
What kind of singularities are known to appear in scattering amplitudes? At tree-
level in perturbation theory, we only encounter poles of the type (
s−m2
)
−1
, together
with their multi-particle generalizations. At low loop level, direct computations give
square-root and logarithmic branch points of the kind
fa/2logbf
for some integers
a, b
and a function
f
of the kinematics [
7
–
11
]. It is also known that scattering amplitudes
can have branching of arbitrary order, for instance,
sα(t)
in the Regge limit (where
s
is large and
α
(
t
) is the Regge trajectory) [
12
,
13
] or
s
in dimensional regularization
around 4
−
2
dimensions [
14
]. Theories of quantum gravity are expected to produce
essential singularities of the form
e−s
in the high-energy limit
s→ ∞
at fixed angle
[
15
,
16
]. Resummation of diagrams can lead to non-holonomic singularities of the
form (1
−fa/2logbf
)
−1
[
17
–
19
]. It is also known that scattering amplitudes can have
distributional support in special limits, such as deep inelastic scattering in QCD [
20
], or
accumulation curves of singularities in unphysical kinematics [
21
]. One may ask if the
list of possible singularities has been exhausted.
Mathematically, analytic functions can certainly have another feature called a
natural boundary of analyticity, formed by a dense set of singularities. As such, they
provide barriers to analytic continuation. Natural boundaries are no strangers to
physicists; for example, elliptic functions such as the Dedekind eta function
η
(
τ
) have a
singularity at every rational
τ
and hence a natural boundary on the real
τ
-axis, see,
e.g., [
22
]. In Sec. 2.1 we review the basic mechanism behind such accumulations of
singularities. A natural question arises, whether scattering amplitudes can have natural
boundaries and where to find them.1
To answer this question, we turn to the simplest non-analytic feature of the 2
→
2
S-matrix: the branch cut for the lightest-threshold exchange, starting at
s
= 4
m2
, where
s
is the center-of-mass energy squared. We take the mass
m
to be strictly positive.
In four space-time dimensions, the threshold is square-root branched, meaning that it
divides the
s
-plane into two sheets glued by a branch cut, see Fig. 1.1. By convention,
they are called the first (or physical ) and second sheet respectively. We will denote the
connected part of the S-matrix on the
i
-th sheet
Ti
(
s, z
), where
z
=
cos θ
is the cosine
of the scattering angle and is held fixed (the analytic structure of the S-matrix at fixed
momentum transfer
t
is less understood, see App. A). It is on the second sheet that we
will encounter a natural boundary, see Fig. 1.1 (right).
The physical S-matrix can be recovered by approaching the branch cut from the
1
In order to avoid circular arguments, we assume that the spectrum of particles in the theory itself
is not dense and it does not have accumulation points. In most of this work we will consider a theory
with a single scalar of mass
m >
0. Note that scattering amplitudes in superstring theory have a
natural boundary in the (non-holomorphic) coupling constant (see, e.g., [
23
]), but it is not a kinematic
singularity.
– 3 –