INTRODUCTION TO PERTURBATIVE QCD∗

P. NASON
INFN, Milan, Italy

Abstract
In this lecture notes I give an introduction to perturbative QCD, that should ad-
dress both theoretical and experimental physics students. I illustrate the basic
features of the theory, by discussing few examples in e+ e− physics, deep-
inelastic scattering, and hard production phenomena in hadron collisions.

Contents

1 STRONG INTERACTIONS 2

2 MOTIVATIONS FOR QCD 2

2.1 Hadron Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 The QCD Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 AN ILLUSTRATION OF ASYMPTOTIC FREEDOM 10

3.1 Renormalization group and asymptotic freedom . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Relation among the couplings with different number of light flavours . . . . . . . . . . . 15
3.3 State of the art in the beta function and R . . . . . . . . . . . . . . . . . . . . . . . . . 16

4 JETS IN e+ e− ANNIHILATION 18
4.1 Sterman–Weinberg jets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2 A comparison with QED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3 Shower Monte Carlo programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.4 More jet definitions and shape variables . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.5 Thrust as an example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5 PROCESSES WITH HADRONS IN THE INITIAL STATE 30

5.1 The naive parton model formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.2 Does the Parton Model survive radiative corrections? . . . . . . . . . . . . . . . . . . . 35
5.3 Derivation of the singular part of the cross section . . . . . . . . . . . . . . . . . . . . . 36
5.4 Effects due to the emission of a collinear gluon . . . . . . . . . . . . . . . . . . . . . . 38
5.5 Failure of the parton model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.6 The evolution equations in the general case . . . . . . . . . . . . . . . . . . . . . . . . 41
5.7 Sum rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
∗
Lecture notes for the XI Jorge André Swieca Summer School, Particles and Fields, January 14-27 2001, Campos do Jordão,
SP, Brazil. This notes are an updated and revised version of the lecture notes for the 1997 European School of High-Energy
Physics, 25 May - 7 Jun 1997, Menstrup, Denmark, report CERN-98-03”
 5.8 Scheme dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.10 How solid is the Factorization Theorem? . . . . . . . . . . . . . . . . . . . . . . . . . . 44

6 DEEP INELASTIC SCATTERING 44

7 QCD IN HADRONIC COLLISIONS 48

7.1 The kinematic variables for hadronic collisions . . . . . . . . . . . . . . . . . . . . . . 49
7.2 Total cross section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.3 Typical inelastic processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.4 Looking for hard processes in hadronic collisions . . . . . . . . . . . . . . . . . . . . . 50
7.5 Jets at Hadron Colliders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.6 Production of W , Z, and Drell-Yan pairs . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7.7 Heavy Flavour production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

8 CONCLUSIONS 56

1 STRONG INTERACTIONS
Strong interactions are characterized at moderate energies by the presence of a single dimensionful scale,
of the order of few hundred MeV, a scale that we will call in the following ΛS . No hint to the presence
of a small parameter, in which to develop a perturbative expansion, is present in the strong interaction
world. Thus, typical cross sections are of the order of 10 millibarns (corresponding roughly to 1/Λ2S ),
the width of hadronic resonances is of order ΛS , and the size of a baryon is typically of the order of
1/ΛS . This is very much different from the case of electromagnetism and of weak interaction, where all
reactions can be viewed as originating from a weakly coupled point-like vertex, the fermion–fermion–
photon vertex in electrodynamics, and the four fermion vertex in weak interactions. The development
of a model of strong interactions has therefore followed a rather intricate path. Aside from what can be
inferred from symmetry properties, S-Matrix models were developed in the 60’s, since the general feeling
prevailed that it was impossible to describe strong interactions using a field theoretical framework similar
to the one used for QED. Dual models, which eventually gave origin to string theories, were discovered
precisely in this context, but failed to give a consistent explanation of strong interaction dynamics.

2 MOTIVATIONS FOR QCD

Today we have a satisfactory model of the strong interaction, which is given in terms of a non–Abelian
gauge theory. The main motivations for this model are essentially the following.

2.1 Hadron Spectrum

The hadron spectrum can be completely classified from the following assumptions

1. Hadrons are made up of spin 12 quarks. The charge and masses of the known quarks are given in
table 1. One usually refers to u, d, s, c, b and t as “flavours”, and commonly refers to u, d and s as
the light flavours, and c, b and t as heavy flavours.

2. Each quark flavour comes in 3 colours. Thus, quark fields are spinors, and carry a flavour and a
(f )←flavour
colour index: ψi←colour .

3. The SU(3) symmetry acting on colour is an exact symmetry.

2
 Electric Charge= 23 e up charm top
m= few MeV ≈ 1.5 GeV ≈170 GeV
Electric Charge= − 13 e down strange bottom
m= few MeV few hundred MeV ≈ 5 GeV

Table 1: Known quarks

4. Observable hadrons are neutral in colour, in the sense that they are colour singlets under the SU(3)
colour group (“singlet” means invariant under the action of the group).

The SU(3) group is the group of 3 × 3 complex unitary matrices U with unit determinant

U †U = 1 , det U = 1 , (1)

that act on the quark fields according to

X
ψi → Uik ψk . (2)
k

Invariants can be easily formed out of quark–antiquark states

à !
X X X X † X
ψi∗ ψi → Uij∗ ψj∗ Uik ψk = Uji Uik ψj∗ ψk = ψk∗ ψk , (3)
i ijk kj i k

which gives us the possibility of forming integer spin color singlet states with a quark and an antiquark.
We can form colour singlet also from three-quark states
X X X ′ ′ ′
ǫijk ψi ψj ψk → ǫijk Uii′ Ujj ′ Ukk′ ψi′ ψj ′ ψk′ = ǫi j k ψi′ ψj ′ ψk′ (4)
ijk ijk,i′ j ′ k′ i′ j ′ k′

where the last equality is a consequence of the identity

X ′ ′ ′
ǫijk Uii′ Ujj ′ Ukk′ = det U ǫi j k (5)
ijk

and det U = 1 for SU(3) matrices. Therefore we have the possibility of forming colour neutral, spin 1/2
hadrons out of three quarks. The most important hadron multiplets are displayed in fig. 1. Multiplets
are classified according their spin, and their transformation properties under the flavour group. Each
multiplet contains particles with similar properties. Observe that we need colour if we want a particle
like the ∆++ , which is made of three up quark with the same flavours and same spin, to have similar
properties to the Σ0 , which has three different flavours. In fact, if we didn’t have colour, because of
the Pauli principle, the spatial wave function of the ∆++ should be antisymmetric, while that of the Σ0
could very well be symmetric. With colour, instead, the colour wave-function itself is antisymmetric, and
so there is no problem to have the particle of the multiplet all in a symmetric spin, flavour, and spatial
wave-function.
It can be shown that in order to form an SU(3) singlet in a system with nq quarks and nq̄ antiquark,
we have the constraint
nq − nq̄ = n × 3 (6)
with n integer. It is a simple exercise to show that because of this condition observable hadrons must
have integer charges.

3
 Fig. 1: Hadron spectrum.

Fig. 2: Deep inelastic scattering.

2.2 Scaling
Scaling was first observed in deep inelastic scattering experiments at SLAC (Stanford Linear Accelerator
Center, Stanford, California), around 1968. The deep inelastic scattering process, depicted in fig. 2, is
the collision of a lepton (an electron in the SLAC case) with a nucleon target, which fragments into a
high multiplicity, massive final state. The scattering process kinematics can be defined by the following
dimensionless variables
Q2 q·p
xBj = y= . (7)
2p · q k·p

4
where Q2 = −q 2 . The value xBj = 1 corresponds to elastic scattering. In fact
2
MX = (q + p)2 = −Q2 + m2p + 2ν = 2ν(1 − xBj ) + m2p . (8)

Scaling means that the differential cross section, when expressed in terms of these dimensionless param-
eters, in the limit of high energy with x and y fixed, scales like the energy in the process, according to its
canonical dimension
dσ 1
∝ 2. (9)
dx dy Q
This property is quite remarkable, since the right hand side does not depend upon ΛS , like most moderate
energy cross sections, and it looks more like the behaviour one may find in a renormalizable field theory
with a dimensionless coupling, like electrodynamics. Even more spectacular scaling phenomena are
observed in e+ e− annihilation, where the total hadron production cross section becomes proportional to
the muon pair cross section at high energies.
The discovery of scaling phenomena in deep inelastic scattering and in e+ e− annihilation, has
given a strong evidence that if a field theory was to describe strong interactions, it had to be weakly
coupled at high energies, that is to say, it had to be “asymptotically free”. The only known asymptotically
free four–dimensional field theories are the non–Abelian gauge theories. It becomes therefore natural to
attempt to describe the hadronic forces by using an SU(3) non–Abelian gauge theory, coupled to the
colour quantum number. This is also hinted by the fact that the condition of colour neutrality of the
hadron spectrum must have a dynamical origin.

2.3 The QCD Lagrangian

The QCD Lagrangian reads
1 X (f ) ¡ ¢ (f )
L = − Faµν Fµνa
+ ∂ − mf )δij − gS taij A
ψ̄i (i/ / a ψj
4
f
X
a
Fµν = ∂µ Aaν − ∂ν Aaµ − gS fabc Abµ Acν . (10)
b,c

Sum over repeated Lorentz and colour indices is always assumed. The sum over different flavours is
explicitly indicated. The symbols taij are the SU (3) generators and the fabc are the structure constant of
the SU (3) algebra. The matrices ta form a complete basis of traceless 3 × 3 matrices. There are 8 such
matrices, and therefore there are 8 gluons. The basis is chosen in such a way that
³ ´ 1
Tr ta tb = δ ab (11)
2
The symbols f are then defined by (square brackets indicate the commutator)

[ta , tb ] = if abc tc (12)

I also give the important property (which follows from completeness, tracelessness and relation (11))
µ ¶
X 1 1
taij takl = δil δkj − δij δkl . (13)
a
2 3

Equation 13 is all we need to compute colour factors for feynman graphs.

The colour structure of the Lagrangian may seem complicated at first sight. One simple way to
look at it, is to think of quarks as objects having 3 colour states. The gluon can be thought as carrying the
combination of a colour and an anticolour, except that out of the nine possible combinations the “neutral”
one, formed by the sum of all equal colour-anticolour pairs is subtracted away. Figure 3 shows how to

5
 fermion
gluon

 
1  1
Fermion-Gluon Vertex (ta )

√  − 
2 3 

 
 
 
1 
3-Gluon Vertex (f abc )

√  − 
2



 

→ =3

 
1  1
→ √  − =0

2 3

 
1 1 1 1 
→  − − + =4
2 3 3 9 

Fig. 3: Colour Feynman rules for QCD

compute colour factors by using this intuitive point of view. The Feynman rules for the QCD Lagrangian
are given in fig. 4.
The QCD Lagrangian is very similar to the QED Lagrangian. The Feynman rules are also very
similar. The most apparent difference is due to the fact that the fermions carry a new quantum number,
the color (the indices i, j = 1, 2, 3 in eq. (10)). Also the gluons carry a colour related quantum number.
Unlike the case of QED, therefore, the gluons are charged, and can emit other gluons.
As in the case of electrodynamics, one defines the strong coupling constant

gS2
αS = . (14)
4π
As we shall see in the following, this coupling constant has a strength that depends upon the energy scale

6
 pα pβ
· ¸
ab αβ i
=δ −g + (1 − λ) 2 2
p + iǫ p + iǫ
i
= δ ab
p2 + iǫ
¯
ik i ¯
=δ ¯
/ − m + iǫ ¯mn
p

h i
= −gS f abc g αβ (p − q)γ + g βγ (q − r)α + g γα (r − p)β

³ ´
= −igS2 f xac f xbd g αβ g γδ − g αδ g βγ
³ ´
−igS2 f xad f xbc g αβ g γδ − g αγ g βδ
³ ´
−igS2 f xab f xcd g αγ g βδ − g αδ g βγ

= gS f abc q α

= −igS taki γmn

α

7
Fig. 4: Feynman rules for QCD
µ of the process in which enters. In leading order
1
αS = µ 2 (15)
b0 log Λ 2

where
11 CA − 4 TF nf
b0 = . (16)
12π
where TF = 1/2 and CA = N for SU(N) (3 for SU(3)) and nf is the number of flavours. Thus Λ is the
parameter that characterizes the QCD coupling constant.

2.4 Symmetries
We know that the strong interaction world has a very good symmetry property, the isospin symmetry.
Particles in the same isospin multiplet, like the proton and the neutron, or the charged and neutral pions,
have nearly the same mass. Furthermore, the Wigner-Eckart theorem can be used to relate decay and
scattering processes which are connected by isospin transformations. This symmetry properties must be
present in some way in the fundamental QCD Lagrangian, whose fermionic sector is given by
X (f ) ¡ ¢ (f )
LF = ψ̄i (i/ ∂ − mf )δij − gtaij A
/ a ψj . (17)
f,ij

An isospin transformation acts on the quark field as a unitary matrix

X ′ ′
ψ (f ) → U f f ψ (f ) (18)
f′

where f and f ′ are restricted to the up and down flavours, and U is a unitary two dimensional matrix.
By a simple exercise, one can verify that, in order for the fermionic Lagrangian to be invariant under
the isospin transformation, we must have either mu = md or mu , md → 0. The distinction of the two
possibilities is a physical one. It can be phrased as follows: if the up and down masses are of the order
of the QCD scale Λ or larger, then they must be nearly equal in order for the isospin symmetry to work.
Alternatively, the up and down masses must be much smaller than Λ. The first possibility is not very
appealing from a theoretical point of view. From what we know from the theory of weak interactions,
particles belonging to different families have different masses. It would be very hard to justify the fact
that two quark flavours have equal masses while all the others are very different. In fact, there is a large
body of evidence that favours the second possibility, that is to say, that the up and down quark masses
are very small. This fact has a few remarkable consequences, due to the fact that, for small masses, the
QCD fermionic Lagrangian has a much larger symmetry than isospin alone. In order to see this fact, let
us define left and right-handed field components
1 1
ψL = (1 − γ5 )ψ , ψR = (1 + γ5 )ψ (19)
2 2
and substituting ψ = ψL + ψR in the fermionic Lagrangian we have (suppressing colour indices)
X n (f ) (f ) (f ) (f )
o
LF = ψ̄L (/ ∂ − gta A
/ a ) ψL + ψ̄R (/ ∂ − gta A
/ a ) ψR
f
³ ´
(f ) (f ) (f ) (f )
X
− mf ψ̄R ψL + ψ̄L ψR . (20)
f

Terms that mix left and right components in the kinetic energy, and terms diagonal in the left and right
component of the mass terms are absent because of the following elementary identities
1 1
ψL = (1 − γ5 ) ψL ψR = (1 + γ5 ) ψR (21)
2 2

8
 1 1
ψ̄L = ψ̄L (1 + γ5 ) ψ̄R = ψ̄R (1 − γ5 ) (22)
2 2
and from the fact that γ5 anticommutes with γµ . If we could neglect the fermion masses the Lagrangian
would have the large symmetry

SUL (N ) × SUR (N ) × UL (1) × UR (1) (23)

where N is the number of flavours. In fact, the transformation

(f )
X f f ′ (f ′ )
ψL → eiφL UL ψ L
f′
(f ) ′ (f ′ )
URf f ψL
X
ψR → eiφR (24)
f′

where UL and UR are (independent) matrices of SU(N ), leaves the Lagrangian invariant. The phase fac-
tors constitute the two U(1) groups. The isospin symmetry group is a subgroup of the above, also called
the vector subgroup, characterized by equal transformation matrices for the left and right components.
Besides the isospin transformations, there are other independent symmetry transformations, in which the
left and right-handed component transform with matrices that are the inverse of each other. These are
called axial transformations (they do not form a subgroup by themselves). In the following, I will only
state what happens of all these symmetries, without giving detailed explanations

• The vector SU(N) subgroup is realized in the spectrum. It is the observed isospin symmetry. The
U(1) vector subgroup is a phase symmetry related to baryon number conservation.

• The axial U(1) symmetry does not survive quantization, because of the so-called triangle anomaly.
This symmetry is simply not there in the full theory.

• The remaining axial transformations are broken symmetries. The Goldstone bosons of these bro-
ken symmetries are the pion fields.

Goldstone bosons are massless particles, while the pions are not. This is a consequence of the fact that
the axial symmetries are only approximate, due to the fact that the quark masses are not strictly zero.
Thus, by assuming that the up and down quark masses are small, we explain the presence of isospin
symmetry, as well as the lightness of the pions. Other dynamical predictions follow, like relations among
the low energy scattering properties of the pions and the pion decay constant. The interested reader can
find many good references where to study this subject [1, 2, 3].

2.5 Summary
In summary, by accepting QCD as the fundamental theory of strong interactions we can

• Explain the low energy symmetry properties, and give a justification of the observed spectrum.

• Explain scaling phenomena at high energies.

• Leave Weak interactions in peace. The QCD colour group commutes with the electroweak group
SU(2)×U(1). Since the electroweak interactions are less symmetric (they break parity and CP),
this guarantees that there is no mixing between electroweak and strong interactions that enhances
the parity–violating effects (giving rise, for example, to parity violating interactions of size αew αS
instead of αew /MW 2 ) or flavour changing neutral current effects.

• Give a description of the hadronic forces which is similar to electroweak forces, thus opening the
possibility of a uniform description of the forces in nature in terms of gauge theories (unification).

9
 There are two common points of view among physicists, with regard to QCD.
Many believe that QCD is an extremely well established theory, much better established than the
Electro-Weak theory. In fact, the Lagrangian is fully specified in term of a single parameter. Remember,
in fact, that quark masses have electroweak origin, and are related to the Yukawa coupling and to the
electroweak symmetry breaking. In Electroweak theories, on the other hand, we have lots of parameters
and quite a few alternatives are possible for the symmetry breaking sector.
Others believe that Electro-Weak theories are much better established. In fact, we can compute
every accessible phenomenon we like with great accuracy, and seek accurate comparisons with experi-
mental results. On the other hand, in QCD, we are unable to explain rigorously even basic phenomena
like colour confinement, and perturbative calculations rely upon unproven assumptions.
The first point of view can be stated by simply saying that QCD must be right because we cannot
think of anything else that is even plausible as a theory of strong interaction. The second point of view is
more humble, and assumes that in order to establish a physical theory one must make testable predictions,
and compare them with experiments.
Thus, we find that essentially no viable alternative to QCD have been formulated so far, and yet
there is a huge ongoing effort in theoretical and experimental physics aimed at testing the predictions of
QCD.
At low energy, QCD is a strongly interacting theory. Besides the phenomenological results that
follow from its symmetry properties, the only known way to perform calculations in this regime is by
computer simulation of QCD on a lattice, that is to say on a finite and discretized model of space-time.
This approach is bound to improve as time goes by, since people become more and more clever, and
computers become more and more powerful.
At high energy, in many cases, standard perturbative methods can be applied. In these lectures I
will deal mostly with the perturbative applications of QCD. We will see that, even at high energy, the
application of perturbative techniques is not straightforward. In fact, we will be able to perform calcula-
tions only when the long distance (low energy) part of the process we examine has no or little influence
upon the quantity we want to compute. In the following, I will illustrate the basics of perturbative QCD
by examining the process of hadrons production via the annihilation of an e+ e− pair at high energy. This
process is particularly simple, since no strongly interacting particles appear in the initial state.

3 AN ILLUSTRATION OF ASYMPTOTIC FREEDOM

We will now introduce the basic features of QCD via the simplest process in which it can be applied, that
is to say the production of hadrons in e+ e− annihilation. By studying this process we will illustrate the
remarkable property of asymptotic freedom, and its physical implications.
We are considering the process depicted in fig. 5. The production of hadrons takes place via the

Fig. 5: Electron–positron annihilation into hadrons.

production of a virtual photon, or of a real or virtual Z boson. From the point of view of QCD, the

10
decay of a virtual photon, or of a W or Z boson, are very similar, and in fact strong corrections to these
processes are given by essentially the same formulae. For simplicity, however, we can always think
about the decay of a virtual photon. We will begin by attempting to compute the total cross section for
the decay of a virtual photon, with a virtuality (q 2 ) much larger then typical hadronic scales. Our attempt
will be extremely crude. We will simply use the QCD Lagrangian and the corresponding Feynman rules,
and try to compute the cross section order by order in the strong coupling constant. The prediction at
zeroth order in the strong coupling comes simply from diagram a of fig. 6. It is usually expressed in

Fig. 6: Diagrams for the QCD calculation of R(e+ e− → Had.) up to the order αS .

terms of the ratio of the hadronic cross section divided by the cross section for the production of a µ+ µ−
pair. It is given by
σ(γ ∗ → hadrons) X
R0 = = 3 c2f (25)
σ(γ ∗ → µ+ µ− )
f

where f runs over the quark flavour species, and cf is the electric charge of the quark of flavour f in
units of the electron charge. The factor of 3 accounts for the fact that there are three colours for each
quark. The sum extends to all the flavours that can be produced at the given energy. The formula is valid
in all cases when we can neglect quark masses. Near the threshold for heavy quark production one must
include a correction factor, which in the general case of a vector boson decay, yelds
s à !
X 4m2f 2m2f
R0 = 3 1− 1+ c2f (26)
s s
f

Corrections of order αS to R can be computed in a straightforward way. The relevant contributions

come from the interference of the virtual diagram b with diagram a, plus the square of the real emission
graphs c + d. There are also diagrams with self–energy on the fermion lines, not shown in the figure,
that should be included with the appropriate weight. The result turns out to be completely finite. All
ultraviolet divergences that arise in intermediate steps of the calculation cancel among each other. This
is a consequence of the fact that the electromagnetic current is a conserved current, and therefore it is
not renormalized by strong interactions. Other kind of singularities arise in intermediate steps of the
calculation, namely soft and collinear singularities. They all cancel in the total. Their meaning will be
discussed further on. The corrected value of R becomes
³ αS ´
R = R0 1 + . (27)
π

11
If we go on, and compute the corrections of order αS2 something new happens. We find ultraviolet
divergences that do not cancel, and the result is

M 2 ³ αS ´2
µ · ¸ ¶
αS
R = R0 1 + + c + πb0 log 2 (28)
π Q π
where M is the ultraviolet cutoff (for those who are familiar with dimensional regularization, the cutoff
scale in d = 4 − 2ǫ dimensions is M = µ exp 1ǫ ), and
33 − 2nf
b0 = (29)
12π
and nf is the number of light flavours. The divergence is dealt with the usual prescription of renormal-
ization. We define a renormalized charge, function of an arbitrary scale µ,

M2 2
αS (µ) = αS + b0 log α (30)
µ2 S
and express the result in terms of αS (µ) instead of αS . We obtain then
à ¶ !
µ2 αS (µ 2
· ¸µ
αS (µ)
+ O αS (µ)3 .
¡ ¢
R = R0 1 + + c + πb0 log 2 (31)
π Q π

The formula for R is now finite. The theory of renormalization guarantees that with this procedure we
can remove the divergences from all physical quantities. This implies that the one loop divergence of
any physical quantity which in lowest order has the value AαSn must have the form nAb0 log M 2 αSn+1 .
Observe that, as a consequence of this procedure, we end up expressing our results in terms of a coupling
constant which is function of a scale.

3.1 Renormalization group and asymptotic freedom

I will now give a general and abstract description of the renormalization group and asymptotic freedom.
From the following discussion it should be clear that the existence of the renormalization group follows
from the property of renormalizability of field theory, and that asymptotic freedom is a possible con-
sequence of the renormalization group. I will not give any technical details on the computation of the
renormalization group flow (i.e. of the so called β function), which can be found in many good textbooks.
In field theories we encounter ultraviolet divergences, which in renormalizable theories can be
removed by a suitable redefinition of the coupling constants and the fields. In the simplest case of a
theory characterized by a single coupling constant, renormalizability can be stated in the following way.
A physical quantity G will be given in such a theory as a power expansion in the coupling α (which we
will assume to be dimensionless), with possibly UV divergent coefficients. We will write:

G = G(α, M, s1 . . . sn ) , (32)

that is to say, G depends upon the coupling, the ultraviolet cutoff M , and some invariants s1 . . . sn
constructed out of the momenta and masses involved in the process in question. Renormalizability means
that I can define a renormalized coupling αren

αren = α + c1 α2 + c2 α2 + . . . (33)

with
ci = ci (M/µ) (34)
in such a way that
G(α, M, s1 . . . sn ) = G̃(αren , µ, s1 . . . sn ) . (35)

12
So, the physical quantity can be expressed in term of the renormalized coupling, the finite scale µ and the
invariants, in terms of a finite function. In other words, all the divergences have been reabsorbed in the
renormalized coupling. The finite scale µ has to be introduced in order for the dimensionless coefficients
ci to depend upon the dimensional quantity M . We will also write

αren = αren (α, M/µ) , α = α(αren , M/µ) . (36)

and
G(α(αren , M/µ), M, s1 . . . sn ) = G̃(αren , µ, s1 . . . sn ) . (37)
Therefore, renormalizability means that by a redefinition of the coupling of the form (36), eq. (37) holds
for all physical quantities. The same redefinition of α makes all physical quantities independent of the
cutoff.
In the redefinition of eq. (36) we are forced to introduce a scale µ. If we change µ and αren by
keeping α and M fixed, the physics remains invariant, because physical quantities, to begin with, are
functions of α and M only. Let us study the infinitesimal transformations dαren dµ2 that leave α and M
fixed. We must have
∂α(αren , M/µ) ∂α(αren , M/µ) 2
dαren + dµ = 0 . (38)
∂αren ∂µ2
Since physical quantities remain the same under this change, we must also have

∂ G̃(αren , µ, p1 . . . pn ) ∂ G̃(αren , µ, p1 . . . pn ) 2
dαren + dµ = 0 . (39)
∂αren ∂µ2
From equations (38) and (39) we get

dαren
2
µ2 ∂µ∂ 2 α(αren , M/µ) µ2 ∂µ∂ 2 G̃(αren , µ, s1 . . . sn )
µ =− ∂ =− ∂ (40)
dµ2 ∂α α(αren , M/µ) ∂α G̃(αren , µ, s1 . . . sn )
ren ren

from which it follows that

dαren
µ2 = β(αren ) (41)
dµ2
where β does not depend upon s1 . . . sn , M or µ. Observe that β does not depend upon M , because M
does not appear on the right hand side of the second equality of (40), it cannot depend upon s1 . . . sn
because they do not occur on the right hand side of the first equality. Finally, it could only depend upon
µ. But µ is dimensionful, while β is obviously dimensionless, and so it cannot even depend upon µ.
Using the expression
2
α(αren , M/µ) = αren + c1 (M/µ)αren + ... (42)

we find
2 ∂
β(αren ) = αren µ2c1 (M/µ) + . . . (43)
∂µ2
Comparing this equation with eq. (30), we immediately get
2
β(αren ) = −b0 αren + ... . (44)

and therefore
d 2
αS (µ) = −b0 αren + ... (45)
d log µ2
which characterizes the evolution of the coupling constant as a function of the scale µ. Equation (45) can
be also written, at the lowest relevant order
d 1
= b0 (46)
d log µ2 αS (µ)

13
which can be easily solved to give

1 µ2 1
= b0 log 2 + . (47)
αS (µ) µ0 αS (µ0 )

Without loss of generality, the solution can be written

1 µ2 1
= b0 log 2 ⇒ αS (µ) = (48)
αS (µ) Λ b0 log µ2 /Λ2

where Λ plays the role of an integration constant. In QCD, b0 is positive, and eq. (48) makes sense only
for µ > Λ. One is tempted to infer that Λ is the value of µ at which the coupling constant becomes
infinite. In fact, this identification is superficial. When the coupling constant starts to be large, we can
no longer trust the perturbative expansion, and the above equation has been derived only at the lowest
order in perturbation theory. It is better therefore to think of Λ as the scale parameter of the theory which
defines the value of αS at large scales. In other words, Λ is defined only through the formula for αS (µ),
and this formula has a meaning only for large µ.
QED is very similar to QCD in many respects, and one may wander why we never talk about a
ΛQED analogous to the Λ in QCD. In fact, the basic difference between QED and QCD is the value of
b0 . We have
4nf
bQED
0 =− , (49)
12π
a negative value. The expression for the running coupling in QED is then

1 µ2
= bQED
0 log . (50)
αQED (µ) Λ2QED

The expression in eq. (50) makes sense only for µ ≪ Λ (so that the right hand side is positive), while the
expression in eq. (48) makes sense only if µ ≫ Λ. In other words, QCD is a weakly coupled theory at
high energy, while QED is weakly coupled at low energy. This is the content of the statement that QCD
is asymptotically free, while QED is not. The scale at which QED becomes strongly coupled is obtained
by solving the equation
1 m2e
= bQED
0 log . (51)
αQED (me ) Λ2QED
which gives
bQED
µ ¶
0
ΛQED = me exp − . (52)
αQED (me )
This formula is valid only if all charged fermions have the same mass, equal to me , and the same charge.
However, even if one does a more accurate job, the basic result is that ΛQED is an astronomic scale,
and this is the reason why we never talk about it. Notice that this fact indicates that QED cannot be a
fundamental theory. The existence of a high scale at which the theory becomes strongly coupled makes
it impossible to measure the basic vertex of QED at short distance, which is somewhat of a contradiction,
since we assume that we know the local Lagrangian of the theory.
We have now discussed the evolution of the coupling constant at the leading order level. The
content of the theory of renormalization is much deeper. It states that up to any order in perturbation
theory, we can remove all ultraviolet divergences from a physical quantity just by a redefinition of the
coupling constant. Furthermore, it states that equation (45) generalizes to all order of perturbation theory,
and the right hand side of the equation is free of ultraviolet divergences. In other words

dαS (µ)
= −b0 αS2 (µ) − b1 αS3 (µ) − b2 αS4 (µ) + . . . . (53)
d log µ2

14
where b0 , b1 , b2 , etc., are ultraviolet-finite.
From eq. (30), we see that αS = αS (M ), that is to say that the original bare αS was in fact the
running coupling evaluated at the cutoff scale. It is not useful to try to express physical quantities in
terms of αS evaluated at a scale which differs widely from the scales involved in the physical quantities
under consideration. In fact, in this case, large logarithms of the ratio of the physical scale to µ arise in
the perturbative expansion, as one cannot trust the truncated (fixed order) result. In order to get a reliable
result, one should instead use µ ≈ Q, so that no large logarithms appear in the perturbative expansion.
Of course, we do not know the precise value of µ we should use. We can use µ = Q, µ = 2Q, µ = Q/2,
without the possibility of arguing what is the best choice. In practice, a difference in the value of the
scale used makes a difference in the result, but this difference is of the order of the neglected terms in
the perturbative expansion. This can be easily verified from formula (31) (students are encouraged to try
this).
It is now tempting to formulate the first prediction of our theory. From the expression of the
running coupling, eq. (48), we see that the strong coupling constant is of order 1 when the scale µ
approaches Λ. It is tempting to set Λ = 300 MeV, the typical hadronic scale, and then predict that
µ ¶
αS (MZ )
R(MZ ) = R0 (MZ ) 1 + = R0 (MZ )(1 + 0.046) (54)
π

in reasonable agreement with the value measured at LEP. Of course, this example is very sloppy, does
not take into account the heavy flavour thresholds, higher order effects, and other important facts. It is
however important to remark that, had we measured R/R0 = 1 + 0.08 at LEP, this would have implied
Λ = 5 GeV, a totally unacceptable value.

3.2 Relation among the couplings with different number of light flavours
Now I will spend a few words concerning the number of light flavours. In order to make the discussion
clearer, let us assume that there is a top quark of 100 GeV, and that all the other quarks are massless.
Intuitively, we should then be able to describe the effects of QCD, for scales much below 100 GeV, but
still much above Λ, in a perturbative fashion, forgetting about the existence of the top quark. The formula
for e+ e− → hadrons contains then b0 evaluated with nf = 5. On the other hand, if the heavy top is
really there, the true description of our phenomenon should be given in terms of the theory with top.
While up to the order αS a top loop never enters our Feynman graphs, at two loops we do have a top
loop contribution, represented in the graphs of fig. 7. In spite of the fact that there is not enough energy

Fig. 7: Top loop contribution to e+ e− → hadrons.

to produce the top, these graphs do contribute. They are always associated to a propagator correction.

15
Neglecting terms suppressed by powers of 1/m2t , their effect is simply to multiply αS by a factor 1 −
αS /(6π)(d + log(M 2 /m2t )), where d is a number which depends upon the particular renormalization
scheme one uses. This result can also be guessed on the basis of the fact that the UV divergence coming
from the top loop must have the same form as the UV divergence coming from any light fermion. We
have then à ¶¸ µ ′ ¶2 !
αS′ M2 1 M2
· µ
αS
R = R0 1 + + c + πb0 log 2 − d + log 2 . (55)
π Q 6 mt π

With αS′ we indicated the true (bare) coupling, of the theory in which the heavy quark is taken into
account properly, instead of the “fake” theory in which the heavy quark is ignored. The renormalization
procedure for the theory including the top requires now the substitution

M2 ′ 2
αS′ (µ) = αS′ + b′0 log α (56)
µ2 S

where b′0 = (33 − 2(nf + 1))/(12π), and the renormalized formula for R becomes
à ¶ !
αS′ (µ) µ2 µ2 αS (µ) 2
· µ ¶¸ µ ′
1
+ O αS (µ)3 .
¡ ¢
R = R0 1 + + c + πb0 log 2 − d + log 2 (57)
π Q 6 mt π

Equation (31) and (57) must be completely equivalent, at least up the order αS2 . It turns out that in the
commonly used MS renormalization scheme, we have d = 0. In this scheme, the equivalence of the two
formulas imply that
αS (µ) = αS′ (µ) for µ = mt . (58)
Therefore, in the MS scheme the relation between coupling constants defined by ignoring a heavy flavour,
and the coupling with the heavy flavour included, is simply stated by saying that the two running cou-
plings should coincide for µ = mh , where mh is the mass of the heavy flavour. In practice, we have three
useful definitions of the coupling constants. One that ignores the charm quark (and heavier flavours),
(3) (4)
which has three light flavours, and may be indicated with αS , one that ignores bottom (αS ) and one
(5)
that ignores top (αS ).
(3) (5) (4) (5)
A plot of the ratios of αS /αS and αS /αS is given in fig. 8. The couplings are correctly
matched at the heavy flavour thresholds according to the MS prescription. From the plot, it appears
that the couplings for four and five flavours are not very different. This is indeed the case. One should
however be careful, because the corresponding value of Λ is in fact very different. The values used in
the figure have Λ3 = 310 MeV, Λ4 = 260 MeV and Λ5 = 170 MeV. A common error is, for example, to
use values of Λ4 where Λ5 should be used. One should never forget that Λ is nothing but a parameter in
the formula for αS . If we change the formula (going for example from one to two loops) the value of Λ
(3) (4)
should be changed. Similarly, if we plug in the same value of Λ in the expression for αS and αS , their
value would be very different, even for µ = mb , while if we use the appropriate value of Λ3 and Λ4 in
the corresponding formulas, their value will be identical at that scale.

3.3 State of the art in the beta function and R

The expression of the beta function known today has the form

∂αS
= −b0 αS2 − b1 αS3 − b2 αS4 − b3 αS5 (59)
∂ log µ2
where the term b2 has been computed in ref. [4], and the term b3 has been very recently computed in
ref. [5]. Here I report below only the values of b0 and b1 , and the corresponding solution of the renor-

16
 Fig. 8: Ratios of the coupling defined for different values of nf .

malization group equation at the two loop level. This is what is commonly used in most applications.
µ2
 
(nf ) 1 b log log Λ 2
αS (µ) = 1 − 1 5 
(60)
b0 log Λµn b20 log µ22
f Λ5
33 − 2nf
b0 = (61)
12π
153 − 19nf
b1 = . (62)
24π 2
The reader can verify that the eq. (60) satisfies equation (59) up to terms of order αS3 .
The accuracy of the β function that is required in phenomenological applications depends upon
the accuracy of the calculations one is using. The rule of thumb is the following:

• if only the leading strong interaction effect is included (LO calculation), one needs one-loop evo-
lution;

• if terms subleading by one power of αS are included (NLO calculation), one needs two-loop evo-
lution;

• if terms subleading by two powers of αS are included (NNLO), one needs two-loop evolution;

• ...

Thus, for example, if we use the O(αS ) formula for R, that is to say R = R0 (1 + αS /π), we need to
include 1-loop evolution. Similarly, if we have a process that starts at order αS2 (like four-jet production in
e+ e− annihilation), we need 1-loop evolution. If we include the O(αS2 ) term in R, we need to use 2-loop
evolution. Notice that the accuracy in the β function that we want is always higher than the accuracy
in the calculation by one unit. So, the leading term in the β function is of order two, but it is needed to
maintain the accuracy of the result for R, which seems strange: if R is known at order αS , why should
its derivative needed at order αS2 ? The answer is that for a large evolution span, an error of order αS2 in

17
the derivative can become of order αS , because a large evolution logarithm log µf /µi can compensate a
power of αS ∝ 1/ log µ/Λ. Consider the case when the final evolution scale is such that µf /µi ≈ µi /Λ.
From the 1-loop renormalization group equation we get:
Z µf
µ2f
µ ¶
2 2 2
αS (µf ) − αS (µi ) = −b0 αS (µ) d log µ ≈ O αS (µ) log 2 ≈ O(αS ) , (63)
µi µi
consistently with the fact that in this case αS (µf ) ≈ αS (µi /2).
If the evolution span is small (i.e. if the scale changes by a factor of order one), one does not need
an extra power of αS in the β function to match the accuracy of the calculation.
Evolution must also be properly adjusted when crossing a flavour threshold. When one uses the 1-
loop β function, the condition α(nf+1) (µ) = α(nf) (µ) for µ = 2m, or µ = m/2 are accurate enough. In
other words, the matching is done at a scale of the order of the flavour mass. The difference of choosing,
for example, µ = 2m or µ = m/2, is simply
α(2µ) = α(µ/2) − 2b0 αS2 log(4) (64)
and is thus a NLO effect. When using a 2–loop β function in the context of an NLO calculation, one
must use a matching condition which is accurate up to terms of order αS3 . In the MS scheme, this is
α(nf+1) (µ) = α(nf) (µ) + O(αS3 ) for µ = m (65)
where we no longer have the freedom of a factor of order 1. Matching conditions appropriate for a 3–
loop β function in the context of a NNLO calculation are given in ref. [6], and consist in a correction of
order αS3 to equation 65.
The radiative corrections to R have been computed up to the order αS3 in ref. [7, 8, 9], a rather
remarkable achievement. The result for nf = 5, expressed in the MS scheme reads
n αS ¡ ¢o
R = R0 1 + 1 + 0.448αS − 1.30αS2 (66)
π
(5)
where αS = αS (Q), Q is the annihilation energy. Besides finding applications in e+ e− annihilation
physics, this formula has found recently a very interesting application to the determination of αS from
the hadronic decay of the τ lepton [10]. After what we have learned in this section about the ratio R, it
should be easy for us to compute the ratio between the hadronic and the leptonic branching ratios of the
τ , at zeroth order in the strong coupling constant. This is depicted symbolically in fig. 9. From the figure,
it is clear that the top and bottom processes only differ by the number of possible final states. Thus, the
top graph has a factor of 3, because of the three colours. Only an up-anti-down, or up-antistrange pair
can be produced, since phase space forbids the production of charmed final states. Neglecting the mass
difference between the down and the strange, one can see that the Cabibbo angle is irrelevant in this
case. Thus, the ratio of the hadronic width to the (for example) electron width is 3 at zeroth order in the
coupling constant. As in the case of R, this ratio will receive strong corrections, and the displacement of
this ratio from 3 can be used to attempt a determination of the strong coupling constant from τ decays.
Observe that the value of αS at the scale of the τ mass is quite large, around 0.35. At LEP1 energy this
value is around 0.12. In table 2 (taken from ref. [11]) the experimental determinations of αS coming from
R below the Z peak, R on the Z peak, and tau decays, are reported. All determinations are performed
at the relevant scale of the process (thus, for example, the τ determination is performed in terms of
αS (Mτ )), and then evolved at the Z mass for comparison. Notice the rather remarkable agreements
among the different determinations.

4 JETS IN e+ e− ANNIHILATION
In the discussion of the previous section, we have left aside a few important issues, that can be summa-
rized in the foloowing questions:

18
 Fig. 9: The ratio between the τ hadronic and leptonic width.

Measurements Q (GeV) αs (Q) αs (mZ )

Rτ √ 1.777 0.323 ± 0.005(exp.) ± 0.030(th.) 0.1181 ± 0.0007(exp.) ± 0.0030(th.)
Re+ e− (20 < s < 60 GeV 42 0.175 ± 0.028 0.126 ± 0.022
Z peak 91.2 0.124 ± 0.004 ± 0.002(Mt , MH ) +0.003
−0.001 QCD

Table 2: Determinations of αS from inclusive hadronic decays, taken from ref. [11]. In the Zpeak determination, the error due
to uncertainties in the Higgs and top mass, and the error due to QCD uncertainties, are separately specified.

1. How can we identify a cross section for producing quarks and gluons with a cross section for
producing hadrons?

2. Given the fact that free quarks are not observed, why is the computed Born cross section so good?

3. Are there any other calculable quantities besides the total cross section?

We will see in the following that question 1 and 2, although unanswerable in QCD, imply no contradic-
tion. We will also see that, under the same assumptions that make 1 and 2 work, also question 3 has an
affirmative answer.
Looking at the lowest order formula, we immediately wonder why a formula describing the pro-
duction of quarks in the final state should also be able to describe the production of hadrons, since we
never observe free quarks in the final state. The structure of the perturbative expansion by itself give us
a hint of how this may happen. Consider in fact the corrections of order αS to the total cross section.
They are given by diagrams in which a real gluon is emitted into the final state, and diagrams in which a
virtual gluon is exchanged (interfered with a Born graph) as depicted in fig. 10. In the previous section I
have just stated that the total of the corrections of order αS is finite, and equals αS /π. I will now show
that the individual real contributions (those with a gluon in the final state) are individually infinite. As
a consequence of the finiteness of the total, also the virtual ones (those with only the quark-antiquark
pair in the final state) must be infinite, with the opposite sign. Let us therefore compute the diagram
of fig. 10. We will perform the calculation under the simplifying assumption that the gluon energy is
much smaller than the total available energy. It turns out that in this approximation the computation will
require very little effort, and the approximation itself contains all the interesting features of the result. It
is easy to convince oneself that the colour factors for all contributing diagrams (after squaring and taking
the colour traces) are one factor of CF = 4/3 relative to the Born term (which has a colour factor of 3,
equal to the number of colours that can flow in the loop), a result whch is illustrated in the last equality

19
 Fig. 10: Soft gluon emission (left graph) and virtual gluon exchange in e+ e− annihilation.

of fig. 3. The amplitude for the Born process is

M = u(k)ǫµ γµ v(k ′ ) (67)

where ǫ is the virtual photon polarization, q is the incoming four momentum, k is the momentum of the
outgoing fermion and k ′ = q − k is the momentum of the outgoing antifermion. Defining

N = ǫµ γµ v(k ′ ) (68)

we have
M = u(k)N . (69)
Consider now the diagram of fig. 10, in which the gluon is emitted from the outgoing fermion. The
amplitude is given by
k/ + /l
M1 = u(k)(−i)γα i N. (70)
(k + l)2
Actually we should have also substituted k ′ = q − k − l in N , but we are assuming that l is small.
Fermion masses are also being neglected, since we assume we are considering a high energy process.
Neglecting l in the numerator, and using the identity u(k)/
k = 0, and expanding the denominator (recall
that l2 = 0, k 2 = 0) we obtain
γα k/ + k/γα 2kα kα
M1 = u(k) N = u(k) N = M. (71)
(k + l)2 2k · l k·l
Analogously, for the amplitude with the gluon emitted from the outgoing antiquark, we obtain
kα′
M2 = − M (72)
k′ · l
and the total is
k′
µ ¶
kα
Mqqg = M1 + M2 = − ′α M (73)
k·l k ·l
which vanishes when contracted with lα , as gauge invariance requires. Taking the square (with the extra
minus for the gluon projector)
k · k′
M2qqg = 2 M2 . (74)
(k · l)(k ′ · l)
From the amplitude square we turn to the cross section by supplying the phase space factor for the gluon

d3 l k · k′
Z
2 Born
σqqg = CF gS σqq . 2 . (75)
2l0 (2π)3 (k · l)(k ′ · l)

20
At this stage I have also included the coupling constant and the appropriate colour factor. Let us now
consider the process in the rest frame of the incoming virtual photon, with q = (q 0 , 0, 0, 0), and ~k = −k~′ .
Let us call θ the angle that the gluon makes with the fermion direction. We have then

k · k′ 4
2 ′
= 2 (76)
(k · l)(k · l) l0 (1 − cos θ)(1 + cos θ)

so that (using αS = gS2 /(4π))

αS Born dl0 4
Z
σqqg = CF σ d cos θ . (77)
2π qq 0
l (1 − cos θ)(1 + cos θ)
The cross section for producing an extra gluon is therefore divergent in three regions:

• when the emitted gluon is in the direction of the outgoing quark (θ = 0)

• when the emitted gluon is in the direction of the outgoing antiquark (θ = π)

• when the emitted gluon is soft (l0 → 0).

The first two kind of divergences are called collinear divergences, while the last one is called a soft
divergence. Both divergences are of infrared (IR from now on) type, that is to say, they involve long
distances. In fact, because of the uncertainty principle, we need an infinite time in order to specify
accurately the particle momenta, and therefore their directions. Unlike UV divergences, there is nothing
like renormalization for the IR divergences. Their meaning is the following: the cross section is sensitive
to the long distance effects, like the fermion masses, the hadronization mechanisms, and so on. In fact, if
we give a fictitious mass to the gluon, the result becomes convergent, but it will be sensitive to the value
of the gluon mass.
It was stated in the previous section that the total of the corrections of order αS to the production
of hadrons in e+ e− annihilation is finite, and equals απS . It follows that also the virtual corrections
must have the same kind of infinities, with opposite sign. If we cutoff these divergences with some
method (like dimensional regularization, or by giving a mass to the gluon), and then sum up real and
virtual contributions, the divergences cancel, and the left-over is finite and equal to αS /π times the Born
cross section, independent of the method we used to regularize the diagrams. This cancellation is a
consequence of the Kinoshita-Lee-Nauenberg theorem [12, 13]. Roughly speaking, this theorem deals
with divergences that arise because of degeneracy in the final state. For example, the final state with
an extra soft gluon is nearly degenerate with the state with no gluons at all, and the state with a quark
split up into a quark plus a gluon, with parallel momenta, is degenerate with the state with no radiation
at all. The theorem states that the cross section obtained by summing up over degenerate states are not
divergent.
We are now ready to show, as promised, that point 1 and 2 imply no contradiction. We have in fact
shown that if we attempt to compute the cross section for the production of a pair of quark–antiquark
alone, while the zeroth order term (the Born term) is finite, the term of order αS is infinite, being collinear
and soft divergent. This means that a perturbative expansion for this quantity does not work, since the
coefficients of the expansion are large (actually infinite). Therefore, even the Born term alone cannot
represent the cross section for producing a quark–antiquark pair. Thus, the fact that a final state with a
quark–antiquark pair and nothing else is not observed is not in contradiction with perturbation theory,
since we have shown that there is no valid perturbative expansion for this quantity. On the contrary, the
cross section for producing strongly interacting particles (no matter how many quarks or gluons) remains
finite even after perturbative corrections are added. One can show that in fact it remains finite order by
order in perturbation theory. Its lowest order approximation is in fact the Born cross section. So, the
Born cross section is the lowest order term in a well defined perturbative expansion with infrared finite

21
coefficients, which is just the cross section for producing strongly interacting particles (no matter how
many and which types). This is why the Born cross section represents quite accurately the total hadronic
cross section. We are now also in the position to answer the third question. We will show that there are
quantities which characterize the hadronic final state, that are infrared finite in perturbation theory, and
therefore should be calculable in perturbative QCD.

4.1 Sterman–Weinberg jets

Sterman and Weinberg [14] first realized that one can define a cross section which is calculable and finite
in perturbation theory, and characterizes in some way the hadronic final state. The definition goes as
follows.
We define the production of a pair of Sterman–Weinberg jets, depending on the parameters ǫ
and δ, in the following way. A hadronic event in e+ e− annihilation, with centre-of-mass energy E,
contributes to the Sterman–Weinberg jets cross section if we can find two cones of opening angle δ that
contain more than a fraction 1 − ǫ of the total energy E. In other words ǫE is the maximum energy
allowed outside of the cones. An example of Sterman-Weinberg jet event is illustrated in fig. 11. We

Fig. 11: Sterman–Weinberg jets.

will now show that the computation of the cross section for the production of Sterman–Weinberg jets, in
the approximation introduced in the previous chapter, is infrared finite. The various contributions to the
cross section (illustrated in fig. 12) are as follows

• All the Born cross section contributes to the Sterman–Weinberg cross section, for any ǫ and δ
(fig. 12a).
• All the virtual cross section contributes to the Sterman–Weinberg cross section, for any ǫ and δ
(fig. 12b).
• The real cross section, with one gluon emission, when the energy of the emitted gluon l0 is limited
by l0 < ǫE (fig. 12c), contributes to the Sterman–Weinberg cross section.
• The real cross section, when l0 > ǫE, when the emission angle with respect to the quark (or
antiquark) is less than δ (fig. 12d), contributes to the Sterman–Weinberg cross section.

The various contributions are given formally by

Born = σ0 (78)

22
 Fig. 12: Contributions to the Sterman–Weinberg cross–section. Born: (a), virtual: (b), real emission: (c) and (d).

4αS CF E dl0 π d cos θ

Z Z
Virtual = −σ0 0 2
(79)
2π 0 l θ=0 1 − cos θ
Z ǫE 0 Z π
4αS CF dl d cos θ
Real (c) = σ0 0 2
(80)
2π 0 l θ=0 1 − cos θ
4αS CF E dl0
· Z δ Z π ¸
d cos θ d cos θ
Z
Real (d) = σ0 0 2
+ 2
. (81)
2π ǫE l θ=0 1 − cos θ θ=π−δ 1 − cos θ

Observe that the expression of the virtual term is fixed by the fact that it has to cancel the total of the real
contribution. Since we are looking only at divergent terms, and since the virtual term is independent of
δ and ǫ, the expression (79) is fully adequate for our purposes. Summing all terms we get
4αS CF E dl0 π−δ d cos θ
Z Z
Born + Virtual + Real (a) + Real (b) = σ0 − σ0 0 2
2π ǫE l θ=δ 1 − cos θ
µ ¶
4αS CF
= σ0 1 − log ǫ log δ 2 (82)
2π
which is finite, as long as ǫ and δ are finite. Furthermore, as long as ǫ and δ are not too small, we find
that the fraction of events with two Sterman-Weinberg jets is 1, up to a correction of order αS .
Now we are ready to perform a qualitative step: we interpret the Sterman-Weinberg cross section,
computed using the language of quarks and gluons, as a cross section for producing hadrons. Thanks to
this qualitative step, we make the following prediction: at high energy, most events have a large fraction
of the energy contained in opposite cones, that is to say, most events are two jet events. As the energy
becomes larger αS becomes smaller. Therefore we can use smaller values of ǫ and δ to define our jets.
Thus, at higher energies jets become thinner.
It should be clear now to the reader that, by the same reasoning, we could show that the angular
distribution of the jets will be very close, at high energy, to the angular distribution one computes using

23
the Born cross section, that is to say, the typical 1 + cos2 θ distribution. These predictions have been
confirmed experimentally since a long time.

4.2 A comparison with QED

The alert reader will have probably realized that the discussion given in this section should also apply to
electrodynamics. In fact, the Feynman diagrams we have considered are present also in electrodynamic
processes, like e+ e− → µ+ µ− , and they differ from the QCD graphs only by the color factor. Thus, from
the previous discussion, we would infer that Sterman-Weinberg jets in electrodynamic processes at high
energy do not depend upon long distance features of the theory. For example, they become independent
from the µ mass when E ≫ µ. Also in electrodynamics, the cross section for producing a µ+ µ− pair plus
a photon is divergent, as is divergent the cross section for producing the pair without any photon. In many
books on quantum electrodynamics these divergences are discussed, and it is shown that a resolution
parameter for the minimum energy of a photon is needed in order to have finite cross section order by
order in perturbation theory. In electrodynamics, we can go even farther, and prove that by resumming
the whole tower of divergent graphs, the infinite negative virtual correction to the production of a µ+ µ−
pair with no photons exponentiates, and gives a zero cross section. In other words, as it is well known,
it is impossible to produce charged pairs without producing arbitrarily soft photons. What is then the
difference with QCD? Why cannot we prove similar results in QCD? The difference arises because of the
different asymptotic properties of QCD and QED. In QED the coupling becomes smaller at low energy,
while in QCD it grows. Thus, when the scale of an emission process approaches a few hundred MeV the
coupling constant becomes of order one, and perturbation theory becomes inapplicable. So, the infrared
problem in QCD is tightly untangled with the confinement problem, and it seems to be unanswerable
in the context of perturbation theory alone. In this sense perturbative QCD is an incomplete theoretical
framework. In order to make predictions we need to assume that the soft phenomena characterized by
scales of the order of few hundred MeV do not spoil completely the computation of the high energy part
of the process. This assumption is consistent with perturbation theory; it is however an assumption, and
it cannot be proven using perturbation theory alone.

4.3 Shower Monte Carlo programs

Perturbation theory can be used to compute radiation processes as long as the energies involved are
safely above the typical hadronic scales. It is then possible to construct event generator programs that
implement the properties of QCD Feynman diagrams for the splitting of partons into more partons, as
long as the splitting involves large transverse momentum, and then use some plausible model for last
step of the splitting process, in which the partons become hadrons. These programs are generally called
shower Monte Carlo event generators [15, 16, 17], and are an invaluable tool for experimental physicists.
They essentially sum a large class of Feynman graphs, precisely the most collinear and (in some cases)
soft-singular ones. In the attempt to describe the full final state, they give up the accuracy that can be
obtained in perturbation theory. They are (until now) compatible with QCD only at the leading order in
the strong coupling. While the QCD part is quite similar in all of them, for the last step of the final state
formation, that is to say the hadronization, they differ widely, since they have to rely on models, like
the so called Lund string model or the Herwig cluster model. Hadronization models are tuned to data.
Nevertheless, one should not forget that there is very little predictivity in these models, since they are
only qualitatively based upon the theory. One can expect in general that the hadronization properties for
which the Monte Carlo has been tuned for will be well reproduced by it, but not much more than this.

4.4 More jet definitions and shape variables

The key property of the Sterman-Weinberg jets, that makes them calculable in perturbation theory, is the
insensitivity of the jet definition to radiation of soft particles, and to the collinear splitting of an particle

24
into two particles that share its momentum. This insensitivity is necessary to guarantee the cancellation
of effects that depend strongly upon long distance phenomena, that is to say, those effects that are infrared
divergent when computed in perturbation theory.
After the paper of Sterman and Weinberg, it was soon realized that it is not difficult to build a
whole class of final state observables that do have the same property of soft and collinear insensitivity,
and can thus be computed in perturbation theory, and compared with experimental measurements: thrust,
oblateness, the C parameter, jet clusters, the mass of the heaviest hemispheres, etc.. The important thing
which is assumed in these definitions is that the same definition must be applied to the final state hadrons
by the experimenter that measures this quantity, and by the theorist that computes this quantity in terms
of quark and gluons. Only if this condition is satisfied, one can assume that in the high energy limit
the computed quantity will agree with the measured one, up to corrections that are suppressed by some
inverse power of the energy.
One of the first of these infrared safe shape variables is thrust. It is defined by the equation
P
i |~pi · ~v |
t = max P . (83)
~v i |~
pi |
In words, one takes an arbitrary vector (in the centre-of-mass frame of the colliding electron-positron
pair) and sums the absolute values of the projection of the momenta of all final state particles onto that
vector, normalized to the sum of all absolute values of the hadron momenta. The vector is rotated until a
maximum is found. The maximum direction is called the thrust axis, and the value at the maximum the
thrust of the event. The maximum value of thrust is one, for a final state of two massless particles in the
back-to-back direction. It is easy to check that thrust is an infrared safe shape variables. In fact, a soft
emission does not alter the thrust abruptly, since all emitted particles enter weighted by their momenta.
Also collinear splitting does not alter the thrust of an event, as one can easily verify. An example of a
quantity which is not infrared safe is the total number of particles in the final state, which changes by
one unit in case of soft emission. Examples of a quantities which are sensitive to collinear splitting are
the axis of the tensor X j
S ij = pil pl (84)
l
which were actually used in the past to classify the “jettiness” of an event.
A modern, and very clever way to define jets is by clustering [18]. For a given events, one forms
the invariant mass of all pairs of particles in the final state. The pair with the smallest mass is merged into
a single pseudoparticles, and then the procedure is continued with the pseudoparticles, and it is stopped
when the smallest mass of a pair exceeds a given cutoff y × S. One ends up with a definite number of
clusters, and one can thus define the cross section for producing two, three, four or more clusters for a
given y cut. It is easy to convince oneself that these cross section definitions are infrared safe. Since the
computation of these cross sections (in terms of partons) should in first approximation give the correct
answer, we see that in perturbative QCD we roughly expect (for not too extreme values of y) that most
events will be made up by two clusters, a fraction of order αS will be made up by three clusters, and a
fraction of order αS2 will be made by four clusters. Analogously, we expect thrust to be near one, and
its departure from one to be of order αS . We also expect that a fraction of events of order αS will have
thrust well below one.
Because of the obvious interest in the determination of αS from jet shape variables, a lot of effort
has gone in the study of jet and shape variables that are directly proportional to αS , which we may call
“three-jet sensitive”, like the thrust distribution, and the fraction of events with three clusters. There are
tens of variables of these kind that have been studied at e+ e− machines.
The present state of the art for jet studies in e+ e− machines mainly relies on the calculation of
Ellis, Ross and Terrano (ERT) [19, 20], which allows to compute any infrared safe 3−jet shape variable
up to the order αS2 . Various computer programs for the computations of these quantities are available, and

25
many of these quantities have been tabulated [21]. Heavy quark mass effects have also been included in
the 3−jets calculation [22, 23, 24]. Three-jet quantities have been intensively studied at e+ e− machines,
The results of LEP1 and SLD have given a quite remarkable contribution to the tests of QCD, and
considerably reinforced our confidence in perturbative QCD.
Recently, the NLO correction to 4-jet partons production have been computed [25, 26, 27, 28],
allowing thus the computation of any 4−jets shape variable in the form αS2 (µ2 )C + αS3 (µ2 )D(µ2 /Q2 ) +
. . .. Phenomenological applications have begun to appear recently [29] [30].
Fixed order calculations of shape variables distributions are sometimes supplemented with all-
order resummation of effects that are enhanced in the limit of thin jets. An example of these effects is
visible in eq. 82; when δ and ǫ become small, the OαS correction becomes large, because of the large
collinear and soft logarithms. These logarithms, called “Sudakov logarithms”, are a general phenomenon
that happens in QCD and QED when we force a process into a region of phase space where radiation
is inhibited. Since soft radiation is infrared divergent, and its divergence cancels againts virtual con-
tributions, when we suppress soft radiation the cancellation becomes unbalanced, and large logarithms
appear at all orders in the perturbative expansion. In some cases, these logarithms can be organized and
resummed [31, 32, 33]
Hadronization and power corrections are believed to be suppressed as 1/Q, but they are still im-
portant at LEP energies. They are usually estimated using Monte Carlo hadronization models. The
renormalon inspired model of ref. [34] provides an alternative method [35].

4.5 Thrust as an example

Let us focus upon the case of thrust as an example. The thrust distribution has the perturbative expansion
¶2 ·
µ2
µ ¸
1 dσ αS (µ) αS (µ)
A(t) 2π b0 log 2 + B(t) + O αS3 .
¡ ¢
= δ(1 − t) + A(t) + (85)
σ0 dt 2π 2π Q

The first term, proportional to a delta function, is the Born contribution, which corresponds to the pro-
duction of two back-to-back massless partons. The functions A(t) and B(t) can be computed from the
ERT results (they are tabulated in ref. [21]). The renormalization scale µ is explicitly indicated in the
formula. As in the total cross section formula, the explicit scale dependence of the term of order αS2 is
related to the coefficient of the term of order αS . Again, using the renormalization group equation at
1 loop (i.e., ∂αS /∂ log µ2 = −b0 αS2 ), one can prove that the scale dependence of the above equation
cancels up to the order αS2 . Of course, if the whole perturbative expansion was included in the right hand
side, no scale dependence would survive, since the left hand side is scale independent. However, only
terms up to the order αS2 are included, and thus one expects a residual scale dependence at order αS3 .
Radiative corrections are generally quite large. For example
1.05
h1 − ti = αS (Q)(1 + 3αS )
π
hoi = 1.29αS (Q)(1 − 4.3αS )
2 1.05
hMD,t i= αS (Q)(1 − 0.025αS ) (86)
π
where the second quantity is oblateness (for a precise definition, see ref. [21]), and the third quantity is
the difference of the square of the masses of the heavy hemisphere with respect to the light hemisphere,
with the hemisphere defined according to the thrust axis. Thus, corrections can be as large as 40% even
at LEP1 energies. Because of this, it is mandatory that corrections of even higher orders (αS3 and higher)
should be at least estimated and included in the theoretical error. There is no universal method to estimate
the theoretical error in this case. A commonly used method is to look at the scale dependence of the result.
Since the remaining terms of the perturbative expansion should compensate the scale dependence, they

26
 Table 3: A summary of measurements of αS from shape variables.

Q ∆αs (MZ0 )
Process [GeV] αs (Q) αs (MZ0 ) exp. theor.
+0.016
e+ e− 22 0.161 −0.011 0.124 +0.009
−0.006 0.005 +0.008
−0.003
+0.012
e+ e− 35 0.145 −0.007 0.123+0.008
−0.006 0.002 +0.008
−0.005
e+ e− 44 0.132 ± 0.008 0.123 ± 0.007 0.003 0.007
Z0 91.2 0.121 ± 0.006 0.121 ± 0.006 0.001 0.006
e+ e− 133 0.113 ± 0.008 0.120 ± 0.007 0.003 0.006
e+ e− 161 0.109 ± 0.007 0.118 ± 0.008 0.005 0.006
e+ e− 172 0.104 ± 0.007 0.114 ± 0.008 0.005 0.006
e+ e− 183 0.109 ± 0.005 0.121 ± 0.006 0.002 0.005
e+ e− 189 0.110 ± 0.004 0.123 ± 0.005 0.001 0.005

must be at least as large as the scale variation of the truncated result. The scale should be varied in a
range around the typical scale of the process. It should not be chosen neither much higher of this typical
scale, nor much smaller, since in these cases the perturbative expansion is not well behaved. A common
choice is mZ /4 < µ < mZ , which accounts for the fact that the typical scale of the process is somewhat
below the Z mass.
Hadronization effects should also be estimated, and included in the theoretical error. For the
observable h1 − ti, for example, we can make a naive estimate in the following way. Let us assume that
the emission of an extra soft pion in the final state has a probability of order one. This emission takes
away from the thrust a value of few hundred MeV (the transverse mass of a soft pion) divided by the total
available energy. To fix the numbers, let us say that at LEP we have δt = 0.5/90 ≈ 0.0055, assuming a
500 MeV average transverse mass for the pion. The perturbative value of h1 − ti is roughly αS /π ≈ .04,
increased by the αS2 correction to roughly 0.055. Thus δt /h1 − ti = 0.1. This means that we can expect
that hadronization effects may have a 10% effect in the determination of αS from h1 − ti.
An instructive example of a QCD study at LEP can be found in ref. [36]. Estimates of hadroniza-
tion corrections are used there to correct the raw data. Theyr typical value is around 10%. Hadronization
corrections are estimated by running a shower Monte Carlo with or without the hadronization stage. The
corrections are determined by looking at the difference between the two runs, and are then applied to the
data. The error on the hadronization corrections are estimated by using different Monte Carlo programs
with different hadronization models. It is quite clear that this procedure is quite risky. The QCD stage is
in fact similar in all shower Monte Carlo. The hadronization step is different, but it is in all cases tuned
to fit the data. This may generate a bias towards determining the same value of αS used in the Monte
Carlo. The size of the radiative correction is reported in ref. [36], and thus, the pessimistic reader may
use the whole hadronization correction as an error on the determination, if he wishes to do so.
Table 3 (from ref. [11]) summarizes the determinations of αS from event shape variables. In all
determinations, NLO calculations are used, together with resummation of soft gluon effects. Power
corrections are estimated using Monte Carlo programs.
Alternative models for the power suppressed corrections have recently appeared, and have been
introduced in phenomenological
√ analysis. In ref. [35], several shape variables have been examined in
the energy range of S = 14 to 189 GeV. QCD NLO prediction, together with the power correction
model of ref. [34] are used to fit the data. I will not try to describe here the features of the model; it is
enough to know that power corrections to shape variables depend upon a universal parameter α0 in this
model. A summary of the results of this analysis is displayed in figs. 13 and table 4.5. We observe

27
 Mean Values Distributions
1
0.8 b) a)
0.9
0.7
0.8
0.6
α0 α0 0.7
0.5 average average
<1-T> 0.6 1-T
<MH2> MH, MH2
0.4 0.5
<BT> BT
<BW> BW
0.3 <C> 0.4 C
0.1 0.11 0.12 0.13 0.08 0.09 0.1 0.11 0.12 0.13
αS(MZ) αS(MZ)
Fig. 13: Simultaneous fits to αS and α0 using mean values of shape variables (left) and distributions (right).

fit syst. Th.

+0.0028
means αS 0.1187 ±0.0014 ±0.0001 −0.0015
+0.065
α0 0.485 ±0.013 ±0.001 −0.043
+0.0044
distr. αS 0.1111 ±0.0004 ±0.0020 −0.0031
+0.099
α0 0.579 ±0.005 ±0.011 −0.071
+0.0032
Comb. αS 0.1171 −0.0020
+0.066
α0 0.513 −0.045

Table 4: Results of the fits to αS (MZ ) and α0 (2 GeV) from ref. [35].

28
that the final value is well in agreement with other determinations [11]. Also, to some extent the data
supports the universality of the non-perturbative parameter. On the other hand, the value determined from
distributions is considerably lower than the value obtained with standard methods (i.e. hadronization
corrections with Monte Carlo models). Furthermore, for some shape variables the consistency of the
determination is quite poor. Thus, as far as power suppressed corrections are concerned, we can certainly
say that they are very poorly understood.
Even if we assume a pessimistic attitude with regard to power corrections, one must recognize
that LEP results do show a remarkable consistency with perturbative QCD results. In figure 14 I try to
give an unbias illustration of the comparison of LEP data with perturbative QCD results. In the figure,

Fig. 14: Bin-by-bin determination of αS for several different shape variables.

a determination of αS is performed for several shape variables. The determination was performed first
using a leading order formula (left plot), and then the full O(αS2 ) formula. No hadronization correction
was applied to the data. Three values of the renormalization scale were chosen for each variable: µ =
mZ /4, mZ /2, and mZ . In the figure, parallel bands correspond to these three choices. The errors on the
various point are experimental errors. If we had a perfect QCD calculation, e.g. all orders in perturbation
theory, and hadronization corrections were truly negligible, we should expect that all experimental point
lie (within errors) on a constant line. If we only have a leading order calculation, we expect instead
large differences among the various points, that should become smaller and smaller as we include higher
order corrections. In the plot, of course, we can only represent the leading and next-to-leading result,
since an O(αS3 ) calculation has never been performed. It is quite striking to see how, by including the
next-to-leading corrections, the various determinations become much closer to each other. It is left to our
fantasy to imagine what would happen if we could include the O(as3 ) effects.

29
5 PROCESSES WITH HADRONS IN THE INITIAL STATE
We will now turn to describe the application of perturbative QCD to processes in which hadrons are
present also in the initial state, like Deep-Inelastic Scattering (DIS), or the production of some objects
of high invariant mass in hadronic collisions. It turns out that cross sections for these processes can be
computed and related to each other. In general the cross section for the production of some final state
with high invariant mass (which could be made of a heavy weak vector boson, a lepton-antilepton pair,
heavy quarks, jets, and the like) will be expressed by the so called improved parton model formula
XZ (H ) (H )
σH1 ,H2 (p1 , p2 ) = dx1 dx2 fi 1 (x1 , µ) fj 2 (x2 , µ) σ̂ij (x1 p1 , x2 p2 , αS (µ), µ) . (87)
i,j

A pictorial representation of formula 87 is given in fig. 15.

Fig. 15: A graphic representation of the improved parton model formula.

For processes with a single incoming the improved parton model formula is even simpler. For
example, in DIS
XZ (H)
σH (p) = dxfi (x1 , µ) σ̂i (x, µ) , (88)
i
Formulae (87) and (88) are applicable to inclusive hard processes. By inclusive, we mean that no detailed

Fig. 16: The improved parton model formula for DIS.

question on the distribution of the final state hadrons is asked in order to measure the cross section. The
generic concept of a hard process is better illustrated by examples. We may, for example, require that a
very large invariant-mass lepton-antilepton pair (the so called Drell-Yan process) is present in the final
state. Or that jets (for example, of the Sterman-Weinberg kind) with large transverse momentum are
observed. In the case of DIS, we simply require |q 2 | to be very large.
The recipe for the improved parton model formulae can be summarized in the following points:

30
 • An incoming beam made of hadrons of type H is equivalent to a beam of constituents (also called
partons), that is to say of quark and gluons, with a longitudinal momentum distribution character-
(H)
ized by the parton density functions (pdf’s from now on) fi (x, µ). More specifically, given the
hadron H with momentunm p, the probability to find in H the parton i with momentum between
(H)
xp and (x + dx)p is precisely dx fi (x, µ). The pdf’s are universal, that is to say, they do not
depend upon the particular process considered.

• The short distance cross section σ̂ is calculable as a perturbative expansion in αS

X (l)
σ̂ij (x1 p1 , x2 p2 , αS (µ), µ) = σ̂ij (x1 p1 , x2 p2 , µ) αSl (µ) . (89)
l

The lowest order term of this expansion is precisely the cross section one would compute naively
at lowest order. For the computation of higher order, a more complex prescription is specified.

• The pdf’s have a mild dependence upon the scale µ, determined by the Dokshitzer-Gribov-Lipatov-
Altarelli-Parisi equation [37]
Z 1
∂ (H) dz X (H)
2
fi (x, µ) = Pij (αS (µ), z)fj (x/z, µ) . (90)
∂ log µ x z j

Using the above equations, given the pdf’s at a specified value of µ, we can compute them at any
other value. The functions P are called splitting function, and have a perturbative expansion in
powers of αS (µ)

αS (µ) 2 (1)
µ ¶
αS (µ) (0)
Pij (αS (µ), z) = P (z) + Pij (z) + O(αS3 ) . (91)
2π ij 2π

The functions P (0) are given in [37], and the functions P (1) are given in [38, 39]. The scale µ is
arbitrary. The µ dependence in the pdf’s is compensate by the µ dependence in the short distance
cross section. As in the case of e+ e− → hadrons, the scale µ is taken to be of the order of the
typical scales in the process, in order to avoid the appearance of large logarithms to all orders in
the short distance cross section. In this way, a truncated expression for the short distance cross
section may be used safely.

The approach described above gives the cross section in terms of a power expansion in αS (µ). Since
αS (µ) ≈ 1/ log µ/Λ, this means that by increasing the perturbative order at which the computation is
performed, one adds corrections which are suppressed by one more inverse power of log µ/Λ. Correc-
tions which are suppressed by powers of Λ/µ are not included in this approach. Thus, for example, the
pdf’s describe the longitudinal momentum distribution of the partons. Since the partons are confined in
a hadron, one knows that they must also have a transverse momentum of the order of the inverse of a
typical hadron size, that is to say 1/Λ. This transverse momentum is neglected, since it would give rise
to power suppressed corrections.
In the following I will try to illustrate and justify the improved parton model approach. I will do
this in three steps.
I will first give a naive argument to show that a somewhat simplified version of formula (87), called
the (naive) parton model formula (i.e., not yet improved), should work. The simplifications consist in
the absence of the scale µ in the pdf’s and in σ̂. Such formula can be used to compute, for example, DIS
cross section, or Drell-Yan pair production cross section. The parton model formula predicts correctly
the existence of scaling in DIS.
In the second step will try to compute QCD corrections in the context of the parton model formula.
I will show that this approach does not survive when radiative corrections are included.

31
 In the third step we will find a modification of the parton model formula that is consistent with
radiative corrections. The main consequence of this improved approach is the appearance of a scale
depedence in the pdf’s. This scale dependence is at the origin of scaling violation phenomena in DIS.

5.1 The naive parton model formula

The basic parton model ideas are based upon a very commonly used intuitive picture of inclusive high
energy scattering of composite systems, when we require a very large momentum transfer. Suppose, for
example, that we collide to hydrogen beams, and require that in the final state we find a pair of electrons
with large transverse momenta. It is clear that the most likely mechanism for producing such an event is
the collision of two electron from the two incoming hydrogen atoms. If the transverse momenta of the
electrons are much higher than the hydrogen binding energy, we may think that, to a good approximation,
the cross section may be computed from the elementary electron-electron cross section, applied to a beam
of incoming free electron. The fact that we want to observe a high transverse momentum scattering
implies that the binding of the electrons to the nuclei cannot have an important effect in this case. In
other words, the electrons behave as free particles in the collision. Observe that the inclusive character of
the reaction, and the presence of high momentum transfer, are both necessary conditions for this approach
to be valid. Inclusiveness is needed, because after the two electron collide, the remaining constituent of
the original atoms (i.e., the protons in the case of hydrogen) are also found in the final state. The high
momentum transfer is instead needed for the reaction to take place in a very short transverse distance. If
this was not the case, like, for example, in the case when we look for small angle scattering, the atoms
may interact coherently. Or, more simply, if the momentum transfer was of the same size as the typical
momentum of the electron in the atom, the binding properties of the system could no longer be neglected.
Assuming now that we have ultra-relativistic monochromatic beams of hydrogen atoms of energy
E, in order to compute the above cross section we would assume that these beams are equivalent to
electron beams with energy Ee = E × me /mp . In reality, even if the atom beams were perfectly
monochromatic, the electron beam would not be perfectly monochromatic. The electrons are moving
inside the atom, with a typical velocity of the order of the electromagnetic coupling v ≈ αem . A simple
exercise in relativistic transformations would show that its energy spread would be of the order vEe . In
fact, the electron energy could be characterized by a pdf fe (x), peaked around the value x = me /mp ,
and a width of order vx. Also the transverse momentum of the electron would be of order vme . However,
while the transverse momentum remains invariant under boost, and thus becomes truly negligible at high
energy, the spread in longitudinal momentum is amplified by the boost, and it thus scales with the energy.
This discussion applies to a boosted, non-relativistic system. We can now try to guess what happens for
a relativistic system, in which all constituents have velocities of order 1, and comparable energies. The
transverse momenta still remain fixed at high energies. Their pdf’s, however, will no longer be peaked
around a particular value. Their spread would be of order 1.
Knowing that the basic building blocks of our hadronic world are quarks and gluons, we thus
expect that for a proton projectile, we will have structure functions for quarks, antiquarks and gluons.
We also naively expect the momentum sum rule
Z 1 X
(p)
dx x fi (x) = 1 , (92)
0 i

because the total momentum of the incoming projectile must be conserved. We also expect that the
proton flavour is conserved. Thus, for example
Z 1 ³ ´
(p)
dx fu(p) (x) − fū (x) = 2 . (93)
0

Since we know that the proton is a relativistic system, we expect that a good fraction of its energy should
be carried by the binding force, that is to say, by the gluons. Thus, the gluon pdf should be sizeable.

32
 Based upon these assumptions, we can now compute various high energy processes involving
hadrons in the initial state. The rules are simple: compute the cross section you are considering for
colliding partons, and then assume that your hadron beam is a beam of partons, with momenta distributed
according to the pdf’s. Always neglect the transverse momenta of the partons, and their masses.
Let us now apply this model to Deep-Inelastic electron scattering. There we collide an electron
with a proton. The kinematical variables of the process are usually defined as

Q2 p·q
q = k − k′ , Q2 = −q 2 , S = (k + p)2 , xBj = , y= . (94)
2p · q k·p
Experimentally, one measures S, y and xBj . One only needs to observe the outgoing electron to obtain
these quantities. The process is an inclusive one, that is to say, no conditions are imposed on the hadronic
final state. The variable y has a simple interpretation in the laboratory frame of a fixed target experiment:
it is the fractional energy loss of the electron.
The corresponding partonic process is the scattering of a charged parton, that is to say a quark
or an antiquark, with the electron. The cross section for this process is easily computed, by using the

Fig. 17: DIS in the parton model.

standard Feynman rules of electrodynamics

dσ̂l ŝ
= c2l 4 2παem
2
1 + (1 − ŷ)2
¡ ¢
(95)
dŷ Q
where l runs over all quarks and antiquarks, and cl is the corresponding electric charge. The kinematics
is given by

p̂ · q
p̂ = xp ŝ = (k + p̂)2 = 2k · p̂ , ŷ = , (p̂ + q)2 = 2p̂ · q − Q2 = 0 . (96)
k · p̂
Observe that eq. 95 is a full cross section, properly normalized, divided by the appropriate flux factors.
Now we write, according to the parton model, the hadronic cross section

dσ dσ̂l
Z X
= dx fl (x) . (97)
dŷ dŷ
l

In order to obtain formula (97) we have only used the composition of probabilities, and the fact that cross
sections are invariant for longitudinal boosts. We now observe that

p·q p̂ · q Q2 Q2
y= = = ŷ , xBj = =x = x, (98)
k·p k · p̂ 2p · q 2p̂ · q

33
and thus we have
dσ X dσ̂l 2 Sx ¡
2παem ¢X 2
Bj 2
= fl (x) = 1 + (1 − y) cl fl (xBj ) . (99)
dy dxBj dŷ Q4
l l

Observe that y has a simple interpretation also in the centre-of-mass of the electron-quark system, where
it is given by y = (1 − cos θel )/2, and θel is the scattering angle of the electron in this frame.
In its simplicity, the parton model makes rather striking predictions. First of all, it shows that the
DIS cross section scales with energy at fixed xBj and y. Furthermore, the y dependence of the cross
section is fully predicted. As we will discuss further on, this y dependence is typical of vector interaction
with fermions, and is thus direct evidence of the fact that charged partons are fermions (this is formally
expressed by the so called Callan-Gross relation, as we will see in subsequent chapters).
The same type of reasoning can be applied also to other processes. For example, in a collision
of two hadrons, a quark from one hadron may annihilate with an antiquark from the other hadron, and
produce a lepton-antilepton pair, provided there are enough antiquarks in the projectile, like in pion-
nucleon collisions, or in proton-antiproton collisions. This is the so-called Drell-Yan process. Its parton

Fig. 18: Drell-Yan pair production in the parton model.

model interpretation is illustrated in fig. 18. As before, we define the partonic variables:

p̂1 = x1 p1 , p̂2 = x2 p2 , S = (p1 + p2 )2 = 2 p1 p2 , Q2 = q 2 = 2 x1 x2 S . (100)

The partonic cross section is given by

(DY) 4παem
σ̂l = c2l , (101)
9Q2

which is very similar to the cross section for e+ e− → µ+ µ− , except for en extra factor of 1/31 . Accord-
ing to the parton model interpretation, the hadronic cross section is
XZ ³ ´ 4πα
(H ) (H ) em
σ (DY)
= dx1 dx2 fl 1 (x1 ) fl̄ 2 (x2 ) + (l ↔ ¯l) c2l , (102)
9Q2
l

for Q2 = ŝ = x1 x2 S. The validity of the above formula is restricted to the range where Q2 is large. It
is therefore usually written as

dσ (DY) X 4παem
Z ³ ´
(H ) (H )
= dx1 dx2 fl 1 (x1 ) fl̄ 2 (x2 ) + (l ↔ ¯l) δ(x1 x2 S − Q2 )c2l . (103)
dQ2 9Q2
l
1
This comes from the colour average for the initial state quark. Its physical meaning is that, in the average, the probability
for the colour of the initial quark to match that of the antiquark is 1/3.

34
Pushing further our parton model interpretation of hard scattering processes, we can go on and compute
the cross section for producing high transverse momentum jets, of heavy bb̄ pairs, of tt̄ pairs, and so on.
In these processes, also gluons could enter in the initial state.
Not all hadronic processes can be computed in this way. For example, Drell-Yan cross sections,
forQ2 approaching typical hadronic scales, cannot be computed. The rule of thumb for deciding if a
process is a hard process or not, in the context of the parton model, is to ask whether it is insensitive
to the initial transverse momentum of the partons, which is of the order of typical hadronic scales. The
parton densities do not carry any information about this quantity.

5.2 Does the Parton Model survive radiative corrections?

We will now try to add perturbative QCD corrections to the Parton Model. As in the case of e+ e− →
hadrons, we will find soft and collinear singularities associated to radiation of gluons from final state
partons, which we expect to cancel for appropriately defined final states. This is the case, for example,
in fully inclusive hadronic final states, like in DIS or in Drell-Yan pair production, or in the production
of Sterman-Weinberg jets.
A new element that can arise in the case of reactions initiated by hadrons, is the appearance of
initial state soft and collinear singularities. We will show that initial state collinear singularities cannot
possibly cancel, and thus spoil the Parton Model interpretation of hard processes. Let us thus consider a
generic hard process initiated by a hadron, and its parton cross section, which we assume for simplicity
to be initiated by a quark

= M(p̂)u(p̂) . (104)

Here M indicates the amplitude for the process, and u is the Dirac spinor. All the complexity of the
process is hidden in M, and we don’t care about it for the moment. The cross section is obtained by
squaring the amplitude, averaging over the initial state spin and colors, and dividing by the appropriate
flux factors
N 1X N p
/̂
σ (0) (p̂) = 0 M(p̂) u(p̂)ū(p̂)M† (p̂) = 0 M(p̂) M† (p̂) (105)
p̂ 2 p̂ 2
where N is whatever normalization factor arises from the rest of the amplitude.
We want to focus upon the initial state corrections

/̂ − /l µ
p
= gs M(p̂ − l) γ u(p̂) ǫµ (l) , (106)
(p̂ − l)2

where ǫµ (l) is the polarization vector of the final gluon. We also observe that this may not be the only

35
correction of order αS . One may also have a process in which an initial gluon splits into a quark-antiquark
pair, and the generated quark gives rise to the reaction

We will assume that this complication does not occur. For example, we may assume that the hard cross
section measures some effect due to the difference of the quark content for two different flavours. Since
the gluon produces equal number of quarks for all flavours, it could not contribute in this case. In
these cases, one says that the cross section is only sensitive to the non-singlet component of the parton
densities. We thus concentrate on the non-singlet case now. Further on we will describe how to treat the
general case.
Experience with the e+ e− case tells us that as l becomes parallel to p̂ we will have a collinear
singularity. It is convenient thus to write l in the following way

l = (1 − z)p̂ + l⊥ + ξη (107)

where η is an arbitrary vector such that η 2 = 0 and η · p̂ 6= 0. For example, in the centre-of-mass frame
of the collision process we can choose

p̂ = (p̂0 , 0⊥ , p̂0 ) , η = (1, 0⊥ , −1) . (108)

The phase space for the emission of the gluon is

d3 l d4 l 2 p̂ · η dξ dz d2 l⊥ ¡ ¯ 2 ¯¢
0 3
= 4
2π δ(l ) = 3
δ 2 p̂ · η (1 − z)ξ − ¯l⊥ ¯
2 l (2π) (2π) (2π)
d2 l⊥ dz
= . (109)
2 (2π)3 1 − z

which yields, from the on-shell condition for the gluon,

¯ 2¯ ¯ 2¯
¯l ¯ ¯l ¯
⊥
ξ= and (p̂ − l) = − ⊥ .
2
(110)
2 p̂ · η (1 − z) 1−z

The most singular part of this cross section can be obtained similarly with what was done in the case
of e+ e− annihilation. It does not make much sense, in this case, to assume that l is small, and thus
the derivation is a little bit more involved. It is nevertheless instructive, so I will report it in the next
subsection. People who are willing to accept the result without discussion, can skip it.

5.3 Derivation of the singular part of the cross section

The amplitude in eq. (106), using our kinematic definitions, can be written as

/̂ − /l
p
gs M(p̂ − l) ¯ ¯
2
γ µ u(p̂)ǫµ (l) . (111)
− l /(1 − z)
¯
⊥
¯

When squared, it seems to give rise to terms of order 1/l⊥ 4 . We will see that these terms, however,
(i)
cancel. The trick is to make careful use the relation lµ ǫµ (l) = 0. The singular region is the one when l
is collinear to p̂, that is to say when l⊥ vanishes. In this region l ≈ (1 − z)p, and thus p ≈ l/(1 − z), up

36
to small corrections. Inserting this expression for p in eq. (111) will lead to simple Dirac algebra, since
by anticommuting l with γ µ we get lµ , which vanishes when dotted into the polarization. We thus write
l − l⊥ − ξη
p= (112)
1−z
and replace it in eq. (111). The term in ξ kills the singularity, and we drop it, since we are only interested
in the singular part. We obtain
z/l − /l⊥ µ
gs M(p̂ − l) ¯ ¯ γ u(p̂)ǫµ (l) ,
2¯
(113)
− ¯l ⊥
which becomes
−zγµ /l − /l⊥ γµ
gs M(p̂ − l) ¯ ¯
2¯
u(p̂)ǫµ (l)
− ¯l ⊥
−zγµ [(1 − z)/
p + /l⊥ ] − /l⊥ γµ
= gs M(p̂ − l) ¯ ¯
2¯
u(p̂)ǫµ (l) (114)
− ¯l ⊥
−zγµ /l⊥ − /l⊥ γµ
= gs M(p̂ − l) ¯ ¯
2¯
u(p̂)ǫµ (l) ,
− ¯l ⊥
µ
−2zl⊥ − (1 − z)/l⊥ γµ
= gs M(p̂ − l) ¯ ¯
2¯
u(p̂)ǫµ (l) , (115)
− ¯l ⊥
where the first step is obtained by anticommuting /l and γ µ , which we can do as explained before. Then
we rewrite l in terms of p. Next, we drop the p / term, since it is in front of the spinor u(p̂), and thus gives
zero, according to the Dirac equation. Finally, we use the anticommutation relation γµ /l⊥ = −/l⊥ γµ +2lµ⊥
In this last form, the singularity appears to be at most of order 1/ |l⊥ |, so that the amplitude squared will
give at most a 1/l⊥ 2 singularity. The rest is simple algebra. We square eq. (115), replace the gluon spin

sum with the transverse projector −gµν ⊥ , replace the fermion spin averaged product u(p̂)ū(p̂) with p̂/2,

and obtain
1 µ ¢p
/̂
gs2 − (1 − z)/l⊥ γ µ ν
− (1 − z)γ ν /l⊥ ) (−gµν
⊥
) M† (p̂ − l)
¡
4 M(p̂ − l) −2zl⊥ (−2zl⊥
l⊥ 2
1 p
/̂ ¡
= gs2 4 M(p̂ − l) 4z 2 ¯l⊥
¯ 2¯
¯ + 4z(1 − z) ¯l2 ¯ + 2(1 − z)2 ¯l2 ¯ M† (p̂ − l)
¯ ¯ ¯ ¯¢
⊥ ⊥
l⊥ 2
2 ¡ p
/̂
= gs2 ¯¯ 2 ¯¯ 1 + z 2 M(p̂ − l) M† (p̂ − l) .
¢
(116)
l⊥ 2

To get the cross section, we should multiply the above expression by N/p̂2 , and integrate over the phase
space. We obtain
2
αS CF 1 + z 2 dl⊥
Z
(1)
σ = σ (0) (zp) 2 dz . (117)
2π 1 − z l⊥
where
/̂ − /l
p p
/̂
σ (0) (zp) = N M(p̂ − l) 0
M† (p̂ − l) = N M(p̂ − l) M† (p̂ − l) . (118)
2(p̂ − l) 2(p̂)0
where we have made use of the relation gs2 = 4παS . The factor CF = 4/3 arises from the colour algebra.
It can be obtained according to the colour Feynman rules of fig. 3, as illustrated in the graphic equation

. (119)

37
There we see a factor of 3 arising in the first term, because of the sum over the colour entering the Born
amplitude, and a factor of 3 in the second because of the colour loop, the net effect being (3 + 1/3)/2 =
4/3.
The result obtained so far arises from the real emission of a gluon. Virtual corrections are also
present, i.e. a gluon can be emitted and reabsorbed by the same line.

5.4 Effects due to the emission of a collinear gluon

The final result is
αS CF
Z h i 1 + z 2 dl2
(1)
σ = σ (0) (z p̂) − σ (0) (p̂) ⊥
2 dz , (120)
2π 1 − z l⊥
where the second term in squared parenthesis is due to the virtual corrections. We see that there is a
singularity at z = 1 which cancels between real and virtual corrections. The region z → 1 corresponds
to soft gluon emission. Thus, soft singularities cancel. There are also collinear singularities, associated
to the small l⊥ region. These do not cancel.
We first make the following remark. In the initial amplitudes, the presence of a denominator of
the form 1/l⊥2 may seem to give rise to divergences like d2 l /l4 . The singularity we find at the end is
⊥ ⊥
instead weaker, of order d2 l⊥ /l⊥2 , because of an l2 we find from the numerator algebra. We can easily
⊥
convince ourselves that this is a consequence of angular momentum conservation. Vector interaction,
in fact, do not change the helicity of a particle. Thus the helicity of the incoming quark must be equal
to the that of the outgoing quark. On the other hand, physical gluons have ±1 helicity. Thus, in the
collinear limit, the total angular momentum contributed by spin is not conserved. This gives rise to the
2 suppression in the cross section. Also, by dimensional analysis, we see that we cannot expect
extra l⊥
divergences stronger than d2 l⊥ /l⊥2 in theories with dimensionless coupling constants.

In the case of e+ e− → hadrons, we made the approximation that z ≈ 1, for simplicity. If we had
been more careful, instead of formula (75), we would have obtained a formula similar to eq. (120). There
would be, however, a very important difference: in the Born cross section for the real emission, under
the integral sign, we would have σ (0) (p̂) instead of σ (0) (z p̂). This property is characteristic of splitting
processes taking place in the final state, rather than in the initial state. Figure 19 illustrate this fact. This

Fig. 19: Collinear processes in the final and in the initial state.

is the reason why collinear singularities cancel in the e+ e− → hadrons case, and do not cancel in this
case.
Equation (120) exhibit a rather intuitive property of collinear emission. Since the singularities are
due to the fact that the intermediate propagator goes near its mass shell, the intermediate particle travels
for a relatively long time and distance. Thus, when it initiates the interaction, it behaves essentially
like an on-shell particle, and the phenomenon can be described in probabilistic terms. In other words,
the total amplitude squared for the splitting process and the hard scattering, becomes the product of the
square of the amplitude for the splitting process, times the square of the amplitude for the hard scattering
(i.e., the cross section).
2 integral is divergent in the lower limit. Its upper limit is instead some scale, of the order
The l⊥
of the typical momenta involved in the hard process, which we now call Q. Equation (120) can then be

38
interpreted intuitively in the following way. In a hard process, taking place in a time of order 1/Q (by
the Heisenberg indeterminacy principle), an incoming parton is also probed for a time of order 1/Q. In
a short period of time, a quantum state may fluctuate into states to which it couples, even if they have
energies that differ by an amount of order Q or less. This is what happens to our incoming quark. This
also explain why the larger is Q, the more likely is the splitting to take place.

5.5 Failure of the parton model

The presence of collinear divergences tells us that there must be something wrong with the parton model.
Of course, we know that divergences, in the real physical world, are never there. In our case, for example,
if we introduce the mass of the quark, the divergence goes away. Or, we may use the known fact that
at low scale confinement effects take place, and thus put a lower cutoff of order Λ in the transverse
momentum integral. Or again, we may remember that the parton is off-shell in the incoming nucleon, by
an amount of order Λ. This also would act as a cut-off. However, neither of these remedies would really
solve the problem. Our cross section would become strongly dependent upon low energy details, like
the quark mass, the off-shellness in the nucleon, or confinement effects, while the Parton Model assumes
that these details do not count. Furthermore, the physics of these details is low scale physics, and is thus
uncalculable in perturbative QCD.
We will now show that, in spite of the collinear divergences, the Parton Model can be rescued,
provided we accept to make some modifications to the original concept.
We begin by introducing some notation. First of all we define
1 + z2
µ ¶
(0)
Pqq (z) = CF (121)
1−z +
where the notation with the + suffix is called the plus prescription. It specifies that the expression in
parenthesis is to be interpreted as a distribution, and its integral against a smooth function f (z) is given
by Z 1µ Z 1
1 + z2 1 + z2
¶
f (z) dz = (f (z) − f (1)) . (122)
0 1−z 0 1−z
We then introduce an infrared cutoff λ, and rewrite formula 120 as
αS Q2
Z
σ (1) = log 2 0
dz Pqq (z) σ (0) (z p̂) , (123)
2π λ
where Q is a characteristic scale in the process. Since The corrected partonic cross section can be written
as Z
σ(p̂) = σ (p̂) + σ (p̂) = dz Γqq (z, Q2 ) σ (0) (z p̂)
(0) (1)
(124)

where
αS Q2 0
Γqq (z, Q2 ) = δ(1 − z) + log 2 Pqq (z) . (125)
2π λ
The form of equation (124) hints to a possible way to resque the parton model approach. In fact, it
has the form of the parton model cross section, except for the Q2 dependence. It is telling us that we
should consider a parton as having a structure, that depends upon the scale at which we are probing it.
This becomes even more apparent if we insert the corrected formula in the parton model formula for the
hadronic cross section. We just replace p̂ = yp, multiply by the parton density fq (y) and integrate in y:
Z
σ(p) = dy dz fq (y) Γqq (z, Q2 ) σ (0) (zyp) . (126)

This formula represents the probablity to find parton q in the hadron, with a fraction y of its momentum,
times the probability to find parton q in parton q with a fraction z of its momentum times the cross

39
section for the final parton, with momentum yzp. It is natural to think that if we have an object that
can be represented as a beam of constituents, and the constituents can be represented as a beam of
subconstituents, the same object can be Rrepresented as a beam of subconstituents. Mathematically this
works as follows. We insert the identity dx δ(x − yz) in equation 126, and obtain
Z
σ(p) = dxf˜q (x, Q2 ) σ (0) (xp) . (127)

where we have defined

Z
f˜q (x, Q2 ) = dydz fq (y) Γqq (z, Q2 ) δ(x − zy) . (128)

We are getting closer and closer to the improved parton model formula. In fact, if instead of using the
process scale Q we introduce an intermediate scale µ, the connection becomes even clearer. We can
write Z
σ(p) = dxf˜q (x, µ2 ) σ̂(xp, µ2 ) , (129)

where
αS Q2
Z
(0) 0
σ̂(p̂) = σ (p̂) + log 2 dzPqq (z)σ (0) (z p̂) . (130)
2π µ
Equation (129) is easily verified by expanding the product of f˜ and σ̂, neglecting terms of order αS2 ,
and combining the logarithms according to the equation log µ2 /λ2 + log Q2 /µ2 = log Q2 /λ2 . It is
the QCD-improved parton model formula we were seeking, and it forms the basis for the application of
perturbative QCD to phenomena initiated by hadrons. A considerable difference with the “naive” Parton
Model formula is the appearance of a scale µ in the parton densities.
Let us summarize we have done so far. We have attempted to compute radiative corrections to
a parton process. We have found that these corrections are large, and depend upon unknown low scale
dynamics, which is represented here by the cutoff λ. However, we have found that these large corrections
can be absorbed into a redefinition of the parton densities. The parton densities redefinition does not
depend upon the hard process in question: it is universal. The physical cross section can then be defined
in terms of these new parton densities. Instead of the partonic cross section, in the QCD-improved parton
model formula we have a so called short distance cross section σ̂. This is obtained by subtracting the
infrared sensitive (or long distance) part from the partonic cross section. Thus, the short-distance cross
section is controlled by high momenta, and is thus calculable in perturbation theory. It is important to
choose the scale µ of the order of the scale Q of the hard process, in order to avoid the appearance of
large logarithms in the perturbative expansion.
Of course, our argument was only carried out at leading order in perturbation theory. There is a
variety of more complex arguments that show that formula (87) actually holds to all order in perturbation
theory. This is called the Factorization Theorem [40]. We will comment later on its present status. For
now, we will assume that the procedure outlined above can in fact be carried out to all orders in the
coupling constant. Thus, the short-distance cross section can be given as a power expansion in αS . If
the scale at which αS is evaluated is near the typical scale of the hard process, no large logarithms can
appear in the coefficients of the expansion, since all the scales entering in the coefficients are of the same
order. Thus, one can improve the accuracy of the short distance cross section by computing higher and
higher orders in perturbation theory. The scale µ introduced in this context is called the factorization
scale. The scale at which αS is evaluated is the renormalization scale, and should be of the same order as
the factorization scale. In principle, they can be taken to be different. Here, for simplicity, I will always
assume that the renormalization and factorization scales are taken equal.
The new pdf f˜(µ) contains uncalculable long distance effects. It has to be measured, by using
formula (87) with some reference hard process, which is typically chosen to be DIS. One then extracts

40
f (µ) at a given scale µ. Its µ dependence is however calculable. In fact, taking the derivative of eq. (128)
we get
∂ dΓqq (z, µ2 )
Z
f˜q (x, µ2
) = dydz f q (y) δ(x − zy)
∂ log µ2 d log µ2
αS
Z
0
= dydz fq (y) Pqq (z)δ(x − zy)
2π
αS
Z
= dydz f˜q (y, µ2 ) Pqq
0
(z)δ(x − zy) + O(αS2 ) , (131)
2π
where we have used eq. 125, and dropped higher order terms in αS in order to identify f with f˜ in the
last step.
Equation (131) is the Altarelli–Parisi (AP) equation (or Dokshitzer–Gribov–Lipatov–Altarelli–
Parisi equation) for the non-singlet case. It is also commonly written in the form
∂ αS 1 dy
Z
2
fq (x, µ ) = fq (y, µ2 ) Pqq
0
(x/y) , (132)
∂ log µ2 2π x y
where we have dropped the tilde sign, since the “naive” parton density disappears in the improved parton
model approach.
The AP equation allows us to compute the parton densities at any scale, once we have measured
them at an initial scale. Thus, in the improved parton model, predictivity is not lost. As before, the
measurement of the pdf’s in one process (at one scale) allows one to extend the computation to any
scale.

5.6 The evolution equations in the general case

We introduce the following symbolic notation for the AP equation
∂ X
µ2 2 fi (µ) = Pij ⊗ fj (µ) , (133)
∂µ
j

where the ⊗ product is defined as

Z 1
f1 ⊗ f2 ⊗ . . . fn (x) = dx1 dx2 . . . dxn f1 (x1 ) f2 (x2 ) . . . fn (xn ) δ(x − x1 x2 . . . xn ) . (134)
0

We have µ ¶2
αS (µ) (0) αS (µ) (1)
Pij (y) = P (y) + Pij (y) + . . . (135)
2π ij 2π
(0) (1)
where the Pij (y) are given in ref. [37], and the Pij (y) in [38, 39]. The terms of order αS3 are not yet
known exactly, although recently approximate expressions have become available [41], based upon some
partial results [42] [43]. Work on an exact calculation is under way [44].
(0)
We report below the formulae for the Pij (y). Its only non-vanishing components are

1 + x2
µ ¶
(0) (0)
Pqq (x) = Pq̄q̄ (x) = CF , (136)
1−x +
(0) (0)
(x) = Pq̄g (x) = Tf x2 + (1 − x)2 ,
¡ ¢
Pqg (137)

(0) (0) 1 + (1 − x)2

Pgq (x) = Pgq̄ (x) = CF , (138)
x
· µ ¶ µ ¶ ¸
(0) 1 1−z 11 nf
Pgg (x) = 2CA z + + z(1 − z) + − δ(1 − x) (139)
1−z + z 12 6CA

41
For a derivation of the above formulae similar to the one given in subsection 5.3, the reader can look in
Appendix B of ref. [45]. For a more intuitive (although less conventional) derivation, the reader can look
directly in the original Altarelli-Parisi paper [37].
(1)
We do not report here the higher order Pij (y) functions. Observe, however, that at higher orders
the components Pqi qj for i 6= j and Pqi q̄j (for any i and j) do arise. Here we limit our discussion, for
simplicity, to leading order evolution only.
We begin by taking the difference of eq. (133) with itself, for two different quark or antiquark
flavour labels i and j. We find
∂ X
µ2 2
(fi (µ) − fj (µ)) = (Pik ⊗ fk (µ) − Pjk ⊗ fk (µ)) . (140)
∂µ
k

As discussed earlier, if i is a quark (or antiquark), then k can only be the same quark (or antiquark) or a
gluon. The gluon contribution cancels among the two terms in parenthesis, and one gets

∂
µ2 (fi (µ) − fj (µ)) = Pqq ⊗ (fi (µ) − fj (µ)) . (141)
∂µ2
Thus, if we have nf light flavours, there are 2nf − 1 independent combinations of the parton densities
that evolve independently from each others. They are called non-singlet components. Next, we take the
sum of eq. (133) for all quark flavours and antiflavours. We get
X ∂ X XX X
fi (µ) = Pik ⊗ fk (µ) = Pik ⊗ fk (µ) + Pig ⊗ fg (µ)
∂µ2
i6=g i6=g i6=g k6=g i6=g
X
= Pqq ⊗ fi (µ) + 2nf Pig ⊗ fg (µ) . (142)
i6=g

On the other hand, eq. (133) for the gluon reads

∂ X X
2
fg (µ) = Pgi ⊗ fi (µ) = Pgi ⊗ fi (µ) + Pgg ⊗ fg (µ) . (143)
∂µ
i i6=g

Thus, defining X
S(µ) = fi (µ) , (144)
i6=g

we get the system of equations

∂
µ2 fg (µ) = Pgq ⊗ S(µ) + Pgg ⊗ fg (µ)
∂µ2
∂
µ2 2 S(µ) = Pqq ⊗ S(µ) + 2nf Pig ⊗ fg (µ) , (145)
∂µ
which define the evolution of the so called singlet component S and the gluon. Thus, while the non-
singlet components evolve independently, the singlet component mixes with the gluon density in its
evolution.

5.7 Sum rules

We said earlier that we expect our parton densities to satisfy certain sum rules. Thus, for example
Z h i
(p)
dx fu(p) (x) − fū (x) = 2 . (146)

42
We must make sure that evolution equations do not spoil the sum rules. Since the difference of the quark
and antiquark parton densities is a non-singlet component, we have
2 ∂ αS
Z h i Z h i
(p) (p) (p) (p)
µ dx fu (x) − fū (x) = dx P qq (y) fu (z) − fū (z) δ(x − yz) dy dz
∂µ2 2π
·Z ¸Z
αS h
(p)
i
= Pqq (y)dy dz fu(p) (z) − fū (z) = 0 (147)
2π
R
because Pqq (y)dy = 0. Similarly, one can show that the momentum sum rule is also preserved by
evolution.

5.8 Scheme dependence

There is some ambiguity in the way one defines the parton densities, This ambiguity is best seen as an
ambiguity in the type of infrared cutoff one uses. For example, one could give a mass to the quark, or
assume it is slightly off-shell. By doing this, the large logarithm does not change, but different finite
pieces can arise in the calculation. In the present context we have only looked at the divergent parts.
When doing next-to-leading QCD calculation, however, one would like to compute precisely the finite
pieces. The reader can find interesting examples in ref. [46, 47] and [48]. There the same processes are
computed (the Deep-Inelastic and the Drell-Yan cross section), but with different infrared cutoffs. Thus,
the finite terms in the various cross sections turn out to be different. However, when expressing the DY
cross section in terms of the DI cross section, both approaches yield the same formula. Thus, to some
extent, the definition of the parton density is a matter of convention, like the definition of αS . It has to
be specified together with a procedure for the computation of short distance cross section. Today, the so
called MS scheme is widely used, and most parton densities are given in the MS scheme.

5.9 Summary
We summarize what we have learned in this chapter.
First of all, by intuitive reasoning, we derived cross sections for high energy inclusive processes,
assuming that the transverse momentum of constituents in hadrons was limited to typical hadronic scales.
We tried to compute radiative corrections to these formulae, and we found inconsistencies, i.e.
uncancelling collinear divergences.
With a procedure very similar to renormalization, we showed that the collinear divergences can be
factorized into the parton densities.
Let us discuss how is the procedure of factorization similar to renormalization. In renormalization,
we hide our ignorance of UV effects into a redefinition of the strong coupling constant. Here, we hide
our inability to compute IR effects into a redefinition of the parton densities.
As a result of this procedure, we find that the parton densities are actually scale dependent. We
may think of a hard process as a probe of transverse dimensions and time of order 1/Q. When we probe
a constituent at higher and higher values of Q, that is to say for shorter and shorted time, because of the
uncertainty principle, we may find it fluctuating into a virtual pair of constituents off the energy shell by
an amount of order Q. The larger is Q the larger is the phase space for virtual particles. This is why
parton densities evolve with the scale at which they are measured.
The original assumption of limited transverse momenta fails in the parton model. We have seen,
in fact, that because of initial state radiation, integrals of the form d2 l⊥ /l⊥
2 arise. Roughly, we expect

Z 2
2 d l⊥ 2 2
hl⊥ i ≈ αS 2 l⊥ ≈ αS Q . (148)
l⊥
Thus the transverse momentum is not limited, but it is “perturbatively” small, i.e. it is suppressed by a
coupling constant factor.

43
5.10 How solid is the Factorization Theorem?
The argument given in this chapter does not certainly pretend to be fully convincing. Thus, we would
like to have a more solid proof of this theorem.
In the case of the DIS process, such proof exists. It relies upon a clever analytic continuation
property of the DIS cross section, that can be used to apply the powerful language of the operator-product
expansion (O.P.E.) to the problem.
For production processes in hadronic collisions, things are much more difficult. Even in the sim-
plest case, the Drell-Yan process, the factorization theorem has a long controversial history, which was
finally settled by the calculation of Lindsay, Ross and Sachrajda [49, 50, 51]. All-order arguments for
factorization have been given in ref. [52]. Today, the factorization theorem is widely accepted in the high
energy physics community.

6 DEEP INELASTIC SCATTERING

Deep-Inelastic Scattering (DIS) is the next-to-simplest QCD process after e+ e− annihilation into hadrons.
It is experimentally quite simple, since in order to define the DIS cross section one does not need to intro-
duce jet definitions. It is enough to measure the momentum of the outgoing lepton in order characterize
the final state.
Deep-Inelastic scattering is also the best place where to measure structure functions, as can be seen
from eq. (99). Thus, QCD predictions for hadronic collisions rely upon the experimental determination
of structure functions performed at DIS experiments.
From a theoretical point of view, DIS (like Re+ e− ) has also a privileged status. There are in fact
good reasons to believe that power corrections in DIS processes behave like 1/Q2 . This is unlike (for
example) jets in e+ e− annihilation, where one expects corrections of the order of 1/Q. Thus, DIS is a
good place where to measure αS .
The most general form of the DIS cross section for electromagnetic processes is given by
2 (S − M )2 xyM 2
·µ ¶ ¸
dσ 4παem 2 2 2
= 1−y− F2 (x, Q ) + y x F1 (x, Q ) , (149)
dx dy Q4 S − M2

where F2 and F1 are called the structure functions for DIS, y corresponds to the variables defined previ-
ously, M is the mass of the target nucleon and x = xBj . I will not illustrate the derivation of this formula,
which is found in many textbooks. It is a simple consequence of electrodynamics at the lowest order
in αem , and of Lorentz invariance. It does not, therefore, contain any dynamical consequence of strong
interactions, aside from its symmetry properties. From formula (99), and after what we have said in
the previous chapter with regard to the factorization theorem, we can now write down the leading order,
QCD-improved parton model formula for DIS

dσ 2 Sx ¡
2παem ¢X 2
Bj 2
= 1 + (1 − y) cl fl (x, Q) . (150)
dy dx Q4
l

In order to have leading order accuracy, it is sufficient to choose µ ≈ Q. For simplicity, I have chosen
µ = Q. From eqs. (149) and (150), neglecting mass effects, we find

F2 (x, Q) = 2xF1 (x, Q) , (151)

which is the so-called Callan-Gross relation, and

X
F2 (x, Q) = x c2l fl (x, Q) . (152)
l

44
The Callan–Gross relation is a prediction of the parton model, and it is a consequence of the fact that
the only charged partons are fermions. It is however only a leading order prediction. When radiative
corrections are included, it is violated. One defines FL = F2 − 2xF1 .
It is useful to focus now upon the y dependence of the parton model formula. We have
p̂ · q p̂ · k ′ 1 − cos θ
y= =1− = , (153)
p̂ · k p̂ · k 2
and thus y is related to the electron scattering angle θ in the CM frame of the electron-parton collision
(sometimes called the partonic CM frame).
The scattering of the lepton on a quark of the same helicity, gives rise to a y dependence propor-
tional to 1, while in the case of a quark of different helicity, the y dependence is (1 − y)2 . Thus, in
the case of spin-averaged cross sections in electromagnetism, the y dependence is 1 + (1 − y)2 . The
verification of these properties is a simple exercise with Feynman graphs.
The vanishing of the cross section in the backward limit (i.e. y = 1) for the quarks and lepton
with opposite helicity has a simple intuitive explanation. The spins of the lepton and the quark are
aligned, since their helicities are opposite, and their momenta are opposite. Thus, they have a total
angular momentum 1 in the collision direction. Vector interactions conserve helicities. Thus, the quark
and lepton will have the same helicity after the interaction. In the case of backward scattering, however,
they have opposite momentum, and thus they have opposite total spin. Thus, conservation of angular
momentum imposes the vanishing of the backward cross section, which is what the (1 − y)2 dependence
predicts.
Parity violating processes contribute anti-symmetrically in the exchange of the helicity of the
incoming lepton. We expect a (1 − (1 − y)2 ) = 2(y − y 2 /2) dependence to be present in this case. Thus,
a third structure function appears. For example, in neutrino charged current DIS (i.e. νµ N → µ− X or
ν̄µ N → µ+ X) we have
"µ
G2F (S − M 2 ) 2 xyM 2
¶
dσ MW
= 1−y− F2cc (x, Q2 )
dx dy 2π (Q2 + MW2 )2 S − M2
#
+y 2 x F1cc (x, Q2 ) ± (y − y 2 /2) xF3cc , (154)

where the sign in front of F3 is chosen positive for ν, and negative for ν̄ interactions. The parton cross
section is given by

G2 ŝ 2
½
dσ MW 1 same helicities
= F . (155)
dy π (Q2 + MW 2 )2 (1 − y)2 opposite helicities

The neutrino is left handed, and charged current interactions involve left-handed quarks and their an-
tiparticles, which are right-handed. Thus, when the neutrino scatters off a quarks, we get the constant y
dependence; when it scatters off an antiquarks, we get the (1 − y)2 dependece. Because of charge con-
servation (i.e., the neutrino goes into an electron, and thus gives one unit of positive charge to the quark)
only negatively charged quarks or antiquarks can be involved. Thus, for example, for νµ p → µ− X,
neglecting for the moment a possible charm or bottom parton density in the proton we have

dσ G2 Sx MW2
(d(x, Q) + s(x, Q)) + (1 − y)2 ū(x, Q) ,
£ ¤
= F 2 (156)
dx dy π (Q2 + MW2)

Here we introduce the notation

(p)
u(x, Q) = fu(p) (x, Q) , d(x, Q) = fd (x, Q) , etc. (157)

45
for the quark densities in the proton. The corresponding densities in the neutron are obtain from isospin
symmetry
(n)
fu(n) (x, Q) = d(x, Q) , fd (x, Q) = u(x, Q) , etc. . (158)
Thus

F2cc (x, Q) = 2xF1cc (x, Q) = 2x(d(x, Q) + s(x, Q) + ū(x, Q)) (159)

F3cc (x, Q) = 2(d(x, Q) + s(x, Q) − ū(x, Q)) . (160)

Similarly, for ν̄p → e+ X

¯ Q))
F2cc (x, Q) = 2(u(x, Q) + s(x, Q) + d(x, (161)
¯ Q) − s̄(x, Q) + u(x, Q)) .
F cc (x, Q) = 2(−d(x, (162)
3

One gets the sum rule

Z 1
dx F3ν̄p (x, Q) + F3νp (x, Q) =
£ ¤
(163)
0
Z 1
¯ Q) + s(x, Q) − s̄(x, Q) + . . . = 6
£ ¤
2 dx u(x, Q) − ū(x, Q) + d(x, Q) − d(x,
0

which is called Gross–Llewellyn Smith sum rule, and expresses the fact that there are three quarks in a
proton.
The phenomenology of DIS scattering is quite complex, and it is really impossible to review it in a
satisfactory way in the context of these lectures. Several complications of experimental nature arise, and
have to be dealt with properly. When extracting the structure functions F1 or F2 from data, it is usually
assumed that F1 and F2 are related on the basis of the Callan–Gross relation
1 + 4M 2 x2 /Q2
2xF1 (x, Q) = F2 (x, Q) × (164)
1 + R(x, Q2 )
where, if the Callan–Gross relation was satisfied exactly, one would have R = 0. Different experiments
are performed on different targets. The structure functions for a nucleon embedded in a nucleus are
distorted (EMC effect). Finally, the size of power suppressed effects (the so called higher twist effects)
should be assessed, especially for low Q2 experiments. In the present context I will not try to explain
how to deal with these complications. I will instead try to give a rough idea of how the strong coupling
constant and the parton densities are extracted from data.
The strong coupling constant can be extracted from DIS data using sum rules, like the Gross–
Llewellyn Smith sum rule. Sum rules are in fact calculable in perturbative QCD, and the difference from
their parton model value can be used to extract αS . For the Gross–Lewellyn Smith sum rule
Z 1
dx F3ν̄p (x, Q) + F3νp (x, Q) =
£ ¤
0
· µ ³ α ´2 ¶ ¸
αS αS
+ O(αS4 ) − ∆HT .
S
6 1− × 1 + 3.58 + 19 (165)
π π π
A recent CCFR determination [53] obtains

αS (1.73GeV ) = 0.280+0.070 +0.009

−0.068 → αS (MZ ) = 0.114−0.012 . (166)

These determinations have the advantage that these quantities have been computed at very high order in
perturbation theory [54] , and thus the theoretical error are reduced. Since, however, they are performed
at a rather low scale, some estimate of higher twist effects (the ∆HT ) are necessary.

46
 Q ∆αS (MZ )
Measurements (GeV) αs (Q) αs (mZ ) exp. theor. Theory
DIS, GLS-sr 1.73 0.280 +0.070
−0.068 0.114 +0.009
−0.012
+0.008
−0.010 ±0.005 NNLO
DIS, ν; xF3 5 0.214 ± 0.021 0.118 ± 0.006 ±0.005 ±0.003 NNLO
DIS, e/µ; F2 2.96 0.252 ± 0.011 0.1172 ± 0.0024 ±0.0017 ±0.0017 NNLO

Table 5: Determinations of αS from DIS data, taken from ref. [11]. GLS-sr stands for Gross-Llewellyn Smith sum rule.

The standard method to measure αS in DIS is based upon the fact that the speed of evolution is
proportional to αS . The logarithmic derivative of the structure functions with respect to Q2 are found
therefore to have a strong sensitivity to the value of αS . It is convenient to use a non-singlet structure
function, in order to avoid uncertainties due to the poor knowledge of the gluon density. Thus, for
example, one can use F3 in neutrino scattering [55]. Alternatively, one can use structure functions at very
large x. Since gluons are not valence particles, their density is quite soft, that is to say, concentrated at
small values of x. In general, there is little gluon content in the hadrons for x > 0.2. Using this fact, one
can also use muon data to determine αS . A summary of αS measurements from DIS is reported in table 5
from ref. [11]. The table deserves some comments. First of all, notice that all these determinations are
performed at the NNLO level. This has become possible because of recent progress in the computation
of moments of the splitting functions at order αS3 [42] [43]. This has allowed NNLO analysis of DIS data
[56] [57]. The heoretical precision of these analysis matches that of Re+ e− . Comparing tables 5 and 2
we see a remarkable consistency in two different determinations, performed with completely different
experimental setups, and at very different scales.
Neutrino scattering allows independent access to the quark and antiquark content of nucleons. It
is generally carried out on heavy, approximately isosinglet targets. F2 measurements in electromagnetic
and charged current experiments give access to the combinations reported in the table 6. In principle,

F2ep /x 4
9 (u+ ū) + 19 (d + d¯ + s + s̄)
F2ed /x 5 ¯ + 2 (s + s̄)
9 (u+ ū + d + d) 9
F2νd 2(u + ū + d + d¯ + 2s)
F2ν̄d 2(u + ū + d + d¯ + 2s̄)
F3νd 2(u − ū + d − d¯ + 2s)
F3ν̄d 2(u − ū + d − d¯ − 2s̄)
Table 6: F2 in various experimental configurations of interest.

strange and antistrange content could be extracted from neutrino and antineutrino data on isosinglet
targets. Or, assuming s = s̄, we can use the combination 5/6F2νd − 3F2ed = x2s. In practice, the strange
content is better constrained by looking at charm production in neutrino DIS. The corresponding signal,
in the case of νµ scattering, is given by an unlike sign muon pair, one arising from the charged current
scattering, and the other from charm decay.
Assuming that we have measured the strange content, we have access to the combinations u + ū,
¯ ¯ These quantities are not independent, since the sum of the first two equals the sum
d + d, u + d and ū + d.
of the last two. Thus, one more input is needed. It was usually assumed that ū = d. ¯ This assumption,
supplemented with sum-rule restrictions, is however in conflict with data. In fact, using the flavour sum
rules Z Z
¯ Q) = 1 , ,
£ ¤
dx [u(x, Q) − ū(x, Q)] = 2 , dx d(x, Q) − d(x, (167)

we obtain
1
dx h (p)
Z i
(n)
F2 (x, Q) − F2 (x, Q)
0 x

47
 1
1£
Z
¯ Q)
¤
= u(x, Q) + ū(x, Q) − d(x, Q) − d(x,
dx
0 3
1 2 1
Z
¯ Q)
£ ¤
= + dx ū(x, Q) − d(x, (168)
3 3 0

which, if ū = d¯ gives the so called Gottfried sum rule. Experimental measurements of the Gottfried sum
favour a negative contribution from the ū − d¯ difference.
In order to access the ū − d¯ difference as a function of x, one has to use different experiments.
Drell-Yan pair production in proton-proton collisions is one example.
The x integrals of F2 are proportional to a combination of the momentum fraction carried by the
quarks and antiquarks. In particular, for example, the integral of F2νd gives the total momentum fraction
carried by quarks. This quantity is measured to be roughly 0.5. Thus, one expects that a large fraction
of the hadron momentum is carried by gluons. This poses a valuable constraint on the gluon density
g(x, Q). From DIS, the traditional way to determine g(x, Q) is from its influence upon the evolution of
the singlet structure functions. This is viable at relatively small values of x, where the gluon density is
not small. At large x, however, one needs to rely upon direct methods, since the gluon density is too
small there to influence evolution. Direct photon production is one such process.
Today’s tendency for structure function studies is to perform global fits to a large variety of data
samples. One recent description of structure functions fits is given in ref. [58], where many aspects are
discussed in detail. The result of these fits is shown in fig. 20.

MRST partons Q2 = 20 GeV2

2
xf(x,Q2)

1
g/10
u
d
c s

0
-4 -3 -2 -1
10 10 10 x 10 1

Fig. 20: Parton distributions by the MRST group.

7 QCD IN HADRONIC COLLISIONS

Perturbative QCD applications in hadronic collisions is extremely important, due to the impact it has had
in the recent past for the discovery of new particles, and the impact it is going to have in the future for
the search of new physics at the LHC. Thus there are essentially two main points of study for QCD at
hadron colliders, and they clearly go hand in hand

• QCD tests in hard processes

48
 • Modeling of particle production processes (computing cross sections for top, higgs, etc.) and
computing backgrounds.

Unlike the case of e+ e− annihilation into hadrons, where each event is a hard process, in hadronic
collisions most events are soft, even if the CM energy is very high. This is because, even if the colliding
energy is high, the momentum transfer involved is not large. However, in the production of very massive
particles, or in processes in which particles at high transverse momentum appear, hard momenta are
actually present, and we can apply perturbative QCD. As a rule of thumb, when we try to compute a
process using the parton model formula, and find that it is dominated by small momenta, this means that
we can no longer neglect low energy details, like the off-shellness of the partons inside the colliding
hadrons, or their mass. In this case, the process is controlled by long distance dynamics, and cannot be
computed using perturbative QCD.

7.1 The kinematic variables for hadronic collisions

Given the two colliding hadron beams, one defines the kinematical variables of any outgoing particles
according to the figure below

.
Thus, the transverse momentum k⊥ is the projection of the particle momentum into the transverse plane
(the plane orthogonal to the collision axis). The azimuthal angle φ is defined with respect to the collision
axis. One usually defines

Transverse energy = ET = sin θE

p
Transverse mass = mT = kT2 + m2
1 k0 + kk
Rapidity = y = log 0 .
2 k − kk
The rapidity has the nice property that under a longitudinal boost it is simply translated by the boost
angle: y → y + log γ. The transverse momentum, and thus the transverse mass, are simply invariant
under longitudinal boosts. Thus, these variables are particularly useful to study hard processes, since in
general the parton centre-of-mass system for the process will be translated with respect to the hadron
CM. For particles of small mass, we have

1 1 + cos θ θ
y≈ log = − log tan , (169)
2 1 − cos θ 2
and thus one defines the pseudorapidity

θ
η = − log tan . (170)
2
It is useful to remember the following formula for the single particle phase space

d3 k 1
0 3
= d2 kT dy . (171)
2k (2π) 2(2π)3

Thus, the single particle phase space is uniform in transverse momentum and rapidity.

49
7.2 Total cross section
The total hadronic cross section is in the range of several 10mb range, and it grows logarithmically with S.
This is roughly the inverse of few hundred MeV squared, the characteristic scale of strong interactions.
We cannot compute the total cross section using perturbative QCD. Phenomenological models based
upon Regge theory are usually employed to describe the data.
If we attempted to estimate the total cross section using parton model concept, we would end up
computing a parton production cross section integrated over the transverse momentum of the parton. On
dimensional ground, this cross section would be divergent at small transverse momenta

dσ 1 dkT2 1
Z
≈ ⇒ σ ≈ ≈ 2 (172)
dkT2 kT4 kT4 Λ

where the last step follows from the fact that some non-perturbative hadronic scale (for example, the
off-shellness of the incoming partons) should act as a lower cutoff of the integral. Thus, perturbation
theory, although incapable to give a definite answer, fails precisely at the point when the cross section
becomes of the order of the total cross section.

7.3 Typical inelastic processes

The typical inelastic events in hadronic collisions are quite complex. Several hadrons are produced, the
average charged multiplicity hnch i being typically of the order of 30 to 40 per event for Ecm = 600
to 1800 GeV, and it grows logarithmically with energy. Fluctuations in multiplicity are large, of the
order of 100%, a typical feature of cascade processes. The transverse momentum distribution of the
produced hadrons are characterized by an average transverse mass of the order of few hundred MeV,
growing slowly with energy. The produced particles are distributed uniformly in rapidity, the distribution
dropping smoothly to zero when approaching the maximum rapidity.

7.4 Looking for hard processes in hadronic collisions

Hadron collider physics is complicated by the fact that interesting events are rare with respect to the
common low pT inelastic events. This is immediately understood if we estimate the cross section for
the production of a 100 GeV object to be of the order of 10−4 GeV−2 , while the typical inelastic cross
section is of the order of 10−4 MeV−2 . We expect roughly 1 hard event every 106 soft ones, and this
estimate ignores eventual suppression due to the coupling constant.
Furthermore, soft events may look like hard ones, because of fluctuations. Thus, with a multiplicity
of 30 and an average pT of few hundred MeV, the average total transverse energy can very well be of the
order of tens of GeV. Fluctuations may favour occasionally even larger transverse momenta.

7.5 Jets at Hadron Colliders

Thus, unlike the e+ e− case, where above a certain energy all events look like jet events, in hadronic
collisions establishing the existence of jets has required the use of an appropriate trigger. In fact, one
has to look only at events with a large total transverse energy. If the total transverse energy is larger than
the typical value for a soft event, the events show the presence of jets. This was the method followed by
the UA2 and UA1 experiments at the CERN Spp̄S collider, to establish the existence of jets in hadronic
collisions. It was found there that requiring a transverse energy larger than 70 GeV, most events look like
jet events.
The description of jet production in QCD follows the lines of the QCD-improved parton model.
At the leading order level, in order to compute jet cross section we only need the Born cross sections for
parton parton scattering, reported in table 7. The 2-jet inclusive cross section can then be obtained from

50
 dσ̂
Process dΦ2
1 4 ŝ2 +û2
qq ′ → qq ′ h ³ 2 2ŝ2 9 t̂22 2 ´ i
8 ŝ2
qq → qq 12 2ŝ 1 4 ŝ +û
9 t̂2 + ŝ +t̂
û 2 − 27 ût̂
1 4 t̂2 +û2
q q̄ → q ′ q̄ ′ h ³ 2ŝ 9 ŝ2 ´ i
1 4 ŝ2 +û2 t̂2 +û2 8 û2
q q̄ → q q̄ 2ŝ 9 + ŝ2
− 27
h t̂2 i ŝt̂
1 1 32 t̂2 +û2 8 t̂2 +û2
q q̄ → gg 2 2ŝ h 27 t̂û − 3 ŝ2 i
1 1 t̂2 +û2 3 t̂2 +û2
gg → q q̄ 2ŝ h 6 t̂û − 8 ŝ2 i
1 4 ŝ2 +û2 2 2
gq → gq 2ŝ − 9 ŝû + û t̂+ŝ 2
³ ´
1 1 9 t̂û ŝû ŝt̂
gg → gg 2 2ŝ 2 3 − ŝ2
− t̂2
− û2

Table 7: Cross sections for light parton scattering. The notation is p1 p2 → k l, ŝ = (p1 + p2 )2 , t̂ = (p1 − k)2 , û = (p1 − l)2 .

the formula
X (H1 ) (H2 ) dσ̂ij→k+l
dσ = dx1 dx2 fi (x1 , µ) fj (x2 , µ) dΦ2 (173)
dΦ2
ijkl

that has to be expressed in term of the rapidity and transverse momentum of the quarks (or jets), in order
to make contact with physical reality. The two particle phase space is given by

d3 k
dΦ2 = 2π δ((p1 + p2 − k)2 ) , (174)
2k 0 (2π)3

and using eq. (171), in the CM of the colliding partons, we get

1
dΦ2 = d2 kT dy 2 δ(ŝ − 4(k 0 )2 ) . (175)
2(2π)2

Here y is the rapidity of the produced parton in the parton CM frame. It is given by
y1 − y 2
y= (176)
2
where y1 and y2 are the rapidities of the produced partons in the laboratory frame (in fact, in any frame).
One also introduces
y1 + y2 1 x1 ŝ
y0 = = log , τ = = x1 x2 . (177)
2 2 x2 s
We have
dx1 dx2 = dy0 dτ . (178)
We obtain
X 1 (H1 ) (H ) dσ̂ij→k+l 1
dσ = dy0 f (x1 , µ) fj 2 (x2 , µ) 2 dy d2 kT (179)
s i dΦ2 2(2π)2
ijkl

which can also be written as

dσ 1 X (H ) (H ) dσ̂ij→k+l
2
= 2
fi 1 (x1 , µ) fj 2 (x2 , µ) . (180)
dy1 dy2 d kT s 2(2π) dΦ2
ijkl

51
The variables x1 , x2 can be obtained from y1 , y2 and pT from the equations
y1 + y 2
y0 = (181)
2
y1 − y 2
y = (182)
2
2pT
xT = √ (183)
s
x1 = xT ey0 cosh y (184)
x2 = xT e−y0 cosh y . (185)

For the partonic variables, we need ŝ = s x1 x2 and the scattering angle in the parton CM frame θ, since
s s
t = − (1 − cos θ) , u = − (1 + cos θ) . (186)
2 2
Since we are neglecting parton masses, rapidity and pseudorapidity are identical, so that the equation
θ
y = − log tan (187)
2
gives us θ.
The Born cross section formulae given here predict the production of back-to-back jets, with op-
posite transverse momenta. Details of the jet distributions depend upon the knowledge of the structure
functions. However, it has been observed that, to a good approximation, scattering processes with gluon
exchange in the t channel dominate, and that they are roughly proportional to each other. More specif-
ically, the gg → gg, qg → qg and qq ′ → qq ′ processes are in the ratio 3 × 3, 3 × 4/3 and 4/3 × 4/3
respectively. This property is exact in the small angle scattering limit, but holds to a good approximation
also at large angles. It can be obtained from Table 7, by keeping only the most enhanced terms when
t → 0 (and u → −s) or when u → 0 (and t → −s). The processes with identical particles in the
final state have an extra factor of 1/2, but on he other hand have enhanced terms when t → 0 and when
u → 0, while those with different particles in the final state have only the t singularity. Thus, at the end,
the qq → qq process at small angle gives the same contribution as the qq ′ → qq ′ process.
Using this property the jet cross section simplifies
dσ 1 dσ̂gg→gg
2
≈ 2
F (H1 ) (x1 , µ) F (H2 ) (x2 , µ) . (188)
dy1 dy2 d kT s 2(2π) dΦ2
with
4 X (H)
F (H) (x, µ) = fg(H) (x, µ) + fi (x, µ) . (189)
9
i6=g

Equation (188) gives a definite prediction for the angular dependence of jet production. It can also be
written, more explicitly, in terms of x1 , x2 and cos θ, where θ is the scattering angle in the rest frame of
the partons.
dσ dσ̂gg→gg
= F (H1 ) (x1 , µ) F (H2 ) (x2 , µ) . (190)
dx1 dx2 d cos θ d cos θ
Early studies of the UA1 and UA2 experiments have confirmed this behaviour [59].
Modern studies of jet physics at colliders are performed at the next-to-leading level in QCD.
Calculations of jets cross sections at next-to-leading level have been available for quite a long time.
Comparisons between data and calculation require agreement on a jet definition to be used. Such a
definition should be of the Sterman-Weinberg type, that is to say, it should be infrared and collinear safe.
Several algorithms have been proposed to define jets. For the purpose of this lectures, it will be enough

52
 10 7

10 6 10 6
DØ Inclusive Jets |η| < 3, present measurement
0.0 ≤ |η| < 0.5
0.5 ≤ |η| < 1.0
5 CDF/DØ Inclusive Jets |η| < 0.7
10
10 5 1.0 ≤ |η| < 1.5

d 2 =(dET d ) (fb=GeV)
ZEUS 95 BPC+BPT+SVTX &
H1 95 SVTX + H1 96 ISR 1.5 ≤ |η| < 2.0
10 4 ZEUS 96-97 & H1 94-97 2.0 ≤ |η| < 3.0
E665 10 4

10 3 CCFR QCD{JETRAD
Q2 (GeV2)

BCDMS
10 3
NMC
10 2 SLAC
10 2
10

10
1

-1 1
10
50 100 150 200 250 300 350 400 450 500
(GeV)
-6 -5 -4 -3 -2 -1
10 10 10
x 10 10 10 1
ET

Fig. 21: The reach of the D0 inclusive jet analysis in the Q2 , x plane for the parton densities (left plot), and the Inclusive jet
cross section as a function of ET , in various rapidity bins, versus theoretical predictions (right plot).

to know that the most commonly used definitions make use of a circle of a given radius R in the φθ plane.
The circle is moved in the plane until one finds a maximum of the transverse energy deposition inside
the circle, and a jet of the given φη and ET values is associated with this point. The single inclusive
distribution of jets found in this way, as a function of ET , is compared with QCD NLO calculation.
An example of a recent measurement of the inclusive jet cross section is given in ref. [60], from the
D0 collaboration. The inclusive jet cross section is measured in a wide rapidity range. By exploring the
high rapidity region, one extends toward smaller values of x the region in the Q2 , x plane where parton
densities are probed, as shown in the left plot of fig. 21. Jets are defined with the ηφ cone algorithm,
with a radius R = 0.7. The D0 results, together with a NLO QCD predictions, are shown in the right
plot of fig. 21, showing a remarkable agreement. A more detailed comparison is shown in fig. 22, where
the ratio (data − theory)/theory is plotted. Theoretical results are obtained with the program JETRAD
[61], using the CTEQ4 [62] (left figure) and MRST [58] (right figure) structure functions. The shaded
band corresponds to one standard deviation on the systematic error. One expects a comparable band for
the theoretical error. The data is therefore in good agreement with theoretical predictions, showing a
preference for the CTEQ4 sets.
Double-inclusive jet cross section (i.e., dijet production) studies at the NLO have also become to
appear. CDF has performed a study of dijet production [63]. They look at the ET of one central jet
(0.1 < η1 < 0.7), while the second jet lies in several different pseudorapidity intervals. In this way, the
sensitivity to the parton densities at large x is enhanced. Qualitatively the theory gives a good description
of data, as can be seen from fig 23. A closer look reveals problems at the quantitative level. Looking at
the (data − theory)/theory ratio in the right plot of fig. 23, one sees that no parton density functions set
fits the data satisfactorily, especially in the high ET region.
We recall that jet studies at the Tevatron is at the frontier of our knowledge on the parton density
functions. In fact, the single inclusive jet cross section [64] was found initially to be higher than QCD
predictions. Further studies have shown that the excess over perturbative predictions is within the cur-
rent flexibility in our parametrization of the parton density. It is however interesting to recall the value
of studies of this kind. Since the QCD jets parton cross sections drop with a the square of the transverse
energy, a contact, 4-fermion interaction (similar, therefore, to weak interactions at low energies) would
stick out at sufficiently high ET . In particular, a 4-fermion interaction with a coupling constant G, would
give rise to corrections to the cross section due to the interference terms with the standard QCD ampli-
tude. On purely dimensional ground, such corrections would be of order G, and would thus overcome the

53
 1.6
1.2 0:0  jj < 0:5 1.6
1.2 0:0  jj < 0:5
0.8 0.8
0.4 0.4
0 0
-0.4 -0.4
1.6
1.2 0:5  jj < 1:0 1.6
1.2 0:5  jj < 1:0
0.8 0.8
(Data - Theory)/Theory

(Data - Theory)/Theory
0.4 0.4
0 0
-0.4 -0.4

 jj < 1:5  jj < 1:5

1.6 1.6
1.2 1:0 1.2 1:0
0.8 0.8
0.4 0.4
0 0
-0.4 -0.4

 jj < 2:0  jj < 2:0

1.6 1.6
1.2 1:5 1.2 1:5
0.8 0.8
0.4 0.4
0 0
-0.4 -0.4

 jj < 3:0  jj < 3:0

1.6 1.6
1.2 2:0 1.2 2:0
0.8 0.8
0.4 0.4
0 0
-0.4 -0.4
-0.8 -0.8
50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500
ET (GeV) ET (GeV)

Fig. 22: Comparison of experimental measurements versus theoretical predictions: CTEQ4HJ (•) and CTEQ4M (◦) (left
figure); MRSTg↑ (•) and MRST (◦) (right figure).

1 2.5
(data - QCD)/QCD

(data - QCD)/QCD

a) b)
2.25
0.8 2
1.75
0.6
1.5
0.4 1.25
1
0.2 0.75
0.5
0
0.25

-0.2 0
-0.25
50 100 150 200 250 300 350 400 50 100 150 200 250 300 350 400
ET (GeV) ET (GeV)

1.6
(data - QCD)/QCD

(data - QCD)/QCD

c) 2.5 d)
1.4
2.25
1.2 2
1 1.75
1.5
0.8
1.25
0.6 1
0.4 0.75
0.2 0.5
0.25
0
0
-0.2 -0.25
-0.4 -0.5
50 100 150 200 250 50 100 150 200
ET (GeV) ET (GeV)

Fig. 23: Dijet cross sections from CDF; ET distribution of one central jet, for the recoiling jet in different rapidity bins (left
plot). A comparison if the dijet cross section to theoretical predictions i showin in the rigth plot. The error bars represent the
statistical errors, while the shaded band represents the correlated systematic error.

54
strong interaction at some ET . Thus, high transverse momentum jets studies can be used to put bounds
on these kind of interactions. Sometimes, these bounds are called, somewhat improperly, compositeness
bounds, since these kinds of 4-fermion interactions would naturally arise in composite models, due to
the exchange of heavy composite particles.

7.6 Production of W , Z, and Drell-Yan pairs

From the point of view of perturbative QCD, the production of W , Z and Drell-Yan pairs are very similar
processes. Some graphs contributing at leading, next-to-leading, and next-to-next-to-leading order in the
strong coupling are shown in fig. 24. The corrections of order αS have been given a long time ago in

Fig. 24: Some graphs contributing to the Drell-Yan partonic cross section in QCD.

refs. [46, 47, 48], while the αS2 corrections have been computed in ref. [65, 66]. In order to get acquainted
with the kinematics, let us compute the parton cross section for the production of a hypothetic massive
vector meson. The amplitude is
M = g v̄(p2 ) γ µ u(p1 ) (191)
and the partonic cross section is
1 11
Z
|M|2 ,
X
σ̂ = dΦ1 (192)
2ŝ 4 9
spin,col.

where we have included a factor of 1/4 for the initial spin average, 1/9 for the initial colour average,
1/2ŝ to go from an amplitude squared to a cross section, and the one-particle phase space dΦ1 . We have

|M|2 = 3g 2 Tr[p/1 γ µ (−p/2 )γµ ] = 12g 2 ŝ ,

X
(193)
spin,col.

and
d3 q
Z
dΦ1 = (2π)4 δ 4 (p1 + p2 − q) = 2π δ((p1 + p2 )2 − MV2 ) (194)
2q 0 (2π)3
so that at the end we get
4π 2
σ̂ = α δ(ŝ − MV2 ) , (195)
3
√
with α = g 2 /(4π). For W ± production, the coupling is g = gem /( 2 sin θW ), and only left handed
quarks, and right handed antiquarks, can contribute. We get

π 2 αem
σ̂W = sin−2 θW δ(ŝ − MW
2
). (196)
3
The full hadronic cross section is then
Z h³ ´ i
(H ) (H )
σW = dx1 dx2 fu(H1 ) (x1 ) fd¯ 2 (x2 ) + fd¯ 1 (x1 ) fu(H2 ) (x2 ) cos2 θC + . . .

π 2 αem 2
× δ(s x1 x2 − MW ) (197)
3 sin2 θW

55
where one should not forget the appropriate CKM factors. A recent summary of W/Z cross section
studies at the Tevatron is given in ref. [67].
From the measured ratio
σW · B(W → eν)
R= , (198)
σZ · B(Z → ee)
assuming that the ratio of the production cross section is accurately calculable, one can extract B(W →
eν̄), and from it ΓW ,
Γ(W → eν̄)
ΓW = , (199)
B(W → eν̄)
assuming that the eν width is correctly given by the standard model.

7.7 Heavy Flavour production

The production of heavy flavour in hadronic collisions involves strong interactions directly. Furthermore,
in many cases of interest, the gluon densities play an important role. This is unlike the case of W/Z
production, in which the main production mechanism does not involve the strong coupling constant. The
search and discovery of the top quark has therefore relied on the whole machinery of perturbative QCD,
factorization, and structure function physics.
The leading order process is proportional to the square of the strong coupling constant. Next-to-
leading (order αS3 ) calculations for the production of heavy flavour production have been available for a
long time. Furthermore, a large amount of work has been performed on resummation of effects enhanced
in particular kinematic regions [68].
Since the top is very heavy, one expects that perturbative QCD should work well in this case.
In fig. 25, taken from ref. [69], I show a comparison of theoretical predictions with the CDF and D0
measurements.
CDF data for bottom production has always shown a tendency to be higher than the theoretical
predictions, as one can see from fig. 26, a problem that is being actively investigated. A large body of
data is available for charm production. Theoretical calculations are, however, not very reliable in these
cases, since the charm mass is only moderately heavy, and thus one cannot safely rely upon perturbation
theory. Some results are shown in fig. 27. A recent review of heavy flavour production is given in [68].

8 CONCLUSIONS
In these lectures I have given an overview of perturbative QCD. As we have seen, the application of
perturbation theory in strong interactions is not straightforward, unlike the case of weak interactions and
electrodynamics. Nevertheless, a consistent and testable framework for the application of perturbation
theory in strong interactions can be defined. This framework has been severely tested in e+ e− , ep, and
hadron-collision physics. It is perhaps true that, after the very extensive work performed at LEP1 and
at the SLD, our confidence in perturbative QCD has become quite solid. Testing QCD remains however
an important activity, due to the large number of applications that heavily depend upon it. The near
future in particle physics research is in hadron collider physics, where the application of QCD is more
complex. We should not forget, for example, that Higgs production at hadronic colliders is essentially
a stong-interaction phenomenon, driven by gluons. Thus, it is important to build more confidence upon
our ability to compute hadronic processes.

REFERENCES
[1] M. E. Peskin, ,. Lectures presented at the Summer School on Recent Developments in Quantum
Field Theory and Statistical Mechanics, Les Houches, France, Aug 2 - Sep 10, 1982.

56
Fig. 25: Top production cross section versus the mass, compared to CDF and D0 measurements. The dashed band correspond
to an O(αS3 ) calculation, while the solid band includes also soft gluon resummation effects to the subleading logarithmic level.

Fig. 26: Comparison of bottom cross section calculations versus CDF measurement.

[2] H. Leutwyler, ,. Lectures given at 30th Int. Universitatswochen fur Kernphysik, Schladming,
Austria, Feb 27 - Mar 8, 1991 and at Advanced Theoretical Study Inst. in Elementary Particle
Physics, Boulder, CO, Jun 2-28, 1991.

[3] G. Ecker, ,. To be published in the proceedings of 4th Hellenic School on Elementary Particle
Physics, Corfu, Greece, 2-20 Sep 1992 and Lectures given at Cargese Summer School on

57
 Fig. 27: Charm and bottom production cross sections in proton-proton collisions at fixed target energies

Quantitative Particle Physics, Cargese, 20 Jul-1 Aug 1992.

[4] O. V. Tarasov, A. A. Vladimirov and A. Y. Zharkov, , Phys. Lett. B93 (1980) 429–432.

[5] T. van Ritbergen, J. A. M. Vermaseren and S. A. Larin, , Phys. Lett. B400 (1997) 379–384,
[hep-ph/9701390].

[6] S. A. Larin, T. van Ritbergen and J. A. M. Vermaseren, , Nucl. Phys. B438 (1995) 278–306,
[hep-ph/9411260].

[7] S. G. Gorishnii, A. L. Kataev and S. A. Larin, , Phys. Lett. B212 (1988) 238–244.

[8] S. G. Gorishnii, A. L. Kataev and S. A. Larin, , Phys. Lett. B259 (1991) 144–150.

[9] L. R. Surguladze and M. A. Samuel, , Phys. Rev. Lett. 66 (1991) 560–563.

[10] E. Braaten, S. Narison and A. Pich, , Nucl. Phys. B373 (1992) 581–612.

[11] S. Bethke, , J. Phys. G26 (2000) R27, [hep-ex/0004021].

[12] T. Kinoshita, , J. Math. Phys. 3 (1962) 650–677.

[13] T. D. Lee and M. Nauenberg, , Phys. Rev. 133 (1964) B1549–B1562.

[14] G. Sterman and S. Weinberg, , Phys. Rev. Lett. 39 (1977) 1436.

[15] T. Sjostrand, , Comput. Phys. Commun. 82 (1994) 74–90.

[16] L. Lonnblad, , Comput. Phys. Commun. 71 (1992) 15–31.

[17] G. Marchesini et. al., , Comput. Phys. Commun. 67 (1992) 465–508.

[18] JADE Collaboration, W. Bartel et. al., , Z. Phys. C33 (1986) 23.

[19] R. K. Ellis, D. A. Ross and A. E. Terrano, , Nucl. Phys. B178 (1981) 421.

58
[20] K. Fabricius, I. Schmitt, G. Kramer and G. Schierholz, , Zeit. Phys. C11 (1981) 315.

[21] Z. Kunszt, P. Nason, G. Marchesini and B. R. Webber, ,. Proceedings of the 1989 LEP Physics
Workshop, Geneva, Swizterland, Feb 20, 1989, published in Lep Physics, v.1:373-453
(QCD161:L39:1989).

[22] G. Rodrigo, M. Bilenky and A. Santamaria, , Nucl. Phys. B554 (1999) 257–297,
[hep-ph/9905276].

[23] A. Brandenburg and P. Uwer, , Nucl. Phys. B515 (1998) 279–320, [hep-ph/9708350].

[24] P. Nason and C. Oleari, , Nucl. Phys. B521 (1998) 237–273, [hep-ph/9709360].

[25] L. J. Dixon and A. Signer, , Phys. Rev. D56 (1997) 4031–4038, [hep-ph/9706285].

[26] Z. Nagy and Z. Trocsanyi, , Phys. Rev. Lett. 79 (1997) 3604–3607, [hep-ph/9707309].

[27] J. M. Campbell, M. A. Cullen and E. W. N. Glover, , Eur. Phys. J. C9 (1999) 245–265,

[hep-ph/9809429].

[28] S. Weinzierl and D. A. Kosower, , Phys. Rev. D60 (1999) 054028, [hep-ph/9901277].

[29] OPAL Collaboration, G. Abbiendi et. al., , Eur. Phys. J. C20 (2001) 601–615,
[hep-ex/0101044].

[30] ALEPH Collaboration, ,. ALEPH 2001-042, CONF 2001-026, July 5, 2001, contributed paper to
this conference.

[31] S. Catani, G. Turnock, B. R. Webber and L. Trentadue, , Phys. Lett. B263 (1991) 491–497.

[32] S. Catani, Y. L. Dokshitzer, M. Olsson, G. Turnock and B. R. Webber, , Phys. Lett. B269 (1991)
432–438.

[33] S. Catani, L. Trentadue, G. Turnock and B. R. Webber, , Nucl. Phys. B407 (1993) 3–42.

[34] Y. L. Dokshitzer, G. Marchesini and B. R. Webber, , Nucl. Phys. B469 (1996) 93–142,
[hep-ph/9512336].

[35] P. A. Movilla Fernandez, S. Bethke, O. Biebel and S. Kluth, , hep-ex/0105059.

[36] DELPHI Collaboration, P. Abreu et. al., , Z. Phys. C54 (1992) 55–74.

[37] G. Altarelli and G. Parisi, , Nucl. Phys. B126 (1977) 298.

[38] G. Curci, W. Furmanski and R. Petronzio, , Nucl. Phys. B175 (1980) 27.

[39] W. Furmanski and R. Petronzio, , Phys. Lett. B97 (1980) 437.

[40] R. K. Ellis, H. Georgi, M. Machacek, H. D. Politzer and G. G. Ross, , Phys. Lett. B78 (1978) 281.

[41] W. L. van Neerven and A. Vogt, , Phys. Lett. B490 (2000) 111–118, [hep-ph/0007362].

[42] S. A. Larin, P. Nogueira, T. van Ritbergen and J. A. M. Vermaseren, , Nucl. Phys. B492 (1997)
338–378, [hep-ph/9605317].

[43] A. Retey and J. A. M. Vermaseren, , Nucl. Phys. B604 (2001) 281–311, [hep-ph/0007294].

[44] S. Moch, J. A. M. Vermaseren and M. Zhou, , hep-ph/0108033.

59
[45] M. L. Mangano, P. Nason and G. Ridolfi, , Nucl. Phys. B373 (1992) 295–345.

[46] G. Altarelli, R. K. Ellis and G. Martinelli, , Nucl. Phys. B143 (1978) 521.

[47] G. Altarelli, R. K. Ellis and G. Martinelli, , Nucl. Phys. B157 (1979) 461.

[48] J. Kubar-Andre and F. E. Paige, , Phys. Rev. D19 (1979) 221.

[49] W. W. Lindsay, D. A. Ross and C. T. Sachrajda, , Nucl. Phys. B222 (1983) 189.

[50] W. W. Lindsay, D. A. Ross and C. T. Sachrajda, , Nucl. Phys. B214 (1983) 61.

[51] W. W. Lindsay, D. A. Ross and C. T. Sachrajda, , Phys. Lett. B117 (1982) 105.

[52] J. C. Collins, D. E. Soper and G. Sterman, , Adv. Ser. Direct. High Energy Phys. 5 (1989) 1–91.

[53] J. H. Kim et. al., , Phys. Rev. Lett. 81 (1998) 3595–3598,

[http://arXiv.org/abs/hep-ex/9808015].

[54] S. A. Larin and J. A. M. Vermaseren, , Phys. Lett. B259 (1991) 345–352.

[55] CCFR Collaboration, W. G. Seligman et. al., , http://arXiv.org/abs/hep-ex/9701017.

[56] A. L. Kataev, G. Parente and A. V. Sidorov, , Nucl. Phys. B573 (2000) 405–433,
[hep-ph/9905310].

[57] J. Santiago and F. J. Yndurain, , Nucl. Phys. B563 (1999) 45–62, [hep-ph/9904344].

[58] A. D. Martin, R. G. Roberts, W. J. Stirling and R. S. Thorne, , Eur. Phys. J. C4 (1998) 463–496,
[http://arXiv.org/abs/hep-ph/9803445].

[59] R. K. Ellis and W. G. Scott, , Adv. Ser. Direct. High Energy Phys. 4 (1989) 131–175.

[60] D0 Collaboration, B. Abbott et. al., , Phys. Rev. Lett. 86 (2001) 1707–1712, [hep-ex/0011036].

[61] W. T. Giele, E. W. N. Glover and D. A. Kosower, , Phys. Rev. Lett. 73 (1994) 2019–2022,
[hep-ph/9403347].

[62] H. L. Lai et. al., , Phys. Rev. D55 (1997) 1280–1296, [hep-ph/9606399].

[63] CDF Collaboration, T. Affolder et. al., , Phys. Rev. D64 (2001) 012001, [hep-ex/0012013].

[64] CDF Collaboration, T. Affolder et. al., , Phys. Rev. D64 (2001) 032001, [hep-ph/0102074].

[65] R. Hamberg, W. L. van Neerven and T. Matsuura, , Nucl. Phys. B359 (1991) 343–405.

[66] W. L. van Neerven and E. B. Zijlstra, , Nucl. Phys. B382 (1992) 11–62.

[67] CDF and D0 Collaboration, F. Lehner, ,. To be published in the proceedings of 4th Rencontres du
Vietnam: International Conference on Physics at Extreme Energies (Particle Physics and
Astrophysics), Hanoi, Vietnam, 19-25 Jul 2000.

[68] S. Frixione, M. L. Mangano, P. Nason and G. Ridolfi, Heavy Flavours II, pp. 609 – 706. World
Scientific, 1998. hep-ph/9702287.

[69] R. Bonciani, S. Catani, M. L. Mangano and P. Nason, , Nucl. Phys. B529 (1998) 424–450,
[hep-ph/9801375].

60