May 11, 2020

version 2.0

Quantum Field Theory II

Lectures Notes
Part III: Non-Abelian Gauge Theories

Prof. Dr. Gino Isidori

ETH & UZH, Spring Semester 2020

i
ii
Contents

3 Non-Abelian Gauge Theories 1

3.1 Geometrical aspects of gauge invariance . . . . . . . . . . . . . . . . . . . . . . . 1
3.2 Non-Abelian Yang-Mills Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.3 Elements of group theory and Lie algebra . . . . . . . . . . . . . . . . . . . . . . 6
3.4 Gauge fixing in non-Abelian gauge theories . . . . . . . . . . . . . . . . . . . . . 9
3.4.1 σ(qq → gg): equality of the coupling constants and role of ghosts. . . . . 11
3.5 The BRST symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.6 One-loop renormalization of non-Abelian gauge theories . . . . . . . . . . . . . . 15
3.6.1 Gauge-boson self-energy: contributions to δ3 . . . . . . . . . . . . . . . . 17
3.6.2 Fermion self-energy: contribution to δ2 . . . . . . . . . . . . . . . . . . . 18
3.6.3 Vertex renormalization: contribution to δ1 . . . . . . . . . . . . . . . . . 18
3.7 Running coupling in QCD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

iii
iv
Chapter 3

Non-Abelian Gauge Theories

3.1 Geometrical aspects of gauge invariance

The purpose of this section is to better understand the idea of gauge invariance, linking it to a
more general geometrical principle. The basic idea is to regard the phases of matter fields, such
as fermion fields or scalar fields, in any point of spacetime, as unobservable quantities. We know
how this works in QED: requiring the theory to be invariant under local transformations on
the phase of the electron field, we deduce the necessity of introducing the eletromagnetic field.
This suggests to assume local phase transformation invariance as a guiding principle to build
interacting field theories. In the rest of this chapter we would like to generalise this concept to
other theories, where the underlying transformation is more complicated, and in particular is
associated to a non-Abelian symmetry. To do so, let’s first revist what happens in the case of
QED.
The starting point is the requirement that the theory is invariant under the transformation
of the spinor field
ψ(x) → eiα(x) ψ(x) (3.1)
Difficulties arise with the derivative of the fields in the direction nµ :
ψ(x + n) − ψ(x)
nµ ∂µ ψ(x) = lim (3.2)
→0 
the problem lies in the fact that the difference of two fields at different space-time points is
not well-defined: it depends on our arbitrary choice of the phases. In order to define a ‘proper’
definition of ‘derivative’ for local operators we introduce a ‘compensator’ field U (y, x) that
transforms as
U (y, x) → eiα(y) U (y, x)e−iα(x) , (3.3)
with
U (x, x) = 1, |U (x, y)| = 1 (3.4)
so that
U (y, x)ψ(x) → eiα(y) U (y, x)ψ(x) (3.5)
The covariant derivative provides us with a good definition of ’derivative’ of local fields in gauge
theories,
ψ(x + n) − U (x + n, x)ψ(x)
nµ Dµ ψ(x) = lim . (3.6)
→0 

1
In fact, the two objects in the numerator, differently from the previous definition, transform
according to the same law and the modulus of the numerator itself is independent of our choice
of the phase. Expanding U we find

U (x + n, x) = 1 − ienµ Aµ (x) + O(2 ), (3.7)

where e is an arbitrary normalization constant and the field Aµ is called a connection. Taking
into account (3.7), Eq. (3.6) becomes
ψ(x + n) − ψ(x) + ienµ Aµ (x)ψ(x)
nµ Dµ ψ(x) = lim , (3.8)
→0 
and we recover the familiar expression

Dµ ψ(x) = [∂µ + ieAµ ]ψ(x) (3.9)

Performing the gauge transformation of U we obtain

1 − ienµ Aµ → eiα(x+n) [1 − ienµ Aµ (x)]e−iα(x)

= eiα (1 + inµ ∂ µ α)[1 − ienµ Aµ (x)]e−iα(x)
 
µ 1 µ
= 1 − ienµ A − ∂ α + O(2 ) , (3.10)
e
which implies that the connection transforms as
1
Aµ → Aµ − ∂ µ α . (3.11)
e
Using this result one recovers
Dµ ψ(x) → eiα(x) Dµ ψ(x) (3.12)
as expected by the transformation properties of (3.6). We have thus recovered most of the
ingredients of the QED Lagrangian. Note that the definition of the covariant derivative and the
transformation law for Aµ follow from the requirement of local phase transformation invariance.
Even the very existence of the vector field Aµ is a consequence of the local symmetry: without
such filed we would have not been able to write a invariant Lagrangian involving derivatives of
ψ.
We can now construct a kinetic term for Aµ and, more generally, all the operators made
in terms of Aµ that are invariant under local gauge transformations (continuous symmetry).
From the definition of U we can verify that

Û (x) := U (x, x + n̂2 )U (x + n̂2 , x + n̂2 + n̂1 )U (x + n̂1 + n̂2 , x + n̂1 )U (x + n̂1 , x) , (3.13)

which we denote as the ‘Plaquette’ in the 1–2 subspace. is invariant under a local phase
transformation. The construction of Û is depicted in Fig. 3.1.
Taking the limit  → 0 we will thus define a locally invariant function of Aµ . To evaluate
Û (x) explicitly, taking into the expansion in (3.7), and given U † (x, y) = U (y, x), we can write
   
µ x+y 2
U (x, y) = exp −ie(x − y)µ A + O((x − y) ) (3.14)
2
and therefore

U (x + n, x) = exp{−ienµ Aµ (x + n) + O(2 )} (3.15)
2
2
 x + ✏n̂2 x + ✏(n̂1 + n̂2 )

x x + ✏n̂1

Figure 3.1: Plaquette in the 1–2 subspace.

Hence, after a couple of steps,

 
  
Û (x) = exp − ie − A2 (x + n2 ) − A1 (x + n2 + n1 ) + A2 (x + n2 + n1 )
2 2 2
 

+ A1 (x + n1 ) + O(3 ) = exp −ie2 [∂1 A2 (x) − ∂2 A1 (x)] + O(3 ) (3.16)

2

The term in the exponent is noting but the element F12 of the field strength tensor Fµν =
∂µ Aν − ∂ν Aµ . We thus rediscovered that Fµν is the basic gauge-invaraint function built in
terms of Aµ .
A complementary way to recover Fµν is to start from the transformation property of the
covariant derivative. Applying a covariant derivative to Dν ψ(x) we find

Dµ Dν ψ(x) → eiα(x) Dµ Dν ψ(x) (3.17)

hence
[Dµ , Dν ]ψ(x) → eiα(x) [Dµ , Dν ]ψ(x) (3.18)
Evaluating explicitly the commutator,

[Dµ , Dν ]ψ(x) = [∂µ + ieAµ , ∂ν + ieAν ]ψ(x) = ieFµν ψ(x), (3.19)

we find that
[Dµ , Dν ] = ieFµν . (3.20)
Since ieFµν (x)ψ(x) → ieiα(x) eFµν (x)ψ(x) we deduce that the field strength Fµν transforms as
Fµν (x) → eiα(x) Fµν (x)e−iα(x) , which in this (Abelian) case means that Fµν is gauge invariant.
We can now write the most general locally invariant Lagrangian for the spinor field ψ(x)
and its associated connection Aµ (x). In the standard four-dimensional case there are only four
possible terms which are both Lorentz and gauge invariant:
1
/ − Fµν F µν − cαβµν Fαβ Fµν − mψ̄ψ
L = ψ̄(iD)ψ (3.21)
4
We notice that mass terms like ma Aµ Aµ would not respect gauge invariance and must therefore
be excluded. The third term violates parity and time reversal and can therefore be excluded
from any theory postulating these symmetries. In this case the only free parameters in L are
e and m. In d > 4 there are many more Lorentz and gauge invariant terms, that in the four-
dimensional case would be non-renormalizable interactions. From the point of view of effective

3
field theories, they are irrelevant to physics at low energies (in four dimensions) in the limit of
a very heavy ultraviolet cutoff.
Summarising, the postulate of local phase invariance together with Lorentz invariance has
allowed us
• to infer the existence of a massless photon field Aµ ;
• to deduce the QED Lagrangian as the only relevant Lagrangian at low energies.

3.2 Non-Abelian Yang-Mills Lagrangian

Yang and Mills proposed that the argument of the previous section could be generalized from
the case of a U (1) symmetry (which is an Abelian unitary symmetry) to any continuous unitary
group.
Let’s consider a set of N fields ψi
 
ψ1
ψ =  ...  (3.22)
 

ψN

and the unitary transformation

ψi → ψi0 = Vij ψj (3.23)
For example, in the case of the neutron-proton doublet the group of symmetry is SU (2) with the
σa
 
a 1 2 3 ψn
three generators t = where σ , σ , σ are the Pauli matrices. The spinor field ψ =
2 ψp
transforms as
a σa a a
ψ → ψ 0 = eiα 2 ψ = eiα t ψ. (3.24)

Coming back to the general discussion, we want to build a Lagrangian imposing local gauge
invariance under ψ(x) → V (x)ψ(x). We then introduce the compensator field U (x, y) that
transforms as
U (x, y) → V (x)U (x, y)V † (y) (3.25)
with U (x, x) = 1, |U (x, y)| = 1, and we decompose it as
a (x)ta
U (x + n, x) = I + ignµ Aaµ ta , V (x) = eiα , (3.26)

where a runs over the number of generators ta of the group. In analogy with the previous
section, defining the covariant derivative as

ψ(x + n) − U (x + n, x)ψ(x)

nµ Dµ ψ(x) = lim . (3.27)
→0 
we find
Dµ = ∂µ − igAaµ ta . (3.28)
So far everything works as in the Abelian case. However, the transformation law of Aµ is
different:
1 + ignµ Aaµ ta → V (x + n)(1 + ignµ Aaµ ta )V † (x) . (3.29)

4
Using the identity

V (x + n)V † (x) = 1 + nµ ∂µ + O(2 ) V (x) V † (x)




= 1 + nµ (∂µ V (x))V † + O(2 )

= 1 + nµ V (x)(−∂µ V † (x)) + O(2 ) (3.30)

we find  
i
Aaµ ta → V (x) Aaµ ta + ∂µ V † (x) (3.31)
g
With this expression it is easy to check the transformation law of Dµ ψ:
 
a a a a i †
Dµ ψ = ∂µ − igAµ t → ∂µ − igV (Aµ t + ∂µ )V V ψ
g
= [∂µ + V (∂µ V † )]V ψ − igV Aaµ ta ψ
= V [∂µ − igAaµ ta ]ψ = V Dµ ψ (3.32)

where we have used V (∂µ V † )V ψ = V ∂µ ψ − ∂µ (V ψ). To get the infinitesimal transformation

(3.31) in terms of α(x) we need to expand V (x):

V (x) = 1 + iαa (x)ta + O(α2 ), ∂µ V † = −i∂µ αa (x)ta + O(α2 ) (3.33)

obtaining
 
i 1
a a
V Aµ t + ∂µ V † = (1 + iαa ta )(Abµ tb )(1 − iαa ta ) + ∂µ αa ta + O(α2 )
g g
a
t
= Aaµ ta + i[ta , tb ]Abµ αa + ∂µ αa + O(α2 )
g
a
t
= Aaµ ta + f abc Abµ αc ta + ∂µ αa + O(α2 ), (3.34)
g

where f abc are the structure constants of the group (see next section). The transformation law
of the connection is therefore

1
Aaµ → Aaµ + ∂µ αa + f abc Abµ αc (3.35)
g

We want now to find the analogue of F µν . We know that under a gauge transformation

[Dµ , Dν ]ψ → V (x)[Dµ , Dν ]ψ (3.36)

and
[Dµ , Dν ] = −ig(∂µ Aaν − ∂ν Aaµ )ta − ig 2 f abc Abµ Acν ta = −igFµν
a a
t (3.37)
with
a
Fµν = ∂µ Aaν − ∂ν Aaµ + gf abc Abµ Acν (3.38)

it is important to notice that Fµν is not invariant under gauge transformations, but it trans-
forms linearly (more precisely, as we shall see later, it transforms as an element in the adjoint
representation of the group):
a a a a †
Fµν t → V (x)Fµν t V (x) . (3.39)

5
For infinitesimal transformations this implies
a a
Fµν → Fµν − f abc αb Fµν
c
. (3.40)

We are now ready to build the generalization of the QED Lagrangian for a set of spinor fields
invariant under a non-Abelian local gauge symmetry. The analogue of the term (Fµν )2 in QED
is
a a
tr[(Fµν t )(F b,µν tb )] = Fµν
a
F b,µν tr[ta tb ] (3.41)
where tr[ta tb ] ∝ δ ab . We can define the normalization of the generators of the gauge group such
that
1
tr[ta tb ] = δ ab (3.42)
2
With such choice we then obtain
1 a a,µν
LYM = − Fµν F / − mψ̄ψ
+ ψ̄iDψ (3.43)
4
As it can easily be checked, differently from the Abelian case, the Yang-Mills Lagrangian
contains also cubic and quartic interactions among the gauge fields.

3.3 Elements of group theory and Lie algebra

A group is a set of elements together with an associative operation such that
• g · h ∈ G ∀g, h ∈ G
• ∃I : g · I = g, ∀g ∈ G
• ∃g −1 : g · g −1 = I = g −1 · g, ∀g ∈ G
We are interested in a rather special category of groups: the Lie groups. In Lie groups the ele-
ments can be parameterised in terms of continuous parameters, g = g(α), and can be connected
to the identity via a relation of the type

g(α) = I + iαa T a + O(α2 ) . (3.44)

Here αa are the continuous parameters and T a are Hermitian operators called the generators
of the group. The generators span the space of infinitesimal group transformations. They form
a closed algebra (Lie Algebra) that is completely specified by the commutation relations

[T a , T b ] = if abc T c . (3.45)

The commutation relations and the Jacobi identity

[T a , [T b , T c ]] + [T b , [T c , T a ]] + [T c , [T a , T b ]] = 0, (3.46)

imply
f ade f bcd + f bde f cad + f cde f abd = 0 (3.47)
We can classify Lie groups according to the following definitions:
• Compact groups: They have a finite number of generators with finite dimensional
representations

6
 • Semi-simple groups: there is not a single generator commuting with all the others (no
U (1) factors).
• Simple group: we cannot divide the group into two separate subsets of mutually com-
muting generators (each forming a group).
For example, U (2) = U (1) × SU (2) is compact but not semi-simple, SU (2) × SU (2) is compact
and semi-simple, and finally SU (2) is compact and simple. Two examples of compact and
simple groups, that we encounter often in particle-physics and condensed-matter systems, are
SU (N ) and SO(N ):

• SU (N ) is the group of unitary N × N matrices with determinant of modulus 1. The

linear transformation of the complex vectors ηa , ξa , under the action of V ∈ SU (N ), is

ηa → Vab ηb , ξa → Vab ξb , a = 1, · · · , N. (3.48)

The fact that V is unitary implies the conservation of the inner product

ηa∗ ξa → ηa∗ ξa . (3.49)

The conservation of the inner product is equivalent to the relations

V † V = V V † = I. (3.50)

The generators of SU (N ) are a set of N × N traceless Hermitian matrices. The number

of independent generators is N 2 − 1.

• SO(N ) is the rotational group on a real N -dimensional vector space. The linear trans-
formation of the complex vectors va , za , under the action of O ∈ SO(N ), is

va → Oab vb , za → Oab zb , a = 1, · · · , N, (3.51)

and it such that the product va za is left invariant. This implies OOT = I. Since there is
an independent rotation for each plane, the number of generators is

N (N − 1)
 
N
= (3.52)
2 2

In the following we concentrate on simple groups only and, most of the type, on the specific
case of SU (N ).
Given a symmetry group G, a finite-dimensional unitary representation of the group’s
Lie algebra is a set of d × d Hermitian matrices ta that satisfy the commutation relations
[ta , tb ] = if abc tc . The size d is the dimension of the representation. An arbitrary representa-
tion can generally be decomposed by finding a basis in which all representation matrices are
simultaneously block-diagonal. Through this change of basis, we can write the representation
as the direct sum of irreducible representations. We denote the representation matrices in the
σa
irreducible representation r by tar . For example (a = 1, 2, 3) constitute a 2-dimensional repre-
2
sentation of SU (2), while the antisymmetric tensors iaij are a three dimensional representation
of SU (2).

7
 If the Lie algebra is semi-simple, then the generators are traceless (i.e. tr(tar ) = 0). Since
the generators are hermitians, tr[tar tbr ] is positive definite, and it is possible to choose a basis
and a normalisation for the operators such that

tr[tar tbr ] = C(r)δ ab , (3.53)

where C(r) is a constant that depends on the specific representation. This relation, together
with Lie algebra, implies
i
f abc = − tr{[tar , tbr ]tcr } (3.54)
C(r)
showing that the structure constants are totally antisymmetric.
The N -dimensional representation of SU (N ) is called the fundamental representation. The
adjoint representation (denoted by the subscript A) of SU (N ) is defined by the generators:

(tbA )ac = if abc . (3.55)

Using the Jacobi identity in terms of the structure constants it is not difficult to check that
this is a good definition: the generators of the adjoint representation satisfy the commutation
relations of the Lie algebra. Using the adjoint representation we have a better understanding
of the transformation law of Aaµ . In fact,

1 1
Aaµ → Aaµ + ∂µ αa + f abc Abµ αc = Aaµ + ∂µ αa − i(tbA )ac Abµ αc =
g g
1 a 1
Aaµ + (∂µ − igtbA Abµ )α = Aaµ + (Dµ α)a

(3.56)
g g
X
The operator T 2 = T a T a commutes with all the generators. In fact,
a

[T b , T a T a ] = [T b , T a ]T a + T a [T b , T a ] = if bac T c T a + if bac T a T c = if bac {T c , T a } = 0 (3.57)

This implies, by Schur’s lemma, that in any irreducible representation r

T 2 = tar tar = C2 (r) · I (3.58)

where C2 (r) is a constant, called the quadratic Casimir operator. Contracting tr[tar tbr ] = C(r)δ ab
with δab we find

d(G)C(r) = tr[tar tbr ]δab = tr[tar tar ] = tr[I]C2 (r) = d(r)C2 (r) (3.59)

where d(r) is the dimension of the representation labelled by r and d(G) is the dimension
of the group, that is also the dimension of the adjoint representation. For the two notable
representations of SU (N ) we have:
• Fundamental representation:

1 C(F )d(G) N2 − 1
d(F ) = N , C(F ) = , C2 (F ) = = . (3.60)
2 d(F ) 2N
• Adjoint representation:

d(F ) = N 2 − 1 , C(A) = C2 (A) = N. (3.61)

8
3.4 Gauge fixing in non-Abelian gauge theories
In order to quantize the theory we need to introduce a gauge-fixing procedure, as in the Abelian
case. We define the generating functional for the pure gauge theory
Z  Z  
4 1 a a,µν µ
W [J] = DAµ exp i d x − Fµν F + Aµ J (3.62)
4

As in the Abelian case the kinetic term cannot be inverted, having a null eigenvalue. The
Lagrangian is unchanged along the infinite number of directions in the space of field configura-
tions corresponding to local gauge transformations. To compute the functional integral we must
factor out the integrations along these directions, constraining the remaining integral to the
smaller subspace of physically inequivalent field configurations. Proceeding as in the Abelian
case we thus insert the identity

δ(G(Aα ))
Z Z  
α
1 = DG(A)δ(G(A)) = Dα(x)δ(G(A ))det (3.63)
δα

where the gauge-fixing condition is G(A) = 0, is obtained with G(A) = ∂ µ Aaµ − ω a (x). Recall
that
1 1
(Aα )aµ = Aaµ + ∂µ αa + f abc Abµ αc = Aaµ + Dµ αa (3.64)
g g
| {z }
in adjoint repr.

Q
and δ(G(A)) is a δ-functional, δ(G(A)) = x δ(G(A(x))), i.e. it is 0 unless G(A) = 0 every-
where. The variables change in the inner integral A → Aα can be regarded as a linear shift
of the Aaµ , followed by a unitary rotation of the various components of the symmetry multi-
plet Aaµ (x) at each point. Both of these operations preserve the measure, thus DA = DAα .
Introducing the Gaussian weight on ω a (x) we thus have

δ(G(Aα )) −i R d4 x (ω2ξa )2
Z Z Z  
A→Aα α iS[Aα ] α
W [J] ∝ Dα Dω DA e δ(G(A ))det e
δα
δ(G(Aα ))
Z Z  
Aα →A iSξ [A]
∝ Dα DAµ e det (3.65)
δα

where Z  
41 a a,µν 1 µ a 2
Sξ [A] = d x − Fµν F − (∂ Aµ ) (3.66)
4 2ξ
Then, following the same steps of the Abelian case we find

d4 k −i
Z  
a b kµ kν
hAµ (x)Aν (y)i = gµν − (1 − ξ) 2 δ ab e−ik·(x−y) (3.67)
(2π)4 k 2 + i k

So far everything works as in the Abelian case. The subtlety of the non-Abelian case is the
δ(G(Aα ))
dependence of det on Aµ . Using the infinitesimal form of the gauge transformation
δα
we find
δ(G(Aα ))
 
δ 1 µ 1
= ∂ Aµ + ∂ Dµ α − ω = ∂ µ Dµ
µ a a a
(3.68)
∂α δα g g

9
  
1
We need to evaluate det ∂ µ Dµ that now depends on Aµ . In analogy to the Grassmann
g
integral identity
YZ
det[B] = dθ̄k dθk e−θ̄i Bij θj , (3.69)
k
Faddeev-Popov proposed to write
  Z  Z 
1 µ 4 a µ ac c
det ∂ Dµ = DcDc̄ exp i d xc̄ (−∂ Dµ )c (3.70)
g
where the coupling has been absorbed inside, and c and c̄ are anticommuting scalar fields. These
special fields that we have introduced have the wrong statistical structure (anticommuting but
with a Klein-Gordon type kinetic term) and should therefore be considered as non-physical par-
ticles, they are usually called ghosts. The role of the ghosts is to provide a simple representation
of the functional determinant from the gauge fixing, that we can easily handle in perturbation
theory (in terms of Feynman diagrams). The ghost term in the Lagrangian is

Lghosts = c̄a [−∂ 2 δ ac − g∂µ f abc Abµ ]cc (3.71)

The first term gives us the ghost propagator ki2 δ ac while the second gives us the gauge-ghost-
ghost vertex i(ipµ )gf abc = −gpµ f abc where pµ is the momentum of c̄. The complete Lagrangian
is therefore
1 a aµν 1 X
L = − Fµν F − (∂µ Aaµ )2 + c̄a (−∂ µ Dµac )cc + / − mf )ψf .
ψ̄f (iD (3.72)
4 2ξ f

As we shall see, the ghost particles serve as negative degrees of freedom to cancel the effects
of the unphysical timelike and longitudinal polarization states of the massless gauge bosons.
They role is essential to enforce the validity of Ward identities in non-Abelian gauge theories
beyond the tree level.
To start this discussion, let’s give a look to the polarisations of the gauge fields. From the
free field equations we know that we can decompose the gauge fields into plane waves:
3
X
Aµ = iµ e−ikx (3.73)
i=0

Of the four independent polarizations, two are physical

Tµ1 = 0, 1, 0, 0 , Tµ2 = 0, 0, 1, 0



(3.74)


and transverse to the momentum kµ = k, 0, 0, k and two correspond to unphysical degrees
of freedom. Given the unit vector nµ with

n · k 6= 0, n · T1,2 = 0, n2 = 1, (3.75)

the other two polarization vector can be constructed as

kµ
0 = nµ (timelike), 3 = nµ − (longitudinal) (3.76)
n·k
the sum over polarizations is therefore
3 2  
X X nµ kν + kµ nν kµ kν
∗(i)
µ ν
(i)
= −g µν
= µ∗(Ti ) (T
ν
i)
+ − + . (3.77)
i=0 i=1
n·k (n · k)2

10
 Figure 3.2: Gauge-boson and ghosts loops.

The term in square brackets is the contribution from the unphysical longitudinal polariza-
tions. In QED this term is always irrelevant because photons interact only with fermions, and
kµ J µ = 0 if J µ is a fermion current. Conversely, this term it gives a non-vanishing contribution
in the non-Abelian case since gluon can interact among themselves. In particular, the unphys-
ical polarizations give a non-zero contribution to the upper loop diagram in Fig. 3.2 but are
cancelled exactly by the corresponding ghost diagram shown below.

3.4.1 σ(qq → gg): equality of the coupling constants and role of

ghosts.
In order to check how the Ward identity works in non-Abelian theories, and the special role
played by ghosts, it is useful to compute explicitly the amplitude for qq → gg in QCD. We
start doing so assuming physical (hence transverse) external gauge bosons. At g 2 order there
are three diagrams, shown in the picture below:

a b a b a b
k1 k2
k1 k2 k1 k2
c
p+ + p + k3
p + p
p+ p

The amplitude for the first two diagrams is

( )
1 1
iMab,µν
12 ∗µ (k1 )∗ν (k2 ) = (ig)2 v̄(p+ ) γ µ ta γ ν tb + γ ν tb γ µ ta u(p)∗µ (k1 )∗ν (k2 )
p/ − k/2 − m k/2 − p/+ − m
(3.78)
µ
If we assume physical external states, the polarization vectors must satisfy ki µ (ki ) = 0. Our
scope is to the Ward identity, namely the relation

k2ν Mfull ab,µν = k1µ Mfull ab,µν = 0 . (3.79)

To do so, let’s check what happens if we substitute ∗ν by k2ν in the amplitude of the first two
diagrams:
( )
ab,µν ∗ µ a 1 b b 1
2
iM12 1µ k2ν = (ig) v̄(p+ ) γ t k/2 t + k/2 t γ t u(p)∗1µ (3.80)
µ a
p/ − k/2 − m k/2 − p/+ − m

Since
(p/ − m)u(p) = 0, v̄(p+ )(−p/+ − m) = 0, (3.81)

11
we can add them to k/2 in both the terms of iMµν ∗
12 1µ k2ν in order to cancel the denominators.
We find
iMab,µν ∗1µ k2ν = −g 2 v̄(p+ ) −iγ µ [ta , tb ] u(p)∗1µ .


12 (3.82)
| {z }
f abc γ µ tc

While this term is zero in the Abelian case, we see that in the non-Abelian an additional
contribution is necessary in order to preserve the Ward identity. In fact, this is the one provided
by the third diagram in the picture (which is present only in the non-Abelian case). The
associated amplitude is
−i ∗ ∗
iMab,µν
3 ∗1µ ∗2ν = igv̄(p+ )γρ tc u(p)  gf abc T µνρ
2 1µ 2ν
k3
T µνρ = [g (k2 − k1 ) + g νρ (k3 − k2 )µ + g ρµ (k1 − k3 )ν ]
µν ρ
(3.83)

with k3 = −k1 − k2 = −p − p+ . Replacing ∗2ν → k2ν in the above expression leads to

k2ν T µνρ = k2µ (k2 − k1 )ρ + k2ρ (k3 − k2 )µ + g ρµ (k1 − k3 ) · k2

k ·=0
= g ρµ k32 − k3ρ k3µ − g ρµ k12 + k1ρ k1µ −→ g ρµ k32 + k1ρ k1µ −→
1
g ρµ k32 (3.84)

The first arrow in the above equation takes into account the fact that terms proportional to
k3µ vanish when contracted with the fermion current, and that k1 is on-shell (k12 = 0). Finally,
if we also assume that the first gluon is transverse, namely k1µ 1µ (k1 ) = 0, we remain with
a single term proportional to k32 . This factor cancels the gluon propagator and we find that
iMab,µν
3 ∗1µ k2ν precisely cancels iMab,µν
12 ∗1µ k2ν .
As can be seen, the cancellation takes place only if the value of the coupling constant in the
three-boson vertex is identical to that in the fermion-boson vertex. In a similar way, the Ward
identity cannot be satisfied among the diagrams for boson-boson scattering unless the coupling
constant g in the four-boson vertex is identical to that in the three-boson vertex. Thus, the
coupling constants of all three nonlinear terms in the Yang-Mills Lagrangian must be equal in
order to: i) preserve the Ward identity and ii) avoid the production of bosons with unphysical
polarization states.
A key element for the above derivation was the assumption of on-shell and transverse ex-
ternal gauge bosons. This is bit strange since we would expect the condition of transversality
to emerge as result of the calculation (as in QED), and something that we need to impose. To
better understand this aspect, let’s look now at the amplitude for producing a pair of ghosts:
a b
k1

ρ
ab −i ρ acb abc k1
iMghost = igv̄(p+ )tc γρ u(p) (−g)k 1 f = g 2
v̄(p + )tc
γρ u(p)f (3.85)
q2 q2
If we do not require µ1 · k1µ = 0 we find
ab,µν
k2ν M1,2,3 − k1µ Mghost
ab
=0 (3.86)

12
This is a sort of generalised Ward identity that holds also for 1 · k1 6= 0. We’ll clarify its
origin in the next lecture, in terms of the BRST symmetry. For the moment we limit ourself
to highlight its important physical interpretation.
From the optical theorem (that, as you have seen in QFT-I, is a consequence of the unitarity
of the S matrix) we know that
X
σ∝ |M (q q̄ → f )|2 ∝ ImM (q q̄ → q q̄) (3.87)
f

Diagrammatically, we can read this relation as the sum of the following four diagrams (to lowest
order in perturbation theory), where the dotted vertical line indicates that we have to take the
imaginary part of each diagram:

+ + +

1) 2) 3) 4)

In the diagrams 1) and 3) we can replace the gµν factors in the propagators with sums over
all four polarization vectors. While the contribution of unphysical polarizations cancel in the
first diagram due to contraction with fermion lines (kµ J µ = 0), in the third diagram this does
not happen. We conclude that the third diagram contains non-vanishing contributions from
unphysical polarization in the gluon loop. On the other hand, |M (q q̄ → f )|2 is a physical
quantity and cannot contain contributions from unphysical polarizations. The generalized
Ward identity and the presence of the ghost loop solves the apparent paradox: the ghost
degrees of freedom cancel the unphysical polarizations coming from the gluon loop. Looking at
this example it is also quite clear while in QED we did not need ghosts: there the unphysical
polarizations of the photons always cancel since the photon couples only to the fermion current.

3.5 The BRST symmetry

A more rigorous way to understand the role of the ghosts in non-Abelian gauge theories is
obtained by means of the so-called BRST symmetry. Let us rewrite the Yang-Mills Lagrangian
by introducing a new auxiliary non-propagating scalar field (B a ):

1 a aµν ξ
L = − Fµν F / − m)ψ + (B a )2 + B a ∂µ Aaµ + c̄a (−∂ µ Dµac )cc
+ ψ̄(iD (3.88)
4 2

The B a as a trivial (non-derivative) kinetic term and therefore we can easily integrating over
it. Doing so we recover the standard gauge-fixing term. Since

ξ a 2 ξ 1 1
(B ) + B a ∂µ Aaµ = (B a + ∂µ Aaµ )2 − (∂µ Aaµ )2 (3.89)
2 2 ξ 2ξ

the first term on the r.h.s. goes away after a shift of B a in the functional integral.

13
 Let’s now consider the following infinitesimal transformation

δAaµ := (Q̂Aµ )a = Dµab cb

δψ := (Q̂ψ) = igca ta ψ
1
δca := (Q̂c)a = − gf abc cb cc
2
δc̄ := (Q̂c̄) = B a
a a

δB a := (Q̂B)a = 0 (3.90)

where  is a global (space-time independent) anti-commuting parameter. The first two transfor-
1
mations can be regarded as gauge transformations of the type δAaµ = (Dµ α)a with αa = gca ,
g
hence they leave invariant the first two terms in Eq. (3.88). The third term in Eq. (3.88) is
trivially invariant. The fourth and last terms in Eq. (3.88) are left invariant by the combined
action of δc̄a and δAaµ . Finally, looking at the transformation of (Dµ c)a , which appear in the
last term in Eq. (3.88), we get:

δ(Dµ c)a = [Dµ (δc)]a + gf abc (δAbµ )cc

g g2
= − ∂µ (f abc cb cc ) − f cde (f abc cb cc )Abµ cd ce + gf abc (∂µ cb )cc + g 2 f abc f bde Adµ ce cc
2 2
2 
g
f abc f cde + f abc f cde + f abc f cde Abµ cd ce = 0 ,

= − (3.91)
2
where we used the anticommuting property of the ghost fields and finally the Jacobi indentity.
Hence we can conclude that the transformations (3.90), discovered by Becchi, Rouet and
Stora in 1976, and since then denoted BRST transformations, are a new symmetry of the
Lagrangian. The remarkable (and somehow not obvious) feature is this symmetry is valid
irrespective of the gauge fixing terms. The generalized Ward identity derived in the previous
lecture can be shown to be a consequence of this new symmetry.
More generally, the symmetry of the Lagrangian implies the existence of a conserved current
µ
jBRST ,
whose conserved charge, Z
Q̂ = d3 x jBRST
0
, (3.92)

is the generator of the transformations (3.90). It is quite simple to show, by explicit calculation,
that Q̂(Q̂φ) vanishes for all the fields φ in our Lagrangian, i.e. that Q̂ is a nilpotent operator

Q̂2 = 0 . (3.93)

This fact, together with the BRST invariance of the Lagrangian has very important conse-
quences. Let us classify the Hilbert space (and the fields) of the theory according to

BRST-closed states/fields, |φclosed i, such that Q̂|φclosed i = 0 ,

BRST-exact states/fields |φexact i, for which it exists a |χi such that |φexact i = Q̂|χi .

According to this definition any exact state is closed, but not viceversa. It can be show that:
i) the physical states (such as the transverse components of the gauge fields) are Q̂-closed states
which are not Q̂-exact states, ii) the unphysical states (ghosts and unphysical polarizations)

14
are not Q̂-closed states. While we do not provide a complete proof of this statement, this can
be understood intuitively the by noting that: i) gauge-invariant operators O(x), such as the
first two terms in in Eq. (3.88), satisfy Q̂O(x) = 0, ii) the gauge fixing part of the Lagrangian
can be written as an exact BRST field:
 
ξ a 2 a aµ a a a aµ ξ a a
Lgauge−f ixing := (B ) + B ∂µ A + ∂µ c̄ (Dµ c) = Q̂ c̄ ∂µ A + c̄ B . (3.94)
2 2

We can then define the cohomology of Q̂ as the quotient Hilbert space

Q̂-closed fields
. (3.95)
Q̂-exact fields
Given a physical state, described by the field φ, the equivalence class [φ] defined by the coho-
mology of Q̂ is the whole set of fields of the form φ + Q̂χ for any χ. This equivalence class is the
set of all physically equivalent field configurations that can be reached by a gauge transforma-
tion starting from φ. The cohomology of Q̂ is what defines the different physical states, i.e. the
matrix element h[φ1 ]|[φ2 ]i is equivalent to the matrix element of its physical components:

h[φ1 ]|[φ2 ]i = hφ1 + Q̂χ1 |φ2 + Q̂χ2 i

= hφ1 |φ2 i + hQ̂χ1 |φ2 + Q̂χ2 i + hφ1 |Q̂φ2 i
= hφ1 |φ2 i + hχ1 |Q̂|φ2 i + hχ1 |Q̂2 |χ2 i + hφ1 |Q̂|φ2 i
= hφ1 |φ2 i (3.96)

This let us to conclude that the exact BRST terms (hence the non-physical states) never enters
into physical amplitudes.

3.6 One-loop renormalization of non-Abelian gauge the-

ories
In order to renormalize at one-loop order the non Abelian-field theory studied so far we split
the Lagrangian
1 a a 1
L = − Fµν Fµν − (∂µ Aaµ )2 + ψ̄(iD
/ − m)ψ + c̄a (−∂ µ Dµac )cc , (3.97)
4 2ξ
into a free Lagrangian, i.e. the quadratic part of the three fields {Aaµ , ψ, ca }

1 1
Lf ree = − (∂µ Aaν − ∂ν Aaµ )2 − (∂µ Aaµ )2 + ψ̄(i∂/ − m)ψ + ∂µ c̄a ∂µ ca , (3.98)
4 2ξ
and interacting Lagrangian, which contain vertices with 3- and 4-field vertices. The three fields
are renormalised in a multiplicative way as
1 1 1
Aµ = Arµ Z32 , ψ = ψ r Z22 , c = Z2c2 cr . (3.99)

Recal that, on general grounds, we define the renormalised coupling λ from a N -field vertex
interaction as
λZV = λ0 (Zfield )N/2 (3.100)

15
where ZV is the vector renormalization factor and λ0 the bare coupling. Looking at the inter-
acting part of the non Abelian-field theory and and introducing a renormalization factor for
each vertex, we then obtain
 1
a a


 gA µ ( ψ̄it ψ) → gZ 1 = g Z
0 3 Z2
2

 1
− gf abc Abµ Acν (∂µ Aaν − ∂ν Aaµ ) → gZ3g = g0 Z32

 3

Lint = 2 (3.101)
1 2 abc ade b c dµ eν 2 2 2

 − g f f A A
µ ν A A → g Z 4g = g Z
0 3
4


 1
+gf abc ∂µ c̄a Abµ cc

→ gZ1c = g0 Z2c Z32


where g is the renormalized coupling. The four vertex normalization factors appearing above
(ZV , with V = 1, 3g, 4g, 1c) are all necessarily connected by the symmetry of the system: the
Ward identities implies the same coupling for all the vertices, i.e. the same renormalization of
g by any of the term in Eq. (3.101). This implies the following relatiuons
1 3  2  12 1
g Z32 Z2 Z32 Z3 Z2c Z32
= = = = , (3.102)
g0 Z1 Z3g Z4g Z1c
that, expanding the Zi as Zi = 1 + δi , can be re-written as
1
δ2 − δ1 = δ3 − δ3g = (δ3 − δ4g ) = δ2c − δ1c . (3.103)
2
In analogy with QED, the beta function can be computed as
1 !  
∂ Z32 Z2 ∂ 1
β(g) = g0 µ = g0 µ δ3 + δ2 − δ1 , (3.104)
∂µ Z1 ∂µ 2

but all other options which can be derived by the other identities in Eq. (3.102) are equally
good.
In QED we found δ2QED − δ1QED = 0 to all orders in perturbation theory, which implies that
the QED beta function depends only on the photon-field renormalization. As we have already
seen, this is a consequence of the Ward identity

∂µ hj µ (x)ψ(x1 )ψ̄(x2 )i = −δ(x − x1 ) hψ(x1 )ψ̄(x2 )i + δ(x − x2 ) hψ(x1 )ψ̄(x2 )i (3.105)

that diagrammatically reads

x1
µ
∂µ J x1 x2 x1 x2
= −δ(x − x1 ) + δ(x − x2 )
x
x2

The strict analogue for a non-Abelian theory, with the current j µ = ψ̄γ µ ψ, holds exactly in
the same manner. However, this is not relevant to the calculation of the beta function of
the theory since the current appearing in the fermion-fermion-gauge interaction vertex is now
j aµ = ψ̄ta γ µ ψ (i.e. the δ1 we need for the beta function is different).
The non-Abelian Ward identities leads to the modified relations (3.103), which are also valid
to all orders in perturbation theory. There we deduce in particular that δ2 − δ1 = δ2c − δ1c
is independent from the number of fermions, and it is something that depends only from the
structure of the gauge group.

16
3.6.1 Gauge-boson self-energy: contributions to δ3
At one-loop level

= +

+ + + tadpoles

It is easy to show that the tadpoles contributions automatically vanish. In fact

Z
= −ie d4 xe−iqx h0|T j µ (x)|0i (3.106)

and the vacuum expectation value of j µ vanishes by Lorentz symmetry.

For the fermion loop diagram everything works as in the QED case, except for the gauge
group generators, resulting in an additional trace factor:

(−g 2 ) 1
Γ(2 − d2 )
Z
a b 2 µν µ ν
= tr[t t ]i(q g −q q ) d dx8x(1 − x) d
(4π) 2 0 (m2 − x(1 − x)q 2 )2− 2
(3.107)
Since tr[ta tb ] = C(r)δ ab we find

X  −g 2 4
  
2 µν µ ν ab d
= i(q g − q q )δ 2
nf C(r)Γ(2 − ) + · · · (3.108)
fermions
(4π) 3 2

where nf is the number of (light) fermions. Notice also that we have a dependence on the choice
of representation. Both the choice of number of fermions and the choice of representation are
going to affect the number of fermionic degrees of freedom, in fact changing either nf or the
representation corresponds to changing the specific type of non-Abelian theory we are looking
at. This contribution is the only term of the beta function that depends on the number of
fermions (and their representation). Simply looking at the color factors, it is easy to understand
that gauge and ghost diagrams in δ3 are proportional to C2 (A). Moreover, as we already
discussed, δ2 − δ1 is also independent from the number of fermions.
Proceeding a similar way we can compute all the other contributions to δ3 . The four-gauge
boson diagram vanishes in d = 4. Combining the other three pure gauge diagrams (that will
be computed in detail in the exercise class) leads to
 2 
2 µν µ ν ab g 5 d
i(q g − q q )δ C2 (A)Γ(2 − ) + · · · . (3.109)
(4π)2 3 2
Taking into account the definition of the counterterm

µ ν
= −i(q 2 g µν − q µ q ν )δ ab δ3 , (3.110)

17
and putting all terms together one finally obtains

g 2 Γ(2 − d2 ) 5
 
4
δ3 = C2 (A) − nf C(r) . (3.111)
(4π)2 (µ2 )2−d/2 3 3

3.6.2 Fermion self-energy: contribution to δ2

In the limit of massless fermions, the divergence of the fermion self-energy

is substracted by the following counterterm

= ip/δ2 . (3.112)

The calculation proceeds in close analogy to the QED case, leading to

g 2 Γ(2 − d2 )
δ2 = − C2 (r) (3.113)
(4π)2 (µ2 )2−d/2

3.6.3 Vertex renormalization: contribution to δ1

The diagrams whose divergences are substracted by δ1 are,

and the corresponding definition of the counterterm is

= igta γ µ δ1 . (3.114)

The explicit calculation yields

g 2 Γ(2 − d2 )
δ1 = − [C2 (r) + C2 (A)] . (3.115)
(4π)2 (µ2 )2−d/2

Finally, putting all ingredients of β(g) together and setting d = 4 leads to

g3
 
11 4
β(g) = − C2 (A) − nf C(r) . (3.116)
(4π)2 3 3

18
3.7 Running coupling in QCD
Let’s finally discuss the consequences of the beta function of non-Abelian gauge theories in
the specific case of QCD, where the gauge group is SU (3), and we have nf quarks in the
fundamental representations. In this case the one-loop beta function can be written as
b0 g 3 2
β(g) = − , b0 = 11 − nf > 0 . (3.117)
16π 2 3
From the Callan-Symanzik equation, we can solve the expression for the running coupling
p
dḡ(t) q2
= β(ḡ) , t = log( ), (3.118)
dt µ
with initial condition ḡ(0) = g, finding
Z t Z ḡ(t)
dg 0 16π 2 1 8π 2 1
   
1 1
t= dt = 0
= − = − (3.119)
0 ḡ(0) β(g ) 2b0 ḡ 2 g 2 b0 ḡ 2 g 2
p
1 1 b0 1 1 b0 q2
− = t =⇒ = + log( ). (3.120)
ḡ 2 g 2 8π 2 ᾱs (t) αs (µ2 ) 2π µ
g2
where we have substituted αs = . This leads us to the expression
4π
αs (µ2 )
ᾱs (t) ≡ αs (q 2 ) = 2 . (3.121)
b0
1 + αs (µ2 ) 4π log( µq 2 )

We can introduce a scale ΛQCD in order to removes from this expression the dependence from
the arbitrary renormalization scale µ2 . The value of ΛQCD is the q 2 value where the expression
(3.121) blows up:
 2 
2 b0
ΛQCD 2
4π
2 − b0 αs (µ2 )
αs (µ ) log = −1 =⇒ Λ QCD = µ e (3.122)
4π µ2
Using this result, we can rewrite (3.121) as
4π
αs (q 2 ) =   . (3.123)
q2
b0 log Λ2QCD

It is easy to check that ΛQCD is µ independent (as we already know from the general discussion
in Chapter 2):
16π 2 d −2 16π 2
   
d 2 2 −3
µ ΛQCD = 2 − µ (g ) ΛQCD = 2 − (−2)g β Λ2QCD = 0 (3.124)
dµ b0 dµ b0
b0 g 3
where we have used β(g) = − .
16π 2
While the expression (3.123) is not trustable when αs /(4π) ∼ 1, given perturbation theory
breaks down in that regime, it is an accurate description of the evolution of αs (q 2 ) at large
values of q 2 , where higher-order corrections are small. From this expression we deduce the
phenomenon of asymptoptic freedom in QCD, i.e. the fact the theory tends to become a free
theory in the deep UV.

19