1

Theory of Strong Interactions

M.A.Braun
Department of High-Energy Physics, St. Petersburg University,
198504 St. Petersburg, Russia

2007 ,
Corrected and complemented in 2020
 2

1 Contents

1. Introduction
2. General properties of scattering amplitudes
3. Analyticity from the local quantum field theory
4. Analytic Properties in the Perturbation Theory
5. Dispersion relations
6. The Froissart theorem
7. Complex angular momenta
8. Multiregge asymptotic and inclusive cross-sections
 3

2 Literature (In Russian)

1. D.Ch’yu, ”Analiticheskaya teoriya S-matritzy”, Moskva, Mir, 1968
2. R.Iden ”Stolknoveniya elementarnykh chastitz pri vysokikh energiyakh”, Moskva,
Nauka, 1970
3. P.Kollinz ”Vvedeniye v redzhevskuyu teoriyu i fiziku vysokikh energiy’, Moskva, At-
omizdat, 1980
 4

3 Introduction
The strong interaction has been known since the beginning of the 20th century, when the
compound structure of atomic nucleus was established. This interaction is responsible for
formation of the nucleus. It showed properties which radically distinguish it from the two
previously known interactions, gravitational and electromagnetic. The first property is that
it is a short-range interaction: it only acts between particles located at distances of the
order 10−13 cm. So it is absent in the world of macroscopic objects. Its second property is
that it is a strong interaction: the strong force acting between two protons at distance of
the order 10−13 cm is roughly a thousand times stronger that the electromagnetic repulsion
between them. Finally, unlike, say, the gravitational interaction, the strong interaction acts
not between all known particles, but between a group of them, called hadrons. Hadrons
include the well-known protons, neutrons, composite structures made of them, nuclei, and a
good many of newly discovered particles. However photons, electrons, muons, gravitons and
some other particles do not take part in the strong interaction.
According to our present knowledge all observed hadrons are in fact not purely elementary.
They are built of certain ’prehadrons’, called quarks. The quarks exist in 6 varieties, called
flavours. They can be grouped in three generations (u, d) (’up’ and ’down’), (c, s) (’charmed’
and ’strange’) and (t, b)(’top’ and ’bottom’, or ’beautiful’). All quarks are fermions with
spin 1/2 and positive parity (by definition). Quarks u, c and t have their electromagnetic
charge equal to (2/3)e, where e is the proton electromagnetic charge. Quarks d, s and b have
their electromagnetic charge equal to −(1/3)e.
In its turn, the quark of each flavour is supposed to exist in three states, which differ by a
certain new quantum number, colour. So the wave function of a quark has a structure qαf (x),
where flavour f = u, d, c, s, t, b and color α = 1, 2, 3. Sometimes, for illustrative purposes, the
three colour states are marked as ’red’, ’yellow’ and ’blue’, which, of course, has nothing to
do with the relevant colours in our everyday life.
It is the presence of colour which is supposed to generate the strong interaction. In a sense,
colour plays the role of charge for the strong interaction. However it is a more complicated
charge than the electromagnetic one. As we shall see roughly speaking it is a vector charge
as compared to the scalar electromagnetic charge. The strong interaction feels only color but
ignores flavour. So quarks of any flavour strongly interact absolutely in the same manner.
It is supposed that the strong interaction is invariant under unitary transformations of the
three color states of each flavoured quark performed by matrices Uαβ with det U = 1, which
form the group S(3)
f
q ′ α = Uαβ qβf , det U = 1 (1)

A peculiar property of quarks is that they cannot be observed as free particles. This
property (’confinement’) is supposed to originate from the fact that strong interaction between
coloured objects infinitely rises with the distance between them. So to split, say, a quark-
antiquark pair into its constituents one needs an infinite amount of energy. Alternatively
one may think that the quark and antiquark splitted from a colorless objects incorporate an
antiquark and quark from the vacuum and turn into two colorless objects. As a result any way
 5

only colourless objects, invariant under transformation (1), can be observed experimentally.
There are two evident ways to build a colourless object from coloured quarks. One is to
take a colourless object composed from a quark-antiquark pair
¯ ¯
M f,f = q̄αf qαf (2)

The resulting hadron has an integer spin and is called a meson. Mesons can have different
flavour contents f, f¯ and also different spins, which may come both from the intrinsic spins
of the pair (0 or 1) and from their relative rotation (orbital angular momentum). In this way
one predicts existence of a multitude of mesons with generally rising spin (and mass).
The second way to build a colorless state is to form a fully antisymmetric state of three
quarks
B f1 ,f2 ,f3 = ǫαβγ qαf1 qβf2 qγf3 (3)

Indeed after a transformation by the SU (3) group we have

B ′ = Uαα′ Uββ ′ Uγγ ′ ǫαβγ qαf1′ qβf2′ qγf3′ = ǫ′α′ β ′ γ ′ qαf1′ qβf2′ qγf3′ (4)

where
ǫ′α′ β ′ γ ′ = Uαα′ Uββ ′ Uγγ ′ ǫαβγ (5)

It is trivial to see that ǫ′α′ β ′ γ ′ is totally antisymmetric in its indeces similar to ǫα′ β ′ γ ′ . So to
find it it is enpough to calculate its matix element (123). We find

ǫ′123 = Uα1 Uβ2 Uγ3 ǫαβγ = det U = 1 (6)

so that ǫ′ = ǫ and consequently B ′ = B and is a colourless state.

This hadron has a semi-integer spin and is called baryon. Again one can have a multitude
of baryons depending on the three chosen flavours and resulting spin.
To discuss the hadrons observed in laboratory we have to say a few words about the quark
masses. Since the free quarks cannot be seen, the notion of their mass is not completely clear.
One may discern two sorts of mass, the current mass, which enters into the Lagrangian of
the strong interaction , and constituent mass, which may be estimated from the mass of the
composite hadron, provided the binding energy is not too large. Leaving aside the complicated
theoretical deliberations, we first mention that for the so-called heavy quarks, c,b and t, the
two definitions practically coincide and qive values 1.5, 4.5 and 180 GeV respectively. For
the rest, light quarks, the current and constituent masses are found quite different. For u,d
and s quarks the current mases are 4, 7 and 150 MeV, respectively, while their constituent
masses are 300, 300 and 450 MeV, respectively.
However, whatever the definition, there is a strong difference in masses between light and
heavy quarks, and, as a consequence, between hadron built of exclusively light quarks and
those which include at least one of the heavy quarks. Naturally most common hadrons are
those built of light quarks, which either exist in nature as such or can be produced already
at small accelerator energies.
The most known hadrons are of course the nucleons, the proton, p=uud, and neutron,
n=udd, with spin 1/2 and masses mp ≃ mn = 940 MeV. They are building elements for all
 6

nuclei. The proton is stable and it is the only stable hadron. The free neutron decays by the
weak interaction into p+e+ν̄. Substituting u or d by s-quarks one obtains a family of hyperons
with spin 1/2: Λ0 =uds (m=1115 Mev),Σ0,± =uus,uds,dds (m=1190 MeV), Ξ0,− =uss,dss
(m=1315 MeV). All of them decay by the weak interaction. More baryons are obtained
adding the three quark spins into 3/2, e.g. ∆−,0,+,++=ddd,udd,uud,uuu (m=1240 MeV), or
Ω− = sss (m=1670 MeV). The first decays by the strong interaction and so is visible only as
a resonance in the scattering amplitudes, the second decays by the weak interaction.
√
¯ ūd, 1/ 2(ūu+ dd)
The lightest meson is the triplet π 0,± = du, ¯ with spin 0, negative parity
and mass mπ = 140 MeV. This anomalously small mass allows to consider a theory with mass
mπ = 0 as a first approximation. In this limit new, chiral, symmetry arises, which allows
to make certain non-trivial predictions, which constitute the content of the so-called chiral
dynamics. Changing u or d quarks or antiquarks into s or anti-s one obtains a family of
K-mesons with mass m = 450 MeV. All these mesons are unstable, decaying by the weak
interaction, with the exception of π 0 which decays by the electromagnetic interaction and so
has a very short lifetime. Summing the spins of the quark-antiquark pair into 1, one obtains
√ √
spin 1 mesons ω 0 = 1/ 2(ūu − dd),¯ m=780 MeV, and ρ0,± = 1/ 2(ūu + dd), ¯ du,¯ ūd, m=760
MeV.
Special attention is to be drawn to spin 1 mesons with hidden non-trivial quantum num-
bers: φ0 = s̄s, m=960 MeV, J/ψ = c̄c, mass=3.1 GeV and Upsilon=b̄b, m=9 GeV. These
mesons are clearly visible in the cross-section for e+ e− annihilation. They served to discover
c and b quarks.
The strong interaction, as mentioned, is only a consequence of the presence of colour.
It does not feel the flavor. This means, that first of all, flavour is conserved in the strong
interaction. As a result we have 6 conserved quantities in the strong interaction: the number
¯ c − c̄, s − s̄, t − t̄ and b − b̄. Second, quarks of all flavours
of the differences u − ū, d − d,
strongly interact in the same way. This means that , if all quarks had the same mass, the
strong interaction would be symmetric under the unitary transformation of all the flavours,
which form the group SU (6)F . One may expect such a full flavour symmetry at energies
much higher than any quark mass. At present energies this symmetry is of course badly
broken by quark masses. However there remains a flavour symmetry SU (2)F due to the
approximately equal constituent masses of u and d-quarks, the isospin symmetry, well known
in the low-energy nuclear physics.
The hadrons are not point-like particles but have a finite spatial extention. Because of
that, in spite of being colorless, they interact strongly due to the fact that strong interaction
between their constituents is not fully compensated. However the resulting strong interaction
between hadrons turns out to be short-ranged.
The fact that the fundamental entities for the strong interaction, quarks, are not observed
experimentally makes the study of the strong interaction very difficult. In fact we observe
only its non-compensated remainder of the interactions inside the observed hadrons. Also
the very strength of the interaction does not allow to indiscriminately use the perturbation
theory, which remains, up to the present, the only effective tool to get predictions from
the quantum field theory. Happily it has been discovered that the strength of the strong
 7

interaction depends on the distance between the interacting particles and goes down (albeit
slowly) as this distance diminishes. So very strong at nuclear distances, the strong interaction
becomes weaker at smaller distances, or transferred momenta much larger that the typical
nuclear ones (of the order 0.2 ÷ 0.3 GeV/c). This opens a possibility to use the perturbation
theory for process dominated by these small distances or large transferred momenta (’hard
processes’)
Accordingly the theory of strong interactions can be split into three parts, which study
different regions of transferred momenta. At small energies and thus transferred momenta
the dominant processes are elastic scattering of protons and neutrons and formation and
scattering of various nuclei. The complicated quark structure of hadrons is irrelevant. So the
basic tool is just the ordinary (non-relativistic) quantum mechanics with certain potentials,
which describe the interaction of nucleons. The upper energy, which allows such a treatment,
is fixed by the possibility of pion production. In the next energy range up to 4 ÷ 5 GeV, the
quark structure is still not felt, but the spectrum of hadrons which are formed is becoming
numerous. Most of them are seen as resonances in the scattering amplitudes. So the standard
tool is to approximate the process by just creating a particle-resonance and its subsequent
decay. Of course this leads to a very phenomenological description of the interaction, which
uses the experimental information about the position and lifetime of various hadrons, which
appear as possible resonances.
A more fundamental approach can be taken at very high energies. Here one can separate
a whole class of processes or events in which hadron and their constituent quarks are brought
to very small relative distances. These hard processes can be treated by the standard per-
turbation method applied to the quantum field theory which supposedly describes the strong
interaction, the Quantum Chromodynamics (QCD). To such processes belong: e+ e− annihi-
lation into hadrons, the so-called deep inelastic e-p scattering (DIS), scattering of hadrons
at large angles etc. Unfortunately a wide class of events is dominated by large distances
between participants (’soft processes’) or involve, apart from small distances, also large ones.
Then one cannot make predictions based only on the fundamental theory, but has instead,
to bring in some amount of phenomenological material.
Our course will consist of two main parts. In the first we shall study some general prop-
erties of strong interaction (any interaction in fact), which constitute a basis for both purely
phenomenological treatment and perturbative approach. In this part the sort of interact-
ing particles, as well as underlying quantum field will be irrelevant. For simplicity we shall
consider fictitious neutral scalar particles with properties resembling those of the physical
nucleons and pions in that the mass of our nucleon is taken to be much larger than that of
our pion and that the interactions proceed by emission of a pion from nucleon: N→N+π.
The second part of our course will be wholly devoted to QCD and its application to the study
of hard processes in the strong interaction.
 8

4 General properties of scattering amplitudes

4.1 Cross-sections and amplitudes
Consider a process of collision of a pair of particles with momenta k1 and k2 in which in
the final state n ≥ 2 particles are produced with momenta ki′ , i = 1, 2, ...n. For simplicity
we assume all particles to be scalar (spinless) but admit some of them to belong to different
species. The differential cross-section for this process is given by
1  2
dσ(ki′ ) = A(ki′ |k1 , k2 ) .dτn′ (k1 + k2 ) (7)

J12
Here q
J12 = 4 (k1 k2 )2 − m21 m22 (8)

is the invariant flux, m1 and m2 are the masses of colliding particles, dτn′ is the invariant
phase volume for the final particles restricted by conservation of energy-momentum:
n n
1 d3 ki′
dτn′ (k1 + k2 ) = (2π)4 δ4 ( ki′ − k1 − k2 )
X Y
, (9)
i=1
n1 !n2 !...ns ! i=1 (2π)3 2ki0

where the factorial factor refers to identical particles of the total number of s different species,
ni = n, and the smooth function A is the scattering amplitude, which will be our main
P

object throughout these lectures.

The amplitude is related to the scattering matrix S by the relation
n
S(ki′ |k1 , k2 ) =< Ψout in ′ 4 4
ki′ − k1 − k2 )A(ki′ |k1 , k2 ). (10)
X
′ |Ψk1 ,k2 >= δ(ki |k1 , k2 ) + i(2π) δ (
k ′ ,...kn
1
i=1

Here
q 1 3 
δ(ki′ |k1 , k2 ) = δ2n (2π)6 16k10 k20 k10
′ k′
20 δ (k1 −k1′ )δ3 (k2 −k2′ )+δ3 (k1 −k2′ )δ3 (k2 −k1′ ) . (11)
2
is the relativistic invariant δ-function, that is the unit matrix in our basis.
An important particular case is when n = 2 and we have what is called a binary reaction.
Then
1 k′
dτ2 (k1 + k2 ) = dΩ′ , (12)
16π 2 W
where W , k′ and Ω′ are the energy, momentum and solid angle of the final particles in the
center-of-mass (c.m.) system.
The cross-section (7) is called exclusive. It refers to the experimental setup in which
momenta of all final particles are measured. In fact, with a multitude of particles produced
at high energies, this cross-section is difficult to obtain and moreover, if found, it depends on
so many variables that its theoretical analysis is hopeless. For this reason a more realistic
setup is to fix momenta of a few final particles allowing for the rest to have any momenta.
A typical example is the single inclusive cross-section, which implies that in the experiment
only the momentum of a single particle of a given sort is measured. Such an experiment is
denoted as a process
1 + 2 → 1′ + X, (13)
 9

where X means all the unobserved particles. The single inclusive cross-section is expressed
via the amplitude by the formula

1 d3 k1′ X 2
Z 
dσ(k1′ , X) = dτ ′
(k + k − k ′ 
) A(k ′
, X|k , k ) . (14)

3 ′ n 1 2 1 1 1 2
J12 (2π) k10 n


The maximally inclusive cross-section is the total one, which experimentally is determined
just by the decrease in the incoming flux:
1 X 2
Z 
tot
dσ = dτn′ (k1 + k2 )A(ki′ |k1 , k2 ) . (15)
 
J12 n

4.2 Relativistic invariance

Let U (Λ) be the unitary operator acting on the states which corresponds to the general-
ized Lorentz transformation Λ (that is including spatial rotations). Then Lorentz-invariance
implies that

< Ψout in out

′ |Ψk1 ,k2 >=< Ψk ′ ,...k ′ |U
k ′ ,...kn n
−1
(Λ)U (Λ)Ψin out
k1 ,k2 >=< ΨΛk ′ ,...Λkn
in
′ |ΨΛk1 ,Λk2 > . (16)
1 1 1

As a result we find
A(Λki′ |Λk1 , Λk2 ) = A(ki′ |k1 , k2 ). (17)

That is the amplitude does not change if all momenta are Lorentz-transformed. Of course, if
particles have spins one has additionally to rotate the spins.
Apart from relativistic invariance the amplitude is symmetric under a few discrete trans-
formations. For the strong interaction these are C (charge conjugation), P (parity) and T
(time reversal). The first two are trivially derived similar to the Lorentz-invariance. Charge
conjugation C consists in changing all particles (initial and final) into antiparticles and vice
versa. Parity P consists in reflecting all three spatial axes, under which the momentum k
changes sign but spin projection σ does not

P : (k, σ) → (−k, σ). (18)

Somewhat non-trivial is the time reversal, which changes sign of both momenta and spin
projections, interchanges the initial and final states and also introduces a certain phase factor
for particles with spins:

T : |Ψin out
(−1)si −σi .
Y
k1 ,σ1 ,... >→< Ψ−k1 ,−σ1 | (19)
i

Here si is the spin of the i-th particle. As a result of T -invariance we find (for spinless
particles)
A(k′i |k1 , k2 ) = A(−k1 , −k2 | − k′i ) = A(k1 , k2 |k′i ). (20)

The second equality follows from the P-invariance. So combined PT invariance implies that
the amplitude does not change if we interchange initial and final states (for spinless particles).
Note that this relation has practical sense only if n = 2, that is for binary reactions. Otherwise
 10

Figure 1: Binary amplitude

we relate the initial amplitude with a non-physical one, which corresponds to the initial state
containing more than two particles.
Relativistic invariance implies that the amplitude is an explicit functions of scalar products
of 4-momenta of all participating particles (initial plus final). However one has to remember
that the number of such products is generally larger than the number of independent variables.
If the total number of participating particles is N = 2+n then the total number of independent
variables is 3N − 10. Indeed the total number of independent 3-momenta is 3N , which
is restricted by the energy-momentum conservation to 3N − 4. We have then 6 Lorentz
transformations which allow to put one of the momentum equal to zero, direct another
momentum along the z axis and the third in the xz plane, which removes 6 variables. This
means that between the scalar products of 4-momenta there exist many algebraic relations.
The important case as before is the binary reaction with n = 2 and N = 4 (a four-legged
amplitude). It depends on 3 · 4 − 10 = 2 independent variables, for which one stndardly
chooses the energy and scattering angle in the c.m. system. However to make the rela-
tivistic invariance explicit it is convenient to introduce different variables (”the Mandelstam
variables”). s, t and u, which are standardly defined as follows (see Fig. 1)

s = (k1 + k2 )2 = (k1′ + k2′ )2 , t = (k1 − k1′ )2 = (k2 − k2′ )2 , u = (k1 − k2′ )2 = (k2 − k1′ )2 . (21)

They are three, so one expects to find a relation between them. Indeed one finds
2 2
s + t + u = m21 + m22 + 2k1 k2 + m21 + m′1 − 2k1 k1′ + m21 + m′2 − 2k1 k2′
2 2 2 2
= 3m21 + m22 + m′1 + m′2 − 2k1 (k1′ + k2′ − k2 ) = m21 + m22 + m′1 + m′2 .

So the sum of s, t and u is equal to the sum of mass squares of participating 4 particles.
2 2
s + t + u = m21 + m22 + m′1 + m′2 (22)

The Mandelstam variables have a transparent physical meaning. In the c.m. system
k1 + k2 = 0 and s = W 2 , that is the square of c.m. energy. Obviously for the process to take
place s has to lie above the threshold

s > max{(m1 + m2 )2 , (m′1 + m′2 )2 }. (23)

The meaning of t is clearer for elastic processes, when the masses of particles do not change:
m′1 = m1 and m′2 = m2 . Then in the c.m. system the momentum does not change its
 11

magnitude and only rotates: |k′ | = |k|. Then k10 = k10

′ and t = −(k − k )2 , that is the
1 2
transferred momentum squared with a minus sign. In terms of the scattering angle
t = −2k2 (1 − cos θ) (24)

and so is negative and satisfies

−4k2 < t < 0. (25)
Variable u has a less transparent meaning, since it includes a contribution from the non-equal
zero components. Its limits can be derived from (22)
q
2(m21 + m22 ) − ( (k2 + m21 )(k2 + m22 ) < u < (m1 − m2 )2 . (26)

Inequalities (23), (24) and (25) restrict what is known as the physical region for the elastic
scattering process
a(k1 ) + b(k2 ) → a(k1′ ) + b(k2′ ). (27)

4.3 Unitarity
In terms of the scattering matrix it is formulated as
SS † = S † S = 1. (28)

Taking these operator relations between two-particle scattering states < k1′ , k2′ | and |k1 , k2 >
we obtain for the corresponding amplitudes
XZ
iA∗ (k1 , k2 |k1′ , k2′ ) − iA(k1′ , k2′ |k1 , k2 ) = dτn (k1 + k2 )A∗ (n|k1′ , k2′ )A(n|k1 , k2 ). (29)
n
Using the PT invariance (for spinless particles) we find

A∗ (k1 , k2 |k1′ , k2′ ) = A∗ (k1′ , k2′ |k1 , k2 ),

which allows to rewrite (29) as
XZ
2 Im A(k1′ , k2′ |k1 , k2 ) = dτn (k1 + k2 )A∗ (n|k1′ , k2′ )A(n|k1 , k2 ). (30)
n
This general unitarity relation gives for the forward scattering
XZ  2
2 Im A(k1 , k2 |k1 , k2 ) = dτn (k1 + k2 )A(n|k1 , k2 ) = J12 σ tot , (31)
 
n
which is the famous ’optical theorem’. Note that this relation allows to find the fully inclusive,
total cross-section from the knowledge of only the 2 → 2 forward elastic amplitude and thus
avoiding to sum all the exclusive amplitudes. Similar relations can be obtained also for single
and multiple inclusive cross-sections.
A particularly simple case is the two-particle unitarity, which follows from the intermedi-
ate states with n = 2 in (30)
k′′
Z
Im A(k1′ , k2′ |k1 , k2 )
= dΩ′′ A∗ (k1′′ , k2′′ |k1′ , k2′ )A(k1′′ , k2′′ |k1 , k2 ). (32)
32π 2 W
For light particles, like pions and nucleons, it is exactly fulfilled for energies below the thresh-
old of new particle production.
One has to stress that in terms of Mandelstam variables s, t and u the unitarity relation
is obtained only in the physical region, where all the momenta are real.
 12

4.4 Crossing
From the perturbation theory it is known that a given Feynman diagram describes different
processes depending on the sign of the zero components of external particle momenta. A
change k → −k for an external line corresponds to the transition from the initial particle to
final antiparticle and vise versa. For particles with spins one has to additionally transform
the relevant spinors. To be concrete consider a binary amplitude for the scattering of scalar
particles (Fig. 1)
a(k1 ) + b(k2 ) → c(k1′ ) + d(k2′ ). (33)
The property of crossing means that the same Feynman diagrams and hence the amplitude
will also describe two more processes

a(k1 ) + c̄(−k1′ ) → b̄(−k2 ) + d(k2′ ) (34)

and
¯ ′ ′
a(k1 ) + d(−k2 ) → c(k1 ) + b̄(−k2 ). (35)
For the first reaction variable s plays the role of the square of the c.m. energy. For the second
one this role is played by variable t, and for the third by variable u. Correspondingly the
three reactions are denoted by channels: s-channel for (33), t-channel for (34) and u-channel
for (35).
In fact the amplitudes in the three channels depend on their variables in different regions
which do not overlap. The s-channel amplitude implies that all zero components of k1 , k2 , k1′
and k2′ are positive. For the t-channel k1′ and k2 have to possess negative zero components.
For the u-channel k2′ and k2 have to possess negative zero components. The Mandelstam
variables s, t, u have to belong to the physical regions of each channel, which do not overlap.
In particular for the s channel conditions (23) - (25) are to be satisfied, which imply that s is
positive and t is negative. However for the t channel one immediately concludes that t has to
be positive and s and u predominantly negative (for elastic scattering their regions are given
by (25)).
Since the physical regions of the three channels do not overlap, the crossing by itself does
not lead to any constructive consequence for the amplitude. In fact one meets with three
different functions A(s, t, u) corresponding to the amplitude in different physical regions.
However if A is given by a certain analytic function, e.g. by the corresponding Feynman
diagrams, one finds a possibility to analytically continue A(s, t, u) from one physical region
to another. Then one discovers a relation between the amplitudes for generally different
physical processes. For a particular case when some of the channels coincide one finds a
symmetry: after continuation one has to find the same function.
Thus the property of crossing acquires its full meaning in relation with the analyticity
properties of the amplitude.
The crossing is of course trivially generalized to multi-legged amplitudes. Each external
particle may be considered both as the incoming particle or outgoing antiparticle and vice
versa. Correspondingly one obtains amplitudes for different processes depending on the values
of the generalized energies and transferred momenta, which now may involve several particles.
 13

Again something constructive may follow if one is able to analytically continue the amplitude
from one physical region to another.

4.5 Analyticity
This property is the most complicated. It involves several points. First, as we shall see
the amplitude indeed allows for an analytic continuation from one physical region to another.
Second, it turns out that the amplitude is in general an analytic function of its variables in the
whole complex plane, except for isolated singularity points of the pole or branch points type.
Third, the position of these singularities and discontinuities around them are determined by
the unitarity relation.
These properties will be derived in the following two sections. Here we only stress its
importance. As one knows, the analytical functions is totally determined by its singularity
points, residues at its poles and discontinuities along the cuts. Once these can be found
from the unitarity, a possibility arises to find the amplitude uniquely only from unitarity for
different channels and analyticity, without recurring to any dynamical background, related
to the Lagrangian approach in the framework of the quantum field theory. However this
maximalistic idea fails due to complexity of analytic properties of multiparticle amplitudes
and the growth of amplitudes for particles with spins at infinitely large momenta.
 14

5 Analyticity from the local quantum field theory

Analyticity is probably a very fundamental and general property of the underlying local
quantum field theory, related to its causality property. However the proof of analyticity from
these first principles is quite difficult and hardly possible at the level sufficient for applications.
We shall trace the logic of such a proof on a simple example of the elastic scattering of two
spinless particles, which for convenience we call ’nucleon’ N and ’pion’ π with masses m and
µ, respectively:
N (k1 ) + π(k2 ) → N (k1′ ) + π(k2′ ). (36)

Introducing the pion field φ(x) we can relate the amplitude to the residue of the corre-
sponding matrix element:
Z
4 4 ′ ′
(2π) δ (k1 + k2 − k1′ − k2′ )A(s, t) =i d4 xd4 x′ eik2 x −ik2 x Kx Kx′ < Ψk1′ |T {φ(x′ )φ(x)}|Ψk1 > .
(37)
Here k12= k1′ 2
= m2 , =k22 = k2′ 2
Kx = µ2 , + ∂2 µ2 ,
Ψk1 is a state of a single nucleon with
momentum k1 (’in’ or ’out’ the same, since the single nucleon state is stable). Our plan is to
establish analyticity in variable s = (k1 + k2 )2 at fixed t = (k1 − k1′ )2 .
To this aim we first insert operators Kx and Kx′ under the sign of the T-product. Ob-
viously we shall find some extra terms, which result from the differentiation in time of the
discontinuity of the T-product. All such terms have a quasi-local structure, that is propor-
tional to δ4 (x − x′ ) or its derivatives. It is easy to find that these terms are analytic in s. In
fact a typical quasi-local contribution has a structure
Z
′
(2π)4 δ4 (k1 + k2 − k1′ − k2′ )AQL (s, t) = i d4 xei(k2 −k2 )x < Ψk1′ |O(x)|Ψk1 >
Z
′ ′
=i d4 xei(k2 −k2 )x ei(k1 −k1 )x < Ψk1′ |O(0)|Ψk1 > . (38)

Integration over x gives the overall δ-function. The matrix element depends only on k1 and
k1′ , that is on the product k1 k1′ , since the squares of momenta are fixed. However we have
t = 2m2 − 2k1 k1′ so that this product depends only on t. Thus we find AQL (s, t) = f (t) and
is trivially analytic in s.
Defining Kx φ(x) = j(x), taking x = X − z/2, x′ = X + z/2, shifting x and x′ by X and
integrating over X to separate the overall δ-function we find
Z
1 ′
A(s, t) = i d4 zei 2 z(k2 +k2 ) < Ψk1′ |T {j(z/2)j(−z/2)}|Ψk1 > . (39)

Our idea is to concentrate all the s dependence in the exponent. For this purpose we
choose the Breit system
k1 + k′1 = 0 (40)

in which the nucleon after collision moves in the opposite direction with the same momentum.
′ we find k k ′ = k 2 +k2 , so that |k | depends only on t. We also have k
Since k10 = k10 ′
1 1 10 1 1 20 = k20
and |k2 | = |k′2 |, which means that the pion momentum does not change its magnitude but
only rotates. The kinematical situation is illustrated in Fig. 2.
 15

Figure 2: Kinematics in the Breit system

′ the pion energy in the Breit system. Obviously k2 = ω 2 − µ2 .

We denote ω = k20 = k20 2
Next we have
t = (k2 − k2′ )2 = −(k2 − k′2 )2 = −2k2 2 + 2k2 k′2 .

So 2k2 k′2 = t + 2k2 2 and

(k2 + k′2 )2 = 4k2 2 + t = 4ω 2 − 4µ2 + t. (41)

Finally we relate ω and s. In the Breit system we find

(k1 + k1′ )(k2 + k2′ )

ω= q .
2 (k1 + k1′ )2

We easily find:

(k1 + k1′ )2 = 4m2 − t, (k1 + k1′ )(k2 + k2′ ) = s − u = 2s + t − 2m2 − 2µ2 . (42)

So
s−u
ω= √ (43)
2 4m2 − t
and is a linear function of s at fixed t.
With these kinematical relations we finally obtain for the amplitude in the Breit system
Z √ 2 2
A(s, t) = i d4 zeiz0 ω−ize ω −µ +t/4 < Ψk1′ |T {j(z/2)j(−z/2)}|Ψk1 > . (44)

As we observe now all the dependence on s is indeed concentrated in the exponent: the
matrix element depends only on k1 which depends only on t.
Inspecting Eq. (44) we immediately see that it cannot be analytically continued from the
real axis to any complex values of ω, since the integral diverges. To overcome this difficulty
we shall pass to different expressions for the amplitude.
We note that one can rewrite the T-product in one of the two following forms:

T {j(z/2)j(−z/2)}

= θ(z0 )[j(z/2)j(−z/2)] + j(−z/2)j(z/2) = −θ(−z0 )[j(z/2)j(−z/2)] + j(z/2)j(−z/2). (45)

Correspondingly we introduce two new amplitudes: the retarded one

Z
1 ′
Aret(s, t) = i d4 zθ(z0 )ei 2 z(k2 +k2 ) < Ψk1′ |[j(z/2)j(−z/2)]|Ψk1 > (46)
 16

and the advanced one

Z
1 ′
Aadv (s, t) = −i d4 zθ(−z0 )ei 2 z(k2 +k2 ) < Ψk1′ |[j(z/2)j(−z/2)]|Ψk1 > . (47)

In the physical region of our reaction we easily find

Aadv = A∗ret . (48)

Indeed
Z
1 ′
A∗ret (s, t) = −i d4 zθ(z0 )e−i 2 z(k2 +k2 ) < Ψk1′ |[j(z/2)j(−z/2)]|Ψk1 >∗ =
Z
1 ′
−i d4 zθ(z0 )e−i 2 z(k2 +k2 ) < Ψk1 |[j(−z/2)j(z/2)]|Ψk1′ >z→−z =
Z
1 ′
−i d4 zθ(−z0 )ei 2 z(k2 +k2 ) < Ψk1 |[j(z/2)j(−z/2)]|Ψk1′ >P T =
Z
1 ′
−i d4 zθ(−z0 )ei 2 z(k2 +k2 ) < Ψk1′ |[j(z/2)j(−z/2)]|Ψk1 >= Aadv (s, t). (49)

Now we pass to the central point: we are going to prove that in the physical region
of reaction (36) the scattering amplitude A in fact coincides with Aret. To this end we
demonstrate that the contribution coming from the additional term in (45) vanishes. Indeed
we can introduce the complete set of intermediatephysical states |n > to write
Z
1 ′
2Au ≡ d4 zei 2 z(k2 +k2 ) < Ψk1′ |j(−z/2)j(z/2)|Ψk1 >
Z
1 ′
d4 zei 2 z(k2 +k2 ) < Ψk1′ j(−z/2)|n >< n|j(z/2)|Ψk1 >
X
= dτn
n

We have from the translateional invariance

′
< Ψk1′ |j(−z/2)|Ψk1 >= e−iz/2(k1 −pn ) < Ψk1′ |j(0|Ψk1 >,

< n|j(z/2)|Ψk1 >= eiz/2(pn −k1 ) < n|j(0|Ψk1 >

so that
 1 1 
dτn (2π)4 δ pn + (k2 + k2′ ) − (k1 + k1′ ) < Ψk1′ |j(0)|n >, n|j(0)|Ψk1 >
X
2Au =
n 2 2

dτn (2π)4 δ(pn − k1 + k2′ ) < Ψk1′ |j(0)|n >, n|j(0)|Ψk1 > .
X
= (50)
n

As we observe, the contribution comes from intermediate states with p2n = (k1 − k2′ )2 = u.
However in the physical region of reaction (36) u < (m − µ)2 and the state with the lowest
mass in (50) is the nucleon with a larger mass p2n = m2 . So there is not a single state which
can kinematically contribute to (50) in the physical region of reaction (36) and Au is zero
in this region. Note for the future that Au given by (50) is just the right-hand side of the
unitarity relation for the u-channel reaction,

N (k1 ) + π(−k2′ ) → N (k1′ ) + π(−k2 ) (51)

 17

for which the c.m.energy is u. So it is not zero in the physical region of this crossed reaction.
Since we have proved that in the physical region of reaction (36) A = Aret we may try to
analytically continue the latter function, which explicitly is
Z √ 2 2
Aret (s, t) = i d4 zθ(z0 )eiz0 ω−ize ω −µ +t/4 < Ψk1′ |[j(z/2)j(−z/2)]|Ψk1 > . (52)

Obviously this function is much better suited for analytic continuation in ω. Since z0 > 0 the
first exponential factor becomes a damping factor if Im ω > 0. The matrix element is different
from zero only for z 2 ≥ 0, which is the consequence of causality. So |z| ≤ z0 . However in the
general case it does not guarantee that the second exponential factor will grow more slowly
that the first factor decreases. In fact for ω = ω1 + iω2 , with both ω1,2 real and small ω2 we
have q
ω1
Im ω 2 − µ2 + t/4 = ω2 q .
ω12 − µ2 + t/4
In the physical region of reaction (36) t ≤ 0 and of course µ2 > 0. So we find
q
Im ω 2 − µ2 + t/4 > ω2 ,

which implies that the second exponential factor in (52) can grow faster that the first de-
creases. This makes analytic continuation impossible.
An exceptional case is when both µ and t are equal to zero. Physically this corresponds
to the forward photon-nucleon scattering. Then we have
Z
Aret (s, t) = i d4 zθ(z0 )eiω(z0 −ze) < Ψk1′ |[j(z/2)j(−z/2)]|Ψk1 > (53)

and since z0 −ze ≥ 0 we can safely continue this amplitude to the whole upper half plane of ω.
This case was historically the first to establish analytic properties of scattering amplitudes.
Demonstration that amplitude Aret (s, t) can also be analytically continued to the upper
half-plane of ω in the general case for the physical values of t and µ can be done but requires
considerable effort and is restricted to certain mass relations between m and µ (fulfilled for
pions, but not for K-mesons). The proof starts from taking first unphysical values of µ2 < 0,
analytic continuation in ω with such µ2 and continuing in µ2 along the real axis to positive
µ2 afterwards. Presentation of this proof lies beyond the scope of these lectures. So we shall
simply take for granted that the amplitude Aret can be analytically continued in ω from the
physical region of reaction (36), that is ω ≥ µ to the whole upper half-plane for physical
values of the pion mass.
Next step is to consider the crossed reaction (51). The amplitude for it will be given
by the same formulas (39) and (44) in which we only have to consider both k20 and k20 ′

negative. As a result in the physical region of the crossed reaction the Breit energy of the
pion will be negative ω < −µ. Now similarly to the previous derivation we prove that in the
physical region of the crossed reaction the physical amplitude A coincides with the advanced
amplitude Aadv . Indeed we find
Z
1 ′
2As ≡ d4 zei 2 z(k2 +k2 ) < Ψk1′ |j(z/2)j(−z/2)|Ψk1 >
 18

Figure 3: Amplitude as an analytic function in the s-plane. Dark initial parts of the cuts
correspond to the unphysical regions

 1 1 
dτn (2π)4 δ pn + (k2 + k2′ ) + (k1 + k1′ ) < Ψk1′ |j(0)|n >, n|j(0)|Ψk1 >
X
=
n 2 2

dτn (2π)4 δ(pn − k1 − k2 ) < Ψk1′ |j(0)|n >, n|j(0)|Ψk1 > .

X
= (54)
n

Now the states which contribute have their mass p2n = (k1 +k2 )2 = s. However in the physical
region of the crossed reaction s ≤ (m − µ)2 and there are no intermediate state with such a
mass. So As = 0 in the physical region of the crossed reaction.
Taking the amplitude Aadv as the starting point for analytic continuation, we observe
that since now z0 < 0 the amplitude Aadv can be continued into the whole lower half-plane
of ω (with the same reservations as before).
On the real axis we easily find

Aret − Aadv = 2i(As − Au ). (55)

This difference is not equal to zero when either As or Au different from zero, which occurs
if either ω > µ or ω < µ. In the interval −µ < ω < µ both As and Au are zero (expect for
certain points to be discussed later), so that in this interval Aret = Aadv . Therefore Aret and
Aadv tepresent in fact a single function A(s, t) analytic in the whole ω (and so s) plane with
two cuts along the real axis, right and left, with discontinuities 2iAs and −2iAu respectively.
The physical amplitudes are given: for the direct reaction by the value of A just above the
right-hand cut and for the crossed reaction just below the left-hand cut. This is ilustrated in
Fig.5.
The discontinuities across the cuts, given by 2iAs and −2iAu , present in fact the right-
hand parts of the unitarity relation for the direct channel (2iAs ) and crossed channel (2iAu ).
They are pure imaginary. Since Aret and Aadv are complex conjugate, outside the cuts the
amplitude is real and the real functions As and Au are in fact the imaginary parts of the
amplitude in the physical region of the direct and crossed channel, respectively. They are
standardly called absorptive parts of the amplitude in the direct and crossed channel. So the
statement that the imaginary part of the amplitude is given by the explicit form of As in the
direct channel and Au in the crossed channel is just the unitarity relation for both channels.
It is important, however, that these unitarity relations are more general than the ones
established from the properties of the physical S-matrix. They are supposed to be valid in
a larger region of variables s and t than the latter, which are derived only strictly in the
 19

physical region of the reaction. The expression for the discontinuity (55) is supposed to be
valid for any real value of s and at any fixed t. From the explicit form of 2iAs and 2iAu one
concludes that both start to be different from zero as soon as there appears an intermediate
state with mass larger than the given s for As or the corresponding given u for Au irrespective
of the fixed value of t. This implies, that the discontinuities may be different from zero also
outside the physical region of the corresponding reaction, in the unphysical region, where the
momenta of the colliding particles become complex.
To see this point, consider As . From its explicit form (54) one finds that the intermediate
state with the lowest mass is the single nucleon state of mass m. Its contribution will appear
as soon as s = m2 . Obviously this value of s is unphysical: the direct reaction cannot go
unless s ≥ (m + µ)2 . However the absorptive part will be non-zero at s = m2 and it is trivial
to find its value at this point. In fact the contribution from the single- nucleon state to As is

1 d3 k
Z
A(1)
s = (2π)4 δ4 (k − k1 − k2 ) < Ψk1′ |j(0)|Ψk >< Ψk |j(0)|Ψk1 > . (56)
2 (2π)3 2k0

The two matrix elements depend on (k − k1 )2 = k22 = µ2 and (k − k1′ )2 = k2′ 2 = µ2 . So they
are just numbers, complex conjugate to one another. It can be proven that they are in fact
real and present the value of the vertex part for pion emission N→ N+π on the mass shell
of all particles. This value is usually taken as the NNπ coupling constant g. So (56) can be
rewritten as
Z
A(1)
s = πg 2
d4 kδ(k2 − m2 )δ4 (k − k1 − k2 ) = πg2 δ(s − m2 ). (57)

Thus the single nucleon contribution to As has the form of a δ function. This corresponds
to a pole of the amplitude A at s = m2 with a residue g2 . Indeed consider the amplitude
multiplied by s − m2 . It will possess the same analytic properties but the contribution (57)
will be zero. returning back to the amolitude we shalldiscover a pole at s = m2 . Similarly the
single nucleon contribution to Au is πg2 δ(u − m2 ) and corresponds to a pole of the amplitude
at u = m2 with the residue g2 .
However these poles do not exhaust all the contributions from the unphysical regions.
For the direct channel, for a given s the physical region of t is bounded from below by the
value −4k2 (s), where k is the c.m. momentum, which depends on s and goes to zero as s
approaches the threshold value (m + µ)2 (see Fig. 4). The expression for As is, however,
different from zero for a given s ≥ (m + µ2 ) and any chosen t < 0. Thus it is also different
from zero in the unphysical region

(m + µ)2 < s < s1 (t), (58)

where the value of s1 is determined by the fixed t: t = −4k2 (s1 ) (Fig. 4).
The expression of the right-hand side of the unitarity relation in the unphysical regions
involves complex particle momenta and implies analytic continuation of the standard uni-
tarity relation. This continuation is technically achieved in the process of the mentioned
continuation in the pion mass squared µ2 .
 20

m2 (m+µ)2
0

-0.05

t
-0.1

unphysical
region
-0.15

-0.2

0 20 40 60 80 100 120 140 160 180 200

s

Figure 4: Physical and nonphysical regions for the unitarity cut at fixed t

Figure 5: Tree diagrams for elastic π-N scattering

6 Analytic Properties in the Perturbation Theory

Derivation of analytic properties of scattering amplitudes from fundamental principles, illus-
trated in the preceding section is difficult even for simplest amplitudes and becomes impossible
for more general ones. So most of the information necessary for practical applications comes
from the study of the expressions obtained in perturbation approach, that is from Feynman
diagrams. Again spins are inessential and we shall restrict ourselves to interaction of spinless
particles.
The simplest Feynman diagrams do not include integration over momenta of intermediate
particles. These are the so-called tree diagrams. Analytic properties of tree diagrams are
easily established, since they are just products of propagators of intermediate particles. Each
one of them is just a pole in an invariant variable which is a sum of a certain combination of
external particle momenta at the point corresponding to the mass squared of the intermediate
particle. So the amplitude is an analytic function of all possible invariant variables, having
poles in sums of external momenta at intermediate particles mass squares.
A simple example is the elastic πN amplitude studied in the preceding section. The two
possible tree diagrams for it are shown in Fig. 5. Each one is just a propagator of the
intermediate nucleon. They give a sum of two poles:

g2 g2
+ 2 . (59)
m2 −s m −u
 21

Figure 6: A tree diagram for pion production in π-N collisions

Figure 7: An one-loop diagram

For a more complicated amplitude for the pion production

N (k1 ) + π(k2 ) → N (k1′ ) + π(k2′ ) + π(k3′ ) (60)

one of a set of possible tree diagrams is shown in Fig. 6. It is a product of two intermediate
nucleon propagators and so a product of two poles
1 1
g3 . (61)
m2 − (k1 + k2 ) m − (k1 + k2 − k3′ )2
2 2

6.1 Loop diagrams. Landau’s rules

More complicated, loop contributions to the amplitude involve integration over intermediate
particle momenta. To study their analytic properties certain practical rules have been estab-
lished . L.D.Landau invented rules which allow to determine positions of singularities of loop
amplitudes in their external momentum variables.
The contribution of a general Feynman diagram to the amplitude can be written in the
form of a multiple integral
L l
1
Z Y
d4 ki
Y
A= (62)
i=1 j=1
Aj

where Aj = m2 − qj2 is the propagator corresponding to the jth internal line. The total
number of internal lines is l. Integration is performed over L loop 4-momenta ki , i = 1, ...L.
In fact the momenta qi of intermediate particles are linear combinations of the loop momenta
and external particle momenta.
To illustrate consider a simple one-loop diagram shown in Fig. 7. It is convenient to
assume all external momenta as incoming to have the maximal symmetry : p12 + p23 + p34 +
 22

p41 = 0. If the loop momentum is k then one may choose q1 = k, q2 = k+p12 , q3 = k+p12 +p23
and q4 = k + p12 + p23 + p34 .
One can present the amplitude in the Feynman parametrization, using the identity
l ∞ l l
1 1
Y Z Y X
= (l − 1)! dαj δ(1 − αj ) Pl . (63)
j=1
Aj 0 j=1 j=1 ( j=1 αj Aj )l

The proof of it starts from the statement that it is true for l = 2

Z 1 dα1 1  1 1  1
= − = . (64)
0 (α1 A1 + (1 − α1 )A2 )2 A1 − A2 A2 A1 A1 A2
Differentiating this identity in A2 l − 1 times we get
1 ∞ dα1 dα2 αl−1
2 δ(1 − α1 − α2 )
Z
=l . (65)
A1 Al2 0 (α1 A1 + α2 A2 )l+1
This enables to make transition from l to l + 1:
l+1 ∞ l l
1 1
Y Z Y X
= (l − 1)! dαj δ(1 − αj ) Pl
A
j=1 j 0 j=1 j=1 ( j=1 αj Aj )l Al+1

l l
∞ dβ1 dβ2 β2l−1 δ(1 − β1 − β2 )
Z Y X
= l! dαj δ(1 − αj ) . (66)
β1 Al+1 + β2 lj=1 αj Aj )l+1
P
0 j=1 j=1

dαj = β2−l dα̃j , δ(1 − αj ) = β2 δ(β2 −

Q Q P
Here we make the substitution β2 αj = α̃j . Then
α̃j ) and we get
P

∞ l l
dβ1 dβ2 δ(1 − β1 − β2 )
Z Y X
l! dα̃j δ(β2 − α̃j ) . (67)
(β1 Al+1 + lj=1 α̃j Aj )l+1
P
0 j=1 j=1

Integration over β2 gives (63) at l + 1.

For our purpose it is desirable to exclude the δ- function without spoiling the symmetry
in all variables. To this end we insert the identity
Z ∞
1= dλe−λ
0

into (63) and change variables αj → λαj Integration over λ with the help of δ-function gives
Ql P
l ∞ −1/ αj
1 j=1 dαj e
Y Z
= (l − 1)! . (68)
Aj αj ( lj=1 αj Aj )l
P P
j=1 0

So finally P
L l Z Ql −1/ αj
∞
j=1 dαj e
Z Y
d4 ki
Y
A = (l − 1)! Pl . (69)
αj ( l
j=1 αj Aj )
P
i=1 j=1 0

Now we have to analyze possible singularities of the amplitude presented in the form (68)
and considered as a function of external momenta.
To start we consider a function f (z) of the complex variable z presented as an integral
over another complex variable ξ along a certain curve:
Z b
f (z) = dξg(z, ξ) (70)
a
 23

Figure 8: Integration contour of g(z, ξ) in the ξ-plane

(see Fig. 8). Let g(z, ξ) be an analytic function of its variables, which has isolated singular
points. One can distinguish between two possibilities. Considered as a function of z function
g(z, ξ) may have singularities at points z = zs which do not depend on ξ (’fixed singular
points’). In this case obviously f (z) given by (70) will also have singularities at the same
fixed points. However this case has no relation with the structure of the integral (68) and so
not very interesting. The second possibility is that the position of the singularity of g(z, ξ) in
variable z depends on variable ξ (’moving singular points’). In this case it is advantageous to
consider g(z, ξ) as a function of the integration variable ξ, which has a singularity at ξ = ξs (z)
depending on z. As z varies the singular point ξs (z) moves in the complex ξ-plane. If ξs (z)
touches the integration contour in (70) this generally does not induce a singularity of f (z),
since the contour can be displaced to avoid the singularity ξs . A singularity arises only when
such a displacement becomes impossible. This may occur in two cases.
1) The singular point ξs (z) arrives at the initial or final integration points a or b. Equations
ξs (z) = a or ξs (z) = b determine the position of this so-called end-point singularity in z.
2) Two different singularities ξs1 (z) and ξs2 (z) touch the integration contour from different
sides. This generates a ’pinch singularity’ whose position is determined by the equation
ξs1 (z) = ξs2 (z). In a particularly important for us case function g(x, ξ) has a form of 1/h(z, ξ)
where h(z, ξ) is an analytic function which may have isolated zeros. Then coincidence of two
zeros at the same point implies two equations

∂h(z, ξ)
h(z, ξ) = 0, = 0, (71)
∂ξ
which determine the pinch singularity.
These considerations can be easily generalized to multiple integrals. Let us consider a
function f (z) presented as a double integral
Z b2 Z b1 1
Z b2
f (z) = dξ2 dξ1 = dξ2 g(z, ξ2 ), (72)
a2 a1 h(z, ξ1 , ξ2 ) a2

where Z b1 1
g(z, ξ2 ) = dξ1 . (73)
a1 h(z, ξ1 , ξ2 )
 24

Function g(z, ξ2 ) becomes singular either if a singularity in ξ1 touches the end points a1 or
b1 or if two singularities in ξ1 coincide, that is if
∂h(z, ξ1 , ξ2 )
h(z, ξ1 , ξ2 ) = 0, = 0. (74)
∂ξ1
In the last case, expressing from the second equation ξ1 = ξ1 (z, ξ2 ) we find the equation for
the singularity of g(z, ξ2 ) as
h(z, ξ1 (ξ2 ), ξ2 ) = 0. (75)

Now a pinch singularity of f (z) arises if two singularities of g(z, ξ2 ) coincide, that is when
additionally
∂h(z, ξ1 (ξ2 ), ξ2 )
= 0. (76)
∂ξ2
However
∂h(z, ξ1 (ξ2 ), ξ2 ) ∂h(z, ξ1 (ξ2 ), ξ2 )  ∂h(z, ξ1 (ξ2 ), ξ2 ) ∂ξ1 (ξ2 ))
= + =0 (77)
∂ξ2 ∂ξ2 ∂ξ1 ∂ξ2

f ixed ξ1

The last derivative is zero, so that the final conditions for the pinch singularity are
∂h(z, ξ1 , ξ2 ) ∂h(z, ξ1 , ξ2 )
h(z, ξ1 , ξ2 ) = 0, = 0, = 0. (78)
∂ξ1 ) ∂ξ2
Any of the two equations with derivatives can be substituted by a condition for the end-point
singularity, ξ1 = a1 , ξ1 = b1 or ξ2 = a2 , ξ2 = b2 .
In the general case when
n
1
Z Y
f (z) = dξi (79)
i=1
h(z, ξ1 , ...ξn )

the conditions for a singularity in z are

∂h(z, ξ1 , ...ξn )
h(z, ξ1 , ...ξn ) = 0, = 0, (ξi = ai or bi ). (80)
∂ξi
Now we apply these conditions to the expression (68) for the amplitude. The integral
involves two sets of variables: α1 , ...αl for each internal line and k1 , ...kL for each integration
loop. The role of function h in (80) is played by function lj=1 αj Aj . Applying (80) to
P

variables αj we obtain conditions:

l
X
h= αj Aj = 0, Aj = 0, or αj = 0, for each line. (81)
j=1

Due to the latter two conditions we do not need to additionally require h = 0. Applying (80)
to variables ki we obtain
l
X ∂Aj
αj = 0. (82)
j=1
∂ki
Since qj is a linear function of ki in a given loop with a coefficient which may be chosen to
be +1 with a suitable choice of direction of qj , this relation can also be written as
X
αj qj = 0, for each integration loop. (83)
 25

Figure 9: Reduced diagrams corresponding to Fig. 6.1

Relations (81)-(83) are the Landau rules to determine the position of the singularity of the
amplitude.
Application of these rules to concrete amplitude will be presented in the last subsection of
this section. Here just a few preliminary comments. Conditions (81) imply that in the study
of singularities each internal article either should be considered as physical, qj2 = m2j , or the
corresponding line should be dropped, αj = 0, with the attached vertices joined together.
In the latter case we get a new, reduced diagram which should be treated by the same
method. The possibility to put any αj = 0 thus implies that a given diagram contains all
the singularities corresponding to its reduced diagrams and also certain proper singularities,
obtained when none of the αj is put to zero. Some reduced diagrams corresponding to the
original diagram in Fig. 7 are shown in Fig. 9. The second condition (83) means that in
the study of singularities all momenta in the main and reduced diagrams should be taken
coplanar.
It is important that the Landau rules are only the necessary conditions for the existence
of singularity. In our derivation we mentioned that the pinch occurs only when the two
singularities touch the integration contour from opposite sides. If they touch it from the
same side there obviously occur no singularity . Thus some of the singularities determined
by the Landau rules may be non-existent on the physical sheet of the amplitude. They can
then be found on unphysical sheets reached by analytic continuation from the physical one.
In particular physical singularities require that the values of αj found from the Landau rules
be non-negative in correspondence with the integration region in (68). Analytic continuation
may demand distortion of this region and then the singularity may appear also with some
negative values of αj .

6.2 Loop diagrams. Cutkosky’s rules

The Landau rules only tell us the position of the singularity. Normally the singularity is a
branch point. The discontinuity of the amplitude across the cut which starts from this branch
point is determined by the Cutkosky rules. They are based on the technique in which all
masses corresponding to internal lines are taken different (although in the physical amplitude
masses take on just a few values).
 26

Figure 10: Pinching two singularities

Suppose we want to study the discontinuity across the cut which is related to the branch
point corresponding to a given reduced diagram, with only certain number n ≤ l lines re-
maining from the original amplitudes which we enumerate 1, ..., n. Present the amplitude in
the form
Z b1 (z) dq12
Z b2 (z,q12 ) dq22
Z bn (z,q12 ,...qn−1
2 ) dqn2
A= ··· An (z, q12 ...qn2 ), (84)
a1 (z) m21 − q12 a2 (z,q12 ) m22 − q22 an (z,q12 ,...qn−1
2 ) m2n − q12

where we separated the first n denominators Ai which correspond to the lines remained in
the reduced diagram and stored all the rest denominators together with the determinant for
the transition from loop momenta to variables qi2 , i = 1, ...n into function An . The external
variable (or variables) are denoted as z.
We begin with the study of integration over q12 presenting (84) as
b1 (z) dq12
Z
A= A1 (z, q12 ). (85)
a1 (z) m21 − q12

Note the characteristic feature of the singularity in question: it should depend on all masses
m1 , ...mn and only on them. Let us study the integral (85) from this point of view. It can
have singularity in three cases:
1) The singularity of the denominator at q12 = m21 may coincide with the endpoints a1 (z)
or b1 (z).
2) A singularity of A1 (z, q12 ) at q12 = q1s
2 may coincide with the endpoints a (z) or b (z).
1 1
3) This singularity at q12 = q1s 2 may coincide with the singularity of the denominator:

2 = m2 .
q1s 1
It is trivial to see that a singularity depending on all n masses m1 , ...mn can only arise in
the third case. In the first case it depends only on m1 and in the second it does not depend
on m1 . So our singularity is a pinch shown graphically in Fig. 10. Obviously we can deform
the integration contour in the q12 plane as shown in Fig. 11 separating the contribution from
the pole at q12 = m21 and the rest integral which has no singularity;

dq12
Z
A = ±2πiA1 (z, m21 ) + 2 A1 (z, q12 ). (86)
C1 m1 − q12

We do not know the true location of singularities, which leads to an undetermined sign of
the pole contribution.
 27

Figure 11: The new integration contour avoiding the pinch

Since the integral in (86) has no singularity, the singular part of the amplitude (and the
branch cut discontinuity) is contained in the pole contribution proportional to A1 (z, m21 ). So
next we study this pole term presenting
Z b2 (z,m21 ) dq22
A1 (z, m21 ) = A2 (z, m21 , q22 ). (87)
a2 (z,m21 ) m22 − q12

We repeat the same study as for the initial amplitude and find that from all three possi-
bility for the singularity only the pinch between the pole at q22 = m22 and a singularity of
A2 (z, m21 , q22 ) at q2s
2 (z, m2 ) gives the desired singularity in z depending on all masses m , ...m .
1 1 n
2
Again we present A2 (z, m1 ) as a sum of the contribution from the pole term and an integral
which has no singularity:

A2 (z, m21 ) = ±2πiA2 (z, m21 , m22 ) + regular term. (88)

Next we present A2 (z, m21 , m22 ) as an integral over dq32 from A3 (z, m21 , m22 , q32 ) and do the
same reasoning after which we repeat this procedure until we arrive at
Z bn (z,m21 ,...m2n−1 ) dqn2
An−1 (z, m21 , ...m2n−1 ) = An (z, m21 , ...m2n−1 ). (89)
an (z,m21 ,...m2n−1 ) m2n − qn2

Now from the three possibilities for the singularity the only suitable is the coincidence of the
singularity at qn2 = m2n with one of the endpoints

an (z, m21 , ...m2n−1 ) = m2n or bn (z, m21 , ...m2n−1 ) = m2n . (90)

Other two possibilities, which involve a singularity of An , depending on masses mj with j > n
which enter An and so correspond to different reduced diagrams.
Let bn (zs ) = m2n (we suppress other arguments in bn ). Then at z close to the singularity
point zs we have
bn (z) = m2n + (z − zs )b′n (zs ) + .... (91)

As z moves around the singular point, bn (z) also moves around m2n (if b′n (zs ) 6= 0), see Fig.
12.
So we find the discontinuity

Disc An−1 (z) = ±2πiAn (z, m21 , ...m2n ). (92)

Again the sign is indeterminate, since the sign of b′n (zs ) is unknown.
 28

Figure 12: Integration over qn2 as bn (z) goes around to m2n when z goes around the singularity
at z = zs

Combining all our results we find the discontinuity of the amplitude

Disc A(z) = (±2πi)n An (z, m21 , ...m2n ). (93)

This means that to find the discontinuity one has to substitute all particle propagators
remaining in the reduced diagram according to
1
→ ±2πiδ(m2 − q 2 ). (94)
m2 − q2
This is the content of the Cutkosky rules.
It is instructive to see how one can determine the position of the singularity point in this
approach. As mentioned the condition for a singularity is

bn (z, m21 , ...m2n−1 ) = m2n (95)

(or an = m2n ). So one encounters the problem to find bn (z, m21 , ...m2n−1 ). To this one has to
find an extremum of qn2 at fixed values of q12 , ...qn−1
2 . The actual variables are the 4 =momenta
k1 . So one has an equation for determination of the extremum
∂qn2 ∂q 2 ∂q 2
+ γ1 1 + · · ·γn−1 n−1 = 0 (96)
∂ki ∂ki ∂ki
with γi i = 1, ...n − 1 being the Lagrange multipliers and under complementary conditions
qi2 = m2i , i = 1, ...n − 1, n. Obviously these equations coincide with the Landau rules.

6.3 Unitary singularities

Let us apply the found rules for the study of some simple examples. For the box diagram
of Fig. 7 the simplest non-trivial reduced diagrams are shown in Fig. 9. Let us consider
one of them shown in Fig. 13 with more natural notations for particle masses and external
momenta The Landau equation for it are

α1 q1 − α2 q2 = 0, q12 = m21 , q22 = m22 (97)

(the negative sign in the first equations comes from the direction of q2 in the integration loop,
opposite to that of q1 ). From the first equation we find q2 = q1 (α1 /α2 ). Squaring this we
find:  α 2
1
m22 = m21 , (98)
α2
 29

Figure 13: Simple reduced diagram for binary scattering

which gives
α2 m1
= . (99)
α1 m2
On the other hand, k1 + k2 = q1 + q2 = q1 (1 + m2 /m1 ). Taking its square modulus we find
the position of singularity
 m2  2
s = (k1 + k2 )2 = m21 1 + ) = (m1 + m2 )2 . (100)
m1
This is precisely the threshold of the reaction according to the unitarity relation with the
intermediate state of two particles with masses m1 and m2 .
Now we try to evaluate the discontinuity across the cut starting from this singularity
using the Cutkosky rules. The amplitude corresponding to Fig. 6.3 is given by
4
d4 k Y 1
Z
A = g4 , (101)
(2π) i j=1 mj − qj2
4 2

where one can choose in accrdance with our new notations q1 = k, q3 = k−k1 , q2 = k−k1 −k2 ,
q4 = k − k1′ . According to the Cutkosky rules the discontinuity across the cut related to the
branch point (100) will be obtained if we substitute the propagators left in the corresponding
reduced diagram as indicated in (94). This gives

d4 k 1 1
Z
Disc A = g4 (2πi)δ(m21 − q12 )(2πi)δ(m22 − q22 ) 2 . (102)
(2π)4 i m3 − q32 m24 − q42
We transform it into
d3 q1 d3 q2 1 1
Z   
4 2 2
Disc A = i (2π) δ(q 1 + q 2 − k1 − k2 ) g g . (103)
(2π)3 2q10 (2π)3 2q20 m23 − q32 m24 − q42
Obviously this is just the lowest order contribution from the two-particle state to the unitarity
relation
Z
Disc A = i dτ2 (k1 + k2 )A∗ (k1′ + k2′ → q1 + q2 )A(k1 + k2 → q1 + q2 ) (104)

with
1 1
A(k1 + k2 → q1 + q2 ) = g2 , A(k1′ + k2′ → q1 + q3 ) = g2 2 . (105)
m23 2
− q3 m4 − q42
Thus the singularity and discontinuity of the reduced diagram of Fig. 9 fully correspond to
those which follow from the unitarity relation.
 30

Figure 14: A ladder diagram and its ’fish’ reduced diagram

This result can be generalized to more complicated amplitudes with the reduced amplitude
of the Fig. 14 type (’fish amplitude’). Consider the ladder diagram and its reduced amplitude
shown in Fig. 10.a and b respectively. The internal momenta which remain in the reduced
amplitude are marked by a dashed line (’a cut’) in the amplitude of Fig. 6.3a. Their
momenta for convenience are denoted q1 , ...q4 . Now the reduced diagram contains 3 loops
and correspondingly we have 3 equations from the Landau rules:

α1 q1 − αi qi = 0, q12 = m21 , qi2 = m2i , i = 2, 3, 4. (106)

Similarly to (99) we find qi = q1 α1 /αi , i = 2, 3, 4

αi m1
= , i = 2, 3, 4 (107)
α1 mi
P4 P4
and, using k1 + k2 = q1 + 2 qi = q1 (1 + 2 mi /m1 ) we find the position of singularity
4
 mi  2
s = (k1 + k2 )2 = m21 1 + = (m1 + m2 + m3 + m4 )2 .
X
) (108)
2
m1

This is a threshold for the transition of two initial particles into 4 intermediate ones k1 +k2 →
P4
1 qi corresponding to the contribution to the unitarity relation from 4 intermediate particles.
Now according to the Cutkosky rules we find the discontinuity related to this threshold
4 4
d4 qj
Z Y
(2πi)δ(m2j 2 4
X
Disc A = − qj )(2π) δ( qj − k1 − k2 )
j=1
(2π)4 i j=1

4 4
×A∗ (k1′ + k2′ →
X X
qj ) A(k1 + k2 → qj ) (109)
j=1 j=1

where we denoted parts of the amplitude suppressed in the reduced amplitude A(k1 + k2 →
P4 ∗ ′ ′ P4
j=1 qj ) to the right of the cutting line and A (k1 + k2 → j=1 qj ) to the left of the cutting
line. Note that the left amplitude should be taken conjugate, which is related to the fact that
it itself may have branch points in s (if the internal masses are adequate, which is usually the
case) and a careful study of interplay of the following branch cuts revels that it has to be taken
on the opposite side of the cut as compared to the right amplitude. In fact this is clear from
the start because the discontinuity should be pure imaginary. Of course discontinuity (109)
is nothing but the contribution to the imaginary part of the amplitude from the 4-particle
intermediate state:
Z 4 4
dτ4 (k1 + k2 )A∗ (k1′ + k2′ →
X X
Disc A = i qj )A(k1 + k2 → qj ). (110)
j=1 j=1
 31

Figure 15: The triangle diagram

Thus the reduced amplitude Fig. 10b contains the singularity of the original amplitude
dictated by the unitarity relation.
This a general result: for any amplitude the singularities following from the unitarity,
their positions and discontinuities are described by reduced amplitudes of the ’fish’ type.

6.4 Non-unitary singularities. Triangle diagram

In contrast, more complicated reduced diagrams correspond to singularities, which are not
related to the unitarity relation. Fortunately, as we shall presently see, these non-unitary
singularities, as a rule, do not appear in the physical amplitudes but only in their analytic
continuation to the second and higher sheets.
To be concrete we study the simplest case of a triangle singularity of the original amplitude
in Fig. 7 corresponding to the reduced amplitude shown in Fig. 15. We redenoted p12 =
p14 +p42 as p12 == p13 +p32 . Our energetic variable s = (p13 +p32 )2 = p212 . We take q1 for the
loop integration momentum. Other internal lines momenta are q3 = q1 + p13 , q2 = q3 + p32 .
The Landau equations are
3
αi qi , αi > 0, qi2 = m2i .
X
(111)
i=1

Multiplying this equation by qj , j = 1, 2, 3 we get three scalar equations

3
X
αi qi qj = 0, j = 1, 2, 3. (112)
i=1

We define
qi qj = mi mj yij , yii = 1, i, j = 1, 2, 3. (113)

Eq. (112) in terms of yij is rewritten as

3
X
βi yij = 0, j = 1, 2, 3, βi = mi αi > 0. (114)
i=1
 32

Figure 16: Vertices N → N + π and π → N + N̄

Note that
(qi − qj )2 = p2ij = m2i + m2j − 2qi qj = m2i + m2j − 2mi mj yij ,
so that for i 6= j
m2i + m2j − p2ij
yij = , i 6= j, i, j = 1, 2, 3. (115)
2mi mj
The Landau condition evidently reduces to

det ||yik || = 0, (116)

which explicitly gives an equation

2 2 2
1 + 2y12 y23 y31 − y12 − y23 − y31 = 0. (117)

However we have to take into account that βi i = 1, 2, 3 solving the system (114) should all
be positive. This immediately implies that from the three non-trivial yij i 6= j at least two
have to be negative. Otherwise the equation which does not contain the only negative yij
cannot be satisfied by positive β.
Precisely this condition prohibits existence of the triangular singularity for most ampli-
tudes. In fact, in our case this means that at least one of the y13 or y23 related to the
vertexes attached to external particles have to be negative. Other negative y may be sup-
plied by y12 , which is not fixed by particle masses. However condition that y13 < 0 or y23 < 0
cannot be satisfied for common amplitudes for the scattering of elementary particles with
realistic masses. For, say, y13 < 0 one should have p213 > m22 + m23 , which never happens
with normal elementary particles. Say for the vertex N→ N+π we have p213 = m2 and
m22 + m23 = m2 = µ2 > p213 (see Fig. 16). The same situation is realized in all amplitudes
with simple hadrons. In all these cases the triangular singularity exists only on the second
or higher sheets of the complex plane and not on the physical sheet.
Still there exist cases when the triangular singularity appears in the physical amplitude.
This happens when one or several external particles are weakly bound systems, for which one
has the mass M with

M 2 > m22 + m23 , although M < m2 + m3 (118)

(the second condition is the requirement of stability of particle M ). The typical example is
the deuteron D with mass MD = 2m − ǫ, ǫ << m and so for the process D→ N+N we have
MD 2 ≃ 4m2 − 4mǫ > 2m2 so that the corresponding y is well negative.
 33

Figure 17: The deuteron form-factor

To illustrate non-unitary singularities consider the deuteron form-factor, the simplest

diagram for which is shown in Fig. 17. The role of p212 = q 2 is in practice played by the
momentum squared transferred from the electron in the reaction

e(k2 ) + D(k1 ) → e(k2′ ) + D(k1′ ). (119)

Compared to Fig. 6.4, q 2 = p212 = (k2 − k2′ )2 , p13 = k1 , p23 = −k1′ . We find
2m2 − M 2 ǫ
y13 = y23 = ≃ −1 + 2 < 0. (120)
2m2 m
The Landau equation (117) determines the singular point in y12
 ǫ 2
 ǫ
1 + 2y12 1 − 4 − y12 −2 1−4 = 0. (121)
m m
We transform it like follows
 ǫ h  ǫ i
(1 − y12 )(1 + y12 ) − 2(1 − y12 ) 1 − 4 = (1 − y12 ) 1 + y12 − 2 1 − 4 = 0. (122)
m m
We find two solutions
 ǫ ǫ
y12 = 1, y12 = 2 1 − 4 −1 =1−8 . (123)
m m
The first solution leads to q 2 = 0. This solution appeared only due to the symmetry of the
diagram. It lies on the second sheet of the complex q 2 plane, since its position does not
depend on the value of M 2 , which can be taken small when the singularity cannot exist in
the physical amplitude. So the triangular singularity in the physical amplitude is determined
by the second solution, which gives
 ǫ
q 2 = 2m2 − 2m2 1 − 8 = 16mǫ. (124)
m
In the complex q 2 plane this singularity lies much lower than the unitary singularity at
q 2 = 4m2 (see Fig. 18).
It can be found that the triangular singularity coincides with the singularity found in
the quantum mechanical treatment of the deuteron as a composite particle made of two
nucleons. There it is provided by the structure of the deuteron wave function. So non-
unitary singularities found in Feynman diagrams for particles with small binding energies
are equivalent to the structure of the corresponding amplitudes in terms of the quantum
mechanical wave functions.
 34

Figure 18: Normal and anomalous cuts in the q 2 -plane for the deuteron form-factor

7 Dispersion relations
The study of Feynman diagrams tells us that scattering amplitudes are analytic functions of
any invariant variable with singularities determined by the Landau equations and discontinu-
ities determined by the Cutkosky rules. In absence of anomalous, non-unitary singularities,
the discontinuities are given by the generalized unitarity relations (which include non-phyiscal
regions).
Returning to our example of πN scattering amplitude we find that at fixed t and thus
varying s and u the amplitude A(s, t) is an analytic function with poles at s = m2 and u = m2
and unitary cuts at s ≥ (m + µ)2 and u ≥ (m + µ)2 . In the complex s-plane the phyisical
amplitude is given by the values of A(s, t) on the real axis above the right-hand cut in the
s-channel and below the left-hand cut in the u-channel. The discontinuities across the cuts
2iAs and −2iAu are given by the unitarity relations in the s- and u- channels, respectively.
As an example, in the elastic region (m + µ)2 < s < (m + 2µ)2 one finds

k
Z
As (s, t) = dΩ1 A∗ (s, t1 )A(s, t2 ), (125)
32π 2 W
where
t = −2k2 (1 − cos θ), t1 = −2k2 (1 − cos θ1 ), t2 = −2k2 (1 − cos θ2 ),

dΩ1 = d cos θ1 dφ1 , cos θ2 = cos θ cos θ1 + sin θ sin θ1 cos φ1

and k2 id the c.m. momentum squared. We stress that this is a generalized unitarity equation,
which is assumed to be valid at any fixed t including values outside the physical region
−4k2 < t < 0 (see Fig. 4). In the unphysical region one finds | cos θ| > 1 so that cos θ2 turns
out to be complex. So in the integrand there appears A(s, t) at complex values of t. This
means that to find As in the unphysical region one has to be able to analytically continue
As, t in the variable t at fixed s.
From the perturbation theory we conclude that this is indeed possible. At fixed s variables
t and u are varying. The amplitude is found to be an analytic function of these variables with
singularities along the real axis at old points u = m2 , u ≥ (m + µ2 ) and new point t > 4µ2
dictated by the unitarity in the t-channel (see Fig. 19). Note that this last singularity lies in
the deep unphysical region, since the physical region starts from t > 4m2 >> 4µ2 .
Likewise the amplitude can be considered as an analytic function of varying variables
s and t at fixed u. In this case we meet with already mentioned singularities at s = m2 ,
s ≥ (m + µ)2 and t ≥ 4µ2 .
 35

Figure 19: Contribution from the intermediate two-pion state to the unitarity in the t-channel

Figure 20: Integration contour in the s′ -plane

It is well-known that the analytic function is completely determined by its singularities.

Dispersion relation is a technical tool to restore the amplitude from given discontinuities
across its cuts.
Let us again consider our A(s, t) for πN scattering. Let us fix t and study A(s, t) as a
function of complex s and u. Let us first assume that A(s, t) vanishes at |s| → ∞:

A(s, t) → 0, at |s| → ∞. (126)

Consider the relation

ds′ A(s′ , t)
Z
= 2πiA(s, t), (127)
C s′ − s
where s is complex and the integral is taken along the closed contour C, shown in Fig. 20,
which embraces the cuts of the amplitude A(s, t) and closes on the large circle. The relation
(127) is valid since A(s, t) is an analytic function inside the integration contour, so that the
only singularity of the integrand is a pole at s′ = s. Due to (126) the integral along the large
circle goes to zero as it radius grows and in the limit we find that only the contribution from
the discontinuities along the cuts remain in (127). So we find the so called dispersion relation

1
Z ∞ ds′ As (s′ , t) 1
Z s(u=m2 ) ds′ Au (u(s′ ), t)
A(s, t) = −
π m2 s′ − s π −∞ s′ − s
 36

1 ∞ ds′ As (s′ , t) 1 ∞ du′ Au (s′ , t)

Z Z
= + . (128)
π m2 s′ − s π m2 u′ − u
Here u(s) = 2(m2 + µ2 ) − s − t and u′ = u(s′ ) = 2(m2 + µ2 ) − s′ − t so that s′ − s = u − u′ .
In fact at s < (m + µ)2 and u < (m + µ)2 the only contribution comes from pole terms:

g2 g2 1 ∞ ds′ As (s′ , t) 1 ∞ du′ Au (s′ , t)

Z Z
A(s, t) = 2
+ 2
+ + . (129)
m −s m −u π (m+µ)2 s′ − s π (m+µ)2 u′ − u

Note that the crossing-symmetry leads to the equality of the two absorptive parts: Au (u, t) =
As (u, t).
If the amplitude does not vanish at |s| → ∞ then one constructs the so called dispersion
relation with subtractions. Let us assume, instead of (126), that there exists a integer N
such than
A(s, t)
→ 0 at |s| → ∞ (130)
(s − s0 )N
where s0 is an arbitrary finite point. Then the procedure employed above is repeated for the
function A(s, t)/(s − s0 )N . As compared to (127) we obtain additional terms coming from
the N -fold pole at s0 :
ds′ A(s′ , t) n A(s, t) 1 h d N −1 A(s′ , t) i
Z o
= 2πi + . (131)
C (s′ − s0 )N (s′ − s) (s − s0 )N (N − 1)! ds′ s′ − s s′ =s0
With the circle included on C extending to infinity we get the dispersion relation with sub-
tractions
(s − s0 )N 
Z ∞ ds′ As (s′ , t)
Z s(u=m2 ) ds′ Au (u(s′ ), t) 
As, t = − + PN −1 (s) (132)
π m2 (s′ − s0 )N (s′ − s) −∞ (s′ − s0 )N (s′ − s)

where PN −1 (s) is a polynomial in s of order N − 1. It represents N first terms in the

expansion of A(s, t) in the Tailor series around the point s = s0 . If one defines u0 by the
equality s − s0 = u0 − u then (132) can be rewritten in a more symmetric form

(s − s0 )N ∞ ds′ As (s′ , t) (u − u0 )N du′ Au (u′ , t) ∞

Z Z

As, t = + + PN −1 (s).
π m2 (s′ − s0 )N (s′ − s) ′ N ′
m2 (u − u0 ) (u − u)
π
(133)
What immediate use may be extracted from dispersion relations? An obvious possibility
is correlation of experimental data. At t = 0 both absorptive parts As and Au are related to
the experimental total cross-sections:
1
As (s, 0) = Im A(s, 0) = J12 σ tot (s). (134)
2
So taking s, say, in the physical region of the s-channel we get

g2 g2 1 ∞ ds′ J12 σ tot (s′ ) 1 ∞ du′ J12 σ tot (u′ )

Z Z
Re A(s, 0) = 2 + 2 + + . (135)
m − s m − u 2π (m+µ)2 s′ − s 2π (m+µ)2 u′ − u

Both left- and right-hand parts of this relation can be determined from the experiment
independently. Thus we obtain a possibility to check causality in our world.
A more ambitious goal is to try to determine the amplitude itself using dispersion relations
and unitarity. It is possible if one restricts to the elastic unitarity. As we have seen, even in
 37

this approximation one needs to be able to analytically continue the unitarity in variable t.
The simplest possibility is to assume certain simple analytical t-dependence of the amplitude.
Let us see, as an example, this scheme in the s-wave approximation, that is when A(s, t) does
not depend on t. In this case we get the dispersion relation (without subtractions)
g2 g2 1 ∞ ds′ As (s′ ) 1 ∞ du′ Au (s′ )
Z Z
A(s, t) = 2
+ 2
+ + . (136)
m −s m −u π (m+µ)2 s′ − s π (m+µ)2 u′ − u
It is convenient to pass to variable ω, which has the meaning of the pion energy in the Breit
system, related to s and u by

s = m2 + µ2 + 2mω, u = m2 + µ2 − 2mω, (137)

so that
g2 g2 2g2 µ2 f2
+ = = . (138)
m2 − s m2 − u 4m2 ω 2 − µ4 ω 2 − ω02
Here ω0 = µ2 /(2m), f 2 = g2 µ2 /(2m2 ). The dispersion relation (136) acquires the form
f2 1 ∞ dω ′ 2 As (ω ′ )
Z
A(ω) = + . (139)
ω 2 − ω02 π µ2 ω′2 − ω2
Now we add to this the unitarity relation in the elastic approximation
k
A(ω) = 4π|A(ω)|2 ≡ ρ(ω)|A(ω)|2 , (140)
32π 2 W
√
where k2 = ω 2 − µ2 , and W = s = m2 + µ2 + 2mω. As a result we get a non-linear
p

integer equation to determine A(ω):

f2 1 ∞ dω ′ 2
Z
A(ω) = 2 + ρ(ω ′ )|A(ω ′ )|2 . (141)
2
ω − ω0 π µ2 ω′2 − ω2
This equation can be solved by a simple trick. Let z = ω 2 . Consider the function
f2
D(z) = A−1 (z). (142)
z − ω02
This is an analytic function with a cut along the real axis µ2 < z < ∞, equal to unity at
z = ω02 and tending to a constant at |z| → ∞. So we can write a dispersion relation with one
subtraction at z = ω02 :
z − ω02 ∞ dz ′
Z
D(z) = 1 + Im D(z ′ ). (143)
π µ2 z′ − z
The unitarity relation gives
f 2 Im A(z) f2
Im D(z) = − = − ρ(z) (144)
z − ω02 |A(z)|2 z − ω02
So we find
z − ω02 ∞ dz ′ ρ(z ′ )
Z
D(z) = 1 − f 2 (145)
π µ2 (z ′ − ω02 )(z ′ − z)
and the amplitude is found as
f2  2
2 z − ω0
∞ dz ′ ρ(z ′ ) −1
Z
A(z) = 1 − f (146)
z − ω02 π µ2 (z ′ − ω02 )(z ′ − z)
 38

Figure 21: π − π elastic scattering

8 The Froissart theorem

8.1 Partial wave expansion
In this section, for simplicity, we shall consider an especially simple case when all masses of
the scattering particles are equal. An adequate physical example is π − π elastic scattering:

π(k1 ) + π(k2 ) → π(k1′ ) + π(k2′ ). (147)

In this case the relation between the c.m. energy and momentum is simplified to
1
k2 = (s − 4µ2 ). (148)
4
The transferred momenta squared are

t = −2k2 (1 − cos θ), u = −2k2 (1 + cos θ), (149)

where θ is the scattering angle in the c.m. system. They vary in the interval

−4k2 < t, u < 0 (150)

with t = 0 for the forward scattering and u = 0 for the backward scattering. For convenience
we denote the cosine of the scattering angle as
t u
z = cos θ = 1 + = −1 − 2 . (151)
2k2 2k
Let us expand the scattering amplitude at fixed s in the Legendre series in z:
  X
A s, t(z) = (2l + 1)al (s)Pl (z). (152)
l=0

Coefficients al (s) are scattering partial waves. They can be found from the amplitude ac-
cording to
1 1 
Z 
al (s) = A s, t(z) Pl (z). (153)
2 −1
where we us the property of the Legendre polinomials
Z +1 2
dzPl (z)Pl′ (z) = δll′ . (154)
−1 2l + 1
 39

Partial wave expansion radically simplifies the elastic unitarity relation:

k
Z
Im A(s, t) = √ dΩ1 A∗ (s, t1 )A(s, t2 ) (155)
32π 2 s

valid in the interval 4µ2 < s < 16µ2 . Here

q
2 2
dΩ1 = dz1 dφ1 , t1 = −2k (1 − z1 ), t2 = −2k (1 − z2 ), z2 = zz1 + cos φ1 (1 − z 2 )(1 − z12 ).

If we put here the expansion (152) and use

4π
Z
dΩ1 Pl1 (z1 )Pl2 (z2 ) = δl l Pl (z) (156)
2l + 1 1 2 1
we find that the unitarity relation (155) is diagonalized in the orbital momentum l
k  2  2
Im al (s) = √ al (s) ≡ ρ(s)al (s) . (157)
8π s
As is well known this relation is solved by introduction of the scattering phases. We find

Im a−1
l (s) = −ρ(s)

, so that one can take

1 iδl (s)
a−1
l (s) = ρ(s)(cot δl (s) − i), or al (s) = e sin δl (s). (158)
ρ(s)
The phases δl (s) remain real while the scattering is elastic and acquire a negative imaginary
part above the threshold of inelastic channels.
Above the inelastic threshold to the right-hand side of (155) inelastic contributions are
added. It is important that at fixed orbital momentum they are positive, since both the
initial and final two-particle states are essentially non-degenerate: at fixed l they only differ
in their z projection lz , which is conserved. So both the initial and final states are given by
Ψl,lz and the contribution from inelastic states has the structure

| < Ψn |Ψl,lz > |2 > 0.

X

As a result we have an inequality

 2
Im al (s) ≥ ρ(s)al (s) , (159)
 

√
which is valid at all values of energy s.
Now we are going to use analytic properties of the amplitude as a function of t. We can
write a dispersion relation (without subtractions) at fixed s:
1 ∞ dt′ At (s, t′ ) 1 ∞ du′ Au (s, u′ )
Z Z
A(s, t) = + , (160)
π 4µ2 t′ − t π 4µ2 u′ − u

where of course u = 4µ2 − s − t and u′ = 4µ2 − s − t′ . Both t and u are just linear functions
of z. Therefore one can rewrite this dispersion relation in terms of z:
1 ∞ dz ′ At (s, t(z ′ )) 1 −z0 (s) dz ′ Au (s, u(z ′ ))
Z Z
A(s, t) = ′
− (161)
π z0 (s) z −z π −∞ z′ − z
 40

Figure 22: Analytic properties of π − π amplitude at fixed s as a function of t or z

where
2µ2
z0 (s) = 1 + . (162)
k2

Putting this representation into (153) we get

1
Z ∞ 1
Z −z0 (s)
al (s) = dzAt (s, t(z))Ql (z) − dzAu (s, u(z))Ql (z), (163)
π z0 (s) π −∞

where Ql (z) is the Legendre function of the second kind, defined by

1 1 dz ′ Pl (z ′ )
Z
Ql (z) = . (164)
2 −1 z − z′

At fixed integer l function Ql (z) is an analytic function with a cut at −1 < z < 1. It has
the property
Ql (−z) = (−1)l+1 Ql (z) (165)

and the following asymptotic properties. At large |z| at fixed l with Re l¿0

Pl (z) ∼ z l , Ql (z) ∼ z −l−1 . (166)

At large |l| with Re l > 0

at |z| < 1, z = cos φ
1   1  
Pl (z) ∼ √ cos (l + 1/2)φ − π/4 , Ql (z + i0) ∼ √ exp i(l + 1/2)φ − π/4 , (167)
l l
at z > 1, z = cosh α
1 1
Pl (z) ∼ √ e(l+1)α , Ql (z) ∼ √ e−(l+1)α . (168)
l l
To avoid negative values of z we can use (165) and rewrite (163) as
1
Z ∞  
al (s) = dzQl (z) At (s, t(z)) + (−1)l Au (s, u(−z)) . (169)
π z0 (s)

It is convenient to introduce two amplitudes with a fixed signature ±:

1
Z ∞
(±)
 
al (s) = dzQl (z) At (s, t(z)) ± Au (s, u(−z)) . (170)
π z0 (s)
 41

(+)
The positive signature amplitude al (s) coincides with the physical amplitude al (s) at even
(−)
values of l, the negative signature amplitude al (s) at odd values of l.
From (169) we can extract, first of all, the threshold behaviour of the partial wave ampli-
tude. We return to variable t in (170)

1
Z ∞ dt  t 
(±)

al (s) = Q l 1 + A t (s, t) ± A u (s, t)) , (171)
π 4µ2 2k2 2k2

where we use u(−z) = t(z). As k2 → 0 we find from (171)

(±)
al (s) ∼ k2l . (172)

This properrty forms the basis for the analysis of hadron scattering at low energies. At
the energies close to the threshold only the partial wave with l = 0 (’s-wave’) gives a non-
vanishing contriburion. As the energy grows other waves with higher and higher waves
gradually appear(’p-wave’ with l = 1, ’d-wave’ with l = 2 etc). Actually this result follows
due to the finite range of the strong interactions determined by the finite value of the pion
mass µ in (171). Should the exchanged particle have its mass equal to zero (the photon) the
situation would change radically and become totally different.
For our purpose the most important conclusion is about the behaviour of the partial wave
amplitudes at large l > 0. From (171) we find

(±) 1
al (s) ∼ √ e−(l+1)α0 f (s), (173)
l
where
2µ2
cosh α0 = 1 + (174)
k2
and f (s) ∼ At (s, 4µ2 ), Au (s, 4µ2 ).

8.2 High-energy behaviour

At s >> µ2 we find k2 = s/4 and obviously α0 << 1. So

1 8µ2
cosh α0 = 1 + α20 = 1 + ,
2 s
from which we find
4µ
α0 = √ , (175)
s
so that at both l and s large we have the asymptotic
1 √
(±)
al (s) ∼ √ e−(l+1)4µ/ s f (s). (176)
l
This asymptotic plus the unitarity relation form the basis for derivation of the restriction
on the high-energy behavour of the scattering amplitude, known as the Froissart theorem.
One additional fundamental ingredient is the assumption that function f (s) in (173) and
(176) may grow with s not faster than a certain power of s. This assumption is equivalent to
 42

the assumption that the amplitude does not have an essential singularity at infinity, which
would spoil the possibility to pass to the Eucledean formulation.
To start the derivation we first use the unitarity property (159). For a reduced amplitude
ãl (s) = ρ(s)al (s) we find an inequality

Im ãl (s) ≥ |ãl (s)|2 . (177)

Introducing real and imaginary parts ãl (s) = x + iy we rewrite it as

y ≥ x2 + y 2 . (178)

Solution of the equation y = x2 + y 2 gives y = 1/2 ±

p
1/4 − x2 , from which we conclude
1
|x| ≤ , 0 ≤ y ≤ 1, x2 + y 2 ≤ 1,
2
which transforms into the inequality
1
|al (s)| ≤ → 16π at s >> µ2 . (179)
ρ(s)

Now we pass to our partial wave expansion (152) at large s. Using the properties of the
Legendre polynomials we have
  X
A(s, t) ≤ (2l + 1)|al (s)|. (180)
 
l=0

Here we split the summation into two parts 0 ≤ l ≤ Λ(s) − 1 and Λ(s) ≤ l < ∞, where we
assume that Λ(s) >> 1. In the first sum we use the unitarity limitation (179) to find
Λ−1 Λ−1
(2l + 1) ≃ 16πC1 Λ2 (s),
X X
S1 ≡ (2l + 1)|al (s)| ≤ 16π (181)
l=0 l=0

where C1 is an inessential constant and we used Λ >> 1.

In the second sum we use the asymptotical behaviour (176):
∞ ∞ √
1
(2l + 1) √ e−(l+1)4µ/ s+ln f (s) .
X X
S2 ≡ (2l + 1)|al (s)| ∼ (182)
l=Λ l=Λ l

Here we put l = l′ + Λ and choose

√
s
Λ(s) = ln f (s) (183)
4µ

to cancel the second term in the exponent in (182). Note that at large s
√
Λ(s) ∼ c s ln s (184)

where c is some constant. We get

∞ √ √ ∞ √ √ √
′ ′
l′ + Λe−l 4µ/ s
Λ)e−l 4µ/ s
X X
S2 = 2 ≤2 ( l′ + , (185)
l′ =0 l′ =0
 43

where we once more used Λ >> 1. The two sums which appear may be estimated by
substituting the sums by integrations:
s√ √
Z ∞ √ √ ′ √  s √  s
S2 ≤ 2 dl ( l′ + Λ)e−l 4µ/ s ∼ 2
′
+ Λ < 16πC1 Λ2 (s). (186)
0 4µ 4µ

So the second sum is less than the first and we obtain the restriction of Froissart
 
A(s, t) ≤ Cs ln2 s. (187)
 

Recalling the optical theorem and the asymptotic J12 (s) → 2s at s >> µ2 we find limitation
on the behaviour of the total cross-section at high energies

σ tot (s) ≤ C ln2 s. (188)

One can prove that at t strictly smaller than zero a more stringent limitation is valid
 
A(s, t) ≤ C ′ s ln s, t 6= 0, t < 0. (189)
 
 44

9 Complex angular momenta

9.1 Definition and properties
For reasons which will be clearer later, we shall study the partial wave expansion in the
t-channel: ∞ X
A(s, t) = (2l + 1)al (t)Pl (zt ), (190)
l=0
where zt is the cosine of the scattering angle in the t-channel:
1
s = −2kt2 (1 − zt ), u = −2kt2 (1 + zt ), kt2 = (t − 4µ2 ). (191)
4
At fixed t the s- dependence is concentrated in Legendre polynomials and is explicit and
simple. The basic idea is to generalize this formula to be valid at large values of s to describe
√
the scattering at high energy s and fixed t. In this form however it does not suit to this aim:
the Legendre expansion (190) converges in the ellipse in the zt plane with focuses at points
±1 and limited by the nearest singularity in zt which corresponds to s = 4µ2 or u = 4µ2 . It
certainly does not converge at large values of |s| and thus large |zt |.
The technical instrument for the extension of representaton (190) to large values of s is
transition to complex angular momenta. Partial waves with complex angular momenta can
be introduced starting from one of the two formulas for their values at integer l, which we
had in the previous section, (153) and (193)
1
Z 1  
al (t) = dzt A s(zt ), t Pl (zt ), (192)
2 −1
1
Z ∞
(±)
 
al (t) = dzt Ql (zt ) As (s(z), t) ± Au (u(−zt ), t) . (193)
π 1+2µ2 /kt2
Both formulas admit analytic continuation from integer values of l to any complex values of l,
since Pl (zt ) and Ql (zt ) are both analytic functions of l at fixed zt 6= ±1. Function Ql (zt ) has
simple poles at negative integer l = −1, −2, .... So in principle both representations may serve
to define partial waves with complex angular momenta. At first sight the simpler formula
(192) is preferable, since the integration in it goes over a finite interval and no problem
with convergence arise. However in reality it is the second formula (193) which leads to
constructive results. In fact one must have in mind that continuation from integer values to
complex ones is not unique. Obviously one can add to any chosen continuation an arbitrary
analytic function proportional to sin(πl) and so vanishing at integer l. So the problem is to
choose a particular continuation suitable for our purpose. As we shall see, this continuation
is realized by Eq. (193).
Eq. (193) defines an analytic function of l provided the integral converges. If at large
values of zt the function in the integrand (the signatured absorptive parts)

A±
s (s, t) = As (s, t) ± Au (u, t) ∼ s
N
(194)

then the total dependence of the integrand on z t at large zt has the form ∼ zt−l−1+N and
obviously the integral converges while

Re l > N. (195)
 45

(±)
Figure 23: Analyticity region of al (t) defined by (193)

So Eq. (193) defines an analytic function of l in the right half-plane restricted by (195).
At very large l in this half-plane the continued partial amplitudes behave as derived in the
previous section
1 2µ2
a(±) (t) ∼ √ e−(l+1)α0 f (t), cosh α0 = 1 + 2 (196)
l kt
and thus decrease in the right half-plane at large Re l.
It is important to have in mind that generally N = N (t) and so the analyticity region
or equivalently position of singularities in the l plane depend on t. We already have some
information about the possible behaviour of the amplitudes (and so their absorptive parts)
as function of s at physical values of t ≤ 0. According to the Froissart theorem, at these
(±) (±)
values of t As < Cs ln2 s. This means that in any case at t ≤ 0 the partial waves al are
defined by (193) and analytic to the right of the line Re l=1.
One can make a somewhat stronger conclusion. Consider a dispersion relation at fixed
t ≤ 0. According to the Froissart limitation we need no more than two subtractions:

(s − s0 )2 ∞ ds′ As (s′ , t) (u − u0 )2 ∞ du′ Au (u′ , t)

Z Z
A(s, t) = as + b + + . (197)
π 4µ2 (s′ − s0 )2 (s′ − s) π 4µ2 (u′ − u0 )2 (u′ − u)

Let us take s → ∞ and so u → −∞. Let us assume for simplicity that at large values of s
and u As ∼ cs s and Au ∼ cu u. Then the leading contribution to the integrals will come from
large values of s′ and u′ . We then approximately have

(s − s0 )2 ∞ ds′ As (s′ , t) s2 ∞ ds′ cs s′ s s1 − s

Z Z
′ 2 ′
∼ ′2
∼ cs ln , (198)
π 4µ2 (s − s0 ) (s − s) π s1 s (s′ − s) π s1

where we have chosen s1 >> µ2 , s0 to be able to use the asymptotic of As . So the leading
contribution from the right-hand cut is real and grows logarithmically with s, that is, faster
than its imaginary part As . A similar contribution will come from the left-hand cut to give
the total amplitude at large s as
s s u  u 
A(s, t) ∼ cs ln + cu ln   + O(s). (199)
π s1 π u1
As we observe the amplitude is generally real and so grows faster that its imaginary part.
This does not seem natural and presumably contradicts the unitarity, which states that the
imaginary part is roughly speaking proportional to the square of its real and imaginary
parts. However there is a possibility to avoid this circumstance. If cs = cu then the leading
logarithmic contribution cancels out, since s + u does not grow at all. This implies that at
 46

large energies one has As ∼ Au . At t = 0 this means that the total cross-sections in the s
and u channels become equal. For particles with charges the channels in questions are

a + b → a + b (s − channel), and a + b̄ → a + b̄, (u − channel). (200)

So our conclusion may be reformulated as a statement that at large energies the total cross-
section for scattering off a particle is equal to that off its antiparticle:
tot
σab (s) ≃ σatot
b̄ (s), s → ∞ (201)

(the Pomeranchuk theorem).

We derived this conclusion assuming a simple behaviour of the absorptive arts at large
energies. However it can be generalized for more general cases, in which this behaviour is
changed by logarithms.
For the signatured amplitudes the Pomeranchuk theorem implies that at high energies the
positive signature absorptive part, defined by (194) grows faster that the negative signature
one:
(−)
As (s, t)
(+)
→ 0, s → ∞ (202)
As (s, t)
(−)
and as a consequence, at t ≤ 0 the negative signature partial waves al (t) are generally
(+)
analytic in a wider region of l than al (t).
The advantage of the continuation by means of (193) follows from the fact that it is unique
in the sense, that it is the only continuation which decreases in the right-half plane of l. This
follows from the Karlsson theorem:
***************************************
Karlsson theorem
If function f (l) is analytic at Re l ≥ N , has asymptotic O(eκ |l|) with κ < π and f (l) = 0
at l = N, N + 1, N + 2, .... then it is equal to zero: f (l) = 0.
Because of this property it is not reasonable to analytically continue the partial waves
using (192).
Uniqueness of the continuation with the mentioned properties allows to demonstrate many
useful properties of the partial waves with complex l defined by (193).
***************************************************
For the following it is helpful to study the analytical properties of the introduced ampli-
tude as a function of complex t
Analytic properties of aξl (t) in the t complex plane
Passing to integration over s we have
1
Z ∞ ds
aξl (t) = Ql (zt )Aξs (t, s) (203)
π 4µ2 2kt2
where ξ = ± is the signature
s
zt = 1 +
2kt2
and
Aξs (t, s) = As (t, s) + ξAu (t, s).
 47

(±)
Figure 24: Analyticity region in the complex t plane of al (t) defined by (193)

For simplicity we study aξl (t) for integer l.

At t away from the real axis aξl is obviously analytic, since the integrand is analytic in
(203). Singularities of aξl (t) lie on the real axis. They come both from Aξs (t, s) and Q(zt ) in
the integrand of (203).
Function
1
Aξs (t, s) = (Aξ (t, s + i0) − Aξ (t, s − i0)) (204)
2i
is the discontinuity of the amplitude on the righthand cut in the s complex plane. It appeared
before at t ≤ 0 as the righthand part of the unitarity relation in the s-channel and was
obviously real. In (203) we actually find its analytic continuation for arbitrary complex t.
So it ceases to be real and according to (204) aquires singularities of Aξ (t, s ± i0) considered
as a function of t. These singularities at fixed s are the cuts at t > 4µ2 and u > 4µ2 that is
s + t < 0 which gives the left cut at t < −s. After integration over s > 4µ2 this produces
generates a righthand cut in the complex t-plane at 4µ2 < t < ∞ and a lefthand cut at
t < −4µ2 .
However there are more singularities which come from Ql (zt ). This function has a cut at
s
−1 < zt < 1, zt = 1 + .
2kt2

The integration over s in (203) goes over s > 4µ2 . If kt2 > 0 we have zt > 1 and we are
outside the cut. So the singularity coming from Ql (zt ) can occur only at negative values of
kt2 . Then zt < 1 and the cut occurs at
s
−1 < 1 + .
2kt2

This inequalty gives

−4kt2 > s, or t < 4µ2 − s.

Since s ≥ 4µ2 we find a lefthand cut covering all the negative axis of t, at t ≤ 0.
In conclusion as a function of t partial amplitude aξl (t) is an analytic function with two
cuts on the real axis, the rigthand one at t ≥ 4µ2 and the left one at t < 0 (see Fig. 24)
Using uniqueness of analytic continuation from the integer l to the right complex l half-
plane we can prove the unitarity relation in the t-channel for complex values of l. To do
this we first define a new function, also analytic in l, using the complex conjugate absorptive
parts in (193):
1 ∞
Z
(±) ∗
āl (t) = dzt Ql (zt )A±
s (s(z), t). (205)
π 1+2µ /kt
2 2
 48

(±) (±)
It is analytic in l in the same region as al (t) and coincides with it when As (s(z), t) is
(±)
real. We have seen that As (s(z), t) is real at s > 4µ2 outside of is right and left cuts at
t > 4µ2 and u > 4µ2 respectively, that is in the interval

−4µ2 < t < 4µ2 .

So on this interval of t
(±) (±)
āl (t) = al (t), at − 4µ2 < t < 4µ2 (206)

for arbitrary complex l. We also note that at real l and zt function Ql (zt ) is real. From this
we conclude that at real zt Ql∗ (zt ) = Q∗l (zt ), and consequently
(±) ∗ (±)
[āl ] (t) = al∗ (t), at real t. (207)

Now consider the elastic unitarity in the region 4µ2 < t < 16µ2 at integer values of l of
the appropriate signature:
(±) (±) (±) (±)
al (t) − āl (t) = 2iρ(t)al (t)āl (t), at 4µ2 < t < 16µ2 , (208)
√
where ρ(t) = kt /2π t and we used (207) at integer l. Considered as functions of complex l
both the left- and right-hand sides of (208) are analytic in the right half-plane and decrease
as Re l → ∞. Therefore they are equal at any complex l. Thus the unitarity relation (208)
is valid at any complex l in the analyticity region.

9.2 Regge poles

Now let us try to understand what kind of singularities may exist in partial waves in the com-
plex l plane, which limit the analyticity region. As in the previous studies we can distingush
between the fixed singularity points l = ls which do not depend on t and moving singularities
l = ls (t) whose position depends on t.
Elastic unitarity in the form (208) prohibits existence of fixed singularities. In fact assume
that the partial amplitude al (t) (of any signature, which we suppress in the following) has a
fixed pole at l = l0
b(t)
al (t) ∼ , (209)
l − l0
where l0 is t-independent. Taking |t| < 4µ2 we also find
b(t)
āl (t) = al (t) ∼ , |t| < 4µ2 .
l − l0
Now we continue this relation in t to the interval t > 4µ2 . Since at t = 4µ2 the amplitude
has a branch point we find that at t > 4µ2
b∗ (t)
āl (t) ∼ , t > 4µ2 (210)
l − l0
with a complex conjugate b but the same l0 . If we put (209) and (210) into (208) we encounter
a contradiction: the left-hand part has a simple pole at l = l0 but the right-hand part has a
double pole. This conclusion can be easily generalized to any type of singularities.
 49

Figure 25: A pole in a±

l (t) on the second sheet of complex t

Thus if we assume (208) the only possible singularities may only be moving l = ls (t).
Moreover if we assume (208) true at any values of t > 4µ2 these singularity may only be
poles at l = α(t) which are called Regge poles. Indeed singularities in l depending on t can
also be considered as singularities in t depending on l. From the definition of al (t), Eq. (193)
one can conclude that in the analyticity region of l and as a function of t on the physical
sheet of complex t plane it cannot have any l-dependent singularities (the only singularities
come either from those of A(±) s(t, s(zt )) or from Ql (zt ) whose position does not depend on
l). However it can have such singularities on the second sheet of the complex t plane related
with the unitarity cut at t > 4µ2 . Expressing from the unitarity the amplitude above the
unitary cut in terms of the same amplitude below the cut
(±)
(±) āl (t)
al (t) = (±)
, (211)
1 − 2iρ(t)āl (t)

The right-hand side is analytic below the cut. So it gives the analytic continuation of the
amplitude to the second sheet of the complex t plane. We observe that the amplitude is
analytic on this second sheet except for possible poles at points where
(±)
1 − 2iρ(t)āl (t) = 0. (212)

This is illustrated in Fig. 25.

This equation defines the so-called Regge trajectory l = α(±) (t), which determines the
position of a moving pole of the amplitude a± l (t) (”Regge pole”), Regge poles stay on the
unphysical sheet of the complex t plane and so are not visible in the analyticity region of l
plane. However with the change of complex l they may appear on the physical sheet and
then become actual singularities of the amplitude.
The situation can be made more illustrative. Assume that in fact the colliding particles
form a true bound state in the t-channel with mass m. This is cerainly not the case for
realistic pions but the t channel, say, with a proton plus neutron indeed contains a stable
bound state, the deuteron with mass md < mp + mn . For better illustration assume that
this bound state has its orbital momentum l = 2. Then in the complex t plane the situation
 50

Figure 26: Movement of a Regge pole in the t and l plane in the presence of a stable bound
state at l = 2 with mass m

will look as shown in the left panel of Fig. 26. In the t plane we observe a moving pole at
t − t0 (l) which at t = t0 (2) crosses the real axis at point m2 below the threshold t = 4µ2 . This
movement is repeated by the Regge pole in the l plane l = α(t) shown in the right panel. At
t = m2 it crosses the real axis at l = 2. At t 6= m2 it stays in the complex plane away from
point l = 2.
As we observe with a change of interaction (the properties of the t channel), Regge poles
may appear on the physical sheet of t from the start. Then at integer l they have to be
situated below the threshold t = 4µ2 and correspond to bound states in the t channel. With
the change of l one obtains a family of bound states of spin l when the trajectory passes
integer values of l of the appropriate signature. If the poles in t remain on the second sheet
at t > 4µ2 the bound states transform into resonances.
(±)
To conclude, a Regge trajectory αl (t) describes a family of particles with rising spins
l=0, 2, 4... for positive signature and l=1, 3, 5,.. for negative signature with different (and
√
normally rising) masses ml = tl determined by the equation

α(+) (tl ) = l = 0, 2, 4, ..., α(−) (tl ) = l = 1, 3, 5...

If tl < 4µ2 these particles are stable but if tl > 4µ2 they are unstable, decaying into two pions
and are in fact resonances, which can be observed in the scattering cross-sections by peaks
in the mass distribution of finite width.
A Regge trajectory α(±) (t) unites into a family a group of particles superficially not
related to one another but in fact belonging to rotational excitations of the basic particle.
From the unitarity equation continued to complex value of l it follows that the contribution of
(±)
the Regge pole to the amplitude al preserves the same structure which is valid for normal
bound states or resonances at integer values of l, that is it has a factorizable residue with
factors which can be interpreted as a coupling of the intermediate particle with the initial
and final ones. Therefore the Regge trajectory may be considered as the position of a pole
corresponding to propagation of a certain quasi-particle with variable spin and mass, which
is called Reggeon, Fig. 27.
Apart from the signature, Reggeons carry all the usual internal quantum numbers cor-
responding to the property of the t-channel in which they appear. In our simple case they
obviously have the vacuum quantum numbers. However if the initial (and final) particle have
 51

Figure 27: The Reggeon exchange

non-trivial quantum numbers, electric charge, parity, isospin, strangeness, charm etc, the
intermediate Reggeon will have the same quantum numbers.
To conclude, we have to mention that the conclusion that the only possible singularities of
the complex angular momentum amplitudes may only be poles heavily rests on the assumption
that the unitarity relation is reduced to its elastic contribution. Inclusion of inelastic channel
allows to admit singularities of the branch-point type. Physically they correspond to the
appearance of more complicated structures than a single Reggeon in the intermediate states,
two, three and more Reggeons.A detailed study of these Regge cut singularities is beyond the
scope of the present lectures. In the following we shall mention situations when inclusion of
these singularities may be important.

9.3 Gribov-Froissart representation and high-energy asymptotic

To convert expansion (190) into an expression suitable for the study of large values of s
and consequently of zt let us transform it into an integral in the complex l plane following
Sommerfeld-Watson. We define the signatured amplitude by a similar expansion in terms of
signatured partial waves:
∞
(±)
A(±) (s, t) =
X
(2l + 1)al (t)Pl (zt ). (213)
l=0
We transform this sum into an integral
i dl(2l + 1)
Z
(±)
A(±) (s, t) = Pl (−zt )al (t), (214)
2 C sin πl
where contour C is shown in Fig. 28. At |zt | < 1 the integrand decreases at large Re l, since
√
function Pl (−zt ) ∼ (1 l) cos(φ) with |φ| < π so that the ratio Pl (−zt )/ sin πl is limited and
(±)
al (t) decreases. This allows to deform the contour to pass parallel to the imaginary axis as
shown in Fig. 28. b, so that (214) transforms into
i
Z η+i∞ dl(2l + 1)
(±) (±)
A (s, t) = Pl (−zt )al (t). (215)
2 η−i∞ sin πl
This new integral turns out to converge at any values of |zt | including large values. In fact,
(±)
since Re l is limited both Pl (−zt ) and al (t) are limited in the integrand and sin πl exponen-
tially rises. So the integrand decreases at large Im l and the integral converges. Representation
(215) is known as the Gribov-Froissart representation for the scattering amplitude.
 52

Figure 28: Original and deformed contour of integration in the complex l-plane

Figure 29: Deformation of the contour in the presence of a pole in the right half-plane

Note that in Fig. 28 it was assumed that the deformation of the contour did not encounter
any singularities of the partal wave a± l (t) in the right half-plane of l. This may be not so. In
±
fact al (t) may have a singularity (for instant, a pole) at some point in the right half-plane of
l. Then the deformed contour has to go round this singularity, as shown in the left panel of
Fig. 29. If the singularity is a pole one can separate its contribution by taking the apprpriate
residue, as shown in the right panel.
The importance of this representation consists in its simple dependence on the energetic
variable s at s → ∞. It is concentrated in the Legendre function Pl (−zt ). We recall that
zt = 1 + s/2kt2 = 1 + 2s/(t − 4µ2 ). In the physical region of the s channel at large energies
t < 0 and we find
2s
−zt ∼ 2 , so that Pl (−zt ) ∼ (−zt )l ∼ sl .
4µ − t
So at large s the behaviour of the amplitude in the Gribov-Froissart form is determined by
(±)
the rightmost singularities of the partial waves al , which prevent moving the contour to
the left. If this leading singularity occurs at l = ls then at large s the amplitude behaves as
sls and grows as sRe ls . So we find that the behaviour of the amplitude at large energies is
related to the analytical properties of the partial waves in the crossed channel considered as
a function of the complex angular momentum and is determined by its singularities with the
maximal value of their real part.
If this leading singularity is a Regge pole
β(t)
A(± ) ∼ (216)
l − α(t)
then its contribution to the amplitude (215) is obviously
(±) Pα (−zt )
AR (s, t) = −π(2α(t) + 1)β(t) = C(t)sα(t) . (217)
sin πα(t)
 53

Up to now we considered the signatured amplitude. Passing to the normal amplitude we

find: ∞ X
A(s, t) = (2l + 1)Pl (zt )al (t)
l=0
∞ ∞
1X  
(+) 1X  
(−)
= (2l + 1) Pl (zt ) + Pl (−zt ) al (t) + (2l + 1) Pl (zt ) − Pl (−zt ) al (t)
2 l=0 2 l=0
∞
1 XX  
(ξ)
= (2l + 1) Pl (zt ) + ξPl (−zt ) al (t). (218)
2 ξ=± l=0

Applying the Sommerfeld-Watson transformation and deforming the contour as before we

get
i X η+i∞ Pl (−zt ) + ξPl (zt )
Z
(ξ)
A(s, t) = dl(2l + 1)al (t) . (219)
4 ξ=± η−i∞ sin πl

The contribution from a Regge pole of a given signature ξ = ± will be

Pα (−zt ) + ξPl (zt )

AR (s, t) = −π(4α(t) + 1)β(t) . (220)
sin πα(t)

As we have seen, in the physical region of the s channel zt = −2s/(4m2 − t) and is negative
with a small negative imaginary part (above the unitary cut in s). At complex l function
Pl (zt ) has a cut at −∞ < zt < −1 and in this region
2
Pl (−zt ± i0) = e±iπl Pl (zt ) − Ql (zt ) sin πl. (221)
π
Function Ql (zt ) decreases at large zt so in the physical region of the s channel at large s and
negative zt we get
Pl (zt ) ≃ e−iπl Pl (−zt ). (222)

Thus the contribution from a Regge pole becomes

1 + ξe−iπα(t)
AR (s, t) = −π(4α(t) + 1)β(t)Pα(t) (−zt ) . (223)
sin πα(t)

The ’signature factor’

1 + ξe−iπα(t)
ζ(α(t)) = (224)
sin πα(t)
guarantees exclusion of the poles of the wrong signature in the amplitude. Indeed at α(ξ) (t) =
n, n=0, 1, 2,.. we have 1 + ξe−iπα(t) = 1 + ξ(−1)n . So if ξ = + the pole occurs only at even
values of n and if ξ = − the pole occurs only at odd values of n. At large s from (223) we
find the asymptotical behaviour

AR (s, t) ∼ −π(4α(t) + 1)β(t)(2µ2 − t/2)−α(t) ζ(α(t))sα(t) . (225)

The Froissart limitation implies that at t ≤ 0 Re α(t) ≤ 1.

 54

Figure 30: The Chew-Frautschi plot

’abelchfr

9.4 Application to physical processes. The Pomeron

According to the previous subsection we derived a relation between the existence of particles
and resonances at t > 0 in the t channel and the asymptotic of the amplitude at high energies
in the s-channel. Knowledge of the spectrum of particles observed in the t channel at t > 0
allows to determine the corresponding Regge trajectories on which the rotational particle
levels lie. The experimental data show that at comparatively small values of |t| all Regge
trajectories are linear functions of t and with a good approximation can be presented as:

α(t) = α(0) + α′ t. (226)

The numbers α(0) and α′ , are called intercept and slope, respectively. For all the channels
except for the vacuum channel, the value of α′ is practically the same and very close to

α′ = 1(GeV/c) −2 .

Intercepts are all smaller than unity and different for different channels. According to the
asymptotic (225) the leading contribution to the total cross-section determined by A(s, t = 0)
comes from the trajectory with the highest intercept. Some representative trajectories derived
from the observed spectrum of mesons are shown in Fig. ?? in the so-called Chew-Frautschi
plot which shows Re α(t) as a function of t. It is remarkable that the trajectories for the two
different signatures coincide: particles appear on them when the trajectory passes through
an integer value, irrespective of whether it is even or odd. The trajectory with the highest
intercept is the ω − ρ trajectory on which the well-known vector mesons ω and ρ lie. Its
intercept is close to 0.5. Next follows the π trajectory with intercept close to zero. Further
down lies the K trajectory with a negative intercept.
As we observe intercepts of all trajectories derived from the observed particles lie below
unity. So the amplitudes which are described by the exchange of the corresponding Reggeons
in the t channels grow with s more slowly than s and the relevant cross-sections fall with s.
 55

Figure 31: The pn exchange amplitude

Note that practically all the observed particles and consequently the corresponding Reggeons
have non-trivial internal quantum numbers (charge, isospin, strangeness etc). The exchange
of such Reggeons implies the exchange of quantum numbers between the colliding particles.
An example of such exchange reaction is the binary collision of the proton and neutron
in which the proton becomes the neutron and vice versa. Such a reaction can indeed go
via the exchange of an ω − ρ Reggeon, as illustrated in Fig. 31. According to (225) the
√ √
corresponding amplitude at large s behaves as s and the cross-section decreases as 1/ s.
Experimental study of this and similar exchange reactions has confirmed that their cross-
sections indeed decrease at high energies in full correspondence with the predictions following
from the picture, in which they go via the exchange of an ω−ρ Reggeon. In general description
of all exchange reaction is well described by exchanges of Reggeons derived from the observed
particle spectrum in the relevant t-channels.
A striking exception is the total cross-section for which the t channel has to bear vacuum
quantum numbers. Experimentally there has been found no sign of particles with vacuum
quantum numbers which may be related to a Regge trajectory with an intercept close to unity
to account for the experimental behaviour of the total cross-section, which certainly does not
decrease but rather grows with energy. This behaviour implies existence of a singularity in the
(+)
positive signature vacuum t channel amplitude al (t) which has been called Pomeranchuk
singularity, or simply Pomeron. The simplest assumption is that the Pomeron is just a Regge
pole with intercept α(0) = 1. In this case the forward scattering amplitude will grow linearly
with s at large s and the cross-sections will tend to a constant value. Experimental data,
as mentioned, rather indicate that cross-sections grow with s. Since such a growth cannot
be powerlike according to the Froissart theorem, but at most be proprtional to ln2 s, this
implies that in fact the Pomeron is not a simple pole but a more complicated singularity. At
present the exact nature of the pomeron is not fully understood. So we shall discuss some
consequences following from the assumption that it is just a simple pole with unity intercept,
that is with the trajectory
αP (t) = 1 + α′P t. (227)

No physical particles have been observed lying on this trajectory. This can be explained by
the fact that α′P is small as compared to non-vacuum trajectories. From some (not very
reliable) experimental data to be discussed in the following one can conclude that

α′P ≃ 0.2 (GeV/c)−2 . (228)

 56

Turning to the expression (224) for the signature factor we find at small t for the Pomeron
′
1 − e−iπαP t
ζP (t) = = −i. (229)
− sin πα′P t

So at small t the contribution from the Pomeron to the amplitude is

π  s 1+α′ t ′
= icseαP t ln(s/s0 ) ,
P
AP (s, t) = i 3βP (0) 2
(230)
4 2µ
where we left only that t dependence which is increased by the factor growing with s. As we
observe the amplitude is pure imaginary and at t = 0 grows with s linearly, which corresponds
to a constant total cross-section (σ tot = c).
It is instructive to calculate the differential cross-section for elastic scattering. According
to the general formulas

dσ el 1 k  2 1  2
= √ 
A P (s, t)

= A P (s, t) ,

(231)
dΩ J12 16π 2 s 64π 2 s
  

√
where we have used that at s → ∞ we have k/ s → 1/2, J12 → 2s. It is convenient to
present
dσ el dσ el s dσ e l
= ≃ , (232)
dΩ 2πdz 4π dt
from which we find
dσ e l 1  2 c2 2α′ t ln(s/s0 )
= A (s, t) = e P . (233)

P
dt 16πs2 16π
 

We see that the width b(s) of the forward cone in t

 
b(s) = 1/ 2α′P ln(s/s0 ) + const

is decreasing with s as
1/ ln s.

With the growth of energy particles are emitted in the cone which becomes narrower. The
experimental observations give contradictory results. The value (228) has been extracted
from some data. But other data are compatible even with α′ = 0 when no change in the cone
width occurs with the growth of s.
The total elastic cross-section is going down at high energies:

dσ el
0 c2 0 c2
Z Z
el
σ ≃ dt = dtet/b = b(s). (234)
−∞ dt 16π −∞ 16π
Thus the contribution of the elastic channel decreases. as compared to the inelastic ones.
The experimental data do not confirm this prediction from the simple form of the Pomeron
singularity (230).
 57

Figure 32: Production amplitude

10 Multiregge asymptotic and inclusive cross-sections

10.1 Production amplitude and its high-energy asymptotic
Let us study a general production amplitude, illustrated in Fig. 32 According to our general
rule it depends on 5 variables. For these variables we may, for instance, choose 3 energetic
variables
s12 = (k1 + k2 )2 , s34 = (k3 + k4 )2 , s45 + (k4 + k5 )2

and two variables of the transferred momentum type

t13 = (k1 − k3 )2 , t25 = (k2 − k5 )2

(the concrete choice of the final momenta k3 , k4 and k5 will be done later). However in the
following sometimes it will be more convenient to choose other sets of 5 variables.
We shall consider it at high energies when the initial energy squared

s12 = (k1 + k2 ) → ∞. (235)

To simplify we assume all particle massless, since at high energies non-zero masses will gen-
erally be unimportant. So
ki2 = 0, i = 1, 2, ...5, (236)

which will simplify the kinematics. In the c.m. system k1 + k2 = 0 we direct the initial
particles along the z-axis k1⊥ = k2⊥ = 0. Then obviously k10 = k20 = k1z > 0.
It is convenient to introduce light-cone variables. For any 4-momentum k we define
1
k± = √ (k0 ± kz ). (237)
2
√
They are positive for physical particles. Then k1+ = 2k1z , k1− = 0 and k2+ = 0, k2− =
√
2k1z = k1+ . Variable s12 in terms of these light-cone momenta is written as

2
s12 = 2k1 k2 = 2k1+ k2− = 2k1+ . (238)

Now let us consider final momenta k3 , k4 and k5 . Their sum is

k3 + k4 + k5 = k1 + k2 . (239)
 58

Since k1+ and k2− tend to infinity, at least one of the three momenta k3 , k4 and k5 has to
possess a large ’+’- component and at least another one has to posses a large ’-’-component.
All the experimental data indicate that at large energies the produced particles have
their transverse momenta much smaller and growing with energy much weaker than the
longitudinal ones. In the first approximation one can safely assume that their k⊥ 2 stay finite

at large energies (in fact it slowly grows with energy as its logarithm). From condition ki2 = 0
we find 2ki+ ki− = −ki⊥ 2 , which means that if k
i+ → ∞ then ki− → 0 and vice versa if
ki− → ∞ then ki+ → 0. So one of the three final momenta, let it be k3 , has the maximal
’+’ component, which tends to ∞ and the minimal ’-’-component, which tends to zero, and
another of these momenta let it be k5 , has the maximal ’-’ component, which tends to ∞ and
the minimal ’+’-component, which tends to zero:

k5+ , k4+ < k3+ → ∞, k5− , k4− > k3− → 0, k3− , k4− < k5− → ∞, k3+ , k4+ > k5+ → 0.
(240)
2
Let us study the relative energetic variable s34 = (k3 + k4 ) . In terms of light-cone
variables  
s34 = 2k3 k4 = 2 k3+ k4− + k3− k4+ + (k3 k4 )⊥ . (241)

Using the mass-shell condition we rewrite it as follows

2 k3+ 2 k4+
s34 = −k4⊥ − k3⊥ + 2(k3 k4 )⊥ . (242)
k4+ k3+

We may distinguish between two cases:

1) If the ratio k3+ /k4+ is finite, which implies k3+ ∼ k4+ , then s34 is finite.
2) If k3+ /k4+ >> 1, which means that k4+ << k3+ , then

2 k3+
s34 ≃ −k4⊥ = 2k3+ k4− → ∞. (243)
k4+

Similarly we consider the energetic variable s45 which we present as

2 k5− 2 k4−
s45 = −k4⊥ − k5⊥ + 2(k3 k4 )⊥ . (244)
k4− k5−

Again two cases are possible

3) If the ratio k4− /k5− is finite, which implies k4− ∼ k5− , then s45 is finite.
4) If k5− /k4− >> 1, which means that k4− << k5− , then

2 k5−
s45 ≃ k4⊥ = 2k5− k4+ → ∞. (245)
k4−

Combining these possibilities we find that cases 1) and 3) are incompatible. So we have
only three possibilities: 1), 3) and 2)+4) together. In other words:
1) s34 is finite, s45 , s35 are of the same order as s12 This region is called fragmentation
region of the projectile. In fact the initial large momentum k1+ is split in two parts of the
same order k3+ +k4+ as if the initial projectile particle splits into two particles with momenta
k3 and k4
 59

Figure 33: Particles 3 an4 combine into a quasi-particle with mass (k3 + k4 )2

2) s45 is finite, s34 , s35 are of the same order as s12 This region is called fragmentation
region of the target. Now the initial large momentum k2− is split in two parts of the same
order k4− + k5− as if the initial target particle splits into two particles with momenta k4 and
k5
3) s34 , s45 → ∞. Then from (243) and (245) we find
2
s34 s35 = −k4⊥ s12 . (246)

The product of the two relative energy squares is of the same order as the initial one. This
region is called central.
Let us study the asymptotic of the amplitude in these three regions. For simplicity we
shall assume that all singularities of the partial waves amplitudes to be introduced are Regge
poles.
We begin with the projectile fragmentation region. Adequate variables in this case are
√
s12 , s34 , t25 , t13 and t14 . Since the total mass of the final particles 3 and 4 s34 is finite, we
can consider the whole diagram as a binary process in which the initial particles 1 and 2 go
√
into the final particle 5 and a compound particle formed by particles 3 and 4 with mass s34 :

1 + 2 → (34) + 5 (247)

(see Fig. 33). In the t-channel in its center of mass the initial and final momenta now are
different, kt and kt′ due to difference in masses. From the lower end of the diagram we have
the standard relation
√
t25 = (k2 − k5 )2 = 4kt2 , so that kt = t25 /2

From the upper end we find

q
t25 = (k1 − k3 − k4 ) = 2
(kt′ + s34 + kt′ 2 )2

or q √
s34 + kt′ 2 = t25 − kt′
or
√
s34 = t25 − 2kt′ t25
, so that finally
√ 1 √ s34 
kt = t25 /2, kt′ = t25 − √ . (248)
2 t25
 60

Figure 34: Single and double Regge limits

(with s34 = 0 this passes into kt ). Now we consider s12 as a function of the scattering angle
and momenta kt and kt′
2
s12 = (k1 + k2 )2 = (kt + kt′ )2 − (kt2 + kt′ + 2kt kt′ zt ), = −2kt kt′ (1 − zt ) (249)

where zt is the cosine of the scattering angle in the t-channel. As before we expand the
amplitude in the Legendre polynomials of zt at fixed s34 , t25 , t13 and t14 :
X
A(s12 (zt ), t25 , s34 , t13 , t14 ) = (2l + 1)Pl (zt )al (s34 , t25 , t13 , t14 ) (250)
l=0

Next we pass to complex angular momenta and transform this expansion into the Gribov-
Froisart representation. At s12 → ∞, as before, −zt → ∞ and the asymptotic will be
determined by the rightmost singularities in the complex l-plane. If we assume that this
rightmost singularity is a Regge pole at l = α(t25 ) then we find an asymptotic similar to
(223)

AR (s12 , t25 s34 , t13 , t14 ) = −π(4α(t25 ) + 1)γR25 (t25 )γR134 (t25 , s34 , t13 , t14 )Pα(t25 ) (−zt )ζ(α(t25 ))
(251)
It is important that the residue at the pole factorizes into a coupling constant λR25 of the
Reggeon with particles 2 and 5, on the one side, and a coupling constant λR134 with particles
1,3 and 4 on the other side, in accordance with the unitarity relation in the t channel. Each
coupling naturally depends only on variables related to the coupled particles. The asymptotic
(251), called a single-Regge limit, is illustrated graphically in Fig. 34, a. Note that the theory
cannot tell much about the structure of the coupling vertexes: it only predicts the asymptotic
dependence on the large s12 .
A similar asymptotic can be derived in the second fragmentation region, that of the
projectile. One then obtains an asymptotic of the same form as (251) with substitutions

t25 s34 , t13 , t14 → t13 , s45 , t25 , t24 . (252)

It is illustrated in Fig. 34 b.
2 .
We finally study the central region. Adequate variables are now s34 , s45 , t13 , t25 and k4⊥
In this case one considers two different t channels, 13̄ with its energetic variable t13 = (k1 −k3 )2
and 25̄ channel with its energetic variable t25 = (k2 − k5 )2 . In the 13̄ channel the initial
 61

Figure 35: Formation of the double Regge limit

particles are 1 and 3 with zero mass and final are 4 and (25) with masses zero and t25
respectively, Fig. 35,a. So in this channel the c.m. momenta are
√ ′
√ √
k13 = t13 /2, and k13 = (1/2)( t13 − t25 / t13 ).

Variable s34 will be expressed by these momenta and the scattering angle as before

′
s34 = −2k13 k13 (1 − z13 ). (253)

In the channel 25̄ we shall find similar formulas with the interchange t13 and t25 and s34 → s54 ,
Fig. 35,b. In particular
′
s54 = −2k25 k25 (1 − z25 ) (254)

with
√ ′
√ √
k25 = t25 /2 and k25 = (1/2)( t25 − t13 / t25 )
2 and
. Now we consider the amplitude as a function of z13 and z25 at fixed t13 , t25 and k4⊥
expand it in the double series of the Legendre polynomials of z13 and z25 :

2 2
X X
A(s34 (z13 ), s54 (z25 )t13 , t25 , k4⊥ )= (2l1 + 1)(2l2 + 1)Pl1 (z13 )Pl2 (z25 )al1 ,l2 (t13 , t25 , k4⊥ ).
l1 =0 l2 =0
(255)
Next we pass to complex angular momenta l1 and l2 and as before transform this expansion
into double Gribov-Froissart representation. The asymptotic at large values s34 and s45
will be determined by the rightmost singularities in the complex l1 - and l2 planes. If these
singularities are Regge poles in the two t channels at l1 = α1 (t13 ) and l2 = α2 (t25 ) we get an
asymptotic (double Regge) limit

2
AR1 R2 (s34 , s45 , t13 , k4⊥ ) = π 2 (4α1 (t13 ) + 1)(4α2 (t25 ) + 1)

2
γR1,13 (t13 )γR2,25 (t25 )γRR4 (t13 , t25 , k4⊥ )Pα1 (t13 ) (−z13 )Pα2 (t25 ) (−z25 )ζ(α1 (t13 )ζ(α2 (t25 )).
(256)
It is illustrated in Fig. 34,c.
 62

Figure 36: Amplitude for the inclusive cross-section

10.2 Inclusive cross-sections. Generalities

In this subsection we shall study the single inclusive cross section for particle production at
high energies. It corresponds to the reaction

a(k1 ) + b(k2 ) → c(k) + X, (257)

where particle c with momentum k is registered in the experiment, all other produced particle
being unregistered and denoted as X. Graphically this process is shown in Fig. 36, where
unregistered particles are combined into the wide arrow. Analytically the inclusive cross-
section is given by the formula (14), which we rewrite in the notation corresponding to (257)
as
(2π)3 k0 dσ(k, X) 1 X 2
Z 
′
I(k) ≡ = dτ (k + k − k) A(k, X|k1 2  .
, k ) (258)
 
n 1 2
d3 k J12 n


We first comment on the kinematics of the inclusive reactions. Already from Fig. 36 one
observes that the reaction is described by the same structure as for any binary reaction with
the only difference that the total mass corresponding to the unobserved produced particles,
the ’missing mass’ M is not fixed and is an extra variable

M 2 = (k1 + k2 − k)2 . (259)

So the process is characterized by three independent variables. For these one may choose
the old Mandelstam variables s = (k1 + k2 )2 , t1 = (k1 − k)2 and M 2 . However this choice
does not take into account the already mentioned fact that produced particles carry their
transverse momenta much smaller than the longitudinal ones. For this reason usually the
inclusive cross-section is characterized by different sets of variables.
For simplicity we shall again take masses of the three observed particles a,b and c all
equal to zero: k12 = k22 = k2 = 0. Then in the c.m. system, using the light-cone variables, we
find as before
2
k1+ = k2− , k1⊥ = k2⊥ = 0, s = 2k1+ . (260)

The observed final particle may be characterized by its transverse momentum k⊥ and
Feynman scaling variables
k+ k−
x+ = , x− = . (261)
k1+ k2−
 63

Both x± are positive. At high energies their product is small:

k+ k− k2
x+ x− = = − ⊥ → 0. (262)
k1+ k2− s

The missing mass is related to x± by the relation

M 2 = 2(k1 + k2 − k)+ (k1 + k2 − k)− + (k1 + k2 − k)2⊥

2
= 2(k1 − k)+ (k2 − k)− + k⊥
2
= s(1 − x+)(1 − x− ) + k⊥ . (263)

From this we conclude that both x± ≤ 1. Taking into account (262) we also see that if one of
the two scaling variables, say x+ is fixed at some positive value different from zero, then at
√ √
large s the other x− → 0 and M 2 ≃ s(1 − x+ ). Since kz = (k+ − k− )/ 2 = k1+ (x+ − x− )/ 2
in this case the produced particle is moving fast along the direction of the incident particle
a with momentum k1 , that is in the forward direction. Momentum k+ is of the order of the
initial particle momentum k1+ , so that the observed particle is born in the fragmentation
region of the particle a.
If the fixed variable is x− then x+ → 0 and the same formula shows that kz < 0 and the
observed particle is moving in the backward direction. The observed particle is born in the
fragmentation region of the target particle b.
However there exists a kinematical situation when both x+ and x− are small. Then on
the one hand k+ , << k1+ and on the other k− << k2− . The particle is born in the central
region. As we shall see, at high energies the majority of particles are born precisely in this
region. Scaling variables are badly suited to describe this region, since they reduce it to a
point x± ∼ 0. A better description is given by the rapidity variable, defined as

1 k+
y= ln . (264)
2 k−

One can transform this expression as

2 √
1 2k+ 1 x2 s x+ s
y= ln = ln + 2 = ln (265)
2 2k+ k− 2 −k⊥ |k⊥ |
or as
1 2k+ k− 1 −k2 |k⊥ |
y= ln 2 = ln 2 ⊥ = ln √ . (266)
2 2k− k 2 x− s x− s
From these formulas it follows that the rapidity varies in the interval
√ √
s s
− ln ≤ y ≤ ln . (267)
|k⊥ | |k⊥ |
√
In the fragmentation regions it is large, of the order ln k⊥s| . It is finite when both x± are
√
small, of the order 1/ s, which corresponds to the central region.
Note that rapidity is a natural generalization of the velocity in the non-relativistic me-
chanics. Under the Lorentz transformation in the direction of the z axis all rapidities change
 64

by the same constant, just as velocities under the Galilean transformation. Indeed under this
transformation with velocity v:
1
k0 = κ(k0′ − vkz′ ), kz = κ(kz′ − vk0′ ), κ = √ , (268)
1 − v2
so that
′ ′
k+ = k+ (1 − v), k− = k− (1 + v) (269)

and consequently
1 1−v
y = y′ +
ln . (270)
2 1+v
Thus in the central region suitable variables for the inclusive cross-sections are s, y and
k⊥ . The phase volume takes an especially simple form in these variables:
d3 k dyd2 k⊥
= . (271)
(2π)3 2k0 8π 3
Let us turn now to the expression (258) for the inclusive cross-section. The right-hand
side of this expression looks very similar to the right-hand side of the unitarity relation for
the crossed process in the forward direction

a(k1 ) + b(k2 ) + c̄(−k) → a(k1 ) + b(k2 ) + c̄(−k). (272)

Indeed repeating the derivation of unitarity for the initial and final states containing the
three particles we get, similarly to (29)

iA∗ (k1 , k2 , −k|k1 , k2 , −k) − iA(k1 , k2 − k|k1 , k2 , −k) = 2Im A(k1 , k2 , −k|k1 , k2 , −k)
XZ  2
= dτn (k1 + k2 − k)A(n|k1 , k2 , −k) . (273)
 
n
Of course in this relation it is assumed that k0 < 0, which corresponds to the initial particle
c̄(−k). Note that from the point of view of the analytic properties the twice imaginary part
(273) gives just the discontinuity divided by i of the amplitude considered as a function of
M 2 = (k1 + k2 − k)2 on its right-hand unitarity cut. So we can also rewrite (273) as
XZ  2
DiscM 2 >0 A(k1 , k2 , −k|k1 , k2 , −k) = i dτn (k1 + k2 − k)A(n|k1 , k2 , −k) . (274)
 
n

If we analytically continue this relation in k0 from negative values to positive ones, then in
the right-hand side the amplitudes A(n|k1 , k2 , −k) describing transition of the initial three-
particle state into intermediate states |n > go over into crossed amplitudes A(k, n|k1 , k2 )
which describe transitions of the initial two-particle state into intermediate states |k, n >
containing particle c(k) and other particles labeled as n. So the right-hand side of (274) goes
over into the right-hand side of Eq. (258) multiplied by iJ12 . Thus we get
1 2
I(k) = DiscM 2 >0 A(k1 , k2 , −k|k1 , k2 , −k) = Im A(k1 , k2 , −k|k1 , k2 , −k), M 2 , k0 > 0.
iJ12 J12
(275)
This relation is known as a generalized optical theorem by A.Mueller. It allows to find the
single inclusive cross-section from the imaginary part of the three-to-three forward amplitude,
 65

Figure 37: Optical theorem for inclusive cross-sections

Figure 38: The region of fragmentation of particle a

in full similarity with the standard optical theorem. Note that (275) implies a possibility of
analytic continuation of the standard unitarity relation in k0 and thus in variables M 2 ,
t1 = (k1 − k)2 and t2 = (k2 − k)2 related to k0 . The question about the branches in all these
variables is not immediately clear, but is actually solved by the condition that the right-hand
side should be pure imaginary with a positive imaginary part. Graphically the generalized
optical theorem can be illustrated as shown in Fig. 37, where it is shown how the square
modulus of the production amplitudes A(k, n|k1 , k2 ) are summed into the discontinuity of
the three-to-three by Cutkosky rules.

10.3 Inclusive cross-sections. High-energy asymptotic

According to the generalized optical theorem to find the asymptotic of the inclusive cross-
section we have to find the asymptotic of the imaginary part of the amplitude
A(k1 , k2 , −k|k1 , k2 , −k) at M 2 > 0 above the cut and divide it J12 /2 ≃ s. Applying our previ-
ous results for the production amplitude, trivially generalized to our three-to-three amplitude
we distinguish between three kinematical regions.
If x+ is finite (smaller than unity) and correspondingly x− → 0 then k+ is of the same
order as k1+ and we are in the fragmentation region of the projectile. In the bb̄ channel
we can consider the process as the transition of the initial bb̄ pair into the quasi-particle ac̄
with the total momentum k1 − k and its antiparticle with a finite mass t1 = (k1 − k)2 . The
 66

Figure 39: Regge description in the fragmentation and central regions

natural variables in this kinematics are the ”energy” squared of these colliding particles in
the s channel M 2 and characteristics of the observed particle c x+ and k⊥ 2 . We recall that

the collision energy s is related to M 2 by M 2 = s(1 − x+ ) in this kinematics. At high M 2

the three-to-three amplitude will be described by the single-Regge limit (Fig. 39,a). Indeed
repeating the same derivation as for the production amplitude we expand the amplitude in
the bb̄ channel in partial waves. considering the particles (ac̄) as a single particle with a finite
mass squared t1 = (k1 − k)2 = k⊥ 2 /x . To avoid a singularity in the bb̄ channel at zero energy
+
t we preserve a non-zero mass
q mb for particle b. Then the two different c.m. momenta in the
2 ′ √
bb̄ -channel are kt = (1/2) t − mb and kt = (1/2) t − t1 . We find

M 2 = −2kt k′ t(1 − zt ), (276)

√
which at t = 0 gives M 2 = −(1/2)mb t1 (1 − zt ), or
√
zt = 1 + 2M 2 /mb t1 . (277)

If the rightmost singularity in the relevant l-plane is the vacuum Regge pole (pomeron) then
similar to (251) we get the asymptotic
 2M 2 αP (0)
2
Im AP = 3πγP bb (0)γP aac (x+ , k⊥ ) √ , (278)
mb t1
where we put αP (0) = 1 in all places except when it appears in the exponent. The inclusive
cross-section in this region results as
2
I(k) = fac (x+ , k⊥ )sαP (0)−1 , (279)

where fac is a function depending on x+ and k⊥ 2 and on the type of particles a and c. If

αP (0) is exactly unity then we get essentially that the imaginary part grows linearly with s
and the inclusive cross-section is a constant. As one may expect, the Regge approach only
allows to find the asymptotic dependence on the large variable, other dependencies remaining
unspecified.
In absolutely a similar manner we may study the case when x− is fixed and smaller than
unity and s → ∞. Then k− is of the same order as k2− and we are in the fragmentation region
of the target. The same derivation as above shows that if at large s and M 2 = s(1 − x− )
2
I(k) = fbc (x− , k⊥ )sαP (0)−1 (280)
 67

Figure 40: Fragmentation and central regions

and does not depend on energy if αP (0) = 1. This situation is illustrated in Fig. 29,b
Now we pass to the case when both x+ and x− are small (central region). In this case it
is convenient to use variables t1 = (k1 − k)2 , t2 = (k2 − k)2 and k⊥
2 . Similarly to the case of

production amplitude we find in this region

2k
k⊥ k2 k2
1+
t1 = −2kk1 = −2k− k1+ = = ⊥ , t2 = ⊥ , (281)
k+ x+ x−

so that both t1 and t2 are large and, using (262)

2
t1 t2 = −k⊥ s (282)

Next we expand the amplitude in partial waves in the two crossed channels aā and bb̄ as was
done for the production amplitude, Fig. 40,b and c.
At zero energy in these channels we get for the cosine of scattering angles
t1 t2
zaā = 1 + 2 , zbb̄ = 1 + 2 . (283)
ma mc mb mc
At large t1 and t2 the asymptotic will be determined by the rightmost singularities in the two
relevant complex angular momentum plane. It both these singularities correspond to Regge
poles (pomerons) then the asymptotic will have the form of the double Regge limit (cf (256)
(see Fig. 29,c)

AP P = 9π 2 γP aa (0)γP bb (0)γP P c (k⊥

2
)PαP (0) (−zaā )PαP (0) (−zbb̄ ), (284)

where we again have put α(0) = 1 everywhere except when it is enhanced by large factors.
Using () we find at large t1 and t2

4t1 t2 2
k⊥
(−zaā )(−zbb̄ ) ∼ = −s (285)
ma mb m2c ma mb m2c

So the asymptotic of (284) is

2
AP P = fcentral (k⊥ )sαP (0) (286)
 68

1.5

dσ/dy d k⊥
2
1

0.5

0
-8 -6 -4 -2 0 2 4 6 8
y

Figure 41: Inclusive cross-sections as a function of rapidity

and the inclusive cross-section in the central region is

2 αP (0)−1
I(k) = fcentral (k⊥ )s . (287)

This distribution does not depend on rapidity of the produced particles. So at fixed k⊥ 2 it is

flat while the rapidity is finite and so has a form shown in Fig. 41 with a central plateau and
decrease in the two fragmentation regions. The plateau height grows with energy as sαP (0)−1
and if αP (0) = 1 stays fixed.
Integrated over k⊥ and rapidity I(k) gives the average multiplicity µ times the total
cross-section. Since at large s the fragmentation regions are relatively unimportant, we find
y2 d2 k⊥ s
Z
µ(s)σ tot (s) = dy 3
I(k) = const sαP (0)−1 ln , (288)
y1 8π < |k⊥ |2 >

where we used (267) for the limits of integration over rapidity y1 and y2 and substituted |k⊥ |2
by its average value < |k⊥ |2 >. Under the same approximation obviously

σ tot (s) = const sαP (0)−1 , (289)

so that we get
s
µ(s) = const ln , (290)
s0
where s0 is some fixed quantity with the dimension of energy squared.

10.4 Triple Regge limit

We finally study a very particular kinematic region for the inclusive cross-section when x+ is
close to unity, so that at s → ∞ M 2 is much smaller than s but still very large:

M2
s → ∞, M 2 → ∞, << 1. (291)
s
For reasons which we shall see presently this region is called triple-Regge region, or high-mass
diffractive region. The amplitude for this process can then be considered in the ac̄, or t1 ,
channel as a transition from the initial colliding particles a and c̄ to final particle b and a
 69

Figure 42: Asymptotic of the inclusive amplitude at M 2 << S

complex of unobserved particles with mass M , which can be considered as a single particle,
since M 2 2 << S ( Fig. 42). At large s we can study its asymptotic as before expanding the
amplitude in the t1 channel in partial waves. The two c.m. momenta in this channel will be

1√ 1 √ M2 
kt1 = t1 , kt′ 1 = t1 − √ , (292)
2 2 t1

where we used Eq. (248) for the c.m. momentum of a pair consisting of a zero mass particle
and another with mass M . So at finite t1 and M 2 >> t1 we find for this amplitude
1 2s
s = −2kt1 kt′ 1 (1 − zt1 ) ≃ M 2 (1 − zt1 ), or zt1 = 1 + 2 (293)
2 M
As a result applying the single Regge-limit we find for the asymptotic behaviour of this
amplitude (see Fig. 42)

AR = −π(3αR (t1 + 1)γRac (t1 )γRbX (t1 , k2 , ki′ )PαR (t1 ) (−zt1 )ζ(αR (t1 )). (294)

Here αR (t1 ) is the Regge trajectory. Note that particles a and c may be different, so that this
Reggeon is not necessarily the vacuum one (pomeron). Function γRbX (t1 , k2 , ki′ ) describes the
lower blob in Fig. 42. It depends on the momenta ki′ of unobserved particles for a concrete
production amplitude.
The inclusive cross-section is obtained by taking the modulus square of this amplitude,
integrating over momenta of unobserved particles and summing over all number of these
unobserved particles. The square modulus of γRbX (t1 , k2 , ki′ ) then passes over in what may
be considered as the forward amplitude of the scattering of the Reggeon on particle b:
XZ  2  
dτn′ γRbX (t1 , k2 , ki′ ) = A R(k1 − k) + b(k2 ) → R(k1 − k) + b(k2 ) (295)
 
n

as illustrated in Fig. 43. The energy for this scattering process is (k1 − k + k2 )2 = M 2 and
is large. So we again can apply the single Regge asimptotic at large M 2 , taking into account
that this time the t channel is the vacuum one and the relevant Reggeon is the Pomeron.
This gives
   M 2 αP (0)
AP R(k1 − k) + b(k2 ) → R(k1 − k) + b(k2 ) = iγRRP (t1 ) , (296)
s0
 70

Figure 43: The inclusive cross-section at M 2 << s

Figure 44: Triple Regge interaction

where s0 is some fixed energy squared and function γRRP (t1 ) has a meaning of the coupling
constant between the two scattering Reggeons and the Pomeron. Combining these formulas
and multiplying the imaginary part of the amplitude by 2/J12 = 1/s we find for the inclusive
cross-section in the triple-Regge region

1  s 2αR (t1 )  M 2 αP (0)

I(k) = c[(2αR (t1 ) + 1)γRac (t1 )]2 γRRP (t1 ) . (297)
s M2 s0
where c is some numerical constant. It can be illustrated in Fig. 44 which shows that
reggeons considered as virtual quasiparticles may interact with each other and not only with
the participant hadrons.

11 Regge cuts. Froissaron

From the elastic unitarity we concluded that the only singularity in the l complex plane
of the partial amplitude al (t) could be only Regge poles l = αt , which at α(t) = 0, 1, 2...
correspond to observed particles or resonances with even or odd orbital momenta depending
on the signature ξ = ± of α(t). The contribution of the Regge pole to the amplitude can be
written as
iA(s, t) = iR(t)Γ21 (t) = iR0 (t)sα(t) Γ21 (t). (298)

Here Γ1 (t) is the coupling of the incident particle to the Reggeon, R0 (t) includes all t-
depending factors in R(t) except the exponential

R0 (t) = −π(2α(t) + 1)(2µ2 − t/2)−α(t) ζ(t),

 71

Figure 45: Single reggeon exchange

Figure 46: Double reggeon exchange

where
1 + ξe−iπα(t)
ζ(t) =
sin πα(t)
and the Regge trajectory is taken as

α(t) = 1 + ∆ + α′ t.

The Regge pole can be interpreted as an exchange of a virtual quasi-particle in the t

channel, as shown in Fig. 45.
It has two complementary manifestations. On the one hand, it is a pole in the partial am-
plitude al (t) and can be found as such. On the other hand, it shows itself in the asymptotical
behavior of the amplitude as a function of s at large s. where it produces a characteristic
dependence sα(t) . So it can be found also in the study of this asymptotical behavior of A(s, t).
This opens a way to study the analytical properties of the partial amplitudes in the complex
l plane from their asymptotical behavior at large s and fixed t.
If one goes beyond the approximation of the elastic unitarity then it turns out that apart
from Regge poles there appear cuts due to multiple Regge pole exchanges.
Consider the contribution from the double Regge pole exchange schematically shown in
Fig 46.
To see it contents in terms of singularities in the complex l plane let us study its asymp-
 72

totical behavior at large s. Explicitly this contribution can be written as

1 d4 q1
Z
(2) 2
iA =i N2 (k1 , q1 , q2 )N2 (k2 , −q1 , −q2 )R(t1 )R(t2 ), (299)
2! (2π)4

where N2 describe the coupling of the two Reggeons to partisipant scattering particles q1 and
q2 are the 4-momenta transferred by the reggeons with q1 + q2 = k1 − k1′ , so that

t1 = q12 , t2 = q22

and it has been taken into account that the contribution to the S matrix is iR and the
symmetry in the iR1 and iR2 introduces factor 1/2!.
To calculate this expression we make an approximation taking all masses equal to zero.
In fact the asymptotical behavior at large s does not depend on µ2 as soon as s >> µ2
Accordingly we use the light-cone variables assuming that the incident particles move along
the z axis. They have
√ √
k1+ = 2k10 , k1− = 0, k1⊥ = 0, k2+ = 0, k2− = 2k20 , k1⊥ = 0,

s = 2k1+ k2− , k1+ , k2− → ∞.

The final particles have finite k⊥ . However from the mass conditions
′ 2
k1⊥
2 2
k1′ = 0 = 2k1+
′ ′
k1− ′
− k1⊥ ′
, so that k1− = ′ →0
2k1+
′ → 0. The energies of particles are conserved in the c.m. system, so that
and similarly k2+

′
k10 = k10

Then
′ ′ ′
k1z ≃ k10 , k1z ≃ k10 and k1+ − k1+ →0

It follows that

t = k2 = (k1 − k1′ )2 = 2(k1+ − k1+

′ ′
)(k1− − k1− ′
) − (k1⊥ − k1⊥ )2 = −(k1⊥ − k1⊥
′
)2

So the total transferred momentum k is purely transversal and t = −k⊥ 2 ≡ −k2 .

We perform integrations over q1+ and q1− passing to variables s1 = (k1 + q1 )2 and s̄1 =
(k2 −q1 )2 . Integrations over s1 and s̄1 are to be performed along the Feynman contours above
the righthand cuts and below lefthand cuts in functions N , as shown in Fig. 47,a. One can
deform the contours to close around the righthand cut (see Fig.47,b) to obtain

1 1 d2 q1
Z
iA(2) = i2 ∆s N (k1 , q1 , q2 )∆s̄1 N (k2 , −q1 , −q2 )R(t1 )R(t2 ), (300)
2 2s (2π)4 1

where ∆s1 N (k1 , q1 |k1′ , −q2 ) is the discontinuity on the righthand cut in variable s1 and sim-
ilarly ∆s̄1 N (k2 , −q1 |k2′ , q2 ). Factor 1/2s = 1/4k1+ k2− appears due to transition to variables
s1 and s̄1 . Note that in this integration both s1 and s̄1 take finite values, so that q1± turns
 73

Figure 47: Integration contours in the s1 complex plane

out to be small. As a result all quantities become independent of longitudinal momenta and
become functions on only transverse 2-dimensional momenta q1⊥ and k⊥ .
We define
1
∆s N (k1 , q1 , q2 ) = Γ2 (k1 , q1 , q2 ),
2π 1
1
∆s N (k2 , −q1 −, q2 ) = Γ2 (k2 , −q1 , −q2 )
2π 2
to obtain the amplitude in the form of a two-dimensional integral over the Euclidean mo-
mentum q1

1 1 d2 q1
Z
(2) 2
iA =i Γ2 (k1 , q1 , q2 )Γ2 (k2 , −q1 , −q2 )R(t1 )R(t2 ). (301)
2! 2s (2π)2

Here
q1 + q2 = k′1 − k1 ≡ k

and
t = −k2 , t1 = −q21 , t2 = −q22 .

Our next simplification is based on the dependence of the exponential factor on the
transferred momentum where its square is multiplied by a large factor ln s:
′ ln s q21
sα(t1 ) = s1+∆ e−α .

So the characteristic values of q21 are small of the order 1/ ln s. In the limit s → ∞ it allows
to put q1 = q2 = 0 everywhere except in this exponential. The amplitude then simplifies to

1  s1+∆ iR 2
0
iA(2) = 2sΓ22 I2 (k), (302)
2! 2s
where the transverse integral can be rewritten as

d2 q1 d2 q2
Z
2 2
I2 (k) = 2 2
(2π)2 δ2 (k − q1 − q2 )e−γ(s)(q1 +q2 ) . (303)
(2π) (2π)

Here we denoted
γ(s) = α′ ln s >> 1.

To factorize the integral I2 we introduce integration over the impact parameter b putting
Z
2 2
(2π) δ (k − q1 − q2 ) = d2 beib(k−q1 −q2 ) .
 74

Figure 48: Exchange of n reggeons

Then we get Z
I2 (k) = d2 beibk I2 (b), I2 (b) = J(b)2 , (304)

where
d2 q −ibq−γq2 1 − b4γ2
Z
J(b) = e = e (305)
(2π)2 2π 2 γ
and we used the standard integral

π b2
Z r
−ax2 +bx
dxe = e 4a . (306)
a
Using (304)and (305) we get

1 2 b2 1 2 2πγ −k2 γ/2

 Z 
I2 (k) = d2 beibk− 2γ = e .
2π 2 γ 2π 2 γ 2
So finally
1  s∆ iR 2
0 2
iA(2) (s, t) = 4πγsΓ22 i e−k γ/2 , t = −k2 . (307)
22! 4π 2 γ
One immediately sees the characteristic dependence on s and t. The dependence on s is
contained in the factor
s1+2∆ s1+2∆
= ′ .
γ(s) α ln s
So as compared to the single reggeon the singularity point shifts by ∆

1 + ∆ → 1 + 2∆.

But there also appears ln s in the denominator. So the singularity is not a pole but a branch
point giving rise to a cut in the complex l plane. This is the simplest Regge cut. As to the t
dependence, in the exponent there appears γt/2. This means that again as compared to the
single reggeon the slope α′ is twice smaller:
1 ′
α′ → α.
2
These conclusions are easily generalized to the exchange of n reggeons (Fig. ??).
 75

The amplitude with the exchange of n Reggeons will be given by the direct generalization
of (299):
Z n−1
Y d4 qi
(n) n1
iA =i Nn (k1 , q1 , q2 ...qn )Nn (k2 , −q1 , −q2 , ...−qn )R(t1 )R(t2 )...R(tn ), (308)
n! i=1
(2π)4

where qi are 4-momenta transferred along the n Reggeons; q1 + q2 + ....qn = k = k1′ − k1 is

the total transferred momentum. Nn represent the coupling to n reggeons of the projectile
and target.
As before integrations over qi + and qi −, i = 1, ...n − 1 are performed by passing to
energetic variables si = (k1 + qi )2 and s̄i = (k2 − qi )2 . The Feynman contours of integration
are deformed to close around the righthand cuts of the amplitudes N in the appropriate
variables giving their discontinuities

∆s1 ∆s2 ....∆sn−1 N (k1 , q1 , q2 , ...qn )∆s̄1 ∆s̄2 ....∆s̄n−1 N (k2 , −q1 , −q2 , −qn ).

All longitudinal momenta become small and as before we are left with the integral over the
transverse momenta. Defining
1 n−1
Γn (k1 , q1 , ...qn ) = ( ) ∆s1 ∆s2 ....∆sn−1 N (k1 , q1 , q2 , ...qn )
2π
and similarly for the projectile we find
Z Y 2
1  1 n−1 d qi 
iA(n) = iR(ti ) Γn (k1 , q1 , q2 , ...qn )Γn (k2 , −q1 , −q2 , −qn ). (309)
n! 2s i=1
(2π)2

As before we make an approximation neglecting small transverse momenta everywhere

except when they are multiplied by the large ln s in the Reggeons. Then we get a simple
expression
1  s1+∆ iR n
0
iA(n) = 2sΓ2n In (k), (310)
n! 2s
where n
d2 qi
Z Y
2 2 2
In (k) = 2
(2π)2 δ2 (k − q1 − q2 − ... − qn ))e−γ(s)(q1 +q2 +...qn) . (311)
i=1
(2π)
This integral is factorized by the transition to the impact parameter space
Z
In (k) = d2 beibk In (b), In (b) = J n (b), (312)

where J(b) is the same integral (305). Doing the final integration over b we get
 1 n Z 2
1 n 2πγ −k2 γ/n
ibk−2 b

In (k) = d2 be 4γ = e ,
2π 2 γ 2π 2 γ n
This brings us to the final expression for the n-Reggeon exchange
1  s∆ iR n
0 2
iA(n) (s, t) = 4πγsΓ2n i e−k γ/n , t = −k2 . (313)
nn! 4π 2 γ
So the n-Reggeon exchange at large s behaves as
s1+n∆
iA(n) ∼ . (314)
lnn−1 s
 76

The intercept is changed by ∆ for each additional reggeon. The slope becomes n times
smaller.
This result can be trivially generalized when the Reggeons are different with intercepts
and slopes
1 + ∆j , α′j , j = 1, 2, ...s ≤ n.

Obviously factor 1/n! will be changed to take into account only the identical Reggeons. Also
the effective slope will depend in a more complicated way on the individual slopes of the
exchanged reggeons. However the s dependence will be simple: at t = 0 each reggeon will
contribute factor
s∆
j
(315)
ln s
and the total behavior of the forward amplitude in s will be just the product of all such
factors.
One immediately sees that the influence of Regge cuts crucially depends on the value of
∆ of the exchanged reggeons. If ∆ < 0 the corresponding branch point will be shifted to the
left and according to (315) its contribution will be damped essentially by power s−|∆| . This
means that all cuts generated by the exchange of non-vacuum Reggeons are not important
at high energies¡ since they diminish faster than without them.
On the other hand if ∆ > 0 the cut shifts the branch point to the right and its contribution
dominates at high energies by factor (315).
In the border case ∆ = 0 the cut starts at the same branch point. Its contribution falls
but slowly, as 1/ ln s.
As a result the only cuts which are to be taken into account are those which come from the
pomeron exchange. The problem becomes especially acute if one assumes that the pomeron
is a pole wit ∆P > 0, which the present experimental evidence suggests. In this case the
n-pomeron exchange will start at 1 + n∆ and its contribution will be proportional to n-th
power of (315). With n → ∞ at t = 0 the overall cut will cover all the positive axis l ≥ 1
and the amplitude will be an infinite sum of terms which grow infinitely. Of course this does
not mean that the total sum will be infinite, since the terms may have different signs an
compensate one another.
To see that this may be so and understand the behavior of the amplitude with any number
of pomeron exchanges one has to sum contributions (310).
X 1  s∆ iR n
0
iA = 2s Γ2n In (k). (316)
n=1
n! 2

Assuming ∆P positive but small we find

3
ζ = −i, iR0 (0) = − ≡ −a, a > 0.
2µ2
So
X 1  s∆ n
iA = 2s Γ2n − a In (k). (317)
n=1
n! 2
 77

Using (312) we present this as an integral over the impact parameter At t = 0

X 1 s∆ aJ(b) n
Z 
iA = 2s d2 be Γ2n − . (318)
n=1
n! 2

Denote
s∆ aJ(b)
ρ≡ , (319)
2
so that Z X 1
iA = 2s d2 b Γ2n (−ρ)n . (320)
n=1
n!
This is an expression which is valid in the general case. It cannot be calculated as such due
to the n-dependence of the couplings Γn
Now we make an assumption

Γn = 1, for all n

(”the eikonal approximation”). Then we can immediately sum the series (321)
Z  
iA = 2πs db2 1 − e−ρ . (321)

Explicitly
as∆ −b2 /4γ
ρ= e , (322)
4π 2 γ
so that
1 dρ
dρ = − ρ db2 , or db2 = −4γ .
4γ ρ
The limits of integration in ρ are

as∆
0<ρ< ≡ r >> 1.
4π 2 γ
So we get
r 1 − e−ρ
Z  
iA = 8πsγ dρ = 8πsγ ln r + C + Ei (−r) . (323)
0 ρ
where Ei is the well-known integral exponential function. Since r >> 1 function Ei (−r) is
exponentially small. We have ln r ≃ ∆P ln s So in the end

iA = 8π∆p αP s ln2 s. (324)

We find a behavior dictated by the Froissart limit. The corresponding cross-section rises as
ln2 s. This amplitude is called ”Froissaron”.