Chapter 8

QCD in e+e− annihilations

Literature:

• Dissertori/Knowles/Schmelling [27]

• Ellis/Stirling/Webber [28]

• Bethke [29, 30]

• Particle Data Group [26]

• JADE, Durham, and Cambridge jet algorithms [31, 32, 33, 34]

• FastJet Package, Fast kT , SISCone [35, 36, 37]

In Chap. 7, QCD is introduced as an SU (3) gauge theory. Here we continue this discus-
sion and consider QCD processes following e+ e− annihilations. The main focus is on the
definition and application of observables linking theoretical predictions with measurable
quantities: Jets and event shapes are discussed; the applications include measurements of
the parton spins, the strong coupling constant, and the QCD color factors. The chapter
is concluded by an outlook to hadronization and non-perturbative QCD.
Some examples of e+ e− colliders and their energies are given in Tab. 6.1. Fig. 8.1(a) maps
the corresponding eras onto the available center of mass energies. A half-logarithmic plot
+ − + − − +
comparing σ e e →hadrons to σ e e →µ µ is given in Fig. 8.1(b). Experimental milestones
include:

• SPEAR (SLAC): Discovery of quark jets.

• PETRA (DESY) & PEP (SLAC): First high energy (> 10 GeV) jets; discovery of
gluon jets (at the PETRA collider, see Fig. 8.2); many pioneering QCD studies.

153
154 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

���������������������������� �����
(a)

(b)

Figure 8.1: Cross sections in e+ e− annihilations. (a) Cross section for e+ e− → hadrons as
a function of the center of mass energy. The ECM dependence is linear because the plot
is double-logarithmic. Source: [38]. (b) Comparison of cross sections for e+ e− → hadrons
and for e+ e− → µ− µ+ . Both cross sections show the same 1/s dependence on the center
of mass energy squared, except at the Z resonance.
8.1. THE BASIC PROCESS: E + E − → QQ̄ 155

(a) (b)

Figure 8.2: Gluon discovery at the PETRA collider at DESY, Hamburg. Event display (a)
and reconstruction (b).

• LEP (CERN) & SLC (SLAC): Large energies (small αs , see later) mean more re-
liable calculations and smaller hadronization uncertainties. Large data samples are
collected: ∼ 3 · 106 hadronic Z decays per experiment. This allows for precision tests
of QCD.

00
8.1 The basic process: e+e− → q q̄
In Sect. 5.10 we calculated the cross section for e+ e− → µ+ µ− and found
2
+ e− →µ+ µ− 4παem 86.9 nbGeV2
σe = = (8.1)
3s s
where the finite electron and muon masses have been neglected. Here, we consider the
basic process e+ e− → q q̄. In principle, the same Feynman diagram contributes:

e− µ− e− q

e+ µ+ e+ q̄
156 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

The only differences are the fractional electric charges of the quarks and the fact that the
quarks appear in Nc = 3 different colors which cannot be distinguished by measurement.
Therefore, the cross section is increased by a factor Nc . For the quark-antiquark case one
thus finds (for mq = 0)
2
+ e− →q q̄ 4παem 86.9 nbGeV2 2
σ0e = e2q Nc = e q Nc . (8.2)
3s s
� e+ e− →qq̄ + −
We assume qσ = σ e e →hadrons , i. e. the produced quark-antiquark pair will
always hadronize.
With Eq. (8.1) and (8.2), neglecting mass effects and gluon as well as photon radiation,
we find the following ratio:
+ −
σ e e →hadrons �
R = e+ e− →µ+ µ− = Nc e2q . (8.3)
σ q

The sum runs over all flavors that can be produced at the available energy. For ECM
below the Z peak and above the Υ resonance (see Fig. 8.3), we expect1
 
� � � � � 2 � �2 � � 2 � �2
 2 2 1 1 2 1 
R = Nc 2 
e q = Nc  + − + − + + −  = Nc 11 .
3 3 3 3 3  9
q � �� � � �� � � �� � � �� � � �� �
u d s c b

0
This is in good agreement with the data for Nc = 3 which confirms that there are three
colors. At the Z peak one also has to include coupling to the Z boson which can be created
from the e+ e− pair instead of a photon. The small remaining difference visible in the plot
is because of QCD corrections for gluon radiation (see later).

8.1.1 Singularities
In order to achieve a better prediction, we have to go beyond the basic QED prediction by
including QCD dynamics: Consider the production of a quark-antiquark pair along with
a gluon:

e− q

e+ q̄

1
Recall that the top quark mass is mt ≈ 171 GeV.
8.1. THE BASIC PROCESS: E + E − → QQ̄ 157

�����������

���������������������
��������������

����������
������������������

����������
������������������
����������
�
�����������������

������������������������������������������
+ − + − + −
Figure 8.3: Ratio R = σ e e →hadrons /σ e e →µ µ as a function of the center of mass
energy. As expected by Eq. (8.3), there is roughly no energy dependence besides various
resonances. The data confirm that there are three quark colors.

We define the kinematic variables

pi · Q Ei∗
xi = 2 = (8.4)
Q2 Ebeam
�
where Q = pe+ + pe− = pγ/Z and Q2 = s. Energy-momentum conservation ( i pi = Q)
requires that, in this case,
xq + xq̄ + xg = 2 (8.5)
xi ≤ 1. (8.6)
One can calculate the differential cross section
d2 σ αs x2q + x2q̄
= σ 0 CF (8.7)
dxq dxq̄ 2π (1 − xq̄ )(1 − xq )
where CF = 4/3 is the color factor of the fundamental representation. Note that this
expression is singular for

• xq → 1, e. g. q̄�g,
• xq̄ → 1, e. g. q�g, and for
• (xq , xq̄ ) → (1, 1), e. g. xg → 0.

Because of the kinematic constrains imposed by energy-momentum conservation (Eq. (8.5)

and (8.6)), the allowed region (part of which we have to integrate Eq. (8.7) over to find a
cross section) for a γ � → q q̄g event is of the form shown in Fig. 8.4.
158 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

q q̄
q q̄

q q̄

Figure 8.4: A Dalitz plot showing the allowed region of the xq -xq̄ plane for a γ � → q q̄g
event with massless partons. The thick lines indicate the singularities where xq = 1 and
xq̄ = 1. Their intersection marks the position of the soft gluon singularity: xg = 0. The
concept of jets will be introduced later, but it is clear that there has to be at least a
certain angle between the gluon and the quarks if the jet in gluon direction is to be
detected separately. Source: [27, p. 74].

So, how does one deal with these singularities to find a meaningful expression for the cross

0000
section to first order? Consider first the two-jet cross section. Two jets are detected when
the gluon is either very soft or almost parallel to the quarks such that only two energy
flows back-to-back can be measured. Including interference terms, the cross section in the
case of an unresolved gluon is given by (integration over two-jet region, see Fig. 8.4)

� �2 � �2
� � � �
� � � �
σtwo-jet (T ) = �� � +�
� �
�
�
� � � �
� �� � � �� �
O(α0s ) O(α1s )
 

+ 2Re  ·  +O(αs2 )

� �� �
O(α1s )
� �
= σ0 1 + αs f (T ) + O(αs2 )

where T stems from the criterion separating the two- and three-jet regions of the Dalitz
plot: max{xq , xq̄ , xg } < T. The singularities of the second and third term cancel and the
result is a function of the parameter T. However, our problem is not yet resolved, since
limT →1 f (T ) = −∞.

If the gluon can be resolved, a three-jet event is detected and the integration is over the
8.2. JETS AND OTHER OBSERVABLES 159

(a) (b)

0
Figure 8.5: Hadronization of quarks and gluons. Diagrams of the processes e+ e− → q q̄ →
hadrons (a) and e+ e− → q q̄g → hadrons (b). The RHS shows the situation in the center
of mass frame. Source: [39, p. 5 and 6].

three-jet region of Fig. 8.4:

� �2
� �
� �
σthree-jet (T ) = �� � + O(αs2 ) = σ0 αs g(T ) + O(αs2 )
�
� �

where limT →1 g(T ) = +∞. Combining the two-jet and three-jet cross sections, one finds
that the dependence on T cancels yielding a finite result for the total cross section:
� �
σtot = σtwo-jet + σthree-jet + · · · = σ0 1 + αs [f (T ) + g(T )] + O(αs2 )
� �
3 αs 2
= σ0 1 + C F + O(αs ) .
4 π

8.2 Jets and other observables

We now focus on entities actually observable in experiment. We do not observe free quarks
but only colorless hadrons produced by the “hadronization/fragmentation” of quarks and
gluons. For instance, the processes discussed so far can be visualized as in Fig. 8.5.
The anatomy of the process e+ e− → Z → hadrons is sketched in Fig. 8.6. The things that
we can do based on such a process include

• Measure αs ,

• Measure the masses of (heavy) quarks,

• Measure gluon self-coupling,

• Study hadronization and particle correlations, and

• Study the transition between the non-perturbative and perturbative regime and the
properties of quark or gluon jets.
160 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

Figure 8.6: The anatomy of the process e+ e− → γ � , Z → hadrons. Source: [40, p. 13,
modified].

8.2.1 Jet algorithms

Let us turn to the question: What is a jet?
Fig. 8.7 shows a multi-jet event recorded by the ALEPH detector. A possible verbal defi-
nition of “jet” would be “cluster/spray of particles (tracks, calorimeter deposits) or flow
of energy in a restricted angular region”. Jets are the connection between the quarks and
gluons of QCD and the signals actually measured in the detectors. If we are to extract
this information from the data, we clearly need some kind of algorithmic definition of this
concept: In the “final state” of many interesting interactions there are quarks and gluons.
These are the fundamental particles of QCD. Confinement (see p. 148) means that in the
detector we see hadrons (together with leptons and photons), but not single quarks or
gluons. At energies much larger than ΛQCD (∼ 1 GeV) these hadrons appear confined into
jets. Our aim is to compare the predictions based on partons (quarks and gluons) with
the measurements on hadrons. Therefore, we need an algorithmic definition of a jet which

1. can be applied both to data and predictions and

2. gives a close relationship between partons and jets of hadrons.

The basic requirement for such an algorithm is applicability at all relevant levels of theory
and experiment: partons, stable particles, measured objects (calorimeter objects, tracks,
etc.) while always finding the same jet. Furthermore, the algorithm has to be independent
of the very details of the detector, e. g. the granularity of the calorimeter, the energy
8.2. JETS AND OTHER OBSERVABLES 161

Figure 8.7: Multi-jet event in the ALEPH detector.

response, etc. Finally, it should also be easy to implement. In order that we can test
QCD predictions, there has to be a close correspondence between the jet momentum (i. e.
energy, momentum, and angle) at the parton level and at the hadron level.

NB: Other requirements might strongly depend on the specific applica-

tion/measurement being performed: For a precision test of QCD there may be
requirements which for an analysis of W decays or searches for new physics might not be
necessary (e. g. infrared safety).
Further requirements come from QCD: We want to compare perturbative calculations
with the data. Therefore, the algorithm has to be insensitive to “soft physics” which
requires infrared safety and collinear safety.
Infrared safety requires that the configuration must not change when adding a further
soft particle. This would be violated by the following behavior2 :

Collinear safety means that the configuration does not change when substituting one
particle with two collinear particles. The problem is visualized in this figure:
2
Source: [41, pp. 4].
162 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

Infrared and collinear safety yield algorithms with the required insensitivity to soft physics:
They guarantee the cancellation (between real and virtual emission diagrams) of the
infrared and collinear divergencies in every order of perturbation theory.

[Link] Examples of jet algorithms

There are two classes of jet algorithms in use. Algorithms of the class “JADE” are used
mainly for e+ e− annihilations (i. e. for the analysis of events with purely leptonic initial
states), but more recently, this class of jet algorithms is also used at hadron colliders. We
will concentrate on this class here. The second class of jet algorithms is called “CONE”
and is mainly used at hadron colliders with some applications also at e+ e− colliders.

JADE class algorithms are characterized by

• a “metric” yij (measure of distance in momentum space),

• a criterion of resolution ycut , and a

• procedure of recombination.

The original definition of the metric from the JADE experiment at PETRA reads

2Ei Ej (1 − cos θij ) m2ij

yij = 2
≈ 2 (8.8)
ECM ECM

where mij is the invariant mass of the particle pair (i, j), see Fig. 8.8(a) Given this metric
and a pre-defined resolution ycut , the algorithm is:
8.2. JETS AND OTHER OBSERVABLES 163

(a) (b)

Figure 8.8: Particle pair (a) and recombination of close particles (b).

����������������

�������������

����������
��� ��

������� ����
������� ����������
��������������

First, all distances yij between pairs (i, j) are calculated. Then we search for the smallest
invariant mass: min(i,j) yij = ykl . The fact that ykl is the smallest distance in momentum
space of all pairs (of particles or, in the subsequent steps, also pseudo-particles) means
that the pair (k, l) is either nearly parallel, θkl = 0, or one or both of the particles are
very soft, see Fig. 8.8(b). If the distance cannot be resolved, ykl < ycut , the two particles
(k, l) are combined (clustered) into one new pseudo-particle with the combined momentum
p(kl) = pk +pl (i. e. momentum is conserved), see Fig. 8.9(a). This is the so-called E scheme.
Applying this algorithm will reduce complex events until there is a certain number of jets
left, as is sketched in Fig. 8.9(b).
The proposed algorithm has some very useful characteristics:

• Infrared safety,

yij → 0 for Ei or Ej → 0,

and collinear safety,

yij → 0 for θij → 0,

164 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

���
���
(a) (b)

Figure 8.9: Recombination of particle pair with small invariant mass (a) and reduction of
particle pattern to jets (b).

(in every order of perturbation theory, see p. 161 and Eq. (8.8)).
• All particles are assigned to one and only one jet.
• The algorithm’s sequence does not depend on ycut .
• The number of found jets is a monotonic function of ycut .

For the discussed algorithm there is no need to stick to the JADE metric of Eq (8.8);
alternative metrics can be introduced. For instance, the DURHAM metric is
� �
2 min Ei2 , Ej2 (1 − cos θij ) k⊥2
yij = 2
≈ 2
(8.9)
ECM ECM
where k⊥ is the transverse momentum of the less energetic particle with respect to the
more energetic one. The introduction of this metric was motivated by perturbative QCD
calculations: It allows for the resummation of large logarithms of the type lnm (ycut ) in all
orders of perturbation theory (see e. g. [27, pp. 139]). These logarithms appear order-by-
order in the expressions for jet cross sections, jet rates, etc.
Now is a good time to recall the Dalitz plot of Fig. 8.4 where we separated a two-jet and a
three-jet region. The algorithmic jet definition we have developed enables us to define the
thee-jet region: Apply the jet algorithm until three jets are left. If the distance between
the jets can be resolved, min(i,j) (yij ) > ycut , there are three jets, else it is a two-jet event.
The shape of the found three-jet region is somewhat different, since yij also depends on
the angle θij , see Fig. 8.10.
In order to compare the analyzed data to the predictions of QCD, we need perturbative
predictions for jet rates. For the reaction e+ e− → hadrons the leading order predictions
are as follows. For the JADE algorithm we have
� �
LO αs 2 π2 5
σthree-jet (ycut ) = σ0 CF 2 ln ycut + 3 ln ycut − + − f (ycut ) (8.10)
2π 3 2
8.2. JETS AND OTHER OBSERVABLES 165

Figure 8.10: A Dalitz plot showing the allowed region of the xq -xq̄ plane for a γ � → q q̄g
event with massless partons. The three-jet region is determined using an algorithmic jet
definition.

where f (ycut ) → 0 for ycut → 0. The prediction for the DURHAM algorithm is the same,
except for the factor “2” in front of “ln2 ycut ”. In simple terms, the logarithm terms arise
because the vertex where the gluon is radiated off contributes � a factor proportional to
αs /Egluon to the integrand which upon integration yields ycut dE/E.
Resummation3 with the DURHAM algorithm looks as follows. First, let
σtwo-jet
R2 (ycut ) = .
σtot
One can show that
 
 �s 2 2
� �
dq CF αs (q ) s 3
R2 = exp − ln 2 −
 q 2 2π q 2 
sycut
�s
dq 2 CF αs (q 2 ) CF α s 2
≈1− 2
ln · · · + · · · ≈ 1 − ln ycut + . . .
q 2π 2π
sycut

where R2 (ycut → 0) = 0. This is an example of the characteristics an algorithm has to

have if you want to perform “high-precision” perturbative QCD calculations. Now there
also exists an algorithm of the kt (DURHAM) type for hadron colliders, see later.
To conclude this section, we turn to the comparison of jet algorithms. There is no such
thing as the best “benchmark” variable which allows to compare algorithms in a general
manner. The suitability and performance of an algorithm depends very strongly on the
performed analysis. Usually we would like to have a good resolution of energies and angles
3
Resummation in QCD is analogous to the treatment of infrared divergencies in QED, see e. g. [14,
pp. 202]
166 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

(a) (b)

Figure 8.11: Visualization of levels at which the algorithms have to deliver good resolution
(a) and comparison of jet algorithms (b). The mean number of jets is displayed as a
function of ycut . The parton level is denoted by squares and the hadron level by circles.
The results were obtained by HERWIG Monte Carlo simulation at ECM = Mz . Source:
[34, p. 28]. For details compare [34, pp. 7].

of the jets at the parton, hadron, and detector levels (see Fig. 8.11(a) for a visualization),
as well as a good efficiency and purity to find a certain number of jets at a certain
level. For some jet algorithms, the mean number of jets as a function of ycut at the
hadron and parton levels, as obtained by HERWIG (Hadron Emission Reactions With
Interfering Gluons) Monte Carlo simulation at ECM = MZ , is compared in Fig. 8.11(b).
Another comparison4 is shown in Fig. 8.12. The fraction of events with 2 jets which have
2, 3, 4, and 5 sub-jets is given as a function of ycut or r2 , the radius fraction sqared,
respectively. The data stem from HERWIG Monte Carlo simulations at ECM = 1.8 TeV
with 75 GeV < Et (jet 2) < 100 GeV. Data from a kt algorithm are shown in Fig. 8.12(a)
while the results in Fig. 8.12(b) come from a CONE algorithm with radius R = 0.7.

8.2.2 Event shape variables

The introduced jet algorithms can be used as a starting point to define more refined
observables that capture the event topologies.
4
More on kt and CONE algorithms can be found in [41].
8.2. JETS AND OTHER OBSERVABLES 167

(a) (b)

Figure 8.12: Comparison of kt (a) and CONE (b) algorithms. Legend: —parton level,
· · · calorimeter level. The fraction of two-jet events with 2, 3, 4, and 5 sub-jets is given as
a function of ycut or r2 . The data is generated by HERWIG Monte Carlo simulations at
ECM = 1.8 TeV with 75 GeV < Et (jet 2) < 100 GeV.

An example for an event shape variable is the differential two-jet rate. The definition goes
as follows: Apply the DURHAM algorithm until exactly three jets are left (in contrast
to the possibility to run the algorithm until a certain resolution is reached). Then take
the minimal distance yij of all pairs (i, j) and call it y23 (or y3 ): min(i,j) yij = y23 = y3 .
This gives one value for each event. The distribution of these values for all events is an
“event-shape distribution”. Therefore, one can plot the differential cross section as in
Fig. 8.13. There is one histogram entry for each event. The data come from hadronic Z
decays at LEP. Observe that two-jet events are more likely than three-jet events. The
perturbative regime is limited to high gluon energies. Hadronization effects that have to
be phenomenologically modeled spoil the perturbative calculations at low y3 values.
As another example for an event-shape variable, let us consider thrust. It was invented
around 1978 and first used at PETRA. The idea is to select the axis that maximizes the
sum of the longitudinal momentum components:

The thrust of an event is then defined as

� #» #»
i | pi · n|
�
T = max #»
i | p i|
#»
n

where | #»
n | = 1 and the sum runs over the three-momenta of all final states. The thrust
axis is defined by the vector #»
n T for which the maximum is obtained. This definition
means that for T = 1 the event is perfectly back-to-back while for T = 1/2 the event is
spherically symmetric:
168 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

Figure 8.13: Differential two-jet rate for hadronic Z decays at LEP.

This point is also illustrated with ALEPH data of Z decays in Fig. 8.14 where Fig. 8.14(a)
corresponds to T → 1 and Fig. 8.14(b) to T → 1/2. The corresponding event-shape
distribution is shown in Fig.8.15 (compare also the differential two-jet rate event-shape
distribution in Fig. 8.13).
There are further event-shape variables suitable for different purposes. Some examples are
given in the following.

• Thrust major Tmajor : The thrust major vector #» n Ma is defined in the same way as
the thrust vector #»
n T , but with the additional condition that #» n M a must lie in the
#»
plane perpendicular to n T :
� #» #»
| p i · n Ma |
i�
Tmajor = #»max#» #» .
i | p i|
n Ma ⊥ n T

• Thrust minor Tminor : The minor axis is perpendicular to both the thrust axis and
8.2. JETS AND OTHER OBSERVABLES 169

(a) (b)

Figure 8.14: Event displays of Z decays recorded at ALEPH. The thrust is nearly 1 for (a)
and close to 1/2 for (b).

Figure 8.15: Thrust for hadronic Z decays at LEP. Observe that the two- and three-jet
events are indicated by thrust values close to 1 and 1/2, respectively. Again, in the non-
perturbative regime hadronization corrections from phenomenological models are needed.
170 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

the major axis: #»

n Mi = #»
n T × #»
n M a . The value of the thrust minor is given by
� #» #»
| p i · n Mi |
Tminor = i� #» .
|p |
i i

• Oblateness O: The oblateness is defined as the difference between thrust major and
thrust minor:
O = Tmajor − Tminor .

• Sphericity S: The sphericity is calculated from the ordered eigenvalues λi=1,2,3 of

the quadratic momentum tensor:
� α β
αβ pi pi
M = � i #» 2
, α, β = 1, 2, 3
i | p i|
λ1 ≥ λ2 ≥ λ3 , λ1 + λ2 + λ3 = 1
3
S = (λ2 + λ3 ).
2
The sphericity axis #»
n S is defined along the direction of the eigenvector associated
to λ , the semi-major axis #»
1 n is along the eigenvector associated to λ .
sMa 2

• Aplanarity A: The aplanarity is calculated from the third eigenvalue of the quadratic
momentum tensor:
3
A = λ3 .
2
• Planarity P : The planarity is a linear combination of the second and third eigenvalue
of the quadratic momentum tensor:
P = λ 2 − λ3 .

• Heavy jet mass ρ: A plane through the origin and perpendicular to #»

n T divides the
event into two hemispheres, H1 and H2 from which the corresponding normalized
hemisphere invariant masses are obtained:
� �2
1 �
Mi2 = 2 pk , i = 1, 2.
ECM k∈H
i

The larger of the two hemisphere masses is called the heavy jet mass,
ρ = max(M12 , M22 ),

and the smaller is the light jet mass ML ,

ML = min(M12 , M22 ).
8.2. JETS AND OTHER OBSERVABLES 171

• Jet mass difference MD : The difference between ρ and ML is called the jet mass
difference:

MD = ρ − ML .

• Wide jet broadening BW : A measure of the broadening of particles in transverse

momentum with respect to the thrust axis can be calculated for each hemisphere
Hi using the relation
�
| #»
k∈Hi p k × #»
nT |
Bi = � #» , i = 1, 2
2 j | p j|

where j runs over all particles in the event. The wide jet broadening is the larger of
the two hemisphere broadenings,

BW = max(B1 , B2 ),

and the smaller is called the narrow jet broadening BN ,

BN = min(B1 , B2 ).

• Total jet broadening BT : The total jet broadening is the sum of the wide and the
narrow jet broadenings:

B T = B W + BN .

• C-parameter C: The C-parameter is derived from the eigenvalues of the linearized

momentum tensor Θαβ :

1 � pα pβ
i i
Θαβ = � #» #» , α, β = 1, 2, 3.
i | p i | i
| p i |

The eigenvalues λj of this tensor define C by

C = 3(λ1 λ2 + λ2 λ3 + λ3 λ1 ).

The discussed event-shape variables have been extensively used to analyze LEP data.
Examples are given in Fig. 8.16: Fig. 8.16(a) shows thrust predictions and measurements;
predictions and data for thrust, heavy jet mass, total jet broadening, wide jet broadening,
and the C-parameter are shown in Fig. 8.16(b).
172 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

(a) (b)

Figure 8.16: Comparison of predictions and LEP data for some event-shape variables.
Thrust data are shown for several center of mass energies (a). The other analyses deal
with heavy jet mass, total jet broadening, wide jet broadening, and the C-parameter (b).

8.2.3 Applications
Examples for applications of the observables discussed above in this section are measure-
ments of the strong coupling constant αs (see later, Sect. 8.3), the discovery of quark and
gluon jets, measurements of the quark and gluon spin, the triple-gluon vertex, and jet
rates or the analysis of differences between quark and gluon jets.
Quark jets were discovered at the SPEAR storage ring (SLAC) [42]. The data are shown
in Fig. 8.17. For higher energies particles cluster around an axis and the Monte Carlo
simulation based on a jet model fits the data better than the simulation based on an
isotropic phase-space model. This is the first observation of a jet structure.
Gluon jets were discovered at PETRA (DESY) [43, 44, 45, 46]. Here, the relevant ob-
servable is oblateness (see p. 170). The first three-jet event seen by TASSO is shown in
Fig. 8.18(a). In Fig. 8.18(b) one can observe that events at ECM ∼ 30 GeV exhibit larger
oblateness (planar structure) than predicted by models without hard gluon radiation.
When it comes to parton spins the question is: How do you measure the spin of unob-
servable particles? For spin-1/2 fermions annihilating into a vector boson, conservation of
angular momentum predicts a distribution
dσ
∗
∼ 1 + cos2 Θ∗
d cos Θ
8.2. JETS AND OTHER OBSERVABLES 173

Figure 8.17: Discovery of quark jets at SPEAR (SLAC). Observed sphericity (see p. 170)
distributions for data, jet model (solid curves) and phase-space model (dashed curves) for
ECM = 3 GeV (LHS) and 7.4 GeV (RHS). Source: [42, 38, p. 1611].

(a) (b)

Figure 8.18: The first three-jet event seen by TASSO (a) and the distribution N −1 dN/dO
as a function of oblateness, measured at MARK-J (b). In both figures of (b) the solid
curves are the predictions based on the q q̄g model and the dashed curve is based on the
standard q q̄ model. Source: [44, p. 832].
174 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

(a) (b)

Figure 8.19: Measurements of quark (a) and gluon (b) spin by ALEPH. Source: [47].

if the final state particles have spin 1/2 and

dσ
∼ 1 − cos2 Θ∗ = sin2 Θ∗
d cos Θ∗
for spin-0 particles in the final state. Therefore, the quark direction has to be measured
to measure the quark spin. At LEP energies the thrust axis in two-jet events to a very
good approximation aligns with the direction of the primary quarks. Thus, one can take
the thrust direction in two-jet events. The exact expression for the spin-1/2 case reads
2
dσ αem e2q πNc � �
∗
= 2 − βq∗2 + βq∗2 cos2 Θ∗ βq∗
d cos Θ 2s
�
where βq∗ = 1 − 4m2q /s → 1 for mq = 0. The resulting angular distribution found by
ALEPH [47] is shown in Fig. 8.19(a).The experimental data are compared to a Monte
Carlo simulation. The data are in perfect agreement with the spin-1/2 assignment for the
quarks while a spin-0 assignment is clearly excluded. The sharp drop in the distribution
around cos Θ∗ ∼ 0.8 is due to the finite detector acceptance.
Let us turn to the gluon spin. Hard gluon radiation leads to three-jet events. So, after
applying a jet algorithm to select the three-jet events, how do we know which one is the
gluon jet? Recall that the probability to radiate off a soft gluon is larger than to radiate
off a hard gluon. Therefore, for three jets
8.2. JETS AND OTHER OBSERVABLES 175

with energies
sin θi
Ei = ECM � ,
j sin θj

if ordered by energy, E1 > E2 > E3 , jet 3 is the gluon jet in 75% of the events. Defining
the variable
1
Z = √ (x2 − x3 )
3
(recall xi = 2Ei /ECM ), the Dalitz plot looks like in Fig. 8.20. The arrow length is pro-
portional to the jet energy. The following cases have to be compared: In the spin-1 case
(“vector gluon”) the prediction reads
� �
d2 σ v x21 + x22
∝ + permutations (1, 2, 3)
dx1 dx2 (1 − x1 )(1 − x2 )

while for spin-0 (“scalar gluon”)

� � 2 �
d2 σ s x23 aq
∝ + permutations (1, 2, 3) − 10 � 2
dx1 dx2 (1 − x1 )(1 − x2 ) aq + vq2

where aq and vq are the axial-vector and vector couplings of the quarks to the intermediate
photon or Z boson and the sums run over all contributing quark flavors. For e+ e− annihi-
lation via a photon only the vector coupling contributes, on the Z resonance both terms

00
have to be taken into account. The ALEPH data shown in Fig. 8.19(b) clearly indicate
that gluons have spin 1.
As we have seen before (see p. 145), the kinematic term of the QCD (SU (3), non-abelian,
gluon) Lagrangian contains a three-gluon term yielding a three-gluon vertex, a feature
not present in QED (U (1), abelian, photon). The splitting of a radiated gluon into two
gluons will lead to a four-jet event, just like the splitting into a quark-antiquark pair:

3 3

4 4
176 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

Figure 8.20: Phase space as function of x1 and Z for energy-ordered jet configurations,
x1 > x2 > x3 . The arrow length is proportional to the energy. Source: [47].

For gluon radiation off quarks one finds that the gluon is preferentially polarized in the
plane of the splitting process. On the other hand, for a gluon splitting into two gluons
there is a positive correlation between the plane spanned by the two new gluons and the
polarization of the branching one. Finally, in case a gluon splits into two quarks, the plane
defined by the momenta of the two quarks is anticorrelated with the polarization of the
splitting gluon. So we conclude that for four-jet events induced by a gluon splitting into
a q q̄ pair, the distribution of the angle between the plane defined by the two primary
quarks and the plane defined by the two secondary quarks should be enhanced around
90◦ (see Fig. 8.21). However, in a non-abelian theory we have contributions also from
the triple-gluon interaction, and in this case the favored angle between the two planes
spanned by the primary and secondary partons is rather small. Therefore, the shape of
the distribution of this angle is sensitive to the color factors (see Sect. 7.4.1). Like in
the three-jet case, it is difficult to distinguish between jets induced by the primary and
the secondary partons. However, because of the 1/E characteristic of radiated gluons we
expect the two secondary particles to be less energetic than the two primary quarks: If the
jets are ordered by energy, E1 > E2 > E3 > E4 , jets 3 and 4 are more likely to come from
the radiated particles. So we arrive at the definition of the angular correlation variable
called Bengtsson-Zerwas angle
( #»
p 1 × #»
p 2 ) · ( #»
p 3 × #»
p 4)
χBZ = ∠ [( #»
p 1 × #»
p 2 ), ( #»
p 3 × #»
p 4 )] = #» #» #» #»
1|( p × p )||( p × p )|
2 3 4

where #»p i , i = 1, . . . , 4 are the energy-ordered momenta of the four partons (jets). In
Fig. 8.21 LEP measurements of χBZ are compared with the predictions by QCD on the
one hand and an abelian model with three quark colors but no three-gluon coupling on
the other. The data agree with QCD being an SU (3) gauge theory rather than an abelian
gauge theory.
At the end of our discussion of jet algorithms Fig. 8.12(a) is shown. It displays the fraction
8.2. JETS AND OTHER OBSERVABLES 177

Figure 8.21: Distribution of χBZ measured by L3. The predictions for QCD and the abelian
model are shown as bands indicating the theoretical uncertainties. Source: [48, p. 233].

of 2-jet events with 2, 3, 4, and 5 sub-jets as a function of ycut . These predictions can
be tested comparing measurements at highest LEP energies to Monte Carlo simulations
which incorporate leading-order matrix elements for two-jet and three-jet production, plus
approximations for multiple soft or collinear gluon radiation. Fig. 8.22 shows the n-jet rate
according to the DURHAM (kt ) algorithm as a function of ycut .

We conclude this section with a discussion of the differences between quark and gluon jets.
Quark and gluon jets have different coupling strengths to emit gluons (see Sect. 7.4.1 and
Fig. 8.23). Therefore, from couplings alone one expects a larger multiplicity in gluon jets
of the order CA /CF = 9/4, and a softening of the momentum distributions for particles
coming from the gluon jet. Thus gluon jets are more “soft” and “fat” than quark jets
(see Fig. 8.24). Also the scaling violations, i. e. change of multiplicities with energy and
momentum scale are different. In Fig. 8.24(d) the CONE algorithm is applied to data
of OPAL (LEP) and compared to CDF data. The variable r denotes the radius of the
considered cone fraction when R is the radius parameter of the cone algorithm:
178 S= CHAPTER
(λ + λ ). 8. QCD IN E + E − ANNIHILATIONS
2

1 ALEPH Ecm = 206 GeV

PYTHIA6.1

n-jet fraction
HERWIG6.1
0.8 1-jet
ARIADNE4.1

2-jet

0.6 3-jet

4-jet

0.4 5-jet

+6-jet

0.2

-5 -4 -3 -2 -1 0
log10(ycut)

Figure 8.22: ALEPH measurements of the n-jet rate (DURHAM) as a function of ycut .

Figure 8.23: Comparison of quark and gluon jets. For a discussion of the difference in
coupling strength see Sect. 7.4.1.
8.3. MEASUREMENTS OF THE STRONG COUPLING CONSTANT 179

Ψ(r) denotes the energy in a fraction of the cone and Φ(r) is defined by Φ(r) = d/drΨ(r).

8.3 Measurements of the strong coupling constant

The QCD Lagrangian is introduced in Sect. 7.4:

b a b

L QCD
a

� ab
b a

gs f abc
c c d
gs2f abef cde

j
i j i

a
flavours �ij gsTija
Except for the quark masses, there is only one free parameter in it: the strong coupling
constant αs which is discussed in Sect. 7.4.2. Recall that the differential equation for the
strong coupling constant depending on the renormalization scale µ, αs (µ2 ), is

∂αs2 (µ2 )
µ2 = β(αs (µ2 ))
∂µ2
180 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

(a) (b)

(c) (d)

Figure 8.24: Comparison of quark and gluon jets. Note that gluon jets are more “soft”
and “fat” than quark jets. The variable xE is the energy fraction of the particles with
respect to the jet energy (c). The variable r in (d) denotes the considered fraction of the
cone.
8.3. MEASUREMENTS OF THE STRONG COUPLING CONSTANT 181

which, retaining only the first term of the power expansion for β and absorbing the factor
of 4π into the coefficient β0 , yields

gs2 (Q2 ) 1
αs (Q2 ) ≡ = .
4π β0 ln(Q /Λ2QCD )
2

At that point we also stressed that

� �
1 2
β0 = 11 − nf > 0 for (the likely case of) nf < 17
4π 3
which makes the effective coupling constant behave like shown in Fig. 7.4.2. The following
expansion holds for αs (µ2 ) (see Eq. (7.43)):
� �
2 2 2 µ2 2 2 2 2 µ
2
3
αs (µ ) ≈ αs (Q ) 1 − αs (Q )β0 ln 2 + αs (Q )β0 ln 2 + O(αs ) . (8.11)
Q Q

To measure the coupling strength one uses as many methods as possible in order to
demonstrate that QCD really is the correct theory of strong interactions by showing that
one universal coupling constant describes all strong interactions phenomena. Consider the
perturbative expansion of the cross section for some QCD process:
� �
pert µ2
σ = αs (µ )A + αs (µ ) B + β0 A ln 2 + O(αs3 )
2 2 2
(8.12)
Q
where the coefficients A and B depend on the specific process. So, if only the leading oder
(LO) expansion is known, the following holds:

pert µ2
σLO = αs (µ2 )A = αs (Q2 )A − αs2 (Q2 )β0 A ln + O(αs2 )
Q2
where in the second step we inserted the expansion from Eq. (8.11). This means that the
result depends strongly on the choice of the renormalization scale µ. Since the corrections
to the cross section can be relatively large, it is possible to find significantly different
values for the measured effective coupling constant αsmeas,eff for two different processes:
Consider two processes, where the LO calculations predict
pert
σLO;1 = αs A1
pert
σLO;2 = αs A2 .

The predictions are compared to the cross sections σ1exp and σ2exp from experiment. Finally,
meas,eff meas,eff
because of the said strong scale dependence, the result may be αs;1 �= αs;1 .
To solve the problem of the correction depending on the renormalization scale being
too large, one has to take the calculation to next-to-leading order (NLO) to reduce the
scale dependence of the prediction. For our example reaction e+ e− → q q̄g this means
 0000
182 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

e− q e− q e− q e− q

g
g g g
g

e+ q̄ e+ q̄ e+ q̄ e+ q̄

Figure 8.25: Feynman diagrams for e+ e− → q q̄g to NLO.

considering the diagrams shown in Fig. 8.25. The NLO expression is again obtained from
the expansion in Eq. (8.12):
� �
pert µ2
σNLO = αs (µ )A + αs (µ ) B + β0 A ln 2 + O(αs3 )
2 2 2
Q
µ2
= αs (Q2 )A + αs2 (Q2 )B + αs3 (Q2 )β02 A2 ln2 2 + O(αs4 )
Q

where in the second line we inserted for αs (µ2 ) the expansion from Eq. (8.11) and the
dependence on ln(µ2 /Q2 ) cancels. Thus, the scale dependence of the prediction is much
smaller than in the LO case. The scale dependence cancels completely at fully calculated
order.
By comparing the NLO prediction for the cross section to experiment, one can extract
αs (Q2 ), e. g. αs (MZ2 ). This information can in turn be used to predict other process cross
sections at NLO. Furthermore, by varying the scale µ2 one can estimate the size of the
NNLO contributions.
This procedure extends analogously to NNLO. Diagrams that have to be included at
NNLO are shown in Fig. 8.26. The prediction reads
� 2
�
pert 2 2 2 3 2 4 3 µ
σNNLO = αs (Q )A + αs (Q )B + αs (Q )C + O αs , ln 2
Q

where the scale dependence is reduced even further. NNLO is the lowest order at which
scale variations at NLO can be tested.
As an example for the scale dependence of the extracted strong coupling constant, see
Fig. 8.27 where αs (MZ2 ) from jet rates at LEP is shown as a function of ln(µ2 /Q2 ). Note
that the scale dependence is reduced by the extension to NLO, as mentioned before.
The theoretical error is taken to be the range of values covered by the projection of
the bands over −1 < ln(µ2 /Q2 ) < 1 on the abscissa. The right figure shows how the
central values and errors obtained this way for three different shape variables converge
with improvements in the theory.
There has been an enormous progress in the measurements of the strong coupling during
the last 20 years. This is due to major improvements on the theoretical and also the
 000
8.3. MEASUREMENTS OF THE STRONG COUPLING CONSTANT 183

e− q e− q e− q
g
g
g g
g
g

e+ q̄ e+ q̄ e+ q̄

Figure 8.26: Feynman diagrams for e+ e− → q q̄g at NNLO.

constructed such that it is reasonable to interpret them like conventional 68%

confidence level intervals.

0.22 0.3
2 – jet rate ALEPH
0.2
(�s) 0.25 �h
Thrust
0.18
�s (M 2z)

R2
�s 0.2
0.16
(�2s )
0.14 0.15

0.12
(�2s +NLLA)
0.1
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2 (�s) (�2s ) (�2s +NLLA)
In (�2 /Q2)
Fig. 8.1. Estimate of theoretical uncertainties for a measurement of the strong
Figure 8.27: Estimate of theoretical uncertainties for a measurement of the strong coupling
constant from event shape variables. NLLA refers to resummation of logarithms. Source:
[27, p. 307].
184 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

���
������ ���������

� �����
�������������������������
��� ������������������
���������������

���

���

���
��� ��������������������������
� �

� �� ���
�������

Figure 8.28: Summary of measurements of αs as a function of the respective energy scale

Q. The curves are QCD predictions. Source: [30, p. 12].

experimental side. A summary of measurements of αs as a function of the respective

energy scale Q is shown in Fig. 8.28.
In general, observables can be classified according to the influence the structure of the
final state has on their value.
Inclusive observables do not look at the structure of the final state. Examples are total
cross sections and ratios of cross sections (see e. g. Eq. (8.3)). Advantages of inclusive
observables are that they do not (or only weakly) depend on non-perturbative correc-
tions (hadronization) and that the perturbative series is now known to NNNLO. The
disadvantage lies in the low sensitivity in some cases.
Non-inclusive (exclusive) observables, on the other hand, look at some structure in the
final state depending on the momenta of the final state particles. Examples are jet rates
and event shape distributions. Advantages of non-inclusive observables are high sensitivity
and that the perturbative series is now known to NNLO (and resummation, see later).
Disadvantages are that in some cases even the NNNLO corrections might be relevant and
that hadronization (non-perturbative) corrections are needed.
As an example for the usage of inclusive observables, consider the determination of αs
from inclusive Z or τ decays. In general, the prediction of the cross section ratio R reads

σ Z,τ → hadrons
R = Z,τ → leptons = REW (1 + δQCD + δmass + δnp )
σ
8.3. MEASUREMENTS OF THE STRONG COUPLING CONSTANT 185

where the overall factor REW depends on the electroweak couplings of the quarks.5 The
corrections are dominated by the perturbative QCD correction δQCD . The other terms
take into account the finite quark masses and the non-perturbative corrections. The per-

0
turbative QCD correction term is given by
αs � α �2 � α �3
s s
δQCD = c 1 + c2 + c3 + ....
π π π
Diagrammatically speaking, the factor REW arises from
q

q̄
while the perturbative QCD corrections come from diagrams like in Fig. 8.25 and 8.26.
For the case of
σ Z→ hadrons
RZ =
σ Z→ leptons
the prediction reads REW = 19.934, c1 = 1.045, c2 = 0.94, and c3 = −15. The correspond-
ing measurement is visualized in Fig. 8.29: Divide the number of hadronic decays by the
number of leptonic decays to find RZ = 20.767 ± 0.025. From this ratio the following value
of the strong coupling at the Z resonance can be extracted:

αs (MZ ) = 0.1226 ± �0.0038

�� � ± 0.0002
� �� � ± �0.0002
�� �
exp., mostly statistical Mt : ±5 GeV renormalization shemes

+0.0058
= 0.1226 .
−0.0038

Finally, we state a new result from 2009, obtained using NNNLO predictions:

+0.0028
αs (MZ ) = 0.1193 ± 0.0005.
−0.0027

We now turn to non-inclusive observables such as event-shapes and jet rates. We have
already seen perturbative predictions for some examples of non-inclusive quantities in
Sect.
� 8.2. There it is stated that the log terms in the predictions are because of the
dE/E integration arising from

dσ q→qg αs 1
∝ σ0
dEgluon 2π Egluon
5
REW is a modified version of the ratio R = Nc 11/9 of Sect. 8.1.
186 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

Figure 8.29: Visualization of RZ measurement.

where σ0 is the Born cross section for Z → q q̄ (see Sect. 8.1). Recall that the perturbative
prediction is given by:
� �
1 dσ µ2
= αs (µ )A(x) + αs (µ ) B(x) + β0 A(x) ln 2 + O(αs3 )
2 2 2
σ0 dx Q

where the coefficients A and B are calculable for the class of observables x which are
infrared and collinear safe, i. e. infrared singularities from real and virtual radiative cor-
rections cancel (thrust, jet rates, C-parameter, etc.). To recall the important example of
thrust, see Fig. 8.15.
Let us take a look at the results obtained by NLO fits. First measurements gave indications
that the missing higher order terms are large: The coupling constant should be the same
for all variables, but the results vary too much (see Fig. 8.30) which indicates that the
expansion to NLO does not suffice. Typical results obtained by NLO fits are

αs (MZ ) = 0.120 ± 0.010.

As we� have seen before, to obtain perturbative corrections, we have to do integrals of the
s LO
type ycut dEgluon /Egluon which gives rise to the logarithm terms in σthree-jet (see Eq. (8.10)):

LO αs � 2 �
σthree-jet = σ 0 CF ln ycut + . . .
2π
8.3. MEASUREMENTS OF THE STRONG COUPLING CONSTANT 187

�������

Figure 8.30: NLO results for αs (MZ ). Source: [40, p. 29, modified].
188 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

where the color factor CF = 4/3—the problem being that for ycut → 0 the series does not
converge.6 The resummation procedure mentioned earlier (see p. 165) also works for the
three-jet rate:

CF α s 2 CF2 αs2 4
R3 = ln ycut − ln ycut + . . .
2π  8π 2 
 �s 2 2
� �
dq CF αs (q ) s 3
= 1 − exp − ln 2 − .
 q 2 2π q 2 
sycut

Combined (to avoid double counting of logarithmic terms in resummed expressions and
in full fixed order prediction) with full NLO calculations this gives theoretically much
improved predictions. Typical results are:

αs (MZ ) = 0.120 ± 0.005.

There are different sources of the remaining uncertainties. Experimental uncertainties

include

• track reconstruction,

• event selection,

• detector corrections (via cut variations or different Monte Carlo generators),

• background subtraction (LEP2), and

• ISR corrections (LEP2).

They amount to about 1% uncertainty. Furthermore, there are hadronization uncertain-

ties arising from the differences in behavior of various models for hadronization such as
PYTHIA (string fragmentation), HERWIG (cluster fragmentation), or ARIADNE (dipole
model and string fragmentation). Theses uncertainties are typically about 0.7 to 1.5%.
Finally, there are also theoretical uncertainties, for instance

• renormalization scale variation,

• matching of NLO with resummed calculation, and

• quark mass effects.

6
Recall that ycut is the resolution parameter deciding if two particles are distinguished or seen as one
pseudo-particle.
8.3. MEASUREMENTS OF THE STRONG COUPLING CONSTANT 189

(a) (b)

Figure 8.31: NNLO fit to ALEPH thrust data (a) and visualization of improvement in
NNLO over NLO (b). Source: [50, p. 11 and 17].

The corresponding uncertainty is typically 3.5 to 5 %.

As we have seen, the perturbative predictions have to be to sufficiently high order if we
are to accurately determine the strong coupling constant: Now a NNLO prediction is
available. Bearing in mind the foregoing, it has to be of the form

1 dσ
(y, Q, µ) = αs (µ)A(y) + αs2 (µ)B(y, xµ ) + αs3 (µ)C(y, xµ ) + O(αs4 )
σ0 dy

where y denotes an event shape variable and xµ = µ/Q. At this level of precision, one
has to take care of additional issues, such as quark mass effects and electro-weak effects
which typically contribute around or below the per-cent range.
The first determination of αs (MZ ) based on NNLO (and NLLA) calculations of event
shape distributions [49, 50] yields

αs (MZ ) = 0.1224 ± 0.0009 (stat) ± 0.0009 (exp) ± 0.0012 (hadr) ± 0.0035 (theo).

The fit to ALEPH thrust data is shown in Fig. 8.31(a). The largely reduced scatter of
values for different variables at NNLO is visualized in Fig. 8.31(b). Note that the reduced
perturbative uncertainty is 0.003.
190 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

The most precise determination of the strong coupling constant is obtained from jet
observables at LEP. Precision at the 2% level is achieved from the three-jet rate [51]:

αs (MZ ) = 0.1175 ± 0.0020 (exp) ± 0.0015 (theo).

The three-jet rate is known to have small non-perturbative corrections and to be very
stable under scale variations (for a certain range of the jet resolution parameter). For a
comparison of LO, NLO, and NNLO predictions to the corresponding ALEPH data, see
Fig. 8.32(a).
The LEP results concerning the determination of the strong coupling constant (see
Fig.8.32(b)) can be summarized as follows (combination by S. Bethke, a couple of years
ago).

• Tau decays (NNLO)

αs (MZ ) = 0.1181 ± 0.0030

• RZ (NNLO)

+0.0058
αs (MZ ) = 0.1226
−0.0038

• Event shapes (NLO + NNLO)

αs (MZ ) = 0.1202 ± 0.0050

• All (not including recent NNNLO results)

αs (MZ ) = 0.1195 ± 0.0035

• Latest world average (S. Bethke, 2009 [30])

αs (MZ ) = 0.1184 ± 0.0007

8.4 Measurements of the QCD color factors

Because they determine the gauge structure of strong interactions, the color factors are
the most important numbers in QCD, besides αs . Discussing the triple-gluon vertex we
concluded that our observables also allow to test the gauge structure of QCD. We have
already learned that the color factors (for SU (3)) CF = 4/3, CA = 3, and TF = 1/2
measure the relative probabilities of gluon radiation (q → qg), triple gluon vertex (g →
gg), and gluon splitting (g → q q̄).
8.5. HADRONIZATION 191

������������������������������

(a) (b)

Figure 8.32: NNLO, NLO, and LO fits to ALEPH data for the thee-jet rate (a) and
summary of LEP results for αs (b).

The cross section prediction for four-jet events at order αs2 can be shown to be
� � �
1 dσfour-jet αs2 CF2 1 CA
= σA (y) + 1 − σB (y)
σ0 dy π2 2 CF
� � � � � � �
CA TF 1 CA
+ σC (y) + nf σD (y) + 1 − σE (y)
CF CF 2 CF
where σi , i = A, . . . , E are kinematic factors independent of the gauge group of QCD.
The combined measurements of the QCD color factors are summarized in Fig. 8.33: Four-
jet and event shape results have been combined accounting for correlations between the
measurements. In addition, constraints on CA /CF from differences between quark and
gluon jets (see p. 177) are included. This yields
CA = 2.89 ± 0.21
CF = 1.30 ± 0.09
which is precise to 7% and agrees with the SU (3) values of CA = 3 and CF = 1.33.

8.5 Hadronization
The trouble with hadronization is that perturbative calculations are no longer useful
since αs ceases to be comparatively small at length scales of about the proton radius.
192 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

CONTENTS

���
�����
���������������
�
���������
�����
��������
���������

���
����������

��
�����������
�����������
�

�����

�
��� ����

���������������������
�����
�
� � � � � � �
��
Figure 8.33: Combined measurements of the color factors CA and CF . The ellipses show
the correlated measurements using four-jet events or event shape distributions while the
lines represent the results of determinations of CA /CF from DELPHI (dashed) and OPAL
(solid). Source: [52, p. 82].
8.5. HADRONIZATION 193

Figure 8.34: Visualization of phenomenological models of hadronization. (LHS) string frag-

mentation: JETSET/PYTHIA; (RHS) Cluster fragmentation: HERWIG. Source: [27, p.
164]

Perturbative QCD is applicable to the transition from the primary partons to a set of
final state partons. This is pictured as a cascading process that is dominated by the
collinear and soft emissions of gluons and mainly light quark-antiquark pairs. By contrast,
phenomenological models are used to describe the non-perturbative transition from these
final state partons to hadrons which then may decay according to further models (recall
Fig. 8.6).
The parameters determining the behavior of the numerical models have to be adjusted
using experimental data. Hadronization can be modeled by string fragmentation (JET-
SET/PYTHIA) or cluster fragmentation (HERWIG). For a visualization of this difference,
see Fig. 8.34.
Fig. 8.35 shows comparisons of simulations to ALEPH data for hadron momentum distri-
butions of the final state: Fig. 8.35(a) shows simulation and data for an inclusive variable
and Fig. 8.35(b) deals with pions, kaons, and protons, respectively.
194 CHAPTER 8. QCD IN E + E − ANNIHILATIONS

(a)

(b)

Figure 8.35: Hadron momentum distributions, ALEPH data and simulation. Inclusive
measurement (a) and differential cross section for pions, kaons, and protons (b) compared
with the predictions of JETSET, HERWIG, and ARIADNE. All observables are shown
as functions of x = phadron /pbeam .