High Energy Physics: from Quantum

Field Theory to Higgs Coupling

Measurement

by
Evan Ranken

Submitted in Partial Fulfillment of the

Requirements for the Degree
Doctor of Philosophy
CERN-THESIS-2020-366

Supervised by
Professor Regina Demina
07/08/2020

Department of Physics and Astronomy

Arts, Sciences and Engineering

University of Rochester
Rochester, New York
2020
 ii

“For Mom, Dad, and Lana”

 iii

Table of Contents

Biographical Sketch viii

Acknowledgements x

Abstract xii

Contributors and Funding Sources xiii

List of Tables xv

List of Figures xvi

List of Symbols and Abbreviations xxiii

1 Introduction 1
1.1 What’s in This Document . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 A Primer on Particle Physics 4

2.1 The Big Picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 When Things Got Weird: Quantum Mechanics . . . . . . . . . . . . . 6
2.3 The Leap to Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Gauge Theories and Particle Interaction . . . . . . . . . . . . . . . . . 11
 iv

2.5 More Gauge Theories: the Nuclear Forces . . . . . . . . . . . . . . . . 16

2.6 The Higgs Mechanism and Electroweak Unification . . . . . . . . . . 20
2.7 The Standard Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.8 Mass in the Standard Model . . . . . . . . . . . . . . . . . . . . . . . . 26
2.9 A Big, Beautiful Mess of Particles . . . . . . . . . . . . . . . . . . . . . 29

3 Unsolved Problems in Particle Physics 31

3.1 Things that are Conspicuously Missing from the Standard Model . . 31
3.2 Quantum Fields: Rough Around the Edges . . . . . . . . . . . . . . . 34
3.3 The Strongly-Coupled Regime . . . . . . . . . . . . . . . . . . . . . . 35
3.4 The Hierarchy Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4 Theoretical Paths Forward 40

4.1 Maybe More is Better . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2 Mathematical Rigor in QFT . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3 Research Project: Continuity of Log-Correlated Gaussian Fields . . . 46
4.4 Toy Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.5 Research project: Nonlinear Wave Solutions in a Strongly-Coupled
Toy Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.6 Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5 Experimental Paths Forward 56

5.1 The Energy Frontier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.2 Other Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

6 Top Quarks and Collider Physics 61

6.1 Why are Top Quarks Interesting? . . . . . . . . . . . . . . . . . . . . . 61
6.2 Top Quark Production at Hadron Colliders . . . . . . . . . . . . . . . 62
 v

6.3 Top quarks: Decay and Detection . . . . . . . . . . . . . . . . . . . . . 64

6.4 Kinematics of dilepton event reconstruction . . . . . . . . . . . . . . . 68

7 The CMS Experiment 73

7.1 The Large Hadron Collider (LHC) apparatus . . . . . . . . . . . . . . 73
7.2 The CMS detector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.3 Collision basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.4 Simulation of CMS events . . . . . . . . . . . . . . . . . . . . . . . . . 82

8 A New Measurement of the Top Quark Yukawa Coupling 85

8.1 Basic Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.2 Electroweak Corrections and tt̄ Production . . . . . . . . . . . . . . . 86
8.3 Simulation of tt̄ Events at CMS . . . . . . . . . . . . . . . . . . . . . . 89
8.4 Formalism of Weak Corrections . . . . . . . . . . . . . . . . . . . . . . 90
8.5 Object & Event Selection . . . . . . . . . . . . . . . . . . . . . . . . . . 93
8.6 Event Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
8.7 Statistical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
8.8 Sources of Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
8.8.1 Experimental sources of uncertainty . . . . . . . . . . . . . . . 107
8.8.2 Theoretical sources of uncertainty . . . . . . . . . . . . . . . . 108
8.8.3 Processing of shape templates . . . . . . . . . . . . . . . . . . 111
8.9 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

9 Concluding Remarks 116

A Continuity of Scalar Fields With Logarithmic Correlations 119

A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
 vi

A.2 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

A.2.1 Continuity of Random Processes . . . . . . . . . . . . . . . . . 123
A.2.2 Quantum Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
A.2.3 Moving Average of a Quantum Field . . . . . . . . . . . . . . 126
A.3 Logarithmically correlated scalar field in 1 dimension . . . . . . . . . 128
A.4 Explicit calculations and further examples . . . . . . . . . . . . . . . . 131
A.4.1 Variance of φ̄s for the log-correlated field . . . . . . . . . . . . 131
A.4.2 Continuity of Brownian Paths . . . . . . . . . . . . . . . . . . . 133
A.4.3 Continuity of φ̄s for logarithmically correlated fields . . . . . 134
A.4.4 Additional Examples for Comparison . . . . . . . . . . . . . . 134
A.5 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

B Highly nonlinear wave solutions in a dual to the chiral model 142

B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
B.2 relation to other models . . . . . . . . . . . . . . . . . . . . . . . . . . 146
B.2.1 The c → 0 Limit and Slow Light . . . . . . . . . . . . . . . . . 146
B.2.2 Sub-Riemannian Geometry and the Strong Coupling Limit . 147
B.2.3 Relation to the WZW model . . . . . . . . . . . . . . . . . . . . 149
B.2.4 Duality with the principal chiral model . . . . . . . . . . . . . 150
B.3 Reduction to A Mechanical System . . . . . . . . . . . . . . . . . . . 152
B.3.1 The reduced system Lagrangian . . . . . . . . . . . . . . . . . 153
B.3.2 Reduction to One Degree of Freedom . . . . . . . . . . . . . . 154
B.3.3 Solution in terms of elliptic functions . . . . . . . . . . . . . . 155
B.4 Mechanical Interpretation and Quantization of the Reduced Systems 159
B.4.1 quantization of the dual reduced system . . . . . . . . . . . . 162
B.5 Conclusions and outlook . . . . . . . . . . . . . . . . . . . . . . . . . . 166
 vii

C Measurement of the top quark Yukawa coupling from tt̄ kinematic dis-
√
tributions in the dilepton final state at s = 13 TeV 168
C.1 Simulation of top quark pair production and backgrounds . . . . . . 172
C.1.1 Simulation of electroweak corrections . . . . . . . . . . . . . . 173
C.2 Event and object selection . . . . . . . . . . . . . . . . . . . . . . . . . 175
C.3 Event reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
C.3.1 Comparison between data and simulation . . . . . . . . . . . 180
C.4 Measurement strategy and statistical methods . . . . . . . . . . . . . 181
C.5 Sources of experimental and theoretical uncertainty . . . . . . . . . . 185
C.5.1 Processing of systematics . . . . . . . . . . . . . . . . . . . . . 190
C.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
C.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193

Bibliography 197
 viii

Biographical Sketch

The author grew up in the town of Bernalillo, New Mexico, in the United States.
He obtained his Bachelors degree at Colorado College in 2012, majoring in Physics &
Mathematics. In 2013, he enrolled in the PhD program at the University of Rochester,
supported by a Sproull Fellowship from the University. From 2014 until 2017, he
worked under the guidance of Professor S.G. Rajeev, studying a variety of research
topics related to quantum field theory in theoretical and mathematical physics.
In the spring of 2014 he received the department’s AAPT Teaching Prize for his
work as a teaching assistant, and went on to serve as instructor for an introductory
physics course that summer. In 2015 he was awarded a fellowship under the NSF
Graduate Research Fellowship Program (GRFP).

After inadvertently attending a conference filled with well-funded experimen-

talists in late 2016, the author was swayed to the dark side and decided to join an
experiment. Following a trip to co-instruct a course at Colorado College in spring
2017, he began working in earnest with his second advisor, Professor Regina Dem-
ina. in collaboration with the CMS experiment. The author then relocated to work
on-site at CERN, where he finished his final doctoral research project measuring
the interaction between the top quark and Higgs boson.
 ix

Publications during doctoral study:

"Measurement of the top quark Yukawa coupling using kinematics of tt̄ dilepton
√
decays at s= 13 TeV " (forthcoming)

with Regina Demina and the CMS collaboration

technical note CMS-PAS-TOP-19-008 available at cds.cern.ch
partially reprinted as Appendix A

"Highly nonlinear wave solutions in a dual to the chiral model" (2016)

with S.G. Rajeev, Phys. Rev. D 93 105016 [arXiv:1603.08256]

partially reprinted as Appendix B

"Continuity of scalar fields with logarithmic correlations" (2015)

with S.G. Rajeev, Phys. Rev. D 92 045021 [arXiv:1508.04655]

partially reprinted as Appendix C
 x

Acknowledgements

I would like to thank all of the teachers and professors who inspired and helped
me, starting during my youth in New Mexico. To name a few: Stuart Lipkowitz,
Danny Packer, Dr. Vraspir, Dr. Buchanan, Dr. Julian, Mr. Montaño... and so many
more! Thanks as well to everyone in the Physics & Mathematics (& Computer
Science) departments at Colorado College circa 2012, with extra special thanks to
Professors Stefan Erickson, Shane Burns, Stephanie Dicenzo, Marlow Anderson,
Mike Siddoway, David Brown, Phil Cervantes, Dick Hilt, and of course to Marita
Beckert! At the University of Rochester, thanks to Laura Blumkin, without whom the
department would fall apart. Thanks, of course, to those who advised me during
my time at Rochester: to Professor S.G. Rajeev for leading me on a wonderful
journey through mathematical physics, and to Professor Regina Demina for offering
to take me into a different world of physics research. Thanks as well to Aran Garcia-
Bellido for all his help and detailed reviews of our paper with CMS. Thanks as well
to the best housemates ever, Joe Murphree and Amanda Davis, and to everyone
else who helped with the self-paced curriculum. Thanks to Joseph Dulemba for
triple-checking this document, and to Rhys Taus for his valuable help exploring the
Galar region.
 xi

Thank you to the many friends from outside of physics who supported me while
I worked towards this dissertation. Special thanks to Kaila Ryan for uncountably
many things–without your friendship I would have found myself very alone in
2019. Thanks as well as John Derksen, Griffin Gorsky, Aku Antikainen and Meghan
Dorn, all of whom helped me greatly as I tried to maneuver the American medical
system. Thanks to everyone else who was supportive during that stressful time.

Thanks to Kyle Leggott and Robbie Martin for visiting me in France and for
taking me on a pleasant trip to the Sword Coast. Thanks to Abigail Kinem for
listening to me ramble. Thanks to Colin Martz for the music recommendations.
And thanks to everyone else who put up with me over the years, including everyone
who I studied alongside at UR and Colorado College. By now I have fallen out
of touch with many of those whose academic paths briefly crossed my own, but
I know I wouldn’t have finished this degree if I hadn’t met so many great people
along the way.

And lastly but most importantly, thanks to my parents Bill & Bev Ranken, who
have supported me through every smart and stupid thing I have ever done. Love
you!
 xii

Abstract

This document summarizes work in theoretical and experimental high-energy

physics completed as part of the author’s doctoral studies, while providing context
and motivation for these projects. Beginning with an overview of the relevant
backgrounds, the motivation for this doctoral research is explored by discussing
the current state of particle physics and many unanswered questions in this field.
Emphasis is placed on providing broad context to readers outside of the particle
physics community. Theoretical work from early PhD studies is briefly summarized,
exploring the continuity of scalar quantum fields and studying a novel toy field
theory with intriguing nonlinear wave solutions. The discussion then turns to the
experimental side of particle physics, covering relevant conceptual material and
experimental techniques. The author’s research in collaboration with the CMS
experiment is presented, detailing the motivation and methodology. This project
produces a measurement of the strength of interaction between the top quark and
Higgs boson, obtaining a Yukawa coupling of Yt = 1.16+ 0.24
−0.35 relative to the standard

model value.
 xiii

Contributors and Funding Sources

This work was supervised by a dissertation committee consisting of Professors

Regina Demina (advisor), S.G. Rajeev, and Arie Bodek of the Physics & Astronomy
Department, as well as Professor Alan Grossfield of the Department of Biochemistry
and Biophysics.

All theoretical research from 2014-2016 was done in close collaboration with
Professor S.G. Rajeev. Experimental research was performed from 2017-2020 with
the CMS collaboration under the guidance of Professor Regina Demina. Research
with CMS used a reconstruction framework developed by Otto Hindrichs, who
also provided useful guidance and tools to assess uncertainties and model detector
effects within that framework. All other work for the dissertation was completed
by the author independently.

This work was made possible by a wide variety of software, including: Mathe-
matica, The CERN ROOT framework with RooFit and RootStat; as well as free and
open source software like the GNU compiler collection, the Linux OS family, and
the Python programming language. Among these, the CERN ROOT framework
caused the author more mental anguish than the others.

Specific particle physics simulation tools are cited individually.

 xiv

Experimental work also relied on computing resources located at Fermi National

Accelerator Laboratory (Fermilab), and the European Center for Nuclear Research
(CERN).

Graduate study was supported by the U.S. Department of Energy Office of

Science, under Award Number DE-SC0008475 as well as the NSF Graduate Research
Fellowship Program under Grant No. DGE-141911, and by teaching assistantships
and the Robert L. and Mary L. Sproull University Fellowship at the University of
Rochester.
 xv

List of Tables

8.1 MC and data event yields for all three years and combined. Statistical
uncertainty on simulated event counts is given. For an estimate of
systematic uncertainty, see Figs. 8.5-8.7. . . . . . . . . . . . . . . . . . 96

B.1 A comparison of results in the dual models . . . . . . . . . . . . . . . 167

 xvi

List of Figures

2.1 Left: Feynman rule for QED diagrams. Feynman diagrams are read
with time flowing left to right, with e representing an electron and γ
a photon. When the arrow attached to the e line is overall moving
leftwards, it is interpreted as a positron ē as indicated here. Right:
a simple process of two electrons interacting via the exchange of a
photon. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.2 Three diagrams are shown in the perturbative expansion for electron-
electron scattering. The full process has two electrons ingoing and
outgoing–but we do not know exactly what happens in between, as
indicated by the grey circle. . . . . . . . . . . . . . . . . . . . . . . . . 15

2.3 A visualization of the Higgs potential, restricted to the one compo-

nent of the doublet φ (courtesy of John Ellis, CERN) . . . . . . . . . . 22

2.4 A visualization of the standard model of particle physics, courtesy of

CERN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.5 Examples of weak interactions involving particles which come to-

gether in SU (2) doublets. . . . . . . . . . . . . . . . . . . . . . . . . . 26
 xvii

2.6 A Yukawa interaction vertex between a fermion f and the Higgs

boson H . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.1 A diagram for a top quark loop correction to the Higgs propagator,
which will affect the Higgs mass renormalization. Because the Higgs
boson couples to all massive fermions, it is influenced by many such
diagrams. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4.1 Two approximately generated Brownian paths W (t) with the same
endpoint, representing sample paths in a path integral. . . . . . . . . 44

4.2 The orbit in the R1 ˘R2 plane (left) and the evolution of R3 over time
(right). This sample solution is plotted for 0 < t < 21 and uses
parameters k = 1, λ = 2.4, and m = 0.5. The initial conditions are
chosen in terms of conserved quantities of the solution set, and are
detailed [32]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

6.1 Example diagrams for leading order tt̄ production via qq and gg
processes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

6.2 Fixed-order theory calculations of dσtt̄ /dMtt̄ are shown at LO, NLO,
√
and NNLO in QCD for proton-proton collisions at s = 8 TeV and a
top mass of mt = 173.3 GeV, which was a popular choice for simu-
lation until recently. This figure is taken from Ref. [40] by Czakon,
Mitov, and Heymes. Uncertainty bands are based on varying the
renormalization and factorization scales, µr and µf , respectively. . . . 65

6.3 A tt̄ pair decaying into a dilepton final state . . . . . . . . . . . . . . . 67

 xviii

6.4 Left: ellipsoidal constraints on neutrino momentum, whose intersec-

tion defines an ellipse. Right: The elliptic constraint projected into
the transverse plane. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

6.5 combination of 2 ellipses in the dilepton case . . . . . . . . . . . . . . 70

6.6 We show the frequency with which the different kinematic con-
straints can be satisfied, comparing the case where the two b jets
are paired with the correct leptons against the case where they are not. 71

7.1 A map showing the layout of LHC. The primary experiment at each
of the 4 bunch crossings is labeled, but there are also smaller experi-
ments associated with these points. . . . . . . . . . . . . . . . . . . . . 74

7.2 a 3-dimensional rendering of the CMS detector is shown, indicat-

ing the scale of the detector and the overall layout of the internal
components with respect to the beam pipe. . . . . . . . . . . . . . . . 76

7.3 A cross sectional slice of the CMS detector is shown with example
particle tracks demonstrating the function of each detector layer . . 77

7.4 A longitudinal slice of the silicon tracker showing the coverage in η.

TIB, TID, TOB, and TEC are sections of the strip tracker. . . . . . . . . 80

7.5 A cross sectional slice of the full detector showing the coverage in η
of the muon chambers as well as other components. . . . . . . . . . . 80

8.1 tt̄ production from qq processes with virtual exchange of a vector or

scalar boson Γ, such as the Higgs . . . . . . . . . . . . . . . . . . . . . 87
 xix

8.2 Effect of the weak corrections on tt̄ differential cross sections for
different values of Yt , after reweighting of generator-level events
from POWHEG +P YTHIA 8. . . . . . . . . . . . . . . . . . . . . . . . . 88

8.3 Examples of background processes for tt̄ dilepton decays are shown.
The left diagram shows a Drell-Yan process involving a Z boson
decaying to a lepton + anti-lepton pair, while the right diagram
shows an example of single-top production. . . . . . . . . . . . . . . . 94

8.4 Effect of the EW corrections on Mb` and ∆yb` kinematic distributions

are shown at generator level (left plots) and reco level (right plots).
The effect is shown relative to the Standard Model EW corrections,
as this is the best way to assess our sensitivity to different Yukawa
coupling values, rather than comparing to a sample with no EW
corrections. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

8.5 2016 Data to MC comparison, with the shaded region indicating

known uncertainty. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

8.6 2017 Data to MC comparison, with the shaded region indicating

known uncertainty. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

8.7 2018 Data to MC comparison, with the shaded region indicating

known uncertainty. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

8.8 Binned data and simulated events prior to performing the statistical
fit. The dashed line divides the two |∆y|b` bins, while the horizontal
axis shows the bins’ Mb` ranges. . . . . . . . . . . . . . . . . . . . . . 103
 xx

8.9 The effect of the yukawa parameter Yt on binned reconstructed events

is shown for two discrete choices of Yt . Data from the full Run 2
period is combined in the kinematic binning. . . . . . . . . . . . . . . 105

8.10 The weak correction rate modifier Rbin

W in two separate [Mb` , ∆yb` ]

bins from 2017 data, demonstrating the quadratic dependence on

Yt . All bins will have an increasing or decreasing quadratic yield
function, with the steepest dependence on Yt found at lower values
of Mb` . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

8.11 The marginal effect of the EW corrections is shown at two values of

top Yukawa coupling. The colored lines show the nominal value of
the effect, while the shaded envelope indicates the included uncer-
tainty on the EW corrections themselves. . . . . . . . . . . . . . . . . 110

8.12 Shape templates are shown for dominant uncertainties in the fit,
associated with the final-state radiation in P YTHIA 8, the jet energy
corrections, the factorization scale, and the renormalization scale. The
shaded bars represent the raw shape template information, while
the lines show the shapes after smoothing and symmetrization have
been applied. In the fit, the jet energy corrections are split into 26
different components, but for brevity only the total uncertainty is
shown here. Variation between data-taking years is minimal for each
of these uncertainties, so their effect is shown on the full dataset in
the kinematic binning. They are treated separately by year in the
actual fit, to ensure accurate modelling. . . . . . . . . . . . . . . . . . 112
 xxi

8.13 Nuisance parameter post-fit constraints and impacts are shown for
the 30 uncertainties in the fit with the largest impact on the best fit
value of Yt . This information is displayed for an Asimov dataset
generated with Yt = 1 (expected) and for the final fit result on data
(observed). Impacts are calculated by repeating the fit with indi-
vidual nuisance parameters varied up and down by one standard
deviation according to the their post-fit uncertainty and held fixed.
Some uncertainty sources are split into a correlated nuisance param-
eter and an uncorrelated parameter unique to each data-taking year.
Uncertainty on corrections to jet energy is broken down into several
sources, whose names are indicated in italics. Technical details on
the meaning of each such component can be found in Ref. [71] . . . . 114

8.14 The agreement between data and MC simulation at the best fit value
of Yt = 1.16 after performing the likelihood maximization. . . . . . . 115

8.15 The result of a likelihood scan, performed by fixing the value of Yt

at values over the interval [0, 3.0] and minimizing the negative log
likelihood (− ln L). Expected curves from fits on simulated data are
shown produced at the SM value Yt = 1.0 (red, dashed) and at the
final best-fit value of Yt = 1.16 (blue, dashed). . . . . . . . . . . . . . . 115

A.1 The behavior of ρ2 (r ) and ρ(r ), setting s = 1. . . . . . . . . . . . . . . 128

 xxii

A.2 We show an approximate φ and two averages φ̄Λ,s where Λ = 4000,

L = 5 and we average over width s = 0.05 (blue/dark) and s =
0.2 (orange/light). We see that despite the factor of 4 difference in
averaging windows, the two appear interchangeable, demonstrating
the scale invariance even for this approximate representation. The
two appear to follow the same law, and the continuity is seen to be
greatly improved. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

B.1 The orbit in the R1 -R2 plane (above) and the evolution of R3 with
time (below). The sample solution is plotted for 0 < t < 21 and uses
parameters k = 1, λ = 2.4, C1 = 0.5, C2 = 1, s = 2, m = 0.5. . . . . . 159

C.4 Data to MC simulation comparison is shown for jet multiplicity, pmiss

T ,

lepton pT , and b jet pT . Uncertainty bands are derived by varying

each known uncertainty source up and down across the full data-
taking run, and summing the effects in quadrature. Here the signal
simulation is divided into the following categories: correctly recon-
structed events (tt̄ correct reco), events with correctly reconstructed
leptons and b jets but incorrect pairing of b jets and leptons (tt̄ b
jets swapped), events with incorrectly reconstructed b jets or leptons
including non-dilepton tt̄ decays (tt̄ wrong reco), and a separate cat-
egory for incorrectly reconstructed dilepton events which have a τ
lepton in the final state (tt̄ τ). . . . . . . . . . . . . . . . . . . . . . . . 180
 xxiii

List of Symbols and Abbreviations

Symbols and abbreviations are defined as they appear in the text, and a few symbols
are used differently in multiple contexts as we survey different fields and eras. We
collect some of the most commonly used notation here, for convenience.

E Energy
x position (usually as a Lorentz 4-vector)
∂
∂µ ∂x µ
p momentum
mf mass of fermion f
gf Yukawa coupling of fermion f
α electromagnetic coupling
αS strong interaction coupling
φ scalar field
ψ fermionic field
QFT Quantum Field Theory
QED Quantum Electrodynamics
QCD Quantum Chromodynamics
EW Electroweak
LHC Large Hadron Collider
CI Confidence Interval
 1

Chapter 1

Introduction

What is this dissertation about, and why should you

care?

Those are both good questions. The goal of this document is to take the author’s
graduate research, which often addresses hyper-specific questions encased in jargon,
and present it in the broader context of current physics knowledge. The research in
question involves both experimental and theoretical projects in an area frequently
labeled high energy physics, a subset of particle physics. In a sense, this branch of
physics aims to answer fundamental questions involving the smallest known pieces
of the universe. The essential nature of these fundamental particles becomes increas-
ing evident at smaller distances and higher energies–which is why the theoretical
scope of high energy physics is quite broadly defined. On the experimental side,
the label is more commonly reserved for experiments at particle colliders, such as
the Large Hadron Collider (LHC) near Geneva, Switzerland.
 2

At its heart, all the research involved in this dissertation shares the common
goal of better understanding these fundamental particles and the rules that govern
them. To contextualize these studies, we will try to summarize the current state
of the field and explain the lingering questions that these research projects seek to
answer. We are guided by a rather bombastic question: What is the universe made
of at the most fundamental level? The author hopes that anyone with an interest in
this question will find something of value in this document, even if the research
involved reveals only a hint of a tiny piece of an answer.

1.1 What’s in This Document

We begin in Chapter 2 with an overview of modern particle physics, followed by a

discussion of open problems in the field in Chapter 3. The content of these chapters
will already be familiar to anyone with experience in particle physics, so they are
written with a non-expert in mind as much as possible, while still covering the
necessary material. We then proceed to discuss theoretical avenues of research in
Chapter 4, including an overview of theoretical research completed by the author
in the early years of doctoral study, rooted in the detailed mathematical study of
quantum fields. Next we set our sights on experiment, giving a brief overview
of experimental avenues in modern particle physics in Chapter 5. We focus in
on the role of top quarks in particle experiments in Chapter 6, before introducing
specifically the LHC apparatus and CMS experiment in Chapter 7. Lastly, we
present a summary of the author’s final doctoral research project, a measurement of
the top quark Yukawa coupling, in Chapter 8.
 3

Effort has been made to keep the majority of this thesis as accessible as possible
to a wide scientific audience, to the best of the author’s ability. A majority of
technical details are confined to Chapters 4 and 8, though other sections may delve
briefly into more precise language to offer clarity to the expert reader while offering
references to the non-expert. It is the author’s hope that the core concepts behind
the research motivation and conclusions are presented as plainly as possible, such
that it is not necessary to understand all technical details in order to appreciate the
bigger picture.
 4

Chapter 2

A Primer on Particle Physics

“From time immemorial, man has desired to comprehend the complexity

of nature in terms of as few elementary concepts as possible.”
— Abdus Salam

2.1 The Big Picture

It is often said that fundamental physics has two pieces which are at conflict with
one another: quantum physics and general relativity. The former describes the
smallest building blocks of the universe, while the latter explains its behavior at
the largest scales. This thesis is concerned with the former, but it is useful to have a
broad idea of how the two contrast with one another.
 5

General relativity provides a unified framework for explaining gravitational

interaction at large scales. It superseded Newton’s theory of gravity, which already
did a fine job at explaining most planetary and lunar orbits, along with earthly
phenomena such as projectile motion and blocks sliding down frictionless ramps.
Meanwhile, general relativity gave us the big bang theory, explaining much about
the universe’s origins and its behavior on large scales. It even predicted the possibil-
ity of extreme objects such as black holes, which have now been observed through
a variety of means.

But general relativity is a theory which operates at almost incomprehensibly dif-

ferent scales from those of particle physics. And rather than provide a mechanism to
explain gravity on small scales, particle physics ignores it completely. Our current
understanding of particle physics fails to provide a small-scale origin for gravi-
tational force, and breaks down upon attempt to forcefully include gravity. This
discrepancy persists in part because it remains essentially impossible to probe the
small scale nature of gravitational interaction in a lab setting. Other forces typically
render gravity obsolete at small scales. Only a few extreme cases directly require a
simultaneous description of quantum physics and gravity, such as the centers of
black holes or the very moment of the big bang. These are of course inaccessible to
direct measurement, and the energies involved are almost unfathomable.

Since these extreme scenarios are so incredibly far away from our lab capabilities,
one might ask why particle physicists bother to build large machines which operate
at a fraction of the energy scale. Luckily, this fundamental contradiction is not the
only mystery remaining in particle physics. While General relativity provides a
streamlined and unified framework for all large scale interactions in our universe,
our model of particle physics is much messier and ad-hoc. It almost begs to be
 6

studied further, presenting such apparent disorganization that one immediately

suspects something to be amiss. And so, within particle physics, there are a number
of avenues that could point to breakthroughs in our understanding of fundamental
physics. To understand these avenues of research, we present a summary of our
current best understanding of particle physics, starting with its predecessors.

2.2 When Things Got Weird: Quantum Mechanics

In the early 20th century, the development of quantum mechanics radically shook
humanity’s understanding of the behavior and composition of matter at small
scales. Evidence-based models for atoms had gained prevalence in the century
prior, but patterns in their structure and behavior remained mysterious. The turn of
the century brought the discovery of electrons and protons, which provided more
fundamental and universal pieces of matter. The idea of all matter consisting of
very small discrete pieces was not new, but the strange behavior of the pieces came
as a surprise.

Before quantum mechanics, tiny pieces of matter were thought of as particles

which behaved like tiny billiard balls, while seemingly separate phenomena like
light ( electromagnetic radiation) were modeled strictly as waves. Einstein’s study of
the photo-electric effect [1] showed that in fact light behaves at times like a particle,
and transmitting energy not continuously but rather in discrete pieces or quanta.
This was the first hint that light itself is made up of many small particles, now
called photons. On the other hand, the famous double slit experiment (though not
demonstrated until the 1961) showed that matter particles like electrons could in
fact behave like waves to produce interference pattern. [2, 3]
 7

In fact, at the smallest scales, matter and light were comprised neither of particles
nor waves in the usual sense. Instead, small-scale physics could only accurately be
modeled by strange objects with a mixture of particle-like and wave-like properties.
This was the dawn of quantum physics. The study of this duality and its impli-
cations has proceeded alongside the development of more advanced models of
particle physics, and continues to this day. In terms of casual description, the term
particle stuck, probably because it conveys their quantized nature more intuitively.
And so today, we build particle colliders–though a pedantic individual could have
dubbed them wave colliders instead.

In quantum mechanics, particles like electrons, protons, or photons are the

fundamental objects. By including the known "shape" of the electrostatic force
F, which decays over a distance r as F ∝ 1
, this theory was able to model what
r2
happens when you take trap a collection of electrons in "orbit" around a positively
charged nucleus made of protons and neutrons. Such a collection of fundamental
objects stuck together by interaction is called a bound state. As a novel consequence
of the theory, electrons orbiting a nucleus were restricted to certain discrete values of
orbital energy and angular momentum. In describing these energy levels, the theory
successfully predicted the structure of the periodic table, and many properties of
the elements therein. For example, this could be used to explain the energies of
emitted photons from many elements, though it did not explain the physical law
that allowed a photon to be emitted in the first place. And this was a hint of what
the theory lacked–it knew that particles could sometimes be created seemingly out
of nothing, as long as energy was conserved. But why this was allowed remained
somewhat mysterious.
 8

In a sense, breaking down quantities like energy into small, finite quanta is at the
heart of quantum mechanics. Of course, the full story is quite a bit more complicated,
and the "true meaning" of consequences is still fiercely debated (especially by
physicists who have run out of more precise research problems to solve). Quantum
physics is inherently probabilistic, a fact which will carry over to modern particle
physics. Second, and instrumental this probabilistic doctrine, is the importance of
operators.

In quantum mechanics, observable quantities are described by mathematical

operators. All particles exist in some quantum state, often written as |ψi. We don’t
think of of observables like position x or momentum p as being simple properties
belonging to this particle; rather, they are promoted to operators denoted x̂ and
p̂ which act on the particle’s quantum state. Probabilistic calculations can be be
performed, such as the expectation value of an observable like x or x2 for a particle
with state |ψi. These would be written as hψ| x̂ |ψi, and hψ| x̂2 |ψi, respectively.
Because the particle represented by |ψi has a distribution of possible measured
x-positions, note that hψ| x̂ |ψi2 6= hψ| x̂2 |ψi. Rather, the two related to calculating
the mean and variance of a statistical distribution. We mention these calculations to
compare with field theory later on.

2.3 The Leap to Fields

Quantum field theory (QFT) is often described as the marriage of quantum mechan-
ics with Einstein’s theory of special relativity. Although many of the new ideas
presented by QFT are a consequence of joining these two theories, it is easy to
understand them without discussing special relativity much at all.
 9

A particle is typically thought of as an object that exists at a specific point in

space at a given time, while a field is something that has a value at each point
in space simultaneously at a given time. The values at each point can be scalars,
vectors, or more exotic objects. Classical field theory was often used to discuss
gravitational and electromagnetic fields, which carry forces felt by classical objects
and even by particles in the context of quantum mechanics. The simplest type of
field to imagine is a scalar field φ : Rd 7→ R. The object φ assigns a real number to
each point in d-dimensional space. Simple attempts to place a quantum scalar field
into relativistic spacetime yield the Klein Gordon equation,

∂2
2
φ( x ) − ∇2 φ( x ) + m2 φ( x ) = 0. (2.1)
∂t

∂2 ∂2 ∂2
Where ∇2 = + + . Here we are writing in so-called natural units, where
∂x2 ∂y2 ∂z2
the speed of light c = 1, to remove excess coupling constants. Constants like mass
the m remain, as well as other field coupling constants governing interactions, as
we shall see.

To shorten things, in field theory we move to relativistic notation. Here x will

represent the spacetime coordinate vector x = ( x0 , x1 , x2 , x3 ), identifying x0 with
the dimension of time t. Greek letter indices such as x µ are used to pick out the
∂
µ-th component of x. We use the shorthand ∂µ = ∂x µ
. Indices can appear as both
upper and lower indices, and are implicitly summed over when repeated. Lastly,
the upper and lower indices are related by contraction with the Minkowski metric
tensor η. For any reader not well-versed in the language of relativity, these details
are not too important–we will not be performing in depth calculations, but mostly
discussing the meaning of equations written with this notation.
 10

In relativistic notation the Klein Gordon equation can be written more compactly
as

∂µ ∂µ φ + m2 φ = 0, (2.2)

Where we have not written the explicit dependence φ( x ) as this dependence is

assumed for all field objects. (2.2) follows directly from an object called a Lagrangian
density L , given by

L = ∂µ φ∂µ φ − m2 φ2 . (2.3)

In field theory, basic properties of fields can be read off from the Lagrangian density
(often just called "the Lagrangian") with minimal effort. This particular Lagrangian
has two terms: first the kinetic term ∂µ φ∂µ φ, also sometimes called the free field
term. Such a term is present for all types of fields, and represents the field’s kinetic
energy. Unsurprisingly then, the remainder is called the potential term, and encodes
the field’s potential energy. In particular, a term of form m2 φ2 is called a mass term.
The mass term, in brief, tells us that the field has some internal energy stored as
a mass m. If we quantize this field, the meaning of the mass term becomes more
interesting.

In QFT, all fields have a ground state or vacuum state with the lowest possible
energy. The excited states of a field with a mass term are particles with mass m.
This encapsulates one of the biggest revolutions brought by QFT: rather than being
taken as immutable objects dropped into force-carrying fields, particles themselves
are excited states of quantum fields. The number of particles is promoted to quan-
tum operator, through a process known as second quantization. In fact, all fields
 11

become operator-valued upon quantization, though this is not generally designated

explicitly. The details of this procedure are beyond the scope of this dissertation,
but the consequence is simple and groundbreaking: particles can be created and
destroyed within the theory.

2.4 Gauge Theories and Particle Interaction

One of the earliest successful models in QFT came from the Dirac equation, describ-
ing a more complicated type of field which assigns an object called a spinor to each
point in spacetime. Spinors describe quantum objects with the attribute of spin,
thought of as an intrinsic angular momentum. The Dirac equation describes a field
ψ with spin 1/2, a type of fermionic field. In standard relativistic notation, the Dirac
equation reads

(iγµ ∂µ − m)ψ = 0, (2.4)

where γµ are the famous Dirac matrices which make possible the relativistic invari-
ance of a spin-1/2 field. The equation follows from the Lagrangian density

L = i ψ̄γµ ∂µ ψ − mψ̄ψ, (2.5)

where ψ̄ = ψ† γ0 is called the adjoint spinor. Though it looks a bit different than
the Lagrangian in (2.3), it has the same fundamental pieces: a kinetic term and a
mass term. The Dirac equation was meant to describe electrons, and had a great
success in predicting the existence of anti-particles called positrons. However, it is
only a starting point for the gauge theories that give us our modern understanding
 12

of how particles interact to generate forces. Examining the Lagrangian, we see it

has a global symmetry under the transformation

ψ → eiα ψ. (2.6)

In the language of group theory, this is a U (1) symmetry–the same mathematical

structure as a symmetry under rotation, though the rotation does not occur in
physical space. If the field is altered globally throughout spacetime by a phase α,
nothing changes. But in relativity, the field cannot communicate instantaneously
from one point to another. So, one could ask how the field “knows" to have the
same phase at all locations. Though this sounds like philosophical nitpicking, there
is a proposed solution with profound consequences. We can promote the global
symmetry to a local symmetry, one which depends on the spacetime coordinates
x. The new problem is, in making α a function of spacetime coordinates α( x ), we
break the global symmetry. Separating out a constant scaling q for later use, and
applying the chain rule, we now have the transformation

ψ → eiqα( x) ψ (2.7)

∂µ ψ → eiqα( x) ∂µ ψ + iqeiqα( x) ψ∂µ α( x ) (2.8)

How can we keep the Lagrangian invariant under such a transformation? We

introduce a new field Aµ , called a gauge field, which transforms under a local
gauge change according to the rule

Aµ → Aµ + iq∂µ α( x ). (2.9)
 13

This field can have the effect of cancelling out the unwanted extra term in (2.8),
effectively enforcing local gauge invariance. We can now construct the locally gauge
invariant Lagrangian,

1
LQED = i ψ̄γµ ∂µ ψ − mψ̄ψ − ψ̄γµ qAµ ψ − F µν Fµν , (2.10)
4

Where Fµν = ∂µ Aν − ∂ν Aµ is introduced to define a gauge-invariant kinetic term

for the new field. The old fermion field ψ still has its kinetic and mass terms. The
new gauge field Aµ is a bosonic field, and has a kinetic term but no mass term.
Most interestingly, there is a new term containing both fields: an interaction term,
modulated by the constant q. Taken as a whole, this Lagrangian describes a massive
fermionic field (whose excitations are electrons) interacting with a massless gauge-
boson field. The excitations of this gauge field are in fact photons. Our construction
of F µν matches the electromagnetic tensor from classical electrodynamics, and the
free gauge term 14 F µν Fµν is the vacuum term from the Lagrangian formulation of
Maxwell’s equations.

In essence, the imposition of local gauge invariance yields a Lagrangian for a

fully interacting field theory, granting our modern understanding of photons as
gauge bosons that mediate the electromagnetic force on the quantum level. This
particular gauge theory is called Quantum Electrodynamics, or QED for short. The
Lagrangian can also be written more compactly in terms of a gauge-covariant deriva-
tive Dµ , which absorbs the gauge field interaction term to create derivative operator
which itself transforms appropriately under local gauge change:

1
LQED = i ψ̄γµ Dµ ψ − mψ̄ψ − F µν Fµν , (2.11)
4
 14

e
e e

γ
γ

ē e e
Figure 2.1: Left: Feynman rule for QED diagrams. Feynman diagrams are read with time
flowing left to right, with e representing an electron and γ a photon. When the
arrow attached to the e line is overall moving leftwards, it is interpreted as a
positron ē as indicated here. Right: a simple process of two electrons interacting
via the exchange of a photon.

Dµ = ∂µ − iqAµ . (2.12)

From the QED Lagrangian, we are able not only to read off the basic properties of
these fields, but also the specific interaction rules. While we won’t explain here
how to perform QFT calculations, it was thankfully noted by Richard Feynman that
these calculations can be codified as little doodles–now called Feynman diagrams.
The doodles are built from individual vertices or Feynman rules, and QED has
just one, shown in Fig. 2.1 (left). The fundamental rule can be thought of as
demonstrating the possibility for electrons to emit a photon, or for a photon to
produce an electron-positron pair, or for electrons and positrons to annihilate into a
photon–all depending how it is rotated. Using these rules, we are able to visualize
the physical processes that this model allows, and an example is shown in Fig. 2.1
(right).

We pause here to note that the deceptive simplicity of Feynman’s doodles hides
some ugly truths about the theory. They are in fact not exactly accurate depictions
of physical processes in the theory–they are terms in a perturbative expansion.
 15

γ e e
e e e e
γ

γ + γ + e + ···

e e γ
e e
e e

e e

=
e e

Figure 2.2: Three diagrams are shown in the perturbative expansion for electron-electron
scattering. The full process has two electrons ingoing and outgoing–but we do
not know exactly what happens in between, as indicated by the grey circle.

Like polynomials in a Taylor series, they each offer us only approximate glimpses
of a true quantum field, translating an incomprehensible sea of interactions into
individual approximations which can be understood intuitively and calculated.
This means also that not all the particles in the Feynman diagrams are real. The real
particles exist in the initial and final states of the diagrams. Anything created and
destroyed in between is called a virtual particle. Virtual particles are not exactly real
and not exactly fake–they are approximations.

For example, Fig. 2.2 shows multiple diagrams for electron-electron scattering
in QED. Virtual particles are seen intersecting in closed paths, called loops. The
more loops and vertices a Feynman diagram has, the harder it is to calculate. Often,
diagrams with two or three loops are already at the cutting edge of evaluation.
But the real process of electron-electron scattering is a combination of all these
possibilities.
 16

The accuracy of perturbation theory relies on the electromagnetic coupling

√
strength α ∝ q. Each vertex in a QED Feynman diagram carries a factor of α for the
purposes of most physical calculations (discussed further in Chapter 6). Different
interactions have different coupling strengths. The more vertices where particles
interact that are present in a diagram, the more factors of α will be associated to it.
Furthermore, in QFT, these couplings vary with the energy scale at which they are
evaluated–they are called running couplings.

Despite the caveats, we are able to approximate many high-energy particle

collisions with perturbative calculations in gauge theories where the coupling
remains small, and the results agree with nature to a perhaps surprising extent.
Luckily, for the electromagnetic interaction, α is generally quite small from the low
energies of idle electrons (α ≈ 1/137) to all the way up to the energies obtained
in particle colliders, so the series converges quickly. QED has produced some of
the most accurately verified predictions ever made, such as the calculation of the
electron’s anomalous magnetic dipole moment, where experiment has verified the
accuracy of the theory within 1 part in a trillion [4, 5].

2.5 More Gauge Theories: the Nuclear Forces

Gauge theories form the basis of our understanding of particle interaction, giving
rise to what are often called fundamental forces in particle physics. The strongest
of these interactions is called, rather imaginatively, the strong interaction or strong
nuclear force. It is a gauge theory with symmetry group SU (3), describing the inter-
action of quarks through a force mediated by gauge bosons called gluons. Quarks
are the tiny pieces which make up protons and neutrons, and the force between
 17

them is so strong that it prevents them from flying apart due to electromagnetic
repulsion.

SU (3) is a more complicated symmetry group than U (1), with no easy analogy
to visualize. In discussing some such terminology from group theory will inevitably
enter our discussion of gauge theories without proper introduction. For a primer on
the essential role that group theory plays in particle physics, the author recommends
Ref. [6] or [7].

Rather than a conserved scalar charge q as seen in electrodynamics, quarks

carry a three-dimensional analogue called color charge. This has led the interaction
of quarks and gluons to be called quantum chromodynamics (QCD). SU (3) is a
non-Abelian group, meaning the generators do not commute. The Lagrangian at
first looks similar to the QED Lagrangian:

1 a
LQCD = i ψ̄(γµ Dµ )ψ − mψ̄ψ − Gaµν Gµν , (2.13)
4

Where ψ is now the fermionic quark field. The big change comes from the new free
gauge field term. This term is related to the commutator of the group elements, so
it becomes more complicated in the non-abelian case, and we have had to define

a
Gµν = ∂µ Aνa − ∂ν Aµa + g f bc
a b c
Aµ Aν . (2.14)

Here a, b, c index the 8 generators of the SU (3) group that form the corresponding
a
Lie algebra su(3), and f bc are called the structure constants of that algebra. These
indices are implicitly summed over when repeated. The coupling constant for
the gauge boson is now labeled g, i.e. Dµ = ∂µ − igAµ . This form of Lagrangian
generalizes to non-Abelian gauge theories, known collectively as Yang-Mills theory.
 18

In fact, the simple case of QED is the exception and not the rule–most physical
gauge theories are non-Abelian.

While this might seem very technical, in one sense the fundamental change is
simple: (2.14) indicates that gluons themselves carry color charge, and its third
term shows that can interact with each other. This is in contrast to photons, which
mediate the electromagnetic interaction but to not self-interact. From a theory
standpoint, this makes calculations difficult, and many known properties of QCD
are difficult to demonstrate with rigor from the fundamental model.

QCD (along with Yang-Mills theory more generally) has a property called asymp-
totic freedom, which meant that the associated coupling constant αs becomes small at
high energies, and thus and perturbative calculations can describe collisions at parti-
cle colliders up to a point (as we will see in Chapter 6). However, the self-interaction
of the gauge bosons leads to a situation where the theory is strongly coupled over
large distances and therefore at low energies. This means that perturbation theory
and its glorious Feynman diagrams break down entirely.

This is also why we don’t observe quarks or gluons sitting around by themselves
in nature–they quickly coalesce to form bound states. In fact, we don’t observe
them alone at particle colliders either–they form bound states before they even
reach the detector. When a quark and anti-quark bind, it is called a meson. When
three quarks bind, the resulting particle is called a hadron, a class of particles that
includes protons and neutrons. Such bound states can be understood approximately
via effective field theories, but their properties are incredibly challenging to deduce
from the fundamental QCD Lagrangian. Only at high energies, like those achieved
at the LHC, do quarks and gluons start to participate in the more fundamental
interactions where perturbation theory holds.
 19

The strong force can hold together protons in a nucleus against the repulsive
electromagnetic force. Recall that the the standard electromagnetic force falls of
1
with distance r as FEM ∼ . However, as a consequence of non-perturbative effects
r2
and gluon self-interaction, the strength of the strong interaction has an additional
1 −κr
exponential dampening term, FQCD ∼ e , where κ is a constant. Thus, its effects
r2
are felt only at the scale of the atomic nucleus. Such a scaling would be expected by
a force with a massive mediator boson, and indeed the strong force was originally
theorized to be mediated via π-mesons by Hideki Yukawa in 1935 [8]. While this is
not true at the most fundamental level, the strong force effectively has a massive
mediator boson due to the gluons’ inability to travel unperturbed. Lone gluons and
quarks will quickly form bound states such as mesons, hadrons, and hypothesized
massive collections of interacting gluons called glueballs.

However, there is another nuclear force, which is mediated by massive bosons

in a more fundamental sense. Understanding how this can occur was a giant leap
in understanding, laying the foundations for what would become known as the
standard model.

The other nuclear force is the weakest of known particle interactions (not count-
ing gravity which is not yet understood on the particle level), so it is usually referred
to simply as the weak interaction, suggesting that perhaps physicists should con-
sider outsourcing the naming of their concepts. The weak interaction is responsible
for many radioactive decays involving nucleons, electrons, and tiny neutral par-
ticles called neutrinos. Like the strong force, the interaction is confined to the
nuclear scale, which might suggest massive mediator bosons. Furthermore, the
famous Wu experiment [9] in 1956 demonstrated that the weak interaction violates
 20

a symmetry called parity1 , to the shock of many in the physics community. This
essentially means that the interaction does not respect a mirror symmetry, and
meaningfully distinguishes right and left directions unlike the electromagnetic and
strong interactions.

Following the Wu experiment, a satisfying explanation for how a gauge theory

could create the observed behavior of the weak interaction remained somewhat
elusive. The full answer would require not only the unification of the weak and
electromagnetic interactions, but also the introduction of a new scalar field to the
model: the Higgs.

2.6 The Higgs Mechanism and Electroweak

Unification

In the early 1960s, Glashow proposed a combination SU (2) × U (1) gauge theory
as a way to to unify QED with the weak interaction, but reproducing the correct
behavior required a forced breaking of the underlying gauge symmetry. Meanwhile,
several authors (including Peter Higgs) were independently demonstrating that
gauge bosons could acquire mass through a mechanism called spontaneous symmetry
breaking. It was then Weinberg and Salam who put all of this together in the mid
1960s, independently, to build our modern understanding of electroweak theory–
arguably the cornerstone of modern particle physics. The resulting interaction had
three massive mediator bosons, the W+ , W− , and the neutral Z, which obtained
their mass through a natural symmetry breaking process involving a scalar field
1
Shamefully, Chien-Shiung Wu who designed and performed this experiment was left out of the
1957 Nobel Prize, awarded to theorists Lee and Yang who proposed the test.
 21

φ. This model was a triumph of theoretical particle physics; the W and Z bosons
would not be observed until the 1980s, while the direct confirmation of the Higgs
field as a physical entity would come only in 2012 at the LHC. We will try to give a
whirlwind tour of this model.
!
φ+
The Higgs field is a scalar doublet φ = where φ+ , φ0 are complex scalars.
φ0
The Higgs Lagrangian should have the standard form

LHiggs = Dµ φ† D µ φ − V (φ). (2.15)

The covariant derivative Dµ must respect the gauge symmetry of electroweak

unification, and will turn out to be somewhat messy, so let us focus first on the
potential term. In addition to the usual mass term, the Higgs potential contains a
fourth order self-interaction term:

V ( φ ) = µ2 φ † φ + λ ( φ † φ )2 (2.16)

where µ and λ are conventional constants and φ† is the Hermitian conjugate of

φ used to take inner products. In the standard model µ2 < 0 while λ > 0. The
resulting shape of the Higgs potential is such that the ground state (or vacuum state)
occurs when | φ |6= 0. The potential can most easily be visualized by restricting
one component, say φ+ , to be zero, as shown in in Fig. 2.3. The resulting shape is
often referred to as the Mexican hat potential by a bunch of European physicists
who don’t know the word sombrero. The Higgs field illustrates a concept called
spontaneous symmetry breaking. Though the potential shown in Fig. 2.3 has a
rotational symmetry, a ball placed at φ = 0 is in unstable equilibrium, and will roll
down in a random direction to break the symmetry. This principle is now essential
 22

Figure 2.3: A visualization of the Higgs potential, restricted to the one component of the
doublet φ (courtesy of John Ellis, CERN)

in our understanding of particle physics, especially in the context of the universe’s

evolution.

The electroweak sector of the standard model starts out as a standard gauge
theory like QED and QCD, with a gauge group that is a product of two symmetries:
SU (2) I × U (1)Y , which are used to define the gauge covariant derivatives in the
Eq. 2.15. The subscripts I and Y stand for the associated conserved quantities weak
isospin and weak hypercharge associated with these symmetries, to distinguish them
from other standard model symmetry groups such as U (1) symmetry of QED. This
gauge theory should have 4 massless mediators, but it is coupled to the Higgs
field. As we can see, the ground state of the Higgs field spontaneously breaks this
symmetry, giving the field a non-zero value.

This breaks the overall gauge symmetry of the model. Gauge bosons mediating
interactions between electrons and neutrinos acquire a mass term through the Higgs
field’s vacuum expectation value. These include the charged W bosons, and the
neutral Z boson. The Z boson in fact is a mixture of the original SU (2) I and U (1)Y
mediators, as is the photon that mediates the remaining U (1) symmetry. Thus, QED
 23

is what remains after the natural breakdown of a greater symmetry, the process of
which separates out the weak force with its massive force carriers. The fact that
these force carriers are massive dampens the effect of the force over large distances,
leaving its effects confined to the scale of the atomic nuclear force–which is why it
is considered one of the nuclear forces.

It doesn’t really matter which way the ball rolls down the hill, as the minimum
it will find is part of a rotationally symmetric ring. We can see looking at (2.16)
|µ|
that the minimum occurs when | φ |= √ ≡ v, called the vacuum expectation
λ
value. The excited states of this field will thus be perturbations about this vacuum.
Because of the symmetry inherent in V (φ), without loss of generality we can take
the ground state to fall such that the first component of φ is zero and the second
component lies on the real axis. A small excitation of the vacuum state will then
look like

! !
0 0
φ= → (2.17)
v v+H

Upon quantization, this excitation H is the famed Higgs boson, discovered in

2012 at the LHC [10, 11]. The discovery of the particle itself was such a landmark
because it verified the much broader model of the Higgs field and the concept of
spontaneous symmetry breaking.
 24

Figure 2.4: A visualization of the standard model of particle physics, courtesy of CERN

2.7 The Standard Model

The standard model is, essentially, a set of gauge field theories within the framework
of QFT. It is the combination in full of the Electroweak interaction (including
the Higgs field) and QCD, and is often described as one big gauge theory with
symmetry group SU (3) × SU (2) × U (1). A summary of the model, and the particles
it contains, are shown in Fig. 2.4
 25

A few things are worth noting that have not yet been mentioned. Notably, the
standard model contains three generations of matter. Stable electrons have heavier
counterparts in the other matter "generations," called muons (µ) and tau leptons
(τ). While these particles are produced in accelerators and in the atmosphere, they
are fairly short-lived and decay into lower-mass particles (though atmospheric
muons survive long enough that many of them are harmlessly passing through
you and your surroundings as you read this!). Similarly, protons and neutrons are
made of up and down quarks, which are stable. However, two more generations
of heavier quarks exist which can combine into many unstable bound states called
exotic hadrons and mesons. The heaviest quark, and the heaviest particle in the
standard model, is the top quark, which will be discussed later in this dissertation.

Another thing to mention is chirality. Fermions can come with two different
types of chirality, separated by applying the left-handed and right-handed projection
operators PL = 21 (1 − γ5 ) and PR = 21 (1 + γ5 ) respectively, where γ5 = iγ0 γ1 γ2 γ3
is the product over Dirac matrices. Thus, we say that all fermions can be either
left-handed or right-handed. There doesn’t have to be a meaningful difference
in how these two behave, but in the parity-violating weak interaction, they are
separated.

Under the the SU (2) I symmetry in electroweak theory, left-handed fermions

transform as a doublet. Specifically, these doublets are formed by combining up-type
and down-type quarks from the same generation:

     
u c t
 ,  ,  ; (2.18)
     
d s b
 26

e b

W− t

νe W+
Figure 2.5: Examples of weak interactions involving particles which come together in SU (2)
doublets.

or a massive fermion and a neutrino from the same generation,

     
 e   µ   τ 
 ,  ,  . (2.19)
     
νe νµ ντ

These same particles are naturally linked by Feynman vertices, such as those shown
in 2.5. All quarks and leptons can also come with right-handed chirality, in which
case they transform only as a singlet under the action of the SU (2) I , and have
a different value of hypercharge Y. While the two chiralities participate in most
interactions equally, this is not true for interactions involving a W boson, and there
is an even more glaring difference: right-handed neutrinos have not been observed
in nature at all, and are not included in the standard model.

2.8 Mass in the Standard Model

On the topic of chirality, there was another profound consequence of the Elec-
troweak unification, which was not originally intended. When writing our the
 27

full standard model Lagrangian, one obviously expects to see mass terms for the
massive fermions. In the Dirac Lagrangian (2.5), such a term could be added triv-
ially. However, in the full standard model, a mass term would have to include in
equal measure left and right-handed fermions, despite their different transforma-
tion properties. A simple number m cannot fulfill this requirement as it does not
obey the correct transformation symmetries. The only candidate for a mass term
would be a simple linear or Yukawa-type coupling between the Higgs field and all
massive particles. Consider, as an example, the electron. It transforms under an
SU (2) doublet L together with νe , and a right handed singlet R all by itself. The
piece of the standard model Lagrangian containing a Yukawa interaction between
an electron and the Higgs field φ would read

LYuk = − ge ( L̄φR + R̄φ† L) (2.20)

After the symmetry breaking in Eq. 2.16, the neutrino field in L drops out and this
becomes

 
H H
LYuk = − ge (ēL v eR + ēR v eL ) − ge ēL √ eR + ēR √ eL . (2.21)
2 2

This shows that the resulting mass term comes directly associated with a Yukawa
coupling to the Higgs boson H. All massive fermions interact with the Higgs boson
in this way,allowing for Feynman vertices such as that shown in Fig. 2.6. Here the
interaction strength for a given fermion is the Yukawa coupling constant g f , and
we obtain the mass-Yukawa relation

gf v
mf = √ . (2.22)
2
 28

Figure 2.6: A Yukawa interaction vertex between a fermion f and the Higgs boson H

It is sometimes said that the Higgs boson thus gives mass to all particles, though
some would argue that this ignores the subtlety of mass-generation in QCD bound
states such as protons, not to mention the mystery of neutrino masses. Either way,
the generation of lepton and quark mass through the Higgs mechanism is a core
feature of the standard model. It also provides an avenue to test the validity of the
standard model, as coupling strengths and mass are typically measured through
very different experimental techniques.

In one sense, the idea of fermions acquiring mass from the non-zero vacuum
energy density of the Higgs field after spontaneous symmetry breaking is a beautiful
and thought-provoking concept. In another sense, it raises more questions than
it answers. For example, the fermion masses suggest that the Yukawa coupling
values vary by up to 5 orders of magnitude. The top quark, in particular, has gt ≈ 1,
while the electron has ge ≈ 3 × 10−6 .

Furthermore, we find that it is impossible for neutrinos to have mass in the

standard model through this mechanism. While neutrino mass must be very small
based on observation (bounded above by around 10−6 times the electron mass),
they firmly are known to have non-zero mass based on neutrino flavor-mixing
 29

observations. There is no agreed upon way to include such a mass in the standard
model, and it remains an open area of study.

2.9 A Big, Beautiful Mess of Particles

Unlike general relativity, the standard model is not a single elegant stroke of genius,
but rather a mysterious and somewhat haphazard collection of objects. It does not
explain why the masses of most particles are what they are, and has at least 26 free
parameters with no clear explanation. Nonetheless, whatever it lacks in elegance,
it makes up for in reliability. The standard model has outlived many attempts to
concoct a more aesthetically pleasing set of objects, and to this date has accurately
described the results of decades of particle physics experiments.

There are many ways to summarize the standard model, but it is unavoidably
rather complex. It is often said to describe three of the four fundamental forces
of nature, with only gravity missing. But this description really steamrolls the
complexity of the electroweak interaction, and ignores the subtle role of the Higgs
field, which generates arguably an entirely new particle interaction. Furthermore,
particle physicists have tried their best to create images (as in Fig. 2.4) which look
as minimal as possible, sometimes paving over the fact that the weak interaction
has three mediators while gluons come in 8 varieties. Even in this chapter, we
have glossed over some subtleties such as CKM mixing and neutrino flavor mixing,
which add another layer of complexity to the story.

In many ways, the standard model of particle physics resembles the periodic
table, which appeared disorganized and random at the turn of the 20th century,
 30

until the theory of quantum mechanics explained its underlying structures. Will
we ever see a similar revolution, that provides a more elegant explanation for the
structure of the standard model? If humanity can prevent catastrophic climate
change and avoid nuclear war, perhaps such a discovery will one day be made.
 31

Chapter 3

Unsolved Problems in Particle

Physics

“I am a mystery wrapped in an enigma wrapped in a pita. Why the pita?

That counts as another mystery.”
— Demetri Martin

3.1 Things that are Conspicuously Missing from the

Standard Model

In many ways, the tension in fundamental physics is driven by apparent incompat-

ibilities between observations at very small scales and observations at very large
scales. This goes beyond the fundamental incompatibility of quantum physics
 32

and gravity, an issue which becomes evident only when examining singularities
appearing in general relativity related to black holes and the big bang. One need
only to look at the large scale motion of our own galaxy to see that something is
missing in current models.

The rotation curve of the Milky Way, and those of many other spiral galaxies,
were found by Vera Rubin and others to differ from their expected behavior in
the 1970s (see Ref. [12] for a review). The most obvious explanation is that a large
quantity of non-luminous, seemingly invisible matter is floating around throughout
our galaxy without interacting much with ordinary matter. Stars and dust that
we can see appear to feel its pull through gravity, but do not interact with this
mysterious invisible substance, called dark matter. If it interacted by the strong or
electromagnetic forces, we would be able to detect it. Furthermore, dark matter
seems to account for more mass in our universe than ordinary matter.

Since the initial arguments for the existence of dark matter, additional evidence
has been found for this substance through gravitational lensing and cosmological
models. Many believe that dark matter is made up of heavy particles which interact
only weakly, and are missing from the standard model. They could in principle be
heavier than what can be directly produced at modern colliders, and interact rarely
enough to evade direct detection, being observed only by their gravitational pull.
While this is the favored explanation, there could always be something unexpected
at play.

On top of this, another mysterious force exists in the universe at even larger
scales. While general relativity posits that the universe is expanding ever since the
big bang, the rate of expansion seems to be accelerating. This was first observed
by the Supernova Cosmology Project [13] and the High-Z Supernova Search [14]
 33

in 19981 , and implies the presence of yet another mysterious substance pervading
the universe, called dark energy. This force, unlike all others, seems to push massive
bodies apart at large scales.

Curiously, a potential explanation for dark energy is already present within QFT,
in the controversial concept of vacuum energy. This type of energy would exhibit
exactly the right type of behavior, except for one thing: Based on the standard
model, the hypothesized energy density appears too large by at least 100 orders of
magnitude. While it is technically possible for as-yet-undiscovered quantum fields
to yield a partial cancellation, the cancellation would have to yield a result so mirac-
ulously close to zero that many view the vacuum energy as a mathematical oddity
and not a viable cosmic force. Still, it is clear that physics is missing something here.

Lastly, a more tangible problem in the standard model is that of neutrino mass.
Neutrinos are observed to have a miniscule mass relative to all other particles. Yet
they have no mass term at all in the standard model, and one cannot be added in
quite the same way as for other fermions, because neutrinos seem to exist only as
left-handed particles. One interesting possibility is the see-saw mechanism, which
posits that right-handed neutrinos do in fact exist and are very heavy, making
them possible candidates for dark matter. As neutrino physics is an active area of
experiment today, there is hope that theorists may receive some guidance on this
issue.

1
These studies garnered a shared Nobel Prize in 2011. However, Vera Rubin was not awarded a
Nobel prize for her seminal work on galactic rotation curves before her death in 2016, in another
obvious injustice that suggests we should not place so much value in these awards.
 34

3.2 Quantum Fields: Rough Around the Edges

We alluded in Chapter 2 to the fact that QFT is rather messy from a mathematical
standpoint. In QFT, true quantum field interactions involve arbitrarily complex
diagrams which often cannot be calculated even with today’s computing resources.
On top of that, most Feynman diagrams hide integrals that yield infinite results
upon first attempt to calculate anything physical. They must be put through a
process called renormalization, whereby physical quantities are redefined specifically
to avoid divergent quantities. While the process of renormalization has had great
success in making physical predictions, and the formulation of the renormalization
group approach by Wilson in the 1970s [15] sheds some additional light on the
once-controversial process, it has one fundamental issue: it is not mathematically
rigorous. This is not a deal-breaker, as Newton’s calculus was wildly successful for
over a century before being made fully rigorous by mathematicians. Nevertheless,
it sets an obvious direction for needed mathematical work.

While the process of renormalization can be quite technical, the conceptual

issue is simpler: renormalization is used to answer a question in physics when
the original phrasing of the question, resulting from the equations in QFT, has
no apparent solution. Renormalization requires one to redefine the quantities in
the problem until divergences cancel and a finite result can be obtained. This
redefinition means that the original question has not been answered in a strict sense,
creating an ambiguous divide between the equations of the theory and the results
of the theory. For an excellent discussion on the concept of renormalization in a
broad sense, see Arnab Kar’s dissertation [16].
 35

Such an issue is not so different from Newton’s use of infinitessimals in his

formulation of calculus–these objects were not defined in a mathematically rigorous
way, but were used to produce many results nonetheless. However, they start to
come up short when one starts to push calculus to its extreme limits, studying
functions that appear bizzare and physically unintuitive. Unfortunately, most
objects in QFT are bizzare and physically unintuitive, suggesting that they may
benefit from more rigorous mathematical formulation.

Because there are situations where renormalization breaks down (strongly-

coupled theories and gravity to name a few), it will ultimately be important to have
a rigorous understanding of the procedure so we may better understand its limiting
cases.

3.3 The Strongly-Coupled Regime

Despite the caveats of perturbative expansion and renormalization, those proce-

dures take place under conditions under which QFT is still able to thrive. The
theory succeeds when the interaction coupling strength is small, as the perturbative
approach of Feynman diagrams and renormalization converges quickly. Then the
contributions of diagrams die off as they become more and more complex. But this
is not always the case, particularly for the strong interaction.

The phenomenon of running couplings in QFT mentioned in Sec. 2.4 is a direct

consequence of renormalization, causing the apparent coupling strength of an
interaction to vary as a function of the energy/length scale at which it is studied.
In the case of the strong interaction, the coupling is small at very high energies,
 36

allowing us to predict the results of high energy particle collisions involving quarks
and gluons. However, the coupling becomes large at low energies, causing quarks
to join together and hadronize into bound states. QFT, for all its success, struggles
to demonstrate exactly how quarks form bound states, or to explain why the
proton mass is so much higher than its constituent valence quarks. This is because
perturbative calculations and renormalization break down at strong coupling values.
Some success has been seen in modelling these non-perturbative effects numerically
through lattice field theories, for example in calculating meson masses (see Ref. [17]
for a review).

Lattice QCD results, coupled with observations and effective field theories,
generally are believed to provide sufficient evidence for the validity of QCD and
the quark model. Nonetheless, the inability to perform calculations involving
real world particles based on first principles is troubling. The strongly-coupled
regime eludes formal description by the tools that we have. This leaves open the
meaning of other hypothesized field theories, such as those which become strongly
coupled at high energies–can such a theory exist in nature? QED was thought to be
such a theory before EW unification, but the additional symmetry group removed
divergences at high energy scales.

To this day, understanding the low energy behavior of QCD directly ties into
one of the famous Millenium Problems [18]–so you could win $1, 000, 000 if you
figure it out! The relevant Millenium Problem is to demonstrate that Yang-Mills
theories, which include QCD, have a mass gap. Theories with a mass gap do not
have massless excitations. In QCD, this would mean that gluons are somewhat a
ruse, existing only as virtual particles in perturbative diagrams at high energies. All
 37

H H

t
Figure 3.1: A diagram for a top quark loop correction to the Higgs propagator, which will
affect the Higgs mass renormalization. Because the Higgs boson couples to all
massive fermions, it is influenced by many such diagrams.

real excitations of the gluon field would have to be a bound state of many interacting
constituents which acquires a mass. This is widely believed to be the case.

3.4 The Hierarchy Problem

The hierarchy problem can be stated in a few ways. First, we note that at present
there is only one independent parameter in the SM which is not dimensionless
in natural units: essentially, the Higgs mass. The mass of the Higgs boson sets
the energy scale for the electroweak interaction, and arguably for much of the
standard model, though we are ignoring a more subtle scale that arises in QCD
here. Quantum Gravity should have a natural scale on the order of the Planck mass
m pl , 16 orders of magnitude larger than the Higgs mass. What can explain this
enormous difference? In one sense, this question itself constitutes the hierarchy
problem. In another sense, there is a more specific issue at play.
 38

The hierarchy problem can also be more technically posed in terms of the Higgs
mass renormalization. The trouble comes from the fact that the Higgs couples to
other particles via loop diagrams, as in Fig. 3.1. Such diagrams must be included
in the Higgs mass renormalization, and their contribution depends on the mass
squared of the particle appearing in the loop. This means that any heavier particles
we have not yet observed ought to produce large corrections to the Higgs mass.
Indeed, if there exist new physics at energies much higher than the EW scale (which
the Planck mass strongly suggests), it seems impossible that the Higgs mass would
remain comfortably within the EW regime, as an extreme fine-tuning is needed
to balance the larger and larger loop corrections. Such fine-tuning is detested
by particle physicists. In general, it seems possible for nature to possess terms
at different scales that bear little relation, and also for many exact cancellations.
However, incredibly unlikely approximate cancellations are frowned upon without a
good explanation.

The favored resolution to this problem has been supersymmetric (SUSY) models,
in which the additional particles create a cancellation effect above the SUSY energy
scale, removing the need for fine-tuning. However, with no direct evidence of
SUSY, the potential energy scale for SUSY has been pushed ever higher, while the
measured Higgs mass that fits comfortably with current SM data. It is not clear how
new physics could eventually come into play at higher energies without greatly
effecting the Higgs mass, unless these effects are wildly different than the particle
interactions we understand today.
 39

Honorable Mentions

We pause here to note a few more mysteries of the standard model that may pique
the interest of some readers: There is the strong CP problem [19], in which a viable
CP-violating term is conspicuously absent from the QCD Lagrangian, a fact that
has been used to conjecture about hypothetical particles called axions. Another
fundamental problem is the discrepancy between observed CP violation in the stan-
dard model and matter anti-matter asymmetry in the visible universe [20], which
seem incompatible. Lastly, for those interested in early universe cosmology, the lack
of a plausible bridge between the standard model and the model of inflationary
cosmology remains an open and intriguing issue [21].
 40

Chapter 4

Theoretical Paths Forward

“Good books tell the truth, even when they’re about things that never
have been and never will be. They’re truthful in a different way”
— Stanisłav Lem (one could say the same of good theories)

4.1 Maybe More is Better

The standard model acquired its name only after decades of stubbornly refusing to
be supplanted by a more elegant theory. In the 1970s, as the pieces of what is today
known as the standard model were coming together, particle physics seemed to be
progressing at breakneck pace on theoretical and experimental fronts. Theorists
had a good track record solving conceptual issues, and experimentalists still had
some heavy quarks and bosons left to discover. On the theoretical side, even more
bold ideas came about into the 80s and 90s, with grand ambition. There was the
 41

proposal of a new type of symmetry between fermions and bosons, referred to as

supersymmetry (SUSY) which would basically double the number of fundamental
particles from those in the standard model. Following the success of Electroweak
unification, there were attempts unify the three known forces under one higher-
dimensional gauge group and create a grand unified theory (GUT). There was also
the idea of string theory, which was able to propose a renormalizable theory of
quantum gravity, at the cost of positing several hidden spatial dimensions existing
in our universe.

In each case, adding more ideas to the standard model seemed to increase the
elegance of particle physics as a whole, and paint a less haphazard picture of our
universe. There is immense mathematical beauty in each of these ideas, which
helped them to gain popularity. And yet, as of June 2020, none of them have seen
any hint of experimental confirmation.

Theoretical research in the field of of high energy physics can today feel some-
what adrift. It took over 50 years for the Higgs mechanism to be verified exper-
imentally since its theoretical inception. But in advance of the Higgs discovery,
there were very convincing arguments that LHC experiments would observe either
the standard model Higgs boson or some other new physics to compensate for its
absence (for a simple overview of pre-LHC Higgs constraints see Ref. [22], or for
more detail on precision Electroweak fits see Refs. [23, 24]).

Models which would take us beyond the standard model, in particular super-
symmetry, tend to have too many free parameters to make similar predictions
about how and where their effects should first be observed. As we speak, LHC
experiments continue to rule out small portions of endless phase spaces in which
supersymmetry could be detected. Yet, whether they will admit it today or not,
 42

many particle physicists were confident 10 years ago that the LHC would have
already seen strong evidence of supersymmetry by now. Some even viewed this
as the primary purpose of the LHC. The lack of such a discovery to date certainly
does not disprove the model, but it has cast a shadow of skepticism on the model’s
strongest proponents.

Some would argue that supersymmetric models have received too much fo-
cus, dominating the conversation due to their prominent proponents in the field.
Perhaps this has indeed distracted from other possibilities and open issues in the
standard model and QFT more broadly. We will examine some avenues of research
which proceed very differently, simplifying QFT and honing in on its individual
characteristics without proposing bold new paradigms.

4.2 Mathematical Rigor in QFT

QFT has an obvious shortcoming in its lack of mathematical rigor. To get a sense
of what mathematical tools are missing to define a more rigorous results, we will
briefly discuss some issues with the path integral formulation of QFT.

The idea of a path integral predates QFT, and is rigorously defined in a select
few cases. One such case is the famous process known as Brownian motion, whose
formal mathematical description is often referred to as the Wiener process. This
refers to a continuous random walk W (t) which results as the limit of independent
increments. In 1-dimension, W (0) = 0 and we have that W is Gaussian distributed
 43

2
at each point pW ( x, t) ∝ e− x /2t
. It then has a variance

Var[W (t2 ) − W (t1 )] = h|W (t1 ) − W (t2 )|2 i ∝ | t2 − t1 | (4.1)

Using the mathematical tools of stochastic calculus, it is possible to define an integral

that allows us to average a function over all such paths leading from one fixed
endpoint to another, called the Wiener integral and often generalized to the Itô
integral. We could write such an average of a function f , using the standard
physicist stylings, as

Z W (t )= x
2 2
f (W )D[W ]
W (t1 )= x1
hfi = Z W (t )= x . (4.2)
2 2
D[W ]
W (t1 )= x1

where D[W ] indicates that we are integrating over all paths of Brownian processes
W (t). This is a far more complex calculation than ordinary integration from Newto-
nian calculus. By integrating over all possible paths, we must integrate an infinite
dimensional system. Additionally, Brownian motion has famously rough paths–
though continuous, they are nowhere differentiable. A sample of two approximately
generated Brownian paths with the same endpoint is shown in Fig. 4.1, which would
be among uncountably infinite sample paths in a Wiener integral.

In the path integral formulation of QFT, we must perform a further generaliza-

tion of this concept, known as a functional integral. Consider a scalar field φ. The
 44

W
1.5

1.0

0.5

t
0.2 0.4 0.6 0.8 1.0

-0.5

Figure 4.1: Two approximately generated Brownian paths W (t) with the same endpoint,
representing sample paths in a path integral.

necessary integration to perform an average of an observable f is written as

Z
f (φ)e−iS(φ) D[φ]
hfi = Z (4.3)
−iS(φ)
e D[φ]

L dx4 . The mathe-

R
Where the action S is the integral of the Lagrangian density,
matical tools to rigorously evaluate such a functional integral generally do not exist.
Even from a physicist’s standpoint, the main such integral which can be evaluated
is that of a non-interacting field, also called a free field, with only a kinetic term. The
purpose of perturbative Feynman diagrams is, in essence, to relate interacting fields
back to this one solvable case.

To get a sense what is so difficult, we consider a typical field configuration. In

QFT, one of the simplest and most important calculations we can perform is called
the two-point function, hφ( x1 )φ( x2 )i, which is related to the correlation of the field
 45

values at these points. For a free scalar field in 4 dimensions, we arrive at

1
hφ( x1 )φ( x2 )i ∝ (4.4)
| x1 − x2 |2

Such a field φ has a divergent two-point function as x1 → x2 , and is a strange

object to contemplate. How can the expected product of the field value at two
points vanish when the points are far away, yet diverge when they are close? The
answer is they cannot–the stochastic sample configurations we are integrating
over do not actually live in 4-dimensional space. They belong to the realm of
distributions, mathematical objects that only make sense inside of an integral–such
as the well-known Dirac δ-distribution. This is a stark difference to the traditional
wavefunction ψ( x ) in quantum mechanics, whose physicality may be debated, but
which is nonetheless mathematically well-defined in most cases.

This is obviously an area ripe for research, but it is mostly of a mathematical

nature. Because the language of formal mathematical rigor is often completely
different from the language used for physical calculation, research in this direc-
tion generally is classified as mathematical physics, an area populated mostly by
mathematicians. To make any rigorous claims, one must usually work on reduced-
dimensional theories with simple interactions. Thus, the work being done by
mathematicians in the field of mathematical physics is often completely disparate
from that being done in the realm of theoretical physics. The two worlds often
diverge in terms of both interest and notation, making it a difficult divide to bridge.

However, when that divide is bridged, it can lead to wondrous results, such
as Martin Hairer’s work with regularity structures in Ref. [25]. Hairer, a physics
PhD from the University of Geneva turned mathematician, has made substantial
 46

progress generating rigorous solutions to problems in stochastic calculus which

map exactly to quantum field theories, in particular 3-dimensional λφ4 theories.
These are scalar field theories with a Higgs-like self-interaction term, but without
the other machinery of the standard model. This work earned him a Fields medal
in 2014. Perhaps if mathematicians and physicists learn one another’s language
and come together more often, we can expect more progress like this in the future.

4.3 Research Project: Continuity of Log-Correlated

Gaussian Fields

As mentioned, Brownian motion has paths which are in a sense barely continuous.
It is not Lipschitz continuous, meaning that for a Brownian path W (t) the equality

|W (t1 ) − W (t2 )|
<C (4.5)
| t1 − t2 |

does not hold for any C as x → y; the quantity is ubounded. However, it is possible
to find an optimal function ω with respect to which W (t) is Lipschitz continuous
(see, for example, Ref. [26]). That is, there exist C such that

|W (t1 ) − W (t2 )|
<C (4.6)
ω (|t1 − t2 |)

is true with probability 1. Such a function is called a modulus of continuity. One can
think of it as measuring the severity of the divergence of the slope of W (t). The
 47

modulus of continuity for W (t), attributed to Lèvy is

s
1
ω (|t1 − t2 |) = |t1 − t2 | log . (4.7)
| t1 − t2 |

Brownian motion is actually related to a free quantum field in 1 dimension,

which is not much of quantum field at all. All quantum fields in higher dimensional
path integrals have distribution-valued sample configurations, and so are divergent.
The most mildly divergent are free fields in 1+1 dimensions (1 space + 1 time).
These fields have a logarithmically-divergent two point function. We can consider,
for simplicity, a 1-dimensional field with the same correlations–this is of interest in
the mathematical community [27], and is related to the ground state wavefunction
of a 1+1 dimensional free quantum field from a physics perspective. For such an
object, a physicist would write

hφ( x1 )φ( x2 )i ∝ − log | x1 − x2 |. (4.8)

To be more mathematically rigorous, we must recognize that φ is distribution-

valued, and makes sense only when integrated against a test function h where

Z Z
φ[h] = φ( x )h( x )dx, h( x )dx = 0. (4.9)

Then we can say, more precisely, that

Z
hφ[h1 ]φ[h2 ]i ∝ − log | x1 − x2 |h1 ( x )h2 (y) dx dy. (4.10)

Working with S.G. Rajeev, the author performed a study to compare such fields
to Brownian motion by implementing simple test functions [28]. The low severity
 48

of logarithmic divergences allows us to use particularly simple test functions. For

constants s and u, we can define a moving average of φ over an interval of width 2s
centered at su as

Z 1
1
φ̄s (u) = φ (s [u − w]) dw (4.11)
2 −1

There is some subtlety here as test functions must retain mean zero, but we preserve
the basic properties of a moving average. This allows us to have in mind a simple
picture for the output of our field φ against the test function–it is as if we are able
to perform a sliding measurement of the field potential across a finite region. The
output is easily shown to be a function, but whether it possesses continuity is not
immediately clear.

Our test function allows us to define a metric ρ on the field

q
ρ(u, v) = h[φ̄s (u) − φ̄s (v)]2 i. (4.12)

It is the existence of this metric which allows us to compute the modulus of con-
tinuity of our averaged field. After a short exercise outlined in [28], we find the
modulus of continuity to be

1
| x − y| log . (4.13)
| x − y|

This is similar to the standard result for W (t) in Eq. 4.7, without the square root.
Our averaged fields are indeed continuous, and not so different conceptually from
the Brownian paths shown in Sec. 4.2.
 49

The goal of this project was to explore the links between simple divergent
quantum fields and well-studied continuous stochastic processes. We see a straight-
forward example of how a simple choice of test function for our quantum field
creates a more familiar mathematical object with more tractable properties. This
project was a fairly short exercise, and it would be interesting to see further study of
the possible translation from divergent quantum fields to more ordinary functions,
as well as the significance of metrics such as that defined in Eq. 4.12, the idea for
which came from work done by Rajeev and Kar [29].

4.4 Toy Models

It is difficult to convey how challenging it is to perform complex calculations in

Yang-Mills theories that describe the most interesting standard model processes.
Even when perturbation theory succeeds, calculations for physical processes at 3rd
order are in many cases considered cutting edge. Despite the fact that we have
exponentially more computing capability afforded to us than at the time these
theories were developed, we are unable to fully automate many calculations. A
deep knowledge of physical properties and various renormalization techniques is
required to extract meaningful results from complicated Feynman diagrams.

Rather than using mathematical techniques to search for rigorous definitions in

reduced-dimensional models as in Sec. 4.3, we can take a more physics-oriented
approach and study so-called toy models. These are simplified interacting models,
 50

often with reduced dimension d < 4, which can be studied via theoretical physics
methods instead of pure mathematics.

Recall the unsolved conjecture of a mass gap in Yang-Mills theories such as

QCD from Sec. 3.3. Another model, called the principal chiral model, has long been
considered a toy model for Yang-Mills theory. The principle chiral model involves
group-valued field g in 1+1 dimensions with a target space of SU (2) (represented
as matrices).

1
LPCM = − 2
Tr( g−1 ∂µ g · g−1 ∂µ g) (4.14)
λ

Tr( X ) is the trace of a matrix X, and λ a coupling constant. This model is a special
case of so called non-linear sigma models arising in string theory, in which a scalar
field theory takes values on any non-linear manifold called a target space.

The principal chiral model shares the property of asymptotic freedom with Yang-
Mills theory. Most notably, it was shown to have a completely factorizable S-matrix,
meaning that the perturbative expansion in QFT can in principle be evaluated at
arbitrary order [30,31]. This led to calculations indicating that a dynamical mass gap
is generated in the theory, despite the apparent fundamental massless excitations–
exactly as is conjectured to occur in QCD. As such, this toy model provides a proof
of principle for what we expect to occur theories like QCD, but the computations
are unfortunately too difficult to confirm this directly in those higher dimensional
theories.
 51

4.5 Research project: Nonlinear Wave Solutions in a

Strongly-Coupled Toy Model

Working with professor S.G. Rajeev, the author examined a toy model related to
the principal chiral model [32]. We consider a field φ in 1+1 dimensions (t and x),
taking values in the Lie algebra su(2). This can be represented as a 3-component
1
decomposition into the standard Pauli matrices σi : φ = 2i [φ1 σ1 + φ2 σ2 + φ3 σ3 ]. We
can construct a model containing a third-order interaction term with Lagrangian
density

1 λ
Ltoy = ∂µ φ a ∂ν φ a η µν + eabc φ a ∂µ φb ∂ν φc eµν , (4.15)
2 6

where e is the Levi-Civita tensor. If we make the indentification

1 −1 1 −1
∂t φ = g ∂ x g, ∂x φ = g ∂t g (4.16)
λ λ

we will find that the classical equations of motion (following the Euler-Lagrange
equations) are equivalent between the theories. Thus, we say that they are classically
dual theories. However, at the quantum level, the two diverge. The principal chiral
model is known to be asymptotically free, meaning that upon renormalization the
running coupling λ → 0 at high energies. However, our toy model of interest
has essentially the opposite behavior. The quantum theory features a Landau pole,
meaning that λ → ∞ at large but finite energies.

Such theories are not unheard of; pure QED has a Landau pole, which is avoided
in the standard model by the emergence of electroweak unification at far lower
 52

energy. Additionally, λφ4 theory, the theory of a solitary Higgs-like scalar field, has
such behavior in 4 dimensions. It remains unclear if QFT Lagrangians which lead
to a Landau pole can describe a physical theory, and thus whether a Higgs-like field
could have a physical realization without its standard-model interactions.

Because the λ → ∞ or strong coupling limit differs between the theories, we

studied first the classical behavior of the field theories in this limit. The theory
of Ltoy in particular has the seemingly troubling behavior that the speed of light
c → 0 in the strongly coupled limit. Nevertheless we were able to find nonlinear
wave-like solutions, which continue to propagate at non-uniform speeds in the
c → 0 limit, defying naive expectations of the field’s behavior. These solutions were
found as part of a general class of solutions following the ansatz

 
i  1 0 
φ(t, x ) = eKx R(t)e−Kx + mKx, K= k . (4.17)
2  
0 −1

The resulting solutions are analogous (but not equivalent) to plane wave so-
lutions. Here m and k are constants, with wavenumber k. The field φ is seen to
have constant energy density, shifting by an internal rotation and a constant shift
under translation. The formula for the solution R(t) is somewhat involved, and
can be found in the paper. The resulting orbit in the R1 − R2 plane, along with the
solution R3 (t), are plotted for a sample solution in Fig. 4.2. We see that this solution
rotates much like an epicycloid in the R1 and R2 components, while increasing at a
non-constant rate in the R3 direction.
 53

Figure 4.2: The orbit in the R1 ˘R2 plane (left) and the evolution of R3 over time (right). This
sample solution is plotted for 0 < t < 21 and uses parameters k = 1, λ = 2.4,
and m = 0.5. The initial conditions are chosen in terms of conserved quantities
of the solution set, and are detailed [32].

We can restrict our field theory to a finite-dimensional system by studying this

class of wave solutions exclusively. This reduces the problem to a finite-dimensional
mechanical system which can be more easily studied, and the rules of ordinary
quantum mechanics can be applied to the reduced system–such an approach is
called a semi-classical approximation.

In the semi-classical approximation, we found that the reduced system ordinarily

had excitations following a dispersion relation of E ∼ |k |, where E is the energy
density and k is analogous to a wavenumber for the system. This is the typical
dispersion relation for a field with massless excitations.

However, in the strongly coupled limit of the semi-classical model, we found the
unusual dispersion relation E ∼ |k |2/3 , which does not correspond to the spectrum
of any normal particle-like excitation. We hypothesize that these nonlinear wave
 54

solutions could thus represent some exotic excitations which appear at strong
coupling in the full theory.

Such a conjecture is somewhat complementary to what is already well-accepted

to occur in Yang-Mills theories like QCD. In that case, easy to understand excitations
in the λ → 0 limit break down at lower energies where the coupling becomes large.
The fundamental excitations of the Lagrangian cease to exist at a certain point,
yielding objects that would not be expected from perturbation theory–e.g., quarks
and gluons give way to hadrons and mesons. Perhaps the reverse can happen
as well–perhaps theories with Landau poles, where the coupling diverges at high
energies instead, have some exotic excitations which survive in this limit that
physicists have yet to understand.

Following this work, the reduced mechanical system introduced in [32] was
further studied by Krishnaswami and Vishnu, who found a Lax pair and a family
of action-angle variables for the system in [33, 34]

4.6 Summary and Outlook

There are many fundamental problems that remain in theoretical high energy
physics. A common approach has been to add new objects in order to cancel out
some difficulties, rather than to examine in detail the minutia of the underlying
framework of QFT. For a long time, similar strategies guided the theory to correct
conclusions in advance of their experimental discovery, such as with the Higgs
model. However, with no exciting experimental results to confirm any grandiose
extensions of the standard model, it is becoming harder to believe that they exist
 55

as originally formulated. Subjects like supersymmetry and string theory have

effectively become matters of faith.

Meanwhile, in the direction of further elucidating our existing theory, problems

like the Yang-Mills mass gap conjecture have stood unsolved for many decades.
Recent progress is hard to assess, as interest and funding have waned somewhat.
There have been many interesting toy models and conjectures over the last 50 years.
It is difficult for a newcomer to sift through them, but any one of these old ideas
could have a sudden resurgence when viewed by the right perspective. Interesting
results in field theory and math can sometimes come seemingly out of nowhere.

In the absence of new experimental guidance for such a long time, theory has
arguably run amok. With that said, the physical utility of QFT as a framework is
irrefutable, and there is interesting work being done on the mathematical side to
better define its axioms and make rigorous its results. Perhaps this is a promising
avenue, hindered only by the fact that mathematical physicists and theoretical
physicists often inhabit different scientific worlds, separated by dense notational
and conceptual barriers.
 56

Chapter 5

Experimental Paths Forward

“Science progresses best when observations force us to alter our preconceptions”

— Vera Rubin

5.1 The Energy Frontier

One obvious path forward is to build higher and higher energy colliders and see
what happens. Particle physics research which takes place at this energy frontier is
often called high energy physics, a rather vague name which aims to set it apart from
other types of particle physics experiment which do not rely on enormous colliders.
The Large Hadron Collider (LHC), discussed further in Chapter 7, is currently the
highest energy collider ever built, and is firmly at the epicenter of experimental
high energy physics today.
 57

It is a guarantee that new physics will emerge eventually, but the energy at which
this will occur has become very unclear since the discovery of the Higgs boson
in 2012. This is a fairly new situation–prior to 2012, there were widely accepted
arguments that if the Higgs boson was not found in a certain mass range upon
turning on the LHC, some other new physics would have to be found around the
same energy energy scale. But in June 2020, there are currently no easily accessible
energy ranges which have an obvious need for new physics. Models for extra
particles and dark matter candidates often have many free parameters that cause
them to become relevant at arbitrary energy scales.

Although discovering new particles is perhaps the most exciting and glamorous
way to advance experimental high energy physics, it is not the only sort of result
the field has to offer, and seems increasingly unlikely in the near future. The lack of
a clear direction for high energy experimental physics pushes us towards an era
of precision measurement, in which every aspect of the existing model and data is
pored over and vetted extensively.

Previously, precision measurements have pointed the way to new physics well in
advance of direct discoveries. For example, the mass of the top quark was predicted
in advance of its discovery by precision fitting in the electroweak sector of the
standard model [23]. However, particle physics at the energy frontier is today a bit
more directionless than in the 1990s, as discussed in Sec. 4.1–we no longer expect to
find new heavy quarks or a long-missing scalar boson. We know that something
is missing from the existing model, but we can’t agree what. Thus, a more broad
approach to precision measurements of standard model parameters is a promising
path forward, as it allows us to scour the data and see if anything in particular is
amiss in our models.
 58

There are many precision measurements being done by experiments at the LHC,
but one obvious target is verifying the expected properties of the Higgs boson. This
direction, specifically with regards to its interaction with the top quark, will be the
final research project presented in this dissertation.

5.2 Other Approaches

Before proceeding into our experimental research, we note briefly for the reader
other exciting avenues of physics research which can crack away at the fundamental
questions addressed in this document.

Neutrino experiments are making slow but steady progress in illuminating the
properties of these often invisible particles. For example, shortly before the writing
of this dissertation, the T2K experiment in Japan published a result regarding the
constraint of CP violation terms involved in neutrino oscillation [35]. Arguably,
neutrinos are the only particles currently seen to disobey the standard model,
because they have mass (albeit very little). In the usual formulation of the standard
model, they are massless. While adding in neutrino mass does not exactly shatter
the standard model into pieces, there is some ambiguity on what mechanism gives
rise to it, and whether they may have heavy unobserved counterparts.

Neutrinos are difficult to detect, and generally we must have trillions of neutri-
nos pass through large liquid-filled detectors to observe even a few. Otherwise, they
simply pass right through everything they encounter. There are many interesting
experiments which are either collecting data now or will begin in the near future,
such as the T2K experiment in Japan and the DUNE collaboration in North America.
 59

These have the potential to revise our understanding of the standard model in
the coming decades. In a completely different approach to study neutrinos, there
are also experimental searches for neutrino-less double beta decay, which could
determine if neutrinos are Majorana fermions and shed light on the problem of
neutrino mass. Overall, this sector of research tends to generate results at a reliable,
steady pace.

Dark matter searches are experiments which try to detect signs of dark matter.
Direct detection experiments can be large tanks of liquid underground, similar to
neutrino detectors, hoping to detect the signature of a new particle passing through.
There are also indirect detection experiments–for example, those that search for
decay products of dark matter annihilation in outer space. Relative to the slow
and steady trickle of information from neutrino experiments, this sector is a bit
of a lottery. At any time, one of them could detect the signature of a dark matter
candidate, completely upending the world of particle physics and surely meriting a
Nobel Prize.

Cosmological experiments are closely linked to particle physics as well–the stan-

dard model governs the evolution of the early. Thus, measurements of the cosmic
microwave background, and perhaps someday the cosmological gravitational wave
background, can have profound implications for dark matter models and particle
physics in general.

These are just a few to highlight the breadth of experimental approaches that
could shed light on the fundamental mysteries of particle physics. There are many
others, from searches for magnetic monopoles, to cosmic ray detection, to the
generation of stable antimatter bound states, to precision studies of radioactive
decay... As the rate of new discoveries from the energy frontier slows, perhaps the
 60

next big breakthrough in particle physics will come from a new type of experiment
which generates little excitement today.
 61

Chapter 6

Top Quarks and Collider Physics

“The top quark was discovered in 1995 and since then the Higgs has
become our obsession, because the standard model was incomplete without
it.”
— Fabiola Gianotti, Director General at CERN

6.1 Why are Top Quarks Interesting?

The top quark is the heaviest known fundamental particle at the time of this dis-
sertation’s writing. It currently has a mass of 172.4 ± 0.7 GeV, a bit higher than the
Higgs boson mass of 125.1±0.14 GeV (the latest values taken from combination
measurements in Ref. [36]). This makes its mass 40 times higher than that of the next
heaviest quark, and over 300,000 times heavier than that of an electron. Its heavy
mass means the top quark is highly unstable, making it the only quark to decay
 62

q t g t
g g

q t g t

g t

g t

Figure 6.1: Example diagrams for leading order tt̄ production via qq and gg processes.

prior to hadronization. This opens up an experimental window into fundamental

physics interactions at the highest energies currently attainable, making the top
quark one of the most useful tools to advance our understanding of the standard
model.

6.2 Top Quark Production at Hadron Colliders

For the purposes of this document, we will focus on top quarks produced as quark
anti-quark pairs. Top quark pair production, or tt̄ production for short, occurs
when a top and anti-top quark are produced together in an inelastic collision. At
hadron colliders, this happens mostly due to strong interactions between quarks
and gluons, as shown in Fig. 6.1. Pair production is generally the easiest way to
observe top quarks, and served as the initial means of discovery for the top quark
at the D∅ and CDF experiments in 1995 [37, 38].
 63

At colliders, the starting point for QFT calculations is a scattering cross section
σ. As the name might suggest, this has dimensions of area, and is often expressed
in terms of the unit Barns b = 10−28 m2 . In classical theory, a scattering cross
section is the size of the area perpendicular to two particle’s motion within which
they must pass to interact and scatter. In quantum theory, it is related to the
probability that two initial state particles interact and product a given final state. For
example, from 2016–2018 the CMS experiment recorded an integrated luminosity
of about Ltot = 137 fb−1 of collision data. The total cross section for tt̄ production
from proton-proton collision at the appropriate energy is calculated to be about
σpp→tt̄ = 840 pb [39]. Multiplying the two, we can estimate that the number of top
quark pairs produced and recorded at the CMS detector during this period was
about Ltot × σpp→tt̄ = Ntt̄ ≈ 108 events.

The basic relationship between Feynman diagrams and cross sections will be
important. While we won’t calculate any in this dissertation, Feynman diagrams
stand in for integrals used to calculate elements in a perturbative expansion of the
scattering matrix. Each vertex of a Feynman diagram carries a factor relating to the
coupling constant governing the relevant interaction, such as αs for QCD diagrams
in Fig. 6.1. The associated matrix element Mi→ f , for initial state i and final state f ,
√
carries a factor of αs for each vertex in the diagram. The cross section follows the
relationship

dσ ∝ | Mi→ f |2 , (6.1)

and so cross section calculations themselves effectively carry one factor of the
coupling αs for each vertex (though things can get more subtle when summing over
initial and final states).
 64

Thus, the diagrams in Fig. 6.1 are said to be of order O(α2s ), which is the leading
order (LO) tt̄ production. Diagrams of order O(α3s ) are then next-to-leading order
(NLO) QCD diagrams, and so on. These higher order diagrams can contain either
loops of virtual particles, or emissions of additional real particles in the final state.

One can also calculate differential cross-sections, to look at the relative probability
of final states as a function of some kinematic variable. This will tell you how many
events to expect as a function of the desired variable. For example, we will be
looking at the differential tt̄ cross section with respect to the invariant mass of the
tt̄ system, dσtt̄ /dMtt̄ . We show a theoretical result for this differential cross section
in Fig. 6.2. We see that pair production begins at threshold mass of 2mt ≈ 345 GeV
and peaks slightly after, trailing off at higher values of Mtt̄ .

6.3 Top quarks: Decay and Detection

Top pair production has a relatively high cross section and leaves unique byproducts
which are easy to distinguish from other processes. When they are produced, top
quarks have an incredibly short lifetime of about 10−25 seconds. Because of this
short lifetime, the top quark is the only quark which decays via a fundamental
Feynman vertex modeled by perturbative calculations. When an unstable particle
decays, each possible decay process has a probability called a branching fraction. To
simplify things further, top quarks in the standard model decay almost exclusively
to a bottom (b) quark and a W-boson, with a branching fraction of over 99%. The
vertex that allows this decay is shown back in Fig. 2.5. We briefly discuss each
byproduct of this decay.
 65

Figure 6.2: Fixed-order theory calculations of dσtt̄ /dM

√tt̄ are shown at LO, NLO, and NNLO
in QCD for proton-proton collisions at s = 8 TeV and a top mass of mt =
173.3 GeV, which was a popular choice for simulation until recently. This figure
is taken from Ref. [40] by Czakon, Mitov, and Heymes. Uncertainty bands
are based on varying the renormalization and factorization scales, µr and µf ,
respectively.
 66

The b quark survives long enough to hadronize, generally forming a B hadron.

This hadron then typically travels a distance on the order of several millimeters
before decaying at a secondary vertex. Like other quarks and unstable particles,
the b quark is ultimately detected as a shower of particles called a jet. Luckily,
information about particles which appear to come from secondary vertices, together
with additional jet kinematics, can be used to distinguish b jets from other jets
with reasonable success [41]. Different algorithms and working points are available
to identify b jets, which have varying rates of sensitivity (the rate of correct iden-
tification fraction of b jets) and specificity (the rate of mis-identification of other
jets). tt̄ decays benefit from having two associated b-jets, which helps greatly in
distinguishing them from other physics processes.

The W boson has two options for decay. First, it can decay hadronically into a
quark anti-quark pair, which both hadronize and are in practice detected as one
or two jets. Secondly, it can decay leptonically into a lepton and neutrino pair. In
this case, the W boson decays into an electron (e), muon (µ), or tau lepton (τ), and
the corresponding neutrino, with equal probability (see the vertex from Fig. 2.5). If
the lepton is an electron or muon, it can be detected and reconstructed with high
quality. Tau leptons decay quickly and can sometimes be reconstructed by their
decay products, though many measurements try to avoid events with taus due to
the added challenge. Meanwhile, the neutrino (ν) escapes undetected, leading to
an imbalance in transverse momentum (pT ) with respect to the beam axis which
contributes to pmiss
T . While this presents a challenge in reconstructing the neutrino

momentum, it also offers another way of distinguishing tt̄ events, as they will
generally have a large value for pmiss
T .
 67

Figure 6.3: A tt̄ pair decaying into a dilepton final state

The process of identifying tt̄ decays is helped by the fact that these decay prod-
ucts will generally have high pT relative to decays of many other physics processes.
The channel with the lowest background rate is the dilepton final state, where both
W bosons decay leptonically. The presence of two b jets, two high pT leptons, and
large pmiss
T makes it easy to distinguish these events and obtain samples of high
purity. However, it does have lower statistics than the other decay channels, with a
branching fraction of only 11% in tt̄ decays.

Because of the high purity and relatively high cross section, tt̄ events offer a
unique ability to measure differential cross sections with high precision. We will
see in chapter 8 how the shape of such measurements can be used to study the
interaction of top quarks with other particles.
 68

6.4 Kinematics of dilepton event reconstruction

The process of two top quarks (tt̄) each decaying leptonically is visualized in Fig. 6.3.
The two undetected final state neutrinos create a challenge in reconstructing the top
quark kinematics. However, if one forces the decays to be strictly on-shell, meaning
that 4-momentum is exactly conserved at each vertex and the mass width of t and
W are ignored, the neutrino momenta can be in theory determined exactly. This
approximation is not exactly valid–recall that intermediate particles in tt̄ decays are
technically virtual. The precise physicality of such particles is a rather philosophical
debate, but what is clear is that they obey conservation of momentum and energy
only in approximation.

If we make this approximation for each top quark decay, conservation of energy
E and momentum ~p at each vertex yield

Et = Eb + EW = Eb + E` + Eν , (6.2)

pt = pb + pW = pb + p ` + p ν . (6.3)

The momenta pb and p` and the corresponding energies are known from measure-
ment. The masses of the top quark and W boson are taken as fixed, entering through
the quadratic relationship

E2 =| p |2 +m2 . (6.4)

With the neutrino’s mass taken to be zero, each vertex of the decay generates a
set of quadratic equations which constrain pν to an ellipsoid in 3-dimensional
momentum space (6.4 left). The neutrino momentum is thus constrained to lie on
 69

Figure 6.4: Left: ellipsoidal constraints on neutrino momentum, whose intersection defines
an ellipse. Right: The elliptic constraint projected into the transverse plane.

the intersection of these two surfaces, which traces out an ellipse. We call this the
mass constraint, and it is not always possible to satisfy, as the two ellipsoids will not
generically intersect. Occasionally, this is due to small deviations from the on-shell
assumption (recalling that intermediate particles are virtual), or from measurement
resolution. More commonly in practice, it occurs in the case of an incorrect event
reconstruction where a jet has been mis-identified or placed at the wrong vertex.

If the mass constraint can be formed, this ellipse can then be projected into
the transverse plane to define a constraining curve ν⊥ , which can be related to
the measured quantity pmiss
T (also known as the MET vector). In the case of semi-
leptonic decays, with only one undetected neutrino, we can simply introduce a
parametrization ν⊥ (θ ) : [0, 2π ] 7→ R2 and find the point ν⊥ (θ ) which minimizes
the distance to pmiss
T .

In dilepton decays, with two final state neutrinos carrying away undetected
momentum, we must repeat this argument twice. This results in two separate
 70

Figure 6.5: combination of 2 ellipses in the dilepton case

transverse momentum constraints, ν⊥ and ν̄⊥ (Fig. 6.5). We wish to find points on
these two ellipses, specified by parameters θ1 and θ2 , such that

ν⊥ (θ1 ) + ν̄⊥ (θ2 ) = pmiss

T (6.5)

This can be rewritten as

ν⊥ (θ1 ) = −ν̄0⊥ (θ2 ) (6.6)

where ν̄0⊥ (θ2 ) ≡ pmiss

T − ν̄⊥ (θ2 ).

This new ellipse maintains the overall shape of ν̄⊥ , but is shifted and rotated
180◦ about its center. The problem then reduces to the intersection of two ellipses.
We call this the MET constraint. The MET constraint can be satisfied if the two
 71

Possible Constraints Satisfied Possible Constraints Satisfied

(correct b-pairing) (swapped b-pairing)

1.5 × 106 1.5 × 106

1. × 106 1. × 106

500 000. 500 000.

none mass mass none mass mass

only & MET only & MET

Figure 6.6: We show the frequency with which the different kinematic constraints can be
satisfied, comparing the case where the two b jets are paired with the correct
leptons against the case where they are not.

ellipses do intersect, which will yield a discrete set of either 2 or 4 possible momenta
for the neutrinos. Because the MET is a difficult quantity to measure in practice,
it is not infrequent that the MET constraint cannot be exactly satisfied even if the
other objects in the event were correctly identified and well reconstructed. In this
case, an approximate solution can be taken from the points of closest approach of
the two ellipses.

We show in Fig. 6.6 the results of examining the two kinematic constraints
when comparing correct and incorrect matching of b jets to leptons in simulated
events. This emphasizes the role of these kinematic constraints as a tool to aid
with proper selection and assignment of b-jets. The reconstruction of tt̄ events
from detected particles can be a non-trivial procedure, particularly in the dilepton
channel. While tt̄ events are easy to identify in this final state which has very low
background rates, it remains difficult to accurately reconstruct the kinematics of
the top quarks themselves due to the undetected neutrinos. The reconstruction
technique outlined here is very sensitive to detector resolution, and deviations from
on-shell behavior of the decays. The kinematic variables used for the measurement
 72

presented in Chapter 8 ultimately do not rely on a full kinematic reconstruction of

the top quarks, but the kinematic constraints outlined here are still used indirectly
to aid in determining which b-jets and leptons are likely to originate from the same
top quark.
 73

Chapter 7

The CMS Experiment

7.1 The Large Hadron Collider (LHC) apparatus

The LHC [42] is the largest and highest-energy particle collider ever constructed,
spanning across the border of France and Switzerland near the city of Geneva.
The main ring, which uses the tunnel from the Large Electron Positron collider
(active 1989-2000), has a diameter of 8.6 km and ranges from 50-175m underground
along its circumference. Its primary function is to store and collide circulating
beams of protons at energies of up to 14 TeV, though it is also used to collide heavy
nuclei during limited data-taking periods. Particle energies are ramped up with
a series of smaller linear and circular accelerators before entering the main ring.
In this document, we will discuss the machine from the proton-proton collision
perspective; for more on the physics resulting from heavy ion collisions at the LHC,
see for example the review [43].
 74

Figure 7.1: A map showing the layout of LHC. The primary experiment at each of the 4
bunch crossings is labeled, but there are also smaller experiments associated
with these points.
 75

Protons circulating in the LHC ring are divided into bunches prior to injec-
tion. Each bunch contains around 1011 protons and is about 30cm long (from
the laboratory reference frame). The main ring has over 2000 bunches circulating
simultaneously when running at its current beam capacity.

There are several experiments designed to detect results of collisions at 4 beam

crossing points, shown in Fig. 7.1. Of these detector experiments, the two largest
are the CMS and ATLAS experiments, both of which are general purpose detectors
aimed at measuring and reconstructing a broad range of collision byproducts and
outcomes. CMS is an acronym for Compact Muon Solenoid, while ATLAS supposedly
results somehow from the words A Toroidal LHC Apparatus (though many scientists
believe that this shortening violates the fundamental rules of an acronym).

The LHC was first turned on in 2008. The first data-taking run for experiments
took place from 2010-2013, where collisions took place at an center of mass energy
of 7 to 8 TeV. After a long shutdown for upgrades and maintenance, the beam
energy was increased to 13 TeV for the second run of data taking, which took place
from 2015-2018. In particle physics, the center of mass energy (total energy of the
√
colliding protons in the rest frame) is denoted by s, due to its relationship to the
traditional Mandelstam variable s. The data-taking period from 2016-2018 was the
second major data-taking effort at the LHC, and this period is referred to collectively
as Run 2. Run 3 is tentatively scheduled to begin in 2021, after which the machine
will begin upgrading to the High-Luminosity LHC (HL-LHC), planned to begin
operating around 2030. This machine, while operating at the same energies, will
produce vastly greater luminosities.
 76

Figure 7.2: a 3-dimensional rendering of the CMS detector is shown, indicating the scale of
the detector and the overall layout of the internal components with respect to
the beam pipe.

7.2 The CMS detector

The detector has 5 main layers, consisting of 4 types of detectors and the solenoid
magnet which is the experiment’s namesake, as shown in Figs. 7.2-7.3. We briefly
outline the detector components here, from innermost to outermost.

1. The tracker is comprised of an inner portion with high-granularity silicon

pixels and an outer level of silicon strips. A particle passing through the
silicon material will create loose electron/hole pairs. A reverse bias voltage
generates an electric field which prevents these pairs from recombining, and
 77

Figure 7.3: A cross sectional slice of the CMS detector is shown with example particle tracks
demonstrating the function of each detector layer

they drift through the bulk of the material to readout sensors where the effect
can be detected.

With the many layers of silicon pixels and strips, we can reconstruct precisely
particle trajectories. Due to the presence of a strong magnetic field, these
trajectories will be curved, and the charge-to-mass ratio of particles can be
estimated. The silicon tracker is essential for estimating the location of the ver-
tex at which particles are produced, which may be displaced from the primary
vertex of a collision. This is essential in b-jet identification, for example.

2. The electromagnetic calorimeter (ECAL) consists of a dense material which

detects electrons and photons, causing them to decay into a well-defined
burst/shower of particles. The energy deposited by this burst can be mea-
sured and used to estimate the energy of detected particles. As the electromag-
netic interaction scales like 1/m2 for particles with mass m, heavier particles
 78

like hadrons/mesons and muons can pass through this layer with minimal
disturbance.

3. The hadronic calorimeter (HCAL) is another calorimeter designed to detect

hadrons/mesons. It can detect charged and neutral particles of this type.
Among standard model particles, only muons and neutrinos can easily pass
through this layer undisturbed.

4. The superconducting solenoid (magnet) is the largest magnet of its kind

ever constructed, producing magnetic field strengths of up to 3.8 T. This
strong magnetic field causes the paths of charged particles to curve based
on their mass and momentum, aiding greatly in particle identification and
reconstruction. This is arguably the central design element of the detector and
its namesake.

5. The muon detector detects only muons, and also contains several layers of
iron return yoke to ensure that any particles other than muons are blocked
before entering this region (excluding neutrinos). While the muons pass
easily through all solid components of CMS, the muon detector primarily uses
small gas chambers to detect them. The muons will knock electrons off of
particles in the gas, creating a detectable signal. Though the spatial resolution
of this type of detector is worse than others, the muons are detected at several
locations and are the most easily identified particle, making them the cleanest
data points in most collisions. The magnet allows their velocity to be easily
computed by the observed radius of curvature of their flight path.

The particle-flow (PF) algorithm [44] aims to reconstruct and identify each
individual particle in an event, with an optimized combination of information from
 79

the various elements of the CMS detector. The energy of photons is obtained from
the ECAL measurement. The energy of electrons is determined from a combination
of the electron momentum at the primary interaction vertex as determined by the
tracker, the energy of the corresponding ECAL cluster, and the energy sum of all
bremsstrahlung photons spatially compatible with originating from the electron
track. The energy of muons is obtained from the curvature of the corresponding
track. The energy of charged hadrons is determined from a combination of their
momentum measured in the tracker and the matching ECAL and HCAL energy
deposits, corrected for zero-suppression effects and for the response function of the
calorimeters to hadronic showers. Finally, the energy of neutral hadrons is obtained
from the corresponding corrected ECAL and HCAL energies. A more detailed
description of the CMS detector can be found in [45].

The standard coordinate frame uses the z-axis as the beam axis, while the x-
axis points inwards towards the center of the ring and the y-axis points vertically
upward. The x − y plane is referred to as the transverse plane, and projections of
momenta p into this plane are labeled pT . Conservation of momentum dictates that
a collision should produce no net momentum in this plane.

The pseudo-rapidity η is frequently used in discussing particle kinematics, as it is

closely related to the polar angle θ measured from the beam axis by the formula η =
− ln(tan(θ/2)). Notably, the tracker covers the angular range | η |< 2.5 (Fig. 7.4),
while the muon detector covers | η |< 2.4 (Fig. 7.5). Though some specialized
forward calorimeters allow some detection of particles up to | η |< 5, high-quality
reconstruction is generally restricted to objects within the tracker/muon detector
coverage.
 80

Figure 7.4: A longitudinal slice of the silicon tracker showing the coverage in η. TIB, TID,
TOB, and TEC are sections of the strip tracker.

Figure 7.5: A cross sectional slice of the full detector showing the coverage in η of the muon
chambers as well as other components.
 81

7.3 Collision basics

Though the LHC generates proton-proton collisions, protons are not the most fun-
damental participants in the inelastic collisions that we aim to study. The collisions
of most interest are those in which pieces of protons break apart and form a variety
of heavy and unstable particles. Recall that protons are a bound state of 3 quarks,
called the valence quarks quarks. Due to the strongly coupled nature of QCD interac-
tions, this bound state is less of an inanimate collection of 3 stable ingredients, and
more of a living organism whose innards are constantly interacting and rearranging.
In addition to the three valence quarks, these protons contain what is often called a
"sea" of non-valence quarks, anti-quarks, and gluons. When two protons collide
at a certain energy, we do not know a priori which fundamental particles start the
interaction, or what fraction of a proton’s energy those fundamental constituents
contain. Instead, this can only be known probabilistically, meaning a large amount
of data or luminosity is needed for most precise observations.

The collision of two proton bunches is referred to as a bunch crossing, and

each of these results in dozens of interactions along the beam axis within the CMS
detector. Due in part to the wide range of possible energies of the fundamental
participants, the vast majority of interactions which occur in any bunch crossing
are physically uninteresting. Some are elastic collisions which do not break apart
the protons, and others are simply inelastic collisions that do not aid in making
new physics measurements. In order to filter out these events from the enormous
amount of data that is produced during each second of data collection, a multi-tier
trigger system selects interesting events on the fly, preventing a wildly impractical
amount of data from being stored permanently. For the purpose of this thesis,
 82

we will only refer explicitly to the high-level triggers, which are used to compile
different datasets used for physics analysis as in Chapter 8. The potential for
overlap between individual interactions taking place very close to each other creates
an issue called pileup. Generally speaking, this will involve a small number of
physically uninteresting events whose byproducts sneak into the data, an effect that
is accounted for in simulation.

When an event is reconstructed, the negative vector sum of all transverse mo-
menta is designated as pmiss
T and is for historical reasons often referred to as the MET
or missing transverse energy. This vector, in principle, corresponds to momentum
that was carried off from an interaction without being detected. While there will
always be some small transverse momentum imbalance due to detector effects, this
quantity can be quite large if undetectable particles such as neutrinos are produced
in the event.

7.4 Simulation of CMS events

Probabilities of different proton constituents interacting, and the energy they carry,
are determined from parton distribution functions. These have been derived and
tuned from many experiments over time, and multiple sets are available.

Starting from the fundamental constituents of quarks and gluons, individual

collision interactions or events are simulated in a series of stages. First, a fixed order
perturbative QFT calculation is performed to evaluate the relevant matrix element.
While this process models the fundamental interaction taking place at short scales,
 83

a variety of QCD effects beyond perturbative hard scattering must be taken into
account in order to make the simulation more realistic.

The first of these is emission of lower energy particles, such as gluons and pho-
tons. Because of the large value of the strong coupling constant αs , there is a strong
tendency for the high-energy participants in an interaction to radiate away some
energy as a shower of lower-energy particles. Luckily, there are some mathematical
methods in QCD that allow one to calculate series of such emissions to partially
model this effect. The second is the non-perturbative effect of hadronization which
affects quarks and gluons. These particles can travel only very short distances before
they form more stable bound-state particles, like exotic hadrons and mesons, which
can go on to interact with the detector. Both showering and hadronization effects
are modeled by parton shower algorithms, which take a hard interaction involving
a small number of fundamental particles and turn it into myriad of more stable
objects capable of reaching the detector. The matching of shower simulations to
matrix element computations is nontrivial, and multiple approaches are available.

Lastly, the detector response to all simulated events is modeled with the GEANT4
software toolkit [46], which simulates the interaction of particles with a variety
of materials. While the previous steps can be run by individuals with access to
reasonable computing resources, this step is very computationally expensive. For
this reason, only essential samples are generated including full detector response,
and we sometimes have to make due with having lower statistics in some samples
than what is desirable.

Event simulation is typically done individually for specific processes and decay
channels. However, the potential effects of generic physics events with overlapping
 84

primary vertices are later added into the simulation by including extra pileup
events.
 85

Chapter 8

A New Measurement of the Top

Quark Yukawa Coupling

8.1 Basic Motivation

The top quark mass is well measured at 172.4 ± 0.7 GeV. Recalling the mass-Yukawa
relationship from Sec. 2.8, we see that the top-Higgs Yukawa coupling, gt , is within
a few percent of gt = 1. Given that the value is completely arbitrary in the stan-
dard model, one might consider it a bit suspicious–especially in relation to other
fermion Yukawa coupling strengths, which are orders of magnitude lower. It is up
to experimentalists to determine that the standard model mass-Yukawa relation
resulting from coupling to the Higgs field holds as expected. To wuantify any
deviation from the standard model, we define the dimensionless relative Yukawa
parameter Yt = gtobs /gtSM , where gt, obs is the observed Yukawa coupling and gt, SM
is the standard model value. Some extensions of the standard model, such as two-
 86

Higgs-doublet and composite Higgs models, can induce altered couplings between
the top quark and the Higgs boson where Yt 6= 1 [47, 48]. This makes the top–Higgs
interaction in particular an interesting quantity to measure at the LHC, especially
because it is experimentally accessible through multiple avenues.

The must obvious way to measure this is through processes containing a top-
Higgs vertex which produce a real Higgs boson, as was recently done in Refs. [49,50].
Indeed, the most precise measurement of the top quark Yukawa coupling comes
from a combination of such measurements in Ref. [51], which yields Yt = 0.98 ± 0.14.
However, this is not the full story. These measurements detect the Higgs via its
decay products, and so they inevitably fold in additional assumptions about its
branching fractions in the standard model, which depend on other Yukawa cou-
plings. This is a typical problem; the standard model is so rich and interconnected,
it is often difficult to isolate a single parameter for measurement. Thus, it is valuable
to measure the same quantity using different techniques. In this section, we out-
line a measurement that places limits on Yt using only information from tt̄ decays,
where the Higgs boson can be exchanged virtually and does not decay. While this
measurement has its own drawbacks, it has the advantage that it does not require
us to model Higgs decay, avoiding dependence on other coupling constants.

8.2 Electroweak Corrections and tt̄ Production

Top pair production at CMS is simulated based on matrix element calculations

performed at order O(α3s ) (as discussed further in Sec. 8.3). This includes only effects
from QCD interactions, which are by far the dominant means of production and
generally much stronger than other interactions at high energy scales. Electroweak
 87

q t q t
g
Γ g Γ

q t q t

Figure 8.1: tt̄ production from qq processes with virtual exchange of a vector or scalar boson
Γ, such as the Higgs

(EW) contributions begin to have a noticeable effect at order O(α2s α). Here we note
that α stands in for the suite of EW interactions, as the weak and electromagnetic
interaction couplings are related and of the same magnitude. Additionally, here
we treat equally at this order diagrams with EW interactions that carry a factor of
the Yukawa coupling Yt rather than an explicit factor of α to be of this order. This
allows us to consider one-loop diagrams such as those in Fig.8.1 which include
Higgs exchange. This process of virtual exchange allows us to isolate the Yukawa
coupling gt from other Higgs couplings.

The leading correction resulting from these diagrams is obtained by including

them only as interference terms, hence the lone factor of α–the one-loop term
diagram interferes with the LO diagram, while its square is ignored in the matrix-
element calculation.

We refer to the diagrams in Fig.8.1 as EW corrections to tt̄ production. Their effect

on the total cross section is small, so they are typically ignored. However, if the
Yukawa coupling entering these corrections via Higgs exchange is anomalously
large, these diagrams can have a measurable effect on differential cross sections. For
example, doubling the value of gt can alter the Mtt̄ distribution by about 10% near
the production threshold as described in [52] and shown in Fig. 8.2 (left). Another
 88

CMS HATHOR Preliminary s = 13 TeV CMS HATHOR Preliminary s = 13 TeV

(dσHATHOR/dMtt)/ (dσLO/dMtt)

(dσHATHOR/d∆y )/ (dσLO/d∆y )
1.06

tt
1.25
Generator level Yt = 3 Generator level Yt = 3
Yt = 2 1.05 Yt = 2
1.2 Yt = 1 1.04 Yt = 1
Yt = 0 Yt = 0
1.15 1.03
1.02

tt
1.1
1.01
1.05
1
1 0.99

0.95 0.98
0.97
0.9
400 600 800 1000 1200 1400 1600 1800 −4 −3 −2 −1 0 1 2 3 4
Mtt [GeV] ∆y
Z Z tt

Figure 8.2: Effect of the weak corrections on tt̄ differential cross sections for different values
of Yt , after reweighting of generator-level events from POWHEG +P YTHIA 8.

1 E+ p
variable sensitive to the value of gt is the difference in rapidity y = 2 ln E− pz
z

between top and anti top quarks. Defining ∆ytt̄ = y(t) − y(t̄) , the effect of the
EW corrections is shown in Fig. 8.2 (right). Together, these two variables serve as
proxies for the traditional Mandelstam variables s and t, and can be used to include
the EW corrections on arbitrary samples via simulated event reweighting without
generating new MC samples.

This is achieved by applying weight corrections to the current standard tt̄ MC

samples used by the CMS TOP group. Contributions to the top quark pair pro-
duction due to the interference of QCD+EW diagrams, shown in Fig. 8.1, with
the leading order (LO) QCD diagrams are evaluated using HATHOR (v2.1) pack-
age [53]. The ratio of these contributions to the LO QCD production is applied as
parton-level weight corrections to events simulated using POWHEG+PYTHIA8,
which incorporates fully NLO QCD calculations and parton shower effects.
 89

8.3 Simulation of tt̄ Events at CMS

Here we provide some technical details on tt̄ Monte Carlo samples used by CMS
during Run 2 data-taking. Simulation of top quark pair production at CMS begins
with a NLO QCD matrix-element computation, using the POWHEG generator
[54–57]. This calculation is performed with the renormalization µr and factorization
q
scale µf set to the transverse top mass, mT = m2t + pT 2 . A top quark mass
of mt = 172.5 GeV is used in all simulations. The matrix element calculations
obtained from POWHEG are then matched with parton shower simulation from
P YTHIA 8 [58–60]. Pythia uses the underlying event tune M 2 T 4 [61] in 2016 and the
CP 5 tune in 2017 and 2018 [62]. The parton distribution functions from NNPDF 3.0 at
NLO [63] are used in 2016 and updated to NNPDF 3.1 [64] at next-to-NLO (NNLO)
in 2017 and 2018. In the near future, samples will be available which update the
simulation in 2016 to match the other years, but these are not available at the time
of this document’s writing.

We note here a few fundamental sources of uncertainty in the modeling. First, the
scales µr and µf are varied up and down by a factor of 2 to obtain a rough model of
uncertainty coming from higher order QCD terms in the matrix element calculation.
This is a standard practice in assessing uncertainty on perturbative QFT calculations.
The choice of scales µr and µf is required by the renormalization procedure for
perturbative calculations, but the dependence of the result on these scales should
vanish as we go to higher and higher orders in perturbation theory. Thus, differences
in simulation at different scale choices give a sense of the sensitivity of the matrix-
element calculation to higher-order effects. The variation by factor of two is rather
arbitrary, but frequently performs well and has become a rule of thumb in the field.
 90

The scale is also varied in the parton shower algorithm, performed separately to
assess both the effects of scale choice on initial state radiation (ISR) and final state
radiation (FSR). Finally, the makers of the NNPDF sets provide variations which
model the uncertainty on the underlying parton distribution functions.

8.4 Formalism of Weak Corrections

The standard jargon of perturbation theory can sometimes be confusing, when

applied to the EW corrections, since perturbative expansions are done in orders
multiple separate coupling constants, principally αs and α. Then it is not immedi-
ately clear what is meant when someone says "NLO EW diagrams," for example. In
the author’s experience, this designation always includes terms of order O(α2s α), fre-
quently includes terms of order O(α3s α), and sporadically includes the diagrams for
Higgs exchange which are technically O(α2s gt2 ). We obviously want to include the
third type, so we treat top-Higgs vertices as contributing an order α for notational
simplicity, as they are part of the full electroweak interaction and the magnitude of
their effect is comparable to diagrams involving W and Z bosons. To clarify further,
we will utilize notation similar to [65], beginning with an observable such as σtt̄ is
expanded in a standard series as

σtt̄ (αs , α) = ∑ αm n
s α σm,n . (8.1)
m + n ≥2

In practice, contributions due to terms with m < 2 or n > 2 are small, and as
such often neglected in the calculation of the differential cross sections. In this
 91

analysis the following contributions are of interest:

σLO QCD = α2s σ2,0 (8.2)

σNLO QCD = α3s σ3,0 (8.3)

σNLO EW = α2s ασ2,1 (8.4)

σextra EW = α3s ασ3,1 (8.5)

In using the Hathor package to compute the differential cross section, as a function
of Yt , we include the terms

σHATHOR (Yt ) = σLO QCD + σNLO EW (Yt ). (8.6)

The EW corrections K-factor can be defined at the LO QCD level:

NLO σLO QCD + σNLO EW

KEW = (8.7)
σLO QCD

Meanwhile, the POWHEG +P YTHIA 8 samples include terms

σPOWHEG = σLO QCD + σNLO QCD +PS
(8.8)

here the subscript +PS indicates that parton shower simulation has been included
as well. Potential variation of these parton shower effects should be covered by the
standard uncertainties discussed in Sec. 8.8.

Since α  αs , it often makes some sense to work at fixed order m ≤ N, n ≤ 1,

rather than at m + n ≤ N (as mentioned in [66]). Thus, the EW corrections to the
NLO QCD calculations are frequently included using the K-factor, defined at the
 92

LO QCD level:

NLO

σNLO QCD×EW ≡ KEW × σLO QCD + σNLO QCD (8.9)

We refer to this as the "multiplicative approximation", which is similar to the

"multiplicative approach" referenced in [65], aside from the missing NNLO QCD
terms, which in our case should be covered by existing modeling uncertainties.
This is the best approximation for EW corrections realistically available to us, and
in some kinematic regimes these corrections do indeed factorize in this way from
their QCD counterparts. Thus, this approach allows us to roughly approximate the
terms in σextra EW in addition to those directly generated in HATHOR. Specific to
this analysis is the dependence of Higgs-based contributions on Yt and the use of
POWHEG, so we can write out our formula for observables more explicitly as

Y NLO
QCD×EW ≡ KEW (Yt ) × σPOWHEG .
t
σNLO (8.10)

In order to estimate the uncertainty on this approach to including EW corrections,

we investigate also the more naive additive approach,

 
Yt NLO
σNLO QCD+EW ≡ K EW × σLO QCD + σNLO QCD , (8.11)

which neglects the term σextra EW entirely. We take the difference between the two
approaches as an uncertainty of the method:

Y Y
QCD×EW − σNLO QCD+EW
t t
EW unc. = σNLO (8.12)

NLO
= σPOWHEG − σLO QCD (KEW (Yt ) − 1) (8.13)

≡ σPOWHEG × δQCD × δEW , (8.14)

 93

where we have introduced the shorthand

σPOWHEG − σLO QCD NLO

δQCD = , δEW = (KEW (Yt ) − 1), (8.15)
σPOWHEG

to illustrate that this uncertainty is effectively a cross term which arises from the
difference in additive and multiplicative approaches. Note that δEW represents the
marginal effect in terms of which EW corrections are often understood. This uncer-
tainty is not a perfect description, but it is important to include near the threshold
where two-loop diagrams of order O(α3s α) can make significant contributions. At
high values of Mtt̄ , large negative EW corrections result from Sudakov logarithms
associated with W/Z exchange [65]. Such diagrams facctorize well, making the
multiplicative approach valid to good approximation, and we thus do not include
this uncertainty in that kinematic regime.

The implementation of this uncertainty is detailed further in Sec. 8.7.

8.5 Object & Event Selection

Backgrounds for tt̄ events in the dilepton channel are low due to the presence
of 2 high-pT leptons and 2 b-jets in the final state, which together are an ample
signature to identify the process with high confidence. Backgrounds can thus be
sufficiently modeled with simulated events, as opposed to data-driven estimation.
The main backgrounds are single top production and Drell-Yan decays, examples
of which are shown in Fig. 8.3. Single top production simulated at NLO with
POWHEG in combination with P YTHIA 8, while rare diboson events are simulated
in P YTHIA 8 with LO precision. Drell–Yan production is simulated at LO using
 94

the M ADGRAPH generator [67] interfaced to P YTHIA 8 with the MLM matching
algorithm [68]. Overall, after event selection, the expected non-tt̄ background rate
in this channel is about 5%.

Events are selected first using single electron or single muon triggers, designed
to pick out events with at least one high-pT lepton from the data. In the case of
muons, we select events with a trigger pT threshold of 24 GeV except in 2017 data,
where the threshold is raised to 27 GeV. For electrons, we select events with a trigger
pT threshold of 27 GeV with the exception of 2018 when a threshold of 32 GeV is
used due to availability. While dilepton triggers are also available with lower pT
thresholds, we found that they have no significant benefit in our case, as most
events which interest us have a leading lepton pT over 30 GeV.

To ensure that all electrons and muons are within the silicon tracker coverage, we
require they have pseudo-rapidity | η |< 2.4. To operate well above the online trig-
ger threshold, we then require at least one isolated electron or muon reconstructed
with pT > 30 GeV, except in 2018, where offline selection of leading electrons is
raised to pT > 34 GeV to match available triggers. A small buffer of a few GeV

ℓ
t
q Z
g b
ℓ̄
q
b b
W
g
q̄ g

b̄
b̄

Figure 8.3: Examples of background processes for tt̄ dilepton decays are shown. The left
diagram shows a Drell-Yan process involving a Z boson decaying to a lepton
+ anti-lepton pair, while the right diagram shows an example of single-top
production.
 95

between the online trigger selection and the offline pT selection helps prevent po-
tential disagreement between simulation and data near the trigger threshold. A
second isolated electron or muon with pT > 20 GeV is required. Any multi-lepton
events with additional isolated leptons having pT > 15 GeV are discarded, helping
to lower diboson background contributions and avoiding events where additional
work would be required due to ambiguity with lepton assignment.

Jets are clustered from particle-flow objects through the anti-kT algorithm [69, 70]
using a distance parameter of 0.4. The jet momentum is calculated by taking the
vector sum of momenta over all constituents identified as part of the jet. Important
corrections to the jet energy are derived as a function of jet pT and η, using both
simulation and measurements of energy balance in data [71]. We select jets with
| η |< 2.4 and pT > 30 GeV. We identify jets originating from b quarks using the
DeepCSV algorithm [41], requiring all events to have at least two b-tagged jets
according to the "loose" working point, which has the highest sensitivity. Events
with more than two b jets are considered only if exactly two of the jets pass a
more stringent selection. In this case these two jets passing the more stringent
requirement are assumed to be those originating from top quark decays.

We find that in tt̄ events, the two b jets are among the candidates passing the
looser b-tagging requirement in roughly 93% of the events.

Drell-Yan background events (as in Fig. 8.3 left), mostly involving the decay of
a Z boson, make a substantial contribution to the ee and µµ channels. To reduce
their prevalence, we reject events with an invariant mass of the two leptons below
50 GeV or within 10 GeV around the Z boson mass of 91.2 GeV.
 96

Source 2016 (X fb−1 ) 2017 (Y fb−1 ) 2018 (Z fb−1 ) All (X+Y+Z fb−1 ) % total MC
tt̄ MC 140 831±134 170 554±95 259 620±146 571 005±220 96.1%
V + jets MC 1944±54 3178±75 4958±133 10 080±162 1.7%
Single t MC 3022±25 3521±24 5827±32 12 370±47 2.1%
Diboson MC 142±9 151±10 250±15 543±20 0.1%
Total MC 145 940±148 177 404±124 270 655±201 593 999±279 100%
Data 144 817 178 088 264 791 587 696 98.9%

Table 8.1: MC and data event yields for all three years and combined. Statistical uncertainty
on simulated event counts is given. For an estimate of systematic uncertainty, see
Figs. 8.5-8.7.

The breakdown of expected signal and background yields for each year is shown
in Tab. 7.1. The Drell–Yan and Single top quark production processes each account
for roughly 2% of the estimated sample composition. Diboson decays, as well as
other multi-jet events typically referred to as QCD background, make negligible
contributions.

Additionally, there are multiple small discrepancies in efficiency when one

applies the same selection criteria to simulation and data. This applies to the single
lepton triggers, lepton identification, and b-jet identification. These disagreements
are addressed via scale factor corrections to the event weights, which can be derived
by comparing data and simulation in special cases–for example, the tag and probe
method on Z decays is used for lepton-associated scale factors [72, 73]. Uncertainty
on these corrective scale factors is addressed in Sec. 8.8.

8.6 Event Reconstruction

While it is possible to completely reconstruct the neutrino momenta as described

in Sec. 6.4, such reconstruction is highly sensitive to the MET (pmiss
T ). Because the

MET computation for each event relies on measurements taken throughout the
 97

detector and involving many different particles, it is vulnerable to large resolution

effects and also systematic uncertainty resulting from the energy scales of various
particles. For this reason, it was decided to forgo the full top quark reconstruction,
instead defining proxy variables for Mtt̄ and ∆ytt̄ . We found that a more sensitive
measurement results from using the variables

¯ ,
Mb` = M(b + b̄ + ` + `) (8.16)
¯ − y(b̄ + `)|,
|∆y|b` = |y(b + `) (8.17)

which were found to have good sensitivity to EW corrections while being less
vulnerable to resolution effects and systematic uncertainties. Mb` has the advantage
of not dependening on the pairing of b jets with leptons at W decay vertices, mean-
ing that it is reconstructed quite accurately. We do however utilize the kinematic
constraints discussed in Sec. 6.4 to help with this pairing of b jets and leptons in
order to reconstruct ∆yb` . Including this variable was shown to add additional
sensitivity, despite some resolution effects.

The full procedure for b-jet pairing is as follows:

1. The mass constraint is checked for both possible pairings. If only one pairing
is found to satisfy the mass constraint, that pairing is used. If both pairings fail
to satisfy the mass constraint, the event is discarded. If both pairings satisfy
the mass constraint, we check the MET constraint.

2. If only one pairing allows for the MET constraint while the other does not, the
pairing yielding an exact solution to the MET constraint is used.
 98

3. If the neutrino kinematics do not suggest a clear pairing, the b-jets, b1 and b2 ,
¯ by minimizing the quantity
are paired with the leptons (l, l)

Σ1(2) = ∆R(b1(2) , l ) + ∆R(b2(1) , l¯)

q
among the two possible pairings, where ∆R(, `) = ( ηb − η ` ) 2 + ( ϕ b − ϕ ` ) 2 .

In the end the correct b-jet pairing is selected for approximately 82% of events in
which both b-jets have been correctly identified.

The sensitivity of our chosen kinematic variables to Yt is shown before and

after reconstruction in Fig. 8.4 . The effect of Yt remains substantial enough, and
unique enough in shape, that a measurement can be extracted. When looking back
at Fig. 8.2, we see that most sensitivity in Mtt̄ is more narrowly confined to the
threshold region, which is particularly difficult to reconstruct accurately in the
dilepton channel. By comparison, the sensitivity of Yt to Mb` is distributed across a
wider region.

We also see that sensitivity to our parameter of interest Yt is not greatly reduced
by the reconstruction process. This stands in contrast to typical attempts to recon-
struct Mtt̄ , where the sensitivity is confined to a narrow peak at the threshold and
much information is lost to resolution effects.

The comparison between data and Monte Carlo is shown in Figs. 8.5-8.7. The
agreement is generally seen to be well within known uncertainty. We do see some
slopes in the ratio of data to MC prediction in the distributions of lepton and b-
jet pT , to varying extent for each year. These are likely related to a well known
disagreement in top quark pT distribution between POWHEG samples and data
(See Ref. [74], for example).
 99

1.14
CMS HATHOR Preliminary s = 13 TeV
1.14
CMS HATHOR Preliminary s = 13 TeV
(dσY /dMbl )/ (dσY =1/dMbl )

(dσY /dMbl )/ (dσY =1/dMbl )

Generator level Yt = 3 Reconstructed Yt = 3
1.12 1.12
Yt = 2 Yt = 2
1.1 Yt = 1 1.1 Yt = 1
t

t
1.08 Yt = 0 1.08 Yt = 0

1.06 1.06
1.04 1.04
1.02 1.02
t

t
1 1
0.98 0.98
0.96 0.96
200 400 600 800 1000 1200 1400 1600 1800 200 400 600 800 1000 1200 1400 1600 1800

Mbl [GeV] Mbl [GeV]

Z Z

1.07
CMS HATHOR Preliminary s = 13 TeV
1.07
CMS HATHOR Preliminary s = 13 TeV
(dσY /d∆y )/ (dσY =1/d∆y )

(dσY /d∆y )/ (dσY =1/d∆y )

bl

bl

Generator level Yt = 3 Reconstructed Yt = 3

1.06 Yt = 2 1.06 Yt = 2
1.05 Yt = 1 1.05 Yt = 1
t

Yt = 0 Yt = 0
1.04 1.04

1.03 1.03
bl

bl

1.02 1.02
t

1.01 1.01

1 1

0.99 0.99
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
∆y ∆y
Z bl Z bl

Figure 8.4: Effect of the EW corrections on Mb` and ∆yb` kinematic distributions are shown
at generator level (left plots) and reco level (right plots). The effect is shown
relative to the Standard Model EW corrections, as this is the best way to assess
our sensitivity to different Yukawa coupling values, rather than comparing to a
sample with no EW corrections.
 100

×10
3 35.9 fb-1 (13 TeV) 2016 35.9 fb-1 (13 TeV) 2016

Events

Events
CMS all dilepton Data CMS all dilepton Data
100 Preliminary
tt correct reco 60000 Preliminary
tt correct reco
reco level tt (b-jets swapped) reco level tt (b-jets swapped)
tt wrong reco tt wrong reco
tt tau 50000 tt tau
80
Single t Single t
Drell-Yan 40000 Drell-Yan
60 di-boson di-boson
30000
40
20000
20 10000

1.4 1.4
Pred.

Pred.
Data

Data
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0 1 2 3 4 5 6 7 8 9 0 50 100 150 200 250 300
Njets b-jet p
Njets b-jet ptT
35.9 fb-1 (13 TeV) 2016 35.9 fb-1 (13 TeV) 2016
60000
Events

Events
30000 CMS all dilepton Data CMS all dilepton Data
Preliminary
tt correct reco Preliminary
tt correct reco
reco level tt (b-jets swapped) reco level tt (b-jets swapped)
50000
25000 tt wrong reco tt wrong reco
tt tau tt tau
Single t 40000 Single t
20000 Drell-Yan Drell-Yan
di-boson di-boson
30000
15000

10000 20000

5000 10000

1.4 1.4
Pred.

Pred.
Data

Data

1.2 1.2
1 1
0.8 0.8
0.6 0.6
0 50 100 150 200 250 0 50 100 150 200 250
leading lepton p trailing lepton p
Leading Lep PtT Trailing Lep PtT
35.9 fb-1 (13 TeV) 2016
Events

14000 Data
CMS all dilepton
Preliminary
tt correct reco
12000 reco level tt (b-jets swapped)
tt wrong reco
10000 tt tau
Single t
Drell-Yan
8000 di-boson
6000

4000

2000

1.4
Pred.
Data

1.2
1
0.8
0.6
0 50 100 150 200 250 300
MET
MET (PT)

Figure 8.5: 2016 Data to MC comparison, with the shaded region indicating known uncer-
tainty.

After reconstruction, events in each year are separated into two |∆y|b` bins and
binned more finely in the Mb` variable. The predominant sensitivity to Yt comes
from shape distortions to the Mb` distribution, but the additional separation by ∆yb`
helps to distinguish these shape distortions from those associated with systematic
uncertainties. The precise binning is chosen to obtain optimal information about
shape variation while maintaining low statistical uncertainty by requiring ' 10, 000
 101

×10
3 41.5 fb-1 (13 TeV) 2017 41.5 fb-1 (13 TeV) 2017

Events

Events
CMS all dilepton Data 80000 CMS all dilepton Data
120 Preliminary
tt correct reco Preliminary
tt correct reco
reco level tt (b-jets swapped) 70000 reco level tt (b-jets swapped)
tt wrong reco tt wrong reco
100 tt tau 60000 tt tau
Single t Single t
80 Drell-Yan 50000 Drell-Yan
di-boson di-boson
40000
60
30000
40
20000
20 10000

1.4 1.4
Pred.

Pred.
Data

Data
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0 1 2 3 4 5 6 7 8 9 0 50 100 150 200 250 300
Njets b-jet p
Njets b-jet ptT
41.5 fb-1 (13 TeV) 2017 41.5 fb-1 (13 TeV) 2017
40000
Events

Events
CMS all dilepton Data 70000 CMS all dilepton Data
35000 Preliminary
tt correct reco Preliminary
tt correct reco
reco level tt (b-jets swapped) reco level tt (b-jets swapped)
tt wrong reco 60000 tt wrong reco
30000 tt tau tt tau
Single t 50000 Single t
25000 Drell-Yan Drell-Yan
di-boson 40000 di-boson
20000
30000
15000
10000 20000

5000 10000

1.4 1.4
Pred.

Pred.
Data

Data

1.2 1.2
1 1
0.8 0.8
0.6 0.6
0 50 100 150 200 250 0 50 100 150 200 250
leading lepton p trailing lepton p
Leading Lep PtT Trailing Lep PtT
41.5 fb-1 (13 TeV) 2017
Events

16000 CMS all dilepton Data

Preliminary
tt correct reco
reco level tt (b-jets swapped)
14000
tt wrong reco
12000 tt tau
Single t
10000 Drell-Yan
di-boson
8000
6000
4000
2000

1.4
Pred.
Data

1.2
1
0.8
0.6
0 50 100 150 200 250 300
MET
MET (PT)

Figure 8.6: 2017 Data to MC comparison, with the shaded region indicating known uncer-
tainty.

events present in each year in each bin. The data and simulation is shown with the
final binning in Fig. 8.8.
 102

×10
3 59.7 fb-1 (13 TeV) 2018 ×10
3 59.7 fb-1 (13 TeV) 2018

Events

Events
200 CMS all dilepton Data 120 CMS all dilepton Data
Preliminary
tt correct reco Preliminary
tt correct reco
180 reco level tt (b-jets swapped) reco level tt (b-jets swapped)
160 tt wrong reco 100 tt wrong reco
tt tau tt tau
140 Single t Single t
Drell-Yan 80 Drell-Yan
120 di-boson di-boson
100 60
80
60 40
40
20
20
1.4 1.4
Pred.

Pred.
Data

Data
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0 1 2 3 4 5 6 7 8 9 0 50 100 150 200 250 300
Njets b-jet p
Njets b-jet ptT
59.7 fb-1 (13 TeV) 2018 ×10
3 59.7 fb-1 (13 TeV) 2018
Events

Events
CMS all dilepton Data CMS all dilepton Data
tt correct reco 100 tt correct reco
Preliminary reco level tt (b-jets swapped) Preliminary reco level tt (b-jets swapped)
50000
tt wrong reco tt wrong reco
tt tau 80 tt tau
40000 Single t Single t
Drell-Yan Drell-Yan
di-boson 60 di-boson
30000

20000 40

10000 20

1.4 1.4
Pred.

Pred.
Data

Data

1.2 1.2
1 1
0.8 0.8
0.6 0.6
0 50 100 150 200 250 0 50 100 150 200 250
leading lepton p trailing lepton p
Leading Lep PtT Trailing Lep PtT
59.7 fb-1 (13 TeV) 2018
Events

25000 CMS all dilepton Data

Preliminary
tt correct reco
reco level tt (b-jets swapped)
20000 tt wrong reco
tt tau
Single t
15000
Drell-Yan
di-boson

10000

5000

1.4
Pred.
Data

1.2
1
0.8
0.6
0 50 100 150 200 250 300
MET
MET (PT)

Figure 8.7: 2018 Data to MC comparison, with the shaded region indicating known uncer-
tainty.

8.7 Statistical Methods

We define a likelihood function L, which itself is defined as a product over separate

likelihood functions evaluated at each bin,

L= ∏ Lbin × p(φ) × ∏ p(θi ), (8.18)

bin∈( Mb` ,|∆y|bl ) i
 Data Events / Bin
Pred.

10000
15000
20000
25000
30000

5000

0.9
1
1.1
100-210
210-230
230-250
250-270

Mb` ranges.
270-290
290-310

CMS
310-340
340-380
380-440 |∆y | < 1
440-3000
100-280 |∆y | > 1
280-320
320-360
360-400
400-460
460-560 2016
560-3000
100-210
210-230
230-250
tt

250-270
270-290
Data

290-310
310-340
Single t

340-380
Drell-Yan
Total unc.

380-440 |∆y | < 1

440-3000
100-280 |∆y | > 1
280-320
320-360
360-400
400-460
460-560 2017
560-3000
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440 |∆y | < 1
440-3000
100-280 |∆y | > 1
280-320
320-360
360-400
400-460
460-560 2018
137 fb-1 (13 TeV)

Mbl range [GeV]

560-3000

Figure 8.8: Binned data and simulated events prior to performing the statistical fit. The
103

dashed line divides the two |∆y|b` bins, while the horizontal axis shows the bins’
 104

where in each bin we have a likelihood based on a Poisson distribution which

describing the statistical fluctuation of the event count,

h i
Lbin = Poisson nbin
obs s bin
({ θ i }) × R bin
EW t( Y , φ ) + b bin
({ θ i })} . (8.19)

Systematic sources of uncertainty are represented by nuisance parameters θi and

φ, which have penalty terms p(θi ) and p(φ) accounting for for each nuisance
parameter’s deviation from its estimated value. In each bin, nbin
obs is the total observed

bin count, with the expected bin count being the sum of the predicted signal sbin
and background bbin . The number of expected signal events is modified by the
additional rate REW , which encodes the EW corrections depends on the Yukawa
coupling ratio Yt and the special nuisance φ. The full expression for the rate Rbin
EW ,

including the modifying factor from this uncertainty, is

Rbin bin bin bin φ

EW (Yt , φ ) = (1 + δEW (Yt )) × (1 + δQCD δEW (Yt )) , (8.20)

bin bin
where δQCD δEW represents the cross term arising from the difference in multiplica-
tive and additive approaches of applying EW corrections as described in Sec. 8.2.
This uncertainty is applied only to bins where EW corrections are increasing as
a function of Yt . The normally distributed nuisance parameter φ modulates this
uncertainty.

Each of the other nuisance parameters, θi , represents an unknown parameter

that can alter the expected bin counts. These are described by a standard normal
Gaussian probability distribution function of p(θi ), often called a penalty term, that
affects the likelihood when this parameter deviates from our best-estimate value.
 105

137 fb-1 (13 TeV)

Events / bin
50000 CMS EW corrections tt (Y = 1)
Simulation t
Preliminary tt (Y = 2)
t
40000
tt (Y = 0)
t

30000

20000

|∆y| < 1.0

|∆y| > 1.0

10000

440-3000

560-3000
100-210

210-230

230-250

250-270

270-290

290-310

310-340

340-380

380-440

100-280

280-320

320-360

360-400

400-460

460-560
Relative effect (tt)

0.05

−0.05
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440

100-280
280-320
320-360
360-400
400-460
460-560
440-3000

560-3000
Mbl range [GeV]

Figure 8.9: The effect of the yukawa parameter Yt on binned reconstructed events is shown
for two discrete choices of Yt . Data from the full Run 2 period is combined in
the kinematic binning.

The effects resulting from nuisance parameters {θi } are modeled by generating
up and down variations of the bin content sbin for each θi in each. These variations
result from changing the underlying physical/technical sources, which are outlined
in Sec. 8.8, usually by an amount of one standard deviation (σ) based on the
uncertainty in our best estimates.

Collectively, the up or down variation across all bins generates a shape template
associated with the nuisance parameter. These shape templates are then enforced as
the bin modifiers associated with θi = ±1, while θi = 0 corresponds to the nominal
estimate at which simulated data is produced.
 106

CMS Simulation Preliminary CMS Simulation Preliminary

EW

EW
Rbin

Rbin
1.35 |∆ y | 0-1.0 0.96
bl
1.3 M bl 100-210 GeV
0.94
1.25
0.92
1.2
0.9
1.15
0.88
1.1
0.86
1.05 |∆ y | 0-1.0
0.84 bl

1 M bl 440-3000 GeV
0.82

0 1 2 3 4 5 0 1 2 3 4 5

Yt Yt

Figure 8.10: The weak correction rate modifier RbinW in two separate [Mb` , ∆yb` ] bins from
2017 data, demonstrating the quadratic dependence on Yt . All bins will have an
increasing or decreasing quadratic yield function, with the steepest dependence
on Yt found at lower values of Mb` .

An example of a shape template for one year of data, compared to the effect of the
parameter of interest (POI), are shown in Fig. 8.9. The Quadratic dependence of the
electroweak corrections on individual bins are shown in Fig. 8.10. Further examples
of shape templates are shown in Sec. 8.9 We impose these shape templates through
vertical template morphing as a function of the underlying nuisance θi , such that
each bin the modifier is interpolated as a cubic spline for values of θi ∈ [−1, 1] and
linearly outside of that region. This is the simplest procedure available to ensure
that sbin (θi ) is a continuous and differentiable function.

After constructing the full likelihood function, the best-fit value of Yt is obtained
by maximizing the likelihood. This maximization is then repeated with Yt fixed over
a fine array of fixed values of the POI and comparing the maximum log likelihood,
we build confidence intervals (CI) at the 68% and 95% level.
 107

8.8 Sources of Uncertainty

8.8.1 Experimental sources of uncertainty

The dominant experimental uncertainty in this analysis comes from the detector’s
jet energy response. Because jets are the result of clustering many individual parti-
cles, accurate measurement of the jet energy is challenging. Corrections are derived
as a function of jet pT and η based on simulation, which are then improved using
data. We follow the standard approach outlined in [71] to consider 26 separate
uncertainties which are involved in determining these calibrations. In this approach,
uncertainty on the resolution of the jet reconstruction is considered as well as a sep-
arate uncertainty. When deriving shape templates for the associated uncertainties,
the altered jet energy is propagated to the calculation of pmiss
T .

Other experimental sources of uncertainty are comparatively minor. The uncer-

tainty on the integrated luminosity has an overall normalization effect on signal
and background processes in each data-taking year, ranging from 2.3% to 2.5%
depending on the year.

Unwanted pileup events are modeled and injected into simulation to include
their effects on event reconstruction. Uncertainty on the number of pileup events
included in this way is assessed by varying the inelastic cross section, 69.2 mb, by
its current uncertainty of 4.6% [75].

Efficiencies in b jet identification and mis-identification are corrected, as a func-

tion of jet pT and η, to better match data [41]. These uncertainties have a normaliza-
tion effect of a few percent and a smaller shape component, and do not have much
effect on the fit.
 108

Similarly, scale factors are applied as functions of pT and η to correct simulated

efficiencies of lepton identification, isolation requirements, and triggers to match
data. These are derived for muons and electrons using the tag-and-probe method
on Z decays [72, 73]. The tag-and-probe fit accounts for uncertainty from statistics
as well as difference in jet multiplicity. Overall, the effect is seen to be below 2%,
and approximately uniform.

8.8.2 Theoretical sources of uncertainty

As a standard technique to estimate potential contributions of higher-order QCD

terms at the matrix element level, the QCD renormalization scale µr and factoriza-
tion scales µf are varied up and down in the POWHEG simulation by a factor of
2. The scales are each varied individually to form templates, and then a same-sign
variation of the two scales together is included as an additional uncertainty. These
three shape templates together cover various higher-order effects, as higher order
perturbative calculations depend less and less on the scale, ultimately in principle
converging to a scale-independent limit. In generating the scale variation templates,
the overall normalization effect is removed, as the NNLO tt̄ cross-section is already
used to improve MC normalization. This allows us to consider mainly shape effects,
which we are far more sensitive to.

A 5% normalization uncertainty is included on the tt̄ cross section, which more

than covers expected contribution from terms beyond NNLO QCD. The back-
grounds in this analysis are small enough (each making up ≈2% or less of the
expected sample composition) that we do not generate shape templates for their
response to individual systematic uncertainties. We do, however, include substan-
 109

tial normalization uncertainties meant to broadly cover uncertainty on background

modelling, which will have an effect on reconstructed distribution shapes. A 15%
sample is included on single top quark production, which was seen to cover the
scale variation and jet energy correction uncertainties in magnitude. A 30% normal-
ization uncertainty is included on Drell-Yan and diboson event simulation, because
these MC samples are generated with LO calculations and have larger uncertainties
associated with scale variation.

We include an uncertainty on the weak corrections, based on our methods for

generating and applying these additional terms, as outlined in Sec. 8.4 and 8.7.
Like the scale variations, this uncertainty stands in for higher-order effects at the
ME level, predominantly those coming from diagrams of order α3s α. This effect is
included only for the region near the threshold where the EW corrections have a
positive effect on the tt̄ cross section. The uncertainty manifests as an envelope
around the EW corrections in this regime, as shown in Fig. 8.11. This is a key
uncertainty in the fit, as it encodes our limited knowledge of the EW interactions
from which we derive our sensitivity.

To model uncertainty in parton distribution functions, NNPDF set [63] provides

100 individual variations (replicas) as uncertainties. As in [76], variations with
similar shape templates are combined to reduce the number of variations to a more
manageable 8 in total. These uncertainties are treated as fully correlated between
2017/2018 datasets, but not in the 2016 data as the simulation uses a different
version of NNPDF with incompatible variations. Variation of the strong coupling
constant αs used by NNPDF is included as well. Though the these uncertainties
can have significant shape effects, they are generally small, on the order of ∼1% or
less.
 110

Effect of Yukawa coupling on binned sample [M , ∆ y ]

bl bl

marginal change
0.06 Yt=2
Yt=0
0.04

0.02

− 0.02

− 0.04 100 2 2 2 2 2 3 3 3 4 1 2 3 3 4 4 5
.0-2 10.0-2 30.0-2 50.0-2 70.0-2 90.0-3 10.0-3 40.0-3 80.0-4 40.0-3 00.0-2 80.0-3 20.0-3 60.0-4 00.0-4 60.0-5 60.0-3
10.0 30.0 50.0 70.0 90.0 10.0 40.0 80.0 40.0 000 8 2 6 0 6 6 0
, 0.0 , 0.0 , 0.0 , 0.0 , 0.0 , 0.0 , 0.0 , 0.0 , 0.0 .0, 0 0.0, 1.0 0.0, 1.0 0.0, 1.0 0.0, 1.0 0.0, 1.0 0.0, 1.0 00.0, 1
-1.0 -1.0 -1.0 -1.0 -1.0 -1.0 -1.0 -1.0 -1.0 .0-1.0 -6.0 -6.0 -6.0 -6.0 -6.0 -6.0 .0-6.0

Figure 8.11: The marginal effect of the EW corrections is shown at two values of top Yukawa
coupling. The colored lines show the nominal value of the effect, while the
shaded envelope indicates the included uncertainty on the EW corrections
themselves.

Dedicated MC samples are generated with the top mass varied up and down
by 1 GeV to model uncertainty on the measured top mass (though at present this
value is an overestimate of that uncertainty). Due to statistical fluctuations in the
dedicated samples, simulations from all three years are combined to generate one
shape template, and the uncertainty is taken as correlated between years. It should
be noted that, although the top mass and Yukawa coupling are generally treated
independently in this measurement strategy, varying the mass will slightly modify
the definition of Yt = 1. However, the effect is below 1%, making it much smaller
than the sensitivity of the measurement and safe to ignore.

The hdamp parameter, which controls the matrix element parton shower matching
in POWHEG + P YTHIA 8, is varied in order to estimate the uncertainty in the
matching algorithm. Dedicated MC samples are also generated with variations of
the P YTHIA 8 tune in order to estimate uncertainty in the hadronization model.
 111

Uncertainty due to initial and final state radiation in the parton shower algo-
rithms is assessed by varying the energy scale in the initial and final states by a
factor of 2.

Lastly, there are two more modelling uncertainties related to b-jet modeling. The
branching ratio of semi-leptonic B decays affects the b-jet response, so this quantity
is varied within its measured uncertainty [36] to model its effects. Additionally, The
momentum transfer from b quarks to B hadrons is dependent on the fragmentatiom
pT ( B )
rate xb = pT (b− jet)
. To include the uncertainty of this ratio, the transfer function
is varied up and down within uncertainty of the Bowler-Lund parameter [77] in
P YTHIA 8 and the resulting effect is applied via event weights.

8.8.3 Processing of shape templates

Shape templates are generated for all uncertainties which are not explicitly normalization-
based uncertainties. These templates are generate separately for each year, due
to minor differences between data taking periods. In many cases the underlying
nuisance parameter is fully or partially correlated between years, which is enforced
in the likelihood fit. Some shape templates are also smoothed to eliminate noise, or
made to be exactly symmetric, both of which help to ensure stability and perfor-
mance of the fit. Templates which do not exhibit an effect of > 0.1% after smoothing
are converted to an overall normalization uncertainty, as shape distortions are not
meaningful at this level. Example shape templates are shown for the 3 dominant
uncertainties in Fig. 8.12, where the effects of smoothing and symmetrizing can be
seen.
 112

Events / bin 137 fb-1 (13 TeV) 137 fb-1 (13 TeV)

Events / bin
CMS final state radiation tt (central) CMS jet energy corrections tt (central)
Simulation Simulation
50000 50000
Preliminary tt (θi +1σ) Preliminary tt (θi +1σ)
tt (θi -1σ) tt (θi -1σ)
40000 40000

30000 30000

20000 20000

|∆y| < 1.0

|∆y| > 1.0

|∆y| < 1.0

|∆y| > 1.0

10000 10000

0.04
440-3000

560-3000

440-3000

560-3000
100-210

210-230

230-250

250-270

270-290

290-310

310-340

340-380

380-440

100-280

280-320

320-360

360-400

400-460

460-560

100-210

210-230

230-250

250-270

270-290

290-310

310-340

340-380

380-440

100-280

280-320

320-360

360-400

400-460

460-560
Relative uncertainty (tt)

Relative uncertainty (tt)

0.04

0.02 0.02

0 0

−0.02 −0.02

−0.04 −0.04
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440

100-280
280-320
320-360
360-400
400-460
460-560

100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440

100-280
280-320
320-360
360-400
400-460
460-560
440-3000

560-3000

440-3000

560-3000
Mbl range [GeV] Mbl range [GeV]

137 fb-1 (13 TeV) 137 fb-1 (13 TeV)

Events / bin

Events / bin

CMS factorization scale tt (central) CMS renormalization scale tt (central)

Simulation Simulation
50000 50000
Preliminary tt (θi +1σ) Preliminary tt (θi +1σ)
tt (θi -1σ) tt (θi -1σ)
40000 40000

30000 30000

20000 20000
|∆y| < 1.0

|∆y| > 1.0

|∆y| < 1.0

|∆y| > 1.0

10000 10000
440-3000

560-3000

440-3000

560-3000
100-210

210-230

230-250

250-270

270-290

290-310

310-340

340-380

380-440

100-280

280-320

320-360

360-400

400-460

460-560

100-210

210-230

230-250

250-270

270-290

290-310

310-340

340-380

380-440

100-280

280-320

320-360

360-400

400-460

460-560
0.04
Relative uncertainty (tt)

Relative uncertainty (tt)

0.02

0.01 0.02

0 0

−0.01 −0.02

−0.02 −0.04
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440

100-280
280-320
320-360
360-400
400-460
460-560

100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440

100-280
280-320
320-360
360-400
400-460
460-560
440-3000

560-3000

440-3000

560-3000

Mbl range [GeV] Mbl range [GeV]

Figure 8.12: Shape templates are shown for dominant uncertainties in the fit, associated
with the final-state radiation in P YTHIA 8, the jet energy corrections, the factor-
ization scale, and the renormalization scale. The shaded bars represent the raw
shape template information, while the lines show the shapes after smoothing
and symmetrization have been applied. In the fit, the jet energy corrections are
split into 26 different components, but for brevity only the total uncertainty
is shown here. Variation between data-taking years is minimal for each of
these uncertainties, so their effect is shown on the full dataset in the kinematic
binning. They are treated separately by year in the actual fit, to ensure accurate
modelling.
 113

8.9 Results

The maximum likelihood fit yields a best fit value of Yt = 1.16+ 0.24
−0.35 , with Yt < 1.54

at a 95% confidence level, with impacts of significant uncertainty sources shown in

Fig. 8.13 Agreement of data and simulation in the final analysis binning is shown for
the final configuration after performing the maximum likelihood fit in Fig. 8.14. We
see that the fit is able to obtain good agreement and remove sloping seen in the pre-
fit data (Fig. 8.8). A likelihood scan visualizing the construction of the confidence
intervals is shown in Fig. 8.15, performed for both data and simulation. The result
is in agreement with the analogous measurement obtained previously in Ref. [76].
While it does not mark a major improvement in sensitivity, this measurement serves
as a valuable complement, particularly due to the use of a different decay channel
with different backgrounds.

Though we achieve lower sensitivity compared to production-based constraints

in ttH [49] and Higgs combination measurements [51], it should be noted that
these measurements depend as well on other Yukawa coupling values through
additional branching assumptions. This is part of the challenge of measurements in
the standard model: the theory is so complex and contains so many free parameters,
it can be challenging to isolate just one. Thus, it is valuable to have measurements
with different approaches which fold in different assumptions. Our result serves
as a compelling orthogonal measurement to other techniques that involve Higgs
production and decay. Another recent measurement which possesses a similar
independence to other branching ratios comes from searches for the production of
four top quarks [78], which yields a 95% CI upper limit of Yt < 1.7 [78], compared
to the 95% upper limit of Yt < 1.54 seen here.
 114

arXiv:2009.07123 +0.24
CMS Supplementary Yt = 1.16−0.35
1 Final state radiation scale (correlated)
2 Electroweak correction uncertainty
3 ME factorization scale
4 Jet energy FlavorQCD
5 Initial state radiation scale (correlated)
6 Final state radiation scale (2016)
7 ME renormalization scale
8 Top quark mass
9 Muon reconstruction efficiency (correlated)
10 Single top normalization
11 NNPDF variation 2 (2016)
12 b tagging miss-ID efficiency (correlated)
13 Jet energy RelativeFSR (correlated)
14 Jet energy TimePtEta (2016)
15 NNPDF variation 4 (correlated)
16 NNPDF αs variation (correlated)
17 NNPDF variation 1 (correlated)
18 Muon reconstruction efficiency (2018)
19 b tagging efficiency (correlated)
20 b tagging efficiency (2018)
21 NNPDF variation 0 (correlated)
22 Jet energy SinglePionECAL
23 NNPDF αs variation (2017)
24 Jet energy resolution (2018)
25 Jet energy RelativeSample (2016)
26 Jet energy AbsoluteMPFBias
27 Jet energy TimePtEta (2018)
28 Jet energy RelativeFSR (2018)
29 tt normalization
30 Drell −Yan normalization
−2 −1 0 1 2 −0.1 0 0.1
Observed: Fit constraint +1σ Impact -1σ Impact (θ-θ0)/ ∆θ ∆Yt
Expected: Fit constraint +1σ Impact -1σ Impact

Figure 8.13: Nuisance parameter post-fit constraints and impacts are shown for the 30 un-
certainties in the fit with the largest impact on the best fit value of Yt . This
information is displayed for an Asimov dataset generated with Yt = 1 (ex-
pected) and for the final fit result on data (observed). Impacts are calculated
by repeating the fit with individual nuisance parameters varied up and down
by one standard deviation according to the their post-fit uncertainty and held
fixed. Some uncertainty sources are split into a correlated nuisance parameter
and an uncorrelated parameter unique to each data-taking year. Uncertainty on
corrections to jet energy is broken down into several sources, whose names are
indicated in italics. Technical details on the meaning of each such component
can be found in Ref. [71]
 115

137 fb-1 (13 TeV)

Events / Bin
30000
CMS Data
tt
25000 Single t
Drell-Yan
Total unc.
20000

15000

10000

|∆y | < 1
|∆y | > 1
|∆y | < 1
|∆y | > 1
5000

2016

2017

2018
|∆y | < 1
|∆y | > 1
Pred.
Data

1.1
1
0.9
440-3000

560-3000

440-3000

560-3000

440-3000

560-3000
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440
100-280
280-320
320-360
360-400
400-460
460-560
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440
100-280
280-320
320-360
360-400
400-460
460-560
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440
100-280
280-320
320-360
360-400
400-460
460-560
Mbl range [GeV]

Figure 8.14: The agreement between data and MC simulation at the best fit value of Yt =
1.16 after performing the likelihood maximization.

CMS 137 fb-1 (13 TeV)

6
-2 ∆ ln L

Data

5 Asimov Y t = 1
Asimov Y t = 1.16
4

1 68% CL

0
0 0.5 1 1.5 2 2.5 3
Yt

Figure 8.15: The result of a likelihood scan, performed by fixing the value of Yt at values
over the interval [0, 3.0] and minimizing the negative log likelihood (− ln L).
Expected curves from fits on simulated data are shown produced at the SM
value Yt = 1.0 (red, dashed) and at the final best-fit value of Yt = 1.16 (blue,
dashed).
 116

Chapter 9

Concluding Remarks

“The goddess of learning is fabled to have sprung full-grown from the

brain of Zeus, but it is seldom that a scientific conception is born in its
final form, or owns a single parent. More often it is a product of a series
of minds, each in turn modifying the ideas of those that came before, and
providing material for those that come after.”
— J.J. Thomson

The standard model of particle physics has been wildly successful at explaining
the behavior of matter at the smallest scales and highest energies which can be
probed in a lab setting. However, for as many successful predictions as it makes,
it raises an equal number of questions. The most famous of these questions arise
when the standard model is compared to large scale behavior of the observed
universe. But, as we have seen, the standard model leaves many open paths worthy
of further pursuit–from neutrino mass generation, to dark matter candidates, to
 117

precision coupling measurement. Even QFT, which forms the basic framework for
the standard model, has mysterious mathematical underpinnings and a host of
unsolved conjectures.

In this document, we have examined two very different approaches to research

in fundamental particle physics pursued by the author, while providing broader
context to each. First, in Chapter 4, we discussed ways of examining reduced-
dimensional quantum fields as a way to study the framework that governs particle
interactions. There is much left to be understood here, and there are unsolved
problems that have stood for over half a century. The path forward remains com-
pletely unknown. Some of the same problems can be approached from either the
field of mathematics or the field of theoretical physics, two groups which see the
issues from surprisingly different perspectives. In attempting to learn pieces of both
perspectives, the author performed detailed studies of reduced dimensional sys-
tems, focusing on both mathematical [28] and physical [32] properties of quantum
fields. While this work hardly answers the most fundamental questions involving
QFT, both projects demonstrate that there are remaining avenues of study that can
produce interesting, unintuitive results.

Secondly, in Chapters 7-8, we looked at the CMS experiment and a novel mea-
surement of the top-Higgs interaction strength from Ref. [79]. Though we see a
measured value consistent with the standard model measurement, we also see
that further study with greater precision is needed. Because the top quark and
Higgs boson are both the newest particles to be discovered and the heaviest known
to exist, it makes sense to study their properties and interactions in detail. This
interaction has only recently begun to be measured, and it is important to continue
performing such measurement with a variety of orthogonal techniques. Historically,
 118

precise measurements at a given energy scale have often provided clues about what
will be seen at higher energies, and even small deviations from what is expected
of the Higgs properties could have huge ramifications.The LHC will continue to
collect data for years to come, and for decades as the upgraded HL-LHC. We will
have to wait and see what the program yields in the long term.

It is impossible to know which direction of research will yield groundbreaking

results. And it is sometimes said that the state of physics in the early 21st cen-
tury bears a striking resemblance the early 20th century. Back then, the story goes
that many believed all fundamental rules of matter and energy had been laid out
by Newton’s gravity, atomic theory, thermodynamics, and Maxwell’s electromag-
netism. Some say that these models were so successful, it created a sense that it
would be difficult to progress much further. And yet, physicists at the time un-
knowingly sat at the edge of a revolution, in which general relativity and quantum
mechanics would upend their understanding of light, matter, and gravity.

Of course, groundbreaking physics results have continued in fields not men-

tioned here. But particle physics of late seems to have slowed its pace. What is to
become of our fundamental understanding of particles, the tiny pieces that make
up the universe and ourselves? Could we be sitting on the edge of a new revolution,
that will upend our understanding in the coming century? If so, where might it
come from? Or has the standard model indeed brought us close to a limit of sorts,
such that particle experiments in the foreseeable future will only serve confirm the
model we already have? Only time will tell, but it’s best to keep an open mind.
 119

Appendix A

Continuity of Scalar Fields With

Logarithmic Correlations

Partial reprinting of Ref. [28] with permission, co-authored with S.G. Rajeev.

A.1 Introduction

In traditional geometry, the distance between two points is the length of the shortest
curve that joins them. This fits well with classical physics, as this shortest path is
the one followed by a free particle. But in quantum physics, the shortest one is only
the most likely of many paths that the particle can take. Moreover, no particle can
follow a path connecting two spacelike separated points. Taking these facts into
account, we should hesitate to associate distance with the length of one particular
curve. Instead, we can average over all paths connecting two points, yielding the
Green’s function of a quantum field (also called the two point function, correlation
 120

function or propagator.) A metric does emerge out of the correlation, but turns out
to be non-Euclidean [29].

The idea of defining a metric from the correlation of a random process is a staple
of modern stochastic analysis [80, 80–82]. This can be illustrated with Brownian
motion. The Brownian paths are continuous, but not differentiable with respect to
the usual time parameter. A particle executing Brownian motion is knocked around
by other particles in the medium. As the time between collisions tends to zero, the
velocity at any instant is no longer a physical quantity. Furthermore, even the speed
| B( x )− B(y)|
cannot be bounded; as x → y the probability of | x −y|
being bounded is zero
(To make comparison with a quantum field easier, we call the time parameter of
the Brownian process x rather than t. Since the diffusion constant has dimension
(length)2 /time, dimensional analysis suggests that

| B( x ) − B(y)|
p (A.1)
| x − y|

would be a better quantity to measure the speed of a Brownian particle. But it

turns out that even this is unbounded with probability 1 (more commonly stated
as “almost surely” or a.s.) as x → y. The proper way to quantify the time that
p
has elapsed between two measurements is not | x − y| or even | x − y|. We seek a
metric ω with respect to which the sample paths are locally Lipschitz continuous,
| B( x )− B(y)|
meaning ω ( x,y)
is almost surely bounded as x → y . The correct such “modulus
of continuity,” attributed to Lévy, is

s
1
ω ( x, y) ∝ | x − y| log (A.2)
| x − y|
 121

for small | x − y|. This quantifies the roughness of Brownian paths (One can bound
√
the variations precisely with a proportionality constant 2, but we will generally
ignore multiplicative constants in discussing continuity/roughness here).

We look at the spatial metric in the simplest relativistic theory, a massless scalar
quantum field in 1 + 1 dimensions. Such logarithmically correlated fields have
generated interest in purely mathematical contexts, and have potential applications
in areas ranging from finance to cosmology (see [27]). It is enough to understand
the continuity of sample fields in the ground state; those in any state of finite
energy will exhibit identical behavior over small distances (see the appendices in
the original publication of Ref. [28] for supplementary discussion of the ground
state wavefunction).

A complication is that scalar quantum fields are random distributions rather

than functions: φ( x ) at some point in space is not a meaningful quantity. But we
will show that a mild smoothing procedure (averaging over an interval) is enough
to get around this difficulty, yielding a continuous but not differentiable function.
This average can be viewed as a model for the potential measured by a device: such
a measurement will always take place over some finite width. A peculiar property
of the logarithmically correlated field is that the probability law of the average is
independent of the size of the interval (size of the measuring device). That is, the
field does not appear any rougher if we average over smaller intervals.

We then obtain a result analogous to that of Levy: a metric in space with respect
to which the scalar field is a.s. Lipschitz (we will use this term exclusively in
the sense of local continuity). Our result is a particular case of the much deeper
mathematical theory of regularity of random processes [80, 80–82] . The idea of
 122

using a moving average (instead of an inner product with smooth test functions)
seems to be new, and yields simple explicit results.

We then apply this moving average technique to other random fields of physical
interest, noting that a new procedure is sometimes needed if the field considered
has more severe divergences. Supplemental material can be found in the original
publication of Ref. [28] discussing the intimate relationship of this work to the
resistance metric on a lattice, connecting to an earlier paper [29], and describing
connections to a functional analytic approach to regularity of random processes.

Although we work with Gaussian fields in this paper, the short distance behavior
is the same for asymptotically free interacting fields (up to sub-leading logarithmic
corrections). The regularity of renormalizable but not asymptotically-free theories
(such as QED or the Higgs model) can be quite different. The strength of interactions
grow as distances shrink, possibly leading to a singularity (Landau pole). In the
case of QED, we know that this is not physically significant, due to unification with
weak interactions into a non-Abelian gauge theory.

But the question of what happens to the self-interaction of a scalar quantum

field (Higgs boson) at short distances is still open. In the absence of evidence at the
LHC for supersymmetry or compositeness of the scalar, we have to consider the
possibility that the Higgs model is truly the fundamental theory. The short distance
behavior is dominated by interactions, necessitating new mathematical methods
beyond perturbative renormalization. The extensive mathematical literature [80,
80–82] on continuity of non-Gaussian processes ought to contain useful tools for
physics. In order to apply this work to a full interacting theory, we must first
know what happens in the simpler case of a free theory. This is part of the physical
motivation for this paper.
 123

A.2 Continuity

A.2.1 Continuity of Random Processes

A random process r ( x ) assigns a random variable to each value of x in some space

X. The quantity

q
d( x, y) = h[r ( x ) − r (y)]2 i (A.3)

satisfies the triangle inequality and so defines a metric on X (provided we identify

any originally distinct points x, y for which d( x, y) = 0).

This metric need not be Euclidean or even Riemannian. A standard example

p
is Brownian motion, where d( x, y) = | x − y|, which is neither. Supplementary
material on another simple example can be found in the appendices of the full
publication, Ref. [28].

One commonly successful approach to the study of continuity is to leave behind

the intuitive structure associated with the space X, and begin instead by looking
at structures related to the process of interest (such as the metric d above). At
first, one might expect that the sample paths for a random process r ( x ) will be
necessarily continuous with respect to d. Although true for Brownian motion,
almost sure continuity with respect to d does not hold in general. For Gaussian
processes, a sufficient condition for continuity is the convergence of the Dudley
integral [80, 80–82]

Z δq
J (δ) = log N ( D, e) de, δ<D (A.4)
0
 124

where N ( D, e) be the minimum number of balls of radius e it takes to cover a ball

of radius D in ( X, d) (We suppress the D dependence of J for simplicity of notation).
The possible divergence comes from the lower limit of the integral e → 0.

Can we go beyond continuity? To speak of differentiable functions, a metric

is not enough: we would need a differentiable structure on X which we do not
have intrinsically. The closest analogue to differentiable functions on a metric space
| f ( x )− f (y)|
( X, d) are Lipschitz functions, for which d( x,y)
is bounded. For comparison,
differentiable functions on the real line are Lipschitz, but not all Lipschitz functions
are differentiable. Of course, all Lipschitz functions are continuous.

Even in cases where J (δ) converges, indicating that the sample paths are contin-
uous, they may still not be Lipschitz with respect to the metric d above. Again, the
Dudley integral comes to the rescue: using it we can define a more refined metric

ω ( x, y) = J (d( x, y)). (A.5)

p
Since log N ( D, e) is a decreasing function of e, J (δ) is a convex function. Thus
J (d( x, y)) satisfies the triangle inequality as well.

The sample paths of a Gaussian process for which J (δ) converges are [83] a.s.
Lipschitz in this refined metric ω. Thus ω, rather than d, is the metric (“modulus of
continuity”) we must associate to a Gaussian random process.

What would one do if the Dudley integral does not converge? There is a more
general theory [81] which gives necessary and sufficient conditions for continuity: a
“majorizing measure” must exist on X . We will not use this theory in this paper, but
 125

hope to return to it, as it can deal with more general cases than Gaussian processes
(e.g., interacting quantum fields).

A.2.2 Quantum Fields

In this paper, we consider a quantum scalar field φ in the continuum limit. In the
trivial case where φ is massless with 1 spatial dimension and no time dimension,
the correlations of Brownian motion are reproduced and φ remains a continuous
function. However, in any fully relativistic field theory, φ lives on a space of
distributions, not functions. To get a sensible random variable, we must then take
the inner product with respect to some test function h with zero average.

Z Z
φ[h] = φ( x )h( x )dx, h( x )dx = 0. (A.6)

We study the case where φ is a distribution with the weakest possible singulari-
ties; one might say we want a field that is “close” to being a function. The obvious
candidate is the case of logarithmic correlations (For a recent review, see [27])

Z
0
hφ[h]φ[h ]i = − log | x − y| h( x )h0 (y)dxdy. (A.7)

The corresponding Gaussian measure can be thought of as the (square of the)

ground state wavefunction of a massless scalar field in 1 + 1 dimensions. (More
 126

precisely, the continuum limit of the resistance metric of a row on an infinite square
lattice, as discussed further in the supplementary material of Ref. [28]).
R
The condition h( x )dx = 0 ensures that the covariance is unchanged if log | x −
y| is replaced by log λ| x − y|, meaning φ is scale invariant. Since φ has the physical
meaning of a potential, observables such as φ[h] must be unchanged under a shift
R
φ( x ) 7→ φ( x ) + a, which equivalently suggests the requirement h( x )dx = 0.

A.2.3 Moving Average of a Quantum Field

Quantum fields which are only mildly singular can act on test functions which are
not smooth or even continuous. It is not necessary to consider the whole space of
test functions as in [82]; in this paper our test functions will be piecewise constant
with compact support and zero mean.

We define a moving average of φ:

Z 1
2
φ̄s (u) = {φ (s [u − w]) − φ (s [0 − w])} dw, s > 0. (A.8)
− 12

This is the inner product of φ with a discontinuous test function h that has support
R
on two intervals of width s based at su and at 0; the sign is chosen so that h( x )dx =
0. The probability law of φ̄s is not translation invariant: the second term ensures the
boundary condition

φ̄s (0) = 0. (A.9)

 127

We will mostly work with the quantity

Z 1
2
φ̄s (u) − φ̄s (v) = {φ (s [u − w]) − φ (s [v − w])} dw (A.10)
− 21

which has a translation invariant law. It is convenient to rescale the coordinate

of the midpoint by the width (as we have already done), so that the variable u is
dimensionless. Then the quantity

q
ρ(u, v) ≡ h[φ̄s (u) − φ̄s (v)]2 i, (A.11)

which is just a special case of (A.3), is finite and defines a metric. Moreover, it
is independent of s in the logarithmically correlated case. This means the process
φ̄s (u) − φ̄s (v) has a probability law that is independent of s: a consequence of scale
invariance, which is specific to logarithmic correlations. As an interesting aside, we
note that sφ̄s (u) produces a solution to the wave equation in u and s.

The moving average does not depart from the essence of the standard idea of
averaging over a test function. It is simply that a piecewise constant test function is
especially convenient for a mildly singular quantum field as opposed to a smoother
function. For more singular fields (e.g. scalar field in four dimensions) we would
have to revert to more regular test functions.
 128

Figure A.1: The behavior of ρ2 (r ) and ρ(r ), setting s = 1.

A.3 Logarithmically correlated scalar field in 1

dimension

In the logarithmically correlated case, we obtain explicit formula

ρ(u, v) = ρ(|u − v|) (A.12)

q
ρ (r ) = L(r + 1) + L(r − 1) − 2L(r ), (A.13)

Where

1
L(r ) = r2 log r2 . (A.14)
2

Being a convex function of r = |u − v|, this ρ(u, v) will satisfy the triangle
inequality (not true of ρ2 , as seen in Fig. A.1). Thus, ρ defines a translationally
invariant metric.
 129

Figure A.2: We show an approximate φ and two averages φ̄Λ,s where Λ = 4000, L = 5 and
we average over width s = 0.05 (blue/dark) and s = 0.2 (orange/light). We see
that despite the factor of 4 difference in averaging windows, the two appear
interchangeable, demonstrating the scale invariance even for this approximate
representation. The two appear to follow the same law, and the continuity is
seen to be greatly improved.

Simple calculations (see Sec. A.4) show that the Dudley integral J converges, so
that φ̄s is a.s. continuous in ρ. Moreover we can construct a refinement

1
J (ρ(u, v)) ≡ ω (u, v) ≈ |u − v| log (A.15)
|u − v|

with respect to which φ̄s is a.s. Lipschitz. This is a “modulus of continuity” for the
moving average of a quantum field, analogous to that of Lévy for Brownian motion.
(Note that there is no square root, however.)

We can obtain a crude picture of the moving average process by generating noise
which has the same power spectrum as a log-correlated field, but with some high
frequency cutoff. This is given by the Fourier series

Λ
1 h  πmx   πmx i
φΛ ( x ) = ∑ m m√ X cos
L
+ Ym sin
L
(A.16)
m =1
 130

where Xm , Ym are independent standard Gaussian variables. For large Λ (ultraviolet

cutoff) and L (the infrared cutoff, − L < x < L ) this creates an intuitive “approxi-
mation” to the divergent field. Such a technique is often used to visualize white
noise. While one must be careful claiming to “approximate” a distribution with a
truncated series, φΛ gives us some sensible object on which to numerically test the
properties of our moving average. This is carried out in Fig. A.2. We see that φ̄s has
the desired properties without requiring the full distribution.

While we focus on the log-correlated field for its mathematical simplicity, it is

worth noting that such objects are not necessarily confined to the realm of math-
ematical fantasy. A free scalar field in one dimension can in principle be a good
approximation for a real physical system, with one possible example being the
electromagnetic field of certain optical fibers. If the refractive index of the fiber is
chosen appropriately, only a finite number of transmission modes will be allowed.
We can think of the wave equation as analogous to the Schrödinger equation, with
the variable refractive index providing an effective potential. This potential can be
chosen to allow only a finite number of bound states. Single-mode fibers have only
one such state, leading to a system with one effective spatial dimension.

Even in the absence of light in the fiber (ground state of the electromagnetic
field), there will be quantum fluctuations in the potential. In the absence of severe
nonlinearities, these fluctuations can be modeled as two noninteracting scalar fields,
one for each polarization mode. If the wire is transparent over a sufficient frequency
(maintaining its single-mode property and minimal dispersion for propagating
waves), then the potential difference between two points will be a Gaussian random
variable whose variance is approximately logarithmic with distance. The measure-
ment of the potential would require a probe of finite size, so the averaging process
 131

employed in this paper provides a convincing model for the potential as seen by
a measuring apparatus at a given instant. The considerations of this paper can
be viewed as a model of the spatial regularity of the electromagnetic potential in
such an optical fiber. This model could also, in principle, describe the ground state
fluctuations of a quantum system confined to a very narrow region in 2 spatial
dimensions, sometimes called a quantum wire.

Perhaps an experimental test of the sample field behavior in Fig. A.2 is indeed
possible. However, the details of realizing such a system and carrying out such
measurements is highly nontrivial and not suited to the themes of this paper; we
include this discussion mainly as a reminder that lower dimensional systems are
often not so unphysical as they seem.

A.4 Explicit calculations and further examples

A.4.1 Variance of φ̄s for the log-correlated field

The calculations that justify the above assertions are straightforward, but worth
outlining as they help illuminate the properties discussed above. Because of the
divergences, we cannot use the standard approach directly to the quantum field,
but only to its moving average. Begin with the observation that

Z b Z d
F ( a, b, c, d) ≡ − dx dy log | x − y| (A.17)
a c

3 1
= ( a − b)(c − d) + [ L(c − a) − L(d − a) − L(c − b) + L(d − b)] , (A.18)
2 2
 132

where L is defined in (A.14). Note this quantity is not quite scale invariant: there is
an “anomaly” proportional to log λ.

L(λr ) = λ2 L(r ) + r2 log λ (A.19)

F (λa, λb, λc, λd) = λ2 F ( a, b, c, d) − (b − a)(d − c) log λ (A.20)

Then

F ( a, b, a, b) F (c, d, c, d) F ( a, b, c, d)
h[φ̄s (u) − φ̄s (v)]2 i = 2
+ 2
−2 , (A.21)
(b − a) (d − c) (b − a)(d − c)

where

F ( a, b, a, b) 3 1
= − log(b − a)2 (A.22)
( b − a )2 2 2

which only depends on the width of the interval [ a, b]. We can then consider two
intervals of equal width s, centered at su and sv, yielding

2  s s s s
h[φ̄s (u) − φ̄s (v)]2 i = 3 − 2 log(s) − F su − , su + , sv − , sv + (A.23)
s2 2 2 2 2

From the scale transformation property above of F we can see that this quantity
is independent of s: the “scale anomaly” of F cancels against 2 log s. So we can
simplify by putting s = 1 and expressing F in terms of L:

h[φ̄s (u) − φ̄s (v)]2 i ≡ ρ2 (u, v)

(A.24)
= L(u − v − 1) + L(u − v + 1) − 2L(u − v)
 133

as was claimed.

A.4.2 Continuity of Brownian Paths

In using the Dudley integral, it is useful to begin with a well-known example. The
most familiar example of a Gaussian process is Wiener’s model of Brownian motion,
p
for which d( x, y) = | x − y| . If an interval [0, 1] is divided into N equal parts,
q  
1
each part is contained in a d−ball of radius e = 2N . Thus N (e) = 1 + Floor 12
2e
and for small δ, [where Floor( a) is the integer part of the real number a]

p
J (δ) ≈ δ −2 log δ. (A.25)

Thus Brownian sample paths B are almost surely continuous. More quantitatively,
we may construct

q
ω (r ) = J (d(r )) = r log(1/r ). (A.26)

to obtain the result of Lévy that, with probability one,

| B( x ) − B(y)|
q <C (A.27)
| x − y| log |x−1y||

as x → y for some constant C.

 134

A.4.3 Continuity of φ̄s for logarithmically correlated fields

We can now show that the sample paths φ̄s (u) are continuous with probability one.
 
Again, if [0, 1] is divided into N intervals, each will have radius e = ρ N1 . To get
small e, we must choose a large N; using the asymptotic behavior

p
ρ (r ) ≈ r − log r (A.28)

for small r,
s  
1 1 1p
e≈ − log =⇒ N (e) ≈ − log e. (A.29)
N N e

The Dudley integral converges:

r
1
J (δ) ≈ δ log , δ→0 (A.30)
δ

J (ρ(r )) ≡ ω (r ) ≈ r log(1/r ) (A.31)

which yields the claimed modulus of continuity.

A.4.4 Additional Examples for Comparison

A.4.4.1 Moving Average of Brownian Paths

It is informative to apply the moving average procedure to the Brownian case,

where the paths B( x ) which we average over are continuous functions to begin
 135

with. Proceeding analogously, consider two intervals with width s with centers su
and sv respectively. Then we can define, analogous to (A.17) but with some added
foresight,

Z su+s/2 Z sv+s/2
 s s s s
F su − , su + , sv − , sv + ≡− dx dy| x − y| (A.32)
2 2 2 2 su−s/2 sv−s/2


   
 1 r3 −3s2 + 3s − 1 + 3r2 s3 + s3


3 0<r<s
= (A.33)
 s3 r

 r > s,

where r = |u − v|. We then have

2
h[ B̄s (u) − B̄s (v)] i ≡ ρ2 (r ) (A.34)
2s 2  s s s s
= − 2 F su − , su + , sv − , sv + . (A.35)
3 s 2 2 2 2

It is easily seen that ρ2 (λr ) = λρ2 (r ), breaking scale invariance. Still for comparison
purposes, we consider averaging over intervals of width s = 1, noting that the
scaling behavior will only change ω (r ) by a constant factor. As before, ρ2 does not
define a metric, but its square root ρ does. In the large-r limit we have

√
ρ (r ) ∼ r, (A.36)

while for small r,

ρ(r ) ∼ r. (A.37)
 136

This short distance behavior suggests by dimensional analysis that B̄s ( x ) might be
Lipschitz in the usual metric |u − v|, but the Dudley integral yields a weaker limit

 
1
ω (r ) ≈ r log (A.38)
r

| B̄s (u) − B̄s (v)|

< C. (A.39)
r log( 1r )

Thus B̄s is just shy of being Lipschitz in the usual metric, but is a.s. Lipschitz with
1
respect to the metric ω (u, v) ≈ |u − v| log |u− v|
. Interestingly, this is the same ω
we obtained in (A.31) for the log-correlated case, even though the short distance
behavior of ρ is not quite the same (the difference in the Dudley integral vanishes
for small δ). However ω for the Brownian sample paths prior to averaging (A.26)
contains a square root not present here.

A.4.4.2 Power Law Correlations in 1D

We can use the same method as with Brownian motion to consider the moving
average of a more general power-law correlated field such that

Z Z
0 α 0
hφ[h]φ[h ]i = sign(α) | x − y| h( x )h (y)dxdy, h( x )dx = 0. (A.40)

When α > 0 this is related to fractional Brownian motion [84]. When α = −1 it is the
restriction to one dimension of a massless scalar quantum field in 2 + 1 dimensions.
The moving average is no longer independent of the width of the intervals. Still, for
purposes of comparison, we consider the average on intervals of fixed width s = 1.
 137

It is not difficult to evaluate the integrals to find that, in the small r limit,


r2




 α>0


2
ρ ( r ) ∼ r α +2 −2 < α < 0, α 6= −1 (A.41)





r log r
 α = −1.

The moving average is Lipschitz with respect to the modulus



r log(1/r )
 α>0
ω (r ) = (A.42)
α
r 2 +1 log(1/r )

 −2 < α < 0.

When α ≤ −2 the divergences are such that the moving average is an insufficient
tool to smooth the quantum field. Note that ω (r ) is the same in the logarithmic case
as the case where α > 0. The logarithmic case can be thought of as the critical case
where the smoothness implied by Dudley’s criterion starts to lessen.

A.4.4.3 Log Correlated Scalar Field in 3D

It is useful to work out a case in higher dimensions as well. The massless scalar
field in n + 1 space-time dimensions has correlation

1
φ( x )φ(y) ∝ . (A.43)
| x − y | n −1

Thus for n > 1 will we get power law, instead of logarithmic correlations. Yet a
logarithmically correlated, nonrelativistic, scalar field in 3 space dimensions is still
of interest in cosmology [27, 85]. As with the log-correlated scalar field in 1D we
 138

must average it over a test function

Z Z
0 0 3 3
hφ[h]φ[h ]i = − log | x − y|h( x )h (y)dx dy , h( x )dx3 = 0. (A.44)

Recall that

1 d3 k
Z
ik· x
− log | x | + const = c e (A.45)
|k|3 (2π )3

where the constant c = 2π 2 . The integral is not absolutely convergent, so we define

it through zeta regularization.

We perform our moving average over the interior of a sphere with radius t,
centered at tu

Z
φ̄t (u) ≡ {φ (t [u − w]) − φ (t [0 − w])} dw (A.46)
|w|≤1

Z
hφ̄t (u)φ̄t (v)i = φ (t [u − w1 ]) φ (t [v − w1 ]) dw1 dw2 (A.47)
|w|≤1

1 d3 k
Z Z
ik·(u−v)t
=c 2
3
e 3
e−ik·(w1 −w2 )t dw1 dw2 . (A.48)
|k| (2π ) |w|≤1

Taking t = 1, this can be reduced to the form

h[φ̄t (u) − φ̄t (v)]2 i = 26 π 4 G (r ) (A.49)

 139

where

Z ∞  
1 2 sin kr
G (r ) = dk [sin k − k cos k] 1− (A.50)
0 k7 kr

and r = |u − v|. We are not able to evaluate the integral analytically, but its
convergence is clear, justifying the independence on t. In the large r limit, the
integral is dominated by small k contribution. We then have

Z ∞  
1 sin kr
G (r ) ≈ dk 1− (A.51)
0 3k kr

∼ log(r ) + O(1). (A.52)

In the case of small r, the dominant contribution comes from the first peak of
1
[sin k − k cos k]2 , which must occur for k < 2π (i.e., k ≈ 5.678). This allows us to
k7
treat kr as small, yielding the behavior

Z ∞
2 1
G (r ) ≈ r dk 5
[sin k − k cos k]2 ∼ r2 . (A.53)
0 3k

The approximation can be verified numerically. This small r behavior dictates

the continuity modulus discussed above. Namely, we have that for the 3D log-
correlated scalar field,

ρ(r ) ∼ r, (A.54)

ω (r ) ∼ r log(1/r ). (A.55)
 140

A similar metric can be obtained for a log correlated field in other dimensions. Note
that, once we have ρ(r ) ∼ r for small r, the logarithm in the Dudley integral ensures
that ω will not depend on the dimensionality (up to proportionality). This is not
true if ρ(r ) has some other short distance behavior.

A.5 Outlook

Gaussian processes correspond to free fields. The most elegant way to introduce
interactions into a scalar field theory is to let it take values in a curved Riemannian
manifold. This is the nonlinear sigma model in physics language, or the wave map
in the mathematical literature. In 1+1 dimensions, such a theory, with a target space
of a sphere or a compact Lie group, is well studied in the physics literature. The short
distance behavior is approximated by free fields with corrections computable in
perturbation theory (asymptotic freedom). The only case for which mathematically
rigorous results are known is that of the Wess-Zumino-Witten model, which has
non-Gaussian behavior at short distances; i.e., a “nontrivial fixed point” for the
renormalization group. The related measure for the ground state of the quantum
field has been constructed by Pickrell. (For a review, see [86]). It is natural to ask for
regularity results analogous to ours in this case.

Looking further out, it would be interesting to quantify the regularity of quan-

tum fields of the nonlinear sigma model in two dimensional space time; and even
further out, λφ4 theory in four dimensions. It is possible that the “naturalness
problem” of the standard model of particle physics has a resolution in terms of
such a deeper understanding of the regularity of scalar quantum field theory. The
“modern” theory [81] of regularity of non-Gaussian processes ought to help with
 141

this daunting task. Even harder is the case of Yang-Mills fields. An analogue
of our moving average is the Wilson loop. The measure of integration over the
space of gauge fields is only known rigorously for the two dimensional case [87].
Regularity of Yang-Mills fields satisfying classical evolution equations (let alone
random processes) is already a formidable problem under active investigation (see
for example [88]).
 142

Appendix B

Highly nonlinear wave solutions in a

dual to the chiral model

Reprinted from Ref. [32] with permission. Co-authored with S.G. Rajeev.

B.1 Introduction

We study the field theory [89–91] with equations of motion

h i
φ̈ = λ φ̇, φ0 + φ00 , (B.1)

where φ is valued in a Lie algebra, φ : R1,1 → su(2). This follows from the action

 
1 2 1 02 1
Z Z
0
S1 ≡ L1 dxdt = Tr φ̇ − φ + φ[φ̇, φ ] dxdt. (B.2)
2λ 2λ 3
 143

In the λ → 0 limit, these equations admit linear wave solutions. But in the high-
coupling regime, the theory is dominated by nonlinear effects.

S1 is closely tied to other models and subjects, which we elaborate on in section

B.2. These include the study of slow light, the Wess-Zumino-Witten (WZW) model,
and the mathematical theory of hypoelliptic operators. Of particular interest in
this paper, the model described by S1 is also classically dual to the well-studied
principal chiral model, described by the action

 2 
1 −1 0 2
Z Z  
−1 2
S2 = L2 dxdt = Tr g ġ −c g g dxdt (B.3)
2f

where g : R1,1 → SU (2). This is a special case of the nonlinear sigma model, with
target space SU (2).

Despite their classical equivalence, S1 and S2 lead to entirely different quantum

theories. S2 gives an asymptotically free theory: at short distances f → 0, giving us
free massless excitations. But the true particles that survive to long distances are
bound states of non-zero mass [92, 93]. For this reason, the principal chiral model
is often used as a toy model for 4-dimensional Yang-Mills theory, notorious for its
mathematical complexity. Not only do the two theories share similar short-distance
behavior, but the existence of a mass gap in the principal chiral model has served as
a proof of concept for the conjectured mass gap in Yang-Mills (though neither can
yet be proven with full mathematical rigor).

S1 , on the other hand, leads to a renormalizable but not asymptotically free

quantum theory. At short distances the coupling constant λ → ∞, while at long
distances we have weakly nonlinear massless excitations. It makes sense to use S1
as a 2-D toy model for strongly coupled theories, in particular 4-dimensional λφ4
 144

theory1 . The behavior of quantum field theories at high coupling is notoriously

intractable, and the physical meaning of such theories is still up for debate. For this
reason, it is still necessary to search for simple examples of such theories and try to
gleam what meaning, if any, they have in the short distance limit.

In addition to sharing short-distance behavior, both the S1 model and λφ4 theory
can be described by hypoelliptic hamiltonian operators with a step-3 nilpotent
bracket algebra, suggesting some algebraic structure in common (see Section B.2.2
and the supplementary material in the original publication of Ref. [32] for more
detail). The S1 model’s relative simplicity makes it a good candidate for attempting
to probe the high coupling regime of field theories in general, but the connection to
λφ4 theory seems the closest. Additionaly, its classical duality to the principal chiral
model motivates a juxtaposition of the two theories in the classical and quantum
formulations.

To glimpse what becomes of our theory in the high coupling limit, we take
the modest approach of finding nonlinear wave-type solutions to the classical
model which survive the λ → ∞ limit (section B.3). This set of solutions defines
a mechanical system or “reduced system” in each of the dual models. While they
physically appear very different, the resulting classical solutions can be mapped
from one system to another. We quantize these collective variables to determine their
dispersion relation (section B.4) in the short distance limit for each theory. We have
in mind the sine-Gordon theory, whose solitons turns out to be the fundamental
constituents which bind to form the scalar particles [94, 95].

These reduced quantum theories yield two different results. In particular, the
reduced model of S1 has an exotic dispersion relation in the short distance limit. We
1
that is, pure λφ4 theory, describing a Higgs-like particle with no coupling to fermions
 145

postulate that its spectrum may hint at the fundamental constituents of the highly
coupled theory, which need not behave like traditional particles at all. In section
B.5 we offer concluding remarks and a side-by-side comparison of our work with
S1 and S2 .

On The Notation

1 1
We regard φ = 2i [φ1 σ1 + φ2 σ2 + φ3 σ3 ] = 2i φ · σ as a traceless anti-hermitian matrix.
Recall then that the commutator and cross product are related by

1
[ X, Y ] = (X × Y) · σ. (B.4)
2i

Also, we define TrX ≡ −2trX so that

Trφ2 = φ12 + φ22 + φ32 . (B.5)

In relativistically invariant notation, (B.1) and (B.2) can be written as

λ
∂ µ ∂ µ φa − eabc eµν ∂µ φb ∂ν φc = 0, (B.6)
2

1 1
Z Z
a a µν 2
S1 = ∂µ φ ∂ν φ η d x + eabc φ a ∂µ φb ∂ν φc eµν d2 x (B.7)
2λ 6

where µ, ν = 0, 1 and a, b, c = 1, 2, 3 ; also, eµν , eabc are the Levi-Civita tensors. This
is a particular case of the general sigma model studied in [96] as the background of
string theory, with a flat metric on the target space and a constant 3-form field eabc .
 146

B.2 relation to other models

B.2.1 The c → 0 Limit and Slow Light

Consider the equations of motion (B.1) where the speed of linear propagation at
low coupling is taken to be c rather than 1:

h i
φ̈ = λ φ̇, φ0 + c2 φ00 . (B.8)

If we rescale φ → λ a φ, t → λb t, this becomes

λ a−2b φ̈ = λ1+2a−b φ̇ × φ0 + c2 λ a φ00 . (B.9)

Set a = 2b and 1 + 2a = b to get

2
φ̈ = φ̇ × φ0 + c2 λ− 3 φ00 . (B.10)

Thus the strong coupling limit λ → ∞ at fixed c is equivalent to the limit c → 0

with λ = 1:

φ̈ = φ̇ × φ0 . (B.11)

The strongly coupled limit can be thought of as the limit in which the waves
move very slowly. It has been noted in that literature [97] that when the speed of
light in a medium is small, nonlinear effects are magnified. Although the specific
equations appearing there are different, it is possible that the solutions of the sort
we study are of interest in that context as well.
 147

From a field theoretic context, the equivalence of these limits seems troubling.
At short distances, the highly-coupled theory will not be relativistic. It is a sort of
“post-relativistic” regime, where c → 0. This is much the opposite of the case in the
theory of S2 ; there the short-distance excitations are massless, but form massive
bounds states which survive to long distances and can be non-relativistic in the
traditional c → ∞ sense. Perhaps some exotic excitations at high coupling are in
fact the fundamental constituents in the S1 -model, forming as bound states the
ordinary massless particles which appear in the long distance limit. As we know
from the quark model, the short distance excitations do not need to be particles in
the usual sense; they could be confined. In any case, it is important to know what
solutions might survive the high coupling limit, whether they be unphysical or
simply unintuitive.

We will see an example of wave solutions which classically survive the c → 0

limit, continuing to propagate through nonlinearity alone. Since the energy density
is constant, these solutions do not violate causality: they are analogous to the
Continuous Wave solutions in a medium where the phase velocity is greater than c.

B.2.2 Sub-Riemannian Geometry and the Strong Coupling Limit

Many physical problems (Yang-Mills, Fluid Mechanics) become intractable in the

strong coupling limit where the non-linearities dominate. It would be nice to
have a unified geometric approach to understanding these systems. We have
such an approach in the weak coupling limit: small perturbations around a stable
equilibrium are equivalent to a harmonic oscillator.
 148

A larger picture emerges if we think in terms of Riemannian and sub-Riemannian

geometry. The orbits of many mechanical systems of physical interest (again, Yang-
Mills or incompressible Fluids) can be thought of as geodesics in some appropriate
Riemannian manifold. In the simplest case, the harmonic oscillator describes
geodesics in the Heisenberg group. The anharmonic oscillator (and many nonlinear
field theories with quartic coupling) can also be thought of as geodesic motion on a
nilpotent Lie group, by introducing an additional generators (see the supplementary
material in the full publication of Ref. [32] for additional discussion).

In the limit of strong coupling, the metric degenerates and becomes sub-Riemannian
[98]. That is, the contravariant metric tensor has some zero eigenvalues so that it can
be written as ∑ j X j ⊗ X j for some vector fields X j whose linear span may be smaller
than the tangent space. Moreover, in the cases of interest, these vector fields satisfy
the celebrated Hörmander condition: X j along with their repeated commutators
span the tangent spaces at every point. In such a case, there are still geodesics
connecting every pair of sufficiently close points (Chow-Rashevskii theorem, [98]).
Thus, we can define a distance between pairs of points as the shortest length of
geodesics.

These ideas came to the notice of many physicists following a model for the self-
propulsion of an amoeba [99], though they have roots in the Carnot-Caratheodory
geometric formalism of thermodynamics and in control theory. Hörmander [100]
discovered independently that the same criterion is sufficient for the sub-Riemannian
Laplace operator ∆ = ∑ j X 2j to be hypoelliptic, meaning the solution f to the inho-
mogenous equation ∆ f = u is smooth whenever the source u is smooth. This can be
thought of the quantum version of the above condition on subgeodesic connectivity.
 149

This kind of sub-Riemannian geometry may present a powerful geometric

framework for strongly-coupled field theories. The example we work out in this
paper is arguably the simplest interesting case of a strongly coupled field theory,
and the solutions we study correspond to sub-Riemannian geodesics in the limit
λ → ∞. We hope to apply such geometric ideas to other cases in the future, using
this as a prototype.

B.2.3 Relation to the WZW model

We can also regard our equations as a limiting case of the Wess-Zumino-Witten

model2 [101]

1 n
Z Z
−1 2
SWZW = µ
tr ∂ g∂µ g d x+ tr ( g−1 dg)3 (B.12)
4λ21 24π M3

2
as n → ∞ and λ1 → 0, keeping λ = λ21 (n/2π ) 3 fixed3 . To see this, let g( x ) =
a
(x)
ebiσa φ and expand in powers of b:

b2 n 3
Z Z
µ a a
SWZW = ∂ φ ∂µ φ + b 2eabc dφ a dφb dφc + · · · (B.13)
2λ21 24π M3

To this order the WZW term is an exact differential, so we can write it as an integral
over space-time

1 b2 n 3
Z Z
µ a a
SWZW = ∂ φ ∂µ φ + b eabc φ a dφb dφc + · · · (B.14)
2 λ21 12π

2
Witten’s Tr is our tr. His λ is our λ1 .
3
Here, M3 is a 3-manifold of which the two-dimensional space-time is the boundary. We do not
require λ21 to take the conformally invariant value 4π
n .
 150

SWZW reduces to S1 if we identify b3 = 2π/n and λ as above. By taking this

limit, we can easily get the renormalization of our model. Recall that [101] the one
loop renormalization group equation of the O( N ) WZW model is

 !2 
dλ21 λ41 ( N
− 2)  λ21 n
=− 1− (B.15)
d log Λ

2π 4π

We need the particular case of N = 4 corresponding to the target space being

S3 ≈ SU (2). Thus in our limit n → ∞, λ1 → 0 keeping λ fixed,

dλ λ4
= (B.16)
d log Λ 4π

It is useful to take this limit rather than calculating loop corrections from scratch,
as the renormalization group evolution of the WZW has been studied to high
order [102,103]. Including these higher order terms does not alter the short-distance
divergence of λ.

B.2.4 Duality with the principal chiral model

We have now seen that the S1 model is strongly coupled in the short-distance limit.
Yet, as a classical field theory, it can be viewed [89,90] as a dual to the asymptotically
free principal chiral model with equation of motion

∂µ [ g−1 ∂µ g] = 0, g : R1,1 → SU (2). (B.17)

 151

To see this, we define the currents

1
I= φ̇, J = φ0 (B.18)
λ

so the equations of motion become

1 0
J̇ = λI 0 , İ = λ[ I, J ] + J. (B.19)
λ

We can solve the second equation with the relations

1 1 −1
I= 2
g −1 g 0 , J= g ġ. (B.20)
λ λ

Then the first equation becomes

h i h i
−1 −1 0
∂0 g ġ = ∂1 g g (B.21)

which is the non-linear sigma model. Thus, the same classical equations of motion
follow from the action

 2 
1 −1 0 2
Z  
−1 2
S2 = Tr g ġ −c g g dxdt (B.22)
2f

if we identify f = λ2 . A summary of relevant correspondences in the dual models

can be found in table B.1 at the end of section B.5.

We also briefly note that our theory is closely related to the sigma model on the
Heisenberg group (see [104]).
 152

B.3 Reduction to A Mechanical System

We will look at propagating waves of the form

 
i  1 0 
[compact]φ(t, x ) = eKx R(t)e−Kx + mKx, K= k  (B.23)
2  
0 −1

for constants k, m. These solutions are equivariant under translations: the “potential”
φ changes by an internal rotation and a constant shift under translation, while the
currents only change only by the internal rotation. Thus, the energy density is
constant. They are to be contrasted with soliton solutions, which have energy
density concentrated at the location of the soliton. They are more analogous to the
plane wave solutions of the wave equation, or a Continous Wave (CW) laser beam.
Moreover, the currents

1 Kx −Kx
I= e Ṙe , J = eKx {[K, R] + mK } e−Kx (B.24)
λ

2π
are periodic in space with wavelength k . Defining

1
L ≡ [K, R] + mK, S ≡ Ṙ + K, (B.25)
λ

We can write the equations of motion and identity (B.19) in a symmetric form

L̇ = [K, S], Ṡ = λ[S, L]. (B.26)

This new choice of variables will allow us to connect to the dual theory, identify the
conserved quantities and to pass to the quantum theory more easily.
 153

B.3.1 The reduced system Lagrangian

Three conserved quantities follow immediately:

s2 k2 ≡ Tr S2

C1 k2 ≡ TrSL (B.27)
 
2 1 2 1
C2 k ≡ Tr L − KS .
2 λ

The quantity s will be of importance in the dual picture, while the other constants
have less obvious roles there. Moreover, we have the identity

TrKL = mk2 . (B.28)

Of the six independent variables in S and L, only two remain after taking into
account these constants of motion. The dynamics are described by the effective
lagrangian density (dropping a total time derivative and an overall factor of volume
of space divided by λ)

 
1 λ   1
L1 = Tr Ṙ2 + R Ṙ, [K, R] + mK − ([K, R] + mK )2
2 3 2

   
1 1 2 λ 1 1 2
= Tr (S − K ) + R S − K, L − L (B.29)
2 λ 3 λ 2

and hamiltonian density

"  2 #
1 1 1 2
H1 = Tr S− K + L , (B.30)
2 λ 2
 154

B.3.2 Reduction to One Degree of Freedom

It is useful to work with the first two components of R as a single complex variable.
Defining Z = R1 + iR2 , we can write explicitly

 
k im Z̄ 
L=  . (B.31)
2 
− Z −im

To describe the third component, we define

1 1
u≡ Ṙ3 − , (B.32)
k λ

allowing us to write a similarly compact expression for S,

 
1  uk Z̄˙ 
S=   . (B.33)
2i 
Ż −uk

The three conserved quantities (B.27) can now be written in terms of Z and u as

s2 k2 = u2 k2 + | Ż |2
ik h i
C1 k2 = Z̄ Ż − Z̄˙ Z − mk2 u
2
k2
 
2 2 2u 2
C2 k = m + + |Z| . (B.34)
2 λ

Using the identity

 2
d
| Z |2 = 4| Z |2 | Ż |2 + ( Z̄˙ Z − Z̄ Ż )2 , (B.35)
dt
 155

we can combine these three equations to eliminate Z and yield an ODE for u(t),

  
2 2 2 22 2 2 2
u̇ = k λ 2C2 − m − u (s − u ) − [mu + C1 ] . (B.36)
λ

B.3.3 Solution in terms of elliptic functions

The ODE for u(t) describes an elliptic curve. Setting u = av + b, we can pick the
constants

2 C2 λ
a= , b= (B.37)
k2 λ 3

to bring our ODE to Weierstrass normal form in terms of v:

v̇2 = 4v3 − g2 v − g3 . (B.38)

The somewhat unsightly expressions for g2 and g3 can be obtained by symbolic

computation:

1 4 2 
g2 = k λ 3C1 λm + C22 λ2 + 3s2 (B.39)
3
1 6 4
k λ 27C12 + 18C1 C2 λm + 4C23 λ2 − 36C2 s + 27m2 s2


g3 = (B.40)
108

The solution to the Weierstrass differential equation (B.38) is then

2 C2 λ
v(t) = ℘(t + α) =⇒ u(t) = 2
℘(t + α) + , (B.41)
k λ 3

where ℘ is the Weierstrass P-function and α is a complex constant determined by

the initial conditions. We can most immediately solve for R3 (t). Recalling (B.32),
 156

we have

 
2 C2 λ 1
Ṙ3 = ℘(t + α) + k + . (B.42)
kλ 3 λ

In order for to obtain a sensible solution, ℘(t + α) must be real and bounded. This
requires Im(α) = |ω2 |, where ω2 is the imaginary half-period of the Weierstrass
P-function (which depends on the elliptic invariants g2 , g3 ). The real part of α
merely shifts our solution in time, so we can take α = ω2 for simplicity. Using the
relationship

Z
℘(u)du = −ζ (u), (B.43)

where ζ is the Weierstrass ζ-function, and taking R3 (0) = 0 gives the solution

 
2 C2 λ 1
R3 ( t ) = [ζ (ω2 ) − ζ (t + ω2 )] + + kt. (B.44)
kλ 3 λ

The solution for the other two components is found by making the substitution
Z = reiθ in (B.34). Writing | Z2 | = r2 quickly yields

4 4
r2 (t) = C2 − m2 − 2 2 ℘(t + ω2 ). (B.45)
3 k λ

Note that the choice of Re(α) = 0 we made earlier implies that t = 0 is a turning
point of the radial variable, as ℘0 (ω2 ) is necessarily 0. It is useful to write

4
r2 (t) = [℘(Ω) − ℘(t + ω2 )], (B.46)
k 2 λ2
 157

where
!
2 2 C2 m2
℘(Ω) = k λ − . (B.47)
3 4

Then we can use the identity

σ(z + Ω)σ(z − Ω)
℘(z) − ℘(Ω) = − , (B.48)
σ2 ( z ) σ2 ( Ω )

where σ is the Weierstrass σ-function, in order to simplify a later result. We obtain

the solution

σ ( t + ω2 + Ω ) σ ( t + ω2 − Ω )
p
2
r (t) = (B.49)
λkσ(Ω) σ ( t + ω2 )

To find θ (t) from (B.34), we substitute ( Z̄Z − Z̄ Ż ) = −2ir2 θ̇, obtaining

C3 kmλ k2 λ2 C3 kmλ
θ̇ = 2 + = + , (B.50)
r 2 4[℘(Ω) − ℘(t + ω2 )] 2

where
" #
m3 λ
C3 ≡ k − C1 − mλC2 (B.51)
2

Using the identity

σ(z − Ω)
 
dz 1
Z
= 0 2zζ (Ω) + log (B.52)
℘(z) − ℘(Ω) ℘ (Ω) σ(z + Ω)
 158

and taking θ (0) = 0, we have

k2 λ2 C3 σ ( t + ω2 − Ω ) σ ( ω2 + Ω )
 
kmλ
θ (t) = 0 2tζ (Ω) + log + t (B.53)
4℘ ( Ω ) σ ( t + ω2 + Ω ) σ ( ω2 − Ω ) 2

We can use the Weierstrass differential equation (B.38) directly to obtain ℘0 (Ω) =
(i/2)k2 λ2 C3 , leading to a seemingly remarkable cancellation. We then have

s
σ ( t + ω2 + Ω ) σ ( ω2 − Ω )
   
iθ (t) ikmλ
e = · exp − ζ (Ω) + t . (B.54)
σ ( t + ω2 − Ω ) σ ( ω2 + Ω ) 2

Finally, a few terms cancel in the overall expression for Z, yielding

" s #
σ ( ω2 − Ω ) σ ( t + ω2 + Ω )
   
2 ikmλ
Z (t) = · exp − ζ (Ω) + t .
λkσ (Ω) σ ( ω2 + Ω ) σ ( t + ω2 ) 2

(B.55)

A sample solution is plotted in Fig. B.1. We can see that, in the R1 -R2 plane,
the solution traces an oscillating curve in between some inner and outer radius.
Meanwhile, the solution propagates in the R3 direction with non-uniform speed.
This behavior is typical over all parameter values we tested.
 159

Figure B.1: The orbit in the R1 -R2 plane (above) and the evolution of R3 with time (below).
The sample solution is plotted for 0 < t < 21 and uses parameters k = 1,
λ = 2.4, C1 = 0.5, C2 = 1, s = 2, m = 0.5.

B.4 Mechanical Interpretation and Quantization of the

Reduced Systems

The equations of motion following from the ansatz (B.26), defining the reduced
system for S1 , can be written as

R̈ = λ[ Ṙ, [K, R] + mK ] + [K, [K, R]]. (B.56)

These are the equations of motion of a particle in a static electromagnetic field, given
by (working in cylindrical polar coordinates where R1 = r cos θ, R2 = r sin θ, z =
R3 )

~B = kr θ̂ + mkẑ, ~E = k2 rr̂, (B.57)

 160

which follow from the vector and scalar potentials

~ = λk mr θ̂ + r2 ẑ ,
  k2 2
A V= r . (B.58)
2 2

The classical Hamiltonian is then

1 2 1 [ p θ − A θ ]2 1
H1 = pr + 2
+ [ p z − A z ]2 + V (r ). (B.59)
2 2 r 2

It is clear that pθ and pz are conserved. This formulation lends some physical
intuition to the solutions found in section B.3. We can pass to the quantum theory
as usual by finding the covariant Laplacian in cylindrical co-ordinates,

h̄2 1 ∂
 
∂ψ 1 1
Ĥ1 ψ = − r + 2 [−ih̄∂θ − Aθ ]2 ψ + [−ih̄∂z − Az ]2 ψ + V (r )ψ.
2 r ∂r ∂r 2r 2
(B.60)

The conservation of pθ , pz leads us to seek a solution to the Schrodinger equation

of the separable form

pz z
1
ψ(r, θ, z) = √ ρ(r )eilθ ei h̄ (B.61)
r

pθ
for integer l = h̄ . The system is then reduced to a one-dimensional Schrodinger
equation

h̄2 ρ00 (r )
− + U (r )ρ(r ) = Eρ(r ) (B.62)
2
 161

with effective potential

h̄2
1 [h̄l − Aθ ]2 1
U (r ) = − 2 + 2
+ [ p z − A z ]2 + V (r )
8r 2 r 2

h i
2 2 1
h̄ l −
" ! #
1 4 h̄kλml k 2 λ2 m2 λ 2 2 4
k r
= − + + p2z + (k − λpz )kr2 + . (B.63)
2 r2 r 4 4

At low coupling (λ → 0), we have

 h i 
 h̄2 l 2 − 41 2
k 2 
−h̄2 ρ00 + + p2z + r ρ = Eρ, (B.64)
 r2 2 

and we see by dimensional analysis

[h̄k] = 1/L2 ⇒ E ∼ |h̄k|. (B.65)

These are weakly coupled massless excitations. But in the high coupling limit
(λ → ∞), we have
( )
k 2 λ2 2
−h̄2 ρ00 + (m + r4 ) ρ = Eρ, (B.66)
4

which yields a much more peculiar spectrum

[h̄2 kλ]2 = 1/L6 ⇒ E ∼ |h̄2 λk|2/3 . (B.67)

If this dispersion relation describes some fundamental constituents of the theory,

then they are certainly not particles in the traditional sense. We propose that this
 162

may be a glimpse of some post-relativistic constituents as mentioned in section

B.2.1.

B.4.1 quantization of the dual reduced system

In the dual picture (nonlinear sigma model), our ansatz picks out a class of solutions
that correspond to a different mechanical system. Though the equations of motion in
each picture can be mapped to one another via the duality, the correspondence is not
immediately obvious, and the systems will appear very different upon quantization.

After the ansatz, the duality relations (B.20) read

 
−1 0 1
g g = λe Kx
S + K e−Kx , g−1 ġ = λeKx Le−Kx . (B.68)
λ

Writing g = h(t, x )e−Kx yields

h−1 h0 = λS h−1 ḣ = λL. (B.69)

We further suppose that h is separable as h(t, x ) = F ( x ) Q(t). Then the equation for
S can be separated as

F −1 ( x ) F 0 ( x ) = λQ(t)S(t) Q−1 (t) (B.70)

Both sides are equal to some constant traceless matrix C. Since Q(t) is only unique
up to multiplication on the left by a constant matrix in SU(2), we can use this to
 163

choose C to be diagonal and thus proportional to K. We then have

Q(t)S(t) Q−1 (t) = sK, (B.71)

implying that TrS2 = s2 k2 . (B.70) is satisfied if

F ( x ) = eλsKx . (B.72)

Thus, the full corresponding ansatz for the field variable in the dual theory is

g(t, x ) = eλsKx Q(t)e−Kx , (B.73)

where Q is related to the previous variables by

1 −1
S = sQ−1 (t)KQ(t), L= Q Q̇. (B.74)
λ

The dual Lagrangian can now be written as

 2 2 
1 −1

−1
L2 = Tr Q Q̇ − λsQ KQ − K . (B.75)
2f2

It is useful to parameterize Q in terms of the Euler angles:

i i i
Q = e 2 σ3 γ e 2 σ1 β e 2 σ3 α . (B.76)

The traces in L2 can then be computed directly, yielding

( )
1 α̇2 + β̇2 + γ̇2
L2 = + cos βα̇γ̇ − V ( β) (B.77)
λ2 2
 164

where (dropping a constant shift)

V ( β) = −2k2 λs cos β. (B.78)

As a mechanical system, this is the well known spinning top (isotropic Lagrange
top). It is instructive to write

 
1 1
L2 = gij α̇i α̇ j − V (B.79)
λ2 2

where
 
 1 0 cos β 
 
 
gij = 
 0 1 0   (B.80)
 
 
cos β 0 1

is the metric of the rotation group and V is the gravitational potential of the top. The
overall constant 1
in the action leads to a rescaling of h̄ 7→ h̄λ2 upon quantization.
λ2

To pass to the quantum theory, we find the Laplacian operator with respect to
the metric g of Eulerian coordinates,

1 h√ i
2 ij
∇ ψ = √ ∂i gg ∂ j ψ (B.81)
g

The Hamiltonian is then

 2 2
h̄2 λ4 ∂α + ∂γ − 2 cos β∂α ∂γ

2
Ĥ2 = − + ∂ β + cot β∂ β ψ + Vψ (B.82)
2 sin2 β
 165

We can again reduce the Schrodinger equation Ĥ2 ψ = Eψ to a one-dimensional

Schrodinger equation with the ansatz

B( β)
ψ(α, γ, β) = eimα α eimγ γ p , (B.83)
sin β

yielding

00
2 4B ( β)
−h̄ λ + U ( β) B( β) = EB( β) (B.84)
2

where

h̄2 λ4 h̄2 λ4
 
2 2 1
U ( β) = − + 2
mα + mγ − 2mα mγ cos β − − 2k2 λs cos β. (B.85)
8 2 sin β 4

This can be studied by standard techniques for periodic potentials (Floquet

theory or Bloch waves etc.) We content ourselves with a quick look at low energy
excitations: small oscillations around the classical equilibrium q = 0 and setting
mα = 0 = mγ . Changing variables β = h̄λ2 q and expanding around the classical
minimum q = 0 gives

( )
1 d2 B 2

2 2 5
 1 2
− + q h̄ k sλ − − 2k s B ≈ EB (B.86)
2 dq2 8q2

The solutions involve Laguerre polynomials and the spectrum is, in this approxima-
√ √ 5
tion En ≈ 2(2n + 1)h̄k sλ 2 . If we remove the zero-point energy (n = 0), we have
√ 5
the energy of n free particles each of energy e1 = h̄k 8sλ 2 . This is the dispersion
relation of massless particles, except for a rescaling of the speed.
 166

B.5 Conclusions and outlook

Because they only exist in the short distance limit, it is difficult to say whether
objects like “preons” we discuss could correspond to directly observable objects in
an experiment. Quarks were not considered at first to be directly observable things
either, as they could not be created as isolated particles. In the S1 -model’s strong
coupling limit, the Minkowski geometry of space-time appears to be lost, and wave
propagation is sustained entirely by the non-linearity. However, these waves do
not appear to transmit information, and perhaps any “post-relativistic” effects are
hidden by some sort of confinement when they form bound states.

It is at least intriguing to question whether highly coupled theories have fun-

damental constituents whose nature is so exotic that they have been overlooked.
Drawing paralells with λφ4 theory, it is tempting to speculate that the Higgs particle
of the standard model is such a composite of some strongly bound preons existing
only at short distances. Were this the case, one could sensibly describe a “pure
Higgs” at short distances.

For a more complete understanding, we must quantize the whole theory rather
than just its mechanical reduction. Since the equations have a Lax pair, it should be
possible to perform a canonical transformation to angle-variables and then pass to
the quantum theory. Such a quantization was achieved for sine-Gordon theory [95],
proving that the solitons are fermions which bind to form the scalar waves. A
similar analysis of our model is a lengthy endeavor, and we hope to return to this
later after laying the groundwork and motivation here.

We present a side by side comparison of comparison of our work with the two
models in Table B.1 below.
 167

Nilpotent Field Theory (S1 ) Principal Chiral Model (S2 )

1 02
n o n o
L1 = Tr 2λ1 2
φ̇ − 2λ φ + 13 φ[φ̇, φ0 ] Lagrangian density L2 = 1 2 Tr ( g−1 ġ)2 − ( g−1 g0 )2
2λ

I= 1
λ φ̇, J = φ0 Currents I = g −1 g 0 , J = g−1 ġ

λI 0 = J̇ Current Identity İ − λ1 J 0 + λ [ J, I ] = 0

İ − λ1 J 0 + λ [ J, I ] = 0 Equation of Motion λI 0 = J̇

Reduced System of S1 Reduced System of S2

φ(t, x ) = eKx R(t)e−Kx + mKx Wave Ansatz g(t, x ) = eλsKx Q(t)e−Kx

n o
I = eKx Ṙe−Kx Wave Currents I = eKx λsQ−1 KQ − K e−Kx

J = eKx {λ[K, R] + mK } e−Kx J = eKx Q−1 Q̇ e−Kx



S = Ṙ + λ1 K Common Variables S = sQ−1 KQ

1 −1
L = [K, R] + mK L= λ Q Q̇

L̇ = [K, S] Current Identity Ṡ = λ[S, L]

Ṡ = λ[S, L] Eqn. of motion L̇ = [K, S]

  2    2 
1 1 1 2 1 1 1 2
H1 = Tr 2 S − λ K + 2 L Hamiltonian H1 = H2 H2 = Tr 2 S − λ K + 2 L
n  2 n o
L1 = Tr 21 S − λ1 K Lagrangian L1 6= L2 L2 = 21 Tr L2 − (S − K )2
o
+ 31 R [S − K, L] − 12 L2

E ∼ |k|2/3 (λ → ∞) short-range dispersion E ∼ |k| ( λ → 0)

Table B.1: A comparison of results in the dual models

 168

Appendix C

Measurement of the top quark

Yukawa coupling from tt̄ kinematic
distributions in the dilepton final
√
state at s = 13 TeV

Partially reprinted from publicly available preliminary result in Ref. [79] with permission.
Co-authored with the CMS collaboration

Since the discovery of the Higgs boson in 2012 [10, 11], one of the main goals
of the CERN LHC program has been to study in detail the properties of this new
particle. In the standard model (SM), all fermions acquire their mass through
interaction with the Higgs field. More specifically, the mass of fermions, mf , arises
√
from a Yukawa interaction with coupling strength gf = 2mf /v, where v is the
vacuum expectation value of the Higgs field. For the purpose of this measurement,
 169

we define for the top quark the parameter Yt = gt /gtSM , which is equivalent to the
modifier κt defined in the κ–framework [105].

Among all such couplings, the top quark Yukawa coupling is of particular
interest. It is not only the largest, but also remarkably close to unity. Given the
measured top quark mass mt = 172.4 ± 0.5 GeV [106], the mass-Yukawa relation
implies a value of the Yukawa coupling gtSM ≈ 0.99. However, physics beyond the
SM, such as two-Higgs-doublet and composite Higgs models, introduce modified
couplings that alter the interaction bestween the top quark and the Higgs field
[47, 48]. This makes the top–Higgs interaction one of the most interesting properties
of the Higgs field to study today at the LHC, especially because it is experimentally
accessible through multiple avenues, both direct and indirect.

Recent efforts have seen the first success in directly probing gt via tt̄H produc-
tion [49, 50], while the most precise determination comes from the κ −framework fit
in Ref. [51], which yields Yt = 0.98 ± 0.14 by combining information from several
Higgs boson decay channels. These measurements, however, fold in assumptions
of SM branching fractions via Higgs couplings to other particles. Another way
to constrain gt , which does not depend on the Higgs coupling to other particles,
was presented in the search for four top quark production in Ref. [78], yielding
a 95% confidence limit of Yt < 1.7. However, it is also possible to constrain gt
indirectly using the kinematic distributions of reconstructed tt̄ pair events, a tech-
nique which recently yielded a similar 95% confidence limit of Yt < 1.67 [76] in the
lepton+jets decay channel from a data set with an integrated luminosity 36 fb−1 .
The measurement presented in this note follows closely the latter approach.

Current simulations of tt̄ production include next-to-leading-order (NLO) preci-

sion in perturbative quantum chromodynamics (QCD). Subleading-order correc-
 170

Figure C.1: Sample diagrams for EW contributions to gluon-induced and quark-induced

top quark pair production, where Γ stands for neutral gauge bosons or the
Higgs boson. (repeated from page 87)

tions arise from including electroweak (EW) terms in a perturbative expansion of the
strong coupling αS and the EW coupling constant α. Such terms begin to noticeably
affect the cross section only at loop-induced order α2s α, and are typically ignored.
While these terms have a very small effect on the total cross section, they can alter
the shape of kinematic distributions to a measurable extent. Such changes will be
more noticeable if the Yukawa coupling affecting the loop correction (Fig. C.1) is
anomalously large. Therefore, these corrections are of particular interest in deriving
upper limits on gt . For example, the distribution of the invariant mass of the tt̄
system, Mtt̄ , will be affected significantly by varying Yt . In particular, doubling
the value of Yt can alter the Mtt̄ distribution by about 9% near the tt̄ production
threshold, as described in Ref. [52] and shown in Fig. C.2 (left). Another variable
sensitive to the value of Yt is the difference in rapidity between top quarks and
antiquarks, ∆ytt̄ = y(t) − y(t̄) , shown in Fig. C.2 (right). Together, Mtt̄ and ∆ytt̄
are proxies for the Mandelstam kinematic variables s and t respectively, and can be
used to include the EW corrections on arbitrary samples via weight corrections.

The effect of EW corrections can thus be evaluated without generating entirely

new Monte Carlo (MC) simulations. After deriving the dependence of these correc-
tions on the parameter of interest (POI) Yt , a measurement is performed. In order
for this analysis to be sensitive to deviations in Yt from unity arising from new
 171

CMS HATHOR Preliminary s = 13 TeV CMS HATHOR Preliminary s = 13 TeV

(dσHATHOR/dMtt)/ (dσLO/dMtt)

(dσHATHOR/d∆y )/ (dσLO/d∆y )
1.06

tt
1.25
Generator level Yt = 3 Generator level Yt = 3
Yt = 2 1.05 Yt = 2
1.2 Yt = 1 1.04 Yt = 1
Yt = 0 Yt = 0
1.15 1.03
1.02

tt
1.1
1.01
1.05
1
1 0.99

0.95 0.98
0.97
0.9
400 600 800 1000 1200 1400 1600 1800 −4 −3 −2 −1 0 1 2 3 4
Mtt [GeV] ∆y
Z Z tt

Figure C.2: Effect of the EW corrections on tt̄ differential kinematic distributions for differ-
ent values of Yt , after reweighting of simulated events. (repeated from page
88)

physics beyond the SM, we avoid correcting our simulated data for any potential
mismodeling with unclear physical origin, such as the discrepancy in top quark
pT distributions observed in Ref. [74]). We target events in the dilepton final state,
for which this type of measurement has not yet been performed. While this decay
channel has a smaller branching fraction than the lepton+jets channel studied in
Ref. [76], it has much lower backgrounds due to the distinguishing presence of two
final-state high-pT leptons. However, two neutrinos are also expected in this final
state, which escape detection and pose challenges in kinematic reconstruction. For
this reason, we do not perform a full kinematic reconstruction as was done in the
previous measurement in the lepton+jets channel. This measurement also utilizes
the full data set of 137 fb−1 collected during Run 2 at the LHC from 2016 to 2018,
allowing us to achieve comparable precision from a decay channel with a much
lower branching rate.

In this note, we will first discuss the data and MC samples (Section C.1), followed
by the methods for event selection (Section C.2) and reconstruction (Section C.3).
We then present an outline of the measurement technique (Section C.4) and the
 172

contributing uncertainty sources (Section C.5), and conclude with the results of the
measurement (Section C.6).

C.1 Simulation of top quark pair production and

backgrounds

The production of tt̄ events is simulated at the matrix-element (ME) level with
NLO QCD precision, using the POWHEG generator [54–57]. This calculation is
performed with the renormalization and factorization scales, µr and µf , set to the
q
transverse top quark mass, mT = m2t + pT 2 , where pT is the transverse momen-
tum of the top quark. The default value of mt is set to 172.5 GeV in all simulations.
The matrix-element calculations obtained from POWHEG are combined with par-
ton shower (PS) simulation from P YTHIA 8 [58–60], using the underlying-event
tune M2T4 [61] to simulate data taken in 2016, and tune CP5 [62] to simulate data
taken in 2017 and 2018 . The parton distribution functions (PDFs) set NNPDF3.0 at
NLO [63] is used in 2016 and updated to NNPDF3.1 [64] at next-to-NLO (NNLO)
in 2017 and 2018. These samples are normalized to a tt̄ cross section calculated
at NNLO in QCD including resummation of next-to-next-to-leading logarithmic
(NNLL) soft gluon terms with TOP++2.0 [39].

A high purity of tt̄ events can be obtained in the dilepton channel. For the
estimation of the remaining small fraction of backgrounds, we rely on simulations.
In particular, we account for dilepton production due to Drell–Yan type processes
and single top quark production. Other SM processes, in particular W boson and
diboson productions, were investigated and found to have negligible contributions.
 173

About 1% of the events identified as tt̄ dilepton decays are misidentified tt̄
lepton+jets decays. EW corrections are applied to all tt̄ events, even misidentified
ones, so their kinematic distributions remain dependent on Yt . Thus, these events
are still considered as signal even though the quality of reconstruction may generally
be poor.

Single top quark events are simulated at NLO with POWHEG in combination
with P YTHIA 8, while diboson events are simulated with P YTHIA 8 with leading-
order (LO) QCD precision. Drell–Yan production is simulated at LO using the
MADGRAPH generator [67] interfaced to P YTHIA 8 using the MLM matching algo-
rithm [68].

The detector response to all simulated events is modeled with the GEANT4
software toolkit [46]. In addition, the effects of multiple pp interactions per event
are included in simulations and the number of these pileup interactions agrees with
what is observed in data.

C.1.1 Simulation of electroweak corrections

Contributions to the top quark pair production arising from QCD+EW diagrams are
evaluated using the HATHOR (v2.1) package [53]. A double-differential cross section
is computed as a function of Mtt̄ and ∆ytt̄ at order α2S α. These diagrams involve
massive gauge boson exchange and are shown in Fig. C.1. The contributions from
photon-mediated interactions, which are comparatively small [107], are ignored.
Formally, these will be referred to as EW corrections, though certain electromagnetic
interactions are not included.
 174

The ratio of this double-differential cross section is evaluated with respect to

the LO QCD computation, in order to obtain a multiplicative weight correction
w( Mtt̄ , ∆ytt̄ ). Applying this weight at parton level to MC samples produced at
NLO QCD approximates the inclusion of EW corrections in the simulation. This
multiplicative approach to including EW corrections was used previously in Ref. [76],
and has the benefit of approximating the inclusion of diagrams at order O(α3S α). Be-
cause EW corrections factorize in some kinematic regimes, this is a better-motivated
approach than the alternative additive approach, in which one adds the fixed-order
result at order O(α2S α) while ignoring all potential contributions of order O(α3S α)
(see Ref. [65] for a detailed discussion). In other words, the additive approach ap-
plies a EW correction factor only to the proportion of POWHEG events present at
LO QCD, while any interplay between EW corrections and higher-order QCD sim-
ulation is ignored. Though the multiplicative approach is clearly favored, neither
approach can account for the effects of two-loop contributions near the tt̄ production
threshold. Accordingly, we take the difference between the two predictions as a
modeling uncertainty in this regime. The implementation of this uncertainty is
discussed further in Section C.4.

The EW correction weights are calculated for discrete integer values of Yt =

0, 1, . . . , 5. Since the dependence of the production rate on Yt is exactly quadratic,
these discrete values are sufficient to parametrize event yields as a continuous
function of Yt (see Section C.4). This allows us to measure which value of Yt best
describes the data.
 175

C.2 Event and object selection

Events are selected using single-electron or muon-triggers. Data taking at CMS is

punctuated by technical stops at the end of each year, leading to minor changes
in configuration and modeling between 2016, 2017, and 2018. For events selected
by the single-electron trigger, we require a trigger pT threshold of 27 GeV with the
exception of 2018, where a threshold of 32 GeV is used. In the case of the single-
muon trigger, we select events with a trigger pT threshold of 24 GeV, which is raised
to 27 GeV in 2017 due to high event rates.

We ensure that all electrons and muons are within the silicon tracker coverage by
requiring a pseudorapidity | η |< 2.4. To operate well above the trigger threshold
we then require at least one isolated electron or muon reconstructed with pT >
30 GeV, except in 2018, where we require leading pT electrons to have pT > 34 GeV
in accordance with the trigger availability. After selecting the leading-pT lepton, a
second isolated electron or muon with pT > 20 GeV is required. Events with three
or more isolated leptons with pT > 15 GeV are discarded.

Jets are clustered from PF objects through the anti-kT algorithm [69, 70] using
a distance parameter of 0.4. The jet momentum is calculated as the vectorial sum
of the momenta of its constituents. Corrections to the jet energy are derived as a
function of jet pT and η in simulation and improved by measurements of energy
balance in data [71]. We select jets with | η |< 2.4 and pT > 30 GeV. Jets originating
from quarks are identified using the DeepCSV algorithm [41]. We require events to
have at least two jet candidates passing the algorithm’s loose working point, using
a tagging criterion with an efficiency of 84%. This is the most efficient and least
stringent working point of the DeepCSV algorithm. Events with more than two jets
 176

Source 2016 (X fb−1 ) 2017 (Y fb−1 ) 2018 (Z fb−1 ) All (X+Y+Z fb−1 ) % total MC
tt̄ MC 140 831±134 170 554±95 259 620±146 571 005±220 96.1%
V + jets MC 1944±54 3178±75 4958±133 10 080±162 1.7%
Single t MC 3022±25 3521±24 5827±32 12 370±47 2.1%
Diboson MC 142±9 151±10 250±15 543±20 0.1%
Total MC 145 940±148 177 404±124 270 655±201 593 999±279 100%
Data 144 817 178 088 264 791 587 696 98.9%

Table C.1: MC and data event yields for all three years and combined. Statistical uncertainty
on simulated event counts is given. For an estimate of systematic uncertainty,
see Figs. 8.5-8.7. (repeated from page 96)

meeting this criterion are considered only if exactly two of the jets pass one of the
more stringent medium or tight working points supplied by the algorithm, which
have efficiencies of 68% and 50%, respectively. In this case, the two jets passing the
most stringent identifier are assumed to be the two b jets originating from top quark
decays. We find that in tt̄ events, the two jets are among the candidates passing the
looser -tagging requirement in roughly 93% of the events.

In order to remove Drell–Yan background events in the ee and µµ channels,

we reject events in which the two leptons together have an invariant mass below
50 GeV or within 10 GeV around the Z boson mass of 91.2 GeV.

The negative vector sum of all transverse momenta, ~/

E T , is generally of large
magnitude in dilepton decays because of the two undetected final-state neutrinos.
To further aid in removing Drell–Yan type decays, we impose an additional selection
requirement on the magnitude of the missing transverse momentum vector of
pmiss
T > 30 GeV in all events with ee or µµ in the final state.

The breakdown of expected signal and background yields is shown by year in

Table C.1. The Drell–Yan background is estimated to be about 2%. Single top quark
production accounts for roughly another 2% of the estimated sample composition.
 177

C.3 Event reconstruction

The EW corrections are calculated based on Mtt̄ and ∆ytt̄ . However, to evaluate
these quantities it is necessary to reconstruct the full kinematic properties of the
tt̄ system, including the two undetected neutrinos. While it is possible to com-
pletely reconstruct the neutrino momenta in the on-shell approximation, such a
reconstruction is highly sensitive to pmiss
T , which introduces large resolution effects

and additional systematic uncertainties. We found that using the proxy variables
¯ and |∆y| = |y( + `)
Mb` = M ( + ¯ + ` + `) ¯ − y(¯ + `)|, where ` represents a final
b`

state electron or muon, results in a more sensitive measurement.

Unlike Mb` , the accurate reconstruction of |∆y|b` requires that each of the two
jets is matched to the correct lepton, i.e., both originating from the same top quark
decay. In order to make this pairing, we utilize the information from the kinematic
constraints governing the neutrino momenta.

If one assumes the top quarks and W bosons to be on-shell, the neutrino mo-
menta are constrained by a set of quadratic equations arising from the conservation
of four-momentum at each vertex. We refer to these kinematic equations, collec-
tively, as the mass constraint. The mass constraint for each top quark decay results
in a continuum of possible solutions for neutrino momenta, which geometrically
can be presented as an intersection of ellipsoids in three-dimensional momentum-
space [108]. For certain values of input momenta of jets and leptons these ellipsoids
do not intersect at all, such that the quadratic equations have no real solution. In
these scenarios the mass constraint cannot be satisfied.

In cases where the mass constraint can be satisfied, one may also constrain pmiss
T

in the event to equal the sum of the pT of the two undetected neutrinos. We call this
 178

the pmiss
T constraint. This constraint reduces the remaining solutions to a discrete
set, containing either 2 or 4 possibilities that fully specify the momenta of both
neutrinos. Similar to the case of the mass constraint, there are some values of the
input parameters, for which the pmiss
T constraint cannot be satisfied exactly.

Assuming the jets are correctly reconstructed and paired, the mass constraint
can be satisfied in 96% of all cases, while the mass and pmiss
T constraints can both
be satisfied in 55% of cases. In contrast, if the jets are correctly reconstructed but
incorrectly paired to leptons, the mass constraint can be satisfied in only 23% of
cases, while both mass and pmiss
T constraints can be satisfied in only 18% of cases.

Pairings with no solution to the mass constraint are thus frequently incorrect.
When the mass constraint can be satisfied, pairings with a solution to the pmiss
T

constraint are more likely to be correct. This information is used as part of the
pairing procedure, which proceeds in three steps.

1. The mass constraint is checked for both possible pairings. If only one pairing
is found to satisfy the mass constraint, that pairing is used. If both pairings fail
to satisfy the mass constraint, the event is discarded. If both pairings satisfy
the mass constraint, we check the pmiss
T constraint.

2. If only one pairing allows for the pmiss

T constraint while the other does not, the
pairing yielding an exact solution to the pmiss
T constraint is used.

3. If the neutrino kinematics do not suggest a clear pairing, the -jets, 1 and 2 , are
paired with the leptons (`, `¯ ) by minimizing the quantity

¯
Σ1(2) = ∆R(1(2) , `) + ∆R(2(1) , `)
 179

among the two possible pairings, where ∆R(, `)2 = (ηb − η` )2 + ( ϕb − ϕ` )2 .

After these steps, we obtain the correct jet pairing in simulation in 82% of the events
for which both jets originating from top quark decays are correctly identified.

The sensitivity of our chosen kinematic variables to Yt , before and after recon-
struction, is shown in Fig. C.3. We see that, in the chosen proxy variables, not
much sensitivity is lost in the reconstruction process. This is especially true for the
proxy mass observable, Mb` , providing an advantage over Mtt̄ , which cannot be
reconstructed as accurately.

1.14
CMS HATHOR Preliminary s = 13 TeV
1.07
CMS HATHOR Preliminary s = 13 TeV
(dσY /dMbl )/ (dσY =1/dMbl )

(dσY /d∆y )/ (dσY =1/d∆y )

bl
Generator level Yt = 3 Generator level Yt = 3
1.12 1.06
Yt = 2 Yt = 2
1.1 Yt = 1 1.05 Yt = 1
t

1.08 Yt = 0 Yt = 0
1.04
1.06
1.03
bl

1.04
1.02
1.02
t

1.01
1
0.98 1

0.96 0.99
200 400 600 800 1000 1200 1400 1600 1800 −4 −3 −2 −1 0 1 2 3 4

Mbl [GeV] ∆y
Z Z bl

1.14
CMS HATHOR Preliminary s = 13 TeV
1.07
CMS HATHOR Preliminary s = 13 TeV
(dσY /dMbl )/ (dσY =1/dMbl )

(dσY /d∆y )/ (dσY =1/d∆y )

bl

Reconstructed Yt = 3 Reconstructed Yt = 3
1.12 1.06
Yt = 2 Yt = 2
1.1 Yt = 1 1.05 Yt = 1
t

1.08 Yt = 0 Yt = 0
1.04
1.06
1.03
bl

1.04
1.02
1.02
t

1.01
1
0.98 1

0.96 0.99
200 400 600 800 1000 1200 1400 1600 1800 −4 −3 −2 −1 0 1 2 3 4

Mbl [GeV] ∆y
Z Z bl

Figure C.3: The ratio of reconstructed kinematic distributions with EW corrections (eval-
uated for various values of Yt ) to the SM kinematic distribution is shown,
demonstrating the sensitivity of these distributions to the Yukawa coupling.
The upper figures show the information at generator level, while the lower
figures are obtained from reconstructed events. (repeated from page 99)
 180

C.3.1 Comparison between data and simulation

Comparisons between data and simulation is shown in Fig. C.4, where the agree-
ment is generally seen to be within the systematic uncertainty. Some slope is

×10
3 137 fb-1 (13 TeV) ×10
3 137 fb-1 (13 TeV)
140
Events / Bin

450

Events / 15 GeV
CMS Data CMS Data
400 Preliminary tt correct reco 120 Preliminary tt correct reco
350 tt b jets swapped tt b jets swapped
tt wrong reco 100 tt wrong reco
300 tt τ tt τ
250 Single t 80 Single t
200 Drell-Yan Drell-Yan
60
Total unc. Total unc.
150
40
100
20
50

1.4 1.4
Pred.

Pred.
Data

Data
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0 1 2 3 4 5 6 7 8 9 0 50 100 150 200 250 300
Njets pmiss [GeV]
Njets TMET (P )
T
×10
3 137 fb-1 (13 TeV) ×10
3 -1
137 fb (13 TeV)
Events / 10 GeV

Events / 10 GeV

120
CMS Data CMS Data
250
Preliminary tt correct reco Preliminary tt correct reco
100 tt b jets swapped tt b jets swapped
tt wrong reco 200 tt wrong reco
80 tt τ tt τ
Single t 150 Single t
60 Drell-Yan Drell-Yan
Total unc. 100 Total unc.
40

20 50

1.4 1.4
Pred.

Pred.
Data

Data

1.2 1.2
1 1
0.8 0.8
0.6 0.6
0 50 100 150 200 250 0 50 100 150 200 250 300
leading lepton p [GeV] b jet p [GeV]
LeadingTLep Pt Tb-jet pt

Figure C.4: Data to MC simulation comparison is shown for jet multiplicity, pmiss T , lepton
pT , and b jet pT . Uncertainty bands are derived by varying each known uncer-
tainty source up and down across the full data-taking run, and summing the
effects in quadrature. Here the signal simulation is divided into the following
categories: correctly reconstructed events (tt̄ correct reco), events with correctly
reconstructed leptons and b jets but incorrect pairing of b jets and leptons (tt̄ b
jets swapped), events with incorrectly reconstructed b jets or leptons including
non-dilepton tt̄ decays (tt̄ wrong reco), and a separate category for incorrectly
reconstructed dilepton events which have a τ lepton in the final state (tt̄ τ).

apparent in the ratio of data to MC prediction in the pT distributions of lepton and

jet. These trends may be related to a previously observed feature of the nominal
POWHEG +P YTHIA 8, in which a harder top quark pT distribution is observed
 181

than in data (see, for example, Ref. [74]). Fixed-order NNLO calculations are avail-
able that generally show a softer top quark pT spectrum than in the POWHEG
+P YTHIA 8 simulation, but these are typically evaluated with a different dynami-
cal scale choice, as described in Ref. [109]. The dynamical scale mT at which ME
calculations are evaluated in POWHEG is generally preferred for modeling Mtt̄ ,
which is more closely related to our measurement sensitivity. The potential effects
of NNLO calculations and the top quark pT spectrum on this measurement are
discussed further in Section C.5.

C.4 Measurement strategy and statistical methods

After reconstruction, events are binned coarsely in |∆y|b` and more finely in Mbl .
The binning is chosen to ensure no fewer than ≈10 000 events in each bin and in
each data-taking year, as seen in Fig. C.5, to ensure a low statistical uncertainty and
improved uncertainty estimation. In each bin, the expected yield is parametrized
as a function of Yt . The effect is exactly quadratic, as a consequence of the order
at which EW corrections are evaluated. Therefore we perform a quadratic fit to
extrapolate the effect of the EW corrections on a given bin as a continuous function
of Yt (Fig. C.6). This correction for each bin can be applied as a rate parameter
Rbin
EW (Yt ) = 1 + δEW (Yt ) affecting the expected bin content.

We construct a likelihood function L, which is defined as a product of a Poisson

probability describing statistical fluctuations in each bin and penalty terms encoding
systematic uncertainty sources as nuisance parameters:
 182

137 fb-1 (13 TeV)

Events / bin
30000
CMS Data tt Single t Drell-Yan Fit unc.

25000 Preliminary

20000

15000

10000

|∆y | < 1
|∆y | > 1
|∆y | < 1
|∆y | > 1
5000

2016

2017

2018
|∆y | < 1
|∆y | > 1
Pred.
Data

1.1
1
0.9
440-3000

560-3000

440-3000

560-3000

440-3000

560-3000
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440
100-280
280-320
320-360
360-400
400-460
460-560
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440
100-280
280-320
320-360
360-400
400-460
460-560
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440
100-280
280-320
320-360
360-400
400-460
460-560
Mbl range [GeV]

Figure C.5: Binned data and simulated events prior to performing the fit. The solid lines
divide the three datataking periods, while the dashed lines divide the two
|∆y|b` bins in each datataking period, with Mb` bin ranges displayed on the x
axis. (repeated from page 103)

CMS Simulation Preliminary CMS Simulation Preliminary

weak correction rate

weak correction rate

1.35 |∆ y | 0-1.0 0.96

bl
1.3 Mbl 100-210 GeV
0.94
1.25
0.92
1.2
0.9
1.15
0.88
1.1
0.86

1.05 |∆ y | 0-1.0
0.84 bl
1 Mbl 440-3000 GeV
0.82

0 1 2 3 4 5 0 1 2 3 4 5

Yt Yt

Figure C.6: The EW correction rate modifier Rbin EW in two separate [Mb` , ∆yb` ] bins from
2017 data, demonstrating the quadratic dependence on Yt . All bins will have an
increasing or decreasing quadratic yield function, with the steepest dependence
on Yt found at lower values of Mb` . (repeated from page 106)
 183

L= ∏ Lbin × p(φ) × ∏ p(θi ), (C.1)

bin∈( Mb` ,|∆y|bl ) i

h i
Lbin = Poisson nbin
obs s bin
({θi }) × Rbin
EW (Yt , φ ) + b
bin
({θi })} . (C.2)

Here nbin
obs is the total observed bin count, with the expected bin count being the

sum of the predicted signal yield sbin and background yield bbin . The number of
expected signal events is modified by the additional rate parameter REW , which
depends on the Yukawa coupling ratio Yt and the special nuisance parameter φ,
discussed further below. Each of the other nuisance parameters, {θi }, that affect
the expected counts in each bin, is described by a probability distribution function
of p(θi ) that describes the likelihood of the deviation of θi from its expected value.
We include an estimate of the uncertainty in the multiplicative application of EW
corrections derived at order O(α2S α). The full expression for the rate Rbin
EW , including

this uncertainty term in the bins near the tt̄ production threshold, is given by

Rbin bin bin bin φ

EW (Yt , φ ) = (1 + δEW (Yt )) × (1 + δQCD δEW (Yt )) , (C.3)

where we have defined

bin nbin bin

HATHOR − n LO QCD bin nbin bin
POWHEG − n LO QCD
δEW = , δQCD = . (C.4)
nbin
LO QCD nbin
POWHEG

Intuitively, δEW represents the marginal effect of EW corrections included in

HATHOR relative to the LO QCD calculation, while δQCD represents the marginal
 184

effect of higher-order terms included in the POWHEG sample relative to the LO

QCD calculation. The multiplicative approach to including EW corrections assumes
bin bin
that these two corrections factorize. The quantity δQCD δEW represents the cross
term arising from the difference in multiplicative and additive approaches. The
normally distributed nuisance parameter φ modulates the uncertainty generated
by this cross term. We note that the uncertainty in the EW corrections is unique
because it depends on the value of Yt at which the EW corrections are evaluated.
Thus, it is described by its own term and nuisance parameter φ separate from other
systematic uncertainties. For bins away from the threshold where EW corrections
decrease as a function of Yt , we do not include this uncertainty. These bins do
not contribute much sensitivity to the measurement, and also enter a kinematic
regime in which this method of uncertainty estimation is no longer meaningful. At
the large values of the Mandelstam variable s that correspond to these bins, the
dominant terms contributing to δEW are Sudakov logarithms resulting from W and
Z exchange. These terms factorize well and do not contribute to the uncertainty we
wish to model [65].

Nuisance parameters corresponding to uncertainties that have an overall normal-

ization effect, that have an overall normalization, such as uncertainty on luminosity
or cross section values, follow a log-normal distribution. Uncertainties with shape
effects associated to nuisance parameters {θi } are handled by generating up and
down variations of the bin content sbin for each θi . These variations result from
changing the underlying theoretical/experimental sources, which are outlined in
Section C.5, usually by an amount of one standard deviation (σ) based on the uncer-
tainty in our best estimates. Since θi is normally distributed, these shapes are then
enforced as the bin modifiers associated with θi = ±1, while θi = 0 corresponds to
 185

the nominal estimate where the simulated data is reproduced. The collection of bin
modifiers for these up and down variations are referred to as shape templates, with
examples shown in Section C.6. A vertical template morphing is applied to alter
the shape as a function of the underlying nuisance parameter θi , where in each bin
the modifier is interpolated as a cubic spline for values of θi ∈ [−1, 1] and linearly
outside of that region, assuring that sbin (θi ) remains continuous and differentiable.

The measurement of Yt is then performed by a maximum likelihood fit. By

repeating this maximization over a fine array of fixed values of Yt and comparing
to the absolute maximum, we evaluate confidence intervals (CI) at the 68% and 95%
level around the best-fit value.

C.5 Sources of experimental and theoretical

uncertainty

The list of uncertainties considered is very similar to that of the previous measure-
ment presented in Ref. [76]. The main differences are the lack of QCD multijet
background and the use of data from the full Run 2 data-taking period. Full or
partial correlations are imposed on the underlying uncertainty sources between
data-taking periods where appropriate, as discussed further in Section C.5.1. Un-
certainties which strictly do not alter the shape of final distribution are treated as
normalization uncertainties, while all others are treated as shape uncertainties on
the binned data. Shape effects are considered for the distributions of tt̄ events only,
as the contribution of background events is small. Correlations of uncertainties
between different data-taking periods are treated on a case-by-case basis. Because
 186

the measurement is more sensitive to shape effects than normalization effects, the
dominant uncertainties are not necessarily those with the largest magnitude of
effect.

The dominant experimental uncertainty in this analysis comes from the cali-
bration of the detector’s jet energy response. Corrections to the reconstructed jet
energies are applied as a function of pT and η. We follow the standard approach
outlined in Ref. [71] to consider 26 separate uncertainties which are involved in
determining these calibrations. In this approach, uncertainty in the resolution of the
jet reconstruction is also considered in addition to the energy response. The effect
of these uncertainties is propagated to the reconstruction of pmiss
T .

Other experimental sources of uncertainty are comparatively minor. The overall

uncertainty in the integrated luminosity of 2.5%, 2.3%, and 2.5% is included as a
normalization uncertainty applied to all signal and background events in 2016, 2017,
and 2018 respectively [110–112]. The uncertainty in the number of pileup events
included in simulation is assessed by varying the inelastic cross section, 69.2 mb, by
4.6% [75].

Efficiencies in jet identification and misidentification are corrected to match

data [41]. The uncertainty in this correction manifests approximately as an overall
normalization effect of around 3%.

Similarly, scale factors are applied in bins of pT and η to correct simulated

efficiencies of lepton reconstruction, identification, isolation, and triggers to match
data. These are derived from a fit using the tag-and-probe method on Z boson
decays [72, 73]. This fit accounts for the uncertainty from the limited number of
 187

events in the data sample as well as differences in performance based on the jet
multiplicity. Overall, the effect is assessed to be below 2%.

As a standard technique to estimate the contributions of higher-order QCD

terms at the ME level, the QCD renormalization scale µr and factorization scale
µf are each varied separately up and down in the POWHEG simulation by a
factor of 2. The same-sign variation of the two scales together is included as an
additional uncertainty. Since an NNLO tt̄ cross section is already used to improve
the normalization of the MC simulation, the normalization effect induced by the
scale variations is overestimated. As our measurement is predominantly sensitive
to shape effects, this overall normalization effect is therefore removed in generating
the corresponding shape templates.

Because a higher-order cross section calculation is used, the uncertainty on sam-

ple normalization coming from scale variations is fictitiously large. In generating
the corresponding shape templates, this overall normalization effect is removed, as
the NNLO tt̄ cross section is already used to improve MC normalization, and we
are predominantly sensitive to shape effects. The shape effects remain significant
and these are among the dominant uncertainties in the fit.

A 5% normalization uncertainty is assumed in the tt̄ normalization, which covers

expected contributions from higher-order terms not included in the NNLO+NNLL
cross section calculation [39]. The backgrounds in this analysis are small enough
(≈2% sample composition each) that we do not generate shape templates for their
response to individual systematic uncertainties. A 15% normalization uncertainty
is included on single top quark MC samples, which was seen to cover expected ME
scale variation and jet energy correction uncertainties associated with these samples.
Drell–Yan and diboson MC samples are assigned a 30% normalization uncertainty,
 188

to cover the larger ME scale variation uncertainties associated with these LO simu-
lations. The background normalizations can alter slightly the expected shape of the
data, but are not among the dominant uncertainties.

We include an uncertainty in the EW corrections, based on our methods for

generating and applying these additional terms, as outlined in Sections C.1 and
C.4. Like the scale variations, this uncertainty is designed to cover higher-order
effects at the ME level, specifically those arising from diagrams of order α3s α. This is
a dominant uncertainty in the fit, and in particular it reduces sensitivity to values
of Yt < 1.

As mentioned in Section C.3.1, fixed-order calculations are available for top

quark pair production at NNLO, and are generally in better agreement with the
top quark pT spectrum observed in data. Some measurements have included
additional uncertainties or reweighting in attempting to account for this effect. To
assess the potential effects of NNLO calculations on this measurement, differential
cross sections in top quark pT as well as Mtt̄ were studied using publicly available
FASTNLO tables [113, 114]. Multidifferential cross sections at NNLO [115] in Mtt̄
and ∆ytt̄ were also studied, as these are the same variables through which EW
corrections are applied in this measurement. The NNLO calculation in Ref. [113]
does not show a significantly softer top quark pT spectrum than the corresponding
NLO calculation evaluated with the same choice of dynamical scale. This suggests
that the differences in the top quark pT spectrum between the POWHEG +P YTHIA 8
and the calculation in Ref. [113] arise partly from the choice of dynamical scale
and not only from higher-order effects. Meanwhile, cross section calculations as a
function of Mtt̄ appear consistent with existing uncertainties. For this reason, no
additional uncertainties are included with the explicit attempt to reflect knowledge
 189

of higher-order simulations, as this would begin to double-count the existing theory

uncertainties.

The uncertainty in modeling the initial- and final-state radiation in the parton
shower algorithm is assessed by varying the value of the renormalization scales in
the initial- and final-state radiation by a factor of two. These are among the domi-
nant modeling uncertainties in the measurement. Uncertainty for other parameters
in the parton shower description are considered separately. The hdamp parame-
ter, which controls the matrix-element parton shower matching in POWHEG +
P YTHIA 8, is set to the nominal value of hdamp = 1.58 mt (1.39 mt ) in 2016 (2017–
2018). Dedicated MC samples are generated with this parameter varied down to
1 mt (0.874 mt ) and up to 2.24 mt (2.305 mt ) in 2016 (2017-18), in order to estimate
the effect of this uncertainty. Dedicated MC samples are also generated with varia-
tions of the P YTHIA 8 underlying-event tune, a set of parameters which impact the
hadronization. The uncertainty in hdamp and the underlying event tune are minor
compared to the parton shower scale variations.

Dedicated MC samples are generated with the top quark mass varied up and
down by 1 GeV from the nominal value m = 172.5 GeV to estimate the effect of
the top quark mass uncertainty. This is a significant uncertainty, but not among the
most dominant. It should be noted that, although the mass and Yukawa coupling
are generally treated as independent in this measurement, varying the mass will
slightly modify the definition of Yt = 1. However, this effect, which is below 1%, is
much smaller than the sensitivity of the measurement and can therefore be ignored.

The NNPDF sets [63] contain 100 individual variations as uncertainties. Following
the approach in Ref. [76], similar variations are combined to reduce the number
of variations to a more manageable set of 10 shape templates. The variation of
 190

the strong coupling αs used by NNPDF is treated separately from the other PDF
variations. The effect of uncertainties in the PDF set is typically smaller than 1%,
and has a minor impact on the fit.

The branching fraction of semileptonic B decays affects the jet response. The
effect of varying this quantity within its measured uncertainty [36] is included as
an uncertainty, which has a small effect relative to other modeling uncertainties.

The momentum transfer from quarks to B hadrons is modeled with a transfer

pT ( B )
function dependent on xb = pT (−jet)
. To estimate the uncertainty, the transfer
function is varied up and down within uncertainty of the Bowler–Lund parameter
[77] in P YTHIA 8. The resulting effect is included by modifying event weights to
reproduce the appropriate transfer function. This has a noticeable shape effect of
the order 4%, but was not found to be a dominant uncertainty in the fit.

C.5.1 Processing of systematics

In this analysis, the effect of the POI Yt manifests itself as a smooth shape distortion
of the kinematic distributions as shown in Fig. C.7. Though the nuisance parameters
describing the sources of uncertainty should induce smooth shape effects as well,
their effects are sometimes obscured by statistical noise or imprecise methods
of estimation. This is noticeable for the uncertainties associated with jet energy
scale, jet energy resolution, parton shower modeling, pileup reweighting, and top
quark mass. For these shape templates only, we apply a one-iteration LOWESS
algorithm [116] to smooth the templates and remove fluctuations that may disturb
the fit. The underlying-event tune and hdamp uncertainties in the parton showering
 191

137 fb-1 (13 TeV)

Events / bin

50000 CMS EW corrections tt (Y = 1)

Simulation t
Preliminary tt (Y = 2)
t
40000
tt (Y = 0)
t

30000

20000
|∆y| < 1.0

|∆y| > 1.0

10000
440-3000

560-3000
100-210

210-230

230-250

250-270

270-290

290-310

310-340

340-380

380-440

100-280

280-320

320-360

360-400

400-460

460-560
Relative effect (tt)

0.05

−0.05
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440

100-280
280-320
320-360
360-400
400-460
460-560
440-3000

560-3000

Mbl range [GeV]

Figure C.7: The effect of the Yukawa parameter Yt on reconstructed events in the final
binning. The POI induces a shape distortion on the kinematic distributions.
(repeated from page 105)
 192

are small enough for their shapes disappear into statistical noise, and are therefore
treated only as normalization uncertainties.

Most shape templates are also symmetrized, by taking the larger magnitude of
the up and down variations in each bin and negating this value to replace the effect
of the opposite variation. This step helps ensure a stable minimum in the likelihood
fit, but is skipped for templates whose natural shape effect is notably asymmetric.
In the few cases where this may be an overly conservative approach, it nonetheless
helps to ensure the performance and reliability of the minimization procedure, and
has little effect on the final result.

Full or partial correlations between the 2016, 2017, and 2018 data samples are
assumed for many uncertainties. In general, all theoretically motivated uncertain-
ties are considered fully correlated between years. Exceptions are made in cases
where modeling methods were altered between years, in particular for initial-state
radiation, final-state radiation, and PDF uncertainties. The modeling of these uncer-
tainties differs in the 2016 simulation, so the associated nuisance parameter in this
year is either partially or fully decorrelated from the other years. Additionally, un-
certainties whose effects disappear into statistical noise due to limited MC sample
size (underlying-event tune, hdamp ) are converted to uncorrelated normalization
uncertainties.

Some experimental uncertainties can be broken into components which are either
fully correlated or uncorrelated between years (large jet energy scale contributions,
luminosity). The uncertainty in the number of pileup events is considered fully
correlated as it is evaluated by varying the total inelastic cross section. For minor
uncertainties from jet and lepton scale factors, which have both correlated and
 193

statistical components, a 50% correlation is assumed between years. Lastly, the jet
energy resolution uncertainties are treated as uncorrelated between years.

C.6 Results

We obtain a best fit value of Yt = 1.16, with a 68% CI of [0.81, 1.40], and a 95%
CI of [0, 1.54]. The scan of the negative log likelihood as a function of Yt , used to
build these intervals, is shown in Fig. C.8. We also show the agreement of data
and simulation after performing the fit in Fig. C.9. The maximum of the likelihood
occurs at a configuration with good agreement between data and simulation. The
shape templates for the four uncertainties with the greatest effect on the fit are
shown in Fig. C.10.

This result is in agreement with the previously obtained measurement in the

lepton+jets final state in Ref. [76], while obtaining a slight increase in sensitivity.
Using a different decay channel and a larger data set provides a measurement
complementary to the previous result.

C.7 Summary

A measurement of the Higgs Yukawa coupling to the top quark is presented, based
on data from proton-proton collisions at the CMS experiment. Data at a center-of-
mass energy 13 TeV is analyzed from the full LHC Run 2, collected from 2016–2018,
corresponding to an integrated luminosity of 137 fb−1 . The result is a best fit
value of Yt = 1.16+ 0.24
−0.35 and a 95% confidence interval (CI) of Yt ∈ [0, 1.54]. This
 194

CMS 137 fb-1 (13 TeV)

6

-2 ∆ ln L
Data

5 Asimov Y t = 1
Asimov Y t = 1.16
4

1 68% CL

0
0 0.5 1 1.5 2 2.5 3
Yt

Figure C.8: The result of a likelihood scan, performed by fixing the value of Yt at values
over the interval [0, 3]. Expected curves from fits on simulated data are shown
produced at the SM value Yt = 1.0 (dashed) and at the final best-fit value of
Yt = 1.16 (dotted). (repeated from page 115)

measurement uses the effects of virtual Higgs boson exchange on tt̄ kinematic
properties to extract information about the coupling from kinematic distributions.
Although the sensitivity is lower compared to constraints obtained from studying
processes involving Higgs boson production in Refs. [49] and [51], this measurement
avoids dependence on other Yukawa coupling values through additional branching
assumptions, making it a compelling orthogonal measurement. This measurement
also achieves a slightly higher precision than the only other Yt measurement that
does not make additional branching fraction assumptions, performed in the search
for production of four top quarks. The four top quark search places a 95% CI of Yt
< 1.7 [78] while this measurement achieves a 95% CI of Yt < 1.54.
 195

137 fb-1 (13 TeV)

Events / bin

30000
CMS Data tt Single t Drell-Yan Fit unc.

25000 Preliminary

20000

15000

10000

|∆y | < 1
|∆y | > 1
|∆y | < 1
|∆y | > 1

5000
2016

2017

2018
|∆y | < 1
|∆y | > 1
Pred.

1.05
Data

1
0.95
440-3000

560-3000

440-3000

560-3000

440-3000

560-3000
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440
100-280
280-320
320-360
360-400
400-460
460-560
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440
100-280
280-320
320-360
360-400
400-460
460-560
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440
100-280
280-320
320-360
360-400
400-460
460-560
Mbl range [GeV]

Figure C.9: The agreement between data and MC simulation at the best fit value of Yt = 1.16
after performing the likelihood maximization, with shaded bands displaying
the post-fit uncertainty. The solid lines divide the three datataking periods,
while the dashed lines divide the two |∆y|b` bins in each datataking period,
with Mb` bin ranges displayed on the x axis. (repeated from page 115)
 196

Events / bin 137 fb-1 (13 TeV) 137 fb-1 (13 TeV)

Events / bin
CMS final state radiation tt (central) CMS jet energy corrections tt (central)
Simulation Simulation
50000 50000
Preliminary tt (θi +1σ) Preliminary tt (θi +1σ)
tt (θi -1σ) tt (θi -1σ)
40000 40000

30000 30000

20000 20000

|∆y| < 1.0

|∆y| > 1.0

|∆y| < 1.0

|∆y| > 1.0

10000 10000

0.04
440-3000

560-3000

440-3000

560-3000
100-210

210-230

230-250

250-270

270-290

290-310

310-340

340-380

380-440

100-280

280-320

320-360

360-400

400-460

460-560

100-210

210-230

230-250

250-270

270-290

290-310

310-340

340-380

380-440

100-280

280-320

320-360

360-400

400-460

460-560
Relative uncertainty (tt)

Relative uncertainty (tt)

0.04

0.02 0.02

0 0

−0.02 −0.02

−0.04 −0.04
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440

100-280
280-320
320-360
360-400
400-460
460-560

100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440

100-280
280-320
320-360
360-400
400-460
460-560
440-3000

560-3000

440-3000

560-3000
Mbl range [GeV] Mbl range [GeV]

137 fb-1 (13 TeV) 137 fb-1 (13 TeV)

Events / bin

Events / bin

CMS factorization scale tt (central) CMS renormalization scale tt (central)

Simulation Simulation
50000 50000
Preliminary tt (θi +1σ) Preliminary tt (θi +1σ)
tt (θi -1σ) tt (θi -1σ)
40000 40000

30000 30000

20000 20000
|∆y| < 1.0

|∆y| > 1.0

|∆y| < 1.0

|∆y| > 1.0

10000 10000
440-3000

560-3000

440-3000

560-3000
100-210

210-230

230-250

250-270

270-290

290-310

310-340

340-380

380-440

100-280

280-320

320-360

360-400

400-460

460-560

100-210

210-230

230-250

250-270

270-290

290-310

310-340

340-380

380-440

100-280

280-320

320-360

360-400

400-460

460-560
0.04
Relative uncertainty (tt)

Relative uncertainty (tt)

0.02

0.01 0.02

0 0

−0.01 −0.02

−0.02 −0.04
100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440

100-280
280-320
320-360
360-400
400-460
460-560

100-210
210-230
230-250
250-270
270-290
290-310
310-340
340-380
380-440

100-280
280-320
320-360
360-400
400-460
460-560
440-3000

560-3000

440-3000

560-3000

Mbl range [GeV] Mbl range [GeV]

Figure C.10: Shape templates are shown for the uncertainties associated with the final-state
radiation in P YTHIA 8, the jet energy corrections, the factorization scale, and
the renormalization scale. Along with the intrinsic uncertainty included on
the EW corrections, these are seen to be the dominant uncertainties in the fit.
The shaded bars represent the raw template information, while the lines show
the shapes after smoothing and symmetrization procedures have been applied.
In the fit, the jet energy corrections are split into 26 different components, but
for brevity only the total uncertainty is shown here. Variation between years
is minimal for each of these uncertainties, though they are treated separately
in the fit. (repeated from page 112)
 197

Bibliography

[1] A. Einstein, “Über einem die Erzeugung und Verwandlung des Lichtes
betreffenden heuristischen Gesichtspunkt”, Annalen der physik 4 (1905).

[2] C. Jönsson, “Elektroneninterferenzen an mehreren künstlich hergestellten

Feinspalten”, Zeitschrift für Physik 161 (1961), no. 4, 454–474.

[3] P. G. Merli, G. Missiroli, and G. Pozzi, “On the statistical aspect of electron
interference phenomena”, Am. J. Phys 44 (1976), no. 3, 306–307.

[4] D. Hanneke, S. F. Hoogerheide, and G. Gabrielse, “Cavity control of a

single-electron quantum cyclotron: Measuring the electron magnetic
moment”, Physical Review A 83 (2011), no. 5, 052122.

[5] T. Aoyama, M. Hayakawa, T. Kinoshita et al., “Tenth-order electron

anomalous magnetic moment: Contribution of diagrams without closed
lepton loops”, Physical Review D 91 (2015), no. 3, 033006.

[6] H. Georgi, “Lie algebras in particle physics: from isospin to unified theories”,
volume 54. Westview press, 1999.

[7] A. Zee, “Group theory in a nutshell for physicists”, volume 17. Princeton
University Press, 2016.
 198

[8] H. Yukawa, “On the interaction of elementary particles. I”, Proceedings of the
Physico-Mathematical Society of Japan. 3rd Series 17 (1935) 48–57.

[9] C.-S. Wu, E. Ambler, R. Hayward et al., “Experimental test of parity

conservation in beta decay”, Physical review 105 (1957), no. 4, 1413.

[10] CMS Collaboration, “Observation of a New Boson at a Mass of 125 GeV with
the CMS Experiment at the LHC”, Phys. Lett. B 716 (2012) 30,
doi:10.1016/j.physletb.2012.08.021, arXiv:1207.7235.

[11] ATLAS Collaboration, “Observation of a new particle in the search for the
Standard Model Higgs boson with the ATLAS detector at the LHC”, Phys.
Lett. B 716 (2012) 1, doi:10.1016/j.physletb.2012.08.020,
arXiv:1207.7214.

[12] Y. Sofue and V. Rubin, “Rotation curves of spiral galaxies”, Annual Review of
Astronomy and Astrophysics 39 (2001), no. 1, 137–174.

[13] Supernova Cosmology Project Collaboration, “Measurements of Ω and Λ

from 42 high-redshift supernovae”, The Astrophysical Journal 517 (1999),
no. 2, 565.

[14] High-Z SN Search Team Collaboration, “Observational evidence from

supernovae for an accelerating universe and a cosmological constant”, The
Astronomical Journal 116 (1998), no. 3, 1009.

[15] K. G. Wilson, “The renormalization group and critical phenomena”, Reviews

of Modern Physics 55 (1983), no. 3, 583.

[16] A. Kar, “Renormalization from classical to quantum physics”, University of

Rochester PhD Thesis (2014).
 199

[17] S. Aoki, Y. Aoki, D. Bečirević et al., “Review of lattice results concerning

low-energy particle physics”, The European Physical Journal C 77 (2017), no. 2,
112.

[18] “Millenium Problems”. https://www.claymath.org/millennium-problems.

[19] R. D. Peccei, “The strong CP problem and axions”, in Axions, pp. 3–17.
Springer, 2008.

[20] M. Dine and A. Kusenko, “Origin of the matter-antimatter asymmetry”,

Reviews of Modern Physics 76 (2003), no. 1, 1.

[21] A. R. Liddle, “An introduction to cosmological inflation”, High energy physics

and cosmology (1998) 260.

[22] D. G., “The pre-LHC Higgs hunt”, Phil. Trans. R. Soc. A. 373 (2015) 20140039,
doi:10.1098/rsta.2014.0039.

[23] A. Blondel, “Precision electroweak physics at LEP”,

doi:10.1016/0920-5632(94)90655-6.

[24] T. A. Collaboration, the DELPHI Collaboration, the L3 Collaboration et al.

Physics Reports 427 (May, 2006) 257–454,
doi:10.1016/j.physrep.2005.12.006, arXiv:hep-ex/0509008.

[25] M. Hairer, “Regularity structures and the dynamical Φ43 model”, arXiv
preprint arXiv:1508.05261 (2015).

[26] R. J. Adler, “An Introduction to Continuity, Extrema, and Related Topics for
General Gaussian Processes”, Lecture Notes-Monograph Series 12 (1990) i–155.

[27] B. Duplantier, R. Rhodes, S. Sheffield et al., “Log-correlated Gaussian fields:

 200

an overview”, arXiv preprint arXiv:1407.5605 (2014).

[28] S. Rajeev and E. Ranken, “Continuity of scalar fields with logarithmic

correlations”, Physical Review D 92 (2015), no. 4, 045021.

[29] A. Kar and S. Rajeev, “Non-Riemannian metric emergent from scalar

quantum field theory”, Physical Review D 86 (2012), no. 6, 065022.

[30] E. Ogievetsky, N. Reshetikhin, and P. Wiegmann, “The principal chiral field

in two dimensions on classical lie algebras: The Bethe-ansatz solution and
factorized theory of scattering”, Nuclear Physics B 280 (1987) 45 – 96,
doi:https://doi.org/10.1016/0550-3213(87)90138-6.

[31] A. Polyakov and P. Wiegmann, “Goldstone fields in two dimensions with

multivalued actions”, Physics Letters B 141 (1984), no. 3, 223 – 228,
doi:https://doi.org/10.1016/0370-2693(84)90206-5.

[32] S. Rajeev and E. Ranken, “Highly nonlinear wave solutions in a dual to the
chiral model”, Physical Review D 93 (2016), no. 10, 105016.

[33] G. S. Krishnaswami and T. R. Vishnu, “On the Hamiltonian formulation,

integrability and algebraic structures of the Rajeev-Ranken model”, Journal of
Physics Communications 3 (feb, 2019) 025005,
doi:10.1088/2399-6528/ab02a9.

[34] G. S. Krishnaswami and T. Vishnu, “Invariant tori, action-angle variables,

and phase space structure of the Rajeev-Ranken model”, Journal of
Mathematical Physics 60 (2019), no. 8, 082902.

[35] “Constraint on the matter–antimatter symmetry-violating phase in neutrino

oscillations”, Nature 580 (Apr, 2020) 339–344,
 201

doi:10.1038/s41586-020-2177-0.

[36] Particle Data Group Collaboration, “Review of Particle Physics”, Phys. Rev.
D 98 (2018) 030001, doi:10.1103/PhysRevD.98.030001.

[37] D0 Collaboration Collaboration, “Observation of the Top Quark”, Phys. Rev.

Lett. 74 (Apr, 1995) 2632–2637, doi:10.1103/PhysRevLett.74.2632.

[38] CDF Collaboration Collaboration, “Observation of Top Quark Production in

pp Collisions with the Collider Detector at Fermilab”, Phys. Rev. Lett. 74
(Apr, 1995) 2626–2631, doi:10.1103/PhysRevLett.74.2626.

[39] M. Czakon and A. Mitov, “Top++: A program for the calculation of the
top-pair cross-section at hadron colliders”, Comput. Phys. Commun. 185
(2014) 2930, doi:10.1016/j.cpc.2014.06.021, arXiv:1112.5675.

[40] M. Czakon, D. Heymes, and A. Mitov, “High-Precision Differential

Predictions for Top-Quark Pairs at the LHC”, Physical Review Letters 116
(Feb, 2016) doi:10.1103/physrevlett.116.082003.

[41] CMS Collaboration, “Identification of heavy-flavour jets with the CMS

detector in pp collisions at 13 TeV”, JINST 13 (2018) P05011,
doi:10.1088/1748-0221/13/05/P05011, arXiv:1712.07158.

[42] L. Evans and P. Bryant, “LHC Machine”, Journal of Instrumentation 3 (aug,

2008) S08001–S08001, doi:10.1088/1748-0221/3/08/s08001.

[43] G. Roland, K. Šafařík, and P. Steinberg, “Heavy-ion collisions at the LHC”,

Progress in Particle and Nuclear Physics 77 (2014) 70–127.

[44] CMS Collaboration, “Particle-flow reconstruction and global event

 202

description with the CMS detector”, JINST 12 (2017) P10003,

doi:10.1088/1748-0221/12/10/P10003, arXiv:1706.04965.

[45] CMS Collaboration, “The CMS experiment at the CERN LHC”, JINST 3
(2008) S08004, doi:10.1088/1748-0221/3/08/S08004.

[46] GEANT4 Collaboration, “GEANT4—a simulation toolkit”, Nucl. Instrum.

Meth. A 506 (2003) 250, doi:10.1016/S0168-9002(03)01368-8.

[47] G. C. Branco, P. M. Ferreira, L. Lavoura et al., “Theory and phenomenology

of two-Higgs-doublet models”, Phys. Rept. 516 (2012) 1,
doi:10.1016/j.physrep.2012.02.002, arXiv:1106.0034.

[48] K. Agashe, R. Contino, and A. Pomarol, “The Minimal composite Higgs

model”, Nucl. Phys. B 719 (2005) 165,
doi:10.1016/j.nuclphysb.2005.04.035, arXiv:hep-ph/0412089.

[49] CMS Collaboration, “Observation of ttH production”, Phys. Rev. Lett. 120
(2018) 231801, doi:10.1103/PhysRevLett.120.231801, arXiv:1804.02610.

[50] ATLAS Collaboration, “Observation of Higgs boson production in

association with a top quark pair at the LHC with the ATLAS detector”,
Phys. Lett. B 784 (2018) 173, doi:10.1016/j.physletb.2018.07.035,
arXiv:1806.00425.

[51] CMS Collaboration, “Combined measurements of Higgs boson couplings in

√
proton-proton collisions at s = 13 TeV”, Eur. Phys. J. C 79 (2019) 421,
doi:10.1140/epjc/s10052-019-6909-y, arXiv:1809.10733.

[52] J. H. Kühn, A. Scharf, and P. Uwer, “Weak interactions in top-quark pair

production at hadron colliders: an update”, Phys. Rev. D 91 (2015) 014020,
 203

doi:10.1103/PhysRevD.91.014020, arXiv:1305.5773.

[53] M. Aliev, H. Lacker, U. Langenfeld et al., “HATHOR: HAdronic Top and

Heavy quarks crOss section calculatoR”, Comput. Phys. Commun. 182 (2011)
1034, doi:10.1016/j.cpc.2010.12.040, arXiv:1007.1327.

[54] P. Nason, “A new method for combining NLO QCD with shower Monte
Carlo algorithms”, JHEP 11 (2004) 040,
doi:10.1088/1126-6708/2004/11/040, arXiv:hep-ph/0409146.

[55] S. Frixione, P. Nason, and C. Oleari, “Matching NLO QCD computations

with parton shower simulations: the POWHEG method”, JHEP 11 (2007)
070, doi:10.1088/1126-6708/2007/11/070, arXiv:0709.2092.

[56] S. Alioli, P. Nason, C. Oleari et al., “A general framework for implementing

NLO calculations in shower Monte Carlo programs: the POWHEG BOX”,
JHEP 06 (2010) 043, doi:10.1007/JHEP06(2010)043, arXiv:1002.2581.

[57] J. M. Campbell, R. K. Ellis, P. Nason et al., “Top-pair production and decay at

NLO matched with parton showers”, JHEP 04 (2015) 114,
doi:10.1007/JHEP04(2015)114, arXiv:1412.1828.

[58] T. Sjöstrand, S. Mrenna, and P. Skands, “PYTHIA 6.4 physics and manual”,
JHEP 05 (2006) 026, doi:10.1088/1126-6708/2006/05/026,
arXiv:hep-ph/0603175.

[59] T. Sjöstrand, S. Mrenna, and P. Z. Skands, “A brief introduction to PYTHIA

8.1”, Comput. Phys. Commun. 178 (2008) 852,
doi:10.1016/j.cpc.2008.01.036, arXiv:0710.3820.

[60] R. Frederix and S. Frixione, “Merging meets matching in MC@NLO”, JHEP

 204

12 (2012) 061, doi:10.1007/JHEP12(2012)061, arXiv:1209.6215.

[61] P. Skands, S. Carrazza, and J. Rojo, “Tuning PYTHIA 8.1: the Monash 2013
tune”, Eur. Phys. J. C 74 (2014) 3024,
doi:10.1140/epjc/s10052-014-3024-y, arXiv:1404.5630.

[62] CMS Collaboration, “Extraction and validation of a new set of CMS

PYTHIA8 tunes from underlying-event measurements”, Eur. Phys. J. C 80
(2020) 4, doi:10.1140/epjc/s10052-019-7499-4, arXiv:1903.12179.

[63] NNPDF Collaboration, “Parton distributions for the LHC Run II”, JHEP 04
(2015) 040, doi:10.1007/JHEP04(2015)040, arXiv:1410.8849.

[64] NNPDF Collaboration, “Parton distributions from high-precision collider

data”, Eur. Phys. J. C 77 (2017) 663, doi:10.1140/epjc/s10052-017-5199-5,
arXiv:1706.00428.

[65] M. Czakon, D. Heymes, A. Mitov et al., “Top-pair production at the LHC

through NNLO QCD and NLO EW”, JHEP 10 (2017) 186,
doi:10.1007/JHEP10(2017)186, arXiv:1705.04105.

[66] D. Pagani, I. Tsinikos, and M. Zaro, “The impact of the photon PDF and
electroweak corrections on tt̄ distributions”, Eur. Phys. J. C 76 (2016) 479,
doi:10.1140/epjc/s10052-016-4318-z, arXiv:1606.01915.

[67] J. Alwall, R. Frederix, S. Frixione et al., “The automated computation of

tree-level and next-to-leading order differential cross sections, and their
matching to parton shower simulations”, JHEP 07 (2014) 079,
doi:10.1007/JHEP07(2014)079, arXiv:1405.0301.

[68] J. Alwall, S. Höche, F. Krauss et al., “Comparative study of various

 205

algorithms for the merging of parton showers and matrix elements in

hadronic collisions”, Eur. Phys. J. C 53 (2008) 473,
doi:10.1140/epjc/s10052-007-0490-5, arXiv:0706.2569.

[69] M. Cacciari, G. P. Salam, and G. Soyez, “The anti-kt jet clustering algorithm”,
JHEP 04 (2008) 063, doi:10.1088/1126-6708/2008/04/063,
arXiv:0802.1189.

[70] M. Cacciari, G. P. Salam, and G. Soyez, “FastJet user manual”, Eur. Phys. J. C
72 (2012) 1896, doi:10.1140/epjc/s10052-012-1896-2, arXiv:1111.6097.

[71] CMS Collaboration, “Jet energy scale and resolution in the CMS experiment
in pp collisions at 8 TeV”, JINST 12 (2017) P02014,
doi:10.1088/1748-0221/12/02/P02014, arXiv:1607.03663.

[72] CMS Collaboration, “Performance of the CMS muon detector and muon
√
reconstruction with proton-proton collisions at s = 13 TeV”, JINST 13
(2018) P06015, doi:10.1088/1748-0221/13/06/P06015, arXiv:1804.04528.

[73] CMS Collaboration, “Performance of electron reconstruction and selection

√
with the CMS detector in proton-proton collisions at s = 8 TeV”, JINST 10
(2015) P06005, doi:10.1088/1748-0221/10/06/P06005, arXiv:1502.02701.

[74] CMS Collaboration, “Measurement of differential cross sections for the

production of top quark pairs and of additional jets in lepton+jets events
√
from pp collisions at s = 13 TeV”, Phys. Rev. D 97 (2018) 112003,
doi:10.1103/PhysRevD.97.112003, arXiv:1803.08856.

[75] ATLAS Collaboration, “Measurement of the inelastic proton-proton cross

√
section at s = 13 TeV with the ATLAS detector at the LHC”, Phys. Rev. Lett.
 206

117 (2016) 182002, doi:10.1103/PhysRevLett.117.182002,

arXiv:1606.02625.

[76] CMS Collaboration, “Measurement of the top quark Yukawa coupling from
tt̄ kinematic distributions in the lepton+jets final state in proton-proton
√
collisions at s = 13 TeV”, Phys. Rev. D 100 (2019) 072007,
doi:10.1103/PhysRevD.100.072007, arXiv:1907.01590.

[77] M. G. Bowler, “e+ e− Production of heavy quarks in the string model”, Z.

Phys. C 11 (1981) 169, doi:10.1007/BF01574001.

[78] CMS Collaboration, “Search for standard model production of four top
quarks with same-sign and multilepton final states in proton-proton
√
collisions at s = 13 TeV”, Eur. Phys. J. C 78 (2018) 140,
doi:10.1140/epjc/s10052-018-5607-5, arXiv:1710.10614.

[79] CMS Collaboration Collaboration, “Measurement of the top quark Yukawa

coupling from tt̄ kinematic distributions in the dilepton final state at
√
s = 13 TeV”, Technical Report CMS-PAS-TOP-19-008, CERN, Geneva,
(2020).

[80] V. I. Bogachev, “Gaussian measures”. Number 62. American Mathematical

Soc., 1998.

[81] M. Ledoux and M. Talagrand, “Probability in Banach Spaces: isoperimetry

and processes”, volume 23. Springer Science & Business Media, 2013.

[82] R. J. Adler, “An introduction to continuity, extrema, and related topics for
general Gaussian processes”, Lecture Notes-Monograph Series (1990) i–155.

[83] R. M. Dudley, “Sample functions of the Gaussian process”, in Selected Works

 207

of RM Dudley, pp. 187–224. Springer, 2010.

[84] F. Biagini, Y. Hu, B. Øksendal et al., “Stochastic calculus for fractional

Brownian motion and applications”. Springer Science & Business Media,
2008.

[85] S. Dodelson, “Modern cosmology”. Academic press, 2003.

[86] D. Pickrell, “Invariant measures for unitary groups associated to Kac-Moody

Lie algebras”, volume 693. American Mathematical Soc., 2000.

[87] A. Sengupta, “The Yang-Mills measure for S 2”, Journal of functional analysis
108 (1992), no. 2, 231–273.

[88] N. Charalambous and L. Gross, “The Yang-Mills heat semigroup on

three-manifolds with boundary”, Communications in Mathematical Physics
317 (2013), no. 3, 727–785.

[89] C. R. Nappi, “Some properties of an analog of the chiral model”, Phys. Rev. D
21 (1980), no. 2, 418.

[90] C. R. Nappi and E. Witten, “Wess-Zumino-Witten model based on a

nonsemisimple group”, Phys. Rev. Lett. 71 (1993), no. 23, 3751.

[91] T. Curtright and C. Zachos, “Currents, charges, and canonical structure of

pseudodual chiral models”, Phys. Rev. D 49 (1994), no. 10, 5408.

[92] E. Ogievetsky, N. Reshetikhin, and P. Wiegmann, “The principal chiral field

in two dimensions on classical lie algebras: the Bethe-ansatz solution and
factorized theory of scattering”, Nucl. Phys. B 280 (1987) 45–96.

[93] A. Polyakov and P. Wiegmann, “Goldstone fields in two dimensions with

 208

multivalued actions”, Phys. Lett. B 141 (1984), no. 3, 223–228.

[94] S. Coleman, “Quantum sine-Gordon equation as the massive Thirring

model”, Phys. Rev. D. 11 (1975), no. 8, 2088.

[95] L. D. Faddeev and V. E. Korepin, “Quantum theory of solitons”, Phys. Rep.

42 (1978), no. 1, 1–87.

[96] C. G. Callan, D. Friedan, E. Martinec et al., “Strings in background fields”,

Nucl. Phys. B 262 (1985), no. 4, 593–609.

[97] J. E. Heebner, R. W. Boyd, and Q. H. Park, “Slow light, induced dispersion,

enhanced nonlinearity, and optical solitons in a resonator-array waveguide”,
Phys. Rev. E 65 (2002), no. 3, 036619.

[98] R. Montgomery, “A tour of subriemannian geometries, their geodesics and

applications”. Number 91. American Mathematical Soc., 2006.

[99] A. Shapere and F. Wilczek, “Self-propulsion at low Reynolds number”, Phys.

Rev. Lett. 58 (1987), no. 20, 2051.

[100] L. Hormander, “Hypoelliptic second order differential equations”, Acta

Math. 119 (1967), no. 1, 147–171.

[101] E. Witten, “Non-abelian bosonization in two dimensions”, Comm. Math. Phys.

92 (1984), no. 4, 455–472.

[102] M. Bos, “Dimensional regularization in the Wess-Zumino-Witten model”,

Phys. Lett. B 189 (1987), no. 4, 435–441.

[103] X. Zheng-Min, “Dimensional regularization and three-loop beta function of

the Wess-Zumino-Witten model”, Nucl. Phys. B 314 (1989), no. 1, 112–128.
 209

[104] B. E. Baaquie and K. K. Yim, “Sigma model Lagrangian for the Heisenberg
group”, Phys. Lett. B 615 (2005), no. 1, 134–140.

[105] LHC Higgs Cross Section Working Group, “Handbook of LHC Higgs Cross
Sections: 3. Higgs Properties”, doi:10.5170/CERN-2013-004,
arXiv:1307.1347.

[106] CMS Collaboration, “Measurement of the top quark mass using

√
proton-proton data at s = 7 and 8 TeV”, Phys. Rev. D 93 (2016) 072004,
doi:10.1103/PhysRevD.93.072004, arXiv:1509.04044.

[107] W. Hollik and M. Kollar, “NLO QED contributions to top-pair production at

hadron collider”, Phys. Rev. D 77 (2008) 014008,
doi:10.1103/PhysRevD.77.014008, arXiv:0708.1697.

[108] B. A. Betchart, R. Demina, and A. Harel, “Analytic solutions for neutrino

momenta in decay of top quarks”, Nucl. Instrum. Meth. A 736 (2014) 169,
doi:10.1016/j.nima.2013.10.039, arXiv:1305.1878.

[109] M. Czakon, D. Heymes, and A. Mitov, “Dynamical scales for multi-TeV

top-pair production at the LHC”, JHEP 04 (2017) 071,
doi:10.1007/JHEP04(2017)071, arXiv:1606.03350.

[110] CMS Collaboration, “CMS Luminosity Measurements for the 2016 Data
Taking Period”, CMS Physics Analysis Summary CMS-PAS-LUM-17-001,
(2017).

[111] CMS Collaboration, “CMS luminosity measurement for the 2017 data-taking
√
period at s = 13 TeV”, CMS Physics Analysis Summary
CMS-PAS-LUM-17-004, (2018).
 210

[112] CMS Collaboration, “CMS luminosity measurement for the 2018 data-taking
√
period at s = 13 TeV”, CMS Physics Analysis Summary
CMS-PAS-LUM-18-002, (2019).

[113] M. Czakon, D. Heymes, and A. Mitov, “fastNLO tables for NNLO top-quark
pair differential distributions”, arXiv:1704.08551.

[114] M. Czakon, D. Heymes, and A. Mitov, “High-precision differential

predictions for top-quark pairs at the LHC”, Phys. Rev. Lett. 116 (2016)
082003, doi:10.1103/PhysRevLett.116.082003, arXiv:1511.00549.

[115] M. Czakon, S. Dulat, T.-J. Hou et al., “A study of the impact of

double-differential top distributions from CMS on parton distribution
functions”, arXiv:1912.08801.

[116] W. S. Cleveland, “Robust Locally Weighted Regression and Smoothing

Scatterplots”, Journal of the American Statistical Association 74 (1979) 829,
doi:10.1080/01621459.1979.10481038.