Even More Perturbative QCD ...

Lorenzo Magnea Universit` di Torino a I.N.F.N. Torino magnea@to.infn.it XI Seminario Nazionale di Fisica Teorica

Abstract These lectures describe the treatment of mass divergences in perturbative QCD, concentrating on hadron production in electron-positron annihilation. I perform an explicit and detailed one-loop calculation, and use it to infer some results and techniques valid to all orders in perturbation theory. I introduce some of the tools necessary to prove factorization theorems, and show how they can be used also to resum certain classes of logarithmic contributions to all orders in perturbation theory.
Ubi maior ... : G. Sterman: An introduction to quantum field theory. P. Nason: http://castore.mib.infn.it/~nason/misc/QCD... M.L. Mangano: http://home.cern.ch/~mlm/talks/cern98... G. Sterman: hep-ph/9606312.

Parma, 05/09/2002

Outline
On mass divergences in perturbative QCD Mass divergences, low energies e long distances. Cancellation for physical observables, KLN theorem. Factorizable and IR/C safe observables in PQCD. An explicit example: Re+ e Denitions, cut diagrams, tree level. O(s ): the calculation. Approximations and observations. Other infrared nite quantities StermanWeinberg jets. Event shapes. Methods for all-order calculations Singularities of Feynman diagrams and trapped surfaces. Landau equations and Coleman-Norton picture. Infrared powercounting and niteness of Re+e . Factorization and resummation Multi-scale problems and large logarithms. From factorization to resummation. Examples: the quark form factor; the thrust.

Parma, 05/09/2002

Mass divergences: qualitative discussion

Fact: in quantum eld theory, two kinds of divergences are associated with the presence of massless particles. Infrared (IR). Emission of particles with vanishing four momentum (DB ); in gauge theories only; they are present also when matter particles are massive. Collinear (C). Emission of particles moving parallel to the emitter; they are present if all particles in the interaction vertex are massless. Example: a massless fermion emits a gauge boson in the nal state.
k

p+k

igu(p)/(k)ta

i(p + k / /) M, (p + k)2 + i

Singularities: 2p k = 2p0k0 (1 cos pk ) = 0 , k0 = 0 (IR); cos pk = 0 (C). Note: p0 = 0 not a problem (singularities are always integrable).
Parma, 05/09/2002 2

 Origin of mass singularities In covariant perturbation theory ( p conserved in every vertex; intermediate particles generally oshell): the emitting fermion is onshell, so that it can propagate indenitely. In time-ordered perturbation theory (all particles are on shell, energy however is not generally conserved in the interaction vertices): the IR/C emission vertex conserves energy, so it can be placed at arbitrary distance from the primary process. Mass divergences originate from physical processes happening at large distances. Available terapies . The sickness is serious. Because of mass divergences, the S matrix cannot be constructed in the Fock space of quarks and gluons Observe. Mass divergences are associated with the existence of experimentally indistinguishable, energy degenerate states. All physical detectors have nite resolution in energy and angle. KLN Theorem. Physically measurable quantities (such as transition probabilities, cross sections, after summing coherently over all physically indistinguishable states) are nite, mass divergences cancel.
Parma, 05/09/2002 3

 KLN theorem. Consider a theory dened by its hamiltonian H , and let D (E0) be the set of eigenstates of H characterized by energies E0 E E0 + , with = 0. Let P (i j) be the transition probability per unit volume and per unit time from eigenstate i to eigenstate j . Then the quantity
P (E0, )
i,jD (E0 )

P (i j)

is nite in the massless limit to all orders in perturbation theory Note. In an asimptotically free eld theory the limit m 0 and the high-energy limit formally coincide. Masses acquire a scale dependences such that m2(2) 0 as 2 . The situation in perturbative QCD Long distance (d > 1fm), or low energy (E < 1GeV) physics is not perturbatively calculable. The KLN theorem is not directly applicable when a sum over initial states is necessary (we have no control on the structure of hadronic initial states). Working at the perturbative, partonic level one identies suciently inclusive cross sections, such that long-distance eects are suppressed thanks to cancellations (IRsafe cross sections); long-distance eects can be isolated in universal factors, depending on the initial state but not on the hard process being studied (factorizable cross sections).
Parma, 05/09/2002 4

The strategy of perturbative QCD

All calculations are performed at partonic level, with an infrared regulator (e. g.: = 2 d/2 < 0), requiring the presence of at least one hard scale Q2. One computes
part = part Q , s (2 ), 2
2

m ( ) , 2

)!

IRsafe quantities are selected, having a nite limit when the IR regulator removed ( 0, m2 (2) 0).
part = part ! Q2 , s (2), {0, 0} + O 2 ( m2 2 !p , )! .

These partonic, inclusive quantities, admitting a perturbative expansion in powers of s (Q2) 1, are interpreted as estimates of the corresponding hadronic quantities, valid ` modulo O (QCD /Q)p corrections. With hadrons in the initial state, the goal is constructing factorizable quantities, such that
part = f m2 2 F ! part b Q2 2 , F 2 2 ! +O m2 2 F !p ! .

Factorization, proved at parton level, is transcribed at hadron level. Distribution functions f are measured, cross sections part are derived with a perturbative calculation. b
Parma, 05/09/2002 5

An explicit example: Re+e

The prototype of IRsafe cross sections is the total annihilation cross section for e+ e hadrons.
Z 1X 1 X dX tot (q 2 ) = |M(k1 + k2 X)|2 , 2 2q 4
X spin

normalized dividing out the total muon-pair production cross section

Re+ e tot e+ e hadrons tot (e+ e + )

In d = 4 2 , to leading order in ,
tot (q ) =
2

1 2 L (k1 , k2 )H (q ) , 2q 2

Current conservation implies qH = q H = 0, so that

H

e2 2 ` L (k1 , k2 ) = k1 k2 + k1 k2 k1 k2 g , q4 X 2 2 2 2 d d 0|J(0)|X X|J (0)|0 (2) (q pX ) . H (q ) = e qf

X

This leads to
g

2 H(q 2 ) . (q ) = q q q g
2

H (q ) = (3 2 ) q H(q ) ,

e2 2 1 2 tot (q ) = g H (q ) . 2q 4 3 2
2
Parma, 05/09/2002 6

A technical interlude: cut diagrams

A useful representation for |M|2 can be constructed in terms of cut diagrams. Pictorially
ui uf ui uf

vi ui

vf

vi ui

vf

vi

p vi

2+ (p2 ) p /

To the right of the cut all explicit is in the Feynman rules and all momentum components change sign. Consistency with spinor and color algebra is easily veried using
` 1 1 2 . . . i 5 . . . . . . n 2 = 2 n . . . . . . i 5 . . . 2 1 1 ,

as well as hermiticity of group generators, (ta)ij

Note that

= (ta )ji.

the rules apply for xed nal state momenta; the loop momentum integral for cut loops, if needed, becomes the phase space integral; for particles with spin = 0 the cut carries the sum over polarizations; cut fermion loops carry the expected minus sign.
Parma, 05/09/2002 7

Exercise: Re+e at treelevel

k 2 2 1 tot (q 2 ) = e 4 32 (g ) 2q q

2 H = e2 2 qf

dd k (2)

In he center of mass frame (k q)2 = q2 2 q2k0 ; use the two + distributions to perform k0 and |k| integrals; the trace evaluates to 4(1 )q2 . Summing over quark colors and avors one nds
X 2 H = 2(1 )e2 2 Nc qf
f

` +(k2 )+ ((kq)2 )Tr k (k q ) . / / / d2 p

q 4

!1

22 . (2)22

The d-dimensional solid angle is given by the classic formula

2dd/2 (d/2) d = . (d)

In d = 4 2 one nds then

(2 ) 2 H = 2 q (2 2 ) 42 q2

whence the famous result, for

X 2 42 tot = Nc qf 3q 2
f

0,

X 2 Nc qf ,
f

X 2 (0) R + = Nc qf . e e f

Parma, 05/09/2002

Radiative corrections
Enumerating graphs contributing to the one-loop correction is easy in terms of cut diagrams
h
H

i(1)

+ c.t.

Summing over positions of the cuts one obtains real and virtual diagrams in turn. We examine them separately.

Real emission
It is convenient to compute separately the transition probability and three-body space. Following the rules for cut diagrams one nds
h i (1,R) H = Z + (p )+(k )+ ((p + k q) ) H (2)2d3 ddpdd k
2 2 2

The transition probability depends on a single polar angle. Let u cos pk in the center of mass frame: one nds
+ 2pk(1 u) , + ((p + k q) ) = p0 s 2 s(p + k)
2

where p = |p|, k = |k|. One can integrate over the energies of p and k using the respective +. All angular integrations are trivial except the one on u.
Parma, 05/09/2002 9

Introduce dimensionless variables (quark and gluon energy fractions)

2k 2p z= , x= , s s and dene y = (1 u)/2. The result is h i 1 22 12 (1,R) H = 8 (2)52

12 Z 1 Z 1 s 12 dxx dzz 12 2 0 0 Z 1 i 1 1z h dy [y(1 y)] x H . 1 yz 1 yz 0

The transition probability can be computed from the Feynman rules.

i 0 h Tr (p + k) p (p k) p / / / / / / X 2 2 2 a 2 2 qf g Tr(ta t ) @ H = 2e (2p k)(2p k) f h i1 (p + k) p Tr (p + k) p / / / / / / A , + 2 (2p k)

and can be simplied using Tr(tata) = NcCF , Cliord algebra identities such as
p / p k / /

= = =

2(1 )p , / 4p k 2 p k , // 2q k / + 2 p k / , / /p / /q

p k / / /q

and with the identications pq = sx/2, pk = sxyz/2, kq = sz/2.

Parma, 05/09/2002 10

One can integrate out the quark energy fraction x, using the remaining . The expression simplies considerably and one gets
h
H

Z 1
0

dzdy (1 )(1 z)2 (1 yz)22 z 12 (1 y)1

12

i(1,R)

X 2 22 12 2 = 2Nc CF qf s (1 ) q (2)34
f

2 q2

!2

(1 z)

(1 yz)

22

12

[y(1 y)]

(1 yz)2 yz 2 (1 y)

+ y 1+ !# .

One recognizes the announced singularities

Infrared: z 12 , gives a pole in when the gluon energy z tends to 0. Collinear: y1 e (1 y)1 , singular when y 0 (gluon collinear to the quark), and y 1 (gluon collinear to the antiquark).
Note: mass singularities are regulated choosing < 0. The y and z integrations yield Euler B functions (typical of oneloop calculations in one-scale problems). Expanding around = 0 one gets the nal result for real emission, displaying the double infraredcollinear pole,
h H

i(1,R)

X 2 s 2 = Nc C F qf q
f

4 q2

!2 .

1 (2 2 )

2
2

19 + + O( ) 2
2

Parma, 05/09/2002

11

Virtual contribution
Purely virtual contributions to the production amplitude are given to all orders by the quark form factor.
p1 2 p 1 , p 2 ; , =

p2

The calculation greatly simplies by taking into account the general properties of the form factor.

For massles quarks, the form factor is expressed in terms of a single scalar function, multiplying the Dirac structure of the threee amplitude.
(p1 , p2 ; 2 , ) p1 , p2 |J (0)|0 q2 , s (2 ), 2 !

= ieqf u(p1 )v(p2 )

As a consequence, the transition probability is proportional to the treelevel result, with an overall factor given by 2 Re. The form factor is renormalization group invariant (it has a vanishing anomalous dimension), as a consequence of the conservation of the electromagnetic current.
+ ( , s ) s Q , s , 2
2

=0.

QCD does not violate the QED Ward identity. Z1 = Z .

Parma, 05/09/2002 12

 Reducible (1PR) graphs on each fermion line, including the respective counterterms, reconstruct the residue R of the quark propagator. Since, according to the reduction formulas, 1/2 every external line must be multiplied times R , it is necessary to include these graphs on one of the two lines only. In Feynman gauge and in dimensional regularisation all 1PR graphs with the loop on the external fermion line vanish because they are expressed by scale-less integrals (p2 = 0). i Note! This is false in general in axial gauges ( n pi); furthermore it depends on a cancellation of IR and UV eects ... At one loop these observations are summarized by the identity

One is left with only one graph to be computed, the vertex correction
(1)

2 p1 , p 2 ; , =

which can be written as

(1) 2 2 = eqf g CF

dd k u(p1 ) (p 1 k (p 2 + k v(p2 ) / /) / /) . k2 (p1 k)2 (p2 + k)2 (2)d

Parma, 05/09/2002

13

The steps to compute this diagram are standard. Summarizing

Sistematically use the massshell conditions (Dirac equation), u(p1 )p 1 = p 2 v(p2 ) = 0, then isolate integrals with dierent / / powers of k.
(1) 2 2 2 = eqf g CF u(p1 ) 2q I0 + 2( p 1 p 2 )I / /

The tensor integrals

I0 =

v(p2 ) .

Z Z Z

1 dd k , (2)d k2 (p1 k)2 (p2 + k)2 dd k k , (2)d k2 (p1 k)2 (p2 + k)2 k k dd k , (2)d k2 (p1 k)2 (p2 + k)2

can be computed with the usual Feynman parametrization. Note: only I0 can have IR divergences, and only I can have UV divergences. I may have collinear poles ...

It may be useful to decompose tensor integrals ` la Passarino a Veltman, in order to get directly the scalar form factor . The result is
I I

= =

p I1 + p I2 , 1 2 g

I3 + p1 p1 I4 + p2 p2 I5 + (p1 p2 + p1 p2 ) I6 .

Parma, 05/09/2002

14

 In terms of the scalar integrals I1, . . . , I6 one nds

(1) 2 2

= g CF 4(1 ) I3 2q (I0 + I2 I1 + (1 )I6 ) . !

The nal result for the form factor is

s (1) = CF 4 4 q 2
2

2 (1 )(1 + ) (1 2 )

2
2

+ 8 + O( ) .

When taking the real part one must use

(q 2 + i) = (q 2 ) ei .

Note! The sign of q2 is dictated by the Cutkosky rules. Because of the double pole, the factor exp(i ) must be expanded to second order in . It generates numerically important contributions.

Result
We observe that, as expected, the virtual contribution has the same IR-C poles as real emission. Summing the two, the poles cancel and one can take the limit 0, with the result
X 2 42 tot = Nc qf 3q 2
f

s 3 2 CF + O(s ) , 1+ 4

For SU (3), where CF = 4/3, the (classical) result is

X 2 (0) R + = Nc qf e e f s + O(2 ) 1+ s .

Parma, 05/09/2002

15

Soft approximation
The cancellation that was just exhibited is possible only because, in the IR and C limits the amplitude for the emission of a real gluon becomes proportional to the Born amplitude, just as was the case for the virtual diagram. This result can be made systematic introducing the soft approximation.
p, i p, i k, a p ,j

Aij =

k, a p ,j

# / /) (k)(p + k / / /) (p + k /(k) a a Aij = gtij u(p) v(p ) . 2p k 2p k

"

When the gluon is soft one can

neglect k in the numerator, and in the denition of p ; commute p and p in order to impose the mass-shell condition, / / p v(p ) = u(p)p = 0. / /
The result is
a Aij = gta ij soft " # p p A0 , pk p k

where A = u(p)v(p ) is the Born amplitude (whatever the 0 explicit form of the vertex ).
Parma, 05/09/2002 16

Remarks

The soft amplitude is gaugeinvariant (it vanishes if k). Soft gluon emission has universal characters. Long-wavelength gluons cannot analyze the short-distance properties of the emitter (spin, internal structure), they only detect the global color charge and the direction of motion These considerations generalize to multiple emission. Identical considerations apply to gluon emission from gluons.
Soft cross section Its easy to recover the singular part of the real emission cross section. The transition probability can be computed summing P = g , which is over colors and polarizations (using allowed in this case).
|Asoft | = g CF |A0 |
2 2 2

2p p . pk p k

To get the cross section one integrates over phase space

soft 2 q q g = g C F q q

d3 k 2p p . 2|k|(2)3 p k p k

In the center of mass frame (q = 0) and in the soft approximation the quark and the antiquark are still back to back. One then recovers the structure of IR and C singularities,
s soft q q g = q q C F Z 1 Z d|k| 2 d cos pk . |k| (1 cos pk )(1 + cos pk ) 1 0

Parma, 05/09/2002

17

Virtual diagrams The soft approximation can be applied to virtual diagrams as well, with some care.

When k pi k in denominators, as well as k in numerators (eikonal approximation). Note: the approximation is not uniformly valid, in some cases it becomes necessary to deform integration contours, or the approximation may break down. Using light-cone coordinates one can set
(p ) = 0, (p ) , 0 . + Note: For a generic four-vector, v = v , v , v , v = (v 0 v 3 )/ 2, v 2 = 2v + v |v|2 .

p q 2 , , one can neglect k2 with respect to

+ p = p , 0, 0 ,

Consider for example the integral I0, containing the virtual double pole. In the eikonal approximation and in d = 4
1 (eik) = I0 324 q 2 Z dk+ dkd2 k . (k + i )(k+ + i )(2k+ k |k |2 + i )

There are three integration regions giving rise to divergences. One can parametrize them introducing a scaling variable , according to q
k

q q2 , k

q 2 , , q q 2 2 , |k | q 2 , q k

IR ;

COLL .

Parma, 05/09/2002

18

Angular ordering
The soft approximation is of great practical relevance in perturbative QCD.

It displays the universal properties of soft color radiation (color transparency, angular ordering) It links perturbative and nonperturbative regimes (resummations, hadronization Monte Carlo)
The simplest example of angular ordering is the dierenzial cross section dq qg in a frame in which the decaying photon has a large momentum q. In that frame the quark and the antiquark are typically emitted forming a small angle (pp ), and one has
soft dq qg

1 cos pp s d|k| dk dq q CF d cos k |k | 2 (1 cos pk )(1 cos p k ) dq q CF s d|k| d 1 ` d cos k k Wq + Wq |k | 2 2 1 1 . (1 cos pk ) (1 cos p k )

where
Wq =

(1 cos pk )(1 cos p k )

1 cos pp

while Wq is found exchanging p p . The full angular distribution is positive denite, but singular for emissions parallel to either the quark or the antiquark. The partial distributions Wi are not positive denite, but they enjoy special properties.
Parma, 05/09/2002 19

Properties of Wq e Wq Wq is singular only when cos pk 1, while the opposite is true for Wq . The azimuthal average of Wq (with respect to the axis dened by p) vanishes if pk > pp .
Z 2 1 2 d Wq () = pp pk , 2 0 1 cos pk

as can be proven using

cos p k = cos pk cos pp + sin pk sin pp cos .

An identical equation applies to Wq . Azimuthal averages are positive denite and can be interpreted as probability distributions for the emission of soft gluons independently of the quark and of the antiquark. Thus: the distribution of soft radiation on average is given by a sum of uncorrelated contributions from the quark and from the antiquark, each vanishing outside the cones built rotating the direction of one of the fermions around that of the other one. Comments This property can be generalized to higher orders. Gluons which are radiated later are forced on average to the inside of the cones dened by previously emitted gluons and quarks. Perturbative evolution in the soft limit is local in phase space, so that color singlet parton clusters are likely to be formed inside collimated particle beams (jets).
Parma, 05/09/2002 20

StermanWeinberg jets
Momentum congurations responsible for singularities and needed for their cancellation are infrared and collinear, as expected. Thus it is not necessary to integrate real emission over the entire phase space in order to obtain a nite result, it is sucient to consider suciently inclusive observables, such that gluon emission be integrated over IR and C congurations. Prototype: two-jet cross section
Eout < s Ein > (1 ) s

Denition: an event is a two-jet event i two opposite cones with opening angle , such that all the energy, except at most a fraction , ows into the cones. At parton level:

All events are two-jet events at leading order. At O(s) two-jet events are those in which the gluon is IR (and emitted in any direction), or C (with any energy). All other events are three-jet events. Virtual contributions are two-jet events. Therefore the partonic two-jet cross section is nite.
Parma, 05/09/2002 21

More precisely:
(0) (0) At LO one nds simply 2j ( , ) = tot = Nc (1) (1) (1)

At NLO one nds only two- or three-jet events, so that

2j ( , ) = tot 3j ( , ) ,
(1) 3j is easily computed from the real emission matrix element with appropriate cuts on phase space. In d = 4

2 2 qf 4 . f 3q 2

h i X 2 s 2 Z 1 (1,R) H = 2Nc CF qf q dz 3j 2
f

Z 1 2 (1z 2 /2)
2

z(1 y) 1z dy + y(1 yz)2 yz(1 y)

It is easy to compute the dominant contributions as , 0. Combining with the leptonic tensor one nds
" # s 2 7 (1) (0) 3j ( , ) = tot CF 4 log() log(2 ) + 3 log() + . 3 4

Observe:

The total cross section is dominated by two-jet events at large q 2 (asymptotic freedom for jets ...). For increasing q2 the perturbative result remains reliable for narrower cones, jets are more collimated. It is possible to compute (and verify experimentally) the angular distribution of two-jet events d2j /d cos 1 + cos2 , as expected for spin 1/2 quarks.
Parma, 05/09/2002 22

Eventshape variables
The mechanism underlying the cancellation of IR and C divergences suggests a further generalization: study distributions of observables constructed with nal state momenta so as to assign equal weights to events diering by IR or C emissions. Given a nal state with m partons, let Em(p1, . . . , pm) be the observable. The corresponding distribution is dened by
Z d 1 X dLIPSm |Mm|2 (e Em (p1 , . . . , pm)) , = 2 de 2q m

and its moments (and in particular the average value) are

e
n

Z emax
emin

de e

n d

de

Note: These are weighted cross sections. At order m1 one must sum contributions with m + 1 partons s in the nal state, with one virtual m real partons, and so on.
Z Z (R) (1V ) (e) dm+1 + dm + . . . . = O m+1
s

IR-C cancellation is preserved if the observable takes the same value for those congurations diering only by the IR/C radiation.
lim Em+1 (p1 , . . . , pj , . . .) = Em (p1 , . . . , pj1 , pj+1 , . . .) , p 0 j lim E (p , . . . , pj , . . . , pk , . . .) = Em (p1 , . . . , pj + pk , . . .) . m+1 1 p p j k

Parma, 05/09/2002

23

Examples of eventshape variables

There is a variety of available IR/C safe event shapes. Thrust Pm
Clearly 0 < Tm 1, and Tm = 1 corresponds to two precisely collimated back to back particle beams.
i=1 Tm = maxn Pm

|pi n| . |pi | i=1

C parameter

Also in this case 0 Cm 1; two-jet eventi have C = 0. The denition can be expressed in terms of the eigenvalues of the space part of the energymomentum tensor of the nal state, C = 3(12 + 1 3 + 2 3 ).

` 2 m pi pj 3 X ` . Cm = 3 2 (pi q) pj q i,j=1

Jet masses m
(H)

H is one of the two hemispheres identied by the thrust axis.

1 = 2@ q

pi H

pi A

12

Observe Perturbative distributions are singular in the two-jet limit (logarithms of the form n log2n1 C ), but expectation values s are nite. Great phenomenological relevance (for example: determination of s , hadronization corrections). Jet algorithms and the related denitions of multi-jet events can be seen as particular event-shapes.
Parma, 05/09/2002 24

A comparison with QED

QED also has IR divergences, as well as collinear divergences in the massless limit. There are similarities and dierences.

In QED, consider for example the process e+ e +. (e+e + diverges starting at O(3). Therefore Born is not a good approximation for . This is not a problem. The true observable is tot() = P + + + n, ), and for tot (), which n (e e is nite, Born is a good approximation. IR divergences in QED can be explicitly resummed
tot () = n exp log " 2 q2 ! f (m2 , q 2 ) # .

so that lim0 tot() = 0. Interpretation: it is not possibile to produce only +; asimptotic states of QED are not isolated fermions. In QCD, considering e+e qq , the situation is almost analogous. (e+e qq diverges starting at O(2s ). Therefore Born is not a good approximation for = 0 (connement). On the other hand it is a good approximation for tot(e+ e hadrons), which is nite. Interpretation: it is not possible to produce only qq ; the asimptotic states of QCD are not quarks and gluons.
Parma, 05/09/2002 25

On singularities of Feynman diagrams

To study IR and C divergences to all orders one must characterize the generic singularity structure of Feynman diagrams. Let us begin with a simple example, I0 .

Example: the scalar form factor

Introduce Feynman parameters y1, y2, y3. Then
I0 = 2 Z (1 y1 y2 y3 ) . dyi y1 k2 + y2 (p k)2 + y3 (p + k)2 + i (2)d 0 i=1 dd k Z 1Y 3

Let D0 be the denominator. The only possible singularities of I0 must lie on surfaces where D0 = 0. Furthermore

The integrand is a function of the complex variables k, yi, with an analytic structure determined by the i prescription. The vanishing of D0 on one integration contour is not sucient to determine a singularity of the integral. The contour can be deformed. Only in two cases the singularity cannot be avoided by a deformation: either the contour is trapped between two poles, or one has end-point singularities, a pole migrates to the edge of an integration contour.
P1 O P2
Parma, 05/09/2002 26

 Singularities of I0 dk integrals cannot have end-point singularities (I0 is UV convergent). D0 however is quadratic in k, so that two poles can trap the contour if they coalesce,
D0 yi , k , p, p = 0 . k

dyi integrals can only have end-point singularities (in yi = 0), since D0 is linear in yi . Alternatively, D0 can be independent of yi on the surface D0 = 0, so that yi becomes useless for the deformation. Landau equations for I0 A necessary condition for a singularity of I0 is that all integration variables be trapped. This is expressed by the Landau equations
y1 k y2 (p k) + y3 (p + k) = 0 yi = 0 or and
2 li = 0 ,

where li is the momentum owing through the line with parameter yi. Solutions of Landau equations Its easy to nd the expected solutions.

k = 0 ; y2 /y1 = y3 /y1 = 0

IR C C

k = p ; y3 = 0 ; y1 = (1 )y2

k = p ; y2 = 0 ; y1 = (1 )y3

We recognize the expected IR and C singularities. Are there other solutions of Landau equations?
Parma, 05/09/2002 27

 ColemanNorton physical picture The search for solutions of Landau equations is simplied by the fact that they admit a simple physical representation. Observe that if line i in a loop is oshell it must be that yi = 0. Let x yili , for each on-shell line li . Then i
xi = xi vi ; vi =
0

li 1, 0 li

Interpretation: x describes classical propagation of a i massless particle with momentum li . The Landau equations for I0 can now be written as
X
i

(i) xi = 0 xi = 0

on shell o shell .

Interpretation: the solutions of Landau equations are given by reduced diagrams, where Oshell lines are contracted to points. Onshell lines describe physically admissible processes for the classical propagation of massless particles.

Parma, 05/09/2002

28

General case
Feynman parametrization Thanks to the identity
P P N a Z 1 N 1 N yi Y 1 a 1 i=1 i i=1 dyi y i , ai = QN i P PN a (ai ) 0 i=1 D i=1 i N y D i=1 i=1 i i=1 i i N Y

an arbitrary Feynman diagram G(pr ) can be written as

G(pi ) = linee 0 Y Z 1 dyi @1 0 X i yi A 1 loops Y Z dd k

where the denominator D is a sum of propagators

X ` 2 (p, k) m2 + i , D yi , kl , pr = yi li i linee

` N yi , kl , pr , l ` D yi , kl , pr N

and the momenta li are linear functions of pr and kl . Landau equations

X i ij x = 0 i x = 0 i on shell , j/i j , o shell .

ColemanNorton picture Solutions are again reduced diagrams (with oshell lines contracted to points), where all remaining loops can be interpreted as classically permitted processes. For any loop, it must be possible to associate to all its vertices coordinates xk , k = 1, . . . , M such that
x12 + . . . + xM 1 = 0 , xij xi xj .

Parma, 05/09/2002

29

Example: the two-point function

Consider an arbitrary 1PI diagram for the two-point function G(q 2 , m2 ), in a theory with only one species of particle with mass m2 .
G(q 2 , m2 ) = q q

Theorem: the only singularities of the diagram (and thus of G(q 2 , m2 )) are normal thresholds q 2 = n2 m2 , n = 1, 2 . . .. Proof:

Normal thresholds are solutions of Landau equations. p 2 In fact, when q > 0 one can choose q = q 2 , 0 . The ColemanNorton process is the creation of n > 1 particles at rest, which do not move, and interact until they are absorbed, for an arbitrarily long time. An example with n = 4 is
q q

No other reduced diagram satises ColemanNorton. If one produced particle has non-vanishing momentum the other ones must compensate moving in the opposite direction. Once separated they cannot meet again in free motion.
Parma, 05/09/2002 30

IR/C power counting

The Landau equations are only necessary conditions for the manifestation of IR/C divergences. If phase space has high enough dimension singularities are suppressed (e.g.: IR divergences in 3). 6 One must develop power counting techniques, similar to those employed in the UV, to determine the strength of the singularities.

Given a diagram, use the CN representation to identify a trapped surface S in the space {ki , yi }. For every S , identify among the {ki } intrinsic coordinates (movement in S ) and normal coordinates (distance from S ).
n 2 i 1 i

Example: for I0 , k p, k+ is intrinsic, {k, k} are normal. Introduce a scaling variable determine the relative weight (integration volume)/singularity. Set ni = ai ni and consider 0, ni nite. p p Example: for I0 , k p, k 2 q2 , |k| q2 . Construct the homogeneous integral for S , taking the dominant power of in every factor of the graph. Example: for I0, the homogeneous integral is the eikonal one.
Parma, 05/09/2002 31

 The degree of divergence is given by the power of associated 2 with the homogeneous integral. If, for every line, li (p, k) n m2 Ai f ( ), then i
nS = X
i

ai

X
i

Ai + nnum .

nS 0 signals a divergence, logarithmic when nS = 0.

Application: niteness of Re+e

The all-order niteness of tot(e+e hadrons) follows from its relationship with the correlator of two elettromagnetic currents, which in turn comes from unitarity,

2 Im pi Z C kj pi kj

generalizing T T = i(T T ). Dene then

(q) ie
2

2 2 = q q q g (q ) .

d4xeiqx 0|T J (x)J (0) |0

Unitarity gives
X 2 4 4 2 Im (q) = e 0|J(0)|n n|J (0)|0 (2) (q pn ) ,
n

Parma, 05/09/2002

32

tot(e e

h i e2 2 hadrons) = 2 Im (q ) . q

It is then sucient to prove the niteness of (q2 ). This follows from the Coleman-Norton representation.

In a frame in which q = one sees that there are no allowed classical processes with non-vanishing momenta such that the photon decays and then reconstitutes. Thus there are no trapped surfaces with non-vanishing momenta. The only trapped surfaces are those with all particles having vanishing momentum, and reduced diagram

q2, 0

S
H
These however are nite by power counting, as may be expected (all lines in H are oshell).

To see it note that fermion lines at zero momentum are less singular than gluon lines. The worst case is then if S contains only gluons. In that case, in d dimensions, with LS loops and gS gluons in S ,
nS = d LS 2gS = 2(1 )gS

which is positive in d > 2.

Parma, 05/09/2002 33

Application: the quark form factor

Considering (q2) one nds collinear divergences associated with observed quarks. The most general reduced diagram is
2 p 1 , p 2 ; , =
J H

Further simplications are possible

Gluons connecting S directly to H are suppressed (one more oshell propagator, plus a new soft propagator dominated by the new soft loop). Fermion lines connecting dierent subgraphs (except the necessary q e q) are suppressed. In an axial gauge, n A = 0, only the (anti)quark line connects the jet J to H . In fact the axial gauge gluon propagator is
ax G (k) =

1 k2 + i

n k + n k 2 k k g + n nk (n k)2

Contracting with k cancels the denominator

k G (k) =
ax

k n 2 , n nk (n k)2

so that the degree of IR/C singularity is reduced. These considerations suggest a factorization = J1 J2SH .
Parma, 05/09/2002 34

Diagrammatics of factorization
Identifying the leading integration regions in momentum space is only the rst step towards factorization. One must then exploit simplications in Feynman diagrams in the leading regions (soft/collinear approximations, Ward identities). organize all-order subtractions a tutti gli ordini so as to avoid double counting (dierent factor functions have operator denitions, in general non local). Microexample: again the one-loop form factor, collinear region k p1 , Feynman gauge.

Kinematics: {k , k }. Grammer-Yennie approximation in the numerator

1 u(p1 ) + (p1 k) (p2 + k)k+ v(p2 ) / / / / + k 1 + u(p1 ) (p1 k) (p2 + k) k v(p2 ) . / / / / / k u2

p1 =

p+ , 0, 0 1

+ , k = k , k , k , con k+

+ u(p1 ) (p1 k) (p2 + k) v(p2 ) u(p1 ) (p1 k) (p2 + k) v(p2 ) / / / / / / / /

Ward identity: k = (p 2 + k p 2 . One nds / / /) /

(coll) Z dd k

u(p1 ) u 2 (p 1 k v(p2 ) / / /) . k2 (p1 k)2 k u2

The gluon-antiquark coupling simplies, recognizing only charge and direction, it becomes eikonal.
Parma, 05/09/2002 35

 Eikonal lines Graphically, one may introduce eikonal Feynman rules

, a = igu ta ij j i j u i
ij = uk+i i

in terms of these rules the previous calculation reads

coll

More than one gluon Summing over all possible insertions of two or more collinear gluons, one nds sistematic cancellations, due to Ward identities. One uses then eikonal identities like
1 1 1 1 1 1 = + . k1 u k2 u (k1 + k2 ) u k1 u (k1 + k2 ) u k2 u

All collinear gluons couple to the same eikonal line. Typical result In an axial gauge the form factor factorizes
q2 2 ! = J1 (p1 n)2 2 n2 ! J2 (p2 n)2 2 n2 ! S (ui n) H q2 2 ! .

Parma, 05/09/2002

36

Multi-scale problems and large logarithms

For observables depending on a single hard scale all logarithms are resummed using the renormalization group
q2 2 , s ( ) 2 ! 2 = 1, s (q ) .

Most problems however have several hard scales. If for example 2 2 q1 q2 QCD , the reliability of perturbation theory is put 2 2 in jeopardy by terms like n logp (q1 /q2 ), with p n (single s logarithms) or p 2n (double logarithms). They all arise from IR/C dynamics. Some examples.

The Sudakov form factor.

q2 2 !

s =1 CF log2 4

q2 2

+ ... .

DIS. The two hard scales are Q2 = q2 and W 2 = (p + q)2 = Q2 (1 x)/x. There are single log(1/x), resummed by the BFKL equation, and double log(1 x), resummed ` la a Sudakov. DrellYan. The process is qq +(q2), the two hard scales 2 2 are s = (p1 + p2) and q . Double log(1 q2/s) are resummed a ` la Sudakov. The transverse momentum distribution in DrellYan. The two 2 2 scales are q2 and q, giving double log(q/q2 ).
Parma, 05/09/2002 37

In a massless theory one may choose 2 = q2 , and be left withIR/C poles. With masses one nds duble logarithms 2 2 2 log q /m .

Factorization and resummation

The deep connection between factorization and resummation, can already be seen from the renormalization group.
n Y 1/2 (n) (n) G0 (pi , , g0 ) = Zi (/, g()) GR (pi , , g()) , i=1

dG0

(n)

=0

d log GR d log

(n)

Solving this equation resums in exponential form the logarithmic dependence on . The AltarelliParisi equation similarly leds to the resummation of (single) logarithms of F . In terms of Mellin moments
e F2 Q2 N, 2 , s (Q2 ) m ! e =C N, 2 , s (Q2 ) F Q2 ! e f N, , s (Q2 ) m2 2 F ! ,

n X i=1

i (g()) .

Double logarithms are more dicult. Ordinary renormalization group is not sucient. Gauge invariance plays a key role. Or: use eective led theory (SCET).

e dF2 =0 dF

e d log f 2 = N s (Q ) . d log F

The quark form factor

Consider the factorization
q 2
2

= J1

(p1 n1 ) 2 n2 1

J2

(p2 n2 ) 2 n2 2

S (ui ni ) H

q , ni 2

Parma, 05/09/2002

38

The form factor is gauge invariant, so that

log log J1 log H log S =0 = . p1 n1 log(p1 n1 ) log(p1 n1 ) log(u1 n1 )

The two functions on the r.h.s. have dierent arguments. Then

log J 2 + GJ = KJ s ( ), log q q2 , s (2 ), 2 ! .

The K function contains all singularities in (H is nite as 0). The function G contains all q2 dependence. The whole form factor obeys an equation of identical form. Furthermore the form factor is renormalization group invariant, so that
dG dK = = K (s ()) , d log d log

with a nite anomalous dimension, independent of q2 . The equation for can be solved. Since is divergent, one needs to keep consistently the dependence on < 0 to all orders. s (2) becomes s (2, ) which implies (q2 = 0, < 0) = 0; then
Q2 , s (2 ), 2 2 2 ! 8 < 1 Z Q2 d2 h 2) = exp K , s ( + :2 0 2 ! , ! Z 2 1 d2 + 2 2 2 K

1,

, s (2 ),

2 , (2 ), 2 s

The exponentiation is non trivial, the exponent has only single poles in of the form n / n+1 . s
Parma, 05/09/2002 39

!!39 = 5 . ;

The thrust distribution

The thrust distribution is singular as T 1, behaving as n log2n1 (1 T )/(1 T ). These logs can be resummed with s similar methods. As T 1 the distribution can be factorized in a manner similar to . The jets J now enter the nal state, so they have a non-vanishing invariant mass m2 (1 T )q2 as T 1. J
Z 1 1 N d (N ) dT T = J1 0 0 dT q 2 (p1 n)2 , N 2 n2 2 ! J2 S H .

In an axial gauge, leading logarithms are in the jet functions J , which satisfy
log J = KJ log q q 2 , s ( ) 2 N
2

+ GJ

q 2 , s ( ) 2

One can then solve for J(N ) in terms of J(1), as

J q2 N 2 ! =J 2 3 ! Z 2 q2 1 q d2 5. exp4 log s (2 ) s (2 ) 2 2 J 2 q 2 /N J

Leading logarithms are determined by J = K + . . .. Neglecting running coupling eects one easily nds J exp(s log2 N ). Inverting the Mellin transform,
1 d s log(1 T ) s 2 = 2CF exp CF log (1 T ) . 0 dT 1T

Note: d/dT 0 as T 1 (Sudakov suppression).

Parma, 05/09/2002 40

The borders of the perturbative regime

Resummations test the limits of the perturbative theory. One gets in fact integrals of the form
fa (q ) =
2

Z q2
0

dk2 2 a (k ) s (k2 ) , 2 k

One sees esplicitly the nonconvergence of perturbation theory at high orders, consequence of the Landau pole. In fact
s (k ) =
2

Letting z log q2 /k2, one observes that The perturbative expansion in powers of s(q2 ) diverges.
2 2 a n=0 0 n=1

` . 1 + 0 s (q 2 ) log k2 /q 2

s (q 2 )

X n n+1 2 Z X az n n fa (q ) = (q ) 0 s (q ) dze z = cn s n! .

The integral is ambiguous, but the ambiguity is suppressed by powers of q2 . In fact

fa (q ) = (q ) s (q )
2 2 a 2

Z
0

eaz dz . 1 0 s (q 2 )z

The Landau pole on the integrazion contour induces an ambiguity of the same parametric size as the residue
2 fa (q ) exp
Parma, 05/09/2002

a 0 s (q 2 )

2 QCD q2

!a

41