrodinger equation.102CHAPTER 6.26) and (6.30) (6.27) (6. SECOND QUANTIZATION AND RELATIVISTIC FIELDS not constant. t) = k2 ψ(x. The operator on the right side of (6.28) are the total energy squared.31) ∂ ∂t ∂ ∂t ψω (x. In this section I would like to examine the motivations behind (6.25). which is just the KG equation with m = 0. The simplest solutions of (6. We have encountered so far three quantum mechanical wave equations. t) ∂t 2m 2 (6. t) 2 ∂ ∂t ˆ ψ(x.29) (6.27) and (6. i i ∂ ∂t i ˆ ∂ k2 ψ(x.26) (6. t) ∂t 2 (6. t) = (k2 + m2 )ψ(x. Suppose that the ψ’s are eigenfunctions of these operators with eigenvalue ω(k). t) = √ ei(±k·x−ωt) (6. and the equation for massless scalar particles. (Each of these three equations will deﬁne a diﬀerent functional relation between k and ω. We will eventually derive a completely relativistic generalization of (6. t) = ω 2 (k)ψω (x. t) 2 Although we don’t usually use this language. which will be our starting point for doing relativistic ﬁeld theory. of course. t) ˆ I have used k = −i x and h = c = 1. we could think of i∂/∂t as a kind of energy operator whose eigengvalues are the total energy of the particle. Suppose now that the ψω ’s are also momentum eigenstates so that ˆ kψω = kψω . t) = ω(k)ψω (x. the KleinGordon equation.29) with ω = k2 /2m are 1 ψk (x. I will write the freeparticle versions of them in a peculiar way to emphasize the complementary roles of time and energy.28) ˆ ψ(x. t) = ω 2 (k)ψω (x.) The equations become i i i ∂ ψω (x. t) ψω (x. t) = ψ(x.25) more carefully and also investigate the many subtleties that arise when we apply these ideas to relativistic wave equations.32) V .
45).34) ∂t Since h = 1.3. the solution has positive frequency (in the sense above).29). the ak ’s should satisfy 2 ak + ωk ak = 0 ¨ (6.36) This is the diﬀerential equation for the harmonic oscillator. Now take an arbitrary timeindependent wave function and expand it in a Fourier series. I will assume that the symbol ω refers to a positive quantity. We can make ψ time dependent by building the time dependence into the ak ’s. As time goes by.34) gives positiveenergy solutions and the lower sign gives negativeenergy solutions whatever that may mean! What about the other sign ambiguity? Think about the solution ei(k·x−ωt) . the ak ’s are complex functions. We call this a positivefrequency solution. ak → ak (t).31) be satisﬁed. and second. If the signs are the same.30) and (6. This is the clue that the upper sign in (6.6. First. t) = √ ei(±k·x V ωt) (6.31) comprise a larger family than those of (6. To avoid ambiguity. Schrodinger’s equation does not ¯ have negative energy solutions. Assume periodic boundary conditions so that k is discretized as in (5.30) and (6.33) √ (The 1/ V is a normalization factor put in for later convenience. If ω and k · x have opposite signs.33) and (6. In some .35) ψ(x) = √ V k At this point ak is a Fourier coeﬃcient and nothing more. and it is this larger family that I want to investigate. or k · x = ωt + φ. Pick out some point on the wave where the phase of the exponential is φ. In order that (6. Then ∂ i e iωt = ±ωe iωt (6.28) and (6. This point is moving in the general direction of k.30) with ω 2 = k2 + m2 or (6. except for two peculiar features. this point will move so that k · x − ωt = φ.27) and (6. INTRODUCTION TO SECOND QUANTIZATION 103 whereas the simplest solutions of (6. 1 eik·x ak (6. one gets the negativefrequency solution. ω has the units of energy. the frequency (and hence the “spring constant”) is a function of k.31) with ω 2 = k2 are 1 ψk (x.) Evidentally the solutions to (6.
and that Hamilton’s equations of motion are satisﬁed. are timeindependent.40) ak and a† are timeindependent.15).37) ˙ The conjugate momentum is given by p(t) = q(t). a† ] = δk. qk (t)] = [ˆk (t). We know from our work in Chapter 5 that ak must be an annihilation ˆ operator with the commutation relations (5. SECOND QUANTIZATION AND RELATIVISTIC FIELDS sense.38) that Hk is timeindependent.40) In (6.k a ˆk [ˆ† . however. but before we take the derivative. [ˆk . The operators in (5.k q ˆ [ˆk (t). Schrodingerpicture operators as is the ﬁeld operator (6. The natural assumption is ak (t) → ak e−iωt ˆ a∗ → a† e+iωt ˆk k (6. the commutation ˆ ˆ relations must be taken at equal times. The commutation relations are then.41) Since pk and qk don’t have this simple time dependence. ˆ ˆk I’ll argue presently that these is a consistent and reasonable assumption. that pk is canonically conjugate to qk . pk (t)] = iδk.35) has a harmonic oscillator associated with it. ak (t) = iωak .39) into a quantummechanical Hamiltonian by simply making ak . We can turn (6. 2 k (6. In order that each term in (6. a∗ . We will want to work in the Heisenberg representation.104CHAPTER 6.25). so we must be careful about the time dependence. k 2ωk (6. a† ] = [ˆk .39) You can verify with the deﬁnitions (6. We know that the Hamiltonian for simple harmonic motion is Hk = 1 2 2 2 p + ωk qk . Schrodingerpicture operators.35) has the form (6. we must decide on whether we are dealing with the positive. ak ] = 0 ak ˆk a ˆ (6. We can tie into the usual harmonic oscillator formalism and avoid the complex coordinates at the same time by deﬁning the real generalized coordinate.37) and (6. qk and pk into operak tors.38) These are real variables oscillating with frequency ω.or negativeenergy solution. For the time being take positive energy (upper sign) ˙ pk (t) = −i ωk [ak (t) − a∗ (t)] k 2 (6.42) . pk (t)] = 0 q ˆ p ˆ (6. 1 qk (t) = √ [ak (t) + a∗ (t)] . [ˆk (t). each term in (6.33).15).
43) by commuting the ak ˆ past the a† in the second term and discarding the commutator. We can check that this Hamiltonian is consistent with the time dependence assumed in (6.” We will say that an operator has been normal ordered if all creation operators are placed to the left of all annihilation operators. The justiﬁcation lies in the canonical quantization procedure from . so a ˆ ˆ ˆa Hˆk = ak (H − ωk ) ˆ ˆ hence as a consequence ak (t) = eiHt ak e−iHt = ak e−iωk t ˆ ˆ ˆ a† (t) = eiHt a† e−iHt = a† eiωk t ˆk ˆk ˆk ˆ ˆ ˆ ˆ (6. INTRODUCTION TO SECOND QUANTIZATION With this substitution (6. The complete Hamiltonian H is a sum over all values.48) The deeper question is why this quantization procedure makes any sense at all. Since this problem will appear again and again in diﬀerent contexts. ˆ ˆ Hk (6.3.47) ˆ ˆ H n ak = ak (H − ωk )n ˆ ˆ (6.44) H= k An inﬁnite number of 1/2’s is still inﬁnity.46) (6.39) becomes 1 1 ˆ ˆk ˆ ˆ ˆk ˆk ˆ Hk = ωk a† ak + ak a† = ωk a† ak + 2 2 105 (6. The last factor of 1/2 in (6. To this end we introduce the concept of “normal ordering. Whenever ˆk we use a Hamiltonian in a practical calculation.25).45) was obtained from (6. It is customary to discard the constant with some weak argument to the eﬀect that in deﬁning energy.43) The same replacement turns (6. (6.35) into (6. This ˆ k is just the Hamiltonian for a single k value.6.40) First note that [ˆk . so for example.45) To put it another way.43) presents something of a dilemma. H] = ωk ak . we will assume that it has been normal ordered. additive constants are meaningless. ˆ : Hk := ωk a† ak ˆk ˆ (6. it is useful to have some formal procedure for sweeping it under the rug. The usual symbol to indicate that an operator has been normal ordered is to place it between colons.
G} is independent of the choice of canonical variables. (6. This seems and G like a leap of faith. That is a fundamental theorem.37) and (6. G] = F G − GF .51) ˆ ˆ where F and G are the quantum mechanical operators corresponding to the ˆ ˆ ˆˆ ˆˆ ˆ classical quantities F and G.3 Now inverting (6. pm } = 0 (6. a† ] = 0 a ˆ ak ˆk (6. qm } = {pn . G} → [F . It uses the Poisson bracket formalism of classical mechanics and then replaces the Poisson brackets with commutator brackets to get the corresponding quantum expression A Poisson bracket is deﬁned as N ∂F ∂G ∂F ∂G {F. but it is valid for all the familiar operators in quantum mechanics. SECOND QUANTIZATION AND RELATIVISTIC FIELDS elementary quantum mechanics. G} ≡ − (6.38) and (6. We get the quantum expression from the classical expression by the h replacement ˆ ˆ h {F. G]/i¯ . and F and G are any two arbitrary functions of qk and pk . The sum runs over the complete set of generalized coordinates. a∗ } = 0.53) (with h = 1).53) are identical. Obviously {qn . Poole and Safko. (6.52) Substituting (6. so that k k [ˆk . the commutator is taken at equal times.49) ∂qk ∂pk ∂pk ∂qk k=1 where qk and pk are any pair of conjugate variables. ak } = k {a† ..49) gives {ak .7 3 .39) gives ak = ipk + ωk qk √ 2ωk a∗ = k −ipk + ωk qk √ 2ωk (6. ¯ The value of the Poisson bracket {F. Classical Mechanics Third Ed. a† ] = δk. a∗ } = −iδk. p] = i¯ .52) into (6. ak ] = [ˆ† . [x. Sec. In the case where F ˆ depend on time.50) This looks like the uncertainty relation in Quantum Mechanics.106CHAPTER 6.k a ˆk [ˆk .k and {ak . Since (6.42) and (6. pm } = δmn {qn . Any choice of variables will do so long as they are related to qk and pk Much of the formal structure of quantum mechanics appears as a close copy of the Poisson bracket formulation of classical mechanics. 9. and [F . See Goldstein.38) are together a canonical transformation.
The fact that the resultant Hamiltonian is that of harmonic oscillators is simply a consequence of the fact that we choose to expand ψ(x.57) ω(k) ck − c† . ˆ ˆk ˆ Everything goes through as before except that pk (t) = qk (t) = +iω qk (t) ˆ ˙ ˆ changes sign.k ck ˆ (6. It ˆ points to an important piece of physics. This is not just a mathematical accident. t) in a Fourier series. Suppose we had chosen the negative energy solution. but I want to concentrate on solutions of the KleinGordon equation. INTRODUCTION TO SECOND QUANTIZATION 107 by a canonical transformation. t) = √ V ˆ e−i(k·x−ω(k)t) d† k k (6.25) as 1 ˆ ψ (+) (x. To see this we deﬁne another pair of creation and annihilation operators. ˆ ˆk 2 (6.56) It seems that the new creation operator c† stands in the place of the old k annihilation operator ak . t) = √ V ei(k·x−ω(k)t) ak ˆ k (6. 1 ˆ ψ (−) (x. ck ] = δk. I can now write Equation (6.58) Substituting this in (6.55) (This automatically becomes a negative frequency solution as well. The functional form of ω(k) is determined by the wave equation it represents.6. The rest followed automatically.) The operators ck and c† annihilate and create these new negative energy particles.55) and changing the sign of the summation variable from k to −k gives 1 ˆ ψ (−) (x. We simply chose qk so that it was real and had convenient units.59) . t) = √ V ei(k·x+ω(k)t) ck ˆ k (6.3. so that (6.41) is [ˆ† .38) becomes pk = i ˆ The counterpart of (6.54) The superscript (+) means the positive energy solution. ˆ dk = c† ˆ−k ˆ d† = c−k ˆ k (6.
61) are relics of our nonrelativistic ﬁelds from Chapter 5. Let’s note in passing that there are several neutral spinzero particles such as the π 0 that have no nonzero additive quantum numbers.60) ˆ 1 ˆ ψ(x.54) and ˆ (6. The complete operatorvalued wave function is the sum of (6. Such particles are thereby identical to their antiparticles. For one thing. If the hole moves in the positive k direction. Those corresponding to charged particles are not. Dirac proposed this mechanism to explain the negative energy states that appear in relativistic electron theory.61) Fields corresponding to neutral particles are Hermitian. Removing one particle from the sea leaves a “hole. then ˆk ˆ dk destroys the same particle.59).60) and (6. so there is no Fermi sea. a real particle must move in the −k direction to backﬁll the hole.” which behaves in some ways like a real particle. the states created by dk have momentum k (rather than −k). In this case there is no point in distinguishing ˆ between ak and dk . SECOND QUANTIZATION AND RELATIVISTIC FIELDS ˆ What is it that the d’s are creating and destroying. . the KleinGordon equation describes bosons. Unfortunately. If a† creates a π 0 . which is a positivefrequency solution. Because they are based on discrete k values and periodic boundary conditions they behave under Lorentz transformations in a most 4 The idea is only relevant to lowtemperature conductors anyway. We write (6.58).108CHAPTER 6.and negativeenergy solutions must be added in just this way. we regard these negativeenergy solutions as representing real positiveenergy antiparticles. we ˆ will show that d† creates a particle of the opposite charge to that created k † by ak . This can be proved formally. Later on when we discuss the interaction of the electromagnetic ﬁeld with bosons.60) ψ(x. 1 ˆ ˆ ei(k·x−ωt) ak + e−i(k·x−ωt) d† ˆ (6. but it is almost obvious from (6. There ˆ are two lines of evidence for this. and what is the signiﬁcance of the change in the sign of the momentum? In Chapter 5 we considered the Fermi sea to be a set of lowlying energy levels all occupied by fermions. t) = √ k V k There are several profound reasons why the positive. t) = √ V ei(k·x−ωt) ak + e−i(k·x−ωt) a† ˆk k (6. These will emerge as we go along. In some ways (6.4 Nowadays.59). The correspondence between holes moving forward and real particles moving backward is a good way of visualizing the signiﬁcance of (6.
The point is that we must also consider the transformation properties of the creation and annihilation operators. dk 0 δ(k 2 − m2 ) = = Equation (6.63) Although this convention is sometimes used. INTRODUCTION TO SECOND QUANTIZATION 109 awkward fashion. t) = ˆ d3 k (2π)3 2ωk ˆ a(k)e−ikx + d† (k)eikx ˆ (6.65) does have the right transformation properties.65) ˆ and and similarly for the d’s. . We are accustomed to passing to the continuum limit through the replacement 1 V → k d3 k .3. Although δ (3) (k − k ) by itself is not a Lorentz scalar. The trouble is that the integration measure d3 k does not transform like a scalar under Lorentz transformations.) The dk 0 integration is done as follows. a† (k )] = δ (3) (k − k ) a ˆ [ˆ(k). The natural generalization of (6.64) together with (6. (2π)3 but this may be a bad idea for relativistic ﬁelds. the ﬁeld deﬁnition (6. a(k )] = [ˆ† (k).62) becomes 1 V → k d(k 0 )2 dk 0 d(k 0 )2 2 δ((k 0 )2 − ωk ) d(k 0 )2 1 2 δ((k 0 )2 − ωk ) = 0 2k 2ωk d3 k (2π)3 2ωk (6. (The symbol k here refers to the usual fourvector k µ → (k 0 .62) which is clearly invariant. a† (k )] = 0 a ˆ a ˆ (6. k).6. A better choice might be 1 V → k d4 k δ(k 2 − m2 ). This will be apparent once we have calculated the propagator.64) where kx = ωt − k · x.41) is [ˆ(k). it is somewhat simpler to use the following ϕ(x. (2π)3 (6.
69) are evaluated at the same time. J. In that equation the positive.66) Suppose we regard ϕ as a generalized “coordinate” perhaps referring to the displacement of some hypothetical continuous medium.69) to be the fundamental postulate of ﬁeld theory and use it to derive the properties of the creation and annihilation operators. Classically. SECOND QUANTIZATION AND RELATIVISTIC FIELDS 6.69) (6.64) satisﬁes (6. As such they are called equal time commutation relations.110CHAPTER 6. even in principle. then a signal can propagate from one to the other only by travelling at inﬁnite velocity. The conjugate ﬁeld is δL π(x) = = ∂0 ϕ(x) (6. t).2. t)] = iδ (3) (x − x ) ˆ ˆ (6. t ). Before we leave the subject of commutation relations. we have no choice in the matter. Consider two spacelike separated points (x. The argument goes like this. The Lagrangian density (6. Nothing I have said so far requires the presence of the negativeenergy term. We believe this to be impossible.65). Classical dynamics: a contemporary approach. We I am using ϕ in this section rather than ψ to refer speciﬁcally to solutions of the KleinGordon equation.4 Field Theory and the KleinGordon Equation The KleinGordon equation and its associated Lagrangian were discussed brieﬂy in Section 6.68) It is easy to verify that (6. In fact. Saletan. There will always be a Lorentz frame such that t = t . Jose and E.67) δϕ ˙ We can use the Poisson bracket approach to quantize these ﬁelds just as we quantized the ak ’s in the previous section.6 {ϕ(x. so our theory must not allow it. t)} = δ (3) (x − x ) Our quantized version of this is [ϕ(x.54) is5 L= 1 (∂ϕ)2 − m2 ϕ2 2 (6.60). let alone that it have the same normalization as the positiveenergy part. there is an issue that has been hanging ﬁre since (6. π(x .V. t). π (x .69) so long as the creation operators are normalized according to (6. If x and x are distinct points. Sec 9. 6 See J. 5 .and negativeenergy solutions appear with equal magnitude. Many texts in fact consider (6.3. t) and (x .1 for a derivation of this rather nontrivial result. It’s crucial that the two ﬁelds in (6.
] is the “timeordered product” deﬁned by T [ϕ(x)ϕ(y)] = θ(x0 − y 0 )ϕ(x)ϕ(y) + θ(y 0 − x0 )ϕ(y)ϕ(x) ˆ ˆ ˆ ˆ ˆ ˆ (6. It can be shown that the requirement proved above only holds for integer spin. In the case of halfodd integer spin. ϕ† (x . t) = ˆ d3 k (2π)3 2ωk a(k)e−ikx + αˆ† (k)eikx ˆ a (6.5. ϕ(x. .5 The Propagator At this point I must anticipate some developments from the next few chapters. An Interpretive Introduction to Quantum Field Theory. i. (∂µ ∂ µ − m2 )D(x − y) = δ (4) (x − y) (6. y) = iD(x − y) = 0T [φ(x). From a mathematical point of view.73) (6. it is the Green’s function of the KleinGordon equation. Timeordered products read like Hebrew. for a careful discussion of this point. There are several ways of interpreting D(x − y). It can be argued that a necessary requirement is that the ﬁelds ϕ(x) and ϕ† (x ) also commute at equal times. Relativity and causality together require an equal admixture of negativeand positiveenergy states.64) with an additional parameter α that we can tweak at our pleasure.6.e. the result only holds if the corresponding creation and annihilation operators anticommute. φ(y)]0 Where T [ . right to left. Chapter 4.74) 7 See Paul Teller. THE PROPAGATOR 111 call this the requirement of causality.72) In a timeordered product the timedependent operators are ordered so that later times stand to the left of earlier times. 6. t).7 ˆ ˆ Let’s rewrite (6. t)] = ˆ ˆ d3 k (1 − α2 )eik·(x−x ) 2(2π)3 ωk (6.71) This is not zero (because of the ω in the denominator) unless α = 1. It will turn out that one of the key ingredients of any perturbation calculation is the Feynman propagator deﬁned by ˆ ˆ G(x.70) A simple calculation now gives [ϕ(x. This argument takes on additional signiﬁcance when spin is taken into account.
77) Notice how the denominator is factored. and for t > 0 the contour is completed in the lower halfplane enclosing the point k0 = ω − i . 2iωk → i and 2 → 0.76) are really the same result.112CHAPTER 6. t)ϕ(0. I need to prove the central result that iD(x − y) = i d4 k e−ik(x−y) (2π)4 k 2 − m2 + i (6. it is the probability amplitude for a particle to propagate from y to x. iD(x) = i d3 k ik·x e (2π)4 dk0 e−ik0 t (k0 − ωk + i )(k0 + ωk − i ) (6. we could derive either form from the other. SECOND QUANTIZATION AND RELATIVISTIC FIELDS From a physical point of view. though this is far from obvious. You see how the i in the denominator displaces the poles so as to pick up the right integrand depending on whether t is positive or negative.75) and (6. but it’s probably easier to start from (6.75) Each kind of particle has its own propagator.75). You will see how D(x−y) becomes a key player in perturbation theory via the interaction picture in the next chapter. Multiplying the two factors and making the replacements. so we can make the algebra a bit simpler by setting y = 0. In principle. The stands for an inﬁnitesimal positive quantity. The dk0 integration is now performed as a contour integration in the complex k0 plane.75).72) and taking the vacuum expectation value gives iD(x) = 0T [ϕ(x. Notice ﬁnally that (6. k 2 and d4 k are all scalar quantities. It’s job is to get the boundary conditions right as you will see in the derivation. gives the same denominator as (6. We know that this will be a function of x−y. Just remember to replace x → x − y at the end of the calculation. Substituting the ﬁelds from (6.73) is a Lorentz scalar since kx. 0)]0 ˆ ˆ = d3 k θ(t)e−i(ωk t−k·x) + θ(−t)ei(ωk t−k·x) (2π)3 2ωk (6.70) into (6. The result is identical to (6. . but the +i term in the denominator is ubiquitous. For t < 0 the contour is completed in the upper halfpane enclosing the point k0 = −ωk + i .76).76) Equations (6.
THE PROPAGATOR 113 Im k0 k0=ω0+iε t<0 t>0 k0=ω0iε Re k0 Figure 6.6.1: The complex k0 plane .5.
SECOND QUANTIZATION AND RELATIVISTIC FIELDS .114CHAPTER 6.
and it only works to ﬁrst order. there are two basic issues: how 115 . In a scattering experiment. There are many subtle issues involved. one starts with a stable system and bombards it with another particle. They are detected on a scale of distance that is many orders of magnitude larger than the characteristic de Broglie wavelength and at times that are vastly longer than the time of interaction. One measures the distribution of momenta among the various particles produced by the reaction and determines the probability that the scattering will lead to a particular ﬁnal state. the relative frequency of the various decay modes. In nonrelativistic quantum mechanics these processes are calculated with ﬁrstorder. one starts with some system such as an atom or nucleus or “elementary” particle and observes the spontaneous transitions that it undergoes. and the distribution of momentum and energy among the resulting particles. it’s hopelessly nonrelativistic. Roughly speaking.” This is not altogether wrong. One common feature of both these experiments is that the particles are detected when they are not interacting with one another. the identity of the decay products. Much of the material in advanced quantum books relates to scattering in one way or another. I say this because it’s easy to lose sight of the goal amidst all the technical diﬃculties. time dependent perturbation theory using a bit of hocus pocus called “Fermi’s golden rule.Chapter 7 The Interaction Picture and the SMatrix Most of what we know about subatomic physics comes from two kinds of experiments: decay experiments and scattering experiments. One can determine the lifetime of the system. In a decay experiment. but it is inadequate for several reasons: it can’t accommodate the creation of new particles. Real scattering theory is diﬃcult.
116CHAPTER 7. You will recall the relation between the Schrodinger and Heisenberg pictures is ψ(t) S = e−iHt ψ H (7.1 The Interaction Picture Path integrals use the Lagrangian. H= d3 xH (7. the interaction picture uses the Hamiltonian. We assume the Hint is “small” in some sense. The reason for this technical assumption will appear later. THE INTERACTION PICTURE AND THE SMATRIX do the quantum ﬁelds ϕ(x) evolve in time. so that the perturbation series converges. how can we calculate the results of these experiments in terms of the momenta of particles measured in the asymptotic regime as explained above? The ﬁrst question is answered neatly by the interaction picture formalism ﬁrst developed by Freeman Dyson.6) QH (t) = eiHt QS e−iHt . In classical mechanics. 7. the two are related by H = pq − L ˙ (7.5) We assume that the Hamiltonian can be split up into two pieces: H = H0 + Hint . and given this information. so I should say a few words about Hamiltonians in general. and Hint is everything left over.4) over all space.3) (7.4) (7.1) In our variety of continuum mechanics the Hamiltonian density becomes H = πϕ − L ˙ For the free KleinGordon ﬁeld L= 1 2 1 (∂µ ϕ)(∂ µ ϕ) − m2 ϕ2 = ϕ − ( ϕ)2 − m2 ϕ2 ˙ 2 2 H= (7.2) 1 2 ϕ + ( ϕ)2 + m2 ϕ2 ˙ 2 We get the Hamiltonian by integrating (7. We also have to assume that Hint is a polynomial of the ﬁelds and their derivatives. The second question is much more diﬃcult and will require the remainder of this chapter. H0 is the Hamiltonian of the free ﬁeld.
[Hint . • If there were no interactions. since in general.1. We deﬁne the interaction state by ψ(t) I ≡ eiH0 t ψ(t) S (7. This is not necessarily the most convenient time to take the picture as we will see. after the particles are separated it settles down to a new Heisenberg state. t0 )U (t. by ψ(t) I = U (t. is a “snapshot” of ψ(t) S taken at t = 0. Now think of a scattering experiment. This is not true of the interaction piece. but one that can be dealt with rigorously.9) Since H is Hermitian and the norm of ψ I is conserved. Actually. Eventually.7) that H0 is the same in the interaction picture as well. I will write I S Hint (t) ≡ eiH0 t Hint e−iH0 t (7. Deﬁne the time evolution operator in the interaction picture. even the vacuum interacts with itself by creating particleantiparticle pairs. they are still interacting with themselves by emitting and reabsorbing virtual particles. THE INTERACTION PICTURE 117 This is the usual convention at any rate.10) . As the particles begin to interact. so the interaction picture state would be stationary. When the particles are far apart. Notice that Hint (t) depends on time. H0 ] = 0. this business of “turning on” and “ turning oﬀ” skirts a profound and diﬃcult problem. and equally clear from (7. we feel entitled by the rules of quantum mechanics to the following basic relations: U † (t. t0 ). U (t.8) S Naturally. In this regime. The trouble is that Hint never really turns oﬀ. Hint is the interaction Hamiltonian in the Schrodinger picI ture.6) the Hamiltonian H is the same in the Schrodinger and Heisenberg pictures. ψ H . Even when the particles are not interacting with one another. This is a diﬃcult problem.7) QI (t) = eiH0 t QS e−iH0 t Some comments: • It’s clear from (7. Worse yet. (7. there is no interaction between them. the interaction picture and Heisenberg pictures are identical. Hint “turns on” and ψ(t) I begins to change. t0 ) = 1 (7. t0 )ψ(t0 ) I . which does not depend on time. ψ(t) I = ψ H .7.
t0 ) = 1 − i +(−i) 2 t t0 t1 t0 I dt1 Hint (t1 ) (7. t0 ) = (−i)n n=0 t t0 t1 tn−1 dt1 t0 dt2 · · · t0 I I I dtn Hint (t1 )Hint (t2 ) · · · Hint (tn ) (7. since U appears on both sides of (7. t0 ) = U −1 (t0 . THE INTERACTION PICTURE AND THE SMATRIX U (t. Hint (t )] = 0. t0 ) (7. [Hint (t).18) The series (7. t0 ).16) paying careful attention to the time dependence.12) (7. Since the Hamiltonians don’t commute. (7. t) U (t2 .16) Well – maybe solve is too strong a word. t) = 1 U (t.11) (7. We would like to claim that t U (t.18) is more diﬃcult than it looks. t0 ) = (?) exp −i t0 I dt Hint (t ) This would be the obvious solution if U and H were not operators. ∂t (7. we must be meticulously careful to keep later times to .7) and (7. We can come up with a valid solution by iterating (7.14) ∂t A quick calculation with (7. t1 )U (t1 . t0 ) = Hint (t)U (t.16).15) We know how to solve equations like this.9) yields i ∂ I U (t. t U (t. The ﬂaw in the reasoning here is that factors of Hint don’t commute at diﬀerent times. t0 ) = 1 − i t0 I dt Hint (t )U (t . t0 ) (7.13) We can derive an equation of motion for U by starting with Schrodinger’s equation ∂ i ψ(t) S = Hψ(t) S .17) + ··· dt1 t0 I I dt2 Hint (t1 )Hint (t2 ) The entire series can be summarized by ∞ U (t.118CHAPTER 7. t U (t. t0 ) = U (t2 .
2. As the particles approach one another.2 The S Matrix We can imagine quantum scattering as taking place in three phases.20) t which can be written in shorthand notation as U (t.21) t0 The reason that this is a workable approximation is that the ﬁelds that make I up Hint are interactionpicture ﬁelds. the interaction between them “turns on” as they exchange virtual particles. (7. THE S MATRIX 119 the left of earlier times. transform as free ﬁelds. and in general t t0 tn−1 dt1 · · · t0 dtn H(t1 ) · · · H(tn ) = 1 n! t t0 t dt1 · · · t0 dtn T [H(t1 ) · · · H(tn )] So our ﬁnal result for the U operator is ∞ U (t. We need some machinery to do this for us. This is diﬃcult to treat theoretically. since the interaction Hamiltonian itself in no way . At some later time t → +∞. T [H(t1 )H(t2 )] ≡ H(t1 )H(t2 )θ(t1 − t2 ) + H(t2 )H(t1 )θ(t2 − t1 ) (7.19) The generalization to three or more Hamiltonians is obvious. This is called time ordering.” You just insert the formulas for the free ﬁelds and do the algebra. You should convince yourself that the following things are true: t t0 t1 t t2 dt1 t0 dt2 H(t1 )H(t2 ) = = 1 2 t t0 t0 dt2 t t0 dt1 H(t2 )H(t1 ) t0 dt1 dt2 T [H(t1 )H(t2 )]. this interaction turns oﬀ and they again become free particles. t0 ) = T exp −i I dt Hint (t ) (7. Deﬁne the time ordering operator. Equation (7.20) is a “recipe. In the limit t → −∞ the particles are separated and not interacting with one another.8).7. 7. which according to (7. t0 ) = n=0 (−i)n n! t t ··· t0 t0 I I I dt1 dt2 · · · dtn T [Hint (t1 )Hint (t2 ) · · · Hint (tn )].
These two views can exist simultaneously because the selfinteractions are virtual processes. According to perturbation theory. they inevitably interact with themselves. Somehow the eﬀect of all virtual processes formulated in terms of the bare particles is to produce the physical particles. There are the physical or “dressed” masses and coupling constants – those that are measured in the lab – and the “bare” masses and coupling constants. but this is a long story. What then are the numerical values of these bare masses and charges? Alas. charge.1 but apart from this. every electron in the universe is an unremarkable little thing with a welldeﬁned mass. and as such they are contained within a region roughly the size of the Compton wavelength. these vacuum ﬂuctuations have very little consequence. So long as one does not probe inside this region. 2 One can calculate the gravitational mass of all the virtual particleantiparticle pairs in the universe. For the time being. and vacuum become profoundly subtle. the coupling constant for any sort of interaction). this is only the tip of the iceberg. From the point of view of theory. This fact is established experimentally (to some extent) by an experiment measuring the socalled Casimir eﬀect. are not well understood. and magnetic moment.120CHAPTER 7. We believe that all these problems can be addressed. we are able to calculate everything that can be Two charged plates suspended parallel to one another in vacuum experience a very small repulsive force because of virtual electronpositron pairs created between them. and the experiment must be interpreted in light of the virtual processes involved. The same sort of ambiguities also complicate our understanding of the vacuum state. and at any rate. spin. They could be calculated in principle. The charge is the measure of the strength with which photons couple to electrons. but this can only be measured by doing a scattering experiment. which are the parameters that go into the interaction Lagrangians and Hamiltonians. virtual particleantiparticle pairs are spontaneously emitted from and reabsorbed into empty space. and probing inside the region means doing a highenergy scattering experiment. 1 . The result come out too large by many orders of magnitude. As soon as we allow particles to interact with other particles. THE INTERACTION PICTURE AND THE SMATRIX turns on or oﬀ. at least for quantum electrodynamics. From the point of view of experiment. we do not know. When this happens. the electron has it’s usual physical properties. but every such calculation yields inﬁnity! Despite this. the notions of free particles. So far as theoretical diﬃculties are concerned. charge. There is no way to measure them directly. let us say that the theory recognizes two kinds of mass and two kinds of charge. a free particle is a boiling stream of selfinteractions. which will occupy us for the next few chapters.2 Finally consider the charge of an electron (or more generally.
22) It is understood that in calculating with this expression. one deﬁnes “out states” β. outα.25) . . The symbol α stands for all the quantum numbers required to completely specify the state. In other words. We need to back up and look at the basic description of scattering processes. We call this state α. For mathematical purposes we say this corresponds to a time “t → −∞” (The actual time for a relativistic particle could be as small as t → −10−23 seconds. We will be particularly concerned with the momentum. A scattering experiment can be idealized as follows. out.2. THE S MATRIX 121 measured (except mass and charge. the bare or the physical mass? It seems that either way we are stuck. is one of the ironic triumphs of modern physics. Consider the state of a scattering experiment before the particles have interacted. The initial target and beam particles are prepared so that they are in a unique eigenstate of all the relevant operators. in make the transition to β. in . . in . . The fact that we have learned to skirt inﬁnities so violent as to give sciﬁ fans nightmares. In view of this. the interaction picture formulation in the previous section is a bit too naive. out.23) If we assume that the in states and out states are both complete sets. It is customary to call such states “in states. out. one uses the physical mass of the particle. k2 . there must be some operator S that transforms in states into out states. The detectors of the scattered particles are so reﬁned that they can specify that the ﬁnal state was in the exact eigenstate β. If there are n particles in the initial state. .” and write them α. kn . In the same way. The ﬁeld operators introduced in the previous chapter can be written ϕin (x) = d3 x 2Ek (2π)3 ain (k)e−ikx + a† (k)eikx in (7. inS (7. of course) using only the physical mass and charge.7. in .) At this time the particles are free in the sense explained above. in (7. What masses are we to put in Hint . “out ﬁelds” ϕout (x). and “out” creation operators a† (k) corresponding to out the limit t → ∞. out = β.24) (7. we can write k1 . we are measuring the probability Pβα = Sβα 2 where the socalled S matrix is deﬁned by Sβα = β. β. The object of the experiment is to ﬁnd what fraction of particles in α.
S is unitary. 2. inp. in = α.3 The LSZ Reduction Scheme What is the relation between the fullyinteracting ﬁeld ϕ and ϕin ? The natural guess is lim ϕ(x) = ϕin (x) t→−∞ . Here are some commonsense properties of S. inS = 0.26) The goal of scattering theory is to calculate the S matrix and the corresponding S operator. p.29) (7. S00 2 = 1. out β. in = 1 3. in (7. in (7.e. Put it another way 0.122CHAPTER 7. inSϕout α. inϕout α. in But β. Proof: β. THE INTERACTION PICTURE AND THE SMATRIX Sβα = β. inSS † α. inϕout is itself an in state. inSα. outϕin Sα. ϕin = Sϕout S −1 Proof: β. in Comparing these two equations gives Sϕout = ϕin S 4. i. in = β.28) 7. in = p. so we can write β. out S † α. inS = β. Oneparticle states are also stable. outp. inSp. outϕout α. outβ. Nothing in – nothing out. 1. out = Sβα (7. in = p. in = β. out = eiθ 0.27) The vacuum is stable. but we must allow the possibility that phase of the vacuum state evolves with time. in = α.
64) ϕin (x) = where fk (x) = 3 † ∗ d3 k fk (x)ain (k) + fk (x)ain (k) (7.24) into an expression involving vacuum expectation values of interacting ﬁelds. This is the LSZ reduction technique. Even this is not quite correct as an operator equation for some subtle mathematical reasons that are discussed in the original article. Since we are only interested in matrix elements in the long run. Brian Hatﬁeld. denoted by k1 · · · kn in = α in . Now we have the apparatus at hand for deﬁning and studying the S matrix deﬁned in (7.3 It is correct in terms of matrix elements. and a ﬁnal state in which m particles emerge denoted by p1 · · · pm out = β out (7.30) Z is understood as the probability for ϕ(x) to produce a singleparticle state out of the vacuum. We start from an initial state of a system with n noninteracting physical particles. for example √ lim αϕ(x)β = Z αϕin (x)β t→−∞ where α and β are arbitrary states. Furthermore. the processes in which ϕ(x) produces multiparticle states out of the vacuum only occur in higher orders of perturbation theory. Let us write the inﬁeld in a way analogous to (6.7.33) e−ikx (2π)3 2ωk (7. When we are working to lowest nontrivial order we can (and will) set Z = 1.34) See also Quantum Theory of Point Particles and Strings. we will assume (7.24). A more careful analysis suggests √ lim ϕ(x) = Z ϕin t→−∞ t→∞ lim ϕ(x) = √ Z ϕout (7.30) without any further comment. The complete ﬁeld ϕ(x) is capable of creating states out of the vacuum with two or more virtual particles.3. . THE LSZ REDUCTION SCHEME lim ϕ(x) = ϕout (x) 123 t→∞ On second thought though.32) (7.31) Our agenda is to convert (7. Chapter 7. this can’t be right.
For any two functions such as fk (x) and ϕ(x).and out. The procedure is in then repeated until all the particles have been removed from the in.30) to replace the in. Now use (7. This is the unscattered wave.124CHAPTER 7.35) to eliminate a† and out in a† .states leaving the vacuum expectation value of the ﬁelds.) The limits have a nasty noncovariant look about them.35) to replace a† (k) with an integral over ϕin (x).36) where α − k in represents the initial assemblage of particles represented by α with one particle of momentum k removed.38) ↔ x0 →∞ x0 →−∞ (The symbol ∼ means that I have temporarily neglected the forward scattering term. Since the instate in (??) α in is made up with a product of creation operators as in (??). we can write Sβα = β outa† (k)α − k in in (7. which we will neglect for the time being.and out.35) The notation a ∂0 b means a∂0 b − (∂0 a)b. The scattered part can be written as follows: i Sβα ∼ √ Z lim − lim d3 x fk (x) ∂0 β outϕ(x)α − k in (7. The ﬁrst term in the last line vanishes unless the initial and ﬁnal out states are identical.37) I have added and subtracted a† (k) and used (7.ﬁelds with the complete interacting ﬁeld. The procedure is now to use (7. THE INTERACTION PICTURE AND THE SMATRIX Equation (7. lim − lim d3 x fk ∂0 ϕ = ↔ ∞ −∞ x0 →∞ x0 →−∞ d4 x ∂0 fk ∂0 ϕ ↔ .33) can be inverted to yield ain (k) = i a† (k) = −i in ↔ ∗ d3 x fk (x) ∂0 ϕin (x) ↔ ↔ d3 x fk (x) ∂0 ϕin (x) (7. Sβα = β outa† (k)α − k in out + β out a† (k) − a† (k) α − k in out in = β−k outα−k in −i β out d3 x fk (x) ∂0 (ϕin (x) − ϕout (x)) α − k in ↔ (7. I will clean up the expression with a series of elegant mathematical tricks.
40) The sense of the right arrow is that the subsequent expression is obtained by integrating by parts twice and discarding the surface terms. THE LSZ REDUCTION SCHEME ∞ 125 ∂2 ∂2 ϕ − ϕ 2 fk (7. We will eventually Fourier transform this expression. The new thing here is the time ordered product.40) into (7. Just repeat this procedure n + m times to get the following result: Sβα = p1 · · · pn outk1 · · · km in ∼ i √ Z m+n m i=1 n d xi fki (xi )(2xi + m ) j=1 4 2 ∗ d4 yj fpj (yj )(2yj + m2 ) × 0T [ϕ(y1 ) · · · ϕ(yn )ϕ(x1 ) · · · ϕ(xm )]0 (7.41) i +√ d4 x fk (x)(2 + m2 ) β outϕ(x)α − k in Z You see the pattern? We have “reduced” an asymptotic particle from the instate and replaced it with an interacting ﬁeld ϕ(x).7.3.43) . whereupon all these complications will disappear leaving only the inverse of the momentum space propagator ∆(k).38) gives the ﬁnal result Sβα = β outin α = β − k outα − k in (7. a wave function fk (x). and a diﬀerential operator (2 + m2 ).39) and (7.42) d4 x † d4 y fk (x)fp (y)(2x + m2 )(2y + m2 ) × β − p. in The ∼ sign again means that we have dropped the forward scattering terms. As an exercise (I am getting tired of typing) you should take the next step and reduce a particle of momentum p out of the outstate. Finally substituting (7. The price we pay for this is an integral. outT [ϕ(x)ϕ(y)]α − k.39) ∂x2 ∂x0 −∞ 0 The fact that fk satisﬁes the KleinGordon equation allows us to write = d4 x fk ϕ ∂2 f = ϕ( ∂x2 0 2 − m2 )f → f ( 2 − m2 )ϕ (7. Now suppose there are m particles in the initial state and n particles in the ﬁnal state. Convince yourself that the result is Sβα = β − p outα − k in ∼ i √ z 2 (7. You should do the calculation carefully to see how this comes about.
For example.” (This was before the days of path integrals. In this simple theory with selfinteracting scalar ﬁelds. Sβα = p1 · · · pn outk1 · · · km in ∼ n i √ Z m+n m i=1 1 2 (ki − m2 ) (2π)3 2Eki ˜ (p2 − m2 )G(k1 · · · km p1 · · · pn ) j (7. THE INTERACTION PICTURE AND THE SMATRIX Equation (7. past and future are unambiguous. After making this assignment.45) There is no distinction between x’s and y’s.44).44). We therefore associate some coordinates with instates and some with outstates. so introduce the Fourier transform.44) into (7. on the other hand.43). The point is that G(x1 · · · xm y1 · · · yn ) is a “general purpose” tool that can be used in many situations including those on a quantummechanical time scale in which there is no clearcut distinction between past and future. We do this by assigning them to k’s (for incoming particles) or with p’s (for outgoing particles). but in (7.) Feynman rules are usually expressed in momentum space. if we were looking at the process e+ + e− → γ + γ.44) There is a possible confusion about this equation. the only diﬀerence is the + or − signs in the exponentials in (7. this choice becomes even more signiﬁcant.46) × j=1 1 (2π)3 2Epj . but this is necessary to get the conservation of momentum right. (positron annihilation) we would need to assign x’s to the electron ﬁelds and y’s to the photon ﬁelds.126CHAPTER 7.43) and do the necessary integrals. to quote Bjorken and Dell. “serves as a cornerstone for all calculation of scattering amplitudes in modern quantum ﬁeld theory. substitute (7. Since G(x1 · · · xm y1 · · · yn ) = 0T [ϕ(y1 ) · · · ϕ(yn )ϕ(x1 ) · · · ϕ(xm )]0 (7. x’s are clearly associated with incoming particles. m G(x1 · · · xm y1 · · · yn ) = i=1 eiki xi d ki (2π)4 4 n d4 pj j=1 e−ipj yj ˜ G(k1 · · · km p1 · · · pn ) (2π)4 (7. and y’s are associated with outgoing particles. In scattering experiments. When there are several diﬀerent kinds of particles involved.
Both ϕ and ϕin satisfy Heisenberg equations of motion.30). ϕ] U −1 + U (U −1 ϕin U )U −1 (7. we must somehow relate them to the “in” ﬁelds deﬁned in (7.43) is the vacuum correlation function G(x1 · · · xn ) = 0T [ϕ(x1 )ϕ(x2 ) · · · ϕ(xn )]0 (7. ϕin (t)] (7.21).52) .4. Let’s hope there exists an operator U (t) such that ϕ(x) = U −1 (t)ϕin (x)U (t) with t = x0 (7.4 Correlation Functions The heart of (7. I have also written ϕ(t). and x0 = t.51) and use (7. t0 ) t0 →−∞ (7. ϕ(t)] ∂t (7. ˙ ∂ ∂ ˙ ˙ ϕin = U ϕU −1 = U ϕU −1 + U ϕU −1 + U ϕU −1 ˙ ∂t ∂t ˙ ˙ = U (U −1 ϕin U )U −1 + U i [H(ϕ).50) to eliminate ϕ. ∂ ϕ(t) = i[H(ϕ). CORRELATION FUNCTIONS 127 7.48) into (7. but I am speciﬁcally concerned with the time development here. If we are to calculate anything in this formalism.49) where U (t. I realize the ϕ is really a function of the four vector xµ . Note that d dU −1 dU −1 U U −1 = U +U =0 dt dt dt In a condensed notation this is ˙ ˙ U U −1 = −U U −1 Now substitute (7.7.47) The ﬁelds ϕ(x) are the fully interacting ﬁelds in the Heisenberg picture.51) ∂t I have written H(ϕ) and H0 (ϕin ) to emphasize that the Hamiltonians are functionals of their respective ﬁelds. t0 ) is time timedevelopment operator given by (7. This is plausible.48) I happen to know that the necessary operator is given by U (t) = lim U (t.50) ∂ ϕin (t) = i[H0 (ϕin ). but it need to be proved.
21) and U (t) by (7.55) we can write n G(x1 . · · · . so for example. U H(ϕ)U −1 = H(ϕin ).57) .15) with the important clariﬁcation that Hint is constructed with in ﬁelds. t1 ) t→−∞ = lim U (t2 . t1 )ϕin (x1 )U (t1 . our correlation function becomes G(x1 . t1 ).53). ﬁelds using the physical masses of the particles involved. Using (7. I have also used the fact that Hamiltonians are polynomials of ϕ (and possibly its derivatives).55) x0 2 Now consider a set of timeordered points x1 . We could think of this constant as a number added to the Hamiltonian. tn )ϕin (xn )U (tn . · · · .53) I have used (7.e. i. t1 ) = lim U (t2 .56) The sequence of times reading from left to right runs from t0 → −∞ to t1 and back again to −∞. xn ) = 0S −1 U (−t0 . −t0 )U (−t0 . With this change.13) we can argue that U (t2 . t2 )ϕin (x2 ) · · · U (tn−1 . We can deform the contour by writing U (t0 . We will simply call U (t0 . We conclude that (to within a meaningless additive constant) dU (t) = Hint (ϕin )U (t) (7. ϕin ] = i [H0 (ϕin ). U ϕU −1 ˙ = U U −1 . xn satisfying x1 > 0 > · · · > x0 . THE INTERACTION PICTURE AND THE SMATRIX ˙ ˙ = U U −1 ϕin − ϕin U U −1 + i (U H(ϕ)U −1 ). An operator of the form U (−t0 . ϕin + i [H(ϕin ). We have had several occasions to discard such numbers before and will do so now. t2 ) · · · U (tn−1 . t0 ) to be given by (7. t0 ) runs from −∞ to +∞.128CHAPTER 7.12) and (7. −t0 ) = S −1 . This is just the role of the S operator deﬁned in (7. t0 ) = U (t2 )U −1 (t1 ) t→−∞ (7. tn )ϕin (xn )U (tn . t1 ) = U (t0 . Using (7.50). It has the eﬀect of transforming asymptotic initial states to asymptotic ﬁnal states. Therefore ˙ U U −1 + iHint (ϕin ) . ϕin ] (7. We can take U (t.54) i dt This is equation (7. · · · . xn ) = 0ϕ(x1 ) · · · ϕ(xn )0 = 0U (t0 . t1 )ϕin (x1 )U (t1 . t0 )U (t0 .49). t0 )0 (7.52) to simplify the second line of (7. t0 )U −1 (t1 . t0 )0 (7. ϕin = 0 The operator in curly brackets commutes with all ϕin . It must therefore be a c number.
· · · .61) . · · · . Consequently 0S −1 = Equation (7. ti ). xn ) = 0U (−t0 . We just need to write it G(x1 .7. In particular. The original deﬁnition of G. xn ) = 0T [ϕin (x1 )ϕin (x2 ) · · · ϕin (xn )S] 0 0S0 (7. t2 ) · · · U (tn−1 . the vacuum state is an eigenstate of S with a complex phase. we started out with the assumption that the sequence of coordinates x1 . Of course. t0 )] 0 0S0 With the timeordering operator in place. t0 )0 0S0 (7. tn )ϕin (xn )U (tn . x2 . U (−t0 .27). t1 )U (t1 .47) already contains the time ordering operator.58) Consider an operator like U (ti−1 . 0 0 constitutes a complete set of states. t1 )ϕin (x1 )U (t1 .59) consists of a vast array of ﬁelds. t2 ) · · · U (tn−1 . we are free to move the factors around in the numerator to suit our convenience. but every term in the sum contains ﬁelds that are timeordered between ti−1 and ti . CORRELATION FUNCTIONS 129 According to (7. xn ) = (7. we can gather all the U ’s together. It contains an inﬁnite sum of products of ﬁelds evaluated at various times.59) 0 0S0 (7. t2 ) · · · U (tn . · · · xn was already time ordered. · · · . so this makes the “scaﬀolding” around which the U ’s are time ordered. The entire numerator of (7. but they are all time ordered.60) 0T [U (−t0 . equation (7. We can write 0S −1 = e−iθ 0 and ﬁnd the phase as follows. t1 )ϕin (x1 )U (t1 . t0 ) = S Our ﬁnal result for the correlation function function is G(x1 . t0 ) = U (−t0 . tn )ϕin (xn )U (tn .4.57) becomes G(x1 . so we need not have made that assumption. 0S0 0S −1 = 0 The point is that since the vacuum is stable.
and there are many subtle physical questions as well.63) By S in (7. The fact is that when you ﬁnally get the results in momentum space. xn ) = 0T [ϕ(x1 )ϕ(x2 ) · · · ϕ(xn )]0 0T [ϕI (t1 )ϕI (t2 ) · · · ϕI (tn )S]0 .63) I mean the operator U (−t0 . This also goes way beyond the limits of conventional mathematics. ∞ S= n=0 (−i)n n! ··· I I I d4 x1 d4 x2 · · · d4 xn T [Hint (x1 )Hint (x2 ) · · · Hint (xn )]. x2 . Itzykson and J. McGrawHill 1980 . In order to do this we need the Fourier transformed Green’s function given (implicitly) by (7.62) Now we need the conﬁguration space Green’s function G(x1 · · · xm y1 · · · yn ). It is also true that equivalent formulas can be derived using the path integral formalism. C.46). = 0S0 (7.5 The Wick Expansion If we can calculate Sβα in (7.130CHAPTER 7. they are often embarrassingly simple. 7. Zuber. This led Feynman to formulate a set of rules that give the right answer in momentum space without having 4 Quantum Field Theory. call this derivation “heuristic. · · · .20) with inﬁnite limits on the integrals.61) G(x1 .20).” The fact is that we just don’t know how to do rigorous mathematics with such badlybehaved operator expansions as (7. m n ˜ G(k1 · · · km p1 · · · pn ) = i=1 d xi e 4 −iki xi j=1 d4 yj eipj yj G(x1 · · · xm y1 · · · yn ) (7. This seems (and is) an excruciatingly complicated set of equations to wade through. (7. The formula for this is (7. You might call this “experimental mathematics. THE INTERACTION PICTURE AND THE SMATRIX Itzykson and Zuber4 in a masterstroke of French understatement.44). but the same result is obtained in the end. t0 ) in the limit t0 → −∞. Then the equation for S is (7.33).B.64) where it is understood that the limits on the time integrations are −∞ < I x0 < ∞ and that Hint is constructed with ϕin given by (7. we can calculate all possible decay rates and cross sections.” We use this formula and it works.
65) to rewrite each term (if necessary) in such a way that all creation operators stand to the left of all annihilation operators. • Multiply the ﬁelds together with the timeordering θ functions. the ﬁrst term on the right is zero. Take a look at the derivation of the propagator at the end of Chapter 6. As a consequence 0T [ϕ(x1 )ϕ(x2 )]0 = 0 : ϕ(x1 )ϕ(x2 ) : 0 + iD(x1 − x2 ) (7. This gives eight terms. I will write ϕ(x) = ϕ+ (x) + ϕ− (x).66) Of course. I could just tell you the rules now. We say that iD(x1 −x2 ) is the contraction of ϕ(x1 ) and ϕ(x2 ). There we were dealing with free ﬁelds so we should rewrite (6. for example ϕ+ (x1 )ϕ− (x2 ) =: ϕ− (x2 )ϕ+ (x1 ) : + d3 k e−ik(x1 −x2 ) (2π)3 2ω The ﬁrst term on the right is normal ordered. and ϕ− is the negativeenergy term with the creation operator. : ϕ(x1 )ϕ(x2 ) : means that the terms between the colons have already been placed in normal order. The usual notation for this is to enclose the expression between colons. i. In this notation (7. T [ϕ(x1 )ϕ(x2 )] = (ϕ+ (x1 ) + ϕ− (x1 ))(ϕ+ (x2 ) + ϕ− (x2 ))θ(t1 − t2 ) + (ϕ+ (x2 ) + ϕ− (x2 ))(ϕ+ (x1 ) + ϕ− (x1 ))θ(t2 − t1 ) • Use the commutation relations (6.65) I am going to describe the steps that led from (6.66) is T [ϕ(x1 )ϕ(x2 )] =: ϕ(x1 )ϕ(x2 ) : + ϕ(x1 )ϕ(x2 ) (7.67) .72) 0T [ϕI (x1 )ϕI (x2 )]0 = iD(x1 − x2 ) (7.5. each one containing a product of two of the creation and/or annihilation operators. This is called normal ordering. THE WICK EXPANSION 131 to calculate anything. In the course of normal ordering there will appear terms without operators since. The second term is a cnumber.75) in rather formal terms that make it easy to generalize to situations in which there are more than two ﬁelds. • Take advantage of the fact that the vacuum expectation value of any normal ordered product is zero.e. I will use the overbrace to indicate contraction. Let’s derive them step by step. but they would be devoid of any intellectual content.72) to (6. In order to keep the notation under control. where ϕ+ is the positiveenergy piece containing an annihilation operator.7.
64) and (6.132CHAPTER 7. • You see the pattern? One term with no contractions plus all possible terms with one contraction plus all possible terms with two contractions plus all possible terms with three contractions and so forth. Relativistic quantum Fields. THE INTERACTION PICTURE AND THE SMATRIX I am now in a position to state Wick’s theorem. each one of which consists of one pair of contracted ϕ’s.4 (7. The last group of terms will consist of n/2 contractions. The last terms will have one normalordered ﬁeld left over This is all much easier to understand that to explain. The following example with n = 4 should make this clear.65). repeat this procedure (n − 1)/2 times. Drell. There will be one term in this group for each way of choosing n/2 pairs out of n objects. If n is odd. D. There is one term in the sum for each way in which two distinct pairs can be made out of n ﬁelds. If n is even. D. • A sum of terms. The n − 2 remaining ﬁelds are normalordered. • A sum of terms. There is one term in the sum for each possible pairing of two of the ﬁelds. each one of which consists of two pairs of contracted ϕ’s.68) . The proof is more work that it’s worth. The time ordered product of n ﬁelds T [ϕ(x1 )ϕ(x2 ) · · · ϕ(xn )] can be written as a sum of terms as follows: • A single term in which all n ﬁelds are normalordered. You will be convinced that the theorem is true. T [ϕ(x1 )ϕ(x2 )ϕ(x3 )ϕ(x4 )] =: ϕ(x1 )ϕ(x2 )ϕ(x3 )ϕ(x4 ) : + ϕ(x1 )ϕ(x2 ) : ϕ(x3 )ϕ(x4 ) : + ϕ(x1 )ϕ(x3 ) : ϕ(x2 )ϕ(x4 ) : + ϕ(x1 )ϕ(x4 ) : ϕ(x2 )ϕ(x3 ) : + ϕ(x2 )ϕ(x3 ) : ϕ(x1 )ϕ(x4 ) : + ϕ(x2 )ϕ(x4 ) : ϕ(x1 )ϕ(x3 ) : + ϕ(x3 )ϕ(x4 ) : ϕ(x1 )ϕ(x2 ) : + ϕ(x1 )ϕ(x2 ) ϕ(x3 )ϕ(x4 ) + ϕ(x1 )ϕ(x3 ) ϕ(x2 )ϕ(x4 ) + ϕ(x1 )ϕ(x4 ) ϕ(x2 )ϕ(x3 ) You should try working out one of these time ordered products by brute force starting with (6. The remaining n − 4 remaining ﬁelds are normalordered.5 5 If you want a proof see J. Bjorken and S. Section 17. repeat this procedure n/2 times.
and calculate the fourpoint Green’s function G(x1 . The ﬁrstorder term has eight ﬁelds each with two terms.68) G(x1 . x2 )G(x3 . x3 ) (7. the particles do not interact . If you were to simply multiply them out there would be 28 = 256 terms. x4 ) to ﬁrst order in λ. As you would expect in this zerothorder term.63) becomes G(x1 . This is a common ¯ problem whenever we make Hamiltonians our of ﬁeld operators. so (7. The Soperator (7. x2 . x3 .3 and 4. AN EXAMPLE 133 This would be a good time to go back to the free ﬁeld Hamiltonian. x4 )G(x2 . we only make contractions of pairs of ﬁelds from diﬀerent Hamiltonians.70) Note that to ﬁrst order 0S0 = 1. x4 ) + G(x1 .71) The zerothorder term can be read oﬀ of (7.69) When we use Wick’s theorem on this expression.4. (6.64) should be written ∞ n=∞ n=∞ S= n=0 (−i)n n! ··· I I I d4 x1 d4 x2 · · · d4 xn T [: Hint (x1 ) :: Hint (x2 ) : · · · : Hint (xn ) :]. In order to insure that this never happens. so here’s where . (7. x2 . x4 ) + G(x1 .6 An Example Let’s take the interaction Hamiltonian Hint = (λ/4!)ϕ4 (x) that we studied in Sections 4.64) to ﬁrst order is S =1− iλ 4! d4 x : ϕ4 (x) : + · · · (7.6. x3 . x4 ) = 0T [ϕI (x1 )ϕI (x2 )ϕI (x3 )ϕI (x4 )]0 − iλ 0T ϕI (x1 )ϕI (x2 )ϕI (x3 )ϕI (x4 ) 4! d4 x : ϕ4 (x) : 0 + · · · I (7. 1 1 H= h ¯ ωn (an a† + a† an ) = h ¯ ωn a† an + n n n 2 n=−∞ 2 n=−∞ We had to discard the hω/2 term to avoid an inﬁnite sum.72) This is the sum of all possible ways two particles can propagate between two pairs of points. we will agree that all Hamiltonians come already normally ordered.43). Eq. x3 )G(x2 .7. 7. Accordingly (7.
ϕ(x3 ). • First do the d4 x integration in (7.74). There are.62) and ﬁnd the Fourier transform G(k1 . x). This simple calculation illustrates most of the rules for calculating more complicated Smatrix elements in the ϕ4 theory. though less obvious. x2 . 4! ways of doing this all leading to the same expression for the ﬁrstorder contribution to G(x1 . We need only be concerned with terms in which all eight ﬁelds are contracted. x)G(x2 . of course. saves a factor of two in scratch paper. Because of the normalordered Hamiltonian.46) to get Sβα . x2 . all the (2xi + m2 ) operators in (7. C. so (2x + m2 )G(x. To start with draw some diagrams with the incoming and outgoing particles labelled with their momenta. Now all the integrals are trivial. x)G(x3 .134CHAPTER 7. and ϕ(x4 ) is contracted with one of the ϕ(x)’s. we don’t contract the four ϕ(x)’s with one another.43) produce factors of −iδ (4) (xi − x) when they operate on G(xi . p2 ). First notice that the vacuum expectation value of any normalordered product of ﬁelds vanishes. (7. x)G(x4 . x3 . x2 . x ) = i(2x + m2 )D(x − x ) = −iδ (4) (x − x ) (7. p1 . k2 . ϕ(x2 ). x4 ) ˜ into (7.73).74) You see. There are two ways of doing this.73) which is exactly (4. Wick earns his keep.75) 1 (2π)3 2Eki 1 (2π)3 2Epj This second technique. Finally ˜ substitute G into (7. x3 . Then substitute G(x1 . G(x1 . G. Now suppose you want to calculate the S matrix element for the scattering of two particles k1 + k2 → p1 + p2 . x4 ) = −iλ d4 xG(x1 . x4 ). x). The only surviving terms are those in which each of ϕ(x1 ). x3 . THE INTERACTION PICTURE AND THE SMATRIX Mr. Sβα = p1 p2 outk1 k2 in = × i=1 i √ Z 2 4 (−iλ)(2π)4 δ (4) (p1 + p2 − k1 − k2 ) 2 j=1 (7. The diagrams will consist of straight lines and vertices where four . • G is a Green’s function after all.
The number of vertices must be less than or equal to the corresponding order of perturbation theory. • A factor 1/ (2π)3 2Ek for each external particle with momentum k. Each one of these gets a propagator i p2 − m2 + i where p is the momentum of the internal particle. Each of the remaining diagrams contributes a term to the perturbation expansion. • Impose momentum conservation at each vertex. These are called “disconnected” diagrams. There are several new features that are only present in higherorder terms. Discard all these.7.e. The self interaction diagrams represents particles that don’t scatter. It can be shown that the vacuum ﬂuctuations only contribute to the 0S0 term. lines that begin and end in the diagram. • A factor (−iλ) for each vertex. AN EXAMPLE 135 lines meet. if you working at the nth order approximation. i. • Integrate over each undetermined loop momentum: d4 p (2π)4 . which cancelled in the formula for the Smatrix. • There will be internal lines.75). • A factor (2π)4 δ (4) (k1 +k2 −p1 −p2 −· · · ) to enforce allover momentum conservation. Each term will have the following factors. These will fall into two categories: “vacuum ﬂuctuation” diagrams in which there is a piece that is disconnected from all the external particles and “self interaction” diagrams in which one of the external particles emits and reabsorbs virtual particles without interacting with the other external particles. and we are presumably not interested in the probability of things not happening. There will be many diagrams in which one or more of the particles don’t interact with the others. you must draw all possible diagrams in which there are n or fewer vertices.6. i. there was only one vertex corresponding to ﬁrstorder perturbation theory. In the example. These factors are all evident in (7. One way to say this is that these diagrams can be separated into two or more pieces without cutting any lines.e.
the symmetry factor. . This last issue. It’s too complicated to explain in the text. That is the subject of the next chapter. I’ll explain it in class when I can draw lots of diagrams on the board. How do you turn an Smatrix into a cross section. There’s still one more step on the road. THE INTERACTION PICTURE AND THE SMATRIX • Divide by the symmetry factor. has to do with redundant diagrams.136CHAPTER 7.
First we prepare a target with nt of these particles per unit volume. We then count the number N of incident particles that are deﬂected as a result of a 137 . There are two easy ways of doing this and one right way. Plane waves occupy inﬁnite volume. We’ll do it the ssecond easy way (at least in this edition). so several key integrals diverge. and bombard this target with Ni incident pointlike particles. the cross section σT = πR2 . we might contrive to measure its cross section by means of a scattering experiment. but of course. The other easy way is to quantize the wave functions in a box of ﬁnite size. If the particle is a sphere of radius R. This avoids the inﬁnite integrals. If the particle were very small. This is usually called the “total” cross section.1 Classical Scattering From the point of view of classical mechanics. We assume that the target is so thin and the target particles so small that no one of them gets in front of another. It is best to acknowledge that the beam and target particles should be represented by wave packets. One easy way is to assume that the initial and ﬁnal state particles can be described by plane waves. but this makes for a long tedious derivation. the cross section of a particle is just its projected area. unfortunately. real experiments aren’t done in a box. 8.Chapter 8 Cross Sections and Decay Rates The last step in the road from Lagrangians to cross sections is this: take the Smatrix computed in Chapter 7 and convert it to a scattering cross section or decay rate. This problem can be ducked with some judicious hand waving.
The relevant equation is dσ 1 dN = (8. where A is the crosssectional area of the beam (assumed smaller than the target. If one of the incident particles hits a particular area dσ it will scatter into a solid angle dΩ. If the beam is a pulse that is turned on for T seconds. Just ﬁre a gazillion arrows at random an observe the distribution of scattering angles. This computed quantity is called the “diﬀerential cross section.4) (8. N/Ni must be equal to the fraction of the target area covered by the projected area of the target particles. but for purposes of comparing with theory. compute dσ/dΩ.3) . We can also determine the shape with a scattering experiment. and the ﬂux I is I = ni nt v (8. then the volume of the beam is vT A. The fraction of particles that are scattered. where L is the target thickness. and pass it along to the theorist. Imagine that the target particles have concentric circles painted on them like an archery target. and L in (8. V = AL. you measure dN/dΩ in your lab.138 CHAPTER 8. nt . The cross section can be written as: dσ 1 dN = dΩ (ni vT A)nt L dΩ (8.1) are directly measurable.2) 1 dR = . then the number of particles in the beam Ni is equal to the density of the beam ni times the volume. nt LσT . If the beam is moving at velocity v toward a stationary target. I dΩ where the transition rate R is the number of scattering events per unit time per unit target volume. Your favorite theorist should then be able to infer the shape of the target particle from the distribution.” The quantities Ni .1) dΩ Ni nt L dΩ If you are an experimentalist. it is more convenient to express them in terms of ﬂux. We just have to make Ni and nt large enough. OK – so your’re not as good with a bow and arrow as Legolas. The cross section is then given by N σT = Ni nt L The idea works even if σT is arbitrarily small. CROSS SECTIONS AND DECAY RATES collision. R = N/V T.) Therefore Ni = ni vT A.
The description of scattering in terms of archery targets is appropriate to lowenergy potential scattering. How do we now ﬁnd the cross sections? You will recall from the end of Chapter 7 that the ﬁrst term of the perturbation theory expansion of the Smatrix describes particles that don’t scatter. We can imagine the momenta as occupying some complicatedshaped region in a 3ndimensional space whose coordinates are the components of the individual momenta. but now dσ might be an 3nfold diﬀerential. but the experiment can still be done exactly as I have described it.2 Quantum Scattering In the quantum mechanics regime. We know how to calculate the Smatrix elements. Rather we have to specify the momenta of each of the particles: p1 . pn . We still use the diﬀerential cross section notation dσ. quantities like size and shape don’t have any direct meaning. It requires more than a scattering angle to describe the ﬁnal state. Not all values of the momenta are allowed of course. . Now let’s put on our theorist’s hats. . The diﬀerential cross section (DCS for short) can’t be interpreted directly in terms of shape. p2 .35) that the S operator is unitary.” If should be clear from (6. where i and f  represent complete 1 Note that phase space in thermodynamics is 6ndimensional. . because of the conservation of momentum and energy. QUANTUM SCATTERING 139 8. since any putative size is bound to be smaller that the corresponding deBroglie wavelength.”1 A typical multiparticle scattering experiment measures the transition probability into some region of phase space.25) and (6. Feynman’s rules decree that all other terms contain the factor (2π)4 δ (4) (Pf − Pi ). . .5) The δf i just represents the particles that don’t scatter. It is customary to deﬁne a transition matrix Tf i as follows: Sf i = δf i − i(2π)4 δ (4) (Pf − Pi )Tf i (8. but at higher energies new particles can be created and the total momentum shared among them. but it does contain most of the information we are entitled to have about particles and other subatomic systems. I will also use the notation ∆Ω to indicate some small region of phase space (not just a region of solid angle).2. This construct is called “phase space. depending on the details of the experiment. where Pi and Pf are the total momentum fourvectors of the initial and ﬁnal state particles.8. The Smatrix in the current notation is Sf i = f Si . because all the measurements are macroscopic and asymptotic. T is the nontrivial part of the Smatrix with the delta function “peeled oﬀ. since it includes space as well as momentum coordinates.
The factor (2π)4 δ (4) (0) is interpreted as a volume of spacetime. Let V be a small arbitrary volume inside the interaction region.5) gives Pf i = (2π)4 δ (4) (0)(2π)4 δ (4) (Pf − Pi )Tf i 2 (8. so I hope you will indulge me while . I’ll get back to that. Squaring (8.10) This is the ﬁrst of several instances in which the arbitrary V ’s and T ’s cancel. Now to the question of counting and normalizing states.6) I say “very schematically” because it’s not at all clear what it means to sum over one of these indices. T is the time required for a beam particle to pass through V . So for example. This is not explained clearly in any book I know of. Since Sf i is a probability amplitude.140 CHAPTER 8.8) (I have ignored the nonscattering term. since in the limit P → 0 (2π)4 δ (4) (P ) = d4 x eiP x → d4 x = V T. (8. this means that ∗ Sf i Sf k = δik f (8. I’ll deal with this last problem ﬁrst.2) as a model.) Here’s the kludge.9) But the time and volume of what? We can use the derivation of (8. we should get the corresponding probability by squaring it. and this is bad news.7) is the probability that the initial state i will scatter into the ﬁnal state f . As such we expect the Smatrix to be unitary as well. Unfortunately this means that we have to square the delta function in (8.3) is Rf i = transition rate per unit volume = = (2π)4 V T δ (4) (Pf − Pi )Tf i 2 VT 4 (4) =(2π) δ (Pf − Pi )Tf i 2 Pf i VT (8. The analogy to (8. Assume that it is small enough to ﬁt inside the beam and target but very large compared the typical quantummechanical dimensions.5). since δ(Pf − Pi )δ(Pf − Pi ) = δ(Pf − Pi )δ(0) and δ(0) is inﬁnite. CROSS SECTIONS AND DECAY RATES sets of states. Pf i = Sf i 2 (8. Very schematically.
First of all. In order to have the right behavior under Lorentz transformations.11) ∂t The reason is that the rate of decrease of particles in a given volume is equal to the total ﬂux of particles out of that volume. not the quantized ﬁeld. KleinGordon wave functions have a conserved norm. ∂ ∂ϕ ∂ϕ∗ i ϕ∗ −ϕ ∂t ∂t ∂t ρ + · [−i(ϕ∗ ϕ − ϕ ϕ∗ )] = 0 j (8. and ← → ϕ∗ ∂0 ϕ ≡ ϕ∗ ∂0 ϕ − (∂0 ϕ∗ )ϕ. that is − ∂ ∂t ρ dV = V S j · n dS = ˆ V · j dV The procedure for ﬁnding this current should be familiar from nonrelativistic quantum mechanics. That is.13) By ϕ(x) I mean the wave function.2.12) As a consequence the norm is ρ=i ← → d3 x ϕ∗ (x) ∂0 ϕ(x) (8. Someone has to do it. but it isn’t dV ϕ2 . there must be a probability density ρ and a probability current j such that ∂ρ + ·j =0 (8. We have been using the quantized ﬁeld operator ϕ= ˆ d3 k (2π)3 2Ek ˆ ak e−ikx + d† eikx ˆ k . Write the KleinGordon equation as − ∂2ϕ + ∂t2 2 ϕ = m2 ϕ Multiply this equation by −iϕ∗ and the complex conjugate equation by −iϕ and subtract. QUANTUM SCATTERING 141 I explain this boring technical detail in excruciating detail. the norm must transform like the timelike component of a conserved current.8.
Only the equaltime commutation relations (6.64).16) In the general case in which two particles go into N particles with initial and ﬁnal states given by i = k2 . CROSS SECTIONS AND DECAY RATES from (6.15) Yes that’s right – there are (2π)−3 particles per unit volume! Equation (8. p2 . but no two books do it the same way! Our choice of normalization for the ﬁeld operator has the advantage of making (6. · · · . we can impose periodic boundary condition to discretize the momenta. First is the factor 1/ (2π)3 2E in the ﬁeld operator (6. quantum theory restricts the number of ﬁnal states in a volume V with momenta in element d3 p to be V d3 p/(2π)3 . This will happen. however. We will ﬁnally arrive at formulas for the cross sections and decay rates that must be independent of all conventions.142 CHAPTER 8. we have agreed to construct singleparticle states as p = a† 0 .65) as simple as possible. We obtain the classical wave function ϕ from the quantized ﬁeld operator ϕ as follows ˆ ϕ(x) = 0ϕ(x)p = ˆ = d3 k (2π)3 2Ek e−ikx 0ˆk a† 0 a ˆp (8. What does this mean? Since we already are committed to the imaginary box.64).10) refers to scattering into a single ﬁnal state f .17) . pN (8. but it results in a rather oddlooking norm. Second is the commutation relations for the creation and annihilation operators (6. so No. For a single particle. Finally.65). k2 f = p1 .69) are nonˆp negotiable. Our V has (2π)−3 V particles.14) e−ipx (2π)3 2Ep This result depends on three normalization conventions that are to some extent arbitrary. ← → ρ = i(ϕ∗ ∂0 ϕ) = 2Eϕ∗ ϕ = (2π)−3 ρ dV = (2π)−3 V V (8. of ﬁnal states/particle = V d3 p = d3 p (2π)3 (2π)−3 V (8.
20) depends. Second. PHASE SPACE the “density of states” is N 143 No. we use this formalism to compute the probability of the decay of a single particle. generally speaking.18) and (8. which as usual is equal to (2π)−3 .10). . First there are the fundamental dynamics.4).21) (Pf − Pi )Tf i  d pj 3 In this case nt is the number of decaying particles per unit volume. 8.18) Now turn to the incident ﬂux (8.19) The generalization of (8. (8. v= k2 1 k1 − = E1 E2 E1 E2 (k1 · k2 )2 − (m1 m2 )2 (8.20) 1 = (2π)10 δ (4) (Pf − Pi )Tf i 2 v d3 pj . of available states = j=1 d3 pj (8. With our norm ni = nt = (2π)−3 . j=1 Finally. and v is the relative velocity of the colliding particles. on three kinds of things.8.3. The diﬀerential decay rate dΓ of an unstable particle is deﬁned by dΓ = = (2π) δ 7 (4) R × density of states nt N 2 j=1 (8.4) is dσ = R × density of states I Using (8. which are encoded in the Hamiltonian of the system.7) we get dσ = (2π)4 δ(Pf − Pi )Tf i 2 (2π)−6 v N d3 pj j=1 N (8.3 Phase Space The outcome of a scattering experiment as described by (8.
the number of available states is N= 1 (2π¯ )3 h d3 x d3 p = V (2π¯ )3 h d3 p. Classical mechanics places no restrictions on the density of these points. I h am concerned here with the integral over d3 p. (x. px . I will explain this again in the next paragraph using a semiquantitative argument based on the uncertainty principle and ﬁnally present a version of the argument that is correct relativistically. y. h Phase space can be thought of as being subdivided into elementary cells of size (2π¯ )3 . the lot will empty quickly. then it will take you a long time to escape. so that N depends on m and E. the construction of the detectors. If there are many momentum states available to the ﬁnalstate particles. There is a third category of inﬂuences that arise because momentum and energy are conserved and because the experiment must be describable in a relativistic invariant way. the scattering cross section will be large or the decay will happen quickly. The factor V /(2π¯ )2 is part of the normalization discussed in the previous section.144 CHAPTER 8. pz ). In classical statistic mechanics the state of motion of a particle is represented as a point in a sixdimensional manifold called phase space consisting of three position and three momentum coordinates. e. These last considerations are incorporated in the notion of phase space. etc.g. ∆xj ∆pj ≥ 2π¯ . Assuming that the particle is contained in a volume V . The limits of integration come about from the conservation of mass and energy. Quantum mechanics on the other hand limits the precision by the uncertainty principle. Imagine leaving the parking lot in front of a movie theater after the show is over. Since pµ pµ = m2 . CROSS SECTIONS AND DECAY RATES there are the details of the experiment itself. the number of particles in the beam and target. If there are freeways leading out in all directions.e. since x and p are supposedly measurable with complete precision. Try to imagine something like this in momentum space and you will have a feeling for what phase space means. z. Phase space in this context is the set of momentum states available to the particles after the interaction. If there is only one narrow road leading out of the lot. . the momentum integration is bounded by a sphere of radius p2 = E 2 − m2 . The number of states available to one particle will therefore be equal to the total volume of phase space divided by the size of the elementary cell. py . The particle’s coordinates cannot be localized to any smaller h region. This is an analogy based on coordinate space. i.
N −1 NN = j=1 d3 pj (8. N −1 k=1 N −1 N N d3 pN δ (3) pN + pk − Pi =1 NN = j=1 d3 pj = j=1 d3 pj δ (3) k=1 pk − Pi Energy is also conserved. that the reaction rate is proportional to the density of states dNN /dEi . L. j=1 We can incorporate this constraint in a more ﬂexible way by introducing a δfunction. dEi δ(Ef − Ei ) = 1 N N N NN = dEi j=1 d pj δ dNN = dEi 3 (3) k=1 N pk − Pi δ k=1 Ek − Ei d3 pj δ (4) (Pf − Pi ) j=1 (8.23) as it stands is not Lorentz invariant.2 This is not the whole story. Liboﬀ. See R. Introductory Quantum Mechanics. 2 .22) There are only N − 1 integrals in (8. N pj ≡ Pf = Pi .22).20). because (8.” 2π wlk = g(Ek )Hkl 2 h ¯ where g(Ek ) is the density of states whose energy is close to Ek . but according to the Feynman rules from Section You will recall from elementary quantum mechanics that the transition probability rate for a system to make a transition from an energy El to Ek is given by “Fermi’s golden rule. PHASE SPACE 145 Now suppose there are N such particles in the ﬁnal state of a scattering experiment. because the N pj ’s are not all independent.3. It appears from this heuristic argument.8. and this constraint can be included with a similar trick.23) These are just the kinematic factors appearing in (8.
j (8.24) and (8. then the corresponding momentum pj would be integrated out. e.20) together with the factors 1/2Ei from  f T i 2 can be combined into a Lorentzinvariant quantity N P= j=1 N d3 pj (4) δ (Pf − Pi ) 2Ej (8. M. CROSS SECTIONS AND DECAY RATES 7.  f T i 2 in (8. P will have a complicated dependence on the momentum variables.24) over some portion of their allowed kinematic range. the square of the scattering amplitude.24) = j=1 d4 pj δ(p2 − m2 ) δ (4) (Pf − Pi ) j The quantity P is called the Lorentzinvariant phase space. In general we are more interested in integrating some of the momenta in (8. The general purpose of an experiment is always to ﬁnd some deviation from phase space. 3 There are entire books written on the subject.20). contains a factor 1/2Ei (2π)3 for each particle in the ﬁnal state. Gordon and Breach (1968) . Nikolic.3 It is often the case that the data from an experiment are determined mostly by P regardless of the dynamics.146 CHAPTER 8. If the momentum of the k’th particle was measured. and P would be left as a function of pk .25) lie on the interface between theory and experiment. since d4 p δ(p2 − m2 ) = d3 p 2E is clearly a Lorentz scalar. Kinematics and Multiparticle Systems. This is good. if you were using (8. the limits on the integrals would be set to represent the details of the experimental apparatus.5. in which case N dP = p∈∆ j=1 d4 pj δ(p2 − m2 )δ (4) (Pf − Pi ).20) to predict the results of an experiment.25) You might say that (8.g. If the apparatus did not detect particle j (it might be a neutral particle). then the corresponding integral would not be done. Such an experiment is a failure almost by deﬁnition. For manyparticle reactions. The point is that the factors N j=1 d3 pj (2π)4 δ (4) (Pi − Pf ) from (8. For example.
In general there will be some complicated function that we may or may not want to integrate.27) The transition matrix Tf i is especially simple in this example. and Ek1 + Ek2 ≡ Ei .4 TwoParticle Scattering To make this more concrete. p1 dp1 = E1 dE1 P= 1 δ(2E1 − Ei ) = δ(E1 − Ei /2) 2 p1 dE1 δ(E1 − Ei /2)dΩ1 8E1 There is still one δfunction. TWOPARTICLE SCATTERING 147 8.75) (with Z = 1) with (8. p1 = −p2 .28) Notice that symmetry together with the conservation of momentum and energy largely predetermine the outcome of such a scattering experiment. Ep2 = Ep1 ≡ E1 . which will enforce energy conservation and hence ﬁx the value of p1 . Comparing (7. I will complete the calculation in the center of mass system assuming equalmass particles. For the remainder of this section only I will use the notation p1  = p1 . P= p1 πp1 dΩ1 = .8. Integrating out d3 p2 gives P= 1 2 2 p dp1 δ(2Ep1 − Ei )dΩ1 4E1 1 It’s convenient to change the variable of integration from p1 to E1 . k1 = −k2 . If the particles are spinless there can be no dependence on the azimuthal .5) gives Tf i = Then (8. let’s calculate the cross section for the elastic scattering of two particles k1 +k2 → p1 +p2 from the example in the previous chapter. In this case k1  = k2  = p1  = p2 . This is why there is nothing inside the integral except dP.4.26) (8.20) gives dσ = λ2 4(2π)2 vEk1 Ek2 λ2 = 4(2π)2 vEk1 Ek2 d3 p1 d3 p2 (4) δ (Pf − Pi ) 2Ep1 2Ep2 dP (2π)6 λ 2Ek1 2Ek2 2Ep1 2Ep2 (8. 8E1 4E1 (8.
19). For this purpose it is best to leave (8.29) We also need v from (8. • Find the transition matrix Tf i from (8. but let’s go ahead and calculate the diﬀerential cross section anyhow. • The diﬀerential cross sections and decay rates are calculated from (8.31) As anticipated.27) becomes E1 p1 λ2 dΩ1 2 E 2 2p 8E 4(2π) 1 1 1 dσ λ2 = dΩ1 (16πE1 )2 dσ = 2p1 E1 (8.5 to calculate the S matrix.30) (8.32) . which in this simple case gives v= Equation (8.5 The General Case Here is a cookbook procedure for calculating scattering cross sections and decay rates for the general case in which there are N particles in the ﬁnal state. and the only way the underlying dynamics could manifest itself is by aﬀecting the diﬀerential cross section as a function of scattering angle. (2π)10 dσ = v N p∈∆Ω j=1 d3 pj δ (4) (Pf − Pi )Tf i 2 (8.19).28) in the form of a diﬀerential dP = p1 dΩ1 8E1 (8. • Use Feynman’s rules from Section 7. Our example is trivial in the sense that there is no θ dependence.21) integrated over the appropriate regions of phase space.20) and (8. • Find the relative velocity v from (8. 8. Only the scattering angle θ1 is unconstrained.5).148 CHAPTER 8. There is no θ1 dependence. CROSS SECTIONS AND DECAY RATES angle. the scattering is isotropic.
you are likely to get the wrong answer. The factors of 2π as well as the energies of the initialstate particles are constants that are gathered together outside the integrals. The Dirac wave functions used for spin. It is best to think of them as part of the phase space deﬁned by (8. Needless to say.33) When there are only two particles in the ﬁnal state.24). THE GENERAL CASE N 7 p∈∆Ω j=1 149 d3 pj δ (4) (Pf − Pi )Tf i 2 dΓ = (2π) (8. When there are more than two particles in the ﬁnal state. one integrates over all the momentum components that are constrained by the delta function and leaves the remaining two angles in the form of a diﬀerential.1 scattering are normalized diﬀerently. Tf i 2 will contain one factor of 1/(2π)3 2E for each external particle in the reaction.5. If you combine formulas from diﬀerent books.8. I’ll return to that point 2 at the end of Chapter 10.5. these rules are tailored to the Feynman’s rules from Section 7. . The rules so far only hold for spinless particles. the decision regarding which variables to integrate and which to leave in the diﬀerential depends on the experiment that the theory is describing. The energies of the ﬁnalstate particles are functions of the variables of integration. That is the meaning of the notation p ∈ ∆Ω that speciﬁes the limits of the integrals Because of our normalization conventions.
150 CHAPTER 8. CROSS SECTIONS AND DECAY RATES .
Ironically.Chapter 9 The Dirac Equation There is a story to the eﬀect that Neils Bohr once asked Dirac what he was working on. it has only positiveenergy solutions. but it does provide a formalism for treating spin 1/2 particles in a way that is consistent with relativity. (I’m talking about freeparticle equations here. but because it has only a ﬁrst derivative with respect to time.” The background to the story is this: the KleinGordon equation is invariant under Lorentz transformations. it has negative energy solutions.) Dirac thought that he could combine the two equations in a way that would preserve the best features of each.1 The Equation Let’s follow in the footsteps of our taciturn master and see what we can come up with. and (3) leads to a covariant equation of motion? The 151 . (2) is Hermitian. it has negativeenergy solutions anyway. This is inevitable with a secondorder equation. but it was known to have a strange pathology. on the other hand. “I’m trying to take the square root of the KleinGordon equation. Start with the Schrodinger equation. replied. Dirac. i ∂ψ = Hψ ∂t Can we come up with a Hamiltonian that (1) is ﬁrst order in space derivatives. who was famous for not saying much. The result is called the Dirac equation. 9. is hopeless so far as relativity is concerned. Bound state problems usually have negative total energy. The Schrodinger equation.
it must satisfy E 2 = p2 + m2 . (αi pi + βm)2 = β 2 m2 + (αi )2 (pi )2 + {αi . If E is the total energy of special relativity. . So αi is traceless. 2. Here is the argument: 1. THE DIRAC EQUATION answer is yes if ψ is a matrix. and 2. and we know that in nonrelativistic QM Pauli spinors have two components. T r[βαi β] = T r[αi β 2 ] = T r[αi ] = −T r[αi ].3) (9.1) here α and β are Hermitian matrices and p = −i .2) 2 I have used the convention that repeated indices are summed.4) We are trying to construct a theory of spin 1/2 particles. In the same way we could prove that T r[β] = 0. B} ≡ AB + BA We would like (9.2) to boil down to pi pi + m2 . {A. αj }i=j = 0 (9. lead us to conclude that the matrices must have even dimension. We will also use the replacement E = i∂/∂t. This will be possible if these matrices obey the following constraints: β 2 = (αi )2 = 1 for each i {αi . They represent the anticommutator. it is not to be. We want α and β to be Hermitian. β} mpi + 1 {αi . The curly brackets are endemic to Diracology. Take the trace of this equation. αj }i=j pi pj (9. βαi +αi β = 0 so βαi β = −αi β 2 = −αi .3) is suﬃcient to prove that their eigenvalues are ±1 3. Try i ∂ψ = Hψ = (α · p + βm)ψ ∂t (9. Arguments 1. This will be possible if the matrices have some special properties.152 CHAPTER 9. This together with (9. β} = {αi . They must be at least 4 × 4. We expect therefore that α and β will be 2 × 2 matrices. Alas.1) E2ψ = i ∂ ∂t 2 ψ = (α · p + βm) i ∂ψ ∂t = (α · p + βm)2 ψ We would like this to equal (p2 + m2 )ψ. Combine this with equation (9.
This is consistent with Zee.10) (9. nonrelativistic look about it. We must have 4 × 4. The traditional choice is β= I 0 0 −I αi = 0 σi σi 0 (9. g µν is the usual metric tensor. σ1 = 0 1 1 0 σ2 = 0 −i i 0 σ1 = 1 0 0 −1 (9. There are only three independent traceless Hermitian 2 × 2 matrices.9) This is Dirac’s equation. so 2×2 won’t do. Starting with (9. and you know what they are: the Pauli spin matrices.1) is (iγ µ ∂µ − m)ψ = 0 (9. we have −i ∂ψ † ∂ψ † =i αi + ψ † βm ∂t ∂xi (9. Gross.6) All this has a nasty. We therefore abandon Dirac’s original notation for the following:1 γ0 = β Now (9.1) and taking the Hermitian conjugate. γµ B µ = γ µ Bµ ≡ B (9. This is true of ψ as well.7) Dirac’s equation ﬁnally is (i ∂ − m)ψ = 0. Wave functions are supposed to have something to do with probability. .9.11) 1 Some authors use diﬀerent conventions. but I will not prove it at this time. Of course ψ is a fourcomponent column matrix.1. THE EQUATION 153 4. covariant.4) becomes {γ µ . It is however. It looks simple. and many other texts. but to see the connection. γ ν } = 2g µν and (9. we need four such matrices. It’s not. There is one other convenient bit of notation.5) Unfortunately. we need to derive the formula for a conserved current. the socalled Dirac spinor. This is far from obvious.8) γ i = βαi = 0 σi −σi 0 (9.
154 CHAPTER 9. the symbol E = p0 will always refer to a positive number.1). THE DIRAC EQUATION Multiply on the right by β 2 = 1 and deﬁne ψ ≡ ψ†β = ψ†γ 0 (9. It often appears in bilinear expressions in place of the usual Hermitian conjugate. multiply (9.13) on the right by ψ and add the two. We will get to the particle interpretation later. Why then are there four components to the Dirac spinor? The answer is that despite Dirac’s best intentions.16) η As a convention. the Dirac equation imposes a relationship between the two. Our job is to ﬁnd it. there are still negative energy solutions. (9.15) The current density j 0 = ψ † ψ is positive and has the same form ψ2 as the probability density in nonrelativistic QM. χ and η are twocomponent spinors. This in turn is related to the existence of antiparticles. Anyhow. They are not independent.2 Plane Wave Solutions We know that electrons have two spin components.16) into (9.13) The arrow indicates that the derivative acts to the left.14) 9. E − m −σ · p −σ · p E + m χ η =0 . For the time being we will just reconcile ourselves to negative energies. To ﬁnd the current. The result is ∂µ ψγ µ ψ + ψγ µ ∂µ ψ = 0 = ∂µ (ψγ µ ψ) So the conserved current is j µ = ψγ µ ψ (9. Start with the upper sign (positive energy) solution. the resulting equation is ∂ψ ∂ψ γi + ψm −i γ 0 = −i ∂t ∂xi or ← ψ(i ∂ +m) = 0 (9.9) on the left by ψ and (9.12) This object ψ is called the Dirac conjugate. We start with χ ψ ± = e ipx (9. Substitute (9.
s) = C 2 so our normalized spinor is χ(s) σ·p (s) E+m χ 2m .22) u(p. this means that E = ± p2 + m2 . s) = C χ(s) σ·p (s) E+m χ (9. which we knew anyhow. it will have the same value in all other frames. It is a fact that combinations like ψψ and uu are Lorentz scalars. so that if we choose the normalization in the rest frame of the particle. In order to appreciate this. s) = u(p.21) Notice that in the nonrelativistic limit.18) so that which is a useful identity to have at your disposal. We conclude that E 2 = p2 + m2 . E 2 − m2 − (σ · p)2 = 0. i. E+m s = 1. s)e−ipx where u(p. A quick calculation shows that u(p.17) (9. Remember that 2 σi = I {σi .9. s)u(p. so we are stuck with negative energy solutions. The normalization constant C has to be chosen according to some convention.2. s)u(p. We will use u(p. χ(1) = 1 0 χ(2) = 0 1 (9. (9. PLANE WAVE SOLUTIONS 155 This will only have a solution if the determinant is equal to zero.20) s = 1. s) = 1. 2 (9.e. σj }i=j = 0 (σ · p)2 = p2 . it is necessary to use one of these group theory results that I have not yet proved. the third and fourth components become negligible.19) The complete positive energy solution to the Dirac equation is written ψ + (x. 2 (9. s) = E+m 2m . At any rate η= σ·p E+m χ We can make χ into a Pauli spinor that satisﬁes χ† χ = 1. Unfortunately.
The electron would fall into a bottomless pit emitting an inﬁnite amount of energy in the process! To sidestep this diﬃculty. THE DIRAC EQUATION 1 0 u(0. 4). 2) = 0 0 (9. For one thing. 4 (9. which I will provisionally call u(p. 1) = 0 0 0 1 u(0. Dirac proposed that under normal circumstances.24) is normalized so that u(p. s) = u(p. In the original version of Dirac theory there is no lowestenergy state since there exists an continuum of negativeenergy states from −mc2 to −∞. 3) and u(p. The minus sign is forced on us.25) 9. it’s a half truth. Like all of our theories I suppose. s) = E+m 2m σ·p − E+m χ(s−2) s = 3. The states are full because no two Fermions can occupy a single state.23) Now back to the negative energy solutions. a disappointment because this wasn’t supposed to happen with a linear wave equation and an embarrassment because of its physical implications. and the KleinGordon . s)eipx s = 3. This agglomeration of ﬁlled states is called the Dirac sea. all negativeenergy states are completely ﬁlled. s) = −1. Unfortunately. s)u(p. Let us consider an atomic electron. it depends on the Pauli exclusion principle.16) brings us to σ·p χ=− η E+m and u(p. the other half is very confusing. The complete negativeenergy solution is ψ − (x. Repeating the above arguments with the lower sign in (9.3 Charge Conjugation and Antiparticles The appearance of negativeenergy states was both an embarrassment and a disappointment. 4 (9. This is not true for integer spin particles.156 so that in the rest frame CHAPTER 9. An excited electron will normally lose energy by emitting photons until it drops down to the lowestenergy or ground state.24) χ(s−2) It is easy to see that in the rest frame (and so everywhere else) (9.
no such particle was known. This is all well and good. Advanced Quantum Mechanics.) When Dirac ﬁrst proposed this in 1930. According to this theory. We have to take it on faith that these inﬁnities are not observable. by J.2 Suppose one of the negativeenergy electrons were to absorb a photon of energy > 2mc2 and become an E > 0 state. all the additive quantum numbers sum to zero. The absence of momentum p in the Dirac sea appears as the presence of momentum −p. Put them together. (It can’t happen in empty space because of energy and momentum conservation. but it will continue to devil us long after we have abandoned the Dirac sea.93.26) The observable charge is then Qobs = Q − Q0 = e. In this way we expect that the absence of a negativeenergy electron appears as the presence of a positiveenergy particle. but it contains a surprising amount of physical insight.10 . Sakurai. It does happen whenever high energy photons pass through matter. then the total charge of the Dirac sea is Q = Q0 − e = Q0 − (−e) = Q0 + e (9. Sections 3. This process actually happens. The observable energy of the Dirac sea is now the energy of the vacuum minus the negative energy of the vacated state. J. As a result a “hole” is created in the Dirac sea. and what you have is – nothing”! Put less cryptically. In the laboratory it looks like γ → e− + e+ . Let Q0 represent the charge of the vacuum and e the negative charge of an electron in the sea. We need to describe antiparticles in a way that does not invoke negative energies. the total energy of the vacuum as well as the total charge is negative inﬁnity. The same argument works with charge. For this reason it was assumed that the theory couldn’t be right.3. The positron was discovered soon thereafter. but we must face the fact that the Dirac sea is really just a metaphor. What is observable are the deviations from inﬁnity corresponding to the energy and charge of a single electron! This is an unsatisfactory aspect of the theory of course. Obviously the Dirac sea can’t be the whole answer. The argument made above regarding charge and energy also works for momentum and spin. AddisonWesley 1967.9. however. This is interpreted as a particle with positive energy and charge. CHARGE CONJUGATION AND ANTIPARTICLES 157 equation also has negative energy states. Similarly the absence of a spinup E < 0 electron is to be interpreted as the presence of a spindown E > 0 positron. What are the 2 The best discussion of this entire subject is still the classic. I once heard Edward Teller explain antiparticles like this: “Take a particle in one hand and an antiparticle in the other. hence a positive quantity.
THE DIRAC EQUATION additive quantum numbers? Well.27). Equation (9. First.158 CHAPTER 9. a 4 × 4 matrix that operates on the space of the γ’s.27) Charge conjugation must change the charge of the electron so that Teller’s dictum comes true. Take the complex conjugate of (9. This rule is based on the idea of gauge invariance.29) At this point we don’t know what C0 is. Electrons have zero baryon number and strangeness anyhow. pz p− σ·p= (9.28) and multiply by C0 . For this we will need C0 γ µ∗ C0 = −γ µ and C0 ψc = ψ.30) is valid only in the the standard or DiracPauli representation we are using. baryon number. We can deﬁne a transformation called “charge conjugation. and strangeness for starters. Let’s concentrate on the matter of charge. I will come to the issue of lepton number presently. The following identities are useful. The transformed equation is (i∂ − e A − m)ψc = 0. ∗ C0 (iγ µ∗ ∂µ + eγ µ∗ Aµ + m)ψc −1 ∗ = (C0 γ µ∗ C0 )(i∂µ + eAµ ) + m C0 ψc = 0 (9. For the time being – just believe it.28) Here ψc is the new “charge conjugated” wave function. it replaces all particles with their antiparticles.31) p+ −pz There are several diﬀerent representations of the gamma matrices that are useful in other contexts.” The rule is everywhere you see the derivative ∂µ simply replace it with ∂µ → Dµ = ∂µ + iqAµ where q is the charge of the particle. which we will discuss in the next chapter. (The i is necessary to make C0 Hermitian. The bottom line is ψc = C0 ψ ∗ = iγ 2 ψ ∗ (9. (9.30) Let us see what this operation does to our prefab spinors. With this replacement the Dirac equation for an electron with charge −e (e > 0) becomes (i∂ + e A − m)ψ = 0 (9.28) −1 ∗ looking like (9. it changes the wave function in such a way that it correctly describes the new situation. 3 . Try C0 = iγ 2 . It’s easy to modify the Dirac equation to include the electromagnetic potential. charge.” It has two eﬀects. It doesn’t take much trial and error to realize that all the γ’s are pure real except γ 2 . three kinds of lepton number. The challenge is to get (9. and second.)3 You −1 2 can check that C0 = I and C0 γ µ∗ C0 = −γ µ as promised. The procedure is called “minimal substitution.
35) (9. (9. Using (9. 4)eipx 0 2m −1 0 0 1 0 In the same way one can prove that + ψc (x. 3)eipx (9.38) Thus the chargeconjugate wave function obtained from the positiveenergy planewave solution is the wave function for a negativeenergy plane wave . 1)e−ipx = pz /(E + m) e 2m p+ /(E + m) 0 −ipx E+m 1 e ψ + (x.9.3. 2) = u(p.37) −p− /(E + m) E + m pz /(E + m) +ipx e = −u(−p. 1).33) gives 0 0 + ψc (x. 3) = u(p. 3)eipx = 1 2m 0 −p− /(E + m) E + m pz /(E + m) +ipx e ψ − (x.30). 4)eipx = 0 2m 1 0 0 C0 = 0 −1 0 0 1 0 159 (9.34) (9.32) (9. 2)e−ipx = 2m p− /(E + m) −pz /(E + m) −pz /(+m) E + m −p+ /(E + m) +ipx e ψ − (x. CHARGE CONJUGATION AND ANTIPARTICLES 0 −1 1 0 0 0 0 0 1 −ipx E+m 0 ψ + (x.32). and (9.33) (9.36) Let’s ﬁnd the chargeconjugated version of ψ + (x. 2) = u(−p. 1) = u(p. 1) = 0 −1 = ∗ 0 −1 1 +ipx 1 0 E+m 0 pz /(E + m) e 0 0 2m 0 0 p+ /(E + m) (9. 4) = u(p.
while v(p. In order to do that. it’s still the wave function of a ﬁctitious negative energy particle.42) u(p. s) = C0 v(p. the rest is logically consistent.40) 2 where I have used the fact that C0 is Hermitian.15). s)∗ = v(p. The spin direction is also reversed since the spinup spinor with s = 1 is changed into the spindown spinor with s = 4. Q=e j 0 d3 x = e ψ † ψd3 x (9. 4) are properly called negativeenergy spinors. 2 s = 1. it is customary to deﬁne spinors for real positrons as follows. s) ∗ (9.160 CHAPTER 9. the observable behavior of the chargeconjugate wave function should be a positron with E > 0 and the same momentum and spin.4 In order to get sensible expectation values we need to introduce second quantization with creation operators that create real positrons. it’s the wave function that is interpreted as the positron wave function. 2) are positron spinors. 1) ≡ −u(−p. . s) = C0 u(p. It’s trivial to show that the expectation values of the momentum and spin do change sign.41) s = 1. If you buy into this bizarre notion of the Dirac sea. Under the skin. 4 Again – thanks to Sakurai for ﬁnally making me understand this point. since Qc = e † ψc ψc d3 x = e (C0 ψ ∗ )† (C0 ψ ∗ )d3 x = e (ψ † ψ)∗ d3 x = Q (9. 1) and v(p. According to Dirac’s hypothesis. THE DIRAC EQUATION with the same E and opposite momentum. 2) ≡ +u(−p. 3) With this deﬁnition uc (p. 4) v(p. 2 (9. Try for example. s) = u(p. to calculate the expectation value of the charge of a positron. and currents are real functions. The electromagnetic current should be proportional to the zeroth component of the probability current (9. 3) and u(p. C0 = I.39) The chargeconjugated wave functions give the same value for Q. but expectation values of the chargeconjugated wave functions do not give the results you would like in terms of antiparticles. s) vc (p. This is consequence of the absence of an electron with negative charge and energy and reversed momentum and spin. v(p. The reason is that ψc is not the positron wave function.
s ) = δss u† (p. 9.60) with the following diﬀerences: . s)e−ipx ψ − (x) = v(p.50) u(p. You can verify them by substituting the representations (9.4 Quantizing the Field We repeat the procedure from Chapter 5 used to quantize the KlineGordon ﬁeld. (9.4.44) (9. and (9. s ) = v † (p. s)v(p. which can be taken as L = ψ(x)(i ∂ − m)ψ(x) The conjugate momentum is π= and the Hamiltonian is H= ˙ d3 x (π ψ − L) = d3 x ψiγ 0 ∂0 ψ (9. s)u(p. s)v(p. s)e−ipx and v(p.48) (9. s)v(p.43) (9. This should look like (6.51) Next we will need the plane wave solutions of the Dirac equation.7). We then expand ψ(x) and π(x) in a Fourier series in which creation and annihilation operators appear as Fourier coeﬃcients. s)v(p.49) (9. First we’ll need a Lagrangian. s) = s These equations will all be useful later on. s ) = 0 u† (−p.45) (9. s)u(p. These are u(p.22). s ) = 0 u(p.47) (9. s)u(−p.53) ∂L = iψ † ˙ ∂ψ (9. s)e+ipx u(p.9. s)v(p. s ) = −v(p. s) = s p+m 2m p−m 2m v(p.52) (9.46) (9. s)eipx . s ) = v(p.24). s ) = E δss m 161 (9. s)u(p. QUANTIZING THE FIELD Here is a summary of the important equations involving u and v: ψ + (x) = u(p. s ) = v † (p. s)u(p.
s)eipx ] (9. (9.54) The analogs to (6. Unfortunately. Put it another way. this leads to some wildly unphysical consequences. ψ † ] doesn’t make any sense. ψβ (y. We must have † ψα (x. each commutator is a 4 × 4 matrix with rows and columns indexed by α and β. ψβ (y. ψβ (y. ψβ (y. we must also sum over the spin components. The commutators in (9. i. s)u(p. ψβ (y. s)e † ipx + d(p. ψβ (y.e. no two particles can occupy the same quantum state.69) we would expect the equaltime commutation relations to be † [ψα (x. THE DIRAC EQUATION 1. The origin of the problem is that electrons are fermions. t). s)e−ipx + d† (p. s) ( Annihilates a positive energy electron Creates a positive energy electron Annihilates a positive energy positron Creates a positive energy positron (9. t). so [ψ. t). particles with halfodd integer spin. u(p. s)u(p. s) b† (p. Mathematically. Be alert to this ambiguity in other contexts.60) are ψ(x) = ψ(x) = m E m E d3 p (2π)3 d3 p (2π)3 s s [b(p. one set for electrons and one for positrons. whereas ψ † ψ is a 1 × 1 matrix. ψψ † is a 4 × 4 matrix.162 CHAPTER 9. t). t) = δαβ δ 3 (x − y) † † {ψα (x. t)} = ψα (x. In the usual notation b(p. s) depend on spin. x) and v(p. s)v(p.55) [b (p. We need two kinds of creation and annihilation operators. ψ is a fourcomponent spinor rather than a scalar like ϕ. Such particles obey the Fermi exclusion principle.56) Pay careful attention to the notation. t). t)] = 0.56) refer to the αth component of one ψ matrix with the βth component of another. must the ﬁelds themselves. so in addition to integrating over all momentum components. and so as a consequence. this means that creation and annihilation operators anticommute. s)e −ipx ] In analogy with (6. s) d(p. t)] = [ψα (x. t)] = δαβ δ 3 (x − y) † † [ψα (x. s)v(p. t) = 0. 3. s) d† p. 2. t). .
and I are consistent. d† (p s ) = δss δ 3 (p − p ) All other combinations of b and d anticommute. The combination †T has the same eﬀect as ∗ on ˆ matrices and the same eﬀect as † on ˆ and d. equations (9. s)e−ipx + b† (p. At least Zee.) Now using (9. s). (9. s) ↔ d† (p.30) for the quantized ﬁelds. s)v(p.59) [d (p. Only (9.56) would come out right.55) was chosen so that (9. Michio Kaku (Quantum Field Theory.24). s)v(p. (9. (9.58) The point is that complex conjugation is not deﬁned for the creation and annihilation operators. This point will be central importance in the case of the weak interactions. s)e † ipx + b(p. The factor of (9. It is only a slight exaggeration to say that no two books do it the same way.56) assuming the following anticommutation relations b(p. One is the normalization of u and v. s) and b† (p.56) is nonnegotiable. . They are implied in the other operator equations in this section.57). b† (p . The second is the abovementioned factor in the Fourier expansion of ψ.22) and (9. (I have put the hats back on b the operators for emphasis. s) with no change in spin. s)u(p. QUANTIZING THE FIELD 163 You should verify that (9. ˆ Deﬁne the charge conjugation operator C by ˆ ˆ ˆ ˆˆˆ ψc ≡ C ψ C −1 = C0 ψ †T = iγ 2 ψ †T (9.4. s)u(p. b(p.9.55) is consistent with (9. A Modern Introduction). s)e −ipx ] The only change is that particle operators have been replaced by antiparticles operators and vice versa. s ) = δss δ 3 (p − p ) d(p. s). s)eipx ] (9. The third is the normalization of the creation and annihilation operators.29).57) m/E in There are at least three places where we have to make a somewhat arbitrary choice of normalization. s) ↔ d(p. It’s easy to generalize charge conjugation (9.55) becomes ψ(x) = ψ(x) = m E m E d3 p (2π)3 d3 p (2π)3 s s [d(p.
64) Where N + and N − are the number operators for positive and negative energy electrons respectively. s) + d(p. the Hamiltonian must be normal ordered to avoid the inﬁnite constant. s)b(p.164 CHAPTER 9. : (a + b† )(c + d† ) := ac − d† a + b† c + b† d† Take another look at the total charge calculated in (9. Again – as an exercise in Diracology. THE DIRAC EQUATION Now insert the Fourier decomposition back into the expression for the Hamiltonian.45). The usual Dirac business gives us Q=e d3 p s [b† (p. . With the anticommutation rules we have : H := d3 pE s [b† (p. : Q := e d3 p s [N − (p. s)d(p. it is necessary to include the minus sign arising from the anticommutators. s)d(p. s)] (9. if a. H= d3 pE s [b† (p. c. s) + d† (p. and d are fermion annihilation operators. s)b(p. s)d† (p. s)b(p.62) where e is the electron charge. you should verify that the following is obtained. If d and d† commuted. s)] (9. s)d† (p.61) As in (6. equation (9. a negative number. s). s) − N + (p.39) Q=e d3 x ψ † (x)ψ(x) (9. s)] (9. so by creating many particles with d† we could create states of arbitrarily large negative energy. s)] (9.60) Here is the ﬁrst of many places where commutation relations lead to unphysical results. the last term could be written −d† (p. s) − d(p. For example.63) This operator also has to be normal ordered. b. When dealing with fermions.53).
Λi i = 1. 3.65) (9.6 Consider the basic Lorentz transformation x = γ(x − vt) 5 (9.66) y =y t = γ(t − vx/c2 ) (9. There are three ways that a transformation might be not proper. Λi i = −1. Here is a brief review of the Lorentz group.9. It is easy to show from (9. The basic equation of Lorentz transformation is x µ = Λµν xν Λµν is any 4 × 4 real matrix that satisﬁes gµν = Λσµ gσγ Λγ ν . or in matrix notation g = ΛT gΛ.66) that this is improper. TP=PT. T.5 All this is preliminary to analyzing the proper Lorentz group and its associated Lie algebra.5. THE LORENTZ GROUP 165 9. (9. 1.5 The Lorentz Group We can’t get much further with the Dirac equation without using some group theory. This simply makes time run backwards. and so it is called the time reversal transformation or T for short. We call it a parity transformation and refer to it as P. This reverses the spatial coordinates. 6 I am following a particularly ﬁne exposition in Quantum Field Theory by Michio Kaku. and all oﬀdiagonal terms equal to zero.67) It is easy to show that this implies det Λ = ±1. The most general Lorentz transformation consists of a product of a proper transformation and P. These are examples of discrete transformations. Any transformation that can be obtained from this by a continuous variation of the transformation parameters (β for example) is called a proper transformation.68) Later on we will use P and T to represent more general operators that have the eﬀect of reversing x or t. and T 2 = P 2 = I. 2. . Λ0 0 = 1. and all oﬀdiagonal terms equal to zero. The identity transformation is given by Λ = I. We could have a product of the two. or PT. Try Λ0 0 = −1.
We usually write the above as K 1 ≡ −i and Λ(ζ1 ) = eiK ∂ Λ(ζ1 ) ∂ζ1 1ζ 1 (9. According to the doctrine of Lie groups.72) it out will appear presently. 0 0 0 0 The reason for putting in the i and then taking With this convention 0 −1 0 −1 0 0 K 1 = −i 0 0 0 0 0 0 (9.70) Transformations of this sort are called boosts and the angle ζ1 that parameterizes the boost in (9.69) then this transformation can be written as: 0 cosh ζ1 − sinh ζ1 x x 1 − sinh ζ1 cosh ζ1 x2 = 0 0 0 0 x3 0 0 1 0 0 x 0 x1 0 0 x2 x3 1 (9.166 CHAPTER 9.65). we can calculate the group generators by diﬀerentiating Λ with respect to its parameters and then setting the parameters equal to zero.71) ζ1 =0 (9. The group can then be “reconstituted” by exponentiating the generators. K1 ≡ ∂ Λ(ζ1 ) ∂ζ1 (provisional deﬁnition) ζ1 =0 The K’s obtained in this way are constant matrices called the “generators” of the transformation. 1 Λ(ζ1 ) = eK ζ1 At this point physicists often muddy the water by inserting i’s in ways that appal the mathematicians.73) . THE DIRAC EQUATION If we make the standard replacement γ= 1 1 − v 2 /c2 = cosh ζ1 βγ = sinh ζ1 β = v/c (9.70) is called the rapidity. In matrix notation x = Λx. The 4 × 4 matrix above represents the tensor Λµν in (9.69) and (9.
7 . K j ] = i 7 ijk k J J i j ijk k ijk (9. The boost generators by contrast. the rotation group is a subgroup of the Lorentz group. The corresponding rotation matrices Γ(θ) = eiJ·θ (9.77) Notice that the J i ’s are antisymmetric and because of the i’s. THE LORENTZ GROUP The generators of boosts in the y and z directions 0 0 1 0 0 0 0 0 0 0 K2 = i K3 = i 1 0 0 0 0 0 0 0 0 1 are 0 0 0 0 0 0 0 0 1 0 0 0 167 (9.76) (9.78) are unitary. K j ] = −i [J . or to put it another way. a rotation through an angle θ1 around the 1 or x axis would yield 1 0 0 0 0 1 0 0 (9. are symmetric and boost matrices are not unitary. Sec.9. The generators have the following commutation relations:7 [K i . The usual parameters are a set of three angles.74) Rotations are also Lorentz transformations.5. 11.75) Λ(θ1 ) = 0 0 cos θ1 sin θ1 0 0 − sin θ1 cos θ1 The corresponding 0 0 J1 = i 0 0 generators are 0 0 0 0 0 0 0 0 1 0 −1 0 0 1 0 0 J2 0 0 0 0 J3 = i 0 1 0 0 0 0 0 0 = i 0 0 0 −1 0 0 −1 0 0 0 0 0 0 0 0 0 (9. For example. J ] = i [J i . Hermitian.79) Kk This is also worked out in Jackson. This is why physicists (who are fond of unitarity) put in the i’s and mathematicians (who would like to keep the structure constants real) do not.
e.82) constitute a matrix representation.80) into (9.84) The other is the spinor representation. Consider inﬁnitesimal Lorentz transformations of the form µ Λµν = δν + M µν (9. and inﬁnitesimal. a member of the algebra.83) also constitutes a valid representation of the Lie algebra.81) This is equivalent in matrix notation to the statement that M (with both indices down) is an antisymmetric matrix. Mµν + Mνµ = 0 (9. Substitute (9. µν is the name of the matrix and α and β index it’s elements. M ρσ ] = i(g νρ M µσ − g µρ M νσ − g νσ M µρ + g µσ M νρ ) (9. One is the diﬀerential operator M µν = i(xµ ∂ ν − xν ∂ µ ).82). but any set of objects that satisfy the commutation relations (9. A particularly useful way of choosing such matrices is µ ν µ ν (M µν )αβ = i(δα δβ − δβ δα ) (9. The matrix named M µν contains all zeros except for two elements. It is easy to check that the maximum number of independent 4 × 4 antisymmetric matrices is six and that the commutator of two antisymmetric matrices is itself antisymmetric. One can derive the following commutation relations: [M µν .82) Just so there’s no confusion.79) must be put in a more convenient form. they operate on fourvectors.80) with M µν a constant matrix.168 CHAPTER 9. The element that sits on the µ’th row and ν’th column is i and the element at the ν’th row and µ’th column is −i. Referring again to (9. There are two other representations that are important. (9. the algebra of the Lorentz group. which I will come to shortly.66) and discard the secondorder terms. there are twelve nonzero 4 × 4 antisymmetric matrices related in a simple way to the K’s and J’s deﬁned previously.83) The M ’s deﬁned by (9. That is to say. It’s easy to remember. We are looking for a representation that transforms Dirac spinors. THE DIRAC EQUATION The matrices above constitute the deﬁning representations of the group and the algebra. i. Thus this set of matrices constitutes a closed Lie algebra. To this end we start by proving a simple theorem about the symmetry properties of the generators. and for this purpose the commutation relations (9. It’s . however.
86). This can also be written S(Λ) to remind ourselves that given the parameters ωµν .90) .85) to build up ﬁnite transformations as usual.83) and hence constitute a representation of the Lorentz group. SPINOR REPRESENTATIONS 169 easy to verify that gM 01 = K 1 .9. but both the point and the components of the ﬁeld are referred to a diﬀerent set of axis. (9. The Dirac spinor ψ(x). σ ρσ ] = (M ρσ )µν γ ν 2 where M ρσ is deﬁned by (9.86) 9. We can iterate (9. The operator i S(ω) ≡ exp − ωµν σ µν 4 (9. I take the passive point of view.85) 2 where ωµν is an inﬁnitesimal antisymmetric tensor. we should think about what it means to transform a vector ﬁeld. like the EM vector potential Aµ (x). Before we apply all this to the Dirac equation. gM 12 = J 3 .8) that the object 1 σ µν satisﬁes the 2 commutation relations (9.82). Inﬁnitesimal transformations are written i α Λαβ (ω) = δβ − ωµν (M µν )αβ .88) is exactly analogous to (9. Deﬁne i σ µν ≡ [γ µ .6 Spinor Representations The anticommutation relations (9. γ ν ] 2 (9. etc. It is also straightforward to show that 1 (9.) These algebras can always be used to deﬁne a new sort of representation. so S(Λ)ψ(x) = ψ (x ).89) [γ µ . we can calculate the deﬁning representation Λµν as well as the spinor representation S. (This works in any even dimension.6. describes some multicomponent property (of the ether. if you like) at the spacetime point x.8) deﬁne what is called a Cliﬀord algebra. (9.87) It is straightforward to show using (9. that a Lorentz transformed ﬁeld describes the same phenomena at the same spacetime point. i Λ(ω) = exp − ωµν M µν 2 (9.
The right side treats γ µ like a fourvector. Thus (9.94) The last line makes use of (9. I’ll leave it as an exercise.94) is also the inﬁnitesimal form of the right side of (9.91) if S −1 (Λ)γ µ S(Λ) = Λµν γ ν (9.90) and pµ = Λµν pν into (9. The left side is a similarity transformation.) This requirement leads to an important equation for S(Λ). It can be proved for ﬁnite transformations by iterating inﬁnitesimal transformations. A glance at (9. Substitute (9. The left side of (9.93) (9. S −1 (Λ)(γ µ Λµν pν − m)S(Λ)ψ(x) = 0 This will be identical to (9.94) is the inﬁnitesimal form of (9. σ ρσ ]) 4 4 2 2 i =(1 − ωρσ M ρσ )µν γ ν 2 (9.93) is true for inﬁnitesimal transformations i i i 1 (1 + ωρσ σ ρσ )γ µ (1 − ωρσ σ ρσ ) = (γ µ − ωρσ [γ µ . THE DIRAC EQUATION x = Λx as usual. ψ † (x )γ 0 = ψ † (x)γ 0 γ 0 S † (Λ)γ 0 Remember that (γ 0 )2 = 1 and (γ 0 )† = γ 0 . ψ (x ) = ψ(x)γ 0 S(Λ)† γ 0 .90) and right multiply by γ 0 .91) This is interesting. it must remain form invariant under this transformation.93). is yes! Or at least we hope so.86) shows that the last line of (9. We are now in a position to prove the claim made earlier that ψψ is a Lorentz scalar.170 CHAPTER 9.93) is valid for inﬁnitesimal transformations. So is γ µ a matrix like S(Λ) or a fourvector like pµ ? The answer of course.92). If the Dirac equation is to make any sense.89). It’s not a new argument. This is how one transforms a matrix like γ µ . so that we must have both (γ µ pµ − m)ψ(x) = 0 and (γ µ pµ − m)ψ (x ) = 0 (9.93). Here its proof: The ﬁrst step is to prove that (9. The covariance of the Dirac equation stands or falls depending on (9. First take the Hermitian conjugate of (9.93) by construction.92) (Here pµ = −i∂µ . and the new ﬁeld components ψ will be determined by S(Λ). but it’s very tedious to write out.
. By the same token.99) (9.87) and (9.93). we should enumerate the various possibilities.101) There are sixteen numbers in a 4 × 4 matrix. so there should be sixteen linearly independent 4 × 4 matrices with welldeﬁned transformation properties.88). ψ (x )ψ (x ) = ψ(x)ψ(x).100) (9. Since these forms are essential in describing interactions.98) (γ 5 )2 = 1 γ5. SPINOR REPRESENTATIONS = ψ(x)S −1 (Λ) 171 (9. ψ (x )γ µ ψ (x ) = ψ(x)S −1 γ µ Sψ = ψ(x)Λµν γ ν ψ(x) (9. We need to deﬁne a new gamma matrix: γ 5 = γ5 = iγ 0 γ 1 γ 2 γ 3 (9. γµ = 0 With our conventions γ5 = 0 1 1 0 (9.95) The last line follows from (9. Notice that unlike ψ and γ µ . 1 scalar 1 γµ vector 4 σ µν tensor 6 γ µ γ 5 pseudovector 4 γ5 pseudoscalar 1 for a total of sixteen. I will return to this important subject later. We can also show that ψσ µν ψ is a secondrank antisymmetric tensor. The terms pseudotensor and pseudoscalar are “pseudo” in the sense that they change sign under parity conjugation. Expressions like ψψ and ψγ µ ψ are called Dirac bilinear forms. they are not matrices but ordinary tensors. Evidentally.6. Here they are. (γ 5 )† = γ 5 (9. Any product of gamma matrices can be reduced to a linear combination of these terms.97) The following properties can be veriﬁed with the help of the anticommutation relations. ψγ µ ψ is a genuine vector. One can show that these sixteen matrices are indeed linearly independent.9.96) The last step used (9.
8 Therefore. it reduces the charge. The argument can be made in a completely covariant way. Zee 8 “Yesterday upon a stair. x) = 0ψ(x )ψ(x)0 θ(t − t) − 0ψ(x)ψ(x )0 θ(t − t ) (9. This emission and reabsorption might take place because of sources.172 CHAPTER 9.102) So why subtract the two terms rather than adding them? The glib answer is that the spinor ﬁelds anticommute.104) SF is a 4 × 4 matrix indexed with α and β. because we cannot annihilate the electron before it has been created. Just as we did in the case of the KlineGordon equation. however.103) . ψ doesn’t just create electrons.” ψ(x )ψ(x) t >t −ψ(x)ψ(x ) t > t (9.55) into (9. so ﬁnally iSF (x .50) to do the spin sums. We get a covariant result in the end anyway.102) and use (9. x) = 0T [ψα (x )ψ β (x)]0 (9. the amplitude for this process is called the propagator. so the product 0ψ(x )ψ(x)0 must have something to do with the propagator. I have argued before that the ﬁeld ψ(x) creates an electron at x and ψ destroys it. THE DIRAC EQUATION 9. / He wasn’t there again today. In any case. eq. This looks noncovariant. we introduce the time ordering operator T [ψ(x )ψ(x)] = Finally i(SF )αβ (x . we should write out propagator as 0ψ(x )ψ(x)0 θ(t − t). but it is logically impossible in any reference frame to make something disappear that doesn’t exist. We must still insist that the positron be created before it is destroyed. (5. This is still not the complete propagator. It’s easy to calculate.49) and (9. / I met a man who wasn’t there.33). It is the basic ingredient in all perturbation theory calculations. or it might come about because the electron interacted with other particles at x and x . / I wish that man would go away. So 0ψ(x )ψ(x)0 also include the process in which negative energy electrons travel backwards in time from x to x . which we discussed in connection with path integrals.7 The Dirac Propagator Consider a physical process in which an electron is created out of the vacuum at the point x and is subsequently reabsorbed at x . simultaneity is relative. but this is just pedantry. We will get more insight into this presently. because there is a distinct process that is physically equivalent. Substitute (9. or equivalently. positrons travel from x to x. True. because or the θ(t −t). This can’t be the whole story.
and boson ﬁelds must commute. We could say that the factors of ( p + m) in the numerator and denominator “cancel” leaving the rather whimsical equation. but electrons don’t interact with themselves like our hypothetical scalar particle.9 iS(x) = In momentum space this is iS(p) = i p−m+i (9.107) i d4 p −ipx e . the propagator violates causality. They do interact with the electromagnetic vector potential and with the potential that carries the weak interactions. Your instructor will become homicidal if you divide by a matrix under any other circumstances. At this point in our treatment of the scalar ﬁeld. An Introduction to Quantum Field Theory page 56 9 . He arrives at the following: d4 p −ipx p+m iS(x) = i e (9. called the spinstatistics theorem. to which we turn in the next chapter. It also reproduces the minus sign in (9. This is an illustration of a deep result. (2π)4 p−m+i (9. 10 Peskin and Schroeder. It can also be shown10 that without that minus sign.) Our next order of business is the electromagnetic ﬁeld. THE DIRAC PROPAGATOR 173 goes through this calculation in detail on pages 108109. but that’s a story for another time.7.105) (2π)4 p2 − m2 + i It turns out that ( p − m)( p + m) = p2 − m2 .102).9. Fermion ﬁelds must anticommute. You can show that (9. It says that in order to preserve causality.105) does have that property.106) Causality requires that the propagator vanishes outside the light cone. (The’re the same ﬁeld actually. we added some interactions and did perturbation theory.
THE DIRAC EQUATION .174 CHAPTER 9.
(3) We set c = 1 as usual.) There are no dielectrics or magnetic materials. ∂ρ + ∂t ·j =0 (10.) In this system. Nothing about them needs to be changed. It arises because: (1) We use HeavisideLorentz units. but we don’t care about that.Chapter 10 The Photon Field 10. (See the appendix in Jackson regarding units. (They retaliate by showing up in Coulomb’s law.2) The vector ﬁelds can be derived from potentials. The electromagnetic current is conserved. so = µ = 1.3) .1 Maxwell’s Equations In some sense Maxwell discovered both quantum electrodynamics and relativity. E = − A0 − ∂A ∂t 175 B= ×A (10. all the factors of 4π are absent from Maxwell’s equation.1) You may never have seen them in this simple form. I mean by this odd remark that Maxwell’s equations are consistent with both. Here they are: ·E =ρ ∂E =j ×B− ∂t ·B =0 ∂B ×E+ =0 ∂t (10.
j) Fµν = ∂µ Aν − ∂ν Aµ Then current conservation is just ∂µ j µ = 0 and Maxwell’s equations can be summarized as: ∂µ F µν =j ν ∂µ Fνρ + ∂ν Fρµ +∂ρ Fµν = 0 (10.8) which is exactly (10.7) To show that this does indeed generate the right equations of motion. Jackson has some obscure things to say about Lagrangians in general. A) j µ = (ρ. it’s really very easy.4) The fact that these can be written in covariant form indicates that relativity is “built in” to Maxwell’s equations. ∂Aν (10.6).6) (10. 1 L = − Fµν F µν − jµ Aµ 4 (10.176 CHAPTER 10. but if you know the answer ahead of time. We need a Lagrangian to do ﬁeld theory. The potential contains some unphysical information in the sense that Maxwell’s equations are invariant under gauge transformations. simplify the expression as follows: 1 1 − Fµν F µν = − (∂µ Aν − ∂ν Aµ )(∂ µ Aν − ∂ ν Aµ ) 4 4 1 = − [(∂µ Aν )(∂ µ Aν ) − (∂µ Aν )(∂ ν Aµ )] 2 ∂L = −∂ µ Aν + ∂ ν Aµ = −F µν ∂(∂µ Aν ) The EulerLagrange equations are ∂µ ∂L ∂(∂µ Aν ) − ∂L = −∂µ F µν + j ν = 0. The familiar form for gauge transformation is A→A+ Λ .5) (10. THE PHOTON FIELD This can all be put in covariant notation by deﬁning the following tensors: Aµ = (A0 .
∂µ Aµ = 0 (10.1 For example. Only the transverse components are meaningful. Aµ (x) → Aµ (x) − ∂ µ Λ(x) (10.” This particular constraint deﬁnes what is called temporal gauge. and hence preserves Maxwell’s equations.9) This leaves the ﬁeld tensor invariant F µν (x) → F µν (x). This poses a dilemma: we can’t fully ﬁx the gauge without making the equations noncovariant. We can’t make the theory covariant without building into the theory unphysical degrees of freedom known as FadeevPopov ghosts. but it does not fully ﬁx the gauge. Imposing constraints like this is called “gauge ﬁxing. Here I am quoting the main results without proof. it cannot depend on the longitudinal components. Another choice is Lorentz gauge.10. 2 Where is Buﬀy now that we need her? 3 Peskin and Schroder 1 .11) This has the advantage of being covariant. Since F ij is basically the curl of A. There are only two This subject is discussed extensively in Jackson and our Ph632 class. A = AT + AL where by deﬁnition × AL = 0 · AT = 0 Furthermore. We can use this symmetry to make Aµ satisfy certain relations.1. We can write this in covariant form as follows. the decomposition is unique up to an additive constant. MAXWELL’S EQUATIONS φ→φ− 177 ∂Λ ∂t where Λ(x) is any arbitrary function of space and time. It is possible to impose an additional constraint ·A=0 (10.10) This is known as Coulomb gauge or radiation gauge.2 The subject is so diﬃcult and convoluted that one distinguished text3 simply refuses to discuss it! Components of Aµ that can be set equal to zero by a gauge transformation cannot have any physical signiﬁcance. we can choose Λ so that A0 (x) = 0. It’s a wellknown result from classical ﬁeld theory that any vector can be decomposed into a longitudinal and a transverse part.
The Lagrangian density for the free ﬁelds is 1 1 L(x) = − F µν Fµν = (E 2 − B 2 ). Hatﬁeld. t)] = iδαβ δ(x − y) [Aα (x. This is my excuse for spending minimal eﬀort on this vexatious subject.15) . and second.12) ∂L = −Ei ∂(∂0 Ai ) The next step is to impose the equaltime commutation relations. πβ (y. 4 (10. It’s ironic that the photon propagator. 5 I am following B. π0 (x) = πi (x) = and the Hamiltonian density is 1 ˙ H = πµ Aµ − L = (E 2 + B 2 ) + E · 2 A0 (10. since π0 = 0. THE PHOTON FIELD physical degrees of freedom.5 First. D. t)] = 0 We are missing one relation. This is an invitation to duck into coulomb gauge and set A0 = · A = 0. We are not out of the woods yet. Drell.178 CHAPTER 10. Perhaps the easiest way out of this mess is to quantize in Coulomb gauge. 4 2 The conjugate momenta to Aµ are.) [Aα (x. Aβ (y. ﬁnd the photon propagator. Quantum Field Theory of Point Particles and Strings. t)] = [πα (x. 10. Bjorken and S. t).13) (10. Relativistic Quantum Mechanics and Relativistic Quantum Fields is the Gold Standard.14) ∂L =0 ∂(∂0 A0 ) (10. t). t).2 Quantization in the Coulomb Gauge We have two goals. πβ (y. D. though even this is not straightforward. is very simple. This is the approach taken in the previous generation of texts. J. to ﬁnd an expansion of the photon ﬁeld in terms of creation and annihilation operators.4 Modern texts do this with path integrals. which is all you need to do perturbation theory in QED. (Compare with (73) in the previous chapter. which correspond to the two polarization states of an electromagnetic wave.
16) tr tr δij (x) has the desired property: ∂i δij (x) = 0. A(x) = d3 k (2π)3 2k0 2 (10. We modify (16) to read tr [Ai (x. Suppose we modify (10. Now we are home safe. • Electric and magnetic ﬁelds calculated using (10. tr δij (x − y) ≡ ki kj d4 k ik·(x−y) e δij − 2 3 (2π) k . (10.10. Eq. t). Lets replace the δfunction with a function f (x−y) such that ∂i f (x−y) = 0. we get the same commutators for E and B regardless whether we use the transverse or ordinary δfunction in (10. we get the odd result 0 = −i∂i δ(x − y) This is time for some creative thinking. so far so good. t)] = −iδij δ(x − y).18). It remains only to expand A in plane waves and calculate the propagator.18) do commute at spacelike separations. (10.3) gives · E = 0. and in fact.15) [Ai (x. Since we are working in transverse gauge. This is Gauss’s law. QUANTIZATION IN THE COULOMB GAUGE 179 however. The following comments are in order: tr • δij (x) = 0 so Ai and Ei don’t commute at spacelike separations. • The underlying idea here is that there are only two degrees of freedom. The transverse δfunction remedies this problem. the photon is its own antiparticle) there is only one set of creation and destruction operators. t). but (10. A must be perpendicular to the direction .15) slightly so that we avoid this contradiction. The usual choice is the transverse δfunction. Ej (y. This in itself is OK since A is not measurable. Ej (y.18) ελ (k) aλ (k)e−ikx + aλ† (k)eikx λ=1 (10. If we take the divergence of (10.17) (10.2. t)] = −iδij δ(x − y).19) Since there is no antiphoton (or more correctly.16) tries to quantize three.
Writing this as a tensor equation gives. ε1 . ε2 .21) Our last mission is to calculate the propagator.24) ˆ The four unit vectors then are η. This is accomplished by deﬁning the photon polarization vectors ελ such that ελ · k = 0 (10. gµν and kµ kν are legitimate fourtensors. and since they have no time component εµ k µ = 0 (regardless of the time component of k).25) is valid in ˆ ˆ one particular reference frame. ˆˆ ηη † − ε1 ε1† − ε2 ε2† − k k † = g. and k. k µ − (k · η)η µ ˆ kµ = (k · η)2 − k 2 (10. where g is the usual metric tensor. We complete our set of four unit vectors by projecting out the part of k µ along η µ . ελ (k) · ελ (k) = δ λλ Equation (10. THE PHOTON FIELD of propagation. they are not covariant. aλ† (k )] = δ λλ δ (3) (k − k ) (10. but ηµ k µ = 0. 2 λ=1 ˆ ˆ ελ (k)ελ (k) = −gµν + ηµ ην − kµ kν . Here is an argument that addresses both issues.180 CHAPTER 10. iDF (x − x)µν = 0T [Aµ (x )Aν (x)]0 =i d4 k e−ik(x −x) (2π)4 k 2 + i 2 ελ (k)ελ (k) µ ν λ=1 (10.25) The usual fourvector argument goes like this: equation (10. We construct a set of orthonormal vectors based on the coordinate system in which the polarization vectors are deﬁned.22) (10. µ ν (10. The process should be quite familiar by now. If you think of them as column matrices. and since they refer to a particular frame of reference. η µ εµ = 0.23) The polarization tensors are not in a convenient form.20) We also require that they be orthonormal. If .18) will be satisﬁed if [aλ (k). We introduce a timelike unit vector η = ( 1 0 0 0 ). In this frame they are perpendicular to k. That is the price of working in Coulomb gauge.
3 . Sec.7 The remaining term in (10. See M. Even the Coulomb interaction doesn’t propagate instantaneously through space.5 and references therein.10.27) or (10. Kaku. this is simply. All the terms above with uncontracted k’s vanish! The last term is something of an embarrassment.23) it gives the term ελ (k)ελ (k) = −gµν − µ ν −ηµ ην δ(t − t ) 4πx − x  2 This is the instantaneous Coulomb interaction. it isn’t.2. D(k) = k2 gµν +i (10. (10. (k · η)(kµ ην + kν ηµ ) k 2 ηµ ην kµ kν + − (k · η)2 − k 2 (k · η)2 − k 2 (k · η)2 − k 2 λ=1 (10. When substituted into (10.28) d4 k e−ik(x −x) . (2π)4 k 2 + i (10. then (24) would be valid in all reference frames. QUANTIZATION IN THE COULOMB GAUGE 181 ηµ ην were also. Let’s go ahead and substitute (10. Eq.24) into (10. Section 4. of course.25) anyhow. so it’s not.27) This argument guarantees that so long as we use (10. 6. It is a well known artifact of Coulomb gauge in classical E&M.5) written in momentum space is kµ j µ = 0. 6 7 See Jackson.26) This is not covariant. but when the photon propagator is used in a actual calculation it is always coupled to a conserved current.28) in our calculations.6 It can’t be right.26) gives the correct propagator. In this case it is cancelled by another spurious term in the Hamiltonian. DF (x − x)µν = gµν In momentum space. Unfortunately. even though the quantization was done in Coulomb gauge. the resulting S matrix elements and Green’s functions will have the correct covariant behavior. Quantum Field Theory.
182 CHAPTER 10. THE PHOTON FIELD .
1 Gauge Invariance Before we can do any realistic calculations.” It’s not what we usually think of as a gauge transformation in E&M. then L → L. This is called a “global gauge the transformation ψ → e transformation. We conclude that H ¯ µ ψAµ . Try an inﬁnitesimal transformation δψ = iqδθψ.2) . If we make iqθ ψ. so the charge ¯ µ ψ. current must be eψγ int = eψγ but there is a profound principle here that we should linger over. ¯ Furthermore. We could guess this as follows: we know that the classical electromagnetic Lagrangian has the term −j µ Aµ . The EulerLagrange equation of motion is δL δL = ∂µ δψ δ(∂µ ψ) 183 (11.Chapter 11 Quantum Electrodynamics 11. This is true. but they are related as you will see. the electron probability current is j µ = ψγ µ ψ. If the Lagrangian is to remain invariant under this transformation we must have δL = 0 = δL δL δψ + δ(∂µ ψ) δψ δ(∂µ ψ) δL δL ψ+ ∂µ ψ δψ δ(∂µ ψ) (11. we must ﬁgure out how the electromagnetic ﬁeld couples to electrons. so we assume that the interaction Hamiltonian must be of the form j µ Aµ . ¯ The freeparticle Dirac Lagrangian is L = ψ(i ∂ − m)ψ.1) = iqδθ Here θ is an inﬁnitesimal and q is something like a charge.
4) and ∂µ j µ = 0.7) is of course.2) into (11.7) We conclude that (11. if we make a local transformation.5). the usual statement of gauge invariance in classical E&M. the quantity in brackets is a conserved current. requires the existence on an additional vector ﬁeld.184 CHAPTER 11. but (∂µ − ieqAµ )ψ = eiqθ ∂µ ψ + iq(∂µ θ)eiqθ ψ − ieqAµ eiqθ ψ 1 Aµ = ∂µ θ + Aµ . jµ ∝ δL ψ δ(∂µ ψ) (11.1) gives iqδθ∂µ δL ψ =0 δ(∂µ ψ) (11. There is a stronger statement to be made if we make a diﬀerent transformation at each point in space.5) Equation (11. quantum gauge invariance. i. which we knew before.5) if our new ﬁelds transform according to (11.6). ψ = eiqθ(x) ψ. (11. The function θ(x) is arbitrary. the electron Lagrangian will be invariant: (∂µ − ieqAµ )ψ = eiqθ (∂µ − ieqAµ )ψ. If the following statement is true. Let’s insist that we replace ∂µ ψ everywhere with (∂µ − ieqAµ )ψ. (11.6) e In other words. but now we see it in a larger context. but these new ﬁelds must also appear in our Lagrangian. Applying this formula to the Dirac Lagrangian gives jµ = ¯ ψγ µ ψ. e (11.3) Evidentally.e. the electron Lagrangian will be invariant under the transformation (11. QUANTUM ELECTRODYNAMICS Substituting (11. and this Lagrangian must be invariant under the transformation Aµ → Aµ + 1 ∂µ θ(x). . Now ∂µ ψ = eiqθ ∂µ ψ + iq(∂µ θ)eiqθ ψ Now L is not invariant by itself. but it can be made invariant by incorporating an additional ﬁeld with special transformation properties.
2 Noether’s Theorem Zee calls Noether’s theorem. Given the Lagrangian density composed of N ﬁelds ϕi and their derivatives. The proof is a simple generalization to that given above. N δS = i=1 N d4 x δL δL δϕi + δ(∂µ ϕi ) δϕi δ(∂µ ϕi ) δL δL δϕi + ∂µ δϕi δϕi δ(∂µ ϕi ) = i=1 d4 x (11. NOETHER’S THEOREM 185 We know that the Lagrangian for classical E&M (excluding sources) is 1 L = Fµν F µν 4 where Fµν = ∂µ Aν − ∂ν Aµ Transformation (11. δϕi . it would contain an additional term m2 A†µ Aµ which would not be invariant.2. You will notice that the “charge” q cancels in the above calculation.8) so that ¯ Hint = −Lint = eψγ µ ψ = j µ Aµ (11.10) . with some sort of conserved “charge.11. As a convention we usually write ¯ L = ψ(iγ µ ∂µ − eγ µ Aµ − m)ψ (11. there is a corresponding symmetry group.7) does leave this invariant. and calculate the action. As a consequence the sign and magnitude of e are undetermined. for every conserved current. We conclude that local gauge invariance requires a massless vector ﬁeld. ∂µ ϕi ). but if this were the Lagrangian for a massive ﬁeld.” The theorem simply states that every continuous symmetry transformation is connected with a conserved current and hence.9) 11. brought about by some small parameter or parameters δ j . Consider an inﬁnitesimal change.” The converse is also true. L = L(ϕi . ”one of the most profound observations in theoretical physics.
Then: d Qj (t) = 0 dt (11. we can deﬁne a current µ Jj = δL ∂ϕi δ(∂µ ϕi ) ∂ j (11. QUANTUM ELECTRODYNAMICS (The repeated µ indices are summed. charm. Theorists then reasoned “backwards” from this discovery to construct the underlying conserved currents. Assume that the ﬁelds vanish at inﬁnity so the last term can be neglected. It’s important to understand that the “charge” so deﬁned usually has no connection with electrical charge. N δS = i=1 d4 x∂µ δL ∂ϕi δ δ(∂µ ϕi ) ∂ j j (11. From this conserved current.) We can use the N EL equations of motion δL δL 0 = ∂µ − δ(∂µ ϕi ) δϕi to combine the two terms in (11.186 CHAPTER 11.10). such as strangeness. Most of the conserved quantities in particle physics. Many theories in particle physics came about because the conserved quantities were ﬁrst discovered experimentally. lepton number. which in turn implies a conserved charge. .12) µ Equation (11. we can also create a conserved charge by integrating over the time component of the current: 0 Q j ≡ d3 x J j (11. come about in this way.13) Now integrate the conservation equation: 0= µ d 3 x ∂ µ Jj = 0 d 3 x ∂ 0 Jj + d3 x · Jj = d Qj + dt s dS · Jj In the last term I have used the divergence theorem to convert the volume integral into a surface integral.14) In summary: a continuous symmetry implies a conserved current. etc.11) says that ∂µ Jj = 0.11) If the action is unchanged by this variation.
ψ = ψ + + ψ − . We call the cnumber part the “contraction” of the two ﬁelds.15) in such a way k that all creation operators stand to the right of all annihilation operators. we introduce the notion of the contraction of two ﬁelds. Rewrite (11.18) where I have borrowed the deﬁnition of the propagator from (9.67). Because fermion ﬁelds anticommute. but because of the additional minus sign we must amend the deﬁnition of normal ordering somewhat. (11.103) T [ψα (x1 )ψ β (x2 )] = ψα (x1 )ψ β (x2 )θ(x0 − x0 ) − ψ(x2 )β ψ(x1 )α θ(x0 − x0 ) 1 2 2 1 (11. all creation operators stand to the left of all annihilation operators. ψ β (x2 )}θ(x0 − x0 ) 1 2 + − + + −ψ β (x2 )ψα (x2 ) + ψα (x1 )ψ β (x2 ) − − − {ψ β (x2 ).104). ψα (x1 )}θ(x0 − x0 ) 2 1 − − + ψα (x1 )ψ β (x2 ) + ψα (x1 )ψ β (x2 ) (11.11. a q number.16) + − You see that the second line of this equation is a c number and the third.19) . Deﬁnition: The ﬁelds : ψ(x1 )ψ(x2 ) · · · ψ(xn ) : are normal ordered if in each term. Remember that ψ + contains the electron annihilation operator ˆk and ψ + b ˆ contains the positron creation operator d† .and negativeenergy parts. The qnumber part is “normal ordered” in the sense that all creation operators stand to the left of all annihilation operators.17) Since the vacuum expectation value of normalordered ﬁelds vanishes. The timeordered product was deﬁned in (9. It’s clear that ψ(x1 )ψ(x2 ) = ψ(x1 )ψ(x2 ) = 0 (11. In a way analogous to (7.3.3 Feynman’s Rules for QED The derivation of Feynman’s rules for electrons and photons proceeds along the same lines as the derivation for scalar particles in Chapter 7. Each term is positive unless it required on odd number of permutations to bring the various creation and annihilation operators into that order. T [ψα (x1 )ψ β (x2 )] + + = {ψα (x1 ). ψα (x1 )ψ β (x2 ) ≡ T [ψα (x1 )ψ β (x2 )]− : ψα (x1 )ψ β (x2 ) : (11. there are some additional minus signs to keep track of. FEYNMAN’S RULES FOR QED 187 11. we have ψα (x1 )ψ β (x2 ) = 0T [ψα (x1 )ψ β (x2 )]0 = i(SF )αβ (x1 − x2 ).15) Decompose the ﬁeld into positive.
1 In analogy with (7.22) 1 Relativistic Quantum Fields. just a lot of niggling details regarding minus signs. There are three such constants in QED. we would be able to calculate them. and nonrenormalizable theories are wrong! Deﬁne the planewave solutions of Dirac’s equation as follows.3. One for the electron wave function. Drell. The next step toward calculating the S matrix requires a fermion version of the LSZ reduction formulas developed in Section 7. Bjorken and S. The derivation is similar to that leading to (7. One of the key ideas to come out of this is that renormalizability is a very special property possessed by only a few theories. D. For example. McGrawHill 1965. and one for the electric charge. J. (compare with (7. the photon renormalization constant can be set equal to unity and the other two can be gathered into the deﬁnition of the physical mass and charge of the electron. This is to say that QED is renormalizable.30) we deﬁne the fermion inﬁeld t→−∞ (11. we must be sure that they do not appear in any ﬁnal results.68)) T [ψ(x1 )ψ(x2 )ψ(x3 )ψ(x4 )] =: ψ(x1 )ψ(x2 )ψ(x3 )ψ(x4 ) : − ψ(x1 )ψ(x3 ) : ψ(x2 )ψ(x4 ) : + ψ(x1 )ψ(x4 ) : ψ(x2 )ψ(x3 ) : + ψ(x2 )ψ(x3 ) : ψ(x1 )ψ(x4 ) : − ψ(x2 )ψ(x4 ) : ψ(x1 )ψ(x3 ) : − ψ(x1 )ψ(x3 ) ψ(x2 )ψ(x4 ) + ψ(x1 )ψ(x4 ) ψ(x2 )ψ(x3 ) Wick’s theorem enables us to calculate the vacuum expectation value of any product of boson and fermion ﬁelds. Uks (x) = m u(k. D.20) lim ψ(x) = Z2 ψin (x) (11.9 . s)e−ikx (2π)3 Ek m v(k. There are no new ideas. As it is. As it turns out. In this case we must include a minus sign for each operator interchange. QUANTUM ELECTRODYNAMICS An additional complication arises when there are more than two ﬁelds to time order.188 CHAPTER 11. I will simply state the results and refer the reader to the classic exposition in Bjorken and Drell.43).21) The constant Z2 is called the electron wave function renormalization constant. Section 16. s)eikx Vks (x) = (2π)3 Ek (11. If it were not for the fact that all Feynman diagrams with closed loops give divergent results. one for the photon.
If we remove a particle from the instate (refer to Section 7.11. s)Uks (x) + d† (k. t) (11. we need to be able to reduce photon ﬁelds. The wave function is 1 Akλ = e−ikx (k. In this case the asymptotic condition is t→−∞ lim A(x.27) After two or more particles have been “reduced. λ)A∗ (x)] in kλ (11.29) (2π)3 2ω so that the Fourier expansion of Ain (x) is Ain = d3 k λ [ain (k.30) . t) = Z3 Ain (x.28) The constant Z3 will turn out to renormalize the electron charge.3 for the signiﬁcation of the notation) Sβα = β outα in i ∼ −√ Z2 ←− − − − − −− d4 x β outψ(x)α − k in (−i∂ − m)Uks (x) (11. s)Vks ] in (11. λ)Akλ (x) + a† (k.” the ﬁelds inside the braket combination are time ordered.3.25) Removing a particle from the outstate leads to i ∼ −√ Z2 −− −→ −−− d4 x U ks (i∂ − m) β − k outψ(x)α in Uks (x) (11. λ) (11. Finally.26) Removing an antiparticle from the outstate gives i ∼√ Z2 ←− − − − − −− d4 x β − k outψ(x)α in (−i∂ − m)Vks (x) (11.23) There are several formulas analogous to (7. FEYNMAN’S RULES FOR QED In this notation the secondquantized Dirac ﬁeld is ψin (x) = d3 k s 189 [bin (k.41) depending on whether we remove a particle or antiparticle from the instate or outstate.24) Removing an antiparticle from the instate leads to i ∼√ Z2 −− −→ −−− d4 x V ks (i∂ − m) β outψ(x)α − k in (11.
In a real experiment. This is a “cookbook” procedure. You should be able to do it while in a coma.20) and (8. The interesting part is Tf i .31) comes from the spacelike nature of the polarization unit vector µ µ =− · = −1 This would be a good point to summarize the results so far.27). (11. α stands for the speciﬁc initial state you have set up in your experiment. After you have done a few such calculations. QUANTUM ELECTRODYNAMICS The reduction formulas are virtually identical to (7. do all the implied diﬀerentiation and integration.69).9) and substitute it into the formula for the Soperator (7. even a modest calculation can involve.43) (with m = 0 of course) except for one minus sign. This is an inﬁnite series.190 CHAPTER 11. implicitly at least. (11. (11. 2 . and (11. which is the crucial ingredient in (8.26). We keep only those terms corresponding to the desired degree of approximation.32) The δf i stands for all those terms appearing in the reduction procedure in which the two initial particles don’t scatter. All this business about calculating cross sections has to do with the inﬁnitesimal fraction that does. We then use (11. These terms are substituted into (??) Sβα = β outSα in .33) In this context.25). you will realize that you are doing the same few things over and over. You have to have patience though. These things can be boiled down to a few simple rules known as Feynman’s rules. According to (8. No problemsolving ability is required.31) The additional minus sign in (11. λ) ←Aµ∗ (x) 2 kλ (11.21) to calculate cross sections and decay rates. tens of thousands of terms. the vast majority of incident beam particles don’t scatter. Sβα = β outα in −i ∼√ Z3 − d4 x β outAµ (x)α − (k.5) the Smatrix can be rewritten Sf i = δf i − i(2π)4 δ (4) (Pf − P − i)Tf i (11.24). (11. Finally. To calculate the Smatrix.2 The (2π)4 δ(Pf − Pi ) appears when we do the last integral over coordinate space. we start with the basic QED Hamiltonian (11.31) to reduce out the various electron and photon ﬁelds. and β represents the particular ﬁnal state whose probability you wish to calculate.
There is a factor given by the following rules for each external line. including loops. with n external legs. s) Z2 (2π)3 2Ep p Outgoing fermion: = u(p. s) Z2 (2π)3 2Ep Incoming photon: = µ (p) Z3 (2π)3 2Ep p . s) Z2 (2π)3 2Ep p Outgoing antifermion: = v(p.3. Ignore vacuumtovacuum graphs. 2. Each vertex must have a continuous electron or positron line and a photon line terminating on the fermion. s) Z2 (2π)3 2Ep Incoming antifermion: = p v(p. Incoming fermion: = p u(p.11. Draw all possible connected. FEYNMAN’S RULES FOR QED 191 1. topologically distinct diagrams.
At each vertex. However. For each internal line there is a propagator Electron propagator: µ p p ν = p2 = i( p + m) − m2 + i −igµν p2 + i Photon propagator: 4. QUANTUM ELECTRODYNAMICS p Outgoing photon: = µ∗ (p) Z3 (2π)3 2Ep 3. As a consequence. For each internal loop.192 CHAPTER 11. 8. integrate over: d4 q (2π)4 (11. . 6. the spinors associated with any vertex must appear in the order uγ µ u (with v replacing u as needed). Electronphoton vertex: = −ieγ µ 5. A relative factor 1 appears between graphs that diﬀer from each other by an interchange of two identical external fermion lines. diagrams that are topologically equivalent are counted only once. Insert an additional factor of 1 for each closed fermion loop. place a factor of −ieγ µ . With this in mind. let’s do some examples. Internal fermion lines appear with arrows in both clockwise and counterclockwise directions.34) 7. Remember that the electronphoton coupling has the form eψγ µ ψAµ .
but they are heavier by a factor of 200.35) as iM = ie2 µ a bµ q2 (11. In terms of momentum. I will write (11. that is p + p → k + k .4 The Reaction e− + e+ → µ− + µ+ Consider the reaction e− + e+ → µ− + µ+ . There are two useful consequences.38) Now make use of the spin information. they all commute and complex conjugation and Hermitian conjugation have the same eﬀect.37) becomes M 2 = e4 v(p )γ µ u(p)u(p)γ ν v(p ) q4 u(k)γµ v(k )v(k )γν u(k) (11. This allows us to write aν† = u† γ ν† γ 0† v = u† γ 0 γ ν v = uγ ν v With that insight (11. In order to obtain a diﬀerential cross section from M we must do three diﬃcult things: square it. and average over initial spins. For another.4. This simpliﬁes the problem in several ways. THE REACTION E − + E + → µ− + µ+ 193 11. r.36) In order to calculate M 2 . s .35) I have added subscripts e and µ to the spinors to indicate that they contain diﬀerent masses. they have the same electromagnetic interactions. we are usually entitled to neglect the mass of the electron. and in terms of spin indices we have s and s → r and r . i.40) 2 s 2 r s r . The diﬀerential cross section will be proportional to 1 1 M (s. The muons are almost identical to electrons. In order to make the notation more compact. r )2 (11.37) Notice that a and b are 1 × 1 matrices in spin space and fourvectors in Minkowski space.e. there are no identical particles in this example. I will get back to spins in a minute. The spin indices are implicit. we will need to ﬁgure out ab2 = aµ aν† bµ b† ν (11.39) (11.11. Feyman’s rules give iM = ie2 v e (p )γ µ ue (p) q2 uµ (k)γµ vµ (k ) (11. For one thing. The momentum of the exchanged photon is q = p + p = k + k . sum over ﬁnal spins.
QED calculations eventually come down to taking traces.39).42) can be rewritten = 1 2m µ ν vd (p . An object like γ µ is a 4 × 4 matrix in spin space. and (11. s)v(p.40). Tensors commute as you know. s ) s I apologize for the notational overload.39) v(p .42). s )v a (p . = 1 2m 2 Tr ( p − m)γ µ ( p + m)γ ν Finally combining (11. I have turned it in to a tensor by adding the matrix indices a and b. s)γ ν v(p . s)u(p. s )γab ( p + m)bc γcd vd (p . QUANTUM ELECTRODYNAMICS Remember from a previous chapter u(p. we get 1 4 M 2 = spins 1 4 e 2mq 4 Tr ( p − me )γ µ ( p + me )γ ν (11. s )γ µ u(p.194 CHAPTER 11. s) = s Let’s do the spin sums on the ﬁrst term of (11. s ) s.43) × Tr ( p − mµ )γµ ( p + mµ )γν There you have it. s )γab ( p + m)bc γcd s Now v and v are in the right relation to one another so that we can use (11. . s)u(p.41) v(p. s) = s p+m 2m p−m 2m (11. (11. so (11.42) = 1 2m µ ν v a (p .s (11.41) 1 2 µ ν = ( p − m)da γab ( p + m)bc γcd 2m Now this is recognizable as the trace of a product of matrices.
1 Tr( a b) = aµ bν Tr(γ µ γ ν ) = aµ bν Tr{γ µ .47) . It has the properties Tr(γ 5 ) = 0 Now the line of argument goes Tr( a1 a2 · · · an ) = Tr( a1 a2 · · · an γ 5 γ 5 ) = Tr(γ 5 a1 a2 · · · an γ 5 ) = (−1)n Tr( a1 a2 · · · an γ 5 γ 5 ) = (−1)n Tr( a1 a2 · · · an ) The trace is zero for n odd. e 1 4 M 2 = spins (11. THE REACTION E − + E + → µ− + µ+ 195 11. This results in some paradoxical formulas like Tr 1 = 4 and Tr m = 4m. In a moment I will use Tr(γ µ γ ν γ ρ γ σ ) = 4(g µν g ρσ − g µρ g νσ + g µσ g νρ ) (11. The proof requires a trick.45) which is obtained from repeated application of (11.11.43) can be evaluated by using (11. Remember the γ 5 matrix.4. Introduction to Quantum Field Theory Appendix A3.2 Kinematics A short calculation using (11. Our result can be summarized Tr(γ µ γ ν ) = 4g µν (11. Traces of an even number of γ’s are generally not zero. Thus in the equation above Tr(γ µ γ ν ) = Tr(γ ν γ µ ). γ ν } = aµ bν Tr(g µν ) = 4a · b 2 I have used the important fact that the trace of a product of matrices such as Tr(abc · · · z) is invariant under any cyclic permutation of the matrices. Here is a simple result. You can check from the representation of the gamma matrices that Trγ µ = 0.4.44) The trace of an odd number of γ’s always vanishes. even when they don’t look like that.46) twice.44). (γ 5 )2 = 1 {γ 5 . There are many other identities that are easily proved.4. At the moment I am looking at Peskin and Schroder.1 Trace Technology The things we are taking traces of are all 4 × 4 matrices. Others can be found in any standard text.46) We can set m2 = 0. Then (11. γ µ } = 0 11.45) yields Tr ( p − m)γ µ ( p + m)γ ν = 4[p µ pν + p ν pµ − g µν (p · p + m2 )] e 8e4 [(p · k)(p · k ) + (p · k )(p · k) + m2 (p · p )] µ q4 (11.
which must be normalized diﬀerently from the KlineGordon ﬁelds. (11. Such an experiment is typically done with a colliding beam machine for which the laboratory is the CM ˆ frame.50) vA − vB  = EA EB dσ dΩ = In our example the M in (11. p = (E. . E z) and ˆ p = (E. 11. k = (E. QUANTUM ELECTRODYNAMICS Equation (11. but in order to compare the result with an experiment. I will therefore state (without proof) the appropriate version of ?? is 1 p1  M (pA + pB → p1 + p2 )2 2EA EB vA − vB  (2π)2 4ECM CM (11. −k).48). Since then we have introduced the Dirac spinors.3 One of the most important discoveries of twentiethcentury physics 3 There are some exceptions in the case of spinzero particles. In our case particles A and B are collinear and assumed massless. The angle between k and the z axis is θ. q 2 = (p + p )2 = 4E 2 p · k = p · k = E 2 − Ek cos θ Eqn. In this case kB kA − =2 (11. This is ﬁne as far as it goes.5 Introduction to Renormalization Every Feynman diagram with one or more closed loops yields a divergent integral.47) is valid in any Lorentz frame. Let us assume then that k = (E. The following kinematic identities follow with me = 0. k).48) p · p = 2E 2 p · k = p · k = E 2 + Ek cos θ Back in Chapter 8 I derived the relationship between M and the diﬀerential cross section. we need to specialize to a deﬁnite frame of reference. but it’s universally true in qed.49) is replaced by the spinaveraged M of (11.47) can be rewritten 1 4 M 2 = e4 spins 1+ m2 µ E2 + 1− m2 µ E2 cos2 θ (11.196 CHAPTER 11. −E z).49) This holds in any coordinate system in which generic particles A and B scatter into a ﬁnal state consisting of two particles with momenta p1 and p2 .
since there are ﬁve powers of k in the numerator and only four in the denominator. since I have ignored the diagram in which the electron emits another photon before reabsorbing the ﬁrst.3.” It is clearly divergent. This is an extremely technical and complicated subject.” Such diagrams present a huge additional complication. The program of extracting useful information from meaningless inﬁnities is called renormalization. Section 11. even to fourth order. iSF ≈ iSF + iSF ΣSF + iSF ΣSF ΣSF This is not quite correct. the amplitude for this process is iSF (p) d4 k (−ieγ µ ) (2π)4 i( p− k + m) (−ieγ ν )iSF (p) (p − k)2 − m2 + i (11. Such as term would arise in perturbation theory as a secondorder correction to the electron propagator. I will concentrate on a subset of divergent diagrams that go under the heading of electron selfenergy. We could deﬁne a “more nearly exact” propagator iSF (p) ≈ iSF (p) + iSF (p)Σ(p)SF (p) I say “more nearly exact” meaning that iSF (p) contains all the secondorder corrections that would arise in the usual perturbation expansion of the propagator. Consider the quantum process in which a single photon of momentum k is emitted and reabsorbed by a virtual electron with initial and ﬁnal momentum p. We will eventually evaluate it. INTRODUCTION TO RENORMALIZATION 197 is that these divergent terms contain real physical content that is part of the “right answer” to any higherorder calculation. To do this.51) γ µ (m+ p− k)γµ d4 k SF (p) (2π)4 (−k 2 − i )(m2 − (p − k)2 − i ) ≡ iSF (p)Σ(p)SF (p) The last equality implicitly deﬁnes the divergent quantity Σ. We could make an even better approximation by including another selfenergy insertion.5. but for the time being I would like to investigate some of its general properties. but “suﬃcient unto the day is the −igµν k2 + i = iSF (p) −ie2 . but I would like to expose you to some of the basic ideas.11. know as a “selfenergy insertion. In the interim between emitting and reabsorbing the photon the electron has momentum q = p − k. According to Feynman’s rules. Diagrams of this sort are called “overlapping divergences.
198 CHAPTER 11. The sum would look like this. so I am tempted to write iS = iS 1 1 − ΣS (11. • It’s clear from (11. Here are some further observations.52) that Σ has the form Σ = mA(p2 )+ pB(p2 ) (11. Therefore the following forms are equivalent. what does it mean to divide by a 4 × 4 matrix? You can regard (11.” I will ignore them.58) (11. QUANTUM ELECTRODYNAMICS evil thereof.53) The question is.56) . S =S 1 S 1 = S= 1 − ΣS 1 − ΣS 1 − ΣS (11. Furthermore p+m S= 2 p − m2 (I will ignore the i ’s for the time being also. I can go on adding diagrams like this ad inﬁnitum.) It follows that [S.57) 1 p+m p+m = . Σ] = 0.54) It’s true that A and B are inﬁnite.52) (I have suppressed the F subscript and well as the (p) to make the equations easier to read. My conclusions will still be qualitatively correct. = p2 − m2 ( p − m)( p + m) p−m (11.) This looks like the familiar geometric series (1 − x)−1 = 1 + x2 + x3 + · · · .52). iS[1 + ΣS + ΣSΣS + (ΣS)3 + · · · ] (11. I don’t want to distract you with details. The propagator can now be rewritten iS = iS S −1 i = −1 1 − ΣS S p−m−Σ (11.55) • Since p p = p2 it makes sense to write in the same spirit S= and ﬁnally S −1 = p − m.53) as notational shorthand for (11. but let’s ignore that for the time being. I promise I haven’t forgotten them and will put them back when necessary.
The agenda of keeping track of these corrections goes by the name “renormalization.58) in such a way that the physical mass appears in the denominator. the same sort of thing happens when we calculate higherorder corrections to the electronphoton vertex.5. There we ﬁnd that the charge e put into the Hamiltonian is not the physical charge of the electron but some “bare” charge. It must be that the series of selfenergy terms we have summed has modiﬁed the mass of the particle. To this end we expand the self energy term as .” It is renormalizable and right! It is also possible to do quantum ﬁeld theory with gravitons. As soon as we include the selfenergy terms. the m that appears in the Hamiltonian is not the physical mass of the particle.” It is a remarkable fact that all the inﬁnite corrections brought about by all loop diagrams to all orders of perturbation theory can swept under the rug of these two constants. we should be able to calculate the physical mass or vice versa. because we have always worked at the lowest nontrivial order of perturbation theory. It seems that given the bare mass. however. is a good example. Fermi’s theory of weak interactions. It is almost axiomatic that a propagator is the probability amplitude for the particle to propagate from x to y such that p2 = m2 on the mass shell. and we have no idea what the true theory might be.11. but it would require an inﬁnite number of inﬁnite constants to take care of all the loops! It was eventually modiﬁed by the addition of the intermediate vector bosons. The resulting theory is called the “standard model. which we will study in the next chapter. It is necessary to rewrite (11. INTRODUCTION TO RENORMALIZATION 199 Leaving aside the fact that Σ is inﬁnite. It is nonrenormalizable. which I will call m. The fact that Σ is inﬁnite makes this permanently impossible.” It is an article of faith that no nonrenormalizable theory can be completely right. the “dressed” or physical mass. and again we must keep track of the e’s and e’s. however useful it might be phenomenologically. Every propagator we have encountered so far has had a pole at the mass of the particle. We have not faced up to this problem so far.57). A theory for which this is possible is said to be “renormalizable. there is still a strange pathology in (11. To this extent it is correct. We have to confess that at least in the context of perturbation theory. The theory at least predicts Newton’s law of gravitation. It explains nuclear beta decay quite nicely. We call the m in the Hamiltonian the “bare” mass and the mass that marks the pole of the exact propagator. we must keep track of both m and m. Worse yet. To that order we can assume the the bare and physical masses are identical. and everything works out OK.
(11.61) Strictly speaking. It is ﬁnite.60) into (11. but for = 0 it is a wellbehaved mathematical expression. and when the time comes to calculate some physical quantity such as a cross section or reaction rate.60) and (11.59) deﬁnes implicitly what is meant by R(p2 ).61) are not well deﬁned.58) iS = −i m + Σ(m)− p + ( p − m)Σ (m) + ΣR (11. Σ(m) and Σ (m) are both inﬁnite. Our strategy for giving them a precise meaning works as follows.51) in such a way the the integrals converge for > 0. but in fact. so the operations in (11. QUANTUM ELECTRODYNAMICS Σ( p) = Σ(m) + ( p − m)Σ (m) + ( p − m)2 R(p2 ) This is the key equation. For example. • It appears that in the term Σ we have diﬀerentiated with respect to a matrix! That is not so strange as it seems. CHAPTER 11. Rather. I will introduce an additional parameter into (11. the A(p2 ) term in (11. The point is that all these calculations are done with ﬁnite (though epsilondependent) terms. and it deserves some comment.60) In which case it deﬁnes ΣR . One part will depend on . there are no epsilondependent terms remaining. (11. p can never equal m. Now substitute (11.62) .59) • It looks like a power series expansion about the point p = m. The other part will depend on p2 but not on . A (m) = dA(p2 ) d p = p=m dA(p2 ) d p2 dp2 d p = 2m p=m dA(p2 ) dp2 p2 =m2 (11.54) can be diﬀerentiated as follows.54) can each be split into two parts. and important. It becomes inﬁnite as → 0. I will sometimes write it Σ( p) = Σ(m) + ( p − m)Σ (m) + ΣR ( p) (11. It will turn out that A(p2 ) and B(p2 ) in (11. I am about to show that all the epsilondependent terms can be lumped together with the bare mass to give the physical mass m and a new epsilondependent term Z2 called the wave function renormalization factor. calculable. which cancels out in any complete calculation of a Feynman diagram. By construction ΣR (m) = 0. The point is that p2 = p2 . since m is diagonal and p never is.200 follows.
11.5. . We need one further constant. The trick is 1 = AB 1 0 dz ≡ [Az + B(1 − z)]2 1 0 dz D2 (11. You can see from (11. Notice that the denominator is the product of two factors that can be labelled A = m2 − (p − k)2 − i B = −k 2 − i Feyman realized that a fourdimensional integral can be simpliﬁed by making it ﬁvedimensional. Then D = C2 − k 2 where C 2 ≡ z[m2 − p2 (1 − z)] 4 This is standard notation.64). e = Z2 e. Then Z2 disappears from the denominator because it is absorbed into Σ.64) The looks like the “bare” propagator iS (with m replaced by m) multiplied by a factor. but since we are only dealing with the electron selfenergy part. There are other Z’s. The Z2 disappears from the numerator when we realize that our virtual electron will not be observable until it couples to a photon at each end. Part of this√ modiﬁcation is due to Z2 . First consider the Z2 in the denominator of (11. if Z2 were not epsilondependent.51).51) that Σ is proportional to e2 . however. these other constants don’t arise.62) can be rewritten iS = i( p + m) Z2 2 − m2 1 + (m− p)Z R(p2 ) p 2 (11. Z2 is ﬁnite.) Each electronphoton vertex comes with a factor of e. Now I will evaluate the very diﬃcult integral (11. (Draw some Feynman diagrams to convince yourself this is true. would be ﬁnite and calculable. which. Before we set = 0. We must confess again the the e that appears in the Hamiltonian is not the measured charge of the electron. and we can show why it does not appear in the ﬁnal result of any complete calculation of a Feynman diagram. but rather a “bare” charge.4 −1 Z2 ≡ 1 − Σ (m) (11. so the entire diagram is proportional to e2 Z2 = e2 . Deﬁne k µ ≡ k µ + zpµ . Let’s tentatively deﬁne a “dressed” charge. INTRODUCTION TO RENORMALIZATION 201 I looks like the “real” mass is m = m+Σ(m). which is modiﬁed by all the loop diagrams to all orders of perturbation theory.63) Equation (11.65) The term D can be simpliﬁed by completing the square.
For our purposes. You will also see that Γ(0) = ∞. z) d4 k 1 4 (C 2 − k 2 − i )2 (2π) (11. say n.69) If α is an integer. We expected that. the numerator of (11. The integral (11. Now of course.67) The integral can be done for integer dimension d using the remarkable formula5 Γ(n − d/2) dd k 1 i = d (C 2 − k 2 − i )n d/2 (2π) Γ(n) (4π) Here Γ is the Euler gamma function deﬁned by ∞ 1 C2 n−d/2 (11. the integral on the left side of (11.68) only makes sense when d is an integer. It’s useful to get Maple to plot this function for you. Even if d is only inﬁnitesimally less than four.66) dzN (p.70) This is proved in Gross. Relativistic Quantum Mechanics and Field Theory. it is still ﬁnite.68) Γ(α) ≡ 0 dt tα−1 e−t (11. but the expression on the right interpolates smoothly between integer values of d. z) = γ µ [m+ p(1 − z)− k ]γµ I can delete the k term using the familiar argument that an odd function integrated over even limits yields zero. page 344 . You will see that Γ interpolates smoothly between the integer values of α. Γ(n − d/2) Γ(3 − d/2) 2 = = Γ(n) 2 − d/2 Σ(p) = Insert the expansion C − = 1 − ln C + O( 2 ) 5 2e2 (4π)2 1 dzN 0 C− (11. N = 2[2m− p(1 − z)] With all these deﬁnitions the selfenergy term becomes Σ(p) = −ie2 0 1 (11.202 CHAPTER 11.67) has to diverge. n = 2. QUANTUM ELECTRODYNAMICS With these substitutions. gives us the ﬁnal form of the numerator function. Now deﬁne ≡ 4 − d. and in general Γ(α) = (α−1)Γ(α−1).51) becomes N (p. Γ(n) = (n−1)!. Using the fact that γ µ γµ = 4.
60). ΣR ( p) is as promised. although the resulting expressions are not very illuminating. . z) + O( ) The integrals in (11. ˜ Σ( p) = (mB1 + pB2 )/ + Σ( p) (11. Thus we can absorb all the inﬁnite terms into the two constants m and Z2 . B1 and B2 are simple constants. INTRODUCTION TO RENORMALIZATION Σ(p) = 2e2 (4π)2 1 203 (11. When we expand Σ( p) using (11. ﬁnite.5. z) 0 1 − ln C(p2 . again as promised.72) ˜ where Σ( p) does not contain and so remains ﬁnite in the limit → 0.11.71) are easy enough to do. and calculate ΣR ( p)and R(p2 ) which are.71) dzN ( p. The important point is that Σ( p) can be divided up as follows. ﬁnite and important.
204 CHAPTER 11. QUANTUM ELECTRODYNAMICS .
and electromagnetic forces. 2 I have written ν rather than ν. it must interact weakly with ordinary matter. electrons. This is harder to recognize.” The particle would have to have zero charge and spin h/2 to conserve charge and angular momentum. so the name was later changed to “neutrino.” ¯ The neutron as we know it is a constituent of nuclei. because in this case. Gravity is too weak to have any eﬀect.Chapter 12 Weak Interactions The ﬁeld of nuclear physics began with the study of radioactive decay. the energy spectrum of the emitted electron looks something like the sketch in Fig. By 1930 the process was believed to be1 N XZ →N XZ+1 + e− . and photons respectively. they manifest the strong. 1. If there were only two particles in the ﬁnal state. say N XZ . however. In 1930 W. weak. So far as one could ¯ tell from the endpoint energy. Unfortunately.2 Now we know that there are actually three kinds of neutrinos all There is a similar decay in which a positron is emitted. In retrospect. because the positron immediately captures an electron forming positronium. it had to have a very small mass.” Consider some generic nucleus that undergoes βdecay. which Pauli called the “neutron. Since it has proved very diﬃcult to detect. This decays quickly into two or three photons. The distinction between neutrinos and antineutrinos is a subtle matter. N XZ →N XZ+1 + e− + ν . to 1 205 . the particle is actually an an¯ tineutrino. the three forces that operate at the level of nuclei and particles.and γdecay corresponding to the emission of alpha particles. ¯ The ν was a hitherto unknown particle. There were three identiﬁable processes called α.β. Several wild hypotheses were put forward to explain this including the suggestion that energy and/or momentum were not conserved in these reactions. all the electrons should have the energy marked Emax . and there has never been any convincing evidence of a “ﬁfth force. Pauli suggested that βdecay was in fact a threebody reaction.
3 It is ironic that whereas the radioactive decay of nuclei was the ﬁrst evidence of the operation of nuclear forces. electromagnetic. Nuclei are very complicated objects. By the mid 1950’s it was believed that neutron decay as well as all the other weak interactions known at the time could be described by the interaction Hamiltonian G † Hint = √ J α (x)Jα (x) 2 which I will return later. but for the time being I will consider only one kind. These turn out to be the purely leptonic decays such as µ → e + ν + ν where the strong interactions play no role. The “scientiﬁc method” dictates that we should understand the simplest instances ﬁrst. This process is complicated slightly ¯ by the eﬀect of the strong interactions.1) . I would ¯ like to postpone that reaction for a bit and follow a more historical line of development. We know that once outside of a nucleus. a neutron will decay. and the strong.1: The βdecay energy spectrum with very small masses. and this was the ﬁrst decay to be studied quantitatively. but no nuclear physics is involved. the socalled electron neutrino and make the very good approximation that it is massless. n → p + e− + ν in about 13 minutes.” (12. and weak forces are inextricably entangled in all of these processes. 3 “Suﬃcient unto the day is the evil thereof.206 CHAPTER 12. WEAK INTERACTIONS Events Emax Electron Energy Figure 12. it is also so very complicated that no ﬁrstprinciples calculation could ever reproduce the lifetimes and branching ratios of these decays.
µ and τ .2) is the universal weak coupling constant or Fermi constant. J α (x) = lα (x) + hα (x).3). The bosons. This is modelled after the electromagnetic interactions that we have already studied.207 µ− γ µ+ e e+ Figure 12. 2 can be described by the matrix element α β Mint = −ie(¯e γ α ue )Dαβ (¯µ γ β vµ ) = −ieJe Dαβ Jµ v u (12. Likewise there are many hadrons that undergo weak decay. semileptonic processes like neutron decay. and many purely hadronic processes such as Λ → p + π and K → π + π.4) The are three diﬀerent leptons. so in principle hα should be a sum of of all the corresponding currents.166 × 10−5 GeV−2 (12. It is now known that the weak interactions are mediated by the exchange of vector bosons just as the electromagnetic interactions are mediated by the exchange of the photon.1) describes many diﬀerent processes. The weak current J α consists of a leptonic part lα and a hadronic part hα . (12. Fermi suggested that protons and neutrons (collectively called nucleons) and electrons and neutrinos (collectively called leptons) could also constitute currents just as the electrons and muons do in (12. the e.3) where Dαβ is the photon propagator.2: e+ + e− → µ+ + µ− via one photon exchange where G = 1. however. are . There will be purely leptonic process such as muon decay. For example the process e+ + e− → µ+ + µ− shown in Fig. so lα consists of a sum of three separate currents. In this way (12.
1). for example. Roughly speaking. but there is much more to the story. however.1). In more mathematical terms. so for ordinary nuclear processes. 12. there is an operation called parity conjugation. so other interaction terms must be considered. is described by the Feynman diagram Fig 13. In 1956 . The spin frequently does ﬂip in nuclear βdecay. The statement that parity is conserved is equivalent to saying that this transformation leaves all the equations of motion form invariant. this means that the mirror image of any allowed process is also an allowed process with all the same reaction rates.” had its successes. For one thing. at least not in the nonrelativistic limit. −x). it replaces every wave function or quantum ﬁeld ψ(x) with a new function ψ (x ) that describes the particle or system in the mirrorreﬂected world. Equation (12.1 Parity Violation By far the most profound modiﬁcation of (12. the socalled “currentcurrent interaction.208 CHAPTER 12. Until 1956 it was assumed that parity was conserved in all interactions. Weak decays therefore are essentially pointlike interactions. Neutron decay. WEAK INTERACTIONS ep ν n Figure 12.3. has to do with parity.3: Feynman diagram for neutron decay extremely heavy. and second. which has two eﬀects: ﬁrst it replaces xµ with x µ = (ct. the propagator collapses into a single constant that is absorbed into G.1) does not allow the nucleon’s spin to ﬂip. (12. It is also nonrenormalizable so it is at best a lowenergy approximation.
In the mirror world it is peaked along J . This is because electromagnetic interactions do conserve parity.6) Multiply the ordinary Dirac equation on the left by γ 0 .6) to be true. parity is not conserved.or axial vector.5) Here J is the nuclear spin and p the momentum of the electron. imagine viewing the experiment in a mirror held perpendicular to the nuclear spin. does change. I showed that ψψ transforms like a scalar ¯ µ ψ transforms like a vector. This in itself is proof that parity is not conserved. it must be that ψ (x ) = γ 0 ψ(x) (12. Since angular momentum has the form L = x × p. The electron distribution.5) is that the distribution is not symmetric between the forward (with respect to the nuclear spin or magnetic ﬁeld) and backward direction. it does not change sign under parity conjugation. 60 Co also undergoes γdecay. Ergo. but incorporating parity into the theory of relativistic electrons requires a some formalism.4 This was done by polarizing 60 Co nuclei in a strong magnetic ﬁeld. To see how this follows.1.4 we touched brieﬂy on the ﬁve bilinear forms that can be ¯ constructed with Dirac spinors.12. Parenthetically. and the photon distribution is symmetric between the forward and backward directions. The important point about (12. It was found that the intensity distribution of the emitted electrons could be described by the formula I(θ) = 1 − J · p/E = 1 − v cos θ (12. 60 Co βdecays emitting an electron. That was a simple argument. Its mirror image is identical to the real thing. γ 0 (iγ µ ∂µ − m)ψ(x) = (iγ µ ∂µ − m)γ 0 ψ(x) = 0 In order for (12. In this world it is peaked in the direction opposite to J . I under Lorentz transformations and that ψγ also introduced the γ 5 matrix and made some vague remarks to the eﬀect 4 I will explain eventually what I mean by this . Our postulate of form invariance requires that the Dirac equation in the mirror reﬂected world should be (iγ µ ∂µ − m)ψ (x ) = 0 (12.8) (12.7) In Chapter 9. PARITY VIOLATION 209 Lee and Yang pointed out that there was no evidence that this was true for the weak interactions. It was shown shortly thereafter that weak interactions violate this principle maximally. however. It is an example of a pseudo.
In the same way it can be shown that ¯ ψγ µ γ 5 ψ is a pseudo. For easy reference. Under parity conjugation. This precise form is tied up with the properties of the neutrino to which we now turn. In principle one could use all ﬁve of the bilinear forms in arbitrary linear combinations.12) 12. this mass is completely negligible. There are a few purely weak decay processes.11) (12. γ 5 = iγ 0 γ 1 γ 2 γ 3 (12.1) to accommodate parity violation? At ﬁrst sight one could replace γ α by γ α γ 5 making the currents into pseudovectors.14) is the electron neutrino whose mass is a tiny fraction of an electron volt. Except for those experiments involving neutrinos from the sun or from outer space.210 CHAPTER 12. as claimed. There is considerable indirect evidence that they all have some small mass. In this limit. and in the case of some nuclear βdecay. however.or axial vector. ¯ ¯ ¯ ¯ ψ(x)γ 5 ψ(x) → ψ (x )γ 5 ψ (x ) = ψ(x)γ 0 γ 5 γ 0 ψ(x) = −ψ(x)γ 5 ψ(x) (12. WEAK INTERACTIONS that it had something to do with parity. the muon neutrino. 5 5 (12. How are we to modify (12. But since the Hamiltonian is a product of two such currents. γ } = 0 ¯ Let’s take a look at ψγ 5 ψ.10) (12. half of the neutrino . a pseudo scalar. After much ¯ experimental work it is clear that the way to construct purely leptonic weak currents is as follows. that possibility must be taken seriously. The neutrino that appears in (12. and it will simplify our work considerably if we ignore it. and the tau neutrino. it is useful to look at reactions in which there are no strong interactions to complicate things. the minus signs cancel. here are its deﬁnition and properties. In order to understand the fundamental nature of weak interactions.14) lµ = ue γ µ 2 (1 − γ 5 )vν ¯ This is called the VA or vector minus axial vector interaction. 1 (12.13) It is. which from a theoretical point of view is both a complication and an embarrassment. µ → e + ν + ν .2 Neutrinos We know that there are three kinds of neutrinos called the electron neutrino.9) (γ 5 )† = γ 5 (γ 5 )2 = 1 {γ . and parity conjugation has no eﬀect. and the most accessible is ordinary muon decay.
2 In order to see what this means. the third and fourth components are proportional to p/(E + m) and so vanish in the lowenergy limit. Concentrate on the positiveenergy solution for the time being.15) The important diﬀerence is that γ 0 is diagonal in the standard representation. . When spinors are constructed with this choice. (12. The gamma matrices are deﬁned by their anticommutation relations.17). it is useful to reconstruct the Dirac spinors according to a diﬀerent plan. There are several diﬀerent sign conventions used by other authors. massless particles are helicity eigenstates.5 γ0 = 0 1 1 0 γi = 0 σi i −σ 0 γ5 = −1 0 0 1 (12. Let us reﬁne our notation somewhat and use λ = ±1 to represent the helicity. NEUTRINOS 211 spinor is “missing. According to (12. Its eigenvalues are called helicity.15) this is 0 E−σ·p E+σ·p 0 φ χ = 0.18) Beware.16) The determinant of the coeﬃcient matrix must vanish.7 is called the Dirac or standard representation. The Dirac equation in the limit m = 0 and in momentum space is simply pu = 0. righthanded particles) or antiparallel (negative helicity. We are left with ˆ χ=σ·pχ (12. and γ 5 is diagonal in the chiral representation. and so E = ±p. The spin is either aligned parallel to the direction of motion (positive helicity.” and the role of the 1 (1 − γ 5 ) in (12.12. lefthanded particles).2. The choice given in Equations 9. The operator σ · p represents the component of spin along the particle’s direction of motion. There is another important representation called the Weyl or chiral representation.14) is to “delete” it. but there are several diﬀerent matrix representations that have the same relations.17) ˆ φ = −σ · p φ In this limit the upper and lower components decouple. Using the representation (12.6 and 9. Introduce the helicity eigenstates deﬁned by σ · p φλ = λpφλ σ · p χλ = λpχλ 5 (12.
and lefthanded refer to helicity not chirality.20) 1 2 (1 Clearly the 1 (1 ± γ 5 )’s are acting like projection operators.” Nonetheless. but somehow only the lefthanded states are allowed to participate.19) The matrix γ 5 is called the chirality operator.7 It is these “missing” or noninteracting helicity states that are directly responsible for parity violation. 2 1 2 (1 + γ 5 )u = − γ 5 )u = 0 0 0 1 1 0 0 0 φ χ φ χ = = 0 χ φ 0 (12. so whether or not there are positive helicity states is something of a metaphysical question.23) The word “chirality” comes from the Greek meaning “hand. If we take the neutrino spinor to be P− uν = φ− 0 (12. P+ = P+ .212 CHAPTER 12.6 In the chiral representation the operator 1 (1 ± γ 5 ) has a simple structure. and P+ + P− = 1. The fact that uν in (12. λ = 1) = 0 φ+ u(p. WEAK INTERACTIONS Massless positiveenergy spinors are then u(p. the terms right. P− uν (λ = 1) = 0 0 P+ uν (λ = −1) = φ− 0 (12. P+ P− = 0. In fact.21) So far as one can tell from the form of the interaction. λ = −1) = φ− 0 (12.22) then the parity conjugation operation (12.8) forces us to conclude that P− uν = P− γ 0 uν = 6 0 0 (12. For example. These operate like any Godfearing projection 2 operators. Let’s be brief 2 1 and write P± ≡ 2 (1 ± γ 5 ). Introduction to High Energy Physics. it might be that neutrinos come in two helicity states. neutrinos only interact weakly and 1 always with the 2 (1 − γ 5 ) in front of the spinor.14) is always multiplied by P− means that only lefthanded neutrinos interact. AddisonWesley 1987 . 7 The fact that the lefthanded rather than righthanded states interact was established in a classic experiment by Goldhaber et al (1958) described in Perkins.
Equations (12. The Dirac equation in our new basis including the mass term is −m E−σ·p φ =0 (12. all the m’s cancel.27) m. Because of our normalization ¯ condition. I meant that the right side of (12. Helicity and Electrons This simple association of chirality with helicity in (12. The statement is not quite correct.25) m The complete spinors are u(λ) = C φλ E+λp m φλ = √1 2m E − λp φλ E + λp φλ (12. The point is that all physical results depend on bilinear products like uu.24) become 1 (E + λp)φλ = χλ (12.24) E+σ·p −m χ (E + σ · p)φ = mχ (E − σ · p)χ = mφ It’s convenient to take φ and χ to be helicity eigenstates.28) which vanishes in the limit m → 0 or equivalently the limit p this limit 1 0 P− u(λ = 1) = √ 2m 0 √ 1 2E φ− P− u(λ = −1) = √ 0 2m 8 Now you see why the neutrino mass is such a problem.3. √ 9 It is still meaningful to take the limit m → 0 despite the factor of 1/ 2m.19) does not hold for massive particles.3 Chirality.8 12.23) is zero without any remaining small components. .26) The normalization constant C is chosen according to our convention that uu = 1.9 In (12. Now look at the relative sizes of the upper and lower components. ¯ E − p E + p = E 2 − p2 m m = ≈ E + p E + p 2E (12.12. HELICITY AND ELECTRONS 213 When I said that parity was violated maximally. CHIRALITY.
31) (Remember that v v = −1. and then only because of a complicated set of fortuitus circumstances.29) so that in this limit.26) becomes 1 − E + λpχλ − m (E + λp)χλ = √1 .59) shows that the chargeconjugate state must also have negative helicity. so the helicity of a parityconjugated state is opposite to that of the original state. According to (12. u1 γ µ P− u2 = P− u1 γ µ u2 = P− u1 γ µ P− u2 ¯ ¯ ¯ (12. WEAK INTERACTIONS The point is that for p m. Well – almost. Two minuses make a plus.30) ˆ χ=σ·pχ The signs have changed. Negative helicity antineutrinos don’t exist. One can construct the spinors for antiparticles by repeating the procedure with E replaced by −E in (12.) For p ¯ m. It is so weak that it only manifests itself in the decays of the neutral K and B mesons. only the lefthanded components of all fermions participate in the interaction. CP violation is one of the most profound discoveries of particle physics in recent years. It is easy to show that if u1 and u2 are any two spinors. the negative helicity components are suppressed by m/2E. It is believed to be the origin of the preponderance of matter over antimatter in the universe. The missing helicity state in the neutrino wave function also causes charge conjugation symmetry to be violated. right? Weak interactions are invariant under CP . Only righthanded antineutrinos participate in weak interactions.20). neutrinos have negative helicity. The combination of parity conjugation and charge conjugation is a diﬀerent story. Equation (12. or if they do.17) becomes ˆ φ = −σ · p φ (12. The helicity operator σ · p is odd under parity conjugation.24). . Equation (9. the P− in the interaction suppresses the positive helicity components by the factor m/2E.16) and (12. Massless righthanded antiparticles have negative chirality. There is a “superweak” component to the weak interactions that does violate CP. they don’t participate. but we have just showed that the helicities of massless antiparticles are opposite to those of particles.214 CHAPTER 12. v(λ) = C 2m χλ E − λpχλ (12. Equation (12.
First of all. requires a neutrino and an antineutrino in the ﬁnal state. The complete set of quantum numbers is given in Table 13. since γ’s are not leptons. so it must carry a tau number assigned in the same way. π 0 → µ+ + e− µ− → e− + γ µ− → e+ + e− + e− µ +p→p+e − − (12. (There is also a tau lepton.4. Similarly. so one looks for the reaction ν +37 Cl →37 A + e− ¯ This can be identiﬁed since 37 Ar is a radioactive gas that can be separated from the liquid and its radioactivity measured. LEPTON NUMBER 215 12.14) that the lepton current should 1 be lµ = ue γ µ 1 (1−γ 5 )vν rather than say lµ = ue γ µ 2 (1−γ 5 )uµ . but it’s been done. There must be electron neutrinos that carry electron number just as electrons do. What are we to make of µ− → e− + ν + ν? We ¯ have to acknowledge that neutrinos also carry lepton number and that all neutrinos are not created identical.) Reactions like e+ + e− → γ + γ are allowed. but none have ever been observed.1. What is “wrong” with these reactions? The (almost) obvious guess is that these leptons carry some sort of additive quantum numbers that in (12.32) just don’t add up. This is a heroic experiment. both of which ¯ ¯ 2 are allowed by Feynman’s rules. Now it’s clear ¯ . so there must be some measure of “muonness” and “electronness” that must be conserved. and the reaction just doesn’t occur.32) ν + n → p + e− ¯ The last reaction seems impossible to realize. and so all their lepton numbers are zero. but there’s a trick. I also ¯ stated (again without explanation) (12.12. Reactors produce mostly antineutrinos. as well as muon neutrinos and tau neutrinos. Park a swimming pool full of cleaning ﬂuid in front of a nuclear reactor. muons can’t decay into electrons. As a convention. How is one to know about these things? The following reactions and decays all conserve energy and charge. negative muons have muon number = +1. and positrons have electron number = −1.4 Lepton Number I mentioned (without explanation) that ordinary muon decay µ− → e− + ν + ν . we say that negative electrons have an electron number = +1. and positive muons have muon number = −1. The correct way to specify muon decay is µ− → e− + νe + νµ .
that both sides of the reaction have nµ = 1 and ne = 0. Antiparticle lepton numbers have the opposite sign. The complete matrix .4 I have illustrated muon decay using solid lines to represent the ﬂow of muon number and dotted lines to represent electron number current. eνµ νe µ− Figure 12.4: Muon decay µ− → e− + νµ + νe .216 e− 1 0 0 CHAPTER 12. Conservation of charge means that you can trace the current along this path following the arrows without lifting your pencil from the page. It represents an electron current that “ﬂows” up the electron leg of the diagram and down the positron leg.2. You will recall that the conservation of charge was a key requirement in constructing currents for electromagnetic α currents. In the same way lepton number represents a kind of “charge” that must be conserved. WEAK INTERACTIONS νe 1 0 0 µ− 0 1 0 νµ 0 1 0 τ− 0 0 1 ντ 0 0 1 ne nµ nτ Table 12. All other particles have zero lepton number.1: Standard lepton number assignments.5) illustrated in ¯ Figure 13. In Figure 13. The dotted lines represent electron number current. Lepton numbers also resolve the question raised above regarding the construction of lepton currents. The solid lines represent the ¯ ﬂow of muon number current. the term je = ve γ α ue in (12. Take for example.
12. LEPTON NUMBER element for muon decay is then 4G M = √ uν γ α 1 (1 − γ 5 )uµ ue γα 1 (1 − γ 5 )uµ ¯ ¯ 2 2 2 217 (12.33) .4.
218 CHAPTER 12. WEAK INTERACTIONS .
” The constant of proportionality will be allimportant. ∝ means “proportional to. 1 j µ (lepton) ∝ u(ν)γ µ 2 (1 − γ 5 )u(e) ¯ (13.” in the sense that the currents change charge in the process of the interaction. the strength of this reaction is almost the same as the purely leptonic process µ → e + ν + ν . Until the 1970’s it was believed that there were no neutral current interactions. ordinary neutron decay. Of course that is important when doing an exact calculation.1) ∝ ν L γ µ eL ¯ j µ (quark) ∝ u(u)γ µ 1 (1 − γ 5 )u(d) ¯ 2 ∝ uL γ µ dL ¯ Notice the condensed notation: uL is the lefthanded projection of the up ¯ quark. By the same token. The currents in (13.2) . n → p + e− + ν . We can construct weak VA ¯ currents for these processes as follows. At this point I am not concerned with whether this is a utype or a vtype spinor. notice that the projection operator. When allowance is made for the diﬀerences ν in mass and phase space. ¯ amounts to a process in which a down quark beta decays into an up quark and a lepton pair. Put it another way. but you have Feynman’s rules to ﬁgure that out.1 Weak interaction of quarks From the point of view of quarks. 219 (13. Finally. they couple to charged intermediate vector bosons. Here I am trying to present an idea with a minimum of notational complications.Chapter 13 The Electroweak Theory 13.2) are both “charged currents. 1 (1 − 2 γ 5 ) projects the lefthanded components of both spinors as shown in class.1) and (13. but not now. d → u+e− +¯.
so the correspondinq currents are j µ (leptonic neutral current) ∝ ν γ µ 1 (1 − γ 5 )ν ¯ 2 j µ (neutral hadronic current) ∝ q γ µ 2 (cV − cA γ 5 )q ¯ 1 (13.5) (13. and the relationship is cemented by the marriage of the righthanded piece of the electron current and the righthanded piece of the weak.2 Weak Isospin The electroweak theory per se does not involve quarks. 13. and these are forbidden. hadronic current.and lefthanded components. and X stands for some hadronic ”stuﬀ. It does this by coupling to a negative .4) The constants cV and cA have been determined experimentally with inelastic neutrino scattering. neutral. but only the lefthanded components participate in the weak interactions. (q stands for any quark.and lefthanded components. THE ELECTROWEAK THEORY since processes like K0 → µ+ µ− seem to be forbidden.) There is no reason to believe that these interactions have the pure VA form. weak neutral currents have both right. Or so it seems. Electrons. To bring this marriage about I will deﬁne + J µ ≡ ν L γ µ eL ¯ − + Jµ = (Jµ )† = eL γµ νL ¯ (13. the following facts established in the previous section are foundational. nonetheless. The K0 carries one unit of strangeness. these are examples of the reaction ν + q → ν + q. also have both right. With the advent of highenergy neutrino beams it became possible to see reactions such as νµ + N → νµ + X. of course. neutral. As a consequence of (13.220 CHAPTER 13. The underlying idea of the electroweak theory is that the weak and electromagnetic interactions are closely related.3) (13. so this is an example of a strangenesschanging interaction. hadronic currents that they couple to have both right. however.and lefthanded components.” In terms of quarks.4). • The weak. • Weak leptonic currents and weak charged hadronic currents have only lefthanded components.6) The notation J + (J − ) means that the initial lepton gains (loses) one unit of charge in the course of the interaction. where N is some heavy nucleus.
deﬁne the “spinor. The third member is 3 Jµ ≡ χL γµ 1 σ3 χL = 1 νL γµ νL − 1 eL γµ eL ¯ 2 2¯ 2¯ (13.9) This is familiar angular momentum physics in an unfamiliar context. but it can’t be the whole story. Then i Jµ ≡ χL γµ 1 σi χL ¯ 2 ± Jµ ≡ χL γµ σ± χL ¯ (13. In (13. Suppose the “missing” righthanded components of the weak neutral current actually come from the righthanded components of the electromagnetic current.2.12) .” The currents ± Jµ are two components of the triplet. Notice in this particular deﬁnition I have not included the electron charge. The key idea is to split up the electron current into right. WEAK ISOSPIN 221 (positive) intermediate vector boson.10) 3 This Jµ is a weak neutral current. We can combine two spin1/2 doublets to make a spin1 triplet and a spin0 singlet. In this case. To this end.7) e− L and raising and lowering operators.8) The σi ’s are the usual Pauli spin matrices. (13. em ¯ 2 jµ = −¯γµ e = −¯γµ 2 (1 − γ 5 )e − eγµ 1 (1 + γ 5 )e e e 1 (13. This is the idea.11) e is just the electron spinor.13. because the complete neutral weak current has righthanded components. We assume that in addition to charge and the third component of isotopic spin. jµ is a neutral current for reasons explained previously.and lefthanded pieces. the underlying quantity is not spin but something conventionally called ‘weak isospin. Think of χL as a Pauli spinor.” ν χL = . there is a third “good quantum number” called hypercharge or Y . but it plays out in ways that are rather subtle. σ± = 1 (σ1 ± σ2 ) 2 σ+ = 01 00 σ− = 00 10 (13. It is customary to use the formalism of spin or isospin to combine these various currents. The three are related by Q = T 3 + Y /2 (13.11) = −¯R γµ eR − eLγµ eL e ¯ em Despite the fact that it contains electron spinors.
15) (13. Similar assignments can be made for the quarks. In the conventional notation they couple to vector boson ﬁelds as follows −ig J i µ i Wµ − i g Y j 2 µ Bµ (13. two heavy charged bosons. the interactions are mediated by the photon ﬁeld Aµ and the basic interaction vertex is −ieJµ .1 The quantum numbers are given in Table 12.13) Y The new current jµ is called the weak hypercharge current. They acquire mass through a mechanism called spontaneous symmetry breaking.16) 3 Aµ = Bµ cos θW + Wµ sin θW 3 Zµ = −Bµ sin θW + Wµ cos θW 1 There is a strong hypercharge also.1: Weak quantum number assignments According to Noether’s theorem. and the even heavier neutral vector boson. which will be explained in a subsequent chapter.15) as follows ± Wµ = 1 1 2 (Wµ 2 iWµ ) (13. THE ELECTROWEAK THEORY T3 1 2 −1 2 0 0 Q 0 1 1 Y 1 1 2 Table 13. is another constant that must be determined from experiment. The ﬁelds W i and B are assumed to be massless. the three components of Jµ and the hypercharge current jµ . In electroweak theory there are i Y four currents.3 The Intermediate Vector Bosons In conventional QED. 13. The particles that are detected in the laboratory are the photon which is massless. . the W ± ’s. Their ﬁelds are related to those in (13.14) The two new coupling constants g and g will be related to the Fermi coupling constant and the electric charge later on.1.222 lepton νe e− L e− R T 1 2 1 2 CHAPTER 13. for every conserved charge there is a conserved current. The currents in this case must be related by em 3 1 Y jµ = Jµ + 2 jµ (13. The socalled Weinberg angle θW . the Z 0 .
17) can be rewritten 3 −ig cos θW Jµ − sin2 θW em g 3 (jµ − Jµ ) Z µ = −i J NC Z µ 2θ cos W cos θW µ NC 3 em Jµ ≡ Jµ − sin2 θW jµ (13.14) can be rewritten using (13.4 The Feynman Rules Now that the smoke has cleared.15) as g + − 1 2 −ig (J 1 )µ Wµ + (J 2 )µ Wµ = −i √ (J µ Wµ + J µ† Wµ ) 2 (13.9). and (13. (13. Typically we write the vertex function −i g γ µ 1 (cV − cA γ 5 ) cos θW 2 (13. handed components form Jµ µ 13. coupling to the weak charged current. The interaction term in (13. the W ± ’s. .18) Whatever it is that couples to the Z 0 must be the complete neutral weak NC current. we are left with the following interactions.13.14) is −ig J 3 µ 3 Wµ − i 223 g Y j 2 µ Bµ (13. THE FEYNMAN RULES The neutral current piece of (13.6).13).4. the current coupled to it must be jµ from (13.17) 3 Y 3 Y = −i(g sin θW Jµ + g cos θW jµ /2)Aµ − i(g cos θW Jµ − g sin θW jµ /2)Z µ em Since Aµ is the actual photon ﬁeld. The last term in (13.19) where (13.22) The constants cV and cA can be determined from based on the charge and isotopic spin of the particles that make up J NC . This can only be so if g cos θW = g sin θW = e (13. • Two charged IVB’s. the Z 0 .20) NC em As promised Jµ inherits its righthanded components from jµ and its left3 and j em . coupled to J NC with coupling constant g/ cos θW .21) • One neutral IVB. which I will call Jµ .
224
CHAPTER 13. THE ELECTROWEAK THEORY
• The photon coupling to the electromagnetic current with the usual constant e. The Feynman rules from Section 11.3 require the following additions. 1. At each vertex at which a W ± couples to a charged current, insert the vertex factor g −i √ γ µ 1 (1 − γ 5 ) (13.23) 2 2 At each vertex at which a Z 0 couples to a neutral current, insert the vertex factor g −i γ µ 1 (cV − cA γ 5 ) (13.24) cos θW 2 2. For each virtual W or Z insert the propagator i(−g µν + pµ pν /M 2 ) p2 − M 2 (13.25)
3. For each free W or Z insert the spin polarization tensor εµ (λ). When summing over the polarization state λ use the completeness relation ε(λ)∗ (p)ε(λ) (p) = −g µν + µ ν
λ
pµ pν M2
(13.26)
The other rules are unchanged.
Chapter 14
Symmetry Breaking
Consider an atom with a spherically symmetric potential. Because of the symmetry, the energy eigenstates are independent of the m and ms quantum numbers. Put it another way, the SU (2) symmetry of the Hamiltonian causes these states to be degenerate. If we put the atom in an external magnetic ﬁeld, the degeneracy is removed and the energies become functions of both m and ms . The system is still symmetric about the direction of the ﬁeld, so the imposition of the external ﬁeld has broken the symmetry from SU (2) down to U (1). This sort of symmetry breaking is called the Wigner Mode. There are several other ways to break symmetry that are more relevant to this chapter. Try standing a straight ﬂexible rod upright and press down on the top of it with a force F . Let’s assume that the rod remains straight for some minimal force, but beyond some threshold force Fc the rod “buckles,” and assumes a curved shape. The symmetry is broken, the rod is no longer straight, but there are an inﬁnite number of equivalent states corresponding to the possible orientations of the rod. We can summarize this as follows: 1. A parameter (in this case F ) assumes some critical value. Beyond that value, 2. the symmetric situation becomes unstable, and 3. the ground state is degenerate. The second example is the ferromagnet. The atoms in a ferromagnet interact through a spinspin interaction H=−
i,j
Jij Si · Sj 225
226
CHAPTER 14. SYMMETRY BREAKING
which is a scalar and therefore invariant under rotations. The ground state however, is one in which all the spins are aligned, and this is clearly not rotationally invariant. The direction of spontaneous magnetisation is random, and all the degenerate ground states may be reached from a given one by rotation. The spontaneous magnetisation disappears at high temperature, when the ground state becomes symmetric. It is clear that the general situation here is the same as in our ﬁrst example, the relevant parameter here being T . These two examples exhibit what is known as “spontaneous symmetry breaking. In both cases the system possesses a symmetry (rotation symmetry) but the ground state in not invariant under that symmetry; rather, it changes into one of the other (degenerate) ground states.
14.1
A Simple Example
Consider a classical selfinteracting real scalar ﬁeld φ with a Lagrangian density 1 1 1 (14.1) L = (∂µ φ)(∂ µ φ) − µ2 φ2 − λφ4 2 2 4 We assume λ > 0 so that there is a ﬁnite minimum energy. This Lagrangian is invariant under the discrete transformation φ → −φ. (14.2)
There are two quite diﬀerent cases depending on the sign of µ2 . The potential for µ2 > 0 has a single minimum, V (φ) = 0 at φ = 0. This point corresponds to the ground state, so in quantum mechanical terms φ
0
= 0φ0 = 0
0
Expanding the Lagrangian about φ
to second order gives (14.3)
1 1 L ≈ (∂µ φ)(∂ µ φ) − µ2 φ2 2 2
which is the Lagrangian density of a free scalar ﬁeld of mass µ. The perturbations induced by the φ4 term can be thought of as oscillations about the origin. Now consider the case µ2 < 0. Now the potential has minima at φ
0
=±
−µ2 ≡ ±v. λ
(14.4)
14.1. A SIMPLE EXAMPLE
227
Now there are two degenerate vacuum states. The minima at φ 0 = ±v are equivalent and either may be chosen as the classical ground (vacuum) state of the system. Once this choice has been made, the vacuum is no longer invariant under the transformation (2). This is a new form of symmetry breaking: the Lagrangian is invariant but the vacuum state is not. This is called the Goldstone mode, spontaneous symmetry breaking, or hidden symmetry. Let’s choose φ 0 = +v. (14.5) We can no longer do perturbation theory by expanding the Lagrangian about φ 0 = 0, since the slightest perturbation about this point would send the system plunging down into one of the two minima. It makes more sense to shift the ﬁeld by deﬁning ξ(x) ≡ φ(x) − φ
0
= φ(x) − v.
0
(14.6) = 0, and the La
In terms of this new variable the vacuum state is ξ grangian density is (neglecting constant terms) 1 L = L0 − λvξ 3 − λξ 4 4
(14.7)
We can think of the ξ 3 and ξ 4 terms as perturbations of the freeﬁeld Lagrangian 1 L0 = (∂ξ)2 − λv 2 ξ 2 (14.8) 2 which is the Lagrangian for a free scalar ﬁeld with mass mξ = −2µ2 . (Remember that µ2 < 0.) This is a toy model, of course, but it illustrates many of the features of spontaneous symmetry breaking as it manifests itself in quantum mechanics.1 1. There is a nonzero expectation value of some ﬁeld in the vacuum state. 2. The resulting classical theory has a degenerate vacuum, with the choice among the equivalent vacuum states completely arbitrary. 3. The transition from a symmetric vacuum to a degenerate vacuum typically occurs as a phase transition as some order parameter (µ2 in the example) is varied.
1 I am copying almost verbatim from Mike Guidry, Gauge Field Theories John Wiley & Sons (1991)
228
CHAPTER 14. SYMMETRY BREAKING
4. The chosen vacuum state does not possess the same symmetry as the Lagrangian. 5. After we have expanded the Lagrangian around the chosen vacuum, the original symmetry is no longer apparent. The degenerate vacua are related to each other by symmetry operations, which tells us that the symmetry is still there, but it is not manifest; it is hidden. 6. The masses of the particles appearing in the theory with and without the spontaneous symmetry breaking may diﬀer substantially. We say that the masses have been acquired spontaneously in the latter case. 7. Once the theory develops degenerate vacua the origin becomes an unstable point. Thus the symmetry may be “broken spontaneously” in the absence of external intervention.
14.2
Goldstone Bosons
The transformation (2) was discrete. The next generalization makes use of a continuous global gauge transformation. For this we will need a complex or nonHermitian ﬁeld. In previous chapters we used Hermitian scalar ﬁelds for spin zero particles. In terms of second quantization, this means that the particle is its own antiparticle. If the particle caries charge or some other quantum number that is diﬀerent for particles and antiparticles, then the ﬁeld must be nonHermitian (as is the case with electron ﬁelds). Take the Lagrangian to be L = (∂µ )† (∂ µ φ) − µ2 φ† φ − λ(φ† φ)2 (14.9)
The λ terms represents self interaction. Under normal circumstances, we would regard µ as a mass. Here µ2 is just a parameter, since we are going to make µ2 < 0. This Lagrangian is invariant under the group of global gauge transformations, φ(x) → φ (x) = eiΛ φ(x). (14.10) Using polar coordinates ρ2 = φ† φ, we may identify a potential V (ρ) = µ2 ρ2 + λρ4 . (14.11)
As before, set µ2 < 0. The minima now occur in the complex φ plane on a circle of radius −µ2 φ2 = ≡ a2 (14.12) 2λ
14.2. GOLDSTONE BOSONS
229
The shape of the potential has been likened to a Mexican hat or perhaps the bottom of a champagne bottle. There is now an inﬁnity of degenerate ground states, corresponding ro diﬀerent positions on the ring of minima in the complex φ plane, and the symmetry operation (10) relates one to another. In the quantum theory, where φ becomes an operator, this condition refers to the vacuum expectation value of φ.  0φ0 2 = a2 . (14.13)
We could choose any point in the minimum and expand the Lagrangian about that point. For simplicity, we choose the real axis Re(φ) = a, and expand around it to investigate the spectrum. It is convenient to work in polar “coordinates” φ(x) = ρ(x)eiθ(x) , (14.14) so the complex ﬁeld φ is expressed in terms of two real scalar ﬁelds ρ and θ. In quantum language, we choose the vacuum state 0φ0 = a where a is real; then 0ρ0 = a, Now let us put φ(x) = [ρ (x) + a]eiθ(x) (14.17) We regard ρ and θ as “small” displacements from the point φ = a. Substitute (17) into (9)and keeping only the quadratic terms gives L = (∂µ ρ )(∂ µ ρ )+(ρ +a)2 (∂µ θ)(∂ µ θ)−4λa2 ρ 2 +O(ρ 3 )+ constants (14.18) The coeﬃcient of the square of the ﬁeld in a freeﬁeld Lagrangian is the mass squared of the corresponding particle. In this case the ρ ﬁeld has a mass given by m2 = 4λa2 (14.19) ρ There is no term in θ2 , so θ is a massless ﬁeld. As a result of spontaneous symmetry breaking, what would otherwise be two massive ﬁelds (the real and imaginary parts of φ), become one massive and one massless ﬁeld. The diﬀerence between the two coordinates is easy to visualize. It requires no energy to displace a particle in the θ direction, but it does cost energy to displace ρ against the restoring force of the potential. The θ particle is known 0θ0 = 0 (14.16) (14.15)
230
CHAPTER 14. SYMMETRY BREAKING
as a Goldstone boson. This phenomenon is general: spontaneous breaking of a continuous symmetry entails the existence of a massless particle. This is the Goldstone theorem. Try the same thing in cartesian coordinates. Introduce two real ﬁelds φ1 and φ2 . φ1 + iφ2 φ=a+ √ , (14.20) 2 so that the vacuum expectation values of both ﬁelds are zero. Then √ 1 1 λ L = (∂µ φ1 )2 + (∂µ φ2 )2 −2λa2 φ2 − 2λφ1 (φ2 +φ2 )− (φ2 +φ2 )2 (14.21) 1 1 2 2 2 2 4 1 The φ2 ﬁeld is massless, but the φ1 has a mass squared of 4λa2 , the same as (19). There is an analogy here to spin waves in a ferromagnet. Spin waves are periodic variations in M . Because the forces in a ferromagnet are of short range, it costs very little energy excite such a wave. Such a wave could carry zero energy in the limit of inﬁnite wavelength, i.e. it’s a massless excitation.
14.3
A More Abstract Look
There is a profound result regarding symmetries known as Noether’s theorem, to the eﬀect that there is a conserved current associated with each generator of a continuous symmetry. The proof is simple: consider a Lagrangian composed of several ﬁelds, which we call φa . Our symmetry transformation changes each ﬁeld ay an inﬁnitesimal δφa . since L does not change, we have 0 = δL = The equations of motion are δL δL = ∂µ δφa δ(∂µ φa ) so we can combine the two and get 0 = ∂µ If we deﬁne Jµ ≡ δL δφa δ(∂µ φa ) δL δφa δ(∂µ φa ) (14.22) δL δL δφa + δ(∂µ φa ) δφa δ(∂µ φa )
(14.23)
14.3. A MORE ABSTRACT LOOK
231
Then (22) says that ∂µ J µ = 0. (We are also assuming that repeated a’s are summed.) (We have already used this argument to get the electron current in Dirac theory.) To make this more concrete, imagine that the N ﬁelds make up a multiplet, φ1 φ2 φ ≡ . , . . φN and the Lagrangian contains terms like (∂φ)2 = (∂µ φa )(∂ µ φa ), and φ2 = φa φa , arranged in the form L= 1 (∂φ)2 − m2 φ2 − c2 (φ2 )2 − c3 (φ2 )3 − · · · 2 (14.24)
(Unless otherwise stated, repeated indices are summed.) This Lagrangian will be invariant under a “rotation,” φa → Rab φb , where R is a member of the rotation group SO(N ). (This is sometimes called an “internal symmetry.”) We can write R = eiθ·X where the X i are the group generators. Under a inﬁnitesimal transformation φa → Rab φb ≈ (1+θi X i )ab φb , or in other words, i we have the inﬁnitesimal change δφa = θi Xab φb . Equation (23) assures us that the N (N − 1)/2 currents
i i Jµ = (∂µ φa )Xab φb
(14.25)
are conserved. So there is one conserved current, J i , for each of the N generators of the symmetry group. For every conserved current there is a conserved “charge” Q≡ d3 xJ 0 = d3 x δL δφa δ(∂0 φa ) (14.26)
We notice that δL/δ(∂0 φa ) = πa , the canonically conjugate ﬁeld. The equaltime commutation relation then gives the elegant result i[Q, φa ] = δφa (14.27)
We are now in a position to prove the Goldstone theorem. The charges deﬁned by (26) are conserved, they are constant in time. Therefore, [H, Q] = 0 (14.28)
29) We make use of the fact that the momentum operator P satisﬁes [P . t)0 d3 xeik·x J 0 (x. J 0 (x.30) (14.31) d3 xeik·x [P . those for which Qi 0 = 0. t)0 (14. or in other words Q0 = 0. it’s a zeroenergy eigenstate of the Hamiltonian. eiθQ 0 = 0 . Then P ψ(k) = = d3 xeik·x P J 0 (x. If it should happen that H0 = c. HQ0 = [H. In the limit that k goes to zero. Q0 ? Clearly. SYMMETRY BREAKING Let the vacuum (or ground state in quantum mechanics) be denoted by 0 .32) Evidentally ψ(k) describes a particle with momentum k. t)]0 = −i Integration by parts then gives P ψ(k) = kψ(k) (14.e. Q]0 = 0 I claim it corresponds to a massless particle. In general there will be as many conserved charges as there are generators in the symmetry group. This relationship between momentum and energy is the signature of a massless particle.33) We know from (29) that Q0 is a state of zero energy. where φ(x) could be any ﬁeld operator including J(x). For each one of these charges that does not leave the vacuum invariant. In the previous examples. (30) becomes k→0 lim ψ(k) = Q0 (14. Q0 = 0. t)0 (14. So what is this state. We can always write H0 = 0.) Normally the vacuum is invariant under the symmetry transformation. i. (The vacuum must be an eigenstate of the Hamiltonian. φ(x)] = −i φ(x). we could always redeﬁne the Hamiltonian H → H − c. there will be one of these massless Goldstone bosons.232 CHAPTER 14. i. .e. Consider the state ψ(k) = d3 xeik·x J 0 (x. we had to choose a speciﬁc vacuum that was not invariant under the symmetry transformation.
We can turn this into Schrodinger’s equation as follows. (14. t) 2m First note that ∂Φ† ∂Φ 1 ∂ ∂ − m2 Φ† Φ = (im + )ϕ† (−im + )ϕ − m2 ϕ† ϕ ∂t ∂t 2m ∂t ∂t ≈ i 2 ϕ† ∂ϕ ∂ϕ† − ϕ ∂t ∂t → iϕ† ∂0 ϕ (14. Then (35) becomes 2 ∂ϕ − ϕ=i . we will have to have a lowenergy version of the φ4 theory discussed previously. Start with the Lagrangian L = (∂Φ)† (∂Φ) − m2 Φ† Φ − λ(Φ† Φ)2 When λ = 0 this gives the KlineGordon equation (∂ 2 + m2 )Φ = 0.34) (14. Spontaneous symmetry breaking can occur in solids and liquids. superconductivity and superﬂuidity are outstanding examples of the mechanism.35) (14.38) . 1 Φ = √ e−imt ϕ(x. The same technique works on the Lagrangian.36) ϕ The lowenergy approximation consists in setting ∂ 2 ϕ/∂t2 = 0.39) (14. In order to study these examples. but it can happen at very low energy.37) 2m ∂t which is Schrodinger’s equation.4 NonRelativistic Field Theory Field theory is necessary whenever particles are created and destroyed. In fact. This is bound to happen at relativistic energies. NONRELATIVISTIC FIELD THEORY 233 14. t) ∂ ∂Φ = e−iωt −im + ∂t ∂t ∂2Φ ∂ = e−iωt −im + ∂t2 ∂t ϕ 2 (14. Deﬁne Φ = e−iωt ϕ(x.14. but we must be careful about the normalization.4. if the “particles” represent some collective phenomena.
number is conjugate to phase angle. (Relativistic is beautiful. δ(∂0 θ) so the equaltime commutation relation is [ρ(x.42) ρ and θ should be thought of as independent ﬁelds. Nonrelativistic is ugly. What sort of interactions do these Lagrangians describe? There is a hint in (42). Let’s use the generating functional formalism from Fall quarter to calculate the propagator for a Lagrangian such as (34).46) In other words.43) (14.) It’s often useful to write the ﬁelds in terms of polar coordinates.234 CHAPTER 14. i. t)] = iδ(x − x ) We can normalize the wave function so that N= then [N. We can learn more about this with a complicated argument from Zee I.44) ρ(x. In this case the canonical momentum is δL = −ρ. The ﬁnal result is L = iϕ† ∂0 ϕ − 1 ∂i ϕ† ∂i ϕ − g 2 (ϕ† ϕ)2 2m (14. SYMMETRY BREAKING The last step required an integration by parts. just like ϕ and ϕ† .e. θ] = i (14. t) dx = number of bosons (14.41) The Lagrangian is i 1 1 L = ∂0 ρ − ρ∂0 θ − ρ(∂i θ)2 + (∂i ρ)2 − g 2 ρ2 2 2m 4ρ (14. t).3.45) (14. bosons repel regardless of the sign of λ. The interaction Hamiltonian is Hint = −Lint = g 2 ρ2 Since ρ represents the density of bosons. θ(x .47) .40) where g 2 = λ/4m2 . √ ϕ = ρeiθ (14. we can see that packing bosons together costs energy. Look at Z= D[Φ]ei R d4 xL(Φ) (14.
so presumably the exchange of a σ will produce a force between two Φ’s. To see how this works. We know that the exchange of such a particle will produce a force of range ∼ m−1 . NONRELATIVISTIC FIELD THEORY Last quarter you proved that dxi e i i 2 235 x·A·x+iJ·x = (2πi)N det[A] 1/2 e− 2 i J·A−1 ·J (14.4. although it’s not clear at this 2 Zee calls the transition from (47) to (52) the HubbardStrantonovic transformation. I will have to take his word for it.52) It contains a scalar Φ particle obeying the KleinGordon equation. Using this in (47) gives Z= D[φ]D[σ] exp i d4 x[(∂Φ)2 − m2 Φ2 + 2σΦΦ† + σ 2 /λ] (14.50) The right side of (50) contains the boson interaction term λΦΦ† . make the following substitutions in (49): J → 2ΦΦ† Then 1 ϕ · A · ϕ + J · ϕ → σ 2 /λ + 2σΦΦ† 2 The identity (49) becomes D[σ]ei R d4 x[2σΦΦ† +σ 2 /λ] ϕ→σ A → 2/λ 1 − J · A−1 · J → 2σΦΦ† 2 = e−i R d4 xλ(ΦΦ† )2 (14.51) To put it another way.14. we have a new Lagrangian2 L = (∂Φ)2 − m2 Φ2 + 2σΦΦ† + σ 2 /λ (14. . D[ϕ]ei R 1 d4 x( 2 ϕ·A·ϕ+J·ϕ) ∝ e− 2 i R d4 x J·A−1 ·J (14. the left side represents a σ “particle” coupling to the Φ.49) Our strategy will be to introduce a new “auxiliary” ﬁeld called σ which has the eﬀect of transmitting the repulsive between the bosons.48) We learned to generalize equations such as this by replacing the coordinates with ﬁelds. It is coupled to a σ particle via the 2σΦΦ† term.
56) (14. It seems that 0e−iHT 0 ought to be something like e−iET where E is the energy due to the presence of the two sources acting on one another.58) Have another look at (55). (58) gives us E=− d3 k eik·(x2 −x1 ) (2π)3 k2 + m2 (14. The argument brieﬂy is this: Z[J] = Z[0]eiW [J] = 0e−iH(T −T ) 0 (14. Introduce the Fourier transform Φ(x) = Then d4 x [(∂Φ)2 − m2 Φ2 ] = ˜ d4 p Φ2 (p)(p2 − m2 ) (14.55) W [J] = − 1 2 d4 xd4 yJ(x)J(y)D(x − y) d4 p eip(x−y) (2π)4 p2 − m2 + i (14.236 CHAPTER 14. T /2 W = −T /2 dx0 d3 k eik·(x2 −x1 ) (2π)3 k2 + m2 (14. Substitute them into (56) W =− dx0 dy0 dp0 ip0 (x−y)0 e 2π d3 p eik·(x1 −x2 ) (2π)3 k 2 − m2 + i Integrate over y0 to get δ(p0 ).59) .34 regarding the Φ. In this spirit then −iRT = iW . Since this is just a qualitative argument that should do for now.54) d4 p −ipx ˜ e Φ(p). J(x) = δ (3) (x − x1 ) + δ (3) (x − x2 ) W [J] will contain terms with δ (3) (x − x1 )δ (3) (y − x2 ) and δ (3) (x − x2 )δ (3) (y − x1 ) representing the interaction between the two sources.57) D(x − y) = Let’s put two delta function sources in (56) and see how they aﬀect one another.53) We have learned to extract the propagator from this integral. SYMMETRY BREAKING point what the range and sign may be. then integrate over p0 . To determine these things I will ﬁrst review Zee’s argument from Section I. (2π)2 (14.
The result is 1 −mr E=− e . . Let’s repeat the steps from (52) to (57) σ(x) = d4 p −ipx e σ (p) ˜ (2π)2 d4 p˜ 2 (p) σ (14. (14.60) 4πr Now go back to the Lagrangian (52). We have just analyzed the free Φ ﬁeld.63) We discussed this in the chapter on quantum electrodynamics. this ﬁeld must appear in the Lagrangian through the replacement ∂µ → ∂µ + ieAµ . but what are we to make of the free σ ﬁeld? There is nothing in the Lagrangian but σ 2 /λ. Moreover. where Fµν = ∂µ Aν − ∂ν Aµ . The momentum space propagator is simply +λ. The Fourier transform of this is the delta function.64) and in the kinetic energy term − 1 F µν Fµν . 14. The force is repulsive and zerorange.5 Spontaneous Breaking of Local Symmetry The situation is quite diﬀerent when the broken symmetry is a local or gauge symmetry. (14. The Lagrangian will remain invariant only if there is an additional massless ﬁeld Aµ . 4 Substituting this into (9) gives 1 L = [(∂µ + ieAµ )φ]† [(∂ µ − ieAµ )φ] − µ2 φ∗ φ − λ(φ∗ φ)2 − Fµν F µν (14. this time with the transformation φ → eiΛ(x) φ. Let’s revisit the model from equations (9) and (10).14. (14. and (2) the