Factorization

l Factorization is the key to

perturbative QCD
◆ the ability to separate the short-
distance physics and the long-
distance physics
l In particular, parton distribution
functions are part of the long-
distance physics
l Factorization tells us that PDFs The calculation of hard scattering processes
at the LHC requires:
determined from one process
(1)knowledge of the distributions of the
(or group of processes) can be quarks and gluons inside the proton, i.e.
used for other processes what fraction of the momentum of the
l So we can determine PDFs from parent proton do they have
experiments whose data was ->parton distribution functions (pdf’s)
taken long before you were born (2) knowledge of the hard scattering cross
(and more recent data as well) sections of the quarks and gluons, at LO,
NLO, or NNLO in the strong coupling
and use them for the LHC
constant αs
Let’s think about this from a space-time perspective

l Partons in the proton are

always emitting virtual gluons/
quark-antiquark pairs which
then recombine (the proton
remains intact)
l The lifetime of these virtual
states depends inversely on
the energy of the partons
l uncertainty principle
l If I can probe smaller and
smaller distances (time-
scales), then I can resolve
more of the radiative structure
inside the proton
l I can probe these smaller
distances by using higher
energies (Q) to probe small time interval,
distance scale
 Consider deep-inelastic scattering (DIS)

here I’m actually

probing these
virtual fluctuations;
l Condition for DIS on protons quark anti-quark pair
◆ Q>mp ~ 1/Rp : scale Q (typically the virtual mass of the photon) cannot recombine
l Using Breit frame

brick wall frame of reference

l Introducing Bjorken variable x

(we’ll call this x from now on, the fraction of the
proton’s momentum taken by a parton)
Consider the timescales involved
 Parton distributions
l The momentum of the proton is distributed among the quarks and
gluons that comprise it
◆ about 40% of the momentum is with gluons, the rest with the quarks
◆ note that the quarks at high x tend to be valence quarks (uud), while the
quarks at low x tend to be sea quarks produced by gluon splitting into
f(x,Q2)
describes quark-antiquark pairs (u-ubar, d-dbar, s-sbar, etc)
the momentum distribution of partons inside a proton

Q is a (factorization)
scale at which PDFs
are evaluated)

momentum fraction
 Parton distributions
l The momentum of the proton is distributed among the quarks and
gluons that comprise it
◆ about 40% of the momentum is with gluons, the rest with the quarks
◆ note that the quarks at high x tend to be valence quarks (uud), while
the quarks at low x tend to be sea quarks produced by gluon splitting into
quark-antiquark pairs (u-ubar, d-dbar, s-sbar, etc)

as we expected from
our simple model

momentum fraction
 Parton distribution functions (PDFs)

l Note the changes in the distributions as Q2 increases

(DGLAP)
l High x distributions decrease, while low x distributions
increase
◆ due to the effects of QCD evolution
Q2=10000 GeV2
Q2=2 GeV2
 Back to DIS

due to DGLAP evolution, number

of partons at low x increases
with Q2; the number of partons
at high x decreases with Q2

note that early measurements of

DIS at SLAC were for x values
around 0.1; at these x values the
number of partons is relatively
stable with respect to Q2, i.e. no
scaling violations
PDFs are non-perturbative
objects. Only evolution is
perturbative (DGLAP).

Global PDF fits are carried out

using data from a variety of
processes, including DIS data
taken long before you were
born. Increasingly, LHC data
are being used in the fits.
 Global fits
l So what do we need
l With the DGLAP equations,
◆ a value of Qo (1.3 GeV for
we know how to evolve PDF’s
CTEQ) lower than the data
from a starting scale Q0 to any
used in the fit (or any
higher scale prediction)
l …but we can’t calculate what ◆ a parametrization for the
the PDF’s are ab initio PDF’s
◆ one of the goals of lattice ◆ a scheme for the PDF’s
QCD ◆ hard-scattering calculations at

l We have to determine them the order being considered in

from a global fit to data the fit
◆ PDF evolution at the order
◆ factorization theorem tells
us that PDF’s determined being considered in the fit
for one process are ◆ a world average value for αs

applicable to another ◆ a lot of data

with appropriate kinematic

◆ extremely important proof
▲
cuts
◆ a treatment of the errors for
the experimental data
◆ MINUIT
 Global fits
l Parametrization: initial form
l What do we know?
◆ f(x)~xα(1-x)β
1. we know that the sum of the
◆ estimate β from quark
momentum of all partons in the
counting rules
proton is 1
▲ β=2ns-1 with ns being the
minimum number of 2. we know the sum of valence
spectator quarks quarks is 3
▲ so for valence quarks in a ◆ and 2 of them are up quarks
proton (qqq), ns=2, β=3 and 1 of them is a down
▲ for gluon in a proton (qqqg), quark
ns=3, β=5 ◆ we know that the net number
▲ for anti-quarks in a proton of anti-quarks is 0
(qqqqqbar), ns=4, β=7
◆ we know that the net number
◆ estimate α from Regge of strange quarks (charm
arguments quarks/bottom quarks) in the
▲ gluons and anti-quarks have proton is 0
α~-1 while valence quarks
have α~=1/2 This already puts a lot of restrictions
on the PDF’s
◆ but at what Q value are these
arguments valid?
 PDF luminosities
l Useful to define PDF luminosities

for any 2 partons, A

and B

PDF luminosity times

partonic cross section
equals hadronic cross
section

note that gluon-gluon

scattering dominates
at low mass and
quark-quark scattering
dominates at high mass

of course, specific PDF

luminosity needed depends
on which process is being
considered
 Factorization theorem
confinement freedom confinement

p3
we’ll discuss
later

p4
 Master formula for cross section calculation
a factorization scale µF a renormalization
scale µR

both PDFs
and matrix
elements
can be at LO,
NLO and
NNLO

We’ll start by calculating the matrix element for W boson production at leading
order.
 W boson production at leading order
l Consider production of an on-shell W+ boson; only 1 diagram

l The corresponding matrix element reads

CKM matrix element weak coupling

quarks have to be a color-anticolor singlet since the W boson has no color

l Summed and squared expression reads

Q=p1+p2=mW2

factor of 3 comes from sum over 3 possible quark-line colors, 1/9 takes care of averaging over
all possible color configurations of the quark and antiquark, and 1/4 takes care of averaging over
the incoming quark spins.
 ...add the decay
l Now let the W+ decay

l Matrix element can be written as

W is no longer a
final state particle,
but instead a
l Squaring yields propagator

average over initial quarks’ spins and colors,

and sum over lepton spins implicit

l Define Mandelstam variables

 Phase space

l Write phase-space integral over final state particles as conservation of

momentum

solid angle of outgoing lepton in rest frame of collision

width of the
l Then can re-write W boson

some kinematic
Breit-Wigner for
W propagator tricks
l And for the final cross section
 Narrow width approximation
l It is often useful to use the narrow width approximation to simplify
the calculation, where the propagator (Breit-Wigner) of mass MX
and width Γx is replaced by

l The cross section then can be written as

l Note this ignores correlations between the initial state particles and
the final state particles (spins for example)
Consider the emission of a gluon (or quark from a gq initial state)

l Three sets of diagrams,each

with two interfering amplitudes
these
◆ identical initial and final intefere
states

these
interfere
these
interfere

color matrix appearing in the quark-quark-gluon vertex

note potentially divergent terms when parton gets soft, or angle

approaches 0
 Matrix elements squared
l Square and average/sum over initial/final states polarizations and colors,
and performing some color algebra, get

l Closer inspection reveals that the squared matrix elements can be written
as the leading order matrix element squared (for W production) times a
QCD emission term, consisting of the strong coupling and a color factor
times an expression representing the kinematics of the extra emission
note the
divergence when
t-hat or u-hat
goes to zero
 Modern life

l Note that this procedure works for simple processes, 2->n, where n
is small (n=1 for W production, n=2 for W+j), but the number of
Feynman diagrams increases (more than) factorially with n
l Squaring the amplitudes, taking traces, is just too complex a
process for large n
l In modern techniques, alas beyond the scope of these lectures, the
focus is on evaluating individual amplitudes as a function of their
internal and external degrees of freedom
◆ helicity amplitude method: any Feynman amplitude

(represented by propagators and vertices for the internal lines

and spinors and polarization vectors for the external particles)
is translated into a complex number dependent on external
helicities and momenta

l Every amplitude becomes just a complex number

l Summation and squaring is then a (more) straightforward exercise
 Let’s start over, in a somewhat more pedagogical way

l Consider Drell-Yan production

where xa is the momentum fraction
◆ write cross section as of parton a in hadron A, and xb the
momentum fraction of parton b in
hadron B, and Q is a scale that
◆ where X=l+l- measures the hardness of the
◆ note we’re back to the parton interaction
model, i.e. no QCD corrections
l Potential problems appeared to
arise from when perturbative
corrections from real and virtual
gluon emissions were calculated
◆ but these logarithms were the
same as those in structure
function calculations and thus
can be absorbed, via DGLAP
equations in definition of parton
distributions, giving rise to
logarithmic violations of scaling
◆ can now write the cross section
as
 …but
l Key point is that all logarithms
appearing in Drell-Yan
corrections can be factored into
renormalized (universal) parton
distributions
◆ factorization

l But finite corrections left behind

after the logarithms are not
universal and have to be
calculated separately for each
process, giving rise to order αsn also depends on µR and
perturbative corrections µF, so as to cancel scale
l So now we can write the cross dependence in PDF’s and αs,
section as to this order

l where µF is the factorization scale An all-orders cross section has no

(separates long and short- dependence on µF and µR; a residual
distance physics) and µR is the dependence remains (to order αsn+1) for
renormalization scale for αs a finite order (αsn) calculation
l choose µR=µF~Q (say,mW/Z)
(see later discussion as well)
 Kinematics
l Double differential cross
section for production of a
Drell-Yan pair of mass M and
rapidity y is given by

◆ where
4πα 2
σˆ o =
3M 2

◆ and

l Thus, different values of M

and y probe different values of
x and Q2
 W/Z production
l Cross sections for on-shell W/Z
production (in narrow width limit)
given by

l Where Vqq’ is appropriate CKM

matrix element and vq and aq are
the vector and axial coupling of LO->NLO is a fairly large (+) correction
the Z to quarks
l Note that at LO, there is no αs NLO->NNLO is a fairly small (+)
dependence; EW vertex only correction
l Quark and anti-quark have to be
color-anticolor pair
◆ factor of 3 suppression

l NLO contribution to the cross

section is proportional to αs;
NNLO to αs2…
 W/Z pT distributions
l Most W/Z produced at low pT,
but can be produced at non-
zero pT due to diagrams such
as shown on the right; note
the presence of the QCD
vertex, where the gluon
couples (so one order higher)
l So an example of a 2->2
process W boson
 W/Z pT distributions
l Most W/Z produced at low pT,
but can be produced at non-
zero pT due to diagrams such
as shown on the right; note
the presence of the QCD
vertex, where the gluon
couples (so one order higher) Mandelstam variables
If this were photon
production, and not
W, then this last term
would not be present

Note that 2->2 matrix elements are

l Sum is over colors and spins singular when final state partons are
in initial state, averaged over soft or collinear with initial state partons
same in final state (soft and collinear->double logarithms)
l Transverse momentum ^ and u=0
^
Related to poles at t=0
distribution is obtained by
convoluting these matrix But singularities from real and virtual
elements with PDF’s in usual emissions cancel when all contributions
way are included, so NLO is finite
 Aside
l Can we say which quark the
gluon is emitted from?
l No, that’s a classical picture
(most often adopted in Monte
Carlos), but doesn’t fit into our
quantum mechanical picture
l In a similar way, if we have a
diagram with a gluon that can
be emitted from either the
initial or final state, we can’t
say from which it was emitted
◆ the two diagrams interfere
with each other
 W/Z pT distributions
l Back to the 2->2
subprocess

◆ where Q2 is the it’s pretty clear that Q~mW is a good choice

virtuality of the W as long as the gluon is reasonably soft
boson
l Convolute with PDFs

phase space momentum

for W and gluon conservation
 W/Z pT distributions
l Transform into differential
cross section

◆where we have one

integral left over, the gluon
rapidity
2 tˆuˆ
l Note that pT = sˆ
◆ thus, leading divergence can
be written as 1/pT2
l In this
€ limit, behavior of cross
section becomes
…diverges unless we apply
a pTmin cut; so we end up
l As pT of W becomes small, with a distribution that
limits of yg integration are depends not only on αs but
given by +/- log(s1/2/pT) on αs times a logarithm:
l The result then is universal theme
 Rapidity distributions
l Now look at rapidity distributions the pT requirement of the gluon
for jet for two different choices of serves as the cutoff
pTmin
l Top diagrams imply that gluon is
radiated off initial state parton at
an early time (ISR)
l With collinear pole, this would
imply that these gluons would be
emitted primarily at forward
rapidities
l But the distributions look central
l The reason is that we are binning
in pT and not in energy, and the
most effective place to convert
from E to pT is at central rapidities
l Suppose I re-draw the Feynman
diagrams as shown to the right
◆ is there a difference from
what is shown at the top of
the page?
◆ hint: no
 Now on to W + 2 jets
l For sake of simplicity,
consider Wgg
l Let p1 be soft
l Then can write

where tA and tB are

◆ so the kinematic structures obtained
from the Feynman diagrams are
color labels of p1 and collected in the function D1,D2 and D3,
p2 which are called color-ordered
amplitudes
l Square the matrix
amplitude to get using tr(tAtBtBtA)=NCF2 and tr(tAtBtAtB)=-CF/2
 W + 2 jets
l Since p1 is soft, can write D’s
(color-ordered amplitudes) as
product of an eikonal term and
the matrix elements containing
only 1 gluon

◆where εµ is the polarization

vector for gluon p1
l Summing over gluon
polarizations, we get

◆ where
Observables and orders

suppose I want to know

dσ/dφjj to NLO; then I
would need the 1 loop
correction to the
diagram on the left, and
the W+3 jets real
correction; both of order
αs3
 Power series
W production

Cross section is a power series in αs

∞
n
dσ = ∑ s f n (…)
α
n = n0

For perturbation theory to work, need αs ≡ gs2/4π <<1

Each vertex has gs in amplitude->gs2 is proportional to αs in cross

section

Higher orders->more vertices->more diagrams (n!)->fn becomes

more difficult to calculate

But if αs<<1, can truncate series (LO, NLO, NNLO,…)

For W production, NNLO corrections are reasonably small; not true

for Higgs production, for example
 Color flow

l The leading term (in number of

colors) contains singularities
along two lines of color flow-one
connecting gluon p2 to the quark
and the other connecting it to the
anti-quark
◆ sub-leading term has
singularities along the line
connecting the quark and
anti-quark
l It is these lines of color that
indicate preferred direction for
emission of additional gluons
◆ needed by programs like
Pythia/Herwig for example …and thus can’t be fed directly into
◆ sub-leading terms don’t the parton shower Monte Carlo
correspond to any unique programs
color flow
 Eikonal factors
l Re-write

l As

l It is clear that the cross section

diverges either as cosθa->1
(gluon is collinear to parton a) or
as E->0
◆ similar for parton b

l Each divergence is logarithmic

and regulating the divergence by
providing a fixed cutoff (in angle
or energy) will produce a single
logarithm from collinear
configurations and another from
soft ones
◆ double logs
Brief interlude: jet definitions and algorithms

l At (fixed) LO, 1 parton = 1 jet

◆ why not more than 1? I have
to put a ΔR cut on the
separation between two
partons; otherwise, there’s a
collinear divergence. LO
parton shower programs
effectively put in such a cutoff

# 1 &
log% (
$ ΔR34 '

l But at NLO,€I have to deal with

more than 1 parton in a jet, and
so now I have to talk about how
to cluster those partons
◆ i.e. jet algorithms
 Jet algorithms at NLO
l At NLO (NNLO), there can be l A jet algorithm is based on some
two(three) partons in a jet, life measure of localization of the
becomes more interesting and expected collinear spray of
we have to start talking about particles
jet algorithms to define jets l Start with an inclusive list of
◆ we will see that the partons (fixed order), particles
addition of the extra (PS shower Monte Carlos, and
parton(s) and virtual terms data)
will cancel the divergence l End with lists of same for each jet
mentioned on the previous
slide l …and a list of particles… not in
any jet; for example, remnants of
the initial hadrons
l Two broad classes of jet
algorithms
◆ cluster according to proximity

in space: cone algorithms

◆ cluster according to proximity

in momenta: kT algorithms
 What do I want out of a jet algorithm?
l It should be fully specified, l It should be IR safe, i.e. adding a
including defining in detail any soft gluon should not change the
pre-clustering, merging and results of the jet clustering
splitting issues
l It should be simple to implement
in an experimental analysis, and
should be independent of the
structure of the detector
l It should be boost-invariant
l It should be collinear safe, i.e.
l It should be simple to implement
in a theoretical calculation splitting one parton into two
collinear partons should not
◆ it should be defined at any order
in perturbation theory change the results of the jet
◆ it should yield a finite cross
clustering
section at any order in
perturbation theory
◆ it should yield a cross section that
is relatively insensitive to
hadronization effects
 Jet algorithms
l The algorithm should behave in a similar manner (as much as
possible) at the parton, particle and detector levels. Note that
differences between levels can unavoidably creep in.
 The kT family of jet algorithms
l p=1 d=distance measure
◆ the regular kT jet algorithm ΔRij2
l p=0 (
dij = min pT2 ,ip , pT2 ,pj ) D2
◆Cambridge-Aachen algorithm
l p=-1 dii = pT2 ,ip size of
jet in
◆ anti-kT jet algorithm
Δy-Δφ
◆ Cacciari, Salam, Soyez ’08 l #1 algorithm for space
◆ also P-A Delsart ’07 (reverse

kT) € ATLAS, CMS

◆ soft particles will first cluster
l Actually, seems to be
with hard particles before
clustering among themselves the only algorithm
◆ no split/merge used
◆ leads mostly to constant area

hard jets
 ATLAS W + 2 jet event

.with the W
boson
decaying into
an electron
and a neutrino

...and the 2 jets

defined with the
antikT algorithm
with R=0.4
 Back to logarithms
l You can keep applying this l Size of L depends on criteria
argument at higher orders of used to define the jets (min pT,
perturbation theory cone size)
l Each gluon that is added l Coefficients cij depend on
yields an additional power of color factors
αs, and via the eikonal l Thus, addition of each gluon
factorization outlined, can results in additional factor of
produce an additional two αs times logarithms
logarithms (soft and collinear) l In many (typically exclusive)
l So can write the W + jets cases, the logs can be large,
cross section as leading to an enhanced
probability for gluon emission
to occur
◆ where L represents the l For most inclusive cases, logs
logarithm controlling the are small and αs counting may
divergence, either soft or be valid estimator for
collinear (Sudakov logs) production of additional jets
◆ note that αs and L appear l For completely inclusive cross
together as αsL sections, the logs vanish
 Specific example
parton 5
l Remember we encounter
parton 1
logs whenever an emitted
gluon becomes soft and/
or collinear

parton 2

l We said the cij were color

factors
l So for emission of parton 5
from parton 1, color factor is not present since have 2 extra gluons,
CF not 1

l For emission of parton 4 from

parton 3, CA
l If parton 5 is soft, and
l If one of the partons is not soft or
collinear with parton 1, and
collinear, then only 3 powers of
parton 4 is soft, and is
logs
collinear with parton 3, have 4
powers of logs l …and so on
l Factors of 2, π, etc ignored
for W + jets
Re-shuffling
each gluon added has an additional
factor of αs and two additional logs
(soft and collinear)
l re-write the term in brackets cij depend on color factors
as

l Where the infinite series has

been resummed into an
exponential form
◆ first term in expansion is
called leading logarithm
term, 2nd next-to-leading
logarithm, etc as jet definitions change, size of the logs
l Now can write out each shuffle the contributions from one jet
contribution as a combination cross section to another, keeping the sum
of terms in powers of αs and over all contributions the same; for example,
as R decreases, L increases, contributions shift
logarithms towards higher jet multiplicities
 Re-shuffling
•Configuration shown to the right
can be reconstructed as an event
containing up to 2 jets (0,1,2),
depending on jet definition and
momenta of the partons.
•For a large value of Rcone, this is
one jet; for a smaller value, it may
be two jets
•The matrix elements for this process
contain terms proportional to
αs log(pT3/pT4) and as log(1/ΔR34),
so min values for transverse
momentum and separation must be
imposed
•Suppose that I consider completely
inclusive cross sections (σW+>=0 jets)
•Then the logs vanish
Reviewing
 NLO calculations
l NLO calculation requires
consideration of all diagrams
that have an extra factor of αs
◆ real radiation, as we have
just discussed
◆ virtual diagrams (with vertex
correction
loops)
l For virtual diagram, have to self-energy
corrections
integrate over loop momentum
◆ but result contains IR

singularities (soft and O(αs) virtual corrections in NLO

collinear), just as found for cross section arise from
tree-level diagrams interference between tree level and
one-loop virtual amplitudes

If we add the real+virtual contributions, we find that the singularities will cancel,
for inclusive cross sections. We have to be more clever for differential distributions.
 Scale choices
l We know that we have two l Often µR and µF are taken
scales, µR and µF equal to each other, but the
l We know that they should be physics associated with each
associated with the relevant is a bit different, so they can
scale in the hard scattering be varied separately…as long
process as the ratio between the two
◆ sometime this scale is
scales is not too large (>2)
evident, like mW for W l For then, we would introduce
production, pTjet for a new log into the calculation,
inclusive jet production the log of the ratio of the two
◆ but what if I have a
scales
process like W+jet(s) l These logarithms would then
▲ there I have both mW and have to be re-summed to
pTjet, and these scales can be restore precision to the
very different->very different measurement
answers
▲ we’ll see that for some
l We don’t want to have to do
cases, general scales like HT that
may work best sum of transverse momenta of all objects in event
 Scale uncertainties
l We try to estimate the uncertainty due to uncalculated
higher order terms by varying µR,µF over some range,
typically a factor of 2
l This is normally the best we can do, but we have to
keep in mind that higher order corrections can arise
from a number of other sources such as Sudakov
effects, large color factors, large π2 terms, the opening
of new channels
l These contributions are not estimated by the variation
of the scale logarithms and can be larger than the
variation
What does the scale dependence for a cross section look like?

l Here, we’re specifically looking at

inclusive jet production, but this holds
for other collider processes
l Write cross section indicating explicit
scale-dependent terms for NLO
l First term (lowest order) in (3) leads to
monotonically decreasing behavior as
scale increases (the LO piece)
l Second term is negative for µ<pT,
positive for µ>pT
l Third term is negative for factorization
scale M < pT
l Fourth term has same dependence as
lowest order term
(1)
l Thus, lines one and four give (2)
contributions which decrease
monotonically with increasing scale (3)
while lines two and three start out (4)
negative, reach zero when the scales
are equal to pT, and are positive for
larger scales
l At NLO, result is a roughly parabolic
behavior
Why does the scale dependence have the shape it does?

l Write cross section indicating explicit

scale-dependent terms Note that
l First term (lowest order) in (3) leads to NLO=LO
monotonically decreasing behavior as for a scale
scale increases (the LO piece) of about pT/2;
l Second term is negative for µ<pT, for other scales
positive for µ>pT NLO>LO, or
l Third term is negative for factorization NLO<LO
scale M < pT
l Fourth term has same dependence as
lowest order term
l Thus, lines one and four give
contributions which decrease
monotonically with increasing scale
(1)
while lines two and three start out
(2)
negative, reach zero when the scales
are equal to pT, and are positive for (3)
larger scales (4)
l At NLO, result is a roughly parabolic
behavior
Look at scale dependence in 2-D

Jet production at the LHC

It’s also useful to use a log-log scale
l …since
perturbative
QCD is
logarithmic
l Note that
there’s a saddle
region, and a
saddle point,
where locally
there is little
slope for the
cross section
with respect to
the two scales
l This is kind of
the ‘golden
point’ and
typically around
the expected
scale (pTjet in
this case)
It’s also useful to use a log-log scale
l Choose pTjet as the
central scale
l The scale variation
represents an
estimate of the
uncalculated higher
orders
l Typically vary both
µR and µF up and
down from their
central values to
estimate the scale
uncertainty
l ...sometimes
making sure that
the ratio of the two
scales is never
larger than two,
creating the
diamond
 Advantages of higher orders
l Less sensitivity to unphysical input consider inclusive jet prod
scales, i.e. renormalization and at LO, NLO, NNLO
factorization scales
l NLO is first level of prediction
where normalization (and
sometimes shape) can be taken
seriously
l At NNLO can take uncertainties
more seriously
uncertainty at
l More physics NNLO<NLO<LO
◆ parton merging gives structure

in jets
◆ more species of incoming

partons
More scale terms in NNLO expression
 Back to W production to NLO
l In 4-dimensions, the contribution
of the real diagrams can be
written (ignoring diagrams with
incoming gluons for simplicity)
+ # uˆ
2 2 tˆ 2Q2 sˆ &
M(ud → W g ~ g CF % + +
$ tˆ ˆu uˆ tˆ ('

2
#* 1 + z 2 - * − sˆ − sˆ - &
~ g CF %, / + ˆ + ˆ . − 2(
$+ 1− z . t u '
◆ where
Q2
€ z = and sˆ + tˆ + uˆ = Q2
s
and don’t sweat the details; I just
l Note that the real diagrams want you to see in general terms
contain collinear singularities, how a NLO calculation is
^
u->0, ^t->0, and soft singularities,
€ carried out
z->1
 Aside: dimensional regularization
l Suppose we have an integral of the form, typical of the integrals in a NLO
calculation
d4k 1
I=∫
(2π ) 4 ( k 2 − m 2 ) 2
l We get infinity if we integrate this in 4 dimensions, so go to 4-2ε
dimensions
€ d4k 2ε d 4 −2ε k 2ε dΩ4 −2ε 3−2ε
∫ (2π )4 → ( µ ) ∫ (2π )4 −2ε → ( µ ) ∫ (2π )4 −2ε dk
∫ EEk

dΩ4 −2ε 2 1
∫ (2π )4 −2ε (4π )2−ε Γ(2 − ε )
=

2ε
∞
k E 3−2ε (µ ) 2ε 1
ε −1 1* µ,
2ε
Γ(ε )Γ(2 − ε )
(µ ) ∫ dkE 2 = 2ε
1−ε
∫ dzz (1− z) =
0 (k E
2
+ m2 ) 2( m) 0
2 + m- Γ(2)

l Using
Γ(1 + z) = zΓ( z);Γ' (1) = −γ E = −0.5772...
€

€
 Dimensional regularization, continued
l Find
2ε
Γ(ε ) & µ ( 1 , 1 & µ( /
I= → + − γ + ln( 4π ) + 2ln + O(ε )
(4π ) 2−ε ' m ) ε->0 (4π ) 2 .- ε E ' m) 1
0

◆ singular bits, plus finite bits as ε->0, plus log singularity as m->0

€
l Define MS scheme: subtract (absorb) 1/ε pole, γE, and ln(4π) bits
 Now do the dimension trick for the real part

l Problem: if I work in 4
dimensions, I get divergences
l Solution: working in 4-2ε
dimensions, to control the
divergences (dimensional
reduction)
ε
α s % µ 2 ( /% 2 3 π 2 ( 2 2 / ln(1− z ) 2 1 + z2 2
σ real = CF ' 2 * cΓ 1' 2 + − * δ (1− z) − Pqq (z) − 2(1− z) + 4 (1 + z ) 1 4 − 2 ln z 4
2π & Q ) 0& ε ε 3) ε 0 1− z 3 + 1− z 3

l with “+ distribution”
(4π )ε # log(1− z) & - log(1− z) 1 0
cΓ = % ( ≡ lim. θ (1− z − β ) + log 2 (β )δ (1− z − β )1
Γ(1− ε ) $ 1− z ' + β->0/ 1− z 2 2

We get 1/ε terms from individual soft and collinear singularities

€
We get 1/ε2 terms for overlapping IR singularities.
€