Digital Comprehensive Summaries of Uppsala Dissertations

from the Faculty of Science and Technology 2383

From Quantum to Classical Scattering

of Kerr Black Holes
A construction of massive higher-spin scattering
amplitudes and their classical limits.

LUCILE CANGEMI

ACTA UNIVERSITATIS ISSN 1651-6214

UPSALIENSIS ISBN 978-91-513-2086-1
2024 urn:nbn:se:uu:diva-525178
Dissertation presented at Uppsala University to be publicly examined in Polhemsalen,
Ångströmlaboratoriet, Lägerhyddsvägen 1, Uppsala, Tuesday, 21 May 2024 at 09:00 for
the degree of Doctor of Philosophy. The examination will be conducted in English. Faculty
examiner: Professor Yu-tin Huang (National Taiwan University).

Abstract
Cangemi, L. 2024. From Quantum to Classical Scattering of Kerr Black Holes. A construction
of massive higher-spin scattering amplitudes and their classical limits.. Digital Comprehensive
Summaries of Uppsala Dissertations from the Faculty of Science and Technology 2383. 93 pp.
Uppsala: Acta Universitatis Upsaliensis. ISBN 978-91-513-2086-1.

Gravitational scattering processes involving black holes as asymptotic states can provide insight
into the classical dynamics of binary black hole systems. The observed gravitational waves
emitted during mergers need to be compared to high-precision theoretical predictions. By
modelling black holes as massive point particles in an effective quantum field theory, one can
take advantage of the advanced computational tools originally designed for collider physics.
For Schwarzschild black holes the natural objects to study are scattering amplitudes involving
massive scalar fields with interactions mediated by gravitons. The classical physics is extracted
by considering limits of the kinematics.
Extending this effective description to rotating Kerr black holes introduces subtleties. To
leading order in the post-Minkowskian perturbation scheme, there now exists candidate three-
point scattering amplitudes for massive higher-spin particles that in the classical limit reproduce
the Kerr metric. For small quantum spins, these are given by familiar theories of interacting
massive fields which have a well-behaved massless limit. These theories are sufficient to
capture the first few spin-multipole orders for the classical observables; however, to capture
more orders one is required to use input from higher-spin theories. The three-point higher-
spin amplitudes were originally introduced without reference to an underlying Lagrangian
description. Lagrangians for interacting higher-spin fields are notoriously complicated as they
necessarily describe composite fields in an effective higher-derivative theory.
This thesis explores the underlying higher-spin effective theories suitable for describing
rotating black holes, and proposes a new spin-s family of Compton scattering amplitudes. We
present two complementary constructions for consistent interacting higher-spin Lagrangians:
the first relies on massive higher-spin gauge symmetry to remove unwanted states, and the
second one manifests the correct degrees of freedom using a chiral field framework. A
significant portion of the thesis discusses how to extract classical physics from quantum
amplitudes, focusing on consistent treatments of the spin degrees of freedom. The resulting
quantum and classical Compton amplitudes are built to be consistent with perturbations of the
Kerr metric, through a combination of constraints from higher-spin considerations and classical
analysis.
In addition to the black-hole amplitudes, we study the scattering of higher-spin fields in
a gauge theory referred to as root-Kerr. The three point amplitudes of this gauge theory are
closely related to the Kerr ones, such that it provides an instructive model for both higher-spin
consistency and classical analysis. Another toy model discussed is the scattering of higher-spin
superstring states on the leading Regge trajectory.

Keywords: Higher-spin theory, higher-spin amplitudes, Kerr black holes, EFT, black hole
scattering, classical limit

Lucile Cangemi, Department of Physics and Astronomy, Theoretical Physics, Box 516,
Uppsala University, SE-751 20 Uppsala, Sweden.

© Lucile Cangemi 2024

ISSN 1651-6214
ISBN 978-91-513-2086-1
URN urn:nbn:se:uu:diva-525178 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-525178)
List of papers

This thesis is based on the following papers, which are referred to in

the text by their Roman numerals.

I L. Cangemi and P. Pichini Classical limit of higher-spin string

amplitudes, JHEP 06 (2023) 167, [arXiv:2207.03947]

II L. Cangemi, M. Chiodaroli, H. Johansson, A. Ochirov, P. Pichini,

E. Skvortsov, Kerr Black Holes From Massive Higher-Spin Gauge
Symmetry, Phys.Rev.Lett 131 (2023) 22, 221401,
[arXiv:2212.06120]

III L. Cangemi, M. Chiodaroli, H. Johansson, A. Ochirov, P. Pichini,

E. Skvortsov, From higher-spin gauge interactions to Compton
amplitudes for root-Kerr, arXiv:2311.14668

IV L. Cangemi, M. Chiodaroli, H. Johansson, A. Ochirov, P. Pichini,

E. Skvortsov, Compton Amplitude for Rotating Black Hole from
QFT, arXiv:2312.14913

Reprints were made with permission from the publishers.

Contents

1 Introduction .............................................................................. 7

Part I: Scattering Amplitudes for Massive Spinning Particles .......... 11

2 Warm-up: modelling Schwarzschild black holes ......................... 13
3 On-shell variables for massive spinning particles . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1 Minimally coupled massive spin-1/2 fermions . . . . . . . . . . . . . . . . . . 17
3.1.1 Spinor variables for massless fermions . . . . . . . . . . . . . . . . . . 18
3.1.2 Spinor variables for massive fermions . . . . . . . . . . . . . . . . . . . 19
3.2 On-shell variables for general spin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2.1 Consistent spin-s candidate amplitudes . . . . . . . . . . . . . . . 23
4 Higher-spin amplitudes in spin-operator variables . . . . . . . . . . . . . . . . . . . . . . 26
4.1 The spin operator and its properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1.1 Relation to the Pauli-Lubanski operator . . . . . . . . . . . . . . 27
4.2 Converting to spin variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2.1 Boosted spinors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2.2 Identifying expectations of spin operators . . . . . . . . . . . . 29
4.2.3 Amplitudes in spin operator basis . . . . . . . . . . . . . . . . . . . . . . . 30

Part II: Constructing Consistent Massive Higher-Spin Theories ....... 33

5 Non-chiral construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.1 Free theory and minimal coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.2 Adding non-minimal interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.2.1 Cubic order: Lagrangian approach . . . . . . . . . . . . . . . . . . . . . . 41
5.2.2 Cubic order: Ward identity approach . . . . . . . . . . . . . . . . . . 41
5.2.3 Adding quartic interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6 Chiral construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.1 Minimal theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.2 Parity invariant non-minimal theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.2.1 Cubic order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.2.2 Quartic order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7 String theory ........................................................................... 57

Part III: Constructing Classical Amplitudes ................................... 61

8 Scattering amplitudes from classical physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
8.1 Classical amplitudes at three points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
8.2 Classical Compton amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
√
9 Classical Kerr and Kerr Amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
9.1 Classical limits for spin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
9.1.1 (i) finite spin classical limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
9.1.2 (ii) Infinite spin classical limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
9.1.3 (iii) Coherent state classical √ limit . . . . . . . . . . . . . . . . . . . . . . . . 75
9.2 Classical constraints for Kerr and Kerr amplitudes . . . . . . . . 77
9.2.1 Candidate
√ classical Compton amplitude for
Kerr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
9.2.2 Candidate classical Compton amplitudes for Kerr 81
10 Summary of Results and Outlook ............................................. 83
11 Acknowledgements ................................................................... 85
12 Svensk Sammanfattning ........................................................... 87
References .................................................................................... 90
1. Introduction

General relativity, formulated by Albert Einstein in 1915, revolutionised

our understanding of gravity by describing it as the curvature of space-
time caused by mass and energy. Among its most intriguing solutions are
black holes. A black hole is characterised by an immense gravitational
pull so strong that nothing, not even light, can escape its gravitational
grip beyond a region called the event horizon. Inside this boundary lies
the singularity, a point of infinite density where the laws of physics as we
understand them break down. The Kerr black hole solution, proposed by
Roy Kerr in 1963, introduced the concept of rotating black holes, which
possess not only mass but also angular momentum. The Kerr black hole
with angular momentum J = m|a| has a ring-shaped singularity with a
radius |a|.
While the static solution for the Kerr black hole has been known since
1963, studying the dynamics of the Kerr black hole is complicated by
the highly non-linear nature of Einsteins field equations, the equations
governing general relativity. Of particular interest is the dynamics of
binary black holes systems; these are systems involving two black holes
orbiting one-another until, ultimately, merging into a single large black
hole. The gravitational waves emitted throughout this process can be
detected by the ground-based detectors LIGO/Virgo/KAGRA [1].
These gravitational waves encode information about their astrophys-
ical sources and extraction of this physical data relies on matching to
high-precision theoretical models. Due to the highly non-linear nature of
general relativity, solving such systems exactly is not tractable. Instead
the theoretical waveforms are usually constructed via a combination of
numerical and analytical methods that are applicable in binary systems
characterised by different physical parameters.
Early in the inspiral, the black holes are still well-separated and the
dynamics are governed by weak gravitational interactions. Therefore we
can solve the equations of motion using a perturbative approach in the
gravitational coupling G, the so-called post-Minkowski expansion (PM).
To further simplify the analysis, we can study the scattering of two black
holes where the gravitational interaction is a transient phenomena, in
contrast to the bound system where black-holes are constantly interact-
ing. In many cases, the scattering data can be related to the bound
physics; a notable example is the effective potential that governs the
conservative interactions of bound systems [42, 15, 16, 49].

7
 In the scattering scenario, we can take advantage of the wealth of high-
precision tools originally designed for collider physics by modelling black
holes as fundamental massive particles in a quantum field theory (QFT).
This is remarkable given we currently do not have a fully consistent
quantum description of gravity. The success of the QFT methods relies
on considering the Einstein-Hilbert action as an effective theory which is
compatible with general-relativity at low energies [32]. We can avoid the
usual inconsistencies related to quantising gravity by restricting ourselves
to energy scales below the Planck scale, this is compatible with the energy
scales of astrophysical black-hole mergers currently detected [1].
The dynamics of non-rotating black holes have been successfully com-
puted from scattering amplitudes involving minimally coupled massive
scalar fields, such that the potential and the scattering angle is currently
known to O(G4 ) [11, 12]. In order to incorporate rotational degrees of
freedom, one can study minimally coupled theories for massive spinning
fundamental fields. This method has been successful for low spins, such
that low orders of the classical spin dependence have also been computed
to O(G4 ) [43]. Observables for classical spinning objects can be series
expanded in the classical spin vector aµ , and contributions O(an ) are
called the n-th spin multipole coefficient. In the case of the Kerr black
hole, the spin parameter aµ is a four-dimensional repackaging of the spin
degrees of freedom of the black hole normalised such that |a| corresponds
to the radius of the ring singularity.
There is a subtle relation between the classical spin aµ , a continuous
parameter, and the quantum spin s, a discrete quantum number. The
guiding principle is that the classical limit of the interactions of a massive
spin-s field can generate, at most, the first 2s classical spin multipoles
[58]. In this framework, computing classical observables to higher orders
in the classical spin necessitates studying theories of interacting higher-
spins.
In gravity theories, minimally coupled theories of massive spinning
tensor fields develop inconsistencies for s > 2 due to propagation of un-
physical degrees of freedom [30]. In order to cancel the unphysical degrees
of freedom one can introduce auxiliary fields [55]. Zinoviev introduced
a streamlined approach that builds from a free theory for general spin
involving the physical spin-s field and a tower of Stückelberg fields all
related by a massive gauge symmetry [61]. Adding interactions con-
sistently requires maintaining the gauge symmetry at each order in the
fields constraining non-minimal interactions in the Lagrangian and in the
gauge variations [61, 62, 64, 65]. However, working at the Lagrangian
level introduces redundancies related to how the fields and parameters
in the theory are defined. To avoid this, in paper II we introduce an
on-shell approach to enforce massive gauge invariance at the level of the
currents in the form of generalised massive Ward identities.

8
 An alternative Lagrangian construction was introduced in ref. [50]
using a massive chiral field as a building block. This field automatically
propagates the correct 2s + 1 degrees of freedom so we do not need to
consider a tower of auxiliary fields. However, the minimally coupled
theories break parity invariance due to the chirality of the fundamental
field. One can impose parity order by order in the Lagrangian at the
cost of introducing non-minimal interactions [23].
In both constructions the theories of massive higher-spins involve non-
minimal interactions. Indeed, due to the no-go theorems for theories
of interacting massless higher-spin fields in flat-space [28, 60], theories
involving interacting massive higher-spins must be effective field theories
(EFTs) that are only valid below a cut-off energy scale.
The focus of this thesis is to study the EFTs that model Kerr dynamics
in the classical limit, which corresponds to a low-energy limit. Previous
work has suggested that Kerr amplitudes exhibit better high energy be-
haviour than a generic higher-spin theory would [4, 39, 26, 24]. Notably,
refs. [39, 26] showed that the linearised energy-momentum tensor of the
Kerr black hole is related to a set of three-point quantum amplitudes
MKerr originally introduced in ref. [4] due to their improved high-energy
behaviour. In ref. [4] these amplitudes were written without specifying
the Lagrangian, although it was known that for s ≤ 2 they correspond
to the interactions of minimally coupled massive fields. In Paper II, we
identified the characteristics of the higher-spin EFTs that uniquely de-
termine MKerr at three point: a combination of massive gauge symmetry
and strict power counting.
Higher-point amplitudes, such as the four-point Compton amplitude
MKerr (1s , 2s , 3, 4) are required for higher-order calculations of classical
observables. One can use on-shell methods to construct these amplitudes,
however the result contains residual freedom in the contact terms [4, 26,
3, 13]. We take an alternative approach, reducing the contact term
freedom by constraining the quartic order of the EFT with higher-spin
constraints. The reduction in the number of contact terms is boosted by
additional constraints coming from the classical limit.
The classical regime of MKerr corresponds to the long-wave length
limit of the gravitons, which is equivalent to a multisoft limit of the
graviton momenta. In order to model classical macroscopic spin, the
classical limit enforces a large quantum spin limit. One formulation of
the classical limit centres on a ‘large charge’ approach [39, 26] where the
spin quantum number scales as s ∼ ℏ−1 while the massless momenta
scale as ℏ. An alternative approach, introduced in ref. [3], involves scat-
tering coherent states of massive spins, as opposed to a single massive
spin-s state, and taking a soft graviton limit ℏ → 0. Notably this ap-
proach is also sensitive to the s → ∞ behaviour of the amplitude since
the coherent state involves an infinite sum over the spin of the massive

9
states. In both methods, the large spin limit is required to consistently
extract the correct spin multipoles even at low orders in the spin multi-
pole expansion. This limit was explored in Paper I when analysing the
classical scattering of leading Regge superstring states.
Throughout this thesis we will also consider a gauge theory which is
only known as the EFT that gives rise to the family
√ of three-point am-
plitudes in ref. [4]. The theory is referred to as Kerr due to the simple
double copy
√ structure that relates both the quantum and classical ampli-
tudes of Kerr and Kerr √ [44, 5]. In paper III we construct the underlying
higher-spin EFTs for Kerr using similar higher-spin constraints that are
relevant for the Kerr EFTs. The structure of the quantum and classical
amplitudes
√ in the gravity and gauge theory is closely related such that
Kerr is a good toy model for Kerr. In this thesis we will only discuss
the electromagnetic gauge theory, see paper III for the full non-abelian
treatment.
In Part I, we begin by considering a massive scalar field minimally
coupled to gravity in order to motivate modelling black holes using inter-
acting fundamental fields. Next we will introduce the on-shell variables
relevant for the scattering of massive spin states, taking the minimally
coupled massive fermion as an example. We will also introduce
√ our pro-
posed four point quantum Compton amplitudes for Kerr and Kerr.
Discussion of the origin of these amplitudes is postponed to Parts II and
III since we first need to introduce the quantum and classical constraints
necessary. The latter half of Part I introduces spin operator variables,
a set of quantum variables that are closely related to the classical spin
parameter aµ .
In Part II we discuss the construction of consistent theories for inter-
acting massive spins. We introduce two approaches based on two dif-
ferent representations of the Lorentz group, the (s, s) tensor fields and
the (2s, 0) chiral fields. At cubic order
√ we highlight the constraints that
uniquely determine the Kerr and Kerr amplitudes. We discuss exten-
sions to quartic order and note that our EFTs are no longer uniquely
determined.
In Part III we discuss classical amplitudes; what computations they
correspond to in classical physics and how they can be generated as limits
of the quantum amplitudes. Section 9 discusses the various classical
limits that appear in the literature to extract classical spin multipoles
from the scattering of massive spin-s particles. Now we have all the tools
necessary to√present the full set of quantum and classical constraints
that fix the Kerr and Kerr EFTs and amplitudes. We will present our
proposal for both the quantum and classical amplitudes in both theories.

10
Part I:
Scattering Amplitudes for Massive
Spinning Particles
2. Warm-up: modelling Schwarzschild
black holes

A priori it may not be obvious how to model interactions between multi-

ple black holes, such as given by the Schwarzschild metric. However, we
may use the approach of effective field theory (EFT), and write down the
simplest action that comes to mind, and then systematically add correc-
tions when needed. We know that for a distant observer a Schwarzschild
black hole is only characterised by its mass, and that its gravitational
pull is similar to that of any other massive source in general relativity.
Hence consider the simplest such effective theory, consisting of two
massive scalars minimally coupled to the Einstein-Hilbert action,
2
√ h2
Z
1X µ i
S = d4 x −g 2 R + ∇ ϕi ∇µ ϕi − m2i ϕ2i . (2.1)
κ 2 i=1

Minimal coupling corresponds to covariantising a free matter theory by

replacing the derivatives with covariant derivatives ∂µ → ∇µ , the flat
metric with the√ dynamical one ηµν → gµν and using the proper vol-
ume form d4 x −g . Of course, for a scalar field the covariant derivative
is trivial ∇µ ϕ = ∂µ ϕ. We may then expose the perturbative interac-
tions relevant to scattering amplitudes by expanding
√ the metric around
Minkowski space, gµν = ηµν + κhµν , where κ = 32πG is the coupling
constant and hµν is the dynamical gravitational field.

Massive two-to-two scattering

The classical two-to-two scattering regime corresponds to the case where
the two black holes are well separated relative to their size 2Gm, or
equivalently, where the deflection qµ = ∆pµ of a black hole’s momen-
tum is small. The dimensionless small parameter is then Gm|q| ≪ 1.
The classical interactions are mediated by graviton exchange (neglecting
pure-graviton loop diagrams). At post-Minkowskian order n-PM one has
n = L + 1 exchanged gravitons that contribute as Gn , where L is the
loop order of the classical Feynman diagram.
Consider the tree-level 1-PM case, consisting of a single Feynman di-
agram. From the above action one can extract the graviton propagator
(in de Donder gauge),
1  1 µρ µσ 1 µσ νρ 1 
∆µν;ρσ (q) = 2 η η + η η − η µν η ρσ , (2.2)
q + i0 2 2 D−2

13
where the dimension is D = 4, and also extract the scalar-graviton in-
teraction,
κ µ ′ν
V µν (ϕi , ϕ′i , h) = pi pi + pνi p′µ µν ′ 2


i + η (pi · pi − mi ) , (2.3)
2
where the momenta are related as p′µ µ µ
i = pi ± q . Then contracting
the vertices with the graviton propagator, gives the following two-to-two
scattering amplitude, at leading order in the |q| → 0 limit:
κ2 m21 m22 − 2(p1 · p2 )2
M2-to-2 = + O(q 0 ) . (2.4)
2 q2
Following the approach of Iwasaki [42] one can compute the leading
order contribution to the relativistic two-body potential by the Fourier
transform
d3 ⃗q −i⃗q·⃗r M2-to-2
Z
Gm1 m2

V1PM = 2
e ∝ 1 − 2σ , (2.5)
(2π) m1 m2 r
2
where σ = (pm12·pm2 2) is the Lorentz factor. In the non-relativistic approxi-
1 2
mation σ = 1+1/2v 2 +. . . where v is the relative velocity and we recover
Newton’s potential in the static limit v = 0.
For a general massive compact object, one would expect corrections
to the potential corresponding to the fact it is an extended object with
finite size. For a black hole the appropriate size is the Schwarzschild
radius rS = Gm, and as such finite-size corrections to the dynamics are
characterised by the mass, and possible dimensionless parameters. So
called tidal-deformability parameters, or Love numbers, are considered
dimensionfull and are expected to vanish for the Schwarzschild case, at
least the ones that are important for long-range dynamics. The min-
imally coupled scalar theory is thus expected to be a good model for
interacting black holes, during the inspiral phase of a merger. For neu-
tron stars one should instead construct an effective theory where the
tidal effects are added as higher-derivative non-minimal interactions to
the above action.
Given the above potential, one can construct an effective Hamilto-
nian in the centre-of-momentum (COM) frame, using the energies and
momenta p1 = (E1 , p⃗), p2 = (E2 , −⃗ p),
q q
H = p⃗2 − m21 + p⃗2 − m22 + V (⃗ p, ⃗r) . (2.6)

The potential has the expansion, V (⃗

p, ⃗r) = V1PM (⃗
p, ⃗r) + V2PM (⃗
p, ⃗r) + . . .,
which correspond to contributions from classical loop diagrams to the
two-to-two scattering amplitude [42, 41, 40]. In the classical regime,
the terms that have inverse powers of |q| in the scattering amplitude will

14
dominate, which corresponds to long-range dynamics. Such non-analytic
terms can be fixed by inspecting unitarity cuts of the loop amplitudes,
which in turn only require the knowledge of tree-level amplitudes in-
volving a single black hole and an arbitrary number of gravitons [14].
Therefore higher-precision classical calculations depend on knowing the
multi-graviton tree-level amplitudes.

Gravitational Compton scattering

Scattering gravitational waves (GWs) against a Schwarzschild background
is the classical process that corresponds to one of the mentioned multi-
graviton tree-level amplitudes. The simplest physical case is Compton
scattering, which involves two on-shell gravitons, one incoming and one
outgoing. In order to do the calculation from first principles in general
relativity, one has to study the asymptotic r → ∞ behaviour of solutions
to the Regge-Wheeler equation [8], which controls the perturbations to
the Schwarzschild metric. Alternatively, in the above point-particle ap-
proximation using scalar fields, this corresponds to taking the character-
istic graviton wavelength λ large enough such that the size of the black
hole is not resolved, λ ≫ rS [28]. In terms of the graviton angular fre-
quency ω ∼ 1/λ, the relevant small parameter is then Gmω ≪ 1. Thus
the gravitons have soft momenta.
The tree-level Compton amplitude for a Schwarzschild black hole can
be worked out from the above action; in the classical limit the graviton
momenta scale as k4 , k3 ∼ ℏ, giving
(p1 · F3 · F4 · p1 )2
M(1, 2, 3, 4) = −4κ2 + O(ℏ) , (2.7)
q 2 (p1 · q⊥ )2
[µ ν]
with Fiµν = 2ki εi and q = k3 + k4 , q⊥ = k4 − k3 [8]. Note that this
kinematic regime is consistent with the one relevant for the computa-
tion of the classical potential, and thus this amplitude can be used in a
unitarity cut where two soft gravitons are being exchanged between two
black holes, which is an intermediate step towards obtaining the 2PM
potential.

Extension to Kerr
Many astrophysical black holes are expected to have non-zero angular
momentum, meaning that they are described by the Kerr metric. We
would like to repeat the successful exercise of the minimally coupled
scalar, except now consider point particles which carry intrinsic angular
momentum, also known as spin.
Remarkably, it turns out that one can archive similar success by us-
ing three-point amplitudes for minimally coupled massive particles of
spins s = 1/2, 1, 3/2, corresponding to a fermion, Proca and Rarita-
Schwinger fields, respectively. These amplitudes matches the three first

15
spin-multipole orders of the linearised energy-momentum tensor of the
Kerr black hole
µν
Mmin. (1, 2, 3) ∼ εµν (q)T̃Kerr (q) , (2.8)

where a Kerr energy-momentum tensor was constructed in ref. [59],

Z
µν µν (µ
ν)
T̃Kerr (q) = d4 xeiq·x TKerr (x) = δ(p1 · q)p1 ea∗q ρ pρ1 . (2.9)

The spin multipole orders are captured by the exponential of the matrix
(a ∗ q)µ ν = iϵµ νρσ aρ q σ , and aµ = S µ /m √
is the ring radius vector of the
Kerr black hole. The magnitude |a| := −a2 is the radius of the ring
singularity [39, 26]. The details of the classical limit are discussed in
Part III.
For s ≥ 2, minimally coupled theories of massive spinning particles
are expected to develop inconsistencies due to propagation of unphysical
degrees of freedom [30]. Nonetheless, Arkani-Hamed, Huang and Huang,
introduced in ref. [4] a natural higher-spin extension of the low-spin min-
imally coupled amplitudes. It is summarised by the elegant formula

⟨12⟩2s
MKerr (1s , 2s , 3+ ) = κ(p1 · ε+
3)
2
, (2.10)
m2s
where the massive particles carry spin s and the graviton carry helic-
ity +2. The bracket denotes massive spinor-helicity variables, which we
will introduce in detail in the next section. For s < 2, the amplitudes
correspond to minimally coupled spin-s fields, and beyond this the inter-
pretation is less clear. While these amplitudes were motivated by their
simplicity and nice high-energy behaviour, later in refs. [39, 26] they were
shown to match the linearised energy-momentum tensor for Kerr up to
multipole order 2s, and thus hold for arbitrary classical spin in the limit
s → ∞.
A central result of this thesis is the explicit proposal of a consistent
Compton amplitude for Kerr,

MKerr (1s , 2s , 3± , 4± ), (2.11)

which we confirm by matching to classical results from general relativity,

independently obtained by solving the scattering problem of GWs in a
Kerr metric background [9]. In order to get there, we first will need to
extensively study and constrain non-minimal interactions contributing
to M(1s , 2s , 3± ), as well as properly implement a consistent classical
limit for spinning particles. We begin by reviewing technical aspects of
the variables used for amplitudes with massive spinning particles.

16
3. On-shell variables for massive spinning
particles

3.1 Minimally coupled massive spin-1/2 fermions

We now introduce on-shell variables for general massive spins. Let us first
discuss the lowest spin case, the spin-1/2 fermion ψ , which transforms as
a spinor in the Lorentz group. We choose to work with Dirac as opposed
to Marojana fermions since we will couple it to electromagnetism. The
on-shell free field satisfies the Dirac equation

(i∂/ − m)ψ(x) = 0 , (3.1)

where ∂/ := γ µ ∂µ . The two plane-wave solutions are ψ(x) = ua e−ip.x and

ψ̄(x) = v a eip.x with p2 = m2 and correspond to an incoming fermion
or anti-fermion respectively. In four dimensions, the massive solutions
transform under the little group SU (2), such that a = 1, 2. The massless
solutions transform under the little group U (1), such that the little group
index a corresponds to two helicity choices for the massless fermion,
u± (p).
The free theory is L = ψ̄(i∂/ −m)ψ , which one can minimally couple to
electromagnetism by promoting the derivative in to covariant derivatives
∂/ → D/ := ∂/ + iQA /,
1
/ − m)ψ − Fµν F µν ,
L = ψ̄(iD (3.2)
4
where Fµν = ∂µ Aν −∂ν Aµ . The resulting three point amplitude between
a fermion, anti-fermion and photon, all incoming, is
√
A(1a , 2b , 3) = − 2Q v b (p2 )/ε3 ua (p1 ) . (3.3)

Likewise, one can minimally couple the fermion to gravity,

√ 
L = −g ψ̄(i∇ / − m)ψ + R . (3.4)

The slashed covariant derivative is defined as ∇ / = eµ ν̂ γ ν̂ ∇µ where eµ ν̂

is the frame field with flat-space index ν̂ . The covariant derivative is
∇µ ψ = ∂µ ψ + 12 ωµ ν̂ ρ̂ Mν̂ ρ̂ ψ with spin connection ωµ ν̂ ρ̂ and spin-1/2
Lorentz generators Mν̂ ρ̂ = 14 [γν̂ , γρ̂ ]. Note from now on most indices
will be flat so we will not use the hatted notation.

17
 The resulting gravitational three-point amplitude is
1
M(1a , 2b , 3) = p1 · ε3 v2b /ε3 ua1 . (3.5)
2
For spin-1/2, the gauge and gravity amplitudes are compact expressions.
However, for general massive spins, the covariant amplitudes can be cum-
bersome and hide structure imposed by the little group covariance [33].
From now on we will work in the Weyl representation, with the gamma
matrices
0 σµ
 
γµ = , σαµα̇ = (1, σ i ) = σ̄µα̇α , (3.6)
σ̄ µ 0
and using the metric η = diag(1, −1, −1, −1). In this chiral representa-
tion, the Dirac equation reduces to two coupled Weyl equations

iσ̄ · ∂ψR = mψL , iσ · ∂ψL = −mψR , (3.7)

where the Dirac field, ψ , has been projected onto the left- and right-
handed chiral fields ψL,R
   
1 5 ψL 1 5 0
P− ψ := (1 − γ )ψ = , P+ ψ := (1 + γ )ψ = (3.8)
2 0 2 ψR

with γ 5 = iγ 0 γ 1 γ 2 γ 3 . We will now introduce spinor-helicity notation

formulated using the Weyl spinors [33, 4].

3.1.1 Spinor variables for massless fermions

For massless fermions, k 2 = 0, we will use the following angle and square
bra-ket notation for the Weyl spinors1

ψL = |k⟩e−ik·x , ψR = |k]e−ik·x , (3.9)

where |k⟩α , |k]α̇ are chiral Weyl spinors constructed from the null mo-
mentum k
kµ σαµα̇ = |k⟩α [k|α̇ . (3.10)
The two helicities of the corresponding massless Dirac spinors decompose
into    
− |k⟩ + 0
u (k) = , u (k) = (3.11)
0 |k]
where we can identify the left-handed spinor |k⟩ with negative helicity
state and the right-handed spinor |k] with positive helicity state.
1
The bra spinors are defined by ⟨k|α = ϵαβ |k⟩β and [k|α̇ = ϵα̇β̇ |k]β̇ where ϵ12 = 1 = ϵ21
is an antisymmetric SL(2, C) Levi-Civita.

18
 Using massless spinors, we can also construct spin-1 polarisations for
a massless photon of momentum k µ with respect to the null reference
momentum rµ
⟨r|σ µ |k] ⟨k|σ µ |r]
εµ+ (k, r) = √ , εµ− (k, r) = √ . (3.12)
2⟨rk⟩ 2[kr]
The spinor contractions are defined ⟨rk⟩ := ϵαβ |r⟩α |k⟩β and [rk] :=
ϵα̇β̇ |r]α̇ |k]β̇ . The graviton polarisations can be constructed as a tensor
product of the spin-1 polarisations εµν µ ν µν µ ν
+ = ε+ ε+ , ε− = ε− ε− , which sat-
isfy the required symmetric, traceless properties of the on-shell massless
spin-2 field.
We will also make use of the alternative notation |i⟩ where i corre-
sponds to the particle label on momentum ki . In these spinor variables,
the three-point amplitude between two massless fermions and a photon
can be written compactly
√ √ [23]2
A(1− + +
m=0 , 2m=0 , 3 ) = 2Qv + (k2 )/ε+
3 u−
(k1 ) = 2Q , (3.13)
[12]
Note that we must use complexified momenta otherwise the amplitude
vanishes due to the restrictive three point kinematics.

3.1.2 Spinor variables for massive fermions

One can construct similar plane-wave solutions for the massive fermions,
p2 = m2 ,
ψL = |pa ⟩e−ip·x , ψR = |pa ]e−ip·x . (3.14)
The SU (2) indices run over a = 1, 2 such that each variable |pa ⟩, |pa ]
is a 2 × 2 matrix. Note that the left- and right-handed spinors are not
independent, since the Weyl equations in eq. (3.7) imposes the relation
p · σ|pa ] = m|pa ⟩ . (3.15)
The parametrisation of the massive spinors is not unique. We use a
parameterisation for the massive momenta p that appears in [24],
m2
p=k+ r, (3.16)
2p · r
where r is a null reference momentum and null momentum k is defined
by solving the on-shell condition p2 = m2 . In this parametrisation the
massive spinors are written explicitly as
 m   
a ⟨kr⟩ |r⟩ a |k]
|p ⟩ = , |p ] = m |r] . (3.17)
|k⟩ [kr]

19
For labelled massive momenta pi we will use the notation |ia ⟩, |ia ]. In
order to avoid working with the free indices, we contract the little group
index with a complex wavefunction zi,a and define the bold massive
spinors
|i⟩ = |ia ⟩zia , |i] = |ia ]zia . (3.18)
Self-contractions of the bold spinors vanish due to symmetrisation im-
posed by the wavefunctions, for example ⟨11⟩ = −m ϵab z1a z1b = 0.
The three point amplitude for the massive fermions coupled to elec-
tromagnetism can then be written in the form
√ ⟨21⟩
A(1s=1/2 , 2s=1/2 , 3+ ) = 2Q p1 · ε+
3 . (3.19)
m
√
This is the lowest spinning case in the Kerr family of amplitudes in-
troduced in ref. [4],
√ ⟨21⟩2s
A(1s , 2s , 3+ ) = 2Q p1 · ε+
3 . (3.20)
m2s
We can generate the negative helicity amplitude A(1s=1/2 , 2s=1/2 , 3− )
−
by swapping ε+
3 → ε3 and |i⟩ ↔ |i]. We can also compute the four-point
Compton amplitudes for the fermion coupled to electromagnetism in the
two independent helicity sectors

m[34]2
A(1s=1/2 , 2s=1/2 , 3+ , 4+ ) = Q2 ⟨21⟩ ,
t13 t14
(3.21)
⟨3|1|4]
A(1s=1/2 , 2s=1/2 , 3− , 4+ ) = Q2 (⟨13⟩[42] − ⟨23⟩[41]) ,
t13 t14
where ⟨3|1|4] = p1µ ⟨3|σ µ |4] and, as before, the poles t1i = 2p1 · ki come
from the massive propagator. Note that the three-point and four-point
amplitudes can be factored into a scalar amplitude and a spin-dependent
coefficient, for example

⟨21⟩
A(1s=1/2 , 2s=1/2 , 3+ ) = A(1s=0 , 2s=0 , 3+ ) , (3.22)
m
Where the scalar amplitudes
√
A(1s=0 , 2s=0 , 3+ ) = 2Q p1 · ε+
3
m2 [34]2 ⟨3|1|4]2
A(10 , 20 , 3+ , 4+ ) = Q2 , A(10 , 20 , 3− , 4+ ) = Q2
,
t13 t14 t13 t14
(3.23)
1
are computed from the Lagrangian L = |Dµ ϕ| − m |ϕ| − 4 (Fµν )2 for
2 2 2

scalars minimally coupled to electromagnetism.

20
 The analogous amplitudes for fermions minimally coupled to gravity
are remarkably similar in structure to the gauge theory,

2 ⟨21⟩
M(1s=1/2 , 2s=1/2 , 3+ ) = κ(p1 · ε+3) ,
m
 κ 2 m3 [34]4
M(1s=1/2 , 2s=1/2 , 3+ , 4+ ) = ⟨21⟩ ,
2 t13 t14 s12
 κ 2 ⟨3|1|4]3
M(1s=1/2 , 2s=1/2 , 3− , 4+ ) = (⟨13⟩[42] − ⟨23⟩[41]) ,
2 t13 t14 s12
(3.24)
where s12 = (p1 + p2 )2 . These gravitational amplitudes can be obtained
trivially from the gauge theory by a double copy prescription,
κ
M(1s , 2s , 3) = A(10 , 20 , 3)A(1s , 2s , 3) Q→1 ,
2
 κ 2 t t (3.25)
13 14
M(1s , 2s , 3, 4) = A(10 , 20 , 3, 4)A(1s , 2s , 3, 4) Q→1
,
2 s12
which holds for s ≤ 2 as discussed in refs. [4, 44].

3.2 On-shell variables for general spin

We will now construct higher-spin states from the spin-1/2 building
blocks. The massive spin-1 polarisation for particle pi can be defined
in terms of massive spinors
⟨i|σ µ |i]
εµi = √ . (3.26)
2m
In this parameterisation, the polarisations are explicitly transverse and
null given the spinors are explicitly symmetrised by their contraction
with zia . Note that one can resolve the three degrees of freedom of
massive spin-1 particle by expanding polarisation as a polynomial in zia
√
εµi = (zi1 )2 εµ−,i − 2zi1 zi2 εµL,i − (zi2 )2 εµ+,i . (3.27)

Where εµL,i = kiµ /m − m rµ /(2pi ·r) is the longitudinal degree of freedom

of the massive spin-1 particle. In order to describe complex massive
vector bosons we take the wavefunctions z to be complex. In which case,
we can introduce the polarisation for the complex conjugated spin-1 field
(⟨i|σ µ |i])∗ ⟨ī|σ µ |ī]
ε̄µi = √ =− √ (3.28)
2m 2m
where the barred spinors |ī⟩ := |ia ⟩z̄ia , |ī] := |ia ]z̄ia are functions of z̄ia =
(zia )∗ , z̄ia = −(zia )∗ . The spinor products between barred and unbarred

21
 a
√ a ⟨1̄1⟩ = [11̄] = mz̄ za , and proportional to
variables are non-vanishing,
the normalisation |z| := z̄ za .
We can construct polarisation tensors for higher-spin states as tensor
products of the spin-1 and spin-1/2 variables
integer spin: εµi 1 ...µs := εµi 1 εµi 2 . . . εµi s , (3.29)
µ
half-integer spin: |i⟩µ1 ...µ⌊s⌋ := |i⟩εµi 1 εµi 2 . . . εi ⌊s⌋ , (3.30)
and similarly for the right-handed polarisation |i]µ1 ...µ⌊s⌋ . These fully-
symmetric tensors satisfy the required transversality and tracelessness
properties that characterise the irreducible representations of the lit-
tle group SO(3). Tracelessness and so-called gamma-tracelessness is a
consequence of the properties ε2i = 0 and εi · σ̄|i⟩ = 0 of the spin-1
polarisations, which follow from ⟨ii⟩ = 0.

Higher-spin amplitudes from on-shell approaches

Ref. [4] provides a study of three-point amplitudes involving two states
of mass m and spin s with a photon or graviton. At three points, the
general amplitudes have the form
2s
√ p1 · ε+ X
A(1s , 2s , 3+ ) = 2 3
cn ⟨12⟩2s−n [12]n ,
m2s n=0
2s
(3.31)
s s + (p1 · ε+
3)
2 X
M(1 , 2 , 3 ) = κ c̃n ⟨12⟩2s−n [12]n ,
m2s+1 n=0
where the coefficients cn , c̃n satisfy the constraints
2s
X 2s
X 2s
X
cn = Q , c̃n = m , nc̃n = 0 . (3.32)
n=0 n=0 n=0
The first two correspond to normalisation of the charge or mass monopole
in the respective gauge and gravity theories. The coefficients for the
gravity amplitudes satisfy an extra constraint due to the universality of
the dipole coupling.
For low spins, s ≤ 1 in gauge theory and s ≤ 2 in gravity, the three
point amplitudes with coefficients cn = Q δn0 , c̃n = m δn0 correspond to
the interactions of minimally coupled fundamental massive fields. For
these low spin theories, one can compute the four-point Compton am-
plitudes from the Lagrangians directly. The resulting the all-plus ampli-
tudes have the form
m2 [34]2 ⟨12⟩2s
A(1s , 2s , 3+ , 4+ ) = ,
t13 t14 m2s
(3.33)
m4 [34]4 ⟨12⟩2s
M(1s , 2s , 3+ , 4+ ) = .
s12 t13 t14 m2s

22
 For
√higher-spins, the Lagrangians that reproduce the three point Kerr
and Kerr three point amplitudes are no longer theories of interacting
fundamental fields. However, one can avoid using the Lagrangian and
bootstrap the higher-point amplitudes using factorisation properties as
done in [4] or BCFW on-shell recursion [19, 20]. In the same-helicity
sector, the resulting amplitudes correspond to those in eq. (3.33).
However the opposite-helicity amplitudes proposed in ref. [4] have the
form
⟨3|1|4]2−2s
AAHH (1s , 2s , 3− , 4+ ) = Q2 (⟨13⟩[42] − ⟨23⟩[41])2s ,
t13 t14
 κ 2 ⟨3|1|4]4−2s
MAHH (1s , 2s , 3− , 4+ ) = (⟨13⟩[42] − ⟨23⟩[41])2s ,
2 s12 t13 t14
(3.34)
such that they develop unphysical poles in the kinematic factor ⟨3|1|4]
for spins s > 1 and s > 2 in the gauge and gravity theories. Although
one can add contact terms to remove the spurious poles, the resulting
amplitude is not unique [26].

3.2.1 Consistent spin-s candidate amplitudes

We take a different approach in this thesis; we construct the effective
theories that generate A√Kerr (1s , 2s , 3) and MKerr (1s , 2s , 3) for higher
spins. At quartic order, we use a combination of higher-spin and classi-
cal constraints to fix the contact terms in the four-point Compton am-
plitudes. We will postpone the construction of the effective theories to
Part II and only present the resulting quantum general-spin amplitudes
in this section.
To avoid writing the couplings, we introduce the notation,
A(1s, 2s, 3, 4) = Q2 A(1s, 2s, 3, 4)
 κ 2 (3.35)
M(1s, 2s, 3, 4) = M (1s, 2s, 3, 4) .
2
√
Compton proposal √ for Kerr
Our proposal for the Kerr Compton amplitude in electromagnetism2
is
⟨3|1|4]2 (2s) ⟨13⟩⟨3|1|4][42] (2s)
A√Kerr (1s, 2s, 3−, 4+ ) = P − P2
t13 t14 1 m2 t13
⟨13⟩⟨32⟩[14][42] h (2s−1) (2s−1) ς3 +ς4

(2s) (2s−2)
i
+ P2 − ς ς P
3 4 4 + P 4 − P 2 .
m4 2
(3.36)
2
see Paper III for the full non-abelian amplitude.

23
We use the helicity-independent variables3

ς1 := ⟨1|4|2]/m2 + [21]/m , ς3 := ⟨21⟩/m ,

(3.37)
ς2 := −⟨2|4|1]/m2 + [21]/m , ς4 := [21]/m ,

as they form a basis for the dependence on the massive spinors of particles
1 and 2. In these variables, the dependence on the quantum spin s
of the particles 1 and 2 is encoded in the family of totally symmetric
√
(k)
homogeneous polynomials Pn . The relevant polynomials in the Kerr
amplitude are
k−1
(k) (k) (k)
X X
P1 = ς1k , P2 = ς1i ς2k−1−i , P4 = ς1i ς2j ς3l ς4r . (3.38)
i=0 i+j+l
+r=k−3

(k)
The degree of the polynomial Pn is k +1−n, so clearly the polynomials
(k) (n−1)
satisfy the identities Pn = 0 for k < n−1 and Pn = 1, which can be
used to identify at what spin each term in eq. (3.36) starts to contribute.
For s ≤ 1 only the first three terms contribute, and the amplitudes agree
with AAHH in eq. (3.34) which are spurious pole free for these spins. At
s = 3/2 the last two polynomials contribute independently but there is
a cancellation such that P43 − P21 = 0 and the amplitude agrees with the
s = 3/2 amplitude proposed in ref. [24]. The amplitude does not develop
any spurious poles and factorises correctly on the massive poles.

Compton proposal for Kerr

One of the central results in this thesis is the spin-s family of Kerr
Compton amplitudes,

⟨3|1|4]4 (2s) ⟨13⟩[42]⟨3|1|4]3 (2s)

M (1s, 2s, 3−, 4+ ) = P − P2
s12 t13 t14 1 s12 t13 m2
⟨13⟩⟨32⟩[14][42] 2 (2s−1) 2 (2s−1)


+ ⟨3|1|4] P 2 + ⟨3|ρ|4] P 4
m4 s12
⟨13⟩⟨32⟩[14][42] (2s−2) (2s−2)

+ 4
⟨3|1|4]⟨3|ρ|4] P2 − ς3 ς4 P4
m s12
⟨13⟩2 ⟨32⟩2 [14]2 [42]2 h
(2s−2) (2s−2)
i
+ ς ς
3 4 (1+η)P5|ς1 + (1−η)P 5|ς2 . (3.39)
2m6
The amplitude is expressed in the same spinor variables ςi and we have
introduced the variable ρµ = 12 (⟨1|σ µ |2] + ⟨2|σ µ |1]). In the last line
we also introduce the limits of the n = 5 polynomial which can also be
3
Note that variables ςi are defined with an extra m−2 factor compare to the definitions
papers I, III and IV such that in this thesis they dimensionless.

24
 (k)
interpreted as a derivative on P4 ,

(k) (k) ∂ (k)

P5|ςi := lim P5 = P , (3.40)
ς5 →ςi ∂ςi 4
(k)
Note that for k+1−n ≥ 0, the general polynomial Pn can be resummed
into the rational form
ς1k
Pn(k) = + cyc(ς1 , ς2 . . . ςn ) . (3.41)
(ς1 − ς2 )(ς1 − ς3 ) . . . (ς1 − ςn )
However, on four-point kinematics there are only four ςi variables, hence
why the higher order polynomials appear with doubled variables. It
would be interesting to investigate what polynomials appear in the five-
point and higher amplitudes, where we expect the spinor structure of the
amplitudes to depend on more than than the four variables ς1 , ς2 , ς3 , ς4 .
The first four terms of MKerr are fixed by the cubic higher-spin theory.
The last term is a contact term that we ansatz as a function of the poly-
(k)
nomials Pn and is fixed by a combination of higher-spin and classical
constraints, which we discuss in Part III.
For s ≤ 2 the first 5 terms contribute such that the amplitudes agree
with MAHH defined in eq. (3.34). At s = 5/2 the contact terms in the
last line do not contribute and the amplitude agrees with the s = 5/2
amplitude proposed in ref. [24]. For general spin the amplitude factorises
correctly onto the Kerr three point amplitudes (2.10).
The parameter η is introduced to match the results in ref. [9], which
studies the classical scattering of gravitational waves off a Kerr black
hole background. The matching is discussed in detail in Part III where
(s)
we will also consider adding possible contact terms Cα .

25
4. Higher-spin amplitudes in
spin-operator variables

As discussed in the introduction, the three-point amplitudes MKerr can

be related to the energy-momentum tensor for a Kerr black hole (2.9).
However, these amplitudes are functions of spinors and the discrete spin
quantum number s, while the energy momentum tensor of Kerr is a
function of the continuous spin vector aµ . In order to relate the two we
express the quantum amplitude in a basis of quantum spin-operators â,
which we will later identify with the classical parameter aµ , discussed in
detail in Part III. Schematically the necessary steps are

⟨12⟩2s
MKerr ∼ −−−−−−−−−→ ⟨eâ·p3 ⟩ −→ ea·p3 ∼ T µν . (4.1)
m2s spin op. basis cl. limit

In this chapter we will discuss the transformation to the spin operator

basis. Note that, while motivated by the classical limit, we can encode
the full quantum amplitude in this basis.

4.1 The spin operator and its properties

The quantum spin operator âµ is defined with respect to the massive
momentum p1 and acts on it’s little group representations,

⃗ 1  
(âµ )⃗a b := ⟨1a1 |σ µ (b1 b2
|1 ]δa2 . . . δ b2s )
a2s + [1 a1 |σ̄ µ (b1 b2
|1 ⟩δ a2 . . . δ b2s )
a2s ,
2m2
(4.2)
⃗
where ⃗a = a1 . . . a2s and b = b1 . . . b2s are multi-indices with ai , bi ∈
{1, 2}. This operator can be considered as a covariantisation of the
quantum mechanical spin operator defined in three dimensions

[Ŝ i , Ŝ j ] = −ϵijk Ŝ k . (4.3)

The normalised spin-s four-dimensional operator, âµ = Ŝ µ /m, satisfies

the following relations
1 µνρσ
[âµ , âν ] = ε pρ âσ m2 ηµν (âµ · âν ) = −s(s + 1)1 (4.4)
m2
26
where 1 is the SU (2) unit-operator. In order to construct expectation
values of the operator we can absorb the SU (2) indices with the spinor
wavefunctions such that

⟨â(µ1 . . . âµn ) ⟩ := (z̄)2s · (â(µ1 . . . âµn ) ) · (z)2s . (4.5)

In particular, the expectation value of a single spin-s operator is

⟨âµ ⟩ = (z̄)2s · (âµ ) · (z)2s

s (4.6)
= (⟨1|σ µ |1̄] + ⟨1̄|σ µ |1])(z̄ a za )2s−1 .
2m
For finite spin-s representations we can take at most 2s products of
the spin operator as higher products collapse down to a polynomial of
highest order (â)2s . The general form of such a product where n > 2s is
s s
(n)
X Y
⟨â(µ1 . . . âµn ) ⟩ = di ⟨â(µ1 . . . âµi (â · â)s−i ⟩ P µk µk+1 ) (4.7)
i=0 k=i+1

µ ν (n)
where P µν = η µν − pmp2 is the spin-1 projector and di is a com-
binatorical factor that can be computed by performing the little group
contractions. For finite s, the self-contraction ⟨â·â⟩ is proportional to the
Casimir and contributes as an overall numerical factor −s(s+1)(z̄ a za )2s .

4.1.1 Relation to the Pauli-Lubanski operator

The Lorentz-covariant spin operator is often defined in terms of the Pauli-
Lubanski operator Ŝ instead,
i µνρσ
Ŝµ = ϵ pν Mρσ . (4.8)
m
µν
The operator Ŝ is defined in terms of the spin-s Lorentz generators M(s) ,
such that the operator acts on the Lorentz representations of the spin-s
particle as opposed to the little group. Thus we take the expectation of
an integer spin-s operator with respect to the spin-s polarisations,

⟨Sˆµ ⟩ := ε̄(s) · Sµ · ε(s) . (4.9)

However, given our parameterisation of the massive polarisations in eq. (3.26),

there exists a simple relation between the two definitions,
2s
⃗ 1 Y ∂ ∂  ˆµ
m(âµ )⃗a b = ⟨S ⟩ . (4.10)
(2s)!2 j=1 ∂ z̄ aj ∂zbj

27
We show some explicit low spin cases
   b 
1 µ b ⟨1a | |1 ⟩
s= 2 : m(â )a = Ŝ
m m
  b1
|1 ⟩ |1b2 ⟩
 
⟨1a1 | ⟨1a2 |
s = 1 : m(âµ )a1 a2 b1 b2 = ⊙ Ŝ ⊙ , (4.11)
m m m m

where the massive spinors are contracted into the SL(2, C) indices of the
Lorentz generators. The ⊙ notation implies a symmetric tensor product
in the SL(2, C) indices such that ⟨1a1 | ⊗ ⟨1a2 | = ⟨1a1 |(α1 ⟨1a2 |α2 ) .
Although the operator mâ acts on the little group SU (2) while Ŝ acts
on the SL(2, Z) indices, the expectation values are equivalent ⟨mâ⟩ = ⟨Ŝ⟩
so long as the expectation values are taken with respect to physical states.

4.2 Converting to spin variables

Converting scattering amplitudes written in covariant notation or mas-
sive spinor variables into the spin-operator basis involves two steps.
1. We first parametrise the outgoing degrees of freedom in terms of the
incoming degrees of freedom. This is necessary as the spin operator
is defined with respect to the little group of a single spin-s state,
which we take to be particle 1.
2. Then we identify spin operators, either from combinations of the
massive polarisations ε1 , ε̄1 or the massive spinors |1⟩, |1̄⟩.

4.2.1 Boosted spinors

The amplitudes considered in this section are generalised Compton am-
plitudes featuring two massive spinning particles and n − 2 massless
states
A(1s , 2s , 3, . . . , n) . (4.12)

This amplitude transforms under the little group of the massive particles
1 and 2. The spin operator, however, acts on the little group of particle
1, for which the massive bra and ket spinors |1̄⟩ = (⟨1|)† , |1⟩ form a
basis. In order to express the amplitude as a function of â we expand
the degrees of freedom of particle 2 in terms of particle 1. This can be
done via a Lorentz boost Λ,

1 1
|2⟩ = Λ|1̄⟩ = (|1̄⟩ + q · σ|1̄]) . (4.13)
cq 2m

28
The boost1 is defined such that
 iζ q p 
µ ν
Λ = exp 2
M µν
, pµ2 = Λµ ν pν1 := −pµ1 − q µ , (4.14)
sinh ζ m
Pn
where q µ = i=3 kiµ and the parameter ζ is defined by its relation to cq ,
r
q2 ζ
cq = 1 − 2
= cosh . (4.15)
4m 2
Note that the Lorentz boost is well-defined for null q (ζ = 0) and
corresponds to the boosts defined in [48] for spin-1/2 and spin-1 rep-
resentations. For the Compton scattering considered in the thesis, we
have q 2 ≤ 0. Thus the boosts are well defined and we do not hit the
singularity at q 2 = 4m2 .

4.2.2 Identifying expectations of spin operators

Covariant identities
Identities exist that map products of spin-s polarisations εµ1 ...µs ε̄ν1 ...νs
to expectation values of the spin-s operators. However, the general form
is cumbersome to derive. In paper I and in ref. [26], some low spin
examples are given explicitly, for compactness we will take p = p1 , ε = ε1
in the following examples. The relevant identity for s = 1 is,
i
ε̄µ εν = −m2 ⟨â(µ âν) ⟩ + ϵµνρσ pρ ⟨âσ ⟩ − P µν |z|2 . (4.16)
2
where P µν is the spin-1 projector and we have used m2 ⟨â · â⟩ = 2|z|2 for
s = 1. For s = 2 the identity already starts to become cumbersome,

m4 (µ1 µ2 ν1 ν2 ) im2 µ1 ν1 κ
ε̄µ1 ε̄µ2 εν1 εν2 = ⟨â â â â ⟩ − ϵ (λ µ2 ν2 )
λ pκ ⟨â â â ⟩
6 6
m2  µ1 µ2 (ν1 ν2 ) 
+ P ⟨â â ⟩ + P ν1 ν2 ⟨â(µ1 âµ2 ) ⟩ + 28P µ1 ν1 ⟨â(µ2 âν2 ) ⟩
36
7i
− P µ1 ν1 ϵµ2 ν2 κ λ pκ ⟨âλ ⟩ + P µ1 µ2 ν1 ν2 |z|4 , (4.17)
18
where P µ1 µ2 ν1 ν2 = 2 P µρ P νσ + P µσ P νρ − 23 P µν P ρσ is the spin-2 pro-
1


jector [24]. Mapping from polarisations to spin operators is not very
flexible given we do not have a general formula, furthermore they are
not valid for half-integer spins. Instead, one can work at the level of the
massive spinors, converting first to the spin-1/2 representation and then
changing to the generic spin-s representation.
1
Strictly speaking this the Lorentz transformation also includes a reflection since in
our conventions the momenta are all incoming.

29
Spin-1/2 identities
The one-particle expectation value of a spin-1/2 spin operator is
1
āµ := ⟨âµ ⟩ = (⟨1̄|σ µ |1] + [1̄|σ̄ µ |1⟩) . (4.18)
2m2
We do not need to consider higher-order products of the spin-1/2 oper-
ator as they reduce down to
(
(µ1 µ2 µn ) |z|2 P µ1 µ2 . . . P µn−1 µn for even n
⟨â â . . . â ⟩ ∝ (4.19)
āµ1 P µ2 µ3 . . . P µn−1 µn for odd n

where the proportionality constants are combinatorical factors depending

on n. The momentum and ā variables are constructed from symmetric
and anti-symmetric combinations of the massive spinors,

|1̄⟩[1| + |1⟩[1̄| = m2 ā · σ , |1̄⟩[1| − |1⟩[1̄| = p1 · σ|z|2 . (4.20)

It is sufficient to consider the expansion of the vectors

1 m i
ρµ := (⟨1|σ µ |2] + ⟨2|σ µ |1]) = |ā| (pµ1 − pµ2 ) − ϵµ (p, q, ā) , (4.21)
2 cq cq
1 q · ā
ρ̄µ := (⟨1|σ µ |2] − ⟨2|σ µ |1]) = 2m2 cq āµ − pµ (4.22)
2 cq
since ρ and ρ̄ form a basis for the spinor variables, where we use ϵµ (p, q, ā) =
ϵµνρσ pν qρ āσ . When√s = 1/2, we can choose freely when to substitute
the identity |ā| := −â2 = |z|2 /(2m) in the expansion of ρ. Our pre-
scription will prioritise keeping an equal count of massless momenta and
ā at leading order, such that ρ · p, ρ · ρ̄ ∼ |z|2 and ρ · k3 , ρ · k4 ∼ |a|. As an
example we can expand the spinor structure of same-helicity amplitude
(3.33),
m2 q2

m⟨21⟩ = p1 · (ρ − ρ̄) = |z|2 + q · ā − |ā| . (4.23)
cq m

We will often choose to normalise the z variables such that |z|2 = 1. Al-
ternatively one can redefine the expectation values ⟨â⟩ → ⟨â⟩/(z̄ a za )2s
and factor out an overall (z̄ a za )2s normalisation in the amplitudes. How-
ever, for now we will avoid either normalisation in order to keep the little
group covariance of the following identities explicit.

4.2.3 Amplitudes in spin operator basis

The identities in eq. (4.21) map the massive spinors to the spin-operator
basis, such that we can express a generic Compton amplitude A(1s , 2s , 3, . . . , n)

30
as polynomial in āµ ,
2s
X
A(1s , 2s , 3, . . . ,n) ∼ cµi 1 ...µi āµ1 . . . āµi (z̄ a za )2s−i . (4.24)
i=0

The tensors ci are the theory dependent factors which depend on the
momenta p1 , k3 . . . kn and the massless polarisations ε3 . . . εn . In order
to maintain little group covariance, the amplitudes must be a polynomial
of order 2s in the wavefunctions z, z̄ .
We can also consider expanding the amplitude into an alternative basis
constructed from spin-operators in the spin-s basis,
2s
X
A(1s , 2s , 3, . . . ,n) ∼ c̃µi 1 ...µi ⟨â(µ1 . . . âµi ) ⟩. (4.25)
i=0

this is a natural choice for an amplitude involving massive spin-s states.

The map between the coefficients ci and c̃i relies on combinatorical
factors generated by the change of the representation of operator âµ .
√
Example: three-point Kerr amplitudes √
We can expand the spin-s three-point Kerr amplitudes into the spin-
1/2 basis where the first step was done in eq. (4.23),
2s
√ √

+ ⟨21⟩  2s
s s +
A(1 , 2 , q ) = 2Q p1 · ε3 = 2Q p1 · ε+
3 1 + q · ā
m
2s
√ +
X 2s  n
= 2Qp1 · ε3 q · ā .
n
n=0
(4.26)
To change to the spin-s representation of the spin operator, we use the
following three-point identity
(2s − n)!
(q · ā)n = (q · â)n , (4.27)
(2s)!
This identity can be derived explicitly from the definition of âµ in eq. (4.6)
and the expectation values eq. (4.5). This identity provides the maps be-
tween the coefficients such that c̃µi 1 ...µi = (2s−n)! µ1 ...µi
(2s)! ci . The resulting
amplitude is
2s
s s +
√ +
X 1
A(1 , 2 , q ) = 2Q p1 · ε3 ⟨(q · â)n ⟩
n=0
n! (4.28)
√ + −2s q·â
= 2Q p1 · ε3 cq ⟨e ⟩ ,
where we consider the exponential to truncate at order 2s in the operator.

31
 The form of eq. (4.27) is governed by the properties of SU (2) group.
On general kinematics, the identity used to change representations has
the form
(2s − n)! (µ1 µ2
ā(µ1 āµ2 . . . āµn ) = â â · · · âµn ) + O(â2 ) , (4.29)
(2s)!

which has corrections proportional to powers of the Casimir â2 ∝ s(s+1),

for example at quadratic order

(2s − 2)!  s ⟨â2 ⟩ 

(q · ā)2 = ⟨(q · â)2 ⟩ − P µν qµ qν . (4.30)
(2s)! 4 s(s + 1)

Note that at three points the â2 correction vanishes as P µν qµ qν = 0,

however for higher-point kinematics it does not. This structure per-
sists for the corrections proportional to higher-order powers of â2 . Each
self-contraction of the spin-operator comes hand-in-hand with the spin-1
projector such that the tensor structure of the O(â2i ) correction is
n−1
Y
(µ1 µn−2i 2 i
⟨â . . . â (â ) ⟩ P µk µk+1 . (4.31)
k=n+1−2i

The Casimir relation highlights a general ambiguity in the decomposi-

tion of the amplitude into products of spin-s operators. Suppose we de-
compose A(1s , 2s , 3, . . . , n) into the spin-s basis, as in eq. (4.25), and we
generate an expression for the quadrupole, c̃µ1 µ2 , and the hexadecapole,
c̃µ1 µ2 µ3 µ4 , tensors. The Casimir identity implies that the amplitude is
invariant under the redefinition of the multipole tensors

c̃µ1 µ2 → c̃µ1 µ2 − s(s + 1)c̃µ1 µ2 ν1 ν2 η ν1 ν2 ,

(4.32)
c̃µ1 µ2 µ3 µ4 → c̃µ1 µ2 µ3 µ4 + s(s + 1)ηµ3 µ4 c̃µ1 µ2 ν1 ν2 η ν1 ν2 ,

such that clearly the quantum multipole tensors are not uniquely defined.
As we will see in part III this introduces ambiguities in the finite spin
classical limit where we need to introduce a prescription to relate the
quantum multipole tensors to the classical coefficients. The identification
turns out to be robust when the classical limits involve taking s → ∞.
In this work we make use of both the spin-1/2 and the spin-s bases;
the choice is dependent on which classical limit procedure we implement.
In general, whenever we discuss finite spin amplitudes we are implicitly
using the spin-s basis.

32
Part II:
Constructing Consistent Massive
Higher-Spin Theories
A massive spin-s particle corresponds to an irreducible representation
of the little group SU (2) and has 2s + 1 physical degrees of freedom in
four-dimensions. In order to construct a field theory describing the inter-
actions of this particle, we package up the degrees of freedom into a field
Φ. The field Φ is taken to transform in one of the spin-s representations
of the Lorentz group. In this work we consider two distinct constructions
based on the (s, s) and (2s, 0) representations of the Lorentz group.
Non-Chiral fields are (s, s)-representations of the Lorentz group and
correspond to totally symmetric tensor fields. We work with a tower of
double traceless fields2 {Φµ(s) , Φµ(s−1) . . . Φµ(0) } related by a massive
gauge symmetry [61]. In principle one can construct theories for any
half-integer representation s, however we will only discuss integer spin,
i.e. massive spinning bosons. Fermionic constructions for s = 32 in gauge
theory and s = 52 in gravity were studied in ref. [24].
Chiral fields are (2s, 0)-representations of the Lorentz group and
correspond to a totally symmetric chiral field Φα1 ...α2s , which contains
2s + 1 degrees of freedom [50]. The chiral construction is well suited to
describing both fermionic and bosonic interactions.
The descriptor ‘higher-spins’ refers to theories where the spin of the
massive field s is greater than the spin |h| of the massless force carrier,

2
For compactness we introduce the shorthand notation for multi-indices Φµ(k) :=
Φµ1 ...µk and trace Φ̃µ(k−1) := Φµ1 µ1 µ2 ...µk . Note Φµ(0) corresponds to a scalar field
and Φ̃µ(k) := 0 for k < 4.
in gauge theory |h| = 1 and in gravity |h| = 2. The distinction between
‘low’ and ‘high’ spins is inherited from the study of massless interacting
spin-s fields. The theories of massless low spins are the well-understood
gauge and gravity theories, however theories of massless higher-spin the-
ories are notoriously hard to construct consistently, see refs. [10, 52] for
a review of the challenges involved.
We expect theories of massive high spins, in flat-space, to develop
divergences in the high-energy limit given there is no corresponding con-
sistent massless theory. In light of this, theories of interacting massive
higher-spin fields should be considered as effective theories describing
composite particles. Early work in refs. [39, 38, 26, 24] suggested that the
theories with improved, albeit not finite, high-energy behaviour generate
the known three-point Kerr amplitudes. Motivated by this connection,
we did a systematic study of the gauge and gravity theories generating
the three point amplitudes A√Kerr and MKerr in papers II and III.
We have labelled the two constructions for the higher-spin theories
by the chirality properties of the primary ingredients, the spin-s fields.
In section 6 we will discuss how to introduce non-minimal interactions
to restore parity in the chiral approach. In this section, we will first
discuss the non-chiral approach making use of some illustrative low spin
examples.

34
5. Non-chiral construction

5.1 Free theory and minimal coupling

We can package the physical spin-s degrees of freedom into a totally
symmetric tensor field Φµ(s) . We will also impose the field to be double-
traceless, Φν1 ν2 ν1 ν2 µ5 ...µs−4 = 0. In four-dimensions, the field Φµ1 ...µs for
s > 0 contains
1 1
(s + 1)3 − (s − 3)3 (5.1)
3! 3!
degrees of freedom, where (b)n := Γ(b + n)/Γ(b) is the Pochammer sym-
bol. The second term corresponds to the degrees of freedom killed by the
double tracelessness property of the fields and it vanishes for s = 1, 2, 3.
In order to reduce to the physical 2s + 1 degrees of freedom, we add
auxiliary fields coupled to the physical field by a generalised gauge sym-
metry.
Following Zinoviev’s approach [61], we add a tower of symmetric,
double-trace-free Stückelberg fields {Φµ(s−1) , Φµ(s−2) , . . . , Φµ(1) , Φµ(0) }.
The fields are subject to a gauge symmetry,

δ0 Φµ(k) = ∂(µ1 ξµ2 ...µk ) + mαk ξµ1 ...µk + mβk η(µ1 µ2 ξµ3 ...µk ) , (5.2)

which depend on the totally symmetric traceless gauge parameters ξµ(k)

and the numerical coefficients
s
(s − k)(s + k + 1) kαk−1
αk = , βk = . (5.3)
2(k + 1)(k + 1) 2k − 2

The degrees of freedom in the full tower of fields is

s
X 1  1 1
1+ (k + 1)3 − (k − 3)3 = (s + 1)4 − (s − 3)4 , (5.4)
3! 4! 4!
k=1

where we factored out the k = 0 term corresponding to the degrees of

freedom of the scalar field. However the gauge symmetry can be used to
remove the unphysical degrees of freedom. The tower of traceless gauge
parameters contain a total of
s−1
X 1  1
1+ (k + 1)3 − (k − 1)3 = (2s)3 , (5.5)
3! 4!
k=1

35
degrees of freedom, where once again the first term corresponds to the
degree of freedom of ξ k=0 . Fixing the gauge symmetry removes double
the degrees of freedom of the gauge parameters. This is a general feature,
which one can motivate by considering the massless gauge variations of
the photon fields, δ̃0 Aµ = ∂µ α where α is the gauge parameter. By
fixing Lorenz gauge ∂ · A = 0, one can remove a single scalar degree
of freedom. The Lorenz condition leaves a remaining scalar degree of
freedom unfixed, given that □α = 0 trivialises δ̃(∂ · A) = □α. This
scalar mode decouples in physical processes.
Thus, the total count of degrees of freedom for the massive spin-s
theory corresponds to
1

fields − 2 × gauge params. = (s + 1)4 − (s − 3)4 − 2(2s)3
4! (5.6)
= 2s + 1 ,

and confirms that the field content of the theory contains the correct
degrees of freedom. There exist other packagings of the 2s + 1 degrees of
freedom where the field content is altered, for example using the partially
gauge fixed approach of Singh-Hagen [55], or using traceful fields [35].
However, we find the Zinoviev approach is well suited to a generic spin
approach and gives simple computational rules.
The original presentation of the free Lagrangian can be found in
ref. [61]. In paper III we worked out a simpler form using complexi-
fied fields. The Lagrangian is decomposed into a Feynman-gauge part
and the corresponding gauge-fixing terms,

L2 = LF − Lgf , (5.7)

where LF is diagonal in the fields

s  
µ(k) k(k − 1) ˜
X
k+1 2 2 µ(k−2)
LF = − (−1) Φ̄µ(k) (□+m )Φ − Φ̄µ(k−2) (□+m )Φ̃ ,
4
k=0
(5.8)
The transverse and off-diagonal quadratic interactions belong to
s−1
X
Lgf = − (−1)k (k + 1)Gµ(k) Gµ(k) , (5.9)
k=0

which is quadratic in the gauge-fixing the functions Gµ(k)

k
Gµ(k) := ∂λ Φλµ(k) − ∂ µ Φ̃µ(k−1) + mαk Φµ(k)
2 (5.10)
k+1 µ(k) k−1 µµ µ(k−2)
− mαk+1 Φ̃ − mαk η Φ̃ .
2 4
36
Both terms are necessary for the free theory to be invariant under the
massive gauge transformations 5.2 such that δ0 L2 = 0. However the
decomposition anticipates our gauge choice, Lgf = 0 is Feynman gauge,
which is closely related to Lorenz gauge Gµ(k) = 0. As discussed previ-
ously, this gauge fixing removes half of the unphysical degrees of freedom.
Since the variation of the constraint, δ0 Gµ(k) = (□ + m2 )ξ µ(k) , is trivi-
alised by (□ + m2 )ξ µ(k) = 0, the residual gauge symmetry removes the
second half of the unphysical degrees of freedom.
In Feynman gauge, the massive propagators ∆(s) for a spin-s state
can be obtained by inverting LF . For unconstrained fields in Feynman
gauge this is well defined. However, recall that the field Φµ(s) is double
traceless, hence the propagator has to be a projector that enforces this
constraint. Combining the two conditions, being an inverse of the kinetic
term and a double-traceless projector, one can show that the propaga-
tor is unique. These propagators can be packaged into the following
generating function
∞
X
s (s) s i 1 − 14 ϵ2 ϵ̄2
∆(ϵ, ϵ̄) = (ϵ) ·∆ ·(ϵ̄) = 2 . (5.11)
s=0
p −m2 +i0 1 + ϵ · ϵ̄ + 14 ϵ2 ϵ̄2

We can extract the spin-1 massive propagator by acting on ∆(ϵ, ϵ̄) with
derivatives,

∂2∆ iη µν
∆(1)µν = =− . (5.12)
∂ϵµ ∂ϵ̄ν ϵ2 →0 p2 −m2 +i0
ϵ̄2 →0

This is the mass deformed version of the usual spin-1 Feynman-gauge

propagator. For spin-2, the massive propagator is

1 ∂4∆ i η µρ η νσ + η µσ η νρ − η µν η ρσ
∆(2)µν;ρσ= = ,
(2!)2 ∂ϵµ ∂ϵν ∂ϵ̄ρ ∂ϵ̄σ ϵ2 →0 2 p2 −m2 +i0
ϵ̄2 →0
(5.13)
which corresponds to a mass deformation of the de Donder propagator
(2.2). The propagator ∆(k) is a valid for both the physical and auxiliary
spin-k fields. This follows from the LF , which makes no distinction
between physical and auxiliary fields.

Minimal coupling
With the free theory at hand, we can investigate what happens when
we couple either to electromagnetism1 or gravity. In order to couple
the theory consistently we have to, at minimum, upgrade all partial
1
See paper II for coupling to a non-abelian gauge field.

37
derivatives in L2 and δ0 to the relevant covariant derivatives

electromagnetism: ∂ν Φµ(k) → Dν Φµ(k) = ∂ν Φµ(k) + iQAν Φµ(k)

δ0 Φµ(k) → δ0 Φµ(k) = D(µ1 ξµ2 ...µk + . . .
gravity: ∂ν Φµ(k) → ∇ν Φµ(k) = ∂ν Φµ(k) + kΓ(µ
ρν Φ
1 ρµ2 ...µk )

δ0 Φµ(k) → δ0 Φµ(k) = ∇(µ1 ξµ2 ...µk ) + . . . (5.14)

There are different prescriptions for minimal couplings since before co-
variantisaton the derivatives commute but not after, i.e. [∂µ , ∂ν ] = 0 but
[Dµ , Dν ] = Fµν . Consistent minimal coupling to gravity also requires
√ the dynamical one ηµν → gµν and using
replacing the flat metric with
the proper volume form d4 x −g .
In this thesis we work with the free Lagrangian presented in eq. (5.9).
The three-point amplitudes generated by minimal Lagrangian are

Amin (1s , 2s , 3+ ) = x⟨12⟩s [12]s ,

(5.15)
Mmin (1s , 2s , 3+ ) = x2 ⟨12⟩s [12]s−1 (s⟨12⟩ − (s − 1)[12]) .

While these amplitudes are well-behaved, given they transform covari-

antly under the little group transformations of the massive and massless
particles, the underlying theories are not consistent. Indeed, the mini-
mal coupling induces an explicit breaking of the massive gauge symmetry
such that
δ0 L2 ̸= 0 . (5.16)
This is a problem as the gauge invariance was used to enforce the correct
degrees of freedom. Thus the breaking induced by minimal coupling
implies unphysical degrees of freedom now propagate.

Massive spin-1 electromagnetism

The breaking of the massive gauge symmetry is easiest to see at low
spins, for example consider a massive spin-1 field, here denoted by W µ ,
minimally coupled to electromagnetism,

(s=1) 1
L2 = − |Wµν |2 + |mWµ − Dµ φ|2 , (5.17)
2
where Wµν = 2D[µ Wν] 2 and, as before, fields that couple to electro-
magnetism are complex. The minimal coupling extension of the massive
gauge variations are

δ0 Wµ = Dµ ξ , δ0 φ = mξ , δ0 Aµ = 0 . (5.18)
2
Note that theis minimal coupling prescription differs from the one used for the
general-spin free Lagrangian in eq. (5.9).

38
Using this gauge symmetry we could fix φ = 0 to obtain the usual
Proca action, however we will keep gauge unfixed in order to use gauge
invariance as a constraint. At cubic order in the fields, massive gauge
invariance requires
δ0 L2 = O(Q2 ) , (5.19)
While order Q0 is guaranteed to vanish by the massive gauge invariance
of the free theory. It suffices to consider the variations of the unbarred
field Φ, giving
(s=1)
δ0,Q1 L2,Q0 = iQξ (∂µ ∂ · W̄ )Aµ − (∂ 2 W̄µ )Aµ − m2 W̄ ·A + m2 (∂ φ̄)·A
 

(s=1) (s=1)
δ0,Q0 L2,Q1 = −δ0,Q1 L2,Q0 + 2iQ ξ ∂[µ W̄ν] ∂ µ Aν .
(5.20)
Where we distinguish the free theory variations and Lagrangian, δ0,Q0
and L2,Q0 , from the non-linear interactions introduced by the covariant
derivatives, δ0,Q1 and L2,Q1 . Thus adding the two contributions gives
δ0 Ls=1
2 = 2iQ ξ ∂[µ W̄ν] ∂ µ Aν , (5.21)
implying that the massive gauge symmetry is broken in the minimally
coupled theory. Since the gauge invariance was necessary to encode the
physical degrees of freedom, the breaking induced by minimal coupling
implies that unphysical degrees of freedom can now propagate. Note,
the propagator is still η µν /(p2 − m2 ).
We can restore the gauge invariance by introducing non-minimal in-
teractions in the Lagrangian and massive gauge variations. The low-
est derivative non-minimal extension of the massive spin-1 theory in
eq. (5.17) that leaves the minimal coupling unchanged is
µ
L3 = −ic1 QW Fµν W ν , δ 1 Wµ = 0 = δ 1 φ , ¯ µ] .
δ1 Aµ = ic2 Q[W̄µ ξ−ξW
(5.22)
Imposing massive gauge invariance at cubic order fixes the coefficients
uniquely to c1 = c2 = 1. This non-minimal contribution corresponds to
the cubic interactions of W bosons in the standard model. The three
point amplitude,
A(1s=1, 2̄s=1, 3) = −2Q(ε1· ε2 ε3·p1 + ε2· ε3 ε1·p2 + ε3· ε1 ε2·p3 ) , (5.23)
√
corresponds to the s = 1 Kerr amplitude in (3.20). Notably, for spin-1,
this lowest derivative non-minimal interaction is required for the theory
to have a smooth massless limit, in which case we obtain the non-abelian
Yang-Mills theory with gauge group SU (2). The Higgs field and self-
interactions of the spin-1 field W are not considered in this set up, see
paper III for details.
The breaking of the gauge symmetry is a property of minimally cou-
pled theories of generic spins s ≥ 1 coupled to gauge theory and s ≥ 2

39
in gravity. In the next section we will discuss how to restore it system-
atically.

5.2 Adding non-minimal interactions

Returning to the spin-s case, we conclude that minimal coupling of the
free Lagrangians introduces propagating unphysical degrees of freedom.
In order to restore the massive gauge invariance, we will introduce non-
minimal interactions order by order in the fields,
L = L2 + L3 + L4 + . . . ,
(5.24)
δ = δ0 + δ1 + δ2 + . . . .
Where L2 and δ0 are minimally coupled extensions of the free theory
Lagrangian (5.9) and gauge variations (5.2). The non-minimal interac-
tions Ln>2 and δn>0 are constructed via an anstaz and constrained by
massive gauge invariance of the full theory δL = 0. We will also impose
further constraints on the interactions, they are
(SI) interactions involve at most two massive fields, as the self-interactions
of massive fields are suppressed in the classical limit of the effec-
tive theory,
(PS) parity symmetry is imposed on the interactions. This implies
we will not include interactions involving the Levi-Civita tensor
ϵµνρσ or (anti-)self-dual field strengths,
(MC) minimal-coupling interactions and linearized gauge transforma-
tions δ0 will not be modified,
(PC) power counting in the L3 interactions have at most s + s′ − |h|
derivatives, where s and s′ are the ranks of the massive fields,
e.g. rank 1 for Wµ and rank 0 for φ and where |h| = 1, 2 is the
spin of the photon or graviton respectively.
(PC2) power counting in the δ1 gauge variations have at most s+s′ −|h|
derivatives, where s and s′ are the ranks δ1 Φ and ξ , respectively.
The last two constraints (PC) and (PC2) correspond to the lowest-
derivative solutions compatible with the other constraints. In order to
describe a general higher-spin theory these power-counting constraints
should be loosened.
These constraints, along with massless gauge invariance, imply the
following structure for the non-minimal interactions and gauge variations
electromag.:Ln ∼ Φ̄Φ(Fµν )n−2 , δn Φ ∼ ξ(Fµν )n ,
δn Aµ ∼ ξ Φ̄(Aν )n−1 , (5.25)
n−2 n
gravity: Ln ∼ Φ̄Φ(Rµνρσ ) , δn Φ ∼ ξ(Rµνρσ ) .
δn hµν ∼ ξ Φ̄(hρσ )n−1 ,

40
where derivatives are not included. Note that Aµ , hµν also picks up
variations that are non-linear in the fields. These depend on the massive
gauge parameters ξ and are not related to the gauge parameters of the
massless gauge symmetry.

5.2.1 Cubic order: Lagrangian approach

By including non-minimal cubic interactions L3 , we now have a hope of
restoring massive gauge invariance at cubic order in the coupling g
(δ0 + δ1 )(L2 + L3 ) = O(g 2 ). (5.26)
√
where g = Q in electromagnetism and g = κ = 32πGN . The right-
hand side is linear in g given L2 and δ0 are fixed by the free theory and
they only L3 and δ1 need to be constructed via ansatz.
In papers II and III, we studied the full off-shell gauge invariance
of massive s = 2, s = 3 fields coupled to electromagnetism and s =
3 coupled to gravity. In each case, it was found that massive gauge
invariance of the full off-shell Lagrangian coupled with the constraints
(SI), (PS), (MC), (PC) and (PC2) is enough to uniquely fix the free
parameters that appear in the three-point amplitudes A(1s , 2s , 3) and
M(1s , 2s , 3). √
In particular these amplitudes correspond to the Kerr amplitudes
(3.20). While the theories describing these amplitudes are not minimal
due to massive gauge invariance constraints, the results from papers II
and III suggest that they correspond to lowest derivative non-minimal
theories.
As the spin s increases the number of free parameters in the interac-
tions and gauge variations increases steeply. This renders it somewhat
inconvenient to study higher-spin theories using the full Lagrangian vari-
ations, in particular when moving on to higher multiplicity. Therefore
we move on to Ward identities that makes studying higher spins more
tractable.

5.2.2 Cubic order: Ward identity approach

The non-linear variations of the free Lagrangian are proportional to the
equations of motion
δL2
δ 1 L2 = δ1Φµ(k) −→ 0 , (5.27)
δΦµ(k) on-shell

such that it vanishes on shell. The massive gauge invariance constraint

in eq. (5.26) simplifies to
δ0 L2 + L3 on-shell = O(g 2 ) .


(5.28)

41
This constraint no longer depends on the non-linear variation δ1 , which
we previously needed to construct via ansatz.
Let us discuss the practical implementation of the above constraints.
We work in momentum space, where vertices V (. . . ) are generated by
the Lagrangian or constructed as an ansatz. At cubic order the vertices
can effectively be read off from the Lagrangian using identifications

∂µ → −ipµ , Φµ(k) → ϵµ1 . . . ϵµk , (5.29)

µ µ µ(k) µ1 µk
A →ϵ , ξ →ϵ ...ϵ . (5.30)

The vectors ϵµ can be considered off-shell placeholders for the physical

polarisations, such that, in general, ϵµ does not satisfy the usual trans-
verse and null constraints required for on-shell states. In momentum
space we put a particle label on the momenta and polarisations, there-
fore there is no ambiguity in what ϵµ means for the different fields. The
on-shell massive state i satisfies

p2i = m2 , pi · ϵi = 0 , ϵ2i = 0 , (5.31)

while the massless state i satisfies

ki2 = 0 , k i · ϵi = 0 , ϵ2i = 0 . (5.32)

Therefore putting the particle i on-shell corresponds to swapping ϵi → εi

or εi depending on whether the particle is massive or not. The notation
V (. . . ) (i) indicates that we put particle i on shell.
The three point vertex receives a contribution from the minimal cou-
pling, in this case to electromagnetism, of the free Lagrangian and the
unfixed non-minimal cubic interactions

V (Φk1 Φs2 A3 ) = Vmin (Φk1 Φs2 A3 ) + Vnon-min. (Φk1 Φs2 A3 ) . (5.33)

We will mainly consider the case where leg 2 is the highest spin-s field
Φs := Φµ(s) and leg 1 is one of the fields in the tower with k ≤ s. The ver-
tices are Lorentz-invariant polynomials in the polarisations (ϵ1 )k , (ϵ2 )s , ϵ3
and momenta p ∈ {p1 , p2 }, with schematic form
p 
Vmin (Φk1 Φs2 A3 ) ∼ m(ϵ1 )k (ϵ2 )s ϵ3 +1 ,
m
s+k−1
X  p n (5.34)
Vnon-min. (Φk1 Φs2 A3 ) ∼ m(ϵ1 )k (ϵ2 )s ϵ3 .
n=0
m

The vertex Vmin is given by the free theory while Vnon-min. is constructed
as an ansatz and constrained by the previously introduced constraints
(MC), (PS), (SI) and (PC).

42
 The variation of the Lagrangian (5.28) corresponds to the constraint
on the cubic interactions V3 = (L2 + L3 )|3pt ,
s
X δV3
δ0 V3 = δ Φµ(k) .
µ(k) 0 (5.35)
k=0
δΦ
Using the ϵ, p variables, the variations translate to
k
δV3 1 Y ∂
→ V (Φk1 Φs2 A3 ) ,
δΦµ(k) k! ∂ϵµ1 l
l=1
1
µ(k) µ(k−1) mαk−1 µk−1 µk µ(k−2)
δ0 Φµ(k) → mαk ϵ1 − ipµ1 k ϵ1 + η ϵ1 .
k 2(k − 1)2
(5.36)
The function δ0 V3 is an inhomogeneous polynomial in (ϵ1 )k , 0 ≤ k ≤ s.
It is useful to extract the homogeneous pieces, which we denote by
1 ∂
V (ξ1k Φ̄s2 A3 ) :=mαk V (Φk1 Φ̄s2 A3 ) − ip1 · V (Φk+1
1 Φ̄s2 A3 )
k+1 ∂ϵ1
(5.37)
mαk+1 ∂ ∂
+ · V (Φk+2
1 Φ̄s2 A3 ) .
2(k+1)2 ∂ϵ1 ∂ϵ1
The gauge invariance constraint in eq. (5.28) corresponds to imposing
δ0 V3 such that we supplement the list of constraints (SI), (PS), (MC),
(PC) with the following:

(WI) Massive Ward identities V (ξ1k Φ̄s2 A3 ) (2,3),ϵ21 =0

= 0, where we set
ϵ21
= 0 in order to impose the tracelessness constraint on the gauge
parameter ξ˜µ(k−2) = 0.

So far we have discussed the Ward identities in the context of elec-

tromagnetism. There is an analogous construction in gravity where we
adjust the power-counting in Vnon-min. to reflect the constraint (PC). All
the equations above hold up to swapping Aµ → hµν , ϵ3µ → ϵ3µν as
the structure of the Ward identity is inherited by the free theory gauge
variations.

Results
Since the Ward identity construction is more economical, we are able to
explore the theories that satisfy the constraints (SI), (PS), (MC), (PC)
and (WI) up to s = 10. Up to s = 1 in electromagnetism, and up to
s = 2 in gravity, this gives unique theories.Free parameters appear in the
three-point amplitudes starting at s = 2 in electromagnetism
( s−1 )
⟨12⟩2s [12]k
X 
s s +
A(Φ1 Φ2 A3 ) = A0 1+ bk −1 , (5.38)
m2s ⟨12⟩k
k=1

43
and s = 3 in gravity
(  s−4
2 X )
2s k
⟨12⟩ [12] [12]
M(Φs1 Φs2 h+
3 ) = M0 1+ 1− b̃k . (5.39)
m2s ⟨12⟩ ⟨12⟩k
k=0
The parameters bk are related to the free parameters
Ps−1 in the general three-
point amplitudes in eq. (3.31) by c0 = 1 − k=1 and ck>0 = bk>0 ,
likewise for c̃k and b̃k . Therefore clearly the higher-spin constraints (SI),
(PS), (MC), (PC) and (WI) fix s + 1 of the 2s − 1 free parameters.
The loss of uniqueness in comparison to the off-shell approach can be
traced back to the translation between off-shell massive gauge invariance
and constraint on the currents in (WI). Notably, we did not construct
′
or impose the Ward identities generated by vertices V (Φk Φ̄k A3 ) with
k, k ′ < s, such that neither massive fields are the top spin field. These
interactions do generate constraints when varying the full off-shell La-
grangian, especially since the interactions are constrained by the power
counting (PC). The obstruction to including such terms is that these ver-
tices include contain no on shell states, hence we always consider k ′ = s.
To constrain the remaining s parameters and obtain unique ampli-
tudes, introduce two the additional constraints:
(CC) Current constraint is motivated by high-energy unitarity, as dis-
cussed in ref. [24]:
∂
p1 · V (Φs1 Φ̄s2 A3 ) = O(m) . (5.40)
∂ϵ1 (2,3),ϵ21 =0

(ND) Near-diagonal interactions of two massive fields. The cubic vertices

differ at most by one unit:
|h|
V (Φs−k
1 Φ̄s2 A3 ) k>|h|
= 0, (5.41)
where |h| = 1, 2 in electromagnetism and gravity respectively.
The first constraint (CC) has been previously discussed in higher spin
literature, in ref. [34] it is sufficient to fix the tree-level gyromagnetic
ratio to g = 2 for theories of general-spin particles. In ref. [27, 51]
it is introduced to avoid violations of tree-level unitarity for theories
of massive higher-spins. As motivation consider the spin-1 four point
amplitude A(1s=1 , 2s=1 , 3, 4). Unitary gauge fixes ϕ = 0 in eq. (5.17)
giving the Proca Lagrangian. Using the unitary gauge spin-1 propagator,
we can construct one of massive channels
µ ν
∂ η µν − pmp2 ∂
V (W 1 W A
p 3 ) V (Wp W 2 A4 ) . (5.42)
∂ϵµp p2 − m2 ∂ϵνp (1,2,3,4)

Adding the constraint (CC) is sufficient to fix all the free parameters
in eq. (5.39) uniquely to b̃k = 0, giving a unique Kerr amplitude. Com-
bining (CC) with the second constraint (ND) is necessary to fix bk = 0

44
 √
in eq. (5.38) giving a unique Kerr amplitude. The latter constraint is a
property of the minimal Lagrangian and amounts to imposing stronger
power counting constraints on certain vertices.
The current constraint is a remarkably powerful constraint at low
spins. In ref. [24], it was used to uniquely fix the three point ampli-
tudes up to s = 3/2 in gauge theory and s = 5/2 in gravity.

5.2.3 Adding quartic interactions

In principle, extending the analysis to quartic order is straightforward
and corresponds to solving

(δ0 + δ1 + δ2 )(L2 + L3 + L4 ) = O(g 3 ) , (5.43)

where we include non-minimal interactions that are quartic in the fields

L4 and quadratic in the fields for δ2 . We can fix all the physical pa-
rameters in δ1 and L3 by cubic order gauge invariance. Any remaining
unfixed parameters at O(g) correspond to field re-definitions and can be
set to zero without loss of generality. Therefore the quartic order massive
gauge invariance is a linear constraint.
In practice, the full off-shell analysis is intractable for high spins; in
paper III we were able to implement up to s = 3 in electromagnetism.
The constraint in eq. (5.43) is a system of s − 1 equations that are highly
coupled and depend tensor structures that increase in complexity as the
spin grows.
The Ward identity formulation provides an efficient approach to the
quartic order analysis. In this approach we can neglect the non-linear
gauge variations and construct currents corresponding to interactions
where all-but-one of the fields in the interactions satisfy on-shell con-
straints.
At three points, the Ward identities act on currents generated from
cubic vertices where two fields are taken on-shell. At quartic order, the
relevant current resembles a four point amplitude with one massive leg
off-shell,

⟨Φk1 Φ̄s2 A3 A4 ⟩ := Γt (Φk1 Φ̄s2 A3 A4 )+Γu (Φk1 Φ̄s2 A3 A4 )
(2,3,4)

+V4 (Φk1 Φ̄s2 A3 A4 ) . (5.44)
(2,3,4)

In electromagnetism, the current has contributions from the t- and u-

channel graphs Γt , Γu and from the quartic vertices V4 (Φk1 Φ̄s2 A3 A4 ) gen-
erated by the minimal coupling of L2 and L3 as well as the unfixed
quartic interactions L4 . In the non-abelian gauge theory and gravity

45
theories there is also a massless channel since the massless fields self-
interact. The four point amplitude corresponds to setting leg-1 on-shell
too, A(1s, 2̄s, 3, 4) := ⟨Φs Φ̄s A3 A4 ⟩ (1,2,3,4) .
Note that the only free parameters in eq. (5.44) appear in contact
terms V4 (Φk1 Φ̄s2 A3 A4 ) which include contributions from the ansatzed the
non-minimal quartic interactions using (MC) and (PS). We will discuss
the relevant power-counting shortly. The contributions of the massive
channels are constructed out of the cubic currents using the massive
Feynman-like propagators defined in eq. (5.11),

s  ∂
X ∂ 
Γt (Φk1 Φ̄s2 A3 A4 ) := ∆ , V (Φk1 Φ̄lP A4 )V (Φl−P Φ̄s2 A3 ) .
∂ϵP ∂ϵ̄P ϵP →0
l=0 ϵ̄P →0
(5.45)
In analogy to the cubic Ward identities, we can now consider the varia-
tion of field Φk and isolate the constraints generated by each independent
gauge parameter ξ k . The resulting quartic Ward identities are

h ip1 ∂
⟨ξ1k Φ̄s2 A3 A4 ⟩ (2,3,4) := mαk ⟨Φk1 Φ̄s2 A3 A4 ⟩ − · ⟨Φk+1
1 Φ̄s2 A3 A4 ⟩
2
ϵ1 =0
k+1 ∂ϵ1
 2
m ∂ i
+ βk+2 ⟨Φk+2
1 Φ̄ s
A
2 3 4A ⟩ .
2 ∂ϵ1 (2,3,4)
2
ϵ1 =0
(5.46)
Imposing the Ward identities corresponds to solving a system of s lin-
ear equations, which constraints the quartic contact terms in A(1s , 2s , 3, 4).
In paper III , we present discuss the calculation for spins s = 2, 3 in the
non-abelian gauge theory.

Spin-2 electromagnetism
The first non-trivial result at quartic order is the spin-2 electromagnetism
analysis. In paper III , there is a detailed discussion of the cubic order
analysis using both off-shell and Ward identity approaches. At quartic
order we ansatz the non-minimal interactions that enter in the massive
Ward identities, V (Φk1 Φ̄22 A3 A4 ) for k ≤ 2.
The lowest-possible derivative counting of the four point amplitude
A(1s=2, 2̄s=2, 3, 4) is fixed by the exchange of the highest spin-state. For
spin-2, the minimum derivative count in the four-point amplitude is 4s−4
since (PC) implies that each cubic vertex can contribute 2s + 1 deriva-
tives, while the propagator in eq. (5.11) removes two.

46
 The power-counting used in our ansatz for the non-minimal interac-
tions is
V (Φ21 Φ̄22 A3 A4 ) ∼ p4 ,
V (Φ11 Φ̄22 A3 A4 ) ∼ p5 , (5.47)
V (Φ01 Φ̄22 A3 A4 ) = 0,
where the last vertex vanishes, satisfying the (ND) constraint. Each
vertex is constructed explicitly out of field strengths for the photons,
Fµν → k3[µ ϵ3ν] . Note that the diagonal spin-2 interaction has the lowest
power-counting 4s − 4 = 4.
Imposing the quartic order Ward identities introduces constraints on
the parameters in the ansatz. However it is not enough to uniquely fix
the four-point amplitudes, such that the resulting amplitudes are

⟨12⟩4 [34]2 (2)

A(1s=2, 2̄s=2, 3+, 4+ ) = 2
+ C++ , (5.48)
m t13 t14
2
⟨3|1|4] (2) ⟨13⟩⟨3|1|4][42] (2)
A(1s=2, 2̄s=2, 3−, 4+ ) = P1 − P2 (5.49)
t13 t14 m2 t13
⟨13⟩⟨32⟩[14][42] (1) (2)
+ P2 + C−+ , (5.50)
m4
(2) (2)
where C++ and C−+ contain unfixed parameters. We note these param-
eters remain unfixed even when we consider the full off-shell variation of
the Lagrangian as done in paper III . The opposite-helicity amplitude
has three unfixed free parameters,
(2) c1 c2
C−+ = 4
⟨13⟩⟨32⟩[14][42](ς3 + ς4 )2 + 4 ⟨13⟩⟨32⟩[14][42](ς3 − ς4 )
m  m
c3
+ 8 m⟨3|1|4]⟨21⟩(ς3 + ς4 )(⟨1|3|2]⟨23⟩[14] + ⟨2|3|1]⟨13⟩[42])
m 
+ ⟨3|1|4]2 [12]⟨12⟩3 + s12 ⟨12⟩[12]⟨13⟩⟨32⟩[14][42] ,
(5.51)

while the same helicity sector has five unfixed parameters,

(2) [34]2
C++ = (c5 ⟨12⟩3 [12] + c6 ⟨12⟩2 [12]2 + c7 ⟨12⟩[12]3 )
m6 (5.52)
⟨12⟩
+ (c8 [12] + c9 ⟨12⟩)([13]2 [24]2 + [14]2 [23]2 ) .
m6
In Part III we constrain the contact terms
√ further by imposing classi-
cal consistency constraints. The spin-2 Kerr amplitudes introduced
in eq. (3.33) and proposed in eq. (3.36) correspond to the fixing the
coefficients c1 = −c2 = 1 and cn>2 = 0.

47
6. Chiral construction

6.1 Minimal theory

In this approach we will use fields in the (2s, 0) chiral representation of
the Lorentz group. In this representation, the fields are totally symmetric
2s-spinor fields Φα(2s) , such that the multi-index α(2s) = α1 . . . α2s is
a string of spacetime SL(2, C) indices each taking values αi = 1, 2. In
contrast to the fields in the (s, s)-representation, the fields in this chiral
representation contain exactly 2s + 1 degrees of freedom. Therefore one
can construct the free theory for a spin-s particles from a single field [50]

L(s) = ⟨∂µ Φ|∂ µ Φ⟩ − m2 ⟨Φ|Φ⟩ . (6.1)

We have introduced a short-hand notation to suppress the spinor indices,
|Φ⟩ := Φα(2s) , |∂ µ Φ⟩ := ∂ µ Φα(2s) , ⟨Φ|Φ⟩ := Φα(2s) Φα(2s) . (6.2)
While in the (s, s) representation the natural on-shell wavefunctions for
a spin-s tensor are the spin-s polarisations Φµ(s) → εµ1 . . . εµ2s , the
natural on-shell wavefunctions for the chiral field are the left-handed
massive spinors Φα(2s) → |p⟩α1 . . . |p⟩α2s /ms [50].
In order to add interactions, we can minimally couple the theory to
electromagnetism and gravity using the same prescription as in section
5.1.

Minimal coupling to electromagnetism

We couple to electromagnetism by promoting the partial derivatives in
the free Lagrangian to ∂µ → Dµ = ∂µ + iQAµ ,
1 m2
L = ⟨Dµ Φ|Dµ Φ⟩ − ⟨Φ|Φ⟩ . (6.3)
2 2
In this thesis I will only introduce the electromagnetic theory and we as-
sume Φ carries electric charges, which is not one-to-one with the complex
properties of the fields given the spinor representation is itself complex.
The general non-abelian treatment was introduced in [50] and discussed
in paper III .
We compute the three-point vertex V2 (Φs1 Φs2 A3 ) as a function of the
massive spinors
⟨21⟩2s
Vmin (Φs1 Φs2 A3 ) = Q(p1 − p2 ) · ϵ3 . (6.4)
m2s
48
where the spinors are assumed to not yet satisfy the coupled Weyl equa-
tions in eq. (3.7). Hence, the three point amplitude for the minimally
coupled theory is practically the same equation,

√ ⟨21⟩2s
A(1s, 2̄s, 3± ) = V2 (Φs1 Φs2 A3 ) = 2Q p1 · ε+
3 . (6.5)
(1,2,3± ) m2s

Unsurprisingly, the three point amplitudes are break chiral symmetry

since
A(1s, 2̄s, 3+ ) ̸= A(1s, 2̄s, 3− ) ε− + . (6.6)
3 →ε3
|i⟩↔|i]

This asymmetry in the dependence on right- and left-handed spinors is

inherited from the Lagrangian, which is constructed from the chiral left-
handed field Φα(2s) . We will restore parity at the level of the amplitudes
by including a tower of non-minimal interactions to L.
Note that Vmin factorises into the scalar vertex and spin-dependent
term, Vmin (Φs1 Φs2 A3 ) = Vmin (Φ01 Φ02 A3 )⟨21⟩2s /m2s . This a feature of the
minimally coupled Lagrangian, which behaves as a scalar Lagrangian
given the contractions of the spinor fields are trivial.

Minimal coupling to gravity

The Lagrangian for the chiral spin-s theory minimally coupled to gravity
is
√ m2
 
1 µ
L = −g ⟨∇µ Φ|∇ Φ⟩ − ⟨Φ|Φ⟩ . (6.7)
2 2

Consistent minimal coupling causes a proliferation of interactions com-

pared to the gauge theory. In this section we will use the parametrisation
of the metric gµν = ηµν
√ + κhµν . There are two cubic interactions, one
from the expansion of −g = 1 + 12 h + . . .

κ
⟨21⟩2s
Vmin,√g (Φs1 , Φs2 , h3 ) = − ϵ23 p1 · p2 − m2 . (6.8)
2 m2s

Note that this vertex exhibits the same factorisation into a scalar vertex
and a spin-dependent factor as the vertex in electromagnetism. However
this is not the case for the other vertex, generated by the expansion of the
spin connection ωµ,α β := 14 ωµ ν̂ ρ̂ σν̂,αγ̇ σ̄ρ̂γ̇β that appears in the covariant
derivative

∇µ Φα1 ...α2s = ∂µ Φα1 ...α2s + 2s ωµ,(α1 β Φα2 ...α2s )β . (6.9)

49
The resulting off-shell three-point vertex
κh
V2,∇ (Φs1 , Φs2 , h3 ) = − (ϵ3 · p1 )2 ⟨21⟩2s − s(ϵ3 · p1 )⟨21⟩2s−1 ⟨2|p3 ϵ3 |1⟩
2
1
+ ⟨21⟩2s−1 (ϵ3 · p3 ) 2(s−1)(ϵ3 · p1 )⟨21⟩−s⟨2|p3 ϵ3 |1⟩


2
1 i
+ s(ϵ3 · p3 )2 ⟨21⟩2s .
2
(6.10)
Putting the graviton on-shell greatly simplifies the vertex such that
only the first line contributes. The full on-shell three point amplitude is
⟨21⟩2s ± ⟨21⟩
2s−1
M(1s , 2s , 3± ) = −κ(p1 · ε±
3)
2
2s
− sκ(p 1 · ε3 ) 2s
⟨2|p3 ε±
3 |1⟩ .
m m
(6.11)
This corresponds to the spin-s Kerr amplitude for positive helicities given
⟨2|p3 ε+
3 |1⟩ = 0. However, as was the case in the minimally coupled gauge
theory, the negative helicity amplitudes are incorrect and are not related
to M(1s , 2s , 3+ ) by a parity transformation. Parity can be restored by
introducing non-minimal interactions to the Lagrangians.

6.2 Parity invariant non-minimal theories

Although this construction discriminates between left- and right-handed
fields, one can still use it to describe parity invariant theories. A concrete
example is given in ref. [23] where the covariant spin-1 Proca action
describing electroweak bosons is mapped to the chiral representation via
a non-linear field redefinition. This example is discussed in detail in
paper III and the resulting chiral Lagrangian
 
← → 1
(1)
L = ⟨Φ| |D|D| ⊗ iQ
|Φ⟩ − m2 ⟨Φ|Φ⟩ + O(Φ4 ) , (6.12)
1+ m 2 |F− |
reproduces the non-chiral physics of the interacting W bosons. The non-
local term should be interpreted as an infinite series of interactions in
the anti-self-dual strength |F− | := F− α β := 12 (σ µ σ̄ ν )α β Fµν . The first
non-minimal interaction is
 
← →
⟨Φ| |D|D| ⊗ |F− | |Φ⟩ := (Dµ Φα1 α2 )(σ µ σ̄ ν )α1 β1 (F− )α2 β2 (Dν Φβ1 β2 ) ,
(6.13)
where the covariant derivatives are contracted into the Pauli matrices
and the arrows indicate which massive field they act on. The notation
⊗ should be interpreted as a tensor product such that, for general spin,
← → ← →
|D|D| ⊗ |F− | ∼ |D|D|α1 β1 (F− )α2 β2 δαβ33 . . . δαβ2s
2s
is 2s-by-2s spinor matrix.

50
This spin-1 gauge theory example illustrates how additional non-minimal
interactions are necessary to restore parity at each order in the fields.
For general spin, the field redefinitions relating representations (s, s) and
(2s, 0) are not known. Instead we will choose to restore parity order by
order in the interactions.

6.2.1 Cubic order

Gauge theory
Working with general spin, we can restore parity at cubic order by adding
a generalisation of the non-minimal term added at spin-1 (6.13),
n←→ 2s−1
µ 2 iQ X
⊙k
o
L√Kerr = ⟨Dµ Φ|D Φ⟩ − m ⟨Φ|Φ⟩ + ⟨Φ| |D|D| ⊗ |F− | |Φ⟩ .
m2k
k=0
(6.14)
We use the notation ⊙ to indicate a symmetrised tensor product such
← → ← → ← →
that |D|D|k ⊗ ∼ |D|D|α1 (β1 . . . |D|D|αk βk ) . This non-minimal interac-
tion generates a vertex of the type
2s−1
X Q
Vnon-min (Φs1 Φs2 A3 ) =− ⟨2|p2 p1 |1⟩k ⟨2|p3 ϵ3 |1⟩⟨21⟩2s−k−1 .
m2s+k−1
k=0
(6.15)
Note that this interaction does not contribute for positive helicity pho-
tons since the spinor |p3 ε3 |α β := p3µ ε3ν (σ µ σ̄ ν )|α β satisfies the helicity
dependent identities
− β
√
|p3 ε+ β
3 |α = 0 and (p3 ε3 )α = 2 |3⟩α ⟨3|β . (6.16)
Therefore Vnon-min only corrects the negative helicity three-point ampli-
tude by
2s−1
X 1
Vnon-min (Φs1 Φs2 A3 ) = −Q⟨23⟩⟨31⟩ [21]k ⟨21⟩2s−k−1
(1,2,3− ) m2s
k=0
[21]2s − ⟨21⟩2s
= −Q⟨23⟩⟨31⟩ .
[21] − ⟨21⟩
(6.17)
Note that this spinor structure is intimately related to the polynomials
first introduced in section 3.2.1,
(2s) [21]2s − ⟨21⟩2s
P1 (ς3 , ς4 ) = . (6.18)
[21] − ⟨21⟩
(2s)
Where the polynomial P2 (ς3 , ς4 ) corresponds to the polynomial defined
in (3.38) but swapping ς1 , ς2 → ς3 , ς4 . This is the first suggestion that

51
the polynomials discussed in section 3.2.1 have a higher-spin origin given
their relation to interactions of the chiral Lagrangian.
In order to reconstruct the full negative helicity three-point amplitude
we employ the following three-point relation
√ √
⟨2|p3 ε−
3 |1⟩ = 2⟨23⟩⟨31⟩ = 2p1 · ε− 3 ([21] − ⟨21⟩) , (6.19)
such that
√ the Lagrangian in eq. (6.14) generates the correct negative-
helicity Kerr amplitudes
√ [21]2s
A√Kerr (1s, 2̄s, 3− ) = 2Q p1 · ε−
.3 (6.20)
m2s
The chiral Lagrangian in eq. (6.14) is not the most general parity
invariant cubic Lagrangian. In principle one could add cubic operators
of the form
2s  n←→
X ← → o
∆L3 = dk m−2k−4 ⟨Φ| |D|D|⊙k−1 ⊗|D|F+ |D| |Φ⟩
k=1 (6.21)
n←→ o 
2k−4s ⊙2s−k
+m ⟨Φ| |D|D| ⊗|F− | |Φ⟩

which depend on |F− | and the self-dual field strength |F+ | := F+α̇ β̇ :=
1 µ ν α̇
2 (σ̄ σ ) β̇ Fµν . The relative coefficient between the operators containing
anti- and self-dual field strengths is fixed by parity. The mass factors are
fixed by dimensional analysis and are a clear indication that the power-
counting for the interactions involving F− and F+ are inversely related
in the chiral Lagrangian.
The overall coefficients dk are free and are related to the 2s − 1 free
parameters in the general three-point amplitude (3.31) via the relation
ck>0 = dk+1 − dk , where dk>2s = 0. The first coefficient, c0 = 1, is fixed
by the minimal coupling. √
In the non-chiral construction, the Kerr amplitudes were generated
by cubic interactions with power counting of 2s − 1. In this case the
√
Kerr chiral Lagrangian discriminates between interactions with |F− |
which have power counting 2s − 1, and those involving |F+ | which are
constrained to vanish.

Gravity theory
In the gravitational case, the cubic order parity-invariant chiral theory
is
√ m2

1
LKerr = −g ⟨∇µ Φ|∇µ Φ⟩ − ⟨Φ|Φ⟩
2 2
2s−2 o  (6.22)
1 X 2s−k−1 n ← → ⊙k
− ⟨Φ| |∇|∇| ⊙ |R− | |Φ⟩ ,
4 m2k
k=0

52
where we introduce the chiral Riemann curvature spinor
1
R−α β γ δ := Rλ̂µ̂ ν̂ ρ̂ σ λ̂ αε̇ σ̄ µ̂,ε̇β σ ν̂ γ ζ̇ σ̄ ρ̂,ζ̇δ . (6.23)
4
At cubic order, this corresponds to including a non-minimal vertex
κ
VR− (Φs1 Φs2 A3 ) = − (⟨2|p3 ϵ3 |1⟩ − p3 · ϵ3 ⟨21⟩)2 ×
8
2s−2
X 2s−k−1
⟨21⟩2s−k−2 ⟨2|p2 p1 |1⟩k . (6.24)
m2k
k=0

The resummed on-shell vertex has the form

κ  ⟨21⟩2s− [21]2s 2s⟨21⟩2s−1 
VR− (Φs1 Φs2 h3 ) = ⟨2|p3 ε3 |1⟩2 −
(1,2,3) 8 (⟨21⟩ − [21])2 ⟨21⟩ − [21]
κ 2 (2s)
= ⟨2|p3 ε3 |1⟩ P3 (ς3 , ς3 , ς4 ) ,
8
(6.25)
(2s)
where P3 (ς3 , ς3 , ς4 ) corresponds to a three-variable polynomial, which
can be constructed as a limit. From the overall factor ⟨2|p3 ε3 |1⟩, it
is clear this vertex vanishes for positive-helicity gravitons, such that
M(1s , 2s , 3+ ) is unchanged.
In the negative helicity case the amplitude M(1s , 2s , 3− ) receives con-
tributions from the three vertices V√−g , V∇ and VR− . Making use of the
identity in eq. (6.19), we find that the second contribution from VR−
cancels exactly against the on-shell contribution of spin-connection V∇ .
Meanwhile the first term of VR− corrects the chiral V√−g term. The
result is the expected Kerr amplitude

[21]2s
MKerr (1s , 2s , 3− ) = κ(p1 · ε−
3)
2
. (6.26)
m2s
In principle one can add further cubic interactions proportional to
LKerr the anti-chiral Riemann tensor, R+ . This would be necessary to
describe a theory of a generic spin-s particle. However the theories cor-
responding to Kerr amplitudes do not require such operators at cubic
order.

6.2.2 Quartic order

We can extend the analysis of the chiral Lagrangian to quartic order.
In order to investigate the parity properties of the quartic Lagrangian
we will compute the four-point Compton amplitude A(1s , 2s , 3± , 4± ) in
electromagnetism. The gravity calculation will not be presented.

53
 In order to compute A(1s , 2s , 3± , 4± ) we need the massive propagator,

(β β )
1 iδα11 . . . δα2s
2s
2 2
= 2 2
. (6.27)
p −m p −m

Given the free chiral Lagrangian (6.1) is effectively a Lagrangian for a

free massive scalar, the massive propagator resembles a scalar propagator
with the SL(2, C) unit operator in the propagator numerator.

Quartic order gauge theory

As discussed previously, the Lagrangian (6.14) generates two cubic ver-
tices Vmin (Φs1 , Φs2 , A3 ), Vnon-min (Φs1 , Φs2 , A3 ) from the minimal coupling
and the non-minimal parity completion. At quartic order, the cubic
Lagrangian generates two quartic vertices,

⟨21⟩2s
Vmin (Φs1 , Φs2 , A3 , A4 ) = Q2 ϵ3 · ϵ4 ,
m2s
Vnon-min (Φs1 , Φs2 , A3 , A4 ) =
2s−1
⟨21⟩2s−k−1

X [2|ε4 |1⟩
− Q2 k k k


⟨2|p3 ϵ 3 |1⟩ [2|1+4|1⟩ −m [21]
m2s+k [2|4|1⟩
k=0

⟨2|ε4 |1] k k k


− ⟨2|1+3|1] −m [21]
⟨2|4|1]
+ (3 ↔ 4) + off-shell in Φ1 , Φ2 , (6.28)

where we only write explicitly the terms in Vnon-min valid for on-shell
fields Φ1 , Φ2 .
The four-point amplitudes have two massive channels Γt and Γu . In
electromagnetism there are no self-interactions of the massless field, so
there is no channel where a photon propagates. We construct the massive
channel Γt

1  ∂2 2s V (Φs , Φs , A4 )V (Φs , Φs , A3 )

1 p p 2
Γt (Φs1 , Φs2 , A+ +
3 , A4 ) =
(2s)!2 ∂ z̄pa ∂zp,a t14
(6.29)
where the differential operator acts on the massive spinors and such
that SL(2, C) indices are trivially contracted, as imposed by the massive
propagator in eq. (6.27). The other massive channel is related by Γu =
Γt |3↔4 .
Since the photons will always be on-shell, the non-minimal vertices
Vnon-min. will only contribute to amplitudes with at least one external
negative helicity photon. Therefore the all-plus amplitude only depends

54
on the interactions of the minimally coupled Lagrangian,
(
A(1s , 2s , 3+, 4+ ) = Γt (Φs1 , Φs2 , A+ + s s + +
3 , A4 ) + Γu (Φ1 , Φ2 , A3 , A4 )
)
+ Vmin (Φs1 , Φs2 , A+ +
3 , A4 )
(1,2,3+ ,4+ )

⟨21⟩2s
= Q2 (t14 p1 · ε+ + + +
3 p2 · ε4 + t13 t14 ε3 · ε4 ) + (3 ↔ 4)
m2s t13 t14
m2 [34]2 ⟨21⟩2s
= Q2 .
t13 t14 m2s
(6.30)
As expected the minimally coupled Lagrangian generates the all-plus
sector of the AHH amplitudes as shown in ref. [50]. Although we did the
explicit calculation, this is clear from the form of the minimally coupled
chiral Lagrangian. The minimally coupled terms in eq. (6.14) have trivial
spinor contractions such that the spin-dependence of the interactions can
be factored out. Therefore, in the all-plus helicity sector, the theory
is effectively the theory of minimally coupled scalars and the n-point
amplitudes
⟨21⟩2s
A(1s , 2s , 3+ , . . . n+ ) = A(10 , 20 , 3+ , . . . n+ ) (6.31)
m2s
factor into the scalar amplitudes A(10 , 20 , 3+ , . . . n+ ) and a spin-dependent
term. The factorisation property does not hold in any other helicity sec-
tor, given the non-minimal interactions in eq. (6.14) have non-trivial
spinor contractions.
The all-negative helicity amplitudes generated by L√Kerr do not cor-
√
respond to the Kerr amplitudes in eq. (3.33), indicating that the chiral
Lagrangian is not parity invariant at quartic order. In order to restore
parity, we need to introduce interactions of form ∝ |F− |2 . However, we
leave the identification of these operators to future work since we are
mostly concerned with the opposite-helicity amplitudes in this thesis.
Remarkably the opposite-helicity amplitudes A(1s , 2s , 3− , 4+ ) gener-
ated by the Lagrangian respect parity directly, without any need to add
additional quartic operators ∝ |F− ||F+ |. The construction of the ampli-
tude follows the approaches of the all-plus amplitudes but in this case
the non-minimal vertices Vnon-min will also contribute. We suppress the
details of the calculation and present the result

⟨3|1|4]2 (2s) ⟨13⟩⟨3|1|4][42] (2s)

A(1s, 2̄s, 3−, 4+ ) = P − P2
t13 t14 1 t13
(2s−1) (2s−1)

+ ⟨13⟩⟨32⟩[14][42] P2 − ς3 ς4 P 4 . (6.32)

55
This amplitude√matches the s = 0 and s = 1/2 AHH amplitudes in
√ (3.34) and Kerr amplitudes in eq. (3.36). The mismatch with the
eq.
Kerr corresponds to the contact term

C (s) := A√Kerr (1s, 2̄s, 3−, 4+ ) − A(1s, 2̄s, 3−, 4+ )

⟨13⟩⟨32⟩[14][42] 
(2s) (2s−2)
 (6.33)
=− (ς 3 + ς 4 ) P 4 − P 2
2m4
The missing contact term corresponds to a specific choice of additional
four-point interactions which have the form
2s−4
X 2s−3−l
X Q2 n
j,k,l
← → o
∆L4 = − ⟨Φ| |D | ⊙ |F − | ⊙ | D|F+ |D| |Φ⟩ ,
j=0
m6
k≤l=0
1 ← → ← → ← → ← →
Dj,k,l = |D|D|+m2 ⊙ |D|D|⊙j ⊙ |D|D+ |⊙k ⊙ |D+ |D|⊙(l−k) .


m2(j+l)
(6.34)
While C (s) , and in turn ∆L4 , is not fixed using solely higher-spin con-
straints, we motivate it by a series of classical constraints introduced in
Part III.

56
7. String theory

In the field theories considered previously, the fundamental object was a

point-particle, while in string theory it is an extended object, the string.
String theory provides a fully consistent theory of interacting higher-spin
states corresponding to vibration modes of the string.
The classical string is described by the Polyakov action [18, 31]
Z
1
S=− d2 σ∂a X µ ∂ a Xµ , (7.1)
4πα′
where the classical space-time coordinate X µ satisfies two additional
constraints. When quantising the bosonic open string, X µ is expanded
into oscillators αnµ corresponding to the creation (n < 0) and annihilation
(n > 0) operators. We can thus construct an arbitrary state in the Fock
µ
space by acting with α−n on the vacuum |0; p⟩
µ1 µs
α−n 1
. . . α−n s
|0; p⟩ . (7.2)

By the state-operator correspondence, we can map the string state to an

operator in the conformal field theory on the worldsheet

: ∂ n1 X µ1 . . . ∂ ns X µs eip·X : (7.3)

constructed from the conformal primary ∂X := ∂1 X + i∂2 X and it’s

descendants. Physical string states must satisfy the Virasoro constraints,
inherited from the constraints on X in the classical theory.
The leading Regge trajectory corresponds
P to taking states of maximum
spin s = n − 1 at a given level n = ni . In the bosonic open string,
this corresponds the states
(µ µ )
ϵµ1 ...µs α−11 . . . α−1
s
|0; p⟩ ←→ ϵµ1 ...µs : ∂X µ1 . . . ∂X µs eip·X : . (7.4)

The Virasoro constraints translate to the expected constraints on the

momentum p and polarisation tensors ϵµ1 ...µs = ϵµ1 . . . ϵµs

p2 = (s − 1)/α′ , pν ϵνµ2 ...µs = 0 , η ν1 ν2 ϵν1 ν2 µ3 ...µs = 0 . (7.5)

Therefore the level-n state in the leading trajectory has the same degrees
of freedom of a spin-s particle, where s = n − 1. Therefore we can
interpret the scattering of string states in the leading Regge trajectory

57
as the scattering of higher-spin particles with non-minimal interactions
that are preselected by the string theory. √
A natural question to ask is whether the Kerr and Kerr three-point
amplitudes have a string theory origin. The quantum three-point am-
plitudes for the scattering of leading Regge states in the bosonic string
and the superstring
√ are known [57, 53, 54] and do not correspond to the
quantum Kerr and Kerr amplitudes, as shown in paper I . Further-
more, we confirmed that the classical
√ limit of the superstring amplitudes
do not generate the classical Kerr and Kerr amplitudes.
However there are also candidate spin-s states lying in the sublead-
ing trajectories. In principle these states can be constructed either in
light-cone gauge [37, 36] or using the Del Giudice - Di Vecchia - Fubini
construction [29, 21], however the analysis of the scattering amplitudes is
complicated by the explicit dependence on a reference momentum. A co-
variant approach was recently introduced in ref. [47], which can construct
whole trajectories at a time. We leave the study of such amplitudes and
their classical limits to future work.
In the bosonic theory, the lowest spin state state in the leading trajec-
tory has m2 < 0. In order to avoid this tachyonic state we studied the
scattering of states in the leading Regge trajectory of the superstring.

Leading Regge states in the open superstring

Introducing supersymmetry removes the tachyon state and also lowers
the critical string dimension to d = 10. In principle, there are several
variations of the superstring corresponding to how the fermions are in-
cluded. In this work we only consider the scattering of bosonic states
which remain unchanged.
In the supersymmetric theory, a spin-s state on the leading trajectory
is
(µ µ s−1 µ )
α−11 . . . α−1 s
ψ−1/2 |0; p⟩ , (7.6)

µ
where ψ−1/2 is a worldsheet fermion. The corresponding vertex operator1
is
(−1) 1
Vs−1 (ϵ, k) = √ : (ϵ · i∂X)s−1 (ϵ · ψ)e−ϕ eik·X : (7.7)
( 2α′ )s−1

where the operator carries a superghost charge −1 [54, 17]. Since tree-
level superstring amplitudes must carry overall ghost charge −2, we also

1
We have suppressed the dependence on the Chan-Paton factor T .

58
need to introduce the zero ghost charge vertex operator,
(0) 1
: (ϵ · i∂X)s−2 (s − 1)ϵ·∂ψ ϵ·ψ+k·ψ ϵ·ψ ϵ·i∂X

Vs−1 (ϵ, k) = √
( 2α′ )s−2
1
+ ′ (ϵ·i∂X)2 eik·X : .

2α
(7.8)
The polarisation tensors corresponding to the level-(s − 1) state is
totally symmetric such that ϵµ1 ...µs = ϵµ1 . . . ϵµs . As in the bosonic
string, ϵµ1 ...µs is constrained by BRST invariance to be transverse and
traceless [17], such that it corresponds to the physical polarisation tensor
for a spin-s particle. In the open string, the states on the leading Regge
trajectory satisfy open string α′ = (s − 1)/m2 , such that the lowest spin
state is massless vector and corresponds to the photon.
The three-point amplitude between two spin-s states and the photon
can be computed from the correlation function
(−1) (−1) (0)
A(1s , 2s , 3) = ⟨cVs−1 (ϵ1 , p1 )cVs−1 (ϵ2 , p2 )cV0 (ϵ3 , k3 )⟩ , (7.9)

and was done in ref. [54]. While the vertex operators are defined in
d = 10, we can trivially compactify down to d = 4 via ϵi = (εi , 0) where
εi are the 4d massive polarisations for the massive legs i = 1, 2. Likewise
we compactify the massless polarisation ϵ3 = (ε3 , 0) and the momenta
pd=10
i = (pi , 0), k3d=10 = (k3 , 0). In four-dimensions we can present the
resulting amplitude in massive spinor variables,

AL.R. (1s , 2s , 3+ ) =
s X s s−n−1
√ X α′
2 Qp1 · ε+ 3 cs,n,l 2n+2
⟨12⟩l+n (⟨12⟩ − [12])2s−l−n , (7.10)
n=0
m
l=0

where the coefficients are

(s − 1)(s − 1)!(s − n − 1)
cs,n,l = − (s l − n(n − 1)) . (7.11)
l!(n − l)![(s − n + 1)!]2
The covariant formula is presented in paper I . Note that taking the
field theory limit, α′ → 0, does not generate A√Kerr amplitudes. We do
find an instance when AL.R. = A√Kerr , corresponding to the scattering of
spin-2 states, where the identity only holds after substituting the relation
m2 α′ = s − 1.
However beyond spin-2, this equality does not hold. Therefore the
corresponding effective chiral Lagrangian in eq. (6.14) thus requires ad-
ditional cubic interactions ∆L3 with coefficients that are explicitly de-
pendent on s. From Ward identities, such amplitudes satisfy (WI) and
(PC) however, they are ruled out by the current constraint (CC). Given

59
the amplitudes are unique from the Lagrangian level approach at power
counting s + s′ + 1 it suggests that the EFT for the leading Regge su-
perstring must have slightly loosened power counting constraints, for
example the power counting on δ1 could be increased or else the grading
of the power counting when looking at the unphysical interactions could
be tailored.

Closed string three-point amplitude

Having studied the open string amplitude, we can use the Kawai-Lewellen-
Tye relations [45] to generate the closed string amplitude,
 2
κ
M(1s , 2s , 3) = A(1s , 2s , 3) ′ ′  . (7.12)
2 α →α /4
Q →1

We will not present an explicit formula, however we note that for s = 4

the amplitude coincides with the Kerr amplitude. As in the opens string
case, this matching no longer holds for spins s > 4.

60
Part III:
Constructing Classical Amplitudes
8. Scattering amplitudes from classical
physics

Scattering amplitudes are typically not direct observables, instead they

are stepping stones towards predictions, also for classical physics. In
section 2, it was shown that the the two-to-two scattering amplitude is
an intermediate step in the computation of the effective potential that
governs the conservative dynamics the binary black hole system.
The Compton amplitudes considered in this thesis do have a corre-
sponding well-defined classical process. In general, one can consider am-
plitudes of the form M(1, 2, 3, . . . n), which involve two massive states
that, in the classical limit, represent the classical source and pertur-
bations are represented by the gravitons. The classical amplitudes are
direct limits of the ones computed using field theory methods. The classi-
cal solutions we are interested in are characterised by a mass parameter,
m, and a spin vector aµ , in particular, the perturbations to the Kerr
solution.
The three-point gravitational amplitude can be computed from the
effective energy-momentum tensor T µν , which sources the classical solu-
tion. The four-point amplitudes correspond to the scattering of a gravi-
tational wave in the classical background solution, as was done explicitly
in ref. [8, 9] for the Kerr solution.
The classical amplitudes in electromagnetism have an analogous defi-
nition. The three-point amplitudes can be computed from to the effective
current j µ , while the four-point amplitude corresponds to scattering of
a single electromagnetic wave off of the classical background.

8.1 Classical amplitudes at three points

We consider a classical solution in electromagnetsm or general relativity
sourced by an effective current j µ or energy-momentum tensor T µν at
linear order in the coupling. The resulting linearised equations of motion
are □Aµ ∼ j µ in electromagnetism and □hµν ∼ Tµν − ηµν T /2 in gravity,
using harmonic gauge. The interactions between the source and the field
are
Aµ j µ , hµν T µν . (8.1)

63
The classical three-point amplitudes are obtained by evaluating these
interactions on-shell,

Acl = εµ j̃ µ , Mcl = εµ εν T̃ µν . (8.2)

Where the sources are Fourier transformed and on-shell pollarisations

introduced.
The amplitudes can be expressed using momenta pµ , q µ , spin aµ and
the massless polarisation εµq . On the three-point kinematics there are two
independent Lorentz contractions p · εq and q · a such that the general
amplitudes have the form,
∞ ∞
X
n κ 2
X
Acl = Q p1 · ε3 cn (q · a) , Mcl = (p1 · ε3 ) cn (q · a)n .
n=0
2 n=0
(8.3)
The spin-multipole coefficients cn describe the nature of the source.

Kerr black hole

Ref. [59] computed the linearised energy-momentum tensor from the Kerr
metric,
Z
µν µν
ν)
T̃Kerr (q) = d4 xeiq·x TKerr (x) = δ(p · q)p(µ ea∗q ρ pρ1 , (8.4)

where (a ∗ q)µ ν = iϵµ νρσ aρ q σ . This exponential structure is reflected in

the three-point amplitudes

MKerr = κ(p · ε± )2 e±q·a , (8.5)

which were originally computed in refs. [39, 26].

√
Electromagnetic Kerr √ solution
The interpretation of the Kerr electromagnetic solution is not properly
understood, although at leading order in Q it is known to correspond to
the electromagnetic part of the Kerr-Newman solution, a charged and
rotating black hole [25]. The authors in ref. [25] derive the effective
µ
current j√ Kerr
(x), such that
Z
µ µ
j̃√ Kerr
(q) = d4 xeiq·x j√ Kerr
(x)
(8.6)
 sinh(q · a) 
= Q δ(p · q) pµ cosh(q · a) + (a ∗ q)µ ν pν1 .
(q · a)
The resulting classical amplitude is

A√Kerr = Q p · ε± e±q·a . (8.7)

64
This amplitude can also be constructed from a Newman-Janis shift of
the Coloumb solution, such that the effective source is a rotating disk
with a certain charge distribution [5]. √
Note that the spin-multipole coefficients of the Kerr and Kerr am-
plitudes are both cn = 1/(n!). The amplitudes also satisfy a classical
double copy relation

MKerr = A√Kerr × A√Kerr . (8.8)

a → a/2
Q2 → κ
2

Classical string solutions

We can construct classical amplitudes for various other types of sources.
For example consider the solution to the Maxwell equations correspond-
ing to a rotating rigid rod of radius r with a charged Q at one end point,
as shown in figure 8.1. The explicit solution is given in detail in paper I
such that the corresponding current is

µ Q
jrod = δ(ρ − r)δ(ϕ − τ /r)δ(z) nµ (8.9)
r
given in polar coordinates (ρ, ϕ) such that the rod is rotating in the
xy -plane, nµ := (1, ϕ̂) = (1, − sin(τ /a), cos(τ /a), 0).
This configuration is of interest as it is a solution of the classical
string action (7.1). In particular, we expect these classical solutions
to be related to the leading Regge trajectory given that the angular
momentum is quadratic in total energy

r r2
E= , J= =⇒ J = α′ E 2 . (8.10)
2α′ 4α′
This corresponds to the large spin limit of the defining relation n = s − 1
for the open string states on the p leading Regge trajectory where the en-
ergy is the rest mass E = m = n/α′ , see section 7. The orbital angular
momentum of the charged rotating rod J is identified with the spin of
the effective point particle, such that aµ = (0, 0, 0, J/E) = (0, 0, 0, r/2)
in the rest frame where the spin is aligned with the z -axis. The corre-
sponding solution for the leading Regge closed string states is a rotating
rigid massive folded loop.
µ
In paper I, we compute the effective current jstr. and energy-momentum
µν
tensor Tstr. explicitly for the open and closed string solutions. This allows
us to compute the full classical amplitudes,

Acl string = (p · ε+ ) I0 (2q · a) + I1 (2q · a)

 2 (8.11)
Mcl string = (p · ε+ )2 I0 (q · a) + I1 (q · a) .

65
Figure 8.1. Rigid open string of length 2r with a charge Q at one endpoint
rotating around its centre point with angular velocity 1/r.

√
As opposed to the simple exponential structure of the Kerr and Kerr
amplitudes, the spin-multipoles for the classical string solutions resum
to two modified Bessel functions of the first kind, defined by the series
expansions
∞
 ∞

x2k x2k+1
I0 (2x) = , I1 (2x) = . (8.12)
(k!)2 k!(k + 1)!
k=0 k=0

The classical amplitudes Acl string and Mcl√string are related by the same
classical double copy relation (8.8) as the Kerr and Kerr amplitudes.
We expect the classical amplitudes Acl. strings and A√Kerr to differ
significantly given the effective currents that source the static solutions
differ, the former corresponding to a single point charge rotating orbit
while the latter corresponds to a rotating disk of charge [5].

8.2 Classical Compton amplitudes

At four point, the classical Compton amplitudes correspond to the lin-
earised scattering of plane waves off an exact background sourced by the
classical solution. This is a tractable computation for Kerr backgrounds,
given the linearised scattering is described by the separable Teukolsky

66
equation. The corresponding classical Compton amplitudes have been
computed up to O(a8 ) in refs. [9, 7].
For a general background, the equations of motion for linearised per-
turbations√may not be separable. As of yet, the classical Compton ampli-
tudes for Kerr have not been computed from linearised perturbations √
of the Kerr-Newman solutions. Furthermore, it is not clear if Kerr
beyond three-point is still related to Kerr-Newman.
√ Therefore there is
less classical data to compare our proposed Kerr amplitudes to.
The classical amplitude M(1, 2, 3, 4) corresponds to the scattering of
an incoming wave with momenta k3µ and the incoming massive black hole
pµ1 to the outgoing outgoing wave with momenta −k4 and the deflected
massive black hole −pµ2 = pµ1 + k3µ + k4µ .
As discussed in section 2, the classical regime corresponds to the low
frequency regime of the waves, such that k3 , k4 are soft. This corresponds
to a small deflection of the black hole. In practice we will scale the
massless momenta with ℏ

pµ := pµ1 ∼ 1 , kµ ∼ ℏ , (8.13)

such that the classical limit corresponds to taking ℏ → 0 in the final

amplitude. At three points the quantum kinematics are essentially clas-
sical given all the momentum invariants vanish. Although the ampli-
tude does not necessarily physical given we use complexified momenta
at three-points.
At four points, there are non-trivial momentum invariants such that
we will need to take an explicit ℏ → 0 limit on the amplitudes. The
relevant soft momenta for the opposite-helicity amplitudes are q µ :=
µ
(k3 + k4 )µ , q⊥ := (k4 − k3 )µ and χµ := ⟨3|σ µ |4]. Thus, the four momen-
tum invariants that parameterise the scalar opposite-helicity amplitude
scale as

p2 = m2 ∼ 1 , 2
q⊥ = 2p·q = −q 2 ∼ ℏ2 , p·q⊥ ∼ ℏ , p·χ ∼ ℏ . (8.14)

Given the Compton amplitudes are defined with all-incoming states, the
opposite-helicity amplitude corresponds to the helicity-conserving clas-
sical process, where both the incoming wave k3 and outgoing wave −k4
have negative helicity.
One can consider non-spinning backgrounds, corresponding to the
Coulomb potential in electromagnetism and the Schwarzschild black hole
in general relativity. As mentioned in section 2, the dynamics of these so-
lutions are modelled by a minimally coupled massive scalar. The scalar,
opposite-helicity, classical Compton amplitudes thus correspond to

(p · χ)2
2
 κ 2 (p · χ)4
A0 = Q , M0 = . (8.15)
(p · q⊥ )2 2 q 2 (p · q⊥ )2

67
 When we consider the scattering of backgrounds with spin, the general
classical amplitude is expressed as a multipole expansion in the classical
spin variable aµ . For the four point helicity-conserving scattering, this
introduces four new spin-dependent variables
p · q⊥ a · χ p · q⊥
x = a · q⊥ , y = a·q, z = |a| , w= . (8.16)
m p·χ

These variables are not independent and are related by a Gram deter-
minant
m2 q 2 (w − x)2 − y 2
ξ −1 := = , (8.17)
(p · q⊥ )2 z 2 − w2
where ξ is the optical parameter [9]. In order for these variables to scale
classically, the spin vector must scale aµ ∼ ℏ−1 .
Extracting these spin-dependent variables from the quantum ampli-
tudes can be subtle and is discussed in detail in section 9. In this sec-
tion, the classical amplitudes are constructed as covariant ansätze that
are matched to classical computations. The form of the general classical
four-point amplitude is discussed in detail in paper III. Here we note
that the classical non-spinning amplitudes can be factored out such that

A(1, 2, 3− , 4+ ) = A0 f (x, y, z, w) ,
(8.18)
M(1, 2, 3− , 4+ ) = M0 f (x, y, z, w) .

The classical amplitudes are invariant under exchange and complex-

conjugation of the massless bosons, implying f is an even function in
variable y . Likewise, if the classical amplitude is symmetric under ex-
change of the massive leg, f is even in z . This is symmetry is broken
for dissipative effects as the incoming and outgoing legs are no longer
related by time-reversal symmetry. Such dissipative effects appear in
the classical amplitudes MTeuk calculated in ref. [9], which we discuss in
detail.

Scattering off a Kerr black hole

The classical amplitude MTeuk was computed up to O(a6 ) by matching
solutions of the Teukolsky equation to an covariant ansatz [9]. The
amplitude will be used both as input and as a check for the classical
amplitude MKerr proposed in paper IV . We present MTeuk in the form
h
MTeuk = M0 ex−w + P(x, y, z, w) + Pη (x, y, z, w)
z2 wz i
+α Pα (x, y, z, w) + ηα Pαη (x, y, z, w) + O(a7 ) ,
ξ ξ
(8.19)

68
where the functions P, Pη , Pα , Pαη are polynomials that are even in vari-
ables y and z . The parameters α and η are bookkeeping parameters that
appear in ref. [9].
The η terms are associated to near-zone effects given they are sensitive
to the boundary conditions at the horizon. In MTeuk such dissipative
terms are odd in variable z ∝ |a|, such that the exchange symmetry on
the massive legs is broken.
The α tag was introduced to keep track of terms that arose from the
expansion of polygamma functions when taking the super-extremal limit
|a| >> Gm. This super-extremal limit was taken in order to identify
the tree-level O(G) Compton dynamics from the Teukolsky solutions,
which are generically Laurent series in G. However there are alternative
definitions of the O(G) physics, for example ref. [7] defines MTeuk after
implementing a near- and far-zone splitting of the Teukolsky solutions,
such that the super-extremal limit is not necessary. Since it is not clear
if this splitting procedure is a unique, there are ambiguities in how to
define the classical amplitude MTeuk .
Nonetheless, in paper IV we present an all-orders-in-spin classical am-
plitude MKerr that predicts the α-independent terms of MTeuk defined
in ref. [9]. The classical amplitude MKerr is generated by the classical
limit of the quantum amplitude MKerr presented in eq. (3.39). There-
fore, before we can present our classical amplitude we have to introduce
how to extract classical physics from the quantum amplitudes

69
 √
9. Classical Kerr and Kerr Amplitudes

√
The goal of this section is to present the classical Kerr and Kerr Comp-
ton amplitudes proposed in papers III and IV. These classical amplitudes
are generated as classical limits of the quantum amplitudes proposed in
section 3.2.1. Therefore, we start by discussing the consistent classical
limits for extracting classical spin variables from the quantum scattering
amplitude.

9.1 Classical limits for spin

Extracting the dependence on classical spin aµ from the quantum scatter-
ing amplitudes introduces subtleties. We will discuss three prescriptions
for obtaining classical amplitudes from quantum scattering amplitudes
of spinning particles,
(i) finite spin, ℏ → 0 limit of a fixed spin amplitude
(ii) large quantum spin, take s → ∞ of the generic spin s amplitude
(iii) coherent spin states as introduced in ref. [3] .
In each case we will make use of the spin-operator basis introduced in
section 4. If we return to the schema,

⟨12⟩2s
MKerr ∼ −−−−−−−−−→ ⟨eâ·p3 ⟩ −→ ea·p3 ∼ T µν , (9.1)
m2s spin op. basis cl. limit

the operation −→ corresponds to implementing one of the three limits

cl. limit
(i), (ii), (iii).
As we will discover, (i) is not, strictly speaking, a valid classical limit
and suffers from ambiguities in defining the classical amplitude. In com-
parison, (ii) and (iii) prove to be more robust approaches for extracting
classical amplitudes with classical spin.

9.1.1 (i) finite spin classical limit

The guiding principle of the finite-spin approach is that one can gener-
ate the first 2s classical multipole orders of a classical solution from the
scattering of a quantum spin-s particle [58]. This suggests there exists a
(s)
map from some suitably defined quantum spin multipoles, e.g. c̃µ1 ...µi in

70
eq. (4.25) to the classical spin multipoles cµ1 ...µi . However, as discussed
in section 4, the expansion of the quantum amplitude into the spin op-
erator basis is ambiguous such that the quantum multipole tensors are
not unique (4.32).
Furthermore, in order for the classical amplitude to be well defined,
we expect the result to be the same whether we scatter spin-s or spin-
s′ states in the quantum theory. This implies the quantum amplitudes
A(1s , 2s , 3h3 , . . . , nhn ) have to exhibit a particular property,

lim c̃(s=1) (s=2)

µ1 ...µi = lim c̃µ1 ...µi = · · · = cµ1 ···µk , (9.2)
ℏ→0 ℏ→0

which we call spin-universality. This is a restrictive property on the

family of spin-s quantum amplitudes and is not true for generic theories.
Neglecting these ambiguities for now, we can implement this limit by,
first expanding the quantum amplitude in the spin-s spin operator basis,
as discussed in section 4 and then scaling the kinematics according to
eq. (8.14).
However, we must scale the spin operators as â ∼ O(ℏ−1 ), in order for
the spin-dependence to survive as ℏ → 0. This scaling is suggested by
the correspondence principle where the macroscopic spin emerges from
a large quantum spin limit. We will identify the expectation values
⟨â(µ1 . . . âµk ) ⟩ with the product of classical ring radius aµ1 . . . aµk .

Three points
Let us study spin-universality in more √ detail at the three point level.
We can study the classical limit of the Kerr amplitudes in eq. (3.20)
and the open string amplitudes in eq. (7.10). At three points, the tensor
(s) (s)
structure is unique such that c̃µ1 ...µk = ck q(µ1 . . . qµk ) .
Using the identities
√ in section 4, we can write the quantum general-
spin three-point Kerr amplitudes as

2s
s s +
X 1
A(1 , 2 , q ) = A0 ⟨(q · â)k ⟩ . (9.3)
k!
k=0

These amplitudes exhibit spin universality given the spin-multipole co-

(s)
efficients ck = 1/k! are independent of the quantum number s. Note
that the Kerr three-point amplitudes are also spin-universal since the
spin-structure is identical.
However the open-string amplitudes, As := Aopen (1s , 2s , q + ) defined
in eq. (7.10), do not exhibit this property. The classical amplitudes
defined by the scattering of the first few massive states in on the leading

71
Regge trajectory are
1 1 1
lim A2 /A0 =1+(q · a)+ (q · a)2 + (q · a)3 + (q · a)4
ℏ→0 2 6 24
9 3 13
lim A3 /A0 =1+(q · a)+ (q · a)2 + (q · a)3 + (q · a)4 + O(a5 )
ℏ→0 30 10 120
5 5 39
lim A4 /A0 =1+(q · a)+ (q · a)2 + (q · a)3 + (q · a)4 + O(a5 ) .
ℏ→0 7 14 280
(9.4)
Clearly scattering spin-2, spin-3 or spin-4 states define three different
classical amplitudes, none of which reproduce the classical amplitude
Acl string , computed from classical field theory methods in eq. (8.11).
(s) (s)
While the monopole and dipole coefficients are universal c0 = c1 =
1, the higher multipole coefficients depend explicitly on the spin of the
scattered state, for example

(s) 4s2 −7s + 4 (s) 2s − 3

c2 = , c3 = . (9.5)
2s(2s − 1) 2(2s − 1)
Given the amplitudes do not exhibit spin universality it is ambiguous
how to define the classical amplitude. Indeed, for the leading Regge am-
plitudes, the s → ∞ limit is required to match classical multipoles after
(s)
taking s → ∞, such that lims→∞ c2 = 14 is the quadrupolar coefficient
in the expansion of the classical amplitude Acl. string in eq. (8.11).

Same-helicity Compton amplitudes for Kerr

Given the three-point Kerr amplitudes exhibit spin universality, we should
investigate whether this property holds at four-points. Let us study the
same-helicity amplitudes in eq. (3.33) given we expect them to be the
correct Kerr amplitudes.
These four-point amplitudes have the same spinorial structure as the
three-points amplitude, which we expand into ā variables in eq. (4.23).
After changing representation using eq. (4.29), the resulting amplitude
is 1 
M(1s, 2̄s, 3+, 4+ ) = M0 ⟨eâ·q ⟩ + O q 2 â2 + O(ℏ) . (9.6)
s
As expected, the same-helicity amplitude has the same exponential struc-
ture as the three-point amplitude plus a correction term proportional to
q 2 â2 . At finite spin there is an ambiguity on how such terms scale due
to the Casimir identity,
q 2 â2 = −s(s + 1)q 2 , (9.7)
which suggests that the term can either scale classically as q 2 â2 ∼ 1,
or be suppressed, given s remains finite q 2 s2 ∼ ℏ−2 . Note that this

72
ambiguity disappear if we allow the spin quantum number to scale as
s ∼ ℏ−1 .
We can choose a prescription that restores spin-universality in the
same-helicity amplitude. This corresponds to always implementing â2 =
−s(s+1)1. Using this prescription we can generate a consistent classical
amplitude
Mcl (1 , 2 , 3+ , 4+ ) = M0 eq·a . (9.8)

Four-point opposite-helicity
The prescription used in the same-helicity amplitude also restores spin-
universality of the opposite-helicity AHH Compton amplitudes in eq. (3.34).
We expand the spin-dependent term into spin-operator variables,
 2s
⟨13⟩[42] + ⟨23⟩[41]
= (1+w̄−x̄)2s +O(ℏ) = ⟨ew−x ⟩+O(q 2 â2 )+O(ℏ) ,
⟨3|1|4]
(9.9)
where the variables w̄ and x̄ defined in terms of ā. In the second line
we change representation to the spin-s operators, picking up a correction
term. However, if we substitute the Casimir identity, this correction is
suppressed when ℏ → 0.
Implementing this substitution universally does not generate the cor-
rect classical physics for generic amplitudes. To illustrate this, we can
(2s) (2s)
study the first two terms of A√Kerr in eq. (3.36), P1 + t14 ρ·χ
p·χ P2 . We
can expand the combination in terms of the spin-1/2 spin operators,
s  
(2s) ρ · χ (2s) X 2s
P1 + t14 P ≈ (1 + q⊥ · ā)2s−2n ((p · q⊥ )2 ā2 )n
p·χ 2 n=0
2n
(9.10)
where ≈ implies we have dropped quantum corrections O(ℏ). In this
case the Casimir appears due to the expansion of the spinor variables
(4.21). However it generates the same ambiguities as we need to decide
whether to keep or drop such terms. We can either use the Casimir to
remove such terms completely such that the term contributes classically
as
(2s) ρ · χ (2s)
P1 + t14 P −→ ex , (9.11)
p · χ 2 ℏ→0
or choose to keep such ā2 terms,
(2s) ρ · χ (2s)
P1 + t14 P −→ ex cosh z . (9.12)
p · χ 2 ℏ→0
Therefore clearly the classical multipoles are dependent on the prescrip-
tion for the â2 terms. A priori the correct prescription is not clear,
however we check it against classical computations and consistency with

73
other classical limits, such that we find that the correct prescription for
this term is given in eq. (9.12).
In principle we find a consistent prescription for the finite spin classical
limit is to use the Casimir identity to neglect O(â2 ) corrections from the
representation change (4.27), this is required by keeping the exponential
form of the classical same helicity amplitude. However we find that we
should not reduce any ā2 terms generated by the spinor expansions, in
order to be consistent with limits (ii) and (iii) where we have better
control of such contributions.
We emphasise that this ambiguity arises since we keep s finite. The
classical limits (ii) and (iii) provide a robust treatment of â2 .

9.1.2 (ii) Infinite spin classical limit

As indicated by the name, this prescription involves taking explicit s →
∞ limits of the quantum amplitude. Therefore, in this case, it is mean-
ingless to speak of the classical limit of a spin-1 amplitude. Instead we
have to work with general spin-s amplitudes, for example the Kerr three
point amplitudes (2.10). Indeed, the original classical analysis of the
Kerr amplitudes did include taking a large spin limit [39, 26].
The large spin limit can be implemented by scaling s ∼ ℏ−1 , which
diverges when we take ℏ → 0. From the definition of ⟨aˆµ ⟩ (4.6) clearly
the classically finite spin-dependent variables are

⟨q · â⟩ ∼ 1 , ⟨q⊥ · â⟩ ∼ 1 , ⟨χ · â⟩ ∼ 1 . (9.13)

In the classical limit, we can make a consistent identification between ⟨â⟩

and the classical ring radius a given the variance of the spin operators
vanishes
Yn Y n
µi ℏ→0
⟨ â ⟩ − ⟨âµi ⟩ −→ 0 . (9.14)
i=1 i=1

In practice we implement this classical limit in two stages; a ℏ → 0 limit

when the amplitude is expressed in spin-1/2 operators āµ , followed by a
s → ∞ limit after we convert to the spin-s operators âµ .
When taking the dual limit we must impose ℏs ∼ O(1) by hand.
Therefore we have to carefully to examine subleading orders in the ℏ
expansion as they can ultimately contribute classically if there are com-
pensating s factors.
The classical spin-multipoles are therefore defined after taking the
s → ∞ limit. This was confirmed in paper I by studying the amplitudes
generated by the scattering of open and closed superstring states. As
discussed in the finite spin analysis, the amplitudes Aopen and Mopen
are not spin-universal. Only in the s → ∞ limit do we generate the

74
correct classical spin multipoles,
(
1
(k!)2 if n = 2k with k ∈ N,
open string: cn = 1 (9.15)
k!(k+1)! if n = 2k + 1 with k ∈ N.
(2k+1)!
(
4k (k+1)!(k!)3
if n = 2k with k ∈ N,
closed string: cn = (2k+1)! (9.16)
4k ((k+1)!)2 (k!)2
if n = 2k + 1 with k ∈ N.

By inserting these multipole coefficients into eq. (8.3) and resumming,

we recover the modified Bessel functions in the classical amplitudes in
eq. (8.11) computed from classical field theory methods.

Quartic same-helicity amplitude revisited

As we saw in the finite spin case, the quartic same-helicity amplitude is
sensitive to corrections O(q 2 â2 ) generated by the representation change
(4.27). We can no longer suppress such terms using the Casimir identity
as s is no longer finite. However, for all the amplitudes considered in
this thesis, the combinatorics of such terms collude such that they are
suppressed when s → ∞. For example, consider the quadrupolar term
generated by ⟨21⟩2s = (1 + q · ā)2s + O(ℏ). Changing the representation
from spin-1/2 to spin-s operators generates the â2 correction

s ⟨â2 ⟩
 
2s 1 
(q · ā)2 = ⟨(q · â)2 ⟩ − P µν qµ qν ,
2 2! 4 s(s + 1) (9.17)
1 2 −1
−→ ⟨(q · â) ⟩ + O(s ) .
s→∞ 2!

However in the s → ∞ limit this correction is suppressed. In paper III

, we investigated up to dotriacontapole O(a5 ) and found that all con-
tributions from the representation change are suppressed. This confirms
that the subleading terms generated by the representation change of the
spin-operator are not classically relevant.

9.1.3 (iii) Coherent state classical limit

As opposed to scattering two massive particles of massive spin, the au-
thors in ref. [3] consider the scattering of coherent spin-states, which we
schematically denote as
∞
2 X 1
|coherent⟩ = e−|z| /2
p |s, {z}⟩ . (9.18)
s=0 (2s)!

Ref. [3] follows the Schwinger construction, where the spin eigenstates
|s, {z}⟩ ∼ (z a a†a )⊙2s |0⟩ are constructed from the repeated action of the

75
creation operator (a†a ) on the scalar state |0⟩. Once again we absorb the
little group SU(2) indices with the wavefunctions z .
The spin-operator in the Schwinger construction is defined (âµ )a b :=
1 † µ b b
2m aa σ a , where a is the annihilation operator. While we are overload-
ing the notation â, however we are permitted to do so given the expecta-
tion of this operator with respect to the spin-eigenstates matches exactly
the definition of ⟨â⟩ in eq. (4.6),

⟨s, {z}|(âµ )a b |s, {z}⟩ ≡ ⟨âµ ⟩ = sāµ (z̄ a za )2s−1 , (9.19)

where we note that â can be constructed from the massive spinors (4.20).
These details are necessary to derive the expectation of the spin-operator
with respect to the coherent states,
∞
2 X 1
⟨coherent|âµ |coherent⟩ = e−|z| ⟨âµ ⟩ = āµ . (9.20)
s=0
(2s)!

We scale the wave functions za , z̄a ∼ O(ℏ−1/2 ), such that the variance of
the operators satisfy a similar relation as in the large spin limit (9.14).
This once again allows us to consistently identify the expectation values
⟨coherent|âµ |coherent⟩ with the classical spin variable aµ .
We will define the classical amplitude as the scattering of two coherent
massive states. This corresponds to taking a weighted infinite sum of the
general spin-s amplitudes,
∞
2 X 1
A(1, 2̄, 3±, 4± ) = lim e−|z| A(1s, 2̄s, 3±, 4± ) . (9.21)
ℏ→0
s=0
(2s)!

In practice, we first resum the general spin amplitudes and then expand
the result in spin-1/2 variables. Given we will take an ℏ → 0 limit we
can identify āµ1 . . . āµn ≈ ⟨coherent|âµ1 . . . âµn |coherent⟩ up to irrelevant
ℏ corrections.
For example the classical limit of the same-helicity quartic amplitude
is almost trivial
∞
2 X 1 m2 [34]2  ⟨21⟩ 2s
A(1, 2̄, 3+, 4+ ) = lim e−|z|
ℏ→0
s=0
(2s)! t13 t14 m
(9.22)
2 2
m [34] −|z|2 +⟨21⟩/m m2 [34]2 q·a
= lim e = e .
ℏ→0 t13 t14 (p · q⊥ )2
In the first line, the amplitudes are resummed and, in the second, ex-
panded in spin-1/2 variables in the second keeping only leading order in
ℏ contributions. There is no ambiguity how to treat â2 terms nor do we
need to implement any representation changes.

76
 Note that the coherent state approach requires knowledge of the scat-
tering of general-spin particles. For generic quantum amplitudes, these
infinite sums could prove intractable. However, for the amplitudes stud-
ied in this paper, we the resummation is trivial. Indeed the amplitudes
proposed in paper III and paper IV have relatively simple dependence
(k)
on the spin number s encoded by the polynomials Pn , which resum to
rational functions of exponentials. For example,
∞ ∞
X 1 (k) X 1 (k) eς1
P = eς1 , P2 = + (ς1 ↔ ς2 ) . (9.23)
k! 1 k! ς1 − ς2
k=0 k=0

√
9.2 Classical constraints for Kerr and Kerr
amplitudes
In this section
√ we will present our candidate classical Compton ampli-
tudes for Kerr and Kerr, which are generated by taking classical limits
of the quantum amplitudes presented in section 3.2.1. We use the co-
herent state classical limit (iii) for ease however, the resulting classical
amplitudes are equivalent to those generated by the infinite-spin limit
(ii) as shown in paper III .
We will also discuss the combination of higher-spin and classical con-
straints used to constrain the contact terms. However, we will first intro-
duce the classical functions that will appear in the classical amplitudes.

Classical functions √
(k)
Since the polynomials Pn form a basis of the Kerr and Kerr quantum
amplitudes, the functional form of the corresponding classical amplitudes
is governed by the classical limits of these polynomials. The pole terms
of the quantum amplitudes A√Kerr , MKerr in eq. (3.36) depend on the
three polynomials,
∞ 2
X e−|z| (2s)
lim P1 = ex+z
ℏ→0
2s=0
(2s)!
∞ 2
X e−|z| (2s) sinh z
lim P2 = e x =: ex sinhcz
ℏ→0
2s=0
(2s)! z
∞ 2
X e−|z| (2s) 2x cosh y + (x2 +y 2 −z 2 ) sinhc y y↔z
x
lim P4 =

2 −z 2 (x + y)2 −z 2

+ x→−x e
ℏ→0
2s=0
(2s)! (x − y)
=: Ẽ(x, y, z) (9.24)
The resulting classical functions are all entire functions that can be ex-
panded as an infinite tower of spin multipoles. For example, the first

77
spin-multipole orders of the Ẽ are
1 1 1 1
Ẽ(x, y, z) = + x+ (3x2 +y 2 +z 2 )+ x(2x2 +y 2 +2z 2 )+O(a4 ) .
6 12 120 360
(9.25)
The Compton amplitudes, A√Kerr and MKerr also depend on polyno-
(2s) (2s−i)
mials with degree shifted by an integer i, Pn → Pn . The clas-
sical limit of these shifted polynomials are identical up to a penalty of
(ℏm)i (2z)−i , such that the leading order is suppressed and appears at
O(ℏi ). Such shifted polynomials can, and do, still contribute classically
if they appear with prefactors that have non-trivial ℏ scaling.
The Kerr amplitude in eq. (3.39) also includes limits of higher-order
(k)
polynomials. Taking a limit of the polynomial Pn+1 is equivalent to
(k)
taking derivatives of the polynomial Pn , such that
(2s) ∂ (2s)
lim P5 = P . (9.26)
ς5 →ς1 ∂ς1 4
Therefore the resulting classical functions can be related to Ẽ by the
partial derivatives 1
∞ 2 
e−|z|

X ∂ ∂ (2s) ∂
lim + P4 = Ẽ(x, y, z)
ℏ→0
2s=0
(2s)! ∂ς1 ∂ς2 ∂x
∞ 2 
e−|z|

X ∂ ∂ (2s) ∂
lim − P4 = Ẽ(x, y, z). (9.27)
ℏ→0
2s=0
(2s)! ∂ς1 ∂ς2 ∂z
√
In the majority of terms in the Kerr and Kerr, the leading order of
the polynomials is the classically significant term. However the electro-
magnetic and gravity amplitudes contain a pair of terms that are individ-
ually divergent but cancel such that the subleading terms are classically
relevant. In the gravity amplitude (3.36), these terms are pole terms
⟨13⟩⟨32⟩[14][42] ρ · χ
(2s−1) (2s−2)

ρ · χp · χ P − ς 3 ς 4 P . (9.28)
m4s−4 s12 p·χ 4 4

Working to leading order in ρ · χ ≈ 2z/(p · q⊥ ), the classical limit is

∞ 2
X e−|z| 1 ρ · χ  ρ · χ (2s−1) (2s−2)

lim P4 − ς3 ς4 P 4
ℏ→0
2s=0
(2s)! 4 p · χ p · χ
ey − ex cosh z + (x−y)ex sinhc z
= + (y → −y)
(x−y)2 − z 2
=: E(x, y, z) ,
(9.29)
1 ∂ ∂
In paper III ∂x Ẽ(x, y, z) is called E and ∂z
Ẽ(x, y, z) is called Ẽ but we will refrain
from using such notation here

78
where individual O(ℏ−1 ) divergences cancel. The rational function E is
once again an entire function and can be related to Ẽ by the following
differential equation
∂ 3
E(x, y, z) = λ Ẽ(λx, λy, λz). (9.30)
∂λ
The full treatment of the term in eq. (9.28) includes taking into account
the subleading term in ρ · χ ≈ 2z/(p · q⊥ ) + ℏ(x − w), which combines
(2s−1)
with the super-classical ℏ−1 piece of P4 such that the contribution
is classical.

√
9.2.1 Candidate classical Compton amplitude for Kerr
We define our classical amplitude as the classical limit of the quantum
√
Kerr amplitude given in eq. (3.36), such that
 w2 − z 2 
A√Kerr (1, 2̄, 3−, 4+ ) = A0 ex cosh z − w ex sinhc z + E(x, y, z) ,
2
(9.31)
where we have implemented either √ classical limit (ii) or (iii).
Therefore fixing the quantum Kerr amplitude is equivalent to fixing
the classical amplitude. The quantum √ amplitude given in eq. (3.36)
follows from the chiral higher-spin Kerr Lagrangian, given in eq. (6.14)
to cubic order. However the resulting general-spin Compton amplitude
(6.32) has unfixed contact terms C (s) . We now list the combination of
higher-spin and classical constraints that are used to fix C (s) .

Higher-spin constraints
Our study of the non-chiral constructions for spin-2 and spin-3 fields pro-
duced a set of constraints on C (s) . Notably by a combination of minimal
coupling (MC), power counting (PC) and Ward identity (WI) constraints
we can constrain C (s<2) = 0, while at C (s=2) has is constrained to be
a linear combination of three possible contact terms (5.51). The same
analysis at spin-3, constrains C (s=3) to be a linear combination of 21
possible contact terms.

Classical consistency constraints

While the above higher-spin constraints restrict the space of possible
contact terms, we reduce the parameter space further by imposing certain
classical consistency constraints.
The first constraint is that A√Kerr must converge in the classical limit.
This is non-trivial given the cubic Lagrangian generates the contact term
(s) ⟨13⟩⟨32⟩[14][42] (2s−1)
CL3 = − ς 3 ς 4 P4 (9.32)
m4
79
which diverges in the classical limit. This divergence appears in both the
(s)
large spin and coherent spin approaches. Studying CL3 in the coherent
state approach we find the spin-dependent term scales as
∞ 2
X e−|z| (2s−1) 1
P4 = ℏ Ẽ(x, y, z) + O(ℏ2 ) , (9.33)
2s=0
(2s)! 2z

(s)
while the coefficient ⟨13⟩⟨32⟩[14][42]ς3 ς4 ∼ ℏ−2 , such that CL3 con-
tributes super-classically O(ℏ−1 ). This suggest we must add further
contact terms C (s) in order to generate a classical amplitude with con-
sistent O(ℏ0 ) scaling.
We construct a general spin contact term with the form
(2s−n) (2s−n)
C (s) ∝ ⟨13⟩⟨32⟩[14][42] × {P4 , P2 }, (9.34)

where we have constrained the helicity structure of the massless spinors

(s)
to match CL3 . Note this is consistent with the massive gauge symmetry
(2)
constraints at spin-2, given two of the terms in C−+ exhibit this structure
(5.51). We also fix the dependence on s to be encoded in the polynomials
(2s−i)
P2,4 where we allow the degree to be shifted by integer i. We will
also impose that C (s) respects charge, parity and time-reversal symmetry,
such that A√Kerr contains no dissipation effects.
Even after imposing the constraints above, it is clear there are many
possible quantum contact terms of this form. However, the majority of
this freedom is washed out in the classical limit such that there is a single
free parameter δ in the classical amplitude
∞ 2
X e−|z| (s) w2 − z 2
CL3 + C (s) = (1 − δ)


lim E(x, y, z) . (9.35)
ℏ→0
2s=0
(2s)! 2

This free parameter δ enters at quadrupole order O(a2 ) in the multipole

expansion of the amplitude. Therefore we can fix it by requiring the
quadrupole to be fixed by the finite spin limit of the spin-1 amplitude,
1
A(1s=1 , 2s=1 , 3− , 4+ )/A0 −→ 1 + (w − x) + (w − x)2 , (9.36)
limit (i) 2

see section 9.1.1. This corresponds to setting δ = 0, such that the

resulting classical amplitude, given in eq. (9.31), is unique.
The simplest quantum contact term that satisfies the combined higher-
spin and classical constraints is
⟨13⟩⟨32⟩[14][42] 
(2s) (2s−2)

C (s) = − (ς 3 + ς 4 ) P4 − P2 . (9.37)
2m4
80
Although one can easily deform the quantum contact term C (s) without
impacting the classical amplitude.
The resulting classical amplitude coincides with the classical limit of
AAHH in eq. (3.34) up to quadrupolar order, a2 . As mentioned in section
8.2, we expect A√Kerr to correspond to the scattering of electromagnetic
waves of a Kerr-Newman background induced by the electromagnetic
charge of the black hole, although the classical field theory calculation is
yet to be done.

9.2.2 Candidate classical Compton amplitudes for Kerr

The classical Kerr amplitude is, likewise, defined as the classical limit of
the quantum Kerr amplitude, proposed in eq. (3.39),

MKerr (1, 2, 3−, 4+ ) =


w2 −z 2
M0 ex cosh z − w ex sinhc z + 2 E + (w2 −z 2 )(x−w)Ẽ

(w2 −z 2 )2  ∂ Ẽ

∂ Ẽ 
− +η + α Cα(∞) ,
2ξ ∂x ∂z
(9.38)
where E = E(x, y, z) and Ẽ = Ẽ(x, y, z) are the entire functions defined
in eq. (9.24) and eq. (9.29). The first line of the classical amplitude is
generated by the pole structure of the quantum amplitude in eq. (3.39).
The second line of the classical amplitude is generated by the contact
term in eq. (3.39),

⟨13⟩2 ⟨32⟩2 [14]2 [42]2 h

(2s−2) (2s−2)
i
C (s) = ς 3 ς 4 (1+η)P 5|ς1 + (1−η)P 5|ς2 .
2m6
(9.39)
This contact term is fixed by the following combination of higher-spin
and classical constraints.

Higher-spin constraints
From the analysis of the non-chiral higher-spin theories, we find that
compatibility with massive gauge symmetry requires C (s<3) = 0.
We also require the amplitude to have improved high-energy behaviour
such that, in the massless limit, MKerr is finite for s ≤ 2 and MKerr ∼
(2s−2)
m−4s+4 otherwise. Given P5|ς1 is degree 2s − 6 polynomial in the
(2s−2)
variables ςi , the combination ς3 ς4 P5|ς1 scales as m−4s+8 in the mass-
less limit. Therefore in the massless limit the contact term defined in
eq. (9.39) scales as C (s) ∼ m−4s+4 .

81
Classical constraints√
In contrast to the Kerr amplitudes, the Kerr amplitude is automati-
cally well-behaved in the classical limit, therefore it is not a meaningful
constraint.
However, in this case we have access to classical data generated by
studying linear perturbations of a Kerr black holes, see section 8.2. The
classical amplitude MTeuk. computed in ref. [9] contains results up to
a6 , which can be split into sectors according to their dependency on the
√ α and η .
book-keeping parameters
In analogy to the Kerr constraint, we impose that the classical hex-
adecapole a4 is fixed by the classical limit of the s = 2 amplitude defined
in eq. (3.34). This is a constraint compatible with the results of ref. [9]
since the s = 2 amplitude matches the first orders of MTeuk. up to order
a4 .
This constraint is enough to fix the η -independent contributions of
contact term C (s) in eq. (9.39). The contact term C (s) generates an
infinite tower of classical multipoles given the classical limit generates
the dependence on the entire function Ẽ(x, y, z), and it’s derivatives.
Remarkably, we find that these classical multipoles correctly predict all
the α = η = 0 terms in MTeuk., which is known up to a6 in ref. [9].
We can also construct quantum contact terms to capture the η -dependent
contact terms in MTeuk. . These contact terms are built from the same
spinor-helicity variables and polynomials, however, since η tags terms are
dissipative, the quantum contact terms must break time-reversal sym-
metry. This corresponds to breaking the symmetry when swapping legs
1 and 2.
We can fix the contact term by fixing the classical a5 multipole given
in MTeuk. , such that the combined α-independent contact term is given
in eq. (9.39). Note how the η -dependent terms are not symmetric under
swapping legs 1 ↔ 2.
Our resulting classical amplitude accurately predicts all given orders
of MTeuk up to the α-dependent terms [9],

MKerr − MTeuk = O(a7 ) . (9.40)

α=0

We can also reproduce the α-dependence given an adequate choice of

(s)
Cα shown in paper IV. However we find that such contributions must
be added at each spin multipole order, such that the classical amplitude
is not predictive.

82
10. Summary of Results and Outlook

√
In Part I we fixed the cubic order EFTs for Kerr and Kerr with a
series of higher-spin constraints. In the non-chiral construction, the con-
sistency requirements on massive higher-spin theories correspond to im-
posing massive gauge invariance, which was either at the level of the
Lagrangian or through Ward identities. This, combined with restrictive
power counting on the non-minimal interactions was sufficient to fix the
cubic order and other constraints was enough to fix the quantum three-
point amplitudes A√Kerr and MKerr , defined in eq. (3.20) and eq. (2.10)
uniquely. In the chiral construction, the amplitudes follow from im-
posing parity invariance on the cubic Lagrangian and not introducing
dependence on the self-dual field strength or Riemann spinor.
At quartic order, we found that the higher-spin constraints were not
enough to fix the four-point Compton amplitudes uniquely. Therefore,
we supplemented them with a set of classical constraints. Uplifting clas-
sical constraints to a constraint on the quantum contact terms is only
possible if we have well-defined classical limits. Therefore, in section 9,
we discussed in detail two consistent formulations of the classical limit,
the infinite spin classical limit and the coherent state classical limit.
While the two approaches agree, we predominantly use the coherent
state limit in this thesis as it is particularly well-suited to amplitudes
(k)
involving the polynomials Pn .
The nature
√ of the classical constraints differs according to whether we
consider Kerr or Kerr since there is a large discrepancy
√ in the state
of knowledge
√ of the classical amplitudes for the Kerr and Kerr cases.
For Kerr there are no explicit four-point results from classical field
theory, therefore we only rely on classical consistency requirements to
fix the free contact terms in the Compton amplitude. The resulting
classical amplitude is unique given the constraints imposed. It would
be interesting to compare the result to future classical computations,
for example the scattering of electromagnetic waves off a Kerr-Newman
black hole.
For Kerr, the classical data up to a6 has been available in the liter-
ature for some time [9]. In ref. [9], the classical amplitude is split into
conservative terms, dissipative terms and terms generated from the ex-
pansion of non-analytic functions, which are bookmarked with α. We fix
the contact terms corresponding the conservative sector at order a4 and
find that they correctly predict the known higher-orders. By introducing

83
time-reversal breaking contact terms, which we fix at a5 , the resulting
amplitude predicts the results in ref. [9], up to the terms proportional to
α.
(s)
While we are able to construct an explicit contact term Cα that
reproduces them up to a6 , it is not clear that the contact term will
capture higher order terms correctly. There are still open questions on
whether these α terms are physically relevant terms or not given that
they are sensitive to the basis of poly-gamma functions used in classical
calculation [9, 7].
Nonetheless, our classical Kerr amplitude is a closed form expression
that contains all-orders in the spin multipole, such that it can be used
to generate new predictions for classical observables, such as the leading
impulse and spin-kick and the leading-order waveform.
Alternatively, we could return to the study of the Lagrangians, with
the goal of treating the dissipative terms seriously. The Lagrangians
studied in this thesis are not able to generate the time reversal breaking
contact terms that captured the dissipative physics of the scattering. A
possible resolution would be to study a theory describing the interactions
of particles with different spin.
We can also extend the study to higher-point amplitudes, which are
necessary for corrections to the observables at higher orders in the post-
Minkowski expansion. We expect the higher-point classical Compton
amplitudes to describe non-linear perturbations of Kerr black holes.

84
11. Acknowledgements

Firstly, I would like to thank my supervisor Henrik. I am so grateful

for all the time and energy you have invested in me and in our projects
together, I am definitely a better researcher for it. You have been so
generous and thanks to your support I have had access to some incredible
opportunities. Thank you.
I have been lucky to meet and work with great people throughout
my PhD. A special mention goes to my collaborator Paolo. Our col-
laborations have always been characterised by a joyful camaraderie. I
still vividly remember the blackboard discussions in the room at KITP
overlooking the palm trees and the ocean. I would also like to thank my
collaborators Marco, Alex and Zhenya, for the stimulating discussions
and humour-filled meetings. Also, thank you Lara; as fellow Swiss fe-
male PhDs in field dominated by men, our collaboration and friendship
has been a reaffirming experience for me.
I found the theoretical physics division in Uppsala to be a warm and
welcoming place, which is due to the people that populate it. Thank you
to members both new and old: Agnese, Alessandro, Alex, Bram, Carlos,
Carlo, Daniel, Daniele, Dima, Fillipo, Francesco, Giulia, Guido, Ingrid,
Jacopo, Joe, Kaiwen, Kays, Kunal, Magdalena, Maor, Marco, Martijn,
Maxim, Michele, Moritz, Lisa, Luca, Lucia, Oliver, Paul, Pietro, Roman,
Sam, Sourav, Simon, Ulf, Vincent, Yacoub, Yoann and more.
A special thank you goes to fellow student and close friend Maor,
whom I have sat next to for the last five years. Thank you for your
support and patience. Thank you Carlos, one of the Atrani three, for
your enthusiasm, Ecuadorian chocolate and help with the ML course.
Thank you to Charel and Simon who started in the same month as me
and have accompanied me through this PhD. Thank you to Joe, Paul,
Roman and all those who initiated the ultimate frisbee games on the
field outside the department.
On the topic of frisbee, thank you to Emma, Hannes, Sia, Vincent
and Andrea, whom I met through playing and who have since enriched
my life. Thank you to Victoria, Cissi and the other women of Cantus
Feminis, it was a joy to sing and drink punsch with you. Thank you to
my far away friends, Claire, Maazin, Fidel, Daniel, Mara, Vici, Caragh,
Yotal, Nerea, to name but a few. You have known me through many life
stages and have always strove to support me in them, which I am forever
grateful for and hope to repay in kind.

85
 Last but definitely not least, I thank my family. Thank you Megan
and Thibault for keeping me humble and reminding me that I am the
dumbest smart person you know. Thank you mum and dad for your
unwavering support and belief in me. I would not be where I am without
you.

86
12. Svensk Sammanfattning

Den allmänna relativitetsteorin, som formulerades av Albert Einstein

1915, revolutionerade vår förståelse av gravitationskraften genom att
beskriva den geometriskt så att rumtidens krökning orsakas av energi och
materia. Bland de mest fascinerande lösningarna till hans fältekvationer
är svarta hål. Ett svart hål kännetecknas av att krökningen är så pass
stor att ingenting, inte ens ljus, kan undkomma gravitations grepp inom
en region som kallas händelsehorisonten. Innanför denna gräns finner
man en matematisk singularitet, en punkt med oändlig densitet där de
fysikaliska lagarna, som vi förstår dem, bryts ned. Roy Kerr föreslog 1963
en lösning som generaliserade Karl Schwarzschilds sfäriskt symmetriska
svarta hål, och introducerade därmed konceptet av ett roterande svart
hål, som inte bara har massa utan även rörelsemängdsmoment.
Den statiska lösningen för Kerrs svarta hål är välkänd sedan drygt 60
år tillbaka, men dynamiken hos dessa svarta hål är mycket komplicerad
och beskrivs av störningar av metriken som uppfyller Einsteins ickelinjära
fältekvationer. Av särskilt intresse är dynamiken hos bundna binära sys-
tem, dvs. system där två svarta hål kretsar kring varandra i krympande
omloppsbanor för att slutligen smälta samman till ett större svart hål.
Gravitationsvågorna som sänds ut under denna process kan observeras
av de markbaserade detektorerna LIGO, Virgo samt KAGRA.
Gravitationsvågorna kodar information om sina astofysikaliska källor
och utrönandet av underliggande parametrar är beroende av matchn-
ing med teoretiska modeller som håller hög precision. På grund av den
allmänna relativitetsteorins ickelinjära natur är det inte möjligt att finna
exakta lösningar som är relevanta för observationer. Istället konstrueras
de teoretiska vågformerna vanligtvis via en kombination av numeriska
och analytiska metoder som uppfyller kravet att de är goda approxima-
tiva beskrivningar.
Under den tidiga delen av den krympande omloppsbanan är de svarta
hålen fortfarande väl separerade och fysiken styrs av svaga växelverkningar.
Därför kan vi lösa fältekvationerna med hjälp av störningsteori serieutveck-
lad i gravitationskonstanten G, som även kallad post-Minkowski (PM)
utvecklingen. För att ytterligare förenkla analysen kan man studera en så
kallad spridningsamplitud för två förbipasserande svarta hål där gravita-
tionens växelverkan är ett övergående fenomen, i motsats till det bundna
systemet där svarta hål ständigt växelverkar. I många fall kan sprid-
ningsprocessen matematiskt relateras till den bundna omloppsbanan, till
exempel genom att beräkna en effektiv gravitationspotential.

87
 I spridningsscenariot kan vi dra nytta av de många analytiska verk-
tyg som ursprungligen utformades för partikelfysik, genom att modellera
svarta hål som elementarpartiklar med stor massa och spinn som beskrivs
av en kvantfältteori. Att detta tillvägagångssätt fungerar är anmärkn-
ingsvärt med tanke på att vi för närvarande inte har någon komplett
beskrivning av gravitationskraften som en kvantmekanisk teori. Kvant-
fältmetodens framgång bygger på att dess verkan modellerar en effektiv
teori som är vid låga energier förenlig med allmänna relativitetsteorin. Vi
kan undvika de oändligheter som förekommer i kvantgravitations ofull-
ständiga formuleringar genom att begränsa oss till energiskalor under
Planckenergin, vilket är fullt tillräckligt för de svarta hål som observeras
genom gravitationsvågdetektorerna.
Dynamiken hos Schwarzschilds ickeroterande svarta hål har framgångsrikt
beräknats från spridningsamplituder som involverar massiva skalärfält
minimalt kopplade till metriken, vilket medfört att den effektiva po-
tentialen och spridningsvinkeln är kända till många ordningar i PM-
utvecklingen. För att införliva frihetsgrader som motsvarar rotation kan
man studera kvantfältteorier för massiva partiklar med spinn. Denna
metod har varit framgångsrik för spinn som inte överstiger det hos kända
elementarpartiklar. Det finns ett subtilt samband mellan det klassiska
rörelsemängdsmomentet, en kontinuerlig vektorvariabel, och kvantspin-
net, ett diskret kvanttal. Den vägledande principen är att processer som
involverar en partikel med spinn s kan generera klassiska spinnmulti-
polsmoment upp till ordningen 2s. För att beräkna klassiska processer
till högre ordningar i det klassiska spinnet från en partikelmodell krävs
det att man studerar växelverkningar för obegränsat högt kvantspinn, så
kallade högre spinnteorier.
Fokus för denna avhandling är att konstruera kvantfältteorier för hö-
gre spinn som ger en effektiv beskrivning av dynamiken för Kerrs svarta
hål. Högre spinnfält är kända för att vara besvärliga att studera eftersom
de ofta leder till matematiska och kvantmekaniska oegentligheter om de
inte behandlas varsamt. För att överkomma dessa problem så studerar
denna avhandling två matematiska konstruktioner för högre spinnfält
som säkerställer att de effektiva teorierna är välartade och har korrekt
antal frihetsgrader. Den första konstruktionen introducerar en ändlig
samling av massiva spinnfält som är relaterade till varandra via en el-
egant gaugesymmetri, och växelverkningarna kan nu begränsas till att
vara invarianta under denna symmetri. Den andra metoden introduc-
erar endast ett spinnfält med kirala frihetsgrader som växelverkar med
gravitationsfältet på ett sätt så att inga oegentliga tillstånd bildas. Ett
resultat av avhandlingen är en uppsättning principer som på ett unikt
sätt ger hur det svarta hålet växelverkar med en graviton, gravitations-
fältets kvanta. Växelverkningar med två gravitoner, som ger upphov till

88
en Comptonamplitud, studeras i detalj och en fullständig formel för alla
kvantspinn presenteras.
Från de kvantmekaniska amplituderna extraheras den klassiska Comp-
tonprocessen för Kerrs svarta hål. Detta är icketrivialt eftersom kvantspin-
ntalet måste omvandlas till den klassiska spinnvektorn vilket kräver att
spinntalet går mot oändligheten samtidigt som Plancks konstant går mot
noll. Summa summarum, från kvantmekaniska högre spinnteorier och
klassiska gränser erhålls dynamiken för ett roterande svart hål, vilket ver-
ifieras genom direkt jämförelse med tidigare kända störningsberäkningar
från allmänna relativitetsteorin.

89
References

[1] B. P. Abbott et al. Observation of Gravitational Waves from a Binary

Black Hole Merger. Phys. Rev. Lett., 116(6):061102, 2016.
[2] Rafael Aoude, Kays Haddad, and Andreas Helset. Searching for Kerr in
the 2PM amplitude. JHEP, 07:072, 2022.
[3] Rafael Aoude and Alexander Ochirov. Classical observables from
coherent-spin amplitudes. JHEP, 10:008, 2021.
[4] Nima Arkani-Hamed, Tzu-Chen Huang, and Yu-tin Huang. Scattering
amplitudes for all masses and spins. JHEP, 11:070, 2021.
[5] Nima Arkani-Hamed, Yu-tin Huang, and Donal O’Connell. Kerr black
holes as elementary particles. JHEP, 01:046, 2020.
[6] V. Bargmann and Eugene P. Wigner. Group Theoretical Discussion of
Relativistic Wave Equations. Proc. Nat. Acad. Sci., 34:211, 1948.
[7] Yilber Fabian Bautista, Giulio Bonelli, Cristoforo Iossa, Alessandro
Tanzini, and Zihan Zhou. Black Hole Perturbation Theory Meets CFT2 :
Kerr Compton Amplitudes from Nekrasov-Shatashvili Functions. 12 2023.
[8] Yilber Fabian Bautista, Alfredo Guevara, Chris Kavanagh, and Justin
Vines. Scattering in black hole backgrounds and higher-spin amplitudes.
Part I. JHEP, 03:136, 2023.
[9] Yilber Fabian Bautista, Alfredo Guevara, Chris Kavanagh, and Justin
Vines. Scattering in black hole backgrounds and higher-spin amplitudes.
Part II. JHEP, 05:211, 2023.
[10] Xavier Bekaert, Nicolas Boulanger, and Per Sundell. How higher-spin
gravity surpasses the spin two barrier: no-go theorems versus yes-go
examples. Rev. Mod. Phys., 84:987–1009, 2012.
[11] Zvi Bern, Julio Parra-Martinez, Radu Roiban, Michael S. Ruf,
Chia-Hsien Shen, Mikhail P. Solon, and Mao Zeng. Scattering
Amplitudes and Conservative Binary Dynamics at O(G4 ). Phys. Rev.
Lett., 126(17):171601, 2021.
[12] Zvi Bern, Julio Parra-Martinez, Radu Roiban, Michael S. Ruf,
Chia-Hsien Shen, Mikhail P. Solon, and Mao Zeng. Scattering
Amplitudes, the Tail Effect, and Conservative Binary Dynamics at
O(G4). Phys. Rev. Lett., 128(16):161103, 2022.
[13] N. E. J. Bjerrum-Bohr, Gang Chen, and Marcos Skowronek. Classical
spin gravitational Compton scattering. JHEP, 06:170, 2023.
[14] N. E. J. Bjerrum-Bohr, Poul H. Damgaard, Guido Festuccia, Ludovic
Planté, and Pierre Vanhove. General Relativity from Scattering
Amplitudes. Phys. Rev. Lett., 121(17):171601, 2018.
[15] N. E. J Bjerrum-Bohr, John F. Donoghue, and Barry R. Holstein.
Quantum gravitational corrections to the nonrelativistic scattering
potential of two masses. Phys. Rev. D, 67:084033, 2003. [Erratum:
Phys.Rev.D 71, 069903 (2005)].

90
[16] N. E. J. Bjerrum-Bohr, John F. Donoghue, and Pierre Vanhove. On-shell
Techniques and Universal Results in Quantum Gravity. JHEP, 02:111,
2014.
[17] Ralph Blumenhagen, Dieter Lüst, and Stefan Theisen. Basic concepts of
string theory. Theoretical and Mathematical Physics. Springer,
Heidelberg, Germany, 2013.
[18] L. Brink, P. Di Vecchia, and Paul S. Howe. A Locally Supersymmetric
and Reparametrization Invariant Action for the Spinning String. Phys.
Lett. B, 65:471–474, 1976.
[19] Ruth Britto, Freddy Cachazo, and Bo Feng. New recursion relations for
tree amplitudes of gluons. Nucl. Phys. B, 715:499–522, 2005.
[20] Ruth Britto, Freddy Cachazo, Bo Feng, and Edward Witten. Direct
proof of tree-level recursion relation in Yang-Mills theory. Phys. Rev.
Lett., 94:181602, 2005.
[21] Richard C. Brower. Spectrum generating algebra and no ghost theorem
for the dual model. Phys. Rev. D, 6:1655–1662, 1972.
[22] I. L. Buchbinder, T. V. Snegirev, and Yu. M. Zinoviev. On gravitational
interactions for massive higher spins in AdS3 . J. Phys. A, 46:214015,
2013.
[23] Gordon Chalmers and Warren Siegel. Simplifying algebra in Feynman
graphs. 3. Massive vectors. Phys. Rev., D63:125027, 2001.
[24] Marco Chiodaroli, Henrik Johansson, and Paolo Pichini. Compton
black-hole scattering for s ≤ 5/2. JHEP, 02:156, 2022.
[25] Ming-Zhi Chung, Yu-Tin Huang, and Jung-Wook Kim. Kerr-Newman
stress-tensor from minimal coupling. JHEP, 12:103, 2020.
[26] Ming-Zhi Chung, Yu-Tin Huang, Jung-Wook Kim, and Sangmin Lee.
The simplest massive S-matrix: from minimal coupling to Black Holes.
JHEP, 04:156, 2019.
[27] A. Cucchieri, M. Porrati, and Stanley Deser. Tree level unitarity
constraints on the gravitational couplings of higher spin massive fields.
Phys. Rev. D, 51:4543–4549, 1995.
[28] W. K. De Logi and S. J. Kovacs. Gravitational Scattering of Zero Rest
Mass Plane Waves. Phys. Rev. D, 16:237–244, 1977.
[29] E. Del Giudice, P. Di Vecchia, and S. Fubini. General properties of the
dual resonance model. Annals Phys., 70:378–398, 1972.
[30] Stanley Deser and Andrew Waldron. Inconsistencies of massive charged
gravitating higher spins. Nucl. Phys. B, 631:369–387, 2002.
[31] Stanley Deser and B. Zumino. A Complete Action for the Spinning
String. Phys. Lett. B, 65:369–373, 1976.
[32] John F. Donoghue. Quantum General Relativity and Effective Field
Theory. 2023.
[33] Henriette Elvang and Yu-tin Huang. Scattering Amplitudes in Gauge
Theory and Gravity. Cambridge University Press, 4 2015.
[34] Sergio Ferrara, Massimo Porrati, and Valentine L. Telegdi. g = 2 as the
natural value of the tree-level gyromagnetic ratio of elementary particles.
Phys. Rev. D, 46:3529–3537, 1992.
[35] Masafumi Fukuma, Hikaru Kawai, Katsuta Sakai, and Junji Yamamoto.

91
 Massive higher spin fields in curved spacetime and necessity of
non-minimal couplings. PTEP, 2016(7):073B02, 2016.
[36] P. Goddard, J. Goldstone, C. Rebbi, and Charles B. Thorn. Quantum
dynamics of a massless relativistic string. Nucl. Phys. B, 56:109–135,
1973.
[37] P. Goddard and Charles B. Thorn. Compatibility of the Dual Pomeron
with Unitarity and the Absence of Ghosts in the Dual Resonance Model.
Phys. Lett. B, 40:235–238, 1972.
[38] Alfredo Guevara, Alexander Ochirov, and Justin Vines. Black-hole
scattering with general spin directions from minimal-coupling
amplitudes. Phys. Rev., D100:104024, 2019.
[39] Alfredo Guevara, Alexander Ochirov, and Justin Vines. Scattering of
Spinning Black Holes from Exponentiated Soft Factors. JHEP, 09:056,
2019.
[40] Suraj N. Gupta and Stanley F. Radford. Quantum field-theoretical
electromagnetic and gravitational two-particle potentials. Phys. Rev. D,
21:2213–2225, Apr 1980.
[41] Barry R. Holstein and John F. Donoghue. Classical physics and quantum
loops. Phys. Rev. Lett., 93:201602, 2004.
[42] Y. Iwasaki. Quantum theory of gravitation vs. classical theory. -
fourth-order potential. Prog. Theor. Phys., 46:1587–1609, 1971.
[43] Gustav Uhre Jakobsen, Gustav Mogull, Jan Plefka, and Benjamin Sauer.
Dissipative Scattering of Spinning Black Holes at Fourth
Post-Minkowskian Order. Phys. Rev. Lett., 131(24):241402, 2023.
[44] Henrik Johansson and Alexander Ochirov. Double copy for massive
quantum particles with spin. JHEP, 09:040, 2019.
[45] H. Kawai, D. C. Lewellen, and S. H. H. Tye. A Relation Between Tree
Amplitudes of Closed and Open Strings. Nucl. Phys. B, 269:1–23, 1986.
[46] S. M. Klishevich and Yu. M. Zinoviev. On electromagnetic interaction of
massive spin-2 particle. Phys. Atom. Nucl., 61:1527–1537, 1998.
[47] Chrysoula Markou and Evgeny Skvortsov. An excursion into the string
spectrum. JHEP, 12:055, 2023.
[48] Ben Maybee, Donal O’Connell, and Justin Vines. Observables and
amplitudes for spinning particles and black holes. JHEP, 12:156, 2019.
[49] Duff Neill and Ira Z. Rothstein. Classical Space-Times from the S
Matrix. Nucl. Phys. B, 877:177–189, 2013.
[50] Alexander Ochirov and Evgeny Skvortsov. Chiral Approach to Massive
Higher Spins. Phys. Rev. Lett., 129(24):241601, 2022.
[51] Massimo Porrati. Massive spin 5/2 fields coupled to gravity: Tree level
unitarity versus the equivalence principle. Phys. Lett. B, 304:77–80, 1993.
[52] Rakibur Rahman and Massimo Taronna. From Higher Spins to Strings:
A Primer. 12 2015.
[53] A. Sagnotti and M. Taronna. String Lessons for Higher-Spin Interactions.
Nucl. Phys. B, 842:299–361, 2011.
[54] Oliver Schlotterer. Higher Spin Scattering in Superstring Theory. Nucl.
Phys. B, 849:433–460, 2011.
[55] L. P. S. Singh and C. R. Hagen. Lagrangian formulation for arbitrary

92
 spin. 1. The boson case. Phys. Rev., D9:898–909, 1974.
[56] Bo Sundborg. Selfenergies of Massive Strings. Nucl. Phys. B,
319:415–438, 1989.
[57] Massimo Taronna. Higher Spins and String Interactions. Other thesis, 5
2010.
[58] Varun Vaidya. Gravitational spin Hamiltonians from the S matrix. Phys.
Rev. D, 91(2):024017, 2015.
[59] Justin Vines. Scattering of two spinning black holes in post-Minkowskian
gravity, to all orders in spin, and effective-one-body mappings. Class.
Quant. Grav., 35(8):084002, 2018.
[60] Steven Weinberg. The Quantum theory of fields. Vol. 1: Foundations.
Cambridge University Press, 6 2005.
[61] Yu. M. Zinoviev. On massive high spin particles in AdS. 8 2001.
[62] Yu. M. Zinoviev. Massive spin two supermultiplets. 2002.
[63] Yu. M. Zinoviev. Massive supermultiplets with spin 3/2. JHEP, 05:092,
2007.
[64] Yu. M. Zinoviev. On massive spin 2 interactions. Nucl. Phys. B,
770:83–106, 2007.
[65] Yu. M. Zinoviev. On massive spin 2 electromagnetic interactions. Nucl.
Phys. B, 821:431–451, 2009.
[66] Yu. M. Zinoviev. On spin 2 electromagnetic interactions. Mod. Phys.
Lett. A, 24:17–23, 2009.
[67] Yu. M. Zinoviev. On spin 3 interacting with gravity. Class. Quant.
Grav., 26:035022, 2009.
[68] Yu. M. Zinoviev. Spin 3 cubic vertices in a frame-like formalism. JHEP,
08:084, 2010.

93
Acta Universitatis Upsaliensis
Digital Comprehensive Summaries of Uppsala Dissertations from
the Faculty of Science and Technology 2383
Editor: The Dean of the Faculty of Science and Technology

A doctoral dissertation from the Faculty of Science and

Technology, Uppsala University, is usually a summary of a
number of papers. A few copies of the complete dissertation
are kept at major Swedish research libraries, while the
summary alone is distributed internationally through
the series Digital Comprehensive Summaries of Uppsala
Dissertations from the Faculty of Science and Technology.
(Prior to January, 2005, the series was published under the
title “Comprehensive Summaries of Uppsala Dissertations
from the Faculty of Science and Technology”.)

ACTA UNIVERSITATIS
Distribution: publications.uu.se UPSALIENSIS
urn:nbn:se:uu:diva-525178 2024