Chapter 2

QCD

Quantum Chromodynamics (QCD) is the theory of the strong interaction. It describes

the ‘color force’ that binds quarks and gluons to colorless hadrons (protons, neutrons,
pions, etc.) and hadrons to nuclei. At hadronic scales, the strong force is ∼ 100 times
stronger than the electromagnetic interaction, extremely short-ranged (the typical in-
teraction range is the size of a hadron ∼ 1 fm = 10−15 m), and its typical energy scale
is the mass of the proton ∼ 1 GeV.
The strong interaction is described by a local, non-Abelian SU (3)C gauge theory
with several peculiar features. While quarks and gluons are asymptotically free at short
distances, they are confined at large distances: only colorless bound states (hadrons)
can be detected in experiments, and no quark or gluon has ever been observed directly.
Nevertheless, nature has given us an abundance of evidence that these constituents ex-
ist, and their theoretical description in terms of a non-Abelian gauge theory has evolved
from being considered a mere mathematical trick to a quite fundamental framework.
In this chapter we will recapitulate the properties of QCD and its fundamental degrees
of freedom and postpone the discussion of hadrons to Chapter 3.

2.1 QCD Lagrangian

Field content. The definition of a quantum field
R theory starts with constructing its
Lagrangian L (or, equivalently, its action S = d4 x L), based on the desired underly-
ing symmetries. The symmetries of QCD are: Poincaré invariance, local color gauge
invariance and various flavor symmetries, and the fields in the Lagrangian should trans-
form under representations of these groups. The QCD Lagrangian contains quark and
antiquark fields, and (as a consequence of color gauge invariance) gluon fields which
mediate the strong interaction:

ψα,i,f (x) , ψ α,i,f (x) , Aµa (x) . (2.1.1)

The quark fields are Dirac spinors (index α) and transform under the fundamental
representation of SU (3)C (color index i = 1, 2, 3 or red, green blue). The additional
index f = 1 . . . Nf labels the flavor quantum number (f = up, down, strange, charm,
bottom, top). The eight gluon fields Aµa (x) are Lorentz vectors; there is one field
2 QCD

for each generator ta of the group (a = 1 . . . 8). In the fundamental representation:

ta = λa /2, where the λa are the eight Gell-Mann matrices; see Appendix A for a
collection of basic SU (N ) relations. Gluons are flavor-blind and carry no flavor labels.

Gauge invariance. A free fermion Lagrangian ψ (i/ ∂ − m) ψ constructed from the

quark and antiquark fields (we leave the summation over Dirac, color and flavor indices
implicit) is invariant under global SU (3)C transformations
P
ψ 0 (x) = U ψ(x), ψ 0 (x) = ψ(x) U † with U = eiε = ei a εa ta , (2.1.2)

where εa = const. and the Uij act on the color indices of the quarks. This invariance
is no longer satisfied if we impose a local SU (3)C gauge symmetry ψ 0 (x) = U (x) ψ(x)
with spacetime-dependent group parameters εa (x). The mass term is still invariant,
but the derivative in the kinetic term now also acts on the spacetime argument of U (x),
and invariance of the Lagrangian (or the action) cannot be satisfied with an ordinary
partial derivative. To ensure local color gauge invariance, we introduce a covariant
derivative and thus gluon fields:

Dµ = ∂µ − igAµ , (2.1.3)
P
where Aµ (x) = Aµa (x) ta is an element of the Lie algebra. From the new Lagrangian
/ ψ must transform in the same way as the quark field itself,
/ − m) ψ we see that D
ψ (iD
which fixes the transformation properties of the gluon fields:
!
ψ0D
/ 0ψ0 = ψ D
/ψ ⇒ Dµ0 ψ 0 = U Dµ ψ = U Dµ U † ψ 0 (2.1.4)
⇒ (∂µ − igA0µ ) ψ 0 = U (∂µ − igAµ ) U † ψ 0
i
⇒ A0µ = U Aµ U † + U (∂µ U † ). (2.1.5)
g

The second term in A0µ is particular to local gauge transformations; for a global sym-
metry we don’t need a covariant derivative and could simply set Aµ = 0. Note also that
we can generate gluon fields out of nothing (Aµ = 0) by a local gauge transformation:
such gauge fields ∼ U (∂µ U † ) are called pure gauge configurations.

Why do we actually impose local gauge invariance in the first place? In fact, only global symmetries
are true ‘symmetries’ which lead to conserved charges and quantum numbers. A local gauge symmetry
reflects a redundancy in the description, which can be seen if we turn the argument around and
start from Eq. (2.1.5), for example in the Abelian case where U (x) = eiε(x) is just a phase. The
action of a free massless vector field contains redundant degrees of freedom which are related to each
other by local gauge transformations A0µ = Aµ + ∂µ ε/g. The standard way to eliminate them is to
modify the Lagrangian and impose a gauge-fixing condition on the state space (cf. Sec. 2.2.3). As a
consequence, longitudinal and timelike photons decouple from physical processes and S-matrix elements
are transverse: qµ Mµ = 0. To preserve this feature when including interactions (e.g., when adding
fermions), the interacting part of the action must couple to a conserved current corresponding to
the global symmetry of the full action, δSint /δAµ = j µ , which is equivalent to imposing local gauge
invariance for the matter fields. Thus, Eq. (2.1.5) is tied to the invariance under ψ 0 (x) = U (x) ψ(x), and
even though we needed an underlying global symmetry in the fermion sector to begin with, the local
gauge invariance is not truly a symmetry but rather a consistency constraint that generates dynamics.
In QCD, it introduces a quark-gluon interaction of the form g ψ A / ψ.
2.1 QCD Lagrangian 3

Another way to motivate the covariant derivative is the following. We can write the ordinary
derivative as
1
nµ ∂µ ψ(x) = lim [ψ(x + n) − ψ(x)] . (2.1.6)
→0 
0
For a local gauge transformation ψ (x) = U (x)ψ(x) the first term becomes U (x + n)ψ(x + n) but
the second U (x)ψ(x), so we are comparing objects at different spacetime points. To remedy this, we
define the parallel transporter or link variable C(y, x) by

C 0 (y, x) = U (y) C(y, x) U † (x) , C(x, x) = 1 , (2.1.7)

because then the quantity C(y, x) ψ(x) has a simple transformation behavior:

[C(y, x) ψ(x)]0 = U (y) C(y, x) U † (x) U (x) ψ(x) = U (y) C(y, x) ψ(x) . (2.1.8)

Now, if we define the covariant derivative as

1
nµ Dµ ψ(x) = lim [ψ(x + n) − C(x + n, x) ψ(x)] (2.1.9)
→0 
and perform a gauge transformation, then U (x+n) can be pulled out so that [Dµ ψ(x)]0 = U (x)Dµ ψ(x),
/ ψ is invariant under the local symmetry.
and thus ψ D
Moreover, we can write down the Taylor expansion of the parallel transporter:

C(x + n, x) = 1 + nµ igAµ (x) + O(2 ) , (2.1.10)

where igAµ is just a name for the coefficient of the linear term. Inserting this into (2.1.9) yields
1
nµ Dµ ψ(x) = lim [ψ(x + n) − ψ(x) − nµ igAµ ψ(x)] = nµ ∂µ ψ(x) − nµ igAµ ψ(x) (2.1.11)
→0 
and therefore Dµ = ∂µ − igAµ . Similarly, the transformation of the gluon field follows from

C 0 (x + n, x) = U (x + n) C(x + n, x) U † (x)

= U (x) + nµ ∂µ U (x) + O(2 ) 1 + nν igAν (x) + O(2 ) U † (x)

  

i
 (2.1.12)
= 1 + nµ ig U (x)Aµ (x) U † (x) − (∂µ U (x)) U † (x) + O(2 )
g
!
= 1 + nµ igA0µ (x) + O(ε2 ) ,

which reproduces the result (2.1.5) since ∂µ (U U † ) = (∂µ U ) U † + U ∂µ U † = 0.

Gluon dynamics. Next, we need a kinetic term that describes the dynamics of the
gluons. To this end we define the gluon field strength tensor as the commutator of
two covariant derivatives:
i
Fµν (x) = [Dµ , Dν ] = ∂µ Aν − ∂ν Aµ − ig [Aµ , Aν ] . (2.1.13)
g

It is then also an element of the Lie algebra and we can write it as

X
a
Fµν = Fµν ta , (2.1.14)
a

where the ta are again taken in the fundamental representation because ∂µ , Dµ and Aµ
act on quark fields in the (three-dimensional) fundamental representation of SU (3)C .
0 = U F U †.
Fµν inherits the transformation properties from (2.1.4): Fµν µν
4 QCD

-1 -1

Fig. 2.1: Tree-level (inverse) propagators and interactions in the QCD action.

The contraction of two field-strength tensors is not gauge invariant; only its color
trace is invariant due to the cyclic property of the trace:
 0 0 µν n o
Tr Fµν F = Tr U Fµν U † U F µν U † = Tr {Fµν F µν } . (2.1.15)

Only the trace can therefore appear in the Lagrangian. We can write it as

Tr {Fµν F µν } = Fµν
a
Fbµν Tr {ta tb } = T (R) Fµν
a
Faµν , (2.1.16)

where T (R) = 1/2 in the fundamental representation of SU (N ), cf. Appendix A. From

Eq. (2.1.5) we also conclude that a gluon mass term ∼ mg Aµ Aµ cannot appear in the
Lagrangian because it would violate gauge invariance: gluons must be massless.
We can work out the components of the field-strength tensor as
a
Fµν = Fµν ta = ∂µ Aaν ta − ∂ν Aaµ ta − ig Aaµ Abν [ta , tb ]

(2.1.17)
= ∂µ Aaν − ∂ν Aaµ + gfabc Abµ Acν ta ,

where we used [ta , tb ] = ifabc tc . Note that in an Abelian gauge theory such as QED this
commutator would vanish, leaving only the linear terms in the gluon fields. The non-
Abelian nature of SU (3)C introduces gluonic self-interactions which lead to significant
complications. Inserting Eq. (2.1.17) into the term Fµν a F µν and partial integration
a
yields
1
a
− 41 Fµν Faµν ∼
= Aaµ (2 g µν − ∂ µ ∂ ν ) Aaν
2
(2.1.18)
g g2
− fabc (∂ µ Aνa − ∂ ν Aµa ) Abµ Acν − fabe fcde Aµa Aνb Acµ Adν ,
2 4
where ∼= means ‘up to surface terms in the action’, e.g. ∂µ Aaν ∂ µ Aνa ∼ = −Aaν 2 Aνa after
2
partial integration. In contrast to the Abelian theory, where the F term only produces
a photon propagator, we can see that in the non-Abelian case we end up with the gluon
propagator, a three-gluon interaction ∼ A3 and a four-gluon interaction ∼ A4 .

Feynman rules. The terms ψ (iD a F µν in the Lagrangian allow us

/ − m) ψ and − 14 Fµν a
to read off the Feynman rules for the tree-level correlation functions of the QFT. In
particular, the action contains the 1PI (one-particle irreducible, see Sec. 2.2.2) quanti-
ties, which means the vertices and inverse propagators that define the theory (Fig. 2.1).
The procedure is as follows: symmetrize the respective term in the action (if necessary),
transform it to momentum space, split off the integrals, fields and symmetry factors,
and multiply with i to get the Feynman rule for the propagator or vertex.
2.1 QCD Lagrangian 5

For example, the inverse quark propagator corresponds to the term ψ (i∂/ − m) ψ.
The Fourier transform of the fields is
Z 4
d p −ip·x
ψ(x) = e ψ(p) . (2.1.19)
(2π)4
R R
Abbreviating p = d4 p/(2π)4 , the term in the action becomes
Z ZZ Z Z
i(p0 −p)·x
d4 x ψ (i∂/ − m) ψ = ψ(p0 ) (p
/ − m) ψ(p) d4
x e = / − m) ψ(p)
ψ(p) (p
p0 p p

and dividing by i, the inverse tree-level propagator is

i (p
/ + m)
S0−1 (p) = −i (p
/ − m) ⇔ S0 (p) = . (2.1.20)
p2 − m2 + i
Likewise, the inverse gluon propagator can be read off from Eq. (2.1.18). Replac-
ing 2 → −p2 and ∂ µ ∂ ν → −pµ pν , we find
 
−1 µν 2 µν pµ pν
(D0 ) (p) = ip g − 2 . (2.1.21)
p
The symmetry factor 1/2 does not enter in the Feynman rule. Here we encounter,
however, a difficulty: the inverse gluon propagator is proportional to a transverse
projector, which is not invertible and thus the gluon propagator does not exist. We
will cure the problem in Sec. 2.2.3 by the Faddeev-Popov method, where we follow
analogous steps as in QED and add gauge-fixing terms to the action (which will also
introduce ghost fields). Before we get there, keep in mind that the gluon propagator is
not yet well-defined.
The quark-gluon vertex comes from the term g ψ A / ψ induced by the covariant
derivative. If we denote the incoming and outgoing quark momenta by p and p0 and
the incoming gluon momentum by q, we have
Z XZ Z Z
4
d x ψ gA/ψ = (2π)4 δ 4 (p0 − p − q) ψ(p0 ) Aaµ (q) gγ µ ta ψ(p) ,
a
p0 p q

so the tree-level vertex is igγ µ ta .

The three-gluon vertex must be fully symmetric under exchange of any two legs,
but this symmetry is not yet manifest in the A3 term of Eq. (2.1.18). To this end, we
abbreviate ∂ µνρ = ∂ µ g νρ − ∂ ν g µρ and write

fabc (∂ µ Aνa − ∂ ν Aµa ) Abµ Acν = fabc (∂ µνρ Aaρ ) Abµ Acν = fabc Aaµ Abν (∂ µνρ Acρ ) = . . .

In the last step we renamed the color indices and used fabc = fbca = fcab . For three
Lorentz indices there are 3! = 6 possible permutations; ∂ µνρ is already antisymmetric
in µ ↔ ν so we only need to add the two remaining cyclic permutations:
1 h i
. . . = fabc Aaµ Abν (∂ µνρ Acρ ) + Aaν Abρ (∂ νρµ Acµ ) + Aaρ Abµ (∂ ρµν Acν )
3 (2.1.22)
1 h i
= fabc Aaµ Abν (∂ µνρ Acρ ) + (∂ νρµ Aaµ ) Abν Acρ + Aaµ (∂ ρµν Abν ) Acρ .
3
6 QCD

In the first line we renamed the Lorentz indices and in the second line the color indices.
Now we can pull out Aaµ (p1 ) Abν (p2 ) Acρ (p3 ) in momentum space and the term in the
action becomes
Z Z Z
ig
− fabc (2π)4 δ 4 (p1 + p2 + p3 ) Aaµ (p1 ) Abν (p2 ) Acρ (p3 )
6
p1 p2 p3 (2.1.23)
h i
ρ µν µ νρ ν ρµ
× (p1 − p2 ) g + (p2 − p3 ) g + (p3 − p1 ) g ,
from where we read off the Feynman rule for the vertex:
h i
Γµνρ
3g,0 = gfabc (p1 − p 2 )ρ µν
g + (p 2 − p 3 )µ νρ
g + (p 3 − p 1 ) ν ρµ
g . (2.1.24)
The symmetry factor 1/6 does again not enter, and p1 +p2 +p3 = 0. The resulting vertex
is Bose-symmetric, i.e., symmetric under a combined exchange of any two momenta
with corresponding Lorentz and color indices.
The same strategy applies to the four-gluon vertex from the A4 term in (2.1.18),
which is also not yet manifestly symmetric:
fabe fcde Aaµ Abν Aµc Aνd = fabe fcde g µρ g νσ Aaµ Abν Acρ Adσ
1 (2.1.25)
= fabe fcde (g µρ g νσ − g νρ g µσ ) Aaµ Abν Acρ Adσ .
2
Denoting Γ µνρσ = g g − g g µσ , then with four Lorentz indices there are 4! = 24
µρ νσ νρ

possible permutations of (µνρσ) ≡ (1234):

1234 3412 2314 1423 3124 2431
1243 3421 2341 1432 3142 2413
. (2.1.26)
2134 4312 3214 4123 1324 4231
2143 4321 3241 4132 1342 4213
The permutations in the first two columns are already covered because Γµνρσ = −Γµνσρ ,
etc., so we only need to add (2314) and (3124):

1
a b µ ν
fabe fcde Aµ Aν Ac Ad = fabe fcde (g µρ g νσ − g νρ g µσ ) Aaµ Abν Acρ Adσ
6
+ (g νµ g ρσ − g ρµ g νσ ) Aaν Abρ Acµ Adσ

ρν µσ µν ρσ a b c d
+ (g g − g g ) Aρ Aµ Aν Aσ
 (2.1.27)
1
= Aaµ Abν Acρ Adσ fabe fcde (g µρ g νσ − g νρ g µσ )
6
+ fbce fade (g νµ g ρσ − g ρµ g νσ )

ρν µσ µν ρσ
+ fcae fbde (g g − g g ) .

Together with −g 2 /4 from (2.1.18), the combined symmetry factor for the A4 term is
indeed 1/24. The resulting four-gluon vertex is Bose-symmetric and given by
h
Γµνρσ
4g,0 = −ig 2
fabe fcde (g µρ g νσ − g νρ g µσ )
+face fbde (g µν g ρσ − g νρ g µσ ) (2.1.28)
i
+fade fcbe (g µρ g νσ − g µν g ρσ ) .
2.1 QCD Lagrangian 7

QCD action. Putting everything together, the resulting QCD action constructed
from the fields ψ, ψ and Aµa has the most general form that is invariant under Poincaré
transformations, invariant under local gauge transformations, and renormalizable:
Z
SQCD = d4 x LQCD , LQCD = ψ(x) (iD a
/ − M) ψ(x) − 14 Fµν Faµν . (2.1.29)

The summation over the Dirac, color and flavor indices of the quarks is again implicit,
and we generalized the quark mass m to a quark mass matrix M = diag (m1 . . . mNf ).
Some further remarks:

 Eq. (2.1.29) also conserves charge conjugation and parity, where the charge

conjugation operation is defined by

ψα0 = ψ β Cβα , ψ 0α = Cαβ ψβ , C = iγ 2 γ 0 (2.1.30)

and the parity transformation by

ψ 0 (x0 ) = γ 0 ψ(x) , ψ 0 (x0 ) = ψ(x) γ 0 , x0 = (t, −x) . (2.1.31)

Since CP T is always conserved, this implies that the QCD action is also invariant
under time reversal.

 In principle, another gauge-invariant and renormalizable (but parity-violating)

term could appear in the Lagrangian, namely a topological charge density:

g2 n o 1
Q(x) = Tr Fµν
eµν
F with Feµν = εµναβ Fαβ , (2.1.32)
8π 2 2

where Feµν is the dual field strength tensor. The resulting ‘θ term’ in the Lagrangian
L + θQ(x) violates parity and would give rise to an electric dipole moment of the
neutron, whose experimental upper limit is however tiny (θ ≤ 10−10 ). So it would
seem that QCD does conserve parity; unfortunately, even if we started with θ = 0 in
QCD, the CP -violating weak interactions would renormalize it to θ 6= 0. There are
several possible scenarios how θ = 0 could be enforced beyond the Standard Model,
e.g. by promoting θ to a field (axions). Then again, CP must have been violated in
the early universe, because otherwise the Big Bang would have created matter and
antimatter in equal portions, which would have annihilated and resulted in a radiation
universe without matter. This leads to the strong CP problem. On the other hand,
since Q = ∂µ K µ can be written as the divergence of the Chern-Simons current K µ ,
it only contributes a surface term to the action and in principle we could discard it
(unless topological gauge field configurations play a role).

 We could have defined the gluon fields so that they absorb the coupling constant
g (i.e., by replacing A → A/g and F → F/g). From Eqs. (2.1.13), (2.1.18) and (2.1.29)
we see that the only place in the Lagrangian where the coupling would then appear is
in front of the gluon kinetic term, as a prefactor 1/g 2 . This shows that the sign of g is
physically irrelevant.
8 QCD

Quark masses and flavor structure. With regard to the flavor structure, we can
simply ignore the gluons since they are flavor independent. The quark-gluon interaction
is flavor-blind, and the distinction between different quarks only comes from their
masses. If the masses of all quark flavors were equal, the Lagrangian would have an
additional SU (Nf ) flavor symmetry. This is not realized in nature, where

mc ∼ 1.3 GeV,
mu ∼ md ∼ 2 . . . 6 MeV, ms ∼ 100 MeV, mb ∼ 4.2 GeV, (2.1.33)
mt ∼ 173 GeV.

These current-quark masses have their origin in the Higgs sector and from the
point of view of QCD they are external parameters that enter through the quark mass
matrix M = diag(m1 . . . mNf ). Because M is diagonal in flavor space, the flavor pieces
P
in the Lagrangian simply add up: ψ M ψ = f mf ψ f ψf . The flavor structure of the
Lagrangian is crucial for the properties of hadrons and we will return to it in Chapter 3.

Infinitesimal gauge transformations. For later convenience it is useful to work

out the infinitesimal transformations of the fields. The covariant derivative as defined
in Eq. (2.1.3) acts on fields that transform under the fundamental representations of
SU (3)C , i.e., the group elements. When acting on elements of the algebra (those
containing the matrix generators ta , for example ε, Aµ or Fµν ), we need an additional
commutator in its definition: Dµ = ∂µ − ig [Aµ , · ], or written in components:

(Dµ ε)a = (∂µ ε − ig [Aµ , ε])a = ∂µ εa − ig Acµ εb ifcba

(2.1.34)
= (∂µ δab − gfabc Acµ ) εb = Dµab εb .

In the fundamental representation, the group generators are the Gell-Mann matrices;
in the adjoint representation they are given by (tc )ab = −ifabc . Inserting this into
Eq. (2.1.3), we see that Dµab is the covariant derivative in the adjoint representation:

(Dµ )ab = (∂µ − igAµ )ab = ∂µ δab − igAcµ (tc )ab = ∂µ δab − gfabc Acµ . (2.1.35)

In an Abelian gauge theory such as QED, the commutator vanishes and Dµab = ∂µ δab
is the ordinary partial derivative.
With U = 1 + iε, the infinitesimal gauge transformation of the fields is given by

ψ 0 = U ψ ≈ (1 + iε) ψ ,
ψ 0 = ψ U † ≈ ψ (1 − iε) ,
i 1 1 (2.1.36)
A0µ = U Aµ U † + U (∂µ U † ) ≈ Aµ + i [ε, Aµ ] + ∂µ ε = Aµ + Dµ ε ,
g g g
0 †
Fµν = U Fµν U ≈ Fµν + i [ε, Fµν ] ,

from where we obtain:

1
δψ = iε ψ , δψ = −iψ ε , δAµ = Dµ ε , δFµν = i [ε, Fµν ] . (2.1.37)
g
2.1 QCD Lagrangian 9

Classical equations of motion. The classical Euler-Lagrange equations of motion

follow from the action principle:
Z Z  
4 4 ∂L ∂L
S[φ] = d x L(φ, ∂µ φ) ⇒ δS[φ] = d x δφ(x) + δ(∂µ φ)
∂φ ∂(∂µ φ)
Z  
4 ∂L ∂L
= d x − ∂µ δφ(x) = 0 ,
∂φ ∂(∂µ φ)
which means that the functional derivative of the action vanishes:
δS[φ] ∂L ∂L
= − ∂µ = 0. (2.1.38)
δφ(x) ∂φ ∂(∂µ φ)
If the action contains several fields, there is one equation of motion for each component:
δS[φ1 , . . . φn ]/δφi (x) = 0.

As a reminder, the functional derivative δF [φ]/δφ(x) of a functional F [φ] is defined as

Z∞
δF [φ]
F [φ + δφ] = F [φ] + δF [φ] = F [φ] + dx δφ(x) , (2.1.39)
δφ(x)
−∞
P
where the last term is the continuum version of i (δF/δφi ) δφi written for one spacetime dimension.
Here are some examples:

δF [φ]
F [φ] F [φ + δφ]
δφ(x)
R R R
dx φ(x) J(x) dx (φ + δφ) J = F [φ] + dx J δφ J(x)

dx [f (φ) + f 0 (φ) δφ] J f 0 (φ(x)) J(x)

R R
dx f (φ(x)) J(x)
h i
dx f (φ, φ0 ) + ∂φ
∂f ∂f 0
R
R 0
δφ + ∂φ 0 δφ ∂f d ∂f
dx f (φ(x), φ (x)) h i −
∂f d ∂f ∂φ(x) dx ∂φ0 (x)
R
= F [φ] + dx ∂φ − dx ∂φ0
δφ

R∞ R∞ R
dx φ(x) = dx φ(x) Θ(x) F [φ] + dx Θ(x) δφ(x) Θ(x)
0 −∞

R 
R  exp i dx (φ(x) + δφ(x)) J(x)  R 
exp i dx φ(x) J(x) R
iJ(x) exp i dy φ(y) J(y)
= F [φ] 1 + i dx J(x) δφ(x)

R R
φ(z) = dx φ(x) δ(x − z) F [φ] + dx δφ(x) δ(x − z) δ(x − z)

f (φ(z)) + f 0 (φ(z)) δφ(z)

f (φ(z)) f 0 (φ(x)) δ(x − z)
= F [φ] + dx f 0 (φ(x)) δφ(x) δ(x − z)
R

φ0 (z) = dx φ0 (x) δ(x − z)

R
dx δφ(x) δ 0 (x − z) −δ 0 (x − z)
R
R 0
F [φ] −
=− dx φ(x) δ (x − z)
10 QCD

Let us work out the

classicala equations of motion of QCD defined by the Lagrangian
L = ψ i∂/ + g A/ − M ψ − 14 Fµν Faµν . Although they are not directly relevant for our
purposes, they will later enter in the quantum equations of motion (Sec. 2.2.2) and
conservation laws (Sec. 3.1). If we take the derivatives of L with respect to ψ and ψ,
∂L ∂L ∂L ∂L
/ − M) ψ ,
= (iD = 0, / − M) ,
= ψ (g A = ψ iγ µ (2.1.40)
∂ψ ∂(∂µ ψ) ∂ψ ∂(∂µ ψ)

we obtain the Dirac equations for the quark and antiquark fields:
δS δS ←−
/ − M) ψ = 0 ,
= (iD = ψ (−i ∂/ + g A
/ − M) = 0 . (2.1.41)
δψ δψ

For the gluons, we first work out the derivatives of the field-strength tensor:
a
∂Fµν   ∂Fµνa

b ρ b ρ ρ σ ρ σ
= gfabc A µ δ ν − Aν δµ , = δ µ δ ν − δ ν δµ δac . (2.1.42)
∂Acρ ∂(∂ρ Acσ )

With the product rule we then obtain

∂L ∂L
= g ψ γ µ ta ψ + gfabc Acν Fbµν , = Faµν (2.1.43)
∂Aaµ ∂(∂ν Aaµ )

and finally
δS
= gfabc Acν Fbµν − ∂ν Faµν + g ψ γ µ ta ψ = 0 . (2.1.44)
δAaν
The first two terms on the r.h.s. can be combined to

− (∂ν δab − gfabc Acν ) Fbµν = −Dνab Fbµν = −(Dν F µν )a , (2.1.45)

whereas the last term is the vector current corresponding

P to the global SU (3)C trans-
formation: Jaµ = ψ γ µ ta ψ. Then the quantity J µ = a Jaµ ta lives in the Lie algebra
and Eq. (2.1.44) becomes
Dν F µν = g J µ . (2.1.46)
These are the classical Yang-Mills equations for the gluon field, i.e., the Maxwell
equations in the non-Abelian theory. They are the direct generalization from electro-
dynamics, where the covariant derivative in the adjoint representation would reduce to
the ordinary derivative.
It is not too much of a stretch to ask whether there is also a generalization of the
Maxwell equation for the dual field strength tensor Feµν . Indeed we find

Dν Feµν = εµναβ Dν Fαβ

1
2
 
= 16 εµναβ + εµαβν + εµβνα Dν Fαβ (2.1.47)
= 1
6 εµναβ (Dν Fαβ + Dα Fβν + Dβ Fνα ) = 0 ,

where the parenthesis vanishes due to the Bianchi identity, which follows from the
Jacobi identity for the generators, Eq. (A.1.3). Similarly, one can establish covariant
current conservation Dµ J µ = 0 for the solutions of the equations of motion.