This is the accepted manuscript made available via CHORUS.

The article has been

published as:

Jet substructure at the Large Hadron Collider

Roman Kogler, Benjamin Nachman, Alexander Schmidt, Lily Asquith, Emma Winkels,
Mario Campanelli, Chris Delitzsch, Philip Harris, Andreas Hinzmann, Deepak Kar, Christine
McLean, Justin Pilot, Yuta Takahashi, Nhan Tran, Caterina Vernieri, and Marcel Vos
Rev. Mod. Phys. 91, 045003 — Published 12 December 2019
DOI: 10.1103/RevModPhys.91.045003
Jet Substructure at the Large Hadron Collider
Roman Kogler (ed.) and Andreas Hinzmann
Universität Hamburg,
Germany

Benjamin Nachman (ed.)

Lawrence Berkeley National Laboratory,
USA

Alexander Schmidt (ed.)

RWTH Aachen University,
Germany

Lily Asquith and Emma Winkels

University of Sussex,
UK

Mario Campanelli
University College London,
UK

Chris Delitzsch
University of Arizona,
USA

Philip Harris
Massachusetts Institute of Technology,
USA

Deepak Kar
University of Witwatersrand,
South Africa

Christine McLean and Justin Pilot

University of California, Davis,
USA

Yuta Takahashi
Universität Zürich,
Switzerland

Nhan Tran and Caterina Vernieri

Fermilab, USA

Marcel Vos
IFIC Valencia, Spain

(Dated: September 26, 2019)

Jet substructure has emerged to play a central role at the Large Hadron Collider (LHC),
where it has provided numerous innovative ways to search for new physics and to probe
the Standard Model, particularly in extreme regions of phase space. In this article
we focus on a review of the development and use of state-of-the-art jet substructure
techniques by the ATLAS and CMS experiments.

CONTENTS A. Inputs 6

Contents 1 B. Calibration 7
 2

A. Definition 8 decays. The following experimental overview describes

B. Mitigation Methods 9 techniques for measuring jets as proxies for hadronic de-
C. Performance Studies 10
cays of W , Z, H bosons and top quarks. However, this
A. Jet Grooming 11
B. Jet Mass 12 review is not limited to these methods but covers also
C. Other Jet Substructure Observables 14 precision jet substructure measurements and the discrim-
A. Quark/Gluon Discrimination 16 ination of quark and gluon jets, reflecting the versatility
B. Vector Boson Tagging 19 of jet substructure. The scientific gains from these mea-
C. Top Tagging 22
D. H → bb̄ Tagging 25 surements are manifold, reaching from precision studies
A. Measurements of Jet Substructure 28 of QCD over the determination of fundamental parame-
1. Jet mass 28 ters of the Standard Model to searches for new physical
2. Jet Charge 30 phenomena at the highest energy scales. A recent re-
3. Other Jet Substructure Observables 31
B. Measurements with Jet Substructure 31
view on the theoretical aspects of jet substructure can
1. Differential tt Cross Section Measurements 31 be found in Ref. (16).
2. W/Z/H Cross Sections 33 Since the first evidence for jets in e+ e− collisions at
A. Diboson Resonances 35 SPEAR (17), jets have had a significant impact on the
B. tt̄ Resonances 36
C. Vector-like Quarks 37
research program of every particle collider since DORIS
D. Leptophobic Z 0 38 through the LHC, and beyond to the design of future
colliders. There is no single, universal definition of a
jet – which particles belong to a jet depend on the al-
I. INTRODUCTION gorithm used to combine particles into jets. In the be-
ginning of jets from the mid 1970’s, there were no jet
Jets are collimated sprays of particles, produced in clustering algorithms; information from the whole event
abundance in high energy particle collisions. They are was used instead of localized energy flows. The spheric-
ubiquitous in particle collider experiments and indespen- ity tensor (18) was typically used to obtain a jet axis
sible to study the underlying dynamics and interactions. for events with a back-to-back dijet topology. Quanti-
Jets have played a central role in the discovery and prop- tative statements about data were obtained from event
erty measurements of many fundamental particles like shapes, like the sphericity or thrust (19; 20). Sphericity
the gluon (g) (1; 2; 3; 4) and the top quark (t) (5; 6). is a measure for the isotropy of the produced particles
They have provided key insights into the structure of and thrust is a measure of the directed energy flow along
the strong force and were indispensable in the study of an axis that maximises this flow in an event. These event
Higgs boson (H) couplings to heavy third generation shapes can be used to characterize how compatible events
quarks (7; 8; 9; 10). Because of their large production are with the assumption of two oppositely directed, colli-
rate at the LHC, jets feature prominently in searches for mated jets. A clear theoretical advantage of these event
new particles and precision measurements of Standard shapes is that they are calculable in perturbative Quan-
Model (SM) properties. However, important information tum Chromodynamics (pQCD). This was realized early
on the underlying particle dynamics is not only carried on and the calculability ultimately resulted in the confir-
by the total four-momenta of jets, but also by their in- mation of the parton
√ model and, with data from exper-
ternal structure. Investigations of this jet substructure iments at higher s, the discovery of the gluon in three
reveal a wealth of physical processes and pose interesting jet events at PETRA (1; 2; 3; 4).
theoretical and experimental challenges. While relatively When studying the dynamics of quark and gluon scat-
young, the field of jet substructure has become an impor- tering, it became necessary to perform quantitative anal-
tant field of research over the last decade and will gain yses and calculations that go beyond event shapes. For
further importance with the future data taking periods these to be possible, it was realized that it is mandatory
at the LHC. to define a deterministic set of rules on how particles are
With the advent of the LHC it was realized that decays combined into jets. A schematic drawing depicting this
of hypothetical, very heavy resonances can lead to highly problem is shown in figure 1. While the sphericity axis
Lorentz-boosted heavy SM particles, W , Z, H bosons is uniquely defined and easily calculable, the direction
and top quarks (11; 12; 13; 14; 15). Since these parti- and magnitude of the jet axes depend on which particles
cles feature the largest branching fractions into hadrons, should be combined into a given jet, and how the particles
final states with fully-hadronic decays have high sensi- are combined to obtain the axes. An intuitive definition
tivity in LHC analyses. The large boost leads to very for a jet algorithm consists of summing the momenta of
collimated decays, where particle masses of O(100) GeV all particles within a cone with fixed size (21). Naive
are not large enough for the outgoing quarks to be suf- cone algorithms are not infared and collinear (IRC) safe
ficiently separated relative to each other to be resolved – the requirement that the resulting jets be insensitive to
into individual jets. It is this small opening angle between arbitrarily low energy particles and collinear splittings.
the decay products which leads to fully-merged particle IRC safety is a useful theoretical requirement for making
 3

results in nearly perfect conical jets the LHC collabora-

tions made a transition to this algorithm. Today, almost
all studies involving jets performed at the LHC use this
algorithm. Even when analyzing the substructure of jets
with advanced grooming or tagging techniques, the ini-
tial step often consists of building an ensemble of particles
that were clustered with the anti-kT algorithm.
So far, it has not been specified what the term particle
refers to when using particles as input to jet clustering. In
fact, in jet physics, the term particle is often used gener-
ically for different sorts of objects, whose ensemble com-
prises the input to a given jet algorithm. Three different
ensembles are commonly used. The partonic final state
includes all particles resulting from the parton shower be-
FIG. 1 Schematic drawing of particles emerging from the
hard scattering of a high energy particle collision. The fore the hadronization starts (which is unphysical). This
sphericity axis is shown as dashed line. also include photons when these were created in the hard
interaction or emitted from charged particles during the
parton shower. The ensemble on the particle level, also
calculations in pQCD and is also a convenient language called hadron level, consists of hadrons and their decay
for describing the experimental robustness to noise and products, including photons and leptons. The detector
detector granularity. level input consists of calorimeter clusters, reconstructed
There exist many variants of cone-type algorithms, de- particle tracks or combinations thereof. Jet algorithms
veloped in the attempt to solve the IRC unsafety of naive using these different ensembles as input result in parton,
cone jet algorithms. This stems from the necessity of an particle or detector level jets, respectively. Ideally, in any
initial axis, which was eventually solved with the formu- given event, the jets obtained on parton, particle and de-
lation of the SISCone algorithm (22). Although this al- tector level are as similar as possible. Realistically, agree-
gorithm is IRC safe, it is not widely used today because it ment can not be achieved, but a close correspondence
was found that sequential recombination algorithms have ensures the possibility to study the underlying partonic
several advantages over cone-type algorithms. First used dynamics with the use of jets. It is this correspondence,
by the JADE Collaboration (23; 24), the initial version paired with calculability in pQCD, which makes jets in-
of a recombination algorithm defined for e+ e− collisions dispensable tools at high energy particle colliders2 .
was improved in several steps (25; 26), to finally arrive at Soon after their discovery, it was realized that not only
the longitudinally-invariant kT -clustering algorithm for the kinematics of jets but also their internal structure
hadron-hadron collisions (27). A generalization of this carry information. The parton shower and subsequent
algorithm leads to three classes, distinct only by the sign hadronization leads to a characteristic multiplicity, as
of the exponent of the transverse momentum pT,i in the well as angular and momentum distributions of hadrons
inter-particle distance measure inside jets, which depend on the parton that initiated
the shower. For example, the probability of a q → qg
∆R2 splitting is proportional to the color factor CF = 4/3 at
dij (pi , pj ) = min(p2k 2k
T,i , pT,j ) , (1)
R2 leading order in QCD, while the probability of g → gg
is proportional to CA = 3. The larger value of CA re-
where1 ∆R2 = ∆φ2 + ∆y 2 and R is typically called
sults in a larger multiplicity of hadrons and in broader
the jet radius. The original kT algorithm, with k = 1
jets. This lead to the suggestion of measuring jet shapes,
in Eq. (1), clusters soft and collinear particles first, the
defined as the fractional transverse momentum profile of
Cambridge/Aachen algorithm (CA) (29; 30), with k = 0,
particles within a concentric inner cone, smaller than the
prioritizes particles in the clustering solely by their an-
jet cone of the original jet, and pointed to their useful-
gular proximity, and the anti-kT algorithm (31), with
ness for distinguishing quark jets from gluon jets (34).
k = −1, combines the hardest particles first. The pro-
Experimental results from LEP (35; 36; 37; 38), Teva-
posal of the latter algorithm is also responsible for the
tron (39; 40) and HERA (41; 42; 43) confirmed this and
disappearance of cone-type algorithms in experimental
can be considered as the starting point of physics with
studies. When it was realized that the anti-kT algorithm
jet substructure in particle physics.

1 Sometimes the rapidity (y) is used and sometimes the pseudo- 2 For a theoretical introduction to jets, we recommend the reviews
rapidity (η) is used depending on the application. See Ref. (28) in Refs. (32; 33) as well as the theory companion this experimen-
for a detailed discussion. tal review, Ref. (16).
 4

sible due to advances in experimental methods. New

technologies, like silicon pixel detectors, high-resolution
tracking detectors in conjunction with strong magnetic
fields, highly granular calorimeters with low electronic
noise and lightweight materials for detector structures
with little dead material inside the active detector volume
have enabled increasingly precise jet measurements and
studies of internal jet structure. Modern particle detec-
FIG. 2 Schematic drawing of particles clustered into a single tors at the LHC are equipped with many layers of high-
jet. Two subjet axes are shown as dashed lines. resolution tracking detectors, strong and very homoge-
neous magnetic fields and finely segmented calorimeters
with an excellent energy resolution. With these tech-
At the LHC, jet substructure is used to identify highly nologies, the ATLAS and CMS detectors3 are equipped
boosted heavy SM particles in fully hadronic decays. An to track and reconstruct individual particles produced in
example of a jet with substructure from a two-prong de- high energy collisions. On average about 60% of a jet’s
cay is shown schematically in figure 2. The difficulty lies momentum is carried by charged hadrons, photons ac-
in identifying the underlying process that led to the fi- count for about 25% of the total jet momentum and the
nal state, for example distinguishing W → q q̄ 0 , Z → q q̄ remaining 15% can be attributed to long-lived neutral
or H → bb̄ from QCD splittings like q → qg, g → gg hadrons (44). With increasing jet energy, the particle
or g → q q̄. Numerous algorithms have been suggested multiplicity increases, and also the fraction of the jet’s
to identify specific decays, which are part of a class of momentum carried by soft particles. For example, on
jet substructure taggers. The idea behind many of these average 50% of the momentum of a 50 GeV jet is car-
algorithms is related to event shapes in e+ e− collisions. ried by particles with a momentum less than 5% of the
By defining N axes within a jet, it is possible to check for jet’s momentum. It is therefore crucial to ensure that
the compatibility of a fully-merged N -prong decay. How particles with energies down to O(100 MeV) can be re-
these axes are found typically differs from algorithm to constructed in order to retain the full information on a
algorithm, and some techniques do not even explicitly jet’s kinematics and internal structure.
require axes. Popular concepts are an exclusive jet clus- As important as the reconstruction of the total jet
tering using the particles inside a jet as input, or the energy is the measurement of the jet constituent mul-
maximization of the projection of the jet constituents’ tiplicity and their angular distributions. While charged
momenta onto the desired number of axes, as illustrated particles can be efficiently reconstructed as tracks, neu-
in figure 2. Since the opening angle between the quarks tral particles develop showers in the calorimeters and the
depends on the momentum of the parent particle and possibility to resolve two separate showers depends on
its mass, larger jets (R ∼ 1) than normally employed in the granularity of the calorimeter and the lateral shower
LHC analyses (R ∼ 0.4) are used to reconstruct boosted development. Hence, it becomes more difficult to sepa-
heavy particle decays. A larger distance parameter is rate two adjacent particles in dense environments, such as
chosen to capture the full kinematics of the decay already high momentum jets, and the situation is aggravated by
at moderate momenta of 200–400 GeV. The drawback of the presence of hadronic showers from charged hadrons.
jets with large areas is unwanted contributions from the Often it is impossible to build one calorimeter cluster per
underlying event and from multiple proton-proton colli- neutral particle. A way to improve the angular resolution
sions in a single bunch crossing (pile-up). These lead to a in jet substructure analyses is to combine measurements
worsening of the resolution in quantities used to identify from the tracking detectors and calorimeters. Using com-
the substructure of jets, like the jet mass. Jet grooming bined detector measurements as input to jet algorithms,
and pile-up removal algorithms have been developed to for example using a particle flow approach, results in im-
mitigate these effects. Grooming algorithms aim at re- proved resolutions of jet substructure observables, com-
moving soft and wide-angle radiation, therefore not only pared to using only tracks or only calorimeter clusters.
reducing the effects from the underlying event but also An important aspect of experimental analyses at the
reducing the sensitivity to the details of fragmentation. LHC is the calibration of jets, necessitated by the non-
Pile-up removal algorithms are designed to identify and
subtract contributions from a different interaction ver-
tex, by eliminating uncorrelated radiation from jets. A 3 The ALICE and LHCb detectors are also well-equipped to per-
combination of these techniques often leads to the best form jet substructure studies. While these experiments do not
overall performance and it is an ongoing effort to un- have access to boosted massive particles due to their data rate
(ALICE) or acceptance (LHCb), the are performing many inter-
derstand the interplay of pile-up removal, grooming and esting QCD studies with jet substructure. This review will be
tagging algorithms. focused on ATLAS and CMS, but the future of jet substructure
The theoretical and algorithmic developments are pos- will involve key contributions from all four LHC experiments.
 5

compensating nature of hadron calorimeters, suppression ployed in ATLAS and CMS analyses are discussed in
of electronic noise, tracking inefficiencies, dead material section IV. In section V we review jet grooming tech-
in front of calorimeters, the influence of pile-up and other niques in use in experimental analyses and discuss their
effects. While the calibration of the total jet energy impact on jet substructure observables. A special em-
scale is an important aspect in all analyses using jets, phasis is given on the jet mass calibration and jet mass
the precise knowledge of the jet mass scale and the de- measurements in different final states. Measurements of
tector response to jet substructure observables and jet other jet substructure distributions are described as well.
tagging algorithms is specific to jet substructure analy- One of the key developments within the field of jet sub-
ses. Calibrating the jet energy scale results in a change structure are tagging algorithms, which are described in
of the magnitude of the jet’s four-momentum, where the detail in section VI. Theoretical and experimental devel-
jet mass scale comprises an additional degree of freedom opments have resulted in large performance gains of sub-
that can not be constrained by the typical methods of structure taggers in the last years, relevant for a large
balancing a jet with a well-calibrated reference object. number of present and future physics analyses. We high-
The jet mass scale is usually calibrated using jets from light the main developments and improvements and give
fully-merged, highly boosted W → q q̄ 0 decays, facilitat- an overview of relevant experimental studies. The use of
ing a calibration of the peak position in the jet mass jet substructure taggers in existing cross section measure-
distribution. Measurements of the jet mass distribution ments is reviewed in section VII. So far, the major ben-
from light quark and gluon jets, as well as from fully- eficiaries of jet substructure methods have been analyses
hadronic highly-boosted W , Z and t decays allow for in search for new physical phenomena. We review the
precise tests of the modelling of perturbative and non- application of these methods to searches for new physics
perturbative effects in jet production. Similar measure- in section VIII and conclude in section IX.
ments can also be used to study the detector response to
jet substructure observables and their modelling in sim-
ulation. A mis-modelling of variables used for tagging, II. ATLAS AND CMS DETECTORS
either in the detector simulation or on the level of the
underlying physics, can result in a wrong estimation of The ATLAS (45) and CMS (46) detectors are designed
the tagging efficiency or the misidentification rate, with to observe leptons, photons, and hadrons resulting from
important consequences for measurements. In order to LHC pp collisions. The physics of the hard reaction
overcome this limitation, measurements of tagging effi- takes place at the point of collision (the primary ver-
ciencies and misidentification rates are performed in sam- tex) within the beam pipe. Beyond the beam pipe4 ,
ples enriched with the particle decays in question. While at 4.4 cm (3.3 cm) in CMS (ATLAS), the first cylindri-
these measurements do not help to understand the cause cal layer of detectors encountered are silicon pixels and
of the mis-modelling or to improve the description of jet strips for identification of charged particles. CMS pro-
substructure distributions, these can be used to correct vides a 3.8 T magnetic field via a solenoid positioned
the efficiencies in simulation. It is these measurements outside the silicon tracking detector, the Electromag-
that have enabled the use of jet substructure taggers in netic Calorimeter (ECAL) and most of the Hadronic
numerous physics analyses since the beginning of data Calorimeter (HCAL). ATLAS has an additional tracking
taking at the LHC. The increased statistics from a data layer composed of straw drift tubes (Transition Radia-
sample corresponding to about 150 fb−1 per experiment tion Tracking or TRT), with a 2 T magnetic field encom-
at a centre-of-mass energy of 13 TeV can now be used to passing the silicon and TRT detectors, while the ECAL
improve our understanding of the detector response to and HCAL are situated outside the solenoidal magnet.
jet substructure algorithms, the underlying physics and The calorimeters are surrounded by muon spectrometers
the performance differences of taggers. These studies and which build the outermost part of the ATLAS and CMS
measurements represent the continuation of an exciting detectors. Both detectors are nearly hermetic and can
physics program at the LHC in a field which reached therefore measure the missing transverse momentum.
its adolescence in the past few years. In the years to The energy and momentum ranges and resolutions for
come, the field of jet substructure will evolve and mature the barrel regions5 of ATLAS and CMS are shown in
through precision measurements and the exploration of table I along with the measurement granularity, which
unknown territory.
We begin this review with a brief overview of the AT-
LAS and CMS detectors in section II, followed by a de- 4 The LHC collaborations are continuously working to improve
scription of the input to jet reconstruction and jet cal- the detectors; the numbers given here are for the detectors that
operated in 2015-2017. Before and after this time, the exact
ibration in section III. An important aspect of jet re-
values are not the same as reported here.
construction at the LHC, and jet substructure in par- 5 For example, the ATLAS ECAL barrel covers the pseudorapidity
ticular, are algorithms to mitigate the effects of pile-up. range |η| < 1.475, the end-caps cover 1.375 < |η| < 3.2 and the
Recent experimental advancements and algorithms em- forward ECAL layer extends the coverage up to |η| < 4.9. The
 6

limits the angular resolution. The better energy res- ATLAS CMS
olution of the CMS ECAL is due to the use of lead
tungstate (PbWO4 ) crystals, as opposed to the Liquid Tracking
Argon (LAr) used by ATLAS. The differences in the AT- 1/pT resolution 0.05% × 0.02%×pT / GeV⊕
LAS and CMS calorimeter designs are a result of the pT / GeV⊕1% (47) 0.8% (48)
different ranking of priorities decided by the two collab- d0 resolution 20 (49) 20 (48)
orations; ATLAS chose a radiation-hard technology with (µm)
sufficient resolution in a fine sampling LAr calorimeter,
ECAL
while CMS prioritized the excellent resolution of a total √ √
absorption crystal calorimeter (the focus was Higgs mass E resolution 10%/ E ⊕ 3%/ E ⊕
reconstruction), and accepted the accompanying limita- 0.2% (45) 12%/E ⊕
0.3% (46)
tions in radiation-hardness associated with this technol-
ogy. The CMS ECAL crystal response varies under ir- granularity 0.025 × 0.025 0.017 × 0.017
radiation, which is partially recovered in a few hours at HCAL
room temperature. √ √
E resolution 50%/ E ⊕ 100%/ E ⊕
The ATLAS ECAL is segmented into three (two) longi- 5% (45) 5% (50)
tudinal layers for |η| < 2.5 (|η| > 2.5). The granularity of
the ATLAS ECAL in table I refers to its second layer (as granularity 0.1 × 0.1 0.087 × 0.087
most of the electromagnetic energy is deposited there);
the first layer has a finer granularity in η. The multi- TABLE I ATLAS and CMS detectors in the barrel regions.
ple layers allow for a finer granularity than the cell size The granularity is in pseudorapidity and azimuth (η × φ) and
d0 is the transverse impact parameter resolution with respect
in any of the individual layers, being advantageous over to the beam-line. The tracker momentum resolution is from
a laterally segmented calorimeter, and additionally pro- muons while the d0 resolution is from generic charged particles
vide pointing information. The difference between AT- (mostly pions) in tt̄ events. The ECAL energy resolution
LAS and CMS for the HCAL resolution is particularly is presented for electrons. The granularity for the ATLAS
large at higher energies: a 1 TeV jet has σ(E) E ∼ 2% in calorimeters are for the middle layers only, which collect the
σ(E) largest amount of energy. For the ATLAS EM calorimeter,
ATLAS, in contrast to E ∼ 5% in CMS. This is one the innermost layer has ∆η = 0.0031 for γ/π 0 separation.
reason why CMS fully adapted a particle flow technique
since the beginning of the LHC (see section III.A below).

III. JET RECONSTRUCTION ation of particle-flow (54), most of this review will focus
on calorimeter-only jets as they are still the most widely
A. Inputs used setup. ATLAS benefits less than CMS from particle
flow because of its weaker magnetic field and longitudi-
Both experiments have dedicated algorithms to recon- nally segmented calorimeter.
struct particle kinematics from calorimeter and tracker ATLAS and CMS combine calorimeter cells using topo-
information designed to minimize the fake rate, max- logical clusters (53; 55). These clusters are three dimen-
imize the efficiency, and minimize the bias and reso- sional in ATLAS as a result of the longitudinal segmen-
lution of the particle candidate parameters. As there tation. Cluster seeds are started from highly significant
is no algorithm that can simultaneously optimize all of energy (high cell signal to average electronic ⊕ pileup
these objectives, the various approaches trade off op- noise) deposits which are combined (or split) based on
timality under one metric for improvements under an- the distribution of the significance of energy in nearby
other. ATLAS and CMS have also developed different cells. Calorimeter-cell clusters in CMS are obtained us-
algorithms that cater to the experiment’s hardware as ing a Gaussian-mixture model, which results in one or
well as the collaboration’s goals for the tradeoffs. By more calorimeter clusters within each topological cluster.
default, CMS combines tracker and calorimeter informa- HCAL clusters can be split according to the number and
tion into unified particle flow objects as inputs to jet re- energy distribution of associated ECAL clusters. Clus-
construction (51; 52; 53). ATLAS has traditionally used ter splitting is critical to achieve a better estimate of the
calorimeter-only information for jet reconstruction, with spatial energy distribution as input to jet substructure
tracking information used to augment/enhance the per- algorithms (56; 57).
formance. While ATLAS is current migrating to a vari- The topological clusters are calibrated using simula-
tions to account for the non-compensating calorimeter re-
sponse to hadrons, signal losses due to energy deposited
CMS ECAL barrel covers |η| < 1.48, the end-caps extend the in inactive detector material and signal losses on clus-
coverage up to |η| < 3. ter boundaries caused by the topological clustering al-
 7

gorithms. In ATLAS, the calibration scheme relies on a

Energy resolution
classification of clusters as hadronic or electromagnetic
0.6 CMS Anti-kT, R = 0.4 Calo
in origin based on the energy and position of the clus-
ter, the longitudinal depth (λclus ) and normalized signal Simulation |ηRef| < 1.3 PF
energy density; hadronic showers tend to occur deeper
in the calorimeter and be less dense (55). Charged and
neutral pions are used to derive this classification and 0.4
calibration, called the Local Cell Weighting (LCW). In
CMS, dedicated ECAL (based on photons) and HCAL
(based on neutral kaons) calibrations are combined to
account for energy and |η|-dependent non-linearities in
the hadron calorimeter response (53). Both ATLAS and 0.2
CMS validate the performance of these calibrations with
single particle studies in data (53; 58).
Different strategies are used by ATLAS and CMS to
reconstruct tracks from their inner detectors. ATLAS fo- 0
cuses first on maintaining a high efficiency with a rather
20 100 200 1000
inclusive first pass through inner detector hits. A second pRef (GeV)
T
step known as ambiguity solving reduces the fake rate. In
contrast, CMS uses a sequential approach with multiple FIG. 3 Jet energy resolution for particle flow (red, lower line)
passes through the remaining inner detector hits. With and calorimeter-only (blue, upper line) jets in the barrel re-
each pass, the efficiency increases while maintaining a gion in CMS simulation, with no pile-up, as a function of the
pT of the reference jet. Taken from (53).
low fake rate. Both procedures are effective at identi-
fying about 90% of charged pions above 1 GeV with a
percent-level (or smaller) fake rate. Lower momentum
particles can be reconstructed, at the cost of a higher ter envelope in the less granular HCAL. Tracks with a
fake rate and lower efficiency. Due to its weaker mag- pT uncertainty in excess of the calorimetric energy reso-
netic field, ATLAS is able to reach low track momentum lution expected for charged hadrons are masked, which
of 100 MeV for physics analysis (59), although most jet allows the rate of misreconstructed tracks at large pT to
substructure measurements and searches use a threshold be reduced.
of 500 MeV. In contrast, the momentum resolution in
The ECAL and HCAL clusters not linked to any track
CMS is excellent up to higher momenta than in ATLAS.
give rise to photons and neutral hadrons. Charged
The TRT can be used to improve the momentum reso-
hadrons are created from the remaining ECAL and
lution of high pT tracks (60), but the weaker magnetic
HCAL clusters, linked to tracks. If the calibrated calori-
field despite a comparable inner detector radius is a fun-
metric energy is compatible with the corresponding track
damental limitation.
momenta under the charged-pion hypothesis, no neutral
Both experiments have implemented dedicated strate- particles are created. Otherwise, the excess energy is in-
gies for track reconstruction in high density environments terpreted to originate from photons and neutral hadrons
such as the core of high pT jets. In such environments, for deposits in the ECAL and HCAL, respectively. The
pixel and strip clusters can merge resulting in a loss in particle flow algorithm in ATLAS is similar to the one
tracking efficiency and degraded resolution. ATLAS has used by CMS and is described in more detail in Ref. (54).
implemented a stacked neural network (NN) approach to
examine pixel clusters to identify multi-particle clusters, The combination of tracking and calorimetric measure-
estimate the position of the particles passing through ments results in an optimal input for jet substructure
the clusters, and also predict the residual resolution of measurements, making use of the superior angular reso-
the position estimates (61; 62; 63; 64; 65). CMS has lution from the tracking detector and calibrated calorime-
introduced a dedicated tracking step in which a cluster ter clusters. Once the calibrated PF objects are clustered
splitting procedure attempts to split merged clusters ex- into jets, their relative momenta and angular distances
ploiting the information of the jet direction, predicting are kept constant, and only the total energy response
the expected cluster shape and charge (66). of jets is corrected with factorized JES calibrations (see
For particle flow in CMS, tracks and calibrated clusters section III.B).
are combined taking the tracking and calorimeter resolu- The particle flow algorithm improves the energy reso-
tions into account. First, a link is created between tracks lution as shown in figure 3. A similar performance gain is
in the central tracker and calorimeter clusters. Links are observed in ATLAS, but the weaker magnetic field means
also created between clusters in the ECAL and HCAL, that the point where calorimetery and tracking are com-
when the cluster position in the ECAL is within the clus- parable is lower (about 100 GeV).
 8

0.1
Fractional JES uncertainty
19.7 fb-1 (8 TeV)
Data 2015, s = 13 TeV ATLAS 6

JEC uncertainty (%)

anti-k t R = 0.4, EM+JES + in situ
0.08 η = 0.0 CMS Total uncertainty
Total uncertainty
Absolute in situ JES Excl. flavor, time
Relative in situ JES 5 Absolute scale
Flav. composition, unknown composition
0.06
Flav. response, unknown composition
R=0.5 PF+CHS Relative scale
Pile-up, average 2015 conditions |η | = 0 Pileup (〈µ〉=20)
Punch-through, average 2015 conditions 4 jet
0.04 Jet flavor (QCD)
Time stability
0.02 3

0 2
20 30 40 102 2×102 103 2×103
jet
p [GeV]
T

1
FIG. 4 ATLAS jet energy scale uncertainty. Adapted
from (68).
0
20 100 200 1000
p (GeV)
B. Calibration T

FIG. 5 CMS jet energy scale uncertainty, from (44). JEC

The ratio of the measured energy Ereco to the deposited means Jet Energy Correction, which has the same meaning
energy Etrue is the jet energy response which depends on as JES.
the energy, pseudorapidity and other features of the jet.
Due to the properties of tracking detectors and calorime-
ters, the average response is not unity. For example observables are constructed, but dedicated corrections
calorimeter jets in ATLAS with Etrue = 30 GeV may are derived as described in section V.B. Figure 5 shows
have responses below 0.3, while jets of higher energies the calibrated JES uncertainty obtained in CMS, which
may have responses above 0.8. For this reason, the Jet is below 1% for jets with pT > 100 GeV in the central
Energy Scale (JES) is calculated in bins of the particle- region with η = 0. Even for jet pT as low as 10 GeV the
level jet energy Etrue and ηdet as the mean of a Gaussian uncertainty is below 3%, owing to the excellent perfor-
fit to the response distribution and a numerical inversion mance of the particle flow reconstruction.
procedure is used to derive calibration factors in bins of A detailed discussion of the different approaches for de-
the reconstructed jet energy from Etrue (67; 68; 69; 70). riving jet energy scale uncertainties in ATLAS and CMS
can be found in Ref. (72).
In ATLAS, the calibration of the JES is undertaken in
several stages, starting from jets either at the electromag-
netic (EM) or LCW (built from calibrated inputs) scale. IV. PILE-UP MITIGATION
Using calibrated inputs bring the JES to within 10% of
unity for E = 30 GeV and |η| < 0.3 (67). The Global A. Definition
Sequential Calibration (68; 71) was introduced for Run 2
and reduces the sensitivity to differences in the responses Pile-up originates from simultaneous proton-proton
of quark versus gluon-initiated jets (quark/gluon separa- (pp) collisions that occur in addition to a hard scatter-
tion is also discussed in section VI.A). This additional ing collision of interest. The hard scattering event of
calibration results in a significant jet pT resolution im- interest is referred to as the Primary Vertex (PV). Pile-
provement of up to 35% depending on the pT and η of up is uncorrelated with the PV and typically consists of
the jet (71). The JES uncertainty varies between 1-6% an admixture of inelastic, elastic and diffractive pp pro-
in the central region with η = 0 as shown in figure 4 (68). cesses which are separated in the longitudinal direction.
In CMS, jets are clustered from calibrated particle As the detector response is not instantaneous, pile-up
flow objects, thus the uncalibrated JES is within 6% of events from both the same (in-time) and neighboring
the expected value of 1 for central jets with η < 0.7 (out-of-time) bunch crossings can contribute. This re-
and pT > 30 GeV (44). To account for deviations from view focuses on the mitigation of in-time pile-up, though
unity, factorized JES calibrations are applied in multi- out-of-time pile-up is also mitigated to differing degrees
ple stages (72) including pile-up corrections, simulation- due to the specifics of the ATLAS and CMS detector
based response corrections and small residual corrections technologies and reconstruction algorithms.
for tracking inefficiencies and threshold effects, derived During the LHC Run 1 the mean number of pile-up
in-situ from γ+jet, Z+jet and dijet samples (69). This interactions reached hµi = 21, and µ values up to 60 were
additional correction is not used when jet substructure attained in certain runs of 2017 (Run 2) with possibly
 9

the whole jet, the individual particles within the jet can
I
19.7 fb−1 (8 TeV) be classified to whether they belong to the actual jet or to
ρ (GeV) 30 the underlying pile-up. Charged particles leave tracks in
CMS high granularity tracking detectors at the heart of multi-
25 purpose detectors like ATLAS and CMS and can be sep-
ρ(µ) = 1.00 + 0.54 µ + 0.0005 µ2 arated based on their longitudinal position ẑ (along the
ρ(µ) = 1.01 + 0.50 µ + 0.0007 µ2
beamline) within the luminous region (see figure 7). The
20 charged hadron subtraction (CHS) (74) method identi-
fies each pile-up track individually. Used in concert with
15 particle flow concepts which attempt to identify each par-
ticle in the event uniquely, CHS can effectively remove
all charged pile-up radiation from the event, including
10 calorimeter signals that are linked to tracks through the
particle flow algorithm. Identification of pile-up jets,
formed predominantly from the energy of one or many
5
Data pile-up vertices, is another technique for removing pile-
MC up using charged particles; by determining the fraction
0 of energy of the jet from the primary vertex, one can
0 10 20 30 40 distinguish such pile-up jets from the PV jets (75; 77).
µ The two methods discussed above can be combined.
First the more precise CHS method subtracts the pile-
FIG. 6 Average pile-up contribution to the jet pT , ρ, as a up contribution from charged particles; in a second step,
function of the average number of pile-up interactions per the remaining contributions from neutral particles are
bunch crossing, µ, for data (circles) and simulation (dia-
monds) at the CMS experiment. Taken from (44).
removed with the area subtraction method.
In a more advanced approach, local, topological infor-
mation is used, as QCD radiation from pile-up vertices is
even higher values in Run 3, and culminating at the high often uncorrelated and soft. It and can thus be removed
luminosity LHC (HL-LHC) reaching up to hµi = 140 − based on the local energy profile, i.e. if the radiation is
200. not consistent with hard scattering radiation from the
Pile-up typically leaves about 0.5 GeV of energy in the PV. This can be done in the transverse plane η, φ and
detector per unit area (η, φ), per pile-up vertex; the ef- also as a function of radiation depth. The jet grooming
fects of this are present in all aspects of LHC physics, technique is such an example to clean the jet of soft and
from detector design and software performance to the fi- wide-angle radiation which incidentally removes pile-up
nal sensitivity of measurements and searches. radiation. It is discussed in more detail in section V.A.
Topoclustering (55), used by the ATLAS Collaboration,
is deployed at the formation of clusters in the calorimeter
B. Mitigation Methods requiring radiation to have a certain topological profile.
In the forward region, where no tracking information is
Properties of pile-up interactions are exploited to dis- available, jet shapes and topological correlations can be
criminate pile-up particles from particles originating from used to identify pile-up (78).
the primary vertex, or to remove energy contributions While the above methods have been successfully de-
from pile-up to the individual jet. ployed in the LHC experiments, they each have some
Pile-up can be approximated as a spatially uniform de- deficiencies as well; ideally, one would hope to effectively
position of energy. The so-called area subtraction uses a combine all pile-up mitigation handles in order to maxi-
pile-up pT density per unit area estimator, ρ, and defines mally distinguish pile-up from PV radiation and to re-
a jet catchment area, A, to remove energy that is assumed move pile-up at the most granular level possible, i.e.
to originate from pile-up interaction. This approach cor- at the particle or constituent level, in order to be as
rects the jet in the following way: pcorrT = porig
T − ρA. generic as possible. For example, while area subtrac-
An example of ρ is shown in figure 6. There are many tion is very effective for correcting the jet pT , it is not
subtleties in defining both ρ and A, which are discussed used to mitigate the pile-up dependence of jet substruc-
in e.g. Refs. (73; 74; 75). An extension to this method is ture observables as it is only able to correctly remove
shape subtraction (76), where randomly distributed ghost pile-up contributions on average. In fact, jet substruc-
particles are used to calculate a jet shape’s sensitivity to ture variables are among the most difficult to correct for
pile-up, which can then be corrected for non-uniformities pile-up because they are so reliant on radiation profiles.
in the spatial distribution of pile-up particles. A number of hybrid methods have been proposed op-
Instead of a global, collective, treatment of pile-up for erating at the event constituent level. One example is
 10

CMS Simulation Preliminary 13 TeV

arbitrary units
Pythia QCD PF+PUPPI
35000 <∆m>=-0.6 GeV
Anti-kT (R=0.8) RMS=10.9 GeV
<nPU> = 40
30000 PF
200 GeV < p < 600 GeV <∆m>=13.7 GeV
T
|η| < 2.5 RMS=17.9 GeV
25000 PF+CHS
<∆m>=-6.2 GeV
RMS=14.6 GeV
20000
PF(Cleansing)
<∆m>=-0.8 GeV
RMS=12.7 GeV
15000
PF+CHS([Link].)
<∆m>=0.5 GeV
10000 RMS=13.7 GeV

5000

0
-100 -80 -60 -40 -20 0 20 40 60 80 100
mreco - mgen(GeV)

FIG. 8 Comparison of different pile-up removal algorithms

for the leading ungroomed jet mass response in simulated
QCD multijet events with CMS. Taken from (74).

FIG. 7 H → 2e2µ candidate event with 25 additional recon-

structed vertices recorded in 2016. Taken from (79).
ber of reconstructed vertices and µ are observed even
with area subtraction methods for the pile-up levels cur-
the PUPPI (80) algorithm which is extensively used in rently observed at the LHC, hµi ∼ 25. To correct
CMS. The PUPPI algorithm uses both event energy den- for these effects, an additional residual correction is ap-
sity and local topological information incorporated in an plied (44; 68). Enhancements are also possible from com-
event-by-event particle-level discriminator to determine bining area subtraction methods with e.g. CHS.
if a particle is from pile-up. The algorithm defines a
shape which attempts to distinguish parton shower-like For jet substructure observables, particle- or
radiation from pile-up-like radiation. The shape is cal- constituent-level pile-up mitigation strategies have
culated from pT , angular distance to nearby particles, been shown to improve performance, especially in
and other information. Particle four-vectors are then simulation studies for up to hµi ∼ 40. An example
weighted proportional to the value of the discriminator is given in figure 8, where the ungroomed jet mass of
value. Ideally, particles from the hard scatter would get a the leading jet in pT in simulated QCD multijet events
weight of one and pile-up particles would get a weight of is corrected with different pile-up removal techniques.
zero. Almost all pile-up particles have values within a few The jet mass resolution can be improved further when
standard deviations of the median and are assigned small using a grooming algorithm. The effect of different
weights. Values that deviate far from the charged pile-up pile-up removal techniques on the groomed jet mass
are indicative of a hard scatter, and these particles are depends however strongly on the choice of the grooming
assigned large weights. This weighting method allows algorithm as discussed in detail in Refs. (74; 86). The
for experimental information, such as tracking, vertexing improved performance observed in simulation has also
and timing information, to be included. been verified in collision data (87).
Other examples of such hybrid methods are Con-
stituent Subtraction (80; 81; 82), SoftKiller (81) and Generally these techniques, particularly those which
PUMML (83). Precursor hybrid methods include jets operate at particle-level, can also be used to improve per-
without jets (84) and jet cleansing (85). formance of non-jet objects such as missing transverse
energy and lepton isolation. In the latter case, where
the energy in a small cone around the lepton is summed,
C. Performance Studies pile-up mitigation techniques help to reduce the isola-
tion’s susceptibility to pile-up.
Pile-up removal algorithms are commissioned for use
in ATLAS and CMS via detailed studies of jet observ- Preliminary studies (detector configurations have not
ables in terms of the resolution and absolute scale, pile-up yet been finalized) into the application of these advanced
dependence, and the background rejection versus signal hybrid techniques at the higher pile-up levels anticipated
efficiency for boosted heavy particle taggers. at the HL-LHC suggest that they are effective in the
For observables like jet pT , dependencies on the num- hµi = 140 − 200 range (88; 89).
 11

V. JET SUBSTRUCTURE METHODS AND ATLAS performed a broad study of the relative per-
OBSERVABLES formance of different grooming techniques for boson-
tagging (86; 94; 95), top-tagging (96; 97) and SM
A. Jet Grooming measurements (98; 99), using the removal of pile-up-
dependence, the jet mass resolution, and the tagging effi-
Jet grooming techniques have seen a particularly high ciency versus background rejection as performance met-
level of interest from the experimental and theoretical rics. The ‘standard’ grooming procedure adopted by
communities alike. Jet grooming is an additional ‘post- ATLAS is trimming with fcut = 0.05 for boson tagging
processing’ treatment of large radius jets, an extra step in both Run 1 (Rsub = 0.3) and Run 2 (Rsub = 0.2).
used to remove unwanted soft radiation and to allow The trimming algorithm with the same parameters was
the underlying hard substructure associated with a two- adopted for top tagging, along with several other tech-
prong (e.g. W boson) or three-prong (e.g. top quark) niques (see section VI.C). Another technique currently in
decay to be identified more efficiently. use by ATLAS is the reclustering of small-R jets (100),
In particular, grooming is the systematic removal of which uses fully-calibrated anti-kT , R = 0.4 jets as inputs
radiation from within a jet, often targeting soft and wide to the anti-kT algorithm with a larger distance parameter
angle radiation. There are a variety of techniques and (typically R = 1.0). This has proven a popular method
each one has tunable parameters which are chosen to in ATLAS analyses due to the flexibility of optimizing
suite the particular needs of the application. The three the jet distance parameter depending on the considered
main algorithms used by ATLAS and CMS are trim- phase-space of the analysis (101; 102; 103). A recent
ming (90), pruning (91), and soft drop (92). In each study of in-situ measurements (104) (including ‘closeby’
of these cases, the constituents of a jet are re-clustered effects) confirm that the differences between data and
and soft/wide angle radiation is rejected in this process. simulation observed with reclustered jets are indeed cov-
For trimming, the kT algorithm is used to re-cluster and ered by simply propagating the uncertainties associated
the radius parameter of the re-clustering is called Rsub . with the input anti-kT , R = 0.4 jets.
Those smaller-radius jets with a momentum fraction CMS studied a large number of grooming techniques
f < fcut are removed to produce the trimmed jet. The in the context of boosted boson-tagging (57; 105), top-
two other algorithms impose a condition on each 2 → 1 tagging (106; 107) and SM measurements (108; 109).
clustering step, by going backwards in the sequence in During Run 1 the grooming techniques were used to-
which the particles were combined in the re-clustering. gether with charged-hadron subtraction for pile-up mit-
The transverse momentum fraction of the softer particle igation (see section IV). All groomers studied showed
to the merged system, z = min(pT,1 , pT,2 )/(pT,1 + pT,2 ), reasonable or good agreement between data and simu-
is a natural choice for determining the scale of the soft lation and the pruning algorithm (R = 0.8, zp = 0.1
radiation, and the angular distance ∆R between the two and dp = 0.5) showed the best performance for boson
particles for identifying wide-angle radiation. The differ- tagging (105). For Run 2, soft drop (zcut = 0.1 and β
ence between pruning and soft drop lies in the way how = 0) is used for jets with R = 0.8 in jet substructure
particles and their combinations get rejected based on analyses in CMS together with the pile-up removal al-
the values of z and ∆R. For pruning, the softer particle gorithm PUPPI (80) (see section IV). Soft drop jets in
of the 2 → 1 clustering step is discarded if z < zp and combination with PUPPI show a similar performance as
∆R < dp . For soft drop, the softer particle is discarded if pruning when comparing signal efficiency versus back-
z < zcut (∆R/R)β , where zcut and the angular exponent ground rejection (87; 107), but allow for better theoret-
β are free parameters6 . ical control. While grooming techniques were found to
The role of grooming has traditionally satisfied two improve the performance (higher background rejection
purposes in ATLAS, being the mitigation of pile-up ef- at fixed signal efficiency) of the jet mass, N -subjettiness
fects on jets, and the removal of soft/wide-angle radia- ratios (110; 111) were found to perform better without
tion. The particle flow algorithm employed in CMS in grooming for boosted boson tagging (105). For top-
conjunction with CHS or PUPPI allows for a correction tagging applications, however, soft drop groomed N -
for pile-up effects. This reduces the usefulness of groom- subjettiness ratios improved the performance with re-
ing for pile-up mitigation, but retains its advantage for spect to ungroomed ones for jets with pT < 400 GeV.
the removal of soft/wide-angle radiation. For higher pT jets there was no significant gain observed
with grooming for N -subjettiness ratios (107).

6 Most applications of soft drop use β = 0, in which case it is

equivalent to an earlier algorithm known as modified mass drop
B. Jet Mass
tagger (mMDT) (93). Since both collaborations call this soft
drop, we also refer to the algorithm by this name, but encourage
the users to cite the mMDT publication in addition to the soft The reconstruction of jet energies mainly relies on the
drop one. capability of a detector to measure the total energy of
 12

all particles deposited in the detector; however, the mea-

Events / 5 GeV
surement of jet mass requires detection of the deposited ATLAS Preliminary
4000
energy with a granularity that is finer than the size of a s = 8 TeV, 20.3 fb-1
jet. The mass of a jet can only be estimated if the en- p > 200 GeV, |η| < 2.0
T

ergy is deposited in at least two detector elements, as it 3000 2012 Data

depends on both the energy and opening angle between Detector-level Simulation

the jet constituents. For jet substructure techniques that Fitted Simulation
Particle-level Simulation
rely on the rejection of soft particles, it is also important 2000
to be able to reconstruct particles with low pT separately
from harder particles in a jet.
1000
The jet mass response distribution Rreco is constructed
from the calibrated, reconstructed jet mass Mreco divided
by the particle-level jet mass Mtrue . The mass response 0
distribution is calculated in bins of reconstructed jet 1.1

Data/MC
Detector-level Simulation
Fitted Simulation
pT,reco and ηreco . In ATLAS, the Jet Mass Scale (JMS) is
1
defined as the mean of this response distribution. The Jet
Mass Resolution (JMR) is then defined as half the 68% 0.9
0 50 100 150 200
interquartile range (IQnR) of the response distribution, Jet Mass [GeV]
as
FIG. 9 The trimmed jet mass before (detector-level) and after
(fitted) determining s and r. The particle-level distribution
r = 0.5 × 68% IQnR(Rreco ). (2) is shown for comparison. Jets are required to have pT > 200
GeV. Adapted from Ref. (113).
This is robust to large non-Gaussian tails but, if the dis-
tribution is Gaussian, is equal to its 1σ width. The frac-
tional JMR is expressed as the JMR divided by the me-
dian of the response distribution. uncertainties are due to the theoretical modeling of jet
ATLAS has recently developed a data-driven approach fragmentation and the cluster energy scale.
to extract the JMS and JMR from an enriched sample As the forward-folding method is currently restricted
of boosted tt events, however the method can also be to jets with pT < 350 and 500 GeV for boosted W
extended to other final states. This forward-folding ap- bosons and top quarks, respectively, the results are com-
proach folds the particle-level mass spectra by a modified bined with the so-called Rtrk method which constrains
response function such that the JMS in a given bin of the mass scale by comparing the calorimeter jet mass
particle-level jet mass and reconstructed jet pT is scaled to the mass calculated from track jets and extends up
by the scale parameter s and the JMR is scaled by the to pT = 3000 GeV (114). The Rtrk method can also be
resolution parameter r: generalized to other variables and is used in ATLAS to
m,pT
constrain the pT scale of large-R jets as well as to derive
Mfold = s × Mreco + (Mreco − hMreco i)(r − s)). (3) systematic uncertainties on jet substructure variables.
The values of r and s for which the Mfold distribution best The concept of a Track-Assisted Mass for trimmed,
matches the data are extracted from a 2 dimensional χ2 large-R jets has been studied in ATLAS (112) to main-
fit as shown in figure 9 and detailed in Ref. (112; 113). tain performance for highly boosted particles due to the
With the forward-folding approach, the JMS and JMR limited granularity of the calorimeter. The track-assisted
for hadronically decaying boosted W bosons with pT & mass is defined as:
200 GeV are determined with 2–3% and 20% system-
atic uncertainties, respectively (see figure 10). As the pcalo
jet mass and its detector-response depend on kinematics mTA = T
× mtrack , (4)
ptrack
T
and jet substructure, the measurement was repeated dif-
ferentially with an increased luminosity for boosted W
and top quarks in Ref. (114). It will be important to where pcalo
T is the transverse momentum of the calorime-
extend the technique to other final states in the future. ter jet, ptrack
T is the transverse momentum of the four-
This may require hybrid data/simulation methods. A vector sum of tracks associated to the calorimeter jet,
detailed study of the various contributions to the JMS and mtrack is the invariant mass of this four-vector sum,
and JMR has been performed in context of the soft drop where the track mass is set to the pion mass mπ . The
mass measurement (98), described in section VII.A.1, by track-assisted mass exploits the excellent angular resolu-
propagating experimental uncertainties on the inputs to tion of the tracking detector and the ratio pcalo
T to ptrack
T
the jet reconstruction to the jet mass. The dominating corrects for charged-to-neutral fluctuations. The Com-
 13

0.3

Fractional jet mass resolution

Relative Jet Mass Resolution ATLAS Simulation Preliminary
ATLAS Preliminary s = 13 TeV, WZ → qqqq
1.4 anti-kt R = 1.0 jets, |η| < 2.0
p > 200 GeV, |η| < 2.0
0.25
T Trimmed (f = 0.05, Rsub = 0.2)
cut
LCW + JES + JMS calibrated

1.2 0.2

0.15
1
0.1 Calorimeter mass
Track assisted mass
Combined mass
0.8
0.05
500 1000 1500 2000 2500
8 TeV, 20.3 fb -1 Data 1σ stat. uncertainty Truth jet p [GeV]
T
0.6 13 TeV, 3.2 fb -1 Data stat ⊕ syst. uncertainty
FIG. 11 The ATLAS combined jet mass resolution. Adapted
0.8 0.9 1 1.1 1.2 from (112).
Relative Jet Mass Scale

2.3 fb-1 (13 TeV)

FIG. 10 The fitted values of the relative jet mass scale (s)

Events / (5 GeV)
300
and resolution (r) for trimmed anti-kT , R = 1.0 calorimeter CMS CMS data tt (unmerged)
jets from the 2012 and 2015 ATLAS datasets√ and the 1σ sta- Preliminary Data fit Single top
tistical and total uncertainty ellipses. The s = 8 and 13 250 MC fit W+jets
TeV selections are similar, although the trimming definition
slightly changed between Runs (Rsub = 0.3 to Rsub = 0.2). tt (merged) WW/WZ/ZZ
Adapted from Ref. (112). 200

bined Mass is defined as: 150

 −2   −2 
σcalo σTA
mcomb = −2 −2 m calo
+ −2 −2 mTA , 100
σcalo + σTA σTA + σcalo
(5)
50
where σcalo and σTA are the calorimeter-based jet mass
resolution and the track-assisted mass resolution, respec- 0
tively. The jet mass resolution for the calorimeter mass, 40 50 60 70 80 90 100 110 120 130
track-assisted mass and combined mass are shown in fig- PUPPI softdrop jet mass (GeV)
ure 11 for W/Z boson jets as a function of jet pT . Similar
techniques that take advantage of the excellent angular FIG. 12 Jet mass distribution in a sample enriched with
resolution of the tracking detector at high pT have been lepton+jets tt events, where the hadronic W jet with pT >
200 GeV is selected, taken from Ref. (87).
developed to correct topoclusters to improve the resolu-
tion of jet substructure variables (56).
It is important to point out that in ATLAS unlike in
CMS, the jet energy scale directly impacts the jet mass jet four-vector is corrected using JES corrections and
scale. As opposed to the description of the JES calibra- small residual differences in the jet mass between data
tion for small-R jets in section III.B, the area subtrac- and simulation are corrected using dedicated samples.
tion, residual correction and Global Sequential Calibra- The residual in-situ jet energy corrections are not ap-
tion (GSC) (see section VI.A) are not applied to large-R plied when reconstructing jet masses. Therefore, dedi-
jets. cated corrections are derived from simulation and data.
In CMS, the jet mass is by default reconstructed as a Firstly, the jet mass response is corrected as a func-
combination of track and calorimeter measurements via tion of pT and η using simulation of W jets from bo-
the virtues of the particle flow algorithm. Thus the strat- son pair production. Secondly, residual corrections are
egy for calibrating the jet mass in CMS differs from the obtained from a data sample enriched in lepton+jets tt̄
one in ATLAS. In CMS, the individual PF objects are in- production where the hadronic W jet can be studied
put to the jet reconstruction, and are locally calibrated in data (87; 105). The selection is optimized for fully-
to account for the detector’s single particle response (see merged hadronic W decays. Large-R jets in this sample
section III.A). After correcting the individual inputs, the show a peak at the W mass in the jet mass distribution,
 14

CMS Simulation 13 TeV one for the combined mass in ATLAS (figure 11), even
JMR 0.3
though quark/gluon jets are compared with W /Z-jets
0.25 p bins and very different technologies are used to reconstruct
T
260-350 GeV the jet mass.
0.2 650-760 GeV
0.15 900-1000 GeV
1200-1300 GeV C. Other Jet Substructure Observables
0.1

0.05 Additional jet substructure observables are used for a

variety of purposes, often to complement the jet mass.
0
30 40 50 100 200 300 1000 Most uses of these observables are within the context of
Ungroomed jet mass mu (GeV)
a dedicated tagger, described in the next section. These
observables can generally be classified into two categories:
FIG. 13 The CMS jet mass resolution as a function of the un- prong-taggers and haze-taggers. The most widely used
groomed jet mass mu in different generated pT bins. Adapted β
prong-taggers are the N -subjettiness ratios τij (110; 111),
from (109). β β β
C2 (117), D2 (118; 119), and N2 (120). The latter three
are ratios of energy correlation functions, which are sums
over constituents inside jets weighted by the momentum
as shown in figure 12 for the soft drop grooming case. fractions and pairwise opening angles to the power β. For
The excellent performance of the PF algorithm results example,
in a JMR of about 10%. The absolute response and the
resolution are well described by the simulation, within (β)
2 e3
1–2% for the JMS and about 10% for the JMR, which is N2 = , (6)
about the same size as the statistical uncertainty of this (1 eβ2 )2
measurement. Residual differences in this distribution where
are used to calibrate the JMS and JMR in simulation, n o
and can also be used for dedicated efficiency corrections (β)
X β β β
1 e2 = zi zj zk min ∆Rij , ∆Rik , ∆Rjk
on other jet substructure observables, such as the N - 1≤i<j<k≤nJ
subjettiness ratio τ21 = τ2 /τ1 . (7)
Since these measurements are performed in samples of (β)
X
2 e3 = zi zj zk
W jets with pT ≈ 200 GeV, additional systematic un-
1≤i<j<k≤nJ
certainties apply at higher pT (115). A detailed study n o
β β β β β β
of the various contributions to the JMS has also been × min ∆Rij ∆Rik , ∆Rij ∆Rjk , ∆Rik ∆Rjk , (8)
performed for fully merged top-jets in the context of an
unfolded top-jet mass measurement (116). To summa- where the sums run over the nJ jet constituents with
rize the impact of the various sources of systematic un- momentum fractions zi and opening angles ∆Rij .
certainty to the measurement of residual corrections for The goal of haze-taggers is to generally characterize
jet substructure observables, we quote here the domi- the radiation pattern within a jet without explicitly iden-
nant uncertainties related to the scale factor measure- tifying the number of prongs. The prong-taggers also
ment of an N -subjettiness ratio τ21 < 0.4 selection (87). are sensitive to the distribution of radiation around the
The statistical uncertainty of 6% (with 2.3/fb of data) is subjet axes and so the distinction is not strict. Popular
comparable to the systematic uncertainties related to the haze-taggers include jet width, nconstituents (or ntracks ),
simulation of the tt̄ topology (nearby jets, pT spectrum) and pD T.
contributing 4%, the choice of method to derive the scale In applications of jet substructure taggers based on
factors contributing 6% and the modeling of the pT de- these variables the description of data by simulation is
pendence that rises from 5% at pT = 500 GeV to 13% at a crucial aspect. Differences in the distributions lead
pT = 2000 GeV. to differences in efficiencies and misidentification rates,
The relative JMR in CMS is shown in figure 13 as which need to be quantified in dedicated measurements.
a function of the ungroomed jet mass mu for anti-kT , Measurements of jet substructure observables, their cal-
R = 0.8 jets. The JMR is obtained from a sample of jets ibration, and improving their description by adjusting
initiated by quarks and gluons. The resolution improves free parameters in event generators is an important step
with increasing mu and is around 9–13% for the most in every analysis.
probable value of mu ≈ 100–150 GeV. For a given value As an example for three-prong taggers, the N -
of mu < 200 GeV, the resolution worsens with increasing subjettiness ratio τ32 = τ3 /τ2 for β = 1 is shown here.
jet pT due to a higher degree of collimation. Remark- It is used in ATLAS and CMS for top tagging and stud-
ably, the resolution obtained in CMS is comparable to the ied in light quark and gluon jets from dijet production,
 15

×106 CMS Preliminary 35.9 fb-1 (13 TeV)

Events
Events / 0.02 1.2 ATLAS Preliminary
Data 2015+2016 2500 Data
-1
Pythia8 dijet (× 0.71) s = 13 TeV, 36.5 fb Merged top t t
1 Herwig++ dijet (× 1.32) Trimmed anti-k t R =1.0 Merged W t t
Dijet Selection
2000
W+jets (× 50) Merged QB tt
0.8 Z+jets (× 50)
pT > 450 GeV Not merged tt
mcomb > 100 GeV 1500 W+Jets
all-had tt (× 50)
Single t
0.6 Stat. uncert.
DY+Jets
Stat. ⊕ syst. uncert. 1000 QCD
0.4
AK8 PUPPI
500 p > 400 GeV
T
0.2

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

DATA / MC
1.5 1.5
MC(sig. + bkg.)

1.25
1
Data

1
0.75 0.5

0.50
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Probe jet τ3/ τ2
Leading Large-R Jet τ WTA
32

FIG. 15 Measured distribution of the N -subjettiness ratio

FIG. 14 Measured distribution of the N -subjettiness ratio τ32
τ32 calculated on anti-kT , R = 0.8 jets with pT > 400 GeV
calculated on trimmed anti-kT , R = 1.0 jets for a dijet selec-
corrected with PUPPI in a tt sample. The data are compared
tion with pT > 450 GeV and pT > 200 GeV for the leading and
to simulated events, where the “Merged QB” tt contribution
sub-leading jet, respectively. The data are compared to sim-
consists of events in which the b quark from the top quark
ulated events, where the dijet samples have been normalized
decay and just one of the quarks from the W boson decay are
to the signal-subtracted data. Taken from Ref. (122).
clustered into the jet. Taken from Ref. (121).

as well as in fully-merged top-quark jets from dedicated

been developed by the theoretical community that can be
tt samples. The distribution of τ32 with Run 2 data is
used along with the jet mass for jet classification. The
shown in figure 14 for a dijet selection and in figure 15 for
term ‘tagger’ indicates the use of one or more of these
a tt selection. Overall good agreement between data and
variables (sometimes after grooming has been applied)
simulation is observed, which leads to data-to-simulation
to discriminate between jets coming from different types
scale factors for top-tagging compatible with unity (121).
of particles.
A rule of thumb for the decay of a massive object such
As an example for an haze-tagger distribution, the pD T as a W/Z/H boson is that the decay products lie within a
distribution is shown in Fig. 16. The distribution from
cone of radius ∆R = 2M/pT in the laboratory rest frame,
Z+jets production is well described by simulation, but
where M and pT are the mass and transverse momentum
a significant discrepancy is observed when selecting dijet
of the object7 . Using this for the example of a W boson
events. This has important consequences for quark/gluon
decay, a W boson with pT = 200 GeV will have its decay
tagging, where dedicated template fits to data are per-
products captured by a jet with a distance parameter of
formed to extract weights to correct the simulation (see
at least 0.8, and the higher the pT of the W boson, the
section VI.A). Similar conclusions are found for the jet
more collimated the decay products. For top quarks, the
width and constituent multiplicity distributions (87).
value of pT for which all decay products are captured by
a jet with R = 0.8 is at least 400 GeV.
VI. JET TAGGING
A. Quark/Gluon Discrimination
Particle identification is an experimental challenge
that is traditionally met using custom-designed charged- Since the first algorithmic definitions of jets, jet sub-
particle detectors, muon chambers and calorimeters with structure observables have been widely used for quark-
granularity fine enough to allow shower shape measure-
ments. Particle identification played an important role
in the design considerations for the ATLAS and CMS
detectors. Jet substructure techniques used for the iden- 7 Note that this rule of thumb gives only a lower bound on ∆R,
tification of the particle origin of jets are a recent de- and it strictly holds only for two-body decays with massless decay
velopment, though. Several substructure variables have products and pT  M .
 16

2.6 fb-1 (13 TeV) 0.25

Normalized Entries
ATLAS Simulation Preliminary
Events/0.02
400
CMS Z+jets, Pythia8 s = 13 TeV
350 Preliminary
Data quark
0.20 Anti k EM+JES R=0.4
| | < 2.1
300
gluon undefined Quark Jet
0.15 Gluon Jet
250 80 GeV < p < 100 GeV 50 < pT < 100 GeV
T
| η| < 2.0 400 < pT < 500 GeV
200 0.10 1200 < pT < 1500 GeV
150

0.05
100

50
0.000 10 20 30 40 50 60
0

2
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ntrack
1.8
1.6
1.4 FIG. 17 The distribution of the number of tracks inside jets
DATA
MC

1.2
1
0.8 for quark and gluon jets in multiple jet pT ranges. Repro-
0.6
0.4
0.2
duced from Ref. (135).
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
pD
T

history (124; 125) to using entirely observable phase-

space regions (126; 127); however, no treatment escapes
FIG. 16 Distribution of pD
T calculated on anti-kT , R = 0.4
jets with 80 < pT < 100 GeV a Z+jets sample. Taken from the problem that the notion of a quark and gluon jet is
Ref. (87). not universal8 : quark and gluon jet radiation depends on
the production mechanism. This means that the calibra-
tion and application of q/g taggers must be treated with
initiated (quark) versus gluon-initiated (gluon) jet tag- additional care compared with more universal classifica-
ging. Most measurements and searches at the LHC target tion tasks such as b tagging.
a final state with a particular partonic structure and the There is a plethora of jet substructure observables that
dominant backgrounds may have a different flavor com- can be used for q/g tagging; see e.g. Ref. (130) for a large
position. Therefore, tagging jets as quark or gluon could survey. Many of these observables exhibit Casimir scal-
increase the analysis sensitivity. For example, jets pro- ing which results in nearly the same, limited discrimina-
duced in vector-boson scattering/fusion (VBF/VBS) are tion power for all the observables (117; 131). The most
quark jets, while many of the background jets are gluon powerful single q/g observable is the particle multiplicity
jets. There are many other applications, ranging from inside a jet (shown in Fig. 17), which does not exhibit
high multiplicity supersymmetry searches, initial state Casimir scaling and recent theoretical advances (132)
jet tagging, etc. have shown that its discrimination power can be largely
The probability for a gluon to radiate a gluon is en- understood from perturbative theory. There is further
hanced by a factor of CA /CF = 9/4 ∼ 2 over the prob- q/g separation possible when using the full radiation pat-
ability for a quark to radiate a gluon of the same en- tern inside a jet, though the combination of multiplicity
ergy fraction and opening angle (123). As a result, gluon and a Casimir scaling observable carries a significant frac-
jets tend to have more constituents and a broader radia- tion of the total discrimination power (133). The mod-
tion pattern than quark jets. There are also more subtle eling of q/g tagging observables has a long history - see
differences due to quark and gluon electric charges and Ref. (134) for a recent and detailed study.
spins. Despite the challenges listed above, both ATLAS and
CMS extensively use explicit or implicit quark versus
There are three key challenges of quark versus gluon
gluon tagging. Explicit taggers are algorithms designed
jet (q/g) tagging: (1) quark and gluon labeling schemes
to directly isolate quark and gluon jets while implicit
are not unique; (2) for a given labeling scheme, quark
techniques are designed for another purpose that also
and gluon jets are not that different; (3) the differ-
happens to perform some quark versus gluon jet tagging.
ences that do exist are sensitive to both perturbative
The explicit taggers developed by ATLAS (135; 136; 137;
and non-perturbative modeling choices. Since quarks
138) and CMS (87; 139; 140; 141) include a variety of
and gluons carry color charge and only colorless hadrons
observables and data-driven calibration and validation
are observed, there is not a unique way to label a jet
in simulation as originating from a quark or a gluon.
Many labeling conventions exist, ranging in simplicity
8
and model-dependence from matching to out-going ma- This can be mitigated by jet grooming; see e.g. Ref. (128). Also,
trix element partons to parsing an entire jet clustering the non-universality may be ‘small’ in practice (129).
 17

techniques. These and related techniques have been suc-

cessfully deployed in a variety of physics analyses (see
(13 TeV)
1

quark-jet efficiency
e.g. (57; 142; 143; 144; 145; 146; 147)). Additionally, it CMS
0.9 Simulation Preliminary
has been shown that an improved W tagger can be con-
dijets, Pythia8
structed by utilizing q/g discrimination on subjets (57). 0.8
Both ATLAS and CMS have developed likelihood-
0.7
based discriminants for explicit q/g tagging. The dis-
criminants are constructed from variables sensitive to the 0.6
radiation pattern of quark and gluon jets, also taking into 0.5
account differences between light (uds) and heavy flavor
(cb) quark jets, where the latter are more similar to gluon 0.4 0 < |η| < 1.3, 80 < p < 100 GeV
T
jets. ATLAS uses the number of tracks ntrk as an approx- 0.3 0 < |η| < 1.3, 40 < p < 50 GeV
T
imation for the number of jet constituents and the jet 2.7 < |η| < 3, 40 < p < 50 GeV
0.2 T
width (138) while CMS utilizes the number of particle-
flow constituents nconst , the jet axes and fragmentation 0.1
functions (87). Since the distributions of these variables
0
depend on η, pT , and ρ, the likelihood discriminators are 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
constructed differentially with respect to these variables. gluon-jet rejection
In Run 2, ATLAS also introduced a simple and robust
FIG. 18 The CMS q/g tagging performance in simulation for
tagger using solely ntrack (135), which has the advantage
two bins in jet pT and two bins in jet |η|. Reproduced from
of a much-simplified uncertainty derivation. Ref. (87).
Figure 18 shows the CMS q/g tagging performance in
simulation. The q/g label is obtained through a matching
of jets on the detector level to outgoing partons from the figure 19 is much more pronounced for MC than for data
matrix-element calculation. For a 50% gluon or quark indicates that the simulation over-predicts the difference
efficiency, the misidentification rate (quark or gluon) is between quark and gluon jets. In contrast, Herwig (not
about 10%. This performance depends slightly on the shown) tends to underestimate the performance observed
jet pT , in part because the particle multiplicity increases in data.
with pT (and therefore the performance improves). Out- Multiple samples with a different (but known) q/g
side the tracking acceptance (|η| & 2.5), q/g tagging sig- composition can be used to extract the distribution of
nificantly degrades due to the coarse calorimeter granu- q/g tagging observables. ATLAS and CMS have both
larity and increased pile-up sensitivity. used Z/γ+jets and dijet samples, which are enriched in
ATLAS (138) and CMS (141) are also actively study- quark and gluon jets, respectively. The extracted average
ing sophisticated approaches based on modern machine ntrack from data is shown using this method in figure 20.
learning. While these methods hold great promise for As expected, gluon jets have more particles on average
their power and flexibility, simple combinations of a small than quark jets and the multiplicity distribution increases
number of features often achieves a similar performance. with jet pT .
Machine learning architecture design and input optimiza- The Run 2 ATLAS tagger is based entirely on dijets,
tion are still an active area of research and development. exploiting the rapidity dependence of the q/g fraction to
extract the track multiplicity separately for quarks and
The modeling of q/g discriminating observables is a key gluons. A Run 1 measurement is used to constrain the
concern for tagging applications. Typically, Pythia (148; particle-level modeling, and dedicated track reconstruc-
149) tends to describe quarks better than Herwig (150; tion uncertainties are used to complement the particle-
151), whereas the opposite is observed for gluons. Pythia level uncertainty with a Run 2 detector-level uncertainty.
tends to overestimate the q/g tagging performance with The uncertainties on q/g tagging are 2-5% over a wide
respect to data, as illustrated quite strikingly in figure 19. range of 200 GeV . pT . 1 TeV at a working point of
This figure shows that gluon jets tend to have more tracks 60% quark jet efficiency (135). The template-based cal-
and have a broader radiation pattern relative to quark ibration can also be used to directly construct the q/g
jets9 . The fact that the hot spot in the bottom left of tagger in data; however, when more than two observables
are used to construct the tagger, it becomes impractical
to extract the high-dimensional templates.
9
The likelihood-based discriminant used for q/g tagging
The jet flavor is obtained as the type of the highest energy
parton from the event record inside the jet cone. This gives
in CMS in Run 2 is calibrated with a template-based
nearly the same result as the CMS definition discussed above for fit using two discriminant distributions obtained from a
the two leading jets in a 2 → 2 calculation, but also works well Z+jets and a dijet sample. The different quark and gluon
for additional jets in the event. fractions in each bin of the discriminant distributions
 18

20

<ntrk>
1 ATLAS Preliminary

L = q/(q+g)
ntrk 18 18
s = 8 TeV 20.3 fb-1
16
0.9 |η| < 0.8
16 ATLAS 14
Discriminant for Data-Driven Tagger
∫ anti-k R=0.4, |η| < 0.8 0.8
-1
12
14 L dt = 4.7 fb , s = 7 TeV
10
t
160 GeV<p <210 GeV
0.7 8
12 T
6
0.6 4 Extracted Validation
10 2
0.5 Gluons Light Quarks
8 1.2 100 200 300 400 500

Validation
Extracted
0.4 1 Jet PT (GeV)
6 0.8
0.3 100 200 300 400 500
4 Jet PT (GeV)
0.2
2 FIG. 20 The average track multiplicity in ATLAS for
0.1
Z/γ+jets (quark-enriched) and dijets (gluon-enriched). The
0 0 dashed lines indicate the measurement on the validation
0 0.05 0.1 0.15
samples : Z/γ+2-jets (quark-enriched) and trijets (gluon-
Track Width enriched). Reproduced from Ref. (137).

1
L = q/(q+g)
ntrk

18 LAS jet calibration procedure (see section III.B). Since

0.9 the calorimeter response is non-linear, a jet with a higher
16 ATLAS Simulation
Discriminant for MC-Based Tagger
0.8 particle multiplicity will have a lower response for the
14 Pythia MC11, s = 7 TeV
same energy. After applying a simulation-based correc-
anti-kt R=0.4, |η| < 0.8
160 GeV<p <210 GeV
0.7 tion to eliminate this inclusive bias in the JES, a resid-
12 T

0.6 ual calibration is applied to correct for the dependence

10 of the bias on the number of tracks associated to the
0.5
8 jet and the jet width (71). After applying this resid-
0.4 ual GSC, the difference in response between quark and
6 gluon jets is reduced. Implicit q/g tagging also appears
0.3
4 in pile-up jet identification (77; 78), boson and top tag-
0.2 ging (57; 145; 147), and elsewhere.
2 Despite its long history, quark versus gluon jet tag-
0.1
0 ging is still a very active topic of research. Since most
0 analyses at the LHC target processes with a known and
0 0.05 0.1 0.15
asymmetric q/g jet composition, q/g tagging holds great
Track Width
promise for improving searches and measurements in the
future. Further studies are required to understand the
FIG. 19 The two-dimensional q/g likelihood with ATLAS limits of q/g tagging performance and to mitigate the
data (top) and simulation (bottom). Reproduced from
sample dependence for universal definitions and calibra-
Ref. (136).
tions. Interestingly, recent studies have shown how mod-
ern machine learning classifiers can be directly trained on
data even though there are no per-jet labels (152; 153).
are determined simultaneously and fitted by polynomial
functions in order to obtain smooth interpolations (87).
Despite its power, the template technique has some B. Vector Boson Tagging
residual non-closure because the resulting calibrated tag-
ger applied to another final state may not have the same The hadronic, two-prong decays of weak vector bosons
performance. This is illustrated in figure 20, which shows V have a distinct radiation pattern compared to indi-
how the average track multiplicities extracted for quark vidual high-pT quarks or gluons. In particular, boosted
and gluon jets (using high-purity Z/γ+jets and dijets bosons tend to have two distinct subjets with relatively
data respectively) differ from the values obtained in the equal momentum sharing. In contrast, most generic
γ+2-jet and trijet samples used for validation. quark and gluon jets will have one prong and if they
Explicit tagging is often the focus of modern q/g dis- have two, the second one tends to be soft. Furthermore,
crimination, but there is a broad program of implicit tag- the mass of quark and gluon jets scales with their pT and
ging as well. One ubiquitous example of this is the AT- is lower than the electroweak boson masses for low jet pT
 19

and higher for ultra-high pT jets. For jets around 200

&T
QCD
GeV, the decay products of a boosted W and Z boson ATLAS s = 8 TeV

1 / ∈G
anti-k t R=1.0 jets Trimmed (f =5%,Rsub =0.2)
are typically only captured by a jet of radius R ∼ 1, while |η| < 1.2, 350 GeV < p < 500 GeV
cut

smaller radii can be used at higher jet pT . Good separa- T

Performance in W bosons from t t (POWHEG+Pythia)
vs. inclusive multijets (Pythia)
tion power between W and Z bosons is also desirable in wta
τ21
( β =1)
C2
( β =1)
D2

a number of analyses, most notably searches for diboson Monte Carlo Predicted

Measured (Medium WP)

resonances (see section VIII.A). Stat. ⊕ Syst. Uncertainty

102 Measured (Tight WP)

Stat. ⊕ Syst. Uncertainty
ATLAS and CMS performed a broad range of studies
during Run 1 and the beginning of Run 2, systematically
identifying the influence of pile-up reduction and groom-
ing techniques on jet substructure observables used for V
tagging (94; 105). Simulated samples containing W jets
(rather than Z jets) are primarily used for these stud-
ies, as W jets are abundant in data thanks to the large 10
quantity of tt events produced at the LHC.
0 0.2 0.4 0.6 0.8 1
The optimization of the V tagging algorithm is gener- ∈G
W
&T

ally based on various factors concerning the tagged jet

mass: (i) a sensible JMS (i.e., tagged jet mass close to FIG. 21 Signal efficiency versus background rejection power
the W mass), (ii) a narrow jet mass response with an compared with measurements from ATLAS for 350 < pT <
approximate Gaussian lineshape, (iii) stability with re- 500 GeV. Taken from Ref. (94).
spect to pile-up and jet pT , and (iv) good background
rejection at a given signal efficiency. Considering all of
these factors, ATLAS decided on using the trimming al- as good a discriminator as τ21 10 as shown in figure 21.
gorithm (90) with fcut = 0.05 and Rsub = 0.2 on anti-kT , This is in contradiction to the study by CMS, where C2β
R = 1.0 jets in Run 2, while CMS opted for using anti- is one of the weaker observables; however, a direct com-
kT , R = 0.8 jets, treating the pile-up first with PUPPI parison is difficult, since in ATLAS groomed substructure
and then applying soft drop grooming with zcut = 0.1 variables are used, calculated for trimmed jets, while in
and β = 0. CMS ungroomed variables are used. Also, the particulars
of particle reconstruction have a large impact on the per-
In addition to the comprehensive studies of grooming formance of individual observables. While a study of the
options (57; 74; 86; 94; 95), ATLAS and CMS both in- performance of D2β at CMS is still pending, the soft drop
vestigated the discrimination powers for a plethora of N2β observable was found to give similar performance to
jet substructure variables, including N -subjettiness (110; τ21 in CMS (160).
111), Qjet volatility (154), ratios of energy correlation CMS studied the quark/gluon likelihood (QGL) dis-
functions C2β (117), D2β (118; 119) and N2β (120), angu- criminator for its potential in V tagging applications in
larities and planar flow (155), splitting scales (12; 156), Run 1 (57), finding that a combination of the groomed
the jet and subjet quark/gluon likelihood, and the jet jet mass and the QGL achieved a similar discrimination
pull angle (157). power as the groomed jet mass and τ21 . When adding the
Both ATLAS and CMS developed simple taggers that QGL to the Run 1 V tagger (pruned jet mass and τ21 ),
rely on the combination of the jet mass with one other the misidentification rate was reduced slightly from 2.6%
variable that improves the discriminating power between to 2.3% at a constant signal efficiency of 50%. A similar
the signal and background. The standard ATLAS V tag- reduction of the misidentification rate was observed when
ger for Run 2 was chosen to be the trimmed jet mass adding C2β=2 , showing that C2β carries additional infor-
and D2β=1 (94), known as ‘R2D2’, while CMS decided to mation with respect to the groomed jet mass and τ21 .
use the soft drop jet mass and the N -subjettiness ratio However, the QGL and C2β exhibit a considerable pile-up
τ21 = τ2 /τ1 . Despite the different choices of tagging ob- dependence, resulting in a degradation of their discrim-
servables and detector design, ATLAS and CMS reach ination power with increasing activity. This pile-up de-
a very similar background rejection at a given tagging pendence is expected to be reduced when using PUPPI
efficiency. An active field of developments is the usage in place of particle flow + CHS.
of multivariate techniques for boosted V identification
which have shown to be able to significantly improve the
background rejection (105; 158). 10 A different axis definition for the subjet axes is used in ATLAS
when calculating τN , known as the-winner-takes-all axis (159),
In the ATLAS studies the variable C2β=1 in combina- which is consistently found to perform slightly better than the
tion with the trimmed jet mass has been shown to be standard subjet axis definition in tagging bosons.
 20

0.9 monly used method is the extrapolation from one or

&T ATLAS Simulation Preliminary more control regions, which are defined orthogonally to
W
∈G
0.8 anti-k t R=1.0 jets
Trimmed (f = 5%, Rsub = 0.2)
the signal region. Usually, these control regions are de-
cut
0.7 |η| < 2.0, W-jets 50% signal efficiency fined by inverting the jet mass window selection, see
25% signal efficiency
0.6
e.g. (161; 162; 163; 164; 165; 166). Transfer functions
are derived from simulation, extrapolating the rates and
0.5 shapes from the control to the signal regions. Even
0.4 though these transfer functions are ratios of distributions,
which results in a reduction of the impact of modeling
0.3 uncertainties, a residual dependence on the simulation
0.2 can not be eliminated. However, additional uncertain-
500 1000 1500 2000 ties in the high-pT tails of the transfer functions can be
Jet p [GeV] eliminated by ensuring a constant behavior as a function
T
of pT . The requirement is thus a flat signal or back-
600 ground efficiency (depending on the needs of the anal-
&T
QCD

ATLAS Simulation Preliminary ysis). In order to achieve a flat signal efficiency, AT-
1 / ∈G

anti-k t R=1.0 jets

500 LAS developed a pT -dependent selection on the value of
Trimmed (f = 5%, Rsub = 0.2)
cut
|η| < 2.0, W-jets 50% signal efficiency
25% signal efficiency
D2β=1 , as this distribution shows a strong dependence
400
w/o D2 systematics on pT (95). In contrast to the Run 1 studies described
300 above, no pT -dependent selection is made on the trimmed
jet mass, as the calibrated jet mass is used to define the
200 V tagging working point. While the jet mass resolution
still increases with pT , a constant window of ±15 GeV
100 around the mean reconstructed W or Z boson mass is
used. This results in a pT -dependent signal and back-
500 1000 1500 2000 ground efficiency, which can also be countered with the
Jet p [GeV] pT -dependent cut on D2β=1 . This leads to a constant
T
signal efficiency, while the background efficiency shows a
FIG. 22 Efficiency (top) and misidentification rate (bot- residual pT dependence, as shown in figure 22.
tom) for tagging boosted W bosons in ATLAS. Adapted from Another possibility has been explored by CMS. Instead
Ref. (95). of introducing pT -dependent selection criteria, a linear
transformation of the ratio τ21 has been studied (87),
DDT
given by τ21 = τ21 − M · log(m2 /pT /1 GeV) (167),
In figure 21 the ATLAS measurements of signal effi- where M is a constant determined from simulation. The
ciencies versus background rejection power are shown for replacement of τ21 with the designed decorrelated tagger
τ21 , C2β and D2β , together with a selection on the trimmed (DDT) τ21 DDT
does not affect the overall performance of
jet mass (in this pT range, the smallest mass window the tagger, but results in an approximately flat misiden-
that captured 68% of the signal jets was found to be 71- tification rate as a function of pT , as shown in figure 23
91 GeV– see Ref. (94), table 7). The measurements are (bottom). The effect of the DDT method on the V tag-
shown with statistical and systematic uncertainties. It ging efficiency is shown in figure 23 (top). The efficiency
is reassuring that the points for all three observables lie increases as function of pT with a slope somewhat smaller
on the predicted performance curves for the two different than the slope for the decreasing efficiency obtained with
working points studied. plain τ21 . The development of decorrelated jet substruc-
In the ATLAS study, the most important systematic ture taggers is an active field with new techniques e.g.
uncertainty is the jet substructure scale, which has been described in Refs. (168; 169; 170).
derived by comparing calorimeter-jets with track-jets. A less-studied possibility to lift the pT -dependence of
Once again, the distributions in data lie between the substructure observables is the application of variable-R
ones derived with Pythia and Herwig, leading to large jets (171). By shifting the pT -dependence to the jet-
modeling uncertainties (94; 95). A similar observation is clustering level with a distance parameter proportional
made by CMS (87; 105). Improving the modeling of jet to p−1T , a stable position of the jet mass and jet sub-
properties and thereby reducing the differences between structure variables with respect to changes in pT can be
different event generators is a major task, but crucial for achieved (172). This can lead to a stable tagging perfor-
future precision studies using jet substructure. mance without the necessity of pT -dependent optimiza-
A crucial aspect of V tagging is the derivation of tion steps, but further experimental studies are needed
background rates from multijet production in real col- to commission this strategy for use in analyses.
lision data when performing measurements. A com- For some analyses the requirement of pT & 200 GeV is
 21

CMS Simulation Preliminary 13 TeV

104

Efficiency

1/(W Efficiency)
1.4 CHS
65 GeV < MPruned < 105 GeV
PUPPI
65 GeV < MSoftdrop < 105 GeV
ATLAS Simulation
1.2 65 GeV < MPruned < 105 GeV + τ21 ≤ 0.45
CHS

65 GeV < MSoftdrop < 105 GeV + τ21 ≤ 0.4

PUPPI

103
1 65 GeV < MSoftdrop < 105 GeV + τDDT
PUPPI
21 ≤ 0.52 No discrimination

+
M (Jet Mass)
0.8 Q (Jet Charge)
B (b-tagging)
102 M+B
0.6
M+Q
Q+B
0.4 M+Q+B
W-jet, AK R = 0.8
10
0.2 p > 200 GeV
T
|η| ≤ 2.4 GeV

0
500 1000 1500 2000 1
Jet p (GeV)
T
CMS Simulation Preliminary 13 TeV 0 0.2 0.4 0.6 0.8 1
0.3
Mistag rate

65 GeV <
CHS
MPruned < 105 GeV Z Efficiency
PUPPI
65 GeV < MSoftdrop < 105 GeV
MPruned < 105 GeV + τ21 ≤ 0.45
CHS
65 GeV <
65 GeV < MSoftdrop < 105 GeV + τ21 ≤ 0.4
PUPPI FIG. 24 Background rejection versus efficiency for discrimi-
0.2 65 GeV < MSoftdrop < 105 GeV + τDDT
PUPPI
21
≤ 0.52 nating Z boson jets from W boson jets for various jet observ-
QCD, Pythia8 ables and their combinations. Reproduced from Ref. (173).
p > 200 GeV
T
|η| ≤ 2.4 GeV

the jet mass resolution is comparable to the difference

0.1
mZ − mW . In order to improve the sensitivity of the tag-
ger, jet charge and b tagging information are combined
with the jet mass. The jet mass distribution depends
on the type of W or Z decay due to semi-leptonic B
0
500 1000 1500 2000 and D decays, so a full likelihood tagger is constructed
Jet p (GeV) by summing over the conditional likelihoods for each fla-
T
vor type. To maximize the discrimination power from
FIG. 23 Efficiency and misidentification rate of various iden-
b tagging, multiple efficiency working points are used
tification techniques for boosted W tagging. Taken from
Ref. (87). simultaneously in the tagger. Figure 24 illustrates the
performance of the boson type-tagger in simulation. A
W + rejection near 8 (corresponding to a misidentifica-
too restrictive, and hadronically decaying V bosons with tion rate of 12.5%) is achieved at a Z boson efficiency
lower pT need to be selected. This poses a particular of 50%. At this moderate Z boson efficiency, all of the
challenge due to the abundance of light flavor jets at the inputs offer useful discrimination information. At low ef-
LHC and their indistinguishability from jets from W /Z ficiencies, below the bb̄ branching ratio for Z bosons, b
decays. An attempt was made by CMS to discriminate tagging dominates over the jet mass and jet charge.
‘resolved’ (non-merged) hadronic W decays from multijet The boson type-tagger was optimized for a relatively
background using the QGL, the sum of the jet charges low boson boost, 200 GeV < pT < 400 GeV. The dis-
of the dijet pair and the jet pull angle. Combining these crimination power of all of the input variables degrades
variables into a Boosted Decision Tree, a misidentifica- with pT due to the worsening jet mass resolution, track-
tion rate of about 25% is achieved for a signal efficiency ing efficiency and momentum resolution, as well b tagging
of 50% (57). While this is a first success, the performance efficiency. However, there are recent developments to ad-
is about an order of magnitude worse than V tagging for dress each of these challenges, such as the track-assisted
fully merged decays, showing the power of substructure jet mass (section V.B), pixel-cluster splitting (61), and
techniques in this field. track-jet b tagging (174).
In addition to developing tools for distinguishing
boosted hadronically decaying W and Z bosons from
generic quark and gluon jets, ATLAS has also built a C. Top Tagging
tagger to further classify a boson jet as either originat-
ing from a W boson or a Z boson (173). While theo- The three-prong decays of highly boosted top quarks in
retically clean due to the color singlet nature of the W the fully hadronic decay channel offer richer phenomenol-
and Z boson, this task is particularly challenging because ogy for their identification than the two-prong decays of
 22

W and Z bosons. This has been exploited in a number ble shower histories that can lead to the observed
of algorithms, which usually aim at an optimal perfor- leading final state anti-kT , R = 1.0 jet are calcu-
mance in a particular kinematic regime. Flavor tagging lated. Each shower history is assigned a probability
also plays a key role for top tagging, which offers its own weight factor based on the aforementioned consid-
challenges because the b jet from the b quark may not be erations (to be signal-like or background-like), then
isolated from the radiation resulting from the associated a likelihood ratio χ(pN ) is constructed, and the
W boson decay. Due to the heavier mass of the top quark log χ(pN ) is used as the discriminating substruc-
compared with the electroweak bosons, top tagging must ture variable. For top quark tagging, efficiencies
also operate in a moderate boost regime where the decay of 80% with misidentification rates of 50% for jets
products may not all be contained inside a single jet with with 500 < pT < 1000 GeV were observed. The
R . 1.0. efficiency increases with increasing jet pT , where a
The techniques for tagging boosted top quarks have plateau is reached for pT > 2000 GeV, with efficien-
evolved as fairly complex methods in comparison to the cies of ∼ 80% at 10% misidentification rate. Re-
V taggers; these techniques include: cently, the Shower Deconstruction algorithm was
optimized for top quarks with pT > 800 GeV in
(a) The Johns Hopkins / CMS top tagger (CMSTT) context of the W 0 to tb hadronic search (182) by
(15) was designed for tagging top quarks with pT using exclusive kT subjets.
> 1 TeV. The algorithm is based on a decomposi-
tion of the primary jet into up to four subjets by In addition to the dedicated techniques described
reversing the CA clustering sequence. It has been above, simpler algorithms using grooming and substruc-
adapted by the CMS Collaboration (175; 176), and ture similar to V tagging methods have been investigated
was adopted as the standard top-tagging algorithm by ATLAS. A performance study at 7 TeV (179) inves-
in CMS in Run 1, where it was typically used in tigated a variety of performance metrics relating to the
the region of pT > 400 GeV, with an average iden- usage of groomed jets. Different grooming algorithms
tification efficiency of 38% at 3% misidentification were investigated for their resilience to pile-up and mass
rate (106). resolution. It was concluded that trimmed anti-kT jets
(b) The HEPTopTagger (HTT) (177; 178) was de- with a distance parameter of 1.0 and trimming parame-
signed to target ttH production in the H → bb de- ters of Rsub = 0.3 and fcut = 0.05 were a good candidate
cay channel. In ttH production the top quark pT for a one-fits-all large-R jet definition. This jet definition
distribution peaks around 150 GeV and is steeply became standard in ATLAS for W /Z/H and top quark
falling towards increasing pT , where it is already tagging in Run 1. A later ATLAS study (96) investi-
an order of magnitude smaller at pT ∼ 400 GeV. gated the various methods available for tagging hadronic,
This results in a requirement of non-zero signal ef- highly boosted top√quarks. The so-called √ Tagger V has
ficiency already at pT ≈ 200 GeV, where the top Mjet >p100 GeV, d12 > 40 GeV and d23 > 20 GeV,
quark decay is only moderately boosted. The HTT where dij is the kT -splitting scale (12). The efficiency
achieves this with a large jet distance parameter of versus rejection is shown for various taggers in figure 25.
1.5 and a sequence of declustering, filtering and re- The difference between√Taggers III and V is the addi-
clustering of the original CA jet. The performance tional requirement on d23 in Tagger V. At √ efficiencies
of the HTT was studied by the ATLAS and CMS smaller than 45%, the W 0 tagger, based on d12 and the
Collaborations on data with a center-of-mass en- N -subjettiness ratios τ21 and τ32 , has better background
√ rejection than Taggers III and V. ATLAS also tested the
ergy s = 7 and 8 TeV (96; 106; 179). Efficiencies
of 10% with misidentification rates of 0.5% for jets HTT and Shower Deconstruction (183), which have been
with 200 < pT < 250 GeV were observed. The ef- found to have good background rejection (larger than 50)
ficiency increases with increasing jet pT , where a for efficiency values smaller than about 35%. However,
plateau is reached for pT > 400 GeV, with efficien- similar as for the CMS experiment, the background ef-
cies of approximately 40% at 3% misidentification ficiencies of the two taggers show a significant rise with
rate, very similar to the performance achieved with increasing pT .
the CMSTT. CMS has focused on enhancing the performance of
CMSTT and HTT by identifying observables which carry
(c) Shower Deconstruction (180; 181) was designed to discriminatory power, but have only small or moderate
be analogous to running a parton shower Monte correlations with the observables used in the main algo-
Carlo generator in reverse, where emission and rithm. Typically, correlation coefficients of about 0.3 or
decay probabilities at each vertex, color connec- less are required for noticeable improvement when aug-
tions, and kinematic requirements are considered. menting an algorithm with additional variables. Exam-
Small-radius (generally R = 0.2) subjets are re- ples for discriminating variables which fulfill this are N -
constructed with the CA algorithm and all possi- subjettiness ratios, energy correlation functions and their
 23

103 CMS
Background rejection ATLAS
SD (C/A 0.2 subjets)

SD (C/A 0.2 subjets) WP

Simulation Preliminary 13 TeV
Simulation s = 8 TeV

εB
350 GeV < pT < 400 GeV
HEPTopTagger (C/A 1.5)
800 < p < 1000 GeV, |η| < 1.5
d12 T
trimmed mass ∆R(top,parton) < 0.6
2
10
flat p and η
10−1
Tagger V (scan d23)

W' Tagger (scan τ32)

T

d23

τ32

Tagger I
10 Tagger II

Tagger III
10−2 CMSTT min. m
CMSTT top m
Tagger IV Filtered (r=0.2, n=3) m
HTT V2 fRec
Tagger V
HTT V2 m
W' Top Tagger Pruned (z=0.1, rcut=0.5) m
1
0 0.5 1 10−3 Q-jet volatility
Softdrop (z=0.1, β=0) m
Top quark tagging efficiency Softdrop (z=0.2, β=1) m
Trimmed (r=0.2, f=0.03) m
Ungroomed τ3/τ2
FIG. 25 Top quark tagging efficiency versus background re- log(χ) (R=0.2)

jection for various substructure variables and combinations in 10−40 0.2 0.4 0.6 0.8 1
ATLAS. Taken from Ref. (96).
εS

FIG. 26 Background versus signal efficiency for the single

ratios, and b tagging. A study by the CMS Collaboration
variables studied in the optimization of top tagging for CMS
showed that at 20% signal efficiency, the background re- Run 2 analyses. Taken from Ref. (107).
jection of the CMSTT can be improved by a factor of 5
when adding information from τ32 and subjet b tagging
information (106). At higher efficiencies, the improve-
ments become smaller. For the HTT, improvements of matched to a generated parton, and the distance between
similar size are observed for pT > 200 GeV, becoming less the top quark and its decay products must be less than
significant at higher pT . 0.6 (0.8) for a reconstructed R = 0.8 (1.5) jet, to ensure
The ATLAS choice of R = 1.0 jets compared to CMS that the top quark decay products are fully merged and
(R = 0.8) results in an earlier rise of the tagging efficiency reconstructed in a single jet. The best single variable in
with increasing jet pT . terms of efficiency versus background rejection is the dis-
The large difference in performance of the single vari- criminator log χ, calculated with Shower Deconstruction.
able τ32 between ATLAS (figure 25) and CMS (figure 26) The second best variables are the N -subjettiness ratio
is due to jet grooming. Although the CMS study shows τ32 at low efficiency and the jet mass calculated with the
only the ROC curves for 800 < pT < 1000 GeV, the over- HTTv2 at high efficiency values. The individual groomed
all picture does not change when studying top quarks in jet masses show similar performance, and the CMS Col-
the region of pT ≈ 400 GeV. Instead, in ATLAS τ32 is laboration moved to using the soft drop mass due to its
calculated from trimmed jets, which results in less dis- beneficial theoretical properties (16). The default for
crimination power when used as sole tagging variable CMS Run 2 analyses was chosen to be the soft drop jet
compared to ungroomed τ32 . However, groomed τ32 can mass combined with τ32 for top tagging at high pT . Gen-
still lead to considerable improvements when combined erally, at high boost, the combination of a groomed mass
with other variables. with τ32 leads to a large gain in background rejection.
As with V tagging discussed above, ATLAS and CMS The CMS study also investigated combining single
took advantage of the LHC shutdown between Run 1 variables with more complex taggers. Combining Shower
and Run 2 to perform broad studies of the different Deconstruction with the soft drop mass, τ32 , and sub-
top-taggers available, with emphasis on their stability jet b tagging can lead to improvements, as shown in fig-
with respect to pile-up and other detector effects, in- ure 27; however, the efficiency and misidentification rate
stead of the utmost gain in performance (97; 107). Single for this combination were found not to be stable as a
variables and their combinations are studied and com- function of jet pT (the combined algorithms were stud-
pared with Shower Deconstruction, CMSTT, HTT, and ied using working points corresponding to a background
an improved version of the HTT with shrinking cone size efficiency of 0.3). At low boosts, the dedicated HTTv2
(HTTv2) (184). shows the best performance. In this kinematic region,
Figure 26 shows a comparison based on simulation of using groomed τ32 , obtained by using the set of particles
the single variable performance in CMS, where signal jets from the soft drop jet instead of the original jet, helps to
are generated through a heavy resonance decaying to tt improve the performance.
and background jets are taken from QCD multijet pro- In the shutdown between Run 1 and Run 2, ATLAS
duction. Note that for this study reconstructed jets are commissioned a single top tagger for use by physics anal-
 24

CMS (13 TeV)

1

Top quark tagging efficiency

Simulation Preliminary 13 TeV
εB 10−1 0.9 CMS Combined
Simulation Preliminary Monojet
800 < p < 1000 GeV, |η| < 1.5 0.8 Dijet
T Top quark tagger efficiency
flat p and η 0.7
Trijet
T measured in T2tt(850,100)
∆ R(top,parton) < 0.6 0.6
10−2 0.5
0.4

CMSTT - min. m, top m, τ3/τ2

0.3
CMSTT - min. m, top m, τ3/τ2, b
HTT V2 - m, fRec, ∆ R, τ3/τ2
0.2
10−3 HTT V2 - m, fRec, ∆ R, τ3/τ2, b
log(χ) 0.1
mSD (z=0.1, β=0) , τ3/τ2
mSD (z=0.1, β=0) , τ3/τ2, b 0
mSD (z=0.1, β=0) , τ3/τ2, log(χ)
0 100 200 300 400 500 600 700 800 900 1000
mSD (z=0.1, β=0) , τ3/τ2, log(χ), b pgen [GeV]
mSD (z=0.1, β=0) , min. m, τ3/τ2 T
mSD (z=0.1, β=0) , min. m, τ3/τ2, b
−4
10 0 0.2 0.4 0.6 0.8 1 FIG. 28 Top tagging efficiency of three different top tagging
methods and the combined efficiency, as a function of the
εS generated top quark pT . Taken from Ref. (188).

FIG. 27 Background efficiency versus signal efficiency for

combined variables studied in the optimization of top tagging
for CMS Run 2 analyses. Taken from Ref. (107). Most top-taggers target either the region of low to in-
termediate boosts, or the highly boosted regime. How-
ever, in typical searches for new physics at the LHC
non-vanishing efficiency for the full kinematic reach is
yses. The rationale behind this approach was the poten- crucial. Several attempts of combining different recon-
tial benefit of having an efficient top tagger with well- struction and identification algorithms have been made.
understood efficiency and associated systematic uncer- A search for resonances decaying to tt by the ATLAS
tainties validated in the Run 1 dataset. Similarly as for Collaboration uses a cascading selection from boosted
Run 1, the supported top-tagger makes use of anti-kT , to resolved (186), where the resolved topology is recon-
R = 1.0 trimmed jets, but with a parameter of Rsub = 0.2 structed and identified using a χ2 -sorting algorithm. To
instead of 0.3 as used in Run 1. Candidate top jets are efficiently identify top quarks over a broad pT range in
required to satisfy a calibrated mass window requirement the search for top squark pair production, reclustered
122.5 < Mjet < 222.5 GeV and a pT -dependent, one-sided variable-R jets are used with R = 0.4 jets as inputs to
cut on τ32 (97). The variable τ32 has been chosen since it the jet reclustering algorithm (102; 187).
shows the best background rejection in combination with A search for supersymmetry in CMS (188) uses
a small correlation with Mjet , a reduced pT -dependence, three distinct topologies: fully-merged top quark decays
and good performance across a large range in pT . (Monojet), merged W boson decays (Dijet) and resolved
A common problem of top-tagging algorithms is the decays (Trijet). The efficiency of the three categories is
rise of the misidentification rate with increasing pT , shown in figure 28, where the turn-on of the combined
which is due to the peak of the mass distribution for efficiency starts at values as low as pT ≈ 100 GeV. The
quark- and gluon-initiated background jets shifting to resolved trijet category is identified using three anti-kT
higher values. For some taggers, for example the CM- jets with a distance parameter of 0.4, where the large
STT, this shift also results in a decrease of the efficiency combinatorial background is suppressed through a mul-
once a very high pT threshold is crossed (larger than tivariate analysis, which achieves a misidentification rate
1 TeV) (176). A possible solution to this is offered by the of approximately 20%. There exist other approaches to
variable-R (VR) algorithm, introduced in section VI.B. cover the transition from low to high Lorentz boosts, us-
The ATLAS Collaboration studied the performance of ing a single algorithm. In the HTTv2 algorithm, the jet
the VR algorithm for top-tagging and reported a stabi- size is reduced until an optimal size Ropt is found, de-
lization of the position of the jet mass peak for a large fined by the fractional jet mass contained in the smaller
range of pT (172). The VR jets are jet. This results in better performance at high pT , while
√ shown to improve
the performance of the jet mass, d12 and τ32 for top keeping a low misidentification rate at low pT .
tagging, when compared to trimmed jets. An interesting An important step towards the commissioning of top
development using VR jets is the Heavy Object Tagger taggers within an experiment are measurements of the ef-
with variable-R (HOTVR) (185), which combines the VR ficiency and misidentification rate in real collision data.
algorithm with a clustering veto, resulting in a single jet Generally, high-purity samples of top-jets in data are ob-
clustering sequence producing groomed jets with subjets. tained using a tight signal selection (an electron or muon,
 25

well-separated from a high-pT large-R jet, and an addi-

Tagging efficiency
ATLAS SD anti-kt R=1.0
tional b-tagged jet) to ensure that events contain a fully- 0.8
∫ L dt = 20.3 fb ,-1 C/A R=0.2 subjets
s = 8 TeV
merged top quark decay in a single large-R jet. This 0.7
correlated data-MC systematic uncertainty
can never be fully achieved, as no requirements on the 0.6 0.7 < |η| < 2.0
substructure of the large-R jet can be imposed without 0.5

biasing the efficiency measurement. This results in an 0.4

efficiency measurement that will be based on a sample 0.3

0.2
also containing partially-merged or even non-merged top Data - BG
0.1 Syst. uncertainty
quark decays. These can be subtracted from the effi- Powheg+Pythia

Data/Sim.
ciency measurement by using simulated events, as done in 1.5
350 400 450 500
a study by the ATLAS Collaboration (96), with the draw- 1
0.5 Large jet p [GeV]
back of relying on a specific simulation and the ambigu- 350 400 450 T 500
ous definition of a fully-merged top quark decay. By not Large-R jet p [GeV]
T
correcting for non-merged top quark decays, efficiency
0.05

Mistag rate
values are obtained smaller than the ones suggested by 0.045
ATLAS
ROC curve studies, see for example (107). Instead of 0.04 ∫ L dt = 20.3 fb ,-1
s = 8 TeV

subtracting the top-backgrounds, the CMS collaboration 0.035 SD anti-kt R=1.0

C/A R=0.2 subjets
performs a simultaneous extraction of the efficiencies for 0.03
0.025
fully- and partially-merged categories (121).
0.02
Measurements of the misidentification rate can be car- 0.015
ried out by selecting a dijet sample, which is dominated 0.01 Data

by light-flavor jets. The disadvantage of this approach is 0.005 Pythia Syst. uncertainty

the high pT threshold of unprescaled jet triggers, which

Data/Sim.
1.5
350 400 450 500 550 600 650 700
results in measurements starting from pT > 400 GeV or 1
higher. A solution to this is the tag-and-probe method, in 0.5 Large jet p [GeV]
350 400 450 500 550 600 650 T 700
which the tagged jet can be required to fail top-tagging Large-R jet p [GeV]
T
selection criteria, resulting in a sample with negligible
contamination of tt production, even after requiring the FIG. 29 ATLAS measurement of the efficiency (top) and
probe jet to be top-tagged (107). Another approach is misidentification rate (bottom) for trimmed jets with a dis-
to use a non-isolated electron trigger, where the elec- tance parameter of 1.0 tagged with Shower Deconstruction.
tron fails offline identification criteria. This yields events Taken from Ref. (96).
mainly from light-flavor multijet production, where a jet
is misidentified as an electron at the trigger level. While
the top-tag misidentification rate can be measured start- for example the presence of a secondary vertex due to the
ing from smaller values of pT with this strategy, a non- long B hadron lifetime of about 1.5 ps.
negligible amount of tt contamination has to be sub- ATLAS and CMS both use dedicated b tagging al-
tracted after requiring a top-tagged jet (96). gorithms that have been developed and optimized over
As an example, the efficiency and misidentification rate more than a decade. Both experiments use multivariate
of Shower Deconstruction with the requirement log(χ) > techniques with various input parameters related to the
2.5, as measured in ATLAS, are shown in figure 29. The secondary vertex or charged particle tracks originating
efficiency of 30% with a misidentification rate of 1% for from the B hadron decay. For Run 2 analyses, CMS uses
350 < pT < 400 GeV agrees well with the values ob- the CSVv2 algorithm (189) and ATLAS uses the MV2c10
tained from figure 25. Note that the largest uncertainty algorithm (190). Typically, efficiencies of around 70%
of the efficiency measurement stems from the choice of with misidentification rates of 1% for light quark and
the Monte-Carlo (MC) event generator used to simulate gluon jets and 20% for charm jets are achieved with these
tt production. The uncertainty of the misidentification algorithms.
rate measurement is dominated by the energy scales and While b tagging in busy hadronic environments plays
resolutions of the subjets and large-R jets. an important role for top tagging, it is the key challenge
for tagging boosted H → bb̄ signatures. Other jet sub-
structure observables can improve performance, but are
D. H → bb̄ Tagging often less powerful once two b tagged jets or subjets are
required (as this necessarily forces the jet to have two-
The identification of jets originating from the fragmen- prongs). The lighter mass of the Higgs boson compared
tation of b quarks (b tagging) is a crucial task in many with the top quark also means that the b-jets from the H
areas of particle physics. Algorithms used for b tagging decay become merged at a lower parent particle boost.
usually rely on the distinct signature of B hadron decays, The boosted H → bb̄ signature is present in many mod-
 26

els of physics beyond the Standard Model: resonant HH 106

Multi-jet rejection
and V H production, searches for boosted mono-H, or ATLAS Simulation Preliminary
vector-like quark searches in the tH and bH final states. 105 calo
p > 250 GeV, No mjet selection
T

Because of the large predicted branching fraction for the Double b-tag
Asymm. b-tag (70% wp)
H → bb̄ decay of about 58%, its coupling to b quarks is 104
Single b-tag
one of the most interesting to study. For a large frac- Leading subjet b-tag
103
tion of Higgs bosons with pT > 300 GeV, the two b quark
jets merge into a single jet for a jet distance parameter of 102
R = 0.8 or 1.0, as used in CMS and ATLAS, respectively.
Several phenomenological studies have explored H → bb̄ 10
tagging algorithms using jet substructure, though ulti-
mately the optimal performance comes from using a com- 1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
bination of substructure information and the track and Higgs-jet efficiency
vertex information related to the B hadron lifetime.
The approaches to identify boosted H → bb̄ candidates FIG. 30 The rejection of inclusive multijets versus Higgs-jet
that have been explored (and used) at CMS and ATLAS efficiency using all large-R jets with pT > 250 GeV for sin-
include: gle, double, asymmetric, and leading subjet b tagging require-
ments. Taken from Ref. (195).
(a) Subjet b tagging (174; 191; 192; 193; 194; 195),
where ‘standard’ b tagging is applied to each of
the subjets (the standard for CMS is the CSVv2 case of asymmetric b tagging, also due to the 70% b tag-
algorithm (196), and for ATLAS is MV2c20 (190)). ging working point requirement on one of the track jets.
Tagging b-jets in dense environments is of particu- The CMS double-b tagging algorithm (189; 198) at-
lar importance here, and was studied by ATLAS in tempts to fully exploit the strong correlations between
Ref. (197). In CMS subjets with R = 0.4 are clus- the b hadron flight directions and the energy flows of
tered with the kT algorithm using the constituents the two subjets, while adapting the variables used in the
of the large-R jet, while for ATLAS track jets with CSVv2 algorithm. The flexibility of the double-b tagger
a radius of 0.2 are matched to the large-R jet us- is ensured by avoiding a strong performance dependence
ing the ghost-association technique. At high pT on the jet pT and mass.
the subjets start to overlap causing the standard b With the double-b tagger, at the same signal efficiency,
tagging techniques to break down due to double- the misidentification rate is uniformly lower by about a
counting of tracks and secondary vertices when factor of two compared to the subjet b tagging approach.
computing the subjet b tag discriminants. Given the different kinematic properties expected for a
bb̄ pair originating from the decay of a massive resonance
(b) Double-b tagging (189; 195; 198), where in AT- compared to gluon splitting, the misidentification rate for
LAS, the term double-b tagging means that the the gluon splitting background reduces from 60% to 50%
two leading pT track jets must pass the same b at 80% signal efficiency and from 20% to 10% at 35%
tagging requirement. In CMS, the double-b tag- signal efficiency. At high pT , even larger performance
ger (189; 198) uses the N -subjettiness axes and the improvements are observed, which is an important gain
pruned anti-kT , R = 0.8 jet mass with a window for searches for heavy resonances, where very high pT
of 50 < M < 200 GeV to reduce the multijet back- jets are expected. In figure 31 the signal efficiencies and
ground. misidentification rates for the double-b tagger are shown
as a function of jet pT for three operating points: loose,
The Higgs-jet efficiency versus the inclusive multijet
medium and tight, which correspond to 80%, 70% and
rejection are shown in figure 30 for ATLAS subjet b tag-
35% signal efficiency, respectively, for a jet pT of about
ging, where the performance curves are shown for double-
1000 GeV. The misidentification rate is mostly flat across
b tagging, leading subjet b tagging, and asymmetric b tag-
the pT range considered while the signal efficiency de-
ging11 requirements. None of the curves reach a Higgs-jet
creases with increasing pT , as expected from the degra-
efficiency of 100% due to the imperfect efficiency to re-
dation of the tracking performance inside high pT jets.
construct the track jets needed for b tagging and, in the
Due to the small cross section of producing events with
boosted H → bb̄ or Z → bb̄ jets, the efficiency of the
11 Asymmetric b tagging means that among the two leading pT
ATLAS and CMS Higgs identification algorithms is mea-
track jets, the track jet with the largest b tagging weight must sured using QCD multijet events enriched in jets from
pass the fixed 70% b tagging working point threshold, while the gluon splitting, g → bb̄ with a topology similar to that of
b tagging requirement of the other jet is varied. boosted H → bb̄ jets.
 27

CMS Simulation Preliminary (13 TeV) One of the major backgrounds for analyses selecting
1
boosted H or Z bosons decaying to bb̄ is tt̄ production.
Mistagging efficiency
double-b-tag
Subjet CSVv2
The misidentification rate for boosted top quark jets fak-
Fatjet CSVv2
ing H jets was measured in data by CMS (189; 198) in
AK8
70 < m < 200 GeV , 300 < p < 500 GeV enriched data samples of lepton+jets tt̄ events.
T
10−1 As previously discussed, for high pT of the Higgs bo-
son, the two subjets from b quarks start overlapping and
the performance of identifying the subjets as fixed-radius
track jets decreases significantly. To improve the perfor-
10−2
mance of the ATLAS standard H → bb̄ identification
algorithm for searches that require the presence of high
pT Higgs bosons, the ATLAS Collaboration studied alter-
native methods like the use of variable-radius track jets,
exclusive kT subjets, calorimeter subjets reconstructed in
10−3 the center-of-mass frame of the Higgs jet candidate (199)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Tagging efficiency (H→bb) and the combination of three jet shape and jet substruc-
ture variables into a multivariate discriminator (200).
FIG. 31 The misidentification rate for inclusive multijets ver- For highly boosted Higgs bosons, these reconstruction
sus Higgs-jet efficiency using jets with 300 < pT < 500 GeV techniques significantly outperform the usage of fixed-
and pruned jet mass 70 < m < 200 GeV for three different b radius track jets.
tagging requirements. Taken from Ref. (198).

VII. STANDARD MODEL CROSS SECTION

MEASUREMENTS
CMS selects topologies as similar as possible to a sig-
nal jet by requiring the jet pT > 300 GeV and pruned The measurement of jet properties is crucial to con-
mass > 50 GeV (189; 198). Each jet has to contain at strain the Standard Model in new energy regimes and
least two muons, each with pT > 7 GeV and |η| < 2.4. constitutes an important test of perturbative calculations
Each pruned subjet is required to have at least one muon of jet structure over a wide region of phase space. More-
among its constituents and within ∆R < 0.4 from the over jet cross section measurements provide constraints
subjet axis (”double-muon tagged”). The double-muon on the parton distribution functions and the strong cou-
tag enriches events with gluons splitting into bb̄ where pling constant, αs . The precise knowledge of jet prop-
both b quarks give rise to a semi-leptonic B hadron decay. erties also improves the precision of other measurements
Such g → bb̄ events are proxies for the signal topology. and searches by constraining the modeling of important
An alternative selection that requires at least one muon background processes. Jet substructure observable mea-
is also examined as a cross-check for the measurement surements are challenging as they require a precise mea-
(”single-muon tagged”). While this single-muon selection surement of the radiation pattern within the jet and thus
allows for a larger dataset in which to perform the tag- a detailed understanding of the jet constituent properties.
ger efficiency measurement, the gluon splitting topology Section VII.A describes measurements of various jet sub-
in this inclusive phase space is less signal-like relative to structure properties, starting from the most widely used
the double-muon selection. Thus, to maximize the simi- and well-understood: the jet mass.
larity between the g → bb̄ and the H → bb̄ topologies, the Jet substructure properties can also be used to ex-
measurement is performed requiring double-muon tagged tend measurements of SM cross sections to higher energy,
jets. It is worth noting however that the jet mass depends where access to the hadronic branching ratios of W/Z/H
on the number of muons and a large fraction of the signal bosons and top quarks is important. Section VII.B intro-
will not contain two muons. duces cross section measurements for SM objects at high
pT . The use of jet substructure in these cases is similar
ATLAS performed a similar measurement selecting
to the application for the searches described in the next
events with at least one anti-kT , R = 1.0 jet with
section (Section VIII).
pT > 250 GeV that has two ghost-associated R = 0.2
track jets (195). As opposed to the measurement from
CMS, only one of the subjets is required to have a muon A. Measurements of Jet Substructure
associated to it. Kinematic and substructure variables
are compared in data and MC after correcting for flavor 1. Jet mass
composition differences of the large-R jet observed be-
tween data and MC simulation and are found to be in The first measurement of the normalized dijet differ-
good agreement. ential cross section as a function of the jet mass was per-
 28
√
0.02 events at s = 13 TeV for groomed anti-kT R = 0.8
σ dm GeV
1dσ 1 0.018
ATLAS 2010 Data,
∫ L = 35 pb -1
jets with the soft drop algorithm with zcut = 0.1 and
0.016
Cambridge-Aachen R=1.2
Split/Filtered with R qq > 0.3
Statistical Unc. β = 0 (β = 0, 1, 2) (98; 109). The soft drop algorithm
Total Unc.
0.014 300 < pT < 400 GeV was chosen as it allows to compare the unfolded measure-
Pythia
0.012
NPV = 1, |y| < 2
Herwig++
ment directly to theoretical calculations which exceed the
0.01 precision of parton shower MC simulations. The jet en-
0.008
ergy of the ungroomed jets used in the ATLAS measure-
0.006
ment are corrected for pile-up effects and calibrated to
0.004
the generator-level while no explicit mass calibration is
0.002
applied to the groomed jets as the unfolding procedure
0
accounts for differences between the reconstructed and
1.80 20 40 60 80 100 120 140 160 180 200 generator-level mass. The CMS Collaboration applied
MC / Data

1.6
1.4 calibration factors derived from simulation and using in
1.2
1 situ techniques (from boosted W bosons) to correct the
0.8
0.6 jet energy and mass scale. Furthermore the jet energy
0.4 and mass are smeared in MC simulation to match the
0.2
0 20 40 60 80 100 120 140 160 180 200
Jet Mass [GeV]
resolution measurements in data. Various sources of sys-
tematic uncertainties, categorized as experimental and
FIG. 32 Normalized differential cross section as a function theoretical uncertainties, that impact the jet mass mea-
of the jet mass for CA jets with R = 1.2 after splitting and surement are taken into account. While CMS evaluated
filtering, taken from Ref. (201). the effect of the jet energy and mass scale uncertainties
on the measurement by varying the energy and mass by
their respective uncertainties, ATLAS evaluated the ex-
formed by the ATLAS Collaboration with a dataset cor- perimental uncertainties based on the accuracy of the
responding to 35 pb−1 of 7 TeV pp collisions (201). Both modelling of the topological cluster energies and posi-
the cross section for groomed and ungroomed CA R = 1.2 tions as well their reconstruction efficiency. Theoretical
jets was measured separately to gain sensitivity to both uncertainties on the physics model are taken into account
the hard and soft jet physics and to gain a deeper under- by comparing the response matrix for various MC gener-
standing of the various effects involved in QCD radiation. ators.
For the ungroomed jet mass, large discrepancies were ob- The comparison of the normalized cross section with
served in the tails of the mass distribution between the two analytical calculations as measured by CMS is shown
predictions from the MC event generators Pythia and in figure 33. ATLAS measured instead the log10 ρ2 distri-
Herwig++, and the data, whereas the core of the mass bution, shown in figure 34, where ρ is the ratio of the soft
distribution agreed within approximately 20% over the drop jet mass to the ungroomed jet pT . Both measure-
considered pT range. The largest discrepancies occur at ments are compared to calculations at next-to-leading
low jet masses which is sensitive to the underlying event order with next-to-leading-logarithm and leading order
description, hadronization model and pile-up effects. The with next-to-next-to-leading-logarithm accuracy. Good
normalized cross section after applying the split filtering agreement between the data and the predictions is ob-
algorithm (14) is shown in figure 32 with the mass drop served in resummation regime −3.7 < log10 ρ2 < −1.7.
parameters µfrac = 0.67 and yfilt = 0.09, and a filter- For higher jet masses, where fixed-order effects play an
ing parameter of Rfilt = min(0.3, ∆R/2). After removing important role, the NLO+NLL calculation provides a
soft radiation from the jet which is difficult to model, better description than the LO+NNLL calculation.
the MC prediction is in excellent agreement with the In addition to generic QCD jets, the jet mass has also
data within statistical precision. The CMS Collaboration been measured for boosted top quarks in lepton+jets
performed a similar measurement with anti-kT R = 0.7 tt̄ events collected by the CMS Collaboration at 8
jets using various grooming
√ techniques in selected dijet TeV (116). This measurement is the first jet mass distri-
events using 5 fb−1 of s = 7 TeV data and found as bution unfolded at the particle level probing three prong
well that the agreement between data and the MC predic- decays. Large-R jets are reconstructed with the CA al-
tion improves significantly after grooming techniques are gorithm using a distance parameter of 1.2. The larger
applied (108). Furthermore a measurement of the cross value of R in this measurement compared to the default
section was performed in V +jet final states which over- R = 0.8 applied for top tagging applications in CMS
all show a slightly better data/MC agreement than that is due to an optimization of of statistical precision ver-
observed in dijet events suggesting that the simulation of sus the width of the jet mass distribution at the parti-
quark jets is better than for gluon jets. cle level and the JMR. The number of fully-merged top
The CMS (ATLAS) Collaboration measured the quarks grows with increasing R, but so does the width of
double-differential jet cross section in balanced dijet the jet mass distribution and the suceptibility to pile-up
 29

d2σ (1/GeV) CMS 2.3 fb-1 (13 TeV) 19.7 fb-1 (8 TeV)
0.016

dmjet GeV
650 < p < 760 GeV 2.5

fb
T
0.014 Data CMS Data
Stat. + syst. unc.
0.012
T

Stat. unc.
dσ/dp dmg dp

MADGRAPH+PYTHIA

dσ
0.01 PYTHIA8
HERWIG++
2
0.008
MC@NLO+HERWIG
POWHEG+PYTHIA8
T

Frye et al
1

0.006 POWHEG+PYTHIA
Marzani et al
1.5
0.004

0.002

0 1
2
Theory
Data

1
0
20 30 40 100 200 1000
0.5
Groomed jet mass mg (GeV)

0
FIG. 33 Normalized differential cross section as a function of 150 200 250 300 350
the mass for jets groomed with the soft drop algorithm in data Leading-jet m [GeV]
jet
and for two theoretical calculations. Taken from Ref. (109).

FIG. 35 Particle-level differential tt̄ cross section measure-

ment as a function of the leading jet mass compared to the
)]
)2]
ungroomed 2

0.7 ATLAS Data

predictions for three different Monte-Carlo event generators.
ungroomed

Data
0.6 s= 13 TeV, 32.9 fb-1
ATLAS Pythia 8.1
Pythia 8.1
lead Sherpa 2.1
2.1 Taken from Ref. (116).
T

anti-k t R=0.8, pT > 600 GeV Sherpa

/p

0.50.6 s= drop,
Soft 13 TeV,
β = 1,32.9 -1
zcut =fb0.1
Herwig++ 2.7
Herwig++ 2.7
T
soft drop

/p

LO+NNLL, large NP effects

LO+NNLL, large NP effects
lead LO+NNLL
anti-k t R=0.8, p > 600 GeV
soft10drop

0.4 LO+NNLL
[(m

T NLO+NLL+NP
0.5 NLO+NLL+NP
19.7 fb-1 (8 TeV)
Ratio to Data (1 / σresum) d σ / d log

0.3 Soft drop, β = 2, z = 0.1

(1 / σresum) d σ / d log [(m

cut
GeV-1
0.4
10

0.2 CMS Data

0.1
0.3 0.015 m t = 178.5 GeV
jet
1 dσ
σ dm

1.5 m t = 172.5 GeV

0.2
1
0.5 m t = 166.5 GeV
0.1 −4 −3 −2 −1
0.01
1.5
1
0.5
−4.5 −4 −4 −3.5 − 3−3 −2.5 22
−− −1.5 −−11 −0.5
soft drop ungroomed 2
log [(m /p )]
10 T
0.005

FIG. 34 Comparison of the unfolded log10 ρ2 distribution for

zcut = 0.1, β = 1 in data to various Monte Carlo particle-
level predictions and theory predictions, normalized to the 0
integrated cross section measured in the resummation regime 150 200 250 300 350
−3.7 < log10 ρ2 < −1.7. Taken from Ref. (98). Leading-jet m [GeV]
jet

FIG. 36 Normalized particle-level differential tt̄ cross section

and the underlying event. The leading jet pT is required measurement as a function of the leading jet mass compared
to be above 400 GeV to ensure the hadronic top quark to predictions using three different top quark mass values.
decay to be fully captured within the large-R jet. No Taken from Ref. (116).
substructure selection is applied on the high-pT large-R
jet in order not to bias the jet mass measurement. A re-
quirement of pT > 150 GeV is imposed on the subleading measurements from the ATLAS and CMS Collaborations
jet to select the b quark from the leptonically decaying at high pT . The shape of the jet mass distribution is well
top quark. A veto on additional jets with pT > 150 GeV described by the simulations. The experimental system-
is applied, which results in a fraction of 65% of fully- atic uncertainties are dominated by the uncertainties on
merged top quark decays within the large-R jet. The the jet mass and energy scale, but are smaller than the
particle-level differential tt cross section as a function of uncertainties due to the signal modeling, coming from
the leading jet mass is shown in figure 35. The shown the choice of the top quark mass, the parton showering
simulations predict a larger cross section than observed and the choice of the factorization and renormalization
in the measurement, consistent with the tt cross section scales.
 30

The normalized mass distribution from boosted top uncertainties are the track pT resolution and the choice
quarks, shown in figure 36, can be used to extract the of MC generator used to construct the response matrix
top quark mass. The normalized distribution is used (Pythia 6 versus Herwig++) for the CMS Collaboration
since only the shape can be reliably calculated, and it has whereas the uncertainties on the unfolding procedure, the
the additional benefit that systematic uncertainties par- jet energy resolution at low pT and uncertainties on the
tially cancel. The top quark mass is measured to be mt = tracking at high pT dominate the measurement of the
170.8±6.0 (stat)±2.8 (sys) ±4.6 (model)±4.0 (theo) GeV ATLAS Collaboration. The unfolded jet charge distri-
in agreement with top quark mass measurement in re- bution (κ = 0.6) of the leading jet in data is compared
solved tt̄ events (see e.g. Refs. (202; 203; 204; 205)), al- to the prediction from Powheg+Pythia8 (PH+P8) and
beit with a much larger uncertainty. This constitutes Powheg+Herwig++ (PH+HPP) in figure 38. The dif-
a proof-of-principle, presenting the possibility to extract ferent hadronization and fragmentation model used by
a fundamental SM parameter from a jet mass distribu- Pythia8 and Herwig++ have the largest impact on the
tion. This is of particular interest, as ambiguities arise in jet charge distribution. Variations of the jet charge can
the interpretation of traditional mt measurements (206), also be observed for different PDF sets however the ef-
which can be circumvented by measurements and analyt- fect of the relative flavor fraction in the dijet samples
ical calculations in the highly-boosted
√ regime (207; 208). is significantly smaller than the choice of the showering
Future measurements at s = 13 TeV will allow for a and fragmentation model. It was further found that the
higher statistical precision and, in combination with jet predicted jet charge distribution has a significant depen-
grooming and pile-up mitigation techniques, lead to a dence on the chosen value of αs that describes final state
large improvement in the total precision of the measure- radiation whereas it is insensitive to NLO QCD effect in
ment. Measurements at higher jet pT will facilitate com- the matrix element calculation, color-reconnection and
parisons with analytical calculations. multiple parton interactions. These findings are consis-
tent between the ATLAS and CMS Collaboration.
In addition to studying the sensitivity to various non-
2. Jet Charge perturbative aspects of hadronization and parton distri-
bution functions, the jet charge measurement by ATLAS
The jet charge (209; 210) is defined as the energy includes the first direct comparison of a jet substructure
weighted sum of the electric charges of the jet con- quantity with a perturbative calculation at the LHC. As
stituents it is not collinear safe, the average jet charge is not cal-
X  pT,i κ culable. However, the pT dependence for a particular jet
Qκ = qi , (9) type has been calculated (212; 213). A new technique
pT,J
i∈J
was introduced in Ref. (209) to separately extract the
where qi is the electric charge of particle i and the free average up and down quark jet charge. For a fixed pT ,
parameter κ that controls the sensitivity to soft particles the more forward of the two dijets has a higher energy
within the jet. The ATLAS (CMS) Collaboration mea- and is therefore more likely to be the scattering parton
sured the jet charge for different values of κ using anti-kT with a higher momentum fraction of the proton. In turn,
jets with a radius parameter of R = 0.4 (R = 0.5) in a the higher momentum fraction parton is most likely to be
sample of dijet events. The ATLAS Collaboration distin- a valence quark. Therefore, the fraction of up quark jets
guishes between the two leading jets using the pseudora- is higher for the more forward dijet than the more central
pidity instead of the pT to avoid cases where the lead- dijet. Assuming further that the jet charge is entirely de-
ing particle-level jet is reconstructed as the sub-leading termined by the jet pT and parton origin, one can then
detector-level jet due to the jet energy resolution and solve a system of equations to extract the average up and
to gain sensitivity to different jet flavors. The average down quark jet charge in each bin of jet pT :
jet charge at detector- and particle-level for the more
forward of the leading jets and for κ = 0.5 is shown hQfJ i = fuf hQuJ i + fdf hQdJ i
in figure 37. Due to the increasing fraction of scatter- hQcJ i = fuc hQuJ i + fdc hQdJ i, (10)
ing valence up quark jets (up quark charge > 0), the
average jet charge increases with pT . The difference of where f = forward, c = central, u = up and d = down.
the average jet charge distribution at detector-level and As expected (though not an input), the average up quark
particle-level in figure 37 shows that the unfolding cor- charge is positive and the average down quark charge
rections are large and growing at high pT , due to the loss is negative; furthermore, the latter is roughly half the
of charged-particle tracks inside jets as a result of track former in absolute value. The pT dependence of hQu,d J i
merging. The average jet charge as predicted by Pythia are fit with a logarithmic scale violating term c: hQii =
8 (148) using the Perugia tunes (211) is smaller than that hQi0 (1 + cκ ln(pT,i /pT,0 )), where i represents the pT bin.
observed in data due to a well-known over-estimation of Figure 39 shows the measured and predicted values of
the multiplicity inside jets. The dominating systematic cκ . The uncertainties are large, but there is an indication
 31

0.15

Average Jet Charge [e]

ATLAS 0.1

Scale Violation Parameter

More Forward Jet, κ = 0.5
ATLAS
s = 8 TeV, 20.3 fb-1
0.1 s = 8 TeV, 20.3 fb-1
0.05
CT10 + Pythia8 Flavor Fractions

0.05 0
Detector-level Data
Detector-level Pythia
Particle-level Pythia
-0.05
0
1.2
Detector/Particle Pythia/Data

1
-0.1 2012 Data
0.8
1.5 QCD (Leading Power)
1
-0.15
0.5
0 500 1000 1500 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Jet p [GeV] κ
T

FIG. 37 The detector- and particle-level average jet charge FIG. 39 The measured and predicted value of the average
as a function of jet pT . Reproduced from Ref. (209). jet charge scale violation parameter cκ . Reproduced from
Ref. (209).

-1
CMS 19.7 fb (8 TeV)
1.4
p > 400 GeV
angularity (201; 225).
1.2 T
|η| < 1.5
[1/e]

Data
1 PH + P8 (CT10)
0.6

PH + HPP (CT10)
0.8 B. Measurements with Jet Substructure
1/N dN/dQ

PH + P8 (HERAPDF)
0.6
0.4 While measurements of jet substructure observables
0.2 such as jet mass, jet charge and event shape variables
0
−1.5 −1 −0.5 0 0.5 1 1.5
have been discussed in section V, the following sections
Leading-jet Qκ =0.6 [e] present measurements of other quantities through the ex-
1.1 ploitation of jet substructure techniques such as top tag-
Data
MC

1 ging.
0.9
−1.5 −1 −0.5 0 0.5 1 1.5
1.5

1. Differential tt Cross Section Measurements

Data
MC

0.5
−1.5 −1 −0.5 0 0.5 1 1.5
The selection cuts applied in traditional tt cross sec-
FIG. 38 Unfolded jet charge distribution for κ = 0.6 in data tion measurements (226; 227; 228; 229; 230; 231; 232)
and MC prediction. Taken from Ref. (210) are chosen to maximize the acceptance and minimize the
. associated uncertainties on the fiducial and total cross
section measurements. The fiducial region is such that
events with top pT below 100 GeV and above 600 GeV
that c < 0 and ∂c/∂κ < 0, as predicted. are under-represented, with the former caused by trig-
ger and reconstruction efficiencies and the latter by colli-
mated decays from large Lorentz boosts. This is evident
3. Other Jet Substructure Observables from figure 40, where a drop in selection efficiency below
100 GeV and above 600 GeV is apparent. This results in
The ATLAS and CMS Collaborations have performed a small number of events being selected with high top
further precision measurements of hadronic jet substruc- quark pT , as seen in the ATLAS Run 1 (7 TeV) mea-
ture in pp collisions, correcting for acceptance and resolu- surement shown in figure 41. This means that a very
tion such as jet and event shapes (214; 215; 216; 217; 218; interesting region in terms of new physics is the least well-
219), charged particle multiplicities (217; 220; 221), the measured. Despite often having similar signal efficiencies
jet fragmentation functions (222; 223), color flow (224) to resolved reconstructed techniques, boosted top tagging
and kT splitting scales, N -subjettiness ratios as well as techniques allow for more precise measurements at high
further substructure variables such as Planar Flow and pT due to their higher background rejection.
 32

10

dσtt/dp t [fb/GeV]
GeV
Fiducial phase-space
s = 7 TeV

1
104 Parton-level hadronic top-quark Data
dEvents Particle-level th POWHEG+PYTHIA CT10+hdamp=∞
POWHEG+PYTHIA CT10+hdamp=mtop
T

T
103
dp
ATLAS Simulation
1 POWHEG+PYTHIA HERAPDF+hdamp=∞
POWHEG+PYTHIA POWHEG+PYTHIA HERAPDF+hdamp=mtop
2
10

10-1
10
ATLAS
s = 8 TeV, 20.3 fb-1
1
-2
10
10-1
2

Pred. / Data
300 400 500 600 700 800 900 1000 1100 1200
10-2 1.5

1.5 1
Normalised ratio

0.5

300 400 500 600 700 800 900 1000 1100 1200
1
Particle top-jet candidate p [GeV]
T

0.5
0 100 200 300 400 500 600 700 800 FIG. 42 Particle-level differential tt̄ cross section measure-
Top p [GeV]
T ment for two different PDF sets and choices of the hdamp
parameters. Taken from Ref. (233).

FIG. 40 The drop in fiducial efficiency at top pT > 600 GeV

when reconstructing top quarks with individual anti-kT , R = √
0.4 jets (resolved reconstruction). Adapted from (229). d12 > 40 GeV (Tagger III, see section VI.C). The recon-
structed pT distribution of the anti-kT R = 1.0 trimmed
jet is unfolded to the parton and particle-level. The mea-
10-1 sured particle-level differential cross section is compared
GeV

ATLAS s = 7 TeV, 4.6 fb-1

pb

Data
stat. ⊕ syst. uncert.
in figure 42 to the predictions of several MC generators
POWHEG+PYTHIA normalized to the NNLO+NNLL inclusive cross section.
dp (th)
dσfid

POWHEG+HERWIG
10-2 Overall good agreement is observed, but a harder pT
T

POWHEG(HERAPDF)+PYTHIA
MC@NLO+HERWIG spectrum is predicted by the simulation than observed
in data with larger discrepancies at high pT . The dif-
ferential cross section measurement is also compared to
10-3 predictions from Powheg+Pythia using either the HER-
APDF (234) or CT10 (235) PDF set and two differ-
ent values for the resummation damping factor hdamp ,
hdamp = mtop and hdamp = ∞. The best data/MC
10-4 agreement is observed when using the HERAPDF set
1.80 200 400 600 800
and hdamp = mtop . For each of the settings, the trend
Expected/Data

1.6 of a harder pT spectrum in simulation compared to data

1.4 persists.
1.2
1 A similar measurement by the CMS Collaboration
0.8 based on 8 TeV data (236) uses the CMSTT algorithm to
0.6 reconstruct boosted top quarks. The unfolded results are
0 200 400 600 800
in agreement with the ATLAS measurement and show a
p (th) [GeV] similar trend between data and simulation, as shown in
T
figure 43.
FIG. 41 The small number of top jets identified at high-pT
results in very coarse cross section measurement when using These measurements extend up to a top quark pT of
the resolved reconstruction technique. Adapted from (229). 1.2 TeV, allowing for higher precision thanks to the usage
of jet substructure techniques. The largest uncertainties
at the highest values of pT in ATLAS and CMS come
The ATLAS Collaboration performed a measurement from the large-R jet energy scale and the extrapolation
of the boosted tt differential cross section as a function of the b-jet calibration to high pT .
of the top quark pT in the lepton+jets channel (233). The parton-level differential cross section in top quark
A least one anti-kT jet, trimmed with Rsub = 0.3 and pT has also been measured in the all-hadronic final state
fcut = 0.05 is required with |η| < 2 and pT > 300 GeV. by the CMS Collaboration using 8 TeV data (237). This
To select events with boosted top quarks, the large-R measurement relies on pruned jets with an N -subjettiness
jet is required to have a mass larger than 100 GeV and and subjet-b tagging requirement to suppress the huge
 33

19.7 fb-1 (8 TeV)

1/σtt ⋅ d σ / d ptt [GeV -1 ]

10
dσ/dp (fb/GeV)
ATLAS Data
CMS Data 10−1 PWG+Py8
Powheg+Pythia6 s = 13 TeV, 36.1 fb-1
Parton level PWG+H7
MadGraph+Pythia6

T
t,2
pt,1 > 500 GeV, p > 350 GeV MG5_aMC@NLO+Py8
MC@NLO+Herwig6 T T
Sherpa 2.2.1
T

1 Stat. uncertainty 10−2 Stat. Unc.

Stat. ⊕ syst. uncertainties

tt
Stat. ⊕ Syst. Unc.

10−3
10-1

10−4
Theory / Data

1.5

1
10−5

Prediction
0.5 1.5
400 500 600 700 800 900 1000 1100 1200

Data
1
Particle-level t jet p (GeV)
T 0.5
0 100 200 300 400 500 600 700 800
FIG. 43 Comparison of the particle-level differential tt̄ cross ptt [GeV]
T
section as a function of the jet pT to three different MC gen-
erators. Taken from Ref. (236). FIG. 44 The normalized differential cross section as a func-
tion of the tt̄ pT as measured by ATLAS in the all-hadronic
channel at 13 TeV. Taken from Ref. (238).
amount of background from QCD dijet production. The
cross section is determined from a maximum likelihood fit
to the jet mass distributions for signal-enriched and sig- cross section as a function of the top jet pT and rapidity
nal depleted regions. This allows for a simultaneous ex- is in good agreement with the different MC predictions.
traction of the tt̄ cross section and the QCD background. Larger discrepancies are observed for the pT of the tt̄
The measurement is in agreement with the results from system as shown in figure 44. The measurement is dom-
the lepton+jets final states, but has somewhat larger sta- inated by the systematic uncertainties on the jet energy,
tistical uncertainties of up to about 40% in the highest mass and substructure scale of the large-R jets, alterna-
pT bin with 0.8 < pT < 1.2 TeV. tive parton shower model and the uncertainties on the b
√
The increased s at Run 2 of the LHC offers the pos- jet identification.
sibility for more precise differential tt̄ cross section mea-
surements in the highly-boosted regime. The tt̄ produc-
tion cross section increased by more than a factor√ of ten 2. W/Z/H Cross Sections
for top quark pT > 400 GeV when going from s = 8 TeV
to 13 TeV. The cross section of boosted W and Z boson pro-
A first measurement based on 3.2 fb−1 of 13 TeV data duction was measured by ATLAS in 4.6 fb−1 of 7 TeV
in the lepton+jets channel has been performed by AT- pp collisions (99). The hadronically decaying W and Z
LAS (232). The measurement extends to pT of 1.5 TeV bosons are reconstructed as one single ungroomed anti-
and a similar trend as at 8 TeV is observed between the kT R = 0.6 jet with pT > 320 GeV, |η| < 1.9 and masses
data and the simulation at high pT . A newer mea- ranging between 50 and 140 GeV. The W and Z sig-
surement of the tt̄ differential cross section in the all- nal is enhanced over the dominating QCD background
hadronic channel is performed by the ATLAS Collabora- by constructing a likelihood discriminant from three sub-
tion with 36.1 fb−1 of 13 TeV data (238). The measure- structure variables; thrust minor (19; 20), sphericity (18)
ment uses trimmed anti-kT R = 1.0 jets with Rsub = 0.2 and aplanarity (18), resulting in a signal efficiency of 56%
and fcut = 0.05. To obtain a flat signal efficiency and a background rejection of 89%. The jet mass distri-
of 50% and a quark/gluon rejection of approximately bution after subtracting the expected background from
17 (10) for pT = 500 (1000) GeV, pT dependent cri- tt̄ events is shown in figure 45. A binned maximum likeli-
teria are applied on the jet mass and τ32 . Further- hood fit to the jet mass distribution is used to extract the
more the two top-tagged large-R jets are required to W/Z jet signal yield and to calculate the inclusive cross
have a b tagged small-R jet within ∆R < 1.0. The section. Only the combined W + Z cross section mea-
event selection results in a signal-to-background ratio of surement is performed in this analysis due to the limited
approximately 3:1. The measured fiducial phase-space jet mass resolution. The combined W +Z cross section is
cross section is σ = 292 ± 7 (stat) ± 76 (sys) fb com- measured to be σW +Z = 8.5 ± 0.8 (stat.) ±1.5 (syst.) pb
pared to the Powheg+Pythia8 prediction of 384 ± 36 fb and is in agreement with the Standard Model prediction
at NNLO+NNLL. The measured normalized differential of σW +Z = 5.1 ± 0.5 pb within 2 standard deviations.
 34

35.9 fb-1 (13 TeV)

Jets / 2 GeV

Events / 7 GeV
Data 2011 8000

Data - Fit bkg

25000
1800
1600
CMS 450 < p < 1000 GeV
T
W
Signal + Background fit
1400
7000 double-b tagger Z
1200 tt
Background fit component
1000 passing region
Multijet
Signal fit component 800 6000
Total background
20000 600
400
H(bb)
200 5000 Data
0
-200
50 60 70 80 90 100 110 120 130 140
4000
15000 Jet Mass [GeV]

ATLAS 3000

s = 7 TeV, 4.6 fb -1 2000

10000
p > 320 GeV |η| < 1.9
T
1000
L > 0.15
5000 0

Data − multijet − tt
10
5

σData
0
50 100 150 200 250 0
−5
40 60 80 100 120 140 160 180 200
Jet Mass [GeV] mSD (GeV)

FIG. 45 Binned maximum likelihood fit to the jet mass distri- FIG. 46 Soft drop jet mass mSD of anti-kT R = 0.8 jets
bution in data for selected W/Z events reconstructed as one in data and for the dominating background processes; mul-
single ungroomed anti-kT R = 0.6 jet. Taken from ref. (99). tijet production and W/Z+jets events. Jets are required to
pass criteria on N21 and to be identified as double-b jets by
the double-b tagger introduced in section VI.D. Taken from
The dominating systematic uncertainties are the jet mass Ref. (241).
resolution and the choice of the QCD background PDF.
The signal significance was furthermore studied when us-
ing groomed jets instead of ungroomed jets. Without an by utilizing the double-b tagger. The W/Z+jets back-
optimization of the analysis for groomed jets, similar sig- ground is estimated from MC simulation and the shape
nificances were observed for groomed and ungroomed jets of the multijet background is determined in a validation
as expected due to the low number of pile-up vertices in region in data with lower values of the double-b tagger
the 7 TeV dataset. discriminator. The soft drop mass distribution of the
As discussed in section VI.D the SM Higgs boson leading jet is shown in figure 46 with a clear resonant
decays with approximately 58% into bb̄. However the structure at the mass of the W and Z boson. The SM
H → bb̄ decay in the resolved channel can only be stud- background processes and the potential signal from SM
ied in associated production with either a vector boson H → bb̄ production are estimated simultaneously. The
(W/Z) (239; 240), top quarks, or via the vector-boson- observed (expected) significance for the H → bb̄ process
fusion production mechanism due to the overwhelming is 1.5(0.7)σ. The measured cross section for the Z+jets
multijet background. To search for H → bb̄ in the gluon- process is 0.85 ± 0.16 (stat.) +1.0
−0.4 (syst.) pb which is in

gluon fusion production mode with an additional high- agreement with the SM prediction of 1.09±0.11 pb. This
pT jet, jet substructure techniques can be employed to is the first observation of Z → bb̄ in the single jet topol-
suppress the enormous multijet background. The CMS ogy.
Collaboration performed a search for the SM Higgs bo- The ATLAS Collaboration also measured the high pT
son using a dijet topology with 35.9 fb−1 of 13 TeV pp Z → bb̄ cross section using two nearby b tagged anti-kT
collisions (241). The analysis uses anti-kT R = 0.8 jets R = 0.4 jets (instead of one large-radius jet) in 19.5 fb−1
corrected with the PUPPI algorithm to reduce the effects of 8 TeV pp collisions (242). The measured fiducial cross
from pile-up, and modified with the soft drop algorithm section was determined to be σZ→bb̄ = 2.02 ± 0.33 pb
(β = 0, zcut = 0.1) to mitigate the effects from the under- which is in excellent agreement with the next-to-leading-
lying event and soft/wide-angle radiation. At least one order theoretical predictions.
large-R jet with pT > 450 GeV and |η| < 2.5 is required.
To distinguish the two prong structure of a jet containing
the full H → bb̄ decay from quark- or gluon-initiated jets, VIII. SEARCHES FOR NEW PHYSICS
the N21 variable, calculated from the generalized energy
correlation functions, is exploited. To ensure a flat QCD Jet substructure methods have been successfully ap-
background rejection of 26% over the considered mass plied in a large variety of searches for physics beyond
and pT range, a decorrelation procedure (167) is applied the SM. The respective exclusion limits are substantially
to N21 . The multijet background is further suppressed improved through the application of these methods. In
 35

some cases the decay signature of heavy BSM particles

35.9 fb-1 (13 TeV)
would not be accessible without the application of jet 105

Events / 100 GeV

substructure methods. CMS CMS data
As the number of such BSM searches is very large, 104 2 par. background fit
only a small subset of the published results can be dis- W'(2 TeV)→WZ (σ = 0.01 pb)
cussed here. The following sections give an overview of 103 WZ, high-purity
a selection of searches for tt resonances (186; 243; 244; |η| ≤ 2.5, pT > 200 GeV
245; 246; 247; 248), diboson resonances (115; 161; 162; mjj > 1050 GeV, |∆ηjj| ≤ 1.3
102
163; 164; 165; 249; 250; 251; 252; 253; 254; 255; 256; 257;
258; 259; 260; 261; 262; 263; 264), vector-like quarks (265; 10
266; 267; 268; 269; 270; 271; 272; 273; 274; 275; 276; 277;
278; 279; 280) and leptophobic Z 0 (160; 281). Further
1
searches using jet substructure techniques can be found
in Refs. (102; 145; 182; 187; 282; 283; 284; 285; 286; 287; 2 1200 14001600 1800 2000 22002400 2600 2800 3000

Data-Fit
[Link]
288; 289; 290; 291; 292; 293; 294; 295; 296; 297; 298; 299; Dijet invariant mass (GeV)
0
300; 301; 302; 303; 304; 305; 306; 307).
−2
1500 2000 2500 3000
Dijet invariant mass (GeV)
A. Diboson Resonances
FIG. 47 Dijet invariant mass distribution in the high purity
Several new physics models predict resonances cou- W Z category of the fully hadronic W W /W Z/ZZ resonance
pling strongly to vector bosons to play a role in the can- search. The fit under the background-only hypothesis is over-
cellation of large corrections to the Higgs mass. These layed. Taken from Ref. (115).
models include extensions of the SM Higgs doublet,
where the simplest realizations are two-Higgs-doublet
models (308) with heavy, neutral Higgs bosons, which have one HP V jet, and are divided into HP and LP
can have large branching fractions to top quarks and events, depending on whether the other V jet is of high
W /Z/H bosons. Alternatives are composite Higgs mod- or low purity. To further suppress the large QCD mul-
els (309; 310; 311; 312; 313; 314; 315; 316) or Randall- tijet background a requirement on the dijet kinematics
Sundrum Kaluza-Klein models (317; 318; 319; 320). |η1 − η2 | < 1.3 is applied.
Searches for new resonances generally focus at high The background is estimated from a sig-
masses with m > 1 TeV such that the SM bosons re- nal+background fit with the function dm dN
jj
= (mjj /P√
0
s)P1
,
ceive high Lorentz boosts. In more than 60% of the where P0 is a normalization parameter and P1 is a pa-
cases, W /Z/H bosons decay into a quark anti-quark rameter describing the shape. This parametrization
pair, which makes the reconstruction of such decays with has been tested and validated on simulated events and
jet substructure techniques an essential ingredient for on data in a control region. As shown in figure 47 the
these searches. In the following, the analysis strategies data in the signal region is well described by the fit
and results from CMS and ATLAS using pp collision data function. Figure 47 also shows that no excess over the
√
with s = 13 TeV are discussed. background-only hypothesis is observed.
The searches for diboson resonances are performed A similar analysis has been performed by the ATLAS
in semi-leptonic (257; 262) and fully hadronic final Collaboration (165). In this analysis events are required
states (115; 163; 164; 165). As the methods of jet sub- to have at least two large-R jets with pT > 200 GeV in
structure analyses exhibit their full strength in hadronic the pseudo-rapidity range |η| < 2.0. These jets are re-
final states, the following discussion gives a summary and constructed with the anti-kT algorithm with a radius pa-
comparison of the ATLAS and CMS results in the search rameter R = 1.0. The trimming algorithm is applied
for W /Z resonances in hadronic final states only. using kT subjets with R = 0.2. The rapidity separation
In an analysis performed by the CMS Collabora- between the two leading jets has to satisfy |∆y12 | < 1.2.
tion (115) events with two anti-kT jets with R = 0.8, cor- The large-R jet mass is computed from the Combined
rected with the PUPPI algorithm, and 65 < msoft drop < Mass (see section V.B), and is required to be within a
105 GeV are selected. The jet is considered to be a W bo- window of the expected W or Z mass value. The window
son candidate if the mass is in the range 65–85 GeV, while width varies from 22 to 40 GeV depending on the jet pT .
it is a Z boson candidate if the mass is in the range 85– In addition, the D2β=1 variable is used to select jets with
105 GeV. This leads to the three signal categories W W , a two-prong structure.
ZZ and W Z. The jets are further categorized according Similar as in the CMS analysis, the background is es-
to τ21 into high purity (HP, τ21 < 0.35) and low purity timated by fitting the dijet invariant mass distribution
(LP, 0.35 < τ21 < 0.75). Events are always required to with the parametric form dn dx = p1 (1−x)
p2+ξp3 p3
x , where
 36

variable isolation cone that changes as a function of the

103 transverse momentum (323). Interestingly, studies per-
Events / 0.1 TeV ATLAS
s = 13 TeV, 36.7 fb-1 formed in CMS for 13 TeV show that the CMS imple-
102
WZ SR mentation of such a variable isolation criterion is not as
χ2/DOF = 8.1/9
10 powerful as the selection based on ∆R(lepton, jet) and
Data
prel
T (324).
1 Fit To reconstruct the boosted hadronic top decay, the
Fit + HVT model B m=1.5 TeV
Fit + HVT model B m=2.4 TeV presence of a single high-momentum, large-R, top-tagged
10−1
jet is required. In CMS (ATLAS) the large-R jet is re-
Significance

1500 2000 2500 3000 3500

2
constructed with the CA (anti-kT ) algorithm with a size
0
parameter of R = 0.8 (1.0). The selection requirement on
−2
1.5 2 2.5 3 3.5 the transverse momentum is pT > 400(300) GeV. ATLAS
mJJ [TeV] applies trimming to the large-R jets with the parameters
fcut = 0.05 and Rsub = 0.3 and the √ jets are required to
FIG. 48 The observed data in the signal region of the W Z have a mass mjet > 100 GeV and d12 > 40 GeV. The
category. Also shown is the fitted background prediction. The
strategy followed by CMS is to apply the CMSTT algo-
gray region represents the uncertainty in the background es-
timate. Taken from Ref. (165). rithm (as defined in section VI.C), where the mass of the
jet has to satisfy 140 < mjet < 250 GeV. In addition, the
N -subjettiness ratio τ32 must be smaller than 0.7.
n is the number of events, x is a dimensionless variable The variable of interest is the invariant mass mtt̄ of the
√ tt̄ system. It is reconstructed from the top-tagged large-
related to the dijet mass mJJ , x = mJJ / s, p1 is a nor-
malization factor, p2 and p3 are dimensionless shape pa- R jet, a b tagged small-R jet as well as the lepton and
rameters, and ξ is a constant chosen to remove the cor- the missing energy. Once the top-pair system is recon-
relation between p2 and p3 in the fit. structed, events are further divided into categories based
The dijet invariant mass distributions for these events on the lepton flavor and the number of b-tagged and top-
are shown in figure 48, where good agreement is found tagged jets. This gives several analysis categories with
between data and the expectations from the background different background compositions: the top-tagged and
fit. b-tagged events are dominated by the SM tt̄ background,
In case of boosted H bosons, different reconstruction while events without top tags and b tags are mostly com-
methods have to be used to benefit from the presence of posed of W +jets events.
b quarks in H → bb̄ decays (see section VI.D). Results Similar methods are applied in case both W bosons de-
have been published on the search for for W H/ZH final cay hadronically (244). To access the region with jets of
states (164; 258; 263) as well as for HH final states (259; lower momenta with 200 < pT < 400 GeV a dedicated al-
260; 264). gorithm with a larger jet size parameter of R = 1.5 (CA15
jets) is applied in CMS. The larger jet size extends the
analysis coverage to the case of intermediate or smaller
B. tt̄ Resonances Lorentz boosts. These low-pT jets are required to be
identified by the HEPTopTagger algorithm (as described
The models of new physics mentioned in the previ- in section VI.C). This approach improves the sensitivity
ous section also predict resonances decaying to pairs of for smaller masses of the hypothetical tt resonance.
top quarks. An example for an alternative model is the Even with the requirement of two top-tagged jets, the
topcolor model which contains a Z 0 boson (321), with event sample is dominated by QCD dijet events. This
exclusive decays to top quarks. background is estimated using a data-driven technique,
In case of boosted t → bW events with leptonic W where an anti-tag and probe method is used. The τ32 re-
boson decays, the lepton may overlap with the associ- quirement is reversed on one jet to select a sample dom-
ated b quark jet. Therefore, the usual lepton-isolation inated by QCD events. The opposite jet is then used to
criteria, which are used to mitigate the contamination measure the misidentification rate for the top-tagging re-
with QCD multijet background, are relaxed. The CMS quirements. The measured misidentification rate ranges
and ATLAS Collaborations follow different strategies from 5 to 10%, depending on the jet momentum, τ32 and
for this purpose. In CMS (244; 248), the lepton must the b tagging requirements applied. This differential rate
have a large angular separation from the associated b is used in a sample of single top-tagged events to pre-
jet candidate of ∆R(lepton, jet) > 0.5 or it must have a dict the double top-tagged event contribution from QCD
transverse momentum relative to the jet axis prel
T above processes in each individual event category. Closure tests
25 GeV. This requirement removes background contri- performed in data and simulation are performed to val-
butions from semi-leptonic B hadron decays. In AT- idate the background estimation for each of the signal
LAS (186; 322), the lepton isolation is achieved by a regions.
 37

µ+jets, 1 t tag 2.6 fb-1 (13 TeV) 19.7 fb-1 (8 TeV)

Events / 100 GeV

Events
CMS Data CMS
Other 106 QCD (MC)
2 Simulation
10 tt tt
Z’ 2.0 TeV, 1% width 105 TT→tHtH (500 GeV/c2)

TT→tHtH (700 GeV/c2)

104
10 TT→tHtH (1000 GeV/c2)
3
10

102
1
10
Data / Bkg

1.5
1
1
10-1
0.5
0 1000 2000 3000 0 1 ≥2
Mtt [GeV] Multiplicity of H tags

FIG. 49 Invariant mass of the reconstructed tt̄-pair in data FIG. 50 Multiplicity of CA15 jets which fulfill the Higgs tag-
and simulation for the lepton+jets channel in the category ging criteria. The solid histograms represent the simulated
with one top-tagged jet, taken from Ref. (248). background processes (tt and QCD multijet). The hatched
error bands show the statistical uncertainty of the simulated
events. Taken from Ref. (266).
No significant excess above the predicted background
is observed in the measured tt̄ invariant mass spectrum.
Figure 49 shows the mtt̄ spectrum in the analysis cate-
The first search for VLQs in the all-hadronic final state
gory with the highest S/B fraction.
(266) targeted the T → tH decay mode. The CA algo-
Depending on the model, narrow tt̄ resonances are ex-
rithm with a large size parameter of R = 1.5 was applied
cluded for masses less than approximately 4 TeV. The
to cluster top quarks and Higgs bosons in single large
exclusion limits are weaker for scenarios with large width
jets. To identify the origin of the large CA jets a top
of the resonance.
tagging algorithm (HEPTopTagger) and a Higgs tagging
algorithm based on subjet-b tagging (see section VI.D)
are used. This was the first time these two algorithms
C. Vector-like Quarks
have been applied in a data analysis by the CMS Collab-
oration. Two subjets must be b tagged and their invari-
Vector-like quarks (VLQs) are predicted by a va-
ant mass must be greater than 60 GeV to fulfill the Higgs
riety of theories introducing a mechanism that stabi-
tagging requirement. The multiplicity of these Higgs tags
lizes the mass of the Higgs particle. Such theories in-
is shown in figure 50 which demonstrates that both the
clude little Higgs models (325; 326), models with ex-
QCD multijet and the tt backgrounds can be suppressed
tra dimensions (327; 328), and composite Higgs mod-
by several orders of magnitudes.
els (327; 328; 329). As VLQs are expected to have large
masses and have top quarks and vector-bosons as decay Extensive use of substructure methods has also been
products, jet substructure analyses have been applied in made by the ATLAS Collaboration, in particular for the
many searches for VLQs. search for single production of VLQs. The single pro-
The first search for VLQs using jet substructure duction modes may have higher cross sections than pair
methods was an inclusive search for pair-produced T production depending on the VLQ mass and the coupling
quarks (265). As VLQs may have many decay modes parameters (330). ATLAS performed an analysis (277)
(T → bW , T → tZ, T → tH, B → tW , B → bZ, where the VLQ is searched for in the decay mode with a
B → bH), a large variety of final states needs to be W boson and a top quark (B → tW ). Final states with
explored. For this reason, an inclusive search has been at least one lepton are considered, where either the W
performed without the attempt to reconstruct a specific boson or the top quark appear in a boosted configura-
decay chain. The CA algorithm was used with a dis- tion. They are identified by the application of a jet mass
tance parameter R = 0.8 (CA8 jets). Boosted W jets are requirement (m > 50 GeV) on a trimmed large-R anti-kT
identified based on the mass of the CA8 jet while boosted jet with a distance parameter R = 1.0.
top jets are identified with the CMSTT, described in sec- A different strategy is followed in another ATLAS
tion VIII.B. search (278), where the decay into the bW final state
 38

is investigated (T /Y → bW ). As the W boson is as- 35.9 fb-1 (13 TeV)

Events / 5 GeV
sumed to decay leptonically, no boosted hadronic W or 2500
Data W(qq)+jets (×3)
CMS
top quark decays are present. Therefore, the analysis Total SM pred. Z(qq)+jets (×3)
uses a veto on the presence of massive (m > 70 GeV), 2000 Multijet pred. t/tt(qq)+jets (×3)
Z'(qq), g =0.17, mZ'=135 GeV
trimmed large-R anti-kT jets with R = 1.0, to suppress q'

the dominant tt background. 1500 p : 700-800 GeV

T

Today, jet substructure methods are widely employed

in almost all VLQ searches published by the LHC Collab- 1000
orations, see e.g. Refs. (269; 270; 271; 272; 273; 274; 279).
The excluded VLQ masses are exceeding 1 TeV for all 500
branching fractions, thanks to jet substructure tech-
niques. 0

Data/Prediction
1.2 60 80 100 120 140 160 180 200 220 240
1
0.8 mSD (GeV)
D. Leptophobic Z 0 50 100 150 200 250
mSD (GeV)

Besides resonaces coupling to heavy SM particles, there FIG. 51 Soft drop jet mass of anti-kt R = 0.8 jets in data and
exist predictions for resonances that couple to quarks and for the dominating background processes; multijet production
gluons (331; 332; 333; 334), including simplified Dark and W/Z+jets events. Taken from Ref. (160).
Matter (DM) models in which resonances couple only to
quarks and DM particles (335; 336; 337). When the new ×10
3

particle (such as a Z 0 ) is sufficiently light (mZ 0  1 TeV),

Events / 2 GeV
ATLAS Data
it can be boosted when produced in association with 140
W + jets
s = 13 TeV, 36.1 fb-1
initial-state radiation and thus entirely captured by a sin- 120 Z + jets
W/Z validation
gle large-radius jet (160; 281). Searching in this mode can Jet channel
Background est.
100 Bkg. stat. uncert.
significantly extend the sensitivity of the existing search
Bkg. syst. uncert.
program, where resolved low-mass resonance searches 80
typically degrade due to high trigger thresholds and the
60
enormous QCD multijet background.
Both ATLAS and CMS have used this strategy to look 40

for boosted Z 0 jets. Jets in the CMS analysis are re- 20

SR
constructed with the anti-kT algorithm with R = 0.8
0
and corrected for effects from pile-up and the under-
Data - background est.

lying event with PUPPI and the soft drop algorithm

4000
(β = 0, zcut = 0.1) whereas anti-kT R = 1.0 jets,
trimmed with Rsub = 0.2 and fcut = 5% are used in
2000
ATLAS. To suppress the dominating QCD multijet back-
ground, CMS applies criteria on N21 (120) and ATLAS
chooses τ21 as discriminator. To avoid distortions of 0

the jet mass spectrum due to large correlation between 50 60 70 80 90 100 110
the jet mass and substructure variables, a decorrelation Large-R jet mass [GeV]
with the DDT method is applied. Data-driven techniques
are used to determine the dominating background from
QCD multijet production. Subdominant processes such FIG. 52 Trimmed jet mass distribution anti-kT R = 1.0 jets
as W/Z+jets events are estimated from MC simulation. in data and for the dominating background processes. Taken
The jet mass distributions of the large-R jet is shown in from Ref. (281).
figure 51 and 52 for the CMS and ATLAS analyses, re-
spectively. No evidence for a resonant structure on top
of the SM background is observed. methods are used to exploit the details of hadronic ac-
tivity detectable by modern particle detectors such as
ATLAS and CMS, and precision Standard Model mea-
IX. CONCLUSIONS surements and searches for physics beyond the Standard
Model at both these experiments increasingly rely on one
Jet substructure is the term used to describe the cal- or more of the tools developed by the jet substructure
culations, algorithms, and analysis techniques developed community. With increasingly sophisticated hardware
over the last decade and reviewed in this article. These and software capabilities, jet substructure techniques of
 39

the future will grow in complexity and utility, further [14] J. M. Butterworth, A. R. Davison, M. Rubin and G. P.
empowering the exploration of the subnuclear properties Salam, Jet substructure as a new Higgs search channel
of nature. at the LHC, Phys. Rev. Lett. 100 (2008) 242001,
[0802.2470].
[15] D. E. Kaplan, K. Rehermann, M. D. Schwartz and
B. Tweedie, Top Tagging: A method for identifying
ACKNOWLEDGMENTS boosted hadronically decaying top quarks, Phys. Rev.
Lett. 101 (2008) 142001, [0806.0848].
Much of the work in this field in recent years has been [16] A. J. Larkoski, I. Moult and B. Nachman, “Jet
galvanized by the Boost Workshop Series (338; 339; 340), substructure at the Large Hadron Collider: A review of
which continues to inspire fruitful collaborations between recent advances in theory and machine learning.” 2017.
[17] G. Hanson, G. S. Abrams, A. M. Boyarski,
experimentalists and theorists.
M. Breidenbach, F. Bulos, W. Chinowsky et al.,
The editors thank CERN and the ATLAS and CMS Evidence for jet structure in hadron production by
Collaborations, the participants and organizers of the e+ e− annihilation, Phys. Rev. Lett. 35 (1975) 1609.
Boost Workshops held in Zurich 2016 (341) and Buffalo [18] J. D. Bjorken and S. J. Brodsky, Statistical model for
2017 (342) for discussions and input, and Jon Butter- electron-positron annihilation into hadrons, Phys. Rev.
worth for suggesting this jet substructure review article. D1 (Mar, 1970) 1416.
We also thank Andrew Larkoski and Ian Moult for the [19] S. Brandt, C. Peyrou, R. Sosnowski and
A. Wroblewski, The principal axis of jets — an
collaboration on the theoretical review.
attempt to analyse high-energy collisions as two-body
processes, Phys. Lett. 12 (1964) 57.
[20] E. Farhi, Quantum chromodynamics test for jets, Phys.
REFERENCES Rev. Lett. 39 (1977) 1587.
[21] G. Sterman and S. Weinberg, Jets from Quantum
[1] JADE Collaboration, Observation of planar three jet Chromodynamics, Phys. Rev. Lett. 39 (Dec, 1977)
events in e+ e− annihilation and evidence for gluon 1436.
bremsstrahlung, Phys. Lett. B91 (1980) 142. [22] G. P. Salam and G. Soyez, A practical seedless
[2] PLUTO Collaboration, Evidence for gluon infrared-safe cone jet algorithm, JHEP 05 (2007) 086,
bremsstrahlung in e+ e− annihilations at high energies, [0704.0292].
Phys. Lett. B86 (1979) 418. [23] JADE Collaboration, Experimental studies on multijet
[3] MARK-J Collaboration, Discovery of Three Jet Events production in e+ e− annihilation at PETRA energies,
and a Test of Quantum Chromodynamics at PETRA Z. Phys. C33 (1986) 23.
Energies, Phys. Rev. Lett. 43 (1979) 830. [24] JADE Collaboration, Experimental investigation of the
[4] TASSO Collaboration, Evidence for planar events in energy dependence of the strong coupling strength,
e+ e− annihilation at high energies, Phys. Lett. B86 Phys. Lett. B213 (1988) 235.
(1979) 243. [25] S. Catani, Y. Dokshitzer, M. Olsson, G. Turnock and
[5] D0 Collaboration, Observation of the top quark, Phys. B. Webber, New clustering algorithm for multijet cross
Rev. Lett. 74 (1995) 2632, [hep-ex/9503003]. sections in e+ e− annihilation, Phys. Lett. B269 (1991)
[6] CDF Collaboration, Observation of top quark 432.
production in p̄p collisions with the Collider Detector [26] S. Catani, Y. Dokshitzer and B. Webber, The
at Fermilab, Phys. Rev. Lett. 74 (1995) 2626, k⊥ -clustering algorithm for jets in deep inelastic
[hep-ex/9503002]. scattering and hadron collisions, Phys. Lett. B285
[7] ATLAS Collaboration, Observation of Higgs boson (1992) 291.
production in association with a top quark pair at the [27] S. Catani, Y. Dokshitzer, M. Seymour and B. Webber,
LHC with the ATLAS detector, Phys. Lett. B784 Longitudinally-invariant k⊥ -clustering algorithms for
(2018) 173, [1806.00425]. hadron-hadron collisions, Nucl. Phys. B406 (1993)
[8] CMS Collaboration, Observation of ttH production, 187.
Phys. Rev. Lett. 120 (2018) 231801, [1804.02610]. [28] J. Gallicchio and Y.-T. Chien, “Quit Using
[9] ATLAS Collaboration, Observation of H → bb̄ decays Pseudorapidity, Transverse Energy, and Massless
and V H production with the ATLAS detector, Phys. Constituents.” 2018.
Lett. B786 (2018) 59–86, [1808.08238]. [29] Y. L. Dokshitzer, G. D. Leder, S. Moretti and B. R.
[10] CMS Collaboration, Observation of Higgs boson decay Webber, Better jet clustering algorithms, JHEP 08
to bottom quarks, Phys. Rev. Lett. 121 (2018) 121801, (1997) 001, [hep-ph/9707323].
[1808.08242]. [30] M. Wobisch and T. Wengler, Hadronization corrections
[11] M. H. Seymour, Searches for new particles using cone to jet cross-sections in deep inelastic scattering, in
and cluster jet algorithms: A comparative study, Z. Monte Carlo generators for HERA physics.
Phys. C62 (1994) 127. Proceedings, Workshop, Hamburg, Germany,
[12] J. M. Butterworth, B. E. Cox and J. R. Forshaw, W W 1998-1999, p. 270, 1998, hep-ph/9907280.
scattering at the CERN LHC, Phys. Rev. D65 (2002) [31] M. Cacciari, G. P. Salam and G. Soyez, The anti-kt jet
096014, [hep-ph/0201098]. clustering algorithm, JHEP 04 (2008) 063,
[13] K. Agashe, A. Belyaev, T. Krupovnickas, G. Perez and [0802.1189].
J. Virzi, LHC Signals from Warped Extra Dimensions, [32] G. P. Salam, Towards Jetography, Eur. Phys. J. C67
Phys. Rev. D77 (2008) 015003, [hep-ph/0612015]. (2010) 637, [0906.1833].
 40

[33] R. K. Ellis, W. J. Stirling and B. R. Webber, QCD performance using particle flow with the ATLAS
and collider physics, Camb. Monogr. Part. Phys. Nucl. Detector, Eur. Phys. J. C77 (2017) 466, [1703.10485].
Phys. Cosmol. 8 (1996) 1–435. [55] ATLAS Collaboration, Topological cell clustering in
[34] S. D. Ellis, Z. Kunszt and D. E. Soper, Jets at hadron the ATLAS calorimeters and its performance in LHC
colliders at order αs3 : A look inside, Phys. Rev. Lett. Run 1, Eur. Phys. J. C77 (2017) 490, [1603.02934].
69 (1992) 3615, [hep-ph/9208249]. [56] ATLAS Collaboration, Improving jet substructure
[35] OPAL Collaboration, A direct observation of quark - performance in atlas using track-caloclusters, ATLAS
gluon jet differences at LEP, Phys. Lett. B265 (1991) PUB Note ATL-PHYS-PUB-2017-015, 2017.
462. [57] CMS Collaboration, V Tagging Observables and
[36] ALEPH Collaboration, Study of the subjet structure of Correlations, CMS Physics Analysis Summary
quark and gluon jets, Phys. Lett. B346 (1995) 389. CMS-PAS-JME-14-002, 2014.
[37] DELPHI Collaboration, Energy dependence of the [58] ATLAS Collaboration, A measurement of the
differences between the quark and gluon jet calorimeter response to single hadrons and
fragmentation, Z. Phys. C70 (1996) 179. determination of the jet energy scale uncertainty using
[38] L3 Collaboration, Ks0 and Λ production in quark and LHC Run-1 pp-collision data with the ATLAS detector,
gluon jets at LEP, Phys. Lett. B407 (1997) 389. Eur. Phys. J. C77 (2017) 26, [1607.08842].
[Erratum: Phys. Lett. B427, 409 (1998)]. [59] ATLAS Collaboration, Charged-particle
√ distributions
[39] CDF Collaboration,
√ A Measurement of jet shapes in pp̄ at low transverse momentum in s = 13 TeV pp
collisions at s = 1.8 TeV, Phys. Rev. Lett. 70 (1993) interactions measured with the ATLAS detector at the
713. LHC, Eur. Phys. J. C76 (2016) 502, [1606.01133].
[40] D0 Collaboration, Transverse √ energy distributions [60] ATLAS Collaboration, Performance of the ATLAS
within jets in pp̄ collisions at s = 1.8 TeV, Phys. Transition Radiation Tracker in Run 1 of the LHC:
Lett. B357 (1995) 500. tracker properties, JINST 12 (2017) P05002,
[41] ZEUS Collaboration, Measurement of jet shapes in [1702.06473].
photoproduction at HERA, Eur. Phys. J. C2 (1998) [61] ATLAS Collaboration, A neural network clustering
61, [hep-ex/9710002]. algorithm for the ATLAS silicon pixel detector, JINST
[42] ZEUS Collaboration, Measurement of jet shapes in 9 (2014) P09009, [1406.7690].
high Q2 deep inelastic scattering at HERA, Eur. Phys. [62] ATLAS Collaboration, Performance of the ATLAS
J. C8 (1999) 367, [hep-ex/9804001]. Track Reconstruction Algorithms in Dense
[43] H1 Collaboration, Measurement of internal jet Environments in LHC Run 2, Eur. Phys. J. C77
structure in dijet production in deep inelastic (2017) 673, [1704.07983].
scattering at HERA, Nucl. Phys. B545 (1999) 3, [63] ATLAS Collaboration, Measurement of track
[hep-ex/9901010]. reconstruction inefficiencies in the core of√jets via pixel
[44] CMS Collaboration, Jet energy scale and resolution in dE/dx with the ATLAS experiment using s = 13 TeV
the CMS experiment in pp collisions at 8 TeV, JINST pp collision data, ATLAS PUB Note
12 (2017) P02014, [1607.03663]. ATL-PHYS-PUB-2016-007, 2016.
[45] ATLAS Collaboration, The ATLAS Experiment at the [64] ATLAS Collaboration, Modelling of Track
CERN Large Hadron Collider, JINST 3 (2008) S08003. Reconstruction
√ Inside Jets with the 2016 ATLAS
[46] CMS Collaboration, The CMS experiment at the s = 13 TeV pp Dataset, ATLAS PUB Note
CERN LHC, JINST 3 (2008) S08004. ATL-PHYS-PUB-2017-016, 2017.
[47] ATLAS Collaboration, The ATLAS Inner Detector [65] ATLAS Collaboration, Measurement of performance of
commissioning and calibration, Eur. Phys. J. C70 the pixel neural network√ clustering algorithm of the
(2010) 787–821, [1004.5293]. ATLAS experiment at s = 13 TeV, ATLAS PUB
[48] CMS Collaboration, Description and performance of Note ATL-PHYS-PUB-2015-044, 2015.
track and primary-vertex reconstruction with the CMS [66] CMS Collaboration, High pT jet tracking with JetCore
tracker, JINST 9 (2014) P10009, [1405.6569]. tracking and pixel cluster splitting, CMS Detector
[49] ATLAS IBL Collaboration, Production and Integration Performance Summary , 2014.
of the ATLAS Insertable B-Layer, JINST 13 (2018) [67] ATLAS Collaboration, Jet energy measurement and its
T05008, [1803.00844]. systematic
√ uncertainty in proton-proton collisions at
[50] CMS Collaboration, The CMS hadron calorimeter s = 7 TeV with the ATLAS detector, Eur. Phys. J.
project: Technical Design Report, Tech. Rep. C75 (2015) 17, [1406.0076].
CERN-LHCC-97-031, Geneva, 1997. [68] ATLAS Collaboration, Jet energy scale measurements
[51] CMS Collaboration, Particle-Flow Event and their systematic
√ uncertainties in proton-proton
Reconstruction in CMS and Performance for Jets, collisions at s = 13 TeV with the ATLAS detector,
Taus, and MET, CMS Physics Analysis Summary Phys. Rev. D96 (2017) 072002, [1703.09665].
CMS-PAS-PFT-09-001, 2009. [69] CMS Collaboration, Determination of Jet Energy
[52] CMS Collaboration, Commissioning of the Calibration and Transverse Momentum Resolution in
Particle-Flow reconstruction in Minimum-Bias and Jet CMS, JINST 6 (2011) P11002, [1107.4277].
Events from pp Collisions at 7 TeV, CMS Physics [70] A. Cukierman and B. Nachman, Mathematical
Analysis Summary CMS-PAS-PFT-10-002, 2010. Properties of Numerical Inversion for Jet Calibrations,
[53] CMS Collaboration, Particle-flow reconstruction and Nucl. Instrum. Meth. A858 (2017) 1, [1609.05195].
global event description with the CMS detector, JINST [71] ATLAS Collaboration, Jet global sequential corrections
12 (2017) P10003, [1706.04965]. with
√ the ATLAS detector in proton-proton collisions at
[54] ATLAS Collaboration, Jet reconstruction and s = 8 TeV, ATLAS CONF Note
 41

ATLAS-CONF-2015-002, 2015. Pruning as a tool for heavy particle searches, Phys.

[72] CMS Collaboration, Jet energy scale uncertainty Rev. D81 (2010) 094023, [0912.0033].
correlations between ATLAS and CMS, CMS Physics [92] A. J. Larkoski, S. Marzani, G. Soyez and J. Thaler,
Analysis Summary CMS-PAS-JME-14-003, 2014. Soft Drop, JHEP 05 (2014) 146, [1402.2657].
[73] M. Cacciari, G. P. Salam and G. Soyez, The catchment [93] M. Dasgupta, A. Fregoso, S. Marzani and G. P. Salam,
area of jets, JHEP 04 (2008) 005, [0802.1188]. Towards an understanding of jet substructure, JHEP
[74] CMS Collaboration, Pileup Removal Algorithms, CMS 09 (2013) 029, [1307.0007].
Physics Analysis Summary CMS-PAS-JME-14-001, [94] ATLAS Collaboration, Identification of boosted,
2014. hadronically decaying W √ bosons and comparisons with
[75] ATLAS Collaboration, Performance of pile-up ATLAS data taken at s = 8 TeV, Eur. Phys. J. C76
mitigation techniques for jets in pp collisions at
√ (2016) 154, [1510.05821].
s = 8 TeV using the ATLAS detector, Eur. Phys. J. [95] ATLAS Collaboration, Identification of boosted,
√
C76 (2016) 581, [1510.03823]. hadronically-decaying W and Z bosons in s = 13
[76] G. Soyez, G. P. Salam, J. Kim, S. Dutta and TeV Monte Carlo Simulations for ATLAS, ATLAS
M. Cacciari, Pileup subtraction for jet shapes, Phys. PUB Note ATL-PHYS-PUB-2015-033, 2015.
Rev. Lett. 110 (2013) 162001, [1211.2811]. [96] ATLAS Collaboration, Identification of high √ transverse
[77] CMS Collaboration, Pileup jet identification, CMS momentum top quarks in pp collisions at s = 8 TeV
Physics Analysis Summary CMS-PAS-JME-13-005, with the ATLAS detector, JHEP 06 (2016) 093,
2013. [1603.03127].
[78] ATLAS Collaboration, Identification and rejection of [97] ATLAS Collaboration, Boosted hadronic top
pile-up jets at high pseudorapidity with the ATLAS identification at ATLAS for early 13 TeV data,
detector, Eur. Phys. J. C77 (2017) 580, [1705.02211]. ATLAS PUB Note ATL-PHYS-PUB-2015-053, 2015.
[Erratum: Eur. Phys. J. C77 (2017) 712]. [98] ATLAS Collaboration, Measurement
√ of the soft-drop
[79] ATLAS Collaboration, Event displays from run 2 jet mass in pp collisions at s = 13 TeV with the
physics analyses, 2016. ATLAS detector, Phys. Rev. Lett. 121 (2018) 092001,
[Link] [1711.08341].
EventDisplayRun2Physics. [99] ATLAS Collaboration, Measurement of the
[80] D. Bertolini, P. Harris, M. Low and N. Tran, Pileup cross-section of high transverse momentum vector
per particle identification, JHEP 10 (2014) 059, bosons reconstructed as single jets
√ and studies of jet
[1407.6013]. substructure in pp collisions at s = 7 TeV with the
[81] M. Cacciari, G. P. Salam and G. Soyez, SoftKiller, a ATLAS detector, New J. Phys. 16 (2014) 113013,
particle-level pileup removal method, Eur. Phys. J. [1407.0800].
C75 (2015) 59, [1407.0408]. [100] B. Nachman, P. Nef, A. Schwartzman, M. Swiatlowski
[82] P. Berta, M. Spousta, D. W. Miller and R. Leitner, and C. Wanotayaroj, Jets from Jets: Re-clustering as
Particle-level pileup subtraction for jets and jet shapes, a tool for large radius jet reconstruction and grooming
JHEP 06 (2014) 092, [1403.3108]. at the LHC, JHEP 02 (2015) 075, [1407.2922].
[83] P. T. Komiske, E. M. Metodiev, B. Nachman and [101] ATLAS Collaboration, Search for pair production of
M. D. Schwartz, Pileup Mitigation with Machine gluinos decaying via stop and sbottom in events with
Learning (PUMML), JHEP 12 (2017) 051, b-jets and large
√ missing transverse momentum in pp
[1707.08600]. collisions at s = 13 TeV with the ATLAS detector,
[84] D. Bertolini, T. Chan and J. Thaler, Jet Observables Phys. Rev. D94 (2016) 032003, [1605.09318].
Without Jet Algorithms, JHEP 04 (2014) 013, [102] ATLAS Collaboration, Search for top squarks in final
[1310.7584]. states with one isolated lepton,
√ jets, and missing
[85] D. Krohn, M. D. Schwartz, M. Low and L.-T. Wang, transverse momentum in s = 13 TeV pp collisions
Jet Cleansing: Pileup Removal at High Luminosity, with the ATLAS detector, Phys. Rev. D94 (2016)
Phys. Rev. D90 (2014) 065020, [1309.4777]. 052009, [1606.03903].
[86] ATLAS Collaboration, Impact of Alternative Inputs [103] ATLAS Collaboration, Search for new phenomena with
and Grooming Methods on Large-R Jet Reconstruction large jet multiplicities and missing transverse
in ATLAS, ATLAS PUB Note momentum using large-radius jets and flavour-tagging
ATL-PHYS-PUB-2017-020, 2017. at ATLAS in 13 TeV pp collisions, JHEP 12 (2017)
[87] CMS Collaboration, Jet algorithms performance in 13 034, [1708.02794].
TeV data, CMS Physics Analysis Summary [104] ATLAS Collaboration, Jet reclustering and close-by
CMS-PAS-JME-16-003, 2017. effects in ATLAS Run 2, ATLAS CONF Note
[88] D. Contardo, M. Klute, J. Mans, L. Silvestris and ATLAS-CONF-2017-062, 2017.
J. Butler, Technical Proposal for the Phase-II Upgrade [105] CMS Collaboration, Identification techniques for highly
of the CMS Detector, Tech. Rep. CMS-TDR-15-02, boosted W bosons that decay into hadrons, JHEP 12
2015. (2014) 017, [1410.4227].
[89] ATLAS Collaboration, Constituent-level pileup [106] CMS Collaboration, Boosted top jet tagging at CMS,
mitigation performance using 2015 data, ATLAS CMS Physics Analysis Summary
CONF Note ATLAS-CONF-2017-065, 2017. CMS-PAS-JME-13-007, 2014.
[90] D. Krohn, J. Thaler and L.-T. Wang, Jet Trimming, [107] CMS Collaboration, Top tagging with new approaches,
JHEP 02 (2010) 084, [0912.1342]. CMS Physics Analysis Summary
[91] S. D. Ellis, C. K. Vermilion and J. R. Walsh, CMS-PAS-JME-15-002, 2016.
Recombination algorithms and jet substructure: [108] CMS Collaboration, Studies of jet mass in dijet and
 42

W/Z+jet events, JHEP 05 (2013) 090, [1303.4811]. [129] S. Bright-Thonney and B. Nachman, Investigating the
[109] CMS Collaboration, Measurements of the differential Topology Dependence of Quark and Gluon Jets, JHEP
jet cross section as a function of the jet√mass in dijet 03 (2019) 098, [1810.05653].
events from proton-proton collisions at s = 13 TeV, [130] J. Gallicchio and M. D. Schwartz, Quark and gluon
JHEP 11 (2018) 113, [1807.05974]. tagging at the LHC, Phys. Rev. Lett. 107 (2011)
[110] J. Thaler and K. Van Tilburg, Identifying boosted 172001, [1106.3076].
objects with N-subjettiness, JHEP 03 (2011) 015, [131] A. J. Larkoski, J. Thaler and W. J. Waalewijn,
[1011.2268]. Gaining (mutual) information about quark/gluon
[111] J. Thaler and K. Van Tilburg, Maximizing boosted top discrimination, JHEP 11 (2014) 129, [1408.3122].
identification by minimizing N-subjettiness, JHEP 02 [132] C. Frye, A. J. Larkoski, J. Thaler and K. Zhou,
(2012) 093, [1108.2701]. Casimir meets poisson: Improved quark/gluon
[112] ATLAS Collaboration, Jet mass reconstruction with discrimination with counting observables, JHEP 09
the ATLAS detector in early Run 2 data, ATLAS (2017) 083, [1704.06266].
CONF Note ATLAS-CONF-2016-035, 2016. [133] P. T. Komiske, E. M. Metodiev and M. D. Schwartz,
[113] ATLAS Collaboration, Measurement√of large radius jet Deep learning in color: Towards automated
mass reconstruction performance at s = 8 TeV using quark/gluon jet discrimination, JHEP 01 (2016) 110,
the ATLAS detector, ATLAS CONF Note [1612.01551].
ATLAS-CONF-2016-008, 2016. [134] P. Gras, S. Höche, D. Kar, A. Larkoski, L. Lönnblad,
[114] ATLAS Collaboration, In-situ measurements of S. Plätzer et al., Systematics of quark/gluon tagging,
large-radius jet reconstruction performance, ATLAS JHEP 07 (2017) 091, [1704.03878].
CONF Note ATLAS-CONF-2017-063, 2017. [135] ATLAS Collaboration, Quark versus gluon jet tagging
[115] CMS Collaboration, Search for massive resonances using charged particle multiplicity with the ATLAS
decaying into W √ W , W Z, ZZ, qW , and qZ with dijet detector, ATLAS PUB Note
final states at s = 13TeV, Phys. Rev. D97 (2018) ATL-PHYS-PUB-2017-009, 2017.
072006, [1708.05379]. [136] ATLAS Collaboration, Light-quark√and gluon jet
[116] CMS Collaboration, Measurement of the jet mass √ in discrimination in pp collisions at s = 7 TeV with the
highly boosted tt events from pp collisions at s = 8 ATLAS detector, Eur. Phys. J. C74 (2014) 3023,
TeV, Eur. Phys. J. C77 (2017) 467, [1703.06330]. [1405.6583].
[117] A. J. Larkoski, G. P. Salam and J. Thaler, Energy [137] ATLAS Collaboration, Discrimination √ of light quark
correlation functions for jet substructure, JHEP 06 and gluon jets in pp collisions at s = 8 TeV with the
(2013) 108, [1305.0007]. ATLAS detector, ATLAS CONF Note
[118] A. J. Larkoski, I. Moult and D. Neill, Power counting ATLAS-CONF-2016-034, 2016.
to better jet observables, JHEP 12 (2014) 009, [138] ATLAS Collaboration, Quark versus gluon jet tagging
[1409.6298]. using jet images with the ATLAS detector, ATLAS
[119] A. J. Larkoski, I. Moult and D. Neill, Analytic boosted PUB Note ATL-PHYS-PUB-2017-017, 2017.
boson discrimination, JHEP 05 (2016) 117, [139] CMS Collaboration, Performance of quark/gluon
[1507.03018]. discrimination in 8 TeV pp data, CMS Physics
[120] I. Moult, L. Necib and J. Thaler, New angles on Analysis Summary CMS-PAS-JME-13-002, 2013.
energy correlation functions, JHEP 12 (2016) 153, [140] CMS Collaboration, Performance of quark/gluon
[1609.07483]. discrimination in 13 TeV data, CMS Detector
[121] CMS Collaboration, W and top tagging scale factors, Performance Summary CMS-DP-2016-070, 2016.
CMS Detector Performance Summary [141] CMS Collaboration, New developments for jet
CMS-DP-2017-026, 2017. substructure reconstruction in CMS, CMS Detector
[122] ATLAS Collaboration, Top and boson tagger Performance Summary CMS-DP-2017-027, 2017.
background performance - Moriond 2017, Tech. Rep. [142] CMS Collaboration, Search for a Higgs boson in the
ATL-JETM-2017-005, 2017. decay
√ channel H → ZZ (∗) → q q̄`− l+ in pp collisions at
[123] G. Altarelli and G. Parisi, Asymptotic freedom in s = 7 TeV, JHEP 1204 (2012) 036, [1202.1416].
parton language, Nucl. Phys. B126 (1977) 298. [143] CMS Collaboration, Measurement of the hadronic
[124] A. Banfi, G. P. Salam and G. Zanderighi, Infrared safe activity in events with a Z and two jets and extraction
definition of jet flavor, Eur. Phys. J. C47 (2006) 113, of the cross section for the electroweak
√ production of a
[hep-ph/0601139]. Z with two jets in pp collisions at s = 7 TeV, JHEP
[125] A. Buckley and C. Pollard, QCD-aware partonic jet 10 (2013) 062, [1305.7389].
clustering for truth-jet flavour labelling, Eur. Phys. J. [144] CMS Collaboration, Search for the standard model
C76 (2016) 71, [1507.00508]. Higgs boson produced through vector boson fusion and
[126] E. M. Metodiev and J. Thaler, Jet topics: decaying to bb, Phys. Rev. D92 (2015) 032008,
Disentangling quarks and gluons at colliders, Phys. [1506.01010].
Rev. Lett. 120 (2018) 241602, [1802.00008]. [145] CMS Collaboration, Search for direct production of
[127] P. T. Komiske, E. M. Metodiev and J. Thaler, An supersymmetric partners of the top quark√in the all-jets
operational definition of quark and gluon jets, JHEP final state in proton-proton collisions at s = 13 TeV,
11 (2018) 059, [1809.01140]. JHEP 10 (2017) 005, [1707.03316].
[128] C. Frye, A. J. Larkoski, M. D. Schwartz and K. Yan, [146] ATLAS Collaboration, Search for the Standard Model
Factorization for groomed jet substructure beyond the Higgs boson produced by vector-boson fusion and
√
next-to-leading logarithm, JHEP 07 (2016) 064, decaying to bottom quarks in s = 8 TeV pp collisions
[1603.09338].
 43
√
with the ATLAS detector, JHEP 11 (2016) 112, with boson-tagged jets in pp collisions at s = 13 TeV
[1606.02181]. with the ATLAS detector, Phys. Lett. B777 (2018)
[147] ATLAS Collaboration, Search for high-mass diboson 91–113, [1708.04445].
resonances with √ boson-tagged jets in proton-proton [166] ATLAS Collaboration, Search for W W/W Z resonance
collisions at s = 8 TeV with the ATLAS detector, production
√ in `νqq final states in pp collisions at
JHEP 12 (2015) 055, [1506.00962]. s = 13 TeV with the ATLAS detector, ATLAS
[148] T. Sjostrand, S. Mrenna and P. Z. Skands, A Brief CONF Note ATLAS-CONF-2017-051, 2017.
Introduction to PYTHIA 8.1, [Link]. [167] J. Dolen, P. Harris, S. Marzani, S. Rappoccio and
178 (2008) 852–867, [0710.3820]. N. Tran, Thinking outside the ROCs: Designing
[149] T. Sjostrand, S. Mrenna and P. Z. Skands, PYTHIA Decorrelated Taggers (DDT) for jet substructure,
6.4 Physics and Manual, JHEP 05 (2006) 026, JHEP 05 (2016) 156, [1603.00027].
[hep-ph/0603175]. [168] C. Shimmin, P. Sadowski, P. Baldi, E. Weik,
[150] J. Bellm et al., Herwig 7.0/Herwig++ 3.0 release note, D. Whiteson, E. Goul et al., Decorrelated jet
Eur. Phys. J. C76 (2016) 196, [1512.01178]. substructure tagging using adversarial neural networks,
[151] M. Bahr et al., Herwig++ Physics and Manual, Eur. Phys. Rev. D96 (2017) 074034, [1703.03507].
Phys. J. C58 (2008) 639–707, [0803.0883]. [169] J. A. Aguilar-Saavedra, J. H. Collins and R. K.
[152] L. M. Dery, B. Nachman, F. Rubbo and Mishra, A generic anti-QCD jet tagger, JHEP 11
A. Schwartzman, Weakly supervised classification in (2017) 163, [1709.01087].
high energy physics, JHEP 05 (2017) 145, [170] I. Moult, B. Nachman and D. Neill, Convolved
[1702.00414]. substructure: Analytically decorrelating jet substructure
[153] E. M. Metodiev, B. Nachman and J. Thaler, observables, JHEP 05 (2018) 002, [1710.06859].
Classification without labels: Learning from mixed [171] D. Krohn, J. Thaler and L.-T. Wang, Jets with
samples in high energy physics, JHEP 10 (2017) 174, variable R, JHEP 06 (2009) 059, [0903.0392].
[1708.02949]. [172] ATLAS Collaboration, Boosted object tagging with
[154] S. D. Ellis, A. Hornig, T. S. Roy, D. Krohn and M. D. variable-R jets in the ATLAS detector, ATLAS PUB
Schwartz, Qjets: A non-deterministic approach to Note ATL-PHYS-PUB-2016-013, 2016.
tree-based jet substructure, Phys. Rev. Lett. 108 (2012) [173] ATLAS Collaboration, A new method to distinguish
182003, [1201.1914]. hadronically decaying boosted Z bosons from W bosons
[155] L. G. Almeida, S. J. Lee, G. Perez, G. F. Sterman, using the ATLAS detector, Eur. Phys. J. C76 (2016)
I. Sung and J. Virzi, Substructure of high-pT jets at the 238, [1509.04939].
LHC, Phys. Rev. D79 (2009) 074017, [0807.0234]. [174] ATLAS Collaboration, Flavor tagging with track jets
[156] J. Thaler and L.-T. Wang, Strategies to Identify in boosted topologies with the ATLAS detector, ATLAS
Boosted Tops, JHEP 07 (2008) 092, [0806.0023]. PUB Note ATL-PHYS-PUB-2014-013, 2014.
[157] J. Gallicchio and M. D. Schwartz, Seeing in color: Jet [175] CMS Collaboration, A Cambridge-Aachen (C-A) based
superstructure, Phys. Rev. Lett. 105 (2010) 022001, jet algorithm for boosted top-jet tagging, CMS Physics
[1001.5027]. Analysis Summary CMS-PAS-JME-09-001, 2009.
[158] ATLAS Collaboration, Identification of [176] CMS Collaboration, Jet substructure algorithms, CMS
hadronically-decaying W bosons and top quarks using Physics Analysis Summary CMS-PAS-JME-10-013,
high-level features as input to boosted decision
√ trees 2011.
and deep neural networks in ATLAS at s = 13 TeV, [177] T. Plehn, G. P. Salam and M. Spannowsky, Fat jets
ATLAS PUB Note ATL-PHYS-PUB-2017-004, 2017. for a light Higgs, Phys. Rev. Lett. 104 (2010) 111801,
[159] A. J. Larkoski, D. Neill and J. Thaler, Jet shapes with [0910.5472].
the broadening axis, JHEP 04 (2014) 017, [1401.2158]. [178] T. Plehn, M. Spannowsky, M. Takeuchi and
[160] CMS Collaboration, Search for low mass vector D. Zerwas, Stop reconstruction with tagged tops, JHEP
resonances decaying into quark-antiquark
√ pairs in 10 (2010) 078, [1006.2833].
proton-proton collisions at s = 13 TeV, JHEP 01 [179] ATLAS Collaboration, Performance of jet substructure
(2018) 097, [1710.00159]. techniques
√ for large-R jets in proton-proton collisions
[161] CMS Collaboration, Search for massive resonances at s = 7 TeV using the ATLAS detector, JHEP 09
decaying into pairs √ of boosted bosons in semi-leptonic (2013) 076, [1306.4945].
final states at s = 8 TeV, JHEP 08 (2014) 174, [180] D. E. Soper and M. Spannowsky, Finding physics
[1405.3447]. signals with shower deconstruction, Phys. Rev. D84
[162] CMS Collaboration, Search for massive W H √ (2011) 074002, [1102.3480].
resonances decaying into the `νbb final state at s = 8 [181] D. E. Soper and M. Spannowsky, Finding top quarks
TeV, Eur. Phys. J. C76 (2016) 237, [1601.06431]. with shower deconstruction, Phys. Rev. D87 (2013)
[163] CMS Collaboration, Search for massive resonances 054012, [1211.3140].
decaying into√WW, WZ or ZZ bosons in proton-proton [182] ATLAS Collaboration, Search for W 0 → tb decays √ in
collisions at s = 13 TeV, JHEP 03 (2017) 162, the hadronic final state using pp collisions at s = 13
[1612.09159]. TeV with the ATLAS detector, Phys. Lett. B781
[164] ATLAS Collaboration, Search for heavy resonances (2018) 327, [1801.07893].
decaying to a W or Z boson and a Higgs √ boson in the [183] ATLAS Collaboration, Performance of top quark and
q q̄ (0) bb̄ final state in pp collisions at s = 13 TeV with W boson tagging in Run 2 with ATLAS, ATLAS
the ATLAS detector, Phys. Lett. B774 (2017) 494–515, CONF Note ATLAS-CONF-2017-064, 2017.
[1707.06958]. [184] G. Kasieczka, T. Plehn, T. Schell, T. Strebler and
[165] ATLAS Collaboration, Search for diboson resonances G. P. Salam, Resonance searches with an updated top
 44

tagger, JHEP 06 (2015) 203, [1503.05921]. the Tevatron.” 2014.

[185] T. Lapsien, R. Kogler and J. Haller, A new tagger for [203] CMS Collaboration, Measurement√of the top quark
hadronically decaying heavy particles at the LHC, Eur. mass using proton-proton data at s = 7 and 8 TeV,
Phys. J. C76 (2016) 600, [1606.04961]. Phys. Rev. D93 (2016) 072004, [1509.04044].
[186] ATLAS Collaboration, A search for tt resonances [204] ATLAS Collaboration, Measurement of the top quark
using
√ lepton-plus-jets events in proton-proton collisions mass in the tt̄ →√ lepton+jets and tt̄ → dilepton
at s = 8 TeV with the ATLAS detector, JHEP 08 channels using s = 7 TeV ATLAS data, Eur. Phys.
(2015) 148, [1505.07018]. J. C75 (2015) 330, [1503.05427].
[187] ATLAS Collaboration, Search for top-squark pair [205] ATLAS Collaboration, Measurement of the √ top quark
production in final states with one lepton, jets, and mass in the tt̄ → dilepton channel from s = 8 TeV
missing
√ transverse momentum using 36 fb−1 of ATLAS data, Phys. Lett. B761 (2016) 350–371,
s = 13 TeV pp collision data with the ATLAS [1606.02179].
detector, JHEP 06 (2018) 108, [1711.11520]. [206] A. H. Hoang, S. Plätzer and D. Samitz, On the cutoff
[188] CMS Collaboration, Search for supersymmetry using dependence of the quark mass parameter in angular
hadronic top quark tagging in 13 TeV pp collisions, ordered parton showers, JHEP 10 (2018) 200,
CMS Physics Analysis Summary [1807.06617].
CMS-PAS-SUS-16-050, 2017. [207] M. Butenschoen, B. Dehnadi, A. H. Hoang, V. Mateu,
[189] CMS Collaboration, Identification of heavy-flavour jets M. Preisser and I. W. Stewart, Top quark mass
with the CMS detector in pp collisions at 13 TeV, calibration for Monte Carlo event generators, Phys.
JINST 13 (2018) P05011, [1712.07158]. Rev. Lett. 117 (2016) 232001, [1608.01318].
[190] ATLAS Collaboration, Optimisation of the ATLAS [208] A. H. Hoang, S. Mantry, A. Pathak and I. W. Stewart,
b-tagging performance for the 2016 LHC Run, ATLAS “Extracting a short distance top mass with light
PUB Note ATL-PHYS-PUB-2016-012, 2016. grooming.” 2017.
[191] CMS Collaboration, Identification of b-quark jets with [209] ATLAS Collaboration,
√ Measurement of jet charge in
the CMS experiment, JINST 8 (2013) P04013, dijet events from s=8 TeV pp collisions with the
[1211.4462]. ATLAS detector, Phys. Rev. D93 (2016) 052003,
[192] CMS
√ Collaboration, Performance of b tagging at [1509.05190].
s = 8 TeV in multijet, tt̄ and boosted topology events, [210] CMS Collaboration, Measurements √ of jet charge with
CMS Physics Analysis Summary dijet events in pp collisions at s = 8 TeV, JHEP 10
CMS-PAS-BTV-13-001, 2013. (2017) 131, [1706.05868].
[193] ATLAS Collaboration, Studies of b-tagging [211] P. Z. Skands, Tuning Monte Carlo generators: The
performance and jet substructure in a high pT g → bb̄ Perugia Tunes, Phys. Rev. D82 (2010) 074018,
√ sample of large-R jets from pp collisions at
rich [1005.3457].
s = 8 TeV with the ATLAS detector, ATLAS CONF [212] W. J. Waalewijn, Calculating the charge of a jet, Phys.
Note ATLAS-CONF-2016-002, 2016. Rev. D86 (2012) 094030, [1209.3019].
[194] ATLAS Collaboration, Performance of b-jet [213] D. Krohn, M. D. Schwartz, T. Lin and W. J.
identification in the ATLAS experiment, JINST 11 Waalewijn, Jet charge at the LHC, Phys. Rev. Lett.
(2016) P04008, [1512.01094]. 110 (2013) 212001, [1209.2421].
[195] ATLAS Collaboration, Boosted Higgs (→ b√ b̄) boson [214] ATLAS Collaboration, Study of jet √ shapes in inclusive
identification with the ATLAS detector at s = 13 jet production in pp collisions at s = 7 TeV using the
TeV, ATLAS CONF Note ATLAS-CONF-2016-039, ATLAS detector, Phys. Rev. D83 (2011) 052003,
2016. [1101.0070].
[196] CMS Collaboration, Identification of b quark jets at [215] ATLAS Collaboration, Measurement of event shapes at
the CMS experiment in the LHC Run 2, CMS Physics large momentum√transfer with the ATLAS detector in
Analysis Summary CMS-PAS-BTV-15-001, 2016. pp collisions at s = 7 TeV, Eur. Phys. J. C72 (2012)
[197] ATLAS Collaboration, b-tagging in dense 2211, [1206.2135].
environments, ATLAS PUB Note [216] ATLAS Collaboration, Measurement
√ of jet shapes in
ATL-PHYS-PUB-2014-014, 2014. top-quark pair events at s = 7 TeV using the ATLAS
[198] CMS Collaboration, Identification of double-b quark detector, Eur. Phys. J. C73 (2013) 2676, [1307.5749].
jets in boosted event topologies, CMS Physics Analysis [217] CMS Collaboration, Shape, transverse size, and
Summary CMS-PAS-BTV-15-002, 2016. charged hadron multiplicity of jets in pp collisions at 7
[199] ATLAS Collaboration, Variable radius, exclusive-kT , TeV, JHEP 06 (2012) 160, [1204.3170].
and center-of-mass subjet reconstruction for [218] ATLAS Collaboration, Measurement of
Higgs(→ bb̄) tagging in ATLAS, ATLAS PUB Note jet-substructure observables in top quark, W boson and
ATL-PHYS-PUB-2017-010, 2017. light
√ jet production in proton-proton collisions at
[200] ATLAS Collaboration, Identification and tagging of s = 13 TeV with the ATLAS detector, Submitted to:
double b-hadron jets with the ATLAS detector, ATLAS JHEP (2019) , [1903.02942].
CONF Note ATLAS-CONF-2012-100, 2012. [219] CMS Collaboration, Measurement of jet substructure
[201] ATLAS Collaboration,
√ Jet mass and substructure of observables
√ in tt events from proton-proton collisions
inclusive jets in s = 7 TeV pp collisions with the at s = 13TeV, Phys. Rev. D98 (2018) 092014,
ATLAS experiment, JHEP 05 (2012) 128, [1203.4606]. [1808.07340].
[202] CDF, D0 Collaborations, Tevatron Electroweak [220] ATLAS Collaboration, Properties of jets measured
Working Group, “Combination of CDF and D0 results from tracks in proton-proton
√ collisions at
on the mass of the top quark using up to 9.7 fb−1 at center-of-mass energy s = 7 TeV with the ATLAS
 45
√
detector, Phys. Rev. D84 (2011) 054001, [1107.3311]. quarks in pp collisions at s = 8 TeV, Phys. Rev.
[221] ATLAS Collaboration, Measurement of the √ D94 (2016) 072002, [1605.00116].
charged-particle multiplicity inside jets from s = 8 [237] CMS Collaboration, Measurement of the differential tt̄
TeV pp collisions with the ATLAS detector, Eur. Phys. cross section with high-p√T top-quark jets in the
J. C76 (2016) 322, [1602.00988]. all-hadronic channel at s = 8 TeV, CMS Physics
[222] ATLAS Collaboration, Measurement of the jet Analysis Summary CMS-PAS-TOP-16-018, 2017.
fragmentation function and transverse profile in [238] ATLAS Collaboration, Measurements of tt̄ differential
proton-proton collisions at a center-of-mass energy of 7 cross-sections of highly boosted top quarks decaying
√ to
TeV with the ATLAS detector, Eur. Phys. J. C71 all-hadronic final states in pp collisions at s = 13
(2011) 1795, [1109.5816]. TeV using the ATLAS detector, Phys. Rev. D98
[223] ATLAS Collaboration, Properties of g → bb̄ at small (2018) 012003, [1801.02052].
√ in pp collisions with the ATLAS
opening angles [239] ATLAS Collaboration, Evidence for the H → bb decay
detector at s = 13 TeV, Phys. Rev. D99 (2019) with the ATLAS detector, JHEP 12 (2017) 024,
052004, [1812.09283]. [1708.03299].
[224] ATLAS Collaboration, Measurement of colour flow [240] CMS Collaboration, Evidence for the Higgs boson
with the jet √pull angle in tt̄ events using the ATLAS decay to a bottom quark-antiquark pair, Phys. Lett.
detector at s = 8 TeV, Phys. Lett. B750 (2015) B780 (2018) 501–532, [1709.07497].
475–493, [1506.05629]. [241] CMS Collaboration, Inclusive search for a highly
[225] ATLAS Collaboration, Measurement√ of kT splitting boosted Higgs boson decaying to a bottom
scales in W → `ν events at s = 7 TeV with the quark-antiquark pair, Phys. Rev. Lett. 120 (2018)
ATLAS detector, Eur. Phys. J. C73 (2013) 2432, 071802, [1709.05543].
[1302.1415]. [242] ATLAS Collaboration, Measurement of the cross
[226] CMS Collaboration, Measurement of differential section of high transverse momentum Z → √bb̄
top-quark
√ pair production cross sections in pp colisions production in proton–proton collisions at s = 8T eV
at s = 7 TeV, Eur. Phys. J. C73 (2013) 2339, with the ATLAS Detector, Phys. Lett. B738 (2014)
[1211.2220]. 25–43, [1404.7042].
[227] CMS Collaboration, Measurement of the differential [243] CMS Collaboration, Search for anomalous tt̄
cross section √ for top quark pair production in pp production in the highly-boosted all-hadronic final state,
collisions at s = 8 TeV, Eur. Phys. J. C75 (2015) JHEP 09 (2012) 029, [1204.2488]. [Erratum: JHEP
542, [1505.04480]. 03 (2014) 132].
[228] CMS Collaboration, Measurement of differential cross [244] CMS Collaboration, Search for√resonant tt̄ production
sections for top quark pair production using the in proton-proton collisions at s = 8 TeV, Phys. Rev.
lepton+jets final state in proton-proton collisions at 13 D93 (2016) 012001, [1506.03062].
TeV, Phys. Rev. D95 (2017) 092001, [1610.04191]. [245] ATLAS Collaboration, Search for tt̄ resonances in the
−1
[229] ATLAS Collaboration, Differential top-antitop lepton plus jets final
√ state with ATLAS using 4.7 fb
cross-section measurements as a function of of pp collisions at s = 7 TeV, Phys. Rev. D88 (2013)
observables constructed
√ from final-state particles using 012004, [1305.2756].
pp collisions at s = 7 TeV in the ATLAS detector, [246] ATLAS Collaboration, Search for resonances decaying
JHEP 06 (2015) 100, [1502.05923]. into top-quark pairs using √fully hadronic decays in pp
[230] ATLAS Collaboration, Measurements of top-quark pair collisions with ATLAS at s = 7 TeV, JHEP 01
differential cross-sections
√ in the lepton+jets channel in (2013) 116, [1211.2202].
pp collisions at s = 8 TeV using the ATLAS [247] ATLAS Collaboration, Search for heavy particles
detector, Eur. Phys. J. C76 (2016) 538, [1511.04716]. decaying to pairs of highly-boosted top quarks using
[231] ATLAS Collaboration, Measurements of top-quark pair lepton-plus-jets
√ events in proton–proton collisions at
differential cross-sections
√ in the eµ channel in pp s = 13 TeV with the ATLAS detector, ATLAS
collisions at s = 13 TeV using the ATLAS detector, CONF Note ATLAS-CONF-2016-014, 2016.
Eur. Phys. J. C77 (2017) 292, [1612.05220]. [248] CMS Collaboration, Search for tt resonances in highly
[232] ATLAS Collaboration, Measurements of top-quark pair boosted lepton+jets and fully√ hadronic final states in
differential cross-sections
√ in the lepton+jets channel in proton-proton collisions at s = 13 TeV, JHEP 07
pp collisions at s = 13 TeV using the ATLAS (2017) 001, [1704.03366].
detector, JHEP 11 (2017) 191, [1708.00727]. [249] CMS Collaboration, Search for heavy resonances in the
[233] ATLAS Collaboration, Measurement of the differential W/Z-tagged dijet mass spectrum in pp collisions at 7
cross-section of highly boosted top √ quarks as a function TeV, Phys. Lett. B723 (2013) 280, [1212.1910].
of their transverse momentum in s = 8 TeV [250] CMS Collaboration, Search for massive resonances in
proton-proton collisions using the ATLAS detector, dijet systems containing jets
√ tagged as W or Z boson
Phys. Rev. D93 (2016) 032009, [1510.03818]. decays in pp collisions at s = 8 TeV, JHEP 08
[234] H1, ZEUS Collaborations, Combined Measurement and (2014) 173, [1405.1994].
QCD Analysis of the Inclusive e+- p Scattering Cross [251] CMS Collaboration, Search for narrow high-mass
√
Sections at HERA, JHEP 01 (2010) 109, [0911.0884]. resonances in proton proton collisions at s = 8 TeV
[235] H.-L. Lai, M. Guzzi, J. Huston, Z. Li, P. M. Nadolsky, decaying to a Z and a Higgs boson, Phys. Lett. B748
J. Pumplin et al., New parton distributions for collider (2015) 255–277, [1502.04994].
physics, Phys. Rev. D82 (2010) 074024, [1007.2241]. [252] CMS Collaboration, Search for a massive resonance
[236] CMS Collaboration, Measurement of the integrated and decaying into a Higgs boson and a W or Z boson in
differential tt̄ production cross sections for high-pt top hadronic final states in proton-proton collisions at
 46
√
s = 8 TeV, JHEP 02 (2016) 145, [1506.01443]. [269] CMS Collaboration, Search for single production of a
[253] CMS Collaboration, Search for heavy resonances heavy vector-like T quark decaying to a Higgs boson
decaying to two Higgs bosons in final states containing and a top quark with a lepton and jets in the final
four b quarks, Eur. Phys. J. C76 (2016) 371, state, Phys. Lett. B771 (2017) 80, [1612.00999].
[1602.08762]. [270] CMS Collaboration, Search for single production of a
[254] ATLAS Collaboration, Combination of searches for vector-like T quark decaying to a Z boson
√ and a top
W
√ W , W Z, and ZZ resonances in pp collisions at quark in proton-proton collisions at s = 13 TeV,
s = 8 TeV with the ATLAS detector, Phys. Lett. Phys. Lett. B781 (2018) 574–600, [1708.01062].
B755 (2016) 285–305, [1512.05099]. [271] CMS Collaboration, Search for electroweak production
[255] ATLAS Collaboration, Search for W W/W Z resonance √ of a vector-like quark decaying to a top quark and a
production in `νqq final states in pp collisions at s = Higgs boson using boosted topologies in fully hadronic
13 TeV with the ATLAS detector, JHEP 03 (2018) final states, JHEP 04 (2017) 136, [1612.05336].
042, [1710.07235]. [272] CMS Collaboration, Search for single production of
[256] ATLAS Collaboration, Searches for heavy ZZ and vector-like quarks decaying to a Z boson and√a top or a
ZW resonances √ in the ``qq and ννqq final states in pp bottom quark in proton-proton collisions at s = 13
collisions at s = 13 TeV with the ATLAS detector, TeV, JHEP 05 (2017) 029, [1701.07409].
JHEP 03 (2018) 009, [1708.09638]. [273] CMS Collaboration, Search for pair production of
[257] ATLAS Collaboration, Searches √ for heavy diboson vector-like T and B quarks in single-lepton final states
resonances in pp collisions at s = 13 TeV with the using boosted√jet substructure in proton-proton
ATLAS detector, JHEP 09 (2016) 173, [1606.04833]. collisions at s = 13 TeV, JHEP 11 (2017) 085,
[258] ATLAS Collaboration, Search for heavy resonances [1706.03408].
decaying into a W or Z boson and a Higgs boson in [274] CMS Collaboration, Search for a heavy resonance
−1
√ states with leptons and b-jets in 36 fb of
final decaying
√ to a top quark and a vector-like top quark at
s = 13 TeV pp collisions with the ATLAS detector, s = 13 TeV, JHEP 09 (2017) 053, [1703.06352].
JHEP 03 (2018) 174, [1712.06518]. [Erratum: [275] CMS Collaboration, Search for pair production of
JHEP11,051(2018)]. vector-like quarks in the bWbW
√ channel from
[259] ATLAS Collaboration, Search for Higgs boson pair proton-proton collisions at s = 13 TeV, Phys. Lett. B
production
√ in the bb̄bb̄ final state from pp collisions at 779 (2018) 82, [1710.01539].
s = 8 TeVwith the ATLAS detector, Eur. Phys. J. [276] C. Collaboration, Search for vectorlike light-flavor
√
C75 (2015) 412, [1506.00285]. quark partners in proton-proton collisions at s =8
[260] ATLAS Collaboration, Search for pair production of TeV, Phys. Rev. D97 (2018) 072008, [1708.02510].
Higgs bosons√ in the bb̄bb̄ final state using proton–proton [277] ATLAS Collaboration, Search for the production of
collisions at s = 13 TeV with the ATLAS detector, single vector-like and excited
√ quarks in the W t final
Phys. Rev. D94 (2016) 052002, [1606.04782]. state in pp collisions at s = 8 TeV with the ATLAS
[261] CMS Collaboration, Search for heavy resonances detector, JHEP 02 (2016) 110, [1510.02664].
decaying into a vector boson and a Higgs boson in final [278] ATLAS Collaboration, Search for single production of
states with charged leptons, neutrinos, and b quarks, vector-like
√ quarks decaying into Wb in pp collisions at
Phys. Lett. B768 (2017) 137–162, [1610.08066]. s = 8 TeV with the ATLAS detector, Eur. Phys. J.
[262] CMS Collaboration, Combination of searches for heavy C76 (2016) 442, [1602.05606].
resonances decaying to WW, WZ, ZZ, WH, √ and ZH [279] ATLAS Collaboration, Search for pair production of
boson pairs in proton-proton collisions at s = 8 and heavy vector-like quarks decaying to high-pT W bosons
13 TeV, Phys. Lett. B774 (2017) 533, [1705.09171]. and b quarks√in the lepton-plus-jets final state in pp
[263] CMS Collaboration, Search for heavy resonances that collisions at s = 13 TeV with the ATLAS detector,
decay into a vector boson √ and a Higgs boson in JHEP 10 (2017) 141, [1707.03347].
hadronic final states at s = 13 TeV, Eur. Phys. J. [280] ATLAS Collaboration, Search for pair production of
C77 (2017) 636, [1707.01303]. vector-like top quarks in events with one
√ lepton, jets,
[264] CMS Collaboration, Search for heavy resonances and missing transverse momentum in s = 13 TeV pp
decaying to a pair of Higgs bosons in the√four b quark collisions with the ATLAS detector, JHEP 08 (2017)
final state in proton-proton collisions at s = 13 TeV, 052, [1705.10751].
CMS Physics Analysis Summary [281] ATLAS Collaboration, Search for light resonances
CMS-PAS-B2G-16-026, 2017. decaying to boosted quark pairs and produced in
[265] CMS Collaboration, Inclusive search for a√vector-like association with
√ a photon or a jet in proton-proton
T quark with charge 32 in pp collisions at s = 8 TeV, collisions at s = 13 TeV with the ATLAS detector,
Phys. Lett. B729 (2014) 149, [1311.7667]. Phys. Lett. B788 (2019) 316–335, [1801.08769].
[266] CMS Collaboration, Search for vector-like T quarks [282] CMS Collaboration, Searches for third-generation
decaying to top quarks and Higgs bosons in the squark production in fully hadronic
√ final states in
all-hadronic channel using jet substructure, JHEP 06 proton-proton collisions at s = 8 TeV, JHEP 06
(2015) 080, [1503.01952]. (2015) 116, [1503.08037].
[267] CMS Collaboration, Search for vector-like √ charge 2/3 [283] CMS Collaboration,
√ Search for supersymmetry in pp
T quarks in proton-proton collisions at s = 8 TeV, collisions at s = 8 TeV in final states with boosted W
Phys. Rev. D93 (2016) 012003, [1509.04177]. bosons and b jets using razor variables, Phys. Rev.
[268] CMS Collaboration, Search for pair-produced √ vectorlike D93 (2016) 092009, [1602.02917].
B quarks in proton-proton collisions at s=8 TeV, [284] CMS Collaboration, Search for direct pair production
Phys. Rev. D93 (2016) 112009, [1507.07129]. of supersymmetric top quarks decaying to all-hadronic
 47
√
final states in pp collisions at s = 8 TeV, Eur. Phys. events with a hadronically decaying W or Z boson√and
J. C76 (2016) 460, [1603.00765]. missing transverse momentum in pp collisions at s =
[285] CMS Collaboration, Search for supersymmetry in the 8 TeV with the ATLAS detector, Phys. Rev. Lett. 112
all-hadronic final
√ state using top quark tagging in pp (2014) 041802, [1309.4017].
collisions at s = 13 TeV, Phys. Rev. D96 (2017) [301] ATLAS Collaboration, Search for dark matter produced
012004, [1701.01954]. in association with a hadronically
√ decaying vector
[286] CMS Collaboration, Search for supersymmetry in boson in pp collisions at s = 13 TeV with the ATLAS
proton-proton collisions at 13 TeV using identified top detector, Phys. Lett. B763 (2016) 251, [1608.02372].
quarks, Phys. Rev. D97 (2018) 012007, [1710.11188]. [302] ATLAS Collaboration, Search for heavy resonances
[287] CMS Collaboration, Search for physics beyond the decaying
√ to a Z boson and a photon in pp collisions at
standard model in events with high-momentum Higgs s = 13 TeV with the ATLAS detector, Phys. Lett.
bosons and missing transverse momentum in B764 (2017) 11–30, [1607.06363].
proton-proton collisions at 13 TeV, Phys. Rev. Lett. [303] ATLAS Collaboration, A search for resonances
120 (2018) 241801, [1712.08501]. decaying into a Higgs boson and a new particle X in
[288] CMS Collaboration, Search for dark matter in the XH → qqbb final state with the ATLAS detector,
proton-proton collisions at 8 TeV with missing Phys. Lett. B779 (2018) 24, [1709.06783].
0
transverse momentum and vector boson tagged jets, [304] ATLAS Collaboration, Search √ for W → tb → qqbb
JHEP 12 (2016) 083, [1607.05764]. [Erratum: JHEP decays in pp collisions at s = 8 TeV with the ATLAS
08 (2017) 035]. detector, Eur. Phys. J. C75 (2015) 165, [1408.0886].
[289] CMS Collaboration, Search for high-mass√Zγ [305] ATLAS Collaboration, Search for squarks and gluinos
resonances in proton-proton collisions at s = 8 and in final states with jets and missing
√ transverse
13 TeV using jet substructure techniques, Phys. Lett. momentum using 36 fb−1 of s = 13 TeV pp collision
B772 (2017) 363–387, [1612.09516]. data with the ATLAS detector, Phys. Rev. D97 (2018)
[290] CMS Collaboration, Search for dark matter produced 112001, [1712.02332].
with an energetic
√ jet or a hadronically decaying W or [306] ATLAS Collaboration, Search for supersymmetry in
Z boson at s = 13 TeV, JHEP 07 (2017) 014, final states with missing transverse momentum √ and
[1703.01651]. multiple b-jets in proton-proton collisions at s = 13
[291] CMS Collaboration, Search for associated production TeV with the ATLAS detector, JHEP 06 (2018) 107,
√ matter with a Higgs boson decaying to bb or γγ
of dark [1711.01901].
at s = 13 TeV, JHEP 10 (2017) 180, [1703.05236]. [307] ATLAS Collaboration, Search for a scalar partner of
[292] CMS Collaboration, Search for new physics in final the top quark in the jets plus
√ missing transverse
states with an energetic jet or a hadronically decaying momentum final state at s=13 TeV with the ATLAS
W
√ or Z boson and transverse momentum imbalance at detector, JHEP 12 (2017) 085, [1709.04183].
s = 13 TeV, Phys. Rev. D97 (2018) 092005, [308] G. C. Branco, P. M. Ferreira, L. Lavoura, M. N.
[1712.02345]. Rebelo, M. Sher and J. P. Silva, Theory and
[293] CMS Collaboration, Search for Zγ resonances using phenomenology of two-Higgs-doublet models, Phys.
leptonic and √ hadronic final states in proton-proton Rept. 516 (2012) 1–102, [1106.0034].
collisions at s = 13 TeV, JHEP 09 (2018) 148, [309] D. B. Kaplan and H. Georgi, SU(2) x U(1) breaking by
[1712.03143]. vacuum misalignment, Phys. Lett. B136 (1984) 183.
[294] CMS Collaboration, Search for dark matter in events [310] D. B. Kaplan, H. Georgi and S. Dimopoulos,
with energetic, hadronically decaying√ top quarks and Composite Higgs scalars, Phys. Lett. B136 (1984)
missing transverse momentum at s = 13 TeV, JHEP 187–190.
06 (2018) 027, [1801.08427]. [311] H. Georgi, D. B. Kaplan and P. Galison, Calculation of
[295] CMS Collaboration, Search√for W 0 → tb in the composite Higgs mass, Phys. Lett. B143 (1984)
proton-proton collisions at s = 8 TeV, JHEP 02 152–154.
(2016) 122, [1509.06051]. [312] T. Banks, Constraints on SU(2) x U(1) breaking by
[296] CMS Collaboration, Search for the production of an vacuum misalignment, Nucl. Phys. B243 (1984)
excited bottom√ quark decaying to tW in proton-proton 125–130.
collisions at s = 8 TeV, JHEP 01 (2016) 166, [313] H. Georgi and D. B. Kaplan, Composite Higgs and
[1509.08141]. custodial SU(2), Phys. Lett. B145 (1984) 216–220.
[297] CMS Collaboration, Searches for W 0 bosons decaying [314] M. J. Dugan, H. Georgi and D. B. Kaplan, Anatomy of
to a top quark and a bottom quark in proton-proton a composite Higgs model, Nucl. Phys. B254 (1985)
collisions at 13 TeV, JHEP 08 (2017) 029, 299–326.
[1706.04260]. [315] H. Georgi, A tool kit for builders of composite models,
[298] CMS Collaboration, Search for Higgs boson pair Nucl. Phys. B266 (1986) 274.
production in events with two bottom quarks √ and two [316] B. Bellazzini, C. Csáki and J. Serra, Composite
tau leptons in proton-proton collisions at s =13 TeV, Higgses, Eur. Phys. J. C74 (2014) 2766, [1401.2457].
Phys. Lett. B778 (2018) 101, [1707.02909]. [317] L. Randall and R. Sundrum, A Large mass hierarchy
[299] ATLAS Collaboration, Search for dark matter from a small extra dimension, Phys. Rev. Lett. 83
produced in association with a Higgs boson
√ decaying to (1999) 3370, [hep-ph/9905221].
two bottom quarks in pp collisions at s = 8 TeV with [318] K. Agashe, A. Delgado, M. J. May and R. Sundrum,
the ATLAS detector, Phys. Rev. D93 (2016) 072007, RS1, custodial isospin and precision tests, JHEP 08
[1510.06218]. (2003) 050, [hep-ph/0308036].
[300] ATLAS Collaboration, Search for dark matter in [319] H. Davoudiasl, J. L. Hewett and T. G. Rizzo, Bulk
 48

gauge fields in the Randall-Sundrum model, Phys. Lett. 094010, [1306.0572].

B473 (2000) 43, [hep-ph/9911262]. [331] U. Baur, I. Hinchliffe and D. Zeppenfeld, Excited quark
[320] A. Pomarol, Gauge bosons in a five-dimensional theory production at hadron colliders, Int. J. Mod. Phys. A2
with localized gravity, Phys. Lett. B486 (2000) 153, (1987) 1285.
[hep-ph/9911294]. [332] J. L. Hewett and T. G. Rizzo, Low-energy
[321] C. T. Hill, Topcolor assisted technicolor, Phys. Lett. phenomenology of superstring inspired E(6) models,
B345 (1995) 483–489, [hep-ph/9411426]. Phys. Rept. 183 (1989) 193.
[322] A. Collaboration, Search for heavy particles decaying [333] U. Baur, M. Spira and P. M. Zerwas, Excited quark
into top-quark pairs using lepton-plus-jets
√ events in and lepton production at hadron colliders, Phys. Rev.
proton proton collisions at s = 13 TeV with the D42 (1990) 815.
ATLAS detector, Eur. Phys. J. C78 (2018) 565, [334] P. Langacker, The physics of heavy Z 0 gauge bosons,
[1804.10823]. Rev. Mod. Phys. 81 (2009) 1199, [0801.1345].
[323] K. Rehermann and B. Tweedie, Efficient identification [335] H. An, R. Huo and L.-T. Wang, Searching for low
of boosted semileptonic top quarks at the LHC, JHEP mass dark portal at the LHC, Phys. Dark Univ. 2
03 (2011) 059, [1007.2221]. (2013) 50–57, [1212.2221].
[324] CMS Collaboration, Search for tt̄ resonances in [336] A. Rajaraman, W. Shepherd, T. M. P. Tait and A. M.
boosted
√ semileptonic final states in pp collisions at Wijangco, LHC bounds on interactions of dark matter,
s = 13 TeV, CMS Physics Analysis Summary Phys. Rev. D84 (2011) 095013, [1108.1196].
CMS-PAS-B2G-15-002, 2016. [337] J. Goodman, M. Ibe, A. Rajaraman, W. Shepherd,
[325] N. Arkani-Hamed, A. G. Cohen and H. Georgi, T. M. P. Tait and H.-B. Yu, Constraints on dark
Electroweak symmetry breaking from dimensional matter from colliders, Phys. Rev. D82 (2010) 116010,
deconstruction, Phys. Lett. B 513 (2001) 232, [1008.1783].
[hep-ph/0105239]. [338] A. Altheimer et al., Jet Substructure at the Tevatron
[326] M. Schmaltz and D. Tucker-Smith, Little Higgs review, and LHC: New results, new tools, new benchmarks, J.
Ann. Rev. Nucl. Part. Sci. 55 (2005) 229, Phys. G39 (2012) 063001, [1201.0008].
[hep-ph/0502182]. [339] A. Altheimer et al., Boosted objects and jet
[327] I. Antoniadis, K. Benakli and M. Quiros, Finite Higgs substructure at the LHC. Report of BOOST2012, held
mass without supersymmetry, New J. Phys. 3 (2001) at IFIC Valencia, 23rd-27th of July 2012, Eur. Phys.
20, [hep-th/0108005]. J. C74 (2014) 2792, [1311.2708].
[328] Y. Hosotani, S. Noda and K. Takenaga, Dynamical [340] D. Adams et al., Towards an Understanding of the
gauge-Higgs unification in the electroweak theory, Correlations in Jet Substructure, Eur. Phys. J. C75
Phys. Lett. B 607 (2005) 276, [hep-ph/0410193]. (2015) 409, [1504.00679].
[329] K. Agashe, R. Contino and A. Pomarol, The minimal [341] Boost 2016 Participants, Boost Zurich 2016, 2016.
composite Higgs model, Nucl. Phys. B 719 (2005) 165, [Link]
[hep-ph/0412089]. [342] Boost 2017 Participants, Boost Buffalo 2017, 2017.
[330] J. A. Aguilar-Saavedra, R. Benbrik, S. Heinemeyer and [Link]
M. Pérez-Victoria, Handbook of vectorlike quarks:
Mixing and single production, Phys. Rev. D 88 (2013)