An Introduction to QED & QCD

F Hautmann

Department of Theoretical Physics

University of Oxford
Oxford OX1 3NP

Lectures presented at the RAL High Energy Physics Summer School

Somerville College, Oxford, September 2012
These lectures present a heuristic introduction to gauge theories of electromagnetic and
strong interactions, focusing on perturbative applications of the S matrix and the use
of Feynman graphs in QED and QCD. They are complementary to the presentations of
field theory in the Standard Model and Quantum Field Theory courses at this School.
The approach followed in these lectures may be found in the textbooks given in Ref. [1].
Quantum field theory treatments may be found in the textbooks given in Ref. [2].
Secs. 1 and 2 discuss relativistic quantum mechanics and spin. Secs. 3 to 5 are devoted to
interactions and scattering processes at tree level in QED. Sec. 6 extends the discussion to
QCD. Secs. 7 and 8 consider loops and give an introductory discussion to renormalization.
1 Relativistic quantum mechanics
In this section we present the Klein-Gordon equation and the Dirac equation as they
arise from attempts to generalize quantum-mechanical wave equations to include rela-
tivity. We discuss the difficulties in interpreting these equations as single-particle wave
equations, and illustrate that the Feynman-Stueckelberg causality argument points to
the resolution of these difficulties by going beyond the single particle interpretation.

1.1 Relativistic wave equations

Let us start with the case of nonrelativistic quantum mechanics, and ask how we can
generalize it to the relativistic case.
The time evolution of the state |ψi of a quantum mechanical system is given by the
Schrödinger equation,
∂
i |ψi = H|ψi , (1.1)
∂t
where H is the hamiltonian operator corresponding to the total energy.
For a free, spinless, nonrelativistic particle we have
p2
H= , (1.2)
2m
where p is the momentum operator, and m is the particle’s mass. In the basis of
eigenstates of the position operator, p is represented by

p = −i ∇ , (1.3)

and therefore the evolution equation reads

∂ 1
i ψ(x, t) = − ∇2 ψ(x, t) , (1.4)
∂t 2m
where ψ(x, t) = hx|ψi is the position-space wave function.
The wave equation (1.4) admits a probabilistic interpretation. By taking the complex
conjugate ψ ∗ times Eq. (1.4) and subtracting ψ times the complex conjugate equation,
we obtain
∂ρ
+∇·j=0 , (1.5)
∂t
with
i
ρ = ψ∗ψ , j = − [ψ ∗ (∇ψ) − (∇ψ ∗ )ψ] . (1.6)
2m
Here ρ and j are interpreted as probability density and current, and the continuity
equation (1.5) expresses probability conservation.
How could we extend this to the relativistic case? To do this, we need to incorporate
the relativistic energy-momentum relation

E 2 = p2 + m2 . (1.7)

If we naively were to take q

H= p2 + m2 , (1.8)
this would yield the correct energy-momentum relation, but would give as a candidate
wave equation
∂ q
i ψ = p2 + m2 ψ , (1.9)
∂t
which contains space derivatives under the square root. This equation has a number of
difficulties, because it treats time and space derivatives on a different footing, contrary
to what one would expect of a relativistic theory, and because it is non-local in space,
as the square root gives rise to an infinite number of spatial derivatives.
One possible way to overcome this is to square the differential operators in Eq. (1.9)
before applying them to ψ. This gives
∂2
− ψ = (−∇2 + m2 ) ψ , (1.10)
∂t2
that is, using covariant notation with ∂ µ = (∂/∂t, −∇), ∂ 2 = ∂ µ ∂µ ,
(∂ 2 + m2 ) ψ = 0 . (1.11)
This is the approach originally proposed by Schrödinger, Klein and Gordon, and Eq. (1.11)
is referred to as the Klein-Gordon equation. This equation describes relativistic spin-0
particles. We will discuss Klein-Gordon in Subsec. 1.2. As we will see, this equation
is a candidate wave equation consistent with relativity, but it runs into problems with
quantum mechanics as a single-particle wave equation, because, due to the second-order
time derivative, it does not lead to a positive-definite probability density.
A second possible way around the naive Eq. (1.9) is to insist on the equation being
first-order in the time derivative but devise a new hamiltonian Hd which is local, linear in
momentum, and such that its square returns the correct relativistic energy-momentum
relation (1.7):
∂
i ψ = Hd ψ . (1.12)
∂t
This is the approach followed originally by Dirac. It turns out that this route is viable
only if the wave function is not one-component but multi-component (which implies
spin), and the new hamiltonian is of the form
Hd = α · p + βm , (1.13)
where α and β are four matrices, in a space to be determined, obeying the relations
αi αj + αj αi = 2δij , βαi + αi β = 0 , β2 = 1 . (1.14)
Eq. (1.12), with Hd given in Eqs. (1.13),(1.14), is the Dirac equation. This equation
describes relativistic spin-1/2 particles. We will discuss it in Subsec. 1.3.
We will see that, unlike the Klein-Gordon equation, the Dirac equation, being first-
order, allows one to construct a positive-definite probability density. However, we will
see that both the Klein-Gordon equation and the Dirac equation have solutions corre-
sponding to states with negative energies. In Subsec. 1.4 we discuss how this issue can be
addressed via the Feynman-Stueckelberg picture of causality. This picture reinterprets
negative energy states by introducing the concept of antiparticle, and leads us to think of
theories that incorporate quantum mechanics and relativity as theories for which particle
number is not conserved.
1.2 The Klein-Gordon equation
We have seen in the previous subsection that the Klein-Gordon equation emerges from
requiring the relativistic energy-momentum relation and taking the square of the hamil-
tonian operator in the position-space wave equation. In covariant form, the Klein-Gordon
equation is given by
(∂ 2 + m2 ) φ(x) = 0 , (1.15)
where
∂2
∂ 2 ≡ ∂ µ ∂µ =
− ∇2 . (1.16)
∂t2
The Klein-Gordon equation is relativistically covariant. That is, if we start with
Eq. (1.15) and make a Lorentz transformation,
xµ → x′µ = Λµν xν , Λµρ Λνσ gµν = g ρσ , (1.17)
φ(x) → φ′ (x) = φ(Λ−1 x) , (1.18)
in the primed coordinate system an equation of the same form holds, because
h i
(∂ 2 + m2 )φ′ (x) = (Λ−1 )ρµ ∂ρ (Λ−1 )σν ∂σ g µν + m2 φ(Λ−1 x)

= (∂ρ ∂σ g ρσ + m2 )φ(Λ−1 x) = (∂ 2 + m2 )φ(Λ−1 x) = 0 . (1.19)

On the other hand, interpreting the Klein-Gordon equation as a single particle rel-
ativistic wave equation leads to difficulties with quantum mechanics. As the equation
is second-order in the time derivative, the norm of φ is not conserved with time. We
can see the difficulty by looking for a continuity equation for Klein-Gordon similar to
Eq. (1.5) for the nonrelativistic case. Following the same steps as described for Eq. (1.5),
we obtain
∂µ j µ = 0 , j µ = (ρ, j) , (1.20)
with " #
∗ ∂φ∂φ∗
ρ=i φ − φ , j = −i [φ∗ (∇φ) − (∇φ∗ )φ] . (1.21)
∂t ∂t
The current j is formally similar to that of the nonrelativistic case in Eq. (1.6). The
density ρ, however, is not. Nor could it be, because φ∗ φ would transform under Lorentz
like a scalar rather than like the time component of a four-vector. Because Eq. (1.15)
contains second-order time derivatives, the density ρ contains terms in ∂/∂t, and is not
positive definite.
If we look for plane wave solutions of the Klein-Gordon equation,
φ(x) = N e−ipx , (1.22)
where pµ = (E, p), another difficulty arises. By substituting Eq. (1.22) into the equation,
we find that Eq. (1.22) is solution if
p2 = m2 , (1.23)
that is, q
E = ± p2 + m2 . (1.24)
The Klein-Gordon equation contains both positive-energy and negative-energy solutions.
Since j µ in Eq. (1.20) is proportional to pµ , negative energy solutions also have negative
probability density. We discuss in Subsec. 1.4 how to interpret negative-energy states.
1.3 The Dirac equation
We have seen in Eqs. (1.12),(1.13) that the Dirac equation has the form
∂
i ψ = (−iα · ∇ + βm) ψ . (1.25)
∂t
Consistency with the relativistic energy-momentum relation requires that by squaring
Eq. (1.25),
∂ 2ψ
− 2 = [−αi αj ∇i ∇j − i (βαi + αi β)m∇i + β 2 m2 ]ψ , (1.26)
∂t
we reobtain the Klein-Gordon equation,
∂ 2ψ
− = [−∇i ∇i + m2 ]ψ . (1.27)
∂t2
Then αi and β must obey anticommutation relations,

αi αj + αj αi = 2δ ij , βαi + αi β = 0 , β2 = 1 . (1.28)

Eq. (1.28) implies that

Tr αi = Tr β = 0 , (1.29)
and that the eigenvalues of αi and β are ±1. Then αi and β must be even-dimensional
matrices. In two dimensions there are no four matrices satisfying Eq. (1.28) (the three
Pauli σ matrices would be three candidate matrices but there is no fourth anticommuting
2 × 2 matrix). So the minimum possible dimension is four. Then ψ in Eq. (1.25) is a
four-component object, referred to as a four-component spinor.
A possible choice of αi and β is given by
   
0 σ 1 0
α= , β= , (1.30)
σ 0 0 −1
where we use block matrix notation.
The Dirac equation can be recast in manifestly covariant form by defining four new
matrices γ in terms of the α and β as

γ 0 = β, γ = βα, (1.31)

and noting that Eq. (1.25), multiplied by β, can be compactly rewritten in terms of the
γ matrices as
(iγ µ ∂µ − m)ψ = 0 , (1.32)
where
γ µ = (γ 0 , γ) . (1.33)
In terms of the γ matrices the anticommutation relations (1.28) become

{γ µ , γ ν } ≡ γ µ γ ν + γ ν γ µ = 2g µν . (1.34)

The representation (1.30) is

   
0 σ 1 0
γ= , γ0 = . (1.35)
−σ 0 0 −1
 Because the Dirac equation is first-order, it gives rise to a positive-definite density,
unlike the Klein-Gordon equation. By manipulations analogous to those seen in the
previous sections, we obtain the continuity equation

∂µ j µ = 0 , j µ = (ρ, j) , (1.36)

with
ρ = ψ † ψ, j = ψ † αψ. (1.37)
On the other hand, because α and β are traceless (Eq. (1.29)), the hamiltonian is
traceless. Then the eigenvalues must be E, −E. Thus the Dirac equation, like the
Klein-Gordon equation, has negative energy solutions.
A first interpretation of negative energy solutions is provided by Dirac’s “sea” picture.
Dirac postulates the existence of a “sea” of negative energy states (Fig. 1), such that
the vacuum has all the negative energy states filled with electrons. The Pauli principle
forbids any positive-energy electron from falling into one of the lower states. Although
the vacuum state has infinite negative charge and energy, this leads to an acceptable
theory based on the fact that all observations only involve differences in energy and
charge. When energy is supplied and one of the negative energy electrons is promoted to
a positive energy one, an electron-hole pair is created, i.e. a positive energy electron and a
hole in the negative energy sea. The hole, i.e. the absence of a negative-energy electron,
is seen as the presence of a positive-energy and positive-charge state, the positron.
This picture led Dirac to postulate (1927) the existence of the positron as the elec-
tron’s antiparticle, which was discovered experimentally five years later.

+m

−m

Figure 1: Dirac sea picture of negative energy states.

But Dirac’s sea picture does not work for bosons, which have no exclusion principle.
A second, more general interpretation of negative-energy states is given by Feynman’s
picture, which we describe in the next subsection.

1.4 The Feynman-Stueckelberg picture

The Feynman-Stueckelberg interpretation of negative-energy states does not appeal to
the exclusion principle but rather to a causality principle. It is based on the observation
that causality ensures that positive energy states, with time dependence e−iEt , propagate
forwards in time, and that if we impose that negative energy states propagate only
backwards in time, with
e−i(−|E|)(−|t|) → e−i|E||t| , (1.38)
we still obtain an acceptable theory, consistent with causality. In this picture the emission
of a negative energy particle with momentum pµ is interpreted as the absorption of a
positive energy antiparticle with momentum −pµ .
Consider for example photon-particle scattering (Fig. 2). In Fig. 2(a) a particle comes
in with energy E1 , and at time t1 and point x1 it emits a photon with energy Eγ < E1 .
It travels on forwards in time, and at time t2 and point x2 it absorbs the initial state
photon, giving rise to the photon-particle final state.
Another process is shown in Fig. 2(b). In this process the particle coming in with
energy E1 emits a photon with energy Eγ > E1 , and is thus forced to travel backwards
in time. Then at an earlier time it absorbs the initial state photon at the point x2 , which
renders its energy positive again.

x
x

E1 E
E γ < E1 1 Eγ > E
1

t t
( a) (b)

Figure 2: Feynman interpretation of positive-energy and negative-energy states in terms

of particle and antiparticle propagation.

The process in Fig. 2(b) can be described by saying that in the initial state we have
a particle and a photon, and that at point x2 the photon creates a particle-antiparticle
pair, both of which propagate forwards in time. The particle ends up in the final state,
whereas the antiparticle is annihilated at a later time by the initial state particle, thereby
producing the final state photon. According to this picture, the negative energy state
moving backwards in time is viewed as a negatively charged state with positive energy
moving forwards in time.
2 Spin
In the previous section we have introduced Dirac spinors, which we will use to describe
spin-1/2 relativistic charges. In this section we further the study of spin in the context
of Dirac theory.
We start from the algebraic description of the group of Lorentz transformations and
associated representations, which parallels the treatment of the group of rotations and
its representations in quantum mechanics. We describe how Dirac spinors emerge from
this point of view, and we discuss solutions of the Dirac equations.

2.1 Algebra of Lorentz transformations

Let us start with the group of rotations in three dimensions. In quantum mechanics,
given a particle with spin s, the matrices that rotate its n-component wave function,
where n = 2s + 1, are constructed from the angular momentum operators Ji , i = 1, 2, 3.
These satisfy the commutation relations

[Ji , Jj ] = iεijk Jk . (2.1)

Rotation operators are obtained by exponentiation as

R = e−iθi Ji , (2.2)

where the parameters θi specify the rotation axis and angle. The angular momentum
operators Ji are the generators of the rotation group, and have matrix representations
for every dimensionality n. The representation for n = 2, corresponding to spin s = 1/2,
is given by
1
Ji → σi , (2.3)
2
where σi are the three Pauli matrices, so that

R1/2 = e−iθi σi /2 . (2.4)

We can generalize this from the group of rotations to the group of Lorentz transfor-
mations. One way to obtain the commutation relations of the Lorentz generators is as
follows. In the case of the rotation group, the relations (2.1) can be obtained by taking

J = x ∧ p = −ix ∧ ∇ , (2.5)

and evaluating the commutators. The components of angular momentum in Eq. (2.5)
can also be rewritten explicitly, using antisymmetric tensor notation, as

J ij = −i(xi ∇j − xj ∇i ) . (2.6)

The generalization of this formula to the four-dimensional case,

Lµν = i(xµ ∂ ν − xν ∂ µ ) , (2.7)

gives the correct generators of Lorentz transformations. As it is antisymmetric in µ

and ν, Eq. (2.7) contains six operators, the three generators of rotations and three
generators of boosts. The commutation relations can be obtained by evaluating directly
the commutators of the operators (2.7), and the result is

[Lµν , Lρσ ] = i (gµσ Lνρ − gνσ Lµρ − gµρ Lνσ + gνρ Lµσ )
= i (gµσ Lνρ − {µ ↔ ν}) − {ρ ↔ σ} . (2.8)

Given the six operators in Eq. (2.7), it is helpful to distinguish the three operators
1
Ji = εijk Ljk (2.9)
2
and the three operators
Ki = L0i . (2.10)
From Eq. (2.8) we get
[Ji , Jj ] = iεijk Jk , (2.11)
that is, the three operators in Eq. (2.9) are the generators of rotations. The three
remaining operators in Eq. (2.10) are the generators of boosts. From Eq. (2.8) we have

[Ji , Kj ] = iεijk Kk , (2.12)

[Ki , Kj ] = −iεijk Jk . (2.13)

The J and K operators in Eqs. (2.9),(2.10) have the clear physical interpretation
of rotation and boost generators, but they are mixed by the algebra of commutation
relations (2.8), as shown in Eqs. (2.12),(2.13). It is useful to disentangle the algebra by
introducing the linear combinations of generators
1 1
Ai = (Ji + iKi ) , Bi = (Ji − iKi ) . (2.14)
2 2
By computing the commutators of the A and B operators in Eq. (2.14), we find

[Ai , Aj ] = iεijk Ak , [Bi , Bj ] = iεijk Bk , [Ai , Bj ] = 0 . (2.15)

That is, the A and the B operators do not mix, and each set obeys commutation relations
of the form (2.1). This means that we can specify a representation of the group of Lorentz
transformations by specifying a pair of rotation-group representations,

(a, b) , where Ai Ai = a(a + 1) , Bi Bi = b(b + 1) , (2.16)

with a and b integer or half-integer. Here a + b gives the spin quantum number: so the
representation (0, 0) is spin-0 (scalar); (1/2, 1/2) is spin-1 (vector); (1/2, 0) and (0, 1/2)
are spin-1/2. The latter are referred to as Weyl spinors (respectively, left-handed and
right-handed). A Dirac spinor is obtained from two Weyl spinors, (1/2, 0) ⊕ (0, 1/2).
Dirac spinors are thus identified by a representation of the group of Lorentz trans-
formations reducible to the sum of two representations (2.16), (1/2, 0) and (0, 1/2).
We discuss Weyl and Dirac spinors in the next two subsections.
2.2 Weyl spinors
A left-handed Weyl spinor is defined by taking a = 1/2, b = 0 in Eq. (2.16). We thus
have
1
Ai = σ i , Bi = 0 . (2.17)
2
Using Eq. (2.14), the representation of the rotation and boost generators is given by
1 i 1 i
Ji = σ (rotation generators) , K i = −i σ (boost generators) . (2.18)
2 2
So a left-handed Weyl spinor is a two-component spinor,
 
ξL1
ξL = , (2.19)
ξL2

which transforms under rotations and boosts as

k σ k /2−η k σ k /2
ξL → e−iθ ξL . (2.20)

A right-handed Weyl spinor is defined by taking a = 0, b = 1/2 in Eq. (2.16). Then

1 i
Ai = 0 , Bi = σ , (2.21)
2
i.e.,
1 i 1 i
Ji = σ (rotation generators) , Ki = i σ (boost generators) . (2.22)
2 2
So a right-handed Weyl spinor is a two-component spinor,
 
ξR1
ξR = , (2.23)
ξR2

which transforms under rotations and boosts as

k σ k /2+η k σ k /2
ξR → e−iθ ξR . (2.24)

Eq. (2.20) and Eq. (2.24) differ by the sign in the boost transformation.

2.3 Dirac spinors

A Dirac spinor is a four-component spinor built out of two Weyl spinors as
 
ξL
ψ= . (2.25)
ξR

We can construct explicitly its Lorentz transformation matrix S

ψ → ψ ′ = Sψ (2.26)
from those for the Weyl spinors in Sec. 2.2. We obtain
µν
ψ → ψ ′ = e−iωµν Σ ψ ≡ Sψ , (2.27)

where
i µ ν
Σµν ≡ [γ , γ ] (2.28)
4
with the γ matrix representation
   
0 σ 0 0 1
γ= , γ = . (2.29)
−σ 0 1 0
Thus the generators of boosts for Dirac spinors are
 
i 0 i i σi 0
Σ0i = [γ , γ ] = − (boost generators) , (2.30)
4 2 0 −σ i
and the generators of rotations are
 
i 1 σk 0
ij
Σ = [γ i , γ j ] = εijk (rotation generators) . (2.31)
4 2 0 σk
An alternative method, equivalent to that given above, for constructing Dirac spinors
is based on observing that if 4 n × n matrices γ µ satisfy

{γ µ , γ ν } ≡ γ µ γ ν + γ ν γ µ = 2g µν , (2.32)

then
i
Σµν ≡ [γ µ , γ ν ] (2.33)
4
obey the Lorentz algebra (2.8). Then, by a reasoning similar to that followed in Sec. 1.3,
one sees that n must be 4. By this method one arrives at Dirac spinors without going
through the construction from two Weyl spinors.
Note the following two properties of the Lorentz transformation matrix S for Dirac
spinors, which follow from Eqs. (2.27),(2.28). The first is that

S −1 γ µ S = Λµν γν . (2.34)

This relation implies that the Dirac equation Eq. (1.32),

(iγ µ ∂µ − m)ψ = 0 , (2.35)

is relativistically covariant, because under a Lorentz transformation we have

h i
(iγ µ ∂µ − m)ψ(x) → iγ µ (Λ−1 )ρµ ∂ρ − m Sψ(Λ−1 x)

= S(iγ ν ∂ν − m)ψ(Λ−1 x) = 0 . (2.36)

The second property is that, because boost generators (2.30) are not hermitian, S is not
unitary; rather, it satisfies
S † = γ 0 S −1 γ 0 . (2.37)
For this reason the product ψ † ψ is not Lorentz invariant. It is thus useful to define the
adjoint spinor
ψ ≡ ψ†γ 0 . (2.38)
Using Eq. (2.37), we see that the product ψψ is Lorentz invariant.
We have seen the transformation law of Dirac spinors under rotations and boosts.
Lorentz transformations also include discrete transformations, space parity and time
reversal. The transformation law of Dirac spinors under these are
P ψ(x, t)P −1 = ηγ 0 ψ(−x, t) , P ψ(x, t)P −1 = η ∗ ψ(−x, t)γ 0 , (2.39)
where η is a phase factor to be fixed, and
T ψ(x, t)T −1 = −γ 1 γ 3 ψ(x, −t) , T ψ(x, t)T −1 = ψ(x, −t)γ 1 γ 3 . (2.40)
A third discrete symmetry is charge conjugation, exchanging particle and antiparticle,
CψC −1 = −iγ 2 ψ ∗ , CψC −1 = −iψ T γ 2 γ 0 . (2.41)
Finally, it is useful to introduce a fifth γ matrix
γ 5 ≡ iγ 0 γ 1 γ 2 γ 3 , (2.42)
obeying
 2  †
γ5 =1 , {γ 5 , γ µ } = 0 , γ5 = γ5 . (2.43)
Then define the projection operators
1 − γ5 1 + γ5
PL = , PR = . (2.44)
2 2
In the representation (2.29) we have
     
−1 0 1 0 0 0
γ5 = =⇒ PL = , PR = . (2.45)
0 1 0 0 0 1
Thus
       
1 − γ5 ξL ξL 1 + γ5 ξL 0
ψL ≡ PL ψ = = , ψR ≡ PR ψ = = . (2.46)
2 ξR 0 2 ξR ξR
Note that, because γ 5 anticommutes with γ 0 , for the adjoint we have
ψ L = ψL† γ 0 = ψ † PL γ 0 = ψ † γ 0 PR = ψPR , (2.47)
and similarly
ψ R = ψPL . (2.48)
5
By including γ , it is possible to see that any combination of the form
ψΓψ , (2.49)
where Γ is any 4 × 4 matrix, can be decomposed into terms with definite transformation
properties under Lorentz, because a basis for 4 × 4 matrices is given by the sixteen,
linearly independent matrices
1 , γ5 , γµ , γ µγ 5 , Σµν , (2.50)
transforming respectively like scalar, pseudoscalar, vector, pseudovector and tensor.
2.4 Solutions of the Dirac equation
We can use Dirac spinors to write plane wave solutions of the Dirac equation. Consider
 
χ(p) −ipx
ψ(x) = e , (2.51)
φ(p)
where pµ = (E, p), and χ and φ are two-components spinors. Substituting (2.51) into the
Dirac equation (2.35) and using the representation (1.35) yields the coupled equations
for χ and φ      
χ m σ·p χ
E = , (2.52)
φ σ · p −m φ
that is,

σ · p φ = (E − m) χ
σ · p χ = (E + m) φ . (2.53)
√
These have solutions for positive and negative energies, E = ± p2 + m2 . We can write
the solution for positive energies as
 
χr
ψ+ (x) = N e−ipx ≡ ur (p)e−ipx , (2.54)
σ·p
E+m r
χ
√
where N = E + m, and the spinors χr for r = 1, 2 are given by
   
1 0
χ1 = , χ2 = . (2.55)
0 1
For negative energies, it is convenient to make the transformation pµ → −pµ , so that we
write the corresponding solution as
 
σ·p
E+m r
χ
ψ− (x) = N eipx ≡ vr (p)eipx . (2.56)
χr
The spinors u and v defined by Eqs. (2.54),(2.56) correspond respectively to particle
and antiparticle solutions. Taking the solutions in the rest frame p = 0, the top two
components of ψ describe electrons with spin up and spin down, while the bottom two
components describe positrons with spin up and spin down. This provides a clear physical
interpretation to the four components of Dirac spinors. For arbitrary p, we can study
the spin content of the solutions by using the explicit expression of the spin operator
 
1 1 σ 0
S= Σ= , (2.57)
2 2 0 σ
corresponding to spin 1/2,
 
1 1 σ·σ 0 3
2
S = Σ2 = = 1 , (2.58)
4 4 0 σ·σ 4
and helicity operator  
S·p 1 σ · p̂ 0
h= = . (2.59)
|p| 2 0 σ · p̂
 The u and v spinors satisfy the Dirac equation in momentum space,

(/p − m)u = 0 , (/p + m)v = 0 , (2.60)

and obey orthonormality and completeness relations. The orthonormality relations are
given by
u†r (p)us (p) = vr† (p)vs (p) = 2Eδ rs , (2.61)
u†r (p)vs (−p) = vr† (p)us (−p) = 0 . (2.62)
Equivalently, in terms of the adjoint spinors ū = u† γ 0 and v̄ = v † γ 0 ,

ūr (p)us (p) = −v̄r (p)vs (p) = 2mδ rs , (2.63)

ūr (p)vs (p) = −v̄r (p)us (p) = 0 , (2.64)

The completeness relations are given by
2
X
ur (p)ūr (p) = (/p + m) , (2.65)
r=1

2
X
vr (p)v̄r (p) = (/p − m) . (2.66)
r=1
3 Perturbation theory and S matrix
We now consider the coupling of the Klein-Gordon and the Dirac equation to electromag-
netism, and study scattering processes (Fig. 3) in quantum electrodynamics, applying
time-dependent perturbation theory.
We will express physical cross sections in terms of invariant scattering matrix ele-
ments. The application of perturbation theory can be encoded in the Feynman rules for
the calculation of the S matrix elements.

φ φ
1 3
p1 p3

p2 p4

φ φ
2 4

Figure 3: The scattering process φ1 + φ2 → φ3 + φ4 .

3.1 Electromagnetic interaction of spinless charges

We consider a relativistic, spinless system described by the Klein-Gordon equation,
Eq. (1.15), and we couple it to electromagnetism via the replacement

∂µ → ∂µ + ieAµ . (3.1)

By including the electromagnetic interaction (3.1) into Eq. (1.15), the equation of motion
of the system can be written

(∂ µ ∂µ + m2 )φ = −ie(∂µ Aµ + Aµ ∂ µ )φ + e2 A2 φ ≡ −Vφ , (3.2)

where in the right hand side we identify the potential

V = V1 + V2 , (3.3)

V1 = ie(∂µ Aµ + Aµ ∂µ ) , (3.4)
V2 = −e2 A2 . (3.5)
Let us apply first-order time-dependent perturbation theory to the transition ampli-
tude A due to the interaction given by V,
Z
A = −i d3 x dt φ∗f Vφi . (3.6)

We will derive the result for the term V1 of the potential. The contribution of V2 can be
treated analogously, and we will include the result later.
 By inserting the potential (3.4) into Eq. (3.6) and doing an integration by parts, we
can recast the transition amplitude in the form of a j · A interaction,
Z
A = −i d4 x jµ Aµ , (3.7)

where the interaction current is given by

jµ = ie[φ∗3 [∂µ φ1 ) − (∂µ φ∗3 )φ1 ] . (3.8)

φ φ
1 3
p1 p3

Figure 4: The φ1 → φ3 transition subprocess by V interaction.

We take the initial and final states φ1 and φ3 to be given by plane waves (Fig. 4)

φ1 = N1 e−ip1 x , φ3 = N3 e−ip3 x , (3.9)

normalized in a box of volume V so that

q q
N1 = 1/ 2E1 V , N3 = 1/ 2E3 V . (3.10)

Inserting Eqs. (3.4),(3.9) into Eq. (3.6) we obtain

Z
A = −iN1 N3 d4 x eip3 x (ie)(∂µ Aµ + Aµ ∂µ )e−ip1 x
Z
= −ieN1 N3 (p1 + p3 )µ d4 x e−iqx Aµ , (3.11)

where q = p1 − p3 , and we have done an integration by parts in the second line.

Let us now determine the electromagnetic potential Aµ in Eq. (3.11) which results
from the transition φ2 → φ4 . Let us use the Lorentz gauge-fixing condition,

∂ ν Aν = 0 . (3.12)

Then the equation of motion for the electromagnetic potential is given by

∂ 2 Aµ = J µ , (3.13)

where the current J µ is of the form (3.8), with the replacements 1 → 2, 3 → 4. The
states φ2 and φ4 are also represented by plane waves, analogously to Eqs. (3.9),(3.10).
By Fourier-transforming Eq. (3.13) with respect to x, we get

−q 2 Aeµ = Jeµ , (3.14)

where Z Z
Aeµ = 4
dxe −iqx
Aµ
, Jeµ = d4 x e−iqx J µ . (3.15)
That is,
−g µν e
Aeµ = Jν . (3.16)
q2
By substituting the explicit expression of the current J and inserting the result (3.16)
into Eq. (3.11), we obtain
1 Z 4 i(p3 −p1 )x i(p4 −p2 )x
A = ie2 N1 N3 N2 N4 (p1 + p3 )µ (p2 + p4 )µ dxe e . (3.17)
q2
Performing the integral in d4 x in Eq. (3.17) gives (2π)4 δ 4 (p3 +p4 −p1 −p2 ), which expresses
four-momentum conservation in the scattering process (Fig. 3). By inserting this result
and the explicit expressions of the normalization factors, we can rewrite Eq. (3.17) as
1 1
A = (2π)4 δ 4 (Pf − Pi ) Q √ Q √ Mf i , (3.18)
f 2Ef V i 2Ei V
where the products over i and f run respectively over initial and final state particles,
Pi and Pf denote the total four-momentum in the initial and final state, and we have
defined the scattering matrix element
−i g µν
Mf i = −e (p1 + p3 )µ e (p2 + p4 )ν . (3.19)
q2
We will interpret Eq. (3.19) as resulting from associating a factor ie(p1 + p3 )µ to the
p p
µ
µ i e (p + p )

p p
µν
2 i e 2 g
µ ν

Figure 5: Feynman rules for interaction vertices of spin-0 particles with photons.

transition vertex in Fig. 4, an analogous factor ie(p2 + p4 )µ to the φ2 → φ4 transition

vertex, and the factor −ig µν /q 2 to the electromagnetic interaction. This gives the Feyn-
man rules in the top row in Fig. 5 and in the top row in Fig. 8 (more comment on this in
Sec. 3.3). An analogous treatment of the potential term in Eq. (3.5) gives the Feynman
rule in the bottom row in Fig. 5.

3.2 Electromagnetic interaction of spin-1/2 charges

The case of spin 1/2 can be treated by an analysis analogous to that of the previous
subsection. In this case, by including the electromagnetic interaction (3.1) into the Dirac
equation (1.32),
∂ − eA
(i/ / − m)ψ = 0 , (3.20)
we arrive at the interaction potential

V = −eγ 0 γ µ Aµ . (3.21)

Then first-order perturbation theory gives

Z
A = −i d3 x dt ψf∗ V ψi
Z
= ie d4 x ψ f γ µ ψi Aµ , (3.22)

that is, a j · A interaction, Z

A = −i d4 x j µ Aµ , (3.23)
with the current given by
j µ = −eψ f γ µ ψi . (3.24)
Now we represent initial and final states by plane-wave solutions
q
ψk = Nk u(pk )e−ipk x , Nk = 1/ 2Ek V , k = 1, 2, 3, 4 . (3.25)

By following the same steps as in the scalar case, we arrive at

" # " #
Y 1 Y 1
4 4
A = (2π) δ (Pf − Pi ) √ √ Mf i , (3.26)
f 2Ef V i 2Ei V

where now the scattering matrix element is given by

−i g µν
Mf i = −eu(p3 )γµ u(p1 ) eu(p4 )γν u(p2 ) . (3.27)
q2
The corresponding Feynman rule for the spin-1/2 transition vertex is given in Fig. 5.

p p
µ
µ i e γ

Figure 6: Feynman rule for the interaction vertex of spin-1/2 particles with photons.

3.3 Green’s functions

In Sec. 3.1 we have treated the equation of motion (3.13) for the electromagnetic potential
by taking the Fourier transformation, and we have solved for the potential in Eq. (3.16) in
terms of the Green function, or propagator, of the ∂ 2 differential operator, proportional
to 1/q 2 . This function has poles for q 2 = q 02 − q2 = 0, that is,

q 0 = ±|q| . (3.28)
Therefore, when we take the inverse Fourier transform, in order to fully specify the solu-
tion we need to specify the prescription for going around these poles on the integration
contour in the complex q 0 plane. Different possible choices are sketched in Fig. 7, and
correspond to different boundary conditions on the solutions of Eq. (3.13):

a) vanishing fields far in the past (radiation case)

b) vanishing fields far in the future (absorption case)

c) propagation of positive frequencies in the future and negative frequencies in the past
(Feynman)

d) propagation of negative frequencies in the future and positive frequencies in the past
(anti-Feynman)

a) ret b) adv

c) F d) antiF

Figure 7: Contours in the complex q 0 plane: a) retarded; b) advanced; c) Feynman;

d) anti-Feynman.

The Feynman contour is obtained by taking

1
(3.29)
q2 + iε
in the denominator of the Green function, where ε is real and positive. We will take the
Feynman prescription for the propagator, according to the causality picture in Sec. 1.4.
The same discussion applies to the Green functions for the the Klein-Gordon equation
and the Dirac equation. The results for the photon, Klein-Gordon and Dirac propagators
are given in Fig. 8, using the Feynman prescription.
Note that Fig. 8 gives the photon propagator for a general covariant gauge-fixing
condition, Eq. (3.12). This depends on the gauge parameter ξ. The value ξ = 1 is the
Feynman gauge choice.
 µν µq ν / q 2
µ q
ν g − (1−ξ) q
−i
q 2 +i ε

q i

q 2 −m 2 +i ε

q i (q+m)

q 2 −m 2 +i ε

Figure 8: Feynman rules for propagators: (top) photon; (middle) Klein-Gordon; (bottom)
Dirac.

3.4 From scattering matrix elements to cross sections

In order to go from transition amplitudes to scattering cross sections, we need to i) con-
struct transition probabilities by squaring the amplitudes, ii) integrate over the final
state phase space, iii) divide by the incident flux of particles.
To carry out step i), we evaluate the square of the δ function, working in volume V
and time interval T , as
Z
4 4 2 4 4
|(2π) δ (Pf − Pi )| ≃ (2π) δ (Pf − Pi ) ei(Pf −Pi )x d4 x

≃ V T (2π)4 δ 4 (Pf − Pi ) . (3.30)

For step ii), we take the phase space element for each particle f in the final state,
V d 3 pf
dφf = . (3.31)
(2π)3
Then the transition probability per unit time is given by
|Af i |2 Y
dwf i = dφf
T f
 Y  !
1Y 1 1 Y V d 3 pf
4 4 2
= (2π) δ (Pf − Pi )|Mf i | V T .(3.32)
T f 2Ef V i 2Ei V f (2π)3

Note that the factor of T cancels; so does each factor of V associated with final state
particles.
Let us now consider the decay process in Fig. 9. The decay rate is given by
!
1 Y d 3 pf
dΓf i = |Mf i |2 V/ (2π)4 δ 4 (Pf − Pi )
2Ei V/ f (2π)3 2Ef
 !
1 Y d 3 pf
= |Mf i |2 (2π)4 δ 4 (Pf − Pi ) . (3.33)
2Ei f (2π)3 2Ef

Note the cancellation of the factor of V for the initial state.

f
1
a

f
n

Figure 9: Decay process a → f1 + f2 + . . . fn .

Next consider the scattering process in Fig. 10. To obtain the cross section we need
to divide the transition probability by the incident particle flux F ,
dwf i
dσf i = , (3.34)
F
1
F = |va − vb | , (3.35)
V
where |va − vb | is the relative velocity of colliding particles. Thus
!
V/ 1 Y d 3 pf
dσf i = |Mf i |2 V/ (2π)4 δ 4 (Pf − Pi )
|va − vb | 2Ea V/ 2Eb V/ f (2π)3 2Ef
!
1 Y d 3 pf
= |Mf i |2 (2π)4 δ 4 (Pf − Pi ) . (3.36)
4Ea Eb |va − vb | f (2π)3 2Ef

Note again the cancellation of all V factors.

a f
1

b f
n

Figure 10: Scattering process a + b → f1 + f2 + . . . fn .

Therefore the scattering cross section has the form

1
dσ = |Mf i |2 dΦ , (3.37)
J
where each of the three factors is relativistically invariant. The invariant matrix element
square |Mf i |2 contains the dynamics of the process, while the factors J and dΦ are
kinematic factors, giving respectively the invariant initial-state flux,

J = 4Ea Eb |va − vb | , (3.38)

and the invariant final-state phase space

!
Y d 3 pf
dΦ = (2π)4 δ 4 (Pf − Pi ) . (3.39)
f (2π)3 2Ef

The invariant flux J can also be rewritten in equivalent forms as

J = 4Ea Eb |va − vb |
(c.m.) √
= 4|pi | s
q
= 4 (pa · pb )2 − m2a m2b , (3.40)
(c.m.)
where s = (pa + pb )2 , and pi is the initial three-momentum in the center of mass
frame.
4 Coulomb scattering
In this section we analyze Coulomb scattering in QED using the formalism developed
in Sec. 3 for S-matrix calculations in perturbation theory. We begin in Subsec. 4.1 by
recalling Coulomb scattering in the nonrelativistic case. In Subsecs. 4.2-4.4 we calculate
elastic electron-muon scattering to lowest order in the QED coupling. Then in Subsec. 4.5
we obtain the cross section for the scattering of a relativistic particle from an external
Coulomb potential. In Subsec. 4.6 we consider the annihilation of electron pairs into
muon pairs, related to eµ scattering by crossing symmetry.

4.1 Nonrelativistic case

The cross section in nonrelativistic quantum mechanics for the scattering of a particle
of mass m from potential V is given in the first-order Born approximation by
dσ m2 e
= 2 |V (q)|2 , (4.1)
dΩ 4π
where Ve is the Fourier transform of the potential,
Z
Ve (q) = d3 x e−iq·x V (x) , (4.2)

q is the momentum transferred in the scattering (Fig. 11), and dΩ is the solid angle
element.

k
q
k
θ

Figure 11: Scattering through angle θ, with q = k − k′ .

Let us take a potential of the form

e−µ|x|
V (x) = C , µ>0 . (4.3)
|x|
We can apply Eq. (4.1), and we can obtain the nonrelativistic cross section for Coulomb
scattering by letting
C → e2 , µ→0 (4.4)
in the result.
By inserting Eq. (4.3) into Eq. (4.2), we get
4π
Ve (q) = C . (4.5)
µ2 + q2
Thus
dσ 4C 2 m2
= 2 . (4.6)
dΩ (µ + q2 )2
Substituting Eq. (4.4) into Eq. (4.6), we obtain that the Coulomb scattering cross section
in the nonrelativistic case is given by
!
dσ 4m2 e4
=
dΩ Coul.
q4
!
α2 dσ
= ≡ , (4.7)
4 k2 v2 sin4 (θ/2) dΩ R

where in the last line we have used q2 = 4k2 sin2 (θ/2), k = mv. The result is given by
the classical Rutherford scattering cross section (dσ/dΩ)R .
In the next few sections we analyze elastic electron-muon scattering in the fully
relativistic quantum theory. From this analysis we will also obtain, in Subsec. 4.5,
the relativistic correction to the result (4.7) for scattering from an external Coulomb
potential.

4.2 The eµ scattering matrix element in QED

Consider elastic electron-muon scattering e(p) + µ(k) → e(p′ ) + µ(k ′ ) (Fig. 12). Let m
be the electron mass and M the muon mass.
Using the Feynman rules for perturbation theory in Sec. 3, the scattering matrix
element Mf i is given by
gµν
Mf i = ie2 ur′ (k ′ )γ µ ur (k) 2
us′ (p′ )γ ν us (p) , (4.8)
q

where the momentum and spin labels are given in Fig. 12, and q 2 = (k − k ′ )2 is the
invariant momentum transfer.
k,r k,r
e e
q
µ µ

p,s p,s

Figure 12: Electron-muon scattering at lowest order in e.

To compute the unpolarized cross section, we need the squared matrix element, av-
eraged over initial spins and summed over final spins:
2 2 X 2 X 2
1X 1X
|Mf i |2 = |Mf i |2
2 r=1 2 s=1 r′ =1 s′ =1
1 e4 X
= 2 2
|ur′ (k ′ )γ µ ur (k)|2 |us′ (p′ )γµ us (p)|2 . (4.9)
4 (q ) spins

In the next subsection we give the basic result that is needed to evaluate such spin sums.
4.3 Fermionic spin sums
For any matrix Γ given by a product of Dirac γ matrices, it can be shown that
2 X
X 2 h i
|uα (p′ )Γuβ (p)|2 = Tr Γ(/
p + m)γ 0 Γ† γ 0 (p/′ + m) . (4.10)
α=1 β=1

To see this, write the square as

X 2
2 X 2 X
X 2
|uα (p′ )Γuβ (p)|2 = [uα (p′ )Γuβ (p)][uα (p′ )Γuβ (p)]∗ (4.11)
α=1 β=1 α=1 β=1

and evaluate the complex conjugate factor:

[uα (p′ )Γuβ (p)]∗ = u†β (p)Γ† (u†α (p′ )γ 0 )†

= uβ (p) γ 0 Γ† γ 0 uα (p′ ) . (4.12)

| {z }
e
Γ

Now write out all products of spinors and γ matrices in Eq. (4.11) in components:

X 2
2 X 2 X
X 2
|uα (p′ )Γuβ (p)|2 = e u (p′ )
uαa (p′ )Γab uβb (p)uβc (p)Γ (4.13)
cd αd
α=1 β=1 α=1 β=1

Next use the completeness relation for u spinors:

2
X
uαj (p)uαk (p) = (/
p + m)jk . (4.14)
α=1

Then from Eq. (4.13) we have

2 X
X 2
|uα (p′ )Γuβ (p)|2 = Γab (/ e (p/′ + m)
p + m)bc Γ cd da
α=1 β=1

= [Γ(/ e p/′ + m)]

p + m)Γ( aa

p + m)γ 0 Γ† γ 0 (p/′ + m)] ,

= Tr[Γ(/ (4.15)

which is the result in Eq. (4.10).

Analogous results hold for spin sums involving v spinors. For instance,
2 X
X 2 h i
|uα (p′ )Γvβ (p)|2 = Tr Γ(/
p − m)γ 0 Γ† γ 0 (p/′ + m) . (4.16)
α=1 β=1

4.4 Elastic eµ scattering cross section

By using the general result (4.10), the matrix element (4.9) can be written as

1 e4 h i h i
α ′ λ β ′ ρ
|Mf i |2 = Tr γ (/
k +m) γ (/
k +m) Tr γ (/
p +M ) γ (/
p +M ) gαβ gλρ . (4.17)
4 (q 2 )2
We can in general evaluate traces of products of γ matrices by using the anticommutation
relations (1.34) and (2.43). To do the calculation in Eq. (4.17) we need the following
traces,
Tr(odd number of γ matrices) = 0 , (4.18)
Tr(γ µ γ ν ) = 4g µν , (4.19)
Tr(γ µ γ ν γ ρ γ σ ) = 4 (g µν g ρσ − g µρ g νσ + g µσ g νρ ) , (4.20)
which can be obtained using Eqs. (1.34),(2.43). By then carrying out the algebra in
Eq. (4.17), we get

2e4 h 2 2 2 2 2 2 2 2 2 2
i
|Mf i |2 = 2 (m + M ) q + (s − m − M ) + (s + q − m − M ) , (4.21)
(q 2 )2

where s is the invariant center-of-mass energy square s = (k + p)2 .

We are now in a position to compute the cross section. This is given in terms of the
scattering matrix element via Eq. (3.37),
1
dσ = |Mf i |2 dΦ , (4.22)
J
where J is the invariant initial-state flux and dΦ is the invariant final-state phase space.
We can compute the cross section by plugging Eq. (4.21) into Eq. (4.22), choosing a
reference frame, and evaluating the flux factor J and the phase space dΦ integration in
this frame.
Consider the center-of-mass reference frame, k + p = 0 (Fig. 13). From Eq. (3.40)
we have √
J = 4|p| s . (4.23)
To carry out the integration over the final state phase space in Eq. (3.39),

d3 k ′ d 3 p′
dΦ = (2π)4 δ 4 (p′ + k ′ − p − k) , (4.24)
(2π)3 2Ek′ (2π)3 2Ep′

we can first use the three-momentum δ function to do the integral in d3 k ′ , so that the
cross section differential in the final-state solid angle dΩ = sin θdθdϕ can be written
dσ 1 1 Z ′2 1
= √ p d|p′ | |Mf i |2 δ(Ep′ + Ek′ − Ep − Ek ) . (4.25)
dΩ 4|p| s (2π) 2 4Ep′ Ek′

Next it is convenient to make the change of integration variable

q q
|p′ | → E ′ = p′2 + M 2 + p′2 + m2
= Ep′ + Ek′ , (4.26)

with jacobian
∂E ′ E ′ |p′ |
= , (4.27)
d|p′ | Ep′ Ek′
by which Eq. (4.25) can be rewritten as
dσ 1 1 Z ′
′ |p |
√
= √ 2
dE ′
|Mf i |2 δ(E ′ − s) . (4.28)
dΩ 4|p| s (2π) 4E
Performing the E ′ integral with the δ function and substituting the explicit expression
(4.21) of |Mf i |2 , we obtain (e2 = 4πα)
dσ α2 h 2 2 2 2 2 2 2 2 2 2
i
= 2 (m + M ) q + (s − m − M ) + (s + q − m − M ) . (4.29)
dΩ 2 s q4

k=( E , k) p =( E p , p)
k

Figure 13: Center-of-mass reference frame.

The result (4.29) takes a simpler form in in the high energy limit s ≫ M 2 , m2 . In
this case we have s → 4p2 , q 2 → −4p2 sin2 (θ/2), and the cross section becomes
dσ α2  
4
≃ 1 + cos θ/2 for s ≫ M 2 , m2 . (4.30)
dΩ 2 s sin4 (θ/2)
The leading behavior of the cross section (4.30) at small angle is given by
dσ 1
∝ 4 for θ ≪ 1 , (4.31)
dΩ θ
where the θ−4 singularity is characteristic of the Coulomb interaction. It comes from the
factor 1/(q 2 )2 , and reflects the long range of the interaction.

4.5 Scattering by an external Coulomb potential

The cross section for the scattering of a relativistic particle from an external Coulomb
potential (Fig. 14) can be obtained as a particular case of the result of the previous
subsection for eµ scattering, by working in the rest frame of µ and letting M → ∞.
To this end, first express the invariant flux J in terms of the electron’s three-
momentum k in the rest frame of µ (Fig. 14),
J = 4|k|M . (4.32)
Next, carry through the final-state phase space integration in terms of rest-frame
variables. This yields
dσ 1 |k′ |
= |Mf i |2 . (4.33)
dΩ 4|k|M 16π 2 M
 k
θ
k=( E , k)
k

Figure 14: Scattering by external Coulomb potential.

Finally, let M → ∞ in the invariant matrix element (4.21). In this limit |k′ | ≃ |k|,
q 2 ≃ −4k2 sin2 (θ/2), and the square bracket in Eq. (4.21) becomes
h i
2 (m2 + M 2 ) q 2 + (s − m2 − M 2 )2 + (s + q 2 − m2 − M 2 )2

≃ [2M 2 q 2 + 2(2M Ek )2 + . . .]
h i
≃ 8M 2 Ek2 1 − (|k|/Ek )2 sin2 (θ/2) . (4.34)

Substituting Eq. (4.34) into Eq. (4.33), we obtain

dσ α2 h i
2 2
= 4 1 − v sin (θ/2) , (4.35)
dΩ 4 k2 v2 sin (θ/2)

where v = |k|/Ek .
Eq. (4.35) can be compared with the nonrelativistic result in Eq. (4.7). Observe that
Eq. (4.35) has the form
!
dσ dσ h i
= 1 − v2 sin2 (θ/2) , (4.36)
dΩ dΩ R

where !
dσ α2
= (4.37)
dΩ R
4 k2 v2 sin4 (θ/2)
is the Rutherford cross section, and the factor in the square bracket is the relativistic
correction. h i
The relativistic correction 1 − v2 sin2 (θ/2) characterizes Coulomb scattering for
spin 1/2. Computing Coulomb potential scattering of spinless charges, one finds
! !
dσ dσ
= , (4.38)
dΩ spin−0
dΩ R

that is, in the case of spinless particles the result does not differ from the nonrelativistic
result. Eq. (4.36) shows that for |v| → 1 the angular distribution of a spin-1/2 particle
differs from the nonrelativistic result as diffusion in the backward direction is strongly
suppressed.
 −− −−
e µ
k q p

k p
e+ µ+

Figure 15: Annihilation of electron pairs into muon pairs.

4.6 Crossing symmetry: e+ e− → µ+ µ− annihilation

The eµ scattering process computed earlier is related by crossing symmetry to the an-
nihilation process e+ (k ′ )e− (k) → µ+ (p′ )µ− (p) (Fig. 15). Let us compute this in the
approximation of massless electrons. The matrix element square, averaged over initial
spins and summed over final spins, is given by
1 e4 h i h i
ρ ′ σ τ ′ λ
|Mf i |2 = Tr γ (/
p − M ) γ (/
p + M ) Tr γ /
k γ k
/ gλρ gστ (4.39)
4 (q 2 )2

where q 2 = s, and we have set the electron mass to zero. Computing the traces yields
8e4 h 2 ′ 2 2 ′
i
|Mf i |2 = (p · k) + (p · k ) + M k · k . (4.40)
s2
Let us work in the center-of-mass reference system, and denote by θ the center-of-
mass scattering angle. In this system the matrix element square (4.40) takes the form
" ! #
4 4M 2 4M 2 2
|2
|Mf i = e 1+ + 1− cos θ . (4.41)
s s

The annihilation cross section can be computed via the general formula (3.37),
1
dσ = |Mf i |2 dΦ . (4.42)
J
Note that for massless electrons
J = 2s , (4.43)
and that the final-state phase space can be written as
|p|
dΦ = √ dΩ , (4.44)
16π 2 s
where dΩ = sin θdθdϕ, and
√ s
s 4M 2
|p| = 1− . (4.45)
2 s
Then the differential cross section dσ/dΩ is given by
! s " ! #
dσ α2 4M 2 4M 2 4M 2 2
= 1− 1+ + 1− cos θ . (4.46)
dΩ e+ e− →µ+ µ−
4s s s s
In the high energy limit 4M 2 /s → 0, from Eq. (4.46) we get
!
dσ α2  
= 1 + cos2 θ for s ≫ M 2 . (4.47)
dΩ e+ e− →µ+ µ−
4s

The total cross section is obtained by integrating Eq. (4.46) over angles,
Z
dσ
σtot = dΩ
dΩ
s " ! ! #
α2 4M 2 4M 2 Z π 4M 2 Z π
= 1− 2π 1+ sin θ dθ + 1 − sin θ cos2 θ dθ
4s s s 0 s 0
s !
4 π α2 4M 2 2M 2
= 1− 1+ . (4.48)
3s s s

In the high energy limit,

4 π α2
σtot ≃ . (4.49)
3s
Remark. In addition to the annihilation into muons and other leptons, experiments
at high-energy e+ e− accelerators measure the cross section for the annihilation of e+ e−
into hadrons. The cross section for e+ e− → hadrons at large s differs from the expression
(4.49) for e+ e− → µ+ µ− by a factor of the squared electric charge of the hadron con-
stituents (quarks), summed over the possible Nc quark “colors” and Nf quark “flavors”,
and by effects of higher order in the strong interaction:
N
+ − 4 π α2 X f

σ(e e → hadrons) = Nc Q2i [1 + O(αstrong )] . (4.50)

3s i=1
5 Compton scattering
Let us analyze electron-photon Compton scattering at lowest order in e. This receives
contribution from the graphs in Fig. 16.

p k
p p

k k k p

( a) ( b)
Figure 16: Electron-photon scattering at lowest order in e.

The scattering matrix element Mf i is given by

" #
(/
p + k/ + m) p − k/′ + m)
(/
Mf i = −ie2 ε′µ (k ′ )εν (k)u(p′ ) γµ γ ν + γ ν γµ u(p), (5.1)
(p + k)2 − m2 (p − k ′ )2 − m2
where ε(k) and ε′ (k ′ ) are the incoming and outgoing photon polarization vectors. The
two terms in the square bracket in Eq. (5.1) correspond respectively to graphs (a) and
(b) in Fig. 16.
The sum of graphs (a) and (b) is gauge invariant, namely, with Mf i = Mµν ε′µ (k ′ )εν (k),
we have
µ
Mµν k ′ = Mµν k ν = 0 . (5.2)
This can be seen by writing
" #
ν 2 ′ 1 1
Mµν k = −ie u(p ) γµ k/ + k/ ′ γµ u(p)
p/ + k/ − m p/ − k/ − m
(
2 ′ 1
= −ie u(p ) γµ p + k/ − m) − (/
[(/ p − m)]
p/ + k/ − m
)
′ 1 ′
p − k/ − m) + (/
+ [−(/ p − m)] ′ γµ u(p) . (5.3)
p/ − k/ − m
Then, from each of the square brackets in the last two lines, the contributions of the first
terms cancel each other, while the contributions of the second terms vanish separately
because of the Dirac equation. Thus
" #
1 1
Mµν k ν = −ie2 u(p′ ) γµ p ′ − m) ′
p − m) + (/
(/ γµ u(p)
p/ + k/ − m p/ − k/ − m
= 0 . (5.4)

Similarly for the dot product of Mµν with k ′µ .

 To compute the unpolarized cross section, we need the squared matrix element, av-
eraged over the initial electron and photon polarizations, and summed over the final
electron and photon polarizations,
1 X
|Mf i |2 = |Mf i |2 . (5.5)
4 polarizations
The electron sums can be dealt with using the general result in Subsec. 4.3. The photon
sums are discussed in the next subsection.

5.1 Photon polarization sums

The sum over photon polarizations can be performed by replacing the sum with −g µν ,
2
X
εαµ (k)εαν (k) → −gµν , (5.6)
α=1

because the amplitude into which ε is dotted is conserved, Eq. (5.2).

To see this, consider a matrix element of the form
Aµ εµ (k) (5.7)
for Aµ such that
Aµ kµ = 0 , (5.8)
and sum the matrix element square over polarizations,
2
X 2
X
|Aµ εαµ (k)|2 = Aµ Aν εαµ (k)εαν (k) . (5.9)
α=1 α=1

Now use that polarizations form an orthonormal set in the plane transverse to the mo-
mentum k,
2
X
εαi (k)εαj (k) = δ ij − k̂ i k̂ j , where k̂ i = k i /|k| = k i /k 0 . (5.10)
α=1

Then the sum (5.9) can be written

2
X 2
X
|Aµ εαµ (k)|2 = Aµ Aν εαµ (k)εαν (k)
α=1 α=1

= Ai Aj (δ ij − k̂ i k̂ j ) = Ai Ai − (Ai k̂ i )(Aj k̂ j )
= Ai Ai − A0 A0 = −Aµ Aν gµν , (5.11)
where in the last line we have used that Eq. (5.8) implies
A0 k 0 − Ai k i = 0 , i.e., Ai k̂ i = A0 . (5.12)
From Eqs. (5.9) and (5.11) we obtain that the sum over polarizations amounts to
2
X
εαµ (k)εαν (k) → −gµν , (5.13)
α=1

as stated in Eq. (5.6).

5.2 The eγ unpolarized cross section
The unpolarized matrix element square (5.5) can now be determined from Eq. (5.1) by
using the result in Eq. (4.10) for the electron polarization sums and the result in Eq. (5.6)
for the photon polarization sums.
To do this calculation, we need to evaluate the traces of γ matrices generated by the
electron spin sums in Eq. (4.10), making use of the γ matrix identities

γµ γ µ = 4 , (5.14)

γµ γ ρ γ µ = −2γ ρ , (5.15)
which follow from the anticommutation relations (1.34). By working out the algebra, we
obtain the result
1 X
|Mf i |2 = |Mf i |2
4 polarizations
 ! !2 
p·k p · k′ 1 1 1 1
= 2e4  ′
+ + 2m2 − + m4 − (5.16)
.
p·k p·k p · k p · k′ p · k p · k′

The eγ cross section is related to the scattering matrix element via the general formula
(3.37),
1
dσ = |Mf i |2 dΦ . (5.17)
J
We can compute it by choosing a reference frame, plugging Eq. (5.16) into Eq. (5.17)
and evaluating the flux factor J and phase space dΦ integration.
k =(ω, k )

p=(m,0)
θ

k=( ω , k)

Figure 17: Compton scattering in the laboratory frame.

Consider the laboratory frame in which the electron is initially at rest, pµ = (m, 0)
(Fig. 17). In the notation of Fig. 17 we have

m2 = p′2 = (p + k − k ′ )2 = m2 + 2m(ω − ω ′ ) − 2ωω ′ (1 − cos θ) , (5.18)

which gives
ω
ω′ = . (5.19)
1 + (ω/m)(1 − cos θ)
By evaluating the right hand side of Eq. (5.16) in the laboratory frame and using
Eq. (5.19), we obtain !
2 4 ω ω′ 2
|Mf i | = 2e + − sin θ . (5.20)
ω′ ω
From Eq. (3.40) for the flux factor we get
J = 4Ea Eb |va − vb | = 4mω . (5.21)
Let us now carry out the integration over the final-state phase space (3.39),
d3 k ′ d 3 p′
dΦ = (2π)4 δ 4 (p′ + k ′ − p − k) . (5.22)
(2π)3 2ω ′ (2π)3 2E ′
By using the three-momentum δ function to perform the integral in d3 p′ , and inserting
the results (5.20) and (5.21) into Eq. (5.17), the differential cross section in the solid
angle Ω of the final photon momentum is given by
" #
dσ 2e4 Z ′ ω′ ′ ′ ω ω′ 2
= dω δ(E + ω − m − ω) + − sin θ , (5.23)
dΩ 4mω 16π 2 E ′ ω′ ω
where √
E′ = ω 2 + ω ′2 − 2ωω ′ cos θ + m2 . (5.24)
Performing the integral in dω ′ , we arrive at the unpolarized electron-photon cross section
!2 " #
dσ α2 ω′ ω ω′
= + − sin2 θ , (5.25)
dΩ 2m2 ω ω′ ω
where α = e2 /4π.
In the low energy limit ω ≪ m, from Eq. (5.19) we have ω ′ ≈ ω, and Eq. (5.25)
reduces to the Thomson cross section,
dσ α2
−→ (1 + cos2 θ) for ω ≪ m , (5.26)
dΩ 2m2
describing the scattering of classical electromagnetic radiation by a free electron.
In the high energy limit ω ≫ m, Eq. (5.25) gives rise to a logarithmic behavior in the
total cross section, arising from the emission of photons at small angles. This is because
for ω ≫ m from Eq. (5.19) we have
m ω
ω′ ≃ for (1 − cos θ) ≫ 1 , (5.27)
1 − cos θ m
which means that in the region
2m
1 ≫ θ2 ≫ (5.28)
ω
the cross section is strongly peaked,
 2  
dσ α2 m 1 ω
≃ (1 − cos θ) − sin2 θ
dΩ 2m2 ω 2
(1 − cos θ) m
α2 1
≃ . (5.29)
2mω 1 − cos θ
Integrating over angles gives
Z
α2 1
σ ≃ 2π d cos θ
2mω 1 − cos θ
Z 1 dθ2 α2 2πα2 ω
≃ 2π 2
≃ ln . (5.30)
2m/ω θ mω mω 2m
The total cross section σ falls like ω −1 , but with a logarithmic enhancement from the
integration over the small-angle, or collinear, region. The collinear region is cut off by
the mass m. The occurrence of collinear logarithms illustrated by this example is a
general feature associated with the massless limit of the theory.

5.3 Photon polarization dependence

The calculation performed above can be redone for fixed ε(k) and ε′ (k ′ ) to obtain the
dependence of the cross section on the initial and final photon polarizations. The result
for the cross section including the polarization dependence is
! !2 " #
dσ α2 ω′ ω ω′
= + + 4(ε · ε′ )2 − 2 . (5.31)
dΩ pol.
4m2 ω ω ′ ω

From Eq. (5.31) we recover Eq. (5.25) through averaging over ε and summing over ε′ by
using Eq. (5.10), i.e., that the sum over polarizations gives the transverse projector with
respect to the momentum,
2
X
εαi (k)εαj (k) = δij − k̂i k̂j , (5.32)
α=1

2
X
ε′β ′ ′β ′ ′ ′
i (k )εj (k ) = δij − k̂i k̂j . (5.33)
β=1

We thus have
2 X 2
!
1X dσ
2 α=1 β=1 dΩ pol.
!  " # 
2 ′ 2 ′ 2 X
2 
α ω ω ω 1X
= 2 + −2 +4 [εα (k) · ε′β (k ′ )]2  , (5.34)
4m2 ω  ω ′ ω 2 α=1 β=1

where
2 X 2 2 X 2
1X 1X
[εα (k) · ε′β (k ′ )]2 = εαi (k)εαj (k)ε′β ′ ′β ′
i (k )εj (k )
2 α=1 β=1 2 α=1 β=1
1 1
= (δij − k̂i k̂j ) (δij − k̂i′ k̂j′ ) = [1 + (k̂ · k̂′ )2 ]
2 2
1 1
= (1 + cos2 θ) = (2 − sin2 θ) . (5.35)
2 2
Substituting Eq. (5.35) into Eq. (5.34) we re-obtain the unpolarized result (5.25).
6 Strong interactions
In this section we extend the discussion given in the previous sections to the case of strong
interactions. The theory of strong interactions, Quantum Chromodynamics (QCD), is
treated systematically in the Standard Model course. Here we give a brief introduction
building on material presented for QED. In Subsec. 6.1 we introduce the gauge symmetry
of the strong interaction as a generalization of that of QED and examine the couplings
that follow from it. In Subsec. 6.2 we contrast features of photon and gluon polarization
degrees of freedom and discuss physical implications. In Subsec. 6.3 we give basic results
on the algebra of QCD charges that serve in practical calculations.
We will use the results presented here to further discuss strong interactions in Sec. 8.

6.1 Basic structure

We can think of QCD as a theory similar to QED but with

• N = 3 charged spin-1/2 particles ψi (quarks, replicated in six families — the

so-called quark “flavors”),

• N 2 − 1 = 8 gauge bosons Aaµ (gluons),

with different couplings to different charges. The couplings are to be thought of as

matrices
T a , a = 1, . . . , N 2 − 1 (6.1)
obeying well-prescribed commutation relations

[T a , T b ] = if abc T c , (6.2)

where f abc are “structure constants”, antisymmetric in all indices. The quantum number
specifying the charge of QCD is called “color”, and T a are the color-charge matrices.
Thus QCD contains multiple vector particles and its charges are non-commuting, or
“non-abelian”.
Eq. (6.2) defines an algebra of color-charge operators whose formal properties can
usefully be thought of along similar lines to the discussion given in Sec. 2 for the algebra
(2.1) of angular momentum operators J i . J i are the generators of the rotation group. T a
are the generators of the color symmetry group (SU (N ) with N = 3), and have matrix
representations for different dimensionalities n. The fundamental representation is the
representation with dimensionality n = N to which quarks belong, ψi , i = 1, 2, 3. The
matrix representation of the generators is given by
1 a
Ta → λ , (6.3)
2
where λa are the eight Gell-Mann 3 × 3 matrices. The adjoint representation is the
representation with dimensionality n = N 2 − 1 to which gluons belong, Aaµ , a = 1, . . . , 8.
The matrix representation of the generators in the adjoint is given by the structure
constants themselves,
(T a )bc → −if abc . (6.4)
 As in QED, the spin-1/2 charged particles satisfy equations of motion of Dirac type,

∂ − m)ψ = 0 ,
(i/ (6.5)

and we can write down their coupling to the vector particles, the gluons, based on
similar reasoning as in QED. While in QED we write the electron-photon interaction by
replacing
∂µ → Dµ = ∂µ + ieAµ (QED) (6.6)
in the equations of motion, in QCD it is not just one term by which we modify ∂ µ but
a sum of terms, one for each of the gluons, and each term is proportional not just to a
number like the electric charge but to a color-charge matrix:

∂µ → Dµ = ∂µ + igs Aaµ T a (QCD) , (6.7)

where gs is the strong-interaction coupling constant. With this interaction term, by going
through the analogous perturbation analysis as for QED, we can extract the Feynman
rule for the quark-quark-gluon coupling. This is given in Fig. 18. It has a similar
structure to the QED vertex of Fig. 6, with the difference being in the color matrix.

j k µ a
µ ig γ (T ) jk
s
a

Figure 18: QCD Feynman rule for quark-quark-gluon vertex.

For internal lines we have the Feynman rules for propagators (Fig. 19), similar to
those of QED except for the additional dependence on color indices.

q jk i (q+m)
δ
j k q 2 −m 2 +i ε

q ab µν
µ ν −ig (Feynman
δ
gauge)
a b q 2 +i ε

Figure 19: QCD Feynman rules for quark and gluon propagators.

The quark-gluon coupling above does not exhaust QCD interactions, though. In a
theory of multiple vector particles such as QCD the vector particles turn out to necessar-
ily be self-interacting. The reason for this lies with the non-abelian nature of the color
charges, i.e., with the nonzero commutator (6.2). The origin and precise form of the gluon
self-couplings can specifically be traced back to the form of the gauge transformations
in QCD and relationship between potentials and fields.
Electromagnetism is invariant under gauge transformations given by changes of the
four-potential Aµ by an arbitrary four-gradient,
Aµ → Aµ + ∂µ λ (QED) . (6.8)
In QCD the gauge freedom involves an additional contribution, namely we can change
Aµ by a four-gradient and/or by a pure rotation of its color indices,
Aaµ → Aaµ + ∂µ λa − gs f abc λb Acµ (QCD) , (6.9)
leaving physics invariant. Eq. (6.9) specifies the form of the gauge transformations in
QCD. Because of the nonvanishing structure constants f abc , while the field strength
tensor Fµν in QED is given by
Fµν = ∂µ Aν − ∂ν Aµ (QED) , (6.10)
in QCD this acquires an extra term,
a
Fµν = ∂µ Aaν − ∂ν Aaµ + gs f abc Abµ Acν (QCD) . (6.11)
a
It is precisely the extra term in Fµν in Eq. (6.11) which is responsible for producing
gluon self-interactions when one constructs a gauge-invariant kinetic energy term for
a
gluons by taking the square of Fµν , analogously to the case of photons. The square of
a
Fµν in Eq. (6.11) gives rise to both cubic and quartic gluon self-interaction terms. The
cubic term is proportional to gs × f and contains derivative couplings, while the quartic
term is proportional to gs2 ×f 2 , with no derivatives. Their precise form is given in Fig. 20.

µ,a ρ,c abc µν ρ

g f [g (k−p)
s
νρ µ
k q +g (p−q)
p ρµ ν
+g (q−k) ]
ν,b

ν,b 2 abe cde

µ, a µρ ν σ µ σ νρ
−i g s f f (g g −g g )

+ permutations

ρ, c σ, d

Figure 20: QCD Feynman rules for cubic and quartic gluon vertices.

The construction above implies in particular that the coupling constant gs in the
gauge boson self-interaction vertices is one and the same as the coupling constant gs
in the quark-gluon interaction vertex. Later in the section we see a specific example
showing that this equality of couplings is necessary for non-abelian gauge invariance to
be satisfied.
6.2 Physical polarization states and ghosts
In this subsection we discuss implications of the non-abelian gauge symmetry by com-
paring features of photon and gluon polarization degrees of freedom. We start from
examining gauge invariance in a simple example, the QCD analogue of Compton scat-
tering, and contrast the case of QCD with the case of QED seen in Sec. 5.

p p
p p
k ,b, µ p

k,a, ν k ,b, µ k,a, ν k ,b, µ

p k,a, ν

( a) ( b) ( c)

Figure 21: Quark-gluon Compton scattering at lowest order in gs .

The QCD analogue of Compton scattering is the quark-gluon scattering depicted in

Fig. 21 at lowest order in the coupling gs . The graphs in Fig. 21(a) and (b) are analogous
to the QED graphs of Fig. 16, while the three-gluon coupling graph in Fig. 21(c) is non-
abelian. The scattering matrix element Mf i can be written as

Mf i = Mµν ε′µ (k ′ )εν (k) . (6.12)

From graphs (a) and (b) in Fig. 21 we have

" #
(a)+(b) 1 1
Mµν = −igs2 u(p′ ) γµ Tb
γν T a + γν T a γµ T b u(p) . (6.13)
p/ + k/ − m p/ − k/′ − m
The sum of graphs (a) and (b) is not by itself gauge-invariant, because by dotting
Eq. (6.13) into k ν we get
(a)+(b) ν
Mµν k = igs2 [T a , T b ]u(p′ )γµ u(p) , (6.14)

which is nonvanishing due to the nonzero commutator of color charges.

From graph (c) in Fig. 21 we have

(c) 1
Mµν = gs2 u(p′ )γ ρ T c u(p) ′ 2
f abc [gνµ (k + k ′ )ρ + gµρ (k − 2k ′ )ν + gρν (k ′ − 2k)µ ] .
(k − k )
(6.15)
ν
By dotting Eq. (6.15) into k we get
(c) ν
Mµν k = gs2 f abc T c u(p′ )γµ u(p) + {. . .} kµ′

= −igs2 [T a , T b ]u(p′ )γµ u(p) + {. . .} kµ′ (6.16)

where in the last line we have used Eq. (6.2) to rewrite f abc T c in terms of the commutator.
The second term in the right hand side of Eq. (6.16) is a term proportional to kµ′ , which
gives zero once it is dotted into physical polarizations ε′ (k ′ ) · k ′ = 0. The first term, on
the other hand, precisely cancels the contribution in Eq. (6.14). Thus gauge invariance
is achieved once graph (c) is added to graphs (a) and (b). This illustrates that gauge
boson self-interactions are required by gauge invariance in the non-abelian case, and that
their coupling constant must equal that of quark-gluon interactions.
The term in kµ′ in the right hand side of Eq. (6.16), however, signifies that gauge
invariance is realized in quite a different manner than in the abelian case. In Eq. (6.16)
we obtain Mµν k ν = 0 only if µ is restricted to physical polarizations, while in the QED
case, Eq. (5.4), we have Mµν k ν = 0 regardless of µ. While this is of no consequence in the
present lowest-order case, since we are entitled to enforce physical gluon polarizations,
it implies a profound difference when we analyze the theory beyond lowest order and
include loops (as we will do in the next two sections).
Recall that in QED as a consequence of Mµν k ν = 0 we arrived at the equivalence
implied by Eq. (5.6),
2
X
εαµ (k)εαν (k) → −gµν (QED) , (6.17)
α=1

in which the sum on the left hand side is over transverse polarizations, while the right
hand side sums over all covariant polarization states, including the unphysical longitu-
dinal ones. This means that the structure of the abelian theory implies that unphysical
polarization states automatically cancel. The result we have just found for QCD indi-
cates that this cancellation is not automatic in the non-abelian theory. Thus, if we are to
restore the equivalence between sum over physical states and sum over covariant states
in the QCD case, further degrees of freedom are to be added in to the theory, which will
have to be such that they cancel the contribution of unphysical gluon polarizations.

q
b c abc µ
g f q
s
a µ

q ab i
δ
a b q 2 +i ε

Figure 22: QCD Feynman rules for ghost vertex and ghost propagator.

The systematic construction of the theory shows that such terms emerge in a well-
prescribed, precise manner. They are referred to as ghosts and correspond to well-
defined, but not physical, degrees of freedom, which can propagate and couple to gluons,
but never be produced in physical final states. Their role is precisely that of canceling
unphysical gluon polarization states. The Feynman rules for ghost propagator and cou-
pling are given in Fig. 22. Ghosts transform under the adjoint representation (6.4) of
the color symmetry group, and transform like scalars under Lorentz (though they obey
anticommuting relations like fermions), hence the form of their propagator and deriva-
tive coupling in Fig. 22. Ghosts are a non-abelian effect. Their coupling is proportional
to f abc . We could introduce ghosts in electrodynamics, but they will just decouple.
It is possible to exploit the gauge freedom in order to devise a gauge-fixing condition
such that the unphysical gluon polarizations are automatically eliminated, and thus no
need arises for ghosts. Such gauges without ghosts are referred to as physical gauges, and
are the gauges in which the non-abelian theory looks the most like its abelian counterpart.
They are distinct from the covariant gauges based on the Lorentz gauge-fixing condition
(3.12) in which we have worked so far. The gauge-fixing relation for physical gauges is
given by assigning a four-vector nµ and setting

Aaµ nµ = 0 . (6.18)

Physical gauges (6.18) present certain advantages, as they do not contain unphysical
degrees of freedom. However, the form of the gluon propagator becomes rather more
complicated in these gauges. This is given in Fig. 23.

q ab
µ ν i µν
δ (−g +
a b q 2 +i ε

µ ν ν µ µ ν
q n + q n 2 q q
+ −n )
q n (q n ) 2

Figure 23: Gluon propagator in physical gauge A · n = 0.

6.3 Color algebra

QCD calculations involve charge factors based on the color algebra in Eq. (6.2). We
here introduce basic invariants of the algebra which occur in practical applications, the
Casimir invariants CR and the trace invariants TR . The subscript R specifies the repre-
sentation of the algebra (6.2).
The Casimir invariant can be defined from the square operator T 2 = T a T a . This
operator commutes with all generators of the algebra (6.2),

[T a T a , T b ] = T a [T a , T b ] + [T a , T b ]T a
= if abc (T a T c + T c T a ) = 0 (6.19)

where the last line vanishes due to f abc being antisymmetric. This is analogous to
the case of the algebra of angular momentum operators, where [J 2 , J i ] = 0, and the
eigenvalue of the square operator J 2 is used to label different representations. In the
case of the color charge operators, Eq. (6.19) implies that T 2 takes a constant value on
each representation, and the matrix representation of T 2 is given by a constant CR times
the identity matrix,
T a T a = CR 1 . (6.20)
Here 1 is the identity matrix in n dimensions, where n is the dimensionality of the
representation. The constant CR is the Casimir invariant, and characterizes the specific
representation.
The trace invariant can be defined from the trace of two generators, Tr(T a T b ), based
on the fact that one can choose a basis such that this trace is proportional to δ ab ,

Tr(T a T b ) = TR δ ab . (6.21)

The constant TR is the trace invariant, specific to any given representation. Convention-
ally we normalize the color generators so that in the fundamental representation R = F ,
specified by the generator matrices (6.3), we have
1
TF = . (6.22)
2
Once this is fixed, all other Casimir and trace invariants are determined in all represen-
tations. The normalization (6.22) for the color generators is analogous to that of the
angular momentum generators in the spin-1/2 representation (2.3), J i → σ i /2, for which
1 ij
Tr(J i J j ) = δ . (6.23)
2
The Casimir invariant CR and the trace invariant TR are related to each other, because
if we multiply Eq. (6.21) by δ ab we get

δ ab Tr(T a T b ) = TR δ ab δ ab = TR d , (6.24)

where d = N 2 −1 is the number of generators, and therefore, using Eq. (6.20) to evaluate
the left hand side of Eq. (6.24), we have

C R n = TR d . (6.25)

Eq. (6.25) implies that the Casimir invariant for the fundamental representation
R = F , for which n = N = 3, is given by

d N2 − 1 4
C F = TF = = , (6.26)
n 2N 3
where we have used Eq. (6.22). In the adjoint representation R = A specified by the
generator matrices (6.4), for which n = N 2 − 1, Eq. (6.25) implies that the Casimir
invariant and trace invariant are equal, CA = TA . By performing the explicit calculation
we get
C A = TA = N = 3 . (6.27)
We will see examples of QCD calculations involving the color factors above in Sec. 8.
7 Renormalization
So far we have considered tree-level effects, that is, Feynman graphs that do not contain
loops. Loop corrections to processes described in earlier sections arise at higher orders of
perturbation theory. For instance, the graph in Fig. 24 is an example of a loop correction
to fermion pair production.

Figure 24: Loop correction to fermion pair production.

In this section we address loop effects. The part of the theory that deals with these
effects is renormalization. We give an introduction to the idea of renormalization and
its physical implications, discussing two specific examples: i) the renormalization of the
electric charge; ii) the electron’s anomalous magnetic moment. In Sec. 8 we continue
the discussion by introducing further concepts and including both electromagnetic and
strong interactions.

7.1 General principles

While the treatment given so far specifies interaction processes at tree level, the method
of renormalization is required to treat processes including loops.
A symptom that renormalization is required is that Feynman graphs with loops may
give rise to integrals containing divergences from high-momentum regions. Renormal-
ization allows one to give meaning to the occurrence of these ultraviolet divergences.
Ultraviolet power counting provides the basic approach to renormalization. For a
Feynman graph involving a loop integral of the form
Z
N (k)
d4 k , (7.1)
M (k)

consider the superficial degree of divergence defined as

D = (powers of k in N + 4) − (powers of k in M ) . (7.2)

If D ≥ 0, the integral is ultraviolet divergent. A first way of characterizing a theory

as “renormalizable” is that the number of ultraviolet divergent amplitudes is finite.
This is the case for instance with QED. There are 3 ultraviolet divergent amplitudes in
QED, depicted in Fig. 25. QCD has a few more, due to the more complex structure of
interactions seen in Sec. 6, but still a finite number. In a renormalizable theory there can
of course be infinitely many Feynman graphs that are ultraviolet divergent, but they are
so because they contain one of the few primitively divergent amplitudes as a subgraph.
 (a)

(b)

(c)

Figure 25: Ultraviolet divergent amplitudes in QED: (a) photon self-energy; (b) electron
self-energy; (a) electron-photon vertex.

The main point about renormalizability is that it implies that all the ultraviolet
divergences can be absorbed, according to a well-prescribed procedure specified below,
into rescalings of the parameters and wave functions in the theory. For a given quantity
φ, the rescaling is of the form
φ → φ0 = Z φ , (7.3)
where φ0 and φ are respectively the unrenormalized and renormalized quantities, and
Z is a calculable constant, into which the divergence can be absorbed. Here Z is the
renormalization constant, possibly divergent but unobservable. Once the rescalings are
done and the predictions of the theory are expressed in terms of renormalized quantities,
all physical observables are finite and free of divergences.
This leads to a characterization of the renormalization program which we can formu-
late as a sequence of steps as follows.
• Compute the divergent amplitudes, by prescribing a “regularization method”. Ex-
amples of regularization methods are a cut-off Λ on the ultraviolet integration
region, where the result diverges as we let Λ → ∞, or, as we will see in explicit
calculations later, dimensional regularization.
• Assign parameter and wave-function rescalings to eliminate divergences. In the
case of QED, these involve the electromagnetic potential A, the electron wave
function ψ and mass m, and the coupling e. Using traditional notation for the
QED renormalization constants Zi , the rescalings can be written as
q
A → A0 = Z3 A ,
q
ψ → ψ0 = Z2 ψ ,
Zm
m → m0 = m ,
Z2
Z1
e → e0 = √ e . (7.4)
Z2 Z3
 Here Z3 and Z2 are the respectively the renormalization constants for the photon
and electron wave function, Z1 is the vertex renormalization constant and Zm is
the electron mass renormalization constant.

• Once the rescalings are done, all physical observables are calculable, i.e., unam-
biguously defined in terms of renormalized quantities, and free of divergences.

Theories for which this program succeeds in giving finite predictions for physical quan-
tities are renormalizable theories. Non-renormalizable theories are theories in which one
cannot absorb all divergences in a finite number of Z: for instance, as we go to higher
orders new divergences appear and an infinite number of Z is needed.
The above program, while it appears quite abstract at first, gives in fact testable,
measurable effects. In the next few subsections we see specific examples of this.
A further, general point is that gauge invariance places strong constraints on renor-
malization, implying relations among the divergent amplitudes of the theory, and thus
among the renormalization constants. Here is an example for the case of QED. Gauge in-
variance establishes the following relation between the electron-photon vertex Γµ dotted
into the photon momentum q µ and the electron propagators S,

q µ Γµ = S −1 (p + q) − S −1 (p) . (7.5)

Eq. (7.5), pictured in Fig. 26, is referred to as the Ward identity and is valid to all orders.
Using the renormalization constants Z1 and Z2 defined by the rescalings in Eq. (7.4),

1 µ Z2
Γµ = γ + ... , S(p) = + ... , (7.6)
Z1 p/ − m
we have
1 1
q/ = p + q/ − m) − (/
[(/ p − m)] . (7.7)
Z1 Z2
Thus in the abelian case
Z1 = Z2 (QED) . (7.8)
As a result, the rescaling relation in Eq. (7.4) defining the renormalized coupling in QED
becomes
e2 = Z3 e20 . (7.9)
That is, the renormalization of the electric charge is entirely determined by the renor-
malization constant Z3 , associated with the photon wave function, and does not depend
on any other quantity related to the electron.

q = −
p p+q p+q p

Figure 26: Relation between electron-photon vertex and electron propagators.

 In the non-abelian case the relation (7.8) does not apply. However, it is still valid
that non-abelian gauge invariance sets constraints on renormalization, leading to other,
more complex relations among the renormalization constants. We will see examples of
this in Sec. 8.3.
In the rest of this section we describe specific calculations of renormalization at one
loop.

7.2 The gauge boson self-energy

Let us consider the gauge boson self-energy. This is one of the divergent amplitudes
shown in Fig. 25. The Feynman graphs contributing to the self-energy through one loop
are given in Fig. 27 for the photon and gluon cases. In the photon case one has the
fermion loop graph only, while in the gluon case one has in addition gluon loop and
ghost loop graphs.
Because of the relations (7.8),(7.9), in the QED case the calculation of the gauge
boson self-energy is all that is needed to determine the renormalization of the coupling.
So the result of this subsection will be used in Subsec. 7.3 to discuss the renormalized
electric charge.

QED: = + +

QCD: = + +

Figure 27: (top) Photon and (bottom) gluon self-energy through one loop.

We now compute the fermion loop graph in Fig. 28. As shown in Fig. 27, in the QED
case the fermion loop is all that contributes to the self-energy, while in the QCD case
this gives one of the required contributions.

q+k
µ, a ν, b ab
= iπ (q)
µν
q k

Figure 28: Fermion loop contribution to the gauge boson self-energy.

 The graph in Fig. 28 is given by
Z
ab 2 a b d4 k Tr[γµ (/
k + q/ + m)γν (/
k + m)]
iπµν (q) = −g Tr(T T ) . (7.10)
(2π) (k − m + i0 )((k + q) − m2 + i0+ )
4 2 2 + 2

This expression is written in general for the non-abelian case. In this case the color-
charge factor is evaluated from Eqs. (6.21),(6.22) and equals
1 ab
Tr(T a T b ) = δ . (7.11)
2
The QED case is obtained from Eq. (7.10) by taking

g 2 −→ e2 = 4πα ,
Tr(T a T b ) −→ 1 . (7.12)

The integral in Eq. (7.10) is ultraviolet divergent. By superficial power counting in the
loop momentum k, the divergence is quadratic. Gauge invariance however requires that
πµν be proportional to the transverse projector gµν q 2 − qµ qν , that is,
 
πµν = gµν q 2 − qµ qν Π(q 2 ) . (7.13)

This reduces the degree of divergence by two powers of momentum. As a result, the
divergence in Eq. (7.10) is not quadratic but logarithmic.
We need a regularization method to calculate the integral (7.10) and parameterize
the divergence. We take the method of dimensional regularization. This consists of
continuing the integral from 4 to d = 4 − 2ε dimensions by introducing the dimensionful
mass-scale parameter µ so that
d4 k 2 ε d
4−2ε
k
g2 −→ g 2
(µ ) . (7.14)
(2π)4 (2π)4−2ε

In dimensional regularization a logarithmic divergence d4 k/k 4 appears as a pole at ε = 0

(i.e., d = 4). We thus identify ultraviolet divergences in the integral (7.10) by identifying
poles in 1/ε.
By carrying out the calculation in dimensional regularization, the result for πµν is
Z 1 !ε
  g2 4πµ2
ab 2 a b
πµν (q) = − gµν q − qµ qν Tr(T T ) Γ(ε) dx 2x(1 − x)
4π 2 0 m2 − x(1 − x)q 2
 
≡ gµν q 2 − qµ qν Π(q 2 ) . (7.15)

We can interpret the different factors in this result. As mentioned above, the first factor
on the right hand side, consistent with the gauge-invariance requirement (7.13), implies
that the gauge boson self-energy is purely transverse,
   
gµν q 2 − qµ qν q µ = gµν q 2 − qµ qν q ν = 0 . (7.16)

Owing to the transversality of the self-energy, loop corrections do not give mass to gauge
bosons in QED and QCD. The factor Tr(T a T b ) in Eq. (7.15) is the non-abelian charge
factor, which just reduces to 1 in the QED case according to Eq. (7.12). Next, g 2 /(4π 2 )
is the coupling factor, which becomes e2 /(4π 2 ) = α/π in the QED case (7.12). The Euler
gamma function Γ(ε) contains the logarithmic divergence, i.e., the pole at ε = 0 (d = 4)
in dimensional regularization:
1
Γ(ε) = − CE + O(ε) , CE ≃ .5772 . (7.17)
ε
The first factor in the integrand of Eq. (7.15) results from the regularization method,
depending on the ratio between the dimensional-regularization scale µ2 and a linear
combination of the physical mass scales m2 and q 2 . The last factor in the integrand,
2x(1 − x), depends on the details of the calculated Feynman graph.
We can extract the ultraviolet divergent part of the self-energy by computing the
integral in Eq. (7.15) at q 2 = 0. Higher q 2 powers in the expansion of Π(q 2 ) give finite
contributions. We have
Z 1 !ε
g2
a b 4πµ2
Π(0) = −Tr(T T ) Γ(ε) dx 2x(1 − x)
4π 2 0 m2
g2 1 1
≃ −Tr(T a T b ) + ... , (7.18)
4π 3π ε
where in the last line we have used the expansion (7.17) of the gamma function and
computed the integral in dx. Specializing to the QED case according to Eq. (7.12) gives
α 1
Π(0) ≃ − + ... (QED) . (7.19)
3π ε
We will next use the results in Eqs. (7.15),(7.19) to discuss the renormalization of the
electromagnetic coupling.

7.3 Renormalization of the electromagnetic coupling

Suppose we consider a physical process occurring via photon exchange, and ask what the
effect is of the renormalization on the photon propagator. Fig. 29 illustrates this effect
by multiple insertions of the photon self-energy,

D0 → D = D0 + D0 πD0 + D0 πD0 πD0 + . . . , (7.20)

where D0 is the photon propagator given in Fig. 8 and π is the photon self-energy
computed in Eq. (7.15). We can sum the series in Eq. (7.20) by applying repeatedly
the transverse projector in π and using that longitudinal contributions vanish by gauge
invariance, and we get
1
D0 → D = D0 . (7.21)
1 + Π(q 2 )
Then the effect of renormalization in the photon exchange process amounts to

e20 e20 1
2
−→ 2 , (7.22)
q q 1 − Π(q 2 )
 + +

+ +

Figure 29: Effect of renormalization in a photon exchange process.

where q is the photon momentum.

Let us now rewrite the denominator on the right hand side of Eq. (7.22) by separating
the divergent part and the finite part in Π. According to the discussion around Eq. (7.18),
this can be achieved by
h  i
1 − Π(q 2 ) = [1 − Π(0)] 1 − Π(q 2 ) − Π(0) + O(α2 ) . (7.23)

Therefore Eq. (7.22) gives

e20 e20 1
2
−→
q q 1 − Π(q 2 )
2

1 e20 1
≃ 2 2
. (7.24)
q 1 − Π(0) 1 − [Π(q ) − Π(0)]
| {z } | {z }
e2 ≡Z3 e20 q 2 −dependence

In the last line of Eq. (7.24) we have underlined two distinct effects in the result we
obtain from renormalization. The first is that the strength of the coupling is modified
to
e20
≡ e2 , (7.25)
1 − Π(0)
from which, by comparison with Eq. (7.9), we identify the renormalization constant Z3 :

Z3 ≃ 1 + Π(0)
α 1
= 1− + ... , (7.26)
3π ε
where in the last line we have used the explicit result for Π(0) in Eq. (7.19). The
coupling e in Eq. (7.25) is the physical coupling, that is, the renormalized coupling. This
is obtained from the unrenormalized one, e0 , via a divergent, but unobservable, rescaling,
according to the general procedure outlined below Eq. (7.3).
The second effect in Eq. (7.24) is that the coupling acquires a dependence on the
momentum transfer q 2 , controlled by the finite part of the self-energy, Π(q 2 ) − Π(0).
This dependence is free of divergences and observable. The q 2 -dependence of the elec-
tromagnetic coupling is a new physical effect due to loop corrections. Using the explicit
expression for Π in Eq. (7.15), we obtain that for low q 2

Π(q 2 ) − Π(0) → 0 for q2 → 0 , (7.27)

and for high q 2
α q2
Π(q 2 ) − Π(0) ≃ ln 2 for q 2 ≫ m2 . (7.28)
3π m
Thus e2 in Eq. (7.25) is the value of the coupling at q 2 = 0; the coupling increases as q 2
increases. Substituting Eqs. (7.25),(7.28) into Eq. (7.24) and rewriting it in terms of the
fine structure, we have for large momenta
α
α(q 2 ) = . (7.29)
1 − (α/(3π)) ln(q 2 /m2 )
The q 2 -dependence of the coupling is referred to as running coupling. We will discuss
this topic further in Sec. 8.
The result for the electromagnetic coupling that we have just found can be viewed as
summing a series of perturbative large logarithms for q 2 ≫ m2 . By expanding Eq. (7.29)
in powers of α, we have
α
α(q 2 ) =
1 − (α/(3π)) ln(q 2 /m2 )
!
α q2 αn n q
2
= α 1+ ln 2 + . . . + ln + ... . (7.30)
3π m (3π)n m2
This is the simplest example of a conceptual framework referred to as resummation in
QED and QCD. The point is that if the result (7.24) for the physical process is expressed
in terms of an expansion in powers of α, as in Eq. (7.30), perturbative coefficients to
higher orders are affected by large logarithmic corrections. On the other hand, one
obtains a well-behaved perturbation series, without large higher-order coefficients, if the
result is expressed in terms of the effective charge α(q 2 ).

7.4 Vertex correction and anomalous magnetic moment

In this section we study the one-loop vertex correction of Fig. 30. In particular we
compute its contribution to the electron’s magnetic moment,
e
µ=g S , (7.31)
2m
where S is the spin operator and g is the gyromagnetic ratio. This computation gives
α
g = gDirac + + O(α2 ) , gDirac = 2 , (7.32)
π
where gDirac = 2 is the prediction from the Dirac equation and α/π is the correction
from the graph in Fig. 30. Higher order corrections arise from multi-loop graphs. The
deviations from the Dirac value are referred to as the electron’s anomalous magnetic
moment.
Let us consider first the Dirac equation coupled to electromagnetism, Eq. (3.20),
and write the magnetic interaction term explicitly. We can recast Eq. (3.20) in the
two-component notation of Subsec. 2.4, including the electromagnetic coupling, as
     
χ m σ · (p − eA) χ
E = . (7.33)
φ σ · (p − eA) −m φ
 q

p p

Figure 30: One-loop vertex correction in QED.

Now substitute the bottom equation in (7.33) into the top equation, use the Pauli σ
matrix relation
σ·a σ·b=a·b+i σ·a∧b , (7.34)
and take the nonrelativistic limit E ≃ m, in which φ ≪ χ. We then obtain that the
action of the hamiltonian on the spinor ψ can be written as
 
e
Hψ ≃ (p − eA)2 − B · 2S ψ , (7.35)
2m
where B = ∇ ∧ A is the magnetic field and S is the spin operator given in terms of the
σ matrices in Eq. (2.57). We recognize that the second term in the right hand side of
Eq. (7.35) is the magnetic interaction
e
−µ · B , with µ = 2S . (7.36)
2m
That is, the Dirac equation prediction for the gyromagnetic ratio g in Eq. (7.31) is
gDirac = 2 . (7.37)
Let us consider now the vertex function Γν (p, p′ ) represented at one loop in Fig. 30.
We can determine the general structure of the vertex function based on relativistic in-
variance and gauge invariance. Because Γν (p, p′ ) transforms like a Lorentz vector, we
can write it as a linear combination of γ ν , pν , p′ν , or equivalently
Γν (p, p′ ) = A γ ν + B(p + p′ )ν + C(p − p′ )ν , (7.38)
where A, B and C are scalar functions of q 2 only (q = p′ − p).
Gauge invariance requires
qν Γ ν = 0 . (7.39)
By dotting qν into Eq. (7.38), the term in B gives zero, and the term in A gives zero
once it is sandwiched between ū(p′ ) and u(p). Thus C = 0. We can further show that
the following identity holds,
1 i
ū(p′ )γ ν u(p) = ū(p′ )(p + p′ )ν u(p) + ū(p′ ) Σνρ qρ u(p) , (7.40)
2m m
where Σ is given in Eq. (2.28), Σνρ ≡ (i/4) [γ ν , γ ρ ]. This implies that the term in (p+p′ )ν
in Eq. (7.38) can be traded for a linear combination of a term in γ ν and a term in Σνρ qρ .
Therefore the vertex function can be decomposed in general as
i
Γν (p, p′ ) = F1 (q 2 ) γ ν + F2 (q 2 ) Σνρ qρ , (7.41)
m
where the scalar functions F1 (q 2 ) and F2 (q 2 ) are the electron’s electric and magnetic
form factors. At tree level, Γν = γ ν , thus F1 = 1 and F2 = 0. In general, F1 and F2
receive radiative corrections from loop graphs and are related to the electron’s charge
and magnetic moment,
F1 (0) = Q , (7.42)
g−2
, F2 (0) =
(7.43)
2
where Q is the electron’s charge in units of e and g is the electron’s magnetic moment in
units of (e/(2m))S where S is the electron spin. F1 (0) is 1 to all orders, that is, radiative
corrections to F1 vanish at q 2 = 0. We next compute the correction to F2 (0) at one loop.
To this end, consider the one-loop graph in Fig. 30. This is given by
Z
d4 k ū(p′ ) γ λ (/k + /q + m) γ ν (/k + m) γλ u(p)
ū(p′ )ieΓν u(p) = e3 . (7.44)
(2π)4 [(k + q)2 − m2 + iε] [k 2 − m2 + iε] [(p − k)2 + iε]

The integral in Eq. (7.44) can be handled starting with the following Feynman’s param-
eterization of the three denominators in the integrand,
1
[(k + q)2 − m2 + iε] [k 2 − m2 + iε] [(p − k)2 + iε]
Z Z Z
1 1 1 2 δ(x1 + x2 + x3 − 1)
= dx1 dx2 dx3
0 0 0 [x1 ((k + q)2 − m2 ) + x2 (k 2 − m2 ) + x3 (p − k)2 + iε]3
Z Z Z
1 1 1 2 δ(x1 + x2 + x3 − 1)
= dx1 dx2 dx3 , (7.45)
0 0 0 (ke 2 − K + iε)3

where in the last line we have set ke = k + x1 q − x3 p, K = m2 (1 − x3 )2 − q 2 x1 x2 .

Next change integration variable k → ke in Eq. (7.44), and note that the numerator
in the integrand can be rewritten according to

γ λ (/k + /q + m) γ ν (/k + m) γλ (7.46)

= γ ν [ke 2 − 2q 2 (1 − x1 )(1 − x2 ) + 2m2 (4x3 − 1 − x23 )] − 4miΣνρ qρ x3 (1 − x3 ) .

Then Eq. (7.44) can be recast in the form

Z 1 Z 1 Z 1
′ ν 2 ′
ū(p )Γ u(p) = −ie ū(p ) dx1 dx2 dx3 2 δ(x1 + x2 + x3 − 1)
0 0 0
Z "
d4 ke ke 2 − 2q 2 (1 − x1 )(1 − x2 ) − 2m2 (1 − 4x3 + x23 )
× γν
(2π)4 [ke 2 − K]3
#
i νρ −4m2 x3 (1 − x3 )
+ Σ qρ u(p) . (7.47)
m [ke 2 − K]3

Comparing Eq. (7.47) with the general decomposition in Eq. (7.41), we see that the two
terms in the second and third line of Eq. (7.47) give one-loop integral representations for,
respectively, the form factors F1 (q 2 ) and F2 (q 2 ). Let us concentrate on the calculation
of F2 :
Z 1 Z 1 Z 1
F2 (q 2 ) = −ie2 dx1 dx2 dx3 2 δ(x1 + x2 + x3 − 1)
0 0 0
Z
d4 ke −4m2 x3 (1 − x3 )
× . (7.48)
(2π)4 [ke 2 − K]3

While the integral for F1 in Eq. (7.47) has divergences that need regularization, the
integral (7.48) for F2 is finite. Let us compute the result for q 2 = 0.
The integration over the four-momentum ke in Eq. (7.48) can be done by using the
transformation of variables ke 0 → −eiπ/2 ke 0 in the integral over the time component of
the momentum. This yields the result
Z
d4 ke 1 i
= − . (7.49)
(2π) [ke 2 − K]3
4 32π 2 K

Then we have (e2 = 4πα)

αZ 1 Z 1 Z 1
m2 x3 (1 − x3 )
F2 (0) = dx1 dx2 dx3 δ(x1 + x2 + x3 − 1)
π 0 0 0 (1 − x3 )2 m2
αZ 1 Z 1−x3
x3
= dx3 dx2
π 0 0 1 − x3
αZ 1 x3 α
= dx3 (1 − x3 ) = . (7.50)
π 0 1 − x3 2π
We thus obtain that the one-loop contribution to the electron’s anomalous magnetic
moment g − 2 = 2F2 (0) is given by
α
g − 2 = 2F2 (0) = . (7.51)
π
8 Renormalization group
Let us discuss renormalization from the standpoint of the renormalization group. We
have seen that renormalization introduces dependence on a renormalization scale µ in
loop calculations. As the value of µ is arbitrary, physics must be invariant under changes
in this scale. This invariance is expressed in a precise manner by the renormalization
group. We will see that by studying the dependence on the renormalization scale µ we
gain insight into the asymptotic behavior of the theory at short distances.

8.1 Renormalization scale dependence and evolution equations

In this section we illustrate how the relation between renormalized and unrenormalized
quantities, applied to a given physical quantity G, can be used to to study the dependence
on the renormalization scale µ and to obtain renormalization group evolution equations.
Renormalizability implies that the divergent dependence in the unrenormalized quan-
tity G0 can be factored out in the renormalization constant Z, provided we re-express
renormalized G in terms of the renormalized coupling and renormalization scale µ,

G0 (pi , α0 ) = ZG(pi , α, µ) . (8.1)

Here pi is the set of physical momenta on which G depends, α is the renormalized

coupling and α0 is the unrenormalized coupling. Because the left hand side in Eq. (8.1)
does not depend on µ,
d
G0 = 0 , (8.2)
d ln µ2
we have
d ∂G ∂G ∂α ∂ ln Z
(ZG) = 0 =⇒ + + G=0 . (8.3)
d ln µ2 ∂ ln µ 2 ∂α ∂ ln µ 2 ∂ ln µ2
By defining
∂α
β(α) = , (8.4)
∂ ln µ2
∂ ln Z
γ(α) = , (8.5)
∂ ln µ2
we can rewrite Eq. (8.3) as
" #
∂ ∂
2
+ β(α) + γ(α) G(pi , α, µ) = 0 , (8.6)
∂ ln µ ∂α
where β(α) and γ(α) are calculable functions of α.
Suppose we measure G at a physical mass-scale Q. Let us rescale by Q the arguments
in G and set
G(pi , α, µ) = F (xi , t, α) , (8.7)
where
pi Q2
xi = , t = ln . (8.8)
Q µ2
In this notation Eq. (8.6) can be written as
" #
∂ ∂
− + β(α) + γ(α) F (t, α) = 0 , (8.9)
∂t ∂α

where from now on we will not write explicitly the dependence on the rescaled physical
momenta xi in F .
Eq. (8.9) is the renormalization group evolution equation, which we can solve with
boundary condition F (0, α) at t = 0, i.e., µ = Q. To do this, we first write the solution
for the case γ = 0 and then generalize this solution to any γ.
For γ = 0 we have
" #
∂ ∂
− + β(α) F (t, α) = 0 (γ = 0) . (8.10)
∂t ∂α

Now observe that if we construct α(t) such that

Z α(t) dα′
t= , (8.11)
α β(α′ )

then any F of the form

F (t, α) = F (0, α(t)) (8.12)
satisfies the equation and the boundary condition.
Eq. (8.11) defines α(t) as an implicit function. To verify that Eq. (8.12) is solution,
note first that the boundary condition at t = 0 is

t=0 , α(0) = α =⇒ F = F (0, α) . (8.13)

Next evaluate the derivative of Eq. (8.11) with respect to t,

1 ∂α(t)
1= , (8.14)
β(α(t)) ∂t

and with respect to α,

1 ∂α(t) 1
0= − . (8.15)
β(α(t)) ∂α β(α)
Then the differential operator in Eq. (8.10) applied to F (0, α(t)) gives
" #
∂ ∂
− + β(α) F (0, α(t))
∂t ∂α
 

∂F  ∂α(t) ∂α(t) 
 
=−  − β(α)  = 0 , (8.16)
∂α(t)  ∂t 
| {z } | ∂α
{z }
β(α(t)) β(α(t))/β(α)

where in the last line we have used Eqs. (8.14),(8.15).

 In the general case γ 6= 0, the solution to Eq. (8.9) is obtained from the γ = 0 answer
(8.12) by multiplication by the exponential of a γ integral, as follows
"Z #
α(t) γ(α′ )
′
F (t, α) = F (0, α(t)) exp dα
α β(α′ )
Z t 
′ ′
= F (0, α(t)) exp dt γ(α(t )) . (8.17)
0

In the second line in Eq. (8.17) we have made the integration variable transformation
using Eq. (8.11). We can verify that Eq. (8.17) is solution by a method similar to that
employed above for the case γ = 0.
Eq. (8.17) indicates that once ultraviolet divergences are removed through renormal-
ization, all effects of varying the scale in F from µ to Q can be taken into account by
i) replacing α by α(t), and ii) including the t-dependence given by the exponential factor
in γ. The latter factor breaks scaling in t, modifying the “engineering” dimensions of
F by γ-dependent terms. For this reason γ is referred to as anomalous dimension. By
expanding the exponential factor in powers of the coupling, we see that this factor sums
terms of the type (αt)n to all orders in perturbation theory. Eq. (8.17) thus provides a
second example, besides that seen in Eq. (7.30) for the electric charge, of perturbative
resummation of logarithmic corrections to all orders in the coupling, giving rise to an
improved perturbation expansion, in which coefficients of higher order are free of large
logarithms.
In QCD the e+ e− annihilation cross section σ(e+ e− → hadrons) is an example of
the γ = 0 case in Eq. (8.12), while deep-inelastic scattering structure functions are an
example of the γ 6= 0 case in Eq. (8.17).

8.2 RG interpretation of the photon self-energy

Let us revisit the analysis of the photon self-energy in Sec. 7 from the standpoint of the
renormalization group. The divergent part of the renormalization constant Z3 computed
in Eq. (7.26) determines the QED β function at one loop.
According to Eq. (8.4), the variation of the coupling α with the energy scale µ is
governed by the β function, calculable as a function of α. In dimensional regularization,
from  ε
α µ 2 = Z3 α 0 , (8.18)
by using Eq. (7.26) we have
  −ε
∂α α 1 2
= −ε 1 − α 0 µ
∂ ln µ2 3π ε
1 2
= α . (8.19)
3π
The leading term of the QED β function at small coupling is given by (Fig. 31),

β(α) = bα2 + O(α3 ) ,

1
b = . (8.20)
3π
Inserting the result (8.20) into Eq. (8.4) gives the differential equation
∂α
2
= bα2 . (8.21)
∂ ln µ
This can be solved by
dα dµ2 1 1 q2
= b =⇒ − + = b ln , (8.22)
α2 µ2 α(q 2 ) α q02
which gives
α
α(q 2 ) = , b = 1/(3π) , (8.23)
1 − bα ln(q 2 /q02 )
that is, the result (7.29) derived directly in Subsec. 7.3.

Figure 31: Small-coupling approximation to the β function in QED.

To sum up, we have found from the analysis of electric charge renormalization in
Sec. 7.3 and in this section that as a result of loop graphs the electromagnetic coupling
is energy-dependent. We can regard this result as illustrating the breaking of scale
invariance as an effect of the quantum corrections taken into account by renormalization.
We start at tree level with a coupling that is scale invariant. Then we include loops.
This implies introducing an unphysical mass scale, such as the renormalization scale
µ, to treat quantum fluctuations at short distances, or high momenta. At the end of
the calculation in the renormalized theory, the unphysical mass scale disappears from
physical quantities. But an observable, physical effect from including loop corrections
remains in the fact that scale invariance is broken. The physical coupling depends on
the energy scale at which we probe the interaction. The renormalization group provides
the appropriate framework to describe this phenomenon, in which the rescalings (7.4) of
the couplings and wave functions, necessary to compensate variations in the arbitrary
renormalization scale, are governed by universal functions, respectively the β and γ
functions (8.4),(8.5) of the theory.

8.3 QCD β function at one loop

We now extend the discussion to the case of renormalization in QCD at one loop, and
determine the one-loop β function.
In the QCD case we assign rescaling relations analogous to those in Eq. (7.4) for the
abelian theory. For wave function and mass renormalization we set
q
A → A0 = Z3 A ,
 q
ψ → ψ0 = Z2 ψ ,
q
c → c0 = Ze3 c ,
Zm
m → m0 = m . (8.24)
Z2

where, in addition to the renormalization constants of the abelian case, we introduce Ze3
for ghost renormalization. For renormalization of quark-gluon, ghost-gluon and gluon
self-coupling vertices we set
q
Z2 Z3 g0 = Z1 g ,
q
Ze3 Z3 g0 = Ze1 g ,
3/2
Z3 g0 = Z1,3 g ,
Z32 g02 = Z1,4 g . (8.25)

As noted in Sec. 6.2, non-abelian gauge invariance requires that the vertices have equal
couplings. This implies relations among the different Z in Eq. (8.25), as follows
s
Ze1 Z1 Z1,3 Z1,4
e
= = = . (8.26)
Z3 Z2 Z3 Z3
In the non-abelian theory, unlike QED, in general one has Z1 6= Z2 . The relations in
Eq. (8.26) can be seen as non-abelian generalizations of the QED result Z1 = Z2 given
in Eq. (7.8).
We can define the renormalized coupling from the quark-gluon vertex. The analogue
of Eq. (8.18) for the QCD case is
 ε Z22
αs µ2 = Z3 αs0 . (8.27)
Z12
Each of the renormalization constants Zi has a perturbation series expansion, with the
coefficients of the expansion being ultraviolet divergent. In dimensional regularization
the ultraviolet divergences appear as poles at ε = 0, so that the Zi have the form
1
Zi = 1 + αs ci + finite , (8.28)
ε
where the coefficients ci of the divergent terms are to be calculated. By using Eqs. (8.27)
and (8.28), the β function is given by
∂αs
β(αs ) =
∂ ln µ2
 −ε
= −εαB µ2 [1 − 2(Z1 − 1) + 2(Z2 − 1) + (Z3 − 1)]
1
= 2αs2 (c1 − c2 − c3 ) . (8.29)
2
 (1)

(2)

(3)

Figure 32: One-loop corrections to (1) quark-gluon vertex; (2) quark self-energy; (3)
gluon self-energy.

The Feynman graphs contributing to c1 , c2 and c3 are the one-loop graphs for, re-
spectively, the quark-gluon vertex renormalization, quark self-energy renormalization,
and gluon self-energy renormalization, and they are shown in Fig. 32. The calculation
of these graphs proceeds similarly to the calculation done in Sec. 7.2 for the fermion
loop contribution. By computing these graphs, working in Feynman gauge ξ = 1 (as in
Fig. 19), we obtain the results for the renormalization constants Zi ,
αs 1
Z1 = 1 − (CF + CA ) , (8.30)
4π ε
αs 1
Z2 = 1 − CF , (8.31)
4π ε
αs 1 5 4
Z3 = 1 + ( C A − Nf T F ) , (8.32)
4π ε 3 3
where Nf is the number of quark flavors (Sec. 6.1), and the color charge factors are given
in Sec. 6.3,
N2 − 1 4 1
CA = N = 3 , CF = = , TF = . (8.33)
2N 3 2
Note from the expression for Z3 that the second term in the bracket in Eq. (8.32) is
the term computed in Sec. 7.2 from the fermion loop graph, which, in the abelian limit
Nf TF → 1, gives the QED contribution −α/(3πε) of Eq. (7.26).
From Eqs. (8.30)-(8.32) we read the coefficients ci to be put into Eq. (8.29) to deter-
mine the β function. We obtain
 
1 α2 1 5 1 4
β(αs ) = 2αs2 (c1 − c2 − c3 ) = 2 s −CF − CA + CF − CA + Nf T R
2 4π 2 3 2 3
 
αs2 11 4 α2
= − CA + Nf TR = − s (11N − 2Nf ) . (8.34)
4π 3 3 12π
Eq. (8.34) shows that for Nf < 11N/2 the β function in the non-abelian case has negative
sign at small coupling (Fig. 33),
β(αs ) = −β0 αs2 + O(αs3 ) , (8.35)
where
1
β0 = (11N − 2Nf ) . (8.36)
12π
This behavior of the β function is opposite to the behavior of the β function in QED,
Eq. (8.20) (Fig. 31).

Figure 33: Small-coupling approximation to the β function in QCD.

The behavior of the β function in Eqs. (8.35),(8.36) implies that QCD is asymp-
totically free, i.e., weakly coupled at short distances. By inserting Eq. (8.35) into the
renormalization group evolution equation,
∂αs
2
= β(αs ) ≃ −β0 αs2 , (8.37)
∂ ln µ
and solving Eq. (8.37), we obtain
αs (µ2 )
αs (q 2 ) = , (8.38)
1 + β0 αs (µ2 ) ln q 2 /µ2
where β0 is given in Eq. (8.36). Eq. (8.38) expresses the q 2 -dependence of the QCD
running coupling at one loop. The QCD coupling decreases logarithmically as the mo-
mentum scale q 2 increases. This property is the basis for the perturbative calculability
of scattering processes due to strong interactions at large momentum transfers.

8.4 The QCD scale Λ

From Eq. (8.38) we also see that QCD becomes strongly coupled in the infrared, low-
momentum region. This behavior is opposite to that in QED. In the QED case, taking
q0 ∼ m in Eq. (8.23), with m the electron mass, we have strong coupling in the ultraviolet
region for
q 2 ∼ m2 e3π/α , (8.39)
corresponding to enormously high energies.
In the QCD case, calling Λ the mass scale at which the denominator in Eq. (8.38)
vanishes, we have
Λ2 2 ))
1 + β0 αs (µ2 ) ln =0 =⇒ Λ2 = µ2 e−1/(β0 αs (µ . (8.40)
µ2
In QED and QCD we thus get the different pictures in Fig. 34 for the scale, referred to
as the Landau pole, at which the coupling becomes strong.

α α

q2 q2

QCD QED

Figure 34: Landau pole pictures in QCD and in QED.

The scale Λ in Eq. (8.40) is renormalization-group invariant, i.e., it is independent of

µ. Under transformations

µ2 −→ µ′2 = µ2 et ,
αs (µ2 )
αs (µ2 ) −→ αs (µ′2 ) = , (8.41)
1 + β0 αs (µ2 )t

we have
′2 ))
Λ2 −→ µ′2 e−1/(β0 αs (µ ,
2 )t)/(β 2 )) 2
= µ2 et e−(1+β0 αs (µ 0 αs (µ
= µ2 et e−1/(β0 αs (µ )) e−t = Λ2 . (8.42)

The scale Λ is a physical mass scale of the theory of strong interaction. Its measured
value is about 200 MeV.

β β

α α

(a) (b)
Figure 35: (a) Trivial and (b) nontrivial ultraviolet fixed points of the β function.

The running coupling (8.38) can be equivalently expressed in terms of Λ,

αs (µ2 )
αs (q 2 ) =
1 + β0 αs (µ2 ) [ln(q 2 /Λ2 ) − 1/(β0 αs (µ2 ))]
1
= . (8.43)
β0 ln(q 2 /Λ2 )
The rewriting (8.43) of Eq. (8.38) makes it manifest that the running coupling αs does
not depend on the choice of the renormalization scale µ.
Remark. The zero of the QCD β function at the origin, sketched in Fig. 35a, is
responsible for the theory being weakly coupled at short distances. This behavior is
referred to as a trivial ultraviolet fixed point. A behavior such as that in Fig. 35b
(nontrivial ultraviolet fixed point), leading to strong coupling at short distances, is in
principle possible but not realized in nature as far as we know. This is the reason why
renormalization can be understood perturbatively and Feynman graphs provide a very
effective method to investigate physical theories of fundamental interactions.
Acknowledgments
I thank the School Director, Mark Thomson, and the School Administrators, Gill Birch
and Jacqui Graham, for the excellent organization and for making this School a very
pleasant event. I am grateful to the other members of the teaching staff and to the
students for the nice atmosphere at the School and for interesting conversations. Many
thanks to Nick Evans for providing teaching material from past editions of the School
and for advice on the content of this course.
References
[1] J D Bjorken and S D Drell, Relativistic Quantum Mechanics, McGraw-Hill 1964
I J R Aitchison and A J G Hey, Gauge theories in particle physics, 2nd edition Adam
Hilger 1989
F Halzen and A D Martin, Quarks and Leptons, Wiley 1984

[2] M E Peskin and D V Schroeder, An Introduction to Quantum Field Theory, Addison

Wesley 1995
F Mandl and G Shaw, Quantum Field Theory, Wiley 1984
C Itzykson and J-B Zuber, Quantum Field Theory, McGraw-Hill 1987