You are on page 1of 148

Physics 232A: Quantum Field Theory I

U.C. Berkeley Fall 2012/2013 Semesters

Professor: Petr Hořava


GSI: Kevin Grosvenor

September 3, 2017
b
Contents

1 Problem Set 1 1
1.1 QM Path Integral with Potential, Zee I.2.1 p.16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Wick’s Theorem, Zee I.2.2 p.16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Discussion: Free Particle on a Circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Problem Set 2 7
2.1 E&M Field Theory, Peskin & Schroeder 2.1, p.33 . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 The Complex Scalar Field, P&S 2.2 p.33 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Discussion 1: Alternative Current Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 Discussion 2: Energy-Momentum Tensor by Coupling to Gravity . . . . . . . . . . . . . . . . . . 20

3 Problem Set 3 21
3.1 Lorentz Group, P&S 3.1, p.71 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Lorentz-Invariant Measure, Zee I.8.1 p.69 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Gordon Identity, P&S 3.2 p.72 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.4 Quadratic Shift Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.5 Discussion 1: Cubic Shift Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.6 Discussion 2: Alternative Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

4 Problem Set 4 31
4.1 Advanced and Retarded Propagators, Zee I.3.3, p.24 . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2 Force Law in Arbitrary Dimension, Zee I.4.1 p.31 . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3 Graviton Propagator, Zee I.5.1 p.39 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.4 Discussion 1: Grassmann Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

5 Problem Set 5 41
5.1 Majorana Fermions, P&S 3.4, p.73 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.2 Supersymmetry, P&S 3.5 p.74 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.3 Discussion 1: Majorana Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.4 Discussion 2: SUSY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

6 Problem Set 6 59
6.1 Amplitude of Figure I.7.11 p.58, Zee I.7.1, p.60 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.2 Amplitude of Figure I.7.10 p.57, Zee I.7.2 p.60 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.3 Two-to-Four Meson Processes, Zee I.7.3 p.60 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.4 Decay of a Scalar Particle, P&S 4.2 p.127 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

7 Problem Set 7 65
7.1 Linear Sigma Model, P&S 4.3 p.127 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
7.2 Rutherford Scattering, P&S 4.4 p.129 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
7.3 Discussion: Goldstone Bosons (being eaten up) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

i
ii CONTENTS

8 Problem Set 8 79
8.1 Massless Tree Diagrams, P&S 5.3 p.170 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
8.2 Physical Amplitude, Zee III.1.1 p.168 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
8.3 Sliding Cut-off, Zee III.1.3 p.168 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.4 Discussion: Naturalness and Renormalizability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

9 Problem Set 9 89
9.1 Fermion Field Dimension, Zee III.3.1 p.181 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
9.2 Degree of Divergence, Zee III.3.2 p.181 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
9.3 Massless Weisskopf Phenomenon, Zee III.3.3 p.181 . . . . . . . . . . . . . . . . . . . . . . . . . . 90
9.4 Form of Anomalous Magnetic Moment, Zee III.6.3 p.198 . . . . . . . . . . . . . . . . . . . . . . . 92
9.5 Discussion 1: Phi-Fourth Dimensional Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
9.6 Discussion 2: Phi-Cubed Dimensional Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

10 Problem Set 10 95
10.1 Exotic Contributions to g – 2, P&S 6.3 p.210 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
10.2 Diagramatics of Effective Potential, Zee IV.3.4 p.244 . . . . . . . . . . . . . . . . . . . . . . . . . 100
10.3 Discussion 1: Coleman-Weinberg Effective Potential . . . . . . . . . . . . . . . . . . . . . . . . . 101
10.4 Discussion 2: A Lower Bound on the Higgs Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

11 Problem Set 11 109


11.1 The Gross-Neveu Model, P&S 11.3 p.390 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
11.2 Discussion 1: Clarification Regarding the Effective Potential . . . . . . . . . . . . . . . . . . . . . 116
11.3 Discussion 2: More on Gross-Neveu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

12 Problem Set 12 119


12.1 QED plus Yukawa, P&S 7.3 p.257 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
12.2 Massive Axial Anomaly, Zee IV.7.4 p.279 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

13 Final Exam 135


13.1 Scalar Electrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
13.2 Non-relativistic Lifshitz Scalar Field Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
13.3 Asymptotic Freedom in 5+1 Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Chapter 1

Problem Set 1

1.1 QM Path Integral with Potential, Zee I.2.1 p.16


Verify (5) p.11 Zee: with H = p̂2 /2m + V (q̂),
Z RT
−iHT dt[ 12 mq̇ 2 −V (q)]
hqF |e |qI i = Dq(t) ei 0 .

SOLUTION: (Thanks to Lenny (’12) and Brian S. (’13) for presenting their solutions.)

We must eventually calculate


p̂2
hqj+1 |e−iδtH |qj i = qj+1 exp −iδt

 
2m + V (q̂) qj .

In circumstances like these, involving exponentials of sums of operators, the Zassenhaus formula is often useful.
This formula is related to the Baker-Campbell-Hausdorff formula and says
1 2 1 3
e(X+Y ) = eX eY e− 2  [X,Y ]
e6 (2[Y,[X,Y ]]+[X,[X,Y ]])
··· . (1.1.1)

Let us apply this with  = −iδt, X = p̂2 /2m and Y = V (q̂). Since we are interested in the continuum limit,
δt → 0, we drop all of the terms on the right side of Eqn. (1.1.1) except for the first two terms. Technically we
should rewrite eX ≈ 1 + X. But, these are equivalent to first order.
2
hqj+1 |e−iδtH |qj i = qj+1 e−iδt p̂ /2m e−iδt V (q̂) qj .

Once the exponential of the potential operator acts on |qj i it simply produces the factor e−iδt V (qj ) . This is no
longer an operator, merely a number, and can be pulled out of the expectation value:
2
hqj+1 |e−iδtH |qj i = qj+1 e−iδt p̂ /2m qj e−iδt V (qj ) .

We already know the result of the remaining expectation value


 1/2
−im qj+1 −qj
2
1
hqj+1 |e−iδtH |qj i = eiδt 2 m δt e−iδt V (qj ) . (1.1.2)
2πδt

The total propagator is then


  N2 NY
−1 Z 
−im PN −1  1 qj+1 −qj
2 
−V (qj )
hqF |e −iHT
|qI i = dqk eiδt j=0 2 m δt . (1.1.3)
2πδt
k=1

In the continuum limit, this becomes


Z RT
dt[ 12 mq̇ 2 −V (q)]
hqF |e−iHT |qI i = Dq(t) ei 0 . (1.1.4)

1
2 CHAPTER 1. PROBLEM SET 1

The boundary conditions, q(0) = qI and q(T ) = qF , are taken for granted.

Note: We could have taken X = V (q̂) and Y = p̂2 /2m in (1.1.1) instead. In this case e−iδt V (q̂) would act on
hqj+1 | instead of |qj i. The result is that everywhere we had V (qj ) before is now changed to V (qj+1 ). However,
the difference between V (qj+1 ) and V (qj ) can be approximated as V (qj+1 ) − V (qj ) ≈ V 0 (qj )(qj+1 − qj ). Recall
that the paths picked out in the path integral are
 those with Hausdorff dimension 2, for which qj+1 − qj ∼ (δt)1/2 .
Therefore, exp −iδt V (qj+1 ) − V (qj ) ∼ exp −iV 0 (qj ) (δt)3/2 , which gives corrections of order larger than 1
 

and thus irrelevant.

1.2 Wick’s Theorem, Zee I.2.2 p.16


Derive (24) p.15 Zee:
X
A−1 · · · A−1
 
hxi xj · · · xk xl i = ab cd
, (1.2.1)
Wick

where
1
dx e− 2 x·A·x xi xj · · · xk xl
R
hxi xj · · · xk xl i = 1 . (1.2.2)
dx e− 2 x·A·x
R

Note that {a, b, . . . , c, d} is a permutation of the set {i, j, . . . , k, l}.

SOLUTION: (Thanks to Jackie (’12) and Eric D. (’13) for presenting their solutions.)

I will take to heart the quotation in the preface to the second edition of Zee: “It is often deper to know why
something is true rather than to have a proof that it is true.” With regards to this problem, a calculation of
hxi xj i and hxi xj xk x` i is sufficient to give one a sense of how Wick’s theorem works. It is actually relatively
easy to extend the first two methods to a rigorous inductive proof; the third is trickier. However, for the third
method, one can recognize that half of the J derivatives have to act on the exponential and half have to act on
the J’s that are pulled down, as suggested by Ken.

The point is that we need to figure out a way to pull down factors of x by taking various derivatives. I can think
of three ways to do this: (1) Take derivatives of the unsourced partition function, Z, (i.e. J = 0) with respect
to x; (2) Take derivatives of the unsourced partition function with respect to A; and (3) Take derivatives of the
sourced partition function with respect to J and then take the J = 0 limit.

Method 1: Taking derivatives with respect to x.

Note that, given implicit summation over repeated indices, and using the fact that A is a symmetric matrix,
 
∂ 1 1 1
− x · A · x = − Aab xb − Aca xc = −Aab xb , (1.2.3)
∂xa 2 2 2

From this, it follows that


∂ − 1 x·A·x 1
e 2 = −Aab xb e− 2 x·A·x . (1.2.4)
∂xa

Left-multiplying this by −A−1


ia (and summing over the repeated index a, of course) yields

X ∂ − 1 x·A·x 1
− A−1
ia e 2 = xi e− 2 x·A·x . (1.2.5)
a
∂xa
1.2. WICK’S THEOREM, ZEE I.2.2 P.16 3

Now, everywhere we see the RHS of the above equation, we will replace it with the LHS. For instance,
Z
1 1
hxi xj i = dx e− 2 x·A·x xi xj
Z
Z
1 X −1 ∂ − 1 x·A·x
=− Aia dx xj e 2
Z a ∂xa
Z
1 X −1 ∂xj − 1 x·A·x
= A dx e 2
Z a ia ∂xa
Z
1 1
= A−1 ij dx e− 2 x·A·x
Z
= A−1
ij . (1.2.6)

We integrated by parts to the get the third line. We dropped the boundary term because it vanishes by virtue
of the Gaussian form of the integrand.
The extension of this argument to the case of four x’s is straightforward. In this case, after the integration
by parts step, ∂x∂ a now acts on xj xk x` instead of only xj . That produces exactly the three terms that you want
depending on which x factor the derivative acts.

Method 2: Taking derivatives with respect to A.

Recall that the sourced partition function is given by


1/2
(2π)N
Z 
1 1 −1
Z[J] ≡ dN x e− 2 x·A·x+J·x = e 2 J·A ·J
. (1.2.7)
det A

With Z ≡ Z[0], as before, observe that taking a derivative with respect to Aij pulls down a factor of xi xj :
 
1 ∂
hxi xj i = − Z. (1.2.8)
Z ∂Aij

We will need to know how to take the derivatives of det A and A−1 . These are given by

∂ det A ∂Aba
= A−1
ab det A = (A−1 −1 −1
ji + Aij ) det A = 2Aij det A, (1.2.9a)
∂Aij ∂Aij
∂A−1 ∂Acd −1
ab
= −A−1
ac A = −A−1 −1 −1 −1
ai Ajb − Aaj Aib . (1.2.9b)
∂Aij ∂Aij db

Then, Eqn. (1.2.8) becomes

1 (2π)N/2
  
1 ∂ det A
hxi xj i = − −
Z 2 (det A)3/2 ∂Aij
N 1/2
 
1 (2π)
= A−1
ij
Z det A
= A−1
ij . (1.2.10)

Now, we can work out the next level up:


  
1 ∂ ∂
hxi xj xk x` i = − − Z
Z ∂Ak` ∂Aij
 
1 ∂
A−1

= − ij Z
Z ∂Ak`
∂A−1
 
1 ij
= A−1
ij A −1
k` − Z
Z ∂Ak`
= A−1 −1 −1 −1 −1 −1
ij Ak` + Aik Aj` + Ai` Ajk . (1.2.11)
4 CHAPTER 1. PROBLEM SET 1

I’ll leave it up to you to consider what happens when two of the indices are equal. One must be a little more
careful, as pointed out by Han Han. However, the result is unchanged.

Method 3: Taking derivatives with respect to J.

Taking a derivative with respect to J will pull down one factor of x at a time. Thus,
1 ∂ 2 Z[J]
hxi xj i =
Z[J] ∂Ji ∂Jj
Z ∂ 1 −1
Ja A−1 2 J·A ·J

= aj e
Z[J] ∂Ji
= Ja A−1 −1 −1
ai Jb Abj + Aij
J=0
−−−→ A−1
ij . (1.2.12)
Now for the four-point correlation function:
1 ∂ 4 Z[J]
hxi xj xk x` i =
Z[J] ∂Ji ∂Jj ∂Jk ∂J`
Z ∂2   1 J·A−1 ·J 
= Ja A−1 −1 −1
ak Jb Ab` + Ak` e
2 . (1.2.13)
Z[J] ∂Ji ∂Jj
2
At this point, note that to get terms that survive in the J = 0 limit, we have two options: (1) ∂J∂i ∂Jj acts on
Ja Jb ; or (2) we drop the term with Ja Jb and one of the derivatives acts on the exponential and the other on the
J that is dropped down as a result. Thus, with · · · denoting the terms that vanish when J = 0,
Z ∂  −1  1 J·A−1 ·J 
hxi xj xk x` i = Ajk Jb A−1 −1 −1 −1 −1
b` + Ja Aak Aj` + Ak` Ja Aaj e
2 + ···
Z[J] ∂Ji
Z  1 J·A−1 ·J
= A−1 A−1 + A−1 −1 −1 −1
ik Aj` + Aij Ak` e
2 + ···
Z[J] i` jk
J=0
−−−→ A−1 −1 −1 −1 −1 −1
ij Ak` + Aik Aj` + Ai` Ajk . (1.2.14)

1.3 Discussion: Free Particle on a Circle


Before we discuss the free particle on a circle, let us review the free particle on a line. We would like to know
the probability amplitude that a free particle of mass M is at some final position xf at some final time tf given
that it was initially at position xi at some initial time ti . This is a straightforward quantum mechanics problem.
In the Schrödinger picture, the initial state is |xi i and the final state is |xf i. The initial state is evolved through
time using the time evolution operator, e−iĤ∆t , where ∆t = tf − ti and Ĥ = p̂2 /2M . Therefore, the desired
amplitude is
hxf , tf |xi , ti i = hxf |e−iĤ∆t |xi i. (1.3.1)
We would like to express the states in the momentum eigenbasis since the Hamiltonian is a function of p̂ and not
of q̂. We write
Z Z Z Z
dp dp −ipxi dk dk ikxf
|xi i = |pi hp|xi i = e |pi , hxf | = hxf |ki hk| = e hk| . (1.3.2)
2π 2π 2π 2π
Then,
p2 ∆t
ZZ Z  
dk dp i(kxf −pxi ) −iĤ∆t dp
hxf , tf |xi , ti i = e hk|e |pi = exp −i + ip∆x . (1.3.3)
2π 2π | {z } 2π 2M
e−ip2 ∆t/2M 2πδ(k−p)

We have done this integral before when ∆t and ∆x were “infinitesimal”. There is no need to break up the
evolution into many steps because the Hamiltonian is just quadratic. The result of the above Gaussian integral
is thus r
1 (∆x)2
 
M
hxf , tf |xi , ti i = exp i M . (1.3.4)
2πi∆t 2 ∆t
1.3. DISCUSSION: FREE PARTICLE ON A CIRCLE 5

When we compactify the line to a circle, let us call the coordinate along the circle θ instead of x. The only
difference now is that not all plane waves are allowed momentum states: only those with integer momentum are
allowed by the periodic boundary conditions, θ ∼ θ + 2π. Customarily, these plane wavefunctions are denoted
eimθ where m ∈ Z (hence the use of M for the mass). Therefore, instead of Eqn. (2.1.6), we get
m2 ∆t
 
1 X
hθf , tf |θi , ti i = exp −i + im∆θ . (1.3.5)
2π 2M
m∈Z

The Gaussian integral was relatively straightforward to perform. This sum is much harder. We will rewrite it
in terms of a slightly different sum, which will be easier to interpret. First, let us perform the following trivial
(admittedly rather unmotivated) manipulation:
p2 ∆t
Z  
dp X
hθf , tf |θi , ti i = exp −i + ip∆θ δ(p − m). (1.3.6)
2π 2M
m∈Z

The point of this is that we can now replace the sum over m using Poisson’s summation formula:
X X
δ(p − m) = e2πinp . (1.3.7)
m∈Z n∈Z

Then, the result is


p2 ∆t
X Z dp  
hθf , tf |θi , ti i = exp −i + ip(∆θ + 2πn) . (1.3.8)
2π 2M
n∈Z

The p integral here is the same as in Eqn. (2.1.6) with the replacement ∆x → ∆θ + 2πn. Therefore,
r
1 (∆θ + 2πn)2
 
M X
hθf , tf |θi , ti i = exp i M . (1.3.9)
2πi∆t 2 ∆t
n∈Z

Okay, now we have calculated the appropriate amplitudes just using quantum mechanics. If you want, you can
perform the standard discretization of the path integral. It is good practice and you should get the same results.
However, we can at least determine the exponential parts of the transition amplitudes using the saddle-point
approximation. Since the Hamiltonian is purely Gaussian in this case, the saddle-point approximation
P iS is actually
exact. Therefore, the result of the path integral of eiS over all paths should be the same as e |c , where the
sum is over all classical paths (solutions to the equations of motion) and eiS is evaluated on the classical paths.
For the free particle, on the line or the circle, the equation of motion is simply that the acceleration is zero.
The classical paths are those that have constant velocity. In other words,
xc (t) = x0 + vt, θc (t) = θ0 + vt, (1.3.10)
where x0 , θ0 , and v are some constants. We only actually care about v, which is determined by the boundary
conditions (e.g. xc (ti ) = xi , etc.) For the line, there is only one solution:
∆x
line: v = . (1.3.11)
∆t
However, for the circle, there are many constant velocity paths, depending on how many times the path winds
around the circle:
∆θ + 2πn
circle: vn = . (1.3.12)
∆t
Now, note that the exponential part of Eqn. (2.1.7) is simply eiS evaluated on the constant-velocity path, where
the velocity is given by Eqn. (1.3.11). Similarly, the exponential part of Eqn. (1.3.9) is eiS evaluated and summed
over all of the constant-velocity paths on the circle with velocities given by Eqn. (1.3.12).
Consider ∆θ very small and zoom in on the region near θi and θf . Sufficiently close, this region will simply
look like a line. The “perturbative” path would be the shortest one between θi and θf . Because you have zoomed
in on this small region, you can imagine that you might be ignorant of the fact that the space eventually curves
back on itself in the shape of a circle and you might miss the other possible paths. They are “non-perturbative”
paths since they go well beyond any small neighborhood of the start and end point. In fact, they explore large
regions of the space and tell you about the global topology of the space rather than just local geometry. In
quantum field theory, the analogous thing would be “non-perturbative” field configurations, which are solutions
to the classical equations of motion. These are also called instantons.
6 CHAPTER 1. PROBLEM SET 1
Chapter 2

Problem Set 2

2.1 E&M Field Theory, Peskin & Schroeder 2.1, p.33


Classical electromagnetism (with no sources) follows from the action
Z
1
S=− d4 x Fµν F µν , where Fµν = ∂µ Aν − ∂ν Aµ . (2.1.1)
4

(a) Derive Maxwell’s equations as the Euler-Lagrange equations of this action, treating the components Aµ (x) as
the dynamical variables. Write the equations in standard form by identifying E i = −F 0i and ijk B k = −F ij .

(b) Construct the energy-momentum tensor for this theory. Note that the usual procedure does not result
in a symmetric tensor. To remedy that, we can add to T µν a term of the form ∂λ K λµν , where K λµν is
antisymmetric in its first two indices. Such an object is automatically divergenceless, so

Tbµν = T µν + ∂λ K λµν (2.1.2)

is an equally good energy-momentum tensor with the same globally conserved energy and momentum. Show
that this construction, with
K λµν = F µλ Aν , (2.1.3)
leads to an energy-momentum tensor Tb that is symmetric and yields the standard formulae for the electro-
magnetic energy and momentum densities:

E = 21 (E 2 + B 2 ), S = E × B. (2.1.4)

SOLUTION: (Thanks to Eric H. (’13) and Haoyu (’13) for presenting parts (a) and (b) respectively.)

(a) Let us first calculate

∂Fµν
= δµα δνβ − δνα δµβ . (2.1.5)
∂(∂α Aβ )

Next, we write the Lagrangian as L = − 41 η µρ η νσ Fµν Fρσ . Thus,

∂L
= − 14 η µρ η νσ δµα δνβ − δνα δµβ Fρσ + Fµν δρα δσβ − δσα δρβ
  
∂(∂α Aβ )
= − 14 F αβ − F βα + F αβ − F βα


= −F αβ . (2.1.6)

The Euler-Lagrange equation of motion reads


 
∂L ∂L
0 = ∂α − = −∂α F αβ . (2.1.7)
∂(∂α Aβ ) ∂Aβ

7
8 CHAPTER 2. PROBLEM SET 2

The β = 0 component of this equation reads

0 = −∂i F i0 = −∂i E i = −∇ · E , (2.1.8)

which is Gauss’ law without sources (charges, in this case).


The β = i component of Eqn. (2.1.7) is

0 = −∂0 F 0i − ∂j F ji = Ė i − ijk ∂j B k = Ė − ∇ × B , (2.1.9)

which is Ampère’s law without sources (currents, in this case).


The two remaining Maxwell equations automatically follow from writing the theory in terms of the 4-
vector potential, Aµ . The fact that B = ∇ × A implies ∇ · B = 0 and the fact that E = −∇φ − Ȧ implies
∇ × E = −Ḃ, which is Faraday’s law.
(b) Under an infinitesimal translation, xµ → xµ − aµ , the field transforms as
δAν = aµ ∂µ Aν . (2.1.10)

Being a scalar, the Lagrangian transforms as


δL = aµ ∂µ L = aν ∂µ (δνµ L). (2.1.11)

On the other hand, actually plugging in the field transformation, Eqn. (2.1.10), gives
 
∂L ∂L
δL = δAν + ∂µ δAν
∂Aν ∂(∂µ Aν )
    

∂L ∂L ∂L

= ∂µ δAν + −
 ∂µ   δAν
∂(∂µ Aν )  ∂A
 ν ∂(∂µ Aν )
= aν ∂µ −F µλ ∂ν Aλ .

(2.1.12)

The canonical energy-momentum tensor is


T µν = −F µλ ∂ν Aλ − Lδνµ . (2.1.13)

We may rewrite this as


T µν = −η νσ F µλ ∂σ Aλ + 14 η µν F λσ Fλσ . (2.1.14)

The first term is not symmetric in (µν). However, let us add ∂λ K λµν , where K λµν = F µλ Aν = η νσ F µλ Aσ ,
as stated in the problem. Then, using the equation of motion,
Tbµν = η νσ F µλ
∂λ A + η νσ F µλ (∂ A − ∂ A ) + 1 η µν F λσ F
σ λ σ σ λ 4 λσ

Tbµν = ηλσ F µλ F σν + 41 η µν Fλσ F λσ . (2.1.15)

The second term is obviously symmetric. The first term is also symmetric because µ ↔ ν is equivalent to
λ ↔ σ and ηλσ is symmetric.
Recall that the energy is precisely the conserved charge associated with time translation symmetry. This
symmetry is the µ = 0 component of Tbµν since the µ index is associated with the symmetry xµ → xµ − aµ .
Since the energy is the charge, it would be the integral over space of the ν = 0 component of T̂ 0ν . Therefore,
the energy density is Tb00 :
Tb00 = ηλσ F 0λ F σ0 + 14 F λσ Fλσ
= −F 0i F i0 + 14 F 0i F0i + 14 F i0 Fi0 + 41 F ij Fij
= −(−E i )E i + 14 (−E i )E i + 41 E i (−E i ) + 14 ijk ij` B k B `
= 12 E 2 + 1
4 · 2δ k` B k B `

E = Tb00 = 12 (E 2 + B 2 ) . (2.1.16)
2.2. THE COMPLEX SCALAR FIELD, P&S 2.2 P.33 9

Now, recall that the Poynting vector can be interpreted as energy flux density or momentum density (these
two things have the same units when c = 1). The former is given by Tb0i (the spacial components of the time-
translation current), while the latter is given by Tbi0 (the charge density associated with spatial translations).
Since Tb is symmetric, it makes no difference which one we calculate. We will calculate the (i0) component:

Tbi0 = ηλσ F iλ F σ0 = −F ij F j0 = −(−ijk B k )E j = ijk E j B k . (2.1.17)

Thus, the Poynting vector is given by S i = Tbi0 and we get the desired result:

S=E×B . (2.1.18)

2.2 The Complex Scalar Field, P&S 2.2 p.33


Consider the field theory of a complex-valued scalar field obeying the Klein-Gordon equation. The action of this
theory is Z
S = d4 x ∂µ φ∗ ∂ µ φ − m2 φ∗ φ .

(2.2.1)

It is easiest to analyze this theory by considering φ(x) and φ∗ (x), rather than the real and imaginary parts of
φ(x), as the basic dynamical variables.

(a) Find the conjugate momentum to φ(x) and φ∗ (x) and the canonical commutation relations. Show that the
Hamiltonian is Z
H = d3 x π ∗ π + ∇φ∗ · ∇φ + m2 φ∗ φ .

(2.2.2)

Compute the Heisenberg equation of motion for φ(x) and show that it is indeed the Klein-Gordon equation.

(b) Diagonalize H by introducing creation and annihilation operators. Show that the theory contains two sets
of particles of mass m.

(c) Rewrite the conserved charge Z


i ∗ ∗
d3 x

Q= φ π − πφ (2.2.3)
2

in terms of creation and annihilation operators, and evaluate the charge of the particles of each type.

(d) Consider the case of two complex Klein-Gordon fields with the same mass. Label the fields as φa (x), where
a = 1, 2. Show that there are now four conserved charges, one given by the generalization of part (c), and
the other three given by Z
i
Qi = d3 x φ∗a (σ i )ab πb∗ − πa (σ i )ab φb ,

(2.2.4)
2

where σ i are the Pauli sigma matrices. Show that these three charges have the commutation relations of
angular momentum (SU (2)). Generalize these results to the case of n identical complex scalar fields.

SOLUTION: (Thanks to Tess, Richard and Chien-I (’12) and Mudassir, Brad and Adrian (’13) for parts (a),
(b) and (d) respectively.)

(a) First, we rewrite the action as


Z
d4 x φ̇∗ φ̇ − ∇φ∗ · ∇φ − m2 φ∗ φ .

S= (2.2.5)

Then, the canonical momenta are

∂L ∂L
π= = φ̇∗ , π∗ = = φ̇ . (2.2.6)
∂ φ̇ ∂ φ̇∗
10 CHAPTER 2. PROBLEM SET 2

The canonical equal-time commutation relations are

[φ(x, t), π(y, t)] = [φ(x, t), φ̇∗ (y, t)] = iδ(x − y),
(2.2.7)
[φ∗ (x, t), π ∗ (y, t)] = [φ∗ (x, t), φ̇(y, t)] = iδ(x − y).

The Hamiltonian density is

H = π φ̇ + π ∗ φ̇∗ − L = ππ ∗ + π ∗ π − ππ ∗ + ∇φ∗ · ∇φ + m2 φ∗ φ
= π ∗ π + ∇φ∗ · ∇φ + m2 φ∗ φ. (2.2.8)

Therefore, the Hamiltonian is


Z Z
d3 x H = d3 x π ∗ π + ∇φ∗ · ∇φ + m2 φ∗ φ .

H= (2.2.9)

The Heisenberg equation of motion for φ is


Z
φ̇(x, t) = −i[φ(x, t), H(t)] = −i d3 y [φ(x, t), H(y, t)]
Z
= −i d3 y π ∗ (y, t)[φ(x, t), π(y, t)]
Z
= −i d3 y π ∗ (y, t) iδ(x − y)

φ̇(x, t) = π ∗ (x, t) , (2.2.10)

which is precisely the second equation in (2.2.6). This agrees with the classical equation, φ̇ = ∂H/∂π.
To calculate the Heisenberg equation of motion for π ∗ , let us first rewrite H by integrating the middle
term by parts and assuming that the corresponding boundary term vanishes:
Z
H = d3 x π ∗ π + φ∗ (m2 − ∇2 )φ .
 
(2.2.11)

Then, we find

π̇ ∗ (x, t) = i[H(t), π ∗ (x, t)]


Z
= i d3 y [φ∗ (y, t), π ∗ (x, t)](m2 − ∇2y )φ(y, t)
Z
= i d3 y iδ(y − x)(m2 − ∇2y )φ(y, t)

π̇ ∗ (x, t) = (∇2 − m2 )φ(x, t) . (2.2.12)

Note that this agrees with the classical equation π̇ ∗ = −∂H/∂φ∗ .


Combining Eqns. (2.2.10) and (2.2.12) yields the Klein-Gordon equation:

φ̈ = π̇ ∗ = (∇2 − m2 )φ =⇒ ( + m2 )φ = 0 , (2.2.13)

where  = ∂µ ∂ µ = ∂t2 − ∇2 .

(b) Split φ up into its real and imaginary pieces:

φ= √1 (A + iB). (2.2.14)
2
2.2. THE COMPLEX SCALAR FIELD, P&S 2.2 P.33 11

Since φ satisfies the KG equation, so do A and B. Being real scalar fields they are expanded as
d¯3 p
Z
αp e−ipx + αp† eipx ,

A(x) = p (1.15a)
2Ep
d¯3 p
Z
βp e−ipx + βp† eipx ,

B(x) = p (1.15b)
2Ep

where d¯3 p = d3 p/(2π)3 and αp and βp satisfy the usual commutation relations of creation and annihilation
operators. Therefore,
d¯3 p
Z
ap e−ipx + b†p eipx ,

φ(x) = p (2.2.16)
2Ep
where
ap = √1 (αp
2
+ iβp ), b†p = √1 (α†
2 p + iβp† ). (2.2.17)

Note that a and b also satisfy the usual commutation relations. For example,
: 0 † :0
[ap , a†q ] = 21 [αp , αq† ] − i † [βp , αq ] + [βp , βq† ]

[α
p , βq ] + i

= 12 (2π)3 δ(p − q) + (2π)3 δ(p − q)


 

= (2π)3 δ(p − q). (2.2.18)

We will verify these commutation relationships again in a moment directly from the assumption of the com-
mutation relationships between φ and π (i.e. ab initio from the basic assumption of canonical quantization).
Below are the expansions of the fields:
d3 p
Z
ap e−ipx + b†p eipx ,

φ(x) = 3
p (1.19a)
(2π) 2Ep
d3 p
Z
φ∗ (x) = a†p eipx + bp e−ipx ,

3
p (1.19b)
(2π) 2Ep
r
d3 p
Z
∗ Ep † ipx
ap e − bp e−ipx ,

π(x) = φ̇ (x) = i (1.19c)
(2π)3 2
r
d3 p
Z
∗ Ep
ap e−ipx − b†p eipx .

π (x) = φ̇(x) = −i 3
(1.19d)
(2π) 2
To solve for the creation and annihilation operators in terms of the fields, we must inverse-Fourier-transform:
d3 q
Z Z Z
−iqx † iqx
d3 x eipx φ(x) = d3 x eipx

p aq e + b q e
(2π)3 2Eq
3
d3 x
Z Z
d q
aq ei(p−q)·x + b†q ei(p+q)·x

= p 3
2Eq (2π)
Z 3
d q h i
= p aq ei(Ep −Eq )t δ(p − q) + b†q ei(Ep +Eq )t δ(p + q)
2Eq
= (2Ep )−1/2 ap + b†−p e2iEp t .

(2.2.20)

Similarly, we find all the following relations:


Z
ap + b†−p e2iEp t = d3 x eipx φ(x),
p
2Ep (1.21a)
Z
a−p e−2iEp t + b†p = 2Ep d3 x e−ipx φ(x),
p
(1.21b)
s Z
† 2
ap − b−p e 2iEp t
=i d3 x eipx π ∗ (x), (1.21c)
Ep
s Z
−2iEp t † 2
a−p e − bp = i d3 x e−ipx π ∗ (x). (1.21d)
Ep
12 CHAPTER 2. PROBLEM SET 2

From these relations, we may solve for ap and b†p and thus their adjoints:

d3 x ipx  ∗
Z

ap = i p e π (x) − iEp φ(x) , (1.22a)
2Ep
d3 x −ipx 
Z
a†p = −i p π(x) + iEp φ∗ (x) ,

e (1.22b)
2Ep
d3 x −ipx  ∗
Z
b†p = −i p

e π (x) + iEp φ(x) , (1.22c)
2Ep
Z 3
d x ipx 
π(x) − iEp φ∗ (x) .

bp = i p e (1.22d)
2Ep

Now, we may use the canonical commutation relations, Eqn. (2.2.7), to calculate the commutation relations
of the above operators. For example,

d3 x d3 y i(px−qy) 
Z 
[ap , a†q ] = p e −iEq [φ∗ (y), π ∗ (x)] − iEp [φ(x), π(y)]
2 Ep Eq
Z
Ep + Eq
= p d3 x d3 y ei(px−qy) δ(y − x)
2 Ep Eq
Z
Ep + Eq
= p d3 x ei(p−q)·x
2 Ep Eq
Ep + Eq i(Ep −Eq )t
= p e (2π)3 δ(p − q)
2 Ep Eq
[ap , a†q ] = (2π)3 δ(p − q) . (2.2.23)

A prefactor was dropped in going from the penultimate to the final line because this factor equals 1 when
p = q, which is the only support of the delta function anyway. This reproduces Eqn. 2.29 p.21 of P&S. It is
easy to see that the two terms, which have the same sign, in the above commutator will have opposite signs
in the commutator of a with itself, or of a† with itself. Hence, these latter commutators vanish.

Next, we calculate the Hamiltonian. We work with the form (2.2.11). The first term is

r
d3 p
Z Z  Z 
3 ∗ 3 Ep −ipx † ipx

d xπ π = d x −i ap e − b p e
(2π)3 2
r
d3 q
 Z 
Eq † iqx
aq e − bq e−iqx

× i 3
(2π) 2
Z 3 3 p
E E d3 x h
Z
d pd q p q
= 3
ap a†q e−i(p−q)·x − ap bq e−i(p+q)·x
(2π) 2 (2π)3
i
− b†p a†q ei(p+q)·x + b†p bq ei(p−q)·x
Z 3 3 p
d p d q Ep Eq h
ap a†q e−i(Ep −Eq )t + b†p bq ei(Ep −Eq )t δ(p − q)

= 3
(2π) 2
i
−i(Ep +Eq )t
+ b†p a†q ei(Ep +Eq )t δ(p + q)

− ap bq e
d3 p Ep h
Z i
† † † † 2iEp t −2iEp t
= ap ap + bp b p − bp a −p e − a p b −p e . (2.2.24)
(2π)3 2
2.2. THE COMPLEX SCALAR FIELD, P&S 2.2 P.33 13

The second term is


Z 3 3
d p d q (m2 + |q|2 ) d3 x h †
Z Z
d3 x φ∗ (m2 − ∇2 )φ = a aq ei(p−q)·x + a†p b†q ei(p+q)·x
(2π)3 p
p
(2π)3 2 Ep Eq
i
+ bp aq e−i(p+q)·x + bp b†q e−i(p−q)·x
Z 3 3
d p d q (m2 + |q|2 ) h †
ap aq ei(Ep −Eq )t + bp b†q e−i(Ep −Eq )t δ(p − q)

= 3
p
(2π) 2 Ep Eq
i
+ a†p b†q ei(Ep +Eq )t + bp aq e−i(Ep +Eq )t δ(p + q)


d3 p m2 + |p|2 h †
Z i
† † † 2iEp t −2iEp t
= a a
p p + b b
p p + a b
p −p e + b a
p −p e
(2π)3 2Ep
d3 p Ep h †
Z i
= 3
ap ap + bp b†p + b†p a†−p e2iEp t + ap b−p e−2iEp t . (2.2.25)
(2π) 2

To get the final line, we transformed p → −p for the last two terms and used the fact that a’s and b’s
commute with each other. The point of this is that, in this form, it is clear that the last two terms of Eqns.
(2.2.24) and (2.2.25) cancel leaving just

d3 p Ep h
Z i
† † † †
H= ap ap + ap ap + bp bp + bp bp . (2.2.26)
(2π)3 2

We may rewrite this as


d3 p
Z Z
Ep a†p ap + b†p bp + d3 p Ep δ (3) (0) .

H= 3
(2.2.27)
(2π)

The last piece may be dropped as usual by normal ordering.


The exact same procedure gives the physical momentum (the integral of T̂ i0 ):

d3 p
Z Z
d3 x π∇φ + π ∗ ∇φ∗ = p a†p ap + b†p bp .
 
P=− 3
(2.2.28)
(2π)

Thus, we may interpret the two sets of creation operators as each creating its own particle of momentum p
1/2
and energy Ep = |p|2 + m2 .
(c) In preparation for part (d), let us show that this charge is indeed conserved.

Conservation proof 1: Take a time derivative:


Z
i
d3 x φ̇∗ π ∗ + φ∗ π̇ ∗ − π̇φ − π φ̇

Q̇ =
2
Z
i ∗
d3 x  + φ∗ φ̈ − φ̈∗ φ −  ∗

= ππ ππ . (2.2.29)
2

2
From Eqns. (1.19a) and (1.19b), we see that φ̈ = −EP φ and φ̈∗ = −Ep2 φ∗ . Thus,
Z
i 2
Q̇ = Ep d3 x φ∗ φ − φ∗ φ) = 0. (2.2.30)
2

Conservation proof 2: Use the Heisenberg equation of motion for Q:

Q̇ = −i[Q, H]
ZZ
i
= −i d3 x d3 y [φ∗x πx∗ − πx φx , πy∗ πy + ∇y φ∗y · ∇y φy + m2 φ∗y φy ]. (2.2.31)
2
14 CHAPTER 2. PROBLEM SET 2

The subscripts remind us of the argument of the various fields.


Note the following commutation identities:

[A, CD] = ACD − CDA = ACD − CAD + CAD − CDA = [A, C]D + C[A, D], (1.32a)
[AB, C] = ABC − CAB = ABC − ACB + ACB − CAB = A[B, C] + [A, C]B. (1.32b)

Using these, we can derive a third identity:

[AB, CD] = A[B, C]D + AC[B, D] + [A, C]DB + C[A, D]B. (1.32c)

We can use this to calculate the various commutators in Eqn. (2.2.26). For example,

[φ∗x πx∗ , πy∗ πy ] = [φ∗x , πy∗ ]πy πx∗ = iδ(x − y)πy πx∗ . (2.2.33)

All in all, Eqn. (2.2.31) gives


ZZ
1
d3 x d3 y iδ(x − y)πy πx∗ − i[∇y δ(x − y)]φ∗x ∇y φy − iδ(x − y)φ∗x φy

Q̇ =
2
− iδ(x − y)πx πy∗ + i[∇y δ(x − y)](∇y φ∗y )φx + iδ(x − y)φ∗y φx

Z
i
ππ ∗ + φ∗ ∇2 φ − 
d3 x  φ∗ ∗
− (∇2 φ∗ )φ + 
φ∗
  
= φ − ππ φ
2
Z
i
= d3 x −∇φ∗ · ∇φ + ∇φ∗ · ∇φ)
2
= 0. (2.2.34)

Note that we used integration by parts at various points in the derivation.

Conservation proof 3: We can derive Q using the Noether procedure. Note, however, that if you are just
given the charge, it might be quite difficult to work out the symmetry from which it derives even though you
may be able to prove conservation using the previous two methods.
The conserved charge, (2.2.3), is associated with the symmetry of the action under the simultaneous phase
transformations

φ → e−iα/2 φ ≈ 1 − i α2 φ, φ∗ → eiα/2 φ∗ ≈ 1 + i α2 φ∗ .
 
(2.2.35)

In other words,

∆φ = − 2i φ, ∆φ∗ = 2i φ∗ . (2.2.36)

The conserved current, given by Eqn. 2.12 p.18 of P&S, is

∂L ∂L i
j µ = ∆φ ∆φ∗ − µ
= (∂ µ φ)φ∗ − (∂ µ φ∗ )φ .

+ ∗
J (2.2.37)
∂(∂µ φ) ∂(∂µ φ ) 2

Note that 0 = ∆L = ∂µ J µ , so we may drop J µ since it is divergenceless and j µ is well-defined only up to a


divergenceless quantity. This gives the charge
Z Z
i i
d3 x φ̇φ∗ − φ̇∗ φ = d3 x φ∗ π ∗ − πφ ≡ Q̃† − Q̃.
 
Q= (2.2.38)
2 2

We switched the order of multiplication of φ̇ = π ∗ and φ∗ in the first term. This does not matter here since
the difference is infinite and vanishes after normal ordering. In part (d), however, the order of multiplication
is important and only one way makes sense. Then, Q is conserved by construction.
2.2. THE COMPLEX SCALAR FIELD, P&S 2.2 P.33 15

Finally, let us actually answer the question and write Q in terms of creation and annihilation operators:
r
d3 p d3 q
Z Z  Z 
† i 3 † ipx −ipx
 Eq −iqx † iqx

Q̃ = d x a p e + bp e −i a q e − bq e
(2π)3
p
2 (2π)3 2Ep 2
Z 3 3 s
d3 x
Z
1 d p d q Eq
a† aq ei(p−q)x + bp aq e−i(p+q)x − a†p b†q ei(p+q)x − bp b†q e−i(p−q)x

=
4 (2π)3 Ep (2π)3 p
Z 3 3 s
1 d p d q Ep h †
ap aq ei(Ep −Eq )t − bp b†q e−i(Ep −Eq )t δ(p − q)

= 3
4 (2π) Eq
i
+ bp aq e−i(Ep +Eq )t − a†p b†q ei(Ep +Eq )t δ(p + q)


d3 p
Z
1
a†p ap − bp b†p + bp a−p e−2iEp t − a†p b†−p e2iEp t .

= 3
(2.2.39)
4 (2π)

Similarly, one finds

d3 p
Z
1
a†p ap − bp b†p + a†p b†−p e2iEp t − bp a−p e−2iEp t .

Q̃ = − 3
(2.2.40)
4 (2π)

The last two terms in Eqns. (2.2.39) and (2.2.40) cancel after a p → −p switch. The result is

d3 p
Z
1
ap a†p + a†p ap − b†p bp − bp b†p .

Q= 3
(2.2.41)
4 (2π)

The a’s can be put into normal ordered form leaving an infinite constant. The same can be done for the b’s.
However, since the a’s and b’s come with opposite sign, the infinite constants cancel! We find

d3 p
Z
1
a†p ap − b†p bp .

Q= 3
(2.2.42)
2 (2π)

Thus, a and b particles have charge + 21 and − 21 , respectively.


(d) Suppose we have n complex scalar fields, φ1 , . . . , φn . Combine these into an n × 1 matrix. Define
  
φ1 φ̇1
φ =  ...  , π ∗ = φ̇ =  ...  ,
   
(2.2.43)
φn φ̇n
φ† = (φ∗1 , . . . , φ∗n ), π = φ̇† = (φ̇∗1 , . . . , φ̇∗n ).

Define
σ µ = (1, σ i ), (2.2.44)

where 1 here is the 2 × 2 identity matrix, and σi are the Pauli matrices
     
1 0 1 2 0 −i 3 1 0
σ = , σ = , σ = . (2.2.45)
1 0 i 0 0 −1

We can unify the notation and write (suppressing the matrix indices)
Z
i
Qµ = d3 x φ† σ µ π ∗ − πσ µ φ .

(2.2.46)
2

Here is an example where it might be more difficult to imagine the symmetry from which this charge derives.
Nevertheless, we are able to prove that it is conserved using the methods in used in part (c):
16 CHAPTER 2. PROBLEM SET 2

Conservation proof 1: The only change from part (c) is that φ∗ is replaced with φ† and σ µ is put in the
middle of each field binomial. Otherwise, the calculation is exactly identical and we find Q̇µ = 0.

µ
Conservation proof 2: Again, this follows through as in part (c) with factors of σab and various indices
flying around.

Conservation proof 3: In this case, the Lagrangian is

L = ∂µ φ† ∂ µ φ − m2 φ† φ. (2.2.47)

Clearly, this is invariant under the transformation φ → U φ and φ† → φ† U † , where U is a unitary matrix
(i.e. U † U = 1). Any unitary n × n matrix, U ∈ U (n), may be written as U = e−iαA TA , where TA forms a
basis for the space of n × n Hermitian matrices. First, let us count how many TA ’s there are. Before any
constraints, a Hermitian matrix, H, is an n×n complex matrix and thus has 2n2 free parameters. Hermiticity
requires H = H † . This implies that the diagonal elements are real and thus cuts the numbers of degrees of
freedom by n. The lower-triangular elements are determined to be the complex conjugates of their mirror
upper-triangular elements, which are otherwise arbitrary. This cuts down the number of parameters by twice
the number of lower-triangular slots (the factor of 2 is due to complexity). This is 2 · n(n−1)
2 = n2 − n. All in
all, we have 2n − n − (n − n) = n free parameters. Thus, there are n matrices, TA . That is, A = 1, . . . , n2 .
2 2 2 2

Infinitesimally, we have

∆A φ = −iTA φ, ∆A φ† = iφ† TA† = iφ† TA , (2.2.48)

where the final equality follows from the fact that TA is Hermitian.
This gives us n2 conserved currents (the order really does matter now):

µ ∂L ∂L
= ∆A φ† ∆A φ = i φ† TA ∂ µ φ − (∂ µ φ† )TA φ ,
 
jA + (2.2.49)
∂(∂µ φ† ) ∂(∂µ φ)

and n2 conserved charges:


Z Z Z
QA = d3 x jA
0
=i d3 x (φ† TA φ̇ − φ̇† TA φ) = i d3 x (φ† TA π ∗ − πTA φ). (2.2.50)

For n = 2, we are supposed to have 22 = 4 matrices TA . This aligns with the fact that any 2 × 2 Hermitian
matrix can be written as a linear combination of 12×2 and the three Pauli matrices. The standard convention
is to let T1 = 21 1, T2 = 21 σ 1 , T3 = 21 σ 2 and T4 = 21 σ 3 . Therefore, it makes more sense to use a spacetime
index instead of the A index. Recalling the definition, σ µ = (1, σ i ), we can write Eqn. (2.2.50) instead as

Z
i
Q = µ
d3 x (φ† σ µ π ∗ − πσ µ φ) . (2.2.51)
2

This is precisely the generalization of (2.2.3) and (2.2.4) with matrix indices suppressed.
Now, we will calculate the commutation relations of the charges QA :
Z
d3 x d3 y (φ† TA π ∗ − πTA φ)(x), (φ† TB π ∗ − πTB φ)(y)]

[QA , QB ] = −
Z  
d3 x d3 y φ† TA π ∗ (x), φ† TB π ∗ (y)] + πTA φ(x), πTB φ(y)] .

=−
2.2. THE COMPLEX SCALAR FIELD, P&S 2.2 P.33 17

Let A(x) = φ† TA π ∗ (x) and B(y) = φ† TB π ∗ (y). Then,

B(y)A(x) = φ∗c (y)TBcd πd∗ (y)φ∗a (x)TAab πb∗ (x)


h i
= TBcd TAab φ∗c (y) φ∗a (x)πd∗ (y) − iδad δ(x − y) πb∗ (x)
h i
= TBcd TAab φ∗a (x)φ∗c (y)πb∗ (x)πd∗ (y) − iδad δ(x − y)φ∗c (x)πb∗ (x)
n h i o
= TBcd TAab φ∗a (x) πb∗ (x)φ∗c (y) + iδbc δ(x − y) πd∗ (y) − iδad δ(x − y)φ∗c (x)πb∗ (x)
h i
= A(x)B(y) + iδ(x − y) φ∗a (x)TAab TBbd πd∗ (x) − φ∗c (x)TBca TAab πb∗ (x)
= A(x)B(y) + iδ(x − y)φ∗a (x)[TA , TB ]ab πb∗ (x)
= A(x)B(y) + iδ(x − y)φ† [TA , TB ]π ∗ .

We find a similar result if we define C(x) = πTAab φ(x) and D(y) = πTBcd φ(y). We organize the results here:

[A(x), B(y)] = −iδ(x − y)φ† [TA , TB ]π ∗ ,


[C(x), D(y)] = iδ(x − y)π[TA , TB ]φ.

Therefore,
Z  
[QA , QB ] = − d3 x d3 y [A(x), B(y)] + [C(x), D(y)]
Z
= i d3 x φ† [TA , TB ]π ∗ − π[TA , TB ]φ .

(2.2.52)

A Lie algebra is defined by its commutation relations:

[TA , TB ] = ifABC TC , (2.2.53)

where fABC are constants, called the structure constants. This encodes the fact that the Lie group (i.e. the
exponentials) must be closed under multiplication. Therefore,
Z
d3 x φ† TC π ∗ − πTC φ = ifABC QC .

[QA , QB ] = ifABC i (2.2.54)

In other words, the charges have the same algebra as the generators of the symmetry Lie algebra.
In the case of n = 2 excluding the time component, the SU (2) algebra reads

[Qi , Qj ] = iijk Qk . (2.2.55)

It should be clear that Q0 commutes with Qi and so the group is SU (2) × U (1) = U (2).
Actually, U (n) is not the full symmetry group. n complex scalars is the same as 2n real scalars. We
can define √12 Φ1 = Re φ1 , √12 Φ2 = Im φ1 , √12 Φ3 = Re φ2 , √12 Φ4 = Im φ2 , . . . , √12 Φ2n−1 = Re φn and
√1 Φ2n = Im φn . Define
2  
Φ1
Φ =  ...  . (2.2.56)
 

Φ2n

The Lagrangian may be written as


L = 21 ∂µ Φ| ∂ µ Φ − 21 m2 Φ| Φ. (2.2.57)

Clearly, this is invariant under the transformation Φ → OΦ, where O is an orthogonal matrix (i.e. O| O = 1).
Since det O| det O = det(O| O) = det 1 = 1 and det O| = det O, an orthogonal matrix has det O = ±1. The
elements in the orthogonal group with determinant −1 are reached from those that have determinant +1 via
18 CHAPTER 2. PROBLEM SET 2

a parity transformation, which is not a continuous transformation and thus excluded from the calculation of
conserved currents. In other words, we only consider SO(2n), those real orthogonal 2n × 2n matrices that
have determinant +1.
One basis for SO(2n) consists of all the plane rotations. That is, pick two orthogonal directions in R2n
and rotate in the plane containing those two directions. In terms of the fields, this simply rotates two of
the fields into each other and leaves the rest unchanged. The number of pairs of orthogonal directions (not
counting the order, of course) is 2n2 = n(2n − 1). This is therefore the dimension of SO(2n) and thus the
number of conserved currents.
To write down the currents, we must pass to the Lie algebra. Consider the rotation in the (1, 2) direction:
 
cos α sin α 0
O =  − sin α cos α 0  .
0 0 1
We can write this as an exponential:
 
0 −1
O = e−αT where T = 1 0 .
0

It should be clear how to generalize this to a rotation in any other pair of directions. This gives us n(2n − 1)
generators, T A . The infinitesimal transformations are thus
|
∆A Φ = −T A Φ, ∆A Φ| = −Φ| T A = Φ| T A , (2.2.58)

|
where we used the fact that T A = −T A .
The corresponding currents are
∂L ∂L
µA = ∆A Φ| |
+ ∆A Φ = 12 Φ| T A ∂ µ Φ − 12 (∂ µ Φ| )T A Φ. (2.2.59)
∂(∂µ Φ ) ∂(∂µ Φ)

In principle, we should be able to write the currents (2.2.49) in terms of (2.2.59). Let us see how this works
in the case of n = 2. In this case, we write Eqn. (2.2.48) as

∆µ φ = − 2i σ µ φ, ∆µ φ† = 2i φ† σ µ . (2.2.60)

Let us consider µ = 0. Then,


     
Φ1 + iΦ2 i Φ1 + iΦ2 1 Φ2 − iΦ1
∆0 =− = . (2.2.61)
Φ3 + iΦ4 2 Φ3 + iΦ4 2 Φ4 − iΦ3

In the four-component form, this reads


      
Φ1 −Φ2 0 −1 Φ1
0  Φ2 
  1 Φ 1
 1 1 0  Φ2 
∆  =−  =−   . (2.2.62)
Φ3 2 −Φ4  2 0 −1 Φ3 
Φ4 Φ3 1 0 Φ4

Organize the generators of the Lie algebra of SO(4) so that T 1 generates rotations in the (1, 2) direction, T 2
in the (1, 3) direction, T 3 in the (1, 4) direction, up to T 6 in the (3, 4) direction. Then, it is clear that

∆0 = 12 (∆1 + ∆6 ). (2.2.63)

Consequently,
j 0µ = 21 (µ1 + µ6 ) and Q0 = 12 (Q1 + Q6 ), (2.2.64)

where QA are the SO(4) charges while Qµ are the U (2) charges. The remaining three U (2) charges, Qi , may
be expressed in terms of the SO(4) charges via a similar procedure.
2.3. DISCUSSION 1: ALTERNATIVE CURRENT DERIVATION 19

Note: The notation often used to express the Lie algebra associated with a Lie group is the name of the Lie
group but in lower-case Fraktur font. For example, the Lie group SO(4) consists of all real orthogonal 4 × 4
matrices with unit determinant whereas the Lie algebra so(4) consists of real skew-symmetric 4 × 4 matrices. As
another example, the Lie group U (2) consists of all unitary 2 × 2 matrices whereas the Lie algebra u(2) consists
of all Hermitian 2 × 2 matrices. Also, the Lie group SU (2) is the subgroup of U (2) containing only those matrices
with unit determinant while the Lie algebra su(2) is the subgroup of u(2) containing only those matrices that are
traceless. This final fact derives from the identity det eX = etr X , which is obviously crucial to the study of Lie
algebras

2.3 Discussion 1: Alternative Current Derivation


For more on the method we are about to describe, see Zee Ed. 1 exercise I.9.1 or Zee Ed. 2 exercise I.10.1.
As is described therein, in the case of a global symmetry, one can promote the infinitesimal field transformation
parameter (called α in P&S) to a spacetime-dependent one (i.e. promote the global transformation to a local
one). Of course, the action will no longer be invariant (i.e. δS 6= 0). However, since δS = 0 when this parameter,
α, is constant, δS must have the form
Z
δS = d4 x J µ (x) ∂µ α(x). (2.3.1)

More precisely,
Z
δS = d4 x (E(x)α(x) + J µ (x) ∂µ α(x)), (2.3.2)

where E is proportional to the Euler-Lagrange equation of motion and vanishes on-shell. In this formalism, the
current is simply the coefficient in δS of the derivative of α. Note that to get δS into the above form may involve
various judicious integration by parts.
Let us derive the (un-symmetrized) energy-momentum tensor in Eqn. (2.1.18) using this formalism. In this
case, Eqn. (2.1.15) is promoted to
δAν = aµ (x) ∂µ Aν . (2.3.3)

Let us find the transformation of the field strength:

Fαβ → ∂α (Aβ + aν ∂ν Aβ ) − ∂β (Aα + aν ∂ν Aα )


= Fαβ + aν ∂ν Fαβ + (∂α aν )∂ν Aβ − (∂β aν )∂ν Aα . (2.3.4)

Let us plug this into the action:


Z
1
d4 x Fαβ + aν ∂ν Fαβ + (∂α aν )∂ν Aβ − (∂β aν )∂ν Aα F αβ + aσ ∂σ F αβ + (∂ α aσ )∂σ Aβ − (∂ β aσ )∂σ Aα
  
S→−
4
Z
1
d4 x aν ∂ν (Fαβ F αβ ) + 4(∂µ aν )F µλ ∂ν Aλ + O(a2 ).
 
=S− (2.3.5)
4

Everything on the right hand side of Eqn. (2.3.5) that is not S is δS. Note my comment about possibly having
to integrate by parts in order to get the form Eqn. (2.3.2). If one were being careless, one might think that what
multiplies aν above is not included in the current. However, we would have to massage δS into a form such that
what multiplies aν is proportional to the equation of motion. Well, the thing that multiplies aν is ∂ν (Fαβ F αβ ),
which is not proportional to the equation of motion, ∂µ F µν = 0. Therefore, we must integrate this term by parts
to get
Z
1
δS = − d4 x (∂µ aν )(−δνµ Fαβ F αβ + 4F µλ ∂ν Aλ ). (2.3.6)
4
Therefore, the current, or energy-momentum tensor is

T µν = −F µλ ∂ν Aλ + 41 δνµ Fαβ F αβ , (2.3.7)

which agrees with the result of Problem 2.1.


20 CHAPTER 2. PROBLEM SET 2

2.4 Discussion 2: Energy-Momentum Tensor by Coupling to Gravity


As mentioned in lecture, there is an alternative definition for the energy-momentum tensor, which makes use
of the fact that in gravity, the energy-momentum tensor acts as a source for the gravitational field, namely the
spacetime metric. Unfortunately, gravity is very much nonlinear and so the source term is not simply L = gµν T µν ,
which would be the thing analogous to the source term for a scalar field, for example, L = φJ. However, there
is still a procedure to get the energy-momentum tensor. One starts with the usual action in flat spacetime of
the system that one wants to couple to gravity. Then, one “covariantizes” this action and asserts the result to

be the action in non-flat spacetime. In practice, one simply replaces d4 x with d4 x −g, where g = det gµν .
R R

The minus sign in the square root is just there because of the signature of the metric; you can replace it with
absolute value if you want. In addition, one replaces derivatives, ∂µ , with covariant derivatives, ∇µ . For our
present purpose, we will not need to know exactly what ∇µ means because it actually does not make a difference
in our case: ∇µ Aν − ∇ν Aµ = ∂µ Aν − ∂ν Aµ . This procedure derives from the “minimal-coupling principle” in
general relativity. For more details, see the beginning of chapter 4 of Sean Carroll’s GR textbook.
Therefore, our minimally-coupled action is

Z
1
S=− d4 x −ggαγ gβδ F αβ F γδ . (2.4.1)
4

To vary S with respect to gµν we need to vary the −g term. We can use Eqn. 2.9a of Problem Set 1 to get
√ 1√
δ −g = −g g µν δgµν . (2.4.2)
2
One can now calculate δS and retrieve the energy-momentum tensor via the formula
2 δS
T µν = − √ . (2.4.3)
−g δgµν
Chapter 3

Problem Set 3

3.1 Lorentz Group, P&S 3.1, p.71


Recall from Eqn. 3.17 p.39 P&S the Lorentz commutation relations,

[J µν , J ρσ ] = i(g νρ J µσ − g µρ J νσ − g νσ J µρ + g µσ J νρ ). (3.1.1)

(a) Define the generators of rotations and boosts as

Li = 21 ijk J jk , K i = J 0i , (3.1.2)

where i, j, k = 1, 2, 3. An infinitesimal Lorentz transformation can then be written

Φ → (1 − iθ · L − iβ · K)Φ. (3.1.3)

Write the commutation relations of these vector operators explicitly. (For example, [Li , Lj ] = iijk Lk .) Show
that the combinations
J± = 21 (L ± iK) (3.1.4)

commute with one another and separately satisfy the commutation relations of angular momentum.
(b) The finite-dimensional representations of the rotation group correspond precisely to the allowed values for
angular momentum: integers or half-integers. The results of part (a) implies that all finite-dimensional
representations of the Lorentz group correspond to pairs of integers or half-integers, (j+ , j− ), corresponding
to pairs of representations of the rotation group. Using the fact that J = σ/2 in the spin-1/2 representation
of angular momentum,  write explicitly the transformation laws of the 2-component objects transforming
according to the 21 , 0 and 0, 12 representations of the Lorentz group. Show that these correspond precisely


to the transformations of ψL and ψR given in Eqn. 3.37 p.44 P&S:

ψL → 1 − iθ · σ2 − β · σ2 ψL , ψR → 1 − iθ · σ2 + β · σ2 ψR .
 
(3.1.5)

(c) The identity σ | = −σ 2 σσ 2 allows us to rewrite the ψL transformation in the unitarily equivalent form

ψ 0 → ψ 0 1 + iθ · σ2 + β · σ2 ,

(3.1.6)

where ψ 0 = ψL σ . Using this law, we can represent the object that transforms as 12 , 12 as a 2 × 2 matrix
T 2


that has the ψR transformation law on the left and, simultaneously, the transposed ψL transformation on
the right. Parametrize this matrix as
 0
V + V 3 V 1 − iV 2

. (3.1.7)
V 1 + iV 2 V 0 − V 3

Show that the object V µ transforms as a 4-vector.

21
22 CHAPTER 3. PROBLEM SET 3

SOLUTION:

(a) We will use the following identity regarding the multiplication of two Levi-Civita symbols:
i`
δ δ im δ in
ijk `mn = δ j` δ jm δ jn

δ k` δ km δ kn
= δ i` (δ jm δ kn − δ jn δ km ) − δ im (δ j` δ kn − δ jn δ k` ) + δ in (δ j` δ km − δ jm δ k` ). (3.1.8)

In particular, this implies the singly-contracted result

ijk i`m = δ j` δ km − δ jm δ k` . (3.1.9)

We first calculate the commutator of two L’s:

[Li , Lj ] = 41 iab jcd [J ab , J cd ]


= 4i iab jcd g bc J ad − g ac J bd − g bd J ac + g ad J bc


= 4i iab −jbd J ad + jad J bd + jcb J ac − jca J bc




= 4i (δ ij δ ad − δ id δ ja )J ad + 4i (δ ij δ bd − δ id δ jb )J bd
+ 4i (δ ij δ ac − δ ic δ ja )J ac + 4i (δ ij δ bc − δ ic δ jb )J bc
= iJ ij . (3.1.10)

Note that all four terms at the fourth equality are equal to each other and the first term in each vanishes
since it involves the trace of J, which vanishes since J is antisymmetric. From the definition of Li , we get

ijk Lk = 12 ijk k`m J `m = 21 (δ i` δ jm − δ im δ j` )J `m = J ij . (3.1.11)

Hence, we may write Eqn. (3.1.10) as


[Li , Lj ] = iijk Lk . (3.1.12)

Next, we calculate the commutator of L with K:

[Li , K j ] = 21 ik` [J k` , J 0j ]
= 2i ik` g `0 J kj −  k0 `j
J − g `j J k0 + g kj J `0

g
= 2i ikj J k0 − 2i ij` J `0
= iijk J 0k .

Therefore, we have
[Li , K j ] = iijk K k . (3.1.13)

Next, we calculate the commutator of K with itself:

[K i , K j ] = [J 0i , J 0j ]
= ig i0 J 0j − ig 00 J ij − ig ij00
J + ig 0j J i0
= −iJ ij .

Using Eqn. (3.1.11), we may write this as

[K i , K j ] = −iijk Lk . (3.1.14)
3.1. LORENTZ GROUP, P&S 3.1, P.71 23

Let us compute the commutator of J+ with J− :


i j
[J+ , J− ] = 14 [Li + iK i , Lj − iK j ]
1
[Li , Lj ] + [K i , K j ] − i[Li , K j ] − i[Lj , K i ]

= 4
1
iijk Lk − iijk Lk + ijk K k + jik K k

= 4
i j
[J+ , J− ] =0. (3.1.15)

Now, we compute the commutator of J± with itself:


i j
[J± , J± ] = 14 [Li ± iK i , Lj ± iK j ]
i j i j
1
i[Li , K j ] ∓ i[Lj , K i ]

= 4 [L , L ] − [K , K ] ±
ijk k
1
L + iijk Lk ± i · iijk K k ∓ i · ijik K k

= 4 i
i ijk
= 2 (Lk ± iK k )
i j
[J± , J± ] = iijk J±
k
, (3.1.16)

which is indeed the same commutation relation as that satisfied by angular momentum.

(b) We can solve for L and K in terms of J± as follows:

L = J+ + J − , K = −i(J+ − J− ). (3.1.17)

1

In the 2, 0 representation, J+ = σ/2 and J− = 02×2 . Thus,
σ
L= 2, K = − iσ
2 . (3.1.18)

The transformation, Eqn. (3.1.3) then reads

1 σ σ
 
2, 0 : Φ → 1 − iθ · 2 −β· 2 Φ, (3.1.19)

which is precisely the transformation of ψL .


In the 0, 12 representation, J− = σ/2 and J+ = 02×2 . Thus,


σ iσ
L= 2, K= 2 . (3.1.20)

The transformation, Eqn. (3.1.3) then reads

0, 21 : σ σ
 
Φ → 1 − iθ · 2 +β· 2 Φ, (3.1.21)

which is precisely the transformation of ψR .

(c) We simply use the identities given to us and (σ 2 )2 = 1 to calculate the transformation of ψ 0 :
| |
ψ 0 = ψL | σ 2 → ψL | 1 − iθ · σ2 − β · σ2 σ 2
2 2 2 2
= ψL | 1 + iθ · σ σσ2 + β · σ σσ 2 σ2
= ψL | σ 2 1 + iθ · σ2 + β · σ2


ψ 0 → ψ 0 1 + iθ · σ2 + β · σ2 .

(3.1.22)

(d) We write the matrix (3.1.7) as


V0+V3 V 1 − iV 2
 
= ηµν V µ σ̄ ν , (3.1.23)
V 1 + iV 2 V0−V3
24 CHAPTER 3. PROBLEM SET 3

where σ̄ µ = (1, −σ). Note that the extra minus sign is necessary given the convention for the metric, which
has the mostly-negative signature.
We are told in the problem that the transformation of this matrix is

ηµν V µ σ̄ ν → 1 − 2i θ · σ + 12 β · σ ηµν V µ σ̄ ν 1 + 2i θ · σ + 12 β · σ
 

= ηµν V µ σ̄ ν − 2i δij θi


[σ j 0
, σ̄
 ]V 0 + 2i δij θi [σ j , σ̄ k ]V k + 12 δij β i {σ j , σ̄ 0 }V 0
− 12 δij β i {σ j , σ̄ k }V k
= ηµν V µ σ̄ ν − 2i δij θi [σ j , σ k ]V k + βi V 0 σ i + 21 βi {σ i , σ j }V j
= ηµν V µ σ̄ ν + δij θi jk` σ ` V k + βi V 0 σ i + βi δ ij V j
= ηµν V µ σ̄ ν + ijk θk V i σ j + βi V 0 σ i + βi V i
= ηµν V µ σ̄ ν − ijk θk V i σ̄ j − βi V 0 σ̄ i + βi V i σ̄ 0 . (3.1.24)

Define the antisymmetric tensor ωµν via

ω0i = βi , ωij = ijk θk , (3.1.25)

so that we may write Eqn. (3.1.24) as

ηµν V µ σ̄ ν → ηµν V µ σ̄ ν − ωij V i σ̄ j − ω0i V 0 σ̄ i − ωi0 V i σ̄ 0


= ηµν V µ σ̄ ν − ωµν V µ σ̄ ν
= ηµν (δρµ V ρ )σ̄ ν − ηµν ωρ µ V ρ σ̄ ν
= ηµν (δρµ + ω µρ )V ρ σ̄ ν . (3.1.26)

This gives the transformation for V µ as

V µ → (δνµ + ω µν )V ν = δνµ − 2i ωρσ (J ρσ )µν V ν ,


 
(3.1.27)


where (J ρσ )µν = i δµρ δνσ − δνρ δµσ is the matrix in Eqn. 3.18 p.39 P&S. This is indeed the transformation of
a 4-vector, as given in Eqn. 3.19 p.40 P&S.

3.2 Lorentz-Invariant Measure, Zee I.8.1 p.69


Derive Eqn. 14 (Zee, p.63):

dD k
Z Z
D+1 2 2 0 0
d k δ(k − m ) θ(k ) f (k , k) = f (ωk , k). (3.2.1)
2ωk

Then verify explicitly that dD k/(2ωk ) is indeed Lorentz invariant. Some authors prefer to replace 2ωk in Eqn.
11 (Zee p.63):
dD k
Z
a(k)e−i(ωk t−k·x) + a† (k)ei(ωk t−k·x)
 
ϕ(x, t) = p (3.2.2)
D
(2π) 2ωk
by 2ωk when relating the scalar field to the creation and annihilation operators. Show that the operators defined
by these authors are Lorentz covariant. Work out their commutation relation.

SOLUTION: (Thanks to Di (’13) for presenting his solution.)

Rewrite the left hand side of Eqn. (3.2.1) as


Z Z ∞
D
d k dk 0 δ[(k 0 )2 − ωk2 ] θ(k 0 ) f (k 0 , k). (3.2.3)
−∞

Write
δ(k 0 − ωk ) + δ(k 0 + ωk )
δ[(k 0 )2 − ωk2 ] = . (3.2.4)
2ωk
3.2. LORENTZ-INVARIANT MEASURE, ZEE I.8.1 P.69 25

Only the first term in Eqn. (3.2.4) contributes to Eqn. (3.2.3) because of the θ(k 0 ) factor. What remains is the
right hand side of Eqn. (3.2.1), as desired.
The measure is obviously invariant under space-time translations and spatial rotations. The only transfor-
mations left are boosts, which we can check explicitly. Consider, for example a boost in the x direction:
ωk0 = cωk + sk 1 , k 01 = sωk + ck 1 , (3.2.5)

with k 2 , k 3 , . . . , k D unchanged, where


c ≡ cosh φ, s ≡ sinh φ, (3.2.6)
and φ is the rapidity parameter related to the familiar γ parameter of relativity via cosh φ = γ.
Then,
 
01 ∂ωk ∂ωk
dk = c + s 1 dk 1 + s i dk i
∂k ∂k
sk 1 i
dk i
 
sk
= c+ dk 1 +
ωk ωk
ωk0 1 sk i dk i
= dk + . (3.2.7)
ωk ωk
Therefore,
ωk0 D
dD k 0 = d k, (3.2.8)
ωk
which proves invariance of the measure.
Note that the scalar field has not changed. The only thing that has changed is the definition of the creation
and annihilation operators. Write the new creation and annihilation operators with a tilde:
dD k
Z
ã(k)e−i(ωk t−k·x) + ㆠ(k)ei(ωk t−k·x) .
 
ϕ(x, t) = (3.2.9)
(2π)D/2 2ωk
Lorentz transform the 4-vector x to Λ−1 x = Λµν xν and k to Λk = Λµν kν . By definition, the scalar field is
contravariant:
U (Λ)ϕ(x)U (Λ)−1 = ϕ(Λ−1 x). (3.2.10)
The left hand side expands to
dD k
Z
−1
 −1 −ikx

U (Λ)ϕ(x)U (Λ) = U (Λ)ã(k)U (Λ) e + h.c. , (3.2.11)
(2π)D/2 2ωk
0
k =ωk

where h.c. stands for Hermitian conjugate.


The right hand side of Eqn. (3.2.10) expands to
dD Λk
Z
−1
 −iΛkΛ−1 x

ϕ(Λ x) = ã(Λk)e + h.c.
(2π)D/2 2ωΛk

(Λk)0 =ωΛk
Z D
d k
ã(Λk)e−ikx + h.c. 0
 
= D/2
, (3.2.12)
(2π) 2ωk k =ωk

where we used the fact that the measure and the inner product kµ xµ are Lorentz invariant.
Since Eqns. (3.2.11) and (3.2.12) are equal, we find that ã is Lorentz covariant:

U (Λ)ã(k)U (Λ)−1 = ã(Λk) . (3.2.13)

We can relate ã to a, the old definition of the operator:



ã(k) = 2ωk a(k). (3.2.14)
Eqn. I.8.12 p.63 of Zee gives the commutation relations of the old operators. Those of the new ones follow
immediately from the above identification:

[ã(k), ㆠ(k0 )] = (2ωk ) δ(k − k0 ) . (3.2.15)


26 CHAPTER 3. PROBLEM SET 3

3.3 Gordon Identity, P&S 3.2 p.72


Derive the Gordon identity,  0µ
p + pµ iσ µν qν

0 µ 0
ū(p )γ u(p) = ū(p ) + u(p), (3.3.1)
2m 2m
where q = (p0 − p). We will put this formula to use in Chapter 6.

SOLUTION:

We will need to use the Dirac equation satisfied by ū. To get this, we will first take the adjoint of the Dirac
equation satisfied by u, Eqn. 3.46 p.45 P&S, then right-multiply by γ 0 :

u† (p)(γ µ† pµ − m)γ 0 = 0. (3.3.2)

We must pass γ 0 accross γ µ† . Note that

σ µ† = σ µ , σ̄ µ† = σ̄ µ . (3.3.3)

Therefore,
σ̄ µ† σ̄ µ
   
µ† 0 0
γ = = . (3.3.4)
σ µ† 0 σµ 0
Left and right-multiplying this by γ 0 gives

σµ
 
0 µ† 0 0
γ γ γ = = γµ =⇒ γ µ† γ 0 = γ 0 γ µ . (3.3.5)
σ̄ µ 0

Therefore, we may write Eqn. (3.3.2) as


ū(p)(γ µ pµ − m) = 0. (3.3.6)
Next, we multiply the left side of Eqn. (3.3.1) by 2m and rewrite it using the Dirac equations:

2mū(p0 )γ µ u(p) = ū(p0 ) γ ν γ µ p0ν + γ µ γ ν pν u(p).



(3.3.7)

We can rewrite the central bracketed term on the right side of the above equation as

γ ν γ µ p0ν + γ µ γ ν pν = 21 (γ µ γ ν + γ ν γ µ )(p0ν + pν ) − 12 (γ µ γ ν − γ ν γ µ )(p0ν − pν )


= 21 {γ µ , γ ν }(p0ν + pν ) − 21 [γ µ , γ ν ]qν
= p0µ + pµ + iσ µν qν , (3.3.8)

where use was made of the defining characteristic of the γ matrics: {γ µ , γ ν } = 2g µν , Eqn. 3.22 p.40 P&S. We
also used the definition of σ µν on p.49 of P&S: −iσ µν = 12 [γ µ , γ ν ].
Plugging this into Eqn. (3.3.7) and dividing by 2m yields the desired result:
 0µ
p + pµ iσ µν qν

ū(p0 )γ µ u(p) = ū(p0 ) + u(p) . (3.3.9)
2m 2m

3.4 Quadratic Shift Symmetry


Consider the following nonrelativistic scalar field theory (known as the “Lifshitz scalar”) in D + 1 dimensions
(which we will parametrize by Cartesian coordinates t and xi , i = 1, . . . , D):
Z
1
dt dD x (∂t φ)2 − (∂i ∂i φ)2 .

S= (3.4.1)
2
This theory is invariant under the following “quadratic shift” symmetry,

φ → φ + aij xi xj + ai xi + a, (3.4.2)

where aij , ai and a are real, spacetime-independent constants.


3.4. QUADRATIC SHIFT SYMMETRY 27

Derive the Noether current for this symmetry (following the strategry outlined in Chapter 2.2 of [PS]), and
prove that it is conserved.

Reference: This symmetry has played a central role in our recent paper, arXiv:1308.5967, Multicritical Symme-
try Breaking and Naturalness of Slow Nambu-Goldstone Bosons (by T. Griffin, K. Grosvenor, Z. Yan & P.H.).

SOLUTION: (Thanks to Ryan J. (’13) for presenting his solution.)

Define

φ̇ ≡ ∂t φ, ∂ 2 φ ≡ ∂i ∂i φ. (3.4.3)

Let us consider the pieces of the transformation separately.

Constant shift: Plugging δφ = a into δL directly clearly gives δL = 0 and so Jt = 0 and Ji = 0. But
∂L ∂L ∂L ∂L
δL = δφ + δ φ̇ + δ(∂i φ) + δ(∂ 2 φ) + · · ·
∂φ ∂ φ̇ ∂(∂i φ) ∂(∂ 2 φ)
       
∂L ∂ ∂L ∂ ∂L ∂L ∂L 
= δφ + δφ −  δφ + ∂i δφ − ∂i   δφ
∂φ ∂t ∂ φ̇ ∂t ∂ φ̇ ∂(∂i φ)  ∂(∂i φ)
       
∂L ∂L ∂L 
+ ∂i ∂i δφ − ∂i ∂i δφ + ∂ 22  δφ + · · · . (3.4.4)
∂(∂ 2 φ) ∂(∂ 2 φ)  ∂(∂ φ)

where · · · stands for derivatives with respect to higher and higher derivatives of φ.
By definition, the terms proportional to δφ combine to give the equation of motion and thus we can drop
them. For the action, Eqn. (3.4.1), we have
∂L
= 0. (3.4.5)
∂(∂i φ)
In addition, all higher derivatives vanish as well. Therefore, we get
     
∂ ∂L ∂L ∂L
δL = δφ − ∂i ∂i δφ − ∂ i δφ . (3.4.6)
∂t ∂ φ̇ ∂(∂ 2 φ) ∂(∂ 2 φ)

When δφ = a, this reads


δL = a∂t φ̇ − a∂i (−∂i ∂ 2 φ). (3.4.7)
Therefore, after peeling off the infinitesimal constant parameter, a, the currents are

Jt = φ̇ , Ji = −∂i ∂ 2 φ . (3.4.8)

Current conservation in this case is identical to the equation of motion:

∂t Jt − ∂i Ji = φ̈ + ∂ 4 φ = 0 , (3.4.9)

where ∂ 4 ≡ (∂ 2 )2 .

Linear shift: Again, direct calculation gives δL = 0 and so (J j )t = 0 and (J j )i = 0. Substitution of δφ = ai xi


into Eqn. (3.4.6) gives
δL = aj ∂t (φ̇xj ) − aj ∂i −(∂i ∂ 2 φ)xj + (∂ 2 φ)δij .
 
(3.4.10)
Therefore,
(J j )t = xj φ̇ , (J j )i = (−xj ∂i + δij )∂ 2 φ . (3.4.11)
Indeed, this is conserved:

∂t (J j )t − ∂i (J j )i = xj ∂ 4 φ) + δij ∂i ∂ 2 φ − ∂ j ∂ 2 φ = 0 ,
 
(φ̈
+ (3.4.12)

where the first term vanishes again by the equation of motion, and the last two terms cancel each other.
28 CHAPTER 3. PROBLEM SET 3

Note the notation here: upper indices inside the parentheses enumerate the different currents (one per in-
finitesimal parameter) while the lower indices are just the space and time components.

Quadratic shift: In this case,


δL = −2(∂ 2 φ)ajk δ jk = −2ajk ∂i (δ jk ∂i φ). (3.4.13)

Therefore,

(J jk )t = 0, (J jk )i = 2δ jk ∂i φ. (3.4.14)

Plugging δφ = ajk xj xk into Eqn. (3.4.6) yields

δL = ajk ∂t (φ̇xj xk ) − ajk ∂i −(∂i ∂ 2 φ)xj xk + (∂ 2 φ)(xj δik + xk δij ) .


 
(3.4.15)

Therefore,

(J jk )t = xj xk φ̇ , (J jk )i = (−xj xk ∂i + xj δik + xk δij )∂ 2 φ − 2δ jk ∂i φ . (3.4.16)

Indeed, this is conserved:

∂t (J jk )t − ∂i (J jk )i = xj xk ∂ 4 φ) + (xj δik + xk δij )∂i ∂ 2 φ − (xj δik + xk δij )∂i ∂ 2 φ


 
(φ̈
+
− 2δij δik ∂ 2 φ + 2δ jk ∂ 2 φ (3.4.17)
= 0.

3.5 Discussion 1: Cubic Shift Symmetry


Let δφ = ajk` xj xk x` , then

δL = −2ajk` (∂ 2 φ)∂i (xj δ k` + xk δ j` + x` δ jk )


= −2ajk` ∂i (xj δ k` + xk δ j` + x` δ jk )∂i φ − (δij δ k` + δik δ j` + δi` δ jk )φ .
 
(3.5.1)

However, using Eqn. (3.4.6) gives

δL = ajk` ∂t (xj xk x` φ̇) − ajk` ∂i −xj xk x` ∂i ∂ 2 φ + (δij xk x` + δik xj x` + δi` xj xk )∂ 2 φ .


 
(3.5.2)

Therefore,
(J jk` )t = xj xk x` φ̇,
(J jk` )i = (−xj xk x` ∂i + δij xk x` + δik xj x` + δi` xj xk )∂ 2 φ . (3.5.3)
− 2(x δj k`
+x δk j` ` jk
+ x δ )∂i φ + 2(δij δ k` + δik δ j` + δi` δ jk )φ

3.6 Discussion 2: Alternative Method


Constant shift: Promote δφ = a to a spacetime-dependent transformation a → a(t, x). Then,
Z
δS = dt d3 x φ̇ ȧ − (∂ 2 φ)∂ 2 a
 

Z
= dt d3 x φ̇ ȧ + (∂i ∂ 2 φ)∂i a .
 
(3.6.1)

Note that we have performed an integration by parts on the second term to get it into the desired form. Then,
J t is the coefficient of ȧ and J i is the coefficient of −∂i a. This derives the same currents as in Eqn.(3.4.8).
3.6. DISCUSSION 2: ALTERNATIVE METHOD 29

Linear shift: Now, δφ = aj (t, x)xj . Then,


Z
δS = dt d3 x φ̇ ȧj xj − (∂ 2 φ)∂ 2 (aj xj )
 

Z
= dt d3 x xj φ̇ ȧj − (∂ 2 φ)(xj ∂ 2 aj + 2δij ∂i aj )
 

Z
= dt d3 x xj φ̇ ȧj − (−xj ∂i ∂ 2 φ − δij ∂ 2 φ + 2δij ∂ 2 φ)∂i aj
 

Z
= dt d3 x xj φ̇ ȧj − (−xj ∂i ∂ 2 φ + δij ∂ 2 φ)∂i aj .
 
(3.6.2)

Again, we read off the same currents as in Eqn. (3.4.11).

Quadratic shift: Now, δφ = ajk (t, x)xj xk . Then,


Z
δS = dt d3 x φ̇ ȧjk xj xk − (∂ 2 φ)∂ 2 (ajk xj xk )
 

Z
= dt d3 x xj xk φ̇ ȧjk − (∂ 2 φ)[xj xk ∂ 2 ajk + 2(xj δik + xk δij )∂i ajk + 2δ jk ajk ]


Z
= dt d3 x xj xk φ̇ ȧjk − [−xj xk ∂i ∂ 2 φ − (xj δik + xk δij )∂ 2 φ + 2(xj δik + xk δij )∂ 2 φ − 2δ jk ∂i φ]∂i ajk


Z
= dt d3 x xj xk φ̇ ȧjk − [−xj xk ∂i ∂ 2 φ + (xj δik + xk δij )∂ 2 φ − 2δ jk ∂i φ]∂i ajk .

(3.6.3)

This gives the same currents as in Eqn. (3.4.16).

Cubic shift: Now, δφ = ajk` (t, x)xj xk x` . Then,


Z
δS = dt d3 x φ̇ ȧjk` xj xk x` − (∂ 2 φ)∂ 2 (ajk` xj xk x` )
 

Z
= dt d3 x xj xk x` φ̇ ȧjk` − (∂ 2 φ)[xj xk x` ∂ 2 ajk` + 2(xj xk δi` + xj x` δik + xk x` δij )∂i ajk`


+ 2(xj δ k` + xk δ j` + x` δ j` )ajk` ]

Z
= dt d3 x xj xk x` φ̇ ȧjk` − [−xj xk x` ∂i ∂ 2 φ − (xj xk δi` + xj x` δik + xk x` δij )∂ 2 φ


+ 2(xj xk δi` + xj x` δik + xk x` δij )∂ 2 φ − 2(xj δ k` + xk δ j` + x` δ j` )∂i φ]∂i ajk`


+ 2(δij δ k` + δik δ j` + δi` δ jk )(∂i φ)ajk`

Z
= dt d3 x xj xk x` φ̇ ȧjk` − [−xj xk x` ∂i ∂ 2 φ + (xj xk δi` + xj x` δik + xk x` δij )∂ 2 φ


− 2(xj δ k` + xk δ j` + x` δ j` )∂i φ + 2(δij δ k` + δik δ j` + δi` δ jk )φ]∂i ajk` .



(3.6.4)

This gives the same currents as in Eqn. (3.5.3).


30 CHAPTER 3. PROBLEM SET 3
Chapter 4

Problem Set 4

4.1 Advanced and Retarded Propagators, Zee I.3.3, p.24


Show that the advanced propagator defined by

d4 k eik(x−y)
Z
DA (x − y) = , (4.1.1)
(2π)4 k 2 − m2 − i sgn(k0 )

(where the sign function is defined by sgn(k0 ) = +1 if k0 > 0 and sgn(k0 ) = −1 if k0 < 0) is nonzero only
if x0 > y 0 . In other words, it only propagates into the future. [Hint: both poles of the integrand are now
in the upper half of the k0 -plane.] Incidentally, some authors prefer to write (k0 − i)2 − |k|2 − m2 instead of
k 2 − m2 − isgn(k0 ) in the integrand. Similarly, show that the retarded propagator,

d4 k eik(x−y)
Z
DR (x − y) = , (4.1.2)
(2π) k − m2 + i sgn(k0 )
4 2

propagates into the past.

SOLUTION:

For the advanced propagator, the roots are located at


p p
k0 = ± |k|2 + m2 + i sgn(k0 ) = ± |k|2 + m2 ± i. (4.1.3)

p ± signs are correlated.  is just an infinitesimal parameter. We are free to redefine it via, for
Note that the two
example,  → 2 |k|2 + m2 . Doing so allows us to expand to linear order in :
p  p
k0 ≈ ± |k|2 + m2 ± i = ± |k|2 + m2 + i. (4.1.4)

As stated in the hint, both roots have a small positive imaginary part and are thus both in the upper half of
the k0 -plane. If x0 < y 0 , then the exponential in k0 reads e−ik0 (y0 −x0 ) . We must complete the countour in
the lower half k0 -plane so that, along the contour at a large radial distance from the origin, the exponential is
e−positive and huge and thus vanishing. This contour does not enclose the roots, which are both in the upper half
k0 -plane. Thus, the residue, the whole integral and the propagator vanish.
This argument is repeated wholesale for the retarded propagator except that the roots are now in the lower
half k0 -plane and, when x0 > y 0 , we must complete the contour in the upper half k0 -plane, once again missing
the roots.

31
32 CHAPTER 4. PROBLEM SET 4

4.2 Force Law in Arbitrary Dimension, Zee I.4.1 p.31


Calculate the analog of the inverse square law in a (2 + 1)-dimensional universe, and more generally in a (D + 1)-
dimensional universe.

SOLUTION:

We want to calculate the integral in Eqn. 6 of Zee p.28 in D spatial dimensions:


dD k eik·(x1 −x2 )
Z
E=− . (4.2.1)
(2π)D |k|2 + m2
Take the spatial Laplacian of E in the m → 0 limit:
dD k |k|2 eik·(x1 −x2 ) m→0 dD k ik·(x1 −x2 )
Z Z
2
∇ E= −−−→ e = δ(x1 − x2 ). (4.2.2)
(2π)D |k|2 + m2 (2π)D
This is Poisson’s equation for a point source. In E&M, this would be Poisson’s equation for a point charge with
charge 1/4π in Gaussian units. Since the force is defined via F = −∇E, the equation for the force is
∇ · F = −δ(x1 − x2 ). (4.2.3)
By spherical symmetry about the point x1 − x2 = 0, we know that the energy, E, and the force, F, can only be
a function of r = |r|, where r = x1 − x2 , and that F must point radially: F = F (r)r̂ (radially away if F (r) > 0
and towards the origin if F (r) < 0). By Gauss’ law, or the divergence theorem, we know that the flux of F
over a sphere of radius r centered at the origin must be equal to −1, which is the volume integral of its source,
−δ(r), inside the region enclosed by the sphere. Since F has constant magnitude over this sphere and points
purely radially, the flux is simply F (r) multiplied by the surface area of the sphere (circumference for a 1-sphere
or a circle, usual area for a 2-sphere, and so on). In D dimensions, this sphere is a (D − 1)-sphere. So, in
two dimensions, it is a circle, S 1 , in three it is a usual sphere, S 2 , in four it is a 3-sphere, S 3 , and so on. The
circumference of S 1 is 2πr, the area of S 2 is 4πr2 . The generalization of area to S D−1 is
D
D−1 2π 2 rD−1
A(S )= . (4.2.4)
Γ( D2)

By the argument above, it follows that


Γ( D
2)
F=− D r̂ . (4.2.5)
2π 2 rD−1
Note that this is always attractive.
Let us write down the two most familiar lower-dimensional cases:
r̂ r̂
D=2: F=− , D=3: F=− . (4.2.6)
2πr 4πr2
For D = 2, the corresponding energy is
1
D=2: E= ln r . (4.2.7)

Of course, r is dimensionfull, so ln r is technically nonsensical. This is the usual problem that appears in E&M
(e.g. electric potential due to a uniformly charged infinite line) where a “boundary” condition, where the potential
vanishes, has to be chosen at some arbitrary finite distance, r0 , and not r → ∞. Then, E ∼ ln(r/r0 ).
For D > 2, the corresponding energy is
Γ( D
2) Γ( D
2)
E=− D =− D . (4.2.8)
2π (D − 2)rD−2
2 4π ( D
2 − 1)r
2 D−2

Using the property of the Gamma function: Γ(x) = (x − 1)Γ(x − 1), we can write E as

Γ( D
2 − 1)
D>2: E=− D . (4.2.9)
4π 2 rD−2
4.2. FORCE LAW IN ARBITRARY DIMENSION, ZEE I.4.1 P.31 33

We can also just calculate the integral (4.2.1). Write the denominator as an integral:
Z ∞
1 2
+m2 )s
= e−(|k| ds. (4.2.10)
|k| + m2
2
0

Doing so allows us to write the energy as



dD k −s|k|2 +ik·(x1 −x2 )−sm2
Z Z
E=− ds e . (4.2.11)
0 (2π)D

Let r = x1 − x2 , r = |r| and k = |k|. We can complete the square in the exponential:

r2 r2 i 2 r2
− sk 2 + ik · r − sm2 = −s k 2 − si k · r − − sm2 = −s k − − sm2 .

4s2 − 4s 2s r − 4s (4.2.12)
√ D
Define q = s(k − i
2s r) and q = |q|. Then, dD k = s− 2 dD q and the energy is

dD q −q2
Z Z
D r2 2
E=− s− 2 e− 4s −sm ds e . (4.2.13)
0 (2π)D

Note that we have been cavalier about the region of integration over q. Relative to k, q has an imaginary part.
However, the integrand is simply Gaussian, which is analytic. Therefore, we can deform the contour of integration
over the q coordinates towards the real line with impunity. The q integral is
∞ D  √ D
dD q −q2
Z Z
dq −q2 π D
e = e = = 2−D π − 2 . (4.2.14)
(2π)D −∞ 2π 2π

Define t = sm2 so that Z ∞


D D (mr)2
E = −2−D π − 2 mD−2 t− 2 e− 4t −t
dt. (4.2.15)
0

An integral expression for the modified Bessel function of the second kind is
Z ∞
zν z2
Kν (z) = ν+1 t−ν−1 e− 4t −t dt. (4.2.16)
2 0

D
Therefore, we can write E using Kν (z) with ν = 2 − 1 and z = mr:

D D D 1  m  D2 −1
E = −2−D π − 2 mD−2 2 2 (mr)1− 2 K D −1 (mr) = − D K D −1 (mr). (4.2.17)
2
(2π) 2 r 2

We are interested in taking the m → 0 limit. Wikipedia gives the small argument limit of K:
(
− ln z2 − γ if ν = 0,

z→0
Kν (z) −−−→ Γ(ν) 2 ν (4.2.18)
2 z if ν > 0.

Here γ is the Euler-Mascheroni constant, which will be irrelevant for our purposes anyway.
Therefore, when D = 2, or in 2 + 1 spacetime dimensions,

1 m→0 1
D=2: E=− K0 (mr) −−−→ ln r , (4.2.19)
2π 2π

1
where we have dropped the constant 2π (γ − ln 2) since the energy is only well-defined up to an overall constant
anyway. For D > 2,

D
Γ D
 m  D2 −1 Γ  
m→0 1 2 − 1  2  D2 −1 2 −1
D>2: E −−−→ − D =− D . (4.2.20)
(2π) 2 r 2 mr 4π 2 rD−2
34 CHAPTER 4. PROBLEM SET 4

4.3 Graviton Propagator, Zee I.5.1 p.39


P (a) (a)
Write down the most general form for a εµν (k)ελσ (k) using symmetry repeatedly. For example, it must be
invariant under the exchange {µν ↔ λσ}. You might end up with something like

AGµν Gλσ + B(Gµλ Gνσ + Gµσ Gνλ ) + C(Gµν kλ kσ + kµ kν Gλσ )


+ D(kµ kλ Gνσ kµ kσ Gνλ + kν kσ Gµλ + kν kλ Gµσ ) + Ekµ kν kλ kσ , (4.3.1)
P (a) (a)
with various unknown A, . . . , E. Apply k µ a εµν (k)ελσ (k) = 0 and find out what that implies for the constants.
Proceeding in this way, derive Eqn. 13 p.35:
X (a) 2
ε(a)
µν (k)ελσ (k) = (Gµλ Gνσ + Gµσ Gνλ ) − Gµν Gλσ . (4.3.2)
a
3

SOLUTION: (Thanks to Byungmin (’12) and Yixing (’13) for presenting their solutions.)

Denote the sum over products of polarizations by Iµνλσ . This is comprised of terms with four indices. To build
it out of Gµν and kµ we either have the product of two G’s, one G and two k’s, or four k’s. The symmetries of
Iµνλσ that we use are that it is invariant under the switches

(a) µ ↔ ν, (b) λ ↔ σ, (c) µ ↔ λ and ν ↔ σ.

All other symmetries are compositions of these three.


Let us consider first the case of two G’s. There are only three independent possibilities (noting that Gµν is
symmetric): Gµν Gλσ or Gµλ Gνσ or Gµσ Gνλ . The first is its own orbit under all the symmetries above and their
compositions whereas the last two form a single orbit. Therefore, these give an expansion term

A Gµν Gλσ + B (Gµλ Gνσ + Gµσ Gνλ ). (4.3.3)

Now, let us consider the case of one G and two k’s. There are six possibilities:

(1) Gµν kλ kσ (3) kµ kλ Gνσ , (5) kν kσ Gµλ ,


(2) kµ kν Gλσ , (4) kµ kσ Gνλ , (6) kν kλ Gµσ .

(1) and (2) form a single orbit excluding the other four which can only be reached from (1) by switches such as
µ ↔ λ, which are not valid symmetries. (4) is reached from (3) by by (b); (5) is reached from (4) by (a); and (6)
is reached from (5) by (b). Thus, the last four form a single orbit.
Therefore, we have an expansion term of the form

C (Gµν kλ kσ + kµ kν Gλσ ) + D (kµ kλ Gνσ + kµ kσ Gνλ + kν kσ Gµλ + kν kλ Gµσ ). (4.3.4)

Finally, the only possible term with four k’s gives an expansion term

E kµ kν kλ kσ . (4.3.5)

Thus, we have derived Eqn. (4.3.1).


Since Gµν = gµν − m12 kµ kν , we have k µ Gµν = 0. Thus, all terms in Iµνρσ where the µ index is in a G vanish
once we multiply with k µ . The remainder is

0 = k µ Iµνλσ = Cm2 kν Gλσ + Dm2 (kλ Gνσ + kσ Gνλ ) + Em2 kν kλ kσ . (4.3.6)

Left multiplying (4.3.6) by g νλ gets rid of the first and second terms. Note that the trace of G is 3: g µν Gµν =
g µν gµν − k µ kµ /m2 = 4 − 1 = 3. Thus, we get

0 = k µ k σ g νλ Iµνλσ = 3Dm4 + Em6 . (4.3.7)

Left multiplying (4.3.6) by g λσ gets rid of the second and third terms and yields the equation

0 = k µ k ν g λσ Iµνλσ = 3Cm4 + Em6 . (4.3.8)


4.3. GRAVITON PROPAGATOR, ZEE I.5.1 P.39 35

The only solution is that C, D and E all vanish:

C=D=E=0. (4.3.9)

At this point, we are left with only the A and B terms in (4.3.3). Now note that
kν kσ kλ kµ µν kλ kµ µν kν kσ
g µν Gµλ Gνσ = gλµ g µν gνσ − gλµ g µν − g gνσ + g
m2 m2 m2 m2
kλ kσ
= gλσ −
m2
= Gλσ . (4.3.10)

By symmetry, Eqn. (4.3.10) is also equal to g µν Gµσ Gνλ . So, left multiplying (4.3.2) by g µν gives

0 = g µν Iµνλσ = (3A + 2B)Gλσ , (4.3.11)


(a)
where the first equality follows from the traceless condition g µν εµν = 0. Thus, A = −2B/3 and we may write
 2 
Iµνλσ = B Gµλ Gνσ + Gµσ Gνλ − Gµν Gλσ (4.3.12)
3
Let us evaluate this at µ = λ = 1 and ν = σ = 2 and in the rest frame, k µ = (m, 0, 0, 0), where Zee gives us the
normalization condition
2  
1 = I1212 = B g11 g22 + 
g1221 − 
g 3 g12 g12 = B. (4.3.13)
Therefore,
X (a) 2
ε(a)
µν (k)ελσ (k) = (Gµλ Gνσ + Gµσ Gνλ ) − Gµν Gλσ . (4.3.14)
a
3

Discussion: Let us clarify why we care about the sum of products of polarizations. The point is in the expansion
of the field in creation and annihilation operators. For a scalar field we simply have
Z
φ ∼ · · · ae··· + h.c. (4.3.15)

Here, h.c. stands for Hermitian conjugate. Really, there is a polarization, but there is only one and it is a scalar:
ε = 1. It couldn’t be simpler. We can write
Z
φ ∼ · · · εae··· + h.c. (4.3.16)

That is, there is one annihilation operator and it kills a particle with polarization ε = 1. Propagation of a particle
from x to y is encoded in the expectation value h0|φ(y)φ(x)|0i. Indeed, the only nonvanishing term is the one
involving the creation operator in φ(x) and the annihilation operator in φ(y) and describes the creation of a
particle at x and its destruction at y (i.e. propagation of a particle from x to y). Since a and a† are multiplied
by the polarization, ε, this expectation value also involves the sum over products εε. This is seems like a silly
statement since there is only one such product and it is just equal to 1. However, it is less trivial for fields with
more components. For example, for the photon, the field is the vector potential. By analogy with Eqn. (4.3.16),
we have Z X
(a) ···
Aµ ∼ · · · ε(a)
µ a e + h.c. (4.3.17)
a

Now, there are three polarizations, one for each value of the superscript a = 1, 2, 3. Each polarization is a 4-
vector having the same spacetime index structure as the field itself. There is one set of creation and annihilation
operators for each polarization since you can create or destroy each type separately. In this case, we would like
P (a) (b)
to calculate h0|Aµ (y)Aν (x)|0i. The only surviving term involves ab εµ εν a(a) a(b)† . After commutation the
creation operator past the annhiliation operator, we pick up a factor of δ (a)(b) , which is the reason why we pick
P (a) (a)
up a factor of a εµ εν .
You can generalize this to gravity. In that case, we write the metric as a perturbation about flat spacetime,
gµν = ηµν + hµν , and then we expand hµν in creation and annihilation operators. In this case, there are five
polarizations and they are 2-tensors since they have the same spacetime index structure as hµν .
36 CHAPTER 4. PROBLEM SET 4

Now, let us discuss the conditions imposed on these polarizations. For polarizations with at least one spacetime
(a)
index, it is customary to impose the transversality condition, k µ εµ··· = 0. Usually, this can be arranged in the
rest frame, but since the inner product is Lorentz invariant, it must hold in any inertial frame. However, it is
possible to be silly and write down polarizations in the rest frame, which do not satisfy these conditions. This
freedom for “silliness” is due to gauge invariance. The massive vector field is described by a 4-vector even though
there are only three degrees of freedom: the theory is over-complete. We have to somehow eliminate one would-be
degree of freedom. This process is called fixing the gauge. A standard gauge for the vector field is the Lorenz
gauge, which imposes ∂ µ Aµ = 0. Since Aµ is written as in Eqn. (4.3.17), this immediately implies k µ εµ = 0.
For gravity, a priori hµν has ten degrees of freedom, but there are only supposed to be five! The analog of the
Lorenz gauge, ∂ µ hµν = 0, gives four constraints (one per value of the ν index). We need one more condition to get
us from ten to five degrees of freedom. The usual choice is the tracelessness condition, g µν hµν = 0 (customarily
we keep only the leading term, which is η µν hµν = 0). Unsurprisingly, this set of gauge fixing conditions is called
the “transverse-traceless” gauge.

4.4 Discussion 1: Grassmann Variables


References:

Steven Weinberg, The Quantum Theory of Fields Vol. 1 Section 4.2 p.173.

A. Zee, Quantum Field Theory in a Nutshell Ed. 2 Section II.5 p.123.

Jean Zinn-Justin, Quantum Field Theory and Critical Phenomena Ed. 4 Sections I.4-7 pp.6-16.

Mikio Nakahara, Geometry, Topology and Physics Ed. 2 Section 1.5 pp.38-51.

Let Gn be the Grassmann algebra of dimension n defined to be the algebra (either real or complex) generated
by {η1 , · · · , ηn }, which satisfy the anticommutation relations

{ηi , ηj } = 0. (4.4.1)

An element f of Gn can be written as


n n
X 1 X
f= fi ···i ηi · · · ηik , (4.4.2)
k! i ,...,i =1 1 k 1
k=0 1 k

where fi1 ···ik are real or complex and fully antisymmetric. Of course, the k = 0 term is just a number (real or
complex).
A function of Grassmann numbers is defined to be the Taylor expansion. Note that this Taylor expansion
must end at order n at most. A famous example is the exponential for n = 1:

eη = 1 + η. (4.4.3)

All the higher order terms vanish because {η, η} = 0 implies η 2 = 0.


Differentiation is defined fairly intuitively:

ηi = δij . (4.4.4)
∂ηj

However, because of the anticommutation relations, the derivative is forced to behave like a Grassmann number
(i.e. it anticommutes with the generators):


(ηj ηk ) = δij ηk − δik ηj . (4.4.5)
∂ηi
You can check that this is consistent with the anticommutation relations among the generators by considering
∂ 2 2
∂η (η ). Since η = 0, this derivative had better vanish. If the derivative were to commute with the generators
and hence give the usual Leibniz rule instead of the modified one above, then we would get 2η instead of 0.
4.4. DISCUSSION 1: GRASSMANN VARIABLES 37

One can use these relations to prove that the derivative operators anticommute (as an operator equation):
 
∂ ∂
, = 0, (4.4.6)
∂ηi ∂ηj

which automatically implies that differentiation is nilpotent:

∂2
= 0. (4.4.7)
∂ηi2

Integration is much weirder. It is defined operationally in terms of the derivative. Firstly, we want the integral
of a derivative to be zero (ignore boundary terms). Secondly, we want the derivative of an integral to be zero.
Thirdly, we want to be able to pull real or complex constants out of an integral. The modified Leibniz rule
and nilpotency of the derivative gives us a natural candidate that satisfies these conditions: the derivative itself!
Thus, we define integration to be differentiation:
Z
∂ ∂
dη1 · · · dηn f (η1 , . . . , ηn ) = ··· f (η1 , . . . , ηn ). (4.4.8)
∂η1 ∂ηn
Note that the order of integrations and differentiations matters: they have the same order. You do the ηn integra
first by doing the ηn derivative first, and so on all the way to η1 .
For n = 1, this gives the same results “derived” differently in Zee:
Z Z
∂ ∂
dη = 1 = 0, dη η = η = 1. (4.4.9)
∂η ∂η

Let n = 1 and consider the integral dη f (η) under a change of variables, which is necessarily of the form aη 0 = η
R

for some real or complex number a. Then,


Z Z
∂ ∂ 0 −1 ∂ 0 −1
dη f (η) = f (η) = f (aη ) = a f (aη ) = a dη 0 f (aη 0 ). (4.4.10)
∂η ∂(aη 0 ) ∂η 0

This is quite strange. Usually, if η = aη 0 , we would expect dη = a dη 0 not a−1 dη 0 , but we get the latter
R R R

because the integral is defined to be the derivative!


More generally,
dη1 · · · dηn = dη10 · · · dηn0 J(η 0 ), (4.4.11)
with

J −1 = det ηi . (4.4.12)
∂ηj0
Again, this is the inverse of what you would expect for normal commuting variables.
We can now compute Gaussian Grassmann integrals:
Z
Z = dη1 dη̄1 · · · dηn dη̄n eη̄i Aij ηj
n
Z Y 
0
= det A dηi0 dη̄i eη̄j ηj
i=1
n
Z Y 
η̄i ηi0
= det A dηi0 dη̄i e (4.4.13)
i=1

In the second line, we have changed variables to ηi0 = Aij ηj . Note that repeated indices are summed over except
when they are explicitly in a product. To get the third line, we first split up the exponential and then redistributed
it among the integration measures. For example, for n = 2:
Z Z
0 0 0 0
dη10 dη̄1 dη20 dη̄2 eη̄1 η1 +η̄2 η2 = dη10 dη̄1 dη20 dη̄2 eη̄1 η1 eη̄2 η2
Z
0 0
= dη10 dη̄1 eη̄1 η1 dη20 dη̄2 eη̄2 η2 . (4.4.14)
38 CHAPTER 4. PROBLEM SET 4

The reason why we can do this is that the exponential is a function of products of two Grassmann variables.
Products of an even number of Grassmann variables commute with everything by the obvious fact that an even
power of −1 is +1. Thus, we can split the exponential (first line above) and move it freely around (second line
above). Now, we simply expand the exponential adn integrate:
Z Yn 
0 0

Z = det A dηi dη̄i 1 + η̄i ηi = det A. (4.4.15)
i=1
R
We used the fact that of 1 is 0 and dη η = 1. Note that the order in which the variables are written is important.
It is important that the η̄ integral is rightmost (done first) and that the η̄ factor in the exponential is written
on the left. If we switch some of these, we pick up corresponding minus signs. This result is contrasted with the
result for a bosonic integral, which yields ∼ (det A)−1/2 .
A source term for η would have to be Grassmann as well and of the same dimension. Therefore, consider a
second copy of the Grassmann algebra with generators θi and θ̄i and consider
Z
Z[θ, θ̄] = dη1 · · · dη̄n eη̄i Aij ηj +θ̄i ηi +η̄i θi . (4.4.16)

Note that the only reason why we insist on writing half of the Grassmann variables with bars is to make contact
with the notation of spinor fields where tersm in the free action contain one copy of ψ̄ and one copy of ψ. Also,
the reason why we insist on writing barred variables on the left of unbarred ones and the reverse order in the
integration measure is because that is the order in which they appear in the path integral for spinor fields. Indeed,
a source term for a spinor field would have to be a spinor and would enter in the Lagrangian as Jψ ¯ + ψ̄J.
0 0
Change variables to η and η̄ defined via

ηi = ηi0 − (A−1 )ij θj , η̄i = η̄i0 − θ̄j (A−1 )ji . (4.4.17)

This serves to “complete the square” after which the integral is straightforward and yields
−1
Z[θ, θ̄] = (det A) e−θ̄i (A )ij θj
. (4.4.18)

From our experience in the Bosonic case, we know that we simply take various derivatives with respect to the
source to pull down factors of the field and get expectation values of products of fields. In fact, we derived
Wick’s theorem in this way. We can do the same thing here. We just have to be very very careful about moving
derivatives past Grassmann variables and taking
care of signs. In the case of the spin-1/2 field, we might be
interested in the propagator, which would be ψ̄ψ . Note that for spinors, the order is really important: ψ̄ψ is a
scalar; ψ ψ̄ is a matrix! Let us take a derivative to pull down a barred field:
Z
∂ ∂
Z= dη1 · · · dη̄n eη̄k Ak` η` +θ̄k ηk +η̄k θk
∂θi ∂θi
Z
∂ η̄k Ak` η` +θ̄k ηk +η̄k θk
= (−1)2n dη1 · · · dη̄n e . (4.4.19)
∂θi
I wrote (−1)2n explicitly to remind you to not take anything for granted. You cannot just move the derivative
past a dη willy-nilly; you pick up a minus sign. However, since there is an even number in the integration measure,
we can actually move the derivative past the whole measure because it produces an even number of minus signs.
As discussed earlier, since the exponential is a function of products of even numbers of Grassmann variables,
we can separate the exponential factors and move them around freely. In the following expression, repeated i
indices are NOT summed over:
Z  
∂ ∂ η̄i θi η̄k Ak` η` +θ̄k ηk +η̄k θk −η̄i θi
Z = dη1 · · · dη̄n e e . (4.4.20)
∂θi ∂θi
Now, we can evaluate the derivative (again, repeated i indices are not summed over):
∂ η̄i θi ∂
e = (1 + η̄i θi ) = −η̄i . (4.4.21)
∂θi ∂θi
We can do something seemingly dumb, but quite useful: we can multiply this by the exponential and it would
not change:
− η̄i eη̄i θi = −η̄i (1 + η̄i θi ) = −η̄i − η̄i2 θi = −η̄i . (4.4.22)
4.4. DISCUSSION 1: GRASSMANN VARIABLES 39

Therefore, we can write


∂ θ̄j ηj
e = ηj eθ̄j ηj , (4.4.23)
∂ θ̄j
which is more remeniscent of standard differentiation in Bosonic variables.
Altogether then, we get Z

Z = dη1 · · · dη̄n (−η̄i ) eη̄k Ak` η` +θ̄k ηk +η̄k θk . (4.4.24)
∂θi
On the other hand, using Eqn. (4.4.18), we find
∂ −1
Z = (det A) θ̄j (A−1 )ji e−θ̄k (A )k` θ` . (4.4.25)
∂θi
Note that there is a cancellation of two minus signs in the above equation. One minus sign comes from the minus

sign already in the exponential and the other comes from the fact that we have to move ∂θ i
past θ̄k first before
it can act on θ` .
It follows that
1 ∂ −1
= −θ̄j (A−1 )ji e−θ̄k (A )k` θ` θ=θ̄=0 = 0.

hη̄i i = − Z (4.4.26)
det A ∂θi θ=θ̄=0
The expectation value of η̄i similarly vanishes:

1 ∂ −1
= −(A−1 )ji θi e−θ̄k (A )k` θ` θ=θ̄=0 = 0.

hηj i = Z (4.4.27)
det A ∂ θ̄j θ=θ̄=0

Please make sure you understand the signs in the above equations.
Let us calculate hη̄i ηj i, which in the case of the spinor field represents the propagator. Consider
Z
∂ ∂ ∂ 
Z = dη1 · · · dη̄n (−η̄i ) eη̄k Ak` η` +θ̄k ηk +η̄k θk ]
∂ θ̄j ∂θi ∂ θ̄j
Z
∂ η̄k Ak` η` +θ̄k ηk +η̄k θk
= dη1 · · · dη̄n η̄i e
∂ θ̄j
Z
= dη1 · · · dη̄n η̄i ηj eη̄k Ak` η` +θ̄k ηk +η̄k θk ] (4.4.28)

On the other hand, using Eqn. (4.4.25), we get


∂ ∂ ∂  −1
θ̄a (A−1 )ai e−θ̄k (A )k` θ`

Z = (det A)
∂ θ̄j ∂θi ∂ θ̄j
−1
= (det A) (A−1 )ji + θ̄a (A−1 )ai (A−1 )jb θb e−θ̄k (A )k` θ` .
 
(4.4.29)

Note that the + sign between the two terms in the square brackets is actually two minus signs cancelling: one
from moving ∂∂θ̄j past θ̄a and one from the minus sign in the exponential.
From Eqns. (4.4.28) and (4.4.29), we find

1 ∂ ∂
hη̄i ηj i = Z = (A−1 )ji . (4.4.30)
det A ∂ θ̄j ∂θi θ=θ̄=0

Note that the order of the subscripts on A−1 is important because A (and therefore A−1 ) is antisymmetric.
Lest we lose sight of the point of all this, when applied to the Dirac field, the above equation implies that
the Dirac propagator is simply the inverse of the operator that sits between ψ̄ and ψ in the Dirac Lagrangian,
namely, i∂/ − m. Well, we have already seen that (i∂/ − m)(i∂/ + m) = −∂ 2 − m2 and so, up to a proportionality
factor, the propagator is
p
/+m
S(p) ∝ 2 . (4.4.31)
p − m2
The Feynman i prescription picks out the Feynman propagator form and then it is conventional to put in a
factor of i. Hence,
d4 p i(p
/ + m)
Z
SF (x − y) = e−ip(x−y) . (4.4.32)
(2π) p − m2 + i
4 2
40 CHAPTER 4. PROBLEM SET 4

Thus, we have shown that by introducing Grassmann variables, we can derive the propagator from the path
integral just as we did for the case of the scalar field. In fact, we can derive higher-point correlation functions as
well by deriving a Fermionic version of Wick’s theorem:
X
hη̄i1 ηj1 · · · η̄in ηjn i = (−1)P (A−1 )jP 1 i1 · · · (A−1 )jP n in , (4.4.33)
P of j’s

where P is a permutation of the j indices and (−1)P is its signature.


For example, for four fields:

hη̄i1 ηj1 η̄i2 ηj2 i = (A−1 )j1 i1 (A−1 )j2 i2 − (A−1 )j2 i1 (A−1 )j1 i2 . (4.4.34)

It would be a very good exercise for you to derive this.


Chapter 5

Problem Set 5

5.1 Majorana Fermions, P&S 3.4, p.73


Denote the two-component spinor, ψL , by χa (a = 1, 2).

(a) Show that it is possible to write an equation for χ(x) as a massive field in the following way:

iσ̄ · ∂χ − imσ 2 χ∗ = 0 (5.1.1)

That is, first show that (5.1.1) is relativistically invariant and that it implies the Klein-Gordon equation,
(∂ 2 + m2 )χ = 0.

(b) Treat χ(x) as a Grassman field and define the action


Z  
4 † im T 2 † 2 ∗

S = d x χ iσ̄ · ∂χ + χ σ χ−χ σ χ (5.1.2)
2

Show that S is real and that varying S with respect to χ and χ∗ yields the Majorana equation (5.1.1).
(c) We may write the four-component Dirac field in terms of two 2-component spinors

ψL (x) = χ1 (x) ψR (x) = iσ 2 χ∗2 (x) (5.1.3)

Rewrite the Dirac Lagrangian in terms of χ1 and χ2 .


(d) Show that the action of part (c) has a global symmetry. Define χ = (χ1 , χ2 )T . Compute the divergences of
the currents

Jbµ = χ† σ̄ µ χ Jcµ = χ†1 σ̄ µ χ1 − χ†2 σ̄ µ χ2 (5.1.4)

for the theories of part (b) and (c), respectively. Relate your results to the symmetries of these theories.
Construct a theory of N free massive two-component fermion fields with O(N ) symmetry.
(e) Quantize the Majorana theory of parts (a) and (b). That is, promote χ(x) to a quantum field satisfying the
canonical anticommutation relation

{χa (x), χ†b (y)} = δab δ (3) (x − y) (5.1.5)

Then, construct a Hermitian Hamiltonian and find a representation of the canonical commutation relations
that diagonalizes the Hamiltonian in terms of a set of creation and annihilation operators.

41
42 CHAPTER 5. PROBLEM SET 5

SOLUTION: (Thanks to Fabio, Paul and Chris (’13) for presenting parts (a), (b) and (c) respectively.)

(a) First, we must show that (5.1.1) is relativistically invariant. From Eqn. 3.37 (P&S, p.44), under infinitesimal
rotations θ and boosts β, the spinors ψL and ψR transform as
 σ σ  σ σ
Λ 12 L ψL = 1 − iθ · − β · ψL , Λ 21 R ψR = 1 − iθ · + β · ψR . (5.1.6)
2 2 2 2

Furthermore, Eqn. 3.29 (P&S, p.42) gives Λ−1 µ µ ν


1 γ Λ 1 = Λ ν γ , which when restricted to the left and right
2
2
spinor spaces gives

Λ−1 µ µ ν
1 σ Λ1R = Λ νσ ,
L 2
Λ−1
1
R
σ̄ µ Λ 12 L = Λµν σ̄ ν . (5.1.7)
2 2

Since χ is a left-handed spinor, σ 2 χ∗ transforms like a right-handed spinor (c.f. comment after Eqn. 3.38
p.44 P&S). Finally, using Eqn. (5.1.7) and some judicious multiplication by 1, we show that (5.1.1) is
relativistically invariant. This procedure is exactly analogous to that on the bottom of PS, p.42:

iσ̄ µ ∂µ χ − imσ 2 χ∗ → iσ̄ µ (Λ−1 )ν µ ∂ν Λ 12 L χ − imΛ 12 R (σ 2 χ∗ )


h i
= Λ 21 R iΛ−11
R
σ̄ µ
Λ 1 (Λ
L
−1 ν
) µ ∂ν χ − imΛ −1
1 Λ 1
R (σ 2 ∗
χ )
2 2R 2
 2 −1 ν ρ 2 ∗

= Λ 2 R i(ΛΛ ) ρ σ̄ ∂ν χ − imσ χ
1

iσ̄ µ ∂µ χ − imσ 2 χ∗ → Λ 21 R iσ̄ ν ∂ν χ − imσ 2 χ∗ .



(5.1.8)

Hence, the same equation is satisfied in frames related via Lorentz transformations. In other words, the
equation is Lorentz invariant.
We must show that Eqn. (5.1.1) implies the Klein-Gordon equation. Left-multiply Eqn. (5.1.1) by −iσ · ∂:

0 = −iσ µ ∂µ iσ̄ ν ∂ν χ − imσ 2 χ∗




= σ µ σ̄ ν ∂µ ∂ν χ − mσ µ σ 2 ∂µ χ∗
= ∂µ ∂ µ χ − mσ 2 σ̄ µ∗ ∂µ χ∗ , (5.1.9)

where use was made of the identities

σ µ σ̄ ν ∂µ ∂ν = ∂µ ∂ µ , σ µ σ 2 = σ 2 σ̄ µ∗ . (5.1.10)

Now, take the complex conjugate of Eqn. (5.1.1) and left multiply by imσ 2 to get

0 = mσ 2 σ̄ µ∗ ∂µ χ∗ + m2 χ, (5.1.11)

where use was made of the identities σ 2∗ = σ 2 and (σ 2 )2 = 1.


Finally, adding Eqns. (5.1.9) and (5.1.11) together gives the Klein-Gordon equation

(∂ µ ∂µ + m2 )χ = 0 (5.1.12)

(b) Let us write the action, (5.1.2) as


Z
i
d4 x χ† σ̄ µ ∂µ χ − ∂µ χ† σ̄ µ χ + m χ| σ 2 χ − χ† σ 2 χ∗ .
 
S= (5.1.13)
2

To do this, we cut the original kinetic term into two and integrated by parts one of the two pieces assuming
that the boundary term vanishes.
Let us write out the first term:
 
∂ −∂ −(∂ −i∂ ) χ1
χ† σ̄ µ ∂µ χ = (χ∗1 , χ∗2 ) −(∂01 +i∂3 2 ) ∂01+∂3 2

χ2

= χ∗1 [(∂0 − ∂3 )χ1 − (∂1 − i∂2 )χ2 ] + χ∗2 [−(∂1 + i∂2 )χ1 + (∂0 + ∂3 )χ2 ]
= −[(∂0 − ∂3 )χ1 − (∂1 − i∂2 )χ2 ]χ∗1 − [−(∂1 + i∂2 )χ1 + (∂0 + ∂3 )χ2 ]χ∗2 (5.1.14)
5.1. MAJORANA FERMIONS, P&S 3.4, P.73 43

Now, let us take the complex conjugate of this:

(χ† σ̄ µ ∂µ χ)∗ = −χ1 [(∂0 − ∂3 )χ∗1 − (∂1 + i∂2 )χ∗2 ] + χ2 [−(∂1 − i∂2 )χ∗1 + (∂0 + ∂3 )χ∗2 ]
  ∗
∂0 −∂3 −(∂1 +i∂2 ) χ1
= −(χ1 , χ2 ) −(∂ 1 −i∂2 ) ∂0 +∂3 χ∗
2
| µ∗ ∗
= −χ σ̄ ∂µ χ
= ∂µ χ† σ̄ µ χ, (5.1.15)

where use was made of the identity η | σ̄ µ∗ ∗ = −† σ̄ µ η. Notice that this term, ∂µ χ† σ̄ µ χ, is precisely negative
of the second term in the action, (5.1.13).
Now, we write out the third term in the action (dropping the m):

χ| σ 2 χ = (χ∗1 , χ∗2 ) 0i −i0 χχ12 = −i(χ1 χ2 − χ2 χ1 ) = 2iχ2 χ1 .


 
(5.1.16)

The complex conjugate of this is

(χ| σ 2 χ)∗ = −2iχ∗1 χ∗2 = −i(χ∗1 χ∗2 − χ∗2 χ∗1 ) = χ† σ 2 χ∗ . (5.1.17)

Again, we recognize this as negative of the last term in the action. Hence, we have shown that
Z Z
i † µ 1
4 | 2
d4 x i χ† σ̄ µ ∂µ χ + mχ| σ 2 χ) + c.c. .
  
S= d x χ σ̄ ∂µ χ + mχ σ χ − c.c. = (5.1.18)
2 2

This is real since it is the sum of complex conjugate terms.


To vary the action with respect to χ† , we revert back to the form of the action in Eqn. (5.1.2) and write
out the mass term explicitly using Eqns. (5.1.16) and (5.1.17):
Z   Z
im
S = d4 x χ† iσ̄ µ ∂µ χ + (2iχ2 χ1 + 2iχ∗1 χ∗2 ) = d4 x[χ† iσ̄ µ ∂µ χ − m(χ2 χ1 + χ∗1 χ∗2 )]. (5.1.19)
2

The variation of the first term with respect to χ† just gives iσ̄ µ ∂µ χ. The variation of the last term with
respect to χ∗1 is −mχ∗2 and with respect to χ∗2 is mχ∗1 . Thus,
 ∗
δS −χ2
0= µ
= iσ̄ ∂µ χ + m = iσ̄ µ ∂µ χ − imσ 2 χ∗ , (5.1.20)
δχ† χ∗1

which is precisely Eqn. (5.1.1).


To take the variation with respect to χ, we first integrate the kinetic term by parts:
Z
S = d4 x[−∂µ χ† iσ̄ µ χ − m(χ2 χ1 + χ∗1 χ∗2 )]. (5.1.21)

Now, χ is on the right, and so when we take variations, we do so on the right. This means that taking the
variation of −mχ2 χ1 with respect to χ1 gives −mχ2 and with respect to χ2 gives mχ1 . Thus,

δS
0= = −i∂µ χ† σ̄ µ + m(−χ2 , χ1 ) = −i∂µ χ† σ̄ µ + imχ| σ 2 , (5.1.22)
δχ

which we note is exactly the adjoint of Eqn. (5.1.20).


(c) Let us first write the Dirac Lagrangian in terms of ψL and ψR :
† †  µ   ψL 
ψ(i∂/ − m)ψ = (ψL , ψR ) 01 10 i σ̄0µ σ0 ∂µ − m ψR
† †  σ µ ∂µ ψR  ψL 
= (ψR , ψL ) i σ̄µ ∂µ ψL − m ψ R

† µ † µ † †
= i(ψR σ ∂µ ψR + ψL σ̄ ∂µ ψL ) − m(ψR ψL + ψL ψR ). (5.1.23)
44 CHAPTER 5. PROBLEM SET 5

Now, let us plug in the forms for ψL and ψR in Eqn. (5.1.3):

ψ(i∂/ − m)ψ = i[(−iχ|2 σ 2 )σ µ ∂µ (iσ 2 χ∗2 ) + χ†1 σ̄ µ ∂µ ∂µ χ1 ] − m[(−iχ|2 σ 2 )χ1 + χ†1 iσ 2 χ∗2 ]. (5.1.24)

Using the identities σ 2 σ µ σ 2 = σ µ∗ , η | σ̄ µ∗  = −† σ̄ µ η and | σ 2 η = η | σ 2 , we get

ψ(i∂/ − m)ψ = i(χ†1 σ̄ µ ∂µ χ1 − ∂µ χ†2 σ̄ µ χ2 ) + im(χ|1 σ 2 χ2 − χ†1 σ 2 χ∗2 ). (5.1.25)

Note that the second term can be integrated by parts to yield the Dirac action (up to total derivative):
Z
d4 x χ†1 iσ̄ µ ∂µ χ1 + χ†2 iσ̄ µ ∂µ χ2 + im(χ|1 σ 2 χ2 − χ†1 σ 2 χ∗2 ) .
 
SDirac = (5.1.26)

(d) The Lagrangian in Eqn. (5.1.26) is invariant under the transformations

χ1 → e−iα χ1 , χ2 → eiα χ2 . (5.1.27)

Infinitesimally, this reads


1 1
∆χ1 = α δχ1 = −iχ1 , ∆χ2 = α δχ2 = iχ2 . (5.1.28)

The associated current is


∂L ∂L
Jµ = ∆χ1 + ∆χ2 = χ†1 iσ̄ µ (−iχ1 ) + χ†2 iσ̄ µ (iχ2 ) = χ†1 σ̄ µ χ1 − χ†2 σ̄ µ χ2 , (5.1.29)
∂(∂µ χ1 ) ∂(∂µ χ2 )

which is the current called Jcµ in (5.1.4).


The equations of motion that follow from Eqn. (5.1.26) are

σ̄ µ ∂µ χ1 = mσ 2 χ∗2 , ∂µ χ†1 σ̄ µ = mχ|2 σ 2 ,


(5.1.30)
σ̄ µ ∂µ χ2 = mσ 2 χ∗1 , ∂µ χ†2 σ̄ µ = mχ|1 σ 2 .

Therefore, the divergence of the current Jcµ is

∂µ Jcµ = ∂µ χ†1 σ̄ µ χ1 + χ†1 σ̄ µ ∂µ χ1 + ∂µ χ†2 σ̄ µ χ2 + χ†2 σ̄ µ ∂µ χ2


= mχ|2 σ 2 χ1 + χ†1 mσ 2 χ∗2 − mχ|1 σ 2 χ2 − χ†2 mσ 2 χ∗1
∂µ Jcµ = 0 . (5.1.31)

Before calculating the divergence of Jbµ = χ† σ̄ µ χ, first note that it would be the current associated with the
transformation χ → e−iα χ. However, this is only a symmetry of the action (5.1.2) when m = 0. Therefore,
we expect that ∂µ Jbµ ought to be proportional to the mass and only vanishes when m = 0. The exact same
procedure as used above yields
∂µ Jbµ = m(χ| σ 2 χ + χ† σ 2 χ∗ ) . (5.1.32)

The theory of N free massive spinors with O(N ) symmetry is

N Z  
im | 2
d4 x χ†i iσ̄ µ ∂µ χi + χi σ χi − χ†i σ 2 χ∗i
X 
S= . (5.1.33)
i=1
2

Clearly, we can rotate the fields arbitrarily. Define the quantities


   µ   2 
χ1 σ̄ σ
 ..  µ . .. 2 ..
χ =  . , Σ = , Σ = . (5.1.34)
   
.
µ 2
χN σ̄ σ
5.1. MAJORANA FERMIONS, P&S 3.4, P.73 45

Then, the action (5.1.33) may be written


Z  
4 † µ im | 2 † 2 ∗

S = d x χ iΣ ∂µ χ + χ Σ χ−χ Σ χ . (5.1.35)
2

The symmetry is χ → Rχ where R ∈ O(N ) is a real orthogonal matrix, so R| R = 1. This is clear because,
µ
in the N × N space enumerating the spinors but not the two components of each spinor, the matrices Σ
and Σ are both proportional to 1. Therefore, R Σ R = Σ and R Σ R = Σ .
| µ
2 µ | 2 2

(e) Looking back at the action we found in part (c), Eqn. (5.1.26), we see that we get exactly twice the Majorana
action if we make the identification χ = χ1 = χ2 . According to the definitions in (5.1.3), this means that ψL
and ψR are related via

χ = ψL = iσ 2 ψR . (5.1.36)

From the quantized Dirac field in P&S Eqn. 3.99 p.58, we write the quantized ψL and ψR :

d3 p 1 √
Z
p · σ asp ξ s e−ip·x + bsp η s eip·x ,

ψL (x) = (4.3.37a)
(2π)3 2Ep
p

d3 p 1 √
Z
p · σ̄ asp ξ s e−ip·x − bsp η s eip·x .

ψR (x) = 3
p (4.3.37b)
(2π) 2Ep

We have left the sum over the repeated index, s, implicit. One should be mindful of the fact that the
anticommutativity (Grassmann nature) of the fields is due solely to the anticommutativity of the creation
and annihilation
 operators. The spinors, ξ and η, are NOT Grassmann-valued! Indeed, a basis for them is
furnished by 10 and 01 .

√ √
The expressions p · σ and p · σ̄ just come from a boost applied to the rest frame solution of the Dirac
equation. This boost was done along the z-direction in Eqns. 3.48 to 3.50 on P&S p.46 with the result that
the boost matrix was
cosh( η2 ) − σ 3 sinh( η2 )
 
i 03 0
e− 2 (2η)S = η η ,
0 cosh( 2 ) + σ 3 sinh( 2 )

where η is the rapidity defined by the relations in P&S Eqn. 3.48 p.46: E = m cosh η and p3 = m sinh η.
The generalization to an arbitrary boost is straightforward. The rapidity is replaced by a vector, η, whose
magnitude is what was formerly called η. Then,

cosh( η2 ) − σ · η̂ sinh( η2 )
 
i 0i 0
e− 2 (2ηi )S = , (5.1.38)
0 cosh( η2 ) + σ · η̂ sinh( η2 )

The relationships between the energy, mass, momentum and rapidity are

E = m cosh η, p = η̂m sinh η. (5.1.39)

Using half-angle formulae, we write


r s r
E
η cosh η + 1 m +1 E+m
cosh = = = , (4.3.40a)
2 2 2 2m
r s r
E
η cosh η − 1 m −1 E−m
sinh = = = . (4.3.40b)
2 2 2 2m

Then, tanh η2 is given by


r √
η E−m E 2 − m2 |p|
tanh = = = . (4.3.40c)
2 E+m E+m E+m
46 CHAPTER 5. PROBLEM SET 5

A more convenient form of this equation is


η p
η̂ tanh = . (4.3.40d)
2 E+m

Then, the boost, Eqn. (5.1.38) may be written as


r  σ·p 
− 2i (2ηi )S 0i E + m 1 − E+m 0
e = σ·p , (5.1.41)
2m 0 1+ E+m

√ ξ

This multiplies the rest frame solution, which is m ξ . The result is

σ·p
r 
1− E+m ξ
 
E+m
u(p) = σ·p
 . (5.1.42)
2 1+ E+m ξ

Evidently,

q
E+m σ·p
 √ q
E+m σ·p

p·σ = 2 1− E+m , p · σ̄ = 2 1+ E+m . (5.1.43)

q q p
E+m E−m
Using the identity 2 + 2 = E − |p|, it can be shown that Eqn. (5.1.43) gives the same results
as in P&S Eqn. 3.49 p.46 when p = (0, 0, p3 ).
√ √ √
Since there are factors of 1/ 2E in ψL and ψR , we might as well combine that with p · σ and p · σ̄. We
p
denote these matrices by M p and M :
√ q
Ep +m
Mp ≡ √1 1 − Eσ·p

p·σ = 4Ep p +m
, (4.3.44a)
2Ep
p √ q
Ep +m
M ≡ √1 1 + Eσ·p

p · σ̄ = 4Ep p +m
. (4.3.44b)
2Ep

The point of writing out these matrices explicitly is to determine what happens to them under certain
algebraic manipulations. For example, it is clear that
p −p
M −p = M , M = M p. (5.1.45)

Using the identity σ 2 σσ 2 = −σ ∗ , we also have


p∗ p
σ2 M p σ2 = M , σ 2 M σ 2 = M p∗ . (5.1.46)


Now, let us return to Eqn. (5.1.36) and write iσ 2 ψR :

d3 p
Z
∗ p∗ s† s∗ ip·x s∗ −ip·x
iσ 2 ψR iσ 2 M − bs†

= 3
ap ξ e p η e
(2π)
d3 p
Z
2 s∗ −ip·x
M p as† 2 s∗ ip·x
− bs†

= 3 p iσ ξ e p iσ η e . (5.1.47)
(2π)

This must be equal to ψL . Focusing on the coefficient of eip·x , we determine the relationship

bsp η s = as† 2 s∗
p iσ ξ . (5.1.48)

Thus, χ = ψL is written as

d3 p
Z
M p asp ξ s e−ip·x + as† 2 s∗ ip·x

χ(x) = 3 p iσ ξ e . (4.3.49a)
(2π)
5.1. MAJORANA FERMIONS, P&S 3.4, P.73 47

Using the fact that M p† = M p and σ 2† = σ 2 , we find

d3 p
Z

as† ξ s† eip·x − asp ξ sT iσ 2 e−ip·x M p .

χ (x) = (4.3.49b)
(2π)3 p

We will calculate the equal-time anticommutation relations at t = 0. Making a p → −p switch in the second
term of χ and χ† and recalling Eqn. (5.1.45), we find

d3 p
Z
p
arp Mac ξc + ar†
p r 2 r∗ ip·x

χa (x) = 3 −p M ac iσcd ξd e , (4.3.50a)
(2π)
d3 q
Z
q  −iq·y
χ†b (y) = 3
as† s∗ q s s 2
q ξe Meb − a−q ξf iσf e M eb e . (4.3.50b)
(2π)

Technically, we should express the creation and annihilation operators in terms of the fields and then use
the canonical anticommutation relations of the latter to derive the anticommutation relations of the former.
However, (a) we have done this before for the scalar field; and (b) the procedure is much more laborious in this
scenario because of the spinor components. Therefore, we will simply posit the standard anticommutation
relations for the creation and annihilation operators,

{arp , as† 3 rs
q } = (2π) δ δ(p − q) , (5.1.51)

and we will show that this leads to the correct anticommutation relations for the fields.
Now, we can calculate the anticommutator:

d3 p d3 q i(p·x−q·y)  r s†
Z 
† p r s∗ q r† s p 2 r∗ s 2 q
{χa (x), χb (y)} = e {ap , aq }M ac ξc ξ e M eb − {a−p , a−q }M ac iσ cd ξd ξ f iσ f e M eb
(2π)3 (2π)3
Z 3
d p ip·(x−y) p s s∗ p p 2 s∗ s 2 p
= 3
e Mac ξc ξe Meb + M ac σcd ξd ξf σf e M eb
(2π)
d3 p ip·(x−y) 
Z
2 p 2 
= 3
e Mp + M ab
, (5.1.52)
(2π)

where we have used the identity on P&S p.49: ξas ξbs∗ = ξas∗ ξbs = δab .

It is actually more convenient to calculate the squares of the M matrices when written in terms of p·σ

and p · σ̄ instead of the explicit expression:

p · (σ + σ̄) 2p0 1
= 1.
2 p 2
Mp + M = = (5.1.53)
2Ep 2Ep

Note that the spatial pieces vanish because σ i + σ̄ i = 0.


Plugging this into the anticommutator gives the desired result:

d3 p ip·(x−y)
Z
{χa (x), χ†b (y)} = δab e = δab δ(x − y). (5.1.54)
(2π)3

By essentially the same calculation, the other anticommutators are found to vanish.
Now, we would like to calculate the Hamiltonian. We will use the Lagrangian in the form (5.1.13):

L = 2i χ† σ̄ µ ∂µ χ − ∂µ χ† σ̄ µ χ + m χ| σ 2 χ − χ† σ 2 χ∗ .
 
(5.1.55)

The canonical momenta are

π= ∂L
∂ χ̇ = 2i χ† , π† = ∂L
∂ χ̇†
= − 2i χ. (5.1.56)

The Hamiltonian density is


H = 2i (χ† χ̇ − χ̇† χ) − L.
48 CHAPTER 5. PROBLEM SET 5

We have elected to keep the expression in terms of the time derivatives instead of the momenta because we
will eventually expand in creation and annihilation operators anyway for which the formality of writing the
Hamiltonian in terms of the momenta is immaterial.
At this point, note that we can use the equations of motion, Eqns. (5.1.20) and (5.1.22), to simplify L to
L = 0! Then, the Hamiltonian density is simply

H = 2i (χ† χ̇ − χ̇† χ) ≡ H1 + H2 . (5.1.57)

where we have defined H1 = 2i χ† χ̇ and H2 = − 2i χ̇† χ.


The time derivatives may be taken, evaluated at t = 0, and then the same p → −p trick used to get

d3 p
Z
p 2 s∗  ip·x
χ̇(x) = 3
(−iEp ) asp M p ξ s − as†
−p M iσ ξ e , (4.3.58a)
(2π)
d3 p
Z
p  −ip·x
χ̇† (x) = 3
(iEp ) as† s† p s
p ξ M + a−p ξ iσ M
sT 2
e . (4.3.58b)
(2π)

The first and second terms in the Hamiltonian density contribute the following to the Hamiltonian:

d3 p d3 q
Z Z
i p q 2 s∗  iq·x
d3 x H1 = d3 x 3 3
ap ξ M p − ar−p ξ rT iσ 2 M e−ip·x (−iEq ) asq M q ξ s − as†
r† r†
−q M iσ ξ e
2 (2π) (2π)
d3 p
Z
1 p s p 2 s∗ 
= 3
Ep ar† r† p r
p ξ M − a−p ξ iσ M
rT 2
ap M p ξ s − as†
−p M iσ ξ , (4.3.59a)
2 (2π)
d3 p d3 q
Z Z
i p  −ip·x s q 2 s∗  iq·x
d3 x H2 = − d3 x 3
(iEp ) 3
ar† r† p r
p ξ M + a−p ξ iσ M
rT 2
e aq M q ξ s + as†
−q M iσ ξ e
2 (2π) (2π)
d3 p
Z
1 r† r† p r rT 2 p s p s s† p 2 s∗ 
= E p a p ξ M + a−p ξ iσ M ap M ξ + a −p M iσ ξ . (4.3.59b)
2 (2π)3

It is clear that when we add these two, the terms with two a’s or two a† ’s cancel. We are left with

d3 p
Z
s† rT 2 p 2 2 s∗ 
Ep ar† s r† p 2 s r

H= 3 p ap ξ (M ) ξ − a−p a−p ξ σ M σ ξ
(2π)
d3 p
Z
Ep ar† s r† p 2 s r s† rT 2 p 2 2 s∗
 
= 3 p ap ξ (M ) ξ − ap ap ξ σ (M ) σ ξ ,
(2π)

where we made a p → −p switch in the second term.


Using Eqn. (5.1.46), we can write
p∗ 2 p 2  ∗
σ 2 (M p )2 σ 2 = M 1
· σ̄ ∗ .

= M = 2Ep p

Therefore,
d3 p  r† s r†
Z
1 s r s† rT ∗ s∗

H= ap ap ξ (p · σ)ξ − ap ap ξ (p · σ̄ )ξ .
2 (2π)3

Recall the identity from Problem Set 3: η | σ̄ µ∗ ∗ = −† σ̄ µ η. The minus sign in this equation is due to the
fact that the spinors were Grassmann-valued. In our present case, ξ is only complex-valued (actually, we can
choose a real basis). Thus, we do not get the extra minus sign:

d3 p  r† s r†
Z
1
ap ap ξ (p · σ)ξ s − arp as† s† r

H= 3 p ξ (p · σ̄)ξ
2 (2π)
d3 p  r† s r†
Z
1
ap ap ξ (p · σ)ξ s − asp ar† r† s

= 3 p ξ (p · σ̄)ξ
2 (2π)

Note that we simply renamed dummy indices r ↔ s in the last term to get the final form.
5.2. SUPERSYMMETRY, P&S 3.5 P.74 49

We use the anticommutation relations of the creation and annihilation operators to put the second term
in normal-ordered form:
d3 p r† s r†
Z Z
1 s 1
H= a a ξ [p · (σ + σ̄)]ξ − d3 p δ(0)ξ s† (p · σ̄)ξ s .
2 (2π)3 p p 2

As before, only the energy  of p · (σ + σ̄) survives and gives 2Ep 1. Then, we use the explicit basis
component
for ξ, namely ξ 1 = 10 and ξ 2 = 01 to derive ξ r† ξ s = δ rs . Finally, we drop the formally divergent second


piece in the Hamiltonian and write the normal-ordered Hamiltonian as

d3 p X
Z
:H:= Ep as† s
p ap . (5.1.60)
(2π)3 s

We have reintroduced the sum over s, which has been implicit throughout.

Note: Please take a moment to read the small section on the Majorana neutrino on p.102 of Zee.

5.2 Supersymmetry, P&S 3.5 p.74


It is possible to write field theories with continuous symmetries linking fermions and bosons; such transformations
are called supersymmetries.

(a) The simplest example of a supersymmetric field theory is the theory of a free complex boson and a free Weyl
fermion, written in the form
L = ∂µ φ∗ ∂ µ φ + χ† iσ̄ · ∂χ + F ∗ F. (5.2.1)

Here F is an auxiliary complex scalar field whose field equation is F = 0. Show that this Lagrangian is
invariant (up to a total divergence) under the infinitesimal transformation

δφ = −i| σ 2 χ,
δχ = F + σ · ∂φσ 2 ∗ , (5.2.2)

δF = −i σ̄ · ∂χ,

where the parameter a is a 2-component spinor of Grassmann numbers.

(b) Show that the term


∆L = [mφF + 21 imχ| σ 2 χ] + (c.c.) (5.2.3)

is also left invariant by the transformation, (5.2.2). Eliminate F from the complete Lagrangian L + ∆L by
solving its field equation, and show that the fermion and boson fields φ and χ are given the same mass.

(c) It is possible to write supersymmetric nonlinear field equations by adding cubic and higher-order terms to
the Lagrangian. Show that the following rather general field theory, containing the field (φi , χi ) i = 1, . . . , n,
is supersymmetric:

i ∂ 2 W [φ] | 2
 
∂W [φ]
L = ∂µ φ∗i ∂ µ φi + χ†i iσ̄ · ∂χi + Fi∗ Fi + Fi + χi σ χj + c.c. , (5.2.4)
∂φi 2 ∂φi ∂φj

where W [φ] is an arbitrary function of the φi , called the superpotential. For the simple case n = 1 and
W = gφ3 /3, write out the field equations for φ and χ (after elimination of F ).

SOLUTION:

(a) Let us start off with a bunch of identities that will be useful later on. The first is

σ 2 σ µ σ 2 = σ̄ µ∗ . (5.2.5)
50 CHAPTER 5. PROBLEM SET 5

This is easily proven using the fact that σ 0 = 1 and σ i σ j = iijk σ k . Thus,

σ 2 σ 0 σ 2 = (σ 2 )2 = 1 = σ 0 ,
σ 2 σ 1 σ 2 = σ 2 iσ 3 = i · iσ 1 = −σ 1 ,
σ 2 σ 2 σ 2 = 1σ 2 = σ 2 ,
σ 2 σ 3 σ 2 = iσ 1 σ 2 = i · iσ 3 = −σ 3 .

The right hand sides of the above four equations may be summarized as σ̄ µ∗ .
Next, for two Weyl spinors, η and ,
η | σ̄ µ∗ ∗ = −† σ̄ µ η. (5.2.6)

This has the important corollary


| σ 2 η = η | σ 2 . (5.2.7)

We prove Eqn. (5.2.6) by brute force:


∗
η | σ̄ 0∗ ∗ = (η1 , η2 ) = η1 ∗1 + η2 ∗2 = −∗1 η1 − ∗2 η2 = −(∗1 , ∗2 ) 10 01 ηη12 = −† σ 0 η,
10
 1
  
01 ∗
2
 ∗ 
η | σ̄ 1∗ ∗ = −(η1 , η2 ) 01 10 ∗1 = −η1 ∗2 − η2 ∗1 = ∗1 η2 + ∗2 η1 = (∗1 , ∗2 ) 01 10 ηη12 = † σ 2 η,
 
2
 ∗ 
η | σ̄ 2∗ ∗ = −(η1 , η2 ) −i0 0i 1∗ = −i(η1 ∗2 − η2 ∗1 ) = −i(∗1 η2 − ∗2 η1 ) = (∗1 , ∗2 ) 0i −i0 ηη12 = † σ 2 η,
 
2
 ∗ 
η | σ̄ 3∗ ∗ = −(η1 , η2 ) 10 −10 1∗ = −η1 ∗1 + η2 ∗2 = ∗1 η1 − ∗2 η2 = (∗1 , ∗2 ) 10 −10 ηη12 = † σ 3 η.
 
2

Indeed, the right hand sides of the above four equations may be summarized as −† σ̄ µ η.
Finally, we will use the identity
σ̄ µ σ ν ∂µ ∂ν = ∂µ ∂ µ . (5.2.8)

This can be proved as follows:

σ̄ µ σ ν ∂µ ∂ν = (∂0 − σ i ∂i )(∂0 + σ j ∂j ) = ∂02 − σ i σ j ∂i ∂j = ∂02 − 21 {σ i , σ 2 }∂i ∂j = ∂02 − ∂i ∂ i = ∂µ ∂ µ .

Note that we symmetrized the product σ i σ j because it is multiplying ∂i ∂j , which is symmetric.


These identities allow us to write δφ and δF in two ways. In components, δφ, looks like

δφ = χ2 1 − χ1 2 .

We can complex conjugate this, remembering that the order of multiplication switches under conjugation:
χ∗
δφ∗ = ∗1 χ∗2 − ∗2 χ∗1 = (∗1 , ∗2 ) 01
= i† σ 2 χ∗ = iχ† σ 2 ∗ ,
 1

−1 0 χ∗
2

where the final equality is just identity (5.2.7) with  → ∗ and η → χ∗ .


Since there is only one spinor on the right side of δχ, it is actually straightforward to take its dagger. We
simply observe that σ µ† = σ µ . Then,

δχ† = F ∗ † + † σ 2 σ µ ∂µ φ∗ .

The hardest one to write out is δF :


 
∂0 −∂3 −(∂1 −i∂2 ) χ1
δF = −i(∗1 , ∗2 ) −(∂

1 +i∂2 ) ∂0 +∂3 χ2

= −i ∗1 [(∂0 − ∂3 )χ1 − (∂1 − i∂2 )χ2 ] + ∗2 [−(∂1 + i∂2 )χ1 + (∂0 + ∂3 )χ2 ]


= i [(∂0 − ∂3 )χ1 − (∂1 − i∂2 )χ2 ]∗1 + [−(∂1 + i∂2 )χ1 + (∂0 + ∂3 )χ2 ]∗2 .

5.2. SUPERSYMMETRY, P&S 3.5 P.74 51

Now, we can complex conjugate this:

δF ∗ + −i 1 [(∂0 − ∂3 )χ∗1 − (∂1 + i∂2 )χ∗2 ] + 2 [−(∂1 − i∂2 )χ∗1 + (∂0 + ∂3 )χ∗2 ]

  ∗
∂ −∂ −(∂ +i∂ ) χ1
= −i(1 , 2 ) −(∂01 −i∂3 2 ) ∂01+∂3 2 χ∗ 2

= −i| σ̄ µ∗ ∂µ χ∗
= i∂µ χ† σ̄ µ ,

where use was made of identity (5.2.6).


Let us collect these transformations below

δφ = −i| σ 2 χ = −iχ| σ 2 , (3.3.9a)


δφ∗ = i† σ 2 χ∗ = iχ† σ 2 ∗ , (3.3.9b)
µ 2 ∗
δχ = F + σ σ  ∂µ φ, (3.3.9c)
† ∗ † | 2 µ ∗
δχ = F  +  σ σ ∂µ φ , (3.9d)
† µ | µ∗ ∗
δF = −i σ̄ ∂µ χ = i∂µ χ σ̄  , (3.3.9e)
∗ | µ∗ ∗ † µ
δF = −i σ̄ ∂µ χ = i∂µ χ σ̄ . (3.3.9f)

Now, we plug these into δL:

δL = ∂µ (i† σ 2 χ∗ )∂ µ φ + ∂µ (−i| σ 2 χ)∂ µ φ∗ + ( F ∗ † + | σ 2 σ ν ∂ν φ∗ )iσ̄ µ ∂µ χ + χ† iσ̄ µ ∂µ (F + σ ν σ 2 ∗ ∂ν φ)




+ (−i| σ̄ µ∗ ∂µ χ∗ )F + ( F ∗( † (
σ̄ µ ∂µ χ)
( (
(−i
((
= i† σ 2 ∂µ χ∗ ∂ µ φ − i| σ 2 ∂µ χ∂ µ φ∗ + i| σ 2 σ ν σ̄ µ ∂µ χ∂ν φ∗ − i| σ̄ µ∗ χ∗ ∂µ F + i† σ 2 χ∗ ∂µ ∂ µ φ − i| σ̄ µ∗ ∂µ χ∗ F

= ∂µ (i† σ 2 χ∗ ∂ µ φ) − ∂µ (i| σ̄ µ∗ χ∗ F ) − ∂µ (i| σ 2 χ∂ µ φ∗ ) + ( σ 2(


i|( ∂ µ φ∗
((
χ∂
(µ(
((
+ ∂µ (i| σ 2 σ ν σ̄ µ χ∂ν φ∗ ) − ( σ 2(
i|( σ(ν (
σ̄ µ χ∂
((
µ ∂ν φ

= i∂µ  σ χ ∂ φ + | σ 2 [(σ ν σ̄ µ − η µν )χ∂ν φ∗ − σ µ σ 2 χ∗ F ]


 † 2 ∗ µ 

Note that the aforementioned identities were all used. Therefore,

δL = ∂µ J µ J µ = i † σ 2 χ∗ ∂ µ φ + | σ 2 [(σ ν σ̄ µ − η µν )χ∂ν φ∗ − σ µ σ 2 χ∗ F ] .
 
where (5.2.10)

It follows that L is supersymmetric.

(b) The fact that ∆L is supersymmetric follows from part (c) using W (φ) = 21 mφ2 . The F -dependent terms in
L + ∆L are F ∗ F + mφF + mφ∗ F ∗ . The F and F ∗ equations of motion amount to the identities

F ∗ = −mφ , F = −mφ∗ . (3.3.11a)

Plugging these back into the Lagrangian L + ∆L gives

L + ∆L = ∂µ φ∗ ∂ µ φ − mφ∗ φ + χ† iσ̄ µ ∂µ χ + im | 2
2 (χ σ χ − χ† σ 2 χ∗ ) (3.3.11b)

Note that −iχ† σ 2 χ∗ is the complex conjugate of iχ| σ 2 χ. Indeed, this is the Lagrangian of a massive spin-0
boson and massive Majorana fermion both with the same mass, m.

(c) Using σ 2T = −σ 2 and σ µT = σ µ∗ , the transformation of χ| is

δχ| = F | − † σ 2 σ µ∗ ∂µ φ. (5.2.12)
52 CHAPTER 5. PROBLEM SET 5

Let L be the free action and ∆L the interaction containing the superpotential. By trivial extension of the
argument in part (a), L is supersymmetric. The transformation of ∆L is
∂W ∂2W i ∂3W
δ∆L = (−i† σ̄ µ ∂µ χi ) + Fi (−i| σ 2 χj ) + (−i| σ 2 χk )(χ|i σ 2 χj )
∂φi ∂φi ∂φj 2 ∂φi ∂φj ∂φk

i ∂2W i ∂2W | 2
+ (Fi | − † σ 2 σ µ∗ ∂µ φi )σ 2 χj + χ σ (Fj + σ µ σ 2 ∗ ∂µ φj ) + c.c.
2 ∂φi ∂φj 2 ∂φi ∂φj i

The dashed terms cancel after we rewrite the third one using χ|i σ 2  = | σ 2 χi . We can use the identity
σ 2 σ µ∗ σ 2 = σ̄ µ , which is simply the complex conjugate of Eqn. (5.2.5), to rewrite the second underlined term:
i ∂2W i ∂W
2nd underlined term = − † σ̄ µ χi ∂µ φj = − † σ̄ µ χi ∂µ .
2 ∂φi ∂φj 2 ∂φi

Note that we also switched i and j indices.


The third underlined term can be simplified by first using σ 2 σ µ σ 2 = σ̄ µ∗ , which is Eqn. (5.2.5), and then
using χ| σ̄ µ∗ ∗ = −† σ̄ µ χ, which is Eqn. (5.2.6). The result is that the third underlined term is actually
identical to the second one. This combines with the first underlined term to give a total derivative:
∂3W
 
∂W i
δ∆L = −i∂µ † σ̄ µ χi + (−i| σ 2 χk )(χ|i σ 2 χj ) + c.c.
∂φi 2 ∂φi ∂φj ∂φk

Clearly, the triple derivative term is not a total derivative. Therefore, it must somehow vanish due to
symmetry in the indices. The triple derivative is symmetric in the ijk indices. Let us write this term as
i ∂3W i ∂3W
| σ 2 χk χ|i (−iσ 2 )χj ≡ | σ 2 (kij),
2 ∂φi ∂φj ∂φk 2 ∂φi ∂φj ∂φk

where we have introduced the notation (kij) for the appropriate terms on the left. By symmetry in the ijk
indices,
i ∂3W i ∂3W
| σ 2 (kij) = | σ 2 [(kij) + (ijk) + (jki)].
2 ∂φi ∂φj ∂φk 6 ∂φi ∂φj ∂φk
It must be the case that the symmetric combination of the (ijk)’s vanishes. The proof of this is just by brute
force calculation:
χi1 χj2 χk1 −χi1 χj1 χk1 
(ijk) = χχi1
 0 −1
 χk1 
i2 (χj1 , χj2 ) 1 0 χk2 = χi2 χj2 χk1 −χi2 χj1 χk2 . (3.3.13a)

Similarly, we can calculate the other two orders. In so doing, we will always reorder the χ factors in the end
in the order ijk and thus we must remember that a switch of two of the χ’s generates a minus sign. The
results are
χ χj1 χk2 −χi2 χj1 χk1 
(jki) = χi1i1 χj2 χk2 −χi2 χj2 χk1
, (3.3.13b)
χi2 χj1 χk1 −χi1 χj2 χk1 
(kij) = χi2 χj1 χk2 −χi1 χj2 χk2 . (3.3.13c)

Now, we can just observe that all the terms pairwise cancel in the sum:

(ijk) + (jki) + (kij) = 00 . (5.2.14)

Therefore, we finally have


 
† µ ∂W
δ∆L = −i∂µ  σ̄ χi + c.c. (5.2.15)
∂φi

This proves that the total Lagrangian is supersymmetric.


When W = gφ3 /3, the F -dependent terms in the Lagrangian are F ∗ F + gF φ2 + gF ∗ φ∗2 , where we have
assumed g to be real out of convenience, not necessity. Thus, the F and F ∗ equations of motion amount to

F ∗ = −gφ2 F = −gφ∗2 . (5.2.16)


5.3. DISCUSSION 1: MAJORANA BASIS 53

Plugging this back into L yields

L = ∂µ φ∗ ∂ µ φ − g 2 φ∗2 φ2 + χ† iσ̄ µ ∂µ χ + ig(φχ| σ 2 χ − φ∗ χ† σ 2 χ∗ ). (5.2.17)

The φ∗ and φ equations of motion read

φ = −2g 2 φ∗ φ2 − igχ† σ 2 χ∗ ,
(5.2.18)
φ∗ = −2g 2 φφ∗2 + igχ| σ 2 χ.

To calculate the equation of the fermionic field, we may treat χ1 , χ∗1 , χ2 and χ∗2 as the independent fields.
Taking the variation of the third term in Eqn. (5.2.17) with respect to χ† gives the two component term iσ̄·∂χ
whose first component is a term in ∂L/∂χ∗1 and whose second component is a term in ∂L/∂χ∗2 . The fourth
term does not contain factors of χ∗1 or χ∗2 . We may write the last term as −gφ∗ (χ∗1 χ∗2 −χ∗2 χ∗1 ). When taking the
variation with respect to χ∗1 , we write this term as −2gφ∗ χ∗1 χ∗2 and the variation is ∂L/∂χ∗1 3 −2gφ∗ χ∗2 . When
taking the variation with respect to χ∗2 , we write this term as 2gφ∗ χ∗2 χ∗1 and the variation is ∂L/∂χ∗2 3 2gφ∗ χ∗1 .
All in all, we have
 ∗
∂L ∗ −χ2
0= µ
= iσ̄ ∂µ χ + 2gφ = iσ̄ · ∂χ − 2igφ∗ σ 2 χ∗ . (5.2.19)
∂χ† χ∗1

Similarly, one finds the variation with respect to χ:

∂L
= −i∂µ χ† σ̄ µ + 2gφ∗ −χ2 χ1 = −i∂χ† · σ̄ + 2igφχ| σ 2 .

0= (5.2.20)
∂χ

5.3 Discussion 1: Majorana Basis


Note: Please take a moment to read the small section on the Majorana neutrino on p.102 of Zee.

For me, the take-away message of P&S Problem 3.4 parts (a)-(c) is that if you start off with the Dirac spinor
with the Dirac Lagrangian and impose the strange-looking condition

ψR = iσ2 ψL , (5.3.1)

then you end up with the Majorana Lagrangian up to√a factor of 2, which you can always just absorb into a
redefinition of the fields (e.g. by absorbing a factor of 2 into the spinor).
We have to remember that we have written all our spinors, γ matrices and the representation Λ 12 of the
Lorentz group all in the so-called Weyl basis. Of course, we could conjugate all of the γ matrices by some
unitary matrix, U , and we would have an equivalent, but different-looking basis and the condition (5.3.1) may
look very different. You have heard in lecture that a Majorana spinor in 3 + 1 dimensions is just the same as the
four-component Dirac spinor, but where each component is now real rather than complex. It is not clear how
condition (5.3.1) achieves this. However, we will find a basis in which this reality condition is made manifest.
We will call this the Majorana basis (more precisely, one possible Majorana basis).
Consider the matrix  
1 σ2 + σ3 i − σ1
U= . (5.3.2)
2 −i − σ1 σ2 − σ3
This is a 4 × 4 matrix written in terms of four 2 × 2 blocks. Note also that i in a 2 × 2 block means i multiplied
by the 2 × 2 identity matrix. Right now I’m just pulling this matrix out of thin air, but after you see what I do
with it, you may want to entertain yourself by trying to work out how I came up with it. You may even try to
think of other examples that will do just as well.
You may check that U is unitary: U † U = 1. Therefore, I can consider a set of γ matrices that are related to
the ones we have used so far (Weyl basis) by conjugation with U . Since we will call this the Majorana basis, we
will put a subscript M on these new matrices:
µ
γM = U γµU †. (5.3.3)
54 CHAPTER 5. PROBLEM SET 5

Explicitly, these matrices are


1 
0 1 2 0 σ2 3 −1 0
= iσ01 −iσ 01
  
γM 0
, γM =i 10 , γM = −σ 2 0
, γM =i 01 . (5.3.4)

If ψ is a Dirac spinor in the Weyl basis (i.e. it satisfies the Dirac equation written in the Weyl basis), then you
can check that
ψM = U ψ (5.3.5)
satisfies the Dirac equation written in the Majorana basis.
Therefore, if we write  
χ
ψ= , (5.3.6)
η
where χ is a left-chiral Weyl spinor and η is a right-chiral Weyl spinor, then the corresponding spinor written in
the Majorana basis is
  
1 σ2 + σ3 i − σ1 χ
ψM = U ψ =
2 −i − σ 1 σ 2 − σ 3 η
 
1 (σ2 + σ3 )χ + (i − σ1 )η
= . (5.3.7)
2 −(i + σ1 )χ + (σ2 − σ3 )η
According to Eqn. (5.3.1), if we impose the condition

η = iσ2 χ∗ , (5.3.8)

then we end up with a Majorana spinor (i.e. one that satisfies the Majorana equation of part (a) of Problem
3.4). What does this condition look like in the Majorana basis? Well, we get
1 (σ2 + σ3 )χ + (i − σ1 )(iσ2 χ∗ )
 
ψM =
2 −(i + σ1 )χ + (σ2 − σ3 )(iσ2 χ∗ )
1 (σ2 + σ3 )χ + (−σ2 + σ3 )χ∗
 
=
2 −(i + σ1 )χ + (i − σ1 )χ∗
 
1 (σ2 + σ3 )χ + c.c.
= , (5.3.9)
2 −(i + σ1 )χ + c.c.
where c.c. means complex conjugate and where we have used the multiplication rule: σi σj = iijk σk + δij .
Therefore, in the Majorana basis, a Majorana spinor is simply a real four-component spinor.
You might ask why it is possible to impose a reality condition on the Dirac spinor in the first place. Why were
we not able to do that directly in the Weyl basis, for example? The answer can be found in the representation
of the Lorentz group given by
S µν = 4i [γ µ , γ ν ]. (5.3.10)
Because three of the γ’s are real and one is imaginary, the representation
i µν
Λ 12 = e− 2 ωµν S (5.3.11)

is complex. Therefore, a reality condition on a Dirac spinor cannot be consistently and uniformly imposed: if you
were to impose a reality condition in one reference frame and then transform the spinor to a new reference frame,
you would end up with a complex spinor! The nice thing about the Weyl basis is that S’s are block diagonal and
so Λ 12 can be split into the two chiral pieces:
 
Λ 12 L 0
Λ 21 = Λ 12 L ⊕ Λ 12 R = . (5.3.12)
0 Λ 21 R

Looking back at (5.3.4), you can see that all of the γM ’s are imaginary. Therefore, in this basis, all of the S’s
are imaginary and Λ 12 is real. Therefore, if we impose a reality condition on the Dirac spinor in this basis in one
reference frame, this reality condition will hold in any other reference frame related by Lorentz transformation.
However, we lose the block diagonal structure of the representation. Indeed, explicit calculation shows
   
01 i σ1 0 03 i 0 σ1
SM = , SM = . (5.3.13)
2 0 −σ1 2 σ1 0
5.4. DISCUSSION 2: SUSY 55

01 03 0i ij
Thus, while SM is block diagonal, SM is not. Furthermore, you can check that SM is anti-Hermitian while SM
is Hermitian, for i, j = 1, 2, 3. Therefore, the representation is still not unitary.
What we are seeing here is actually a manifestation of an important point that Petr made during his lecture
on spinors in arbitrary dimension. In 3+1 dimensions, you can have a Weyl spinor (2 complex components) or a
Majorana spinor (4 real components), but you cannot have a Majorana-Weyl spinor (2 real components). That
would imply the existence of a basis in which the S’s are both block diagonal and purely imaginary so that Λ 12 is
simultaneously block diagonal and real. Of course, we have not proven that no such basis exists here. But, what
we have done confirms this fact: we now have two bases, the Weyl and the Majorana bases. In the Weyl basis
Λ 12 is block diagonal, but not purely real, and in the Majorana basis it is purely real but not block diagonal.

5.4 Discussion 2: SUSY


The following is a discussion on the topic of the example of supersymmetry in this problem. Recall that the
/ Let us write Ψ as
kinetic term for a Dirac spinor is iΨ∂Ψ.
 
χ
Ψ= , (5.4.1)
ξ
where χ is a left-chirality (chirality −1) spinor and ξ is a right-chirality (chirality +1) spinor, as suggested by
P&S Eqn. 3.36 p.43. Then, the kinetic term may be written
 χ
0 σ µ ·∂
/ = i(χ† , ξ † ) 01 10 σ̄·∂ † †

iΨ∂Ψ 0 ξ = i(χ σ̄ · ∂χ + ξ σ · ∂ξ). (5.4.2)

The kinetic term for χ in the Lagrangian of this problem, (5.2.1), is precisely the same as the kinetic term for χ
contained in the Dirac kinetic term above. Apparently, the χ in this problem has been chosen to be a chirality-left
spinor. Presumably, the construction could equally well have been done using a right-chirality spinor, ξ. We are
allowed to use only one of the Weyl spinors and not both in the form of a Dirac spinor because we have not
included a Dirac mass term −mΨΨ, which is the only thing that mixes the two.
But, eventually, χ will have a mass, as we found in part (b). The point here is that the only way to get
a Lorentz-invariant product of two spinors is to have one be chirality-left and the other chirality-right. For
example, the Dirac mass term is −mΨΨ = −m(χ† ξ + ξ † χ) and χ and ξ have opposite chirality. Part (c) of
Problem 3.4 shows that a chirality-right spinor can be built out of a chirality-left spinor by transposing and then
right-multiplying by σ 2 : namely, if χ is left-chirality, then χ| σ 2 is right-chirality. Therefore, χ| σ 2 χ is Lorentz-
invariant. Since σ 2 is imaginary, we make it real by multiplying by i and then we must add the Hermitian
conjugate, which is −iχ† σ 2 χ∗ , to make the whole mass term real. The factor of 1/2 just gets the normalizations
† 2 ∗
right. This gives the mass term im | 2
2 (χ σ χ − χ σ χ ).
By the way, another way to motivate the combination χ| σ 2 χ is that it equals −χ1 χ2 + χ2 χ1 , which is like the
anti-symmetric combination of a spin-up and spin-down state if we make the analogy χ1 ∼↑ and χ2 ∼↓. This is
precisely the spin-0 combination, which is rotationally-invariant.
One complex two-component spinor contains four real fields off-shell and two real fields on-shell (the Dirac
equation kills off two of the modes). Thus, a theory containing this spinor has a chance of being supersymmetric
on-shell if we pair it with one complex scalar field since this gives a total of two bosonic and two fermionic degrees
of freedom. However, we usually want symmetries to hold even off-shell because intermediate virtual particles
will be off-shell. The simplest way to make the counting work is to introduce an auxiliary complex scalar field,
which in the problem was called F . This gives a total of four bosonic and four fermionic degrees of freedom.
The Lagrangian, (5.2.1), must have dimension +4 so that the action is dimensionless. This implies the
following dimensions for the fields

[φ] = 1, [χ] = 32 , [F ] = 2. (5.4.3)

Supersymmetry should turn bosonic fields into fermionic ones and vice versa. The simplest scenario would be
if the transformation is linear in the fields. This implies δφ ∼ χ. But the right hand side must be a scalar and
so we must take the inner product with some infinitesimal spinor, , but in a way that is Lorentz invariant. We
have already discussed how to produce Lorentz invariant combinations of left-chirality spinors:

δφ = −i| σ 2 χ. (5.4.4)

This is certainly not unique: χ could have appeared complex-conjugated, the factor of −i is certainly not unique,
etc. But, this choice presumably more or less fixes the form of the other transformation by requiring that the
56 CHAPTER 5. PROBLEM SET 5

action be invariant. A different choice would alter the other transformations as well, but should nevertheless be
tenable. No matter the choice, dimensional analysis requires

[] = − 12 . (5.4.5)

We must also have δF ∼ χ. However, [δF ] = 2 whereas [χ] = 1. We don’t want to add any more fields and so
the only thing we can add with dimension 1 is a derivative. We are then forced to contract ∂µ with something.
We already know how to construct a Lorentz-invariant “kinetic” term out of two left-chirality spinors. Recall
that we need to contract ∂µ with σ̄ µ . Thus,

δF = −i† σ̄ · ∂χ. (5.4.6)

δχ is much more difficult, but one can at least partially motivate it by observing that it is linear in the scalar
fields and , and has the correct dimension. Ultimately, the point is that it is what it is in order to keep the
action invariant.
Now, let us calculate the commutator of two supersymmetry transformations:

δη δ φ = −i| σ 2 (ηF + σ µ σ 2 η ∗ ∂µ φ) = −i| σ 2 ηF − i| σ̄ µ∗ η ∗ ∂µ φ = −i| σ 2 ηF + iη † σ̄ µ ∂µ φ. (5.4.7)

Therefore,
[δη , δ ]φ = i(η | σ 2  − | σ 2 η)F + i(η † σ̄ µ  − † σ̄ µ η)∂µ φ. (5.4.8)
The first term on the right vanishes by virtue of identity (5.2.7). By writing out the components, one finds that
−i† σ̄ µ η is in fact the complex conjugate of iη † σ̄ µ . Therefore,

[δη , δ ]φ = aµ ∂µ φ where aµ = iη † σ̄ µ  + c.c. (5.4.9)

This is the result of a translation xµ → xµ − aµ .


Let us repeat this for F :

δη δ F = −i† σ̄ µ ∂µ (ηF + σ ν σ 2 η ∗ ∂ν φ) = −i† σ̄ µ η∂µ F − i† σ 2 η ∗ ∂µ ∂ µ φ. (5.4.10)

Therefore,
[δη , δ ]F = i(η † σ̄ µ  − † σ̄ µ η)∂µ F + i(η † σ 2 ∗ − † σ 2 η ∗ )∂µ ∂ µ φ. (5.4.11)
Once again, the second term vanishes and we get the same translation:

[δη , δ ]F = aµ ∂µ F where aµ = iη † σ̄ µ  + c.c. (5.4.12)

Finally, let us repeat this for χ:

δη δ χ = (−iη † σ̄ µ ∂µ χ) + σ µ σ 2 ∗ ∂µ (−iη | σ 2 χ) = −iη † σ̄ µ ∂µ χ + i∂µ [σ µ (−iσ 2 )∗ η | (−iσ 2 )χ]. (5.4.13)

For the moment, define

χi ≡ σ µ (−iσ 2 )∗ , χj ≡ η, χk ≡ χ. (5.4.14)

Then, using the notation of Eqn. (3.3.13a), we may write

δη δ χ = −iη † σ̄ µ ∂µ χ + i∂µ (ijk). (5.4.15)

Using the identity (5.2.14) and σ µT = σ µ∗ , the second term above may be written

i∂µ (ijk) = −i∂µ [(jki) + (kij)]


= −i∂µ [ηχ| (−iσ 2 )σ µ (−iσ 2 )∗ + χ† (iσ 2 )σ µ∗ (−iσ 2 )η]
= i∂µ [ηχ| σ̄ µ∗ ∗ − χ† σ̄ µ η]
= −iη† σ̄ µ ∂µ χ − i∂µ χ † σ̄ µ η. (5.4.16)

In the last term, we can pass ∂µ χ through the scalar † σ̄ µ η without any change in sign because it has to pass
through two Grassmanns. Hence,

δη δ χ = −i(η † + η† )σ̄ µ ∂µ χ − i† σ̄ µ η∂µ χ. (5.4.17)


5.4. DISCUSSION 2: SUSY 57

The opposite order gives


δ δη χ = −i(η† + η † )σ̄ µ ∂µ χ − iη † σ̄ µ ∂µ χ. (5.4.18)
Notice that the first term cancels in the commutator and we get the same translation result:

[δη , δ ]χ = aµ ∂µ χ where aµ = iη † σ̄ µ  + c.c. (5.4.19)

We see that if we did not have the auxiliary field, F , then we would have gotten

[δη , δ ]χ = aµ ∂µ χ + i(η † − η† )σ̄ µ ∂µ χ. (5.4.20)

The algebra would only close on-shell, when χ satisfies the massless Dirac equation σ̄ µ ∂µ χ = 0. By counting
degrees of freedom, we should have known this would have to be true from the beginning.
58 CHAPTER 5. PROBLEM SET 5
Chapter 6

Problem Set 6

6.1 Amplitude of Figure I.7.11 p.58, Zee I.7.1, p.60


Work out the amplitude corresponding to Figure I.7.11 p.58 in Eqn. 24 p.58.

SOLUTION:

Wick Method: The amplitude for four external particles and four interaction points is
 4 Z Z
1 1 iλ i
R 4 4 2
M= − d w1 · · · d w4 Dφ e 2 d x[(∂φ) −m φ] φ(x1 ) · · · φ(x4 )φ(w1 )4 · · · φ(w4 )4 .
4 4
(6.1.1)
Z(0, 0) 4! 4!
We must count the various ways we can contract the φ’s in order to produce the desired diagram. There are
16 contraction choices for φ(x1 ). Then, φ(x2 ) must be contracted to the same point as φ(x1 ), so there are only
3 choices for this. φ(x3 ) must be contracted to a different point, of which there are 12 choices. φ(x4 ) must be
contracted to the same point as φ(x3 ), of which there are now only 3 choices. The two points that have yet to
be contracted to anything (called intermediate points) constitute 8 φ’s, one of which gets contracted to the same
point as φ(x1 ) and φ(x2 ). This leaves 4 φ’s at the other intermediate point, one of which also gets contracted to
the same point as φ(x1 ) and φ(x2 ). Then, there are 6 remaining intermediate φ’s, one of which gets contracted to
the same point as φ(x3 ) and φ(x4 ), and then 3 φ’s from the other intermediate point, one of also gets contracted
to the same point as the former. Finally, there are two ways to cross-contract the remaining four intermediate
φ’s. This gives a total of 16 · 3 · 12 · 3 · 8 · 4 · 6 · 3 · 2 = 6 · 4!4 equivalent diagrams. When multiplied by the prefactor
4!−5 , this gives an overall factor of 41 (or a symmetry factor of 4). Therefore, Eqn. (6.1.1) simplifies to

λ4
Z Z
2
M= ··· D1w1 D2w1 D3w4 D4w4 Dw1 w2 Dw1 w3 Dw 2 w3
Dw2 w4 Dw3 w4 . (6.1.2)
4 w1 w4

Label the diagram as follows


k'$
5 R k7
k@
1
R  k3
@ k6 ??
k9 
k2 R k4
@
 kR &%
8 k10 @

We write out Eqn. (6.1.2):
10 4
λ4 d4 ki
Z Y
i Y
M= 4 k 2 − m2 + i
d4 wj eik1 (x1 −w1 ) eik2 (x2 −w1 ) eik3 (w4 −x3 ) eik4 (w4 −x4 )
4 i=1
(2π) i j=1

× eik5 (w1 −w2 ) eik8 (w1 −w3 ) eik6 (w2 −w3 ) eik9 (w2 −w3 ) eik7 (w2 −w4 ) eik10 (w3 −w4 )
7
λ4 d4 ki
Z Y
i
= 4 2 2 + i
ei(k1 x1 +k2 x2 −k3 x3 −k4 x4 ) (2π)4 δ(k1 + k2 − k3 − k4 )
4 i=1
(2π) k i − m
i i i
× (6.1.3)
(k1 + k2 − k5 )2 − m2 + i (k5 − k6 − k7 )2 − m2 + i (k1 + k2 − k7 )2 − m2 + i

59
60 CHAPTER 6. PROBLEM SET 6

After amputating the external legs and the overall delta function, what remains is

λ4 d4 k5 d4 k6 d4 k7
Z
i i i
M= 2 2 2
4 (2π) (2π) (2π) k5 − m + i k6 − m + i k7 − m2 + i
4 4 4 2 2

i i i
× . (6.1.4)
(k1 + k2 − k5 )2 − m2 + i (k5 − k6 − k7 )2 − m2 + i (k1 + k2 − k7 )2 − m2 + i

Up to the symmetry factor, this is the same as Eqn. 24 of Zee p.58 with the identifications p = k5 , q = k6 and
r = k7 .

Schwinger Method: The amplitude may be written as


 4 Z Y4  4   ZZ 10
1 iλ δ 1 i
M= − d4 wi − d4 x d4 y J(x)D(x − y)J(y) . (6.1.5)
4! 4! i=1
iδJ(wi ) 10! 2

Now, we must count the number of ways to act the derivatives on the J’s. In the language of Zee, there are
10 strings with J’s on each end and D’s connecting the two endpoints. There are 20 choices for string ends to
place at x1 , then 18 for x2 , then 16 for x3 , then 14 for x4 . The other ends of the strings placed at x1 and x2
are attached to each other at, say, a, as are the free ends of the strings placed at x3 and x4 at, say, b. Next,
there are 12, then 10 choices of two string ends to place at a and 8, then 6 choices of two strings to place at b.
At this point, we have two crosses. There are 4 ways to attach the remaining two free strings at the free ends
of one of the crosses and then 2 ways to attached the remaining two components together. This alone accounts
for a factor
RR of 20!! = 210 10!, which nicely
 cancels the same factor in the denominator from the expansion term in
i
exp − 2 d x d4 y J(x)D(x − y)J(y) .
4

We have not accounted for the fact that the vertices could have any order of labeling from w1 to w4 . This
gives a factor of 4!. We must also account for the fact that the four derivatives at each vertex may be acted on
the string ends in any order, which gives a factor of 4! per vertex, or 4!4 . Overall, we have 4!5 , which neatly
cancels the same factor in the denominator out front in Eqn. (6.1.5).
However, we must be careful because we have overcounted the number of diagrams. The following switches
of the internal lines leave the diagram unchanged: (1) k5 ↔ k8 and k7 ↔ k10 ; or (2) k6 ↔ k9 . This gives an
internal symmetry structure of dimension 4 (i.e. a symmetry factor of 4). Note that this is the same factor as
was found using the Wick method.
At this point, the result is actually exactly the same as Eqn. (6.1.2) since the derivatives simply collapse to
the same product of propagators. The only possible difference is a sign, which is immaterial since M is squared
in any physically sensible quantity.

Symmetry factors formula: For φ4 theory, there is a formula for the symmetry factor:

Y
S=g 2β (n!)αn , (6.1.6)
n=2

where g is the number of permutations of vertices that leave unchanged the diagram with fixed external lines,
αn is the number of vertex pairs connected by n identical lines, and β is the number of lines connecting a vertex
with itself. For the diagram in this problem, g = 2 since we have the identity permutation (in other words, no
permutation at all), which is always obviously allowed, and the permutation that switches the two intermediate
vertices. No vertex is connected to itself, and so β = 0. Finally, the two intermedate vertices are connected to
each other with 2 identical lines, and so α2 = 1 (and αn = 0 for n > 2). This gives the same symmetry factor of
S = 4 as we found using the previous two methods.

6.2 Amplitude of Figure I.7.10 p.57, Zee I.7.2 p.60


Work out the amplitude corresponding to Figure I.7.10 p.57 in Eqn. 23 p.57.

SOLUTION: (Thanks to Kate and Melanie (’12) for presenting their solutions.)
6.3. TWO-TO-FOUR MESON PROCESSES, ZEE I.7.3 P.60 61

Wick Method: The amplitude for four external particles and four interaction points is
 2 Z Z
1 1 iλ i
R 4 4 2
M= − d w1 d w2 Dφ e 2 d x[(∂φ) −m φ] φ(x1 ) · · · φ(x4 )φ(w1 )4 φ(w2 )4 .
4 4
(6.2.1)
Z(0, 0) 2! 4!
We must count the various ways we can contract the φ’s in order to produce the desired diagram. There are
8 contraction choices for φ(x1 ). Then, φ(x2 ) must be contracted to the same point as φ(x1 ), so there are only
3 choices for this. φ(x3 ) must be contracted to a different point, of which there are 4 choices. φ(x4 ) must be
contracted to the same point as φ(x3 ), of which there are now only 3 choices. Now, we just have two more φ(w1 )’s
and two φ(w2 )0 s, which can be cross-contracted with each other in 2 ways. This gives a total of 8 · 3 · 4 · 3 · 2 = 4!2
equivalent diagrams. When multiplied by the prefactor 2!−1 4!−2 , this gives an overall factor of 21 (or a symmetry
factor of 2). Therefore, Eqn. (6.2.1) simplifies to

λ2
Z Z
2
M=− D1w1 D2w1 D3w2 D4w2 Dw 1 w2
. (6.2.2)
2 w1 w2
Plugging in the expressions for the propagators, integrating over w1 and w2 and amputating the external legs
and an overall momentum delta function, as in the previous problem, exactly reproduces Eqn. 23 of Zee p.57:
λ2 d4 k
Z
i i
M=− , (6.2.3)
2 (2π)4 k 2 − m2 + i (k1 + k2 − k)2 − m2 + i
where k is the momentum along one of the internal lines.

Schwinger Method: The amplitude may be written as


 2
2 Z  Y  4   ZZ 6
1 iλ δ 1 i
M= − d4 w i − d4 x d4 y J(x)D(x − y)J(y) . (6.2.4)
2! 4! i=1
iδJ(wi ) 6! 2

Now, we must count the number of ways to act the derivatives on the J’s. In the language of Zee, there are 6
strings with J’s on each end and D’s connecting the two endpoints. There are 12 choices for string ends to place
at x1 , then 10 for x2 , then 8 for x3 , then 6 for x4 . The other ends of the strings placed at x1 and x2 are attached
to each other at, say, a, as are the free ends of the strings placed at x3 and x4 at, say, b. Next, there are 4, then
2 choices of two string ends to place at a, at which point the other ends are attached to b. This alone accounts
for a factor
RR of 12!! = 26 6!, which nicely cancels the same factor in the denominator from the expansion term in
i
exp − 2 d x d4 y J(x)D(x − y)J(y) .
4

We have not accounted for the fact that the vertices could have any order of labeling from w1 to w2 . This
gives a factor of 2!. We must also account for the fact that the four derivatives at each vertex may be acted on
the string ends in any order, which gives a factor of 4! per vertex, or 4!2 . Overall, we have 2! · 4!2 , which neatly
cancels the same factor in the denominator out front in Eqn. (6.2.4).
However, we must be careful because we have overcounted the number of diagrams. Switching the two internal
lines leaves the diagram unchanged. This gives an internal symmetry structure of dimension 2 (i.e. a symmetry
factor of 2). Note that this is the same factor as was found using the Wick method.
As in the previous problem, this procedure is identical to the first henceforth.

Symmetry factors formula: In this case, g = 1, β = 0 and α2 = 1, which gives S = 2, as previously found.

6.3 Two-to-Four Meson Processes, Zee I.7.3 p.60


Draw all the diagrams describing two mesons producing four mesons up to and including order λ2 . Write down
the corresponding Feynman amplitudes.

SOLUTION: (Thanks to Ken (’12) for presenting his solution.)

There is only one diagram at order λ and seven at order λ2 :


We will not treat (b) separately from (a). There is no point in calculating the contribution of the vacuum bubble
because when one includes the contributions of all possible vacuum bubbles, one eventually divides out by that
contribution anyway.
62 CHAPTER 6. PROBLEM SET 6

@ @ @ @
@ i@ @ i @
i i
1 1
(a) O(λ) (b) O(λ2 ) (c) O(λ2 ) (d) O(λ2 )

3 3 3 3 4
@i @ @ @
@ i @ @
@ @
1 1 1 @2 1 @2
(e) O(λ2 ) (f) O(λ2 ) (g) O(λ2 ) (h) O(λ2 )

Figure 6.1: Diagrams describing two-to-four meson processes up to and including order λ2 .

1
Diagram (a) comes from the term 2! [W (J, λ)]2 in the expansion of Z = eW . Here, W is called the Wightman
1
function. However, the factor of 2! is cancelled by the fact that there are two ways to have each fully connected
piece (one is a line and the other is a trident) come from one of the two factors of W . Therefore, the amplitude
is simply Ma = −iλ.
In all the loop diagrams, the momentum in the loop is taken to be q. We pick up an integral over q and one
propagator for the internal line connecting the loop to the tree vertex. There are two integrals in diagram (f),
but there is also a momentum delta function, which fixes one of the internal lines to k1 − k3 − q. This diagram
also has a symmetry factor of 2, corresponding to the interchange of the two internal lines.
The amplitudes are
Ma = −iλ, (5.3.1a)
Z 4
i d q i
Mc = Md = (−iλ)2 , (5.3.1c)
k12 2
− m + i (2π) q − m2 + i
4 2

d4 q
Z
i i
Me = (−iλ)2 2 2
, (5.3.1e)
k3 − m + i (2π) q − m2 + i
4 2

d4 q
Z
i i
Mf = (−iλ)2 , (5.3.1f)
(2π)4 q 2 − m2 + i (k1 − k3 − q)2 − m2 + i
i
Mg = (−iλ)2 , (5.3.1g)
(k1 + k2 − k3 )2 − m2 + i
i
Mh = (−iλ)2 . (5.3.1h)
(k1 − k3 − k4 )2 − m2 + i
Of course, there are many diagrams with many different ways of assigning momenta for just one of the diagrams
drawn above. For example, we could change the assignments of momenta 3,4,5,6 in diagram (h). This is due to
the fact that we cannot distinguish in an experiment which particle came from which vertex.

6.4 Decay of a Scalar Particle, P&S 4.2 p.127


Consider the following Lagrangian, involving two real scalar fields Φ and φ:
L = 12 (∂µ Φ)2 − 12 M 2 Φ2 + 21 (∂µ φ)2 − 12 m2 φ2 − µΦφφ. (6.4.1)
The last term is an interaction that allows a Φ particle to decay into two φ’s, provided that M > 2m. Assuming
this condition is met, calculate the lifetime of the Φ to lowest order in µ.

SOLUTION: (Thanks to Jacob (’12) for presenting his solution.)

The only appropriate diagram is the trivalent diagram where one Φ turns into two φ’s. Putting a source for each
scalar field, J for Φ and j for φ, and expanding to linear order in J, quadratic order in j, and linear order in µ,
we may write the amplitude as
Z Z
µ
M= d4 w DΦ Dφ eiS0 Φ(x1 )φ(x2 )φ(x3 )Φ(w)φ(w)2 , (6.4.2)
Z(0)
6.4. DECAY OF A SCALAR PARTICLE, P&S 4.2 P.127 63

where S0 is the quadratic (free) part of the action, Z(0) is the free partition function with vanishing sources, and
the interaction occurs at point w, over which we integrate.
There are two connected ways to contract the φ’s. The contractions turn into propagators and integration
over the interaction point simply imposes overall momentum conservation. After amputating the external legs,
which are all of the legs, the only term left is
M ≈ 2µ, (6.4.3)
where ≈ means up to ±1 or ±i, which is irrelevant since physical quantities depend on |M|2 , not M.
The differential decay rate is given by Eqn. 4.86 P&S p.107. However, there is an extra factor of 21 due to
the fact that the outgoing particles are identical and thus we should actually count the diagram with the two
outgoing legs switched as distinct.

1 1 d3 p1 1 d3 p2 1
dΓ = · (2µ)2 (2π)4 δ(P − p1 − p2 ), (6.4.4)
2 2M (2π)3 2E1 (2π)3 2E2

where P, p1 , p2 are the 4-momenta of the parent and daughter particles, respectively. Since we are in the rest
frame of the parent particle, P = (M, 0, 0, 0), and the 3-momentum delta function imposes p2 = −p1 . Since the
daughter particles have the same mass, integration over d3 p2 just sets E2 = E1 and so

(2µ)2 (2π)4 1
Z 3
d p1
Γ= δ(M − 2E1 ). (6.4.5)
4 6
2 (2π) M E12
p
Write d3 p1 = |p1 |2 d|p1 | dΩ = E1 E12 − m2 dE1 dΩ and δ(M − 2E1 ) = 12 δ E1 − M

2 . Then,
p
µ2 E12 − m2 1
Z Z
M

Γ= dΩ dE1 δ E1 − 2
16π 2 M E1 2
r
µ2  2m 2 Z
= 1 − dΩ
32π 2 M M
r
µ2  2m 2
Γ= 1− . (6.4.6)
8πM M

As expected, this result is only applicable when M ≥ 2m. Furthermore, Γ is the decay constant; the actual
lifetime is τ = Γ−1 , since this is the only decay channel available to Φ. If we want to reintroduce factors of c
and ~, note that, prior to doing this, µ has units of mass, and so Γ has units of mass. Multiplying it by c2 gives
energy and dividing by ~ gives inverse time, as desired.
64 CHAPTER 6. PROBLEM SET 6
Chapter 7

Problem Set 7

7.1 Linear Sigma Model, P&S 4.3 p.127


The linear sigma model consists of N real scalar fields coupled by a φ4 interaction that is symmetric under
rotations of the N fields (O(N ) symmetry). The Hamiltonian is
Z  2 2 
H = d3 x 21 Πi + 12 ∇Φi + V Φ2 (7.1.1)

2
where Φi = Φ · Φ and
2 2 2
V (Φ2 ) = 12 m2 Φi + λ
4 Φi (7.1.2)

(a) If m2 > 0, show that the propagator is

Φi (x)Φj (y) = δ ij DF (x − y) (7.1.3)

where DF is the Klein-Gordon propagator for mass m, and that there is one type of vertex given by
k l
@ r
@ = −2iλ(δ ij δ kl + δ il δ jk + δ ik δ jl ) (7.1.4)
@
i j

Compute, to leading order in λ, the differential cross-sections dσ/dΩ in the center of mass frame for the
scattering processes

Φ1 Φ2 → Φ1 Φ 2 Φ1 Φ1 → Φ2 Φ2 Φ1 Φ1 → Φ1 Φ1

as a function of the center of mass energy.


(b) If m2 = −µ2 < 0, the potential V gains minima away from 0. By rotation invariance, we can choose the shift
to be only in the N th direction:

Φi (x) = π i (x) i = 1, . . . , N − 1
N
Φ (x) = v + σ(x) (7.1.5)

Show that now we have a theory of one massive σ field and N − 1 massless pion fields, interacting through
cubic and quartic potential energy terms which all become small as λ → 0. Construct the Feynman rules by
assigning values to the propagators and vertices.
(c) Compute the scattering amplitude for the process

π i (pA )π j (pB ) → π k (p1 )π l (p2 ) (7.1.6)

to leading order in λ. There are now four Feynman diagrams that contribute. Show that, at threshold
(pi = 0), these diagrams sum to 0. Show that if N = 2, the term O(p2 ) also cancels.

65
66 CHAPTER 7. PROBLEM SET 7

(d) Add to V a symmetry-breaking term ∆V = −aΦN where a is a small constant. Find the new value of v that
minimizes V , and work out the content of the theory about that point. Show that the pion acquires a mass
such that m2π ∼ a, and show that the pion scattering amplitude at threshold is now nonvanishing and also
proportional to a.

SOLUTION: (Thanks to Lenny (’12) for presenting his solution.)

(a) Notice that for λ = 0, the Hamiltonian (7.1.1) is exactly N copies of the Klein-Gordon Hamiltonian. Recall
that propagators are given by contractions of interaction fields. Thus, if the masses are all the same, then
the propagators are the same as the Klein-Gordon propagators except that we must make sure to distinguish
between the field species (hence, the Kronecker delta):

Φi (x)Φj (y) = δ ij DF (x − y) . (7.1.7)

In momentum space, this would be given by

iδ ij
Φi Φj = . (7.1.8)
p2 − m2 + i

Vertices follow from the perturbation term


Z   XZ Z
λ 2 2 λ 4 X λ 2 2
Hpert = d3 x Φ = d3 x Φi + d3 x Φi Φj . (7.1.9)
4 i
4 i<j
2

Thus, there are two vertex types:

(1) 4 lines of same field type with factor −i λ4 × 4! = −6iλ


(2) 2 lines of Φi and two of Φj6=i with factor −i λ2 × (2!)2 = −2iλ

This is indeed equivalent to the vertex factor (7.1.4) which would give −6iλ if i = j = k = l and −2iλ if
two pairs of the indices are separately equal.
To lowest order, we only have the one vertex diagram (7.1.4) for the scattering process Φi Φj → Φk Φl ,
which we notationally simplify to ij → kl. The scattering amplitudes are thus

M(12 → 12) = M(11 → 22) = −2λ, M(11 → 11) = −6λ. (7.1.10)

Now, we can use Equation 4.85 (PS p.107):

|M|2
 

= 2 , (4 identical masses). (7.1.11)
dΩ CM 64π 2 ECM

However, we must divide by 2 in the case when the final particles are of the same type:
 dσ 12→12 λ 2
= ,
dΩ CM 4πECM
 dσ 11→22  λ 2
= , (7.1.12)
dΩ CM 4πECM
 dσ 11→11  3λ 2
= .
dΩ CM 4πECM

(b) We rewrite our potential in terms of the fields (7.1.5):


−1 N −1 N −1
N
!
1 X λ X i 2 j 2 λ X i 2 λ
V (π i , σ) = − µ2 (π i )2 + (v + σ)2 + (π ) (π ) + (π ) (v + σ)2 + (v + σ)4 .
2 i=1
4 i,j=1
2 i=1
4
7.1. LINEAR SIGMA MODEL, P&S 4.3 P.127 67

We choose v so as to minimize V when π i and σ all vanish. First we would differentiate with respect to π i
and set that equal to 0 when π i = 0. Since each term will have a π i in it, this is automatically satisfied.
The derivative with respect to σ gives
−1
  N
!
∂V 2
X
i 2 3
0= = −µ (v + σ) + λ (π ) (v + σ) + λ(v + σ)
∂σ πi =σ=0 i=1 π i =σ=0
= λv 3 − µ2 v. (7.1.13)

v = 0 is a solution but it is a local maximum, not a minimum, since the second derivative, 3λv 2 − µ2 is
negative when v = 0. Therefore, the minimum occurs at
µ
v=√ . (7.1.14)
λ

The Lagrangian is now


1 1 1 1 λ 1 λ
L= (Dπ)2 + (Dσ)2 + µ2 π 2 + µ2 σ 2 − µ2 vσ − (π 2 )2 − λv 2 π 2 − λvσπ 2 − π 2 σ 2
2 2 2 2 4 2 2
3 3 2 2 3 λ 4
− λv σ − λv σ − λvσ − σ + const.
2 4
1 2 1 2 1 1
= (Dπ) + (Dσ) − ( λv− µ2 )π 2 − (3λv 2 − µ2 )σ 2 − v(
2 2 
− µ2 )σ

 λv
2 2 2 2
λ
− λvσ(π 2 + σ 2 ) − (σ 2 + π 2 )2 + const.
4
1 1 λ
= (Dπ) + (Dσ) − µ2 σ 2 −λvσ(σ 2 + π 2 ) − (σ 2 + π 2 )2 + const.
2 2
(6.2.15a)
2
| 2 {z } 4
free part

PN −1
where π 2 = π · π = i (π i )2 and (DΦ)2 = Π2 + (∇Φ)2 . Indeed,

λ 2 λ→0
Lint = −λvσ(σ 2 + π 2 ) − (σ + π 2 )2 −−−→ 0 . (6.2.15b)
4

The free term describes N − 1 massless real scalar fields π i (x). We expect this because the exact continuous
O(N ) symmetry was spontaneously broken by one process (m2 acquiring a negative value), which should
thus give N − 1 Goldstone massless particles with O(N − 1) symmetry. There is also one massive real scalar
field σ(x) of mass

mσ = 2 µ . (7.1.16)

Since the free part again looks like the Klein-Gordon Lagrangian, the propagators are the same (just one
with different mass):

i
σ(x)σ(y) = DF (x − y)m=mσ , σ σ = , (6.2.17a)
p2 − 2µ2 + i

iδ ij
π i (x)π j (y) = δ ij DF (x − y)m=mπ =0 , πi πj = . (6.2.17b)
p2 + i

The last two terms in the Lagrangian (6.2.15a), excluding the irrelevant constant, give the interaction
Z  
λ λ λ
Hpert. = d3 x λvσ 3 + λvσπ 2 + σ 4 + σ 2 π 2 + (π 2 )2 . (7.1.18)
4 2 4

As in part (a), we have


k l
@ r
@ = −2iλ(δ ij δ kl + δ il δ jk + δ ik δ jl ). (6.2.19a)
@
i j
68 CHAPTER 7. PROBLEM SET 7

λ 4 −iλ
The 4σ contributes a factor 4 × 4! = −6iλ:

@r
@
@
@ = −6iλ. (6.2.19b)
@
@

λ 2 i j −iλ
The 2σ π π contributes a factor 2 × 2 × 2 × δ ij = −2iλδ ij :

@
@
@@r = −2iλδ ij . (6.2.19c)
@ @
i j

The λvσ 3 term contributes −iλv × 3! = −6iλv:

r = −6iλv. (6.2.19d)
@
@ @
@

Finally, the λvσπ i π j term contributes −iλv × 2δ ij = −2iλvδ ij :

r = −2iλvδ ij . (6.2.19e)
@ @
i j

(c) The π 4 diagram (the first one treated in part (b)) has the value (7.1.4). The others have

π k (p1 ) π l (p2 )
@r
@
i
= (−2iλvδ ij ) (−2iλvδ kl ), (6.2.20a)
(pA + pB )2 − 2µ2 + i
r
@ @
π i (pA ) π j (pB )

π k (p1 ) π l (p2 )
i
@r
@ r = (−2iλvδ ik ) 2 2 + i
(−2iλvδ jl ), (6.2.20b)
@ (p A − p1 ) − 2µ
@
π i (pA ) π j (pB )

π k (p1 ) π l (p2 )
HH 
i
rH
Hr = (−2iλvδ il ) 2 2 + i
(−2iλvδ jk ). (6.2.20c)
@ (p A − p2 ) − 2µ
@
i
π (pA ) π j (pB )

This gives the following value for M:


2λv 2 2λv 2
    
M = −2λ δ ij δ kl 1 + + δ ik jl
δ 1 +
(pA + pB )2 − 2µ2 + i (pA − p1 )2 − 2µ2 + i
2λv 2
 
+ δ il δ jk 1 +
(pA − p2 )2 − 2µ2 + i
(pA + pB )2 + i (pA − p1 )2 + i (pA − p2 )2 + i
 
ij kl ik jl il jk
= −2λ δ δ +δ δ +δ δ .
(pA + pB )2 − 2µ2 + i (pA − p1 )2 − 2µ2 + i (pA − p2 )2 − 2µ2 + i
7.1. LINEAR SIGMA MODEL, P&S 4.3 P.127 69

where we used that λv 2 = µ2 from Equation (7.1.14).


Let us drop the ’s or else make it implicit. Let us also define the standard Mandelstam variables as usual.
Then, the scattering amplitude is
 
s t u
M = −2λ δ ij δ kl + δ ik δ jl + δ il δ jk . (7.1.21)
s − 2µ2 t − 2µ2 u − 2µ2

At the threshold (initial momenta = 0), the outgoing momenta must also vanish by momentum and energy
conservation. Hence, all Mandelstam variables vanish and thus

M=0 at threshold. (7.1.22)

Let us consider small momenta. Then the Mandelstam variables are small compared to µ2 . Factor −2µ2 out
from all the terms in the parentheses in (7.1.21). Then, we get a prefactor λ/µ2 = 1/v 2 using (7.1.14). But,
the Mandelstam variables are all order p2 and hence we just consider the numerators and set the denominators
equal to 1 (this is lowest order in the Mandelstam variables). Thus,
1 ij kl
δ δ s + δ ik δ jl t + δ il δ jk u + O(p4 /m2σ ) .

M≈ (7.1.23)
v2

In general the terms written above are order p2 . However, if N = 2 for which there is only one species of
pion, all the Kronecker deltas are 1. The subscript indicates MN :
1 1
s + t + u + O(p4 /m2σ ) = 2 4m2π + O(p4 /m2σ ) = 0 + O(p4 /m2σ ),
 
M2 ≈ (7.1.24)
v2 v

where we have used the standard result s + t + u = 4m2 and the fact that the pions are massless.
Thus, if N = 2, then the term of order p2 also vanishes.

(d) Once again, we write the fields as in (7.1.5) except that v is now slightly different from before (slightly
because a is small). There is a term −a(v + σ) added to V . When we take the derivative of V with respect
to σ, this contributes a term −a to Eqn. (7.1.13) giving the equation

λv 3 − µ2 v − a = 0. (7.1.25)

Write v as a small perturbation, δ, around the old value:


r
µ2
v= + δ. (7.1.26)
λ

Keeping only up to linear order in δ in Eqn. (7.1.25) yields


r
 µ2 
3/2
 µ2 µ2 a
2 
λ  + 3λ δ − µ  − µ2 δ − a = 2µ2 δ − a = 0 =⇒ δ= . (7.1.27)
 λ λ  λ 2µ2

Thus, we have our new v to first order in a:


r
µ2 a
v= + 2 . (7.1.28)
λ 2µ

p
We have λv 2 = µ2 + a λ/µ2 + O(a2 ). From the work in (6.2.15a), we now have
s
λ
m2π 2 2
= λv − µ ≈ a . (7.1.29)
µ2
70 CHAPTER 7. PROBLEM SET 7

Again, from the work of (6.2.15a), we have the mass of the σ:

m2σ = 3λv 2 − µ2 = 2µ2 + 3(λv 2 − µ2 ) = 2µ2 + 3m2π . (7.1.30)

But, the free part of the theory still looks like massive Klein-Gordon fields. Hence,

i
σ(x)σ(y) = DF (x − y)m=mσ , σ σ = , (6.2.31a)
p2 − m2σ + i

iδ ij
π i (x)π j (y) = δ ij DF (x − y)m=mπ , πi πj = . (6.2.31b)
p2 − m2π + i

In complete analogy to the work of part (c), M will be given by

2λv 2 2λv 2
    
ij kl ik jl
M = −2λ δ δ 1+ +δ δ 1+
(pA + pB )2 − m2σ (pA − p1 )2 − m2σ
2λv 2
 
+ δ il δ jk 1 + .
(pA − p2 )2 − m2σ

Then, using (7.1.28), this simplifies to

− m2π 2 2
 
ij kl s ik jl t − mπ il jk u − mπ
M = −2λ δ δ + δ δ + δ δ . (7.1.32)
s − m2σ t − m2σ u − m2σ

As we did in part (c), we factor our −m2σ from the denominator and set the denominator to 1 just to see the
order p2 terms:
 
2λ 1
δ ij δ kl (s − m2π ) + δ ik δ jl (t − m2π ) + δ il δ jk (u − m2π ) + O(p4 /m2σ ) .

M= 2
≈ 2 (7.1.33)
mσ v

Again, at threshold, t = u = 0 and s = 4m2π and we have

m2π a  λ 3/2
M= = = a + O(a2 ) . (7.1.34)
v2 v3 µ2

As an aside, we can actually solve the cubic equation, Eqn. (7.1.25), with some neat tricks and it is a bit
simpler than solving a general cubic equation because the first step is usually to shift the variable by a
constant in order to eliminate the quadratic term; well, our cubic equation already has vanishing quadratic
term! Let us rewrite the equation as

µ2 a
v 3 − 3pv − 2q = 0, p= , q= . (7.1.35)
3λ 2λ

Then we perform a Vieta substitution:


p
v=w+ . (7.1.36)
w
After a bit of algebra, we can rewrite the cubic equation in v in terms of w as follows:
 p 3
w3 + − 2q = 0. (7.1.37)
w

Multiplying by w3 turns this equation into a simple quadratic equation for w3 , which I choose to divide by
2 for convenience:
1 6 1
w − qw3 + p3 = 0. (7.1.38)
2 2
7.2. RUTHERFORD SCATTERING, P&S 4.4 P.129 71

The quadratic equation gives the solution


p
w3 = q ± q 2 − p3 . (7.1.39)

The solution is therefore, q


3
p p
v= q± q 2 − p3 + q p . (7.1.40)
3
q± q 2 − p3

This is not particularly illuminating. Also, we have a rather large degeneracy of possible solutions here
q→0 √ √
depending on the ± sign and how we take the cube-root. We want the solution such that v −−−→ 3p = µ/ λ.
It turns out that the correct choice for the ± is +. If q = 0, then, our expression reads
q=0 p √
= (e±iπ/6 − e±5iπ/6 ) p = 3p,
p
v −−→ (−p3 )1/6 + 3 1/6
(7.1.41)
(−p )

where we used the expressions


√ √
1/6 ±iπ/6 3 i −1/6 5/6 ±5iπ/6 3 i
(−1) =e = ± , (−1) = −(−1) = −e = ∓ .
2 2 2 2

We can see why the choice of − for ± in Eqn. (7.1.40) fails. It multiplies the first term in Eqn. (7.1.41) by
e±iπ/3 and the second by (−1)−1/3 = −(−1)2/3 = −e±2iπ/3 . None of the possible combinations of
(−1)1/3 = √
signs give 3p; they either give 0 or else something complex.
Now, we look for the lowest order term in q. The q 2 term in the square-root in w3 (Eqn. (7.1.39)) is
irrelevant to this order. Thus, we find
1/3  q  q
w ≈ q + (−p3 )1/2 ≈ (−p3 )1/6 1 + 3 1/2
= (−p3 )1/6 + .
3(−p ) 3(−p3 )1/3

We can use this to write 1/w as well. Below are the simplified expressions for w and 1/w:

√ 1 q 1 e±5iπ/6 1 q
w = e±iπ/6 p − e±2iπ/3 , =− √ + e±iπ/3 2 . (7.1.42)
3 p w p 3 p

Plugging these into v (Eqn. (7.1.36)) gives precisely the same result as Eqn. (7.1.28):
√ 1 q p q µ a
v = (e±iπ/6 − e±5iπ/6 ) p + (e±iπ/3 − e±2iπ/3 ) = 3p + = √ + 2. (7.1.43)
3 p 3p λ 2µ

The next term in the expansion can be computed with some effort:

µ a 3 λ a2
v=√ + 2− . (7.1.44)
λ 2µ 8µ5

In turn, this can be used to work out the quadratic correction to the pion mass:

a λ a2 λ
mπ = − 4. (7.1.45)
µ 2µ

The σ mass is still related to the π mass according to Eqn. (7.1.30).

7.2 Rutherford Scattering, P&S 4.4 p.129


The cross section for scattering of an electron by the Coulomb field of a nucleus can be computed, to lowest
order, without quanitizing the electromagnetic field. Instead, treat the field as a given, classical potential Aµ (x).
The interaction Hamiltonian is Z
HI = d3 x eψγ µ ψAµ , (7.2.1)

where ψ(x) is the usual quantized Dirac field.


72 CHAPTER 7. PROBLEM SET 7

(a) Show that the T -matrix element for electron scattering off a localized classical potential is, to lowest order,

hp0 |iT |pi = −ieū(p0 )γ µ u(p) · A


eµ (p0 − p), (7.2.2)

where A
eµ (q) is the four-dimensional Fourier transform of Aµ (x).

(b) If Aµ (x) is time independent, its Fourier transform contains a delta function of energy. It is then natural to
define
hp0 |iT |pi ≡ iM · (2π)δ(Ef − Ei ), (7.2.3)

where Ei and Ef are the initial and final energies of the particle, and to adopt a new Feynman rule for
computing M:
J
]
J
J h = −ieγ µ A eµ (q), (7.2.4)

where A eµ (q) is the three-dimensional Fourier transform of Aµ (x). Given this definition of M, show that the
cross section for scattering off a time-independent, localized potential is

1 1 d3 pf 1
dσ = |M(pi → pf )|2 (2π)δ(Ef − Ei ), (7.2.5)
vi 2Ei (2π)3 2Ef

where vi is the particle’s initial velocity. This formula is a natural modiciation of Eqn. 4.79 P&S p.106.
Integrate over |pf | to find a simple expression for dσ/dΩ.
(c) Specialize to the case of electron scattering from a Coulomb potential (A0 = Ze/4πr). Working in the
nonrelativistic limit, derive the Rutherford formula,

dσ α2 Z 2
= . (7.2.6)
dΩ 4m2 v 4 sin4 (θ/2)

(With a few calculational tricks from Section 5.1, you will have no difficulty evaluating the general cross
section in the relativistic case; see Problem 5.1.)
SOLUTION: (Thanks to Melanie for presenting her solution to this problem.)

(a) The transition amplitude in general is


d4 x ψγ µ ψAµ
R R
hp0 |T e−i H dt
|pi = hp0 |T e−ie |pi.

The zeroth order term is obviously just hp0 |pi = (2π)4 δ(p0 − p). We do not care about this term since it
contains no interactions. The first order term is
Z
hp0 |iT |pi = −ie d4 x Aµ (x) p0 ψ(x)γ µ ψ(x) p

Z
0
= −ie d4 x Aµ (x) ū(p0 )γ µ u(p)eix(p −p)

hp0 |iT |pi = −ieū(p0 )γ µ u(p)A


eµ (p0 − p) . (7.2.7)

Contracting p0 with ψ and p with ψ would instead describe a positron interacting with the classical field.
(b) As in Eqn. 4.68 P&S p.103, we set the “in” state to be a superposition of plane wave states with some
distribution, ψ(k) e−ik·b over momenta and impact parameters

d3 k
Z
1
|ψiin = 3
√ ψ(k) e−ik·b |kiin . (7.2.8)
(2π) 2Ek
7.2. RUTHERFORD SCATTERING, P&S 4.4 P.129 73

So that in hψ|ψiin = 1, the distribution must be normalized to 1, as in Eqn. 4.66 P&S p.102:
d3 k
Z
|ψ(k)|2 = 1. (7.2.9)
(2π)3

The probability for the initial state to scatter to a final state containing one electron with momentum pf is
d3 pf 1 2
P(b) = 3 out hpf |ψiin

(2π) 2Ef
d3 pf 1 d3 k ψ(k) d3 q ψ ∗ (q) −i(k−q)·b
Z
= 3
√ e (out hpf |kiin )(out hpf |qiin )∗ . (7.2.10)
(2π)3 2Ek (2π)3 2Eq
p
(2π) 2Ef

As instructed, we write
out hpf |kiin = iM(k → pf )2πδ(Ef − Ek ). (7.2.11)
Consider first the q|| integral:
dq|| ψ ∗ (q) ∗
Z
p M (q → pf )2πδ(Ef − Eq ). (7.2.12)
2π 2Eq

Using the rule for delta functions of functions, we write


q −1  q  1
||
δ q|| − q||0 ,

δ(Ef − Eq ) = δ q|| − |pf |2 − |q⊥ |2 = (7.2.13)
Eq | {z } v(q)
0
q||

where v(q) is the speed of the incident particles with momentum q.


The cross-section is given by the integral over all impact parameters of P(b), as in Eqn. 4.75 P&S p.105.
The only part of the integrand that depends on b is the exponential whose integral yields a delta function
over the transverse components of the momenta:
Z
d2 b e−i(k−q)·b = (2π)2 δ(k⊥ − q⊥ ). (7.2.14)

Now consider the q⊥ integral:


1 ψ ∗ (q)
Z 2
d q⊥ 2 ∗

(2π) δ(k − q ) M (q → p ) . (7.2.15)

2 ⊥ ⊥ p f
(2π) v(q) 2Eq
0
q|| =q||

Taken together, the delta functions δ(Ef − Ek ), δ(k⊥ − q⊥ ), and δ(q|| − q||0 ) impose the equality q = k. This
collapses the integral, Eqn. (7.2.15), via evaluation at q = k. Therefore, the k integral is
d3 k 1 |ψ(k)|2
Z
|M(k → pf )|2 2πδ(Ef − Ek ). (7.2.16)
(2π)3 v(k) 2Ek

As per the discussion on P&S p.106 around Eqns. 4.78 and 4.79, we can pull all the factors except ψ out of
the integral and evaluate them at the peak of the distrubution at k = pi . After doing so, the k integral just
becomes 1 by the normalization condition, Eqn. (7.2.9). What remains is exactly what we want:

d3 pf 1 1 1
dσ = |M(pi → pf )|2 (2π)δ(Ef − Ei ) . (7.2.17)
(2π)3 2Ef vi 2Ei
q
To integrate over pf , we change variables to Ef = |pf |2 + m2 . The relationship between the differentials
q
is |pf |2 d|pf | = Ef Ef2 − m2 dEf . Thus,
Z q 1
σ = dΩ dEf  E f Ef2 − m2 |M|2 δ(Ef − Ei )
16π 2 vi Ei
E
f
p
Ei2 − m2
Z
= dΩ |M|2 . (7.2.18)
16π 2 vi Ei
74 CHAPTER 7. PROBLEM SET 7

|pi |
1
p
Note that Ei Ei2 − m2 = Ei = vi , which cancels the vi in the denominator. Therefore,

dσ |M|2
= . (7.2.19)
dΩ 16π 2

(c) We must Fourier transform the Coulomb potential:


Z Z Z
ik·x Ze ik0 t Ze
e 4
A0 (k) = d x e = dt e d3 x e−ik·x = 2πδ(k 0 ) A
e0 (k). (7.2.20)
4πr 4πr

We have done this already when we considered the generalization of the inverse square law in arbitrary
dimension in problem set 4:
d3 k eik·x e−mr
Z
Ze
lim Ze 3 2 2
= lim Ze = = A0 . (7.2.21)
m→0 (2π) |k| + m m→0 4πr 4πr

Therefore,
Ze Ze
A
e0 (k) = lim = , (7.2.22)
m→0 |k|2 + m2 |k|2
and the four-dimensional Fourier transform is
e0 (k) = Ze 2πδ(k 0 ).
A (7.2.23)
|k|2

Combining Eqns. (7.2.2), (7.2.3) and (7.2.23) gives the amplitude:


Ze
Mrs = −eūr (pf )γ 0 us (pi ) . (7.2.24)
|pf − pi |2

Note that, until now, we have kept the spinor indices implicit. The spinor product may be simplified in the
non-relativistic limit to
ūr (pf )γ 0 us (pi ) = ur† (pf )us (pi )
q q 
= (pf · σ ∗ )(pi · σ) + (pf · σ̄ ∗ )(pi · σ̄) ξ r† ξ s
E>>|p|
−−−−−→ 2 Ei Ef δ rs .
p
(7.2.25)

In the non-relativistic limit, the energy is much greater than the momentum and thus only the terms in
the square roots involving energy will matter; these are just the dot products with σ 0 = 1 and its bar
and complex conjugates, which are obviously all still equal to the identity. Since there is an overall energy
conserving delta function, we might as well set Ei = Ef = E, in which case
2Ze2 Eδ rs
Mrs = − . (7.2.26)
|pf − pi |2

In order for the amplitude to be non-vanishing, the incoming and outgoing electrons must have the same
spin state. There are two such spin states, giving a factor of 2. Then, we average over these states, which
means we divide by 2. These two factors of 2 cancel. Finally, since Ei = Ef = E, the scattering is elastic
and |p|f = |pi | = P . Thus, the differential scattering cross section, given by Eqn. (7.2.19), is
dσ 4Z 2 e4 E 2 4Z 2 e4 E 2 Z 2 e4 E 2
= = = . (7.2.27)
dΩ 16π 2 |pf − pi |4 16π 2 [2P 2 (1 − cos θ)]2 16π 2 P 4 (1 − cos θ)2

Note that we used the fact that |pf − pi |2 = |pf |2 + |pi |2 − 2pf · pi = 2P 2 (1 − cos θ). Here, θ is the angle
between pf and pi . We write P 4 = E 4 v 4 . Two factors of E cancel between the numerator and denominator
leaving a factor of E 2 in the denominator, which we replace by m2 in the non-relativistic limit:
dσ Z 2 e4
= . (7.2.28)
dΩ 16π 2 m2 v 4 (1 − cos θ)2
7.3. DISCUSSION: GOLDSTONE BOSONS (BEING EATEN UP) 75

Next, we use the trigonometric identity sin θ2 = 12 (1 − cos θ) and the definition of the fine structure constant,
α = e2 /4π, to write
dσ α2 Z 2
= . (7.2.29)
dΩ 4m v 4 sin4 (θ/2)
2

As an aside, let us try to complete this calculation in the fully relativistic case. As before, |pf − p1 |2 =
2P 2 (1 − cos θ) = 4P 2 sin2 θ2 . The amplitude is

Ze2
Mrs = − ūr (pf )γ 0 us (pi ). (7.2.30)
4P 2 sin2 θ2

Therefore, the unpolarized amplitude squared is

1 XX e4 Z 2 e4 Z 2
A≡ |M|2 = 4 ūs (pi )γ 0 ur (pf )ūr (pf )γ 0 us (pi ) = tr[(p 0
/i + m)γ (p
0
/f + m)γ ].
2 s r 4
32P sin 2 θ
32P 4 sin4 θ
2
(7.2.31)
Let us expand out the trace:

/f ] + m tr 1.
0 0 0 0 2
tr[(p
/i + m)γ (p
/f + m)γ ] = tr[p
/i γ p/f γ ] + m tr[p
/i + p (7.2.32)

The middle terms vanish because they involve traces of just one γ matrix. The last term is simply 4m2 . For the
first term, we move the γ 0 past one of the slashed momenta to combine the two γ 0 ’s together to 1:
0 0 0ν ν 0 0
tr[p
/i γ p /i pf ν (2g − γ γ )γ ]
/f γ ] = tr[p
0
/i γ ] − tr[p
= 2Ef tr[p /i p
/f ]
= 8Ei Ef − 4pi · pf
= 4(E 2 + P 2 cos θ), (7.2.33)

where Ef = Ei = E and |pf | = |pi | = P by energy conservation.


Therefore, we find

e4 Z 2 2 2 2 e4 Z 2 2 2 e4 Z 2 E 2  P2 2 θ

A= (E + P cos θ + m ) = [2E − P (1 − cos θ)] = 1 − sin . (7.2.34)
8P 4 sin4 θ2 8P 4 sin4 θ2 4P 4 sin4 θ2 E2 2

If we replace P/E with v and E 2 with m2 /(1 − v 2 ), we find

dσ A α2 Z 2 2 2 2 θ

= 2
= 4 θ
(1 − v ) 1 − v sin 2 . (7.2.35)
dΩ 16π 4m2 v 4 sin 2

7.3 Discussion: Goldstone Bosons (being eaten up)


Whenever a symmetry holds for the Hamiltonian, but not for the ground state, the symmetry is said to have
been spontaneously broken. Each spontaneously broken continuous symmetry produces a massless excitation
(particle) called a Goldstone boson. In the example of the previous problem, prior to symmetry breaking, the
fields enjoyed an O(N ) rotation symmetry. Afterwards, only the π fields enjoy rotation symmetry and there are
only N − 1 of them, giving an O(N − 1) symmetry group. We worked out the dimension of O(N ) in problem set
2 and found |O(N )| = 21 N (N − 1). Therefore, the number of symmetry generators that have been broken is

|O(N )| − |O(N − 1)| = 21 N (N − 1) − 12 (N − 1)(N − 2) = N − 1. (7.3.1)

As expected, there should be N − 1 massless Goldstone bosons (i.e. the π fields).


An interesting phenomenon occurs when the symmetry is a gauge symmetry (local symmetry), meaning that
the symmetry parameter depends on spacetime. In the previous problem, if the O(N ) rotation symmetry were
gauge, then it would mean that we could perform different rotations at different points in spacetime (varying
smoothly over spacetime, of course). This is NOT a symmetry of the previous problem; for that problem, the
76 CHAPTER 7. PROBLEM SET 7

rotation had to be constant and so the symmetry is called a global one. Sometimes, we can gauge a global
symmetry (i.e. make it local) by introducing extra fields to compensate. The complex scalar is an example of
this. Recall from problem set 2 that the free Lagrangian for a massive complex scalar is

L = ∂µ φ∗ ∂ µ φ − m2 φ∗ φ. (7.3.2)

Again, recall from problem set 2 that this has a global symmetry with corresponding current given by

φ → eiα φ, j µ = i(φ ∂ µ φ∗ − φ∗ ∂ µ φ). (7.3.3)

We can turn this into a gauge symmetry if we simultaneously add a vector field, Aµ , which has a compensating
transformation. This is the analog of Eqn. 4.6 P&S p.78 in the case of a complex scalar field:

φ(x) → eiα(x) φ(x), Aµ (x) → Aµ (x) − 1e ∂µ α(x), (7.3.4)

as long as we also alter the Lagrangian by turning the derivatives into covariant derivatives as in Eqn. 4.5 P&S
p.78, i.e. ∂µ → Dµ = ∂µ + ieAµ . Of course, we take the complex conjugate when this derivative acts on φ∗ :

L = Dµ∗ φ∗ Dµ φ − m2 φ∗ φ. (7.3.5)

This turns out to be equivalent to adding an interaction term to the original Lagrangian of the form Aµ j µ , where
j µ is the same as in Eqn. (7.3.3) with the derivatives turned into covariant ones. This suggests including the
usual Yang-Mills term for Aµ :
L = − 41 Fµν F µν + Dµ∗ φ∗ Dµ φ − m2 φ∗ φ, (7.3.6)
where Fµν = ∂µ Aν − ∂ν Aµ , so that the Euler-Lagrange equation of motion for Aµ is

∂ν F µν = j µ , (7.3.7)

which is simply the statement that the current associated with the scalar field sources the electromagnetic field
(we can now associate the vector field with the massless Maxwell (photon) field).
Finally, we add a quartic φ interaction potential just as in the previous problem:

L = − 14 Fµν F µν + Dµ∗ φ∗ Dµ φ − m2 φ∗ φ − λ2 (φ∗ φ)2 . (7.3.8)

Then, we consider the situation when m2 = −µ2 < 0. For convenience, we add a constant µ4 /2λ to L in order
to write it in the form 2 2
L = − 14 Fµν F µν + Dµ∗ φ∗ Dµ φ − λ2 φ∗ φ − µλ . (7.3.9)
As usual, the potential is minimized at |φ|2 = µ2 /λ ≡ v 2 /2. Write

φ= √1 (φ1 + iφ2 ).
2
p
By rotation symmetry in (φ1 , φ2 ) space, we can choose to expand φ1 around 2µ2 /λ = v and φ2 around 0:

φ1 = v + h, φ2 = h̃. (7.3.10)

Plugging this in gives the Lagrangian

L = − 41 (Fµν )2 + 12 (∂µ h)2 + 12 (∂µ h̃)2 − eh̃Aµ ∂ µ h + e(v + h)Aµ ∂ µ h̃ + 21 e2 Aµ Aµ [(v + h)2 + h̃2 ]
− µ2 h2 − 2λvh(h2 + h̃2 ) − λ8 (h2 + h̃2 )2 . (7.3.11)

This is a rather complicated Lagrangian with many interactions and even derivative √ interactions between the
various fields. Nevertheless, we see that h is a Klein-Gordon field with mass mh = 2 µ and h̃ has zero mass.
However, we note that the gauge symmetry, φ(x) → eiα(x) φ(x) allows us to fix the gauge (i.e. choose a specific
α(x)) in order to force φ to be real. If we write φ = ϕeiθ , then this is accomplished by setting α(x) = −θ(x).
Note that it is important that the symmetry be gauge. If the symmetry is global, we can always make φ real at
at least one spacetime point, x0 , by setting α = −θ(x0 ). However, φ cannot be made real throughout spacetime.
When α can vary locally, this can be done. Fixing the gauge in this way allows us to get rid of h̃ altogether
leaving only
L = − 41 (Fµν )2 + 12 (∂µ h)2 + 12 e2 Aµ Aµ (v + h)2 − µ2 h2 − 2λvh3 − λ8 h4 . (7.3.12)
7.3. DISCUSSION: GOLDSTONE BOSONS (BEING EATEN UP) 77

Note that the vector field now has a mass, mA = ev. What has happened here? We started off with 2 degrees of
freedom in the complex scalar field and 2 degrees of freedom in the massless vector field (massless vectors have
only the two transverse polarization states and no longitudinal one since there is no rest frame). However, we have
ended up with one massive scalar degree of freedom and one massive vector field with three degrees of freedom
since now it must have a longitudinal polarization state. The language often used to describe this phenomenon is
that the vector field has eaten the would-be Goldstone boson, h̃, which then becomes the longitudinal polarization
of the resulting massive vector field.
In principle, this is the story of the Ginzburg-Landau theory of superconductivity. In that case, the phe-
nomenon of magnetic flux exclusion in the superconductor (Meissner effect) is a manifestation of the photon
gaining a large effective mass in the bulk of the superconductor. Nambu took this story as his inspiration behind
electroweak symmetry breaking. In this case, the weak interactions enjoy an SU (2) symmetry with the corre-
sponding massless vector fields being the weak force carriers, the W ± and Z 0 bosons. One adds a Higgs scalar
field into the mix in a way very similar to what we did above. The difference is that there are actually 4 real
Higgs degrees of freedom: the Higgs field is an SU (2) doublet the two components of which are complex, and
the covariant derivatives now include all three vector fields corresponding to the W ± and Z 0 instead of just the
one Aµ in our example above. Again, one adds a quartic interaction for the Higgs and flips the sign of the mass
squared. Rotation symmetry in the components of the Higgs field allows us to choose one to expand around the
minimum of the potential and the other three to expand around 0. This should produce one massive Higgs scalar
and three massless Goldstone modes. However, just as in our example above, the SU (2) symmetry is gauge and
one can fix the gauge so as to get rid of the three would-be Goldstone bosons altogether. What remains is one
massive Higgs scalar and the three vector fields, W ± and Z 0 , which are now massive. Thus, we say that the W ±
and Z 0 bosons ate three of the massless Higgs modes, which become the longitudinal modes of the now-massive
W ± and Z 0 .
The fact that the SU (2) symmetry is broken but the U (1) symmetry of electromagnetism is not means that
the W ± and Z 0 become massive, but the photon remains massless. Of course, as a result, the Coulomb force
remains a long-range force, but the weak force becomes a short-range force with a range roughly equal to the
inverse mass of the W ± and Z 0 .
78 CHAPTER 7. PROBLEM SET 7
Chapter 8

Problem Set 8

8.1 Massless Tree Diagrams, P&S 5.3 p.170


The spinor product formalism introduced in Problem 3.3 provides an efficient way to compute tree diagrams
involving massless particles. Recall that in Problem 3.3 we defined spinor products as follows: Let uL0 , uR0 be
the left- and right-handed spinors at some fixed lightlike momentum k0 . These satisfy
 1 − γ5   1 + γ5 
uL0 ūL0 = k/0 , uR0 ūR0 = k/0 . (8.1.1)
2 2
P
(These relations are just the projections onto definite helicity of the more standard formula u0 ū0 = k/0 .) Then
define spinors for any other lightlike momentum p by
1 1
uL (p) = √ p
/uR0 , uR (p) = √ p
/uL0 . (8.1.2)
2p · k0 2p · k0
We showed that these spinors satisfy p
/u(p) = 0; because there is no m around, they can be used as spinors for
either fermions or antifermions. We defined

s(p1 , p2 ) = ūR (p1 )uL (p2 ), t(p1 , p2 ) = ūL (p1 )uR (p2 ), (8.1.3)

and, in a special frame, we proved the properties

t(p1 , p2 ) = s(p2 , p1 )∗ , s(p1 , p2 ) = −s(p2 , p1 ), |s(p1 , p2 )|2 = 2p1 · p2 . (8.1.4)

Now let us apply these result.


(a) To warm up, give another proof of the last relation in Eqn. (8.1.4) by using Eqn. (8.1.1) to rewrite |s(p1 , p2 )|2
as a trace of Dirac matrices, and then applying the trace calculus.
(b) Show that, for any string of Dirac matrices,

tr[γ µ γ ν γ ρ · · · ] = tr[· · · γ ρ γ ν γ µ ], (8.1.5)

where µ, ν, ρ . . . = 0, 1, 2, 3, or 5. Use this identity to show that

ūL (p1 )γ µ uL (p2 ) = ūR (p2 )γ µ uR (p1 ). (8.1.6)

(c) Prove the Fierz identity

ūL (p1 )γ µ uL (p2 )[γµ ]ab = 2[uL (p2 )ūL (p1 ) + uR (p1 )ūR (p2 )]ab , (8.1.7)

where a, b = 1, 2, 3, 4 are Dirac indices. This can be done by justifying the following statements: The right-
handed side of this equation is a Dirac matrix; thus, it can be written as a linear combination of the 16 Γ
matrices discussed in Section 3.4. It satisfies

γ 5 [M ] = −[M ]γ 5 , (8.1.8)

79
80 CHAPTER 8. PROBLEM SET 8

thus, it must have the form


 1 − γ5   1 + γ5 
[M ] = γµ V µ + γµ W µ , (8.1.9)
2 2
where V µ and W µ are 4-vectors. These 4-vectors can be computed by trace technology; for example,
1 h ν  1 − γ5  i
Vν = tr γ M . (8.1.10)
2 2

(d) Consider the process e+ e− → µ+ µ− , to the leading order in α, ignoring the masses of both the electron and
the muon. Consider first the case in which the electron and the final muon are both right-handed and the
positron and the final antimuon are both left-handed. (Use the spinor vR for the antimuon and ūR for the
positron.) Apply the Fierz identity to show that the amplitude can be evaluated directly in terms of spinor
products. Square the amplitude and reproduce the result for

(e− e+ → µ− +
R µL ) (8.1.11)
d cos θ R L

given in Eqn. 5.22 p.143. Compute the other helicity cross sections for this process and show that they also
reproduce the results found in Section 5.2.
SOLUTION: (Thanks to Tess, Richard, Ken and Edgar (’12) for parts (a), (b), (c) and (d) respectively.)

(a) Define the left- and right-projection operators

1 − γ5 1 + γ5
PL = , PR = . (8.1.12)
2 2

From the definitions,


s(p1 , p2 ) 2 =
1
4(p1 ·k0 )(p2 ·k0 ) ūL0 p
/1 p
/2 uR0 ūR0 p
/2 p
/1 uL0
1

= 4(p1 ·k0 )(p2 ·k0 ) tr uL0 ūL0 p/1 p
/2 uR0 ūR0 p
/2 p
/1
1

= 4(p1 ·k0 )(p2 ·k0 ) tr PL k/0 p /2 PR k/0 p
/1 p /2 p
/1 . (8.1.13)

Since γ 5 anticommutes with γ µ , each time we pass γ 5 past anything that is slashed, it picks up a minus sign.
This means that PL turns into PR and vice versa when they are moved past slashed objects. Thus,
s(p1 , p2 ) 2 = 2
1
 1

/ /1 p
4(p1 ·k0 )(p2 ·k0 ) tr k 0 p /2 k/0 p /1 k/0 p
/1 = 4(p1 ·k0 )(p2 ·k0 ) tr p
/ 2 PR p /2 k/0 p
/1 p / 2 PR , (8.1.14)

where the last step uses the cylic property of the trace and the fact that PR is a projection operator and is
therefore idempotent: PR2 = PR .
Slashed objects inherit a contracted version of the Clifford algebra:

{p
/, k/} = 2p · k. (8.1.15)

2
/ = p2 . The momenta involved in this problem are all light-like, or null, and thus
A special case of this is p
2 2
all satisfy p
/ = p = 0. Therefore, we can write
2
/ = 2(p · k)p
/k/p
p / − k/p
/ = 2(p · k)p
/. (8.1.16)

Therefore, we can write Eqn. (8.1.14) as


 
tr 2(p1 · k0 )p
/1 2(p2 · k0 )p
/ 2 PR
s(p1 , p2 ) 2 =

= tr p
/1 p
/ 2 PR . (8.1.17)
4(p1 · k0 )(p2 · k0 )

Eqn. 5.5 P&S p.134, namely, tr(γ µ γ ν ) = 4g µν and tr(γ µ γ ν γ 5 ) = 0, imply tr γ µ γ ν PR = 2g µν . Thus, we
derive the desired result:
s(p1 , p2 ) 2 = 2p1 · p2 .

(8.1.18)
8.1. MASSLESS TREE DIAGRAMS, P&S 5.3 P.170 81

(b) Let us follow the hint on P&S p.135. Define C = γ 0 γ 2 and notice that C 2 = γ 0 γ 2 γ 0 γ 2 = −(γ 0 )2 (γ 2 )2 =
−1(−1) = 1. By explicit calculation, we find that

Cγ µ C = −γ µT , Cγ 5 C = γ 5T . (8.1.19)

Let n be the number of γ’s in the trace that are NOT γ 5 . Then,

tr[γ µ γ ν γ ρ · · · ] = tr[Cγ µ CCγ ν CCγ ρ C · · · ]


= (−1)n tr γ µT γ νT γ ρT · · · ]


= (−1)n tr (· · · γ ρ γ ν γ µ )T
 

= (−1)n tr[· · · γ ρ γ ν γ µ ]. (8.1.20)

The last equality follows from the fact that the trace is invariant under transposition. As stated on P&S
p.134, the trace of γ 5 times an odd number of γ µ ’s vanishes. Also, the trace of an odd number of γ µ ’s
vanishes. Therefore, n must be even. This continues to hold even if there is more than one γ 5 in the product
because we can simply move the γ 5 ’s until they are all side-by-side, at which point, if there is an even number
of them, their product is identity, and if there is an odd number of them, their product is γ 5 . Both cases,
still require n to be even. This derives the desired result,

tr[γ µ γ ν γ ρ · · · ] = tr[· · · γ ρ γ ν γ µ ] . (8.1.21)

Let us write out the desired products:


ūR0 p
/1 µ p / uR0 /1 γ µ p
tr[uR0 ūR0 p /2 ] /1 γ µ p
tr[PR k/0 p /2 ]
ūL (p1 )γ µ uL (p2 ) = √ γ √2 = p = p , (8.1.22a)
2p1 · k0 2p2 · k0 2 (p1 · k0 )(p2 · k0 ) 2 (p1 · k0 )(p2 · k0 )
ūL0 p
/2 µ p / uL0 / γµp
tr[p /1 uL0 ūL0 ] / γµp
tr[p /1 PL k/0 ]
ūR (p2 )γ µ uR (p1 ) = √ γ √1 = p 2 = p 2 . (8.1.22b)
2p2 · k0 2p1 · k0 2 (p1 · k0 )(p2 · k0 ) 2 (p1 · k0 )(p2 · k0 )

We chose to combine the uL0 ’s at the back of the trace rather than the front for convenience. In Eqn.
(8.1.22a), move PR past k/0 , which turns it into a PL . Then, use Eqn. (8.1.21) to derive the equality
µ µ µ
tr[PR k/0 p
/1 γ p/2 ] = tr[/
k 0 PL p
/1 γ p/2 ] = tr[p
/2 γ p/1 PL k/0 ]. (8.1.23)

This establishes the desired identity,

ūL (p1 )γ µ uL (p2 ) = ūR (p2 )γ µ uR (p1 ) . (8.1.24)

(c) As in the hint, define


M = 2[uL (p2 )ūL (p1 ) + uR (p1 )ūR (p2 )]. (8.1.25)

This is a product of Dirac matrices and is thus a Dirac matrix itself. Therefore, we can expand it in terms
of the 16 Γ’s (P&S p.49). Since there is an odd number of slashed quantities in each term of M (i.e. p/1 , p
/2
and k/0 ),
γ 5 M = −M γ 5 . (8.1.26)

The only Γ’s that anticommute with γ 5 are γ µ and γ µ γ 5 . Therefore,

M = γ µ Aµ + γ µ γ 5 B µ , (8.1.27)

for some sets of constants, Aµ and B µ . Instead, define

V µ = Aµ + B µ , W µ = Aµ − B µ , (8.1.28)

in terms of which, M is written


 1 − γ5   1 + γ5 
M= γµ V µ + γ µ W µ = PL γ µ V µ + PR γ µ W µ . (8.1.29)
2 2
82 CHAPTER 8. PROBLEM SET 8

Since PL PR = 0, PL2 = PL , and PR2 ,

γ µ P L M = γ µ PL γ ν V ν , γ µ P R M = γ µ PR γ ν W ν . (8.1.30)

We take the  trace of both sides of these equations. Using the cyclic property of the trace and the results
tr γ µ γ ν PR = tr γ µ γ ν PL = 2g µν found immediately before Eqn. (8.1.18), we find

Vµ = 1
tr γ µ PL M , Wµ = 1
tr γ µ PR M .
 
2 2 (8.1.31)

The corresponding results for Aµ and B µ are

Aµ = 12 (V µ + W µ ) = 41 [γ µ (PL + PR )M ] = 1
4 tr(γ µ M ), (8.1.32a)
µ 1 µ µ 1 µ µ 5
B = 2 (V −W )= 4 [γ (PL − PR )M ] = − 14 tr(γ γ M ). (8.1.32b)

We first compute the trace in B µ by anticommuting γ µ past γ 5 , then moving M to the front of the trace by
the cyclic property, and then using the result Eqn. (8.1.21) to reverse the order of multiplication:

− tr(γ µ γ 5 M ) = tr(γ 5 γ µ M ) = tr(M γ 5 γ µ ) = tr(γ µ γ 5 M ). (8.1.33)

Since this quantity is equal to its negative, it must vanish. Therefore,

B µ = 0. (8.1.34)

Now, let us compute the trace in Aµ :

tr[γ µ M ] = √ 1
tr[γ µ (p
/2 PR k/0 p /1 PL k/0 p
/1 + p /2 )]. (8.1.35)
(p1 ·k0 )(p2 ·k0 )

Let us work on the first term


tr[γ µ p
/2 PR k/0 p
/1 ] /1 γ µ ]
/ PR k/0 p
tr[p tr[p /1 γ µ ]
/ uR0 ūR0 p
p =p 2 =p 2 = 2ūL (p1 )γ µ uL (p2 ). (8.1.36)
(p1 · k0 )(p2 · k0 ) (p1 · k0 )(p2 · k0 ) (p1 · k0 )(p2 · k0 )

The second term is identical:

tr[γ µ p
/1 PL k/0 p
µ
/2 ] = tr[γ p/1 k/0 PR p /2 PR k/0 p
/2 ] = tr[p
µ
/1 γ ]. (8.1.37)

Therefore,
Aµ = 1
4 tr(γ µ M ) = ūL (p1 )γ µ uL (p2 ). (8.1.38)

Finally,
2[uL (p2 )ūL (p1 ) + uR (p1 )ūR (p2 )] = M = Aµ γµ = ūL (p1 )γ µ uL (p2 )γµ . (8.1.39)

(d) For e− e+ → µ− µ+ scattering, there is only the s-channel diagram since we do not have e− µ− γ coupling.
Assign momenta as in the diagram below.

µ− µ+
HH 
H 
k HH k 0
Y *

(8.1.40)

p HH p0
* H
YH

e−
H
e+
8.1. MASSLESS TREE DIAGRAMS, P&S 5.3 P.170 83

Now, consider e− + − +
R eL → µR µL scattering. A subtlety here is that for antiparticles, ūL is right-chirality and
vice versa! Then, the amplitude is given by
e2 0 0
MRL→RL = µ
s ūR (k)γ uR (k )ūR (p )γµ uR (p)
4πα 0 µ 0
= s ūL (k )γ uL (k)[γµ ]ab [ūR (p )]a [uR (p)]b
8πα 0 0 0
= s [uL (k)ūL (k ) + uR (k )ūR (k)]ab [ūR (p )]a [uR (p)]b
8πα 0 0 0
 
= s ūR (p ) uL (k)ūL (k ) + uR (k )ūR (k) uR (p) (8.1.41)

We used Eqn. (8.1.24) to get the second line and the Fierz identity in the form Eqn. (8.1.39) to get the
third.
The second term in Eqn. (8.1.41) vanishes because it involves two factors of PL separated by an odd
number of slashed objects:
0 0
ūR (p0 )uR (k 0 )ūR (k)uR (p) ∝ tr[p
/PL k/0 p
0
/ k/ PL k/0 k/] = tr[p
/k/0 p
0
/ k/ PR PL k/0 k/] = 0. (8.1.42)

The first term in Eqn. (8.1.41) can be simplified using s and t defined in Eqn. (8.1.3):
0 0 0 0 ∗
MRL→RL = 8πα
s s(p , k)t(k , p) = 8πα
s s(p , k)s(p, k ) . (8.1.43)

Using Eqn. (8.1.18) for the square of s, we get


RL→RL 2 8πα 2
2(p0 · k)2(p · k 0 ).

M = (8.1.44)
s

Since we neglect mass, p2 = p02 = k 2 = k 02 = 0. Thus, the Mandelstam variables are given by

s = (p + p0 )2 = (k + k 0 )2 = 2p · p0 = 2k · k 0 , (8.1.45a)
2 0 2 0 0
t = (k − p) = (p − k) = −2p · k = −2p · k , (8.1.45b)
0 2 0 2 0 0
u = (k − p) = (p − k) = −2p · k = −2p · k, (8.1.45c)

where we also used conservation of momentum, p + p0 = k + k 0 .


In the center of momentum frame, let θ be the angle between the muons and the electrons. Then,
Ecm Ecm
p= 2 (1, 0, 0, 1), k= 2 (1, 0, sin θ, cos θ), (8.1.46a)
0 Ecm 0 Ecm
p = 2 (1, 0, 0, −1), k = 2 (1, 0, − sin θ, − cos θ). (8.1.46b)

Therefore,
Ecm 2 2

s=2 2 [(1)(1) − (1)(−1)] = Ecm , (8.1.47a)
2
−2 Ecm [(1)(1) − (1)(cos θ)] = − 2s (1 − cos θ),

t= 2 (8.1.47b)
2
−2 Ecm [(1)(1) − (1)(− cos θ)] = − 2s (1 + cos θ).

u= 2 (8.1.47c)

Thus, we can write Eqn. (8.1.44) as


RL→RL 2 8παu 2
= [4πα(1 + cos θ)]2 .

M = (8.1.48)
s

Therefore, the differential cross-section is


 dσ RL→RL |MRL→RL |2 α2
= 2
= (1 + cos θ)2 . (8.1.49)
dΩ 64π s 4s

Indeed, this agrees with Eqn. 5.22 P&S p.143.


Consider the LR → LR case. This simply changes all the R subscripts to L subscripts in the first line of
Eqn. (8.1.41). Using Eqn. (8.1.24), we can change these back to R’s while also switching the momenta in
84 CHAPTER 8. PROBLEM SET 8

each coupled pair. Thus, MLR→LR = MRL→RL except with p0 and k switched in the first s and p and k 0
switched in the second. However, using Eqn. (8.1.4), we can switch them back, incurring two minus signs,
which cancel. Therefore, the cross-section is the same:
 dσ LR→LR α2
= (1 + cos θ)2 . (8.1.50)
dΩ 4s

Consider the RL → LR case. As just described, the only difference in this case is that
RL→LR 2
= 8πα 2 |s(p0 , k 0 )|2 |s(p, k)|2 = 8πα 2 2(p0 · k 0 )2(p · k) = 8παt 2 = [4πα(1 − cos θ)]2 . (8.1.51)
  
M
s s s

Therefore,
 dσ RL→LR  dσ LR→RL α2
= = (1 − cos θ)2 . (8.1.52)
dΩ dΩ 4s

dσ dσ
This agrees with Eqn. 5.23 P&S p.143. Finally, d cos θ = 2π dΩ .

8.2 Physical Amplitude, Zee III.1.1 p.168


Work through the manipulations leading to Eqn. 9 without referring to the text:
s0 t0 u0
M = −iλP + iCλ2P log + O(λ3P ). (8.2.1)
stu
SOLUTION:

We have written down the the one-loop fully connected diagram at order λ2 before (problem set 5). The result
in the s-channel is
(−iλ)2 d4 k
Z
i i
Ms = , (8.2.2)
2 (2π)4 k 2 − m2 + i (K − k)2 − m2 + i
where K is the total momentum, that is, K 2 = s.
We write
Z 1
1 1 dx
=
k 2 − m2 + i (K − k)2 − m2 + i 0 {x(k 2 − m2 + i) + (1 − x)[(K − k)2 − m2 + i]}2
Z 1
dx
= 2 2 2 2 2 2 2 2
0 [k − 2(1 − x)k · K + (1 − x) K − (1 − x) K + (1 − x)K − m + i]
Z 1
dx
= 2 2 2 2
. (8.2.3)
0 {[k − (1 − x)K] + x(1 − x)K − m + i}

Define

` = k − (1 − x)K, ∆ = m2 − x(1 − x)K 2 , D = `2 − ∆ + i. (8.2.4)

Then, we write the amplitude as


1 1 ∞
λ2 d4 ` 1 λ2 `3 d`
Z Z Z Z
Ms = dx = dx . (8.2.5)
2 0 (2π)4 D2 16π 2 0 0 (`2 − ∆)2
By integration by parts, we find
Z ∞ ∞ Z ∞ ` d`
`3 d` `2 1 1 ∞
= − + = − + log(`2 − ∆) . (8.2.6)

(`2 − ∆)2 2(`2 − ∆) 0 `2 − ∆ 2 2 0
0 0

Unfortunately, the final integral is divergent. Let us regularize it by imposing a cut-off, Λ:


Z Λ
`3 d` 1 1 Λ2 − ∆
2 2
= − + log . (8.2.7)
0 (` − ∆) 2 2 −∆
8.3. SLIDING CUT-OFF, ZEE III.1.3 P.168 85

In the limit Λ → ∞ and s = K 2 >> m2 , we have ∆ ≈ −x(1 − x)s and Λ2 >> ∆. Thus,
Z ∞
`3 d` 1 1 Λ2 1 Λ2 
≈ − + log ≈ log − 1 − log x − log(1 − x) . (8.2.8)
0 (`2 − ∆)2 2 2 x(1 − x)s 2 s
Integration over x yields
λ2  Λ2  λ2 Λ2
Ms = 2
log −3 ≈ 2
log . (8.2.9)
32π s 32π s
2
The most important part is Ms ∼ λ2 log Λs since it is the divergent piece. We could also extract this divergent
piece by observing that the divergence is a UV divergence and thus is dominated by the large k region of the
integrand in Eqn. (8.2.2), that is, k 2 >> K 2 , m2 . This justifies dropping K and m from the integrand and
restricting the integral to the region k >> K. To do this, let the lower bound of the integral be ξK, where ξ is
some sufficiently large, but finite number. Then,
Z Λ 3
(−iλ)2 1 2 k dk Λ Λ2
Ms ≈ 2π ∼ log − log
ξ ∼ log . (8.2.10)
2 (2π)4 ξK k
4 K  s

Note: Even though we introduced the large number ξ to parametrize the fact that lower bound of the integral
should be much greater than K, it ends up not mattering because it does not contribute to the divergence (i.e. is
subleading in the Λ → ∞ limit). Also, we can square the argument of the log because that just incurs an overall
factor of 2 and we are not keeping track of the exact proportionality constant anyway.
To get the t- and u-channel results, we simply change s to t and u. This derives the result
Λ6
M = −iλ + iCλ2 log . (8.2.11)
stu
At (s, t, u) = (s0 , t0 , u0 ), the physical coupling constant is λp = λp (s0 , t0 , u0 ), and by definition, M(s0 , t0 , u0 ) =
−iλp . Hence,
Λ6
λp = λ − Cλ2 log . (8.2.12)
s0 t0 u0
We can invert this to order λ3p :
Λ6
λ = λp + Cλ2p log . (8.2.13)
s0 t0 u0
Therefore,
Λ6 Λ6 s0 t0 u0
M = −iλp − iCλ2p log + iCλ2p log = −iλp + iCλ2p log . (8.2.14)
s0 t0 u0 stu stu

8.3 Sliding Cut-off, Zee III.1.3 p.168


Change Λ to e Λ. Show that for M not to change to the order indicated λ must change by δλ = 6Cλ2 + O(λ3 ),
that is,

Λ = 6Cλ2 + O(λ3 ). (8.3.1)

SOLUTION:

With Λ → e Λ and λ → λ + δλ,


 Λ6 
M → M + iδλ 2Cλ log + 12Cλ − 1 + 6iCλ2 . (8.3.2)
stu
At this order, there is nothing that can cancel the extra log term if it indeed exists to order λ2 . The only way to
get rid of it is to force it to be of higher order by having δλ = Aλ2 + O(λ3 ). This gives

M → M + (6C − A)iλ2 + O(λ3 ). (8.3.3)

Therefore, in order that M not change to this order, we must have A = 6C. Thus,

δλ = 6Cλ2 + O(λ3 ) . (8.3.4)


86 CHAPTER 8. PROBLEM SET 8

Infinitesimally, Λ → e Λ is δΛ = Λ. Thus,

dλ 6Cλ2
Λ =Λ = 6Cλ2 + O(λ3 ) . (8.3.5)
dΛ Λ

8.4 Discussion: Naturalness and Renormalizability


We can consider the question of the naturalness of the smallness of a parameter (e.g. mass, or coupling constant)
by looking at the possible Feynman diagrams and their degrees of divergence. For example, in lecture, we
considered the Yukawa theory coupling one fermion and one real scalar:

L = 12 ∂µ φ ∂ µ φ − 12 µ2 φ2 − λ 4
4! φ + ψ(i∂/ − m)ψ − f φψψ. (8.4.1)

Since the λ → 0 limit alone is not accompanied by increased symmetry, ’t Hooft’s naturalness “theorem” says
that we should not artificially get rid of the φ4 term. How can we see this at the level of diagrams? The answer is
that, even if we were to get rid of the φ4 interaction, an effective φ4 interaction will be produced by the Yukawa
coupling via a square of virtual fermions:

p4 p3
I 

q ? 6 (8.4.2)
-

p1
 p2I

The superficial degree of divergence of this diagram is a log divergence:

D = 4 − BE − 32 FE = 0. (8.4.3)

Indeed, the amplitude is, up to constants,

d4 q
Z
4 1 1 1 1
M∼f · · · , (8.4.4)
(2π)4 /q − m /q + p
/1 − m q + p +
/ /1 /2 p − m / /1 /2 − p
q + p + p /3 − m

which is indeed log divergent. In fact,


Λ2
M ∼ f 4 log max(s,t,u) . (8.4.5)

On the other hand, a λφ4 vertex would have M ∼ λ. So, we see that the Yukawa diagram above effectively
contributes an infinite φ4 term proportional to f 4 . In order to renormalize the theory, we would have to add a
divergent φ4 counterterm to cancel most of the divergence in Eqn. (8.4.5) leaving only a finite piece that gives the
correct value for the coupling constant at some particular experimentally verified energy scale. More importantly,
the counterterm would have to be proportional to f 4 and not λ. If it had been the latter scenario, then λ = 0
would be a fixed point, whose renormalization vanishes since it is proportional to λ, which is set to zero. This is
why it is unnatural to set λ = 0: even if we did, the Yukawa interaction would produce an effective φ4 diagram
that is divergent and not proportional to λ. Of course, it would be trivially natural to set both λ and f to zero!
The Yukawa interaction can also produce φ2n diagrams via a 2n-gon of virtual fermions for n = 1, 2, . . .. The
n = 1 case (two external bosons) contributes a quadratically divergent piece to the boson mass. Therefore, it is
unnatural to set µ = 0. The n = 2 case is the previously-discussed λφ4 case. For n ≥ 3, the diagram has BE ≥ 6,
which means that the diagrams are not divergent and require no renormalization. Hence, it is natural to drop
φ6 , φ8 , and so on.
Now, let’s consider effective vertices coupling 2n fermions. These can be produced by essentially the same
types of diagrams as above, but where the external lines are now fermionic and the internal lines alternate between
fermionic and bosonic types. By Eqn. (8.4.3), the only divergent diagram is the n = 1 case, which contributes
to the fermion mass term. As found in lecture, this gives a renormalization of m that is proportional to m itself.
Therefore, m = 0 is natural. As discussed in lecture, the associated extra symmetry is separate conservation of
left- and right-chirality fermions.
8.4. DISCUSSION: NATURALNESS AND RENORMALIZABILITY 87

You can keep playing this game, determining all of the types of effective vertices you can produce that mix φ
and ψ. However, all of them, except for the Yukawa term are convergent and require no renormalization. Hence,
this theory requires only a finite number of divergent counterterms and is therefore renormalizable.
Contrast this with the case of the four-Fermi theory: Lint = −GF (ψψ)2 . We can ask the question, is it
natural to drop the (ψψ)n term for n ≥ 3? Below, on the left, is a diagram involving only quartic interactions
that nevertheless produces an effective sextic interaction, proportional to G2F . By the superficial degree of
divergence formula, D = 4 − 23 FE + 2V , this diagram is not divergent and so we might rejoice in thinking that
it is natural to drop the sextic interaction term. However, since we can add any number of vertices without
changing the number of external lines, it turns out that we can make arbitrarily divergent diagrams that would
contribute to an effective sextic interaction, such as the one on the right below.

@ @ 
@ @
@ @ (8.4.6)
@ 
@
@ @
@ @
Therefore, it is not natural to set the sextic term to zero. Indeed, we pick up no extra symmetry from doing
so. We would need a divergent sextic counterterm in order to renormalize the theory. The problem is, we would
need a counterterm for (ψψ)n for all n = 1, 2, . . ., and therefore the theory is not renormalizable.
However, we can think of the theory as an effective theory, which is only sensible at low energies. The (ψψ)n
counterterm must be proportional to Gn−1
F . Note that GF has mass dimensions −2 in 4D and so the counterterms
have coupling constants with dimension −2(n − 1) = 2 − 2n. For n ≥ 3, this is quite a large negative dimension.
Therefore, if we set GF ∼ M −2 , where M is the scale at which the theory breaks down, then the contributions
of such vertices would be suppressed by factors of (µ/M )n−1 , where µ is the scale at which we are observing the
system (the experimental scale). As long as µ << M , the theory should be well-behaved.
88 CHAPTER 8. PROBLEM SET 8
Chapter 9

Problem Set 9

9.1 Fermion Field Dimension, Zee III.3.1 p.181


Show that in (1+ 1)-dimensional spacetime the Dirac field ψ has mass dimension 12 , and hence the Fermi coupling
is dimensionless.

SOLUTION: (Thanks to Tova (’12) for presenting her solution.)

The dimensional analysis determining the dimension of ψ proceeds as follows:

0 = [d2 x ψi∂ψ]
/ = 2[x] + [∂] + 2[ψ] = −2 + 1 + 2[ψ] =⇒ [ψ] = 1
2 . (9.1.1)

The dimensional analysis determining the dimension of GF proceeds as follows:

0 = [d2 x GF (ψ̄ψ)2 ] = 2[x] + 4[ψ] + [GF ] = −2 + 4 12 + [GF ]



=⇒ [GF ] = 0 . (9.1.2)

9.2 Degree of Divergence, Zee III.3.2 p.181


Derive Eqn. 11 p.178 and Eqn. 13 p.179. The first says that the degree of divergence in Yukawa theory in 3 + 1
dimensions is D = 4 − BE − 23 FE , where BE and FE are the numbers of external bosonic and fermionic legs, re-
spectively. The second says that the degree of divergence in Four-Fermi theory in 1+1 dimensions is D = 2− 12 FE .

SOLUTION: (Thanks to Diana and Shreyas (’12) for presenting their solutions.)

Yukawa: Let L be the total number of loops, Vf and Vλ be the numbers of f -vertices and λ-vertices, respectively,
V ≡ Vλ + Vf be the total number of vertices, and BE , BI , FE and FI be the numbers of external and internal
bosonic and fermionic lines, respectively. The superficial degree of divergence is

D = 4L − 2BI − FI , (9.2.1)

since each loop contributes d4 q for some unspecifiied momentum, q, each internal bosonic line contributes
R

a factor of the bosonic propagator ∼ q −2 , and each internal fermionic contributes a factor of the fermionic
propagator ∼ /qq −2 ∼ q −1 .
The number of loops is equal to the number of internal lines minus the number of internal lines whose
momenta are fixed by momentum conservation. The latter are due to momentum-conserving delta functions at
every vertex. However, some combination of these delta functions just gives an overall momentum-conserving
delta function, which does not constrain any of the internal momenta. Hence, the number of internal momenta
that are constrained by momentum conservation is V − 1. Hence,

L = BI + FI − (V − 1) =⇒ D = 4 − 4V + 2BI + 3FI . (9.2.2)

Conservation of bosonic and fermionic ends yields

Vf + 4Vλ = BE + 2BI , 2Vf = FE + 2FI . (9.2.3)

89
90 CHAPTER 9. PROBLEM SET 9

The first follows from the fact that each external bosonic line has only one of its endpoints attached to a vertex
while both ends of an internal line must be attached, and each f -vertex accommodates only one bosonic end
while each λ-vertex accommodates four. The same argument pertains to the second equation for fermionic ends.
These two equations combine to the following useful form:

4V = 4(Vλ + Vf ) = BE + 2BI + 23 FE + 3FI . (9.2.4)

Putting this all together, we get


D = 4 − BE − 32 FE . (9.2.5)

4-Fermi: In 1 + 1 dimensions, each loop contributes two powers of momentum while each fermion propagator
still contributes one inverse power of momentum. Therefore,

D = 2L − FI . (9.2.6)

The same argument that gave Eqn. (9.2.2) gives

L = FI − (V − 1) =⇒ D = 2 − 2V + FI . (9.2.7)

Conservation of ends reads


4V = FE + 2FI . (9.2.8)
Combining this with Eqn. (9.2.7) yields
D = 2 − 21 FE . (9.2.9)

9.3 Massless Weisskopf Phenomenon, Zee III.3.3 p.181


Show that B(p2 ) in Eqn. 14 p.180, reproduced below, vanishes when we set m = 0. Show that the same behavior
holds in quantum electrodynamics.

d4 k / + k/ + m
p
Z
1
(if )2 i2 ≡ A(p2 )p 2
/ + B(p ). (9.3.1)
(2π) k − µ (p + k)2 − m2
4 2 2

SOLUTION: (Thanks to Jonathan (’12) for presenting his solution.)

The term in the left hand side that does not involve gamma matrices (i.e. slashed objects) is proportional to m.
Hence, the same must be true for the right hand side. Hence,

m→0
B(p2 ) ∝ m −−−→ 0 . (9.3.2)

If the virtual scalar is replaced by a virtual photon (of fiducial mass µ), then the amplitude becomes

d4 k −gµν / + k/ + m
p
Z
(ie)2 i2 4 2 2
ū(p)γ ν γ µ u(p). (9.3.3)
(2π) k − µ (p + k)2 − m2
kµ kν
Recall that we can safely drop the piece of the photon propagator involving µ2 by the appropriate Ward
identity, which encodes gauge invariance. Let us simplify the numerator:

−gµν γ ν (p µ ν µ µ
/ + k/ + m)γ = −gµν γ [2(p + k) − γ (p
/ + k/ − m)]
= −2(p / + k/ − m)
/ + k/) + 4(p
/ + k/ − 2m).
= 2(p (9.3.4)

Here, we used the identity γµ γ µ = 4. Therefore, up to numerical factors, Eqn. (9.3.3) is the same as Eqn. (9.3.1).
By the way, you can actually keep the massive term in the photon propagator. It can only possibly contribute
to A and not to B.
9.3. MASSLESS WEISSKOPF PHENOMENON, ZEE III.3.3 P.181 91

Aside: It might seem peculiar that the integral involving k/ end up being proportional to p /. However, it must
do because it must be proportional to a relativistic-invariant matrix and since it contains γ matrices, the only
possibility is p
/. This is a non-constructive proof of this fact.
We can also prove it by brute force calculation using the Feynman trick. Define the integral

d4 k k/
Z
I= . (9.3.5)
(2π)4 (k 2 − µ2 )[(p + k)2 − m2 ]

Then, using the Feynman trick, we have


Z 1 Z
d4 k k/
I= dx 4  2
0 (2π) x[(p + k) − m ] + (1 − x)(k 2 − µ2 )
2 2
Z 1 Z
d4 k k/
= dx
(2π) k + 2xpk + x p − x p + xp2 − xm2 − (1 − x)µ2 2
4  
0 2 2 2 2 2
Z 1 Z
d4 k k/
= dx  . (9.3.6)
(2π) (k + xp)2 + x(1 − x)p2 − xm2 − (1 − x)µ2 2
4 
0

Define
`µ = k µ + xpµ , (9.3.7)
in terms of which, we may write the integral as
1
d4 ` /̀ − xp
Z Z
/
I= dx , (9.3.8)
0 (2π)4 D2

where
D = `2 − ∆, (9.3.9)
and
∆ = xm2 + (1 − x)µ2 − x(1 − x)p2 . (9.3.10)
µ
The integral of the /̀ term vanishes because it is the integral of an odd function of ` . The only term left is
indeed proportional to p/ and is a function only of p2 (since ∆ is a function only of p2 ).

Another aside: It was suggested that we write the intermediate fermion propagator in the m = 0 limit as

/ + k/
p 1 / − k/ ? p
p / − k/
2
= · = 2 . (9.3.11)
(p + k) / + k/
p / − k/
p p − k2

The reason for the question mark above the final equality is that it is not actually correct. In fact,
2 2 2 2 2 2
/ − k/) = p
/ + k/)(p
(p / − k/ + k/p
/−p
/k/ = p − k + [/ /] 6= p − k .
k, p (9.3.12)

This spoils our ability to use this method to massage the denominator into a form that depends only on p2 and
k 2 . The point of the p2 is that the result is supposed to be a function only of p2 and the point of k 2 is that it is
even and so if it is multiplied by an odd function of k, such as k/ in the numerator, then that integral vanishes
identically. Even though this doesn’t work, the idea is spot on. The method of Feynman parameters achieves
this goal.
92 CHAPTER 9. PROBLEM SET 9

9.4 Form of Anomalous Magnetic Moment, Zee III.6.3 p.198


By Lorentz invariance, the right hand side of Eqn. 7 p.196, reproduced below, has to be a vector. The only
possibilities are ūγ µ u, (p + p0 )µ ūu, and (p − p0 )µ ūu. The last term is ruled out because it would not be consistent
with current conservation. Show that the form given below is in fact the most general allowed.
iσ µν qν
 
0 0 µ 0 0 µ 2 2
hp , s |J (0)|p, si = ū(p , s ) γ F1 (q ) + F2 (q ) u(p, s), (9.4.1)
2m
where q ≡ p0 − p.

SOLUTION:

Current conservation reads qµ J µ = 0, which means that the only terms that can show up on the right hand side
of Eqn. (9.4.1) are vectors with vanishing dot product with q. Since q 2 6= 0 for the virtual photon, we cannot
have a term proportional to ūq µ u. On the other hand, q · (p + p0 ) = p2 − p02 = m2 − m2 = 0 and ūqµ γ µ u = 0 since
0 0
/u(p) = mu(p) and ū(p0 )p
p / = mū(p0 ) and so ū/qu = ū(p / −p /)u = ū(m − m)u = 0. Therefore, current conservation
cannot get rid of the other two terms. Thus,

hp0 , s0 |J µ (0)|p, si = ū(p0 , s0 ) γ µ A + (p + p0 )µ B u(p, s),


 
(9.4.2)
0
where, A and B are Lorentz scalars, but could be proportional to slashed momenta, p / and p
/ . Since the latter are
0
equivalent to m when acting on u(p) or ū(p ), we may assume A and B to just be functions and be proportional
to the identity in spinor space. A and B must be functions of Lorentz invariants (and constants, of course).
Certainly, q 2 is an invariant, and since p2 = p02 = m2 and 2p · p0 = 2m2 − q 2 , there is no other independent choice
for a variable on which A and B may depend. Hence,

hp0 , s0 |J µ (0)|p, si = ū(p0 , s0 ) γ µ A(q 2 ) + (p + p0 )µ B(q 2 ) u(p, s).


 
(9.4.3)

Finally, recall the Gordon identity from problem set 3:


 0µ
p + pµ iσ µν qν

0 µ 0
ū(p )γ u(p) = ū(p ) + u(p). (9.4.4)
2m 2m
Then, define

F1 (q 2 ) = A(q 2 ) + 2mB(q 2 ), F2 (q 2 ) = −2mB(q 2 ), (9.4.5)

in terms of which, Eqn. (9.4.3) may be written as

iσ µν qν
 
0 0 µ 0 0 µ 2 2
hp , s |J (0)|p, si = ū(p , s ) γ F1 (q ) + F2 (q ) u(p, s) . (9.4.6)
2m

9.5 Discussion 1: Phi-Fourth Dimensional Analysis


Recall that a theory with a coupling constant of negative mass dimension is non-renormalizable (see Zee p.170
for a heuristic argument). Thus, we can find out in which dimensions φ4 theory is renormalizable. In d spacetime
dimensions, the Lagrangian density must have mass dimension [L] = d so that the action dd x L be dimensionless.
R

Since [∂] = +1,


d = 12 ∂µ φ∂ µ φ = 2[∂] + 2[φ] = 2 + 2[φ] [φ] = d−2
 
=⇒ 2 . (9.5.1)
Therefore, the dimension of λ is
 λ 4 d−2

d = − 4! φ = [λ] + 4[φ] = [λ] + 4 2 =⇒ [λ] = 4 − d. (9.5.2)

Therefore, φ4 theory is renormalizable for d ≤ 4 and non-renormalizable for d > 4.


We know another method for determining renormalizability: calculating the superficial degree of divergence.
Recall that in d = 4, for φ4 theory, this is given by D = 4 − BE . The only difference in the equations used to
derive this is in the number of momenta contributed by each loop: it is now d instead of 4. Thus,

D = dL − 2BI , L = BI − (V − 1), 4V = 2BI + BE . (9.5.3)


9.6. DISCUSSION 2: PHI-CUBED DIMENSIONAL ANALYSIS 93

Together, these give


D = d − d4 BE + 2 d

4 − 1 BI . (9.5.4)
This indeed gives D = 4 − BE when d = 4. Note that the BI term exactly vanishes in this dimension. This is nice
because it means that we can enumerate the divergent diagrams by the number of external lines and only finitely
many cases are divergent and require renormalization. Hence, the theory is renormalizable. This continues to
hold when the coefficient of BI is negative since adding more and more internal lines just makes diagrams more
and more convergent. On the other hand, if the coefficient of BI is positive, then we can make diagrams with
any number of external lines arbitrarily divergent by adding sufficiently many internal lines. In this case, the
theory is non-renormalizable. Note that this exactly reproduces the result from dimensional analysis because the
coefficient of BI is zero or negative exactly when d ≤ 4 and it is positive when d > 4.
In principle, the secret agent behind all of this is Weinberg’s theorem, which has to do with the behavior of
correlators under dilations of momenta. You can look it up if you like, but we probably won’t get to it in class.

9.6 Discussion 2: Phi-Cubed Dimensional Analysis


λ 3
If we repeat the dimensional analysis for a theory with Lin = − 3! φ , instead, we find
d−2 6−d
[φ] = 2 , [λ] = 2 . (9.6.1)

Therefore, the theory is renormalizable for d ≤ 6 and non-renormalizable for d > 6.


We also find the superficial degree of divergence

D = d − d3 BE + 2 d6 − 1 BI .

(9.6.2)

As before, the coefficient of BI vanishes when d = 6, is negative when d < 6 and is positive when d > 6. This
confirms the fact that the theory is renormalizable when d ≤ 6 and non-renormalizable when d > 6.
In accordance with the discussion last week, we can see this break-down at the diagramatic level. For
simplicity, consider diagrams that are divergent at the one-loop level. Thus, L = 1 and the middle equation in
(9.5.3) implies that BI = V . The third equation is replaced with 3V = 2BI + BE and so this also implies that
BE = BI = V when L = 1. In this case, D simplifies to

D|L=1 = d − 2BE . (9.6.3)

In d = 2 dimensions, the only divergent diagram has BE = BI = V = 1. This diagram is the tadpole diagram:

(9.6.4)


This contributes a source term to φ and therefore a divergent constant vacuum expectation value, hφi. This is
like an infinite zero-point energy and is simply subtracted out since an overall constant value in φ is irrelevant.
Therefore, this is not actually of any interest.
In d = 4 dimensions, the tadpole diagram is still divergent, but now BE = BI = V = 2 is also divergent. This
diagram is

(9.6.5)

This gives the wavefunction renormalization (of φ) and the mass renormalization. Since these are already in the
action, the theory is certainly one-loop renormalizable.
In d = 6 dimensions, we get a new divergent diagram when BE = BI = V = 3:


(9.6.6)

Q
 Q

This renormalizes the vertex. That is, it renormalizes λ. Again, nothing other than the terms originally in the
Lagrangian get renormalized. Thus, the theory is one-loop renormalizable.
94 CHAPTER 9. PROBLEM SET 9

Finally, this must break down in d = 8 dimensions. BE = BI = V = 4 is divergent. This diagram is

@
@
(9.6.7)

@
@
This would renormalize a φ4 interaction, but there is no such thing in the original Lagrangian. Thus, it generates
an infinite φ4 interaction proportional to λ4 . This signals one-loop non-renormalizability. The φ4 interaction
together with the φ3 interaction will now generate further divergent one-loop diagrams involving higher and
higher numbers of external particles.
Renormalizability must break down when d = 7 as well. However, it must do so at the two-loop level.
Chapter 10

Problem Set 10

10.1 Exotic Contributions to g – 2, P&S 6.3 p.210


Any particle that couples to the electron can produce a correction to the electron-photon form factors and, in
particular, a correction to g − 2. Because the electron g − 2 agrees with QED to high accuracy, these corrections
allow us to constrain the properties of hypothetical new particles.
(a) The unified theory of weak and electromagnetic interactions contains a scalar particle h called the Higgs
boson, which couples to the electron according to
Z
λ
Hint = d3 x √ hψψ. (10.1.1)
2

Compute the contribution of a virtual Higgs boson to the electron g − 2, in terms of λ and the mass mh of
the Higgs boson.
g−2
(b) QED accounts extremely well for the electron’s anomalous magnetic moment. If a = 2 ,

|aexpt. − aQED | < 1 × 10−10 . (10.1.2)

What limits does this place on λ and mh ? In the simplest version of the electroweak theory, λ = 3 × 10−6
and mh > 60 GeV. Show that these values are not excluded. The coupling of the Higgs boson to the muon
is larger by a factor of (mµ /me ): λ = 6 × 10−4 . Thus, although our experimental knowledge of the muon
anomalous magnetic moment is not as precise,

|aexpt. − aQED | < 3 × 10−8 , (10.1.3)

one can still obtain a stronger limit on mh . Is it strong enough?


SOLUTION: (Thanks to Richard (’12) for presenting his solution.)

(a) The Higgs contribution to the electron g − 2 comes from the diagram
0
Jp
]J
J k+q
]
(10.1.4)
J
p−k 6 J 

q
k


p



The amplitude for the diagram (10.1.4) is

d4 k −iλ i(/
k + /q + m)
   
k + m) −iλ
Z
0 i µ i(/
iM = ū(p ) √ (−ieγ ) 2 √ u(p) · · · , (10.1.5)
(2π)4 2 (p − k)2 − m2h (k + q)2 − m2 k − m2 2

95
96 CHAPTER 10. PROBLEM SET 10

where the · · · involves the rest of the diagram to which the vertex is attached, including the photon propa-
gator, or the photon polarization state if it is external. The i is taken for granted.
Let us clean this up a little bit:

iλ2 d4 k ū(p0 )(/


k + /q + m)γ µ (/k + m)u(p)
Z
iM = (−ie) 2
2 (2π) [(p − k) − mh ][(k + q) − m2 ](k 2 − m2 )
4 2 2

d4 k ū(p0 )(/
k + /q + m)γ µ (/
k + m)u(p)
Z Z
= (−ie)iλ2 , (10.1.6)
(2π) x(k 2 − m2 ) + y[(k + q)2 − m2 ] + z[(p − k)2 − m2 ] 3
4 
F h

R
where F
stands for the integral over the Feynman parameters. In this case,
Z ZZZ 1
= dx dy dz δ(x + y + z − 1). (10.1.7)
F 0

Before we embark on complicated and tedious mathematics, let us pause for a moment to think about the
goal. We want the correction to g − 2, which depends only on the F2 (q 2 ) term. Thus, we can neglect all the
terms in Eqn. (10.1.7) that do not contribute to F2 (such as terms proportional to γ µ ).
Now, let us first work on the denominator:

D ≡ x(k 2 − m2 ) + y(k 2 + 2q · k + q 2 − m2 ) + z(k 2 − 2p · k + p2 − m2h )


= (x + y + z)k 2 + 2(yq − zp) · k + yq 2 + zp2 − (x + y)m2 − zm2h
= k 2 + 2(yq − zp) · k + (yq − zp)2 − (yq − zp)2 + yq 2 + zp2 − (x + y)m2 − zm2h
= (k + yq − zp)2 − (x + y)m2 + zm2h − y(1 − y)q 2 − z(1 − z)p2 − 2yzp · q .
 
(10.1.8)
| {z } | {z }
`2 ∆

Replace p02 and p2 with m2 . Then, note that m2 = p02 = (p + q)2 = m2 + 2p · q + q 2 . Thus, we may write

p2 = m2 , 2p · q = −q 2 . (10.1.9)

Using these identities, we simplify ∆ to

∆ = (x + y)m2 + zm2h + y(1 − y)q 2 − z(1 − z)m2 + yzq 2 . (10.1.10)

Since x + y + z = 1, we may write the coefficients of m2 and q 2 as

x + y − z(1 − z) = (1 − z) − z(1 − z) = (1 − z)2 , (10.1.11a)


−y(1 − y) + yz = −y(x + z) + yz = −xy. (10.1.11b)

Therefore, this leaves us with


∆ = (1 − z)2 m2 + zm2h − xyq 2 . (10.1.12)

This is exactly the same as Eqn. 6.44 P&S p.191 except for an extra factor of zm2h . Therefore, the same
argument that showed that Eqn. 6.44 P&S p.191 is positive, namely that q 2 < 0 for a scattering process,
implies that Eqn. (10.1.12) is positive as well.
We want to express the numerator of the integrand in Eqn. (10.1.6) in terms of `, which is related to k via

`µ = k µ + yq µ − zpµ . (10.1.13)

The numerator is

num = ū(p0 ) /̀ + z p
  µ 
/ + (1 − y)/q + m γ /̀ + z p/ − y /q + m u(p)
= ū(p0 ) /̀ + z − (1 − y) p 0 0
   µ 
/ + m γ /̀ + (y + z)p
/ + (1 − y)p / − yp
/ + m u(p)
= ū(p0 ) /̀ − xp 0
  µ 
/ + (2 − y)m γ /̀ − y p / + (2 − x)m u(p). (10.1.14)
10.1. EXOTIC CONTRIBUTIONS TO G – 2, P&S 6.3 P.210 97

0
We have turned any p/ on the right and p/ on the left into m because they act on u(p) and ū(p0 ), respectively.
We also used the identity x = 1 − y − z twice, to turn the coefficients of p
/ on the left and m on the right into
−x and (2 − x), respectively.
The following identities (Eqns. 6.45 and 6.46 P&S p.191) will help us simplify this:

d4 ` `µ d4 ` `µ `ν d4 ` 14 g µν `2
Z Z Z
= 0, = . (10.1.15)
(2π)4 D3 (2π)4 D3 (2π)4 D3

The first identity implies that we can discard all terms in the numerator that are linear in `. The second
identity implies that we can replace `µ `ν in the numerator with 14 `2 g µν . In the sequel, we will take the ū(p0 )
and u(p) factors for granted. The quadratic term in num is
2
num(2) = /̀γ µ /̀ = 2`µ /̀ − γ µ /̀ = 2 14 `2 g µν γν − γ µ `2 = − 21 `2 γ µ . (10.1.16)

As we mentioned earlier, we only care about terms that contribute to F2 . This quadratic piece is proportional
to γ µ and thus only contributes to F1 . We keep it here only for the sake of completeness.
The terms of zeroth order in ` are
0
num(0) = −xp
  µ 
/ + (2 − y)m γ −y p
/ + (2 − x)m
µ 0 µ µ 0 2 µ
= xy p / −x(2 − x)mp
/γ p /γ − y(2 − y)mγ p
/ +(2 − x)(2 − y)m γ . (10.1.17)
| {z } | {z }
(0) (0)
num3 num2

0
Now, let us massage this into a convenient form by moving p/ to the left and p
/ to the right and then replacing
0
each with m since they act on ū(p ) and u(p), respectively:
(0) 0µ 0 µ
num2 = −x(2 − x)m(2pµ − γ µ p
/) − y(2 − y)m(2p − p
/γ )
= −2x(2 − x)mpµ + x(2 − x)m2 γ µ − 2y(2 − y)mp0µ + y(2 − y)m2 γ µ . (10.1.18)

Define

q µ = (p0 − p)µ , q 0µ = (p0 + p)µ . (10.1.19)

Inverting this yields

p0µ = 21 (q 0 + q)µ , pµ = 21 (q 0 − q)µ . (10.1.20)

Write Eqn. (10.1.18) in terms of q 0 and q:


(0)
num2 = [x(2 − x) + y(2 − y)]m2 γ µ − x(2 − x)m(q 0 − q)µ − y(2 − y)m(q 0 + q)µ
= [x(2 − x) + y(2 − y)]m2 γ µ − [x(2 − x) + y(2 − y)]mq 0µ + [x(2 − x) − y(2 − y)]mq µ . (10.1.21)

(0)
Now, let us work on num3 :
(0) 0
num3 = xy(2pµ − γ µ p
/)p
/
= 2xympµ − xyγ µ (2p0 · p − p
/p
0
/)
= 2xympµ − 2xyp0 · pγ µ + xym(2p0µ − p0 µ
/γ )
= 2xympµ − 2xyp0 · pγ µ + 2xymp0µ − xym2 γ µ
= 2xymq 0µ − xy(2p0 · p + m2 )γ µ . (10.1.22)

Let us rewrite the dot product using the following identity:

q 2 = (p0 − p)2 = p02 + p2 − 2p0 · p = 2m2 − 2p0 · p. (10.1.23)


98 CHAPTER 10. PROBLEM SET 10

This gives
(0)
num3 = xy(q 2 − 3m2 )γ µ + 2xymq 0µ . (10.1.24)

Putting Eqns. (10.1.16) (10.1.17), (10.1.21) and (10.1.24) together, Eqn. (10.1.14) becomes
A
z }| {
num = − 12 `2 + xy(q 2 − 3m2 ) + [x(2 − x) + y(2 − y)]m2 + (2 − x)(2 − y)m2 γ µ


+ 2xy − [x(2 − x) + y(2 − y)] m q 0µ + [x(2 − x) − y(2 − y)]m q µ .



(10.1.25)
| {z } | {z }
B C

Let us simplify some of these coefficients. First, we simplify A.


 2
A = − 12 `2 + xyq 2 + −3xy  − x2 + 2y − y 2 + 4 −   m2

 +
 2x  − 2y + 
2x xy
= − 12 `2 + xyq 2 + [4 − (x + y)2 ]m2
= − 12 `2 + xyq 2 + [2 − (1 − z)][2 + (1 − z)]m2
= − 12 `2 + xyq 2 + (1 + z)(3 − z)m2 . (10.1.26)

Next, we simplify B.
B
= 2xy−2x+x2 −2y+y 2 = (x+y)2 −2(x+y) = −(x+y)(2−x−y) = −(1−z)(1+z) = −(1−z 2 ). (10.1.27)
m

Finally, we can write the amplitude as

d4 ` Aγ µ + Bq 0µ + Cq µ
Z Z
2 0
iM = (−ie)iλ ū(p ) u(p) · · · . (10.1.28)
F (2π)4 D3

Note that ∆ in Eqn. (10.1.12), and therefore D, is even under the interchange x ↔ y. Meanwhile, C is odd.
Therefore, after integration over the Feynman parameters, the term proportional to q µ vanishes. We proved
this in problem set 9 using current conservation (the Ward identity).
Recall the Gordon identity, which we proved in problem set 3. We will write it in the form

ū(p0 )q 0µ u(p) = ū(p0 ) 2mγ µ − iσ µν qν u(p).


 
(10.1.29)

Then, already dropping the C term, we get

d4 ` (A + 2mB)γ µ − Biσ µν qν
Z Z
2 0
iM = (−ie)iλ ū(p ) u(p) · · · . (10.1.30)
F (2π)4 D3

Define
d4 ` Fi
Z Z
2
F1 = A + 2mB, F2 = −2mB, Fi = iλ . (10.1.31)
F (2π)4 D3

Then, we can write the amplitude as

iσ µν qν
 
0 µ
iM = ū(p )(−ie) γ F1 + F2 u(p) · · · . (10.1.32)
2m

Compare this to the vertex at tree level:

iM = ū(p0 )(−ieγ µ )u(p) · · · . (10.1.33)

Therefore, we recognize that the correction to the vertex due to the Higgs is
iσ µν qν
δΓµ = γ µ F1 + F2 . (10.1.34)
2m
10.1. EXOTIC CONTRIBUTIONS TO G – 2, P&S 6.3 P.210 99

Only F2 affects g. It is given by


d4 ` −2m[−m(1 − z 2 )]
Z Z
F2 = iλ2 . (10.1.35)
F (2π)4 D3

We can use Eqn. 6.49 P&S p.193 to compute the ` integral:


d4 ` 1 i(−1)3
Z
1 1 i 1
4 3
= =− . (10.1.36)
(2π) D (4π)2 (3 − 1)(3 − 2) ∆3−2 2(4π)2 ∆

Therefore,
2 Z
1 − z2

λm
F2 = . (10.1.37)
4π F ∆
Anticipating the fact that mh >> m, write ∆ as
m 2 q 2
∆ = m2h z + (1 − z)2
   
mh − xy m . (10.1.38)

As expected, since ∆ is a function just of q 2 , so is F2 . Define


2
 ≡ mmh . (10.1.39)

g−2
Then, the correction to a = 2 due to the higgs is
2 Z Z Z 1
1 − z2

λ
ah = F2 (0) =  dx dy dz δ(x + y + z − 1) . (10.1.40)
4π 0 z + (1 − z)2 

Change the x and y variables to

u = x + y, v = y − x. (10.1.41)

We can invert these relations:

x = 21 (u − v), y = 21 (u + v). (10.1.42)

Therefore, the Jacobian matrix and determinant are


 ∂x ∂x   
∂u ∂v
1 1 −1 1
J = ∂y ∂y = , det J = . (10.1.43)
∂u ∂v 2 1 1 2

Therefore, dx dy = 21 du dv. The u integral goes from 0 (minimum of x + y) to 2 (maximum of x + y). The
v limits depend on u. When 0 ≤ u ≤ 1, the v limits are −u ≤ v ≤ u. When 1 ≤ u ≤ 2, the v limits are
u − 2 ≤ v ≤ 2 − u. Therefore,
 2 Z 1 Z 1 Z u Z 2 Z 2−u 
λ  1 − z2
ah = dz du dv + du dv δ(u + z − 1) . (10.1.44)
4π 2 0 0 −u 1 u−2 z + (1 − z)2 

The delta function sets u = 1 − z. Since 0 ≤ z ≤ 1, we have 0 ≤ 1 − z ≤ 1 and so the u integral in the range
[1, 2] vanishes. Thus,
 2 Z 1 Z 1−z  2 Z 1
λ  1 − z2 λ (1 − z)(1 − z 2 )
ah = dz dv =  dz . (10.1.45)
4π 2 0 −(1−z) z + (1 − z)2  4π z + (1 − z)2 
|0 {z }
I

Note that the integrand behaves like z1 near z = 0 when  = 0, which signals a logarithmic divergence in .
Therefore, we peal off a piece from the integral which will give a logarithm:
Z 1 
1 − 2(1 − z) 2(1 − z) − z(1 + z − z 2 )

I= dz + . (10.1.46)
0 z + (1 − z)2  z + (1 − z)2 
100 CHAPTER 10. PROBLEM SET 10

The first piece can be integrated exactly and just gives


Z 1
1 − 2(1 − z) 1
dz 2
= ln[z + (1 − z)2 ] 0 = ln 1 − ln  = − ln . (10.1.47)
0 z + (1 − z)

The integrand in the remaining piece is a much better behaved function. Its limit as z → 0 is just 2 and
it has no poles. Thus, we can simply set  = 0 in that term. The integrand becomes −1 − z + z 2 and the
integral is easy to compute:
Z 1
2(1 − z) − z(1 + z − z 2 )  z2 z 3  1 7
dz 2
≈ − z + − =− . (10.1.48)
0 z + (1 − z)  2 3 0 6

Therefore, we finally have


 2    2  
λ 7 λm 7 m
ah ≈ −  + ln  = − + 2 ln . (10.1.49)
4π 6 4πmh 6 mh

Since  << 1, its logarithm is large and negative. Therefore, it makes more sense to write
 2  
λm mh 7
ah = 2 ln − . (10.1.50)
4πmh m 6

(b) Well, all we can say is


|ah | < 10−10 . (10.1.51)

→0 →∞
ah −−−→ 0 and ah −−−→ −∞. Meanwhile, there is one point where da d = 0, namely  = e
h −19/12
≈ 0.2, and
2
d ah
d2 < 0 and so there is one maximum. Therefore ah must increase from 0, reach a maximum, then decrease
forever beyond that point. Clearly,  << 0.2 since m ≈ 0.5 MeV and mh > 60 GeV. Therefore, in the region
mh > 60 GeV, a bigger and bigger mh just makes ah smaller and smaller. Therefore, if mh = 60 GeV is not
excluded when λ = 3 × 10−6 , then the whole region mh ≥ 60 GeV is not excluded. Indeed, plugging in these
values gives
ah ≈ 9 × 10−23 . (10.1.52)

This is indeed much smaller than the upper bound in Eqn. (10.1.51) and thus is not excluded.
If we plug in λ = 6 × 10−4 instead, for the muon, and the same mh , we get

ah ≈ 4 × 10−18 . (10.1.53)

This is still not excluded.

10.2 Diagramatics of Effective Potential, Zee IV.3.4 p.244


Understand Eqn. 14, Zee p.240, reproduced below, using Feynman diagrams. Show that Veff is generated by an
infinite number of diagrams. [Hint: Expand the logarithm in Eqn. 14, Zee p.240, as a series in V 00 (ϕ)/k 2 and
try to associate a Feynman diagram with each term in the series.]

d4 k k − V 00 (ϕ)
Z  2 
i~
Veff (ϕ) = V (ϕ) − log + O(~2 ). (10.2.1)
2 (2π)4 k2

SOLUTION: (Thanks to Melanie (’12) for presenting her solution.)

Let us expand the quantum corrections to the potential in powers of V 00 :


∞ n
d4 k V 00 (ϕ)
Z 
i~ X 1
Veff (ϕ) − V (ϕ) = . (10.2.2)
2 n=1 n (2π)4 k2
10.3. DISCUSSION 1: COLEMAN-WEINBERG EFFECTIVE POTENTIAL 101

In massless φ4 theory, with V (ϕ) = λ 4


4! ϕ , we have V 00 (ϕ) = λ2 ϕ2 and so
∞ n
λn d4 k ϕ2
X Z 
Veff (ϕ) − V (ϕ) = i~ . (10.2.3)
n=1
2n+1 n (2π)4 k2

We recognize each term in the sum as describing a Feynman diagram with n vertices, each contributing a factor
of λ and having two external legs (the (ϕ2 )n term) and connected by a circle of n scalar propagators (the (k −2 )n
term). There is only one undetermined internal momentum since there are n vertices and n internal lines. The
internal lines other than the one with the undetermined momentum, k, do not have momentum k; they have
momenta that involve the external momenta. However, the integral just captures the divergence in k, when it is
much greater than any of the external momenta.
J

Q  Q   J

Q
Q + Q
Q 
 + + ··· (10.2.4)
   


  Q
Q
  Q J

J

J
In this divergent limit (large k 2 ), the external momenta do not matter. Each pair of external lines attached at a
vertex can be switched, which gives a symmetry factor of 2 for each vertex. In addition, the internal lines can be
thought of as forming an n-gon and since the momenta in all the lines are all approximately k, the symmetries
of the internal lines are the symmetries of the n-gon, which is the nth dihedral group, whose dimension is 2n (n
rotations including the identity and these rotations multplied by one reflection). This gives an extra factor of 2n
in the symmetry factor. The grand total is 2n (2n) = 2n+1 n, which shows up in Eqn. (10.2.3).

10.3 Discussion 1: Coleman-Weinberg Effective Potential


Consider minimally coupling E&M to a complex scalar field as we did in my notes in problem set 6. We will
slightly change the normalization of the quartic scalar interaction term for convenience

L = − 14 Fµν F µν + |Dφ|2 − m2 |ϕ|2 − λ 4


3! |ϕ| . (10.3.1)

where D = ∂µ + ieAµ .
Naı̈vely, if m2 > 0 then the potential for ϕ is minimized at ϕ = 0 and there is no spontaneous symmetry
breaking. Meanwhile, if m2 < 0, then ϕ is minimized at |ϕ| 6= 0 and there is spontaneous symmetry breaking.
Classically, the situation at m2 = 0 is undetermined. How do quantum corrections alter this behavior? To answer
this question, we will look at the Coleman-Weinberg effective potential at the one-loop level for m = 0.
Firstly, rewrite ϕ = √12 (φ + iψ) and use the gauge symmetry of this theory to force ψ = 0 (i.e. to force ϕ to
be real). Similarly, we will need some sort of gauge-fixing condition on the vector potential. The details of this
gauge-fixing condition will not matter. The important part, if you recall from my notes in problem set 6, is that
the photon becomes massive after we expand φ around the minimum of the potential, and thus the photon will
have three physical polarization
√ states.
When we plug ϕ = φ/ 2 into the Lagrangian (with m = 0) and write it out using the vector field, we get

L = 21 Aµ (∂ 2 + e2 φ2 )Aµ + 21 ∂µ φ ∂ µ φ − λ 4
4! φ . (10.3.2)

In order to isolate the scalar field whose effective potential we seek, we will integrate out the vector field. The
action is a functional of both fields, S[A, φ]. However, we can define the action with A integrated out, which is
a functional only of φ:
Z R 4 1
Z
µ λ 4 i
R 4 µ 2 2 2
eiS[φ] = DA δ(gauge-fix) eiS[A,φ] = ei d x( 2 ∂µ φ ∂ φ− 4! φ ) DA δ(gauge-fix)e 2 d x A (∂ +e φ )Aµ , (10.3.3)

where δ(gauge-fix) is some delta function which collapses the path integral to only those A satisfying the gauge-
fixing condition. The components of the vector potential are independent of each other in the action (the only
dependence comes in through the gauge-fixing delta function). Therefore, each independent component (of which
there are three) contributes a factor of
−1/2
det(∂ 2 + e2 φ2 ) = exp − 12 tr ln(∂ 2 + e2 φ2 ) ,
  
(10.3.4)
102 CHAPTER 10. PROBLEM SET 10

where we used the identity ln det = tr ln. Thus,


Z  
1 λ 4 3i
S[φ] = d x ∂µ φ ∂ φ − φ + tr ln(∂ 2 + e2 φ2 ).
4 µ
(10.3.5)
2 4! 2

Now, define the Wightman function as usual:


Z
eiW [J] = Dφ ei(S[φ]+Jφ) , (10.3.6)

d4 x Jφ, using the notation of Zee. As usual, define


R
where Jφ is short-hand notation for

δW [J] h0+ |φ̂|0− i


φc (x) ≡ = , (10.3.7)
δJ(x) h0+ |0− i

where |0− i and |0+ i are vacua in the infinite past and future, respectively. Then, define the effective action

Γ[φc ] ≡ W [J] − Jφc . (10.3.8)

Once again, Jφc here means d4 x Jφc . Just as we must replace q̇ by its functional form in terms of p when we
R

transform from the Lagrangian to the Hamiltonian, here J must be replaced by its functional form in terms of
φc by inverting Eqn. (10.3.7). One virtue of Γ is that

δΓ[φc ]
= −J(x). (10.3.9)
δφc (x)

We can expand Γ in the derivatives of φc (Taylor expansion). The term that does not contain any φc derivatives
is called the effective potential : Z
Γ[φc ] = − d4 x Veff [φc (x)] + · · · . (10.3.10)

The minus sign is there just because potentials enter into the Lagrangian and the action with a minus sign. Then,
we can extract the effective potential by plugging in a constant field for φc (x), which we call ρ, since doing so
sets all the terms involving the derivative of φc to zero. This makes Γ[ρ] proportional to the volume of spacetime,
which we denote by V :
Γ[ρ] = −V Veff (ρ). (10.3.11)
Define φ0 to be the saddle-point of the action S (with a source):

δS[φ]
= −J(x). (10.3.12)
δφ(x) φ=φ0

Expand S[φ] about φ0 :

δ 2 S[φ]
Z ZZ
4 δS[φ] 1 4 4
S[φ0 + φ] = S[φ0 ] + d x φ(x) + d x d y φ(x)φ(y) +··· . (10.3.13)
δφ(x) φ=φ0 2 δφ(x)δφ(y) φ=φ0
| {z } | {z }
−J(x) S (2) [φ0 ;x,y]

The Wightman function, Eqn. (10.3.6), is then given by


Z
eiW [J] = Dφ exp i{S[φ0 + φ] + J(φ0 + φ)}
Z
 + 21 S (2) [φ0 ]φ(x)φ(y) + Jφ0 + 

= Dφ exp i S[φ0 ] −  Jφ Jφ)

Z
= ei(S[φ0 ]+Jφ0 ) Dφ exp i 12 S (2) [φ0 ]φ(x)φ(y) .

(10.3.14)

Again, we have suppressed the integrals in the argument of the exponential. Using the same method as in Eqn.
(10.3.4), we get
W [J] = S[φ0 ] + Jφ0 + 2i tr ln S (2) [φ0 ]. (10.3.15)
10.3. DISCUSSION 1: COLEMAN-WEINBERG EFFECTIVE POTENTIAL 103

In the absence of quantum corrections,

φ0 (x) = φc (x) (at tree level). (10.3.16)

Therefore, we will package the quantum corrections into the difference, which we call φq :

φc (x) = φ0 (x) + φq (x). (10.3.17)

Using Eqn. (10.3.13) just up to the first order, we find

S[φc ] = S[φ0 + φq ] = S[φ0 ] − Jφq + · · · . (10.3.18)

Therefore,
S[φc ] + Jφc = S[φ0 ] + J(φc − φq ) = S[φ0 ] + Jφ0 . (10.3.19)
(2)
To the order to which we have W in Eqn. (10.3.15), we can replace φ0 in the S by φc . Then,

W [J] = S[φc ] + Jφc + i


2 tr ln S (2) [φc ]. (10.3.20)

Then, the effective action, Eqn. (10.3.8), is

Γ[φc ] = W [J] − Jφc = S[φc ] + i


2 tr ln S (2) [φc ]. (10.3.21)

Then, Eqn. (10.3.11) reads


− V Veff (ρ) = Γ[ρ] = S[ρ] + i
2 tr ln S (2) [ρ]. (10.3.22)
After all this work keeping S (2) around, we will now just drop it because it turns out to constitute corrections of
higher order in λ and e than are contained in S[ρ]. You might be wondering how this can be. After all, usually,
the S (2) is the one that contains quantum corrections and S is just the classical action. However, recall that S[φ]
is not the classical action, but is instead the action we get once we integrate out the vector potential. Thus, it
already contains radiative corrections. We will just focus on these corrections. Using S[φ] in Eqn. (10.3.5), we
get
− V Veff (ρ) = S[ρ] = −V 4!
λ 4
ρ + 3i 2 2 2
2 tr ln(∂ + e ρ ). (10.3.23)
Let us be more precise about the operator in the last term. It is really

tr ln(∂x2 + e2 φ(x)φ(y))δ(x − y) φ=ρ .



(10.3.24)

Therefore, taking the trace over momentum states simply turns ∂x2 into −k 2 and the delta function into V . Then,
φ is replaced with ρ. Thus, dividing Eqn. (10.3.23) by −V yields

d4 k
Z
λ 4 3i
Veff (ρ) = ρ − ln(−k 2 + e2 ρ2 ). (10.3.25)
4! 2 (2π)4

Going to Euclidean signature by defining k 0 = ikE


0
gives
Z 4
λ 4 3 d kE 2
Veff (ρ) = ρ + ln(kE + e2 ρ2 ). (10.3.26)
4! 2 (2π)4
Define
d4 kE
Z
λ 2 3
ν ≡ e2 ρ2 =⇒ Veff (ν) = ν + 2
ln(kE + ν) . (10.3.27)
4!e4 2 (2π)4
| {z }
I

The integral in I is divergent. In Zee, it is regularized using a sharp cut-off. We will use a different trick. If we
differentiate I with respect to ν three times, we get something convergent:
Z ∞
d3 I(ν)
Z 4 3
d kE 1 3 kE dkE
3
= 3 4 2 3
= 2 . (10.3.28)
dν (2π) (kE + ν) 8π 0 (kE + ν)3
2

Define ∞
2
d3 I(ν)
Z
kE 3 2u du
u≡ =⇒ 3
= . (10.3.29)
ν dν 32π 2 ν 0 (1 + u)3
104 CHAPTER 10. PROBLEM SET 10

We can perform this integral by partial fractions:


Z ∞ Z ∞ 
2u du 2 2
= − + du
0 (1 + u)3 0 (1 + u)3 (1 + u)2
 ∞
1 2
= 2

(1 + u) 1+u 0
 ∞
1 + 2u
=−
(1 + u)2 0
= 1. (10.3.30)

Therefore,
d3 I(ν) 3
= . (10.3.31)
dν 3 32π 2 ν
Integrate this with respect to ν once to get

d2 I(ν) 3
= ln ν + C 00 , (10.3.32)
dν 2 32π 2
where C 00 is some infinite constant. Integrate again to get

dI(ν) 3 3 B 3 B
= 2
ν ln ν − 2
ν + C 00 ν + 2 ≡ 2
ν ln ν + 2 + C 0 ν. (10.3.33)
dν 32π 32π e 32π e
Integrate one more time to get
3 2 3 1 B 3 2 B C
I(ν) = ν ln ν − ν2 + C 0ν2 + 2 ν + A ≡ ν ln ν + A + 2 ν + 4 ν 2 . (10.3.34)
64π 2 128π 2 2 e 64π 2 e e
Therfore, we have
λ 4 3e4 4
Veff (ρ) = ρ + ρ ln(e2 ρ2 ) + A + Bρ2 + Cρ4 . (10.3.35)
4! 64π 2
As usual, define the floating renormalization scale µ and the renormalized λR such that

3e4 e2 ρ2
  
λR (µ) 1
Veff (ρ) = ρ4 + − + ln . (10.3.36)
4! 64π 2 2 µ2

The factor of − 12 is convenient, but of course could be changed by redefining µ.


The derivative of the effective potential is

3e4 e2 ρ2
 
dVeff (ρ) 3 λR (µ)
= 4ρ + ln 2 . (10.3.37)
dρ 4! 64π 2 µ

This has a nonzero root at


2
3e4 e2 hφi 4π 2 λR (µ)
 
λR (µ) µ
2
ln =− =⇒ hφi = exp − . (10.3.38)
64π µ2 4! e 9e4

We can thus write Eqn. (10.3.36) as

3e4 ρ4 ρ2
 
1
Veff (ρ) = − + ln 2 . (10.3.39)
64π 2 2 hφi

The mass of the scalar particle is given by

d2 Veff 3e4

2
m2S = = hφi . (10.3.40)
dρ2 ρ=hφi
8π 2

The mass of the photon is given by


2
m2V = e2 hφi . (10.3.41)
10.4. DISCUSSION 2: A LOWER BOUND ON THE HIGGS MASS 105

Let us be clear about the logic here. Without experimental input, all we can say is that both the scalar and
vector particles gain mass. But these masses are theoretically arbitrary since they are proportional to the as-
yet-undetermined vev hφi. However, if we can perform an experiment in an energy scale where the value of e is
known and that can be sensitive only to something like the ratio of the masses, then this could really make or
break the theory because the mass ratio is independent of the unknown hφi.
Even though there is no natural mass scale to begin with, since we started with a massless theory, we were
forced to introduce a floating renormalization scale, µ. If the theory is really scale-free, then we should expect
that all renormalization points be equivalent. However, we find that this is not the case. Instead, the scale µ
related to hφi is singled out. This singling out of a mass scale from a theory that originally has no preferred mass
scale is called dimensional transmutation. The name is apt because the cut-off or renormalization scale has been
transmuted into a physical mass.
Something of this sort happens in QCD. We start off with massless u and d quarks and end up with the
massive spectrum of hadrons as bound states. The story there is a fair bit more complicated, however.
Finally, before we move on to the Higgs, we should motivate, in hindsight, introducing the photon to begin
with. In the regular massless scalar theory, we get a very similar form for the effective potential roughly via the
identification λ ∼ e2 . Looking at Eqn. (10.3.38), we see that this would force us to equate a term of order λ
with one of order λ2 thereby taking us outside the regime of validity of perturbation theory. By coupling to the
photon, we introduce the second coupling constant, e, such that e4 is basically able to play the role of λ2 and
this problem vanishes since these two coupling constants are independent.

Alternative derivation:

In the pure massless scalar case, the one-loop effective action was found to be (Eqn. 12 Zee p.239)
Z    
4 1 µ λ 4 i 2 λ 2
Γ[φc ] = d x ∂µ φc ∂ φc − φc + tr ln ∂ + φc . (10.3.42)
2 4! 2 2
If we drop the S (2) term, Eqn. (10.3.21) reads
Z  
4 1 λ 4 3i
∂µ φc ∂ φc − φc + tr ln ∂ 2 + e2 φ2c .
µ

Γ[φc ] = S[φc ] = d x (10.3.43)
2 4! 2
0
Recall that Eqn. (10.3.42) leads to a solution to Veff = 0 that requires equating a tree level result (order λ) with
2
a quantum correction (order λ ). The equation at some appropriately chosen renormalization scale, µ, can be
written as
2
3 2 λ hφi
λ=− λ ln . (10.3.44)
32π 2 2µ2
The λ on the left comes directly from the action and the λ2 on the right comes from the quantum correction in
Eqn. (10.3.42) (i.e. the tr ln term). We see how coupling to the photon avoids this problem: the λ on the right
is replaced by ∼ e2 . Since λ and e are independent coupling constants, the equation can be solved within the
2
regime of validity of perturbation theory. That is, both λ and e can be small while producing a solution for hφi .
To get the equation in the photon-coupled case, we replace λ with 2e on the right hand side of Eqn. (10.3.44) and
multiply by 3 due to the extra factor of 3 in the trace term of Eqn. (10.3.43). Then, we multiply the numerator
and denominator by 2 for convenience:
2 2
2 3 2 2 e2 hφi λ 3e4 e2 hφi
λ=− ·3· (2e ) ln =⇒ =− ln . (10.3.45)
2 32π 2 µ2 4! 64π 2 µ2
This reproduces Eqn. (10.3.38).

10.4 Discussion 2: A Lower Bound on the Higgs Mass


Letus now think of the scalar field in the previous section as the one component of the SU (2) doublet Higgs field
that cannot be gauged away. This time, we have three vector fields corresponding to the W ± and Z 0 bosons. If
we start of with a “mass” term for the Higgs, the the generalization of the effective potential in Eqn. (10.3.36)
would be
3
3ρ4 X 2 e2i ρ2
 
λ 1 1
Veff (ρ) = ρ4 + m2 ρ2 + e − + ln , (10.4.1)
4! 2 64π 2 i=1 i 2 µ2
106 CHAPTER 10. PROBLEM SET 10

where ei is the coupling appropriate to the three vector bosons. Since we are not starting from scratch, let us
consider these parameters to already be renormalized.
If the Higgs develops a nonzero vev, hφi, then the vector masses are
2
m2i = e2i hφi . (10.4.2)

Redefine µ0 to be the scale at which


3
λ 3 X 4 e2i ρ2
=− e ln 02 . (10.4.3)
4! 64π 2 i=1 i µ
Then,
3
3ρ4 X 4 µ02 ρ2
 
1 2 2 1
Veff (ρ) = m ρ + e − + ln . (10.4.4)
2 64π 2 i=1 i 2 µ2

Redefine µ first to be µ/µ0 and then to absorb the − 21 factor. Then,


3
1 2 2 3ρ4 X
4 ρ2
Veff (ρ) = m ρ + 4 mi ln . (10.4.5)
2 64π 2 hφi i=1 µ2

It is convenient to define
3
3 X 3
α≡ 4 m4i = 4
4 (2mW + m4Z ). (10.4.6)
64π 2 hφi i=1 64π 2 hφi
Then,
1 2 2 ρ2
Veff (ρ) = m ρ + αρ4 ln 2 . (10.4.7)
2 µ
The first and second derivatives are
ρ2
 
dVeff (ρ) 2 3 1
= m ρ + 4αρ + ln 2 , (10.4.8a)
dρ 2 µ
2
ρ2
 
d Veff (ρ) 2 2 7
= m + 12αρ + ln 2 .
dρ2 6 µ
Vanishing of the first derivative yields the equation
 2
2 1 hφi
m2 = −4α hφi + ln 2 , (10.4.9)
2 µ
which, when plugged into the potential and its second derivative gives
 2
4 hφi
Veff (hφi) = −α hφi 1 + ln 2 , (10.4.10a)
µ
2
d2 Veff (ρ)

2 3 hφi
= 8α hφi + ln . (10.4.10b)
dρ2 ρ=hφi 2 µ2

Veff (0) = 0 implies that if Eqn. (10.4.10a) is positive then it is a local minimum, but higher than the minimum
at ρ = 0. It is at best a metastable state. Therefore, in order to have spontaneous symmetry breaking, we must
have
2
hφi
ln 2 ≥ −1. (10.4.11)
µ
Eqn. (10.4.10b) gives the physical Higgs mass on which we therefore get a lower bound:

2 3
m2H ≥ 4α hφi = 4
2 (2mW + m4Z ). (10.4.12)
16π 2 hφi
Often, hφi is called v when it pertains to the Higgs. This is called the electroweak scale. It is related to the Fermi
coupling constant via
−1/2
v = 2−1/4 GF . (10.4.13)
10.4. DISCUSSION 2: A LOWER BOUND ON THE HIGGS MASS 107

We have experimental values for GF (e.g. via neutron beta decay), and thus for v. We also have experimental
values for the W and Z masses (e.g. leptonic decays and the neutral K decay). The accepted values are

v = 246 GeV, mW = 80.4 GeV, mZ = 91.2 GeV. (10.4.14)

Plugging these numbers into Eqn. (10.4.12) gives

mH ≥ 6.92 GeV. (10.4.15)


108 CHAPTER 10. PROBLEM SET 10
Chapter 11

Problem Set 11

11.1 The Gross-Neveu Model, P&S 11.3 p.390


The Gross-Neveu model is a model in two spacetime dimensions of fermions with a discrete chiral symmetry:
2
/ i + 12 g 2 ψ i ψi ,
L = ψ i i∂ψ (11.1.1)
with i = 1, . . . , N . The kinetic term of two-dimensional fermions is built from matrices γ µ taht satisfy the
two-dimensional Dirac algebra. These matrices can be 2 × 2:
γ 0 = σ2 , γ 1 = iσ 1 , (11.1.2)
where σ i are Pauli sigma matrices. Define
γ 5 = γ 0 γ 1 = σ3 ; (11.1.3)
µ
this matrix anticommutes with the γ .
(a) Show that this theory is invariant with respect to
ψi → γ 5 ψi , (11.1.4)

and that this symmetry forbids the appearance of a fermion mass.


(b) Show that this theory is renormalizable in 2 dimensions (at the level of dimensional analysis).
(c) Show that the functional integral for this theory can be represented in the following form:
Z Z  Z  
R 2 1
Dψ Dψ ei d x L = Dψ Dψ Dσ exp i d2 x ψ i i∂ψ / i − σψ i ψi − 2 σ 2 , (11.1.5)
2g

where σ(x) (not to be confused with a Pauli matrix) is a new scalar field with no kinematic energy terms.
(d) Compute the leading correction to the effective potential for σ by integrating over the fermion fields ψi . You
will encounter the determinant of a Dirac operator; to evaluate this determinant, diagonalize the operator by
first going to Fourier components and then diagonalizing the 2 × 2 Pauli matrix associated with each Fourier
mode. (Alternatively, you might just take the determinant of this 2 × 2 matrix.) This 1-loop contribution
requires a renormalization proportional to σ 2 (that is, a renormalization of g 2 ). Renormalize by minimal
subtraction.
(e) Ignoring two-loop and higher-order contributions, minimize this potential. Show that the σ field acquires a
vacuum expectation value which breaks the symmetry of part (a). Convince yourself that this result does
not depend on the particular renormalization condition chosen.
(f) Note that the effective potential derived in part (e) depends on g and N according to the form
Veff (σcl ) = N · f (g 2 N ). (11.1.6)

(The overall factor of N is expected in a theory with N fields.) Construct a few of the higher-order contri-
butions to the effective potential and show that they contain additional factors of N −1 which suppress them
if we take the limit N → ∞, with g 2 N fixed. In this limit, the result of part (e) is unambiguous.

109
110 CHAPTER 11. PROBLEM SET 11

SOLUTION: (Thanks to Lucia (’12) for presenting parts (a), (b) and (c), and Dillon (’12) for part (d).)

(a) Let us write these matrices out explicitly. We find


     
0 0 −i 1 0 i 5 1 0
γ = , γ = , γ = . (11.1.7)
i 0 i 0 0 −1

Also, by explicit calculation, we find


{γ 5 , γ µ } = 0. (11.1.8)

If ψ → γ 5 ψ, then
ψ = ψ † γ 0 → ψ † γ 5 γ 0 = −ψ † γ 0 γ 5 = −ψγ 5 , (11.1.9)

where we used Eqn. (11.1.8) and the fact that γ 5† = γ 5 , as can be seen from (11.1.7).
Therefore, the Lagrangian transforms as
2 2
L → −ψ i γ 5 i∂γ
/ 5 ψi + 12 g 2 −ψ i γ 5 γ 5 ψi / i + 21 g 2 ψ i ψi
= ψ i i∂ψ =L, (11.1.10)

where we used Eqn. (11.1.8) again on the kinetic term and (γ 5 )2 = 1, as can be seen from (11.1.7). This
shows that the Lagrangian is invariant. The path integral measure is Dψ Dψ gains a factor of det γ 5 = −1
for ψ and ψ separately. These two minus signs cancel, thereby leaving the path integral measure invariant
as well. Therefore, the theory is invariant .
A fermion mass term takes the form Lm = −mψψ which transforms nontrivially under (11.1.4):

Lm = −mψψ → mψγ 5 γ 5 ψ = mψψ = −Lm . (11.1.11)

Hence, this symmetry forbids the appearance of a fermion mass.


(b) The dimension of ψ in d dimensions is

0 = [dd x] + [∂] + 2[ψ] = −d + 1 + 2[ψ] =⇒ [ψ] = d−1


2 . (11.1.12)

Therefore, the dimension of the coupling constant is

0 = [dd x] + 4[ψ] + 2[g] = −d + 4 d−1 2−d



2 + 2[g] =⇒ [g] = 2 . (11.1.13)

Hence, the theory is renormalizable for d ≤ 2 . For d > 2, g has negative mass dimension and the theory is
non-renormalizable.
We can also take the superficial degree of divergence perspective. As in problem set 9, with D being the
degree of divergence, L the number of loops, FI the number of internal fermion lines, FE the number of
external fermion lines, and V the number of vertices, we get the following three equations;

D = dL − FI , L = FI − (V − 1), 4V = 2FI + FE . (11.1.14)

Together, these give


D = d − d4 FE + d

2 − 1 FI . (11.1.15)

If d ≤ 2 then the FI contribution is ≤ 0 producing only finitely many divergences. If d > 2, then the FI
contribution is positive and we can make a diagram arbitrarily divergent by adding more internal lines.
(c) We can integrate out σ to get the original path integral, which we call I:
Z  Z Z  Z 
2 i 2 2 2 2 2 2 2

I = Dψ Dψ exp i d x ψ i i∂ψi / Dσ exp − 2 d x σ + 2g ψ i ψi σ + (g ψ i ψi ) − (g ψ i ψi )
2g
Z  Z Z  Z 
2
 1 2
2 i 2 2
2
/
= Dψ Dψ exp i d x ψ i i∂ψi + 2 g ψ i ψi Dσ exp − 2 d x σ + g ψ i ψi . (11.1.16)
2g
11.1. THE GROSS-NEVEU MODEL, P&S 11.3 P.390 111

One can shift and rescale σ and since the shift and rescaling is independent of σ, the path integral measure
changes by at most an overall constant, which we can simply absorb into the fermion integral measure or
neglect since it drops out when taking expectation values.
Z
I= Dψ Dψ eiS . (11.1.17)

(d) Denote the action containing fermions and σ by S[ψ i , ψi , σ]. Let S[σ] be the action for σ when the ψi are
integrated out. Let I now denote the integration over the fermion fields:
Z  Z 
I = Dψ Dψ exp i d2 x ψ i (i∂/ − σ)ψi = DetN (i∂/ − σ) = exp N Tr ln(i∂/ − σ) ,
 
(11.1.18)

where we used the identity ln Det = Tr ln. The trace involves integration over phase space as well as a trace
over the spinor indices, which we denote by tr. Analogously, we use Det to mean the total determinant and
we reserve det for the determinant over spinor space:
Z 2 2 Z 2 2
d xd p d xd p
Tr ln(i∂/ − σ) = 2
tr ln(p/ − σ) = / − σ).
ln det(p (11.1.19)
(2π) (2π)2

The latter determinant is


−i(p0 + p1 )
 
−σ
/ − σ) = det
det(p = σ 2 − (p0 )2 + (p1 )2 = −p2 + σ 2 . (11.1.20)
i(p0 − p1 ) −σ

Therefore,  Z 2 2 
d xd p 2 2
I = exp N ln(−p + σ ) . (11.1.21)
(2π)2

Wick-rotate the p0 integral by defining p0 = ip0E so that


 Z 2 2 
d x d pE 2 2
I = exp iN ln(pE + σ ) . (11.1.22)
(2π)2

Therefore, we have the action just in terms of σ:


Z  Z 2 
2 1 2 d pE 2 2
S[σ] = d x − 2 σ + N ln(pE + σ ) . (11.1.23)
2g (2π)2
| {z }
L(1)

We can use several different methods to calculate L(1) :

Method 1 (Sharp cut-off ): As in Zee, we will subtract a constant (infinite, but constant and thus
irrelevant) to make the argument of the logarithm unitless.

ln(p2E + σ 2 ) → ln(p2E + σ 2 ) − ln p2E . (11.1.24)

Then, we write this in a convenient form as an integral:


Z σ2
du
ln(p2E + σ 2 ) − ln p2E = . (11.1.25)
0 p2E+u

Then, we can switch the order of integration (between u and pE ):


Z Λ Z σ2 Z σ2 Z Λ
d2 pE
 2
pE + σ 2
Z 
(1) 1 du 1 2pE dpE
L = ln 2 = dpE pE 2 = du . (11.1.26)
(2π) 2 pE 2π 0 0 pE + u 4π 0 0 p2E + u
112 CHAPTER 11. PROBLEM SET 11

The pE integral gives


Z Λ
2pE dpE
= ln(Λ2 + u) − ln u ≈ ln Λ2 − ln u. (11.1.27)
0 p2E + u

The u integral yields


σ2
Λ2
Z  
(1) 1 2 1 2
σ 2 1 2
L = du(ln Λ − ln u) = u(ln Λ − ln u + 1) 0 =
σ ln 2 + 1 . (11.1.28)
4π 0 4π 4π σ

To lowest order in g, the effective action is just the action with the fermions integrated out:

σc2
Z   
2 1 2 N 2
Γ[σc ] = S[σc ] = d x − 2 σc − σ ln 2 − 1 . (11.1.29)
2g 4π c Λ

The effective action is extracted by plugging in a constant field σc = σ = constant:

σ2
Z  
2 1 2 N 2
Γ[σ] = − d x Veff (σ) =⇒ Veff (σ) = 2 σ + σ ln 2 − 1 . (11.1.30)
2g 4π Λ

We write this in terms of a renormalized coupling gR (µ), which is a function of the renormalization scale, µ:

σ2
 
1 2 N 2
Veff (σ) = 2 σ + σ ln 2 − 1 . (11.1.31)
2gR 4π µ

The relationship between the renormalized and bare couplings is

1 1 N µ2
2 = + ln . (11.1.32)
gR g2 2π Λ2

Since gR must be Λ-independent, the Λ-dependence of g must be just so as to cancel the last term:

d 1 d 1 N
Λ 2 =0 =⇒ Λ = . (11.1.33)
dΛ gR dΛ g 2 π

The same equation holds for gR with Λ replaced with µ.

Method 2 (Dimensional Regularization): This is the method employed by Peskin and Schroder in Eqns.
11.72 and 11.73 p.374. Copying the result directly from there, we get

Γ − d2
Z d 
(1) d pE 2 2 1
L = ln(pE + σ ) = − . (11.1.34)
(2π) d (4π) d/2 (σ )−d/2
2

Set d = 2(1 − ) so that


Γ(−1 + ) 2 1−
L(1) = − (σ ) . (11.1.35)
(4π)1−

Recall the definition of the Gamma function:


Z ∞
Γ(z) = sz−1 e−s ds. (11.1.36)
0

This sastisfies the recursion relation


Γ(z + 1)
Γ(z + 1) = zΓ(z) =⇒ Γ(z) = . (11.1.37)
z
11.1. THE GROSS-NEVEU MODEL, P&S 11.3 P.390 113

This shows that Γ(z) has a simple pole at z = 0 with residue 1. We can extend Γ(z) into the region Re z > −n
by iterating the recursion relation:
Γ(z + n)
Γ(z) = . (11.1.38)
z(z + 1) · · · (z + n − 1)

Near z = −n, Γ(z + n) has the same simple pole as Γ(z) has near z = 0. Therefore, Eqn. (11.1.38) implies

(−1)n 1
Γ(−n) ∼ . (11.1.39)
n! z + n

This picks up the pole, but not the constant term. Unfortunately, that is much harder to capture. Thankfully,
Mathematica saves the day:
1
Γ(−1 + ) = − − 1 + γ + O(). (11.1.40)

Now, let us expand the rest of L(1) :
 2 1−
σ2 σ2
 2
σ σ
= − ln  + O(2 ). (11.1.41)
4π 4π 4π 4π

Therefore,
  2   2 
1 σ σ
L(1) = − − − 1 + γ + O() 1 − ln  + O(2 )
 4π 4π
2 2
 
σ 1 σ
= + 1 − γ − ln + O()
4π  4π
σ2 1 σ2
 
= + 1 − γ + ln 4π − ln σ 2 , (11.1.42)
4π  4π

and the effective potential is


 
1 N 1 N 2
Veff (σ) = 2 σ 2 − + 1 − γ + ln 4π σ 2 + σ ln σ 2 . (11.1.43)
2g 4π  4π

N 1

Let us define a generalized MS renormalization scheme by absorbing a factor of 4π  − γ + c into the
coupling constant, where c is some arbitrary constant. To do this, we define a renormalization scale, µ, and
a renormalized coupling gR (µ) such that

σ2
 
1 N 2
Veff (σ) = 2 σ 2 + σ ln 2 − 1 − ln 4π + c . (11.1.44)
2gR 4π µ

The standard MS scheme has c = 0. The MS scheme has c = ln 4π. Therefore,

σ2
 
MS 1 2 N 2
Veff (σ) = 2σ + σ ln 2 − 1 − ln 4π ,
2gR 4π µ
 2
 (11.1.45)
MS 1 N σ
Veff (σ) = 2 σ 2 + σ 2 ln 2 − 1 .
2gR 4π µ

We see that the MS scheme coincides with our result using a sharp cut-off, (11.1.31).

Method 3 (Slick): Note that the only difference between the results of the different renormalization schemes
−2
is a constant times σ 2 , which is equivalent to an overall constant in the definition of gR relative to g −2 .
That is, we didn’t really need to know exactly how to define µ in terms of Λ and g, in the case of method
1, or  and g in the case of method 2. Whatever µ has to be in order to cancel the appropriate terms that I
want to cancel, that is what µ will be.
114 CHAPTER 11. PROBLEM SET 11

With this perspective, we realize that we don’t really need to know the exact constants in the divergences
that we need to cancel. All we really need to know is their functional form in order to check that we can
actually cancel them by adding appropriate counterterms.
With this in mind, let us define new variables and rewrite L(1) slightly:

d2 k
Z
2 (1)
ν≡σ , k ≡ pE =⇒ L = ln(k 2 + ν). (11.1.46)
(2π)2

Differentiate L(1) twice with respect to ν:

dL(1) d2 k d2 L(1) d2 k
Z Z
1 1
= , =− . (11.1.47)
dν (2π)2 k 2 + ν dν 2 (2π)2 (k 2 + ν)2

This last integral is convergent:


∞ ∞
d2 L(1) −2k dk
Z
1 1 =− 1 .

= = (11.1.48)
dν 2 4π 0
2
(k + ν) 2 2
4π(k + ν) 0 4πν

Now, we integrate this twice with respect to ν. Once gives

dL(1) 1
=− ln ν + B 0 , (11.1.49)
dν 4π

where B 0 is a constant of integration, which is formally infinite. A second time gives


1 1
L(1) = − (ν ln ν − ν) + B 0 ν + A ≡ − ν ln ν + Bν + A, (11.1.50)
4π 4π

where A and B are more infinite constants.


An overall constant in the Lagrangian is irrelevant and so we can just drop A. Replacing ν with σ 2 , we
find the effective potential
1 N 2
Veff (σ) = 2 σ 2 − N Bσ 2 + σ ln σ 2 . (11.1.51)
2g 4π

This takes the same form as Eqn. (11.1.43) except that we just don’t know the exact form of B and how
it depends on the regulator (Λ or ). That’s okay, we will just say that the counterterms involving the
renormalization scale µ will be whatever they need to be in order to essentially cancel the B term. At the
end of the day, we will again end up with

σ2
 
1 2 N 2
Veff (σ) = 2 σ + σ ln − 1 , (11.1.52)
2gR 4π µ2

where the factor of −1 in the parenthesis is chosen so as to coincide with previous results, but it is arbitrary
in all of the methods anyway.
(e) In taking the derivative of Veff , we see the virtue of having the factor of −1. Henceforth, we will drop the
subscript R in gR and consider the renormalization of g to be done already.

σ2 σ2
   
0 1 N N 22 1 N
Veff (σ) = 2 σ + σ ln 2 − 1 + σ = + ln σ. (11.1.53)
g 2π µ 2π σ g2 2π µ2

Setting this equal to zero gives a nontrivial solution


2
m ≡ hσi = µe−π/N g . (11.1.54)

We call this expectation value m because, if you look back at the action S[ψ, ψ, σ] in Eqn. (11.1.5), you will
see that a vev for σ will give a fermion mass term, − hσi ψψ, thereby breaking the symmetry in part (a).
11.1. THE GROSS-NEVEU MODEL, P&S 11.3 P.390 115

Different renormalization schemes will change the value of g 2 slightly, but will not negate the existence of
00
a solution. In order for this to be a local minimum, we need Veff > 0 at the above solution. Indeed,

00 1 N σ2 N V 0 (σ) N
Veff (σ) = 2
+ ln 2 + = eff + . (11.1.55)
g 2π µ π σ π

Therefore, when evaluated at the solution, (11.1.54), the second derivative is positive:

00 N
Veff (m) = > 0. (11.1.56)
π

Finally, in order for this to be the actual vacuum, Veff < 0 at the solution. Otherwise, it is just a metastable
vacuum and the true vacuum would be at hσi = 0. Well, we can write

σ 0 N 2
Veff (σ) = V (σ) − σ . (11.1.57)
2 eff 4π

At the solution, the first term vanishes leaving the second term, which is negative as desired.

(f) N is playing the role of ~−1 here. Recall that when we reintroduced ~ by eiS → eiS/~ , we showed that the loop
expansion could be interpreted as an ~ expansion in the sense that the amplitude of diagram with L loops
contained a factor of ~L−1 . Thus, we can see that summing more and more complicated Feynman diagrams
really does just mean systematically including more and more quantum corrections, which are irrelevant in
the classical limit when ~ → 0.
We can say the same thing here. This time, the action has a factor of N in front of it, instead of ~−1 .
It is important, of course, that the rest of the parameters in the action, namely λ = g 2 N , are constant.
Therefore, the amplitude of a diagram with L loops contains a factor of N 1−L and higher-loop diagrams
become irrelevant in the N → ∞ limit.
Recall how this power counting works. If the action has an overall factor of N , then every single interaction
term contains a factor of N . Since the propagator is just the inverse of the operator between the two factors
of the field in the quadratic term, every propagator gets a factor of N −1 . Therefore, a diagram with V
vertices and I internal lines comes with a factor of N V −I . Recall the relationship between the number of
loops, vertices and internal lines: L = I − (V − 1), since each internal line gets an integral, except that some
of these integrals collapse due to momentum conservation delta functions at the vertices. Then, one delta
function just imposes overall momentum conservation and does not restrict any internal momentum. Thus,
the power of N in a diagram with L loops is

N V −I = N 1−L . (11.1.58)

To actually construct some higher-loop diagrams than those that were included in the previous work, we
need to know what types of interactions we now have after integrating out the fermions. We could interpret
L(1) , defined in Eqn. (11.1.23), just as we did in the case of φ4 theory in problem set 10. We would find that
the integral just amounts to summing all one-loop 1PI diagrams with only fermions in the loop:
TT 
 @ 
@
T 
+ + + ··· (11.1.59)
  
@  T
@  TT

Note that we have drawn σ as a double line because it is supposed to play the role of a ψψ condensate.
We could also work backwards and say that we want to consider just 1PI diagrams where the internal
lines contain only fermions (lowest order for the σ field). The diagrams above are the only ones that fit the
bill! There are no diagrams with more loops where the loops only include fermions. Therefore, we see that
the action for σ alone contains vertex interactions for any even number of σ lines. Furthermore, we easily
see why each vertex gets a factor of N : each diagram above has N copies, one for each fermion that can
116 CHAPTER 11. PROBLEM SET 11

run around the loop. Therefore, the one-loop diagrams in the σ theory would be the following. Since the
fermions are gone now, we will just use single lines to denote σ.

Q  Q   @ 
Q Q  aa
Q + Q  + ··· @
a
! + ··· (11.1.60)
   !! 
  Q
 Q
  Q

Three examples of two-loop diagrams are


 Q 
   
Q  
Q   (11.1.61)
   
 Q Q
Q Q
 Q Q

11.2 Discussion 1: Clarification Regarding the Effective Potential


Before we move on to discuss the Gross-Neveu model specifically, I want to correct (or qualify) something I said
during last discussion section with regards to the Zee problem. We were supposed to interpret the one-loop
effective potential for φ4 theory as a sum over one-loop 1PI diagrams. We saw that the internal propagators all
had momentum k and the external lines could be interchanged in pairs and swapped within each pair that is
attached to the same point on the loop. This makes the diagram considerably more symmetric than it usually
is and gave us the symmetry factor 2n (2n). I said that we only care about the divergence of the diagram, as far
as renormalization is concerned, and in this limit, we can neglect all momenta except the internal momentum,
k. While this is true, it happens that when we calculate the effective potential, we are actually supposed to set
all the external momenta to zero. Therefore, the statement about only caring about the large k divergence is
actually superfluous. Let us see how this works.
The point is to compare the two ways we can expand the effective action: (a) in powers of the field; and (b)
in powers of derivatives of the field. It will be more convenient to consider the expansion in powers of the Fourier
transform of the field instead:
∞ n
d4 ki
Z
X 1 Y
Γ[φc ] = Γn (k1 , . . . , kn ) (2π)4 δ(k1 + · · · + kn ) φec (ki ) , (11.2.1a)
n=0
n! i=1
(2π)4
Z
Γ[φc ] = d4 x −Veff (φc ) + higher-derivatives .
 
(11.2.1b)

To isolate the effective potential, we plug in a constant value for φc , which we call ρ. In this case,
Z
φec (k) = d4 x eikx ρ = (2π)4 ρ δ(k). (11.2.2)

Therefore,
∞ Z n
X 1 Y
Γ[ρ] = Γn (k1 , . . . , kn ) (2π)4 δ(k1 + · · · + kn ) ρ δ(ki ) d4 ki
n=0
n! i=1

X ρn
= Γn (0) (2π)4 δ(0). (11.2.3)
n=0
n!

Using (2π)4 δ(0) = d4 x, and the fact that everything in Eqn. (11.2.3) is independent of spacetime, we can write
R

the effective action with constant field as



ρn
Z X Z
Γ[ρ] = d4 x Γn (0) = − d4 x Veff (ρ), (11.2.4)
n=0
n!

where we have used the fact that the higher derivative terms in Eqn. (11.2.1b) vanish when φc = ρ is constant.
Therefore, when calculating the effective potential, the external lines are supposed have zero 4-momentum.
11.3. DISCUSSION 2: MORE ON GROSS-NEVEU 117

11.3 Discussion 2: More on Gross-Neveu


The Gross-Neveu model is actually just the 2D version of a theory of dynamical symmetry breaking proposed by
Nambu and Jona-Lasinio to explain spontaneous chiral symmetry breaking. The virtue of the 2D version is that
the 4-Fermi interaction is renormalizable in 2D whereas it is non-renormalizable in 4D.
The difference between this type of symmetry breaking and the Higgs mechanism is that the latter requires
the introduction of a separate scalar field to explain the symmetry breaking and, as often happens, the masses
that result. On the other hand, dynamical symmetry breaking does not require coupling to a separate field.
Instead, interactions in the field(s) that is already present in the description of the system force composite
“particles”, built out of the pre-existing particles, to form a condensate (i.e. a non-vanishing vev). This is
exactly what happens in the BCS theory of superconductivity in which pairs of electrons of opposite spin form
bound Cooper pairs. Therefore, the Higgs mechanism in particle physics is analogous to the Ginzburg-Landau
theory of superconductivity while the Gross-Neveu model and other theories exhibiting dynamical symmetry
breaking are analogous to the BCS theory of superconductivity.
Since σ is supposed to play the role of the bound state of ψψ that gains a vev, and since this mechanism gives
a mass to the fermions, m, we would expect the σ bound state to have mass as well. The difference between mσ
and 2m ought to be interpreted as a binding energy. We will find that this binding energy is a ≥ 2-loop effect
and vanishes even at one-loop.
We learned in part (e) of the problem that σ gains a vev, which we called m. Then, when we shift σ → m + σ,
the Lagrangian becomes
m2 m 1 2
L = ψ(i∂/ − m)ψ − σψψ − 2g 2 − g 2 σ − 2g 2 σ . (11.3.1)
The constant and linear term in σ are unimportant and we neglect them. We would like to see the effect of the
presence of the fermions on the σ propagator. Thus, we want to integrate out the fermions, but only focus on
the term that contributes to the σ propagator, which is the first diagram in (11.1.59).
At tree level, the amplitude for the two-point function in σ (inverse propagator) is
(0)
Γ2 = − gi2 . (11.3.2)

The amplitude for the first diagram in (11.1.59) is

d2 k i(/
k+p / + m) i(/
Z  
k + m)
iM = −i tr ·
(2π)2 (k + p)2 − m2 k 2 − m2
Z 1 Z
d2 k tr[(/
k+p/ + m)(/
k + m)]
=− dx 2  2 . (11.3.3)
0 (2π) x[(k + p) − m ] + (1 − x)(k 2 − m2 )
2 2

The denominator is written as D2 where

D = k 2 + 2xp · k + x2 p2 − x2 p2 + xp2 − m2 = (k + xp)2 + x(1 − x)p2 − m2 = `2 − ∆, (11.3.4)

where

` = k + xp, ∆ = m2 − x(1 − x)p2 . (11.3.5)

In simplifying the numerator, we will automatically drop terms linear in ` since they vanish upon integrating over
`. Furthermore, we will drop traces of a single slashed object since the gamma matrices are traceless. Finally,
2
/ = p2 , which is a result of the defining relation {γ µ , γ ν } = 2g µν . Thus,
we use the identity p

num = tr[`2 − x(1 − x)p2 + m2 ] = tr(`2 + ∆) = 2(`2 + ∆). (11.3.6)

Therefore, altogether, we have


1 1
d2 ` `2 + ∆ d2 `
Z Z Z Z  
1 2∆
iM = −2 dx = −2 dx + . (11.3.7)
0 (2π)2 (`2 − ∆)2 0 (2π)2 `2 − ∆ (`2 − ∆)2

Wick rotate to
1
d2 `E
Z Z  
2∆ 1
iM = −2i dx − 2 . (11.3.8)
0 (2π)2 (`2E + ∆)2 `E + ∆
118 CHAPTER 11. PROBLEM SET 11

Let us calculate each ` integral one at a time:



∆ ∞ 2`E d`E
Z 2 Z  
d `E 2∆ ∆ 1 ∆ 1 1
2 2 2
= 2 2
=− 2 =− 0− = . (11.3.9)
(2π) (`E + ∆) 2π 0 (`E + ∆) 2π `E + ∆ 0
2π ∆ 2π

The next integral is log divergent and requires regularization:


Λ
d2 `E Λ2
Z Z
1 1 2`E d`E 1 2
Λ 1 1 ∆
− 2 2 =− 2 = − ln(`E + ∆)
0
= − ln = ln 2 . (11.3.10)
(2π) `E + ∆ 4π 0 `E + ∆ 4π 4π ∆ 4π Λ

Therefore,
1 Z 1
m2 − x(1 − x)p2
Z  
i ∆ i i
iM = − dx 2 + ln 2 = − − dx ln . (11.3.11)
2π 0 Λ π 2π 0 Λ2
Remember that there are N diagrams: one per fermion. Thus, adding this one-loop term to Eqn. (11.3.2) gives
1
m2 − x(1 − x)p2
Z
(1) i iN iN
Γ2 =− 2 − − dx ln . (11.3.12)
g π 2π 0 Λ2

Using Eqn. (11.1.32), we write this using the renormalized coupling constant:
1
m2 − x(1 − x)p2
Z
(1) i iN iN
Γ2 =− 2 − − dx ln . (11.3.13)
gR π 2π 0 µ2

This is a strange equation because gR is a function of the arbitrary scale µ and m is naı̈vely a function of both
µ and g, as in Eqn. (11.1.54). However, m is actually independent of µ:
   
dm π d 1 π N
µ =m 1− µ 2 = m 1 − · = 0, (11.3.14)
dµ N dµ gR N π

where we made use of Eqn. (11.1.33) with g and Λ replaced with gR and µ.
Thus, the µ-dependence must fall out of Eqn. (11.3.13). This happens because of the definition of m in Eqn.
(11.1.54), or more simply as the root of Eqn. (11.1.53):

1 N m2
2 + ln 2 = 0. (11.3.15)
gR 2π µ

Using this, we can write Eqn. (11.3.13) as


1
m2 − x(1 − x)p2
Z
(1) iN iN
Γ2 =− − dx ln . (11.3.16)
π 2π 0 m2
Set
p2 = αm2 , (11.3.17)
so that Z 1
2πi (1)  
Γ =2+ dx ln 1 − αx(1 − x) . (11.3.18)
N 2 0
Using Mathematica, we find
 1/2  −1/2
2πi (1) 4 4 α→4
Γ2 = 2 −1 tan−1 −1 −−−→ 0. (11.3.19)
N α α

Therefore, to one loop, the inverse propagator has a zero at p2 = 4m2 . A zero for the inverse propagator is
equivalent to a pole for the propagator, which occurs at the physical mass. Hence, the physical mass of the σ
particle is twice the mass of the fermion:
mσ = 2m . (11.3.20)

As we said earlier, σ is a bound state of ψ and ψ. Therefore, mσ ≤ 2m. What we have found is that the energy
that binds this state is a ≥ 2-loop effect!
Chapter 12

Problem Set 12

12.1 QED plus Yukawa, P&S 7.3 p.257


Consider a theory of elementary fermions that couple both to QED and to a Yukawa field φ:
Z Z
λ
Hint = d3 x √ φψψ + d3 x eAµ ψγ µ ψ. (12.1.1)
2
(a) Verify that the contribution to Z1 from the vertex diagram with a virtual φ equals the contribution to Z2
from the diagram with a virtual φ. Use dimensional regularization. Is the Ward identity generally true in
this theory?
(b) Now consider the renormalization of the φψψ vertex. Show that the rescaling of this vertex at q 2 = 0 is not
canceled by the correction to Z2 . (It suffices to compute the ultraviolet-divergent parts of the diagrams.) In
this theory, the vertex and field-strength rescaling give additional shifts of the observable coupling constant
relative to its bare value.
SOLUTION: (Thanks to Ka Hei and Bo (’12) for presenting parts (a) and (b) respectively.)

(a) Scalar contribution to Z2 : At one-loop level, the scalar contribution to the electron self-energy is due to
the diagram below.
−k
p-

- - - (12.1.2)
p k p

Following Eqns. 7.15 and 7.16 P&S p.217, the amputated amplitude for this diagram is
2 Z
d4 k i(/

λ k + m0 ) i
−iΣ2 (p) = −i √
2 (2π)4 k 2 − m20 (p − k)2 − m2φ
λ2 1 d4 k k/ + m0
Z Z
= dz . (12.1.3)
2 0 (2π) {z[(p − k) − mφ ] + (1 − z)(k 2 − m20 )}2
4 2 2

As usual, write the denominator as D2 where

D = k 2 − 2zp · k + z 2 p2 − z 2 p2 + zp2 − (1 − z)m20 − zm2φ


= (k − zp)2 + z(1 − z)p2 − (1 − z)m20 − zm2φ . (12.1.4)

Define

` ≡ k − zp, ∆φ ≡ −z(1 − z)p2 + (1 − z)m20 + zm2φ , (12.1.5)

in terms of which D may be written as usual as D = `2 − ∆φ . Now, the numerator is

N = /̀ + z p
/ + m0 ∼ z p
/ + m0 , (12.1.6)

119
120 CHAPTER 12. PROBLEM SET 12

where we drop the linear term in ` since, being odd in `, its integral vanishes. Thus,
1
λ2 d4 `
Z Z
1
− iΣ2 (p) = dz(z p
/ + m0 ) . (12.1.7)
2 0 (2π) (` − ∆φ )2
4 2

Wick rotate and multiply both sides by i to get


1
λ2 d4 `E
Z Z
1
Σ2 (p) = − dz(z p
/ + m0 ) 2 . (12.1.8)
2 0 (2π) (`E + ∆φ )2
4

Method 1 (Pauli-Villars): First, perform the angular integral to get


1 ∞
λ2 `3E d`E
Z Z
Σ2 (p) = − dz(z p
/ + m0 ) . (12.1.9)
16π 2 0 0 (`2E + ∆φ )2

Next, define
ν ≡ `2E , (12.1.10)

so that
1 ∞
λ2
Z Z
ν dν
Σ2 (p) = − dz(z p
/ + m0 ) . (12.1.11)
32π 2 0 0 (ν + ∆φ )2

We can perform the divergent integral:


Z ∞ ∞ Z ∞  ∞
ν dν ν dν ν
=− + = ln(ν + ∆φ ) − . (12.1.12)
0 (ν + ∆φ )2 ν + ∆ φ 0 0 ν + ∆φ ν + ∆φ 0

Replace the scalar propagator by


1 1 1
→ − . (12.1.13)
(p − k)2 − m2φ (p − k)2 − m2φ (p − k)2 − Λ2

The effect of this is


d4 `E d4 `E
Z Z  
1 1 1
2 → − , (12.1.14)
(2π) (`E + ∆φ )2
4 (2π)4 (`2E + ∆φ )2 (`2E + ∆Λ )2

where
Λ→∞
∆Λ = −z(1 − z)p2 + (1 − z)m20 + zΛ2 −−−−→ zΛ2 . (12.1.15)

Therefore,
Z ∞  ∞
ν dν ν + ∆φ ν ν ∆φ Λ→∞ ∆φ
→ ln − + = − ln −−−−→ − ln 2 . (12.1.16)
0 (ν + ∆φ )2 ν + ∆Λ ν + ∆φ ν + ∆Λ 0 ∆Λ zΛ

Then, Eqn. (12.1.11) reads


1
λ2
Z
∆φ
Σ2 (p) = dz(z p
/ + m0 ) ln . (12.1.17)
32π 2 0 zΛ2

We can now take the derivative:


1
λ2 zp
/ + m0 d∆φ
Z  
dΣ2 ∆φ
= dz z ln 2 + . (12.1.18)
dp
/ 32π 2 0 zΛ ∆φ dp
/

2
/ = p2 , we have
Using p
d∆φ
= −2z(1 − z)p
/. (12.1.19)
dp
/
12.1. QED PLUS YUKAWA, P&S 7.3 P.257 121

Thus,
1
λ2 zp
/ + m0
Z  
dΣ2 ∆φ
= dz z ln 2 − 2z(1 − z)p
/ . (12.1.20)
dp
/ 32π 2 0 zΛ ∆φ

Define
∆0φ ≡ ∆φ |p/=m0 = (1 − z)2 m20 + zm2φ . (12.1.21)

Then,
∆0φ
Z 1 
λ2 2z(1 − z 2 )m20

dΣ2
δZ2 = = dz z ln − . (12.1.22)
dp
/ p/=m0 32π 2 0 zΛ2 ∆0φ

Method 2 (Dimensional Regularization):


Return to the expression for Σ2 (p) in Eqn. (12.1.8) and write it in arbitrary spacetime dimension, d:

λ2 1
Z Z d
d `E 1
Σ2 (p) = − dz(z p
/ + m0 ) . (12.1.23)
2 0 (2π)d (`2E + ∆φ )2

The `E integral is computed and expanded for  = 4 − d around zero in Eqn. 7.84 P&S p.250:
Z 1
λ2
 
2
Σ2 (p) = − dz(z p
/ + m0 ) − ln ∆φ − γ + ln(4π) . (12.1.24)
32π 2 0 

Now, differentiate with respect to p


/, recalling Eqn. (12.1.19) for d∆/dp
/, and evaluate at p
/ = m0 :
Z 1  
λ2 2z(1 − z 2 )m20
 
dΣ2 2 0
δZ2 = = dz z − + γ − ln(4π) + ln ∆ φ − . (12.1.25)
dp
/ p/=m0 32π 2 0  ∆0φ

Scalar contribution to Z1 : At one-loop level, the scalar contribution to the electron-photon vertex is due
to the diagram below.
0
Jp
]J
J k+q
]
(12.1.26)
J
p−k 6 J 

q
k




p


We calculated the correction to the fermion-photon vertex due to a virtual Higgs in problem set 10. Thank-
fully, Hint for that problem is identical to this one with the Higgs, h, replaced by the scalar field φ. Therefore,
the result still holds. In problem set 10, what I called F1 should really be called δF1 . Let us collect the
results from that problem set that are useful here. First, Eqn. 1.11 in pset 10, adapted to the notation of
this problem, reads (we distinguish this from ∆φ above by the overline):

∆φ = (1 − z)2 m20 + zm2φ − xyq 2 . (12.1.27)

Here, m0 is the bare electron mass, mφ is the scalar field mass, q is the photon 4-momentum, and x, y, z are
the Feynman parameters. Recall the terms A and B in Eqns. 1.26 and 1.27 of pset 10:

A = − 12 `2 + (1 + z)(3 − z)m20 + xyq 2 , B = −m0 (1 − z 2 ). (12.1.28)

Then, Eqn. 1.31 in pset 10 (adapated) reads

F1 = A + 2m0 B = − 12 `2 + (3 + 2z − z 2 − 2 + 2z 2 )m20 + xyq 2 = − 21 `2 + (1 + z)2 m20 + xyq 2 , (12.1.29)

and
d4 ` − 12 `2 + (1 + z)2 m20 + xyq 2
Z Z
2 2
δF1 (q ) = iλ . (12.1.30)
F (2π)4 (`2 − ∆φ )3
122 CHAPTER 12. PROBLEM SET 12

Wick rotating yields


d4 `E 12 `2E + (1 + z)2 m20 + xyq 2
Z Z
2 2
δF1 (q ) = λ . (12.1.31)
F (2π)4 (`2E + ∆φ )3

∆φ evaluated at q 2 = 0 is the same as ∆φ evaluated at p


/ = m0 , which we called ∆0φ in Eqn. (12.1.21):

∆φ |q2 =0 = (1 − z)2 m20 + zm2φ = ∆0φ . (12.1.32)

Let us now evaluate δF1 at q 2 = 0. In this case, the integrand in Eqn. (12.1.31) is independent of x and y.
Therefore, the x integral just gives 1. After this, y is constrained to be y = 1 − z − x whose minimum value,
−z, occurs at x = 1. However, since y is also constrained to be between 0 and 1, the actual lower bound for
the y integral is 0. On the other hand, the upper bound is 1 − z, occuring when x = 0. Thus, the y integral
gives a factor of (1 − z):
1
d4 `E 12 `2E + (1 + z)2 m20
Z Z
2
δF1 (0) = λ dz (1 − z) . (12.1.33)
0 (2π)4 (`2E + ∆0φ )3

Method 1 (Pauli-Villars): Changing variables to ν = `2E yields


1 ∞ 1 2
λ2 2ν + (1 + z)2 m20 ν
Z Z
δF1 (0) = dz (1 − z) dν . (12.1.34)
16π 2 0 0 (ν + ∆0φ )3

Only the first term in the ν integral requires regularization since the second term is convergent. Let us
simplify the first term:
Z ∞ 1 2 ∞ Z ∞
2 ν dν ν2 +1
ν dν
0 = − 0
0 (ν + ∆φ )3 2
(ν + ∆φ ) 0 2 0 (ν + ∆0φ )2
∞
ν2 1 ∞ dν
 Z
ν
=− + +
(ν + ∆0φ )2 2(ν + ∆0φ ) 0 2 0 ν + ∆0φ
∞
ν2

1 ν
= ln(ν + ∆0φ ) − − . (12.1.35)
2 (ν + ∆0φ )2 2(ν + ∆0φ ) 0

Pauli-Villars regularization amounts to


Z ∞ 1 2 Z ∞ 1 2 1 2 
2 ν dν 2 ν dν 2 ν dν
→ −
0 (ν + ∆0 )3 0 (ν + ∆0φ )3 (ν + ∆Λ )3
∞
1 ν + ∆0φ ν2 ν2

ν ν
= ln − − + +
2 ν + ∆Λ (ν + ∆0φ )2 2(ν + ∆0φ ) (ν + ∆Λ )2 2(ν + ∆Λ ) 0
1 ∆0φ
= − ln 2 . (12.1.36)
2 zΛ

Now, let us compute the convergent integral:

∞ 1 ∞
Z ∞  Z ∞
ν dν ν  dν 1 = 1 .

0 = − 0
+
0 = − 0 (12.1.37)
0 (ν + ∆φ ) 3 2
2(ν + ∆φ ) 0
 2 0 (ν + ∆φ )2 2(ν + ∆φ ) 0 2∆0φ

Thus, we get
λ2 1 ∆0φ (1 + z)2 m20
Z  
1
δF1 (0) = dz (1 − z) − ln 2 + . (12.1.38)
16π 2 0 2 zΛ 2∆0φ

Rewrite this as
λ2 1 ∆0φ (1 + z)(1 − z 2 )m20
Z  
δF1 (0) = dz −(1 − z) ln 2 + . (12.1.39)
32π 2 0 zΛ ∆0φ
12.1. QED PLUS YUKAWA, P&S 7.3 P.257 123

Method 2 (Dimensional Regularization):


Now, return to the expression for δF1 (0) in Eqn. (12.1.33) and generalize to arbitrary spacetime dimension,
d:
d `E d−2
Z 1 Z d 2 2 2
d `E + (1 + z) m0
δF1 (0) = λ2 dz(1 − z) . (12.1.40)
0 (2π)d (`2E + ∆0φ )3

The term with `2E in the numerator requires some explanation. Recall how that term arose in the first place.
It came from simplifying the term /̀γ µ /̀ in δF1 :
2
/̀γ µ /̀ = 2`µ `ν γν − γ µ /̀ ' 24 `2 g µν γν − γ µ `2 = − 12 `2 = 21 `2E . (12.1.41)

We used the fact that when integrating over `, we can make the replacement `µ `ν → 14 `2 g µν . In d dimensions,
this replacement turns into `µ `ν → d1 `2 g µν since gµν g µν = d. Therefore,
2
/̀γ µ /̀ = 2`µ `ν γν − γ µ /̀ ' d2 `2 g µν γν − γ µ `2 = 2−d 2
d ` = d−2 2
d `E . (12.1.42)

Using Eqns. 7.85 and 7.86 P&S p.251, we get


Z 1 d 2 2 d
 Γ(2 − 2 ) 0  d2 −2 (1 + z) m0 Γ(3 − 2 ) 0  d2 −3
 
2 d−2 1 d
δF1 (0) = λ dz(1 − z) ∆φ + ∆φ . (12.1.43)
0  (4π)d/2 2 Γ(3)
d (4π)d/2 Γ(3)

Expanding this in  = 4 − d around  = 0 gives


Z 1
λ2 (1 + z)2 m20
 
2
δF1 (0) = dz(1 − z) − ln ∆0 − γ + ln(4π) − 1 + . (12.1.44)
32π 2 0  ∆0

Rewrite this as
1
λ2 (1 + z)(1 − z 2 )m20
Z    
2 0
δF1 (0) = dz (1 − z) − γ + ln(4π) − (1 − z) ln ∆φ + . (12.1.45)
32π 2 0  ∆0

Proof of Z1 = Z2 :
Method 1 (Pauli-Villars): Add Eqns. (12.1.22) and (12.1.39):

∆0φ
Z 1 
λ2 (1 − z)(1 − z 2 )m20

δZ2 + δF1 (0) = dz −(1 − 2z) ln + . (12.1.46)
32π 2 0 zΛ2 ∆0φ

Rewrite the first term as


1 ∆0φ 1 ∆0φ
Z Z  
− dz (1 − 2z) ln = dz (1 − 2z) ln z − ln 2 . (12.1.47)
0 zΛ2 0 Λ

Integrate the first term in Eqn. (12.1.47) by parts:


Z 1 Z 1 Z 1
((1 1
dz (1 − 2z) ln z = (
z(1(−(z) ln z|0 −
( ( dz z(1 − z) = − dz (1 − z). (12.1.48)
0 0 z 0

Now, integrate the second term in Eqn. (12.1.47) by parts:


1 Z 1
∆0φ 0φ −2(1 − z)m20 + m2φ
Z 1

− dz (1 − 2z) ln 2 = −z(1 − −

z)ln + dz z(1 z)
Λ Λ2 0 ∆0φ
 
0 0

(1 − z)2 m20 + zm2φ − (1 − z)2 m20 − 2z(1 − z)m20


Z 1
= dz (1 − z)
0 ∆0φ
Z 1 
(1 − z)(1 − z 2 )m20

= dz 1 − z − . (12.1.49)
0 ∆0φ
124 CHAPTER 12. PROBLEM SET 12

The first term above, (1 − z), cancels Eqn. (12.1.48) simplifying Eqn. (12.1.47) into
1 ∆0φ 1
(1 − z)(1 − z 2 )m20
Z Z
− dz (1 − 2z) ln =− dz , (12.1.50)
0 zΛ2 0 ∆0φ

which indeed cancels the last term in Eqn. (12.1.46) leaving

δZ2 + δF1 (0) = 0 . (12.1.51)

Since Γµ = Z1−1 γ µ , we have δZ1 = −δF1 (0) and so Eqn. (12.1.51) implies δZ1 = δZ2 . Since both Z1 and Z2
are 1 at tree level, we have
Z1 = Z2 . (12.1.52)

Method 2 (Dimensional Regularization): Add Eqns. (12.1.25) and (12.1.45):

1 
λ2 (1 − z)(1 − z 2 )m20
Z    
2  0
δZ2 + δF1 (0) = dz (1 − 2z) − γ + ln(4π) − (1 − z) − (1 − 2z) ln ∆φ +
32π 2 0   ∆0φ

1
λ2 (1 − z)(1 − z 2 )m20
Z  
= dz −(1 − z) − (1 − 2z) ln ∆0φ + . (12.1.53)
2
32π 0 ∆0φ

Note that the first term, containing the divergence, vanishes because the integrand is proportional to (1−2z),
whose integral vanishes. Meanwhile, the integral of (1 − 2z) ln ∆0φ does not vanish because ∆0φ depends on z.
As in the previous method, integrating the log term by parts shows that Eqn. (12.1.53) vanishes.
Ward-Takahashi identity:
The proof of the Ward-Takahashi identity will work as long as each vertex involves a ψ and a ψ. We are still
attaching a photon line to a fermion line or loop but now the lines that are already attached to the fermion
can be either photons are scalars. All this does is get rid of some γ µ ’s and some minus signs, which does not
invalidate the proof. This certainly ensures gauge invariance classically, but since the proof goes through,
gauge invariance holds to all orders of quantum correction.

(b) We must compare the one-loop corrections to the fermion self-energy and the fermion-scalar √ vertex due to
the photon and the scalar. At tree level, the fermion-scalar vertex comes with the factor, −iλ/ 2. Write the
exact vertex as  
1 λ
V ≡ fermion-scalar vertex = 0 −i √ , (12.1.54)
Z1 2

so that Z10 is 1 at tree level and


δZ10 = −δV. (12.1.55)

We are trying to show that


Z10 6= Z2 . (12.1.56)

Photon contribution to Z2 : The one-loop photon contribution to the fermion self-energy is

p−k
-
- - - (12.1.57)
p k p

Define the two quantities,

∆γ = −z(1 − z)p2 + (1 − z)m20 + zm2γ , (12.1.58a)


∆0γ = (1 − z) 2
m20 + zm2γ . (12.1.58b)
12.1. QED PLUS YUKAWA, P&S 7.3 P.257 125

According to Eqn. 7.16 P&S p.217, the above diagram gives

d4 k µ i(/ −i
Z
2 k + m0 )
− iΣ2 = (−ie) 4
γ 2 2 γµ . (12.1.59)
(2π) k − m0 (p − k)2 − m2γ

Method 1 (Pauli-Villars): This integral is worked out in Eqn. 7.31 P&S p.221:

e2 1 ∆0γ 2z(2 − z)(1 − z)m20


Z  
δγ Z2 = dz z ln 2 + . (12.1.60)
8π 2 0 zΛ ∆0γ

Method 2 (Dimensional Regularization): In d spacetime dimensions, the only thing that changes in
Eqn. (12.1.59) is the integration measure: (2π)−4 d4 k → (2π)−d dd k.
We will need some gamma matrix identities. By repeated use of the Clifford algebra, {γ µ , γ ν } = 2g µν , one
can prove the following identities:

γ µ γµ = d, (12.1.61a)
γ γ γµ = (2 − d)γ ν ,
µ ν
(12.1.61b)
γ µ γ ν γ ρ γµ = 4g νρ + (d − 4)γ ν γ ρ . (12.1.61c)

Thus,
1
dd k (2 − d)/
Z Z
k + dm0
Σ2 = −ie2 dz (12.1.62)
0 (2π)d {z[(p − k)2 − m2γ ] + (1 − z)(k 2 − m20 )}2

As usual, the denominator can be written as D2 , where

D = `2 − ∆γ , ` = k − zp. (12.1.63)

Then, dropping linear terms in ` and Wick rotating, we get


1
dd `E (2 − d)z p
/ + dm0
Z Z
2
Σ2 = e dz . (12.1.64)
0 (2π)d (`2E + ∆γ )2

Again, using Eqn. 7.84 P&S p.250, we get


Z 1 
e2

2  
Σ2 = 2
dz − ln ∆γ − γ + ln(4π) −2z p / + 4m0 + (z p / − m0 )
16π 0 
Z 1 
e2
 
2
= dz − γ + ln(4π) − ln ∆γ (−2z p
/ + 4m 0 ) + 2(z p
/ − m 0 ) . (12.1.65)
16π 2 0 

Now, we take the derivative with respect to p


/ and evaluate at p
/ = m0 to get
Z 1  
e2 2z(1 − z)(2 − z)m20
 
dΣ2 2 0
δγ Z2 = = dz z − + γ − ln(4π) + 1 + ln ∆γ + . (12.1.66)
dp
/ p/=m0 8π 2 0  ∆0γ

Photon contribution to Z10 : Next, we calculate the diagram


0
Jp
]J
J k+q
]
(12.1.67)
J
p−k 6J 
q




p


126 CHAPTER 12. PROBLEM SET 12

Now, we calculate the electron-scalar vertex. The diagram is the same as for the electron-photon vertex (e.g.
P&S p.189), but with the photon with momentum q replaced by a scalar or both photons replaced by a
scalar. Let us first calculate the more difficult case where the loop includes a photon. The amplitude, Mγ ,
of this diagram is
0
d4 k
 
−igµν
Z
0 µ i(/
k + m0 ) λ i(/ k + m0 )
iMγ = ū(p )(−ieγ ) 2 −i √ (−ieγ ν )u(p)
(2π) 4 2 2 02
(p − k) − mγ k − m0 2 k 2 − m20
0
d4 k ū(p0 )γ µ (/
  Z Z
λ 2 k + m0 )(/
k + m0 )γµ u(p)
= −i √ (−2ie ) 4 {x(k 2 − m ) + y[(k + q)2 − m2 ] + z[(p − k)2 − m2 ]}3
2 . (12.1.68)
2 F (2π) 0 0 γ

Let us first simplify the denominator by writing it as D3 , where


D = k 2 + 2(yq − zp) · k + (yq − zp)2 − (yq − zp)2 + yq 2 + zp2 − (x + y)m20 − zm2γ
= (k + yq − zp)2 + y(1 − y)q 2 + 2yzp · q + z(1 − z)p2 − (1 − z)m20 − zm2γ . (12.1.69)

Using the identities p2 = m2 and 2p · q = −q 2 (Eqn 1.9 of problem set 10),


D = `2 − ∆γ , (12.1.70)

where
` ≡ k + yq − zp, ∆γ ≡ (1 − z)2 m20 + zm2γ − xyq 2 . (12.1.71)

When d = 4, the identities (12.1.61a), (12.1.61b) and (12.1.61c) read


γ µ γµ = 4, (12.1.72a)
µ ν ν
γ γ γµ = −2γ , (12.1.72b)
γ µ γ ν γ ρ γµ = 4g νρ , (12.1.72c)

Using these identities, we may simplify the numerator to


0
N = ū(p0 )[4k 0 · k − 2m0 (/
k + k/) + 4m20 ]u(p). (12.1.73)

Let us simplify this one term at a time:


4k 0 · k = 4[` + (1 − y)p0 − xp] · [` − yp0 + (1 − x)p]
= 4`2 + 4(1 − x − y + 2xy)p0 · p − 4(y − y 2 + x − x2 )m20
= 4`2 + 4(1 − x − y + 2xy − y + y 2 − x + x2 )m20 − 2(1 − x − y + 2xy)q 2
= 4`2 + 4[(1 − z)2 − 1 + 2z]m20 − 2(z + 2xy)q 2
= 4`2 + 4z 2 m20 − 2(z + 2xy)q 2 . (12.1.74)

Note that we dropped linear terms in ` and made use of the identity
2p0 · p = 2m20 − q 2 . (12.1.75)

Using the Dirac equations and taking the spinors for granted, the middle term in Eqn. (12.1.59) gives
0 0
k + k/) = −2m0 [2/̀ + (1 − 2y)p
−2m0 (/ / + (1 − 2x)p
/]
= −2m0 (1 − 2y + 1 − 2x)m0
= −4zm20 . (12.1.76)

Putting these pieces together and taking the spinors for granted, Eqn. (12.1.73) simplifies to
N = 4`2 + 4(1 − z + z 2 )m20 − 2(z + 2xy)q 2 . (12.1.77)

Everything in iMγ except the usual tree level value −iλ/ 2 gives δV . Hence, evaluating at q 2 = 0, we have
d4 ` `2 + (1 − z + z 2 )m20
Z Z
δγ Z10 = −δγ V (0) = 8ie2
F (2π)4 (`2 − ∆0γ )3
Z 1 Z 4
d `E −`2E + (1 − z + z 2 )m20
= 8e2 dz (1 − z) . (12.1.78)
0 (2π)4 (`2E + ∆0γ )3
12.1. QED PLUS YUKAWA, P&S 7.3 P.257 127

Method 1 (Pauli-Villars): As usual, define ν = `2E so that


1
e2 −ν 2 + (1 − z + z 2 )m20 ν
Z Z
δγ Z10 = dz (1 − z) dν . (12.1.79)
2π 2 0 (ν + ∆0γ )3

Using Eqns. (12.1.36) and (12.1.37), we get

e2 1 ∆0γ (1 − z + z 2 )m20
Z  
δγ Z10 = dz (1 − z) ln 2 + . (12.1.80)
2π 2 0 zΛ 2∆0γ

Method 2 (Dimensional Regularization): In d dimensions, Eqn. (12.1.73) becomes


0 0
N = ū(p0 )[4k 0 · k + (d − 4)/ k + k/) + dm20 ]u(p).
k k/ + (2 − d)m0 (/ (12.1.81)
0
Let us simplify the extra k/ k/ term using the Dirac equations and taking the spinors for granted:
0 0 0
k/ k/ = [/̀ + (1 − y)p /][/̀ − y p
/ − xp / + (1 − x)p
/]
0
= [/̀ − xp
/ + (1 − y)m0 ][/̀ − y p
/ + (1 − x)m0 ]
0
= `2 + xy p
/p
2
/ + [(1 − x)(1 − y) − x(1 − x) − y(1 − y)]m0
= `2 + xy(m20 − q 2 ) + [1 − x − y + xy − x + x2 − y + y 2 ]m20
= `2 + [1 − 2x − 2y + (x + y)2 ]m20 − xyq 2
= `2 + [(1 − z)2 − 1 + 2z]m20 − xyq 2
= `2 + z 2 m20 − xyq 2 , (12.1.82)

where, to get the fourth line, we used


2p0 · p = 2m20 − q 2 , (12.1.83)

to derive the identity


0 0 0 2 2 2 2 2
/p/ = 2p · p − p
p /p / = 2m0 − q − m0 = m0 − q . (12.1.84)

Note that
0
4k 0 · k + (d − 4)/
k k/ = d`2 + dz 2 m20 − (2z + dxy)q 2 . (12.1.85)

Thus, the numerator is


N = d`2 + [dz 2 + 2(2 − d)z + d]m20 − (2z + dxy)q 2 , (12.1.86)

and, after evaluating at q 2 = 0, we have


Z 1
dd ` d`2 + [dz 2 + 2(2 − d)z + d]m20
Z
0 2
δγ Z1 = 2ie dz (1 − z) . (12.1.87)
0 (2π)d (`2 − ∆0γ )3

The second piece is convergent at d = 4 and gives precisely the same contribution as the second (non-log)
piece in Eqn. (12.1.80). Only the first term is actually divergent and requires regularization. Thus, we write
1 1
dd `E `2E e2 (1 − z)(1 − z + z 2 )m20
Z Z Z
δγ Z10 = −2de 2
dz (1 − z) + dz . (12.1.88)
0 (2π)d (`2E + ∆0γ )3 4π 2 0 ∆0γ
| {z }
A

Using Eqn. 7.86 P&S p.251, we find, after expanding around  = 4 − d ≈ 0,


1
d Γ(2 − d2 ) 0  d2 −2
Z
1
A = −2de2 dz (1 − z) d/2
∆γ
0 (4π) 2 Γ(3)
1
e2
Z  
2 0
= dz (1 − z) − + γ − ln(4π) + 1 + ln ∆γ . (12.1.89)
2π 2 0 
128 CHAPTER 12. PROBLEM SET 12

Therefore, we have

1
e2 (1 − z + z 2 )m20
Z   
2
δγ Z10 = 0
dz (1 − z) − + γ − ln(4π) + 1 + ln ∆γ + . (12.1.90)
2π 2 0  ∆0γ

Scalar contribution to Z10 : Next, we calculate the diagram


0
Jp
]J
J k+q
]
(12.1.91)
J
p−k 6 J 
q




p


2
Compared to Eqn. (12.1.68), the factor (−ie)2 is replaced with −i √λ2 and there is an extra minus sign
because the scalar propagator’s numerator is i whereas that of the photon is −igµν . We drop the two factors
of γ associated with the fermion-photon vertices. Also, mγ is replaced with mφ , or ∆γ with ∆φ . Finally, we
simplify the numerator:
0
N = (/
k + m0 )(/
k + m0 )
0 0
= [/̀ + (1 − y)p / + m0 ][/̀ − y p
/ − xp / + (1 − x)p
/ + m0 ]
0
= [/̀ − xp
/ + (2 − y)m0 ][/̀ − y p
/ + (2 − x)m0 ]
0
= `2 + xy p
/p
2 2 2
/ − x(2 − x)m0 − y(2 − y)m0 + (2 − x)(2 − y)m0
= `2 + xy(2p · p0 − m20 ) + (−2x + x2 − 2y + y 2 + 4 − 2x − 2y + xy)m20
= `2 + xy(m20 − q 2 ) + (x2 + xy + y 2 + 4 − 4x − 4y)m20
= `2 + [(x + y)2 + 4(1 − x − y)]m20 − xyq 2
= `2 + [(1 − z)2 + 4z]m20 − xyq 2
= `2 + (1 + z)2 m20 − xyq 2 . (12.1.92)

Therefore,
d4 ` ū(p0 )[`2 + (1 + z)2 m20 − xyq 2 ]u(p)
  Z Z
λ
iMφ = −i √ (iλ2 ) . (12.1.93)
2 F (2π)4 (`2 − ∆φ )3

Then,
1
d4 `E `2E − (1 + z)2 m20
Z Z
δφ Z10 = λ2 dz (1 − z) . (12.1.94)
0 (2π)4 (`2E + ∆0φ )3

Method 1 (Pauli-Villars): As usual, define ν = `2E so that


1 ∞
λ2 ν 2 − (1 + z)2 m20 ν
Z Z
δγ Z10 = dz (1 − z) dν . (12.1.95)
16π 2 0 0 (ν + ∆0φ )3

Again, using Eqns. (12.1.35) and (12.1.36), we get

1 (φ)
λ2 (1 + z)2 m20
Z  
∆0
δφ Z10 =− dz (1 − z) ln + . (12.1.96)
16π 2 0 zΛ2 2∆0φ

Rewrite this as
λ2 1 ∆0φ (1 + z)(1 − z 2 )m20
Z  
δφ Z10 =− dz (1 − z) ln 2 + . (12.1.97)
16π 2 0 zΛ 2∆0φ
12.1. QED PLUS YUKAWA, P&S 7.3 P.257 129

Method 2 (Dimensional Regularization): Again, only the first term in Eqn. (12.1.94) is divergent and
needs regularization:
Z 1 Z d Z 1
0 2 d `E `2E λ2 (1 + z)(1 − z 2 )m20
δφ Z1 = λ dz (1 − z) − dz . (12.1.98)
0 (2π)d (`2E + ∆0φ )3 32π 2 0 ∆0φ

Again, using Eqn. 7.86 P&S p.251 and expanding around  = 4 − d ≈ 0, we get

1
λ2 (1 + z)(1 − z 2 )m20
Z    
2 1
δφ Z10 =− 0
dz (1 − z) − + γ − ln(4π) + − ln ∆φ + . (12.1.99)
16π 2 0  2 2∆0φ

Proof that Z10 6= Z2 :


Method 1 (Pauli-Villars): Subtract Eqn. (12.1.60) from (12.1.80):

∆0γ
Z 1 
e2 2(1 − z)2 (1 − 2z)m20

0
δγ Z1 − δγ Z2 = dz (4 − 5z) ln 2 + . (12.1.100)
8π 2 0 zΛ ∆0γ

The usual trick would be to integrate the first term by parts. Separating out a ln z term, we get
Z 1
1 1 1 1 1
Z Z
1 (( 1
− − 2(z(8
dz(4 − 5z) ln z = ( ( − 5z) ln z 0 +
( ( ( dz z(8 − 5z) = dz(8 − 5z). (12.1.101)
0 2 0 z 2 0

The rest of the first term of Eqn. (12.1.100) gives


1 ∆0γ ∆0γ 1 1 1 −2(1 − z)m20 + m2φ
Z Z
1
dz (4 − 5z) ln 2 = z(8 − 5z) ln 2 −
dz z(8 − 5z)
0 Λ 2 Λ 0 2 0 ∆0γ
3 m2γ 1 1 (1 − z 2 )m20
Z  
= ln 2 − dz(8 − 5z) 1 − , (12.1.102)
2 Λ 2 0 ∆0γ

where the same trick was used here as in Eqn. (12.1.38). As usual, the first term above cancels Eqn. (12.1.73).
The rest is finite. The remainder is

m2γ
Z 1
3e2 (4 + z)(1 − z)2 m20
 
δγ Z10 − δγ Z2 = ln 2 + dz 6= 0 . (12.1.103)
16π 2 Λ 0 ∆0γ

The scalar contribution also does not cancel. Subtract Eqn. (12.1.22) from Eqn. (12.1.97):

λ2 1 ∆0φ (1 − z)(1 − z 2 )m20


Z  
δφ Z10 − δφ Z2 = − dz (2 − z) ln 2 + . (12.1.104)
32π 2 0 zΛ ∆0φ

The analog of Eqn. (12.1.102) in this case is


1 ∆0φ 3 m2φ 1
(1 − z 2 )m20
Z Z  
1
dz(2 − z) ln 2 = ln 2 − dz(4 − z) 1 − . (12.1.105)
0 Λ 2 Λ 2 0 ∆0φ

The final result is

m2φ
Z 1
3λ2 (2 − z)(1 − z 2 )m20
 
δφ Z10 − δφ Z2 = − ln + dz 6= 0 . (12.1.106)
64π 2 Λ2 0 ∆0φ

As stated in the problem, these non-cancellations imply extra contributions to the renormalization of λ and
e. One might worry about whether or not the cancellations that occurred for the fermion-photon vertex, due
to gauge-invariance, still hold after renormalization. There is no need to worry because those cancellations
did not depend on the specific value of the coupling constants.
130 CHAPTER 12. PROBLEM SET 12

Method 2 (Dimensional regularization):


The finite pieces in the differences are precisely the same as were found in Eqns. (12.1.103) and (12.1.106).
Therefore, we will not bother to calculate those. We will just focus on the divergent pieces. From Eqns.
(12.1.25), (12.1.66), (12.1.90) and (12.1.99), we find

e2 e2
δγ Z10 = − , δγ Z2 = − , (12.1.107a)
2π 2  4π 2 
λ2 λ2
δφ Z10 = , δφ Z2 = − . (12.1.107b)
16π 2  32π 2 

Therefore,
e2
δγ Z10 − δγ Z2 = − 6= 0 , (12.1.108)
4π 2 

and
3λ2
δφ Z10 − δφ Z2 = 6= 0 . (12.1.109)
32π 2 

Aside: In part (a), we found that δφ Z1 − δφ Z2 = 0. This is true regardless of the regularization procedure. Even
though both integrals are divergent, and therefore ill-defined, one can still prove that they are equal without
actually evaluating the integrals and therefore without recourse to a regularization procedure. Reproduced below
are the integral expressions for δφ Z1 and δφ Z2 :
1
λ2 (1 − z)`2E 2(1 + z)(1 − z 2 )m20
Z Z  
δφ Z1 = − dz d¯4 `E
+ ,
0 2 (`2E + ∆0φ )3 (`2E + ∆0φ )3
λ2 1 4z(1 − z 2 )m20
Z Z  
4 z
δφ Z2 = − dz d¯ `E + .
2 0 (`2E + ∆0φ )2 (`2E + ∆0φ )3

Note that the expression for δφ Z2 is found by differentiating the integral expression for Σ2 in Eqn. (12.1.8) before
actually evaluating the integral explicitly. One procedure for showing the equality of these two expressions is to
integrate the first term in δφ Z1 by parts with respect to `E , and use the integral
Z
1 1
d¯4 `E 2 0 = , (12.1.110)
(`E + ∆φ ) 3 32π 2 ∆0φ

to write
`2 ∆0φ
Z Z Z
4 4 1
d¯ `E 2 E 0 3 = d¯ `E 2 − d¯4 `E . (12.1.111)
(`E + ∆φ ) (`E + ∆0φ )2 (`2E + ∆0φ )3
Finally, in δφ Z1 − δφ Z2 , integrate the term involving (`2E + ∆0φ )−2 by parts with respect to z. Using the same
sort of trick as was used in the derivation of Eqn. (12.1.49), for example, one can show that the result vanishes.

12.2 Massive Axial Anomaly, Zee IV.7.4 p.279


Take the Pauli-Villars regulated ∆λµν (k1 , k2 ) and contract it with qλ . The analog of the trick in chapter II.7 is to
write /qγ 5 in the second term as [2M + (p
/ − M ) − (p / − /q + M )]γ 5 . Now you can freely shift integration variables.
Show that
qλ ∆λµν (k1 , k2 ) = −2M ∆µν (k1 , k2 ), (12.2.1)
where
d4 p
Z  
µν 3 5 1 ν 1 µ 1
∆ (k1 , k2 ) = (−1)i tr γ γ γ + {µ, k1 ↔ ν, k2 }. (12.2.2)
(2π)4 / − /q − M p
p / − k/1 − M p /−M

Evaluate ∆µν and show that ∆µν goes as 1/M in the limit M → ∞ and so the right hand side of Eqn. (12.2.1)
goes to a finite limit. The anomaly is what the regulator leaves behind as it disappears from the low energy
spectrum: It is like the smile of the Cheshire cat. [We can actually argue that ∆µν goes as 1/M without doing
12.2. MASSIVE AXIAL ANOMALY, ZEE IV.7.4 P.279 131

a detailed calculation. By Lorentz invariance and because of the presence of γ 5 , ∆µν must be proportional to
µνλρ k1λ k2ρ , but by dimensional analysis, ∆µν must be some constant times µνλρ k1λ k2ρ /M . You might ask why
we can’t use something like 1/(k12 )1/2 instead of 1/M to make the dimension come out right. The answer is
that from your experience in evaluating Feynman diagrams in (3 + 1)-dimensional spacetime you can never get
a factor like 1/(k12 )1/2 .]

SOLUTION:

We take {µ, k1 ↔ ν, k2 } for granted. We have

d4 p
Z  
5 1 1 µ1 1 1 1
qλ ∆λµν = (−1)i3 tr /q γ γ ν
γ − q
/ γ 5
γ ν
γ µ
. (12.2.3)
(2π)4 / − /q p
p / − k/1 p / / − /q − M p
p / − k/1 − M p /−M

Let A and B be the first and second terms, respectively. Rewrite the first term as

d4 p
Z  
  5 1 ν 1 µ1
A=i tr p/ − (p −
/ / q ) γ γ γ
(2π)4 / − /q p
p / − k/1 p /
4
Z  
d p 5 1 ν 1 µ 5 ν 1 µ1
=i tr γ γ γ +γ γ γ . (12.2.4)
(2π)4 / − /q p
p / − k/1 / − k/1 p
p /

Rrewrite the second term as


d4 p
Z  
  5 1 ν 1 µ 1
B = −i tr 2M + (p / − M ) − (p/ − /q + M ) γ γ γ
(2π)4 / − /q − M p
p / − k/1 − M p /−M
4
Z 
d p 1 1 1 1 1
= −i 4
tr 2M γ 5 γν γµ + γ5 γν γµ
(2π) / − /q − M p
p / − k/1 − M p /−M / − /q − M p
p / − k/1 − M

1 1
+ γ5γν γµ . (12.2.5)
/ − k/1 − M p
p /−M

Note that the first term in Eqn. (12.2.5) is precisely −2M ∆µν . Therefore, reintroducing the terms implied by
{µ, k1 ↔ ν, k2 }, we get

d4 p
Z 
λµν µν 1 1 1 1
qλ ∆ + 2M ∆ = i 4
tr γ 5 γν γµ − γ5 γν γµ
(2π) / − /q p
p / − k1
/ / − /q − M p
p / − k1 − M
/
1 1 1 1
+ γ5γν γµ − γ5γν γµ
/ − k1 p
p / / / − k1 − M p
p / /−M
1 1 1 1
+ γ5 γµ γν − γ5 γµ γν
p −
/ / /q p − /
k 2 p −
/ / q − M p
/ − /
k 2−M

1 1 1 1
+ γ5γµ γν − γ5γµ γν . (12.2.6)
/ − k/2 p
p / / − k/2 − M p
p /−M

These terms cancel in pairs after we shift the integration variable. We will have to justify this shift, but let us
first show that it produces the appropriate cancellations. For example, define

d4 p
Z  
5 1 ν 1 µ 5 ν 1 ν1
C≡i tr γ γ γ + γ γ γ
(2π)4 / − /q p
p / − k/1 / − k/2 p
p /
4
Z  
d p 5 1 ν 1 µ 5 1 ν1 µ
=i tr γ γ γ −γ γ γ . (12.2.7)
(2π)4 / − /q p
p / − k/1 / − k/2 p
p /

Then, making the shift p → p + k1 in the first integral will turn it into the second integral, but with a plus sign.
Hence, the two terms cancel and C = 0. This can be done with all the appropriate pairs in Eqn. (12.2.6). The
example above was the cancellation of the first and seventh terms. The second and eighth terms cancel, the third
and fifth, and the fourth and sixth.
Now, we just need to justify our ability to shift the integration variable. Note that the first and second
terms are shifted in the same way in order to cancel the seventh and eighth terms, respectively. This means that
132 CHAPTER 12. PROBLEM SET 12

Pauli-Villars regulated terms cancel in pairs. The reason why we are allowed to shift the integration variable is
because Pauli-Villars regularization makes the integrand decay faster at large p than it would otherwise do.
However, we do still have to make sure that the integrand decays sufficiently quickly as to justify the shift.
How fast would it have to decay? Well, it would have to decay faster than the rate at which the surface area
of the 3-sphere in four spacetime dimensions grows, which is ∼ p3 . Therefore, the integrand must decay faster
than p−3 . This is really not that obvious because, as written, each term in the integrand decays as p−2 . Naı̈vely,
Pauli-Villars regularization might make this decay faster, say as p−3 , but even that would be insufficient. It turns
out that the O(p−3 ) term vanishes! Let us see how this works with the first line of Eqn. (12.2.6). Define

1 1 1 1
I µν = γ 5 γν γµ − γ5 γν γµ. (12.2.8)
p − q
/ / / p − /
k 1 p −
/ / q − M p
/ − /
k 1−M

Let us expand this for large p:

1 1 ν1 1 1 1 1 1
I µν = γ 5 qγ
/ /
k
γµ − γ5 q +M
γν /
γµ
p
/1− p p / 1 − p1 p
/1− p/
/ 1 − 1 +M
p k
p
/ / / /
       
1 q ν1 k1 µ 1 q M 1 k1 M
= γ5 1+ γ 1+ γ − γ5 1+ + γν 1+ + γµ
p
/ p p
/ p p
/ p p
/ p
/ p p
/
51 ν1 M µ 51 M ν1 µ −4
= −γ γ γ −γ γ γ + O(p )
p
/ p / p/ p
/ p
/ p
/
 
M 5 1 ν µ 1
=− 2γ γ γ + γ ν γ µ + O(p−4 )
p p
/ p
/
ν
2M p
= − 4 {γ 5 , γ µ } + O(p−4 )
p
= 0 + O(p−4 ). (12.2.9)
−1
We used the fact that {p / , γ ν } = p−2 {p
/, γ ν } = 2p−2 pν and {γ 5 , γ µ } = 0. We conclude that the integrand of
−4
Eqn. (12.2.6) decays at least as fast as p and hence we need not worry about any boundary terms incurred by
shifting the variable of integration.
Therefore, we have finally proven that

qλ ∆λµν (k1 , k2 ) = −2M ∆µν (k1 , k2 ) . (12.2.10)

Once again, taking {µ, k1 ↔ ν, k2 } for granted, we write ∆µν as

d4 p tr[γ 5 (p
/ − /q + M )γ ν (p
/ − k/1 + M )γ µ (p
/ + M )]
Z Z
µν
∆ = 2i . (12.2.11)
F (2π)4 {x(p2 − M 2 ) + y[(p − k1 )2 − M 2 ] + z[(p − q)2 − M 2 ]}3

The denominator is written D3 where

D = p2 − 2(yk1 + zq) · p + (yk1 + zq)2 − (yk1 + zq)2 + yk12 + zq 2 − M 2


= (p − yk1 − zq)2 + y(1 − y)k12 − 2yzk1 · q + z(1 − z)q 2 − M 2 . (12.2.12)

Define

` ≡ p − yk1 − zq, ∆ ≡ M 2 − y(1 − y)k12 − z(1 − z)q 2 + 2yzk1 · q, (12.2.13)

so that we can write


D = `2 − ∆. (12.2.14)
Let us simplify the numerator. This is simpler than it seems. Because of the presence of the γ 5 , all combinations
of the successive γ matrices containing an odd number of gamma matrices or just two gamma matrices vanish!
Therefore, only the terms containing four gamma matrices matter:

N µν = M tr γ 5 [(p ν
/ − /q)γ (p
µ
/ − k/1 )γ + (p
ν µ
/ − /q)γ γ p
ν µ
/ − k/1 )γ p
/ + γ (p /]. (12.2.15)
12.2. MASSIVE AXIAL ANOMALY, ZEE IV.7.4 P.279 133

Again, because of the γ 5 , we are free to anticommute all of the gamma matrices. The anti-commutator terms,
being proportional to the metric leaves γ 5 and two gamma matrices whose trace vanishes. Thus,

N µν = M tr γ 5 γ µ γ ν [(p
/ − /q)(p
/ − k/1 ) − (p
/ − /q)p / − k/1 )p
/ + (p /]
= M tr γ 5 γ µ γ ν [p2 − p
/k/1 − /qp
2
/ + /qk/1 − p + /qp
2
/ + p − k/1 p
/] (12.2.16)

/ terms obviously cancel. The p2 terms, being proportional to identity just leave the trace of γ 5 γ µ γ ν , which
The /qp
/ = 2p · k1 also vanishes. This leaves only
/k/1 + k/1 p
vanishes. Similarly, the term involving p

N µν = 4iM µνρσ qρ k1σ . (12.2.17)

Thus, since we are allowed to freely shift the integration variable from p to `,

d4 `
Z Z Z
µν µνρσ 1 iM µνρσ 1
∆ = −8M  qρ k1σ 4 (`2 − ∆)3
= 2
 qρ k1σ . (12.2.18)
F (2π) 4π F ∆

In the large M limit, ∆ ≈ M 2 . Then,


Z
i i
∆µν = µνρσ qρ k1σ = µνρσ qρ k1σ . (12.2.19)
4π 2 M F 8π 2 M

Plug in q = k1 + k2 and discard the term k1ρ k1σ , which is symmetric in ρ ↔ σ whereas µνρσ is anti-symmetric.
Do not forget the extra {µ, k1 ↔ ν, k2 } term:

i i
∆µν = (µνρσ k2ρ k1σ + νµρσ k1ρ k2σ ) = − 2 µνρσ k1ρ k2σ . (12.2.20)
8π 2 M 4π M

Finally,
i µνρσ
qλ ∆λµν = −2M ∆µν =  k1ρ k2σ , (12.2.21)
2π 2
which agrees with Eqn. 9 Zee p.275.
134 CHAPTER 12. PROBLEM SET 12
Chapter 13

Final Exam

13.1 Scalar Electrodynamics


Consider minimally coupled scalar electrodynamics in 3 + 1 spacetime dimensions, whose action is given by
Z  
4 µ
† 1 µν 2 † † 2
S1 = d x D φ Dµ φ − Fµν F − m φ φ − g(φ φ) , (13.1.1)
4

where Dµ φ = (∂µ + ieAµ )φ.

(a) Derive and explain momentum-space Feynman rules for this theory.

(b) Draw all superficially divergent diagrams, and indicate their degree of divergence.

(c) Set g = 0 in the classical theory, and demonstrate conclusively using Feynman diagrams whether or not the
self-interaction term (φ† φ)2 will be generated by quantum corrections anyway.

(d) Which regularization procedure would you use to show that the Ward identity associated with the gauge
symmetry in this theory is respected by quantum corrections?

SOLUTION:

(a) Let us expand out the action:


Z  
4 † µ 2 † 1 µν
 µ † † µ
 2 µν † λ † 2
S1 = d x ∂µ φ ∂ φ − m φ φ − Fµν F + ieAµ (∂ φ )φ − φ ∂ φ + e g Aµ Aν φ φ − (φ φ) . (13.1.2)
4 4

After canonical quantization, we have

d¯3 p
Z
ap e−ip·x + b†p eip·x ,

φ(x) = p (13.1.3a)
2Ep
d¯3 p
Z
φ† (x) = a†p eip·x + bp e−ip·x .

p (13.1.3b)
2Ep

a†p creates the scalar particles and b†p the scalar antiparticles. In this theory, these are the scalar versions of
the electron and positron in regular QED (i.e. the superpartners). Therefore, we will call them selectrons
and spositrons, respectively.

The first things we will consider are the propagators. These depend only on the free part of the action.
Therefore, these propagators are the same as in regular scalar field theory and regular QED. These are in
P&S p.118 and p.123.

135
136 CHAPTER 13. FINAL EXAM

Propagators:
i
φ(x)φ† (y) = -
q = (13.1.4a)
q 2 − m2 + i

−igµν
Aµ Aν = µ  ν = (13.1.4b)
q q 2 + i

Next, we will consider external leg contractions. These are also basically the same as in P&S p.118 and p.123.
The only difference is that there are now scalar particles as well as antiparticles. This merely doubles the
possible external scalar legs, but they all still take the value 1. In the following diagrams, the dot is meant
to indicate the presence of a vertex. Heading towards a vertex is incoming and going away is outgoing.
External leg contractions:

φ |qi = r  =1 hq| φ† =  r =1 (13.1.5a)


|{z} q |{z} q
selectron selectron

φ† |qi = r -
 =1 hq| φ = -
 r =1 (13.1.5b)
|{z} q |{z} q
spositron spositron

Aµ |pi = r  µ = µ (p) hp| Aµ = µ  r = ∗µ (p) (13.1.5c)


p p

Next, we will consider the vertices. There is one vertex that couples two scalar fields and one photon, one
vertex that couples two scalar fields and two photons, and one vertex that couples four scalar fields. First,
the interaction Hamiltonian is
Z  
λ
Hint = d3 x −ieAµ ∂ µ φ† φ − φ† ∂ µ φ − e2 g µν Aµ Aν φ† φ + (φ† φ)2 .
  
(13.1.6)
4
R
Vertices are contained in the first order term in the expansion of e−i Hint dt , which is simply
Z Z  

− i Hint dt = d4 x eAµ φ† ∂ µ φ − ∂ µ φ† φ + ie2 g µν Aµ Aν φ† φ − (φ† φ)2 .
  
(13.1.7)
4

Number the vertex types 1,2, and 3 according to their order from left to right in Eqn. (13.1.7). Let us work
out the factor associated with the first vertex type when the scalars are selectrons. Consider the first term:

hp0 | eAµ φ† ∂ µ φ |pi = hp0 | eAµ φ† (−ipµ )φ |pi . (13.1.8)

∂ µ φ gets converted into −ipµ φ because ap is multiplied by e−ip·x in φ. Similarly, since a†p is multiplied by
eip·x in φ† , the second term in the first vertex, −eAµ (∂ µ φ† )φ, picks up a factor of −ipµ . This gives a vertex
factor of −ie(p + p0 )µ . The same factor, but with a minus sign, pertains to the case when the scalars are
spositrons. In this case, φ† must first be moved to the right of φ, but the two terms, having opposite sign,
contribute opposite-signed divergent pieces, which cancel. Then, the minus sign is due to the fact that the
b’s have opposite frequency from the a’s.
Next, we consider vertex type 2. For selectrons, there are two ways to contract the photons: either Aµ is
outgoing and Aν is incoming, or the other way round. Therefore, this vertex contains an extra factor of 2
and is given by 2ie2 gµν . Finally, the quartic vertex gives a factor −iλ, as usual.

Vertices:
p0 6 p0 6
?
r µ = −ie(p + p )0 µ r µ = ie(p + p0 )µ (13.1.9a)
p6 p6?

µ r ν = 2ie2 gµν r = −iλ (13.1.9b)


13.1. SCALAR ELECTRODYNAMICS 137

(b) Let us calculate the superficial degree of divergence in arbitrary spacetime dimension, d. Let L be the number
of loops, I the number of internal lines, E the number of external lines, and Vi the number of vertices of type
i. Type 1 is the γss (photon-scalar-scalar) vertex, type 2 is γγss, and type 3 is ssss. A subscript γ or φ on
I or E indicates a photon or scalar line, respectively. Conservation of scalar and photon endpoints imply

2Iφ + Eφ = 2(V1 + V2 ) + 4V3 , 2Iγ + Eγ = V1 + 2V2 . (13.1.10)

d
Multiply both equations by 4, add them together, and then add ( d4 − 1)V1 to both sides of the resulting
equation. This gives
(d − 1)V1 + d(V2 + V3 ) = d2 I + d4 E + d

4 − 1 V1 . (13.1.11)

The number of loops is related to internal lines and vertices via

L = I − (V1 + V2 + V3 − 1). (13.1.12)

The superficial degree of divergence is

D = dL − 2I + V1 = dI − d(V1 + V2 + V3 − 1) − 2I + V1 = d + (d − 2)I − (d − 1)V1 − d(V2 + V3 ). (13.1.13)

Using Eqn. (13.1.11), we can simplify this to


d→4
D = d − d4 E + d

4 − 1 (2I − V1 ) −−−→ 4 − E. (13.1.14)

In the following diagrams, a solid line will denote the scalar field because there are no fermions in the problem
anyway.

E = 0 : These are vacuum bubbles. They are quartically divergent. An example is the clam:

E = 1 : These have one external photon leg (it is impossible to have an odd number of external scalar legs).
These are cubically divergent. To lowest order, O(e), we have the tadpole diagram. Of course, this can be
dressed up with arbitrarily complicated internal lines.

E = 2 : These have two external photon legs or two external scalar legs. These are quadratically divergent.
To order e2 or λ, the diagrams are

E = 3 : This is linearly divergent and to order e3 the diagram is

E = 4 : These are logarithmically divergent. There are quite a few diagrams. Here are a few:
138 CHAPTER 13. FINAL EXAM

(c) It is technically unnatural for g to be small because there is no enhancement of symmetry in the g → 0 limit.
Therefore, we should expect that the quartic term would be generated dynamically anyway by quantum cor-
rections. The one loop diagram below is an example of a diagram that contributes a quartic vertex correction,
and would produce such a vertex even if one did not exist to begin with. This diagram is logarithmically
divergent since it has E = 4 or D = 0.

(13.1.15)

(d) I would use dimensional regularization because, unlike Pauli-Villars or the sharp cut-off method, dimensional
regularization never breaks gauge-invariance.

13.2 Non-relativistic Lifshitz Scalar Field Theory


Consider the simplest theory of a Lifshitz scalar, whose action is given in D spatial dimensions by
Z
1
dt dD x (∂t Φ)2 − (∆Φ)2 ,

S2 = (13.2.1)
2
PD
where ∆ ≡ i=1 ∂i ∂i is the spatial Laplacian. This defines a free field theory with anisotropic scalaing, charac-
terized by dynamical exponent z = 2.

(a) Determine the classical scaling dimension of Φ in this theory, as a function of D.


(b) List all independent classically relevant and classically marginal terms that can be added to the action of
this theory in the special case of D = 3 spatial dimensions.
SOLUTION:

(a) Spatial coordinates have dimension −1 and time has dimension −z = −2. Therefore, in order to make S2
dimensionless, we must have
D−2
[Φ] = . (13.2.2)
2
(b) To maintain spatial isotropy, ∂i must appear in like pairs, ∆. Let λabc be the coefficient of a term containing
a time derivatives, b spatial Laplacians or 2b spatial derivatives, and c scalar fields. The dimension of λabc is

[λabc ] = 5 − 2(a + b) − 2c . (13.2.3)

λabc is irrelevent if its dimension is negative. Therefore, there are three cases: (1) a = b = 0 and c = 0, . . . , 10;
(2) (a, b) = (1, 0) or (0, 1) and c = 0, . . . , 6; or (3) (a, b) = (2, 0) or (0, 2) or (1, 1) and c = 0, 1, 2. If we want
to maintain the φ → −φ symmetry, then c must be even.
13.3. ASYMPTOTIC FREEDOM IN 5+1 DIMENSIONS 139

13.3 Asymptotic Freedom in 5+1 Dimensions


Consider the relativistic scalar field theory with the λφ3 interaction,
Z  
1 d µ 2 2 λ 3
S3 = d x ∂ φ ∂µ φ − m φ − φ . (13.3.1)
2 3

Even though this theory does not have energy bounded from below and therefore no non-perturbatively stable
ground state, it still makes sense to study it to all orders in perturbation theory using Feynman diagrams. People
say that this theory is asymptotically free in d = 5 + 1 dimensions.

(a) What quantity would you calculate to convince yourself that this theory is asymptotically free at one loop?

(b) Draw all Feynman diagrams that you would need in order to show asymptotic freedom at one loop. What
degree of divergence do you expect from those diagrams?

(c) (Bonus) Using dimensional regularization, complete the one-loop calculation in the λφ3 theory to the point
where you prove its asymptotic freedom in d = 5 + 1 at one loop. (You must show all intermediate steps,
not just the final result.)

SOLUTION:

(a) The sign of the beta function for λ. Asymptotic freedom would be implied by β(λ) < 0.

(b) The number of loops in a diagram is related to the number of internal lines, I, and vertices, V via L =
I − (V − 1). Conservation of endpoints implies 3V = 2I + E, where E is the number of external lines. Finally,
the superficial degree of divergence is related to the number of loops and internal lines via D = dL − 2I.
Combining these three equations gives

D = dL − d3 E + 2 d6 − 1 I.

(13.3.2)

At the one-loop level, L = 1. The loop and endpoint-conservation equations imply V = I = E in this case
and Eqn. (13.3.2) simplifies to
D|L=1 = d − 2E. (13.3.3)

When d = 6, there are four types of divergent diagrams: (1) E = 0 vacuum bubbles, which are unimportant;
(2) E = 1 tadpole diagram, which is also unimportant since it just gives a source term for φ; (3) E = 2
self-energy diagram, which gives the wavefunction and mass renormalizations and which is quadratically
divergent; and (4) E = 3, which gives the vertex renormalization and is logarithmically divergent.

(c) Rewrite the action in Eqn. (13.3.1) with 0 subscripts on φ, m and λ to indicate that they are bare quantities.
Quantities with no subscripts are taken to be renormalized. Then,
Z  
1 λ 6−d Cλ 6−d 3
S3 = dd x ∂ µ φ ∂µ φ − m2 φ2 − µ̄ 2 φ3 − A∂ µ φ ∂µ φ + Bm2 φ2 − µ̄ 2 φ + Jφ . (13.3.4)
2 3 3

The first two terms are the free action, the third is the interaction term, and the last four are the counterterms.
We have added a φ counterterm to cancel the tadpole diagram in principle, but we will not actually calculate
it since it does not affect the rest of the renormalization or whether or not the theory is asymptotically free.
6−d
The term µ̄ 2 requires some explanation. The dimension of φ is [φ] = d−22 . Therefore, without this µ̄
term, the dimension of λ would depend on the dimension of spacetime as [λ] = 6−d
2 . We must absorb this
dimension into an arbitrary scale, µ̄, in order to keep λ dimensionless.
A and B introduce a new quadratic vertex with vertex factor −i(Ap2 − Bm2 ), where p is the incoming
and outgoing momentum. C introduces a new cubic vertex with vertex factor −iCλ. These new vertices are
denoted by a cross:

-
p
-
p = −i(Ap2 − Bm2 ), = −iCλ. (13.3.5)
140 CHAPTER 13. FINAL EXAM

The exact propagator, De , may be expanded in terms of the free propagator, Df , and successive 1PI inter-
mediates as

De = Df + Df (−iΣ)Df + Df (−iΣ)Df (−iΣ)Df + · · ·


= Df 1 + (−iΣDf ) + (−iΣDf )2 + · · ·
 

Df
=
1 + iΣDf
i
= 2 , (13.3.6)
p − m2 − Σ
i
where we recall that Df = p2 −m2 . To one-loop order, or quadratic order in λ,

− iΣ = −iΣ2 + O(λ4 ) = m + (13.3.7)

Denote the first term by −iΣint


2 and the second by −iΣct 2 . The superscripts stand for “interaction” and
“counterterm”. The counterterm contribution is simply its vertex factor in Eqn. (13.3.5). The interaction
contribution is
Z
1 6−d 2 i i
−iΣint
2 = −iλµ̄ 2 d¯d k 2
2 k − m2 (k − p)2 − m2
Z 1 Z
1 1
= λ2 µ̄6−d dz d¯d k 2 − m2 ] + (1 − z)(k 2 − m2 )}2
. (13.3.8)
2 0 {z[(k − p)

Note that the factor of 21 out front is due to the symmetry factor of the diagram, which is 2. Write the
denominator as D2 where

D = k 2 − 2zp + z 2 p2 − z 2 p2 + zp2 − m2 = `2 − ∆, (13.3.9)

where

` ≡ k − zp, ∆ ≡ m2 − z(1 − z)p2 . (13.3.10)

After Wick-rotating, we have


Z 1 Z
i 2 6−d 1
−iΣint
2 = λ µ̄ dz d¯d `E
2 0 (`2E + ∆)2
1
Γ(2 − d2 ) d −2
Z
i 1
= λ2 µ̄6−d dz d ∆2 , (13.3.11)
2 0 (4π) Γ(2)
2

6−d
where use was made of Eqn. 7.85 P&S p.251. Now substitute in  = 2 :
Z 1
i 2 2 Γ(−1 + )
− iΣint
2 = λ µ̄ dz ∆1− . (13.3.12)
2 (4π)3− 0

Expand around  = 0 using


n
(−1)n
 
1 X 1
Γ(−n + ) = −γ+ + O() , a = 1 +  ln a + O(2 ). (13.3.13)
n!  k
k=1

The result is
1
iλ2 4π µ̄2
Z  
1
− iΣint
2 =− dz ∆ − γ + 1 + ln . (13.3.14)
2(4π)3 0  ∆

Let us follow the MS renormalization scheme and absorb the factor of γ and ln 4π into a redefinition of the
RG scale by defining
4π µ̄2
µ2 = γ . (13.3.15)
e
13.3. ASYMPTOTIC FREEDOM IN 5+1 DIMENSIONS 141

Also, we can calculate the integral of ∆ easily:


Z 1
dz ∆ = m2 − 61 p2 . (13.3.16)
0

Finally, for notational brevity, define


λ2
α≡ . (13.3.17)
2(4π)3

Therefore,
 Z 1
µ2
  
1 1 2
−iΣint
2 = −iα +1 m − p + 2
dz ∆ ln
 6 0 ∆
 Z 1
µ2 m2
  
1 1
= −iα + 1 + ln 2 m2 − p2 + dz ∆ ln . (13.3.18)
 m 6 0 ∆

Set
µ2 µ2
     
1 1 1
A=α + 1 + ln 2 + a , B=α + 1 + ln 2 + b . (13.3.19)
6  m  m

Then,
1
m2
 Z 
2 2
− iΣ2 = −iα ap − bm + dz ∆ ln . (13.3.20)
0 ∆

For convenience, define

1 p2 ∆
Σ2 ≡ αm2 Σ2 , χ≡ m2 , ∆≡ m2 = 1 − z(1 − z)χ. (13.3.21)

Then, Eqn. (13.3.20) reads


Z 1
Σ2 = aχ − b − dz ∆ ln ∆. (13.3.22)
0

By definition, the exact propagator has a pole at p2 = m2 with residue 1. This implies

dΣ2
Σ2 |p2 =m2 = 0, = 0. (13.3.23)
dp2 p2 =m2

These conditions are equivalent to



dΣ2
Σ2 |χ=1 = 0, = 0. (13.3.24)
dχ χ=1

These two conditions fix the values of a and b. Consider the derivative condition first because this isolates a:
Z 1 Z 1
1
dz z(1 − z) ln(1 − z + z 2 ).

0=a+ dz z(1 − z)(1 + ln ∆) χ=1 = a + + (13.3.25)
0 6 0

We won’t actually need the integral, but it can be done in Mathematica and yields the result
1

a = 18 (14 − 3 3 π). (13.3.26)

Similarly the first condition in (13.3.24) reads


Z 1
0=a−b− dz (1 − z + z 2 ) ln(1 − z + z 2 ). (13.3.27)
0
142 CHAPTER 13. FINAL EXAM

The integral can again be done in Mathematica and yields the result
1
√ √
b = 18 (33 − 6 3 π) = 16 (11 − 2 3 π). (13.3.28)

Define
√ µ2
 
1 1 17
Zφ = 1 − A = 1 − α + − π 3 + ln 2 , (13.3.29a)
6  3 m

µ2
 
1 17 π 3
Zm = 1 − B = 1 − α + − + ln 2 . (13.3.29b)
 6 3 m

Also define Zλ = 1 + C, which we will calculate next. The action, (13.3.4) takes the form
Z  
1 Zλ λ 6−d 3
S3 = dd x Zφ ∂ µ φ ∂µ φ − Zm m2 φ2 − µ̄ 2 φ + Jφ . (13.3.30)
2 3

Now, we will calculate the one-loop vertex correction. Call the exact vertex −iV and let −iV2 be the order
λ2 correction to the vertex. Then,

− iV = −iλ − iV2 + O(λ4 ) = −iλ + m + (13.3.31)

Again, call the first term the interaction piece, −iV2int , and the second one the counterterm, −iV2ct . The
counterterm is just given by its vertex in Eqn. (13.3.5). We must calculate the amplitude of the interaction
diagram. Denote the momenta to look like the vertex corrections we have calculated in the past:

q
?

k R k0 (13.3.32)
-  -0
p p

-
p−k

So that we do not have to keep carrying µ̄ around, we will take it for granted and only write it explicitly
near the end. The amplitude is
Z
i i i
−iV2int = (−iλ)(−iλ)2 d¯d k 2
k − m2 k 02 − m2 (p − k)2 − m2
Z Z
1
= (−iλ)2iλ2 d¯d k . (13.3.33)
F {x(k − m ) + y[(k + q) − m2 ] + z[(p − k)2 − m2 ]}3
2 2 2

Write the denominator as D3 where

D = k 2 + 2(yq − zp) · k + (yq − zp)2 − (yq − zp)2 + yq 2 + zp2 − m2


= (k + yq − zp)2 + y(1 − y)q 2 + z(1 − z)p2 + 2yzq · p − m2 . (13.3.34)

Using 2q · p = −q 2 = −m2 and (p + q)2 = p02 , we find that D = `2 − ∆, where

` ≡ k + yq − zp, ∆ ≡ m2 − xyp2 − yzp02 − zxq 2 . (13.3.35)

Then, after Wick rotating and writing d = 6 − 2, we get


Z Z
1 int 2 1
V
λ 2 = 2λ d¯d `E
F (`2E + ∆)3
Γ(3 − d2 ) d −3
Z
1
= 2λ2 d ∆2
F (4π) Γ(3)
2
Z
2 Γ()
=λ ∆− . (13.3.36)
(4π)3− F
13.3. ASYMPTOTIC FREEDOM IN 5+1 DIMENSIONS 143

We reintroduce µ̄ on the right (there is no point in writing it on the left) by inserting a factor µ̄6−d = µ̄2 .
Then, we expand around  = 0 to get

λ2 4π µ̄2 MS µ2
Z   Z  
1 int 1 1
λ V2 = − γ + ln −−→ 2α + ln . (13.3.37)
(4π)3 F  ∆ F  ∆

R1 R 1−z
dy = 12 , we get
R
Using F
= 0
dz 0

µ2
 Z 
1 int 1
λ V2 =α + ln 2 − 2 ln ∆ , (13.3.38)
 m F

where we recall that ∆ = ∆/m2 .


Set
µ2
 
1
C = −α + ln 2 + c . (13.3.39)
 m
1 ct
Then, since λ V2 = C, we get
 Z 
1
λV =1−α c+2 ln ∆ . (13.3.40)
F

We must simply choose a value for c because there is no analog of the conditions in Eqn. (13.3.23). We
will fix c such that V = λ when the external legs are at zero momentum. This is somewhat unphysical
since the scalar is massive and thus cannot be physical at zero 4-momentum. In this case, ∆ = 1 and thus,
V = (1 − αc)λ. Therefore, we must set c = 0. Finally, recalling that Zλ = 1 + C, we have

µ2
 
1
Zλ = 1 − α + ln 2 . (13.3.41)
 m

Compare the expression of S3 in terms of renormalized and bare quantities:


Z  
1 Zλ λ  3
S3 = dd x Zφ ∂ µ φ ∂µ φ − Zm m2 φ2 − µ̄ φ + Jφ , (13.3.42a)
2 3
Z  
1 λ0
S3 = dd x ∂ µ φ0 ∂µ φ0 − m20 φ20 − φ30 − J0 φ0 . (13.3.42b)
2 3

Therefore, λ0 is related to λ via

−3/2 λ20
λ0 = Zφ Zλ µ̄ λ =⇒ α0 = = Zφ−3 Zλ2 µ̄2 α. (13.3.43)
2(4π)3

Let us focus only on the divergent pieces in the Z’s because it turns out that the non-divergent pieces only
contribute terms of order  or higher, which vanish in the  → 0 limit. Thus,
α α
Zφ = 1 − , Zλ = 1 − , (13.3.44)
6 

and  α  2α   3α 
α0 ≈ 1 − 1− (4πe−γ ) µ2 α = α 1 − (4πe−γ ) µ2 + O(α3 ). (13.3.45)
2  2
Take the log of this:
 3α  3α
ln α0 = ln α + ln 1 − + 2 ln µ +  ln 4π − γ ≈ ln α − + 2 ln µ +  ln 4π − γ. (13.3.46)
2 2

Recall what happens in Pauli-Villars regularization or the sharp cut-off method: the bare parameters are
functions of the cut-off, but not of the RG scale. Instead, the renormalized parameters are functions of the
144 CHAPTER 13. FINAL EXAM

RG scale, but not of the cut-off. Roughly speaking, the  parameter in dimensional regularization plays the
role of the cut-off, Λ. Therefore, we must have
d ln α0 1 3  dα
0= = − + 2. (13.3.47)
d ln µ α 2 d ln µ

Therefore,
 3α  dα
1− = −2α. (13.3.48)
2 d ln µ
Then, the β function is taken in the  → 0 limit:

dα  3α  →0
β(α) = ≈ −2α 1 + = −2α − 3α2 −−−→ −3α2 . (13.3.49)
d ln µ 2

Since the sign of the beta function is negative, the theory is asymptotically free . It is easy to solve for α:

α
e
α(µ) = , (13.3.50)
1 + 3e
α ln(µ/µ̃)

µ→∞
e when µ = µ̃. Indeed, α −−−−→ 0.
where α = α

Aside: Note that the calculation of a, b and c is immaterial insofar as asymptotic freedom is concerned since we
ended up caring only about the  poles. However, just in case you dislike the unphysical limit that we took to
set c = 0, let us consider the more physical limit of on-shell external particles for which p02 = p2 = q 2 = m2 .
Looking back at the expression for ∆ in Eqn. (13.3.35) and recalling that ∆ = ∆/m2 , in this case

∆ = 1 − xy − yz − zx. (13.3.51)

You can check for yourself that this quantity is always positive in the integration domain taking on its minimum
value of 23 when x = y = z = 13 . The required integral can be calculated numerically and yields
Z
ln ∆ ≈ −0.1456. (13.3.52)
F

Therefore, Eqn. (13.3.40) reads


1
λV = 1 − α(c − 0.2912) =⇒ c = 0.2912. (13.3.53)

This means that we have chosen V = λ to coincide with the external particles being on-shell.

You might also like