QCD Amplitudes in Coordinate Space

Universiteit van Amsterdam
Master Thesis
QCD Amplitudes in Coordinate Space
Alexandre Salas Bernárdez
Supervised by Professor Eric Laenen
July 10, 2019

Contents
Introduction 1
Soft and collinear divergences in momentum-space 3

1.1 Landau Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Coleman-Norton picture . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Infrared power counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 All-order quark electromagnetic form factor . . . . . . . . . . . . . . . . . . . . . 10
1.4 Wilson Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 Factorization of the quark EM form factor . . . . . . . . . . . . . . . . . . . . . . 15
1.5.1 One loop soft and jet functions . . . . . . . . . . . . . . . . . . . . . . . . 16
Coordinate-space description 19
2.1 Analysis of singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Power Counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.1 Power counting for a single jet . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.2 Overall power counting for the vertex function . . . . . . . . . . . . . . . 26
2.3 Factorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.1 Hard-collinear aproximation . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.2 Soft-collinear approximation . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.3.3 Eikonal approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3.4 One-loop Jet function in coordinate-space . . . . . . . . . . . . . . . . . . 32
2.3.5 One-loop jet function with internal gluon emission in coordinate-space . . 35
2.4 Imaginary parts of Wilson line correlators . . . . . . . . . . . . . . . . . . . . . . 36
2.5 Largest time equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.6 Time Ordered Perturbation Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.6.1 Momentum-space TOPT . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.6.2 Unitarity and cut diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.6.3 Coordinate-space TOPT to all orders . . . . . . . . . . . . . . . . . . . . 46
2.6.4 Quark electromagnetic form factor in coordinate-space TOPT . . . . . . . 58
2.7 Lightcone-ordered perturbation theory . . . . . . . . . . . . . . . . . . . . . . . . 59
2.8 Cross Sections in coordinate-space . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.9 Unitarity and Causality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Conclusions 64
CONTENTS
Appendices
A Quadratic Casimirs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
A.1 Special Unitary Groups SU (N ) . . . . . . . . . . . . . . . . . . . . . . . i
A.2 Special Orthogonal Groups SO(N ) . . . . . . . . . . . . . . . . . . . . . ii
A.3 Simplectic Groups Sp(N ) . . . . . . . . . . . . . . . . . . . . . . . . . . ii
B Massless Scalar Propagator in coordinate-space . . . . . . . . . . . . . . . . . . . iv
C Massive propagator in 3+1 dimensions in coordinate-space . . . . . . . . . . . . . v
D TOPT propagators for fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
E Factors in the one-loop radiated jet function contribution . . . . . . . . . . . . . . vii
Introduction
Heisenberg’s uncertainty principle enunciates the impossibility to measure both the position and
the momentum of a particle with arbitrary accuracy at the same time. The heuristic way of under-
standing this is that, in quantum mechanics, the position and momentum probability amplitudes of a
particle are related via a Fourier transformation of the wavefunction in either one of the representa-
tions (momentum or coordinate-space), so that a very peaked distribution in coordinate-space will
produce a smeared distribution in momentum-space and viceversa. Hence, to describe a physical
problem, one can choose either one of the two representations and these will produce complemen-
tary interpretations and perspectives. In the momentum picture one is describing the degrees of
freedom of the theory as plane waves or superpositions of them. Here, to realize how these interact
between themselves is not an inmediate task. I have the feeling that, on the other hand, the idea of
point-like particles and how they interact or collide is much more transparent in the coordinate view-
point since it is much closer related with intuition. When explaining my research to other fellow
scientists, usually the coordinate picture is rather easy to communicate, whereas its complementary
description remains relatively obscure to imagine. Nonetheless, momentum-space description of
physical processes is ubiquitous in theoretical inquiries whilst, although more straight to grasp, the
position characterization has not received the attention it deserves. These reasons among others
made me choose to study the coordinate-space descriptions in high energy physics.
In the context of Quantum Field Theories (QFTs), the mathematical entities used to relate the
theoretical framework to physical processes are the so called Green’s functions. In principle, if
all the Green’s functions of a particular theory are known, then the theory is solved completely,
since these are introduced to solve the equations of motion for the fields. Usually, the computation
of Green’s functions is performed in momentum-space due to mathematical simplicity. Further-
more, scattering and collider experiments, which are one of the main methods and tools to unravel
the structure of elementary particles, are formulated and studied in momentum-space. Here, the
requirement of unitarity (forcing conservation of probabilities) naturally arises as the optical the-
orem and helps identifying resonances or bound states that the theory possesses. Moreover, huge
efforts were made at the end of last century to justify the use of perturbation theory in Quantum
Chromodynamics (QCD) consistently. These resulted in the so called factorization theorems for
amplitudes in momentum-space, which, by means of Resummation, ensured the convergence of
the perturbation series in the weak coupling regime (see [7] or [8]). On the other hand, analogue
results in coordinate-space only appeared in recent years [13], signaling the few attention that the
coordinate-space description has received in high-energy physics research.
1
2 INTRODUCTION
Due to this relatively scarce number of inquiries on the position-space description of quantum
field theories, in this thesis we will survey the general state of the art of coordinate-space QFTs. To
do so we review the main results of the factorization theorems in momentum-space, which provide
a very neat and simple way of reproducing the infrared (IR) singularities of the electromagnetic
form-factor of a colored particle. The fact that this form-factor presents singularities in different
regions of the internal momenta, i.e. the jet, soft and hard pinch surfaces, is easily shown to appear
at one-loop when looking at the unavoidable singularities of it. These are identified via the Landau
equations, and thanks to the Coleman-Norton picture the true solutions are found. Giving, at the
same time, a very clever interpretation of the Feynman parameters and a physical elucidation of the
occurrence of the singularities (these are associated with classical free particle trajectories!). Next,
via power-counting, the divergencies in the form factor are seen to be logarithmic at worst in four
spacetime dimensions. Finally, thanks to the definition of the Wilson lines, the jet, soft and hard
functions are defined, allowing to show that the IR divergencies of the form factor are all encoded
in the product of these simple functions (Factorization).
In the second and principal part of this thesis we turn to study the coordinate-space description
of QFTs, reviewing the main results of factorization. These come along in a similar way as in the
momentum-space picture, offering a complementary view of the IR singularities of amplitudes.
Subsequently we present original computation of the renormalized coordinate-space jet function at
one-loop, reproducing exactly the momentum-space answer after LSZ reduction, and of one con-
tribution to the one-loop radiated jet function in coordinate-space. Afterwards we carry on with a
bibliographic analysis of the imaginary parts of Wilson line correlators and the clear physical inter-
pretation of them. Following the intuition that new perspectives may arise in the coordinate-space
description we study the Largest Time equation. This equation is the graphical representation of
the causality requirement for QFTs. With these considerations at hand we review Time Ordered
Perturbation Theory (TOPT) in momentum-space, which offers a neat perspective of the different
time-orderings of events in amplitudes. Furthermore, here it emerges a clear interpretation of the
Feynman parameters as the ratio of the “flying time” of the internal-line particles and the on-shell
energies of these, in deep connection with the Coleman-Norton picture. We also present an original
proof of unitarity in momentum-space TOPT for non-gauged theories by means of mathematical
induction. Thereafter, we present an original algebraic proof of coordinate-space TOPT to all or-
ders. Which instead of a time-ordering aspect as in the momentum-space case, it brings about a
requirement on the space-configurations of internal vertices in order to produce singularities in the
amplitudes. We power count the possible divergencies for some diagrams and reproduce the known
results in the covariant momentum-space picture provided we force the vertices to comply with the
Coleman-Norton picture. We compare these results with the ones in lightcone coordinates.
For completeness, we review how the cancellation of IR divergencies in inclusive cross sections
comes along and we try to wrap up making a connection between causality and unitarity.
As one can see, this thesis resulted in a very wide and diverse collection of knowledge about
coordinate-space QFTs, even though there is a thin weaving and discursive element throughout the
work. I believe this thesis is formative and of benefit to anyone interested in the coordinate-space
description.
Soft and collinear divergences in
momentum-space
In perturbative QCD, the appearance of large logarithms (usually called Sudakov logarithms [7])
in the invariant mass of the system order by order makes perturbation theory unreliable for large
enough invariant mass. However, it is possible to “re-sum” these (i.e. reorganize the series expan-
sion in a clever way) and to obtain solutions in terms of quantities with perturbative expansions
with no large logarithms. To achieve this, it is necessary to first analyze which regions of loop
momenta give rise to divergencies in general amplitudes and to be able to separate the different
contributions by a method called Factorization. Once this is achieved, the RG evolution equations
for the factorized amplitudes will allow the resummation of large logarithms.
When performing integrals in scattering amplitudes one usually encounters singularities which
make the computation problematic. However, scattering amplitudes must be thought as complex
valued integrals and therefore, by Cauchy’s theorem, much information is contained in their sin-
gularity structure. Using the well know renormalization methods, it is possible to treat ultraviolet
(UV) divergences arising at high values of loop momenta. Instead, if we think of amplitudes in-
cluding massive particles, divergences regarded as threshold singularities emerge when the energy
is enough to generate internal particles on-shell. On the other hand, with massless particles we will
encounter the so called infrared divergences.
The latter divergences fall into two types. To see this take the scalar propagator
1 1
= (1.1)
(p + k)2 − iη 2p · k − iη
where k µ and pµ are both massless and on-shell. IR divergences emerge when, fixing pµ 6= 0,
the denominator vanishes: if k µ = 0 we will have a so called soft divergence and if p · k =
|~k||~p|(cosθ − 1) = 0 called collinear since it occurs when ~k is parallel to p~.
1.1 Landau Equations

To employ a more systematic approach, take a general Feynmann diagram G(p1 , ..., pn ) given
by
L Z I
Y dD ki Y N (ki , pr )
G(p1 , ..., pn ) = , (1.2)
i=1
(2π)D j=1 lj2 + m2j − iη
3
4 SOFT AND COLLINEAR DIVERGENCES IN MOMENTUM-SPACE
where the pr are the external momenta, ki the loop momenta, lj the line momenta which are linear
combinations of the loop and external momenta and mj the current mass for the particle propagating
in line j. L is the number of loops, N (ki , pr ) is an arbitrary polynomial in the momenta and I the
number of internal lines. Introducing the Feynman parametrization we obtain
L Z
Y I Z
Y 1 I
X I
X −I
D 2 2
G = (I − 1)! d ki N (ki , pr ) dαj δ(1 − αj ) αj (lj + mj − i) . (1.3)
i=1 j=1 0 j=1 j=1
P −I
I 2 2
We see that the graph G will be infrared divergent if D ≡ j=1 α j (lj + m j − i) = 0
unavoidably. To see what unavoidably means notice that D is a quadratic function in the momenta
and therefore it has no more than two zeroes in any momentum component liµ . If these zeroes are at
real values of liµ we will encounter singularities of the integrand in (1.3). However, if the contour of
integration can be deformed in the (αj , kiµ ) plane, then by virtue of Cauchy’s theorem the integral
will be well defined. There can be two cases where this will not happen:
• If the two zeroes in kiµ merge at the same point and “pinch” the contour of integration. This
condition amounts to
∂D(pr , ki , αj )
=0, (1.4)
∂kiµ

D=0
which by the definition of D means

X ∂lj2 (pr , ki ) X
αj = αj ljµ j,i = 0 (1.5)
j=1
∂kiµ i∈loop j
where the incidence matrix element j,i is +1 if the line momentum lj in the loop i flows in the
same direction as the loop momentum ki , −1 if in the opposite direction and zero otherwise.
• If the singularity is at the endpoints of the contour of integration then we will not be able to
apply Cauchy’s theorem if we change these points. Since kiµ ∈ R this type of singularities
corresponds to UV divergences and are taken care by renormalization. For αj integrations
these singularities are important when αj = 0 or, if D does not depend on αj (meaning that
lj2 = −m2j ). Either one of these two conditions on all the αs has to apply in order to have an
unavoidable singularity.
Hence, we can condense the necessary conditions for unavoidable divergences in the Landau Equa-
tions  X

 αj ljµ j,i = 0 ∀µ, i
j∈loop i
 α (l2 + m2 ) = 0

∀j . (1.6)
j j j
The proof that these conditions are necessary but not sufficient can be found in [2] (pg. 98). To ver-
ify that the solutions are indeed unavoidable singularities we have to resort to the method of power
counting and the Coleman-Norton picture, which will be described below. We can also note that
there are two classes of solutions: the solutions which constrain external momenta (i.e. there is a
singularity for a specific kinematical configuration) called threshold singularities and solutions for
a given set of masses and external momenta being fixed regarded as soft and collinear divergences.
1.1. LANDAU EQUATIONS 5
1.1.1 Coleman-Norton picture

Finding solutions to the Landau Equations (1.6) by hand is not an easy task, specially for higher-
order diagrams. However, owing to Coleman and Norton [3], there is a much easier and intuitive
procedure to solve the equations.
Recall that, in a solution to the Landau Equations, for off-shell lines we have αj = 0 whether
for an on-shell internal line we will have that αj 6= 0 and ∂D/∂kiµ = 0. If we identify now the
products αj lj for each on-shell line with a spacetime vector
∆xµj ≡ λαj ljµ (1.7)
and λαj = ∆x0j /lj0 as the Lorentz invariant ratio of the time component of ∆x0j to the energy lj0 .
Then we have that
∆xµj = ∆x0j vjµ (1.8)
with viµ = (1, ~lj /lj0 ). The introduction of the parameter λ has a subtle meaning. For collinear
divergences it can be set to unity but it is necessary in the soft case to keep the displacement ∆xµ
finite even when all components of lµ go to zero. Since soft gluons have almost infinite wavelength,
is natural to think that they will have a finite displacement as classical particles. Notice also that the
λ parameter has dimensions of length squared to keep the Feynman parameter dimensionless.
Summarizing, ∆xµj may be thought as a four-vector describing the free propagation of a classical
on-shell particle with momentum lj and the Landau Equations become
 X

 ∆xµj j,i = 0 if lj2 = −m2j
j∈loop i
∆xµj = 0 if lj2 6= −m2j .


 (1.9)
This means that the “pinch” condition for on-shell lines amounts to the condition that every loop
made out of these lines is a closed classical path and that off-shell lines are shrunk to a point (i.e.
they do not propagate). To illustrate this method we now turn to an example.
Quark Electromagnetic Form Factor
Let’s now work on an illustrative example of the application of the Coleman-Norton trick to find
possible “pinch” singularities. The only relevant diagram contributing to the QCD one loop cor-
rection to the quark-quark-photon vertex in D dimensions is (D = 4 − 2)
q
p2 + k k − p1 ≡ −ie Γµ(1) (p1 , p2 )
p2 −p1
k
dD k v̄(p2 )γ ν (p/2 + k/)γ µ (k/ − p/1 )γν u(p1 )

Z
Γµ(1) (p1 , p2 ) 2
= g CF µ 2
, (1.10)
(2π)D (k 2 − iη)((k − p1 )2 − iη)((k + p2 )2 − iη)
where here CF denotes the fundamental Casimir CF δij ≡ (T a )ik (Ta )kj (with T a being the generators
of the color gauge group). See appendix A for the derivation of these for different Lie groups.
We will regard the external quark lines as massless fields p21 = p22 = 0. If we think of these
as distinguishable final states, p1 and p2 cannot be collinear. This means that the photon must be
off-shell, q 2 = (−p1 − p2 )2 = 2p1 · p2 6= 0.
What reduced diagrams of the diagram (1.10) are allowed by the classical free propagation picture
between vertices? We can see all the reduced diagrams in Fig. 1.1. Thanks to the Coleman-Norton
Trick we can immediately rule out the diagrams (e)-(g) since a particle that leaves a vertex cannot
come back to the same vertex in free flight. Diagram (d) is also ruled out since the photon is
off-shell and the particles leaving the vertex have different directions and can never meet again.
Diagram (h) corresponds having all the internal particles off-shell, meaning that this solution to the
Landau Equations represents an UV divergence. In this way we are left only with diagrams (a)-(c)
as possible candidates to IR soft and collinear divergences.
sof t
(a) (b) (c) (d)
(e) (f) (g) (h)
Figure 1.1: Reduced diagrams for the one loop QCD correction to the qqγ vertex (1.10).
1.2 Infrared power counting

We have stated that Landau Equations provide necessary but not sufficient conditions for a diagram
to be singular. In fact, numerator factors or a small phase-space volume of the singularities can
make solutions non-singular. Is it therefore necessary to come up with a method to power-count
the infrared divergences of a general Feynman diagram.
In a given diagram, every solution to the Landau equations defines a surface S in the space (ki , αj )
that we will call a “pinch surface”. At each point on a pinch surface S, we look for the coordinates
that are intrinsic to the surface, i.e., when varied, the resulting point is still in S, and those that
parameterize how close we are to the singularity surface, which we call “normal”. In this way, it is
the vanishing of the normal variables which makes the integrand singular, and a singularity in the
integral will occur if the degree of divergence of the integrand is greater than the vanishing of the
normal variables’ integration measure.
1.2. INFRARED POWER COUNTING 7
The power counting analysis is done rescaling the normal variables with a parameter λ
norm
klnorm ≡ λal k 0 l , (1.11)
such that we approach the singularity when λ → 0. The coefficients ai with i = 1, ..., N norm are
introduced to include the possibility of different scalings for different normal variables (N norm is
the total number of normal coordinates).
We will proceed expanding the denominator of each of the i propagators to leading power Ai in
λ. The remaining integral, which we will regard as the homogeneous integral, will scale as λnS ,
where norm
NX X I
nS = al − Ai + anum . (1.12)
l=1 i=1
The first term in the RHS comes from the measure of the normal variables, the second from all the
leading powers in λ of the denominators and the third from numerator contributions to powers of
λ. In this way we can see that a sufficient condition for a diagram to be divergent is
nS ≤ 0 . (1.13)
In most cases the homogeneous integral does not generate new IR divergences and has the same
pinch surface S of the original integral. This does not hold for UV divergences where the homoge-
neous integral can develop unwanted new UV singularities [1]. In case the homogeneous integral
has a subset of normal variables which vanish faster than others, we have to put bounds verifying
the power counting for these pinch subsurfaces [4].
Quark Electromagnetic Form Factor
To illustrate the power counting technique let us come back to the example above. We already
identified the allowed solutions to Landau equations as the diagrams (a)-(c) in Fig. 1.1. Using the
Feynmann parametrization, the triangle diagram (1.10) reads
Z 1
dD k
Z
µ 2
Γ(1) (p1 , p2 ) = 2g CF µ dα1 dα2 dα3 δ(1 − α1 − α2 − α3 )
(2π)D 0
v̄(p2 )γ ν (p/2 + k/)γ µ (k/ − p/1 )γν u(p1 )
× (1.14)
(α1 k 2 + α2 (k − p1 )2 + α3 (p2 + k)2 − iη)3
and the Landau equations
µ µ

α1 k + α2 (k − p1 ) + α3 (k + p2 ) = 0

α = 0 or k 2 = 0

1

 α2 = 0 or (k − p1 )2 = 0

α3 = 0 or (k + p2 )2 = 0 .

Thanks to the Coleman-Norton trick we know that diagram (a) gives the soft solution
α2 α3
kµ = 0 , = 0, = 0.
α1 α1
Diagram (b) and (c) in Fig. 1.1 give instead two collinear solutions

 µ µ −(ξ1 − 1)
k = ξ1 p1 α3 = 0 α1 =

 α2 ξ1 ∈ R − {0}
ξ1
µ µ −ξ2 − 1
α3 ξ2 ∈ R − {0} .

k = ξ2 p2 α2 = 0 α1 =


ξ2
Due to the Coleman-Norton trick we also know that there are no other solutions to the Landau
Equations.
Now, before employing the power counting technique for these solutions, we need to identify the
normal coordinates for each case. Clearly for the soft pinch surface k µ = 0, the four k µ components
are the normal coordinates while there is no intrinsic coordinates. For the collinear power counting
we will need to introduce light-cone coordinates.
Recall that in light-cone coordinates (x± ≡ 2−1/2 (x0 ± x1 )) the scalar product of four-vectors
with components (x− , x− , x⊥ ) becomes
x · y = −x− y + − x+ y − + x⊥ · y⊥
and the line and volume element are
ds2 = −2dx+ dx− + dx2⊥

dD x = dx+ dx− dD−2 x⊥ = dx+ dx− d|x⊥ | |x⊥ |D−3 dΩD−2 . (1.15)
where dΩD−2 is the differential solid-angle in D − 2 dimensions. For simplicity we will choose a
reference frame where p1 = (p+ , 0, 0⊥ ) and p2 = (0, p− , 0⊥ ) so that
(k − p1 )2 = 2p+ k − − 2k + k − + k⊥
2
(k + p2 )2 = −2p− k + − 2k + k − + k⊥2
(1.16)
The collinear singularity where (k − p1 )2 → 0 is achieved when k − and k⊥ become 0, so these

variables are the normal ones while k + and ΩD−2 are intrinsic.
Let us power count for the soft case rescaling every loop momentum component k µ → λk µ so
that the parts in the integral scale as
dD k ∼ λD , k 2 ∼ λ2 , (k − p1 )2 ∼ λ , (k + p2 )2 ∼ λ,
v̄(p2 )γ ν (p/2 + k/)γ µ (k/ − p/1 )γν u(p1 ) ' −v̄(p2 )γ ν p/2 γ µ p/1 γν u(p1 ) = −4p1 · p2 v̄(p2 )γ µ u(p1 ) ∼ λ0 .
(1.17)
Where in the last line we approximated (k − p1 ) by −p1 since we are in the soft limit and we
used the Dirac equation v̄(p2 )p/2 = 0 and p/1 u(p1 ) = 0 to simplify v̄(p2 )γ ν p/2 = v̄(p2 ){γ ν , p/2 } =
2pν2 v̄(p2 ).
Keeping this leading behavior in λ when λ → 0 is regarded as the eikonal approximation and
amounts to
2
−p1 · p2 v̄(p2 )γ µ u(p1 )|k⊥ |D−3
Z
eik D−4 g CF 2 + −
I =λ µ dk dk d|k⊥ |dΩD−2 2
.
(2π)D (−2k + k − + k⊥ − iη)(−k + p− − iη)(k − p+ − iη)
(1.18)
1.2. INFRARED POWER COUNTING 9
We can see that I eik has a soft pinch surface at k µ = 0 and two collinear ones at k ± = k⊥ = 0
so that the divergent behaviour of the original integral is also present in the homogeneous eikonal
integral. We can then conclude that the superficial degree of IR divergence is
nS = 4 − D . (1.19)
For the collinear power counting we want to rescale the normal coordinates keeping their ratio
fixed so that, for the case k k p1 , the propagator in the first line of (1.16) is kept on shell. This
means
k − → λ2 k − , k⊥2
→ λ2 k⊥ 2
;. (1.20)
The different terms in the integral will therefore scale as
dk − ∼ λ2 , d|k⊥ ||k⊥ |D−3 ∼ λD−2 , k 2 ∼ λ2 , (k − p1 )2 ∼ λ2 , (k + p2 )2 ∼ λ0 . (1.21)
Note also that now since k k p1 we have (k/ − p/1 )γν u(p1 ) = 2(k − p1 )+ η+ν u(p) (remember
p1 = (p+ , 0, 0⊥ )). In this way we can change the numerator in (1.10) by
v̄(p2 )γ ν (p/2 + k/)γ µ (k/ − p/1 )γν u(p1 ) = 2v̄(p2 )γ − (p/2 + k/)γ µ (p1 − k)+ u(p1 ) , (1.22)
which scales as λ0 .
In this way, the collinear homogeneous integral for k k p1 becomes
2 Z
D−4 g CF 2
coll
I =λ µ dk + dk − d|k⊥ |dΩD−2
(2π)D
v̄(p2 )γ − (p/2 + k/)γ µ (p1 − k)+ u(p1 )|k⊥ |D−3
× . (1.23)
(−2k + k − + k⊥2
− iη)(−k + p− − + + − 2
2 − iη)(2k p1 − 2k k + k⊥ − iη)
We see that this integral has the collinear pinch surface k − = |k⊥ | = 0 and the and also the soft
one k − = k + = |k⊥ | = 0 so that the homogeneous integral gives the same divergent behavior as
the original integral. We can conclude that the superficial degree of divergence is
nS = 4 − D . (1.24)
For the other collinear singularity the power counting is analogous.
Summarizing, both collinear and soft divergences are logarithmically divergent for D = 4. We
also see that the soft and collinear pinch surfaces intersect giving rise to a soft-collinear pinch
surface k µ = 0 which corresponds to the collinear subsurface of the soft surface in (1.18). Hence,
we expect that this form factor will have a double pole in coming from the intersection of the two
collinear and the soft surface and a single pole for each of the individual pinch surfaces. This is
indeed the case since, writing
µ2
Γµ (p1 , p2 ) = Γ , αs (µ), v̄(p2 )γ µ u(p1 ) , (1.25)
2p1 · p2
the one-loop expression for the form factor Γ in the MS scheme is [8]
µ2 α 4πµ2 Γ2 (1 − )Γ(1 + ) h 2 3 i
s 0
Γ(1) , αs (µ), = CF − 2 − + O( ) . (1.26)
2p1 · p2 4π 2p1 · p2 Γ(1 − 2)
We will now apply the Landau equations and the power counting technique to classify the sources
of infrared divergence in the quark electromagnetic form factor to all orders in perturbation the-
ory.
1.3 All-order quark electromagnetic form factor

Thanks to the picture of reduced diagrams, we can study the IR divergences in the electromagnetic
form factor at higher orders without explicitly considering the Landau equations. It turns out that
the structure of singularities we have already studied at one-loop is also present at higher orders.
The possible reduced diagrams associated with pinch surfaces are all of the form shown in figure
1.2.
p1
J1
H S
J2
p2
Figure 1.2: General reduced diagram for a virtual photon decaying into massless quark and anti-
quark with no radiation.
This reduced diagram corresponds to physical processes in which the photon decays into two jets
J1 and J2 each with the total momenta, p1 and p2 , as the two final state particles. Between these
two jets the only interaction is via zero-momentum soft particles, labeled S. This is due to the fact
that since, once the jets are formed, they travel in different directions at the speed of light and hence
no finite momentum transfer can occur between the two. Higher order off-shell, short-distance
contributions coming from shrunk lines are encoded in the subdiagram H. The full derivation of
this characterization of general pinch surfaces is presented in [5]. In what follows we will only
introduce the main arguments which give the general reduced diagram in Fig. 1.2.
To begin we can justify that no lines connect the regions H and S since a particle coming from
the hard region will be highly off-shell and its propagator will give a strong suppression, making
the diagram subleading.
In this way we are left only with lines connecting all the other regions S, J1 and J2 . The behavior
of lines connecting H to the jets Ji will depend in the choice of gauge (see below). To analyze the
degree of divergence of the general graph S we only need to find the minimum value of the lower
bound of nS . This lower bound can be found [7] imposing the conditions:
• Ji and H are connected only through one fermion line and longitudinal polarized gluons.
• Ji and S are only connected by longitudinal polarized soft gluons.
• every vertex inside a jet is a three or four gluon vertex.
To complete this treatment is necessary to discuss the choice of gauge. In a physical gauge as the
axial gauge (n · A = 0, where n is a spacelike vector), the gluon propagator takes the form
−i nµ kν + nν kµ kµ kν
∆µν (k) = 2 gµν − + n2 , (1.27)
k − iη n·k (n · k)2
In this case, when contracted with the momentum of the gluon
inµ in2 kµ
k ν ∆µν (k) = − (1.28)
n · k (n · k)2
1.4. WILSON LINES 11
we see that the resulting expression has no pole at k 2 = 0 and therefore vanishes except when
k µ = 0. This means that reduced diagrams which give infrared divergences have only a single line
connecting each jet to H. In this case we find that
nS ≥ 0 , (1.29)
which means that the diagram in Fig. 1.2 has at worst a logarithmic divergence in the axial or any
other physical gauge [8].
On the other hand, in covariant gauges like the Feynman gauge, longitudinally polarized gluons
between H and Ji do propagate. Nevertheless, if one sums all the contributions coming from all
the different gluon attachments, they happen to cancel out via Ward identities. Therefore also in
Feynman gauge, after summing all the possible gluon insertions to the jet, the form factor pinch
surfaces have no lines other that the fermion line connecting Ji and H and nS ≥ 0. However this
is not true when treating with individual diagrams.
Summarizing, we have seen that all reduced diagrams associated with logarithmic IR divergences
in the electromagnetic quark form factor are of the form in Fig. 1.2, described by a two-jet structure
connected to one hard scattering function by one fermion lines and connected to the soft part by
soft longitudinally polarized gluons only.
1.4 Wilson Lines

We have seen that in the axial gauge, S, H and Ji are disconnected regions allowing the form factor
to be factorized in terms of sub-diagrams defined as functions of loop momenta restricted to the
leading region they belong to (soft or collinear restrictions). Nonetheless, we want to look for a
gauge independent framework where also the momenta in every region are not constrained to that
region.
The tool to solve this problem is the so called Wilson line. The Wilson line is is related with
the parallel transport in fibre bundles and the uniqueness of the so called “horizontal lift” [9] of a
spacetime curve γ(t) with t ∈ [t1 , t2 ] between the two points y = γ(t1 ) and z = γ(t2 )
t2
dγ µ
n Z o

Φ(t2 , t1 ) ≡ P exp − igµ dt Aµ (γ(t)) (1.30)
t1 dt
where the symbol P is the path ordering operator which orders the Aµ (γ(t)) so that the ones with
higher t stand to the left (remember that the A’s are matrices).
To see how this object can be used for factorization first let us show how soft and collinear
interactions reduce to effective vertices with the same Feynman rules as the ones related to Wilson
lines. We will illustrate this with the triangle diagrams already presented and then proceed to
generalize to higher orders.
Consider the absorption /emission of a soft gluon from a fermion coming from a a jet,
p−k p
J
k
S i(p/ − k/)
∼ ū(p)(gT a γ µ ) J({lJ })Sµ (k, {lS }) (1.31)
(p − k)2
where the lS and lJ are all the other line momentum variables inside the soft and jet surfaces re-
spectively. The jet and soft structures are encoded in the functions J({lJ }) and Sµ (k, {lS }).
Taking now the soft limit k → 0 and using the Dirac equation we find
a µ
i(p/ − k/) kµ →0

a p
µ
ū(p)(gT γ ) J({l J })Sµ (k, {lS }) ∼ ū(p)J({l J }) (−igT ) Sµ (k, {lS }) .
(p − k)2 p·k
(1.32)
We can see from this equation that the soft gluon only knows about the color and the jet direction.
This is why we will regard their interaction as eikonal. How is this related to the Wilson line?
Recall that internal lines in reduced diagrams are in free-flight so that their velocity v µ is constant
along their trajectory and we can hence write (1.30) for these particles as
n Z t2 o

Φv (t2 , t1 ) = P exp − igµ dtv µ Aµ (tv) . (1.33)
t1
Considering now Φ(∞, 0) and expressing Aµ (x) as the sum of its Fourier coefficients we get
Z ∞
dD k
n Z o
µ itk·v
Φv (∞, 0) = P exp − igµ dtv Ãµ (k)e . (1.34)
0 (2π)D
To carry out the integration in t we will Wick rotate k 0 → ik 0 so that the contribution from t = ∞
vanishes. In this way, after going back to real energy, we obtain
dD k µ eiλk·v λ=∞ o
n Z

Φv (∞, 0) = P exp − igµ v Ãµ (k)
(2π)D ik · v λ=0

idD kE v µ dD k v µ
Z Z
k0 →ik0 ik0 →k0
= exp gµ Ãµ (kE ) = exp gµ Ãµ (k) .
(2π)D kE · v (2π)D k · v
(1.35)
If we expand now in powers of g
dD k v µ g 2 µ2
Z D
d k1 dD k2 v µ v ν
Z

Φv (∞, 0) = 1 + gµ Ãµ (k) + Ãµ (k1 )Ãν (k2 ) + ...
(2π)D k · v 2 (2π)D (2π)D k1 · v k2 · v
(1.36)
we can see that the Wilson line Φv (∞, 0) reproduces order by order the eikonal Feynmann rule of
soft gluon emissions from the jet line in (1.32).
1.4. WILSON LINES 13
Consider now the triangle diagram for the quark one-loop form factor. In the soft limit, both
vertices in (1.10), become eikonal. For this reason, in the soft limit, we can replace the two quark
lines with two Wilson lines (Fig. 1.3a).
This procedure can be extended to the collinear case. Take k k p1 , the integrand in (1.10) has the
leading contribution in the normal variables k − and k⊥2
v̄(p2 )γ ν (p/2 + k/)γ µ (k/ − p/1 )γν u(p1 ) v̄(p2 )γ µ (k/ − p/1 )γν u(p1 ) pν2
' . (1.37)
k 2 (k − p1 )2 (k + p2 )2 k 2 (k − p1 )2 k · p2
Thus, in the collinear case we have that the interaction of the p2 antiquark line with the gluon
is eikonal, while with the p1 quark is not. Hence, the interaction of the collinear gluon can be
described as a Wilson line in the same direction as p2 (Fig. 1.3b).
(a) (b)
Figure 1.3: Separation of (a) soft and (b) collinear corrections from the hard part of the one-loop
quark form factor.
In summary, we realize that the one-loop soft and collinear interactions eikonalize and can there-
fore be described by Wilson lines.
One last feature of the Wilson line is that will also allow us to separate the hard part from the
jets. Gluons being absorbed by a jet Ji can either come from the fermion line connecting H to Ji or
directly from the hard region H. However, we can relate these two cases via Ward identities since
we talk about longitudinally polarized gluons. For simplicity consider a single, longitudinally po-
larized, soft gluon attached to the hard surface H shown in Figure 1.4 (note that the longitudinally
polarized soft gluon cannot resolve the internal structure of H). Thanks to a Ward identity this dia-
gram is opposite to the diagram representing the soft gluon insertion to the fermion line connecting
Ji and H. This diagram amounts to
p/i − k/
H({lH }) µ (k)γ µ u(p) , (1.38)
(pi − k)2
where µ (k) is the gluon polarization vector, pi the jet Ji total momentum and {lH } the internal
hard part line momenta. Taking the jet J1 in the + direction we know that only the − component
of the gluon polarization vector will couple to the jet since, by virtue of the Dirac equation and the
fact the k is soft, (1.38) becomes
p+1 β1 · (k)
H({lH })u(p) η µ+ µ (k) = −H({lH })u(p) . (1.39)
p1 · k β1 · k
Here β1 is an arbitrary 4-vector in the + direction.
H = − H = H
Figure 1.4: Factorization of the jet and hard regions in one gluon emission obtained in (1.39).
This last result represents the decoupling of the hard and jet regions and can be generalized to
higher orders. In this case the number of insertions in the Ward identities increases rapidly. This
issue is resolved by the eikonal identity, which represents the fact that at leading order in softness,
soft emissions factorize and are expressed in terms of independent emissions with eikonal vertices
[1] (see Fig. 1.5).
k1 kn ki
P Qn
... = i=1
all {ki } permutations
Figure 1.5: Eikonal identity for the emission of n soft gluons (all vertices are eikonal here).
Thanks to the identity, all soft emissions are eikonal and yield a result analogue to (1.39). There-
fore, if there are n soft gluon emissions from the hard part H to the jet, this process is encoded in
the nth order term in the expansion (1.36) of a Wilson line in the jet direction.
We can now safely assert that all the general pinch surfaces for the electromagnetic quark form
factor in Fig. 1.2 can be factorized in the hard contribution to the vertex, each jet connected by
soft gluons to a Wilson line in the direction of the other jet and the soft part connected through
longitudinally polarized soft gluons to two Wilson lines each in the direction of one of the jets (Fig.
1.6).
J1
H S
J2
Figure 1.6: Factorization in the pinch surfaces of the quark EM form factor.
1.5. FACTORIZATION OF THE QUARK EM FORM FACTOR 15
1.5 Factorization of the quark EM form factor

Let us now study how the covariant gauge factorization of the quark form factor in Fig. 1.6 takes
place. We will follow the treatment in [1] which was originally developed by Collins [7].
Thanks to the Wilson lines we will be able to construct composite operators which reproduce the
pinch surfaces of the form factor. Nevertheless, these composite operators generate new singulari-
ties not present in the original amplitude which will need to be somehow removed.
Each pinch region we have identified above has its momentum variables restricted to a region
in the phase space: in the soft and jet subdiagrams all the lines are soft and collinear respectively.
Hence, the functions we want to define that encode the singularities in each subdiagram will also
have to have their momenta restricted. This fact complicates a lot the factorization and this is why
we will seek a formula in which line momenta are not restricted.
Let us begin defining the soft function as
S(β1 · β2 , αs (µ2 ), ) ≡ h0| Φβ1 (∞, 0)Φβ2 (∞, 0) |0i , (1.40)
where β1 and β2 are proportional to the jet momenta p1 and p2 respectively, µ2 is a renormalization
scale and the dimensional regulator. We define now each jet leg i as
(p · n )2 h0| Φni (∞, 0)ψ(0) |pi i
i i 2
Ji 2 2
, αs (µ ), u(pi ) = , (1.41)
ni µ h0| Φni (∞, 0) |0i
where ψ is the fermion field operator and ni is the direction of the Wilson line. To avoid spurious
collinear singularities it is customary to choose n2i 6= 0. We introduce the denominator to cancel
graphs with eikonal self interactions.
Now we need to take into account the overlap of the soft and collinear regions to avoid double
counting. The overlapping can be seen as a jet function whose collinear gluons become soft or a
soft function whose gluons become collinear [1]. In either case, it can be described by the eikonal
jet function defined as
(β · n )2
i i 2
Ji , αs (µ ), ≡ h0| Φni (∞, 0)Φβi (∞, 0) |0i . (1.42)
n2i µ2
Q
Defining the hard function H as the result of dividing the form factor Γ by S i (Ji /Ji ), we can
finally write the formula for the factorized form factor

(pi ·ni )2 2
µ2 µ2 2 J
Y i n2i µ2 , αs (µ ),
Γ 2 , αs (µ2 ), = H 2 , αs (µ2 ), S(β1 · β2 , αs (µ2 ), ) . (1.43)
Q Q i=1 J
(βi ·ni )2 2
, α (µ ),
i 2 ni µ2 s
Here it is implied that, since the soft and jet functions present in (1.43) can generate new UV
divergences, UV counterterms are introduced to cancel UV divergences present in the soft and jet
functions. Note that these functions depend only on general properties of the external particles like
spin, charge or color, and collect all soft and collinear divergences. This dependence and some
issues concerning the so called cusp anomaly of is studied in detail in [10]. The factorization
formula (1.43) can be extended to more generic amplitudes. In cases with more legs, the color
dependence of the amplitude is non-trivial but remains treatable. However, the presence of the
so called Glauber gluons might spoil factorization. This is still an open topic of research (see for
example [11]-[12] for more recent and refined results).
1.5.1 One loop soft and jet functions

To convince ourselves that the factorization formula (1.43) captures the singularity structure of the
quark EM form factor in (1.26), let us compute the soft and jet functions at one loop.
Expanding the soft function as
∞
2
X αs CF n
S(β1 · β2 , αs (µ ), ) ≡ S (n) (1.44)
n=0
4π
we see immediately that, from the definition of the Wilson line, S (0) = 1. At one loop, we can see
that we recover the soft eikonal integral (1.18) from the definition of the soft function
α C Z ∞
s F (1) 2 2 µ ν
S = −gs µ β1 β2 dt1 dt2 h0| T {Aµ (t1 β1 )Aν (t2 β2 )} |0i
4π 0
Z ∞
dD k −ik·(t2 β2 −t1 β1 ) −ig µν
Z
2 2 µ ν
= −gs CF µ β1 β2 dt1 dt2 e
0 (2π)D k 2 − iη
∞
dD k −g µν
Z Z
= −igs2 CF µ2 β1µ β2ν D 2
dt1 eit1 (k·β1 +iη) dt2 eit2 (−k·β2 +iη)
(2π) k − iη 0
D
β1 · β2
Z
d k
= −igs2 CF µ2 D 2
. (1.45)
(2π) (k − iη)(k · β1 + iη)(−k · β2 + iη)
Here in the third

R ∞ equality we introduced iη regulators to be able to use the Schwinger parametriza-
isA
tion i/A = 0 dse with Im(A) > 0 in the last equality. Note that this integral is scaleless in
D = 4, therefore we expect it to vanish in dimensional regularization. This is indeed the case since,
after introducing Feynmann paramenters in (1.45) and carrying out the momentum integration, it
becomes
Z 1 Z 1
dD k 4β1 · β2
α C Z
s F (1) 2 2 dy y
S = 2iCF gs µ dx 3 i3
4π 0 (1 − y) (2π)D 2
h
0 y
k + 1−y (2xβ1 · k − 2(1 − x)β2 · k)
D
−2
Γ(3 − D2 ) 1
Z 1
2 2 (2β1 · β2 )
Z
2 D
−3
= −2CF gs µ D dx[x(1 − x)] 2 dyy D−5 (1 − y)3−D .
(4π) 2 0 0
(1.46)
The last expression happens to vanish in D = 4 since the last integral in y multiplied by 1 =
y + (1 − y) equals
Z 1
dyy D−5 (1 − y)3−D = Γ(D − 3)Γ(4 − D) + Γ(D − 4)Γ(5 − D) (1.47)
0
= Γ(D − 3)Γ(2) + Γ(−2)Γ(5 − D) (1.48)
and, recalling Γ(1 − 2) = −2Γ(−2), that lim→0 Γ(−2) = lim→0 −1/2 will cancel with
Γ(2) = 1/2 + O(0 ). This result represents an exact cancellation between an UV pole and an IR
pole. This result is valid at all orders [1] and means that, in an unrenormalized bare calculation,
the soft function is equal to the identity. Nonetheless, recall that composite operators need UV
1.5. FACTORIZATION OF THE QUARK EM FORM FACTOR 17
renormalization. The UV divergence comes from the point there the two Wilson lines originate
since here we get an expectation value of a composite operator (the two Φ-operators evaluated at
the same spacetime point).
To identify the UV divergence note that in the first line of (1.46), if y = 0 the momentum inte-
gration has the scaling
dD k
y = 0 → S (1) ∼ 6 , (1.49)
k
which correspond to the IR divergence. If y = 1 the momentum integral has the scaling
dD k
y = 1 → S (1) ∼ , (1.50)
k3
which is indeed the UV divergence. Thus, we can conclude that the UV divergence is opposite to the
contribution that was multiplied by (1 − y) in (1.47), which is Γ(4 − D)Γ(D − 3). The counterterm
must cancel this and hence behaves as −1/2 in the MS scheme. Adding the UV counterterm we
finally obtain
4πµ2 Γ(1 + )Γ2 (−) 1
S (1) = 2 , (1.51)
2β1 · β2 Γ(−2) 2
which expanded in gives
2 2 4πµ2
(1)
S = − 2 + log + O(0 ) . (1.52)
2β1 · β2
(1)
For the jet function at one loop we will have the contributions from a self energy correction Jp
(1)
of the quark line and a gluon exchange vertex correction JV between the Wilson and the fermion
line. All these contributions including the UV counterterms are plotted in Fig. 1.7. The eikonal
self interaction graphs do not appear due to the denominator in equation (1.41).
Figure 1.7: Diagrams contributing to the first-order jet function. The solid dots represent UV
counterterms.
The first diagram in Fig. 1.7 vanishes in dimensional regularization since it is scaleless (remem-
ber the external momentum is lightlike p2 = 0) and there is no available quantity with non-zero
mass dimension. This comes from a cancellation of UV and IR poles. Therefore, after introducing
(1)
the UV counterterm we will obtain that Jp = 1/. The vertex correction in the second diagram
in Fig. 1.7 amounts to
dD k
Z
(1) 2 1
JV = 8πiµ D 2 2
, (1.53)
(2π) (k − iη)(k − 2p · k − iη)(2n · k − iη)
which in the MS scheme and after substraction of the UV pole by adding the counterterm in the
fourth diagram in Fig. 1.7, results in [1],
h n2 4πµ2 1 1 5π 2 2 i
(1)
JV = − + + 2 + + . (1.54)
(2p · n)2 2 12
The one loop contribution to the eikonal jet function comes from diagrams like the second and
fourth in Fig. 1.7 but instead of a fermion and a Wilson line we will have two Wilson lines. This
contribution yields [1]
h1 1 n2 4πµ2 i
(1)
JV = − 2 + log . (1.55)
2(β · n)2
Finally gathering all first order corrections,
h 2 2 q2 3 i
(1) (1)
2(JV + Jp(1) − JV ) + S (1) = − 2 + log 2 − , (1.56)
µ
we see that the singular part of the one loop quark form factor (1.26) is correctly reproduced by the
universal functions encoding soft and collinear divergences. Similar calculations but with different
conventions have been carried out in [6] producing the same results for the case when n2 = 0. We
will study this condition in the same diagrams from the coordinate-space picture in the next chapter
below.
The presence of the so called Sudakov “logarithmcally enhanced” singularities like in (1.56) may
be thought as a problem since for high center of mass energies q the logarithm also gets large and the
convergence of the perturbation series seems to be spoiled. However, to give sense to perturbative
calculations Factorization comes to our aid by means of Resummation (see for example [7]). Here
it is stated that the renormalization group equations are used to absorb all logarithms of large mass
ratios into the integral of the anomalous momentum.
We will not go into the details of resummation of large logarithms in this thesis since the topic is
only tangentially relevant with our program of studying coordinate-space features of QFTs which
we turn to study now.
Coordinate-space description
In this chapter we will study the singularities of Green’s functions with massless lines in coordinate-
space, aiming at finding similar results as in the momentum-space picture. We will also compute
some QCD amplitudes in coordinate-space that are relevant for this thesis. We have already stressed
on the importance of Wilson lines to describe soft and collinear divergences, ordering fields in
coordinate-space. The expectation values of these objects gives rise to perturbative coordinate-
space integrals, we will study their analytic structure to understand in which physical scenarios they
they develop an imaginary part. Other tools like time-ordered perturbation (TOPT) theory will also
give us deeper insight on the meaning of the Feynman parameters and the Coleman-Norton picture,
and a causality interpretation of divergences as well. Unitarity in TOPT will be also proven in an
original way to this work by means of induction. TOPT in coordinate-space to all orders will be
also developed in this work and will bring about a very clear picture of how vertices must be places
in space in order to give divergencies.
2.1 Analysis of singularities

Extending the analysis of the previous chapter to coordinate-space is rather easy, applying the Lan-
dau equations will provide us with a correspondence with the momentum-space results and their
physical interpretations already presented. Following Erdogan’s work, [13], we will see how fac-
torization of amplitudes in coordinate-space is possible, enlightening the short and long-distance
behaviour of field theories.
A graph representing an amplitude in coordinate-space will amount to the integral,
µ
Y Z Y 1
I({xi }) = dD yk P 0 2 pj
× F ({xi }, {yk }) , (2.1)
vertices lines
[( k 0 jk 0 Xk ) + iη]
k j
where the {ykµ } are the position of internal vertices and {xµi } the positions of external points.
F ({xi }, {yk }) is a numerator factor containing all color factors, constant and numerator factors.
For each line j, the sum over {Xk0 } = {yk , xi } includes all vertices, internal and external, and
jk0 is an incidence matrix which takes values +1 and −1 when the line j ends or begins at ver-
tex k respectively and zero otherwise. The power of the denominator pj depends if the line is
bosonic, pj = 1 − (see Appendix A for derivation), or fermionic, where, pj = 2 − . This can be
seen using that the massless fermion propagator S(x) is related to the massless scalar propagator
∆(x) ≡ i∆F (x) by
Γ(2 − ) −/x
S(x) = ∂/∆(x) = 2−
. (2.2)
2π (x + iη)2−
2
19
20 COORDINATE-SPACE DESCRIPTION
As in the previous chapter we use Feynmann parametrization and identify the true singularities in
(2.1) using the Landau equations. For this case, remembering the discussion in the first chapter
above, the equations for a pinch surface read
 X


 αj zjµ kj = 0 and
lines j
 at vertex k
αj zj2 = 0 ,

 (2.3)
where zjµ denotes the argument of the denominator in propagator of line j. It is of course intended
that these equations must be satisfied together with the vanishing of the overall denominator ob-
tained after Feynman parametrization. To connect with the Coleman-Norton picture, identify the
product αj zjµ with a momentum vector lµ ≡ λαj zjµ and λαj ≡ lj0 /zj0 . Note that in the coordinate-
space picture the Feynman parameters are interpreted in the inverse way as in the momentum-space
picture. This means that the soft singularities will have αj = 0 and λ can be set to one. For a hard
singularity, i.e. zj0 = |~zj | → 0 (since αj 6= 0) the λ parameter helps α to remain finite and to
have a non-zero momentum even if z µ → 0. In this way we can see that each pinch singularity
corresponds to massless particles propagating freely on the lightcone between vertices with their
momenta satisfying momentum conservation at each vertex.
Take for example the relevant integral for a three point vertex at y with endpoints at x1 , x2 , x3 in
a scalar φ3 theory,
Z 3
D
Y 1
I({xi }) = d y 2 + iη]
. (2.4)
i=1
[(x i − y)
In this case ziµ = (xi − y)µ and the Landau equations read
α1 z1µ + α2 z2µ + α3 z3µ = 0 . (2.5)
To solve this equation, all the vectors zi must be either lightlike or have αj = 0. Also it cannot be
satisfied if all vectors have positive (negative) entries. Hence, at least one external point must have
x+ + +
i > y and another one xi < y , so that there is at least one incoming and one outgoing line. We
see then how this picture gives a natural time (or lighcone) ordering of the vertices. If all αj 6= 0
and all lines are on the lightcone, zj2 = 0, we have that zj · zi = 0. Meaning that all these lines are
parallel. Also, if one of the lines is off the lightcone, αi = 0, then the other two must be on the
lightcone and again parallel to each other. These cases correspond to the merging or splitting of
two particles occurring at point y µ with the ratios of their momenta determined by the ratios of the
αj . These results hold also for gauge theories and can be generalized to n-point vertices [13].
Let us study the vertex function in (1.10) but this time in coordinate-space. By translational
invariance we can set the position of the electromagnetic vertex at the origin. All the coordinate
placements are showed in Fig. 2.1.
The one loop contribution to the vertex amounts to
Z
µ 2 2 1 1
Γ(1) (x1 , x2 ) = CF gs µ dD y1 dD y2 × /
∂ x2 −y2 ×
[(y2 − y1 )2 + iη]1− [(x2 − y2 )2 + iη]1−

ν / 1
µ / 1
ν / 1
× γ ∂ y2 2 γ ∂ y1 2 γ ∂ y1 −x1 . (2.6)
[y2 + iη]1− [y1 + iη]1− [(y1 − x1 )2 + iη]1−
2.1. ANALYSIS OF SINGULARITIES 21
• x2
y2
•
0
• ≡ −ie Γµ(1) (x1 , x2 )
µ
γ
•
y1
• x1
Figure 2.1: One loop contribution to the quark EM vertex function in equation (2.6).
We can use Feynman parametrization before or after the action of the derivatives, in either case the
pinch singularities originate from the two poles of a single propagator or poles from different ones.
The denominator after the parametrization and before the derivatives will be
D({xi }, {yk }; {αj }) = α1 (y1 − x1 )2 + α2 y1 + α3 y22 + α4 (x2 − y2 )2 − α5 (y2 − y1 )2 + iη . (2.7)
Using the coordinate-space analog of (1.4), we get the Landau conditions for pinches in the inte-
gration over the positions of the internal vertices y1 and y2
α1 (y1 − x1 )µ + α2 y1µ − α5 (y2 − y1 )µ = 0 (2.8)
α4 (x2 − y2 )µ + α3 y2µ + α5 (y2 − y1 )µ = 0 (2.9)
The simplest solution to these equations is taking α5 = 0, yielding
α1
y1µ = xµ1 (2.10)
α1 + α2
µ α4
y2 = xµ . (2.11)
α3 + α4 2
In other words, y1µ becomes parallel to xµ1 and y2µ to xµ2 and the condition D = 0 imposes x21 =
x22 = 0. Remembering the identification of the momentum of a line as pµ5 = λα5 (y2 − y1 )µ we see
that this solution corresponds to the exchange of a zero momentum gluon between the two fermion
lines. It propagates a finite invariant distance not changing the direction of the outgoing fermions,
i.e., a soft gluon.
In the case that none of the αj is equal to zero we can see that a general solution to the Landau
equations (2.10) and (2.11) is such that y1 and y2 are given as linear combinations of x1 and x2
[13]. Under these conditions, to satisfy D = 0, x2i = 0 and also x1 · x2 = 0, so that x1 and x2
become lightlike separated. This is also the case if α2 = α3 = 0. These are not solutions we want to
describe since it means that x1 is parallel to x2 and what we want is real physical scattering process
where the two final state particles travel in different directions.
The first UV-type singularities we find, i.e. processes happening at a zero distance, is taking
y1µ = 0 or y2µ = 0. These give the previously identified jet solutions that by equations (2.11) and
(2.10) are, respectively,
α4
y2µ = xµ if y1µ = 0 (2.12)
α3 + α4 + α5 2
α1
y1µ = xµ if y2µ = 0 (2.13)
α1 + α2 + α5 1
with x21 = x22 = 0 to satisfy the vanishing of the denominator D. These are clearly the collinear
pinch surfaces identified in the previous chapter. In [13] it is shown that the only remaining solution
satisfying that x1 and x2 are not lightlike separated is the hard solution identified in momentum-
space.
We see therefore that the pinch surfaces identified here have the same structure as in momentum-
space, with two jets, a hard function and a soft surface (as in Fig. 1.2). This is indeed the case and
is proven to all orders in Erdogan’s work [13]. It is also proven there by power counting that these
pinch singularities of massless gauge theories have at worst a logarithmic degree of divergence, just
as in the momentum-space description. Once we have observed such analogies between the two
pictures is natural to think that factorization will also occur in coordinate-space.
2.2 Power Counting

We will apply, following [13], the power counting technique to identify and study the behaviour of
the pinch surfaces of vertex functions in coordinate-space. We see that all the singularities studied
in the previous section have the external points x1 and x2 on the lightcone. Since the external
propagators are not truncated in coordinate-space, even the zeroth order results are singular; for
example the fermionic vertex diverges as 1/(x21 x22 )2 [13]. This is why we consider all the external
lines on the lightcone, and study the degree of divergence of the vertex function with respect to the
lowest order results. In other words, we look for the divergences in the residues of the lightcone
poles, like it is usually done in momentum-space with the residues of single particle poles in external
momenta for renormalization.
We will do the power counting by comparing the normal variables’ phase space volume element
with that of the integrand. To do so we seek for the homogeneous integrals keeping the terms of
lowest order in the normal variables as before, and then compare the power of normal variables in
these integrals with the normal volume element. This will provide us with bounds to the original
integrals.
Denote z1 , ..., zn and w1 , ..., wm as the normal and intrinsic variables for a pinch surface S of an
integral I. We rescale the normal variables with λ and take the leading behaviour when λ → 0,
while the intrinsic coordinates remain finite. This scale λ measures the distance of the contour of
integration H from the pinch surface S. Bounding the normal phase space with an n dimensional
sphere with radius λ means introducing an integral in λ2 and a delta function having the sum of the
squares of the absolute values of the normal variables in its argument. In this way, the homogeneous
integral I¯ near the pinch surface S will be
Z Z Y n Xn Z Ym
¯
I ∼ dλ 2
dzi δ(λ −2 2
|zi | ) dwj f¯({zi }, {wj }) , (2.14)
H i=1 i0 j=1
where the homogeneous integrand f¯({zi }, {wj }) is obtained by keeping terms of lowest order in λ
in each factor of the original integrand such that
f ({zi }, {wj }) = λ−dH f¯({zi0 }, {wj })(1 + O(λ)), (2.15)
for each normal variable zi = λ zi0 and dH being the degree of homogeneity of f¯({zi0 }, {wj }). That
is to say, dH equals the sum of the lowest powers in λ coming from denominator factors minus that
2.2. POWER COUNTING 23
in the numerator factors in f ({zi }, {wj }). Now we scale out λ from each factor to count the overall
power in the scale, to find the dominant behaviour as λ → 0,
Z n
Z Y n
X m
Z Y
I¯ ∼ γ−1
dλλ dzi0 δ(1 − |zi0 |2 ) dwj f¯({zi0 }, {wj }) . (2.16)
H i=1 i0 j=1
This overall degree of divergence γ is given by
γ = n − dH . (2.17)
For γ = 0 the divergence is logarithmic, while for γ < 0 we will encounter power divergences.
Here we were considering that all normal variables scale as λ. In case a set of normal coordinates
vanished faster than others, say λ2 , this would only increase the power of λ in the volume element
of the normal space, giving a non-leading contribution. This is however not the case if taking the
homogeneous integrals generates new pinches in the sense that poles that were originally separated
by non-leading terms, after the dropping of these, become coalescent enhancing the divergence.
Nonetheless we will see that only pinches of the hard-jet-soft type arise in the homogeneous inte-
grals. Also, if every normal variable in a subregion vanishes faster that those of other regions, the
power counting will still be the same for each subregion. In this way, we can choose a single scale
for all normal variables for the power counting of a vertex function near a pinch surface.
Let us study now the one loop quark EM form factor in (2.6). Since we are dealing with lightlike
external lines we will use lightcone coordinates for simplicity in the Landau equations, choosing
xµ2 = X2 δ +µ and xµ1 = X1 δ −µ .
For a particular pinch surface (i.e. a solution to the Landau Equations) we have to identify the
intrinsic and normal coordinates. Take the soft solution identified in the previous section (equations
(2.10) and (2.11)) where y2+ and y1− are the intrinsic variables, while y2− , y1+ ,y2 2⊥ and y1 2⊥ are the
normal ones vanishing at the pinch surface. For the hard solution, all components of y1µ and y2µ
are vanishing and therefore all of them are normal coordinates. Finally, for the collinear solutions
(2.12) and (2.13) y2+ and y1− are, respectively, the intrinsic variables while all the other components
are normal. We define the jet function to be the set of collinear gluons together with the external
lines. In the first collinear case y2+ goes from zero up to X2 , for y2+ > X2 the Landau equations are
not solved and it does not correspond to a physical process where all lines move forward in time.
In general, the limits of integration over the intrinsic variables in a given jet pinch surface are set
by the time ordering of the vertices [13].
In a general homogeneous integral, the denominators of jet lines are linear, the denominators of
hard lines quadratic, and the denominators of soft lines are zeroth order since these denominators do
not have to vanish (remember the example above where α5 = 0 and therefore the soft denominator
(y1 − y2 )2 did not have to vanish).
The approximation made to obtain the homogeneous integrals amounts in dropping terms that
are higher order in the normal variables. Such terms only pertain to lines connecting different
subdiagrams, like lines connecting the jets to the hard and soft functions or those connecting the hard
and soft function. We need to ensure that no new pinches are introduced after dropping these, i.e.
that the Landau equations obtained have the same solutions leading to the same physical pictures.
These new pinches can only arise in lines connecting the jet to the hard part since we know that
soft lines are not pinched. To address this issue consider the integral in a jet function,
Z
1 1
I = d4 y , (2.18)
(x − y) + iη (y − z)2 + iη
2
which connects a hard vertex z µ to the endpoint of a jet xµ through an internal vertex y µ , where we
omit derivatives at each vertex for simplicity. The Landau equations read
− α1 (x − y)µ + α2 (y − z)µ = 0 (2.19)
along with (x−y)2 = (y−z)2 = 0. The only solution here is that all three vertices are aligned along
the jet direction, say the plus direction. The equations allow z µ to be hard, i.e. z + can also vanish
in (2.19) in the same rate as all its other components. Approximating (2.18) by the homogeneous
integrand with (y − z)2 ∼ y 2 + 2y + z − , gives the same pinch in the y − integral as in (2.19) with
z + = 0. The integrals over transverse components can only be pinched at y⊥ 2
= x2⊥ = 0. This
2
pinches are also present in the original integral and, after the change of variables y1 , y2 → y⊥ ,φ
they become endpoint singularities in the integrals over the new variables. This is something that
is shared in general by all the pinches in homogeneous integrals which correspond to pinches of the
original integral but with some variables moved to their endpoints.
This approximation we made happens to fail if z + becomes comparable to y + and x+ , or if y +
goes to zero like z + . These are different solutions to Landau equations corresponding to different
pinch surfaces. However, we will see that the power counting will be the same in either case,
whether z µ is taken as part of the jet or included in the hard part. For a pinch surface we will
identify the normal variables, group each vertex in a certain subdiagram depending on the size of
the components of its position, and power count the divergence of that particular pinch surface. It
is possible to show that the effect of removing a vertex from a region and its inclusion in another
subdiagram gives the same overall degree of divergence.
z• •
• x
y
Figure 2.2: The pinch surface for the single jet is reached when y lies on the lightcone line between
the hard point z and the endpoint x.
2.2.1 Power counting for a single jet

Let us now consider a “dressed” (i.e. with several self energy corrections) ultrarelativistic fermion
moving in the plus direction (see Fig. 2.2 with z µ = 0), so that all the plus components of the inter-
nal jet vertices are intrinsic coordinates, while their minus coordinates and, by azimuthal symmetry
of the jet, the squares of transverse positions are the normal variables. With these coordinates, all
normal variables appear linearly in the jet line denominators.
In D = 4 − 2 dimensions, for every integration over the positions of the three- and four-point
vertices inside the jet we have to add +2(1 − ) to the overall degree of divergence of the jet γJ ,
+(1 − ) from each of the normal variables. Every internal gluon propagator gives −(1 − )
contribution while internal fermion lines give −(2 − ). Also remember that we wanted to study
the lowest order result after extracting the divergence from lightlike external fermion lines, so we
add +(2 − ) to γJ to cancel the lightcone divergence of the zeroth order diagram.
Now we need to take into account numerator factors, keeping only terms of lowest order in normal
variables. From each fermion-gluon vertex at a point ynµ we get the numerator factor,
(y/n+1 − y/n )γ µ (y/n − y/n−1 ) = 2(yn+1 − yn+1 )µ (y/n − y/n−1 ) − γ µ (y/n+1 − y/n )γ µ (y/n − y/n−1 ) .
(2.20)
The first term here must form an invariant with some other vector but it is still unsuppressed. The
second term proportional to γ µ is either zero by (γ ± )2 = 0, or vanishes as the transverse coordinates
of the vertices, which are of order λ1/2 .
µ
There are also numerator factors coming from three-gluon vertex functions. A vertex at zm com-
bines with gluon propagators ∆(z − yi ) ending at points yi to give, up to overall factors and taking
the pure scalar part of the propagators,
v3g (zm , {y}) = ijk g µi µj ∆(zm − yi )∆(zm − yj )∂zµmk ∆(zm − yk ) . (2.21)
The derivatives acting on gluon lines give vectors in the numerator, while they increase the power
of the denominator by one. These vectors must again form invariants, either between them or with
Dirac gamma matrices. If zi denotes the position of the ith three-gluon vertex and yj the position of
the jth fermion-gluon vertex, then the numerator will have invariants in the form, z/i , y/j , zi · zi0 and,
referring to (2.20), also yj · zi . All these invariants are linear in the normal variables. In this way we
can see that each fermion-gluon vertex gives a suppression of λ1/2 , while every pair of three-gluon
vertices produces a suppression of λ, and if a three-gluon vertex is contracted with a Dirac matrix
it gives a suppression of λ1/2 at least. In summary, the contribution from the numerators γnJ has a
lower bound of
1 1
γnJ ≥ (V3f + V3g ) ≡ V3 , (2.22)
2 2
where V3f is the number of fermion-gluon vertices and V3g the number of three-gluon vertices.
Recovering all the contributions considered we arrive at
1
γJ ≥ 2(V3 + V4 ) + 2 − Ng − 2Nf − V3g + V3 + O() . (2.23)
2
Here, V3 (V4 ) is the number of three-point (four-point) vertices, while Nf (Ng ) is the number of
internal fermion (gluon) lines. Take the graphical identity
X
2N = E + iVi , (2.24)
i=3,4
which can be obtained thinking the E external lines as one-vertices and that all lines are connected
to vertices so that counting the lines entering on every vertex gives two times the total number of
lines, and use it into (2.23),noting that the jet line has E = 2, to get
γJ ≥ 1 + V3f − Nf + O() . (2.25)

Now, since at each fermion-gluon vertex one fermion line enters and one exits, the number of
fermion lines in the jet is equal to one plus the number of fermion-gluon vertices in the jet, we
finally find
1
γJ ≥ O() = ( V3 + V4 ) . (2.26)
2
Thus, a fermion jet with the topology of a self-energy diagram like in Fig. 2.2 has at worst a
logarithmic degree of divergence in four dimensions. This is also the case for scalar jets [13]. This
shows that the collinear singularities are regulated with > 0 in coordinate-space.
The power counting described above holds in the presence of self-energies inside the jet as well.
For example, inserting a fermion loop in a gluon line does not change γJ , since the contribution
from the two normal variables of the two new vertices gets cancelled with the contribution from the
two new fermion propagators, while the contribution from the new gluon line gets cancelled with
the numerator factor of the two fermion propagators y/(−y/) = −y 2 ∼ O(λ). Similar cancellation
occurs in the cases with a ghost or gluon loop being inserted. A different power counting is needed
when these self energies shrink to a point like taking y + → 0 in the example above. However, these
divergences are removed by UV-counterterms [13].
2.2.2 Overall power counting for the vertex function
x•2
J(+)
| H S
0
J(−)
x•1
Figure 2.3: Representation of the hard (H), the soft (S), and jet J(±) pinch surfaces for the quark
electromagnetic vertex at all orders before power counting. All the hard vertices are concentrated
in the origin, while the jets travel in the plus and minus directions. The dotted lines represent all
the possible lines (fermionic or gluonic) connecting different pinch surfaces.
Now we want to study the overall behaviour of the vertex function with two jets, one soft function
and the hard function. Happens that the soft function is independent of normal variables (recall
that all variables are intrinsic in soft vertices) and that by dimensional counting it is finite for fixed
H H
external points. We define J(±) and J(±) to denote the number of gluon and fermion, respectively,
g f
lines connecting the hard part to the two jets in the plus and minus direction. We also define
H H H
Jg,f ≡ J(+) + J(−) , and J H ≡ JgH + JfH as the total number of lines connecting the jets to the
g,f g,f
J J
hard part. Similarly we define S(±) and S(±) as the number of gluon or fermion lines connecting
g f
J J J
the jets to the soft region, also Sg,f ≡ S(+) + S(−) and S J ≡ SgJ + SfJ .
g,f g,f
Recall now that all components of the vertices in the hard function vanish together, and that hard
gluon (fermion) lines are quadratic (cubic) in normal variables. Since we already now the power
counting for a single jet we see that the overall degree of divergence of the vertex, γΓ , is
γΓ ≥ 4(V3H + V4H ) − 2NgH − 3NfH − V3gH +

X h J(i) J(i) J(i) J(i) J(i)
i
J(i)
+ 2(V3 + V4 ) + 2 − Ng − 2Nf − V3g + n + O() , (2.27)
i=+,−
where nJ(±) are the jet numerator factors and the terms with H correspond to contributions from
the hard surface. In this surface, every three gluon vertex increases the power of a gluon line
denominator while giving a vector of first order in the denominator. Contributing hence −1 to the
power counting, giving the −V3gH in (2.27).
The numerator power counting for the jets is different from the one outlined in the previous
section. Now numerator vectors in one jet can attach to soft or hard vectors coming from three
gluon vertices or also attach to the other jet numerators. Invariants resulting from contracting a jet
three-vertex numerator with a soft vertex numerator are zeroth order in normal variables (remember
all soft variables are intrinsic), while those from a jet and a hard vertex are linear at lowest order in
normal variables, but this we already took into account in the contributions from the hard part in
(2.27). The maximum number of such vectors is JgH , as many as the number of lines connecting the
jets to the hard part. The polarization of the SgJ soft gluons connecting the jets to the soft function
(+) (−)
do not produce an invariant contributing to nJ ≡ nJ + nJ , and the fermion-gluon vertices in
the jets where a soft fermion line attaches does not always give a suppression in the numerator [13].
If we want to count the minimum numerator suppressions we can thus subtract JgH + SgJ + SfJ from
the total number of vertices in nJ ,
1
nJ ≥ (V3J − JgH − SgJ + SfJ ) . (2.28)
2
Now we apply the same graphical identity in (2.24) for the hard function, subtracting the number
of external lines coming out of the hard subdiagram , E H = S H +J H (where for the vertex function
we know JfH ≥ 2), since they are not counted here. We obtain
2N H = 2V2H + 3V3H + 4V4H − E H . (2.29)
Here we consider the external fermion current as a two point vertex, so that V2H = 1.
Similarly, the number of jet lines is related to the number of vertices in both jets as
2N J = 2 + J H − S J + 3V3J + 4V4J , (2.30)
here we count the start and endpoint as one-vertices. The lines escaping to the hard part can be
regarded as one-vertices, as if they were external points. The lines connecting the jets to the soft
surface need to be subtracted since these are counted in the soft part.
Now think of a similar relation but regarding only the number of fermions inside the hard part.
Remember now that each fermion-gluon vertex gives two fermion lines and that we have to take
into account that two fermion lines escape to the jets, giving the contribution of two one-vertices.
Subtracting also the fermion lines that connect the hard part to the other regions, since we count
them there, we hence come up with the relation
2V3fH + 2 − SfH − JfH = 2NfH . (2.31)
Similarly, for jets we find
2V3fJ + 2 − SfJ + JfH = 2NfJ . (2.32)
We have now reached a point where we can make explicit the lower bound for γΓ in (2.27). We
apply here the equation (2.29) to the H terms and (2.30) to the J terms. Now we use the two
relations just derived about fermion lines, and the numerator inequality (2.28) to obtain
3 1
γΓ ≥ SgH + SfH + (SfJ + JfH − 2) + O() . (2.33)
2 2
From this last equation we can see, since in the EM vertex we are considering JfH ≥ 2, that no
line connects the hard subdiagram directly to the soft subdiagram and that no femionic line must
connect a jet to the soft part. We also note that there can be as many gluons connecting the hard
and soft part to the jets (see Fig. 2.3). This result is very similar to the momentum-space picture of
the pinch surfaces.
x•2
J(+)
| H S
0
J(−)
x•1
Figure 2.4: General pinch surface corresponding to logarithmic divergences. Only gluons connect-
ing the hard and soft to the jet surfaces can be present to give this divergence.
In [13] it is proven, studying the O() terms, that this logarithmic divergence is regulated by
> 0 in coordinate-space. It is also treated the effect of migrating a vertex from one subdiagram
to another, concluding that doing so does not affect the leading behaviour.
We have then seen that the vertex function is at worst logarithmically divergent times the overall
lowest-order behaviour, and needs dimensional regularization with D < 4.
2.3 Factorization
As in momentum-space, we will see that the structure of the pinch surfaces of the vertex in Fig. 2.4
factorizes as well. This is seen at all orders by the use of the same Ward identities as before.
2.3. FACTORIZATION 29
2.3.1 Hard-collinear aproximation

Let us now begin with the factorization of the hard part from the jet part. To do so, as an example,
take the integral Z
I(x) = dD yJ ν (x)gνρ Dρµ (x − y)Hµ (y) . (2.34)
Here J ν denotes a jet function with direction β ν , Dρµ (x −y) the propagator of the line connecting a
jet vertex at x and a hard vertex at y, and Hµ is a hard function. This integral will diverge when the
jet moves to the plus or minus lightcone direction and all coordinates of the hard function vanish.
In this limit we can replace the Minkowski metric as gµν → β 0 µ βν , where β µ = δ µ+ and β 0 µ = δ µ− ,
and approximate hence
Z
I(x) ∼ dD yJ ν (x)β 0 ν βρ D̄ρµ (x − y)Hµ (y) . (2.35)
In the approximated gluon propagator D̄ we neglect the small terms coming from the hard vertex.
This means that if we take the jet in the plus direction, then the dependence on y µ in the argument
of the propagator will be governed by y − , the component of the hard vertex in the opposite direction
since,
(x − y)2 = −2x+ (x− − y − ) + x2⊥ + O(λ3/2 ) , (2.36)
for x+ y + and x2⊥ y⊥2
. Now we re-express the −+ component of the approximated propagator
as
Z y−
−+ ∂ ∂
D̄ (x − y) = − dσD−+ (−2x+ (x− − σβ 0− ) + x2⊥ ) ≡ − D(x, y − ) . (2.37)
∂y ∞ ∂y
We can now integrate by parts, noting that in the hard function Hµ (z) there is at least one propagator
vanishing at z − = ±∞ and hence there are no boundary terms, to obtain
Z
I(x) ∼ dD yJ + (x)D(x, y − )(−∂y− H − (y)) (2.38)
We can now add the derivatives of all the other components of y µ , since the jet function and D do
not depend on y + and y⊥ and these added terms are total derivatives that vanish after integration,
to get Z
I(x) ∼ dD y(J ν (x)βν0 )D(x, y − )(−∂µ H µ (y)) . (2.39)
Summarizing, our approximation has replaced the gluon propagator escaping from the jet by Dµν →
D(x, y)β 0ν ∂zµ , where β 0 is a vector in the opposite lightcone direction of the jet. This turns out to
be the position space analogue of the longitudinally polarized gluons associated with the scalar
operator ∂µ Aµ (x).
Consider now this approximation in the one-loop vertex function in (2.6), omitting the incoming
currents, integrations over the jet vertex, and numerical factors, it amounts to
Z y1−
y/2 − x
/2 −y/2
Z
(1) D 0 1
I ∼ d y1 β/ 2 dσ + − 2
(x2 − y2 ) + iη) 2− (y + iη) 2−
∞ (−2y2 (y2 − σβ 0 ) + y⊥ + iη)1−
∂ y/1 µ
/ 1 − y/1
x
× − µ γ . (2.40)
∂y1 (y12 + iη)2− ((y1 − x1 )2 + iη)2−
Computing the derivative and using the massless Dirac equation in coordinate-space ∂/SF (x) =
δ D (x) we find
Z y1−
y/2 − x/2 −y/2
Z
(1) D 0 1
I ∼ d y1 β/ 2 dσ + −
(x2 − y2 ) + iη)2− (y + iη)2−
∞ (−2y2 (y2 − σβ 0 ) + y2 2⊥ + iη)1−

D
/ 1 − y/1
x y/1 D

× δ (y1 ) − δ (y1 − x1 ) . (2.41)
((y1 − x1 )2 + iη)2− (y12 + iη)2−
Now we integrate over y1 , the position of the attachment of the longitudinally polarized gluon, using
the delta functions. The two resulting terms differ only in the upper limits of the σ integral and can
be combined to get
y/2 − x
/2 y/2 x−
x
Z
(1) 0 /1 1 1
I ∼ β
/ dσ .
((x2 − y2 ) + iη) 2− 2 2
(y2 + iη) (x1 + iη)2−
2−
0 (−2y2+ (y2− − σβ 0 ) + y2 2⊥ + iη)1−
(2.42)
y•2 • x2
0 •
y1
• x1
Figure 2.5: One-loop hard-collinear approximation. The arrow represents the action on the hard
function H(y2 ), which is just a fermion propagator here.
We see therefore that performing the hard-collinear approximation amounts to the factorization
of the longitudinally polarized gluon into an eikonal line in the opposite direction from the jet (see
Fig. 2.5). The integration over the eikonal line is a scaleless integral and hence, after renormaliza-
tion, when x−1 → ∞ it will be defined only by minus the ultraviolet pole.
As in momentum-space, we can use the Ward identity
hout| T {∂µ1 Aµ1 (x1 )...∂µn Aµn (x1 )} |ini = 0 (2.43)
to factorize the longitudinally polarized gluons from the jets at higher orders [13] in the same way
as in Fig. 1.4.
2.3.2 Soft-collinear approximation

We will now follow the same steps for the gluons attaching the soft to the jet parts. Consider the
integral Z
I(x) = dD ySµ (x)Dµν (x − y)gνρ J ρ (y) , (2.44)
where Sµ (y) denotes the soft function. In this integral, the only singularities of the integrand come
from the jet function. We will keep the leading behaviour in normal variables in the argument of
the gluon propagator D, and keep only the large numerator component of the jet,
Z
I(x) ∼ dD ySµ (x)D̄µν (x − y)βν βρ0 J ρ (y) , (2.45)
where β 2 = β 02 = 0 and β · β 0 = 1. If the jet is in the plus direction, we will have, by power
counting with x− y − and x2⊥ y⊥2
, we will have that
(x − y)2 = −2(x+ − y + )x− + x2⊥ + O(λ1/2 ) . (2.46)
Defining now
Z y+
+− ∂
D̄ (x − y) = + dσD+− (x − σβ) ≡ ∂y+ D(x, y + ) (2.47)
∂y ∞
we arrive as before to the leading contribution
Z
I(x) ∼ dD ySµ (x)β µ D(x, y)(−∂ν J ν (y)) . (2.48)
In this way we see that the jets are connected to the soft part only by scalar polarized gluons (lon-
gitudinally polarized), which can be factored out from the jet by the same Ward identity (2.43) in
the same way as in momentum-space [13].
2.3.3 Eikonal approximation

The two approximations presented above help us factorize the contribution from different subdia-
grams at leading singularity. We will now present an approximation to compute the leading term
easily. It relies on the fact that we can approximate the integrands keeping only the leading terms
by imposing Landau equations.
In the fermionic vertex function the solutions to the Landau equations of the collinear fermions
in the plus (minus) directions set the transverse and the minus (plus) coordinates of the positions
of the fermion-gluon vertices on the plus-line (minus-line) to zero. These conditions approximate
the fermion propagators with
SF (x2 ) = ∂/∆F (x2 ) → θ(x+ )δ(x− )δ (2) (~x⊥ )γ · β , (2.49)
where β µ = δ µ+ for the plus fermion line. This could have been derived from taking the Fourier
transform of the eikonal propagator in momentum-space of a massless scalar field
d4 k −i
Z
4
eik·x = θ(x+ )δ(x− )δ (2) (~x⊥ ) , (2.50)
(2π) β · k − iη
where we have used the integral definition of the θ-function, θ(x) = (2πi)−1 dτ eiτ x /(τ − iη). So
R
that this approximation is the coordinate-space version of the already studied eikonal approxima-
tion.
Now we will employ this approximation to the one-loop vertex diagram in (2.6) to get,
Z
(1) − − − − (2) 1
Γeik = d4 y1 d4 y2 θ(x+ + + (2)
2 − y2 )θ(y2 )θ(y1 )θ(x1 − y1 )δ(y2 )δ (y2 ⊥ )δ (y1 ⊥ )
(y2 − y1 )2 + iη
Z x+2 Z x−1
1
= dλ dσ 0
(2.51)
0 0 (2β · β λσ + iη)
where we have suppressed numerator and numerical factors and introduced the parameters λ and
σ with β 0 µ = δ µ− . The parameters λ and σ in (2.51) indicate at which points the gluon is attached
to the Wilson lines and are integrated over. This result coincides with the g 2 term of the vacuum
expectation value of the time ordering of two Wilson lines, one starting at x1 and ending at the
origin and the other one starting at the origin and ending at x2 ,
(1)
Γeik = h0| T {Φβ (x2 , 0)Φβ 0 (0, x1 ) |0iorder g2 . (2.52)
It is proven in [14] that the vertex function can be approximated by a Wilson line calculation at all
orders in perturbation theory. In [13] it is also proven that this calculation gives the same bound
for the overall degree of divergence of the vertex. We will further study the origin of the imaginary
parts of Wilson line correlators below.
2.3.4 One-loop Jet function in coordinate-space

Once we have seen that factorization of the vertex function in coordinate-space comes along pretty
much as in the momentum-space picture, let us now compute the one-loop jet function in coordinate-
space and see if, using the LSZ reduction formula, we can recover the results known in momentum-
space.
x y
· ·
(1)
≡ JCS (x)
· tn
Figure 2.6: Diagram representing the one-loop jet function (without UV conterterms) in coordinate-
space.
Reading from Fig. 2.6 we have that, taking the Wilson line to be lightlike (n2 = 0),
(1) Γ(1 − )Γ2 (2 − )

JCS (x) = ×
(2π 2− )2 4π 2−
Z +∞
−(/x − y/) −y/ (−igT a )nµ
Z
D a
× d y (−igT γµ ) dt .
((x − y)2 + iη)2− (y 2 + iη)2− 0 ((y − tn)2 + iη)1−
(2.53)
We can introduce Feynman parameters in two steps combining firstly the second and third de-
nominators above and secondly the resulting denominator with the first one above to identify the
different configurations giving rise to unavoidable divergences. We will also employ this Feynman
parametrization to solve the whole integral. In this case, using the delta functions in the Feynman
parameters, the Landau equations read
α2 (x − y)2 + (1 − α2 )α1 y 2 + (1 − α2 )(1 − α1 )(y 2 − 2tn · y) = 0 (2.54)

α2 (x − y)µ + (1 − α2 )α1 y µ + (1 − α2 )(1 − α1 )(y µ − tnµ ) = 0 . (2.55)
To study the case when the gluon line becomes soft we take α1 = 1. In this case we do not expect
the gluon to change de direction of the external line since it carries no momentum. This is indeed
the case because the solution to the Landau equations tell us that
α2
yµ = xµ , (2.56)
2α2 − 1
and that x and y must lie in the lightcone for α2 6= 0, 1. For the end-point singularities in the α
parameters we see that for α2 = 1 the vertex y migrates to the external point. This makes sense
since the fermion line from 0 to y will be carrying all the momentum while the fermion line from
y to x shrinks. Taking α2 = 0 means that the y vertex migrates to the origin, signaling that we are
dealing with an UV divergence. However in both cases there is not restriction to y or x to lie on the
lightcone.
For the case when α1 = 0, so that the gluon is not soft, the Landau equations tell us that
(1 − α2 )tnµ − α2 xµ
yµ = . (2.57)
(1 − 2α2 )
As we will see below, the only contribution to this amplitude will come when t = 0 so that we will
have a collinear pinch surface with the vertex y obeying (2.56) but this time the gluon emerges from
the Wilson line cusp (the origin), making the gluon collinear. Both the fermion external line and
gluon lines are lightlike for α2 6= 0, 1. Again if α2 = 1 the internal vertex y migrates to the external
point x and if α2 = 0 we will have that y µ = tnµ signaling again a UV divergence for t = 0.
To carry out the whole integral we will introduce Feynmann parameters in two steps as above,
first combining the second and third denominators and then the resulting denominator with the first
one in (2.53). In this way
(1) Γ(1 − )Γ2 (2 − ) Γ(5 − 3)

JCS (x) = −g 2 CF 6−3 2
×
16π Γ (2 − )Γ(1 − )
/ y/)α11− (1 − α1 )− α21− (1 − α2 )2−2
Z +∞ Z 1
(/
xn/ y/ − y/n
Z
D
× d y dt dα1 dα2 2 .
0 0 (y − 2y · (α2 x + (1 − α1 )(1 − α2 )tn) + α2 x2 + iη)5−3
(2.58)
Now we shift the integration variable y → y − (α2 x + (1 − α1 )(1 − α2 )tn) and, by parity, drop
odd powers of the new y in the numerator. This yields
(1) Γ(5 − 3)

JCS (x) = −g 2 CF ×
16π 6−3
/ y/)α11− (1 − α1 )− α21− (1 − α2 )2−2
Z +∞ Z 1
(α2 (1 − α2 )/
xn / − y/n
/x
Z
D
× d y dt dα1 dα2 2 .
0 0 (y + α2 (1 − α2 )x2 − 2(1 − α1 )(α2 − α22 )tn · x + iη)5−3
(2.59)
It is possible (taking y to be non-collinear with n so that n · y 6= 0, which is arguably not a possible

condition since we are integrating in all y, we will hence consider we do not integrate y in the region
y · n = 0) at this point to carry out the integration in t, producing

(1) Γ(5 − 3)
JCS (x) = g 2 CF ×
16π 6−3
/ y/)α11− (1 − α1 )−1− α2− (1 − α2 )1−2
Z 1
(α2 (1 − α2 )/
xn / − y/n
/x
Z
D
× d y dα1 dα2 .
0 (4 − 3)2n · x(y 2 + α2 (1 − α2 )x2 + iη)4−3
(2.60)
Using standard Dimensional Regularization formulas we carry out the integral in y and identify the
Euler Beta functions in the Feynmann parameters integrations to get
(1) D (x2 )2−2 Γ(1 − 2)Γ(2 − )Γ(−) Γ(1 + ) Γ()
JCS (x) = −iπ 2 g 2 CF + ×
(2x · n)16π 6−3 Γ(2 − 2) Γ(2 + ) Γ(2 + )

2
× 2x · n/ x(1 − 2) + /nx . (2.61)
Where in the Euler Beta function for the parameter α2 we used the trick of multiplying by α2 + (1 −
α2 ) to separate the finite part when α2 = 1, from the UV divergence arising when α2 = 0, so that
the denominators y 2 and (y − nt)2 are active in producing the UV divergence when y µ , t → 0 (this
is also due to the analysis we have made above of the Landau equations where α2 = 0 signaled a
UV divergence). This means that the renormalized one-loop jet function amounts to
2 2 2−2
(1) D g CF (x ) Γ(1 − 2)Γ(2 − )Γ(−) Γ(1 + ) 2

JCS r (x) = −iπ 2 2x · n/
x (1 − 2) + /
n x .
(2x · n)16π 6−3 Γ(2 − 2) Γ(2 + )
(2.62)
This expression, expanded to zeroth order in , has the behavior
(1) iαs CF −/x 2 / x2
−1 n
JCSr (x) = − + 4(1 − γE ) − (/
x ) + O() . (2.63)
4π 2π 2 (x2 + iη)2 n·x
This is a very nice result since the most divergent part is proportional to a Fermion propagator S(x)
(one can think of this result as follows: the gluon merges collinearly with the fermion producing the
divergence times the propagator) and this will allow us to use the LSZ formula to get the momentum-
space expression in a very simple manner. This formula relates the Jet function in momentum-space
with the one in coordinate-space as
Z
(1)
JM S (p) = −i dD xe−ip·x (−i∂/)JCSr (x) , (2.64)
By using the Dirac equation ∂/S(x) = iδ (D) (x) we find that the most divergent part of the jet function
is
αs CF 2
lim JM S (p) = lim − . (2.65)
→0 →0 4π
So that we reproduce exactly the same pole structure as the renormalized one-loop gluon exchange
(1)
vertex correction JV in (1.54) when n2 = 0, disregarding the term containing n2 . However one
could be doubtful of this ad hoc “renormalization” and also from chucking the zeroth order in in
the expression (n2 µ2 /2(p·n)2 ) in (1.54). Hence we present the full result of the jet function,
(1) iαs CF −/ x 2 2(1 − 2γE ) / x2 1
−1 n
JCS (x) = − + − (/
x ) + O(0 ) , (2.66)
4π 2π 2 (x2 + iη)2 2 n·x
where the double pole expected from the the soft-collinear pinch surface appears.
2.3.5 One-loop jet function with internal gluon emission in coordinate-space

Let us now compute the diagram in Fig. 2.7 contributing to the one-loop radiated jet function in
coordinate-space. We consider a lightlike Wilson line, i.e. n2 = 0.
x·
y w
· · z·
(1)µb
≡ Jrad (x, y)
· tn
Figure 2.7: Diagram representing the one-loop internal gluon emission from a jet.
This diagram amounts to

Z ∞
(1)µ b 3 a b
Γ(2 − ) 3 Γ(1 − ) 2 Z
a D D
−(y/ − w)
/
Jrad (x, y) = (ig T T T ) 2− 2−
d zd w dt ×
2π 4π 0 ((w − y) + iη)2−
2
−(w/ − z/) −/z 1 1

×n/ 2 2−
γµ 2 .
((w − z) + iη) (z + iη) ((w − tn) + iη) ((x − z) + iη)1−
2− 2 1− 2
(2.67)
The color structure above is easily computable and is presented in Appendix A, it gives T a T b T a =
−T b /2N for SU (N ). This factor is also computed for SO(N ), where is just T b /4 (giving no sup-
pression of the diagram for large number of colors), and for Sp(N ) where it amounts to −T b /4.
As in the previous subsection we introduce Feynman parameters in two steps for combining the
second, the third, and last denominators above such that
Z ∞ Z 1
(α11− (1 − α2 )1− α23−2 (1 − α1 )− )
Z
(1)µ b b D
Jrad (x, y) = F d w dt dα1 dα2 ×
0 0 ((w − y)2 + iη)2− ((w − tn)2 + iη)1−
/ nz/γ µ z/ − (/
(y/ − w)/ nw2 + y/n / µ z/
/ w)γ
Z
× dD z 2 , (2.68)
(z − 2z · (α1 α2 w + (1 − α2 )x) + α1 α2 w2 + (1 − α2 )x2 )5−3
where F b encodes the color structure and numerical factors. Now, to compute the integral in z, we
shift it as z → z − (α1 α2 w + (1 − α2 )x) and then drop odd powers of z in the numerator due to
parity. This integration results in
Z ∞ Z 1
(α11− (1 − α1 )1− α23−2 (1 − α1 )− )
Z
(1)µ b 0b D
Jrad (x, y) = F d w dt dα1 dα2 ×
0 0 ((w − y)2 + iη)2− ((w − tn)2 + iη)1−

× (w nγ µ (m2 ) + K µ (m2 )2−3 ,
/ − y/)/ (2.69)
where m2 = α1 α2 (1−α1 α2 )w2 +α2 (1−α2 )x2 −2α1 α2 (1−α2 )x·w and K µ = (y/ − w)/ / n(α1 α2 w
/+
x)γ µ (α1 α2 w
(1 − α2 )/ x) − (2w2 n
/ + (1 − α2 )/ / + y/n / µ (α1 α2 w
/ w)γ / + (1 − α2 )/
x). Next we combine the
second denominator above with the last one again with Feynman parameters and then compute the
integral in t assuming we are not integrating w in the region where w · n = 0. From this integration
we will obtain a denominator factor and a 2x · w. We will combine these with the propagator
denominator that was left in two steps of Feynman parametrization, which gives
Z Z 1
(1)µ b 00b
Jrad (x, y) = F D
d w dα1 ...dα5 α11− (1 − α1 )1− α23−2 (1 − α2 )− α3−1− (1 − α3 )2−2 ×
0
α4 (1 − α4 )1− α54−4 (1 − α5 )0
2−3
× (w
/ nγ µ (m2 ) + K µ .
− y/)/ (2.70)
(αw2 + 2w · () + κ2 )6−4
Here α = α3 α4 α5 + α5 (1 − α3 )α1 α2 (1 − α1 α2 ) + α5 (1 − α4 ), ()µ = (1 − α5 )nµ − α1 α2 (1 −

α2 )xµ − α5 (1 − α4 )y µ , and κ2 = α5 (1 − α3 )α2 (1 − α2 )x2 + α5 (1 − α4 )y 2 . Finally we shift the
1 1
integration variable as w → wα 2 + ()α− 2 and carry out the integration finding
Z 1
(1)µ b ig 3 π D T b
Jrad (x, y)
= 8 10−5 × dα1 ...dα5 α11− (1 − α1 )1− α23−2 (1 − α2 )− α3−1− (1 − α3 )2−2
2π N (3 − 3)
0
2−3 1− 4−4 0
× α4 (1 − α4 ) α5 (1 − α5 ) g ρσ Aµρσ Γ(3 − 3)(M 2 ) + B µ Γ(4 − 3) (M 2 )3−4 , (2.71)
for SU (N ). The expressions for the Aµρν , B µ , and M 2 factors are shown in Appendix E. This result
is not very enlightening since I have not been able to untangle the α parameters integrals while the
tensor part is rather cumbersome and long. I expect that a better parametrization of the integral
may give nicer results.
2.4 Imaginary parts of Wilson line correlators

Recall that amplitudes develop imaginary parts when virtual particles in the amplitude can go on-
shell. Following [16], we will study related interpretations for the soft function (defined above as
a correlator of Wilson line operators) from the coordinate-space picture. This view will give as a
new insight based on causality considerations.
Remember that the long distance singularities of the EM form factor of a colored particle were
encoded in the soft function defined in eq. (1.40). Using the result presented in Appendix B for
the massless scalar propagator in position space in D = 4 − 2 dimensions, the position-space
representation of a soft gluon exchange between two Wilson lines is obtained by direct perturbative
expansion to order gs2 of eq. (1.40). This gives
Γ(1 − ε) ∞ ∞ dt1 dt2 v1 · v2

Z Z
0(1) 2 2
F = −gs µ CF 2−ε
. (2.72)
4π 0 0 [(t1 v1 − t2 v2 )2 + iη]1−ε
Here we choose timelike Wilson lines to avoid collinear singularities, that is v12 = v22 = −1. t1
and t2 have dimension of time and multiplied by v1µ and v2µ respectively represent the positions
of the attachment points of the soft gluon on the Wilson lines. The integrals in (2.72) give an
infrared divergence which can be extracted by the change of variables (t1 , t2 ) = (λx, (1 − x)λ)
with x ∈ [0, 1] and λ ∈ [0, ∞) with Jacobian λ. This yields (omitting color factors and the Γ(1−ε)
4π 2−ε
term), Z ∞ Z 1
(1) 2 dλ dx (−v1 · v2 )
F =µ 1−2 2 2 1−
. (2.73)
0 λ 0 [−x − 2x(x − 1)v1 · v2 − (1 − x) + iη]
2.4. IMAGINARY PARTS OF WILSON LINE CORRELATORS 37
We see that, in four dimensions, the integral in λ gives a logarithmic divergence when the wave-
length λ goes to infinity, i.e. the IR divergence. To regulate this integral we will multiply the
numerator by the regulator e−Λλ so that we extract the divergence
Z ∞
2 dλe−Λλ µ 2 1 µ 2
µ = Γ(2) = + O(0 ) . (2.74)
0 λ1−2 Λ 2 Λ
Now we compute the integral in x in the timelike and spacelike kinematics to obtain, respec-
tively,
(1) 1 µ 2 (γ − iπ) coth γ for v1 · v2 < 0
F = × (2.75)
2 Λ γ coth γ for v1 · v2 > 0
where we define the angle γ as cosh γ = |vi · v2 |. Note that the timelike result can be obtained from
the spacelike one using the analytic continuation γ → πi − γ and that the timelike result develops
an imaginary part not present in the spacelike case. These facts were also the case for the massive
scalar propagator in Appendix C. They can be explained as follows: in the timelike case there are
regions t1 /t2 = e±γ where the denominator (t1 v1 − t2 v2 )2 vanishes and the −iη prescription is
necessary. This gives an imaginary contribution to the integral. In the spacelike case v1 · v2 > 0
the equation (t1 v1 − t2 v2 )2 = 0 has no real solution and therefore the −iη can be dropped in (2.73),
producing a pure real result.
Let us interpret these results in terms of causality considerations. In the spacelike case, the
two partons are never lightlike separated and therefore cannot exchange lightlike massless gauge
bosons. Nonetheless, in the timelike case, the partons can become lightlike separated and therefore
exchange gauge bosons (see Fig. 2.8). We can see how the coordinate description gives us a nice
interpretation of the imaginary parts of correlators from a causality point of view.
Figure 2.8: One loop eikonal diagrams embedded in spacetime diagrams. (a) represents the case
the timelike kinematics where the two Wilson lines are inside the future lightcone representing two
final state partons interchanging an on-shell lightlike gluon, generating thus an imaginary part of
the amplitude. (b) represents the spacelike case where an incoming parton in the past lightcone
and an outgoing parton in the future lightcone. These are never lightlike separated and can only
interchange virtual gluons, hence not generating an imaginary part of the amplitude. (From [16]).
The authors in [16] suggest that there is a direct interpretation for the imaginary part of the
Wilson lines correlator as an interparton potential. In the non-relativistic limit, the initial and final
R∞ −Λt
two-particle states are related as |f iI = ei 0 dte VI t |iiI , where VI denotes the potential in the
interaction picture and Λ a cutoff scale. This interpretation can be made concrete for a configuration
where the particles carry no color charges, since in an Abelian gauge theory (CF = 1) the correlator
of two Wilson lines can be written as the exponential
W ≡ hΦv1 Φv2 i = exp F (1) + O(g 4 ) .

(2.76)
For timelike kinematics, the object that describes the behaviour of the cusp of the Wilson lines
(i.e. short distance contributions to the correlator) is the anomalous cusp dimension defined as
[16]
d log W g2
Γcusp (γ) ≡ − lim = − 2 (γ − πi) coth γ . (2.77)
→0 d log µ 4π
Here the non-relativistic scenario corresponds to the small angle limit γ ' 0 where the velocities
of the outgoing particles are almost collinear so that their relative velocity is also small. Taking
hence this limit, the imaginary part of the cusp anomalous dimension is
g2
Im Γcusp (γ) = + O(γ 0 ) . (2.78)
4πγ
This implies that the imaginary part of the cusp anomalous dimension in timelike non-relativistic
kinematics resembles a non-relativistic Coulomb potential (if one replaces γ by the distance be-
tween the two fermions). The possible results and outcomes of this analisis in non-Abelian gauge
theories are discussed also in [16].
2.5 Largest time equation

We will now follow [15] to derive an equation useful for causality considerations that will help us
interpret further results. Taking the propagator of a scalar particle in four dimensions (see Appendix
C) and performing the energy integrals we can express it in the form
−
∆F (x) = θ(x0 )∆+
F (x) + θ(−x0 )∆F (x) , (2.79)
where
eipx
Z
∆±
F (x) = d4 p θ(±p0 )δ(p2 + m2 ) . (2.80)
(2π)3
Note that ∆− is the complex conjugate of ∆+ , so that
∆∗F (x) = θ(x0 )∆− +
F (x) + θ(−x0 )∆F (x) . (2.81)
−
The form of ∆F says that for x0 > 0 it is equal to ∆+ F and for x0 < 0 equal to ∆F . As we will see,
the Largest Time Equation (LTE) is a generalization of this fact to an arbitrary diagram.
Consider a diagram with n vertices place at x1 , ..., xn connected by propagators with arguments
xj − xi , and let the time component of one of these to be the largest time (xm )0 > (xi )0 for all
i 6= m. We can thus replace the propagators ∆F containing xm − xi (or xi − xm ) as an argument
by ∆+ (or ∆− ).
To apply this reasoning to a general diagram we will introduce the notation of circled vertices
and define the rules:
2.5. LARGEST TIME EQUATION 39
• Each circled vertex will give a minus sign.

• A line connecting two uncircled vertices has a Feynman propagator ∆F associated.
• A line connecting an uncircled and a circled vertex corresponds to a ∆+F.
• A line connecting a circled and an uncircled vertex corresponds to a ∆−F.
∗
• A line connecting two circled vertices corresponds to a ∆F .
These rules can be used to derive the diagrammatic equalities for a single line connecting two
points x1 and x2 where (x2 )0 > (x1 )0 presented in Fig. 2.9.
=− · · =−· ·
∆F (x2 − x1 ) = −∆+
F (x2 − x1 )(−1) (−1)∆− ∗
F (x2 − x1 ) = −(−1)∆F (x2 − x1 )(−1)
Figure 2.9: Simplest LTE for a scalar propagator where the right endpoint, x2 , has the largest time
component.
Another slightly less trivial example is the self energy graph in Fig. 2.10.
= − · · = − · ·
Figure 2.10: Self-energy LTEs where the right vertices have the largest time components.
Now we wish to derive an equation that holds when we do not know which time is the largest.
If we take an arbitrary diagram and sum all diagrams obtained by circling vertices in all possible
ways, this sum vanishes since terms will cancel two by two. Take an arbitrary vertex xm with the
largest time, then any diagram with the vertex xm not circled will cancel against the same diagram
with the vertex xm circled. Take the self energy example in Fig. 2.11, if the time component of the
rightmost vertex is the largest, we know that the first and third diagram cancel against the second
and the fourth diagram respectively. On the other hand, if the leftmost vertex has the largest time,
then the first and second diagram will cancel with the third and fourth diagrams respectively.
+ · + · + · · =0
Figure 2.11: General self-energy LTE for any time ordering of the vertices.
To set down this equation more formally, consider a diagram with x1 , ..., xn vertices, G(x1 , ..., xm ).
The same diagram with some vertices circled will be represented underlining the labels of the cir-
cled vertices, e.g. if the vertex x2 is circled then the diagram will be represented by G(x1 , x2 ..., xm ).
The above mentioned equation will be
X
G(x1 , ..., xm ) = 0 , (2.82)
underlinings
and can be reexpressed as

X
G(x1 , ..., xm ) + G∗ (x1 , ..., xm ) = − G(x1 , ..., xm ) . (2.83)
underlinings
0,all
Now we want to consider diagrams contributing to scattering amplitudes, i.e. with external lines
attached, with integrations over the positions of the internal vertices, and with the Fourier transforms
of the various propagators. The equation above remains valid in this case. However, we will now see
that not all underlinings of internal vertices do not need to be taken into account. The θ functions in
the ∆± in (2.80) will imply that positive energy must flow from the uncircled to the circled vertex,
giving zero otherwise. This means that positive energy can only move from an uncircled towards a
circled vertex and that energy through lines connecting circled or uncircled vertices may go either
way.
Figure 2.12: Circlings of a diagram that are zero due to the θ functions in the ∆± .
Thanks to these considerations we can see that a contribution will be nonzero only if the diagram
has a region where all the vertices are circled on one side of the diagram and uncircled on the other.
We can introduce the notation of shadowing on one side of a cutting line of the diagram as in Fig.
2.13 representing the circling of all vertices in the shaded region. A region on the shadowed side
must be connected to one or more outgoing lines since the energy flow will always be directed from
the unshadowed to the shadowed part of the diagram.
·
≡
Figure 2.13: The shadowed side of the cut stands for the circling of all the internal vertices there.
Now we can modify the “sum over all circlings” in equation (2.83) by the ”sum of all cuttings”
to get X
G + G∗ = − G. (2.84)
cuttings
Where the diagram G is now a function of the external momenta. We will return to this topic when
studying unitarity and cut graphs in the next section.
+ + + =0
Figure 2.14: Expression of the LTE in Fig. 2.11 with external lines in terms of the cutting equation
(2.84). The shaded side of the cutting line has circled all internal vertices.
2.6. TIME ORDERED PERTURBATION THEORY 41
2.6 Time Ordered Perturbation Theory

In quantum mechanics, transition amplitudes between unperturbed initial and final states repre-
sented by Green’s functions are constructed from (virtual) unperturbed states through which the
system passes during the transition. This picture is also present in QFT Feynman diagrams even if
it is not obvious, since we know that internal lines are usually off-shell.
To make this connection with on-shell unperturbed states, recall that in the Coleman-Norton
picture giving rise to the IR leading contributions, intermediate states traveling a noticeable dis-
tance are on-shell and travel freely. Feynman integrals can be re-expressed in a way such that this
connection becomes transparent.
2.6.1 Momentum-space TOPT

Any Green’s function is a sum of all the covariant Feynman graphs contributing to the amplitude,
this sum is equivalent to the sum of time ordered (TO) diagrams. These are topologically equivalent
to covariant diagrams but with their vertices ordered in time. Hence, a covariant graph with V
vertices corresponds to up to V ! TO diagrams (we do not need to count permutation of vertices
giving identical configurations [4]).
Let’s illustrate this fact with a simple example. Take the scalar self energy Feynman diagram and
carry out the energy integrals reexpressing the energy conservation delta function at each vertex
in the integral representation of it, ordering the times of each vertex and finally carrying out the
energy integrals of each line. This amounts to
Z ∞
dk0 (2π)δ(p0 − k0 − k00 ) ∞ dk00 (2π)δ(k0 + k00 − p00 )
Z
0
I(p, p ) =
−∞ 2π −k0 + ~ k 2 + m2 − iη −∞ 2π −k00 2 + (~k − p~ )2 + m2 − iη
2
Z ∞ Z ∞
1 i(p0 τ1 −p00 τ2 )
h
i(ω+ω 0 −iη)(τ2 −τ1 ) i(ω+ω 0 −iη)(τ1 −τ2 )
i
=− dτ1 dτ 2 e θ(τ 2 − τ 1 )e + θ(τ1 − τ 2 )e ,
2ω2ω 0 −∞ −∞
(2.85)
p p q
where ω = ~k 2 + m2 and ω 0 = ~k 02 + m2 = (~k − p~ )2 + m2 . A similar calculation is pre-
sented in the Appendix C. Finally, carrying out the τ integrations using the coordinates τ± = τ2 ±τ1
and using the Schwinger parametrization we obtain
0 1 h 1 1 i
I(p, p ) = −(2π)δ(p0 − p00 ) + . (2.86)
2ω2ω 0 p0 − ω − ω 0 + iη −p0 − ω − ω 0 + iη
How to interpret this result in terms of TO diagrams? In this case we have two vertices and
therefore we can time-order them in two different ways. The first case corresponds to an incoming
particle with energy p0 which decays into two virtual states with on-shell energies ω and ω 0 which
recombine into the state with energy p00 . The second case corresponds to a free particle with energy
p0 that annihilates with two out of three particles originating from the vacuum with energy ω and
ω 0 with their total energy adding up to −p00 . The third particle coming from the vacuum with
momentum p00 travels freely to the final state. These two cases are illustrated in Fig. 2.15 (a) and
(b) respectively.
p0
p0
(a) (b)
Figure 2.15: First order scalar (with cubic interaction) self energy TO diagrams. It is intended that
the time ordering goes from left to right. (a) has an intermediate state of 2 particles while (b) has
4 particles in the intermediate state.
It is easy to see that this pattern extends to all orders such that any Green’s function ∆ can
be seen as either a sum of Feynman G diagrams or a sum of distinct TO diagrams Γ such that,
schematically,
X X
∆= G= Γ. (2.87)
cov. graphs T O graphs
Here, following the example already presented, the TO graphs with external momenta {p} are
defined as
Y Z d3~li
N ({p}, {li }) Y
Y 1
Γ({p}) = −(2π)δ(Ein − Eout ) ,
loops i
(2π)3
int. lines j
2ω j ({li }, {p}) states a Ea − Sa + iη
(2.88)
where the set of lines between the ath vertex and the (a + 1)th vertex define the “state” a. Ea is the
“energy of the state a”, i.e. the energy that has flowed into the diagram up to the ath vertex. Sa is
the “on-shell” energy of state a, which is the sum of of the mass-shell energies of each of the lines
in a,
X Xq
Sa = ωj = ~k 2 + m2 . (2.89)
j j
lines j in a j in a
The factor N ({p}, {li }) represents numerator factors coming from, e.g. fermion propagators and
three gluon vertices, and the {kj } are momenta in each internal line.
We see therefore that, in Feynman diagrams, energy is conserved at each vertex but having lines
off the mass shell while in time-ordered perturbation theory all lines are on-shell but energy is not
conserved at the vertices.
Let us see how from the TO point of view we can gain new insight into the Landau equations and
the Coleman-Norton picture. Manipulating the general TO diagram in (2.88) we obtain
∞ τ3 τ2
d3~li
Z Z Z Y Z Y N
Γ({p}) = −i dτn ... dτ2 dτ1 ×
−∞ −∞ −∞ loops i
(2π)3 int. lines j
2ωj
h n−1
X i
× exp − i (Sa ({li }) − Ea − iη)(τa+1 − τa ) − i(Ein − Eout )τn . (2.90)
states a=1
Using the stationary phase conditions to identify the dominant terms in these integrals with respect
to the loop momentum variables, we get the equations

∂ X X X (a)
Sa ({li })(τa+1 − τ a ) = 0 ⇒ vjµ (τa+1 − τa )j,i = 0 (2.91)
∂liµ a a lines j in a
(a)
with vj = pj /ωj being the usual relativistic velocity and ij being the incidence matrix of state
a. In other words, the conditions given by the stationary phase method yield the Landau equations
(1.6) if we identify the αj coefficients as
X τa+1 − τa
= λ αj . (2.92)
ωj
a (j in a)
With this result at hand it is easy now to interpret things in terms of the Coleman-Norton picture.
An off-shell particle in line j has αj = 0, that is, its flying time is really small or/and its spatial
momentum is huge such that ωj is very large. In collinear cases we will obtain a relationship
between the ωs of the jet particles. Finally, we see that for soft cases, ω goes to zero and we need
the scale λ to infinity to keep the flying time and α finite. Remember we introduced a parameter
λ to keep the displacement of these soft particles and their corresponding Feynman parameter α
finite, in the other cases it can be set to unity (but still with dimensions of length squared).
2.6.2 Unitarity and cut diagrams

The unitarity of the S matrix has a fundamental importance since it implies the conservation of
probability in the field theory. This matrix is related to measurable on-shell scattering amplitudes
(the matrix T ) that are in turn associated to general Green’s functions, G, through reduction for-
mulas. Following [4] we will now employ time ordered perturbation theory in momentum-space
to prove the unitarity of the scattering matrix S for a non-gauged theory which, remembering the
relation with the transition matrix T , S ≡ I + iT , implies,
i(T − T † ) = −T † T . (2.93)
Take now a general sum of TO diagrams, G,

X Y Z d3~li Y NG ({p}, {li }) Y 1
G(p1 , ..., pm ; k1 , ..., kn ) = i ,
TO of G loops i
(2π)3 int. lines j
2ωj ({li }, {p}) states a Ea − Sa + iη
τ
(2.94)
where we have introduced an extra i so that if its truncated according to the usual reduction for-
mula, it contributes to the T matrix. NG contains all numerator factors and also the total energy
conservation delta function, −(2π)δ(Ein − Eout ). We will denote a cut graph by Gγ obtained from
cutting internal lines, i.e. choosing the internal cut lines as final states, with cut γ from the overall
graph G (see Fig. 2.16). The cut γ is basically replacing the TOP “propagator” (Eγ − Sγ + iη)−1 by
−2πiδ(Eγ − Sγ ), i.e. taking the discontinuity of it. Here Sγ = nm=1 γ
ωm is the sum of energies of
the final states in the cut γ. Lγ and Rγ are the subgraphs to the left and right of γ and the momentum
flow is from left to right. Note that the time orderings of Gγ are the same as the time orderings of
k1 p1
.. ..
. Lγ Rγ .
kn γ pm
Figure 2.16: General cut diagram Gγ of equation (2.95).
G and that the denominators in the right part Rγ have to be complex conjugated. In this way we
have
X Z Y d3 ~ql Y NG ({pj }, {ki }, {ql })

Gγ ({pj }, {ki }) ≡ 3
×
TO of Gγ loops of G
(2π) int. lines
2ω b
τ l b
−1 nγ −1
Y X Y
× Es − Ss − iη (2π)δ(Eγ − ωk ) Es0 − Ss0 + iη .
states in R k states in L
s s0
(2.95)
Now we will sum over all the possible cuts γ of the diagram G. To do so remember the distribution
identity
(x + iη)−1 − (x − iη)−1 = −2πiδ(x) , (2.96)
and use it into (2.95) interchanging the sum over final states and time orders of Gγ to get
X X Z Y d3 ~ql Y NG ({pj }, {ki }, {ql })
Gγ ({pj }, {ki }) = 3
×
γ orders of G loops of G
(2π) int. lines of G
2ω b
γ
τ l b
X Y −1 Y −1
× Es − Ss − iη Es0 − Ss0 + iη
final states of order τ states of Rγ states of Lγ
γ s s0
h −1 −1 i
× i Eγ − Sγ + iη − Eγ − Sγ − iη . (2.97)
Let’s prove unitarity using induction, consider the A cuts of a diagram G with A + 1 vertices and
the integrand in (2.97) for a fixed time ordering of Gγ . Remember that the cut γ separates the left
part Lγ with states from 1 to γ − 1, and the right part with states from γ + 1 to A. Assume now
that the relation
XA YA h i γ−1
Y
−1 −1 −1
(Ej − Sj − iη) (Eγ − Sγ + iη) − (Eγ − Sγ − iη) (Ei − Si + iη)−1
γ=1 j=γ+1 i=1
A
hY A
Y i
= (Ej − Sj + iη)−1 − (Ej − Sj − iη)−1 (2.98)
j=1 j=1
holds (here, if the starting index of the product or the sum is bigger that the ending index we suppose
there is no product). Defining (±iηj ) ≡ (Ej − Sj ± iη), let us prove that holds for three vertices,
i.e. A = 2. Equation (2.98) in this case amounts to
2
X 2
Y h i γ−1
Y
−1
−1 −1
(−iηj ) (+iηγ ) − (−iηγ ) (+iηi )−1 =
γ=1 j=γ+1 i=1

−1 −1 −1 −1 −1
= (−iη2 ) (+iη1 ) − (−iη1 ) + (+iη2 ) − (−iη2 ) (+iη1 )−1
= (+iη1 )−1 (+iη2 )−1 − (−iη1 )−1 (−iη2 )−1 . (2.99)
Now we move on to prove this relation for a diagram with A + 2 vertices supposing it holds for
A + 1 vertices,
A+1 A+1
X Y h i γ−1
Y
−1 −1 −1
(−iηj ) (iηγ ) − (−iηγ ) (+iηi )−1 = (−iηA+1 )−1 ×
γ=1 j=m+1 i=1
A
X A
Y h i γ−1
Y h A
iY
−1 −1 −1 −1 −1 −1
× (−iηj ) (iηγ ) − (−iηγ ) (+iηi ) + (iηA+1 ) − (−iηA+1 ) (iηi )−1
m=1 j=γ+1 i=1 i=1
A
hY i h A
iY
(2.98)
= (−iηA+1 )−1 (iηj )−1 − (−iηj )−1 + (iηA+1 )−1 − (−iηA+1 )−1 (iηi )−1 =
j=1 i=1
h A+1
Y A+1
Y i
−1 −1
= (iηj ) − (−iηj ) , (2.100)
j=1 j=1
so that the proof of equation (2.98) is complete. Using this into (2.97) we obtain
X X Z Y d3 ~ql Y NG ({pj }, {ki }, {ql })
Gγ ({pj }, {ki }) = 3
×
γ orders of G loops of G
(2π) int. lines of G 2ωb
γ
l τ b
h Y Y i
×i (Es − Ss + iη)−1 − (Es − Ss − iη)−1 (2.101)
states of G states of G
s s
and hence to the well known unitarity relation

X
Gγ ({pj }, {ki }) = 2Im G({pj }, {ki }) . (2.102)
γ
This equation is the graphical representation of (2.93). Nevertheless, it is much more general since
it does not require the external lines of the cut graph to be on the mass shell.
All things said, this proof of unitarity, although general, does not include gauge theories. It
assumes that all propagator poles correspond to physical states so that (2.102) is an equation for
the physical S matrix. However, since in gauge theories unphysical degrees of freedom (unphysical
polarizations or ghosts) propagate, these can appear in the final states of the cut. Hence our proof
does not apply for these theories. The proof of unitarity for QED and QCD can be found in [4], it
relies on the use of Ward identities to show that the sum over physical polarizations equals the sum
over gauge polarizations.
2.6.3 Coordinate-space TOPT to all orders

Now let us study TOPT in coordinate-space. A general expression for an arbitrary connected scalar
diagram with E fixed external points and L lines in four dimensional coordinate representation is,
remembering the calculation in Appendix B,
N
(−ig)N
Z Y Y 1
Ge ({xa }) = d4 y i , (2.103)
(4π 2 )L vertices
z 2 (yi , xa )
lines j j
+ iη
i=1
where the xa , a = 1, ..., E, denote the external points and the yi the internal vertices, and the vectors
zjµ , j = 1, ..., L, defined by
zjµ = (ji yi + 0ja xa )µ . (2.104)
The incidence matrix elements ji take the value +1 (−1) when zj is defined to end (begin) at
vertex i, and zero otherwise, and similarly for 0ja in terms of external vertices xa . Now we insert
the identity in (2.103) as
N
(−ig)N δ(zj0 − jk yk0 − 0jc x0c )
Z Y Y Z
4
Ge ({xa }) = d yi dzj0 2 , (2.105)
(4π 2 )L vertices lines j
−zj0 + ~zj2 + iη
i=1
where summation over k = 1, ..., N and c = 1, ..., E is intended. Next we re-express each delta
function as an integral over an auxiliary variable with units of energy, Ej0 ,
N 0 0 0 0 0
(−ig)N +∞ +∞ dEj0 eiEj (zj −jk yk −jc xc )
Z Y Y Z Z
4
Ge ({xa }) = d yi dzj0 , (2.106)
(4π 2 )L vertices lines j −∞ −∞ 2π −zj0 2 + ~zj2 + iη
i=1
and carrying out the integrations in zj0 we get

N
(−ig)N iL
Z Y Y 1
Ge ({xa }) = d4 yi ×
2|~zj | + iη
i=1
Z +∞
0 0 0 0 0 0 0 0
× dEj0 θ(Ej0 )e−iEj (jk yk +jc xc −|~zj |−iη) + θ(−Ej0 )e−iEj (jk yk +jc xc +|~zj |+iη) . (2.107)
−∞
Finally, integrating over the time component of the internal vertices, yj0 , we obtain
N L Z +∞
(−ig)N iL
Z Y Y
Ge ({xa }) = (2π)N 3
d yi dEj0 ×
(4π 2 )L vertices lines j −∞
i=1
0 0 0 0 0 0 N L
θ(Ej0 )e−iEj (jc xc −|~zj |−iη) + θ(−Ej0 )e−iEj (jc xc +|~zj |+iη) Y X
× δ El0 li . (2.108)
2|~zj | + iη i=1 l=1
From this equation we can see that there is energy conservation at each vertex and that the θ-
functions produce the same result as in the study of the LTE, where energy can flow forward or
backwards in the time ordering. This is the main difference with the lightcone ordered perturbation
theory where, as we will see below, energy can only flow forward in the diagrams. It will be useful
to note that the number of lines L of a diagram, and hence the number of energy integrals in (2.108),
will satisfy the identity
2L = E + 3N (2.109)
where E is the number of external points and N the number of three-vertices. This equation gives
the restriction that N − E must be an even number. In a connected diagram, we must have that the
smallest number of internal vertices is N = E −2 (corresponding to the tree-level diagrams).
In (2.103), each of the vertices gives an energy-conservation δ-function. Each of these reduce
by one the number of energy integrals. Due to the form of the integrands, each of the indepen-
dent integrals will give a denominator depending on the time components of the external lines and
the modulus of the displacement vectors, |~z|, between vertices due to the Schwinger parametriza-
tion. We will regard these denominators as TOPT propagators. Thus, an arbitrary diagram can
be expressed as a sum of terms, each having in the integrand a product of Np TOPT propagators,
where
E+N
Np = . (2.110)
2
In this way, given a diagram, we can choose a set of Np independent energies. Of course, we
cannot choose energies as independent if their three respective lines emerge or end at the same
three-vertex. The choice of independent energies must be such that any cut in lines of the original
diagram does not produce a set of diagrams with more independent energies than the maximum
allowed in each resulting diagram.
To go further with equation (2.108) let us define the function

−ixα−
 e
 j if x > 0
fj (x) = where αj± ≡ 0jc x0c ± (|~zj | + iη) .
 e−ixα+j

if x < 0
Then (2.108) becomes

N L Z +∞ N L
(−2iπg)N iL
Z Y
3
Y fj (Ej ) Y X 0
Ge ({xa }) = d yi dEj0 δ El li0 . (2.111)
(4π 2 )L vertices lines j −∞ 2|~zj | + iη i0 =1 l=1
i=1
PN
By means of the δ-functions we can express each dependent energy {Ek }k∈Dep as Ek = l p ckl El
where the sum runs over all the independent energies {El }l∈Indep . The coefficients ckl will be ±1
or 0 depending on the expression of Ek in terms of the other independent energies. We will then
have
( P −
e−i(Pl ckl El )αk θ( l ckl El )
P
if Ek flows forward in its line
fk (Ek ) = −i( l ckl El )α+
P
e k θ(−
l ckl El ) if Ek flows backwards in its line .
Now we can express (2.111) as a sum of all possible energy flows in all the lines. Note however
that not all energy flows in an arbitrary connected diagram give a non-zero contribution (this is the
same as in the LTE section above). First, to produce a non trivial result, no vertex in the diagram
can have all energies incoming or all outgoing. Second, the energies flowing in the external lines
must not all flow into or out of the diagram. Also note that inverting all energy flows at the same
time will produce the same result but with all time components getting an extra minus sign. In this
way we have
Z N L
Y
3
Y 1
Ge ({xa }) = KL,N d yi ×
vertices j=1
2|~zj | + iη
i=1
XZ +∞ Y −dl P −d0k Y X
× dEl0 e−iEl dl [αl + k ckl αk ]
θ(d0 k ckl dl0 El0 ) . (2.112)
energy 0 l∈Indep k∈Dep l0
flows
N L
Where we have defined KL,N ≡ (−2iπg)
(4π 2 )L
i
and dl = +1 (d0 k = +1) if the independent (dependent)
energy El (Ek ) flows forward in line l (k) and dj = −1 (d0 k = −1) if it flows backwards. Also
−d
understand that, for dj = ±1, we define αj 0 j ≡ αj∓ (same with d0 ).
N
Due to the presence of the θ-functions, the region of integration will be a subset of R+p and
theP use of Schwinger parametrizations is not possible yet. Defining the restrictions vk∗ ({El }) ≡
d0k l ckl dl El (which have to be positive for giving non-zero in (2.112)), we see that these are just
N N
linear operators acting on R+p and therefore living in the dual space of R+p (which is the space
itself). From this fact we draw the conclusion that there will be m ≤ Np independent restrictions
on the region of integration for each energy flow. We will call the restricted region of integration
N
A = {x ∈ R+p |vk∗ (x) ≥ 0 ∀k}. We can construct now the m × Np matrix (M |M̂ ) (with M being
and m × m matrix such that det M 6= 0) with the elements ak0 l0 = d0k0 ck0 l0 dl0 with k 0 = 1, ..., m and
l0 ∈ Indep selecting the independent restrictions. The matrix (M |M̂ ) has range m by definition,
and we can always arrange the columns such that det M 6= 0. Now we construct the matrix B
as  
a11 ... a1m a1m+1 ... a1Np
 .. .. .. .. .. .. 
 . . . . . . 
M M̂
 
B≡ = am1 ... amm am m+1 ... amNp 
 
0 I  
 
 0 I 
Clearly det B = det M 6= 0, and hence B has an inverse. Also by definition B(x) ≥ 0 ⇒
N N
B(x) ∈ R+p when x ∈ A. Now we take y ∈ R+p so that the vector with independent energies,
PNp
E, arranged in the suitable order is E ≡ B−1 (y). In this way Ej = l=1 bjl yl where the bjl are
−1
the coefficients of B with bjl = δjl if j ≥ Np − m. To be able to carry out the Schwinger
parametrization afterwards, we have to choose the Ej with j ≥ Np − m such that the sum dj +
0
P
c d
k kj k has the same sign as dj to have the correct sing in the iη part. It is possible to check that
the integrations in the variables yk with k = 1, ..., m always yield the correct sign for the iη part so
that the Schwinger parametrization is always possible.
To complete the procedure let us define the function in the integrand of (2.112) as
Y −d −d0
−iEl dl [αl l + k ckl αk k ]
P
0
H({dk }, {dl }, {El }) ≡ e . (2.113)
l∈Indep
Now we are ready to use the change of variables theorem to apply afterwards the Schwinger parametriza-
tion
Z Y Z
I≡ H dEl = N | det(B−1 )|[H ◦ B−1 ](y)dNp y =
0
A l∈Indep R+p
Np
1 Y −i
= −d0
. (2.114)
| det M | l0 =1 P −dl P
l bll0 dl [αl + k ckl αk k ]
Thus, we have been able to carry out all the energy integrals for a particular energy flow in a diagram
finding that any connected scalar (cubic) diagram can be written as
N L Np
−i
Z Y
3
Y 1 X 1 Y
Ge ({xa }) = KL,N d yi −d0k
.
2|~zj | + iη energy | det M | l0 =1 P
b 0 d [α −dl
+
P
c α ]
vertices j=1 l ll l l k kl k
i=1 flows
Note that, of course, all the coefficients bl0 l , dl and d0l depend on each energy flow. Let us clarify
this procedure with some examples below.
Scalar one-loop self-energy diagram
Let us work out an example already studied in momentum-space TOPT, the one loop scalar self
energy diagram with cubic interaction in Fig. 2.17.
x1 • 1 •y1 y2 • 4 • x2 ≡ iΣS (x1 , x2 )

3
Figure 2.17: Coordinate representation of the scalar self energy diagram. The arrows represent
where the line starts and end. With this choice the non zero elements of the incidence matrices are:
11 = 22 = 32 = 042 = +1 and 011 = 21 = 31 = 42 = −1. The numbers indicate the labeling
of the lines.
Reading from Fig. 2.17 and using (2.108) we have that

2 4 Z +∞
g2
Z Y Y
3
ΣS (x1 , x2 ) = − d yi dEj0 ×
(2π)6 i=1 j=1 −∞
−iEj0 (0jc x0c −|~

zj |−iη) 0 0 0 2 4
θ(Ej0 )e + θ(−Ej0 )e−iEj (jc xc +|~zj |+iη) Y X
× δ El0 li ,
2|~zj | + iη i=1 l=1
(2.115)
Now we study all possible energy flows for the diagram. These are,
plus all the same diagrams with all flows inverted. Of course the second and third flows from the
left will give the same answer. In the first case, choosing E2 and E3 as independent energies, it is
not necessary to invert the matrix B since the region of integration is already R2+ . This energy flow
gives the TOPT propagator
−1
, (2.116)
(−∆τ + |~z1 | + |∆~y | + |~z4 | + iη)2
where ∆τ ≡ x02 −x01 is the difference in time between the initial and final points and ∆~y ≡ ~z2 = ~z3 .
For the second energy flow above we have only one independent restriction, namely E2 − E3 ≥ 0.
We construct the positive definite variables yi as

−1 1 E3 −1 1 y1
y = BE = ⇒E= .
0 1 E2 0 1 y2
Notice that we could

P not have chosen a positive definite variable yi = E3 since d3 does not have the
0
same sign as d3 + k ck3 dk and the iη part in the E3 integration would not have the correct sign to
carry out the Schwinger parametrization.
We are ready to use (2.114) to obtain the TOPT propagators as
Np
Y −i −i −i
−d0k
= ,
P −dl P (α3+ − − − − −
+ α1 + α4 ) (α2 + α1 + α4 ) − (α3+ + α1− + α4− )
l0 =1 l bll0 dl [αl + k ckl αk ]
(2.117)
giving
−1
. (2.118)
2(−∆τ + |~z1 | − |∆~y | + |~z4 | + iη)(|∆~y | + iη)
Hence, adding all the contributions from the different energy flows
4
g2
Z
1Y h 1
ΣS (x1 , x2 ) = d3 y 1 d3 y 2 +
(2π)6 j=1
2|~z j | + iη (−∆τ + |~
z1 | + |∆~y | + |~
z 4 | + iη)2
1 i
+ + (∆τ ↔ −∆τ ) , (2.119)
(−∆τ + |~z1 | − |∆~y | + |~z4 | + iη)(|∆~y | + iη)
To interpret this result let us assume now that x1 and x2 are lightlike separated, i.e. ∆τ = |x~2 | =
|~z1 + ∆~y + ~z4 | (where we set x1 = 0, and assume that x2 has the largest time). To get a divergence
from the first term between square brackets in (2.119), the condition ∆τ = |~z1 | + |∆~y | + |~z4 | must
hold. These two conditions will be satisfied at the same time if and only if ~z1 k ∆~y k ~z4 and, to
have that |~x2 − ~x1 | = |~z1 | + |∆~y | + |~z4 |, if the two internal vertices are between the external ones
with the one at ~y1 closer to ~x1 than the one at ~y2 . So that all lines flow forward, corresponding to
the momentum-space TOPT configuration in Fig. 2.15 (a). Hence, this divergence is caused by
the collinear configuration that we expect from the Coleman-Norton picture for this diagram, with
the vertices ordered in time so that no lines emerge from the vacuum. In addition, considering that
the divergency arises when all vertices are collinear and since we expect conservation of energy
we can think that particles travel in lightlike intermediate trajectories between the vertices (See
Fig. 2.18). This is just what we expect from the Coleman-Norton picture. In the timelike case,
∆τ > |~x2 |, the only restriction for a singularity is that the lines must not be collinear, however
these divergences do not correspond to physical configurations in agreement with conservation of
momentum. The spacelike case has ∆τ < |~x2 | which can never be satisfied at the same time as
∆τ = |~z1 | + |∆~y | + |~z4 |, not giving rise to divergences coming from denominators.
·
∆τ
·~x1·· · · ~x2
Figure 2.18: Space representation of some of the configurations for the scalar self energy diagram
with lightlike separated endpoints ~x1 and ~x2 . The red path does not give a divergence for the first
term inside square brackets in (2.119), while the green path does. Seeing therefore that configura-
tions not conserving spatial momentum do not give rise to divergences.
Let us study the second term in (2.119), which we will call Σhsc
S , using the Feynman parametriza-
tion. This amounts to
4 1
g2
Z Z
Y 1 dα
Σhsc
S = 3 3
d y1 d y2 .
(2π)6 j=1
2|~zj | + iη0 [(1 − α)(−∆τ + |~ z1 | − |∆~y | + |~z4 |) + α|∆~y | + iη]2
(2.120)
Notice that the type of denominators appearing in coordinate-space TOPT are linear in the coor-
dinates of the vertices so that we will only have endpoint singularities, i.e. when αj = 0, when
the argument of each denominator vanishes, or when the vertices are placed at infinity. Look-
ing at (2.120), one could argue that the denominators 1/2|zj | should also appear in the Feynman
parametrization. And this is true, but since we have only endpoint singularities these will represent
two vertices going hard, |~zj | = 0 if the corresponding Feynman parameter is one (forcing also all
the other parameters to be set to zero due to the delta function δ(1−Σi αi ) from the parametrization),
and will not contribute to the denominator when the corresponding αj is zero. The parametrized
term in (2.120) tells us that, when α = 1, we will have a hard singularity if |∆y| = 0, while the path
between the external points is not restricted to be lightlike and to have collinear internal vertices.
If α = 0 we are dealing with a possible soft divergence (remember the identification λα∆~y = p~,
so that the line will carry zero momentum but with a finite displacement). In this case, taking the
external points as lightlike separated (i.e. ∆τ = |x~2 |), a collinear pinch surface appears in the
configurations in Fig. 2.19.
~0 ~y2 ~y1 ~x2

• • • • |~x2 | = −|∆~y | + |~z1 | + |~z4 |
Figure 2.19: Soft-collinear configurations giving rise to possible divergences of the coordinate-
space TOPT self energy diagram in equation (2.119). One can see that these configuration corre-
sponds to the one in Fig. 2.15 (b) studied in the momentum-space TOPT, where lines emerged from
the vacuum.
At this point, we have to power-count the different divergences to analyze if they correspond to
true singularities. Recall that from the momentum-space covariant version of the scalar self-energy
diagram we expect logarithmic UV hard and collinear divergences but no soft divergences. Let us
analyze the first type of divergence that arises when ∆~y → 0. Assuming that now ∆τ 6= |~z1 |+|~z4 | to
avoid the zeroth-order lightcone divergences (coming from the first denominator in the second term
of (2.119)), we have that the term from (2.119) encoding the UV hard divergence is proportional
to
f ({yij })
Z 1 3
I≡ dy11 dy12 dy13 dy21 dy22 dy23 1
| 3i=1 (y2i − y1i )2 | 2 (−∆τ + |~z1 | + |~z4 | + iη)
P
f ({yij })
Z
1 2 3 1 2 3
1 3
= dy+ dy+ dy+ dy− dy− dy− (2.121)
i 2 21
P3
2| i=1 (y− )| (−∆τ + |~z1 | + |~z4 | + iη)
where we define y± i
= y2i ± y1i (the y i are the spatial coordinates of each vertex) and f ({yij }) is a
finite function at the pinch surface containing all remaining terms for this contribution in (2.127).
i i
We rescale y− → λy− so that we reach the hard limit when λ = 0. It is easy now to see that with
this rescaling the integral scales as λ0 so that the overall degree of divergence of the diagram is
logarithmic as we had for the covariant diagram. We will not power count the other divergencies in
this case, since we do in the next example.
Next, let us use the LTE to identify the imaginary parts of the diagram. Taking x02 to be the largest
time, the following diagrammatic identities hold
= − · ; · · · · = − · · · .
(2.122)
Summing the first diagram and its complex conjugate, which is the third diagram above, we will
obtain two times the imaginary part of the diagram (remember all the diagrams have an extra i to
contribute to the T matrix). So that
2Im = − · − · · · ∝
Z
∝ d4 y1 d4 y2 ∆(y1 − x1 )∆2 (y2 − y1 )∆+ (x2 − y2 ) + ∆∗ (y1 − x1 )∆∗ 2 (y2 − y1 )∆− (x2 − y2 ) .
(2.123)
Now let us study the two possible time orderings of the internal vertices assuming x01 < y10 , y20 < x02 .
In the case y20 > y10 we can substitute the Feynman propagators by the ∆+ and their complex
conjugates by ∆− , obtaining
2Im ∝y20 >y10

Z
2 2
∝ d4 y1 d4 y2 ∆+ (y1 − x1 )∆+ (y2 − y1 )∆+ (x2 − y2 ) + ∆− (y1 − x1 )∆− (y2 − y1 )∆− (x2 − y2 ) .
(2.124)
For the other time ordering y20 < y10 we will have
2Im ∝y20 <y10

Z
2 2
∝ d4 y1 d4 y2 ∆+ (y1 − x1 )∆− (y2 − y1 )∆+ (x2 − y2 ) + ∆− (y1 − x1 )∆+ (y2 − y1 )∆− (x2 − y2 ) .
(2.125)
Remembering that the ∆± are propagators of on-shell particles with positive or negative frequency,
we see that the imaginary part of this diagram arises when particles travel on-shell between vertices.
Also, the ∆± tell us how the energy flows with respect to the time ordering. In the first term of
(2.124) we see how all the energy flows forward in time, corresponding to the configuration in
Fig. 2.15 (a), and that the second term has the opposite energy flow. In the other case, the first
term in (2.125) tells us that the energy flows forward in time from x1 to y1 and from y2 to x2 , and
backwards from y1 to y2 , corresponding to the configuration in Fig. 2.15 (b) and the coordinate-
space counterpart in Fig. 2.19 where three particles emerged from the vacuum and two of them
annihilate with the incoming particle and the other travels to the final state.
Scalar one-loop vertex diagram
Let us move on to a more relevant example that has connections with all the previous discussions,
i.e. the triangle diagram in Fig. 2.20.
Reading from Fig. 2.17 and using (2.108) we have that
3 6 Z +∞
−ig 2
Z Y Y
3
ΓS (x1 , x2 , x3 ) = d yi dEj0 ×
(2π)9 i=1 j=1 −∞
0 0 0 0 0 0 3 6
θ(Ej0 )e−iEj (jc xc −|~zj |−iη) + θ(−Ej0 )e−iEj (jc xc +|~zj |+iη) Y X
0

× δ El li ≡
2|~zj | + iη i=1 l=1
3
−g 2
Z Y
≡ d3 yi ΥS ({yi }) . (2.126)
(2π)9 i=1
After a very lengthy calculation, from the possible 64 terms ( corresponding to the possible energy
flows) we are left with only 24 non-trivial energy flows. Using the Schwinger parametrization,
• x3
6
y3
•
3
x1 • 1 y1
• 4 ≡ igΓS (x1 , x2 , x3 )
2 •
y2
5 • x2
Figure 2.20: Coordinate representation of the scalar triangle diagram. With this choice of where
the lines start and end, the non zero elements of the incidence matrices are: 11 = 22 = 33 =
43 = 063 = 052 = +1 and 011 = 21 = 31 = 52 = 42 = 63 = −1.
these yield the result

3 6
g2
Z Y
3
Y 1
ΓS (x1 , x2 , x3 ) = − d y i ×
(2π)9 i=1 j=1
zj | + iη
2|~
n 1 1
× ×
−∆τ31 + |~z1 | + |~z3 | + |~z6 | + iη −∆τ31 + |~z1 | + |~z2 | + |~z4 | + |~z6 | + iη
1 1
× +
−∆τ2,13 + |~z5 | + 12 (|~z1 | + |~z2 | − |~z4 | − |~z6 |) + iη ∆τ2,13 + |~z5 | + 21 (−|~z1 | − |~z2 | + |~z4 | + |~z6 |) + iη
1 1
+ ×
−∆τ21 + |~z1 | + |~z2 | + |~z5 | + iη −∆τ21 + |~z1 | + |~z3 | + |~z4 | + |~z5 | + iη
1 1
× +
−∆τ3,12 + |~z6 | + 12 (|~z1 | + |~z3 | − |~z4 | − |~z5 |) + iη ∆τ3,12 + |~z6 | + 21 (−|~z1 | − |~z3 | + |~z4 | + |~z5 |) + iη
1 1
+ ×
−∆τ32 + |~z4 | + |~z5 | + |~z6 | + iη −∆τ32 + |~z2 | + |~z3 | + |~z5 | + |~z6 | + iη
1 1
× +
−∆τ1,32 + |~z1 | + 12 (|~z2 | − |~z3 | + |~z5 | − |~z6 |) + iη ∆τ1,32 + |~z1 | + 21 (−|~z2 | + |~z3 | − |~z5 | + |~z6 |) + iη
1 h 1
+ 1 ×
|~z2 | + |~z3 | + |~z4 | + iη −∆τ31 + |~z1 | + |~z6 | + 2 (−|~z2 | + |~z3 | − |~z4 |) + iη
1 1
+ +
−∆τ32 − |~z4 | + |~z5 | + |~z6 | + iη −∆τ21 + |~z1 | − |~z2 | + |~z5 |
1 1
+ 1 +
−∆τ32 + |~z5 | + |~z6 | + 2 (−|~z2 | − |~z3 | + |~z4 |) −∆τ12 + |~z1 | − |~z2 | + |~z5 | + iη
1 1
+ + 1 ×
−∆τ31 + |~z1 | − |~z3 | + |~z6 | + iη −∆τ12 + |~z1 | + |~z5 | + 2 (|~z2 | − |~z3 | − |~z4 |) + iη
1 1 i o
× + + (τ ↔ −τ ) .
−∆τ32 − |~z4 | + |~z5 | + |~z6 | + iη −∆τ13 + |~z1 | − |~z3 | + |~z6 | + iη
(2.127)
Where we define, ∆τi,jk = x0i − 21 (x0j + x0k ), and the last symbol in brackets represents the terms
shown before it with an extra minus sign in all the time components.
Let us interpret the divergences coming from the different terms above assuming all external
points are lightlike separated between them. The first term between curly brackets above can be
expressed as,
1 1
×
−∆τ31 + |~z1 | + |~z3 | + |~z6 | + iη −∆τ31 + |~z1 | + |~z2 | + |~z4 | + |~z6 | + iη
2
× . (2.128)
(−∆τ21 + |~z5 | + |~z1 | + |~z2 |) + (−∆τ23 + |~z5 | − |~z4 | − |~z6 |) + iη
We see that the first two TOPT propagators, in case ∆τ31 > 0, will give rise to divergences whenever
the three internal vertices lie in the direction between ~x1 and ~x3 ordered as ~y1 -~y2 -~y3 , in this direction.
The third propagator will give another divergence when ~x2 lies in the direction of ~x3 − ~x1 and after
~x3 if ∆τ23 > 0, being all lines of the diagram collinear. See Fig. 2.21.
~x1 ~y1 ~y3 ~x3

• • • • |(~x3 − ~x1 )| = |~z1 | + |~z3 | + |~z6 |
~x1 ~y1 ~y2 ~y3 ~x3
• • • • • |(~x3 − ~x1 )| = |~z1 | + |~z2 | + |~z4 | + |~z6 |
~x1 ~y1 ~y2 ~y3 ~x3 ~x2
• • • • • •
|(~x2 − ~x1 )| = |~z1 | + |~z2 | + |~z5 |
|(~x2 − ~x3 )| = |~z5 | − |~z4 | − |~z6 |
Figure 2.21: From top to bottom, the space collinear configurations giving rise to divergences in
the first, second and third term in (2.128) respectively. The external vertices are time-ordered from
left to right. Notice that the first configuration can happen independently of the other two, if we
have the second configuration we also have the first, and having the third means having also the
previous two. The dotted line is depicted to represent the vertices being collinear.
~x2
~x3 ~x3 ~x3 •
~y3 • ~y3 • ~y3 •
• ~y2 • ~y2 •
~y1 ~x2 ~y1 • ~x2 ~y1 •
~x1 • • • ~x1 • • ~x1 •
•
~y2 • •
Figure 2.22: The three configurations for the vertex diagram giving rise to divergences in (2.128).
The dashed lines are free while the bold lines are fixed in that particular configuration. Meaning
that in the first diagram from the left ~y2 and ~x2 can be placed anywhere in space, still giving rise to
a divergence in the first term in (2.128) no matter the time ordering of ~x2 . In the second diagram
only ~x2 is free (and also its time ordering with respect to the other vertices), giving divergences
in the first and second term in (2.128). In the third diagram all vertices must be collinear and the
external vertices are time ordered from left to right, giving rise to a divergence in all the terms in
(2.128).
Let us power-count these collinear divergences to see if they correspond to true singularities.
Remember that from the Coleman-Norton picture for this diagram we expect divergences coming
from configurations with the topologies in Fig. 1.1 (b)-(c) and we will see appear also (d) since
in our scalar theory it is the same as the other ones. We will require that ~x2 is not collinear to
~x3 − ~x1 , so that the divergence from the third configuration in the figure above is not present, and
by translational invariance we can set ~x1 = ~0. To power-count we will use cylindrical coordinates
with ~x3 oriented in the z-axis in Fig. 2.23.
φ3 ~x3
•
φ2
y3 cl y3 ⊥
• ~y3
φ1 y2 cl
~y1• y1 ⊥ y1 cl y2 ⊥
• • ~x2
~x1 ≡ ~0 • ~y2
Figure 2.23: Coordinates chosen to analyze the collinear divergences in the first two terms in
(2.128). The yi ⊥ measure the distance to the collinear line, φi represent the angles formed in the
plane perpendicular to ~x3 , and yi cl the collinear separation from the origin. The red dotted line
depicts the triangle diagram.
In this reference frame, the TOPT term in (2.128) contributing to (2.127) is proportional to
3
Z Y 6
Y 1
I= dyi cl yi ⊥ dyi ⊥ dφi θ(yi cl ) θ(y3 cl − y1 cl )θ(y3 cl − y2 cl )×
i=1 j=1
zj | + iη
2|~
θ(y2 cl − y1 cl )θ(x3cl − y3 cl )f ({yi cl }, {yi ⊥ }, {φi })
× , (2.129)
(−∆τ31 + |~z1 | + |~z3 | + |~z6 | + iη)(−∆τ31 + |~z1 | + |~z2 | + |~z4 | + |~z6 | + iη)
where we introduce the θ-functions to have the configuration in the middle diagram in Fig. 2.22
and to avoid for now divergences arising from vertices going hard. f ({yi cl }, {yi ⊥ }, {φi }) is a finite
function at the pinch surface containing all remaining terms for this contribution in (2.127). It is
easy to see that the intrinsic variables for the pinch surface are the {yi cl }, and the {φi }. While the
normal variables are the {yi ⊥ }. We have to study how the denominators in (2.129) scale under
the rescaling yi ⊥ → λyi ⊥ . For that we have that, assuming that we are in the region where the
θ-functions in (2.129) are nonzero,
1
∆τ31 = |~x3 | = |(x3cl −y3 cl )2 +(y3 cl −y1 cl )2 +(y1 cl )2 | 2 = |x3cl −y3 cl |+|y3 cl −y1 cl |+|y1 cl | (2.130)
and that, taking all φi to be equal (since they are just intrinsic variables at the pinch surface),
|~z1 | + |~z3 | + |~z6 | =

1 1 1
= ((x3cl − y3 cl )2 + λ2 (y3 ⊥ )2 ) 2 + ((y1 cl )2 + λ2 (y1 ⊥ )2 ) 2 + ((y3 cl − y1 cl )2 + λ2 (y3 ⊥ − y1 ⊥ )2 ) 2
λ2 (y3 ⊥ )2 12 λ2 (y1 ⊥ )2 21 λ2 (y3 ⊥ − y1 ⊥ )2 21
= (x3cl − y3 cl ) 1 + + y1 cl 1 + + (y3 cl − y1 cl ) 1 +
(x3cl − y3 cl )2 (y1 cl )2 (y3 cl − y1 cl )2
λ2 (y3 ⊥ )2 λ2 (y3 ⊥ − y1 ⊥ )2
= (x3cl − y3 cl ) 1 + + ... + (y 3 cl − y 1 cl ) 1 + + ... +
2(x3cl − y3 cl )2 2(y3 cl − y1 cl )2
λ2 (y1 ⊥ )2
+ y1 cl 1 + + ... , (2.131)
2(y1 cl )2
due to the θ-functions in (2.129). This calculation is similar for the other denominator. Having
hence that in the collinear limit λ → 0 the TOPT propagators behave as
1 1
∼λ→0 4 . (2.132)
(−∆τ31 + |~z1 | + |~z3 | + |~z6 | + iη)(−∆τ31 + |~z1 | + |~z2 | + |~z4 | + |~z6 | + iη) λ
Note now that each volume element yi ⊥ dyi ⊥ scales as λ2 so that the integrand in (2.129) scales as
λ2 . Giving therefore no divergence whatsoever. Remember that in the Coleman-Norton picture, in
the two collinear configurations, the vertices ~y2 and ~y3 migrated towards ~y1 respectively due to the
shrinking of their respective lines. Let us discuss one of these cases next.
Note that in the integral encoding the collinear singularities (2.129) we can safely take the limit
y2 ⊥ → y1 ⊥ , this can be done introducing a δ(y2 2⊥ − y1 2⊥ ) in this integral. Now we can take the
collinear limit while the vertex ~y2 migrates to ~y1 , as (y2 cl −y1 cl , y1 ⊥ , y3 ⊥ ) → λ(y2 cl −y1 cl , y1 ⊥ , y3 ⊥ ).
This makes that the volume element in the normal variables scale as λ5 while in the two TOPT prop-
agators still give λ−4 and the 1/2|z~2 | produces a λ−1 . Hence, we find that the collinear configuration
predicted by the Coleman-Norton picture (where the line ~z2 shrinks) gives rise to a logarithmic di-
vergence. This is precisely the same result we found in momentum-space in the first chapter.
For the other collinear divergence the treatment is the same. We see therefore how, in order to
have a divergence, the Coleman-Norton configurations in Fig. 1.1 (b)-(c) are necessary. However,
we can not find a reason here why the diagram in Fig. 1.1 (d) is excluded since the same divergence
is present here. We could nonetheless have expected this since we are not making any difference
between all the three external lines and we do not have anything that relates us to a picture where
we can identify the external lines as final or initial on-shell states. It is possible that the last con-
figuration gets excluded when using the LSZ formula in this context to take the external points as
final (initial) on-shell states. This would require further investigation.
The next term in (2.127) has the same form as (2.128) but with the substitution of (∆τ21 , ∆τ23 ) ↔
(−∆τ21 , −∆τ23 ). So that the configurations giving rise to divergences are the same as in Fig. 2.21
but with ~x2 all the way to the left, meaning that it has the smallest time compared to the other ex-
ternal vertices. The following four terms in (2.127) have analogous configurations giving rise to
divergences.
We move on to analyze the first term in (2.127) giving rise to an UV hard divergence,
1 1
× ×
|~z2 | + |~z3 | + |~z4 | + iη −∆τ32 − |~z4 | + |~z5 | + |~z6 | + iη
2
× . (2.133)
(−∆τ31 + |~z1 | + |~z6 | + |~z3 |) + (−∆τ31 + |~z1 | + |~z6 | − |~z2 | − |~z4 |) + iη
The first denominator above encodes the hard divergence when |~z2 | = |~z3 | = |~z4 | = 0. In this case
the other two terms give additional divergences whenever the three external vertices lie in the same
line with x3 having the largest time. The second term gives a divergence for the collinear ordering
~y2 -~x2 -~y3 -~x3
~x2 ~y3 ~y2 ~x3

• • • • |(~x3 − ~x2 )| = −|~z4 | + |~z5 | + |~z6 |
The third term cannot give rise to a divergence since the first term in its denominator requires the
collinear ordering ~x1 -~y1 -~y3 -~x3 while the second requires ~x1 -~y3 -~y2 -~y1 -~x3 . All the remaining hard
terms have analogous divergences. The terms with opposite time differences are the same as those
already discussed but the time ordering of the external vertices must be read from right to left. From
power counting it is possible to show that in the hard limit (|~z1 |, |~z2 |, |~z3 |) → λ(|~z1 |, |~z2 |, |~z3 |) the
integral scales as λ2 (avoiding the zeroth order lightcone singularity when ∆τ32 = |~z5 | + |~z6 |) so
that no hard divergence is present. This is the same scaling that one can derive from the momentum-
space diagram.
We have therefore seen that our newly developed coordinate-space TOPT description for the
scalar cubic theory offers a very neat causality and space configuration picture, predicting the same
degree of divergence as in the momentum-space picture for Coleman-Norton-like configurations.
As we will see next, these results are easily generalizable to gauge theories such as QED since
we will have the same topologies in our diagrams. We will apply these results to the one-loop
electromagnetic form factor.
2.6.4 Quark electromagnetic form factor in coordinate-space TOPT

Let us now extend our treatment of TOPT to fermions. We have to recover the expression of an
arbitrary Green’s function in (2.105) but distinguishing now between fermion and scalar (these will
be vectors in QED) lines. The fermion lines will require special treatment, we have to study the
expression
0 0 0 0 0
z µ δ(zj0 − jk yk0 − 0jc x0c ) dEj0 z µ eiEj (zj −jk yk −jc xc )
Z Z +∞ Z
µ 0 j 0 j
Ij = dzj 2 = dzj 2 . (2.134)
(−zj0 + ~zj2 + iη)2 −∞ 2π (−zj0 + ~zj2 + iη)2
Now we use the Cauchy’s integral theorem to carry out the integral in zj0 obtaining
iẑjµ
0 ∂
X 3
Ijµ = 2ẑj − 2 δ µi
×
(2|~zj | + iη)3 ∂ ẑj0 i=1
dEj0
Z +∞
0 −iEj0 (jk yk0 +0jc x0c −|~
zj |−iη) 0 −iEj0 (jk yk0 +0jc x0c +|~
zj |+iη)
× θ(Ej )e + θ(−Ej )e , (2.135)
−∞ 2π
2.7. LIGHTCONE-ORDERED PERTURBATION THEORY 59
where we have defined the lightlike vector ẑjµ = (|~zj |, ~zj ) (see Appendix D for the details of this
calculation). With this representation of the fermion propagator we can use all results obtained
in coordinate-space TOPT to all orders including fermion lines. These will only add the term in
big brackets in the first line, which do not depend on y0 and the delta functions of conservation of
energy are still obtained in the same way as in (2.108).
• x2
y2
• 6
3
x1 1 y 1
• • γµ 4 ≡ −ie Γµ(1) (x1 , x2 , x3 )
2 •
y2 5
• x3
Figure 2.24: One loop contribution to the quark EM vertex function in equation in coordinate-space.
The numbers refer to the labeling of the displacement vectors ~zj .
Let us apply this result to the one-loop quark electromagnetic form factor in Fig. 2.24. We can
readily see that, remembering the minus sign from the fermion loop,
3 3
g 2 CF ẑlµl
Z Y
µ 3
Y 1 Y
0 ∂
X
µl i

Γ(1) (x1 , x2 , x3 ) = d yi 2ẑl 0
− 2 δ ×
(2π)9 i=1 j=1,4
2|~
z j | + iη l6=1,4
(2|~
zl | + iη)3 ∂ ẑl i=1
× γµ6 γν γµ3 γ µ γµ2 γ ν γµ5 ΥS ({yi }) , (2.136)

where ΥS ({yi }) was defined in (2.126) and contains the TOPT propagators calculated in (2.127).
It is very interesting to see now how the hard logarithmic singularity that we expect from the
momentum-space picture arises in our coordinate-space diagram. Recall that in the scalar case,
under the rescaling (|~z1 |, |~z2 |, |~z3 |) → λ(|~z1 |, |~z2 |, |~z3 |), the integrands containing the hard singu-
larities in (2.127) scaled as λ2 . Now two of the lines that shrink, |~z2 | and |~z3 |, are fermion lines
and each produce a factor of λ−1 extra with respect to the scalar case. Giving therefore a scaling of
λ0 corresponding to the logarithmic singularity we expect from power counting in the momentum-
space description. In collinear cases the scaling goes along as in the scalar case, giving a logarithmic
divergence strictly linked with the Coleman-Norton picture. However the same problem as in the
scalar case arises: the diagrams in Fig. 1.1 (d) are not excluded to give a divergence. As in the
scalar case, here we should make a connection between external points and on-shell final or initial
states through an adapted version of the LSZ formula. Summing up, our newly developed TOPT
treatment is in agreement with the Coleman-Norton picture but some further investigation is re-
quired to get a full answer that excludes the non-Coleman-Norton cases. Also, the interpretation of
soft lines becomes not always apparent in the coordinate-space TOPT treatment.
2.7 Lightcone-ordered perturbation theory

Following [17], let us first review the main results of lightcone-ordered perturbation theory (LOPT)
in momentum-space to compare with the results in time-ordered perturbation theory. Take a momentum-
space Green’s function G({ld }, {kc }) with incoming momenta kc and outgoing momenta ld . We
will integrate over the minus component of all loop momenta in each of the covariant diagrams
contributing to this correlation function. As in TOPT, the resulting expression will be a sum of
different orderings of the V vertices, P, in which the plus momentum flows forward in each line.
The states, s = 1, ..., V − 1, are also defined as the set of lines between two vertices in a precise
ordering. For a scalar diagram, each of such ordered diagram to order N in perturbation theory is
of the form [18]
+ V −1
dqi+ d2 qi⊥ Y θ(pj ) Y
Z Y
N 1
GP = ig , (2.137)
loops i
(2π)3 lines j
2p+
j states s P
− (s)
−
P p2⊥ +m2
+ iη
ext p∈s 2p+
where the pj are the momenta of each of the lines, being linear combinations of the internal, external
− (s)
and loop momenta. Pext is the total external minus momentum flowing into the diagram before
state s. The main difference between the lightcone and the time ordered descriptions is that no lines
“emerge from or disappear into the vacuum” in LOPT as in the second diagram in Fig. 2.15.
As before we can think that the lines in one of the states all go on shell, i.e. a cut γ = 1, ..., V − 1
between any two vertices, and replace their minus momentum denominators by their discontinuity
and complex conjugating the right part of the diagram, as in the TOPT case,
+ V −1
dqi+ d2 qi⊥ Y θ(pj ) Y
Z Y
(γ) N 1
GP ({ld }, {kc }) =g 3 + p2⊥ +m2
×
loops i
(2π) lines j
2pj states P − (t) − P
− iη
ext p∈t 2p+
t=γ+1
γ
X p2 + m2 Y 1
− (γ) ⊥
× 2πδ Pext − p2⊥ +m2
. (2.138)
p∈γ
2p+ states s=1 P
− (s)
−
P
+ iη
ext p∈s 2p+
Using the same proof by induction in (2.98) we can prove again unitarity in lightcone ordered
perturbation theory
X
Gγ ({pj }, {ki }) = 2Im G({pj }, {ki }) . (2.139)
γ
In [17] it is proven the general form of LOPT to all orders in perturbation theory, studying also the
case where there are charged scalar lines. They are able to prove that it is possible to decompose a
general Feynman diagram as a sum of path orderings of LOPT diagrams.
Coordinate-space LOPT
Now we will follow similar steps in the coordinate-space description. Remembering the general ex-
pression for an arbitrary connected scalar diagram in (2.103) but this time in lightcone coordinates,
we insert the identity as
N
(−ig)N δ(zj− − jk yk − 0jc xc )
Z Y Y Z
Ge ({xa }) = 4
d yi dzj− , (2.140)
−2zj− zj+ + zj⊥
2
+ iη
i=1
2.7. LIGHTCONE-ORDERED PERTURBATION THEORY 61
where summation over k and c is intended, and then express the delta function as an integral of a
conjugate light-cone momentum, Ej+ , as
N Y Z +∞ dEj+ Z +∞ iEj+ (zj− −jk yk− −0jc x−
c )
(−ig)N
Z Y
4 −e
Ge ({xa }) = d yi dzj . (2.141)
(4π 2 )L vertices lines j −∞
2π −∞ −2zj− zj+ + zj⊥ 2
+ iη
i=1
Now we carry out the integrations in zj− closing the contour in the upper (lower) half complex plane
for Ej+ positive (negative), here we pick residues only when zj+ has the same sign as Ej+ . We then
get
2

+ − 0 − z⊥ +iη
−iEj jk yk +jc xc −
N L Z +∞ 2z +
(−ig)N L
Z Y
+ + e
Y j
4 +
Ge ({xa }) = i d y i dE j θ(zj Ej ) + .
(4π 2 )L vertices lines j −∞ 2|zj |
i=1
(2.142)
Finally, we integrate over the minus component of all the internal vertices obtaining delta functions
of conservation of the lightcone momentum for each of these vertices,
(−ig)N L
N
Ge ({xa }) = (2π) i ×
(4π 2 )L
2 +iη

zj⊥
−iEj+ 0ja x−
a−
N L Z 2z + N L
+∞ dEj+ e
Z Y Y j Y X
× 3
d yi θ(zj+ Ej+ ) δ El+ li
vertices lines j −∞ 2π 2|zj+ | i=1 l=1
i=1
(2.143)
where no summation over repeated indices is intended.
From (2.143) we can see that the step functions make each lightcone energy Ej+ flow in the same
direction as the plus component of the line vector zj+ . At fixed yj+ , the Ej+ flow only “forward” in
every diagram, once we have ordered all the vertices from smallest to largest values of yj+ , and are
conserved at each vertex. Also, whenever any yi+ is earlier or later, i.e. lower or higher, than all
vertices it is connected to (internal or external), there is no contribution to the amplitude. So that
no sets of lines emerge or disappear into the vacuum as in LOPT in momentum-space.
Let us come back to the example studied in TOPT of the scalar self-energy diagram in Fig. 2.17
assuming the lightcone ordering x+ + + +
1 < y1 , y2 < x2 . Due to the theta functions in (2.143), the
amplitude will vanish if y1+ > y2+ . This means that configurations where lines emerge from the
vacuum as in Fig. 2.15 (b) are not present in LOPT. On the other hand if y1+ < y2+ after similar
calculations as in TOPT we find
4
g2 θ(y2+ − y1+ )
Z
+ 2 + 2
Y 1
ΣS (x1 , x2 ) = − dy1 d y1⊥ dy2 d y2⊥ ,
(2π)6 j=1
2|zj+ | + iη (−∆µ + z1⊥ 2
+ +
2
z2⊥
+ +
2
z4⊥
+ + iη)
2
2z 2z 2z 1 2 4
(2.144)
where ∆µ ≡ x− −
2 − x1 . We see therefore how the denominator resulting from LOPT only captures
the first type of denominator that appeared in the self energy graph in TOPT in equation (2.119)
with the same power and mathematical shape. Also, the denominator giving rise to configurations
where particles emerge from the vacuum is not present.
2.8 Cross Sections in coordinate-space

For completeness of this thesis, following [19], we will present briefly how the cancellation of IR
divergences occurs in fully inclusive cross sections in coordinate-space. In momentum-space, the
contribution of a final state f to a fully inclusive cross section is proportional to
Y Z dD kj X
2 0 ∗ D (D)
D
δ(k j )θ(kj )A ({kj })(2π) δ P i − k A({kj }) , (2.145)
j
(2π)
k∈{kj }
where A({kj }) is the amplitude for the production o a final state with particles with momenta {kj },
δ(kj2 )θ(kj0 ) is the momentum-space “cut propagator” of the final states (which is actually the same
propagator ∆+ we had in the LTE if Fourier transformed). The dependence on the initial state i is
suppressed by simplicity.
If we want to express the cross section in coordinate-space we will need therefore the Fourier
transform of the cut propagator which is [19]
dD k ik·(y−w) Γ(1 − )
Z
+ 1
∆ (y − w) = D
e (2π)δ(k 2 )θ(k 0 ) = 2−
(2π) 4π ((y − w) + iη(y 0 − w0 ))1−
2
(2.146)
0 0 0 0
Note that for y > w we have a propagator in the amplitude, but if y < w we will have a propaga-
tor in the complex conjugate amplitude. This is something we already knew from the decomposition
in equation (2.81) and the fact that (∆+ )∗ = ∆− .
Consider now that two vertices x and y lie collinear in the plus direction (x− = x⊥ = y − = y⊥ =
0) connected by two lines that pass through the vertex w and take the cut in the line connecting w
to y as in Fig. 2.25.
w
•
x y
• →+ •
∗
A A
Figure 2.25: This configuration illustrates the role of a cut propagator in cross sections. When w
lies on the + direction a collinear pinch surface is reached.
To have a pinch in the amplitude A we know that w0 > x0 . An the imaginary part of the cut
propagator has imaginary part η(y 0 − w0 ) > 0. When the vertices are aligned in the lightcone the
cut propagator has a pole w− = −iη(y 0 − w0 )/2(y + − w+ ), which is in the lower half plane since
plus components have the same ordering as time components. This is the same pole as if the vertex
y was part of the amplitude and no cut were applied. The pole in the w − x propagator is in the
upper half plane, so that in both cases (with or without the cut in the propagator) the integration in
w− is pinched. In the case that w0 > 0, the cut propagator has imaginary part η(y 0 − w0 ) < 0, but
since y + −w+ < 0, the pole remains in the lower half plane and the pinch in the w− integration
stays.
This means that cross sections have the same pinch surfaces as amplitudes with the same physical
configurations, but with the energy flow reversed in the complex conjugate part. All vertices in both
2.9. UNITARITY AND CAUSALITY 63
A and A∗ are time ordered at any pinch surface, and the vertex with the largest time is always next
to the final stated (i.e. the cut). This vertex, w, will then satisfy
w0 > x0i and w0 > yj0 (2.147)
for all {xi } ({yj }) being the vertices in A (A∗ ). We must sum over the two cuts, γ and γ 0 , for which
w is in A or A∗ respectively (see Fig. 2.26). The relevant terms in this sum giving pinch surfaces
when integrating in w are (here V is just a vertex operator)
Z Y 1 Y 1
dD w w 2 1−
[−iV (∂ w )] w 2 0 − x0w ))1−
+
j
((y j − w) − iη) i
((x i − w) + iη(w i
Y 1 Y 1
+ w 2 + iη(y 0w − w 0 ))1−
[iV (∂ w )] w 2 + iη)1−
. (2.148)
j
((y j − w) i i
((x i − w)
So that whenever we have pinch surfaces (when any pair xw w
i and yj become collinear to w on the
lightcone) giving rise to infrared divergences, the sum vanishes because w has the largest time. This
is how IR divergences cancel in coordinate integrals for an inclusive cross section.
γ γ0
xw
1
w
y1 •
•
•• •
• • ••
w
•w •
w
x m yn
A A∗
Figure 2.26: Cuts that are adjacent to the vertex with largest time w. The superscripts w refer to
the vertices that are connected to w by lines.
2.9 Unitarity and Causality

M. Veltman notes in [20] that the LTE is the expression of causality in the Bogoliubov’s sense.
It is also interesting to recall the expression of the LTE in (2.84), which stated that the imaginary
part of an arbitrary connected coordinate-space G({xa }) diagram (times i) was proportional to the
sum over all cuttings or shadings of the diagram (here {xa } denote the external points). It is quite
striking the similarity between the expression of causality in coordinate-space diagrams with the
expression of unitarity such as (2.102). In fact the mathematical shape of these equations is identical
except for a minus sign and the meaning of the cuts is different. This topic seems rather interesting
and will be left for future research due to time limitations. I just want underline how two concepts
that are of fundamental importance in field theory are so closely related. I have the feeling that a
bit of research in this line will for sure make this enlightening connection possible.
It would be also interesting to carry out a systematic analysis on which particular space configura-
tions of vertices in coordinate-space TOPT give rise to imaginary parts in the amplitudes. Already
in the scalar self-energy graph we can see that the two collinear configurations had different powers
in the diverging denominators, changing the way the iη prescription would produce an imaginary
part.
Conclusions
After this long journey through the coordinate-space description, the impression left is that it pertains
a wide variety of new and intuitive interesting perspectives about QFTs.
The factorization results we review in coordinate-space come along pretty much in the same way as in
momentum-space and are quite enlightening. Showing clearly the eikonalization of soft and collinear
gluon emissions in coordinate-space. In this context we have computed several contributions to the
one-loop jet function and its radiated version, finding a very nice result for the non radiated case. Here,
a collinear divergence arose in the shape of a fermion propagator “dressed” with the collinear gluon
producing a simple pole in . This result was easily transported to momentum-space by means of LSZ
reduction, matching perfectly with previous calculations. For Wilson-line correlators, we analyzed
how a very clear causality picture of their imaginary parts emerges in coordinate-space.
Next, after introducing the LTE and seeking for further interpretations on the Coleman-Norton (CN)
picture, we presented TOPT in momentum-space. It succeeded in connecting the on-shell classical
propagation of intermediate states with the CN view of free classical propagation of particles in di-
vergent configurations. Also, a stationary phase approximation showed that the Feynman parameters
contain information of the ratios of flying times versus on-shell energies of intermediate state particles.
This was again in deep connection with the CN picture where hard lines had a null α parameter associ-
ated with a very huge on-shell energy (nonetheless, in the CN picture hard particles were off-shell) or
an almost vanishing flying time in TOPT. Aiming at finding the dual picture of TOPT in momentum-
space, we developed an original algebraic proof of TOPT in coordinate-space to all orders. Here we
took a general connected scalar Green’s function and, by introducing auxiliary energies and expressing
the graph as a sum of energy flows (in connection with the LTE), we carried out the time-component
integrals of all internal vertices in the diagram. This produced denominators that vanished in specific
space configurations of the internal vertices. When analyzing examples, we found a nice correspon-
dence with the momentum-space TOPT results in the case of the self-energy graph, where the power
counting of divergencies matched the known results, and we studied the scalar vertex graph finding the
divergent collinear configurations of vertices that we expected from the CN picture. In this last case
the power-counting required that vertices should have the exact same configurations as in the CN in
order to have a logarithmic divergence, i.e. the shrinking of hard lines. We were also able to extend our
treatment to fermionic lines, reproducing again the correct power-counting of collinear and hard sin-
gularities. However in this section a connection between the external points in coordinate-space with
incoming momenta lines was underlined as crucial in order to make the full connection with the CN
picture. We also studied briefly the alternate description in lightcone coordinates, LOPT, and noted the
differences with TOPT. Finally we saw how IR cancellations occur in coordinate space full inclusive
cross sections and discussed in a shallow manner the connection between causality and unitarity.
64
Appendices
A. QUADRATIC CASIMIRS i
A Quadratic Casimirs
In this section we will present some of the calculations carried out to obtain the quadratic Casimirs
of different Lie groups. These quadratic Casimirs are elements in the Lie Algebra which commute
with all the other elements.
We will focus on the Casimir invariant in the fundamental representation of the group G, CF δ ij =
(T a T a )ij , and the Casimir invariant in the adjoint representation, CG δ ab = f acd f bcd . Normalization
of the algebra generators is chosen as T r(T a T b ) = 21 δ ab .
A.1 Special Unitary Groups SU (N )

We start with the special unitary family SU (N ). Its generators T a are traceless hermitian. Therefore
every Hermitean N × N matrix A can be written as,
A = A† = c0 I + ca T a . (A.1)
From this we find

T r(A)
c0 = ca = 2 T r(AT a ) . (A.2)
N
Having then
1
Aij = Alk δ li δ kj = Alk 2(T a )ij (T a )kl + δ kl δij (A.3)
N
1
⇒ Alk 2(T a )ij (T a )kl + δ kl δij − δ li δ kj = 0 . (A.4)
N
Since A is arbitrary, we find for the generators the useful relation
1 1
(T a )ij (T a )kl = δli δkj − δkl δij . (A.5)
2 N
Contracting j and k we obtain the fundamental representation Casimir or Color Factor
a a 1N2 − 1
(T T )ij = δij = CF δij . (A.6)
2 N
Now we compute the following combination,
1 b 1
(T a )ji (T b )jk (T a )kl = (T )jk δli δ kj − (T b )jk δlk δij
2 N
1
=− (T b )il . (A.7)
2N
Noting the following identity and using the results already computed, we obtain the adjoint Casimir
for SU (N ),

acd bcd a c b c
f f = −2 T r [T , T ][T , T ]

= −2T r 2T a T c T b T c − (T a T b + T b T a )T c T c
= N δ ab = CG δ ab . (A.8)
ii APPENDICES
A.2 Special Orthogonal Groups SO(N )

We will follow now the same steps for the Special Orthogonal family SO(N ). Its generators are
antisymmetric and traceless and they form a basis for the antisymmetric N × N matrices. Thus,
taking an arbitrary antisymmetric matrix A, we have

A = −AT = ca T a ⇒ ca = 2T r AT a . (A.9)
Then we have
1
Aij = Akl δik δjl = Akl δik δjl − δjk δil = Alk 2(T a )ij (T a )kl . (A.10)
2
Finding
h1 i
Akl δik δjl − δjk δil + 2(T a )ij (T a )kl = 0 , (A.11)
2
and since A is an arbitrary antisymmetric matrix we get
a a kl 1 k l k l

(T )ij (T ) = δ δ − δi δj . (A.12)
4 j i
Here, since the group is real there is no need for distinction between upper and lower indices.
Contracting in the previous expression j with k we obtain the Color Factor
1 j l N −1
(T a )ij (T a )jl = δj δi − δij δjl = δil = CF δil . (A.13)
4 4
As before, we compute
1 b jk
(T a )ij (T b )jk (T a )kl = (T ) δil δkj − (T b )jk δik δlj
4
1 1
= − (T b )li = (T b )il . (A.14)
4 4
We are able now to obtain the adjoint Casimir for SO(N ). Similar to (A.8)
acd bcd
1
a b N − 1
a b b a
f f = 2 T r T T − (T T + T T )
2 4
1 ab ab
= (N − 2)δ = CG δ . (A.15)
2
A.3 Simplectic Groups Sp(N )

The elements M ∈ Sp(N ) (with N even) are N × N matrices which preserve the antisymmetric
tensor !
0 IN ×N
Ω= 2 2
, (A.16)
−I N × N 0
2 2
in the sense
Ω = M T Ω M ⇒ M −1 = ΩT M T Ω . (A.17)
A. QUADRATIC CASIMIRS iii
Using this relation it is possible to prove that the generators of the group take the form

a T a T a A B
− T = Ω (T ) Ω ⇒ T = , (A.18)
C −AT
where B and C are symmetric matrices. It is now possible to show that the generators satisfy
1
(T a )ij (T a )kl = δil δjk + Ωik (Ω−1 )jl . (A.19)
4
Therefore
1
(T a )ij (T a )jl = (N + 1)δij = CF δij . (A.20)
4
Noticing ΩT = Ω−1 = −Ω, we compute the usual combination
a b jk a 1 b jk b jk T

(T )ij (T ) (T )kl = δil (T ) δjk + Ωik (T ) (Ω )jl
4
1 1
= (ΩT )ik ((T b )T )kj Ωjl = − (T b )il , (A.21)
4 4
where in the last equality we have used (A.18). The adjoint Casimir now falls down easily
1 N + 1
f acd f bcd = −2 T r − T a T b − (T a T b + T b T a )
2 4
1 ab ab
= (N + 2)δ = CG δ (A.22)
2
To obtain the Color Factors and adjoint Casimirs for G2, F 4, E6 and E7 we refer to the article of
P. Cvitanović [?]. The results obtained are presented in Table I.
Group Color Factor (CF ) Adjoint Casimir (CG ) Nc

1
SU (N ) 2
N − N1 N ∀N ∈ N

1 1
SO(N ) 4
N −1 2
N −2 ∀N ∈ N

1 1
Sp(N ) 4
N +1 2
N +2 N = 2n ∀n ∈ N

1 1
G2 4
N −3 2
N −3 N =7

1 1
F4 18
N −8 18
N +1 N = 26

1
E6 12
N − 29
3
1
12
N −3 N = 27

1 1
E7 48
N +1 48
N + 16 N = 56
Table A.1: Quadratic Casimirs for different Lie Groups.

iv APPENDICES
B Massless Scalar Propagator in coordinate-space

The free propagator for a massless scalar field φ in D = 4−2ε space-time dimensions in coordinate-
space is (using the mostly positive Minkowski metric)
dD p eip(x−y)
Z
∆F (x − y) = h0| T {φ(x)φ(y)} |0i = −i . (B.1)
(2π)D p2 − iη
R∞
Using that i/A = 0 dteitA with Im(A) > 0, we get the expression
∞
dD p
Z Z
2 −iη)+ip(x−y)
∆F (x − y) = dt e−t(p . (B.2)
(2π)D 0
Now we perform a Wick rotation p0 → ip0 (x0 → ix0 ) and completing squares to carry out
gaussian integrations in the momentum to get (E subscript refers to euclidean signature)
D Z ∞ D Z ∞
iπ 2 −D −t−1
(x−y)2E iπ 2 D s 2
∆F (x − y) = dt t2 e 4 = ds s 2 −2 e− 4 (x−y)E
(2π)D 0 (2π)D 0
D ∞ D ∞
41−ε
Z Z
iπ 2
−ε − 4s (x−y)2E iπ 2
= ds s e = dr r− e−r . (B.3)
(2π)D 0 (2π)D [−(x − y)2 ]1−ε 0
Using the definition of the Euler gamma function we finally find that i times the free massless
propagator is (restoring the regulator −iη),
Γ(1 − ε) −1 Γ(1 − ε) 1
i∆F (x − y) = 2−ε 2 1−ε
= 2−ε
. (B.4)
4π [−(x − y) − iη] 4π [(x − y) + iη]1−ε
2
Figure B.1: Massless free scalar Propagator. We can observe that outside the lightcone the propa-
gator is heavily dumped in both spacelike and timelike regions.
C. MASSIVE PROPAGATOR IN 3+1 DIMENSIONS IN COORDINATE-SPACE v
C Massive propagator in 3+1 dimensions in coordinate-space

Now we will derive the expression for the massive propagator in 3+1 spacetime dimensions. The
massive scalar free propagator is
d4 p −ieip(x−y) dp0 d~p −iei~p(~x−~y)−ip0 (x0 −y0 )

Z Z
∆F (x − y) = = . (C.1)
(2π)4 p2 + m2 − iη (2π)4 −p20 + p~ 2 + m2 − iη
p the denominator of integrand in (C.1) has two simple poles for p0 = ±(ωp − iη)
Observe that
(here ωp ≡ p~ 2 + m2 ). This fact can be seen from
(−p0 + ωp − iη)(p0 + ωp − iη) = −p20 + (ωp − iη)2 ' −p20 + p~ 2 + m2 − 2iωp η , (C.2)
where we can replace 2iωp η with simply iη since we just need a positive infinitesimal number for
the regulator. To compute the integral in dp0 we will employ Cauchy’s residue theorem. If x0 > y0
we will enclose the integration contour in the lower half complex plane since the exponential factor
e−ip0 (x0 −y0 ) becomes small for large negative imaginary part of p0 . Therefore, in this case, we pick
up the residue of the pole at p0 = ωp − iη. In case x0 < y0 we enclose the pole in the upper half
complex plane getting the residue of the other simple pole at p0 = −ωp + iη taking into account a
minus sign from the clockwise contour integral. Therefore the integration in dp0 yields
ei~p(~x−~y) ei(y0 −x0 )(ωp −iη) ei(x0 −y0 )(ωp −iη)

Z
3
∆F (x − y) = d p~ Θ(x0 − y0 ) + Θ(y0 − x0 )
(2π)3 2(ωp − iη) 2(ωp − iη)
x−~y ) −i|x0 −y0 |(ωp −iη) Z ∞
i~
p(~
dp p sin(p|~x − ~y |)e−i|x0 −y0 |(ωp −iη)
2
Z
e e
= d3 p~ = , (C.3)
(2π)3 2(ωp − iη) 0 4π 2 p|~x − ~y |(ωp − iη)
where in the last equality we performed spherical coordinates integrations.
Now we would like to study the behavior of the propagator for timelike and spacelike separations.
In the first case we can choose a reference frame where (x − y)µ = (x0 − y 0 , ~0) and noting that
limx→0 sin x/x = 1, the propagator becomes
Z ∞ √
2 −i|x0 −y0 |( p~ 2 +m2 −iη)
dp p e
∆F (|x0 − y0 |) = p . (C.4)
0 4π 2 ( p~ 2 + m2 − iη)
(1)
This result is well known and it is proportional to a Hankel function H1 (m|x0 − y0 |) which de-
scribes an outgoing wave at long lightlike distances (see Figure C.1a). For the spacelike separation
we can choose a frame where (x − y)µ = (0, ~x − ~y ) getting
Z ∞
dp p sin(p|~x − ~y |)
∆F (|~x − ~y |) = 2
p . (C.5)
0 4π |~x − ~y |( p~2 + m2 − iη)
This is proportional to the modified Bessel function K1 (m|~x − ~y |), which is proportional to the
Hankel function but with complex arguments. Therefore instead of an oscillatory behaviour we
will obtain exponentially damped at great spacelike distances (see Figure C.1b). It is worth noting
that for lightlike separations (x − y)2 = 0 the propagator can only be computed in the massless
limit and its proportional to a δ(x2 ).
vi APPENDICES
(a) Timelike Propagator (b) Spacelike Propagator
Figure C.1: Behaviour of the Timelike and Spacelike Propagators. We can see that the spacelike
propagator is heavily damped, while the timelike propagator has an oscillatory nature and is much
less suppressed than the spacelike one. The timelike propagator will have an asimptotic oscillatory
behaviour since the Hankel function is H1 (x) = J1 (x) + iN1 (x) where J1 is the first Bessel func-
tion, (which has an asimptotic behavior as cos x) and N1 the first Neumann function (which has an
asimptotic behavior as sin x).
D TOPT propagators for fermions

In the TOPT treatment in coordinate-space we needed to carry out the zj0 integrations for each line
after introducing the unity in form of an integration in the auxiliary energy Ej and a delta function
δ(zj0 −jk yk0 −0jc x0c ). The contribution from fermion lines will be proportional to the integral
0 0 0 0 0
z µ δ(zj0 − jk yk0 − 0jc x0c ) +∞ dEj0 zjµ eiEj (zj −jk yk −jc xc )
Z Z Z
0 j
Ijµ = dzj 2 = dzj0 .
(−zj0 + ~zj2 + iη)2 −∞ 2π (−zj0 + ~zj + iη)2 (zj0 + ~zj + iη)2
(D.1)
We see that the integrations in the zj0 will pick up, after proper closing of the contour of integration,
the residues of the two double poles zj0 = ±(|~zj | + iη). These are:
• For spatial components of the numerator the integrand f has residues
iE (2|~z | + 2iη) ∓ 2 0 0 0
j j
Res(f, ±(|~zj | + iη)) = θ(±Ej )zji 3
e±iEj (|~zj |∓jk yk ∓jc xc ) (D.2)
(2|~zj | + 2iη)
• For the time component of the numerator

iE (2|~z | + 2iη) 0 0 0
j j
Res(f, ±(|~zj | + iη)) = θ(±Ej )|~zj | 3
e±iEj (|~zj |∓jk yk ∓jc xc ) . (D.3)
(2|~zj | + 2iη)
E. FACTORS IN THE ONE-LOOP RADIATED JET FUNCTION CONTRIBUTION vii
In this way, taking into account the sign of the orientation for the contour integrals and dropping
iηs in numerators,
3
zji
Z +∞ hX
µ 0 0 0
Ij = i dEj0 δ µi 3
θ(Ej )(2i|~zj |Ej − 2)eiEj (|~zj |−jk yk −jc xc ) +
−∞ i=1
(2|~zj | + 2iη)
0 0 0
|~zj |
+ θ(−Ej )(−2i|~zj |Ej − 2)eiEj (|~zj |−jk yk −jc xc ) + δ µ0 ×
(2|~zj | + 2iη)3

zj |−jk yk0 −0jc x0c )
iEj (|~ zj |−jk yk0 −0jc x0c )
iEj (|~
× θ(Ej )(2i|~zj |Ej )e + θ(−Ej )(−2i|~zj |Ej )e . (D.4)
This becomes, after defining the lightlike vector ẑ µ = (|~zj |, ~zj ),
iẑjµ
0 ∂
X 3
Ijµ = 2ẑj − 2 δ µi
×
(2|~zj | + iη)3 ∂ ẑj0 i=1
dEj0
Z +∞
0 0 0 0 0 0 0 0
× θ(Ej0 )e−iEj (jk yk +jc xc −|~zj |−iη) + θ(−Ej0 )e−iEj (jk yk +jc xc +|~zj |+iη) . (D.5)
−∞ 2π
This expression is ready to employ all the know results of scalar TOPT in coordinate-space.
E Factors in the one-loop radiated jet function contribution

Here we present the factors that appear in (2.71):
M 2 = m2 − ( + (1 − α5 n))2 α−2 (E.1)
h
/ −1 + y/) n
B µ = (α / (1 − α2 )/x − α1 α2 α−1 x − α1 α2 α−1 (
/ γ µ (1 − α2 )/ / + (1 − α5 )/
n) −
i
µ 2 2
− α2 n
/ γ α1 (1 − α1 α2 )( + (1 − α5 )n)) + (1 − α2 )x + 2α1 (1 − α2 )x · ( + (1 − α5 )n))

+ α−1 ( + (1 − α5 )n)2 n/ + y/n
/ x − α−1
/ γ µ (1 − α2 )/ / + (1 − α5 )/
n) (E.2)

Aµρν = −α1 α2 α−2 (1 − α1 α2 )(2( + (1 − α5 )n)ν ) + 2α−1 (1 − α2 )/ /γ µ+
n γρ n

/ −1 + y/) α1 α2 (1 − α1 α2 )/
+ (α / (α1 α2 )2 α−1 γρ γ µ γν −
nγ µ gρν + n

/ α1 α2 α−1 γν γ µ (1 − α2 )/ x − α1 α2 α−1 (

− γρ n / + (1 − α5 )/ n)

x − α1 α2 α−1 / γ µ (α1 α2 α−1 γν ) −

+ (1 − α2 )/

− α−1 gρν n
/ γ µ
(1 − α 2 )/
x − α α
1 2 α −1 /
( + (1 − α 5 )/
n ) +

+ α1 α2 2α−2 ( + (1 − α5 )n)ρ n / + α−1 y/n
/ γρ γ µ γν (E.3)
Bibliography
[1] D. Bonocore, Next-to-soft factorization and unitarity in Drell-Yan processes.

http://hdl.handle.net/11245/1.532061 (2016).
[2] I.T. Todorov, Analytic Properties of Feynman Diagrams in Quantum Field Theory. Interna-
tional Series of Monographs in Natural Philosophy Volume 38, Pergamon Press (1971).
[3] S. Coleman and R.E. Norton Singularities in the physical region. N. Cim. 38 438-442 (1965).
[4] G. Sterman, An Introduction to Quantum Field Theory. Cambridge University Press (1993).
[5] G. Sterman, Mass divergences in annihilation processes. I. Origin and nature of divergences
in cut vacuum polarization diagrams. Phys. Rev. D, v- 17 n. 10, 2773-2788 (1977).
[6] D. Bonocore, E. Laenen et al., A factorization approach to next-to-leading-power threshold
logarithms. arXiv:1503.05156v2 (2015).
[7] J.C. Collins, Sudakov Form Factors. Adv.Ser.Direct.High Energy Phys., 573-614, (2003).
[8] G. Sterman, Partons, Factorization and Resummation. TASI 95 (1996).
[9] M. Nakahara, Topology, Geometry and Physics. Graduate Student Series in Physics (2003).
[10] L.J. Dixon, E. Gardi, L. Magnea, On soft singularities at three loops and beyond. High Energ.
Phys. (2011).
[11] S. Catani, D. de Florian, and G. Rodrigo, Space-like (versus time-like) collinear limits in
QCD: Is factorization violated?. High Energ. Phys. (2012). arXiv:1112.4405.
[12] J. R. Forshaw, M.H. Seymour, and A. Siodmok, On the breaking of collinear factorization in
QCD. High Energ. Phys. (2012). arXiv:1206.6363.
[13] O. Erdogan, Coordinate-space singularities of massless gauge theories. P. Rev. D 89 (2014).
[14] G. Korchemsky and A. Radyushkin, Nucl. Phys. B283, 342-364 (1987).
[15] M. Veltman, Diagrammatica. Cambridge Lecture Notes in Physics 4, Cambridge University
Press (1994).
[16] E. Laenen, K.J. Larsen, and R. Rietkerk, Imaginary parts and Discontinuities of Wilson Line
Correlators Phys. Rev. Lett. 114, 181602 (2015).
[17] O. Erdogan, and G. Sterman; Path description of coordinate-space amplitudes, Phys. Rev. D
95, (2017).
[18] S. J. Chang, and S. K. Ma; Feynman rules and quantum electrodynamics at infinite momentum,
Phys. Rev. D 180 1506 (1969).
[19] O. Erdogan, and G. Sterman, A coordinate description of partonic processes. Proceedings of
Science, https://doi.org/10.22323/1.235.0027, (2017).
[20] M. Veltman, Unitarity and causality in a renormalizable field theory with unstable particles,
Physica, vol. 29, iss. 3, 186-207, (1963).

QCD Amplitudes in Coordinate Space

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

QCD Amplitudes in Coordinate Space

Hochgeladen von

Copyright:

Verfügbare Formate

Universiteit van Amsterdam

QCD Amplitudes in Coordinate Space

Alexandre Salas Bernárdez

Supervised by Professor Eric Laenen

July 10, 2019

Soft and collinear divergences in momentum-space 3

1.1 Landau Equations

which by the definition of D means

1.1.1 Coleman-Norton picture

∆xµj ≡ λαj ljµ (1.7)

∆xµj = 0 if lj2 6= −m2j .

Quark Electromagnetic Form Factor

dD k v̄(p2 )γ ν (p/2 + k/)γ µ (k/ − p/1 )γν u(p1 )

(a) (b) (c) (d)

(e) (f) (g) (h)

1.2 Infrared power counting

Quark Electromagnetic Form Factor

and the line and volume element are

ds2 = −2dx+ dx− + dx2⊥

The collinear singularity where (k − p1 )2 → 0 is achieved when k − and k⊥ become 0, so these

1.3 All-order quark electromagnetic form factor

1.4 Wilson Lines

If we expand now in powers of g

1.5 Factorization of the quark EM form factor

1.5.1 One loop soft and jet functions

Here in the third

2.1 Analysis of singularities

α1 z1µ + α2 z2µ + α3 z3µ = 0 . (2.5)

2.2 Power Counting

This overall degree of divergence γ is given by

− α1 (x − y)µ + α2 (y − z)µ = 0 (2.19)

2.2.1 Power counting for a single jet

v3g (zm , {y}) = ijk g µi µj ∆(zm − yi )∆(zm − yj )∂zµmk ∆(zm − yk ) . (2.21)

γJ ≥ 1 + V3f − Nf + O() . (2.25)

2.2.2 Overall power counting for the vertex function

γΓ ≥ 4(V3H + V4H ) − 2NgH − 3NfH − V3gH +

2N H = 2V2H + 3V3H + 4V4H − E H . (2.29)

2N J = 2 + J H − S J + 3V3J + 4V4J , (2.30)

2.3.1 Hard-collinear aproximation

hout| T {∂µ1 Aµ1 (x1 )...∂µn Aµn (x1 )} |ini = 0 (2.43)

2.3.2 Soft-collinear approximation

2.3.3 Eikonal approximation

2.3.4 One-loop Jet function in coordinate-space

(1) Γ(1 − )Γ2 (2 − )

α2 (x − y)2 + (1 − α2 )α1 y 2 + (1 − α2 )(1 − α1 )(y 2 − 2tn · y) = 0 (2.54)

(1) Γ(1 − )Γ2 (2 − ) Γ(5 − 3)

(1) Γ(5 − 3)

It is possible (taking y to be non-collinear with n so that n · y 6= 0, which is arguably not a possible

y · n = 0) at this point to carry out the integration in t, producing

2.3.5 One-loop jet function with internal gluon emission in coordinate-space

This diagram amounts to

−(w/ − z/) −/z 1 1

Here α = α3 α4 α5 + α5 (1 − α3 )α1 α2 (1 − α1 α2 ) + α5 (1 − α4 ), ()µ = (1 − α5 )nµ − α1 α2 (1 −

2.4 Imaginary parts of Wilson line correlators

Γ(1 − ε) ∞ ∞ dt1 dt2 v1 · v2

2.5 Largest time equation

• Each circled vertex will give a minus sign.

and can be reexpressed as

2.6 Time Ordered Perturbation Theory

2.6.1 Momentum-space TOPT

to the loop momentum variables, we get the equations

2.6.2 Unitarity and cut diagrams

Take now a general sum of TO diagrams, G,

Figure 2.16: General cut diagram Gγ of equation (2.95).

v3g (zm , {y}) = ijk g µi µj ∆(zm − yi )∆(zm − yj )∂zµmk ∆(zm − yk ) . (2.21)

γJ ≥ 1 + V3f − Nf + O() . (2.25)

(1) Γ(1 − )Γ2 (2 − )

(1) Γ(1 − )Γ2 (2 − ) Γ(5 − 3)

(1) Γ(5 − 3)

−iEj0 (0jc x0c −|~

ei~p(~x−~y) ei(y0 −x0 )(ωp −iη) ei(x0 −y0 )(ωp −iη)