QFT 2

Quantum Field Theory II
Babis Anastasiou
Institute for Theoretical Physics,
ETH Zurich,
8093 Zurich, Switzerland
E-mail: babis@phys.ethz.ch
March 4, 2018
Contents
1 Path integral quantization in Quantum Mechanics 4
1.1 The propagator . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 The path integral . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 An “adventurous” transition . . . . . . . . . . . . . . . . . . . 11
1.4 Functional differentiation . . . . . . . . . . . . . . . . . . . . . 15
1.5 Vacuum to vacuum transitions . . . . . . . . . . . . . . . . . . 17
1.6 The simple harmonic oscillator . . . . . . . . . . . . . . . . . . 19
2 Path integrals and scalar fields 23

2.1 Functional Integration . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Path integrals and interacting fields . . . . . . . . . . . . . . . 30
2.3 Perturbation theory for −λφ4 /4! interactions . . . . . . . . . . 32
2.4 Fermionic path integrals . . . . . . . . . . . . . . . . . . . . . 35
3 Non-abelian gauge theories 43

3.1 Gauge invariance . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2 Non-abelian (global) SU (N ) transformations . . . . . . . . . . 45
3.3 Local non-abelian gauge symmetries . . . . . . . . . . . . . . . 48
1
4 Quantization of non-abelian gauge theories 55
4.1 Perturbative QCD . . . . . . . . . . . . . . . . . . . . . . . . 64
5 BRST symmetry 69
5.1 Application to the free electromagnetic field . . . . . . . . . . 74
6 Quantum effective action and the effective potential 79

6.1 The quantum effective action as a generating functional . . . . 81
6.2 The effective potential . . . . . . . . . . . . . . . . . . . . . . 88
7 Symmetries of the path integral and the effective action 92

7.1 Slavnov-Taylor identities . . . . . . . . . . . . . . . . . . . . . 92
7.2 Symmetry constraints on the effective action . . . . . . . . . . 93
7.3 Contraints on the effective action from BRST symmetry trans-
formations of the classical action . . . . . . . . . . . . . . . . 95
7.4 Slavnov-Taylor identities in QED . . . . . . . . . . . . . . . . 97
8 Spontaneous symmetry breaking 102

8.1 Goldstone theorem . . . . . . . . . . . . . . . . . . . . . . . . 102
8.2 General broken global symmetries . . . . . . . . . . . . . . . . 106
8.3 Spontaneous symmetry breaking of local gauge symmetries . . 108
9 Renormalization: counting the degree of ultraviolet diver-

gences 110
9.1 Subdivergences . . . . . . . . . . . . . . . . . . . . . . . . . . 114
9.2 Cancelation of superficial divergences with counterterms . . . 115
9.3 Nested and overlapping divergences . . . . . . . . . . . . . . . 118
10 Proof of renormalizability for non-abelian gauge theories 120
Abstract
The subject of the course is modern applications of quantum field
theory with emphasis on the quantization of non-Abelian gauge theo-
ries. The following topics are discussed:
• Classical gauge transformations.
• Quantization for fermionic and bosonic fields and perturbation
theory with path-integrals is developed.
2
• Quantization of non-Abelian gauge-theories. The Fadeev-Popov
method. BRST symmetry.
• The quantum effective action and the effective potential.
• Classical symmetries of the effective action. Slavnov-Taylor iden-
tities. The Zinn-Justin equation.
• Physical interpretation of the effective action.
• Spontaneous symmetry breaking. Goldstone theorem. Sponta-
neous symmetry breaking for theories with local gauge invari-
ance.
• Power-counting and ultraviolet infinities in field theories. Renor-
malizable Lagrangians.
• Renormalization and symmetries of non-Abelian gauge theories.
Renormalization group evolution.
• Infrared divergences. Landau equations. Coleman-Norton phys-
ical picture of infrared divergences. Soft and collinear singulari-
ties.
3
1 Path integral quantization in Quan-
tum Mechanics
We have carried out a quantization program for simple field theories in
the course of QFT I by means of “canonical quantization” (imposing
commutation and anti-commutation relations on fields). Here we will
quantise gauge invariant field theories with a different method, us-
ing a formalism based on path integrals. The formalism is somewhat
imperative to develop. While the canonical formalism can be suc-
cessfully applied to Quantum Electrodynamics, it is not understood
how it can be applied to the gauge field theories which describe the
unified strong and electroweak interactions. As a warm-up we revisit
quantum mechanics, formulating quantisation using path integrals.
1.1 The propagator

We consider a quantum mechanical state |ψi which satisfies the Schrödinger
equation:
ih̄∂t |ψi = Ĥ |ψi . (1)
The solution of this equation
i
|ψ(t2 )i = e− h̄ Ĥ(t2 −t1 ) |ψ(t1 )i , (2)
determines the evolution of this state from an initial moment t1 to a

later moment t2 . The wave function ψ (x, t) ≡ hx| ψ(t)i is then
i
ψ (x, t2 ) = hx| e− h̄ Ĥ(t2 −t1 ) |ψ(t1 )i (3)
The sum of the Hamiltonian eigenstates
Ĥ |βi = Eβ |βi (4)
is unitary, X
1= |βi hβ| . (5)
β
Inserting it in the rhs of the evolution equation we find

 
− h̄i Ĥ(t2 −t1 ) 
X
ψ (x, t2 ) = hx| e |βi hβ| |ψ(t1 )i
β
− h̄i Eβ (t2 −t1 )
X
= e hx| βi hβ| ψ(t1 )i . (6)
β
4
We now insert another unit operator
Z
d3 x0 x0

0
1= x , (7)
obtaining
Z
− h̄i Eβ (t2 −t1 ) 3 0 0
0
X
ψ (x, t2 ) = e hx| βi hβ| d x x x |ψ(t1 )i
β
 
Z X 
− h̄i Eβ (t2 −t1 )
d3 x0 hx| βi hβ| x0 ψ x0 , t1 . (8)

= e
 
β
In other words, if we know the wave-function at one time, we can

determine it fully at a later time by integrating it,
Z
d3 x0 K(x, x0 ; t2 − t1 )ψ x0 , t1 ,

ψ (x, t2 ) = (9)
with a kernel
i
K(x, x0 ; t2 − t1 ) = e− h̄ Eβ (t2 −t1 ) hx| βi hβ| x0 ,
X
(10)
β
which depends on the Hamiltonian of the system and the elapsed time
t2 − t1 during the evolution of the quantum state. This integration
kernel is called the “propagator”. For t2 = t1 the propagator is a delta
function
K(x, x0 ; t1 − t1 ) = e0 hx| βi hβ| x0 = δ 3 (x − x0 ).

X
(11)
β
Exercise: Prove that the propagator K(x, x0 ; t−t0 ) satisfies the Schrödinger
equation in the variables x, t for times t > t0 .
Exercise: Prove that
Z ∞ r
−ax2 π
dxe = .
−∞ a
Exercise: Compute the propagator K(x, x0 ; t − t0 ) for

• a free particle,
• the simple harmonic oscillator.
5
The propagator is the amplitude for a particle measured at position
|x0 i at time t1 to propagate to a new position |xi at time t2 > t1 . We
can verify this easily. Consider a particle measured at a position x0 .
After time t2 − t1 it will be evolved to a new state
i
e− h̄ Ĥ(t2 −t1 ) x0 .

(12)
The probability amplitude to be measured at a position x is

i
hx| e− h̄ Ĥ(t2 −t1 ) x0

 
− h̄i Ĥ(t2 −t1 )
|βi hβ| x0
X
= hx| e 
β
− h̄i Eβ (t2 −t1 )
hx| βi hβ| x0
X
= e
β
= K(x, x0 ; t2 − t1 ). (13)
We can attempt to compute the propagator for a transition which

takes a very small time δt → 0.
i
K(x, x0 ; δt) = hx| e− h̄ Ĥδt x0

i
= hx| 1 − Ĥδt x0 + O δt2

h̄
i
Z
d3 p |pi hp| x0 + O δt2

= hx| 1 − Ĥδt
h̄
i
Z
3 0 0
= d p hx| pi hp| x − δt hx| Ĥ |pi hp| x + O δt2
h̄
(14)
We now specialise to Hamiltonian operators of the form Ĥ = f1 (p̂) +

f2 (x̂). Then
hx| Ĥ |pi = H hx| pi , (15)
where H is not an operator anymore but the classical Hamiltonian,
i.e. a real-valued function of position and momentum. Recall that the
position and momentum states are related via a Fourier transform,
i
e h̄ px
hx| pi = √ . (16)
2πh̄
For simplicity let us consider one only dimension; the three-dimensional
case is a faithful repetition of the same steps. Then we find for the
6
propagator at small time intervals:
dp i p(x−x0 ) i
Z
K(x, x0 ; δt) = e h̄ 1 − Hδt + O δt2
2πh̄ h̄
dp i
Z
p x − x0 − Hδt + O δt2 (17)

= exp .
2πh̄ h̄
An interesting form for the propagator for small time intervals arises
when the Hamiltonian is of the form
p2
H= + V (x). (18)
2m
Then,
( )!
dp i p2
Z
0 0
K(x, x ; δt) = exp p x−x − δt − V (x)δt + O δt2
2πh̄ h̄ 2m
( 2 2 )!
dp i δt x − x0 1 x − x0
Z
= exp − p−m + δt m − V (x)δt + O δt2
2πh̄ h̄ 2m δt 2 δt
( 2 )!
dp̃ iδt p̃2 1 x − x0
Z
= exp − + m − V (x) + O δt2 , (19)
2πh̄ h̄ 2m 2 δt
where in the final step we performed a trivial change of integration

variable. Our final result reads:
( 2 )!
1 iδt 1 x − x0

0
K(x, x ; δt) ' exp m − V (x) , (20)
N (δt) h̄ 2 δt
with ! 1/2
1 dp −iδtp2 m
Z
= exp = (21)
N (δt) 2πh̄ 2mh̄ 2πih̄δt
1.2 The path integral

We consider now the transition from an initial position (xi , ti ) to a
final position (xf , tf ), which has a propability amplitude given by the
propagator:
K(xf , xi ; tf − ti ).
We can take a “snapshot” at an intermediate time t1 during this tran-
sition:
ti < t1 < tf .
7
If taking the “snapshot” particle is a measurement of the position
x1 of the particle at the moment t1 , then the amplitude for the full
transition will be:
K(xf , xi ; tf − ti ) = K(xf , x1 ; tf − t1 )K(x1 , xi ; t1 − ti ). (22)
If taking the “spanshot” is only a thought experiment and we don’t

actually determine the position x1 of the particle at t1 with a real
measurement, we should integrate over all probability amplitudes for
the particle to have performed this transition via any point. We then
have:
Z ∞
K(xf , xi ; tf − ti ) = dx1 K(xf , x1 ; tf − t1 )K(x1 , xi ; t1 − ti ). (23)
−∞
We are allowed decide to take several “snapshots” during the transi-

tion from xi to xf , in times ti < t1 < t2 < . . . < tn < tf . Using again
the superposition principle we must write:
Z ∞
K(xf , xi ; tf − ti ) = dx1 . . . dxn K(xf , xn ; tf − tn )K(xn , xn−1 ; tn − tn−1 ) ×
−∞
. . . K(x2 , x1 ; t2 − t1 )K(x1 , xi ; t1 − ti ). (24)
For simplicity, we now consider infinitesimal equally fast tk+1 − tk =

tf −ti
δt = n+1 intermediate transitions. Then we obtain
Z ∞
K(xf , xi ; tf − ti ) = lim dx1 . . . dxn K(xf , xn ; δt)K(xn , xn−1 ; δt) ×
n→∞ −∞
. . . K(x2 , x1 ; δt)K(x1 , xi ; δt). (25)
We have discretised and taken the infinite limit, which is the defining
procedure of an integration. However, this new integral is rather un-
usual. Think of all the paths which connect the initial and final points
xi and xf . The points which belong to these paths are accounted for
by the limit of infinitesimal n → ∞ transitions which we have taken
in Eq. 25. Therefore, the rhs of this equation is an integral over all
paths that a particle may take in going from xi → xf .
We can insert the expressions for the propagator at small time
intervals of Eq. 17 or Eq. 20 into Eq. 25. Notice that in the limit
δt → 0,
xn − xn−1
→ x˙n .
δt
8
Then Eq. 25 becomes:
1 ∞ Z
K(xf , xi ; tf − ti ) = lim dx1 . . . dxn
n→∞ N (δt)n −∞
iδt m 2

exp x˙1 − V (x1 )
h̄ 2
iδt m 2

exp x˙2 − V (x2 )
h̄ 2
...
iδt m 2

exp x˙f − V (xf ) . (26)
h̄ 2
Equivalently,
1 ∞ Z
K(xf , xi ; tf − ti ) = lim dx1 . . . dxn
n→∞ N (δt)n −∞
i

exp {L(x1 (t1 ))δt + L(x2 (t2 ))δt + . . . L(xf (tf ))δt} ,
h̄
(27)
where
m 2
L(x) = ẋ − V (x),
2
is the Lagrangian of the system. We now have a more concrete under-
standing of the above integral. This is an integration over “all paths”
connecting the fixed points xi and xf , which we write symbolically as
Z tf
1 i
Z
K(xf , xi ; tf − ti ) = Dx exp dtL[x(t)] , (28)
N h̄ ti
and the sum (integral) in the exponential is the classical action as it

is evaluated in each of the paths. Even shorter, we can write:
1 i
Z
K(xf , xi ; tf − ti ) = Dx exp S[x] . (29)
N h̄
Exercise: Consider a Lagrangian of the form
1
L = f (x)ẋ2 + g(x)ẋ − V (x).
2
1. Compute the Hamiltonian
2. Compute the propagator for a small transition
9
3. Write the path-integral expression for the propagator at large
time integrals. Notice that the measure of the path integration
is modified
1
Dx → Dxf (x) 2
10
1.3 An “adventurous” transition
We look now at a more eventful transition amplitude. We first prepare
a particle on an initial position xi at a time ti and let it evolve for
some time t − ti according to a Hamiltonian Ĥ:
i
e− h̄ Ĥ(t−ti ) |xi i
At the time t, something abrupt occurs (e.g. an interaction with an-

other particle which was originally far away) and modifies the particle
state. We will see later how we can describe interactions of particles
using path integrals; now, let us consider an “easy” modification of
the state where the state is “mixed up” in a simple way, acting on it
with the position operator:
i
x̂e− h̄ Ĥ(t−ti ) |xi i
Then we allow the particle to evolve undistracted for a time tf − t,

i i
e− h̄ Ĥ (tf −t) x̂e− h̄ Ĥ(t−ti ) |xi i ,
and then we place a detector at xf :

i i
hxf | e− h̄ Ĥ (tf −t) x̂e− h̄ Ĥ(t−ti ) |xi i .
We can compute this matrix-element as a path integral. Before we

proceed, we should use some language which is more convenient to
describe “eventful” transitions. We can write the same transition am-
plitude as:
i i i i
n on on o
hxf | e− h̄ Ĥtf e h̄ Ĥt x̂e− h̄ Ĥt e h̄ Ĥtf |xf i
= hxf , tf | x̂(t) |xi , ti i . (30)
We have defined states

i
|ψ, ti ≡ e h̄ Ĥt |ψi , (31)
which refer to a fixed moment t only and do not evolve

∂ ∂ i Ĥt
ih̄ |ψ, ti = ih̄ e h̄ |ψi
∂t ∂t
∂ ∂ i Ĥt

i
= e h̄ Ĥt ih̄ |ψi + ih̄ e h̄ |ψi
∂ ∂t
i

= e h̄ Ĥt Ĥ − Ĥ |ψi = 0. (32)
11
We have defined operators which do change with time
i i
Ô(t) = e h̄ Ĥt Ôe− h̄ Ĥt . (33)
As you recognize, this is the “Heisenberg picture” of evolution. For

us, it is convenient to assign a time date on a state which denotes a
particle or a collection of particles at the beginning of an experiment
or at the end of it. However, one could have equally well chosen to
work in the probably more familiar “Schrödinger picture”.
Lets us now compute this “eventful” transition:
hxf , tf | x̂(tj ) |xi , ti i with ti < t < tf ,
following the method we used for the simple transition hxf , tf | xi , ti i =

i
hx | e h̄ Ĥ (tf −ti ) |x i, and subdividing the transition in small time inter-
f i
vals. We will find a similar/related path integral for the new case as
well.
Z
hxf , tf | x̂(tj ) |xi , ti i = dx1 . . . dxj−1 dxj . . . dxn
× hxf , tf | xn , tn i . . . hxj+1 , tj+1 | xj , tj i
× hxj , tj | x̂(tj ) |xj−1 , tj−1 i
× hxj−1 , tj−1 | xj−2 , tj−2 i . . . hx1 , t1 | xi ti(34)
i.
The subdivision of time is carefully chosen. We have

i i i

x̂(tj ) |xj , tj i = e h̄ Ĥtj x̂e− h̄ Ĥtj e h̄ Ĥtj |xj i
i
= e h̄ Ĥtj (x̂ |xj i)
i
= xj e h̄ Ĥtj |xj i
= xj |xj , tj i
; hxj , tj | x̂(tj ) |xj−1 , tj−1 i = xj hxj , tj | xj−1 , tj−1 i . (35)
We then obtain for the transition amplitude the same succession of

propagators as for the simple transition multiplied with an additional
factor xj :
Z
hxf , tf | x̂(tj ) |xi , ti i = dx1 . . . dxj xj . . . dxn
× hxf , tf | xn , tn i . . . hxj+1 , tj+1 | xj , tj i hxj , tj | xj−1 , tj−1 i . . . hx1 , t1 | xi ,(36)
ti i .
12
Introducing, as before, the explicit form for the propagator during a
small time transition
1 " ( 2 ) #
m i m xb − xa

2
hxb , tn+1 | xa , tn i ≈ exp − V (xa ) (tm+1 − tm )
2πih̄ (tn+1 − tn ) h̄ 2 tn+1 − tn
and taking equal time intervals, we obtain the path integral

Z ∞
1
hxf , tf | x̂(tj ) |xi , ti i = lim dx1 . . . dxj xj . . . dxn
n→∞ N (δt)n −∞
n
!
X tf − ti
× exp L(xr , ẋr )δt with δt = . (37)
r=1
n+1
In compact notation we can write

Z
i
hxf , tf | x̂(τ ) |xi , ti i = Dxx(τ )e h̄ S[x] . (38)
Exercise:
1. Prove that
Z ∞
hxf , tf | x̂(τ ) |xi , ti i = dxx hxf , tf | x, τ i hx, τ | xi , ti i .
−∞
2. Evaluate the above integral explicitly for a free particle

3. Observe the dependence of the result on the intermediate time τ
4. You (could) have considered the simpler matrix element hxf , tf | xi , ti i
and written down an analogous integral. Observe how the inter-
mediate time τ drops out from the final expression, when nothing
special occurs then!
We now elaborate further on the form of the path integral that
we have just found. Recall the method for computing exponential
integrals of the form
Z ∞
2
In = dxxn e−ax with n = 1, 2, . . . (39)
−∞
from the result of the integral

Z ∞ r
−ax2 π
I0 ≡ dxe = . (40)
−∞ a
13
We can compute In with a rather very simple differentiations, after
we add a “source” term on the exponent of the integrand
Z ∞ Z ∞ 2 2
J
2 +Jx + J4a
IJ = dxe−ax = dxe−a(x− 2a )
−∞ −∞
Z ∞ r
2
−ax̃2 + J4a π J2
= dx̃e = e 4a (41)
−∞ a
To compute the I1 it is sufficient to differentiate the last expression
with respect to the source and substitute J = 0.
2

de−ax +Jx
Z ∞ Z ∞
dIJ 2
= dx = dxxe−ax .
dJ J=0 −∞ dJ
J=0 −∞
Similarly,
dn IJ
Z ∞
2
= dxxn e−ax .
dn J J=0 −∞
Adding a source term to the exponent does not increase the difficulty
of the computation and it allows us to calculate all integrals where the
integrand is multiplied with a polynomial in the integration variable.
This is very suggestive, and we will do the same trick for path integrals
such as the one that we found in Eq. 37. We then add a linear term
(source) in the Lagrangian; this is only a computational trick and
eventually we will compute all interesting physical quantities as in the
above example with the source term set to zero. The simple transition
from a state |xi , ti i to a state |xf , tf i in the presence of the source has
a probability amplitude:
1
Z
i
P
dtk (L(xk ,ẋk )+Jk xk )
hxf , tf | xi , ti iJ = lim dx1 . . . dxn e h̄ k (42)
n→∞ N (δt)n
We can then compute

h̄ ∂
hxf , tf | x̂(tl ) |xi , ti i = hxf , tf | xi , ti iJ . (43)
i ∂Jl Jl =0
Let us now differentiate two times with respect to the source.

2
h̄ ∂2 1
Z
hxf , tf | xi , ti iJ = lim dx1 . . . dxl xl . . . dxq xq . . . dxn

i ∂Jl ∂Jq n→∞ N (δt)n
J l,q =0
i
P
dtk (L(xk ,ẋk )+Jk xk )
×e h̄ k . (44)
14
We can recognize the rhs as the expectation value for the product of
two position operators x̂(tl )x̂(tq ) if tq is earlier than tl or x̂(tq )x̂(tl )
otherwise. We then write
2
∂2
h̄
hxf , tf | T (x̂(tl )x̂(tq )) |xi , ti i = hxf , tf | xi , ti iJ ,

i
∂Jl ∂Jq
J=0
(45)
where we have introduced the notation T (Ô(t1 )Ô(t3 )Ô(t2 ) . . . Ô(tn ))
to remind us that we should put the operators in the correct time
order once the sequence of the moments ti is known. For example, if
t1 < t2 < t3 < . . . < tn we should write
T (Ô(t1 )Ô(t3 )Ô(t2 ) . . . Ô(tn )) = Ô(tn ) . . . Ô(t3 )Ô(t2 )Ô(t1 ).
Exercise: Compute hxf , tf | T (x̂(tl )x̂(tq )) |xi , ti i for a free particle.
1.4 Functional differentiation

It is cumbersome to work with path integrals by writing explicitly
the infinite limit of discretised paths. We introduced earlier a more
compact notation,
R tf
1 i
Z
dt[L(x(t),ẋ(t))+J(t)x(t)]
hxf , tf | xi , ti iJ = Dxe h̄ ti
. (46)
N
We can write neatly expressions for the expectation values of opera-
tors,
R tf
1 i
Z
dtL(x(t),ẋ(t))
hxf , tf | x̂(t1 ) |xi , ti i = Dx x(t1 ) e h̄ ti
N
or
R tf
1 i
Z
dtL(x(t),ẋ(t))
hxf , tf | T (x̂(t1 )x̂(t2 )) |xi , ti i = Dx x(t1 ) x(t2 ) e h̄ ti
N
by defining a functional derivative. We consider an integral F[f] over a
function f(t) (for example a path-line). We define a functional deriva-
tive by changing slightly the function f (y):
δF [f (y)] F [f (y) + δ(y − t)] − F [f (y)]

= lim . (47)
δf (t) →0
15
For example, consider the derivative of the action integral in the pres-
ence of a source with respect to the source:
R
δ dy {L (x(y), ẋ(y)) + J(y)x(y)}
Z
= dyδ(t − y)x(y)
δJ(t)
= x(t). (48)
Practically, we need the chain rule and to remember that
δf (x)
= δ(x − y).
δf (y)
The expectation values of time-ordered operators can then be written

as
n
δn

h̄
hxf , tf | T (x̂(tl ) . . . x̂(tn )) |xi , ti i = hxf , tf | xi , ti iJ .
i δJ(t1 ) . . . δJ(tn ) J=0
(49)
END OF WEEK 1
16
1.5 Vacuum to vacuum transitions
Field theory allows us to compute transitions between states with
different particle content. Interestingly, we can build particle states
acting with creation (or field) operators on the ground state which
contains no particles (vacuum). All transition amplitudes can be de-
scribed as expectation values of operators in the vacuum:
h0, tf | T (. . .) |0, ti i .
We can compute expectation values of operators in the vacuum with

path integrals. First, we try a direct approach
Z
dxdx0 h0, tf | x, ti hx, t| T (. . .) x0 , t0

0 0
h0, tf | T (. . .) |0, ti i = x , t 0, ti i .(50)
This formula is complicated. It requires that we know the wave func-

tion of the vacuum and that we are able to convolute it with the result
for a path integral. There is a rather simpler way with less integra-
tions.
Consider a Hamiltonian Ĥ with eigenstates |ni,
Ĥ |ni = En |ni (in the Scrhödinger picture)
and a general state X

|ψi = cn |ni .
n
Taking a “Heisenberg photograph” of the state in the very past (t →

∞), we find: X
|ψ, −ti = eiH̄(−t) cn |ni . (51)
n
Now, we play a mathematical game; we give a very small imaginary

part to the Hamiltonian
Ĥ → Ĥ(1 − i), ( ≈ 0),
which has the same energy eigenstates as the physical Hamiltonian.

The general state in the very past with the modified Hamiltonian is
X
|ψ, −ti = eiĤ(−t)(1−i) cn |ni
n
−(+i)En t
X
= cn e |ni . (52)
n
17
For a very long time t in the past the exponential e−En t vanishes; it
vanishes faster for larger energy eigenvalues. For the vacuum, (E0 =
0) 1 , there is no such suppression.
∞
!
−t(i+)E0 −(i+)(En −E0 )t
X
|ψ, −ti = e c0 |0i + e cn |ni
n=1
≈ e−t(i+)E0 c0 |0i = c0 |0, −ti . (53)
This is a very convenient. An arbitrary Heisenberg state in the very
past with the slightly complex Hamiltonian is, essentially, the vacuum
state in the very past. Higher energy eigenstates do not contribute to
the superposition since the small imaginary part forces them to decay
as we take the time −t further back in the past.
The same happens for a general Heisenberg state hψ, t| in the fu-
ture.
c∗n hn| e−iHt(1−i)
X
hψ, +t| =
n
≈ c∗0 h0| e−iE0 t(1−i)
≈ c∗0 h0, t| , t→ +∞. (54)
Vacuum-to-vacuum transition over very long times can therefore
be replaced by transitions over arbitrary states on equally long times
1
hψ, t| T (. . .) ψ 0 , −t

h0, t| T (. . .) |0, −ti =
c∗0 c00
hψ, t| T (. . .) |ψ 0 , −ti
= t → ∞, (55)
hψ| 0i h0| ψ 0 i
as long as we set H → H(1 − i) in the right hand side. We can
choose as states two position states; one in the very past and one in
the very future. Then we immediately write the expectation values of
operators in the vacuum state as a path integral
hx2 , t| T (. . .) |x1 , −ti
h0, t| T (. . .) |0, −ti = . (56)
hx2 | 0i h0| x1 i
The wave function of the vacuum still appears in this expression, but
only as a pre-factor. We will see that the overall normalization of the
1
The value of the vacuum energy depends on how we have chosen to calibrate the
energy. If the ground state for a Hamiltonian H is not calibrated to be zero, then we shift
the Hamiltonian by a constant H → H − h0| H |0i. The effect on the path integral is to
change its constant prefactor. As we shall see, this has no impact on physica predictions.
18
path integral is not important for the physics questions that we are
interested in.
1.6 The simple harmonic oscillator

An instructive example is the simple harmonic oscillator, with a Hamil-
tonian
p2 1
H= + mω 2 x2 . (57)
2m 2
For vacuum-to-vacuum transitions over large times we modify H →
H · (1 − i). For the path integral expression we need the Lagrangian:
L = pẋ − H(1 − i)
mẋ2 1
= mẋ2 − (1 − i) + mω 2 x2 (1 − i)
2 2
1 1
= m(1 + i)ẋ2 − mω 2 (1 − i)x2 , (58)
2 2
The i prescription of the Hamiltonian results to giving a small positive
imaginary part to the mass m and a small negative imaginary part to
the combination mω 2
m → m(1 + i),
mω 2 → mω 2 (1 − i).
We will now compute a so called “generating functional integral”
Z R
W [J] = Dx ei dt(L+J(t)x(t))
,
Z R
dt[ 12 m(1+i)ẋ2 − 12 mω 2 (1−i)x2 +J(t)x(t)]
= Dx ei , (59)
which also includes a source term in the action. Functional differ-

entiation with respect to the source J will permit as to compute a
large variety of vacuum-to-vacuum transitions including interactions
or perturbations.
Things become easier if we perform a Fourier transformation of all
quantities which depend on time:
Z ∞
dE −iEt
x(t) = e x̃(E), (60)
−∞ 2π
Z ∞
dE −iEt ˜
J(t) = e J(E). (61)
−∞ 2π
(62)
19
The action integral in the exponent of the generating functional be-
comes
1 1
Z
S[x] = m(1 + i)ẋ2 − m(1 − i)ω 2 x2 + J(t)x(t)
dt
2 2
1 1
Z
0
nh i
= 2 dEdE 0 dt e−it(E+E ) × −mEE 0 (1 + i) − mω 2 (1 − i) x̃(E)x̃(E 0 )
4π 2 o
˜ 0 ˜ 0)
+J(E)x̃(E ) + x̃(E)J(E
dE n h 2
Z i
= m E − ω 2 + i E 2 + ω 2 x̃(E)x̃(−E)
4π o
˜
+J(E)x̃(−E) ˜
+ x̃(E)J(−E)
dE n h 2
Z i o
= ˜
m E − ω 2 + i x̃(E)x̃(−E) + J(E)x̃(−E) ˜
+ x̃(E)J(−E)
4π (
˜ ˜
! !
dE J(E) J(−E)
Z h i
2 2
= x̃(E) + m E − ω + i x̃(−E) +
4π m [E 2 − ω 2 + i] m [E 2 − ω 2 + i]
˜ J(−E)
˜
)
J(E)
− (63)
m [E 2 − ω 2 + i]
In the above we simplified (E 2 + ω 2 ) → . We can also define

˜
J(E)
ỹ(E) = x̃(E) + (64)
m [E 2 − ω 2 + i]
The second term of the rhs corresponds to the Fourier transform of
the solution of the Euler-Lagrange equations; i.e. it corresponds to
the classical path. We can verify it easily. The classical equation of
motion for the harmonic oscillator with a source term and a small
negative imaginary part for the frequency is
" #
d2 xcl 2
m + ω − i xcl = J(t), (65)
dt2
and taking the Fourier transform
Z ∞
dE −iEt
xcl = e x˜cl ,
−∞ 2π
we obtain
h i
˜
m E 2 − ω 2 + i x̃cl (E) = −J(E)
˜
J(E)
; x̃cl (E) = − . (66)
m [E 2 − ω 2 + i]
20
Then the path integral over paths x(t), under a shift
x(t) = xcl (t) + y(t), (67)
becomes
R ˜
dE J(E)i ˜
J(−E)
− 21
Z R
i dE
W [J] = e 2π m[E 2 −ω 2 +i]
Dye 2 2π
ỹ(E)m[E 2 −ω2 +i]ỹ(−E)(68)
The dependence of the path integral on the source terms becomes a

simple pre-factor. It is easy to see that the new path integral corre-
sponds to the harmonic oscillator Hamiltonian with no sources. We
can therefore write the above expression as,
i
R ˜
dE J(E)(−1) ˜
J(−E)
2 2π m[E 2 −ω 2 +i]
hxf , +t| xi , −tiJ = e hxf , +t| xi , −ti , (69)
where t → ∞. Performing the inverse Fourier transformations, we

have:
Z ∞ ˜ J(−E)
dE −J(E) ˜
−∞ 2π E 2 − ω 2 + i

Z ∞ − dte −iEt J(t) dt0 eiEt0 J(t0 )
dE
=
2π E 2 − ω 2 + i
Z−∞
= dtdt0 J(t)G(t − t0 )J(t0 ), (70)
with 0
ei(t−t )E
Z +∞
0 dE
G(t − t ) = − (71)
−∞ 2π E 2 − ω 2 + i
We then write:
i
R
dtdt0 J(t)G(t−t0 )J(t0 )
hxf , t| xi , −tiJ = e 2 hxf , +t| xi , −ti . (72)
We can now differentiate this expression with respect to the sources

as many times as we need. We then obtain:
1 δn
R
i
hxf , +t| T x̂(tn ) . . . x̂(t1 ) |xi , −ti = e 2
in δJ(t1 ) . . . δJ(tn )
× hxf , +t| xi , −ti . (73)
Taking the limit of i → 0 will select the ground state on both sides of
the equation as asymptotic states (the normalization factors are also
21
the same on both sides of the equation). We therefore have for the
vacuum expectation value of time ordered position operators:
h0, t| T x̂(tn ) . . . x̂(t1 ) |0, −ti 1 δn

R
i
= e 2
h0, t| 0, −ti in δJ(t1 ) . . . δJ(tn )
(74)
Exercise:
1. Show that
h0, t| T x̂(tn ) . . . x̂(t1 ) |0, −ti = 0,
for n odd.
2. Compute h0, t| T x̂(t2 )x̂(t1 ) |0, −ti and h0, t| T x̂(t4 ) . . . x̂(t1 ) |0, −ti
in terms of G(t − t0 ).
3. Find a general expression for h0, t| T x̂(t2n ) . . . x̂(t1 ) |0, −ti in terms
of G(t − t0 ).
We now compute the propagator G(t1 , t2 ) explicitly.
dE e−i(t2 −t1 )E
Z
G(t1 − t2 ) = = −
2π E 2 − ω 2 + i
dE e−i(t2 −t1 )E
Z
= −
2π (E − ω + i) (E + ω − i)
dE e−i(t2 −t1 )E 1 1
Z
= − − (75)
.
2π 2ω E + ω − i E − ω + i
We can compute this integral using the residue theorem. If t2 > t1 we
can close the contour of integration on the lower complex half plane
where the exponential vanishes at infinity. If t1 < t2 we can only
close the contour on the upper complex half-plane. For both cases, we
obtain:
i −iω|t2 −t1 |
G(t1 − t2 ) = e . (76)
2ω
22
2 Path integrals and scalar fields
In this Chapter, we wil apply the quantization formalism of path inte-
grals in field theory. Earlier, for Quantum Mechanics, we found that
the canonical quantization procedure for a particle gives rise to an
equivalent path integral formulation.
Z R
[x̂(t), p̂(t)] = i ; Dx(t)ei dtL
. (77)
In quantum field theory, the canonical quantization procedure postu-

lates commutation relations to scalar fields
[φ(~x, t), π(~y , t)] = iδ 3 (~x − ~y ). (78)
The role of the position is now played by the field φ(x) and the role
of the time-coordinate (parameterizing paths) is played by the space-
time coordinates.
We will not attempt to derive a path integral formulation start-
ing with canonical quantization. Instead we will define quantum field
theory in terms of path integrals from the beginning, relying on the
analogies observed between Eq. 77 and Eq. 78. We will test the equiv-
alence of the two approaches in some cases by verifying that vacuum
expectation values or equivalently Feynman rules are the same for both
the path integral formalism and the canonical quantization formalism
of QFT I.
In analogy to quantum mechanics, we postulate a generating func-
tional for fields
Z R
d4 x(L+Jφ+iφ2 )
Z [J] = N Dφei (79)
where Z
L= d3 ~xL,
and the iφ2 is in order to dissipate the contributions from expectation

values in between states other than the vacuum. (Recall that in the
harmonic oscillator the H → H(1 − i) substitution yielded equivalent
Green’s function with adding a quadratic imaginary part in the poten-
tial) We start with the Lagrangian for a free real scalar field. We
will see that the results are as simple as in the case of the harmonic
oscillator in quantum mechanics. The Lagrangian density is
1
L= ∂µ φ∂ µ φ − m2 φ2 . (80)
2
23
The action integral in the exponent is:
1
Z
4 µ 2 2 2
S = d x ∂µ φ∂ φ − m φ + iφ + Jφ
2
1
Z h i
4 µ 2 2 2 2
= d x ∂µ (φ∂ φ) − φ∂ φ − m φ + iφ + Jφ
2
1 h 2
Z i
= − d4 x φ ∂ + m2 − i φ − Jφ (81)
2
The path integral for this action
Z
Z [J] = N DφeiS ,
is a straightforward generalization in four dimensions of the path-

integral of the simple harmonic oscillator. We can then compute it
repeating faithfully the steps of the previous lecture. An important
step in this computation is to shift the field by the solution of the
classical equations of motion.
φ → φ + φclassical
We then find a path integral free of source terms:

i
R
d4 xd4 yJ(x)∆F (x,y)J(y)
Z [J] = e 2 ×
Z R
− 2i d4 xφ ( ∂ 2 +m2 −i )φ .
N Dφe (82)
with the propagator
d4 k e−ik·(x−y)
Z
∆F (x, y) = − . (83)
(2π)4 k 2 − m2 + i
Exercise:
a) Write the Euler-Lagrange equations for the free real scalar field.
b) Evaluate the generating path integral for the free real scalar field
working in Fourier space and following the analogous derivation
of the simple harmonic oscillator.
c) Find the Fourier representation of the propagator.
d) Integrate the Fourier representation of the propagator over the
energy using the Cauchy theorem. Pay attention to the condi-
tions on the time variable in order to be able to use a closed
contour of integration.
24
The path integral for the free scalar field is no more difficult than
for the harmonic oscillator. We can act with functional derivatives
with respect to the sources on the above expressions. We then derive
expressions which, in analogy to Quantum Mechanics, are, likely, vac-
uum expectation values of field operators. From here, we can easily
develop a method for performing perturbative calculations and de-
rive Feynman rules. Before doing so, we will develop some additional
mathematical tools.
2.1 Functional Integration

We start with the familiar integral
Z ∞ r
−ax2 π
dxe = , a > 0. (84)
−∞ a
We then obtain
Z ∞ n
−
Pn
a x2 π2
dx1 . . . dxn e i=1 i i = , ai > 0. (85)
1
−∞ (a1 . . . an ) 2
We can define
A = diag (a1 , a2 , . . . , an ) , (86)

T
x = (x1 , . . . , xn ). (87)
We then rewrite the above integral as

Z ∞ !
Y dxi T Ax 1
√ e−x = 1 . (88)
−∞ i
π (detA) 2
Let us now perform a transformation
xi = Rij yj , or x = Ry. (89)
The integral can be written as

n
Z ∞ !
dyi T T 1
(detR) e−y (R AR)y =
Y
√ 1 . (90)
−∞ i=1
π (detA) 2
We define the matrix

B = RT AR. (91)
25
We can easily verify that B is symmetric.
T
B T = (RT AR)T = RT AT RT = RT AR = B.
The determinant of B is then
detB = det(RT AR) = (detA)(detR)2 .
We then find
n
Z ∞ !
dyi 1 T 1
e− 2 y
Y
By
√ = 1 , (92)
−∞ i=1 2π (detB) 2
where B is any real, positive definite (positive eigenvalues), symmetric
matrix.
We can add a linear term (source) in the exponent.
n
Z ∞ !
dyi 1 T
By+y T J
e− 2 y
Y
√ , (93)
−∞ i=1 2π
The minimum (classical field) of the exponent (action) is at
y0 = B −1 J. (94)
Substituting
y = y0 + x,
we find
n
Z ∞ !
dyi 1 T
By+y T J
e− 2 y
Y
√
−∞ i=1 2π
n
Z ∞ !
dx 1 T
Bx+ 21 J T B −1 J
√ i e− 2 x
Y
=
−∞ i=1 2π
1 T B −1 J
e2J
= 1 (95)
(detB) 2
Let us finally take the limit n → ∞. The exponent in the rhs is
originally a double sum over the two indices of the n × n matrix B.
In the infinite limit, sums turn into integrals and the inverse matrix
turns into an integration kernel which is the inverse of an operator.
We can then write:
1
R
d4 xd4 yJ(x)∆(x−y)J(y)
e2
Z R
d4 x 12 φ(x)Â(x)φ(x)−J(x)φ(x)
[ ]=
Dφe− p . (96)
detÂ
26
Â may be even a differential operator which has an inverse. For zero
sources, we obtain the identity
1
Z R
1
Dφe− 2 φ(x)Â(x)φ(x)
=p , (97)
detÂ
which we can consider as the definition of the determinant of a differ-
ential operator.
Applying the above to the generating functional for a free real
scalar field we find:
Z R
d4 x{− 12 φi[∂ 2 +m2 −i]φ+iJφ}
Z[J] = N Dφe
i
R
d4 xd4 yJ(x)∆F (x−y)J(y)
e− 2
= N p (98)
det [i (∂ 2 + m2 − i)]
The generating functional should have a normalization such that for
zero sources it describes a vacuum to vacuum transition; the corre-
sponding probability is unity in a free field theory and the correspond-
ing amplitude can also be taken to be unity:
Z[0] = h0, +∞| 0, −∞i = 1.
This fixes the normalization to be

q
N = det [i (∂ 2 + m2 − i)].
To conclude, we identify the propagator with the inverse of the

Klein-Gordon differential operator,
−1
∆F (x) = − ∂ 2 + m2 − i , (99)
or, more precisely,

− ∂ 2 + m2 − i ∆F (x) = δ 4 (x). (100)
Writing the Fourier transform of it
d4 k −ik·x ˜
Z
∆F (x) = e ∆F (k) (101)
(2π)4
and substituting in Eq. 100 we find
d4 k e−ik·x
Z
∆F (x) = . (102)
(2π)4 k 2 − m2 + i
27
What is the physical meaning of ∆F (x − y)? It is a transition
amplitude for a particle generated at a point x to propagate to a
point y if y 0 > x0 or from y to x if x0 > y 0 , i.e. the propagator for a
free particle. Indeed:

δ 2 Z[J]
− = h0, +∞| T φ(x)φ(y) |0, −∞i = . . . = ∆F (x − y).
δJ(x)δJ(y) J=0
(103)
This result for the propagator agrees with the canonical quantization
formalism. Exercise: Prove the above using canonical quantization.
You will need the Fourier integral of the propagator after you integrate
out the energy.
It is useful to know one more technical definition. We can define
the determinant of a Hermitian operator via path integration over
complex fields. Consider
Z ∞
π 2 +y 2
= dxdye−a(x )
a −∞
Z ∞
= dxdye−a(x−iy)(x+iy) . (104)
−∞
We define
z + z∗ z − z∗
x= , y=
2 2i
We obtain
dzdz ∗ −azz ∗ 1
Z
e = . (105)
2πi a
Following the same steps as before, we can write the determinant of a
Hermitian operator as:
1
Z R
d4 xφ∗ (x)Â(x)φ(x)
DφDφ∗ e− = . (106)
detA
28
END OF WEEK 2
29
2.2 Path integrals and interacting fields
The only computation needed for a free scalar field in the path integral
formalism is the evaluation of the propagator ∆F . The generating
functional requires this propagator as a kernel and it yields Green’s
functions by taking functional derivatives with respect to the sources.
It is extremely difficult or even impossible problem to compute exactly
the path integral for a Lagrangian with interacting fields. However,
the formalism can be easily adapted in order to enable the use of
perturbation theory.
Let us consider a real scalar field which can interact with itself,
L = L0 + LI , (107)
where
1
L0 = − φ(x) ∂ 2 + m2 − i φ(x), (108)
2
and
λ
LI = − φ(x)4 . (109)
4!
Vacuum expectation values of operators will then be given via func-
tional derivatives of
Z R
d4 x{L0 +φ(x)J(x)+LI }
Z[J] = N Dφei . (110)
We fix the normalization of the generating functional by requiring that

the amplitude for a vacuum to vacuum transition over infinitely long
times is the unity 2 :
Z[0] = 1.
We then write:
R
R i d4 x{L0 +φ(x)J(x)+LI }
Dφe
Z[J] = R . (111)
d4 x{L0 ++LI }
Dφei
R
We consider the case with λ 1. We separate the action into a part

corresponding to the free scalar particle with source terms and the
interaction part
S = S0 + SI , (112)
2
We have assumed that if necessary a constant is subtracted from the Hamiltonian, so
that the ground energy is zero; in canonical quantization this is achieved with a normal
ordering of the operators.
30
with Z
S0 = d4 x {L0 + φ(x)J(x)} , (113)
and Z
SI = d4 xLI . (114)
We can perform a perturbative expansion in λ,

∞
(iSI )n iS0
Z Z X
iS0 +iSI
Dφe = Dφ e
n=0
n!
h R in
Z ∞ i d4 y −λ φ4
X 4!
= Dφ eiS0
n=0
n!
4 n
−λ δ
i d4 y
R
∞
X 4! iδJ(y)
Z
= DφeiS0 . (115)
n=0
n!
In the last step, we used the fact that we can generate powers of φ in
the integrand of the path integral by acting with functional derivatives.
We now sum back the sum into an exponential. Our result for the
generating functional reads:
R 1 δ

i d4 xLI
DφeiS0
R
e i δJ(x)
Z[J] = R 1 δ
. (116)
i d4 xLI
DφeiS0
R
e i δJ(x)
J=0
Let us recall the expression for the generating path integral in the free
field theory:
Z R
d4 x(L+J(x)φ(x))
Z0 [J] = N Dφei
i
R
d4 xd4 yJ(x)∆F (x,y)J(y)
= e2 . (117)
The generating functional in the interacting theory is then related to

the generating functional in the free theory via,
R 1 δ

i d4 xLI
e i δJ(x) Z0 [J]
Z[J] = R 1 δ
. (118)
i d4 xLI
e i δJ(x) Z0 [J]
J=0
31
2.3 Perturbation theory for −λφ4 /4! interac-
tions
Let us consider the numerator of the generating functional Z[J],
R 1 δ

i d4 xLI
N um[J] = e i δJ(x) Z0 [J]
!
iλ δ4
Z R
i
d4 xd4 yJ(x)∆(x−y)J(y)
= 1− d4 z e2 + O(λ2 ).
4! δJ(z)4
(119)
We introduce a Feynman diagram notation by defining:
x y ≡ ∆(x − y),
Z
y ≡ d4 xJ(x)∆(x − y),
and
Z
≡ d4 xd4 yJ(x)∆(x − y)J(y).
With this notation, we write:
!
iλ δ4
Z
i
N um[J] = 1− d4 z e2 + O(λ2 ).(120)
4! δJ(z)4
We need to differentiate the exponential four times. It is easy to see

that
δ
= z + z =2 z ,
δJ(z)
and
δ
δJ(w) z = w z .
32
We find,
i
δe 2 i
=i z e2 , (121)
δJ(z)
 
 
i
δ2e 2
 
  i
i
= z −  e2 , (122)
δJ(z)2  z


 
and so on. We can then compute the generating function, and you
will find that:
!
N um[J] λ

i
Z[J] = = 1− 6 +i e2 +O(λ2 ) .
N um[0] 4!
(123)
Notice that in the above expression, all diagrams are connected to a
source. This is due to the normalization of Z[0] = 1. The denomina-
tor, after expanding in λ, removes all vacuum graphs. We can now
compute Green’s functions using this expression. It is useful to cast
the generating functional as an exponential. We can write,
Z[J] = eiW [J] , (124)
with
1 iλ λ
W [J] = + − + O(λ2 ). (125)
2 4 4!
Interestingly, only W [J] generates Green’s functions which contribute
to the S-matrix. We find:

1 δZ[J] δW [J]
= = h0| T φ(x1 ) |0i ,
i δJ(x1 ) J=0
δJ(x1 ) J=0

1 δ 2 Z[J] 1 δ 2 W [J]
= +h0| T φ(x1 ) |0i h0| T φ(x2 ) |0i ,

i2 δJ(x1 )δJ(x2 ) J=0

i δJ(x1 )δJ(x2 ) J=0
33

1 δ 3 Z[J] 1 δ 3 W [J]
=

i3 δJ(x1 )δJ(x2 )δJ(x3 ) J=0 i2 δJ(x1 )δJ(x2 )δJ(x3 ) J=0

+ h0| T φ(x1 )φ(x2 ) |0i h0| T φ(x3 ) |0i

+ h0| T φ(x1 )φ(x3 ) |0i h0| T φ(x3 ) |0i
+ h0| T φ(x3 )φ(x2 ) |0i h0| T φ(x1 ) |0i
−2 h0| T φ(x1 ) |0i h0| T φ(x2 ) |0i h0| T φ(x3 ) |0i ,
and so on. The diagrams which contribute to a scattering are the

ones where all derivatives act on a single W . The remaining terms
produce disconnected diagrams where subsets of particles scatter in-
dependently.
34
2.4 Fermionic path integrals
We have discussed the path integral for scalar fields which are bosonic.
To describe fermions, we need to formulate path integration over anti-
commuting functions.
We need to introduce Grassmann numbers, which are defined to
obey:
{ci , cj } = ci cj + cj ci = 0. (126)
This defintion means that any function constructed out of Grassmann
variables is at most linear in any of them, since:
c2i = 0.
Exercise:The product of two Grassmann variables a = c1 c2 commutes

wth a Grassmann variable c3 and normal numbers x:
[a, c3 ] = 0 [a, x] = 0.
Let us consider a concrete example of a function of two Grassmann

variables c1 , c2 . Its general form is:
f (c1 , c2 ) = a0 + a1 c1 + a2 c2 + a12 c1 c2 , (127)
where the coefficients a0 , a1 , a2 , a12 are normal commuting numbers.

We will now develop our calculus for functions of Grassmann variables.
We start by defining the derivative:
∂ci
= δij . (128)
∂cj
However, rearrangements of the Grassmann variables give rise to mi-

nus signs. For example, we can decide to write the c2 variable at the
left of any other Grassmann variable. The derivative with respect to
c2 is then
∂Lf ∂L
= (a0 + a1 c1 + a2 c2 + a12 c1 c2 )
∂c2 ∂c2
∂L
= (a0 + a1 c1 + a2 c2 − a12 c2 c1 )
∂c2
= a2 − a12 c1 . (129)
35
This is defined to be a “left” derivative (as the superscript denotes).
We could have defined a “right” derivative, where we always anti-
commute a Grassmann variable to the right of any other variable be-
fore we differentiate with it. In our example, we have:
∂Rf
= a2 + a12 c1 . (130)
∂c2
( )
∂
Ci , = δij , (131)
∂Cj
and ( )
∂ ∂
, = 0. (132)
∂Cj ∂Cj
We will always use left derivatives, unless it is explicilty stated oth-
erwise.
Our next step is to define an integration over Grassmann variables
ci . We only require to know two integrals
Z
dc 1 =?
and Z
dc c =?
since quadratic and higher terms in c vanish.

We require that Grassmann integrals are translation invariant:
Z Z
dc1 f (c1 + c2 ) = dc1 f (c1 ). (133)
This yields that Z

dc 1 = 0. (134)
We require that the remaining integral,

Z
dc c = x.
is an arbitrary (but fixed) commuting number x. We can always divide

the integration measure by x. It is then convenient to define:
Z
dc c = 1. (135)
36
We should be careful in integrating multiple variables, since
Z Z
dc1 c2 c1 = − dc1 c1 c2 = −c2 .
We can make a striking observation: “differentiation and integration

are identical operations”.
Z Z
dc2 f = dc2 (a0 + a1 c1 + a2 c2 + a12 c1 c2 )
Z
= dc2 (a0 + a1 c1 + a2 c2 − a12 c2 c1 )
= a2 − a12 c1
∂f
= . (136)
∂c2
Exercise: Show that linear combinations of Grassmann variables are
also Grassmann variables We must also investigate how we can per-
form a change of variables in integrals over Grassmann variables. We
consider with the integral
Z
dc1 dc2 . . . dcn f (c1 , c2 , . . . , cn ),
and perform a transformation (which is always linear)
ci = Mij bj . (137)
We shall have
Z Z
dc1 dc2 . . . dcn f (ci ) = (Jacobian) db1 db2 . . . dbn f (Mij bj ) . (138)
What is the Jacobian? In the case of a double integral, we have:

Z Z
dc1 dc2 c1 c2 = (Jacobian) db1 db2 (M11 b1 + M12 b2 ) (M21 b1 + M22 b2 )
Z
= (Jacobian) db1 db2 (M11 M22 b1 b2 + M12 M21 b2 b1 )
Z
= (Jacobian) db1 db2 b1 b2 (M11 M22 − M12 M21 )
Z
= (Jacobian)det(M ) db1 db2 b1 b2
1
; Jacobian = . (139)
det(M )
37
Exercise: Prove that the same results holds for multiple integrals of
arbitrary dimensions.
If we recall that Grassmann integration is in reality differentiation,
it is not surprising that the Jacobian of the transformation on Grass-
mann variables is the inverse of what emerges in integrations of normal
commuting variables.
Exercise: Consider the complex linear combinations
c1 + ic2 c1 − ic2
y= √ , ȳ = √ ,
2 2
of two real Grassmann variables c1 , c2 . Show that:
{y, ȳ} = 0
y + ȳ y − ȳ
Z Z
dc1 dc2 f (c1 , c2 ) = i dydȳf √ , √
2 2i
Exercise: For two complex Grassmann variables
After we have defined integration over Grassmann variables we
can study multiple exponential integrals which, in analogy with the
bosonic case, could be used to define a fermionic path-integral. Let
us consider two independent vectors of Grassmann variables xT =
(x1 , x2 ) and y T = (y1 , y2 ). We have
xT y = x1 y1 + x2 y2 , (140)
and
2
xT y = (x1 y1 + x2 y2 )(x1 y1 + x2 y2 )
= x1 y1 x2 y2 + x2 y2 x1 y1
= 2x1 y1 x2 y2 . (141)
We can easily see that

n
xT y = 0, n > 2. (142)
Therefore, using a Taylor expansion, we have

Z Z
Ty
dx1 dy1 dx2 dy2 e−x = dx1 dy1 dx2 dy2 [1 − (x1 y1 + x2 y2 ) + x1 y1 x2 y2 ]
= 1. (143)
We can now perform linear transformations on both x and y.
x → M x0 ,
38
y → N y0.
We find,
Z Z
−xT y T −1 −1 T MT Ny
1= dxdye = det(M ) det(N ) dxdye−x
Z
T MT Ny
= det(M T N )−1 dxdyex . (144)
Defining A = M T N , we obtain that

Z
T Ay
dxdye−x = det(A). (145)
Exercise: Prove the above for xT = (x1 , x2 , x3 ) and y T = (y1 , y2 , y3 ).

Generalize to arbitrasy dimensionality
Recall that for normal commuting variables we have
1
Z
† Ax
dxdx† e−x ∼ . (146)
detA
Let us now define a path integral over Grassmann variables by
considering an infinite number of them and taking the continuous
limit. This path integral will quantize the field ψ as in the scalar field
case. It will differ however from the Green’s functions of the bosonic
field. Let us consider the Lagrangian of a free Dirac fermion.
L = ψ̄ (i6∂ − m) ψ. (147)
We cab write a generating functional

Z R
d4 x[L+āψ+ψ̄a]
Z0 [a, ā] = N DψDψ̄ei . (148)
In this path integral the integration variables ψ, ψ̄ and the sources a, ā

are all independent Grassmann functions. The constant N is fixed as
usual by requiring that
Z0 [0, 0] = h0| 0i = 1.
Everything works in an analogous way as in the case of the path in-

tegral for the scalar field. For a theory without interactions, we can
“complete the square” and compute the generating functional explic-
itly if we shift he fields ψ, ψ̄ by a constant corresponding to their
39
classical value which minimizes the action. We need the inverse of the
Dirac wave operator,
−(i6∂ − m1)S(x − y) = δ 4 (x − y) 1. (149)
As for the scalar field and the harmonic oscillator we can write a
Fourier representation,
d4 k 6k + m −ik·(x−y)
Z
S(x − y) = − e (150)
(2π)4 k 2 − m2
Exercise:Write the Euler-Lagrange equations for the action with sources

a, ā. Solve these equations in Fourier space. The values of ψ, ψ̄ which
minimize the action are now given by
Z
ψcl = d4 yS(x − y)a(y),
Z
ψ¯cl = d4 yā(y)S(x − y),
and the change of variables
ψ = ψcl + η, ψ̄ = ψ̄cl + η̄,
yields R
d4 xd4 yā(x)S(x−y)a(y)
Z0 [a, ā] = ei . (151)
40
END OF WEEK 3
41
We obtain expectation values of time-ordered products of field op-
erators from the generating functional via
1 δ δ
h0, +∞| T . . . ψi (x) . . . ψ̄j (y) . . . |0, −∞i = . . .
...i . . . Z0 [a, ā],
i δāi (x) δaj (y)
(152)
where the indices i, j are spinor indices. Exercise:Calculate the ex-
pectation values
• h0, +∞| T ψi (x1 )ψ̄j (x2 ) |0, −∞i
• h0, +∞| T ψi (x1 )ψ̄j (x2 )ψk (x3 )ψ̄l (x4 ) |0, −∞i
•
and compare them with your results from canonical quantization. We
can also develop a perturbative expansion repeating faithfully the
steps we performed in the φ4 scalar field theory. It is not hard to
convince ourselves that a completely analogous formula should be
valid here for the generating functional when an interaction term LI
is present in the Lagrangian:
R δ

i d4 zLI i δa (x), 1i δ
,
e δā(x) Z0 [a, ā]
Z [a, ā] = R δ
. (153)
i d4 zLI i δa (x), 1i δā(x)
δ
,
e Z0 [a, ā]
a=ā=0
Besides the main similarities in the appearance of the formulae
there are also very important differences which are encoded in the
Grassmann algebra of the functions which we integrate upon. What
is very important to remember, is the fact that functional derivatives
anticommute, generalizing the result that we found for the derivatives
of discrete Grassmann variables:
δ δ δ δ δ δ

, = , = , = 0. (154)
δa(x) δa(y) δā(x) δā(y) δa(x) δā(y)
We should also remember that these derivatives are left (by conven-
tion) derivatives, and the order in which sources appear in the inte-
grand matters indeed.
Exercise: Consider a theory with a fermion a real scalar and a
Yukawa interaction λψ̄ψφ. Calculate the fermion and scalar propa-
gators through O(λ2 ). Derive the Feynman diagrams contributing to
the scattering of four fermions. Derive the Feynman diagrams con-
tributing to the scattering of two fermions and two scalars. 3
3
This exercise is very important for checking your understanding of the material cov-
ered so far. Please spend as much time as needed until you get it right.
42
3 Non-abelian gauge theories
3.1 Gauge invariance
We will introduce a new principle which is found in realistic quan-
tum field theories, i.e. theories such as QED, QCD and the Standard
Model. These theories describe nature to the experimentally accessible
accuracy. Their Lagrangian is invariant under local gauge transforma-
tions. We start with the familiar case of Quantum Electrodynamics.
You have already seen the Lagrangian in the course of QFT I. Here
we will construct it by imposing invariance under local U(1) transfor-
mations. We consider a free fermion field:
L = ψ̄(x) (i6∂ − m) ψ(x) (155)
and define a transformation:
ψ(x) → ψ 0 (x) = exp(igθ)ψ(x). (156)

∂θ
We first consider a global gauge transformation ∂x = 0 . The free
Lagrangian is clearly invariant under the gauge transformation of
Eq. 156. However, under a local gauge transformation,
U (x) = exp (igθ(x)) , (157)
the Lagrangian is no longer invariant.

h i
ψ̄6∂ ψ → ψ̄6∂ ψ + ψ̄e−igθ 6∂ eigθ ψ. (158)
We cannot make any local gauge transformation which leaves in-

variant the Lagrangian of a single field. The problem was that
6∂ ψ → eigθ 6∂ ψ + something else .
We will modify the derivative so that it transforms more conveniently.

We will look for a new derivative which, under a local gauge transfor-
mation transforms as:
6Dψ → U (x)6Dψ. (159)
We add to the normal derivative a new function (field):
Dµ = ∂µ − igAµ (x). (160)
43
We need:
Dµ ψ(x) → Dµ0 ψ 0 = U (x)Dµ ψ

; ∂µ − igA0µ (U (x)ψ) = U (x) (∂µ − igAµ ) ψ
; U (x)∂µ ψ + [∂µ U (x)] ψ − igA0µ U (x)ψ = U (x)∂µ ψ − igAµ U (x)ψ
i
; A0µ = Aµ − U −1 (x)∂µ U (x) (161)
g
The covariant derivative transforms as:
Dµ → Dµ0 = ∂µ − igA0µ
i −1

= ∂µ − ig Aµ − U ∂µU
g
−1
= ∂µ − igAµ − U (∂µ U )

= ∂µ − igAµ + U ∂µ U −1
= U (x) (∂µ − igAµ ) U −1 (x) (162)
Therefore:
Dµ → Dµ0 = U (x)Dµ U −1 (x) (163)
We can now replace the free Lagrangian of the spin-1/2 field with
a new Lagrangian which is also gauge invariant.
L = ψ̄ [i6D − m] ψ
→ ψ̄U −1 U [i6D − m] U −1 U ψ
ψ̄ [i6D − m] ψ.
If Aµ is a physical field, we need to introduce a kinetic term in the

Lagrangian for it. We will insist on constructing a fully gauge invariant
Lagrangian. To this purpose, we can use the covariant derivative as a
building block. Consider the gauge transformation of the product of
two covariant derivatives:
Dµ Dν → Dµ0 Dν0 = U Dµ U −1 U Dν U −1
= U Dµ Dν U −1 .
This is not a gauge invariant object. Now look at the commutator:

h i
[Dµ , Dν ] → Dµ0 , Dν0
= U [Dµ , Dν ] U −1 (164)
44
This is gauge invariant. To convince ourselves we write the commu-
tator explicitly:
[Dµ , Dν ] = (∂µ − igAµ ) (∂ν − igAν ) − [µ ↔ ν]

= ∂µ ∂ν − ig (∂µ Aν ) − igAν ∂µ − igAµ ∂ν + (ig)2 Aµ Aν − [µ ↔ ν]
= −ig [∂µ Aν − ∂ν Aµ ] . (165)
Inserting Eq. 165 into Eq. 164, we find that the commutator of covari-
ant derivatives (in the abelian case) is gauge invariant. We have also
found that it is proportional to the field strength tensor of the gauge
(photon) field:
i
Fµν = [Dµ , Dν ] = ∂µ Aν − ∂ν Aµ (166)
g
We now have invariant terms for a Lagrangian with an “electron”
and a “photon” field. The Lagrangian for QED reads
1
L = ψ̄ (6D − m) ψ − Fµν F µν . (167)
4
Exercise: Find the Noether current and conserved charge due to
the invariance under the U (1) gauge transformation of the QED La-
grangian.
Exercise: Think of at least five more operators that one can add to
the QED Lagrangian without spoiling gauge invariance. Thesre is an
infinite number of them. Which of them do not have mass dimension
four? As we shall see, operators of higher dimension spoil renormal-
izability
3.2 Non-abelian (global) SU (N ) transformations

We can think of more complicated transformations than a simple phase
on a complex field. Let’s consider as an example a collection of N
scalar fields. A simple Lagrangian for them could be:
λ
L = (∂µ φi ) (∂ µ φ∗i ) − m2 (φi φ∗i ) − (φi φ∗i )2 , i = 1 . . . N, (168)
4
where we have used Einstein’s double index summation. The La-
grangian is symmetric unde transformations of the φi fields. Let us
transform
φi → φ0i = Vij φj , (169)
45
where once again a summation convention is assumed (this will always
be the case unless explicitly stated otherwise). The following term is
invariant:
φ∗i φi → φ0∗ 0 ∗ ∗
i φi = Vij φj Vik φk
= Vji† Vik φ∗j φk = φ∗i φi , (170)
if
Vji† Vik = δjk . (171)
This is a unitary U (N ) transformation under which our example La-
grangian is invariant. The Lagrangian is also invariant under a stricter
SU (N ) transformation. We can write
V = eiα U,
such that
U †U = 1 and detU = 1. (172)
Exercise: Think of a concrete example for a unitary 2 × 2 matrix V
where this can be done. The N × N matrices U represent the group
of special unitary transformations SU (N ).
It is sufficient to study small SU (N ) transformations. Due to them
forming a group, large transformation can be obtained by repeating
(infinitely) many small ones. We write:
Uij = δij − iθa Tija + O(θ2 ), (173)
where we choose θa to be real parameters. The N × N matrices T a

are generators of SU (N ) matrices. They are N 2 − 1: An arbitrary
N × N complex matrix has 2N 2 real elements. For a unitary matrix
U † = U −1 , only N 2 elements are independent. The specialty condi-
tion detU = 1 adds one more constraint, leaving N 2 − 1 independent
elements. Remember the dimensionality of the indices concerning the
generators.
a
Ti,j : a = 1 . . . (N 2 − 1) and i, j = 1 . . . N.
They will be needed in various situations.

Exercise SO(N ) group: N × N matrices with
Rij Rkl δjl = δik .
46
Find the number of generators.
Exercise The symplectic group Sp(2N ) can be defined as 2N × 2N
matrices S with
Sij Skl δjl = ηik ,
where,
ηij = −ηji and η 2 = −1.
Find the number of generators.
The SU (N ) generators are hermitian:
†
U

U =1
; 1 + ig(T a )† θa 1 − igT b θb = 1 + O(θ2 )
; T a† = T a (174)
and traceless:
detU = 1 ; log (detU ) = 0
; T r (log U ) = 0
; T r log (1 − iθa T a ) = 0
; T r (iθa T a ) = 0
; T r (T a ) = 0 (175)
We can choose a normalization condition for the SU (N ) generators.
By convention, we choose:
δab
T r T a T b ≡ Tija Tjib = . (176)
2
A very basic property of the generators is that they satisfy a Lie
algebra: h i
T a , T b = if abc T c , (177)
where f abc are the structure constants of the algebra.
Exercise: Prove Eq. 177 by considering a transformation U 0−1 U −1 U 0 U ,
with U, U 0 independent SU (N ) transformations.
From Eq. 177, we can derive
h i h i
T a , T b = if abd T d ; T a , T b T c = if abd T d T c
h i
; f abc = −2iT r T a, T b T c . (178)
Exercise: Prove that the structure constants are fully antisymmetric

and real
47
3.3 Local non-abelian gauge symmetries
We now consider N fields φi (scalar or spinor). We are interested in
Lagrangians which are invariant under a local SU (N ) transformation:
φi (x) → φ0i (x) = Uij (x)φj (x). (179)
As in QED, the building block for the construction of a gauge invariant

Lagrangian will be a covariant derivative:
Dµ = ∂µ − igAµ , (180)
such that
Dµ → Dµ0 = U Dµ U † , with U † = U −1 . (181)
For a scalar field φ we have
(Dµ0 φ0† )(D0µ φ0 ) → (Dµ φ† )(Dµ φ)
Similarly, the kinetic term with the same covariant derivative for a
fermion field is invariant. There are many similarities with QED,
however there are also many important differences. Let’s start by
pointing out that the gauge field Aµ is an N × N matrix.
We can easily find the transformation for the gauge field:
Dµ → Dµ0 = U (x)Dµ U † (x)

; ∂µ − igA0µ = U (∂µ − igAµ ) U †

; ∂µ − igA0µ = ∂µ + U ∂µ U † − U igAµ U †
i
; A0µ = U (x)Aµ U † (x) + U (x) ∂µ U † (x) (182)
g
This formula is analogous to the gauge transformation of the photon in
QED. However, here the gauge field Aµ and the gauge transformation
U are complex N × N matrices rather than complex numbers.
We now compute the corresponding commutator:
[Dµ , Dν ] = (∂µ − igAµ ) (∂ν − igAν ) − [µ ↔ ν]

= ∂µ ∂ν − ig (∂µ Aν ) − igAν ∂µ − igAµ ∂ν + (ig)2 Aµ Aν − [µ ↔ ν]
= −ig {∂µ Aν − ∂ν Aµ − ig [Aµ , Aν ]} . (183)
The commutator term was absent in the case of QED.
48
The gauge field strength:
i
Gµν ≡ [Dµ , Dν ] = ∂µ Aν − ∂ν Aµ − ig [Aµ , Aν ] , (184)
g
is no longer gauge invariant:
Gµν → G0µν = U (x)Gµν U † (x).
However, the trace
T r(Gµν Gµν )
→ T r(U Gµν U † U Gµν U † )
→ T r(U † U Gµν U † U Gµν )
→ T r(Gµν Gµν )
is gauge invariant.
We can expand the gauge field in the basis of generators:
Aµ = Aaµ T a , (185)
Equivalently,
Aaµ = 2T r(Aµ T a ) (186)
We also have
Gµν = Gaµν T a , (187)
with
Gaµν = 2T r(Gµν T a ). (188)
It is
Gaµν = 2T r(Gµν T a )
= 2T r (∂µ Aν T a − ∂ν Aµ T a − ig [Aµ , Aν ] T c )
= ∂µ Aaν − ∂ν Aaµ + gf abc Abµ Acν . (189)
A concrete example is the Lagrangian for QCD which is invariant

under SU (3) gauge transformations.
1
L = ψ̄ (i6D − m1) ψ − Gcµν Gcµν . (190)
4
Exercise:Expand all terms in the QCD Lagrangian using the explicit
expressions in terms of the gauge field for the covariant derivative and
the gauge field strength. Sketch the interactions (the precise Feynman
49
rules will be derived in forthcoming lectures)
Exercise:Write a gauge invariant Lagrangian under SU (N ) trans-
formations for a scalar field. This case appears in supersymmetric
theories for the scalar partners of quarks. Sketch the interactions.
Adjoint representation:
By expanding the commutators we can prove that
hh i i hh i i h i
T a, T b , T c + T b , T c , T a + [T c , T a ] , T b = 0. (191)
From this we derive a relation for the structure constants:

h i h i h i
; f abd T d , T c + f bcd T d , T a + f cad T d , T b = 0
; f abd f dce + f bcd f dae + f cad f dbe = 0 (192)
We define the matrices

b
T̃ac = if abc . (193)
Then the above relation can be written as
h i
T̃ b , T̃ c = if bcd T̃ d . (194)
Therefore, the matrices T̃ furnish a representation of the same Lie

algebra. This is called the adjoint representation.
We can now consider fields ψa which transform in the adjoint rep-
resentation. An example is the gluino, the supersymmetric partner of
the gluon, which transforms in the adjoint.
c T̃ c
ψa → ψa0 = UA ab ψb with UAab = e−iθ ab and a, b, c = 1 . . . (N 2 −1).
(195)
Or, for small transformations,
ψa → ψa0 = ψa − iθb T̃ac

b
ψc

; ψa0 = ψa − iθb if abc ψc
; ψa0 = ψa + f abc θb ψc (196)
We can find a covariant derivative for the transformations of the ad-

joint representation in a complete analogy as for the “fundamental”
representation:
Dµ ψa = ∂µ ψa − igAbµ T̃ac
b
ψc
= ∂µ ψa + gf abc Abµ ψc . (197)
50
Now consider a general member of the represenation ψ = ψa T a . The
covariant derivative acts on it like:

Dµ ψ = ∂µ ψa + gf abc Abµ ψc T a
h i
= ∂µ ψ − ig T b , T c ψc Abµ , (198)
or equivalently
Dµ ψ = ∂µ ψ − ig [Aµ , ψ] , ψ ≡ ψa T a , Aµ ≡ Aaµ T a . (199)
Exercise: Take an element ξ = ξ a T a of the Lie algebra, transform-

ing in the adjoint representation, and a field χ transforming in the
fundamental representation. Prove the Leibniz rule for the covariant
derivative:
Dµ (ξχ) = (Dµ ξ) χ + ξ (Dµ χ) . (200)
Euler-Lagrange equations/ conserved currents:

We consider the variation of the gauge field
Aµ → Aµ + δAµ
and
∂ν Aµ → ∂ν Aµ + δ (∂ν Aµ ) .
The corresponding variation of the gauge field strength is
δGµν = δ (∂µ Aν − ∂ν Aµ − ig [Aµ , Aν ])

= ∂µ δAν − ∂ν δAµ − ig [δAµ , Aν ] − ig [Aµ , δAν ]
; δGµν = Dµ (δAν ) − Dν (δAµ ) . (201)
We can now look at the variation of this term in the QCD action
Z
δ d4 xT r (Gµν Gµν )
Z
= 2 d4 xT r (Gµν δGµν )
Z
= 2 d4 xT r (Gµν [Dµ (δAν ) − Dν (δAµ )])
Z
= 4 d4 xT r (Gµν Dµ (δAν )) antisymmetry
Z
= 4 d4 x {Dµ [T r (Gµν (δAν ))] − T r [Dµ (Gµν ) δAν ]} (202)
51
The first trace is gauge invariant. We can then find a gauge transfor-
mation for the gluon field so that Dµ → U Dµ U † = ∂µ , and drop the
surface term. So, we have:
−1
Z Z
4
δ d xT r Gµν Gµν = 2 d4 xT r [Dµ (Gµν ) δAν ] . (203)
2
The fermionic term in the Lagrangian varies as:
Z Z
δ d4 xψ̄ (i6D − m) ψ = δ d4 xψ̄ (i6∂ + g6A) ψ
Z Z
4
= g d xψ̄δ6Aψ = g d4 xψ̄γ µ T a ψδAaµ =
Z
= 2g d4 x ψ̄i γ µ Tija ψj T r [δAµ T a ]
= −2gT r [Jµ δAµ ] , (204)
with
Jµ = Jµa T a , (205)
and
Jµa = −g ψ̄γ µ T a ψ. (206)
Combining the variation of both fermionic and gauge boson terms in
the Lagrangian of Eq. 190 we derive the Euler-Lagrange equation:
Dµ Gµν = J ν . (207)
We can act with a second covariant derivative on the above equation:
Dν Dµ Gµν = Dν J ν (208)
The lhs is:
Dν Dµ Gµν = Dν (∂µ Gµν − ig [Aµ , Gµν ])

= ∂ν ∂µ Gµν
−ig [∂ν Aµ , Gµν ]
−ig [Aµ , ∂ν Gµν ] − ig [Aν , ∂µ Gµν ]
−g 2 [Aν [Aµ , Gµν ]] (209)
Using that Gµν = −Gνµ and
[A, [B, C]] + [B, [C, A]] + [C, [A, B]] = 0
52
you can prove that
−ig
Dν Dµ Gµν = [Gµν , Gµν ] = 0 (210)
2
Therefore:
Dµ J µ = 0. (211)
The fermionic current is thus no longer (as in QED) a conserved cur-
rent. It is rather covariantly conserved!
Exercise: In an abelian gauge theory, consider the dual tensor
1
F̃ µν = µνρσ Fρσ .
2
Show that
F µν F̃µν = ∂µ K µ , (212)
with
K µ = µνρσ Aν Fρσ
Exercise: In a non-abelian gauge theory, consider the dual tensor
1
G̃µν = µνρσ Gρσ .
2
Show that
Gµν G̃µν = ∂µ K µ , (213)
with
2

µ ρσµν
K = T r Gσµ Aν + Aσ Aµ Aν
3
.
53
END oF WEEK 4
54
4 Quantization of non-abelian gauge
theories
We now have a sufficient formalism to quantize a non-abelian gauge
theory. The classical Yang-Mills Lagrangian is,
1
LYM = − Gaµν Gaµν . (214)
4
with
Gaµν = ∂µ Aaν − ∂ν Aaµ + gf abc Abµ Acν . (215)
Proceeding in analogy with the quantization of the scalar field theory,

we write the following path integral:
Z
d4 x[LYM (Aµ µ
R
a )+Ja Aaµ ]
Z[Jaµ ] = N DAaµ ei (216)
We would like to develop a perturbation theory program. Remember

4
what we required in the −λ φ4! case. We could try to repeat the same
steps here:
- find the propagator of the free-field by inverting the differential
operator in the quadratic part of the Lagrangian. This would
give us an expression for the path integral when all interactions
are switched off (g = 0)
d4 xd4 yJaµ (x)∆µνab (x−y)Jbν (x)
R
Z0 [Jaµ ] = N 0 ei . (217)
- derive the perturbative expansion from

R
1 δ
i d4 zLint i δJ µ (z)
Z[J] ∼ e a Z0 [Jaµ ]. (218)
The free-field action (g = 0) is

1
Z
Sf ree = − d4 x ∂µ Aaν − ∂ν Aaµ (∂ µ Aνa − ∂ ν Aaµ )
4
1
Z h i
= − d4 x (∂µ Aaν ) (∂ µ Aνa ) − ∂ν Aaµ (∂ µ Aνa )
2
1
Z h i
= − d4 x ∂µ (Aaν ∂ µ Aνa ) − Aaν ∂µ ∂ µ Aνa − ∂ν Aaµ ∂ µ Aνa + Aaµ ∂ν ∂ µ Aνa
2
1
Z h i
= − d4 xAaµ δ ab −∂ 2 gµν + ∂µ ∂ν Aνb . (219)
2
55
We now need to find the inverse of the operator
h i
−∂ 2 gµν + ∂µ ∂ν δ ab .
However, it turns out that there is none! This operator has zero
eigenvalues, and its determinant is zero. In particular we obtain zero
when it acts on any function that can be written as a total derivative:
h i
−∂ 2 gµν + ∂µ ∂ν ∂ ν Λ(x) = −∂ 2 ∂µ Λ(x) + −∂ 2 ∂µ Λ(x) = 0. (220)
Our naive attempt to establish a perturbation expansion in g using

a path integral formalism has failed at the first step. However, there
is a property of the theory, gauge invariance, which we have not yet
used and we can exploit it to remove the zero-modes of the operator
in the free part of the Lagrangian.
Let us start by defining a δ-functional, in analogy to a δ-function,
which we will need in a while. The integral over a δ-function is
Z
df δ(f ) = 1.
We can change variables, f = f (w), and we obtain

∂f
Z
dw δ(f (w)) = 1.
∂w
The multidimensional generalization of this equation is:
!
∂fi
Z
1= dw1 . . . dw1 det δ(f1 (w1 , . . . , wn )) . . . δ(fn (w1 , . . . , wn )).
∂wj
We can take the limit n → ∞ which yields a functional integral over

w(x), where x is the continuous variable corresponding to the index
i = 1 . . . n. We define the infinte product of delta functions as a delta
functional. We write:
δf (x)
Z
Dwdet δ [f (w)]
δw(y)
!
∂fi
Z
≡ lim dw1 . . . dw1 det δ(f1 (w1 , . . . , wn )) . . . δ(fn (w1 , . . . , wn ))
n→∞ ∂wj
= 1. (221)
Notice the emergence of a functional determinant, due to changing

variables in the measure of a functional integral.
56
We now return to the gauge-theory; the action is of course invariant
under gauge transformations
i
Aµ → A0µ = U (x)Aµ U † (x) + U (x) ∂µ U † (x) , (222)
g
with Aµ ≡ Aaµ T a . The transformation matrices U are determined by
as many independent parameters as the generators of the Lie group,
a (x)T a
U (x) = e−iθ = 1 − iθa (x)T a + O(θ2 ). (223)
DAaµ
R
The integration in the path integral formalism does not dis-
criminate among the fields which are connected via a gauge transfor-
mation.
a(independent)
Consider all the fields Aµ which cannot be connected
via a gauge transformation. We never know them explicitly, but we
can impose that they satisfy gauge-fixing conditions of the form
F1 (Aaµ ) = 0,
which remove the superfluous degrees of freedom. For example, a very
common choice is a Lorentz gauge fixing condition,
∂µ Aaµ = 0.
Of course, we are allowed to choose other conditions to fix the gauge,
F2 (Aaµ ) = 0,
The gauge fields Aµa (2) and Aµa (1) satisfying the two different gauge
conditions are related via gauge transformations; for each of the solu-
tions of the second gauge-fixing condition there is a unique set of gauge
transofrmation parametetes θa which maps it to a solution of the first
gauge-fixing condition. Therefore, if we consider all infinite possibili-
ties for gauge-fixing conditions, we enumerate all infinite members of
the Lie algebra parameters θa .
Let us now go back to the path integral for the gauge theory with-
out sources: Z
µ
R 4
Z = N DAaµ ei d x[LYM (Aa )] (224)
This integrates over all fields including the ones related by gauge trans-
formations. In other words, had we thought of all possible gauge fix-
ings, it integrates over all these possibilities. We can write:
Z
d4 x[LYM (Aµ
R
a )]
Z = N DAaµ ei ×1
Z Z
d4 x[LYM (Aµ
R
a )]
= N DAaµ ei × DF a δ[F a (Aaµ )], (225)
57
where
F a (Aµa ) = 0 ; Aaµ = Aaµ (θb ).
Different gauge fixings F a correspond to different group parameters, so
we may change variables F a → θa in the second functional integration.
δF a (Aaµ )
Z Z
d4 x[LYM (Aµ
R
a )]
Z = N DAaµ ei Dθ det b
δ[F a (Aaµ (θb ))]
δθb
δF a (Aaµ )
Z Z !
µ
R
b a i d4 x[LYM (Aa )]
= N Dθ DAµ e det δ[F a (Aaµ (θb ))]
δθb

a aµ b
F (A (θ ))=0
(226)
Now there is a crucial observation to be made. No term in the inner

functional integral depends on θb . Let us justify this statement. We
have,
Z
1= DF a δ [F a (Aaµ )]
1
Z
; a
aµ
= Dθb δ(F a (Aaµ (θb )))(227)
det δF δθ(Ab )

F a (Aaµ (θb ))=0
The rhs is gauge invariant. We have

aT a
U = eiθ ; θa T a = ln U ; θa = 2T r(T a ln U ).
b
Under a gauge transformation U 0 = eiT bθ we have
θa → θc = θa + θb , (228)
and
δθc

c a
Dθ = Dθ det = Dθa .
δθa
A gauge transformation only reshuffles the elements of the Lie algebra;
we can compute the measure over all possible Lie algebra elements in
any gauge. Therefore:
1
Z
a aµ
= Dθb δ(F a (Aaµ (θb )))
det δF δθ(Ab )

F a (Aaµ (θb ))=0
Z
= Dθc δ(F a (Aaµ (θc )))
1
. (229)
δF a
det

δθc
F a (Aaµ (θc ))=0
58
This determinant is therefore just a gauge-invariant number. The
exponential of the Yang-Mills action is of course gauge invariant:
R R
d4 xLYM (Aµ ) d4 xLYM (Aa b
ei = ei µ (θ )) .
Finally, the gauge-boson fields are also members of the Lie algebra
and only get reshuffled by gauge transformations. The measure DAaµ
can be evaluated at any gauge.
DAaµ = DAaµ (θb ). (230)
Exercise: Consider a gauge transformation Aaµ → Aaµ 0 . Prove that

DAaµ = DAaµ 0 . You only need to consider an infinitesimal gauge trans-
formation.
We then compute all terms in the inner path inegral at the specific-
gauge chosen by the delta-functional.
δF a (Aaµ )
Z Z !
d4 x[LYM (Aµ
R
a )]
Z = N Dθ b
DAaµ ei det δ[F a (Aaµ (θb ))]

δθb

F a (Aaµ (θb ))=0
δF a (Aaµ (θb ))
Z Z !
d4 x[LYM (Aµ
R b
a (θ ))]
= N Dθb DAaµ (θb )ei det δ[F a (Aaµ (θb ))]
δθb
δF a (Aaµ )
Z Z !
d4 x[LYM (Aµ
R
a )]
= N Dθ b
DAaµ ei det δ[F a (Aaµ )] (231)
δθb
In the last line we noticed that the gauge-fixed field variable Aaµ (θb ) is
a dummy integration variable. The path integration over all possible
gauge transformations (corresponding to all possible gauge fixings) is
an overall normalization factor. This is an infinite integration over
the measure of infinite Lie algebra parameters. However, this is not a
problem if we want to compute physical quantities, since the overall
normalization will cancel. We therefore end up with the gauge-fixed
path integral
δF a (Aaµ )
Z Z !
d4 x[LYM (Aµ
R
0 a )]
Z=N Dθ b
DAaµ eidet δ[F a (Aaµ )].
δθb
(232)
It is very important to notice that (up to an irrelevant infinite nor-
malization) this path integral does not depend on the gauge-fixing
condition F (Aaµ ) that we may choose!
This new-path integral in Eq. 232 integrates over fields which can-
not be related via gauge transformations; it should therefore be fine
59
to derive Green’s functions for fields which are physically distinct.
However, the new expression has two features, the gauge-fixing delta-
functional and the determinant, which were absent in our formulation
of perturbation theory. It is therefore unclear at first sight how to
establish a calculable perturbative expansion with the mathematics
that we now. Two very clever tricks will come to our rescue.
Without loss of generality we write
F a (Aaµ ) = G a (Aaµ ) − wa (x). (233)
We are allowed to multiply the path-integral with an overall constant

without any physical consequences. We then multiply with the factor,
wa (x)2
Z R
−i d4 x
C= Dwa e 2ξ . (234)
We can do this because Z[Jaµ ] in Eq. 232 does not depend on w(x).
We have
δG a (Aaµ )
!
wa (x)wa (x)
Z R Z
d4 x[LYM (Aµ
R
a −i d4 x a )]
Z ∼ Dw e 2ξ DAaµ ei det δ[G a (Aaµ ) − wa (x)]
δθb
a (Aa ) Z
!
δG w(x)2
Z R
LYM (Aµ
R
d4 x [ ] det µ a −i d4 x
= DAaµ ei a)
Dw e 2ξ δ[G a (Aaµ ) − w(x)]
δθb
δG a (Aaµ )
Z !
d4 x LYM (Aµ
R 1 a aµ ))2

i a )− 2ξ (G (A
= DAaµ e det (235)
δθb
With this trick, we remove the delta-function from the integrand add
modify the exponent of the path-integral. This also yields a well-
defined propagator for the gauge boson field. If we choose for example
the gauge-fixing condition
G a (Aaµ ) = ∂µ Aaµ ,
the free-part (g = 0) of the action in the exponent becomes:
1 1
Z
4 aµ ab 2
Sf ree = − d xA δ −∂ gµν + 1− ∂µ ∂ν Abν .(236)
2 ξ
Now, the new differential operator has an inverse:
d4 k e−ikx kµ kν
Z
∆ab
µν =δ ab
4 2
gµν − (1 − ξ) 2 . (237)
(2π) k + i k + i
60
Exercise: Find the inverse of the operator:
1

2 2
−∂ + M gµν + 1− ∂µ ∂ν.
ξ
This will be the case of a massive gauge-boson such as W, Z.
Exercise: Find the gauge boson propagator in an axial gauge
G(A) = nµ Aaµ ,
where n is a light-like vector n2 = 0.
We now need to deal with the determinant in the integrand of the

path integral. Here we will use a result that we found from infinite
integration over Grassmann variables. We proved earlier that:
Z
T Ay
dx1 . . . dxn dy1 . . . dyn e−x = det(A). (238)
where A is an n × n matrix and x, y are Grassmann variables. We can

take the limit of n → ∞. We then obtain express a functional deter-
minant as a fermionic path integrals over two independent Grassmann
functions x and y:
Z R
d4 x1 d4 x2 y(x1 )(gA(x1 −x2 ))x(x2 )
ig det(A) = DyDxei . (239)
The Fadeev-Popov idea was to introduce two new fields with odd spin-
statistics (Grassmann variables in the path-integral), a ghost and an
anti-ghost, and write the determinant as functional integral over an
exponential. We write,
δG a (Aaµ )
Z !
d4 x LYM (Aµ
R 1 a aµ ))2

i a )− 2ξ (G (A
Z ∼ DAaµ e det ig
δθb
Z
d4 x LYM (Aµ
R 1 a aµ ))2

i a )− 2ξ (G (A
∼ DAaµ Dη̄ a Dη a e ×
δG a (Aa (θ a ))
R
µ
i d4 x1 d4 x2 η̄ a (x1 ) g η b (x2 )
δθ b
×e . (240)
We need not to worry about computing precisely the overall normal-

ization of the path-integral. This will drop out when we require that
vacuum to vacuum transitions have a unit amplitude. The factor −ig
is convenient, in order to later combine easierly all terms under a
common exponential.
61
Let us consider a local gauge transformation
aT a
U (x) = e−iθ ,
under which a gauge-field transforms as

i
Aµ (θ) = U (x)Aµ U † (x) + U (x) ∂µ U † (x) , (241)
g
where, as usual, Aµ = Aaµ T a . For a small transformation, we obtain
1 h µ ab i
Aµ,a (θ) = Aµ,a − ∂ δ − gf abc Aµ,c θb . (242)
g
We recognize in the above expression the covariant derivative in the
adjoint representation:
Dµab ≡ ∂µ δ ab − gf abc Acµ , (243)
so we can write:
1
Aµ,a (θ) = Aµ,a − Dµ,ab θb . (244)
g
We then have,
δAµ,a (θ(x)) 1 δ (Dµ,ac (x)θc (x)) 1

b
= − b
= − Dµ,ab (y)δ(x − y). (245)
δθ (y) g δθ (y) g
We can now compute the functional derivative
δG(Aµ,a (θ(x))) δG(Aµ,a (θ(x))) δ(Aν,c (θ(z)))

Z
g = g d4 z
δθb (y) δ(Aν,c (θ(z))) δθb (y)
Z µ,a
δG(A (x)) ν,cb
= − d4 z D (z)δ(z − y)
δ(Aν,c (z))
δG(Aµ,a (x)) ν,cb
= − D (y) (246)
δ(Aν,c (y))
We can then write the following expression for the path integral:
Z
d4 x LYM (Aµ
R 1 a aµ ))2

i a )− 2ξ (G (A
Z ∼ DAaµ Dη̄ a Dη a e ×
δG a (Aa

µ (x1 )) cb
R
i d4 x1 d4 x2 η̄ a (x1 ) − δAc
Dν (x2 ) η b (x2 )
×e ν (x2 ) . (247)
This is valid for any arbitrary gauge fixing condition G a (Aaµ ).
62
It will be instructive and practical (this is what we need to do when
we compute elements of the S-matrix) to choose a gauge. A common
choice is the Lorentz gauge:
G a (Aaµ ) = ∂µ Aµ,a , (248)
where
δ∂µ Aµ,a (x1 )
= ∂µ gνµ δ ac δ(x1 − x2 ) = ∂ν δ ac δ(x1 − x2 ). (249)
δAν,c (x2 )
Therefore, in the Lorentz gauge, the path integral is (up to a normal-

ization):
Z R
i d4 x − 14 Ga a,µν − 1 (∂ Aaµ )2 −η̄ a (x)∂ D µ;ab η b (x)
µν G
Z= DAaµ Dη̄ a Dη a e 2ξ µ µ
.
(250)
After a partial integration, we obtain:
Z R
i d4 x − 41 Ga a,µν − 1 (∂ Aaµ )2 +(∂ η̄ a )D µ;ab η b
µν G
Z= DAaµ Dη̄ a Dη a e 2ξ µ µ
.
(251)
In this last expression, we have exponentiated all terms which arose
from gauge-fixing. The argument of the exponential is a new ac-
tion, modified by a gauge-fixing term and contributions from the ghost
fields.
Notice that the form of the quadratic term in the ghost fields is the
same as for a complex scalar field. However, the variables η, η̄ in the
path-integral are anti-commuting Grassmann variables. Therefore,
the ghost field is a scalar field with the “wrong” spin-statistics.
We also observe that for an abelian gauge theory f abc = 0 we
have Dµab = δ ab ∂µ and there is no interaction term for the ghost field
and the gauge-boson. In this case, the ghost field is a free field and
we can integrate out its contribution to the path integral (changing
the irrelevant overall normalization). This is why in QED you never
needed to introduce it.
Exercise: Find the expression of the path-integral for an SU (N )
Yang-Mills theory in an axial gauge
G(A) = nµ Aaµ ,
where n is a light-like vector n2 = 0.
63
4.1 Perturbative QCD
After gauge-fixing and the Fadeev-Popov method, we can formulate a
path integral for QCD, where the path-integral action is:
Z
S= d4 xL,
with
L = LYANG−MILLS + LFERMION + LGAUGE−FIXING + LFADEEV−POPOV .
The classical Yang-Mills Lagrangian is:

1
LYANG−MILLS = − Gaµν Ga,µν ,
4
and
Gaµν = ∂µ Aaν − ∂ν Aaµ + gf abc Abµ Acν .
The gauge fixing and the Fadeev-Popov terms, in the Lorentz gauge,
are:
1 µ a 2
LGAUGE−FIXING = − ∂ Aµ .
2ξ
LFADEEV−POPOV = (∂ µ η̄ a ) Dµab η b ,
and the fermion term:

LFERMION = ψ̄ i iγ µ Dµij − mδ ij ψ j .
The covariant derivtives in the adjoint and fundamental representation

are
Dµab = ∂µ δ ab − gf abc Acµ
and
Dµij = ∂µ δ ij − igTijc Acµ
accordingly.
We would like to compute Green’s function using perturbation the-
ory. If we switch off the coupling, g = 0, then we are left with terms
which are quadratic in the fields, and we can compute the correspond-
ing path integral for the free action. We define,
L = Lf ree + Linteracting , (252)
with
Lf ree = L|g=0 .
64
Explicitly,
1 1 µ a ν a
Lf ree = − ∂µ Aaν − ∂ν Aaµ (∂ µ Aa,ν − ∂ ν Aa,µ ) − ∂ Aµ (∂ Aν )
4 2ξ
+ (∂µ η̄ a ) (∂ µ η̄ a )
+ψ̄ i (iγ µ ∂µ − m) ψ i (253)
It is convenient to use integration by parts and cast the free Lagrangian

in a “standard form”:
1
− (FieldA ) Ô (FieldA ) + ∂µ (. . .)
2
or, for terms with independent fields (appearing separately in the mea-
sure of the path-integral),
− (FieldA ) Ô (FieldB ) + ∂µ (. . .) .
We have
1
Lf ree = − Aaµ K ab,µν Abν
2
−η¯a K ab η b
−ψ̄ i Λij ψ j
+∂µ (. . .) , (254)
with (∂ 2 ≡ ∂µ ∂ µ )
1

ab,µν ab µν 2
K =δ −g ∂ + 1 − ∂µ∂ν , (255)
ξ
K ab = δ ab ∂ 2 , (256)
ij ij
Λ = δ (m − i6∂ ) . (257)
An essential step in order to compute the generating functional for

the free path-integral is to find the inverse of the above operators, i.e.
objects which diagonilize them in all indices. We define:
ac
Kµρ (x)Dcb,ρν (x − y) = δ ab gµν δ (4) (x − y),
K ac (x)Dcb (x − y) = δ ab δ (4) (x − y),
Λik (x)S kj (x − y) = δ ij δ (4) (x − y). (258)
Exercise: Solve these equations. The solutions are given next in the
text and it is easy to verify whether they are correct or not by insertibg
65
them above. However, it will be instructive to think how to find them
if nobody told you the answer!
We find that
d4 k e−ik·x kµ kν
Z
ab
Dµν (x) = δ ab 4 k 2 + i gµν − (1 − ξ) k 2 + i (259)
(2π)
d4 k e−ik·x
Z
Dab (x) = δ ab (−1) (260)
(2π)4 k 2 + i
d4 k e−ik·x
Z
S ij (x) = δ ij 4 k 2 − m2 + i (−1) (6k + m) . (261)
(2π)
The generating functional for the free-Lagrangian is
h i Z Z
Z0 Jψ , Jψ̄ , Jη , Jη̄ JA = DψDψ̄DηDη̄DAµ exp i d4 x
"
1
− Aaµ K ab,µν Abν − η¯a K ab η b − ψ̄ i Λij ψ j
2
#!
+JAa,µ Aaµ + Jψ̄i ψ i + ψ̄ i Jψi + Jη̄a η a + η̄ a Jηa
(262)
where we have included independent sources for all fermion, anti-
fermion, ghost, anti-ghost, and gauge fields. We should keep in mind
that only the source JAa,µ for the gauge field is a bosonic (commut-
ing) variable; all other sources are Grassmann variables. We can now
“complete squares” by shifting the fields as,
Z
Aa,µ (x) → Aa,µ (x) + d4 yDµν
ab
(x − y)JAb,ν (y) (263)
Z
η a (x) → η a (x) + d4 yDab (x − y)Jηb (y) (264)
Z
η̄ a (x) → η̄ a (x) + d4 yJη̄b (y)Dba (x − y) (265)
Z
ψ i (x) → ψ i (x) + d4 yS ij (x − y)Jψj (y) (266)
Z
ψ̄ i (x) → ψ̄ i (x) + d4 yJψ̄j (y)S ji (x − y). (267)
We then obtain (with an undetermiend overall constant factor),

"
1 a,µ
h i Z
Z0 Jψ , Jψ̄ , Jη , Jη̄ , JA = N exp i 4
d xd y 4 ab
J (x)Dµν (x − y)JAb,ν (y)
2 A
66
#!
+Jη̄a (x)Dab (x − y)Jηb (y) + Jψ̄i (x)S ij (x − y)Jψj (y) (268)
We will now deal with the interaction Lagrangian; this is defined

as
Linteraction = L − Lf ree . (269)
We find
Linteraction = g ψ̄ i Tija 6Aa ψ j

−g (∂µ η̄ a ) f abc Aa,µ η b

−2gf abc Ab,µ Ac,ν ∂µ Aaν − ∂ν Aaµ
g 2 abc ade b c d,µ e,ν
− f f Aµ Aν A A . (270)
4
The generating functional for the full theory can be obtained pertur-
batively, expanding in g,
h i
Z Jψ , Jψ̄ , Jη , Jη̄ , JA
( Z !)
δ δ δ δ δ
= N exp i d4 zLinteraction −i ,i , −i ,i , −i
δJA δJψ δJψ̄ δJη δJη̄
h i
Z0 Jψ , Jψ̄ , Jη , Jη̄ , JA , (271)
where we must replace in the interaction Lagrangian the field variables

with functional derivatives with respect to their corresponding sources.
Exercise:
h i
• Compute Z Jψ , Jψ̄ , Jη , Jη̄ , JA through O(g 2 ) in the expansion
around g = 0.
• Compute the generating functional of connected diagrams
h i h i
W Jψ , Jψ̄ , Jη , Jη̄ , JA = −i log Z Jψ , Jψ̄ , Jη , Jη̄ , JA
through the same order.

• Compute the gauge-boson propagator through the same order
h0, +∞| T Aa,µ (x)Ab,ν (0) |0, −∞i
• Transform this expression in momentum space
67
END OF WEEK 5
68
5 BRST symmetry
We recall here the path-integral for a non-abelian gauge theory with
a fermion field,
Z R
i 1
d4 x LYM +Lf ermion − 2ξ (G a (Aµa ))2 +η̄ a ∆a
Z ∼ DAaµ Dψ̄ i Dψ i Dη̄ a Dη a e ,
(272)
where we have defined,
δG a (Aaµ (x)) cb
Z !
a 4
∆ (x) = − d y Dν (y) η b (y). (273)
δAcν (y)
It is convenient to linearize the action in the gauge-fixing term. If we

introduce (yet) another (bosonic) field, wa , we can re-write,
R h a 2 a Ga
i
ξ
d4 x wa + G2 −G
Z Z
a i2
R
d4 x 2ξ wa wa +wa G a
( ) =
Dwa ei Dw e ξ2
R a a
i d4 x −G2ξG
= Nx e (274)
We will denote,
Lcl = LYM + Lfermion , (275)
the classical Lagrangian which satisfies local gauge-invariance. The
path-integral, up to a normalization, is equal to
Z R
d4 x[Lcl +η̄ a ∆a +wa G a + 2ξ wa wa ]
Z= DAaµ Dψ̄ i Dψ i Dη̄ a Dη a Dwa ei .
(276)
If we want to go back to the original form of Z we simply need to
integrate out the field wa . The exponent is not gauge invariant, except
the term with the classical Lagrangian Lcl .
We find a closely related symmetry which is called Becchi-Rouet-
Stora; Tyutin or, shorter,BRST symmetry which leaves invariant not
only the classical Lagrangian but also the sum gauge-fixing and Fadeev-
Popov terms. The BRST symmetry transformations are gauge trans-
formations of a special form for the gauge boson Aaµ and fermion ψ
fields which enter the classical Lagrangian Lcl . the gauge boson Aaµ
and fermion ψ. The gauge parameter is made out of the ghost field
and a Grassmann variable. Specifically, the fermion transforms as:
δθ ψ = −iT a (θη a ) ψ. (277)
69
This is equivalent to classical gauge transformation with the replace-
ment
θa (x) → θη a .
where both η and the ghost field η a are Grassmann variables. Of
course, their product is a bosonic variable as expected by a classical
gauge transformation. We take the parameter θ to be global
∂µ θ = 0.
For the gauge boson, we require that the BRST transformations is
also a gauge transformation with gauge parameter θη a :
θ
δθ Aaµ = − Dµab η b . (278)
g
Similarly, the anti-fermion field transforms as:
δθ ψ̄ = ψ̄iT a (θη a ) .
Notice that the Grassmann variable entering here is η, so that we
perform the same gauge-transformation on all classical fields ψ, ψ̄, Aaµ .
Before we present the trasnformations for the remaining fields, we
state a characteristic property of the transformation: Two succes-
sive BRST transformations on an arbitrary function of fields
leave the function invariant (nilpotent transformation).
δθ2 δθ1 F (A, ψ, ψ̄, η, η̄) = 0. (279)
If we insist on this property, we obtain:
0 = δθ2 (δθ1 ψ)
= δθ2 (−iT a θ1 η a ψ)
= −iT a θ1 [(δθ2 η a )ψ + η a (δθ2 ψ)]
h i
= −iT a θ1 (δθ2 η a )ψ + η a −iT b θ2 η b ψ
h i
= −iT a θ1 (δθ2 η a ) − iη a θ2 η b T b ψ (280)
Equivalently,
T c δθ2 η c = iT b T c η b θ2 η c
; tr(T a T c )δθ2 η c = −iθ2 tr(T a T b T c )η b η c beware of Grassmann!
δ ac ηbηc − ηcηb
; δθ2 η c = −iθ2 tr T a T b T c
2 h i 2
; δθ2 η c = −iθ2 tr T a T b , T c η b η c
; δθ2 η c = f bcd tr(T a T d )η b η c , (281)
70
Therefore we require the ghost field to transform as:
θ
δθ η a = f abc η b η c . (282)
2
Two successive transformations on the gauge field produce,
θ1 h i
δθ2 δθ1 Aaµ = − δθ2 ∂µ η a − gf abc η b Acµ
g
θ1 h i
= − ∂µ (δθ2 η a ) − gf abc δθ2 η b Acµ − gf abc η b δθ2 Acµ
g
θ1 1 abc b

ab a cd d
= − Dµ (δθ2 η ) + f η θ2 Dµ η = . . . = 0. (283)
g 2
Exercise:Fill the dots... Prove the above statement using the anti-
commutation of Grassmann variables and the Jacobi identity for the
structure constants
For the remaining two independent fields in the action of the path
integral, we have no unabiguous guidance in order to construct their
BRST transformations. We will make two very simple choices. We
perform no transformation on the auxiliary bosonic field,
δθ wa = 0.
For the anti-ghost we require that under a a BRST transformation it

gets shifted by a a constant.
1 a
δθ η̄ a = θw .
g
This choice as we will see guarantees BRST invariance of the quantum
action. Notice that
δθ1 δθ2 η̄ a = δθ1 δθ2 wa = 0.
Let us now compactify the notation. We consider any field F from

{Aµ , ψ, ψ̄, η, η̄, w}. We will introduce the short-hand notation:
δθ F ≡ θ(sF )
We have pulled an explicit prefactor of the BRST transformation pa-

rameter, and the notation sF denotes the remainder of the expression
after we transformed the field F . For example, we write
δθ ψ = −igT a θη a ψ ; sψ = −igT a η a ψ.
71
Notice that if the field F is a bosonic field then sF is Grassmannian
and vice versa. If we perform two consecutive BRST transformations,
we have:
δθ1 δθ2 F = θ1 θ2 s2 F.
Nilpotency means that:
s2 F = 0. (284)
Nilpotency is a property valid for a product of two variables as well.
Performing a BRST transformation on the product of two fields we
have:
δθ1 (F1 F2 ) = (δθ1 F1 )F2 + F1 (δθ1 F2 )
θ1 (sF1 )F2 + F1 θ1 (sF2 )
θ1 [(sF1 )F2 ± F1 (sF2 )] , (285)
where the minus sign arises if F1 is a Grassmann variable. We have
used here that the field F and sF have always opposite spin-statistics.
If we perform a double BRST transformation on the product F1 F2 we
then find,
δθ2 δθ1 (F1 F2 ) = θ1 δθ2 [(sF1 )F2 ± F1 (sF2 )]
= θ1 [(sF1 )θ2 (sF2 ) ± θ2 (sF1 )(sF2 )]
= θ1 θ2 [∓(sF1 )(sF2 ) ± (sF1 )(sF2 )]
= 0. (286)
We can continue in the same spirit. We find that:
δθ2 δθ1 (F1 F2 . . . Fn ) = 0. (287)
In fact every functional of these fields satisfies,
δθ2 δθ1 G[(F1 , F2 , . . . , Fn )] = 0.
We will return to this shortly.
We should investigate the effect of a BRST transforamtion on the
gauge fixing term G a in the Lagrangian. G a is a function of the gauge
field and we should use the chain-rule.
δG a (x)
Z
δθ G a (Aaµ (x)) = d4 y b δθ Abµ (y)
δAµ (y)
1 δG a (x) ab b
Z
= − θ d4 y D η (y)
g δAbµ (y) µ
θ a
= ∆ (288)
g
72
Let us now consider the variation:
ξ ξ

a
δθ η̄ a
G + wa = (δθ η̄ ) G + wa + η̄ a (δθ G a )
a a
2 2
θ ξ a a

a a a a
= η̄ ∆ + w G + w w . (289)
g 2
In other words, the non-classical part of the Lagrangian is already a
total variation under a BRST-transformation. Due to the property of
nilpotency, such terms remain invariant under a BRST-transformation.
ξ ξ

a a a a
δθ Lcl + η̄ ∆ + w G + wa wa = (δθ Lcl )+gδθ s η̄ a G a + wa =0
2 2
(290)
We remind here that the classical-part of the Lagrangian is BRST-
invariant due to its gauge-invariance.
Exercise:Prove that the Jacobian of a BRST transformation is unit.
This completes a proof that the full path-integral is BRST-invariant.
The BRST symmetry provides us with the asymptotic states of
the S-matrix. Let us compute the S-matrix element
hα| βi ,
in a two different gauges, G1 and G2 , where the two gauge fixing con-
ditions differ by little:
G2 = G1 + δ̃G.
We require that the initial |βi and final hα| states are physical and
thus the same in both gauges:
hα|G1 = hα|G2 , |βiG1 = |βiG2 .
The matrix-element is computed through the path integral Z, which

has an apparent but fake dependence on the gauge. A difference due
to the different gauge-fixing conditions shows up due to the change of
the action in the exponent of the path-integral. We will have:
Z R Z R
i d4 Lcl+gs K|G i d4 Lcl+gs K|G
δ̃Z = ZG2 −ZG1 = DA . . . e 1 − DA . . . e 2
Since we have considered infinitesimally different gauge-fixing condi-

tions, we have that:
δ̃Z = ZG1 igsδ̃K.
73
Demanding that this difference has no physical consequences leads to a
selection criterion for physical states. Promoting δ̃K into an operator,
we would like that it yields a zero expectation value when inserted in
a matrix-element for the transition in between physical states:
0 = hα| sδ̃K |βi .
We can construct an operator Q which is the generator of the BRST

transformations for canonical fields.
δθ (F ield) = i [θQ, F ield]±
where a commutator for a field with even spin and an anti-commutator

for a field with odd spin are understood with the ± subscript: [A, B]± =
AB ± BA. We can write:
h i h i
0 = s(s F ield) = Q, [Q, F ield]∓ = Q2 , F ield .
± −
For the above to be satisfied for every field we need:
Q2 = 0.
Using the BRST generator we write:

h i
hα| sδ̃K |βi = hα| Q, δ̃K |βi = hα| QK |βi ± hα| KQ |βi .
±
The matrix-element is the same in all gauges if the above vanishes;

this provides us with a condition for physical fields. Computations of
matrix-elements in different gauges yield the same result if the physical
external states annihilate the generator of BRST transformations:
hα| Q = Q |βi = 0. (291)
We have discovered a new symmetry transformation which leaves

the gauge-fixed path-integral of a non-abelian gauge theory invariant.
The BRST transformation is very important for the renormalization
of the theory since it constrains the UV divergencesof S-matrix ele-
ments.
5.1 Application to the free electromagnetic field

If we trun off the coupling constant g non-abelian gauge theories re-
duce to the free QED Lagrangian. Also, the in and out states of
74
the S-matrix are constructed from the requirement that asymptotic
states at times very far in the future and the past are states of the
interaction-free Lagrangian. It is therefore very useful to study how
BRST symmetry can be used to select physical states in the case of
the free electromagnetic field. The generating functional for QED in
the Lorentz gauge takes the form:
ξ
Z Z
Z = DAµ Dη̄DηDw exp i 4
d x Lcl + w2 + w∂µ Aµ + (∂µ η̄)(∂ µ η) ,
2
(292)
We recover a form which is non-linear in the gauge-fixing condition if

we integrate the bosonic field w:
(∂µ Aµ )2
Z ( Z " #)
4
Z = DAµ Dη̄Dη exp i d x Lcl − + (∂µ η̄)(∂ µ η)(293)
,
2ξ
The classical Euler-Lagrange equation of motion for the auxiliary field

w is
1
w = − ∂ µ Aµ (294)
ξ
Notice that we can obtain the part-integral with the field integrated
out in Eq. 293 by substituting in the path-integral of Eq. 292 the
equation of motion of Eq. 294.
The BRST transformations with the w field set to its classical value
(or integrated out) are:
θ θ
δθ η = 0, δθ η̄ = − ∂ µ Aµ , δθ Aµ = − ∂ µ η. (295)
gξ g
We now write the quantum fields as a superposition of plane waves:
d3 k
Z X µh i
∗ −ik·x
Aµ (x) = aλ e ik·x
+ aλ e (296)
(2π)3 2ωk λ λ
d3 k
Z h i
η̄(x) = beik·x + b∗ e−ik·x (297)
(2π)3 2ωk
d3 k
Z h i
η(x) = ceik·x + c∗ e−ik·x , (298)
(2π)3 2ωk
where the operators b, b∗ and c, c∗ are not necessarily Hermitian con-
jugates and µλ are polarization vectors. We take them to satisfy the
normalization
0
λ · λ0 = g λλ . (299)
75
Substituting these equations into the BRST transformations of Eq. 295,
we obtain (exercise):
[Q, aλ ] = −(k · λ ) c (300)

[Q, a∗λ ]
= (k · λ ) c ∗
(301)
1X
{Q, b} = (λ · k) aλ , (302)
ξ λ
1X
{Q, b∗ } = (λ · k) a∗λ , (303)
ξ λ
{Q, c} = 0. (304)
∗
{Q, c } = 0. (305)
(306)
A state is physical if it annihilates the BRST operator:
Q |physi = 0.
Let us consider a state
|λ, physi = αλ∗ |physi ,
which has in adddition a photon with polarization λ . Acting with

the BRST generator on it, we obtain:
Q |λ, physi = Qαλ∗ |physi

= [Q, αλ∗ ] |physi
= (k · λ ) c∗ |physi , (307)
which is zero if k · λ = 0. We conclude that a photon state is physical

if the polarization vector is transverse to the momentum, in agreement
with our expectations.
Now, we consider a state
b∗ |physi
which also contains an anti-ghost field. Acting on it with the BRST

generator we obtain
Qb∗ |physi = {Q, b∗ } |physi

1X
= (k · λ ) a∗λ |physi , (308)
ξ λ
76
The polarizations which are transverse to the momentum yield a zero
contribution to the sum. However, a polarization vector µII will sur-
vive. We thus have that the BRST generator transforms an anti-ghost
state into a photon-state with a longitutidal polarization.
Qb∗ |physi = k · k |k, physi . (309)
Reading this equation in the opposite direction,
|k, physi ∼ Qb∗ |physi (310)
a state with a longitutidal photon is a “BRST exact” state, meaning

that it can be written as the BRST generator acting on a different
state. While such a state satisfies the physical condition
Q |k, physi ∼ Q2 b∗ |physi = 0,
it does not contribute to the S-matrix. Indeed,
phys0 k, physi ∼ phys0 Qb∗ |physi = 0.

(311)
Similarly, we find that a state with a ghost is unphysical
.... (312)
Suggested further reading:

“A Brst Primer”
D. Nemeschansky, C. R. Preitschopf and M. Weinstein.
10.1016/0003-4916(88)90233-3
Annals Phys. 183, 226 (1988).
77
END OF WEEK 6
78
6 Quantum effective action and the ef-
fective potential
We have started to collect essential tools for the renormalization of
gauge field theories by proving the existence of the BRST symmetry.
In the following lectures we will convince ourselves that gauge theories,
such as QCD, are renormalizable. It turns out that for renormalization
we need to worry only about one-particle-irreducible (1PI) Feynman
diagrams. If these are rendered finite, then the full S-matrix elements
will possess no other ultraviolet singularities. In this Section, we wll
introduce a new functional, the quantum effective action, which gen-
erates only 1PI graphs. The quantum effective action is also a very
important tool in order to define the ground state of the quantum field
theory, and to study symmetry breaking via quantum effects.
We have worked with the generating functional
Z R
d4 xJ(x)φ(x))
Z[J] = Dφei(S[φ]+ ;
Green’s functions are obtained via:

δ n Z[J]

1 1
h0| T φ(x1 ) . . . φ(xn ) |0i = n .
i Z[J] δJ(x1 ) . . . δJ(xn ) J=0
We found that if we require only connected graphs to be generated,

which are relevant for computing S-matrix elements, we should use
the generating functional W [Z], with
Z[J] = eiW [J] ; W [J] = −i log Z[J].
Then,
δ n W [J]

1
h0| T φ(x1 ) . . . φ(xn ) |0iCONECTED = n−1 .
i δJ(x1 ) . . . δJ(xn ) J=0
An important Green’s function is the one-point function

δW [J]
h0| φ |0i (x) = ,
δJ(x) J=0
which we typically find it to vanish for physical fields that can create
asymptotic states. However, there are situations that a field (external
field) fills the vacuum h0| φ |0i (x) 6= 0. Such a field cannot be in the
79
final or initial state of a scattering process, but it can be a background
where scattering of other fields takes place. For example, the Higgs
field may have such a role. The field vacuum expectation value does
not have to vanish either when the corresponding source term in the
functional is not set to zero. In the presence of sources, we have
δW [J]
hφiJ ≡ h0| φ |0iJ (x) = .
δJ(x)
The above equation defines a relation between the source J(x) and
the vacuum expectation of the corresponding field. We can then trade
source functions in the path-integral and in functional derivatives with
the corresponding vev’s (vacuum expectation values).
hφiJ = function(J), J = function−1 (hφiJ ).
Considering it as differential equation, we solve

Z
W [J] = d4 x hφij (x)J(x) + Γ [hφiJ ] (313)
where the last term is a constant of integration and does not depend on
the source J(x). This constant of integration is the quantum effective
action Z
Γ [hφi] = W [J] − d4 xJ(x) hφi J(x). (314)
and it is a functional of field vacuum expecation values with very

interesting properties. From the above we find the simple equation,
δΓ [hφi]
= −J(x). (315)
δ hφi (x)
Recall the role of the classical action S[φ]. The equations of motion
for the classical field are found by requiring that the action takes a
minimal value
δS[φ]
= 0.
δφ φ=φclassical
At the quantum level, fields are promoted into operators. The ana-
logue of the classical fields in quantum field theory is the expectation
value of the field operator in the state, usually ground state, of the
system. In the absence of sources, the quantum effective action yields
the equations of motion for the average values of quantum fields:
δΓ [hφi]
= 0. (316)
δ hφi0 (x)
80
6.1 The quantum effective action as a gener-
ating functional
We may take a next step and promote the quantum effective action to
generate new Green’s functions. We use it in the exponent of a path
integral: Z R
eiWΓ [g,J]
= D hφi e
i
g { Γ[hφi]+ d4 xJ(x)hφi(x)}
. (317)
We can establish perturbation theory using this path integral. It is

possible to derive propagators by identifying the quadratic terms in
the action and inverting the corresponding operator. We will not do
this explicitly here; we are rather interested in counting powers of the
arbitrary constant g.
Every propagator in a graph (since it is produced by inversion),
will contribute a single power of g. Vertices are derived from the non-
quadratic terms in the Lagrangian without any inversion. Thus, each
vertex will contribute a power 1/g to a Feynman graph. For a graph
with NI propagators and NV vertices the overall power of the coupling
is:
g NI −NV .
All connected graphs generated by WΓ 4 with NI propagators and NV
vertices have L = 1 + NI − NV loops. It only takes trying a few ex-
amples out in order to convince ourselves about the above statement.
Otherwise, assign NI unconstrained momenta for each propagator.
Each vertex will provide one constraint, of which one combination is
an overall delta function stating that the sum of momenta of incoming
an outgoing particles is zero. NI − NV + 1 momenta are left uncon-
strained and they are thus loop momenta. The power of g for a graph
is therefore determined exclusively from the number of loops that it
posseses:
g L−1 .
We can then perform an expansion:
∞
X (L)
WΓ [g, J] = g L−1 WΓ [J]. (318)
L=0
Of course, we can still be interested in the case with g = 1. What the

4
they are connected because WΓ is the logarithm of a path integral
81
above expression tells us,
∞
X (L)
WΓ [1, J] = WΓ [J], (319)
L=0
is that the generating functional WΓ [1, J] can be decomposed as a

(L)
sum of independent generating functionals WΓ [J] corresponding to
(L)
different loop orders. The functionals WΓ are independent in the
sense that shifts in the measure do not mix them; symmetries of the
full action should therefore be symmetries of each one of the loop
contributions sepearately.
Let us explore the possibility of a very small parameter g. We can
expand the exponent in the path integral around the value:
δΓ[hφi]
hφi = hφiJ + η, with = −J(x).
δ hφiJ (x)
We have for the exponent of the path integral:

Z Z
4
Γ[hφi] + d xJ hφi = Γ[hφiJ ] + d4 xJ hφiJ +
" #
δΓ[hφi]
Z
4
+ d xη(x) +J
δ hφiJ
δ 2 Γ[hφi]
Z
+ d4 xd4 yη(x) η(y) + . . .
δ hφiJ (x)δ hφiJ (y)
(320)
The linear term in η vanishes. We therefore have,

Z Z
4
Γ[hφi]+ d xJ hφi = Γ[hφiJ ] + d xJ hφiJ +O(η 2 ) = W [J]+O(η 2 ).
4
(321)
We then have,
i
P∞ (L) i
Z
i
R
g L WΓ [J] W [J] O(η 2 )
e g L=0 =e g Dηe g . (322)
The path integral over η can be computed perturbatively. The factor

1 0 1/2 . Then we are left with a
g can be eliminated by redefining η = η g
“canonically” normalized quadratic part. The important observation
to make is that this perturbative expansion will start at order O(g 0 )
82
the earliest. After taking the logarithm of both sides of the above
equation, and by comparing the g1 coefficients, we find that:
(0)
W [J] = WΓ [J]. (323)
In other words, the full generating functional of connected graphs

W [J] can be obtained by a generating function where we have replaced
the classical action with the quantum effective action:
Z R Z R
i[S[φ]+ d4 xJ(x)φ(x)] i[Γ[φ]+ d4 xJ(x)φ(x)]
−i log Dφe = Dφe ,
TREE
(324)
and keeping only the tree-contributions (denoted with the sub-
script in the integral symbol).
This is a remarkable result; it states that it is possible to re-
organize the perturbative expansion, which gives rise to both tree
and loop graphs, into a new expansion where only tree-graphs appear.
(0)
Of course, W [J] and WΓ [J] are equal. The corresponding perturba-
tive expansions are therefore equivalent; the apparent lack of loops in
(0)
the expansion from the path integral wth the effective action WΓ [J]
should be explained by a re-writing of the usual expansion from W [J]
with modified vertices and propagators. These new exotic vertices and
propagators should be “dressed” to account for all loop effects that
we have encountered in the path integral of the classical action.
The above result is of great importance for renormalization. “Trees”
do not have any ultraviolet divergences. Therefore, we only need to es-
tablish a renormalization procedure which renders finite the “dressed”
propagators and vertices of the quantum effective action.
Let us take the “tree-only” statement seriously, and write down
all possible graphs that we might have for two-, three-, and four-point
functions. This will be sufficient to establish a pattern for the Green’s
functions derived from the effective action. Actually, we can only have
a very small number of tree-graphs for small number of external legs.
We can figure out the propagators and vertices of the tree dia-
grams of Fig 1 by comparing with the usual Feynman diagrams which
we obtain by expanding W [J]. The two-point function in Fig 1 must
be equivalent to the full propagator, computed at all orders in pertur-
bation theory from W [J]:
= ( this is the full propagator )
83
Figure 1: Possible tree-graphs for two-, three-, and four-point functions.
This very small number of connected graphs, which arises from the pertur-
bative expansion of the path integral WΓ [J] with the effective action Γ[hφi],
should contain in the propagators and vertices all loop effects found in the
usual path integral W [J].
= + 1PI + 1PI 1PI
+ 1PI 1PI 1PI + ... (325)
where we sum all possible Feynman diagrams with two external legs.
We can write the sum of all graphs contributing to the full propa-
gator as a geometric series of one-particle-irreducible two-point loop
Feynman diagrams.
1PI = + + +(326)
...
One-particle irreducible diagrams are these which cannot be separated

into two diagrams after we cut one of the propagators. Knowledge of
the 1PI propagator graphs is sufficient to determine the full propaga-
tor. Let us work, as an example, with a scalar field theory. We denote,
84
in Fourier-space,
i
2
i
1PI = −iΣ(p ) .
p2 − m2 p2 − m2
From Eq. 325 we find,
∞
" #n
i X Σ(p2 ) i
= = 2 = .
p − m2 n=0 p2 − m2 p2 − m2− Σ(p2 )
Indeed, the full propagator is then computed using only 1PI graphs.
We proceed to compare the three-point Green’s functions of Fig. 1
with the result of the full perturbative expansion from W [J]. We now
that the propagators connected to the triple-vertex are full propaga-
tors.
The triple vertex must then be only the sum of all 1PI three-point
functions.
= + + +
Three-point graphs which are one-particle reducible, always contribute

a two-point subgraph to the full propagators of the external legs and
an 1PI 3-point subgraph to the vertex.
85
We can now convince ourselves easily that the four point vertex in
Fig. 1 contains all one-particle irreducible four point functions.
The same of course holds for higher multiplicities. Our conclusion is

that we can always rearrange the sum of graphs in the perturbative
exansion, derived via W [J] and containing both loops and trees, to an
equivalent “tree-level graphs only” expansion where the propagator
is the full “two-point” function and the vertices are all one-particle
irreducible graphs with the same number of external legs as in the
vertex.
(0)
The important statement is that W [J] = WΓ [J] and that all
Green’s functions can be obtained automatically from the tree-level
expansion of a generating functional with the effective action replacing
the classical action. Let us verify that the two, three, and four point
functions are derivable from the effective action Γ[J].
We first introduce the short notation
hφiJ (x) ≡ φx , J(x) ≡ Jx .
We start from the equation,

δΓ
= −Jx .
δφx
Differentiating with a source, we obtain:
δ δΓ
δ(x − y) = −
δJy δφx
δφz δ 2 Γ
Z
= − d4 z
δJy δφz δφx
" #" #
δ2Γ δ2W
Z
4
= d z . (327)
iδφx δφz iδJz δJy
From the above we see that

1 δ2Γ
i δφx δφy
86
is the inverse of the full two-point function
1 δ2W
∆(x1 − x2 ) ≡ h0| T φ(x1 )φ(x2 ) |0i = .
i δJx δJy
Before we compute the three-point function we need two tricks.

- Chain rule:
δ δφz δ
Z
= d4 z
δJx δJx δφz
δ 2 W [J] δ δ
Z Z
= d4 z = i d4 z∆(x − z) . (328)
δJx δJz δφz δφz
- Differentiation of an inverse matrix
1 = M M −1
∂ M M −1 ∂M −1 ∂M −1
; 0= = M +M
∂λ ∂λ ∂λ
∂M −1 ∂M
; = −M −1 M −1 . (329)
∂λ ∂λ
We have:
1 δ 3 W [J]
=
i2 δJx1 δJx2 δJx3
" #
δ δ2W
Z
4
= d y1 ∆(x1 − y1 )
δφy1 iδJx2 δJx3
" #−1
δ δ2Γ
Z
4
= d y1 ∆(x1 − y1 )
δφy1 iδφx2 δφx3
" #−1 " #−1
δ2Γ δ3Γ δ2Γ
Z
4 4 4
= d y1 d y2 d y3 ∆(x1 − y1 )
iδφx2 δφy2 iδφy1 φy2 δφy3 iδφy3 δφx3
δ3Γ
Z
= d4 y1 d4 y2 d4 y3 ∆(x1 − y1 )∆(x2 − y2 )∆(x3 − y3 ) (330)
iδφy1 δφy2 δφy3
Now we may compare the graph on the lhs and the rhs of this equa-
tion. We have explicitly found that the full three-point function is the
convolution of propagators, one for each external leg, and the third
derivative of the effective action. From our earlier discussion we now
87
that after we factor out full propagators for the external legs, the
remainder is the sum of one-particle irreducible three-point diagrams.
δ4Γ
iδφy1 δφy2 δφy3 δφy4
is the sum of 1PI 4-point functions.
In summary, we can deduce from the Quantum Effective Action
all physical predictions in a quantum field theory.
• The second derivative of Γ[hφi] is the inverse propagator. The
zeros of the inverse propagator yield the mass values of the phys-
ical particles in the theory.
• Higher derivatives of the effective action are 1PI Green’s func-
tion. Connecting them with full propagators to from trees we can
derive akk connected amplitudes which are required for S-matrix
element computations.
Additionally, solving the equation
δΓ
=0
δ hφi
yields the values of vevs where the effective action is minimal. This will
serve to define the true ground-state of the theory. The location of the
minimum will also reveal whether any symmetries of the Lagrangian
are broken spontaneously.
6.2 The effective potential

We have just observed that by differentiating the effective action
functional with respect to the field vevs, we generate one-particle-
irreducible Feynman diagrams. All functional derivatives of Γ[hφi] are
therefore represnted in terms of Feynman diagrams; if we could com-
pute all these diagrams we could compute the full effective action by
summing up all the terms of a Taylor series expansion.
Specifically, we can expand
∞
X 1 δΓ [hφi]
Γ [hφi] = hφ(x1 )i . . . hφ(xn )i
n=1
n! δ hφ(x1 )i . . . δ hφ(xn )i
∞
X i
= Γ(n) (x1 , . . . , xn ) hφ(x1 )i . . . hφ(xn )i , (331)
n=1
n!
88
where
1 δΓ [hφi]
Γ(n) (x1 , . . . , xn ) ≡ . (332)
i δ hφ(x1 )i . . . δ hφ(xn )i
are one-particle-irreducible Green’s functions (in coordinate space).
We consider the case where the ground state (vacuum) is transla-
tion invariant; it does not distinguish among different points in space-
time. There are situations where this is not true (instantons), however
the space-time blind vacuum case is interesting and common. We then
have:
hφ(x)i = constant ≡ φ. (333)
In this case, the Green’s functions simplify enormously if we use a
Fourier transformation ( momentum space):
Z
d4 x1 . . . d4 xn Γ(n) (x1 , . . . , xn ) hφ(x1 )i . . . hφ(xn )i
Z
= φn d4 x1 . . . d4 xn Γ(n) (x1 , . . . , xn )
Z
n
= φ d4 x1 . . . d4 xn
d4 k1 d4 kn −ik1 x1
Z
. . . e . . . e−ikn xn (2π)4 δ (4) (k1 + k2 + . . . + kn )
(2π)4 (2π)4
×Γ̃(n) (k1 , . . . , kn )
Z
= φn d4 k1 . . . d4 kn δ (4) (k1 ) . . . δ (4) (kn )(2π)4 δ (4) (k1 + k2 + . . . + kn )
×Γ̃(n) (k1 , . . . , kn )
= φn (2π)4 Γ̃(n) (0, 0, . . . , 0)δ (4) (0). (334)
Notice that we have explicitly shown the delta-function which imposes
momentum conversation. The multiple integrations over space-time
xi are simple because of the assumption of x-independent vev φ. The
factor Z Z
(2π)4 δ(0) = d4 xe−i0·x = d4 x .
We then have for the effective action,
∞
φn (n)
Z X
Γ[φ] = d4 x Γ̃ (0, 0, . . . , 0). (335)
n=1
n!
The effective potential is defined from the effective action, factoring
out the space-time volume:
Γ[φ]
Vef f (φ)] ≡ − R 4 (336)
( d x)
89
We obtain:
X φn
Vef f (φ) = − Γ̃(n) (0, 0, . . . , 0). (337)
n=1
n!
Therefore, the recipe to compute the effective potential is:
• Compute all 1PI Green’s fucntions with increasing number of ex-
ternal legs in momentum-space and setting all external momenta
to zero.
• For each external leg include a power of the classical vev of the
corresponding field.
• Sum the series up without forgetting to include the i/n! from
the Taylor series expansion.
Let us consider the Lagrangian of a real scalar field with a quartic
interaction,
1 m2 2 λ 4
L = (∂m uφ)2 − φ − φ . (338)
2 2 4!
A computation of the effective potential including all orders in pertur-
bation theory is impossible. We can compute the effective potential
easily in the tree and one-loop approximation.
We observe that the only two 1PI Green’s functions that we can
write in the tree approximation are the 2-point (amputated propaga-
tor) and 4-point (vertex). From Eq. 337 we find,
tree m2 2 λ 4
Vef f =+ φ + φ . (339)
2 4!
It turns out that the effective potential at tree-level is the same as the
potential of the classical Lagrangian.
The 1-loop computation of the effective potential will be discussed
in the exercise class.
90
END OF WEEK 7
91
7 Symmetries of the path integral and
the effective action
Our guiding principle in constructing realistic theories for particle in-
teractions is invariance of the classical action under certain symme-
tries (e.g. BRST symmetry for Yang-Mills theories). Symmetries of
the classical action S may not be automatically symmetries of the ef-
fective action Γ. However, the effective action Γ satisfies very general
equations (Slavnov-Taylor identities) due to these classical symmetry
constraints.
7.1 Slavnov-Taylor identities

Consider a theory of φi interacting fields with arbitrary (bosonic or
fermionic) spin-statistics. We assume that this theory is symmetric
under an infinitesimal symmetry transformation:
φi → φi 0 = φi + Fi (x, φi ),
where is a small parameter and F i is usually an ordinary function
of the fields φi and their derivatives. Then, we require that both the
action and the path-integral measure of the fields are invariant under
this transformation:
S[φi + Fi (x, φi )] = S[φi ]
D (φi + Fi (x, φi )) = Dφi .
After transforming the fields, the generating path-integral is
Z R
0 d4 xφ0i Ji
Z[Ji ] = Dφ0i eiS[φi ]+i
Z R
d4 x(φi +Fi (x,φi ))Ji
= D (φi + Fi (x, φi )) eiS[φi +Fi (x,φi )]+i
Z R
d4 x(φi +Fi (x,φi ))Ji
= Dφi eiS[φi ]+i .
We can now expand in the small parameter ,

Z R
d4 xφi Ji
Z[Ji ] = Dφi eiS[φi ]+i 1 + iFi (x, φi ) + O(2 )
Z Z R
d4 xφi Ji
= Z[Ji ] + i Dφi d yFi (y, φi )Ji (y) eiS[φi ]+i
4
Z Z R
d4 xφi Ji
; Dφi d yFi (y, φi )Ji (y) eiS[φi ]+i
4
= 0, (340)
92
or, dividing by the path inegral,
R R 
iS[φi ]+i d4 xφi Ji
Dφi Fi (y, φi )e
Z
d4 y   Ji (y) = 0 (341)
Z[Ji ]
In the square brackets we recognize the average of the transformation

over all field configurations,
R
d4 xφi Ji
Dφi Fi (y, φi )eiS[φi ]+i
R
hFi (y, φi )iJ ≡ . (342)
Z[Ji ]
We then find the identity,
Z
d4 y hFi (y, φi )iJ Ji (y) = 0, (343)
concluding that if there exists an infnitesimal symmetry transforma-

tion of the classical action, then there is a constraint on the “average”
value of the transformation. Eq. 343 depends on abritrary sources, and
by differentiating multiple times with the sources, we can obtain an in-
finite number of identities. These are called Slavnov-Taylor identities;
we shall consider an example soon.
7.2 Symmetry constraints on the effective ac-

tion
The generating Slavnov-Taylor identity of Eq. 343 identity tells us
that there exists a symmetry for the effective action. Substituting
δΓ
Ji (y) = − ,
δ hφi (y)iJ
we obtain:
δΓ
Z
d4 y hFi (y, φi )iJ = 0. (344)
δ hφi (y)iJ
Equivalently,
Γ [hφi iJ ] δΓ Γ [hφi iJ ]
Z
+ d4 y hFi (y, φi )iJ =
δ hφi (y)iJ
δΓ
Z
; Γ [hφi iJ ] + d4 y hFi (y, φi )iJ = Γ [hφi iJ ]
δ hφi (y)iJ

; Γ [hφi iJ + hFi (y, φi )iJ ] = Γ [hφi iJ ] + O 2 . (345)
93
Therefore, the effective action is symmetric under the transformation
hφi i → hφi i0 = hφi i + hFi (φi )i . (346)
We recall that the classical action is symmetric under the transforma-

tion
φi → φ0i = φi + Fi (φi ) . (347)
Are these two transformations the same? Otherwise, is it Fi =
hFi i? In general they are not! The symmetries of the classical action
are usually no symmetries of the quantum effective action. Consider
an example of a classical action symmetric under a field transformation
φ(x) → φ(x) + φ2 (x)
The quantum action should be symmetric under the transformation

D E
hφ(x)i → hφ(x)i + φ2 (x) .
Given that D E
hφ(x)i2 6= φ2 (x) ,
the two transformations are different.
Nevertheless, we can identify many symmetries in classical actions
for realistic field theories which are linear:
Z
Fi [φi , x] = ci (x) + d4 yT ij (x, y)φj (y). (348)
The equivalent symmetry transformation for the effective action is

Z
hFi [φi , x]i = ci (x) + d4 yT ij (x, y)φj (y)
Z
= ci (x) + d4 yT ij (x, y) hφj (y)i
= Fi [hφi i , x], (349)
and it is identical to the classical transformation. It is usefull to

remember that linear symmetry transformations of the classical
action, are automatically symmetry transformations of the
effective action.
94
7.3 Contraints on the effective action from BRST
symmetry transformations of the classical ac-
tion
The BRST transformations are not linear; therefore they are only a
symmetry of the classical action 5 and not of the effective action. Nev-
ertheless, the effective action is constrained by the BRST symmetry
of the classical Lagrangian (Eq. 343). For these nillpotent transfor-
mations Eq. 343 takes a very special form, the so called Zinn-Justin
equation.
We start with a classical action S[φi ] of fields φi which is symmetric
under the BRST transformation
δθ φi = θBi . (350)
Since Bi is nilpotent we also have
δθ0 δθ φi = 0
; δθ0 Bi = 0 (351)
We realize that, because of Eq. 351, there is a more general action

which has the same symmetry as the original S[φi ]. It is easy to verify
that the action,
Z
S[φi , Ki ] = S[φi ] + d4 xBi (x)Ki (x), (352)
is indeed symmetric under the same transformation.

The functions Ki are arbitrary (sources). We recall, however, that
the functions Bi have the opposite spin-statistics of the corresponding
field φi . Since the product Bi Ki must have even spin-statistics (the
same as the action S), we deduce that the source Ki and the field φi
have also opposite spin-statistics.
We can write the generating functional W for connected graphs,
Z R R
d4 xBi Ki +i d4 xφi Ji
eiW [Ji ,Ki ] = Dφi eiS[φi ]+i . (353)
The fields φi may be bosonic or fermionic, therefore the order that we

have chosen in writing the integrands in the exponential is important.
Conventionally, we have placed source terms Ji , Ki to the right.
5
From now on, by “classical action” for a gauge theory we mean the action obtained
after gauge-fixing using the Fadeev-Popov method.
95
The vacuum expecation value hφi i is given by
R R
Dφi (y)eiS[φi ]+i
R
hφi (y)i = R R
Dxi eiS[φi ]+i
R
δR W [Ji , Ki ]
= . (354)
δJi (y)
This is an implicit relationship among Ji , Ki , hφi i, and we will consider

Ki , hφi i as independent variables, and the sources Ji as being expressed
in terms of these two variables:
Ji = Ji (hφi i , Ki ).
The exact form of Ji (hφi i , Ki ) can be found if we evaluate explicitly

the effective action,
Z
Γ [hφi i , Ki ] = W [Ji (hφi i , Ki ), Ki ] − d4 x hφi i Ji (hφi i , Ki ), (355)
and take a left derivative,
δL Γ [hφi i , Ki ]
= −Ji (hφi i , Ki )(x). (356)
δ hφi (x)i
We now compute the derivative of the effective action with respect

to the sources Ki .
δR Γ [hφi i , Ki ] δR
Z
= W [Ji (hφi i , Ki ), Ki ] − d4 x hφi i Ji (hφi i , Ki )
δKi (x) δKi (x)

δR W [Ji , Ki ]
=
δKi (x) Ji =Ji (hφi i,Ki )
!
δR Jm (hφi i , Ki )(y)

δR W [Ji , Ki ]
Z
4
+ d y
δJm (y) Jm =Jm (hφi i,Ki ) δKi (x)
Z
− d4 y hφm i (y)
δKi (x)

δR W [Ji , Ki ]
=
Z
+ d4 y hφm i (y)
δKi (x)
Z
− d4 y hφm i (y)
δKi (x)
96

δR W [Ji , Ki ]
=
δR
Z R R
iS[φi ]+i d4 xφi Ji + d4 xBi Ki

= −i ln Dφi e
δKi (x)
Ji =FIXED
= hBi i . (357)
From the general Slavnov-Taylor identity we have,

Z
d4 x hBi i Ji = 0
δL Γ
Z
; d4 x hBi i =0
δ hφi i
δR Γ δL Γ
Z
; d4 x = 0. (358)
δKi δ hφi i
This is a constraint which depends only on the effective action Γ[hφi i , Ki ]

(Zinn-Justin equation). It is a very useful form in order to study the
consequences of symmetry for the effective action, especially in con-
nection with renormalization proofs and studying anomalies.
For later use, we define the product
δR F δL G δR F δL G
Z
(F, G) ≡ d4 x − . (359)
δKi δ hφi i δ hφi i δKi
for functionals F [hφi i , Ki ], F [hφi i , Ki ] of the functions hφi i , Ki . Re-

call that hφi i and Ki have opposite spin-statistics. Then, the Zinn-
Justin equation can be written as
(Γ, Γ) = 0. (360)
7.4 Slavnov-Taylor identities in QED

We now consider an example of Slavnov-Taylor identities in QED. The
classical Lagrangian is, in the Lorentz gauge,
1 1
L = − Fµν F µν + ψ̄ (i6D − m) ψ − (∂µ Aµ )2 . (361)
4 2λ
The corresponding path integral is,
Z R
µ i d4 x[L+Aµ Jµ +ψ̄ρ+ρ̄ψ ]
Z [J , ρ̄, ρ] = DAµ Dψ̄Dψe . (362)
97
An infiniterimal local gauge transformation is:
1
Aµ → Aµ − ∂µ Θ(x),
q
ψ → (1 − iqΘ(x))ψ,
ψ̄ → (1 + iqΘ(x))ψ̄.
The path integral measure is invariant under the gauge transforma-

tion. In the integrand of the path-integral exponent, we can identify
a part which is invariant under these transformations, while the re-
maining, which includes tha gauge-fixing term and the source term, is
not invariant.
1
Lnon−invariant = − (∂µ Aµ )2 + Aµ Jµ + ψ̄ρ + ρ̄ψ. (363)
2λ
By performing a gauge-transformation on the path-integral we can
derive, as before, the Slavnov-Taylor identity,
Z
d4 x hδLnon−invariant i = 0. (364)
We can work out what is the change in the non-invariant part of the
Lagrangian. The gauge-fixing transforms as:
2
1 1 1

− (∂µ Aµ )2 → − ∂ µ Aµ − ∂ µ Θ
2λ 2λ q
1 1
= − (∂µ Aµ )2 + ∂µ Aµ ∂ 2 Θ(x) + O Θ2
2λ qλ
1 1

; δ − (∂µ Aµ )2 = ∂µ Aµ ∂ 2 Θ(x). (365)
2λ qλ
Adding the variation of the source terms, we obtain:
1 1
δLnon−invariant = ∂µ Aµ ∂ 2 Θ(x)−iqΘ(x)ρ̄ψ+iqΘ(x)ψ̄ρ− J µ ∂µ Θ(x).
qλ q
(366)
The Slavov-Taylor identity is:
1 1
Z
4
d x ∂µ Aµ ∂ 2 Θ(x) − iqΘ(x)ρ̄ψ + iqΘ(x)ψ̄ρ − J µ ∂µ Θ(x) = 0
qλ q
1 1 µ
Z
; 4 µ 2

d x ∂µ hA i ∂ Θ(x) − iqΘ(x)ρ̄ hψi + iqΘ(x) ψ̄ ρ − J ∂µ Θ(x) = 0
qλ q
1
Z
; d4 xΘ(x) ∂µ ∂ 2 hAµ i − iq 2 ρ̄ hψi + iq 2 ψ̄ ρ + ∂µ J µ = 0,

(367)
λ
98
where we have used integration by parts. The above should be valid
for arbitrary Θ(x), therefore the kernel of the integration in the square
brackets should be identically zero.
1
∂µ ∂ 2 hAµ i − iq 2 ρ̄ hψi + iq 2 ψ̄ ρ + ∂µ J µ = 0.

(368)
λ
Substituting vacuum expectation values with functional derivatives of
the path-integral for connected graphs, W = −ilnZ, we write:
1 δW δW δW
∂µ ∂ 2 − iq 2 ρ̄ − iq 2 ρ + ∂µ J µ = 0, (369)
λ δJµ δ ρ̄ δρ
where the functional derivatives are left derivatives for the fermionic
sources.
Eq. 369, provides constraints for Green’s functions in QED at all
orders in perturbation theory. We find the simplest example, by differ-
entiating this equation with a photon source and then set all sources
to zero,
!
δ 1 δW δW δW
∂µ ∂ 2 − iq 2 ρ̄ − iq 2 ρ + ∂µ J µ = 0,

δJ ν (y)

λ δJµ δ ρ̄ δρ
J,ρ,ρ̄=0

1 δ2W
; ∂µ ∂ 2 + ∂µ δ(x − y) = 0.

λ δJµ (x)δJν (y) J,ρ,ρ̄=0
1
; ∂µ ∂ 2 h0| T Aµ (x)Aν (y) |0i = −∂µ δ(x − y). (370)
λ
We now write the Fourier representations,
d4 k −ik(x−y)
Z
δ(x − y) = e ,
(2π)4
and
d4 k −ik(x−y) µν
Z
µ ν
h0| T A (x)A (y) |0i = e D (k).
(2π)4
Substituting into Eq. 370 we find that the photon propagator in mo-
mentum space should satisfy,
kν
kµ Dµν (k) = λ (371)
k2
We can write, in complete generality,
kµ kν
Dµν = A(k 2 )gµν + B(k 2 ) . (372)
k2
99
Substituting in Eq. 371, we find
λ
A(k 2 ) + B(k 2 ) = . (373)
k2
Thus, the photon propagator in momentum space has the form,
kµ kν λ kµ kν

Dµν (k) = A(k 2 ) g µν − + (374)
k2 k2 k2
This is a result valid at all orders in perturbation theory.
Notice that the term which depends on the gauge-fixing parameter
is fully known. We can compare this with the result at leading order
in perturbation theory,
−1 kµ kν λ kµ kν

µν
D (k) = 2 g µν − 2 + + O(g 2 ). (375)
k k k2 k2
We can see that the gauge-fixing contribution is accounted fully in
the leading order result, and therefore it is not modified at higher
orders in perturbation theory. Higher order corrections modify only
the function A(k 2 ). For this reason, the gauge-fixing parameter λ does
not receive any renormalization.
A second important observation to make is that the part of the
propagator which does not depend on λ,
kµ kν

DTµν 2
= A(k ) g µν
− 2 ,
k
is transverse to the photon-momentum. Indeed, we easily find that
DTµν (k)kµ = 0.
100
END OF WEEK 8
101
8 Spontaneous symmetry breaking
Symmetry transformations that leave the effective action invariant
may not be symmetries of the physical states and the vacuum state.
These symmetries are “spontaneously broken”. Spontateous symme-
try breaking is associated with a degeneracy of the ground state (vac-
uum). Let’s assume that an effective action is symmetric under
hφi → − hφi ; Γ[hφi] = Γ[− hφi]
a symmetry which is inherited unchanged from the classical action (as

we have seen earlier). Assume also that the physical vacuum expec-
tation value of the field,
δΓ
hφia : = 0, (376)
δ hφia
with hφia 6= 0. In other words, there exists a state |va i with
hva | φ̂(x) |va i

hφia ≡ 6= 0, (377)
hva | va i
for which Γ[hφia ] ≡ Γa is a minimum. Then, there should be a second

value of the field vev in a different state which also gives the same
value for the effective action:
∃ |bi : hφib = − hφia with Γ[hφia ] = Γ[hφib ] = minimum .
So, while the transformation φ → −φ preserves the action and the

effective action, it does not preserve the states and transforms one
state into another:
|va i → |vb i .
The symmetry is broken as long as the system is in one of the degen-
erate states.
8.1 Goldstone theorem

Consider a symmetry transformation
φn (x) → φ0n (x) = φn (x) + i

X
tnm φm (x) (378)
m
102
with a small parameter and tnm generators of the transformation.
The transformation leaves the effective action intact:
Γ [hφn (x)i] = Γ φ0n (x) .

(379)
The symmetry of the effective action gives rise to Slavnov-Taylor iden-

tities:
δΓ
Z
d4 x tnm hφm (x)i = 0. (380)
δ hφn (x)i
Taking a second derivative, we have:
δ2Γ δΓ
Z
0 = d4 x tnm hφm (x)i + tnl . (381)
δ hφl (y)i δ hφn (x)i δ hφn (y)i
For physical systems with zero external sources, Jn (x) = 0, we have

that:
δΓ
=0 (382)
δ hφl (y)i
and we arrive to the equation
δ2Γ
Z
0 = d4 x tnm hφm (x)i . (383)
δ hφl (y)i δ hφn (x)i
We now make an important assumption that the vacuum state |Ωi is

translation invariant. We then find that the vacuum expectation value
of the field is the same in all space-time. Since we can find the value
of the field operator at a space-time point x from the value of the field
at the origin with a translation using the momentum operator as a
generator, we can write:
hφ(x)i = hΩ| φ̂(x) |Ωi

= hΩ| eiP̂ x φ̂(0)e−iP̂ x |Ωi
= hΩ| φ̂(0) |Ωi = constant ≡ hφi . (384)
Then, the effective action can be written as:

Z Z
Γ [hφn i] = − d4 xVef f (hφn i) = − d4 x Vef f (hφn i) (385)
Eq. 383 then yields for the effective potential Vef f the following con-
straint:
X ∂ 2 Vef f
tnm hφm i = 0. (386)
nm ∂ hφl i ∂ hφn i
103
This is a constraint on the mass spectrum of the theory. To see that,
we recall that the second derivative of the effective action is the inverse
of a two-point Green’s function:
δW [J]
= hφm (x)i
δJm (x)
δ 2 W [J]
; = δ(x − y)δnm
δ hφn (y)i δJm (x)
...
δ2Γ
Z
; d4 z hΩ| T φn (y)φk (z) |Ωi = δ(x − y)δnm
δ hφk (z)i δ hφm (x)i
δ2Γ
Z
; d4 zd4 y hΩ| T φn (y)φk (z) |Ωi = δnm
δ hφk (z)i δ hφm (x)i
∂ 2 Vef f
Z
; d4 zd4 y hΩ| T φn (y)φk (z) |Ωi δ(z − x) = −δnm
∂ hφk i ∂ hφm i
∂ 2 Vef f
Z
; d4 y hΩ| T φn (y)φk (x) |Ωi = −δnm (387)
Substituting the Fourier transformation of the 2-point function:
d4 p
Z
hΩ| T φn (y)φk (x) |Ωi = Dnk (p2 )e−ip·(x−y) (388)
(2π)4
the integration over the y variable yields a delta function setting the
momentum pµ = 0. We therefore have:
∂ 2 Vef f
Dnk (0) = −δnm . (389)
Or, equivalently,
∂ 2 Vef f −1
= −Dnm (0) (390)
∂ hφn i ∂ hφm i
Eq. 386 yields that
−1
X
Dln (0)tnm hφm i = 0. (391)
nm
When is this equation satisfied? Let’s write the combination m tnm hφm i =
P
δ hφn i as the variation of the vev under the symmetry transformation.

Then Eq. 391 becomes:
−1
X
Dln (0)δ hφn i = 0. (392)
n
104
If the transformation leaves the vacuum state and, thus, the vacuum
expectation value of the fields invariant, δ hφn i = 0, then Eq. 392 is
fulfilled. What if the symmetry is broken and the symmetry transfor-
mation of the effective action changes the vacuum, so that there are
some δ hφi i =
6 0? Let us rewrite Eq. 392 in a matrix notation:
    
−1
Dln (0)   hφn i  = 0  hφn i  (393)
    

−1
We observe that the matrix Dnl (0) has zero eigenvalues, as many
as the independent vectors δ hφn i which are non-vanishing. In the
simplest case of only one field, the inverse propagator of the field at
zero momentum is proportional to the mass of the particle excitation
of the field:
iZ
D(p) = + continuum ; D−1 (0) ∝ m2 .
p2 − m2
−1
In general, Dnl (0) is the mass-matrix of the theory. Redefining appro-
priately the fields, φn = Rnm φ̃m eliminates non-diagonal terms and
the diagonal terms, the eigenvalues of the matrix, are the masses of
the physical particle excitations of the fields φ̃i .
We have just proven Goldstone’s theorem. Namely, for each inde-
pendent δ hφn i = m Tnm hφm i = 6 0 there exists a massless particle
P
in the spectrum of the theory. The symmetry generators Tnm which

change the vev of the fields are called “broken” generators. There is
an alternative proof 6 of Goldstone’s theorem due to Weinberg. This
proof also demonstrates that
• The massless states are one-particle states.
• They are also invariant under rotations and correspond to spin-0
particles, the so called Goldstone bosons.
• The Goldstone bosons have the same “quantum numbers” as the
conserved currents corresponding to the broken generators.
Goldstone’s theorem seems very powerful and its proof appears
to leave no room for exceptions. Nevertheless, we will be able to
find a loophole soon: it is possible to have spontaneous symmetry
breaking without giving rise to massless particles. We note that our
6
to be taught in the course of The physics of Electroweak Symmetry Breaking
105
proof requires translation invariance of the vacuum states as well as
positive norms. These requirements cannot be satisfied simultaneously
for quantum theories with local gauge invariance.
8.2 General broken global symmetries

Let’s assume a pattern of spontaneous symmetry breaking:
G → H,
where G is the symmetry group that leaves invariant the effective

action and H a subgroup of G which leaves invariant the vacuum. We
will also assume that the symmetry group is global. In other words,
the effective action Γ remains invariant Γ[ψn ] = Γ[ψn0 ] for
∂gnm
ψn0 =
X
gnm ψm , = 0, gnm ∈ G, (394)
m ∂xµ
and the vacuum remains invariant

X
hnm hψm i = hψn i , ∀ hnm ∈ H. (395)
m
According to Goldstone’s theorem, the mass matrix of the theory has

zero eigenvalues for the eigenvectors:
X
a
Tnm hψm i ≡ δ hψn i , (396)
m
a is a broken generator.
where Tnm
Which independent linear combinations of the fiels in the La-
grangian of the theory correspond to Goldstone fields and which are
not? We shall prove that all fields ψn (including non-Goldstones) can
be obtained from Goldstone-free fields ψ̃n by performing a local group
transformation: X
ψn (x) = γnm (x)ψ̃m (x). (397)
m
We start by observing that Goldstone-free field combinations ψ̃n (the

“heavy” fields of the theory) must be orthogonal to the vectors of
Eq. 396, that is: X
a
ψ̃n (x)Tnm hψm i = 0. (398)
nm
106
Without loss of generality, we will assume that the elemebts g ∈ G
belong to a real and orthogonal representation of the group which is
compact. Then, the quantity:
Vψ (g) = ψn gnm hψm i (399)
is a bounded, continuous, real-valued function.

Exercise: . . . Let us now find an appropriate g = γ for which Vψ(x) (g)
is a maximum at every space-time point x. Then, under a small
variation of the group parameter
X
δγnm = i a γnl Tlm
a
, (400)
a
Vψ (g) is stationary:
X X
0 = δVψ (g) = i a a
ψn (x)γnl (x)Tlm hψm i (401)
a nlm
Recalling that we have chosen an orthogonal and real representation

of the group, we have:
[γnl ] = [γln ]−1 . (402)
Thus, " #
−1
X X X
0=i a γln a
ψn (Tlm hψm i) (403)
a lm n
Therefore, the field combinations:

−1
X
ψ̃l = γln ψn (404)
n
are orthogonal to the vectors

X
a
Tlm hψm i = δ hψl i (405)
m
and they are not Goldstone bosons.

Let’s rewrite the Lagrangian of the theory by making the substi-
tution which we have just found:
ψ(x) = γ(x)ψ̃(x), (406)
rewriting the fields of the theory as explicit non-Goldstones ψ̃() and

the remaining Goldstone fields contained in γ(x). We remind that
the Lagrangian is only invariant under a global gauge transformation,
107
while the above transformation is a local gauge transformation which
does not leave the Lagrangian invariant. We have:
h i h i
L γ(x)ψ̃(x) = L γ(x0 )ψ̃(x)
+derivatives of γ(x), ψ̃(x) (407)
h i
Due to the global gauge invariance of the theory, L γ(x0 )ψ̃(x) =
h i
L ψ̃(x) , we have that:
h i h i
L γ(x)ψ̃(x) = L ψ̃(x)
+derivatives of γ(x), ψ̃(x) (408)
where the first term does not have any Goldstone bosons. Goldstone
bosons appear only as derivatives. This forbids mass terms:
m2B B(x)B(x)
for them. Also, ar low energies, Goldstone interactions vanish. In-

deed, the Feynman rules for fields that appear as derivatives will be
proportional to the momenta of the particles:
∂µ γ(x) → ∂µ B(x) → pµ (in Ferynman rules)
and vanish for zero momenta pµ to0.
8.3 Spontaneous symmetry breaking of local

gauge symmetries
Let us now assume that our Lagrangian is invariant under a local
gauge symmetry. Repeating the reasoning of the previous section and
rewriting
ψ = γ ψ̃,
we have that h i h i
L γ(x)ψ̃(x) = L ψ̃(x) . (409)
In other words, our carefully selected gauge transformation eliminates
all Goldstone boson fields from the Lagrangian. We have just found
an exception of Goldstone’s theorem in theories with local gauge in-
variance, where the symmetry is spontaneously broken but there are
108
no physical massless Goldstone fields due to the breaking of the sym-
metry. The rewriting ψ = γ ψ̃ is equivalent to choosing a gauge fixing
condition:
ψ̃ · (T a hψi) = 0. (410)
Lagrangians which are locally invariant under a continuous sym-
metry transformation require gauge bosons in order to form covariant
derivatives. Let us look at the quadratic terms in the covariant deriva-
tives:
109
9 Renormalization: counting the de-
gree of ultraviolet divergences
Consider a Lagrangian with fisdimensionalityelds f and generic inter-
action operators Oi .
L = kinetic terms + g1 O1 + g2 O2 + . . . + gN ON . (411)
Each operator Oi is a product of fields and/or their derivatives. In

QED for example, we have one such interaction term:
ψ̄6Aψ;
In QCD more operators emerge, e.g.
f abc ∂µ Aaν Aµ,b Aν,c , f abe f cde Aaµ Abν Aµ,c Aν,d , . . .
We would like to keep this discussion as general as possible; At the

end, we will be able to make statements on whether we can remove
via renormalization ultraviolet infinities from arbitrary Lagrangians.
Most of our arguments will be derived using simple dimensional anal-
ysis.
We consider a generic one-particle irreducible Feynman diagram
in perturbation theory. We will first find a simple formula to test
whether it has the most obvious of all possible divergences, the so
called superficial ultra-violet divergence. If the diagram has L loops,
a superficial divergence corresponds to an infinity of the diagram in
the limit
|k1 | = |k2 | = . . . = |kL | = κ → ∞,
where ki are the loop-momenta. A Feynman diagram might have di-
vergences in other limits, where only some momenta or linear combi-
nations of them are taken to infinity while the remaining independent
momenta remain fixed. A Feynman diagram in the superficial ultra-
violet limit behaves as Z ∞
dκκD−1 , (412)
where D is an integer, called the superficial degree of divergence.

• If D > 0 the Feynman diagram has a powerlike divergence,
• if D = 0 it diverges logarithmically,
• if D < 0 it is convergent (only in the superficial limit, since it
might have other divergences).
110
We can compute D (or an upper bound of it) for any Feynman dia-
gram on general grounds. We assume that our 1PI Feynman diagram
has
• If internal propagators for each of the fields f ,
• Ef external legs for each of the fields f and
• Ni vertices corresponding to the term gi Oi in the Lagrangian.
Recall, as examples, the Feynman rules for propagators in gauge the-
ories, and how they behave at the limit of infinite momentum.
• a photon propagator,
kµ kν
−gµν + k2
∼ ∼ κ−2 ;
k2
• a fermion propagator,
6 +m
k
∼ ∼ κ−1 ;
k 2 − m2
• for a scalar,
1
∼ ∼ κ−2 .
k2 − m2
For each of the internal propagators of the field f in the Feynman
diagram there is a contribution to the superficial divergence,
∆f ∼ k −2+2sf ,
where, sf = 0 for a boson and sf = 12 for a fermion. The total

contribution to the asymptotic limit from propagators is then
P
2If (sf −1)
κ f . (413)
The contribution from vertices is easy to find if we know how many

loop momenta appear in the corresponding Feynman rules. For a
vertex due to an operator Oi this number is equal to the number of
space-time derivatives di which can be found in the expression for
Oi . Recall that a Feynman rule for a vertex is esentially the Fourier
transform of the expression of its operator and therefore momenta
arise only from derivatives. The total contribution from vertices to
the superficial ultraviolet limit is
P
Ni di
κ i . (414)
111
Finally, due to the integration measure d4 ki for each loop, the total
contribution from the loop-momenta to the superficial UV limit is
κ4L , (415)
where L is the number of loops in the graph. L is known if we are

given the number of internal propagators If and the number of vertices
Ni in the graph. The number of loop-momenta carried from internal
P P
propagators is f If . The vertices provide i Ni constraints of which
one is not for loop-momenta but for the external momenta. Therefore,
the number of loop-momenta is
X X
L= If − Ni + 1.
f i
Putting together the contributions from the loop integration mea-

sures, vertices, and internal propagators, we find that the asymptotic
behavior at infinity has a superficial degree of divergence
X X
D= 2If (1 + sf ) − Ni (4 − di ) + 4. (416)
f i
We can express the number of internal propagators in terms of the

number of external legs. Let us assume that we have Nif particles
f in the vertex corresponding to the operator Oi in the Lagrangian.
The total number of (internal) legs of the particle f connected in all
the vertices of the graph are
X
Ni Nif .
i
From these Ef are external and the remaining are internal. Every
propagator of f has two edges, so the number of internal legs is
2If .
We then have the identity

X
2If + Ef = Ni Nif . (417)
i
We can therefore write the degree of divergece as

 
X X X
D =4− Ef (1 + sf ) − Ni 4 − di − Nif (1 + sf ). (418)
f i f
112
Notice that the square bracket in the last expression depends only
on the functional form of the operator Oi . If this operator is multiplied
with a coupling constant gi in the Lagrangian, i.e. L = gi Oi + . . . we
can prove that this square bracket is exactly the mass dimensionality
of the coupling gi :
X
[gi ] = 4 − di − Nif (1 + sf ). (419)
f
Indeed. The mass dimensionality of each term in the action should be

zero. We then have that
h i
d4 x + [gi ] + [Oi ] = 0,
where the operator Oi has di derivatives and Nif fields f . Thus,

X
−4 + [gi ] + di − Nif [f ] = 0,
f
and [f ] = 1 + sf is the mass dimensionality of the field. Indeed,

Z
h0| T f (x1 )f (x2 ) |0i ∼ d4 kk −2+2sf e−ikx
k→∞
; 2[f ] = 4 − 2 + 2sf ; [f ] = 1 + sf . (420)
In conclusion, we can write a very suggestive expression for the super-

ficial degree of divergence:
X X
D =4− Ef (1 + sf ) − Ni [gi ]. (421)
f i
If the Lagrangian does not contain any couplings with negative mass
dimensions, [gi ] ≥ 0, we find a superficial ultraviolet divergence, D ≥
0, only in Feynman diagrams with a small number of external legs.
D≥0;
X
Ef (1 + sf ) ≤ 4.
f
In particular, superficial divergences do not appear in (one-particle-

irreducible) Feynman diagrams with five external legs or more.
Examples of theories where superficial divergences may appear in

only a limited number of Green’s fucntions are QED and QCD. All
interaction operators have dimension four and their coefficients are
113
dimensionless. Superficial ultraviolet divergences are limited in 1PI
Green’s functions, such as h0| T ψ̄(x1 )ψ(x2 ) |0i , h0| T Aaµ (x1 )Abν (x2 ) |0i , h0| T ψ̄(x1 )Abν (x2 )ψ(x3 ) |0i , . . .
On the contrary h0| T ψ̄(x1 )ψ(x2 )ψ̄(x3 )ψ(x4 ) |0i1P I is (superficially) fi-
nite.
Theories with [gi ] ≥ 0 are called renormalizable. As we shall see,
these superficial divergences in a finite number of Green’s functions
can be removed by adding a finite number of extra terms in the original
Lagrangian (counterterms).
If the Lagrangian contains a coupling with negative mass dimen-
sion [gj ] < 0, then from Eq. 421 we see that all Green’s functions,
at some loop-order, will develop a superficial divergence. It is there-
fore impossible to cancel the infinities by adding a finite number of
counterterms. Such theories are called non-renormalizable.
9.1 Subdivergences
We should stress that the counting of the superficial degree of diver-
gence is not sufficient to prove that a Feynman is finite. Consider the
two example graphs of Fig. 2. Both two-loop graphs have the same
Figure 2: Both two-loop graphs have the same superficial degree of divergence
D = −2. However the Feynman diagram on the right has a self-energy one-
particle-irreducible subgraph which has a superficial degree of divergence
D = 1. A necessary condition for a graph to be UV finite is that the graph
and all its subgraphs have D < 0
superficial degree of divergence D = −2. One could naively conclude

that both Feynman two-loop graphs are likely to be finite. We know,
though, that this is not the case. Let us compute the superficial de-
gree of divergence for all one-loop subgraphs that we can spot in the
two diagrams. For the left diagram, we find that all subgraphs have
a negative superficial degree of divergence. It also turns out with an
114
explicit calculation (beyond the scope of this lecture) that the dia-
gram is indeed UV finite. However the two-loop Feynman diagram on
the right has an one-loop self-energy subgraph; this has a superficial
degree of divergence D = 1. The self-energy is 1/(d − 4) divergent,
where d is the space-time dimensionality. Such a divergence remains
even after we embed the one-loop self-energy as a subgraph inside a
two-loop graph (there is no mechanism to cancel it). Against our naive
counting for the global superficial degree of divergence, the two-loop
diagram on the right is divergent. The lesson from the above examples
is that for a diagram to be UV finite it is necessary that the supere-
ficial degrees of divergence for the full graph and all of its sub-graphs
must be negative.
9.2 Cancelation of superficial divergences with

counterterms
We derived a criterion to decide whether a Green’s function will de-
velop the most “obvious” type of divergence (superficial) in the limit
where the magnitudes of all loop momenta tend simultaneously to
infinity. We also found that for renormalizable theories this type of
divergence appears in only a finite number of Green’s functions with
a small number of external legs.
Infinities are not acceptable for physical theories. A way out of this
problem is to recognize that the Lagrangian that we started with has a
certain degree of arbitrariness. The guiding principle for constructing
a Lagrangian is to respect a set of symmetries (e.g. BRST symmetry).
However, this is not a tight enough constraint to fix, for example, the
actual values of independent mass and coupling parameters. It may
be possible to redefine the parameters and fields of the Lagrangian
or even add more operators to it without destroying the symmetries
of the Lagrangian. How can we fix the fields and parameters of the
Lagrangian, choosing among their various possible redefinitions? In
renormalizable theories, we fix (partially) this arbitrariness so that
all Green’s functions calculated with the redefined (“renormalized”)
fields, couplings and masses are finite.
We have seen that for “renormalizable theories” the “disease” of
infinities is only spread to a few Green’s functions. Redefinining the
fields and parameters of the Lagrangian (ψ = ZψR = ψR + δZψR , . . .)
gives rise to a few new terms (counterterms) with coefficients engi-
neered to cancel exactly the UV infinities which emerge order by order
115
in perturbation theory. But, is it possible mathematically that we can
cancel the infinities from loop diagrams with counterterms? For this
method to work, it is essential that diagrams with counterterms at a
loop order have the same kinematic dependence as the UV infinities
of loop diagrams without counterterms at higher orders. At the first
two orders in perturbation theory, this statement means that tree-
diagrams with counterterms must have the same kinematic depenence
as the infinities of one-loop diagrams without counterterms.
If, for example, the 1/ terms of a Green’s function at the one-loop
order (where d = 4 − 2 in dimensional regularisation) are logarithms
of external momenta,
ln(p2 )

such a contribution cannot be cancelled by the tree-level contribution
of a countertem. Recalling the Feynman rules for vertices which enter
tree-level calculations in all theories that we have examined so far, we
find no such logarithms in their expressions. Feynman rules always
yield simple polynomial expressions for the vertices of tree-diagrams.
For such tree-level expressions mde out of counterterms to cancel the
infinities of one-loop diagrams the latter have also to be constants or
simple polynomials of momenta. A necessary condition for the coun-
terterm program to be succesfull is that one-loop infinities are “local”,
i.e. they appear as simple polynomials in the external momenta as the
usual Feynman rules do.
Let us look at the fuctional form of the superficial infinities at-one
loop order in perturbation theory, and convince ourselves that indeed
this is exactly what happens in practice. Take, as an example, the one-
λ 4
loop correction to the four-point function in the − 4! φ theory. The
superficial degree of divergence is D = 4 − 4 × 1 = 0 and the graph
is indeed divergent. In dimensional regularization, the corresponding
Feynman parameter integral yields,
1
Z
h0| T φ(x1 )φ(x2 )φ(x3 )φ(x4 ) |0i|1−loop ∼ λ2 dd k
(k 2 − m2 ) [(k + p)2 − m2 ]
Z 1
δ(1 − x1 − x2 )
∼ λ2 Γ() dx1 dx2
0 (m2 − x1 x2 s)
λ2
Z 1
∼ − λ2 γ − λ2 dx ln m2 − x(1 − x)s (422)
0
with s = (p1 + p2 )2 = (p3 + p4 )2 .
116
Loop integrals, in general, contain logarithms or integrals of log-
arithms (polylogaritms) with arguments kinematic invariants formed
from external momenta. Our example result is not an exception and
we indeed find logarithmic contributions in the finite part. We cannot
escape to observe however, that the divergent part is very simple; it
is just a constant. We can then modify the interaction terms in the
Lagrangian,
−λ 4 −λ 4 ]λ2 4
φ → φ + φ
4! 4!
and adjust the coefficient ] so that it cancels exactly the divergent
part of this one-loop integral.
The divergent parts of one-loop integrals are simple polynomials
of the external momenta, as in the above example. If we use Feynman
parameters, any one-loop integral may be written as,
Z 1
d dx1 . . . dxn δ(1 − x1 − . . . xn )

I1−loop ∼Γ N− N − d
2 0 m21 x1 + . . . + m2n xn −
P
si...j xi xj − iδ 2
(423)
where N is an integer (equal to the number of propagators) and
d = 4−2 the dimension. The denominator contains a sum over masses
and kinematic invariants of the external momenta. Divergences may
arise from two terms; the Gamma function Γ(N − 2 + ) and the de-
nominator of the integrand. The argument of the Gamma function
N − d/2 = D/2 is proportional to the superficial degree of divergence
D of the integral. The denominator of the Feynman integral does
not have any ultraviolet divergences. It could become divergent when
masses or invariants become zero, but it is finite when all the propaga-
tors are massive. If this is not the case, and there are massless particles
propagating in a loop, singularities from the denominator are of in-
frared nature connected to the small or zero values of |k| rather than
the UV |k| → ∞ limit. Infrared singularities can also be regulated
by attributing a small mass to massless particles and/or considering
them to be slighlty off-shell. We shall not worry here about infrared
one-loop singularities and focus on the ultraviolet divergences which
can be found, at one-loop order, in the Gamma function pre-factor of
the Feynman parameters integral:
Γ(N − d/2) = Γ(−1 + ), Γ()
for N = 1, 2. Using the identity,
Γ(1 + x)
Γ(x) = ,
x
117
and
Γ(1 + ) = 1 − γ + O(2 ),
and the fact that for N = 1, 2 (responsible for the UV divergences)
the denominator of Eq. 423 turns into a numerator N − d/2 < 0 in
four dimensions, we can see that the coefficient of the 1/ pole can
only be a polynomial in the external momenta. A loop diagram with
superficial degree of divergence D
Z
|k|D−1 , (424)
|k|→∞
has a mass dimensionality D. Therefore, the polynomial can only be

of rank D in the external momenta. Each term in this polynomial
multiplying 1/ must be cancelled by a separate counterterm opera-
tor with a different number of derivatives. Naturally, the number of
derivatives must be:
di ≤ D. (425)
Exercise:Prove that for the cancelation of UV divergences we need at

most as many counterterms in the Lagrangian as the divergent Green’s
functions.
9.3 Nested and overlapping divergences

It can be proven that we only to worry about removing superficial
divergences from loop integrals. Nested and overlapping singularities
are “automatically” removed with this procedure as well. We refer to
original literature for this topic:
• Hepp:1966eg K. Hepp, “Proof of the Bogolyubov-Parasiuk the-
orem on renormalization,” Commun. Math. Phys. 2 (1966) 301.
118
END OF WEEK 11
119
10 Proof of renormalizability for non-
abelian gauge theories
Consider a theory with action S[φ] which is invariant under BRST
trnsformations of the fields φi : δθ φi = θBi . We can add to the clas-
sical action source terms which preserve the invariance under BRST
transformations due to their nilpotency.
Z
S[φi , Ki ] = S[φi ] + d4 xBi Ki . (426)
We now split the action into two terms,
S[φi , Ki ] = SR [φi , Ki ] + S∞ [φi , Ki ]. (427)
The first term is the action with the fields, masses and coupling con-
stants set to their renormalized values. The second term contains the
counterterms. S and SR have the same functional form. Therefore,
they possess the same set of symmetries. It also follows that S∞ must
also possess the same set of symmetries.
The effective action can be cast as an expansion in loops:
∞
X
Γ[φi , Ki ] = ΓL [φi , Ki ]. (428)
L=0
We recall that all terms in the expansion are seperately symmetric

and that we can perform independent shifts to the measure of the
path integral for each one of them.
The Slavnov Taylor identities for the BRST symmetry transfor-
mations result to the Zinn-Justin equation (Eq. 358) which is written
in a short notation as
(Γ, Γ) = 0. (429)
Inserting the loop expansion of the effective action, we obtain:
0 = (Γ0 , Γ0 )
+(Γ0 , Γ1 ) + (Γ1 , Γ0 )
+(Γ0 , Γ2 ) + (Γ1 , Γ1 ) + (Γ2 , Γ0 )
+... (430)
Every line in the above expression must be separetely zero, since it

corresponds to a different order in the loop expansion (equivalently,
120
the h̄ expansion). For the N −th term of the expansion we have
N
X
(ΓL , ΓN −L ) = 0. (431)
L=0
At each loop order we find UV infinities. We decompose the L-loop

effective action into a finite and a divergent part:
ΓL = ΓL,f in + ΓL,∞ . (432)
At zeroth order we only find tree-graphs and there are no infinities. In

addition, the tree-level effective action is equal to the classical action.
We therefore have
Γ0,f in = SR , Γ0,∞ = 0. (433)
We will prove using induction that we can removing all infinities from
the effective action, rendering all ΓL,∞ = 0, with the counterterms in
SR . Let’s assume that we have achieved this for all loops up to N − 1,
ΓL,∞ = 0, L = 1 . . . N − 1. (434)
Then, taking the infinite part of Eq. 431 we obtain that
(Γ0,f in , ΓN,∞ ) + (ΓN,∞ , Γ0,f in ) = 0. (435)
Or, equivalently,
(SR , ΓN,∞ ) = 0. (436)
As we have discussed in a previous section, we expect the infinties of
momentum space Green’s functons in ΓN,∞ to have a simple polyno-
mial dependence in the momenta, given that all divergences at the
previous loop orders are cancelled. We now make two observations:
• As we have shown earlier, the infinities of ΓN,∞ arise in Green’s
functions with a small number of external legs. As we have as-
sumed that the infinities of all loop previous orders have been
cancelled, at the N −th loop order we cannot have any subdiver-
gences. Thus, the N −th loop order divergences correspond to
the superficial limit where all loop momenta are taken to infinity.
For the superficial divergences we have derived that they should
originate from local field operators (products of fields and their
derivatives as well as sources Ki ) in ΓN,∞ whose dimensionality
is less than or equal to four.
121
• ΓN,∞ has all the linear symmetries of SR . These are:
– Lorentz transformations
– Global gauge transformations
– Anti-ghost translations
– Ghost phase-transformations (; ghost number conserva-
tion)
The last two symmetries are apparent by inspecting the ghost-
terms of the Lagramgian:
LFDEEV−POPOV = (∂ µ η̄ a ) Dµab η b . (437)
The anti-ghost field enter the Lagrangian only with its derivative,
and thus the Lagrangian is invariant if we shift globally the field
by a constant. In addition, the Lagrangian is invariant under a
phase-transformation of the ghost and anti-ghost fields:
η a (x) → ei(+1)ρ η a (x), η a (x) → ei(−1)ρ η a ,
Aaµ → ei(+0)ρ Aaµ , ψ → ei(+0)ρ ψ.

The conserved charge of this symmetry is called the ghost num-
ber. The ghost numbers of the φi = {Aaµ , ψ, η a , η̄ a } fields are
γi = {0, 0, +1, −1} respectively. The above phase-transformations
leave the action S[φi ] invariant. For the extended action S[φi , Ki ]
to be invariant, we need to assign ghost numbers to the sources
Ki as well. From the BRST transformations we see that if a field
φi has a ghost number γi , the variation under the transformation
Bi of the field has a ghost number γi + 1. The term d4 xBi Ki
R
ought to remain invariant under the pghost-phase transforma-

tion. We must therefore assign ghost numbers −γi − 1 for the
sources Ki . Specifically, the ghost numbers for KA , Kψ , Kη̄ , Kη
are −1, −1, 0, −2 respectively.
Lemma: ΓN,∞ is linear in the sources Ki .
Proof: To prove this we shall use dimensional analysis and symmetries.
First we determine the mass dimension of the sources Ki . For a field
φi with dimensionality di the operators Bi have dimensionality di + 1,
as can be Rseen from the expressions of the BRST transformations.
The term d4 xBi Ki ought to have zero dimensionality. Therefore we
conclude that the sources Ki have 3 − di dimensionality. Therefore
the dimensionality of BRST sources corresponding to scalar and vector
122
fields, KA , Kη , Kη̄ is 2 while the dimensionality of the BRST source
corresponfing to fermion fields Kψ is 3/2.
Since the operators of ΓN,∞ are of dimensionality four at most, we
can have operators with at most two sources Ki :
• Kscalar/vector Kscalar/vector ,
• Kfermion Kfermion ,
• Kfermion Kfermion field, with a dimensionality [field] ≤ 1.
All quadratic terms in the sources Ki have a non-zero ghost-number
and they are therefore excluded, with the exception of
Kη̄a Kη̄a
operator which has a zero ghost-number. We can exclude this operator

for a different reason. The BRST symmetry transformation of the
classical action for an anti-ghost is linear and not quadratic,
δθ η̄ a = −θω a . (438)
Therefore,
δL ΓN [hφi i , Ki ] δL ΓN [φi , Ki ]
= − hω a i ; = −ω a (439)
δKη̄ a δKη̄a
where in the last step we used that the transformation is linear so

that, in that case, the transformation of the “average” is equal to the
“average” of the transformation. The above differential equation tells
us that ΓN is at most linear in the source Kη̄a . We have just shown
that ΓN,∞ is at most linear in all sources Ki . We write
Z
ΓN,∞ [φi , Ki ] = ΓN,∞ [φi , 0] + d4 xB̃i Ki . (440)
Recall that the classical action is also linear:

Z
SR [φi , Ki ] = SR [φi ] + d4 xBi Ki . (441)
Substituting in (SR , ΓN,∞ ) = 0 we obtain two equations for the zeroth

and the first order term in Ki . Namely,
δL ΓN,∞ δL SR
Z
d4 x Bi + B̃i = 0, (442)
δφi δφi
123
" #
δL B̃j δ L Bj
Z
4
d x Bi + B̃i = 0. (443)
δφi δφi
We now define:
Γ() ≡ SR + ΓN,∞ , (444)
Bi ≡ Bi + B̃i , (445)
where is a very small parameter. With the Eqs 442, we can prove
that under a field transformation:
φi → φi + θBi , (446)
• Γ() is invariant
• The tranformation is nilpotent (up to O()).
We leave the proof of the above statements to the reader as an exer-
cise.
From the above equations we infer that the dimensionality of B̃i
is at most the dimensionality of Bi . From Eq 443, we also infer that
Bi , B̃i and thus Bi have all the same ghost number (exercise). With
these constraints, the allowed form of the transformations Eq. 446 is
ψ → ψ + i (θη a ) Ta ψ (447)
h i
Aµ,a → Aµa + θ B ab ∂ µ η b + Dabc Aµ,b η c (448)
1
η a → η a − θE abc η b η c (449)
2
with E abc = −E acb due to the ghost field η b being a Grassmann vari-
able.
We can place more constraints on the coefficients of Eqs 447 by
exploiting that the transformations are nilpotent:
• From δθ1 δθ2 η a = 0, we find that
E abc E bde + E abe E bcd + E abd E bec = 0, (450)
which reveals that E abc must be a structure constant of some
Lie algebra. It would not be a surprise if this Lie algebra is the
same as the one of the non-Abelian gauge group of the classical
action SR . Indeed, if we set the small parameter exactly to
its zero value then Γ() = SR . The structure constants E abc

=0
must therefore be proportional to the structure constants of the
non-abelian gauge group of the classical action:
E abc = λf abc . (451)
124
• The nilpotency of the transformation of the gauge field Aµ,a
yields two constraints. Namely
Dabc Dbde − Dabe Dbdc = E bec Dadb = λf bec Dadb (452)
B ab E bcd = Dabd B bc (453)

Eq. 452 tells us that the the matrices t̃abc = iDbca satisfy the
commutation relation of generators in some representation of
the non-abelian gauge group:
c e
t̃ , t̃ = if ceb t̃b .
The only representation of the Lie group with the dimensionality

of Dabc is the adjoint representation. Therefore, the solution of
Eq. 452 is:
Dabc = λf abc . (454)
Eq 453 reveals that the matrix Bab commutes with the structure
constants which can be chosen to be totally antisymmetric. The
only possible solution is therefore a diagonal matrix (exercise).
You can verify this easily in the special case of an SU (2) group
where the structure constants are the totally antisymmetric Levi-
Civita symbol. Eq 453 takes the form:
B ab bcd = B bc abd
which, as examples, for (a, c, d) = (1, 2, 3) yields B 11 = B22

and for (a, c, d) = (1, 2, 2) yields B 23 = 0. Similarly, one finds all
diagonal terms to be equal and the non-diagonal terms to vanish.
We write
B ab = N λδab . (455)
• Nilpotency of the fermion field transformation yields for the ma-
trices T a that the also satisfy the Lie algebra of the non-abelian
group of the classical action,
h i
T b , T c = iE abc T a = iλf abc T a . (456)
Therefore, as suggested from the = 0 limit, we have
T a = λta , (457)
where ta are the generators of the representation for the fermions

in SR .
125
We have therefore found that the Γ() is symmetric under the same
BRST symmetry transformation as SR up to some re-scalings. Ex-
plicitly, the BRST symmetry transformations of Γ() take the form:
ψ → ψ + i (λθη a ) ta ψ (458)
h i
µ,a µa µ a abc µ,b c
A → A + λθ N ∂ η + f A η (459)
1
η a → η a − λθf abc η b η c (460)
2
η̄ a → η̄ a − θω a (461)
ωa → ωa . (462)
The last two transformations are linear symmetry transformations of

the classical action and they are automatically symmetry transforma-
tions of the effective action and Γ as well.
Recall that we expect Γ() to be made out of local operators. We
write Z
Γ = d4 xL() .
()
(463)
The dimensionality of the operators is bounded by the power-counting

arguments of the previous chapter. In addition, L() should consist of
a combination of operators that they respect the BRST symmetry
which we have just discovered (Eqs 458). Finally, L is constrained
further to respect all the linear symmetries of the classical Lagrangian:
1 ξ
L = Lfermion − Gaµν Gaµν −(∂µ η̄ a ) (∂ µ η a )+f abc (∂µ η̄ a ) Ab,µ η c +ω a ∂µ Aa,µ + ω a ω a .
4 2
These linear symmetries are:
• Lorentz invariance
• Global gauge invariance. Explicitly, the global symmetry trans-
formations are:
δψ = ia ta ψ, δAbµ = fbca a Acµ , δη b = fbca a η c , δ η̄ b = fbca a η̄ c , δω b = fbca a ω c .
• Anti-ghost translation invariance: η̄ a → η̄ a + c,

• Ghost-number conservation.
Can we write a Lagrangian density L() with additional operators than
the ones that we find in the classical L and still satisfying the list of
constraints that we have found above? If such operators exist, then we
can establish some simple rules for them. To preserve ghost-number,
126
the ghost and anti-ghost fields must appear in pairs or not appear at all
in such novel operators. Because of anti-ghost translation invariance,
the anti-ghost must always be differentiated. We therefore conclude
that the ghost fields should appear in the form
(∂µ η̄ a ) (464)
Let us recall the dimensionalities of the fields
[Aa,µ ] = [η µ ] = [η̄ a ] = 1, [ω a ] = 2.
The combination of fields in Eq. 464 has a dimensionality three. Oper-

ators must have a dimensionality less than four, thus they can include
at most one such combination of ghost fields. Altogether, we can have
the following operators:
• ghost-operators

(∂µ η̄ a ) ∂ µ η b , (∂µ η̄ a ) Ac,µ η b , (465)
• auxiliary field operators

ω a ∂µ Ab,µ , ω a Acµ Ab,µ , (466)
• and operators which contain only fermion and gauge boson fields.
We denote the sum of them as
LψA . (467)
Therefore, the most general Lagrangian density L() is
ξ0 a a
L() = LψA + ω ω + Cω a (∂µ Aa,µ )
2
−eabc ω a Abµ Ac,µ − Zη (∂µ η̄ a ) (∂ µ η a ) − dabc (∂µ η̄ a ) Ac,µ(468)
ηb,
where ξ 0 , C, dabc , eabc are unknown constants with eabc being symmetric
in the last two indices: eabc = eacb , which are however constrained by
global gauge invariance.
We now use that L() is invariant under the BRST transformations
of Eqs 458. We recall that for fermion and gauge boson fields, the
BRST transformation has the same functional form as a classical local
gauge transformation with a local gauge parameter made out of a
Grassmann constant and the ghost field. Thus, the LψA part of the
127
Lagrangian has to be not only globally gauge invariant but also locally
gauge invariant with a gauge parameter:
a → λN θη a .
and with a gauge coupling gs → gs

/N (equivalently,
replacing the gen-
a
erators and structure constants by t , f abc → t̃ = ta /N, fãbc = f abc /N
a
). BRST invariance of the ghost and auxiliary field part of L() leads
to a determination of the constants. Specifically, we find (exercise):
Zη
C = (469)
λN
Zη
dabc = − (470)
N
eabc = 0. (471)
Summarising the effect of all constraints, we can cast L() in the

form:
h i
L() = −ZA G̃aµν G̃aµν − Zψ ψ̄γ µ ∂µ − it̃a Aaµ ψ
ξ0 Zη

+ ωaωa + ωa ∂µ Aaµ − Zη (∂µ η̄ a ) (∂ µ η a )
2 Nλ
+Zη fãbc (∂µ η̄ a ) Ac,µ η b , (472)
where the field strength tensor G̃aµν is evaluated as in Gaµν with the
replacement f abc → fãbc .
This is a Lagrangian which is vert similar to the classical La-
grangian L, differing only in multiplicative constants. This tells us
that the two Lagrangians describe the same physics, since we are al-
lowed to rescale at wish the definitions of fields, couplings and masses.
We can exploit this freedom to remove all ultraviolet divergences.
With explicit calculations of a few Green’s functions at the N −th
loop order, we can find how the constants that emerged in L() (which
contain necessarily the infinities of all matrix-elements) are related
to the original parameters and field definitions of the classical La-
grangian. With this information at hand and reverse-engineering we
can redefine the fields, fermion masses, and coupling constant so that
at the N −th loop order we have Γ() = SR , which renders ΓN,∞ = 0.
We have proven that non-abelian gauge theories are renormaliz-
able, in the sense that multiplicative redefinitions of fields and param-
eters order by order in the loop expansion can remove all ultraviolet
128
infinities from Green’s functions. This is one of the biggest successes in
Quantum Field Theory since we have realistic theories with predictive
power for physical (i.e. finite) observables.
129
END OF WEEK 12
130
References
[1] The Quantum Theory of Fields, Volume I Foundations, Steven
Weinberg, Cambridge University Press.
[2] The Quantum Theory of Fields, Volume II Modern Applications,
Steven Weinberg, Cambridge University Press.
[3] Gauge Field Theories, Stefan Pokorski, Cambridge monographs
on mathematical physics.
[4] An introduction to Quantum Field Theory, M. Peskin and D.
Schroeder, Addison-Wesley
[5] An introduction to Quantum Field Theory, George Sterman,
Cambridge University Press.
[6] Quantum Field Theory, Mark Srednicki, Cambridge University
Press.
[7] Modern Quantum Mechanics, J.J. Sakurai, Addison-Wesley.
131

QFT 2

Uploaded by

Copyright:

Available Formats

QFT 2

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

QFT 2

Uploaded by

Copyright:

Available Formats

Quantum Field Theory II

2 Path integrals and scalar fields 23

3 Non-abelian gauge theories 43

6 Quantum effective action and the effective potential 79

7 Symmetries of the path integral and the effective action 92

8 Spontaneous symmetry breaking 102

9 Renormalization: counting the degree of ultraviolet diver-

10 Proof of renormalizability for non-abelian gauge theories 120

1.1 The propagator

determines the evolution of this state from an initial moment t1 to a

The sum of the Hamiltonian eigenstates

Ĥ |βi = Eβ |βi (4)

Inserting it in the rhs of the evolution equation we find

In other words, if we know the wave-function at one time, we can

K(x, x0 ; t1 − t1 ) = e0 hx| βi hβ| x0 = δ 3 (x − x0 ).

Exercise: Compute the propagator K(x, x0 ; t − t0 ) for

The probability amplitude to be measured at a position x is

We can attempt to compute the propagator for a transition which

We now specialise to Hamiltonian operators of the form Ĥ = f1 (p̂) +

where in the final step we performed a trivial change of integration

1.2 The path integral

K(xf , xi ; tf − ti ) = K(xf , x1 ; tf − t1 )K(x1 , xi ; t1 − ti ). (22)

If taking the “spanshot” is only a thought experiment and we don’t

We are allowed decide to take several “snapshots” during the transi-

For simplicity, we now consider infinitesimal equally fast tk+1 − tk =

and the sum (integral) in the exponential is the classical action as it

2. Compute the propagator for a small transition

At the time t, something abrupt occurs (e.g. an interaction with an-

Then we allow the particle to evolve undistracted for a time tf − t,

and then we place a detector at xf :

We can compute this matrix-element as a path integral. Before we

We have defined states

which refer to a fixed moment t only and do not evolve

As you recognize, this is the “Heisenberg picture” of evolution. For

hxf , tf | x̂(tj ) |xi , ti i with ti < t < tf ,

following the method we used for the simple transition hxf , tf | xi , ti i =

The subdivision of time is carefully chosen. We have

We then obtain for the transition amplitude the same succession of

and taking equal time intervals, we obtain the path integral

In compact notation we can write

2. Evaluate the above integral explicitly for a free particle

from the result of the integral

We can then compute

Let us now differentiate two times with respect to the source.

T (Ô(t1 )Ô(t3 )Ô(t2 ) . . . Ô(tn )) = Ô(tn ) . . . Ô(t3 )Ô(t2 )Ô(t1 ).

Exercise: Compute hxf , tf | T (x̂(tl )x̂(tq )) |xi , ti i for a free particle.

1.4 Functional differentiation

δF [f (y)] F [f (y) + δ(y − t)] − F [f (y)]

Practically, we need the chain rule and to remember that

The expectation values of time-ordered operators can then be written

We can compute expectation values of operators in the vacuum with

This formula is complicated. It requires that we know the wave func-

Ĥ |ni = En |ni (in the Scrhödinger picture)

and a general state X

Taking a “Heisenberg photograph” of the state in the very past (t →

Now, we play a mathematical game; we give a very small imaginary

Ĥ → Ĥ(1 − i), ( ≈ 0),

which has the same energy eigenstates as the physical Hamiltonian.

δF [f (y)] F [f (y) + δ(y − t)] − F [f (y)]

Ĥ → Ĥ(1 − i), ( ≈ 0),

In the above we simplified (E 2 + ω 2 ) → . We can also define

and the iφ2 is in order to dissipate the contributions from expectation

We consider the case with λ 1. We separate the action into a part