Sei sulla pagina 1di 12

DGLAP: equazione di evoluzione

I. EQUAZIONE DI EVOLUZIONE
Lindipendenza di unosservabile dalla scala di fattorizzazione =
f

dO
d
= 0 +O(
2
S
)
porta allequazione di evoluzione. Infatti se si sostituisce O = F
2
0 =
2
dF
2
(x, Q
2
)
d
2
=
2
d
d
2

q
e
2
q
x
_
1
x
d

q(,
2
)
_
(1
x

) +

S
2
_
P(
x

)ln
Q
2

2
+ (C(
x

) C

(
x

))
__
e chiamiamo t =
2
si ottiene lequazione DGLAP (DokshitzerGribovLipatovAltarelli
Parisi) oppure GLAP o AltarelliParisi o equazione di evoluzione (1972-1977):
t
q(x, t)
t
=

S
2
_
1
x
d

P
qq
(
x

) q(, t)
In notazione compatta con il simbolo per identicare lintegrale di convoluzione A(x)
B(x) =
_
1
x
dy
y
A(
x
y
)B(y):
t
q(x, t)
t
=

S
2
P
qq
q
Le funzioni P
qq
sono dette splitting functions o anche evolution kernels.
La DGLAP `e unequazione integrodierenziale del primo ordine, quindi appena sia
nota la distribuzione partonica ad una certa scala (=la condizione iniziale per lequazione),
`e possibile calcolare il risultato ad ogni altra scala, purche nellintervallo di scale in consid-
erazione sia valida la QCD perturbativa.
La QCD perturbativa non pu`o predire i valori delle pdf, intrinsecamente non perturba-
tive, ma predice la loro variazione con la scala (levoluzione).
Dallequazione si vede che levoluzione della pdf ad un certo x dipende da tutti i suoi
valori a x pi` u elevati (lintegrale `e esteso da x ad 1). Quindi `e necessario conoscere a t
0
= Q
2
0
le pdf q(, Q
2
0
) per tutti i valori > x per poi valutare la q(x, Q
2
) con Q
2
> Q
2
0
. Questa `e
una fortuna dal momento che le misure a piccolo x sono quelle pi` u dicili da realizzare.
Se si sostituisce lespressione della splitting function:
P(x)
qq
= C
F
_
1 +x
2
1 x
_
+
si vede che la variazione di q con t `e data da tutti i quark che riducono il loro impulso
per emissione di un gluone: da una parte c`e lingresso in x dei quark che provengono da
x pi` u alti, e dallaltra luscita da x dei quark che perdono impulso andando a x pi` u bassi.
I due contributi sono entrambi divergenti in = 1 (emissioni di gluoni soci), ma le due
divergenze si cancellano tra loro:
t
q(x, t)
t
= C
F

S
2
_
1
x
d

q(
x

, t)
_
1 +
2
1
_
C
F

S
2
q(x, t)
_
1
x
d
_
1 +
2
1
_
1
N.B.:
x

e possono essere scambiati: se y =


x

allora
dy
y
=
d

.
Le splitting functions P(x) (allordine leading) hanno infatti linterpretazione sica di prob-
abilit` a che un partone vada in un altro partone avente una frazione x del momento longitu-
dinale del partone genitore (anche dette probabilit`a di branching).
II. SISTEMA DI EQUAZIONI
Bisogna considerare tutti i processi per cui un partone entra o esce dallelemento dxdt,
cio`e bisogna considerare i contributi da parte di tutti i dierenti tipi di partoni (quark,
antiquark e gluoni). Lequazione di evoluzione `e in realt`a, e pi` u in generale, un sistema di
(2N
f
+ 1) equazioni accoppiate:
t

t
_
q
i
(x, t)
g(x, t)
_
=

S
2
_
1
x
d

_
P
q
i
q
j
(
x

,
S
(t)) P
q
i
g
(
x

,
S
(t))
P
gq
j
(
x

,
S
(t)) P
gg
(
x

,
S
(t))
__
q
j
(, t)
g(, t)
_
t
q
i
t
=

S
2
[P
q
i
q
j
q
j
+P
q
i
g
g]
t
q
i
t
=

S
2
[P
q
i
q
j
q
j
+P
q
i
g
g]
t
g
t
=

S
2
[P
gq
i
(q
i
+ q
i
) +P
gg
g]
La coniugazione di carica e SU(N
f
) di avour implicano:
P
q
i
q
j
= P
q
i
q
j
P
q
i
q
j
= P
q
i
q
j
P
q
i
g
= P
q
i
g
P
qg
P
gq
i
= P
g q
i
P
g q
A ordini superiori le equazioni di evoluzioni mantengono la stessa forma, con delle funzioni
di splitting che contengono le correzioni perturbative in
S
:
P(x,
S
) = P
(0)
(x) +

S
2
P
(1)
(x) +. . .
Si pu` o dimostrare la validit`a della fattorizzazione a tutti gli ordini. in particolare
P
q
i
q
j
(x,
S
) =
ij
P
(0)
qq
(x) +

S
2
P
(1)
q
i
q
j
(x) +. . .
P
qg
(x,
S
) = P
(0)
qg
(x) +

S
2
P
(1)
qg
(x) +. . .
P
gq
(x,
S
) = P
(0)
gq
(x) +

S
2
P
(1)
gq
(x) +. . .
P
gg
(x,
S
) = P
(0)
gg
(x) +

S
2
P
(1)
gg
(x) +. . .
2
Allordine leading P
(0)
q
i
q
j
= 0 se q
i
= q
j
. Termini nulli al LO in generale iniziano a contribuire
a ordini superiori. Come gi`a detto, allordine leading le P hanno linterpretazione di proba-
bilit`a di emissione di un partone da parte di un altro partone. Inoltre soddisfano alle regole
di somma:
_
1
0
dx P
(0)
qq
(x) = 0 conservazione numero di quark di valenza
_
1
0
dx x [P
(0)
qq
(x) +P
(0)
gq
(x)] = 0 conservazione momento dei quark
_
1
0
dx x [2N
f
P
(0)
qg
(x) +P
(0)
gg
(x)] = 0 conservazione momento dei gluoni
Infatti il numero totale di quark di valenza `e indipendente dalla risoluzione con cui si osserva
il protone:
d
dQ
2
_
1
0
dx q
V
(x, Q
2
) =

S
2
_
1
0
d q
V
(, Q
2
)
_
1
0
dxP
qq
(x) = 0
Allordine leading le funzioni P sono (AltarelliParisi (1977), Altarelli (1982)):
P
(0)
qq
(x) = C
F
_
1 +x
2
(1 x)
+
+
3
2
(1 x)
_
,
P
(0)
qg
(x) = T
R
_
x
2
+ (1 x)
2

,
P
(0)
gq
(x) = C
F
_
1 + (1 x)
2
x
_
,
P
(0)
gg
(x) = 2C
A
_
x
(1 x)
+
+
(1 x)
x
+x(1 x)
_
+
0
(1 x)
dove
0
=
1
6
(11C
A
4N
f
T
R
), C
A
= 3, C
F
=
4
3
e T
R
=
1
2
.
Il risultato al leading order LO prevede il calcolo di 18 diagrammi a 1 loop (sono
sucienti carta e penna).
I risultati al Next to Leading Order NLO prevedono il calcolo di 350 diagrammi a
due loop, quindi `e necessario laiuto di programmi algebrici. (Curci, Furmanski, Petronzio;
Floratos, Ross, Sachrajda; Gonzales Arroyo, Lopez, Yndurain.)
I risultati al NNLO prevedono il calcolo di 9607 diagrammi a 3 loop. (Larin, Van
Ritbergen, Vermaseren, Nogueira; Moch, Vermaseren, Vogt.) Sono necessari programmi
algebrici e tecnologie di calcolo particolari.
3
III. SINGOLETTO E NON SINGOLETTO . . .
Per calcolare levoluzione delle distribuzioni partoniche `e utile introdurre le combi-
nazioni di non singoletto di avour:
V
i
= q

i
= q
i
q
i
T
3
= u
+
+d
+
T
8
= u
+
+d
+
2s
+
T
15
= u
+
+d
+
+s
+
3c
+
T
24
= u
+
+d
+
+s
+
+c
+
4b
+
T
35
= u
+
+d
+
+s
+
+c
+
+b
+
5t
+
dove
q

i
= q
i
q
i
e le combinazioni di singoletto di avour:
q
S
= =

i
q
+
i
=

i
(q
i
+ q
i
)
Ci sono 6 combinazioni di tipo V , 5 di tipo T, 1 di tipo .
Per le pdf nonsinglet q
NS
= V, T vale unequazione di evoluzione semplicata, dove
non c`e mixing con i gluoni:
t
q
NS
t
=

S
2
P
qq
q
NS
Per il singlet ci sono invece due equazioni accoppiate:
t

t
=

S
2
[P
qq
+ 2N
f
P
qg
g]
t
g
t
=

S
2
[P
gq
+P
gg
g]
IV. MOMENTI: TRASFORMATE DI MELLIN
Lequazione diventa pi` u semplice da risolvere nello spazio dei momenti: la con-
voluzione in questo caso diventa un prodotto ordinario, e lequazione integro-dierenziale
diventa unequazione dierenziale ordinaria.
Si denisce trasformata di Mellin o momento nesimo di f:
f(n, t) =
_
1
0
dx x
n1
f(x, t)
Lantitrasformata di Mellin o trasformata inversa :
f(x, t) =
1
2i
_
C
dn x
n
f(n, t)
4
`e un integrale nel piano complesso n, sul percorso C parallelo allasse immaginario che passa
a destra di tutte le singolarit`a dellintegrando. Si pu`o vericare che la trasformata di una
convoluzione diventi eettivamente un prodotto di momenti:
_
1
0
dxx
n1
__
1
x
dy
y
f(y)g(
x
y
)
_
=
_
1
0
dxx
n1
_
1
0
dy
_
1
0
dz(x zy)f(y)g(z)
=
_
1
0
dy
_
1
0
dz(zy)
n1
f(y)g(z)
= f(n)g(n)
Lequazione DGLAP non-singlet q
NS
= V, T nello spazio dei momenti diventa :
t
q
NS
(n, t)
t
=

S
2

qq
(n,
S
) q
NS
(n, t)
dove la funzione
qq
, detta dimensione anomala, `e la trasformata di Mellin o momento
nesimo della funzione di splitting:

qq
(n,
S
) =
_
1
0
dx x
n1
P
qq
(x,
S
)
Come le splitting functions anche le dimensioni anomale si espandono in
S
:

ij
(n,
S
) =

m=0

(m)
ij
(n)
_

S
2
_
m

(m)
ij
(n) =
_
1
0
dx x
n1
P
(m)
ij
(x)
Allordine leading...
V. SOLUZIONI ALLORDINE LEADING
A. Nonsinglet
La trasformata di Mellin ci permette di studiare le soluzioni analiche dellequazione
DGLAP. A meno di casi semplici lantitrasformata `e unoperazione complicata, che si esegue
numericamente.
Consideriamo una combinazione q
NS
di distribuzioni di partoni che sia nonsinglet: per
esempio la V = q
i
q
i
, e lequazione allordine leading:
t
q
NS
(n, t)
t
=

S
2

(0)
qq
(n) q
NS
(n, t)
5
Anomalous Dimensions and Wilson Coefcients
-0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
5 10 15
N
!
qq
(N)
LO
NLO
NNLO
ps
1
N
!
gg
(N)
"
S
= 0.2, N
f
= 4
1
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
5 10 15
Vermaseren, Moch, Vogt 2004
J. Bl umlein Dedicated to the memory of Jiro Kodaira KEK Symposium, March 2007
p.9
FIG. 1: Dimensioni anomale
Se assumiamo
S
costante con la scala t, la soluzione
q
NS
(n, t) = q
NS
(n, t
0
)
_
t
t
0
_
d
d =

S

qq
(n)
2
porta a correzioni a potenza in t, in contrasto con levidenza sperimentale di dipendenza
pressoche logaritmica, e di scaling approssimato da parte delle funzioni di struttura. In
generale per una teoria con un punto sso non nullo
(t)
t

0
= 0
`e problematico riconciliare la teoria dei campi con le violazioni di scaling.
Se invece consideriamo
S
running con la scala t (formula allordine leading):

S
(t)
2
=
1

0
log
t

0
11
6
C
A

2
3
T
R
N
f
allora la soluzione
q
NS
(n, t) = q
NS
(n, t
0
)
_

S
(t
0
)

S
(t)
_
d
qq
(n)
d
qq
(n) =

(0)
qq
(n)
2
0
descrive correttamente le violazioni di scala: i momenti delle distribuzioni variano con
potenze di ln(t).
I valori
d
qq
(1) = 0
d
qq
(n) < 0 n 2
6
10
-3
10
-2
10
-1
1
10
10
2
10
3
10
4
10
5
10
6
1 10 10
2
10
3
10
4
10
5
Q
2
/ GeV
2
F
2

!


2
i
x = 0.65, i = 0
x = 0.40, i = 1
x = 0.25, i = 2
x = 0.18, i = 3
x = 0.13, i = 4
x = 0.080, i = 5
x = 0.050, i = 6
x = 0.032, i = 7
x = 0.020, i = 8
x = 0.013, i = 9
x = 0.0080, i = 10
x = 0.0050, i = 11
x = 0.0032, i = 12
x = 0.0020, i = 13
x = 0.0013, i = 14
x = 0.00080, i = 15
x = 0.00050, i = 16
x = 0.00032, i = 17
x = 0.00020, i = 18
x = 0.00013, i = 19
x = 0.000080, i = 20
x = 0.000050, i = 21
H1 e
+
p
ZEUS e
+
p
BCDMS
NMC
H1 PDF 2000
extrapolation
H
1

C
o
l
l
a
b
o
r
a
t
i
o
n
ZEUS
0
1
2
3
4
5
1 10 10
2
10
3
10
4
10
5
F
2

e
m
-
l
o
g
1
0
(
x
)
Q
2
(GeV
2
)
ZEUS NLO QCD fit
tot. error
ZEUS 96/97
BCDMS
E665
NMC
x=6.32E-5
x=0.000102
x=0.000161
x=0.000253
x=0.0004
x=0.0005
x=0.000632
x=0.0008
x=0.0013
x=0.0021
x=0.0032
x=0.005
x=0.008
x=0.013
x=0.021
x=0.032
x=0.05
x=0.08
x=0.13
x=0.18
x=0.25
x=0.4
x=0.65
J. Bl umlein Dedicated to the memory of Jiro Kodaira KEK Symposium, March 2007
p.14
FIG. 2: Evoluzione della funzione di struttura con lequazione DGLAP
rendono conto del fatto che quando t cresce V decresce a grande x, e cresce a piccolo x:
F
(NS)
(n, t)
_
lnQ
2
/
2
lnQ
2
0
/
2
_
d
qq
(n)
La scoperta della libert`a asintotica nel 1973-74 (Gross-Wilczek,Politzer) ha permesso la
spiegazione della propriet`a di scaling approssimata.
7
4: DIS and Hadron Collisions
1
10
F
2
x=1.3E-05 (x 18)
x=2E-05 (x 10)
x=3.2E-05 (x 4)
x=5E-05 (x 2)
x=8E-05 (x 1)
H1SVX95
H194
H195NVX prel.
H196NVX prel.
E665
NMC
BCDMS
H1 QCDFit96 prel
1
10
x=0.00013 (x 10)
x=0.0002 (x 5)
x=0.00032 (x 3)
x=0.0005 (x 1.8)
x=0.0008 (x 0.8)
x=0.0013 (x 12)
x=0.002 (x 8)
x=0.0032 (x 4)
x=0.005 (x 2.2)
x=0.008 (x 1.3)
1
10
10
-1
1 10 10
2
10
3
10
4
x=0.013
(x 18)
x=0.02
(x 10)
x=0.032
(x 6)
x=0.05
(x 3.5)
x=0.08
(x 2)
10
-1
1 10 10
2
10
3
10
4
Q
2
[GeV
2
]
x=0.13
(x 32)
x=0.2
(x 16)
x=0.32
(x 8)
x=0.5
(x 6)
25
FIG. 3: Evoluzione della funzione di struttura con lequazione DGLAP
Nelle gure (2) e (3) sono riportati i risultati sperimentali della funzione di struttura F
2
del protone sovrapposti alle curve di evoluzione dellequazione DGLAP. La scala verticale
`e adattata per separare le curve a x dierente. Nonostante lintervallo in x e Q
2
coperto
dai dati sperimentali sia enorme, la descrizione dellevoluzione in Q
2
dei dati mediante t
di QCD `e spettacolare, e rappresenta il test pi` u importante per la QCD perturbativa.
La soluzione dellequazione di evoluzione rappresenta la risommazione delle emissioni
collineari a tutti gli ordini. Levoluzione dellequazione allordine leading risomma i con-
tributi (
S
lnt)
n
(leading logs), mentre al NLO sono compresi termini del tipo
S
(
S
lnt)
n1
Le soluzioni analitiche dellequazione DGLAP sono complicate, e spesso si preferisce
risolvere numericamente lequazione integrodierenziale direttamente nello spazio x.
8
12 Siegfried Bethke: The 2009 World Average of s
of the measurements with the others, exclusive averages,
leaving out one of the 8 measurements at a time, are cal-
culated. These are presented in the 5
th
column of table 1,
together with the corresponding number of standard de-
viations
5
between the exclusive mean and the respective
single measurement.
As can be seen, the values of exclusive means vary only
between a minimum of 0.11818 and a maximum 0.11876.
Note that in the case of these exclusive means and ac-
cording to the rules of calculating their overall errors,
in four out of the eight cases small error scaling factors
of g = 1.06...1.08 had to be applied, while in the other
cases, overall correlation factors of about 0.1, and in one
case of 0.7, had to be applied to assure
2
/n
df
= 1. Most
notably, the average value
s
(M
Z
0) changes to
s
(M
Z
0) =
0.11860.0011 when omitting the result from lattice QCD.
5 Summary and Discussion
In this review, new results and measurements of
s
are
summarised, and the world average value of
s
(M
Z
0), as
previously given in [7, 28, 6], is updated. Based on eight
recent measurements, which partly use new and improved
N3LO, NNLO and lattice QCD predictions, the new av-
erage value is

s
(M
Z
0) = 0.1184 0.0007 ,
which corresponds to

(5)
MS
= (213 9 ) MeV .
This result is consistent with the one obtained in the pre-
viuos review three years ago [28], which was
s
(M
Z
0 ) =
0.11890.0010. The previous and the actual world average
have been obtained from a non-overlapping set of single
results; their agreement therefore demonstrates a large de-
gree of compatibility between the old and the new, largely
improved set of measurements.
The individual mesurements, as listed in table 1 and
displayed in gure 5, show a very satisfactory agreement
with each other and with the overall average: only one
out of eight measurements exceeds a deviation from the
average by more than one standard deviation, and the
largest deviation between any two out of the eight results,
namely the ones from decays and from structure func-
tions, amounts to 2 standard deviations
6
.
There remains, however, an apparent and long-standing
systematic dierence: results from structure functions pre-
fer smaller values of
s
(M
Z
0) than most of the others, i.e.
those from e
+
e

annihilations, from decays, but also


those from jet production in deep inelastic scattering. This
issue apparently remains to be true, although almost all of
the new results are based on signicantly improved QCD
5
The number of standard deviations is dened as the
square-root of the value of
2
.
6
assuming their assigned total errors to be fully uncorre-
lated.
predictions, up to N3LO for structure functions, and Z
0
hadronic widths, and NNLO for e
+
e

event shapes.
The reliability of measurements of
s
based on ex-
periments on the lattice have gradually improved over
the years, too. Including vaccum polarisation of three light
quark avours and extended means to understand and cor-
rect for nite lattice spacing and volume eects, the overall
error of these results signically decreased over time, while
the value of
s
(M
Z
0 ) gradually approached the world aver-
age. Lattice results today quote the smallest overall error
on
s
(M
Z
0 ); it is, however, ensuring to see and note that
the world average without lattice results is only marginally
dierent, while the small size of the total uncertainty on
the world average is, naturally, largely inuenced by the
lattice result.
QCD ! (" ) = 0.1184 0.0007
s Z
0.1
0.2
0.3
0.4
0.5
!
s
(Q)
1 10 100
Q [GeV]
Heavy Quarkonia
e
+
e

Annihilation
Deep Inelastic Scattering
July 2009
Fig. 6. Summary of measurements of s as a function of the
respective energy scale Q. The curves are QCD predictions for
the combined world average value of s(M
Z
0), in 4-loop ap-
proximation and using 3-loop threshold matching at the heavy
quark pole masses Mc = 1.5 GeV and M
b
= 4.7 GeV. Full sym-
bols are results based on N3LO QCD, open circles are based on
NNLO, open triangles and squares on NLO QCD. The cross-
lled square is based on lattice QCD. The lled triangle at
Q = 20 GeV (from DIS structure functions) is calculated from
the original result which includes data in the energy range from
Q =2 to 170 GeV.
In order to demonstrate the agreement of measure-
ments with the specic energy dependence of
s
predicted
by QCD, in gure 6 the recent measurements of
s
are
shown as a function of the energy scale Q. For those results
which are based on several
s
determinations at dierent
values of energy scales Q, the individual values of
s
(Q)
FIG. 4:
S
vs Q
VI. MISURE DI
s
Per confrontare le misure di
S
estratte da diversi esperimenti `e necessario evolverle con
il gruppo di rinormalizzazione alla stessa scala, tipicamente alla massa dello Z
0
. Levoluzione
nello schema MS `e nota a quattro loop: NNNLO(!) (Larin,Van Ritbergen,Vermaseren). la
beta function `e forse la quantit`a meglio nota in QCD perturbativa. La denizione di massa
dei quark pu`o introdurre errori (m
b
= 4.3 0.2GeV m
c
= 1.3 0.3GeV )
Le determinazioni di
s
vengono
Studio del DIS La prima determinazione di
s
(1979) sfrutt`o proprio le predizioni del
DIS al NLO. Oggi sono disponibili molte misure sperimentali, su un grande intervallo
per Q
2
: Q
2
1 10000GeV
2
. Si usano fasci di elettroni, muoni e neutrini su
bersaglio sso, o collisioni di fasci elettrone(positrone)protone
Si studiano le funzioni di struttura nonsinglet
e la produzione dei jet.
Studio della larghezza adronica dello Z
0
R
Z
=
(Z
0
adroni)
(Z
0
e
+
e

)
9
Studio della larghezza di decadimento del leptone
R

=
(

adroni

)
(

)
Studio della spettroscopia dei mesoni pesanti nella QCD su reticolo
Studio dei decadimenti dei quarkonia pesanti
Sono i mesoni composti da quark-antiquark dello stesso avour pesante (c, b). In par-
ticolare `e studiato il decadimento radiativo della (stato legato bottom antibottom)
R =
( gg)
( ggg)
Studio delle congurazioni dellevento adronico (event shape) e della produzione dei
jet nelle collisioni e
+
e

Il test pi` u importante della QCD perturbativa tuttora rimane quello della rottura
dello scaling di Bjorken nel DIS. I t globali delle funzioni di struttura per esempio da
MRST04(Martin-Roberts-Stirling-Thorne tting group) rendono
S
(M
Z
) = 0.1205 NLO e

S
(M
Z
) = 0.1167 NNLO. Laccordo dei due risultati `e indicazione che gli errori sistematici
sono sotto controllo.
Le funzioni di struttura non singlet orono il test pi` u preciso, poiche levoluzione `e
indipendente dalla distribuzione del gluone, che `e quella meno conosciuta.
[1] C. Amsler et al. (Particle Data Group), Physics Letters B667, 1 (2008)
[2] K.Ellis, J.Stirling and B.Webber, QCD and Collider Physics Cambridge Monographs on Par-
ticle Physics, Nuclear Physics and Cosmology, volume 8, Cambridge University Press, (1996)
[3] S. Bethke, arXiv:0908.1135 [hep-ph].
10
9. Quantum chromodynamics 7
0.1 0.12 0.14
Average
Hadronic Jets
Polarized DIS
Deep Inelastic Scattering (DIS)
decays
Z width
Fragmentation
Spectroscopy (Lattice)
ep event shapes
Photo-production
decay
e
+
e
-
rates

s
(M
Z
)
Figure 9.1: Summary of the value of
s
(M
Z
) from various processes. The values
shown indicate the process and the measured value of
s
extrapolated to = M
Z
.
The error shown is the total error including theoretical uncertainties. The average
quoted in this report which comes from these measurements is also shown. See text
for discussion of errors.
consistent with the theoretical estimates. If the nonperturbative terms are omitted from
the t, the extracted value of
s
(m

) decreases by 0.02.
For
s
(m

) = 0.35 the perturbative series for R

is R

3.058(1+0.112+0.064+0.036).
The size (estimated error) of the nonperturbative term is 20% (7%) of the size of the
order
3
s
term. The perturbation series is not very well convergent; if the order
3
s
term
is omitted, the extracted value of
s
(m

) increases by 0.05. The order


4
s
term has been
estimated [47] and attempts made to resum the entire series [48,49]. These estimates
can be used to obtain an estimate of the errors due to these unknown terms [50,51].
Another approach to estimating this
4
s
term gives a contribution that is slightly larger
than the
3
s
term [52].
R

can be extracted from the semi-leptonic branching ratio from the relation
R

= 1/B( e) 1.97256; where B( e) is measured directly or extracted


from the lifetime, the muon mass, and the muon lifetime assuming universality of lepton
couplings. Using the average lifetime of 290.6 1.1 fs and a mass of 1776.99 0.29
July 24, 2008 18:04
FIG. 5: Misure di
S
Siegfried Bethke: The 2009 World Average of
s
11
The presence of correlated errors, if using the equations
given above, is usually signalled by
2
< n
df
. Values of

2
> n
df
, in most practical cases, are a sign of possibly
underestimated errors. In this review, both these cases are
pragmatically handled in the following way:
In the presence of correlated errors, described by a
covariance matrix C, the optimal procedure to determine
the average x is to minimise the
2
function

2
=
n

i,j=1
(x
i
x)(C
1
)
ij
(x
j
x) ,
which leads to
x =

ij
(C
1
)
ij
x
j

ij
(C
1
)
ij

1
and
x
2
=

ij
(C
1
)
ij

1
.
The choice of C
ii
= x
2
i
and C
ij
= 0 for i = j re-
tains the uncorrelated case given above. In the presence
of correlations, however, the resulting
2
will be less than
n
df
= n 1. In order to allow for an unknown common
degree of a correlation f, the method proposed in [61] will
be applied by choosing C
ij
= f x
i
x
j
and adjusting
f such that
2
= n 1.
For cases where the uncorrelated error determimation
results in
2
> n
df
, and in the absence of knowledge which
of the errors x
i
are possibly underestimated, all individ-
ual errors are scaled up by a common factor g such that
the resulting value of
2
/n
df
, using the denition for un-
correlated errors, will equal unity.
Note that both for values of f > 0 or g > 1, x
increases, compared to the uncorrelated (f = 0 and g = 1)
case.
4.2 Determination of the world average
The eight dierent determinations of
s
(M
Z
0 ) summarised
and discussed in the previous section are listed in ta-
ble 1 and are graphically displayed in gure 5. Apply-
ing equations 14, 15 and 16 to this set of measurements,
assuming that the errors are not correlated, results in
an average value of
s
(M
Z
0 ) = 0.11842 0.00063 with

2
/n
df
= 5.4/7.
The fact that
2
< n
df
signals a possible correlation
between all or subsets of the eight input results. Assuming
an overall correletion factor f and demanding that
2
=
n
df
= 7 requires f = 0.23, inating the overall error from
0.00063 to 0.00089.
In fact, there are two pairs of results which are known
to be largely correlated:
the two results from e
+
e

event shapes based on the


data from JADE and from ALEPH use the same theo-
retical predictions and similar hadronisation models to
0.11 0.12 0.13
! (" )
s Z
Quarkonia (lattice)
DIS F
2
(N3LO)

#-decays (N3LO)
DIS jets (NLO)
e
+
e

jets & shps (NNLO)

electroweak fits (N3LO)

e
+
e

jets & shapes (NNLO)

$ decays (NLO)
Fig. 5. Summary of measurements of
s
(M
Z
0 ). The vertical
line and shaded band mark the nal world average value of

s
(M
Z
0 ) = 0.1184 0.0007 determined from these measure-
ments.
correct these predictions for the transitions of quarks
and gluons to hadrons. While the experimental errors
are uncorrelated, the theoretical uncertainties may be
assumed to be correlated to 100%. The latter accounts
for about 2/3 to 3/4 of the total errors. An appropriate
choice of correlation factor between the two may then
be f = 0.67.
the QCD predictions for the hadronic widths of the
-lepton and the Z
0
boson are essentially identical, so
the respective results on
s
are correlated, too. The
values and total errors of
s
(M
Z
0) from decays must
therefore be correlated to a large extend, too. In this
case, however, the error of one measurement is al-
most entirely determined by the experimental error
(Z
0
-decays), while the other, from -decays, is mostly
theoretical. A suitable choice of the correlation factor
between both these results may thus be f = 0.5.
Inserting these two pairs of correlations into the error
matrix C, the
2
/n
df
of the averaging procedure results
in 6.8/7, and the overall error on the (unchanged) central
value of
s
(M
Z
0 ) changes from 0.00063 to 0.00067. There-
fore the new world average value of
s
(M
Z
0 ) is dened to
be

s
(M
Z
0 ) = 0.1184 0.0007.
For seven out of the eight measurements of
s
(M
Z
0),
the average value of 0.1184 is within one standard devi-
ation of their assigned errors. One of the measurements,
from structure functions [45], deviates from the mean value
by more than one standard deviation, see gure 5.
The mean value of
s
(M
Z
0 ) is potentially dominated
by the
s
result with the smallest overall assigned un-
certainty, which is the one based on lattice QCD [26]. In
order to verify this degree of dominance on the average
result and its error, and to test the compatibility of each
FIG. 6: Misure di
S
11
10 Siegfried Bethke: The 2009 World Average of s
Table 1. Summary of recent measurements of s(M
Z
0). All eight measurements will be included in determining the world
average value of s(M
Z
0 ). The rightmost two columns give the exclusive mean value of s(M
Z
0 ) calculated without that particular
measurement, and the number of standard deviations between this measurement and the respective exclusive mean, treating
errors as described in the text. The inclusive average from all listed measurements gives s(M
Z
0) = 0.11842 0.00067.
Process Q [GeV] s(Q) s(M
Z
0) excl. mean s(M
Z
0 ) std. dev.
-decays 1.78 0.330 0.014 0.1197 0.0016 0.11818 0.00070 0.9
DIS [F2] 2 - 15 0.1142 0.0023 0.11876 0.00123 1.7
DIS [e-p jets] 6 - 100 0.1198 0.0032 0.11836 0.00069 0.4
QQ states 7.5 0.1923 0.0024 0.1183 0.0008 0.11862 0.00114 0.2
decays 9.46 0.184
+0.015
0.014
0.119
+0.006
0.005
0.11841 0.00070 0.1
e
+
e

[jets & shps] 14 - 44 0.1172 0.0051 0.11844 0.00076 0.2


e
+
e

[ew] 91.2 0.1193 0.0028 0.1193 0.0028 0.11837 0.00076 0.3


e
+
e

[jets & shps] 91 - 208 0.1224 0.0039 0.11831 0.00091 1.0


where the rst error is experimental and the second is
theoretical, estimated by the dierence of the results in
NNLO and in N3LO QCD.
4 The 2009 world average of
s
(M
Z
0)
The new results discussed in the previous section are sum-
marised in table 1 and in gure 5. Since all of them are
based on improved theoretical predictions and methods,
and/or on improved data quality and statistics, they su-
persede and replace their respective precursor results which
were summarised in a previous review [28]. While those
previous results continue to be valid measurements, they
are not discussed in this review again, and they will not be
included in the determination of a combined world average
values of
s
(M
Z
0 ), according to the following reasons:
1. from each class of measurements, only the most ad-
vanced and complete analyses shall be included in the
new world average;
2. older measurements not being complemented or super-
seded by the most recent results listed above, as e.g.
results from sum rules, from singlet structure functions
of deep inelastic scattering, and from jet production
and b-quark production at hadron colliders, are not
included because their relatively large overall uncer-
tainties, in general, will not give them a sizable weight
but will complicate the denition of the overall error
of the combined value of
s
(M
Z
0 );
3. restricting the new world average to the most recent
and most complete (i.e. precise) results allows to ex-
amine the consistency between the newest and the pre-
vious generations of measurements and reviews.
4.1 Numerical procedure
The average x of a set of n dierent, uncorrelated mea-
surements x
i
of a particular quantity x with individual
errors or uncertainties, x
i
is commonly dened using
the method of least squares (see e.g. [6]: For x
i
being
independent and statistically distributed measures with a
common expectation value x but with dierent variances
x
i
, the weighted average is dened by
x =

n
i=1
w
i
x
i

n
i=1
w
i
(14)
and the variance x of x is minimised by choosing
x
2
=
1

n
i=1
1
x
2
i
, i.e. w
i
=
1
x
2
i
. (15)
The quality of the average is dened by the
2
variable,

2
=
n

i=1
(x
i
x)
2
x
2
i
(16)
which is, for uncorrelated data, expected to be equal to
the number of degrees of freedom, n
df
:

2
= n
df
= n 1 .
The results summarised in table 1, however, are not
independent of each other. They are, in the most general
sense, correlated to an unknown degree. While the statis-
tical errors of the data and the experimental systematic
uncertainties contained in the errors are independent and
uncorrelated, the theoretical uncertainties are very likely
to be (partly) correlated between dierent results, because
they are all based on applying the same underlying the-
ory, i.e. QCD, and similar methods to obtain estimates of
theoretical uncertainties are being used.
For some observables, like e.g. the hadronic widths of
the Z
0
boson and the lepton, the theoretical predictions
and hence, their uncertainties, are known to be correlated
by almost 100%. For other cases, like the results based on
lattice QCD and those based on QCD perturbation the-
ory, it can be assumed that their theoretical uncertainties
are not correlated at all. In addition to the inherent lack of
knowledge of theoretical correlations, estimates of theoret-
ical uncertainties, in general, are performed in widely dif-
ferent ways, using dierent methods and dierent ranges
of parameters.
FIG. 7: Misure di
S
12