EUROPEAN ORGANISATION FOR NUCLEAR RESEARCH (CERN)

JINST 19 (2024) P06029 CERN-EP-2023-299

DOI: 10.1088/1748-0221/19/06/P06029 26th July 2024
arXiv:2401.06630v2 [hep-ex] 25 Jul 2024

The ATLAS Trigger System for LHC Run 3 and

Trigger performance in 2022

The ATLAS Collaboration

The ATLAS trigger system is a crucial component of the ATLAS experiment at the LHC. It
is responsible for selecting events in line with the ATLAS physics programme. This paper
presents an overview of the changes to the trigger and data acquisition system during the
second long shutdown of the LHC, and shows the performance of the trigger system and its
components in the proton–proton collisions during the 2022 commissioning period as well as
its expected performance in proton–proton and heavy–ion collisions for the remainder of the
third LHC data-taking period (2022–2025).

© 2024 CERN for the benefit of the ATLAS Collaboration.

Reproduction of this article or parts of it is allowed as specified in the CC-BY-4.0 license.
Contents

1 Introduction 4

2 The Run-3 ATLAS detector 4

2.1 Overview of ATLAS Trigger and Data Acquisition System 6
2.2 The ATLAS run structure 7

3 Changes to the Trigger/DAQ system for Run 3 7

3.1 Level-1 calorimeter trigger 7
3.2 Level-1 muon trigger 9
3.3 Level-1 topological trigger 12
3.4 HLT software 13
3.5 DAQ system 15
3.6 Validation strategy 16

4 Trigger menu 17
4.1 Baseline physics trigger menu for proton–proton collisions 19
4.2 Baseline physics trigger menu for heavy–ion collisions 21

5 High-level trigger reconstruction 23

5.1 Inner detector tracking 23
5.1.1 Tau lepton tracking 24
5.1.2 Muon tracking 24
5.1.3 Electron tracking 25
5.1.4 Full scan tracking and vertex finding 26
5.1.5 Tracking in the jets containing 𝑏-hadrons 30
5.1.6 Large radius tracking 31
5.1.7 Execution time 32
5.2 Calorimeter reconstruction 37
5.3 Tracking in the muon spectrometer 41

6 Physics triggers 44
6.1 Electrons and photons 44
6.1.1 Electron and photon trigger reconstruction and selection 44
6.1.2 Electron and photon trigger menu 45
6.1.3 Electron and photon trigger performance 45
6.1.4 Trigger level analysis with photons 45
6.1.5 Electron and photon triggers for heavy-ion collisions 47
6.2 Muons 48
6.2.1 Muon trigger menu 48
6.2.2 Muon trigger performance 49
6.3 Taus 49
6.3.1 Tau trigger reconstruction and selection 49
6.3.2 Tau trigger menu 51
6.3.3 Tau trigger performance 52

2
 6.4 Jets 53
6.4.1 Jet trigger reconstruction 53
6.4.2 Jet trigger menu 54
6.4.3 Jet trigger performance 55
6.4.4 Trigger level analysis with jets 57
6.4.5 Jet triggers for heavy-ion collisions 58
6.5 Jets containing 𝑏-hadrons 58
6.5.1 𝑏-jet trigger reconstruction and selection 58
6.5.2 𝑏-jet trigger menu 60
6.5.3 Trigger level analysis with 𝑏-jets 61
6.6 Missing transverse momentum 62
6.6.1 𝐸 Tmiss trigger reconstruction 62
6.6.2 𝐸 Tmiss trigger performance and menu 63
6.7 𝐵-physics and light states 63
6.8 Minimum-bias and forward signatures 65
6.8.1 Minimum-bias triggers 66
6.8.2 Forward triggers 69
6.9 Unconventional tracking signatures 70
6.9.1 Long-lived charged particles that partially or fully traverse the inner detector 70
6.9.2 Long-lived particle decays into jets 72
6.9.3 Long-lived particle decays into SM leptons 75

7 Auxiliary triggers 76
7.1 Non-collision background 76
7.1.1 Cosmic ray triggers 77
7.1.2 Beam-induced background triggers 77
7.2 Zero-bias trigger 78
7.3 Triggers for tracking performance studies 79
7.4 Triggers for detector performance studies (calibration, noise) 80
7.4.1 LArNoiseBurst algorithm 80
7.5 Beam-spot algorithms 81

8 Trigger software performance 83

8.1 Trigger software performance scaling 83
8.2 HLT processing time 85

9 Conclusion 88

3
1 Introduction

The trigger system [1, 2] is an essential component of the ATLAS experiment [3] as it is responsible
for deciding whether or not to permanently record data from a given bunch-crossing interaction (also
referred to as an event) for further study. It aims to minimise the rate of the recorded events while
maintaining an excellent and unbiased efficiency for physics processes of interest. During the Run 1
(2009 – 2013) and Run 2 (2015 – 2018) operational periods at the Large Hadron Collider (LHC), the
ATLAS trigger system collected data from proton–proton (𝑝 𝑝) collisions at a centre-of-mass energy up to
√
𝑠 = 13 TeV, a peak instantaneous luminosity of 2.1 × 1034 cm−2 s−1 and up to a mean of 60 interactions
per bunch crossing (pile-up). As part of the LHC heavy–ion (HI) programme, lead-lead nuclei collisions
√
(Pb+Pb) at a centre-of-mass energy up to 𝑠 = 5 TeV per nucleon and a peak instantaneous luminosity of
6.2 × 1027 cm−2 s−1 were achieved, with other ion species and collision configurations used in addition.
During the second LHC Long Shutdown (LS2, 2019 – 2021), the ATLAS trigger system underwent a
major upgrade. The purpose of this upgrade was to enhance the physics reach of the experiment for the
ongoing operation in Run 3 (2022 – 2025). In Run 3 the centre-of-mass energy of the 𝑝 𝑝 collisions is
√
𝑠 = 13.6 TeV with an instantaneous luminosity planned to be kept constant at its peak value, approximately
2.4 × 1034 cm−2 s−1 for a duration as long as ten hours per LHC fill and 60–70 interactions per bunch
crossing.
The ATLAS Run-3 trigger system, its performance in the 𝑝 𝑝 collisions during the 2022 commissioning
period, including rates and efficiencies, as well as expected trigger algorithm modifications for HI data
taking are described in this paper. After a brief introduction to the ATLAS detector in Section 2, Section 3
summarises the changes to the trigger and data acquisition during LS2. Section 4 gives overviews of trigger
selections implemented for Run 3 followed by an introduction to the reconstruction algorithms used at the
High Level Trigger (HLT) in Section 5. The implementation of different triggers and their performance are
discussed in Sections 6 and 7. The trigger software performance is presented in Section 8.
The results presented in this paper are based on the 𝑝 𝑝 collision data recorded by the ATLAS experiment
and the corresponding simulated Monte Carlo (MC) events. The rates are quoted for an instantaneous
luminosity value of 1.8 × 1034 cm−2 s−1 . The integrated luminosity used varies over the figures and is
measured to be at most 31 fb−1 for the 2022 data set [4]. Due to the absence of HI physics data taking in
2022, there are no HI results to show in this paper.
Trigger efficiencies are calculated with respect to either reconstructed or true physics objects from MC
simulation. Typical methods used to determine trigger efficiency in data [5] are either the tag-and-probe or
bootstrap methods. The tag-and-probe method is based on a sample of events with a known resonance,
such as 𝑍 boson or 𝐽/𝜓, decaying to two charged leptons. The bootstrap method involves successive
measurements of trigger efficiency with respect to unbiased or lower-threshold triggers. MC generator
information about the physics objects can be used to determine trigger efficiency in the MC simulation.

2 The Run-3 ATLAS detector

The ATLAS detector at the LHC has undergone a substantial upgrade with improvements to various
detector subsystems and their electronics, in order to enable the broad physics programme planned for the
Run-3 data taking. The original configuration of the detector (as it was built for the start of Run 1 of the
LHC) is described in Ref. [6].

4
The ATLAS detector covers nearly the entire solid angle around the collision point1 . It consists of an inner
tracking detector surrounded by a superconducting solenoid, electromagnetic and hadron calorimeters, and
a muon spectrometer incorporating three large superconducting air-core toroidal magnets.
The Inner Detector (ID) system is immersed in a 2 T axial magnetic field and provides charged-particle
tracking in the range |𝜂| < 2.5. The high-granularity silicon pixel detector (PIX) covers the vertex region
and typically provides four measurements per track, the first hit normally being in the insertable B-layer
(IBL) installed before Run 2 [7]. It is followed by the silicon microstrip tracker (SCT), which usually
provides eight measurements per track. These silicon detectors are complemented by the outermost of the
three tracking subsystems, the transition radiation tracker (TRT), which enables radially extended track
reconstruction up to |𝜂| = 2.0. The TRT also provides electron identification information based on the
fraction of hits (typically 30 in total) above a higher energy-deposit threshold corresponding to transition
radiation. The TRT gas configuration has a significant impact on the particle identification. Due to a
number of leaks in flexible active gas exhaust pipes that developed during Run 1 and Run 2, it became too
costly to operate the entire detector with the baseline xenon-based gas mixture. For Run 3 an argon-based
gas mixture is used in the entire barrel and within a few endcap wheels on one side of the detector [3].
Due to poor absorption of transition radiation photons by the argon gas, the particle identification function
is significantly reduced in the barrel region. However, in combination with d𝐸/d𝑥 measurements, it still
contributes to the ATLAS electron identification, particularly at particle energies below 10 GeV. The
particle identification performance of the endcaps is largely preserved.
ATLAS uses two sampling calorimeter technologies covering the pseudorapidity range |𝜂| < 4.9. Within
the region |𝜂| < 3.2, electromagnetic (EM) calorimetry is provided by barrel and endcap high-granularity
lead/liquid-argon (LAr) calorimeters, consisting of three layers with varying granularities. In addition,
a LAr presampler covering |𝜂| < 1.8 corrects for energy loss in material upstream of the calorimeters.
Hadron calorimetry is provided by the steel/scintillator-tile calorimeter (Tile calorimeter), segmented
into three barrel structures within |𝜂| < 1.7, and two copper/LAr hadron endcap calorimeters. The solid
angle coverage is completed with forward copper/LAr and tungsten/LAr calorimeter modules optimised
for electromagnetic and hadronic energy measurements respectively. During LS2, the LAr calorimeter
electronics was augmented with a new digital trigger path [8] providing finer granularity inputs to the
upgraded trigger system discussed in Section 3.
The muon spectrometer (MS) comprises separate trigger and high-precision tracking chambers measuring
the deflection of muons in a magnetic field generated by the superconducting air-core toroidal magnets. The
field integral of the toroids ranges between 2.0 and 6.0 T m across most of the detector. Three stations of
precision chambers, each consisting of layers of monitored drift tubes (MDTs), cover the region |𝜂| < 2.7.
In the innermost station of the endcaps, |𝜂| > 1.3, the detectors used in Runs 1 and 2, Small Wheels, have
been replaced by the New Small Wheels (NSWs) [3]. The NSWs use two technologies: small-strip thin gap
chamber (sTGC) and Micromegas (MM) detectors, both with high-rate tolerance and improved resolution.
The muon trigger system covers the range |𝜂| < 2.4 with resistive plate chambers (RPCs) in the barrel
(|𝜂| < 1.05), and thin gap chambers (TGCs) in the endcap regions (1.05 < |𝜂| < 2.4). In response to the
increasing number of gas leaks due to cracks in the gas inlets of the RPC system that developed over time
during Run 2, significant work was undertaken during LS2 to reinforce the RPC gas inlets and recover a
large number of channels that had become inactive. Inlets were repaired and no-return valves installed.

1 ATLAS uses a right-handed coordinate system with its origin at the nominal interaction point (IP) in the centre of the detector
and the 𝑧-axis along the beam pipe. The 𝑥-axis points from the IP to the centre of the LHC ring, and the 𝑦-axis points upwards.
Polar coordinates (𝑟, 𝜙) are used in the transverse plane, 𝜙 being the azimuthal angle around the 𝑧-axis. The
√︁ pseudorapidity is
defined in terms of the polar angle 𝜃 as 𝜂 = − ln tan(𝜃/2). Angular distance is measured in units of Δ𝑅 ≡ (Δ𝜂) 2 + (Δ𝜙) 2 .

5
 Calorimeter detectors

LAr TileCal
Muon detectors (including NSW) LHC collision rate & event size
40 MHz 3.0 MB
Detector
Level-1 Calo Level-1 Muon Read-Out
Pre-processor TileCal Endcap Barrel
FE
nMCM TREX via TREX sector logic sector logic FE FE
... Level-1 accept rate
FELIX
100 kHz 300 GB/s
CP (e,γ,τ) JEP (jet, E) e/j/g
MUCTPI ROD ROD

Level-1 Accept
CMX CMX
CMX FEX

DataFlow
L1Topo
L1Topo
Legacy
Read-Out System
CTP
(ROS / Software ROD)
CTPCORE
CTPOUT
...
Central Trigger
Level-1 Trigger
Data Collection Network
RoI

HLT output to storage

Data Storage
High Level Trigger 3 kHz 6 GB/s
(HLT)
Accept
Processors
Event
Data
Tier-0

Figure 1: The ATLAS TDAQ system in Run 3 with emphasis on the components relevant for triggering as well as the
detector read-out and data flow. Level-1 Calo, Level-1 Muon and the Central Trigger all send data to the Read-Out
System (or FELIX/Software ROD), described in Section 3.5, primarily for the purposes of offline validation and
error checking. This is not shown in the diagram for simplicity. Abbreviations used are defined in Sections 2.1 and 3.

Such maintenance work is expected to continue throughout Run 3 in periods with no data taking. The
resulting impact on the muon trigger efficiency is described in Section 3.

2.1 Overview of ATLAS Trigger and Data Acquisition System

The selection and recording of events is handled by the Trigger and Data Acquisition (TDAQ) system
shown in Figure 1.
The first-level (Level-1, L1) trigger is mainly based on two independent systems which use custom
electronics to trigger on reduced-granularity information from either the calorimeters (L1Calo) or the
muon detectors (L1Muon). The L1 topological processor (L1Topo) system uses kinematic information
from objects reconstructed in the L1Calo and L1Muon systems and applies topological selections. Changes
to these systems for Run 3 are described in detail in Section 3.
The L1 trigger decision is formed by the Central Trigger Processor (CTP), based on inputs received from
the L1Calo trigger system, the L1Muon trigger system through the Muon-to-Central Trigger Processor
Interface (MUCTPI) [9], the L1Topo system as well as several other subsystems. These subsystems are
the Minimum Bias Trigger Scintillators (MBTS) [10], the ATLAS Forward Proton (AFP) [11] detector
and ALFA detector [12] discussed further in Section 6.8; the LUCID-2 Cherenkov Counter [13], used for
the primary luminosity measurements [14] complemented by measurements using the inner detector and
calorimeters; and the Zero Degree Calorimeter (ZDC) [15] which is installed for heavy ion data taking (see
Section 4.2). The CTP is also responsible for applying dead time, a mechanism to limit the number of
close-by L1 accepts [16].

6
As detailed in Section 4, events that satisfy the trigger menu requirements based on object type, threshold
and multiplicity are accepted at a rate up to the maximum detector read-out rate of 100 kHz (down from
the bunch crossing rate of about 40 MHz) at a fixed latency (detector read-out time window) below 2.5 𝜇𝑠.
Up to 512 distinct L1 trigger items may be configured in the CTP.
If accepted by the L1 trigger, events are then sent to a software-based HLT. Here online algorithms
reconstruct the event at progressively higher levels of detail than at L1, either in the full detector volume or
in restricted Regions-of-Interest (RoIs), which are detector regions in which candidate trigger objects have
been identified by the L1 trigger. The HLT software is incorporated in the same software framework as is
used offline to reconstruct recorded events. For Run 3, this software framework was redesigned to support
multi-threaded execution [17] as detailed in Section 3.4. The physics output rate of the HLT during an
ATLAS data-taking run is expected to be 3 kHz on average (see Section 4 for more details).
The Data Acquisition (DAQ) system [18], described in Section 3.5, transports data from custom subdetector
electronics to offline processing, according to the decisions made by the trigger. Data are compressed prior
to processing from the original event size of about 3 MB to below 2 MB. An extensive software suite [19]
is used in the reconstruction and analysis of real and simulated data, in detector operations, and in the
trigger and data acquisition systems of the experiment.

2.2 The ATLAS run structure

An ATLAS run is a period of data acquisition with a stable detector configuration. In the case of collecting
data for physics analyses (physics data taking), it usually coincides with an LHC fill, which can last many
hours. A unique number is assigned to every run at its beginning by the DAQ system. A run is divided
into Luminosity Blocks (LB), defined as intervals of approximately constant instantaneous luminosity and
stable detector conditions (including the trigger system and its configuration), with a nominal length of
one minute. To define a data sample appropriate for physics studies, quality criteria are applied to select
LBs where conditions are acceptable for a particular analysis. The instantaneous luminosity in a given
LB is multiplied by the LB duration to obtain the integrated luminosity delivered in that LB. From a data
quality point of view, the LB represents the smallest quantity of data that can be declared good or bad for
physics analysis. Further details on the LHC fill cycle, fill patterns and ATLAS run structure can be found
in Ref. [20].

3 Changes to the Trigger/DAQ system for Run 3

This section describes changes in the L1Calo, L1Muon and L1Topo systems as well as the HLT software
and DAQ system implemented for the Run 3, including some initial performance data where available.
More details about the hardware upgrades can be found in Ref. [3].

3.1 Level-1 calorimeter trigger

During LS2, the L1Calo trigger was upgraded to perform on-detector digitisation of transverse energies
from the LAr calorimeters. With this change in Run 3, L1Calo can receive finer-granularity input data from
the LAr calorimeter. While in Runs 1 and 2, these inputs consisted of trigger towers spanning 0.1 × 0.1 in
𝜂 and 𝜙, the electromagnetic calorimeter information is now provided in the form of SuperCells containing

7
Figure 2: The trigger granularity from each 0.1 × 0.1 trigger tower after the upgrade of the LAr Calorimeter electronics.
Ten 𝐸 T values are provided from “1-4-4-1” longitudinal/transverse samples, each forming a SuperCell.

sums of four or eight calorimeter cells. Each trigger tower contains ten SuperCells as shown in Figure 2.
These trigger towers are distributed to Feature Extraction (FEX) processors. In this system, the digitisation
and calibration of the LAr calorimeter data are now performed in the LAr calorimeter electronics. The Tile
calorimeter data are still received in analog format, digitised by the pre-processors as shown in Figure 1
and detailed below, and transmitted by the new Tile Rear EXtension (TREX) modules to both the FEXs
and the Run-2 L1Calo system (hereafter referred to as the legacy system).
The upgraded L1Calo system includes new electromagnetic (eFEX) and jet (jFEX) feature extractors
as well as a global feature extractor (gFEX). The full SuperCell granularity is available at the eFEX to
reconstruct EM objects and hadronically decaying 𝜏 leptons, as well as shower-shape variables used for
their identification [3]. Three programmable threshold parameters (commonly referred to as loose, medium
and tight) are available for three shower-shape variables for the EM triggers and one shower-shape variable
for tau triggers. The resulting threshold pass bits for each shower shape variable are sent to the Run-3
L1Topo system and allow for the configuration of different levels of background rejection for these triggers.
In the region with |𝜂| < 2.5 jFEX receives towers with a granularity of 0.1 × 0.1 in 𝜂 − 𝜙 space, which is a
factor of two better in both dimensions than the inputs used for the Run-2 L1-jet-trigger system. In the
region with |𝜂| > 2.5 the granularity is similar to Run 2 with some improvements in the far forward region.
The jFEX is used to reconstruct small-radius (small-𝑅) jets with 𝑅 = 0.4. Its performance is expected to be
similar to that of the legacy L1Calo system for the single jet triggers, while improving the reconstruction for
the nearby jets in the multi-jet triggers. In addition, jFEX brings new capability to reconstruct large-radius
(large-𝑅) jets with 𝑅 = 0.8, hadronically decaying 𝜏 leptons in the range of |𝜂| ≤ 2.5, and electromagnetic
objects in the forward region of 2.3 ≤ |𝜂| ≤ 4.9. The latter two triggers have an optional isolation
requirement. Missing transverse momentum (𝐸 Tmiss ) is computed by summing transverse energy values in
slices of constant 𝜙 and then weighted by cos(𝜙) and sin(𝜙) in order to determine the 𝑥- and 𝑦-components,
respectively. The gFEX has been designed with a coarser granularity than jFEX (similar to the Run-2
system) so that the data from the entire calorimeter can be processed on a single module, facilitating
the identification of boosted objects and global observables. The gFEX identifies large-𝑅 jets within a
“circular” 1.8 × 1.8 area (𝑅 < 0.9) and provides local pile-up density and substructure information. The
gFEX computes 𝐸 Tmiss by separating the transverse energy (𝐸 T ) sums into “hard” and “soft” where the hard
term consists of the 𝐸 T sum of towers satisfying 𝐸 T > 25 GeV and the soft term consists of the 𝐸 T sum of
the remaining towers. The 𝐸 Tmiss is then computed as a linear combination of the hard and soft terms.

8
Both jFEX and gFEX have energy sum algorithms based on vector and scalar sums of 𝐸𝑇 . The pile-up
subtraction algorithms introduced in jFEX and gFEX improve L1 jet and energy sum triggers efficiency,
rates and purity. More details on all these algorithms can be found in Ref. [3]. Complementary information
on the same types of objects (e.g. hadronically decaying 𝜏 leptons from eFEX and jFEX or 𝐸 Tmiss from jFEX
and gFEX) can be used in the Run-3 L1Topo selection to further improve their L1 trigger performance.
The legacy L1Calo system [2, 21] can be operated in parallel to the upgraded L1Calo system in Run 3. In
the legacy system, analogue detector signals are digitised and calibrated in the multi-chip modules (nMCM)
in the pre-processor system and sent in parallel to the Cluster Processors (CP) and the Jet/Energy-sum
Processors (JEP). The CP system identifies electron, photon, and 𝜏-lepton candidates above a programmable
threshold, and the JEP system identifies jet candidates and produces global sums of total and missing
transverse energy.
During the commissioning and validation of the upgraded L1Calo system in 2022, calorimeter-based
physics triggers were provided by the legacy L1Calo system. In parallel, the electromagnetic objects for
the new Run-3 L1Calo trigger were commissioned to be used from the start of 2023 data taking. After
final tuning, they allow for a significant reduction in the L1 trigger rate of up to 10 kHz and an improved
efficiency for electron triggers requiring isolation. Figure 3 shows the 2022 performance of the legacy
trigger in comparison to the new one for the barrel region.2 At the time of submission of this manuscript,
the final optimisation of the Run-3 L1Calo trigger system was still ongoing.

70 400
eFEX EM ET [GeV]
L1 EM efficiency

Count
65 ATLAS
1 350
60
|η| ≤ 0.8 300
0.8 55
ATLAS 50 250
0.6 Data 2022, s=13.6 TeV 45
Z→ ee tag-and-probe, |η| ≤ 0.8 200
40 150
0.4 legacy E >22 GeV
T
legacy E >22 GeV, isolated 35
T 100
eFEX ET>26 GeV 30
0.2 eFEX ET>26 GeV with identification
25 50

0 20 0
20 30 40 50 60 70 20 25 30 35 40 45 50 55 60 65 70
Offline electron ET [GeV] Legacy EM E [GeV]
T

Figure 3: (left) Efficiency of the L1 single-object electromagnetic (EM) trigger and (right) the correlation of the
L1Calo transverse energies measured by the legacy Cluster Processor system and the Run-3 eFEX. Both measurements
are done in the inner EM barrel within |𝜂| < 0.8. Isolated legacy trigger [5] and eFEX trigger with identification [3]
have the same rate. Only statistical uncertainties are shown.

3.2 Level-1 muon trigger

The L1Muon trigger uses hits from the RPCs (in the barrel) and the TGCs (in the endcaps) to determine
the deviation of the hit pattern from that of a muon with infinite momentum [22]. For Run 3 the L1Muon

2 For Run 3, the L1 threshold nomenclature is changed so that the threshold value corresponds to a 50% efficiency point, not the
efficiency plateau as was the case in Run 2. Thus L1 Run-3 triggers with thresholds of 14 GeV for a single muon and 8 GeV for
a pair of muons have the same performance, respectively, as a 20 GeV single muon and a 10 GeV di-muon L1 triggers in Run 2.

9
transverse momentum (𝑝 T ) thresholds have been redefined to improve the performance of both the high-𝑝 T
single muon and the low-𝑝 T multi-muon triggers.
To reduce the rate of the low-momentum charged particles in the endcap regions, the L1Muon trigger in
Run 2 applied coincidence requirements between the outer TGC station and either the inner TGC stations
or the tile calorimeter. In Run 3, the replacement of the Small Wheels by the NSWs allows for a further
rate reduction. Additional RPC modules have also been deployed in the inner barrel station for Run 3.
These new Barrel Inner Small (BIS78) chambers are located in the transition 1.0 < 𝜂 < 1.3 region between
the barrel and endcap on one side of the detector only and will be utilised to reduce the trigger rate of the
endcap region which is not covered by inner endcap TGC stations. These new chambers are considered to
be a pilot project for the inner barrel upgrade in preparation for Run 4 (2029–2032) when the second side
of BIS78 will be installed.
L1 efficiency

L1 efficiency
1.2 ATLAS 1
Data 2022, s=13.6 TeV
1 Z→µµ tag-and-probe, 1.05<|η|<2.4
0.8

0.8 L1 p threshold
T
ATLAS
3 GeV (2 or 3 stations)
4 GeV (2 or 3 stations)
0.6 Data 2022, s = 13.6 TeV
5 GeV (2 or 3 stations) Z → µµ tag-and-probe
0.6
1.05 < |η| < 2.4
6 GeV (2 or 3 stations)
7 GeV (3 stations)
8 GeV (3 stations)
9 GeV (3 stations)
0.4
0.4 10 GeV (3 stations) L1 p threshold
T
11 GeV (3 stations)
12 GeV (3 stations) 3 GeV
13 GeV (3 stations)
0.2
0.2 14 GeV (3 stations)
15 GeV (3 stations)
3 GeV with correctly identified charge
18 GeV (3 stations)
20 GeV (3 stations)
0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Offline muon p [GeV] Offline muon p [GeV]
T T

Figure 4: (left) Efficiency of L1 muon triggers in the endcap region for various 𝑝 T thresholds. (right) The efficiency
of the charge identification for L1 muons with 𝑝 T > 3 GeV. Only statistical uncertainties are shown.

The upgrade of the endcap trigger processor board, called Endcap Sector Logic, increases the number of
TGC thresholds from 6 to 15 and supports the attaching of up to four flags to muon candidates. Figure 4 (left)
shows the efficiency of single-muon triggers for the endcap region. The finer granularity of low-𝑝 T TGC
thresholds results in a better resolution of the invariant mass of two muons in the Run-3 L1Topo system,
crucial for 𝐵-physics and light states triggers discussed in Section 6.7. The rate of these low-𝑝 T multi-muon
triggers can be further reduced by using information on the charge of the muon candidate, which is provided
by one of the new flags. As can be seen from Figure 4 (right), the accuracy of muon charge identification is
close to 100% at low 𝑝 T , although it gets worse with increasing 𝑝 T , due to the reduction of curvature of
the muon path in the magnetic field. The other three flags indicate the following conditions: the muon
candidate satisfies the coincidence in all three outer stations of the TGC (the Full station flag); the Inner
Coincidence flag described in the previous paragraph (in 2022 only the inner endcap TGC stations in the
region of 1.05 < |𝜂| < 1.3 were used to satisfy it, outside this region the flag was always set); the muon
candidate does not pass through regions of weak magnetic field, where the momentum resolution of such
candidates is poor (the Hot-RoIs flag). All three flags were required by default for the 2022 single L1 muon
triggers. The di-muon triggers use only the Full station flag.
Muon triggers in the barrel region are provided by three concentric layers of RPC doublets, with the trigger
decision relying on coincidence logic. For the low-𝑝 T thresholds, the coincidence of three out of four
layers in the two inner doublets of the RPC, which was required in Run 2, was relaxed during 2022 by
requiring at least one hit in each of the two doublets. This increased the trigger efficiency with a minimal

10
 L1 efficiency
L1 efficiency
ATLAS
1 ATLAS combination 1
Data 2022, s=13.6 TeV
Data 2022, s=13.6 TeV one L1 BOM muon p >10 GeV
Z→µµ tag-and-probe, |η|<1.05
T

0.8 J/ψ → µµ (|ηµ|<1.05) two L1 muons p >10 GeV 0.8

T

0.6 0.6
L1 p threshold
T

4 GeV (2 stations)
0.4 0.4
6 GeV (2 stations)
8 GeV (2 stations)

0.2 0.2 10 GeV (3 stations)

12 GeV (3 stations)
14 GeV (3 stations)
0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0 5 10 15 20 25 30 35 40 45 50
∆Rµµ Offline muon p [GeV]
T

Figure 5: (left) The efficiency of trigger reconstruction with respect to offline muon selection as a function of distance
(Δ𝑅 𝜇𝜇 ) between two close-by muons from 𝐽/𝜓 decay in the barrel region for a L1 trigger with two muon candidates
of 𝑝 T > 10 GeV and a L1 trigger with a single Barrel-Only (BO) muon candidate of 𝑝 T > 10 GeV and close-by flag (M).
(right) Efficiency of L1 muon triggers for the barrel region for various 𝑝 T thresholds. Only statistical uncertainties
are shown.

impact on the rates. The high-𝑝 T trigger thresholds require the low-𝑝 T trigger logic to be satisfied, as
well as an additional hit in one of two layers in the outer barrel station. The number of RPC thresholds
remains the same as in Run 2, with a total of six thresholds: three low-𝑝 T (2-station) and three high-𝑝 T
(3-station) thresholds. The firmware of the RPC Pad and of the Barrel Sector Logic boards is upgraded
to flag the possibility of two muons in a single tower3 , which would otherwise be identified as only one
candidate. By propagating this information to the HLT, the identification of close-by muons (e.g. from
boosted 𝐽/𝜓 decays) is enhanced, as shown in Figure 5 (left). Figure 5 (right) shows the L1Muon efficiency
of various single muon triggers in the barrel region with respect to offline muon selection. Their maximum
efficiencies are about 5% lower than at the end of Run 2, due to inefficiencies in the RPC detectors mainly
caused by leaks in the gas distribution system. The lower L1 efficiency of the 3-station triggers is due to
the additional hit requirements at the outer station and the detector coverage.
The MUCTPI is upgraded to provide full-granularity muon RoI information to the L1Topo system and to
be able to interface with the new Endcap Sector Logic. A special SL2MuCTPI board [3] is introduced
to interface the Barrel Sector Logic with the new MUCTPI. The MUCTPI receives muon candidate
information from the barrel and endcap Sector Logic and calculates the overall muon candidate multiplicity
for each 𝑝 T threshold, taking into account the possible overlap between trigger sectors in order to avoid
double counting of muon candidates. Additional logic in the MUCTPI allows for the exclusion of candidates
from the RPC feet areas, or to flag Barrel-Only and Endcap-Only muon candidates. The MUCTPI sends to
the CTP information for up to 32 combinations of the 𝑝 T thresholds and flags, which is an increase from
the six available in Run 2. Combinations with different thresholds in the barrel and endcap regions can be
also implemented.
The commissioning of the L1Muon trigger proceeded systematically in several steps. It started with
the commissioning of the upgraded endcap trigger system using test pulses injected into the on-detector
electronics. This allowed for the detection of hardware problems and incorrect fibre connections, as well as
the adjustment of the trigger timing and the validation of the trigger logic. Subsequently, the use of cosmic
rays provided a full integration test with the barrel and endcap trigger systems with the new MUCTPI
3 A tower is a region of approximately 0.2 × 0.2 in 𝜂 − 𝜙 space comprising four possible RoI positions which are read out by a
Pad Logic board.

11
and the CTP. The commissioning of the upgraded L1Muon system (without the NSW) was finalised in
2022, while the commissioning of the NSW and the BIS78 RPC chambers is still ongoing at the time of
writing.

3.3 Level-1 topological trigger

The addition of the L1Topo trigger in 2016 allowed for both the reduction of energy thresholds without an
increase in trigger rates as well as a higher complexity of the algorithms available at L1 through topological
selections. This resulted in a significant improvement of the background rejection and enhanced acceptance
of physics signal events, despite the increase of luminosity during Run 2. A technical description of the
Run-2 L1Topo system can be found in Ref. [23].
The principle of operation of the L1Topo system remains the same between Run 2 and Run 3. The L1Topo
system receives Trigger OBjects (TOBs) containing kinematic (e.g. 𝐸 T , position) and further qualifying
information (e.g. flags) from the L1Calo and L1Muon systems and applies topological selections. The
Run-3 upgrades of both the L1Calo and the L1Muon systems have an impact on the L1Topo, since they
provide the input data directly for the topological selections. The outputs from the L1Calo and the MUCTPI
are remapped in the Topo-Fiber Optic eXchange (TopoFOX) module before arriving at the L1Topo, contrary
to the Run-2 system where the Common Merger Module (CMX) and MUCTPI2Topo [24] were needed for
e.g. TOB sorting. In Run 3, the L1Topo hardware is upgraded with three new modules (TOPO1, TOPO2
and TOPO3) designed to deal with the new input format. Each module has two Field Programmable Gate
Array (FPGA) processors on which to run the algorithms.

Figure 6: Distribution of L1 TOB inputs for TOPO2 and TOPO3 modules.

In Run 3, the selections based on the multiplicities of the L1Calo objects passing pre-defined energy
thresholds are performed in the TOPO1 module. These algorithms count the number of objects passing a
given threshold, e.g. 𝐸 T , or located within a given region in 𝜂. The output multiplicity bits are transmitted to
the CTP. The number of available thresholds and multiplicity bits per L1Calo object type is configurable in
L1Topo firmware and could be modified in the course of Run 3 according to the trigger menu requirements
discussed in Section 4.
The TOPO2 and TOPO3 modules are used for the topological selections as in the Run-2 L1Topo system [23].
Available inputs for these two L1Topo models, shown in Figure 6, impose constraints on the possible

12
combinations of L1Calo and L1Muon TOBs in topological algorithms. The L1Topo algorithms are
distributed across the FPGAs as evenly as possible to achieve optimal resource usage. The configuration
of algorithms with their parameters is stored in the trigger menu. The L1Topo algorithm decisions are
transmitted to the CTP and are limited to 32 bits per FPGA.
The L1Topo legacy system consists of two Run-2 L1Topo modules with their original Run-2 firmware
configuration, but with inputs only from the legacy L1Calo system and not from L1Muon. The decisions
of only 22 out of 113 topological algorithms that were implemented in Run 2 [23] are sent to the CTP in
Run 3. The L1Topo legacy system was used to collect physics events in 2022, while the upgraded system
underwent commissioning. When the three new L1Topo modules are fully validated, the two L1Topo
legacy modules will be disconnected.

3.4 HLT software

The ATLAS software framework Athena [25] was used in Runs 1 and 2 at all stages of the event data
processing path, from detector simulation to event reconstruction and physics analysis (referred to as
“offline”), as well as for the HLT selections (referred to as “online”). Athena is based on the inter-experiment
framework Gaudi [26]. In Run 2, memory limitations of the simulation and reconstruction workflows
led to the development of a multi-process event processing model (AthenaMP), which was adopted as
an intermediate solution, also at the HLT. In this approach, the main process is forked after initialisation
into a number of worker processes equal to the number of events which should be processed in parallel.
Each worker processes events independently using a single thread, sharing read-only memory with other
workers via the copy-on-write mechanism. Thanks to this mechanism, a reduction of the overall memory
requirements was possible, with respect to the instantiation of totally independent processes.
A typical online reconstruction sequence makes use of dedicated fast trigger algorithms to provide early
background rejection, followed by more precise and CPU-intensive algorithms that are similar or identical
to those used for offline reconstruction to make the final selection. Reconstruction algorithms process
detector data to extract features and hypothesis algorithms test the selection hypothesis for all active trigger
chains, where a chain consists of a L1 trigger item and a series of HLT algorithms organised into distinct
steps that reconstruct physics objects and apply kinematic selections to them.
For Run 3, the ATLAS software framework was adapted to support multi-threaded execution (AthenaMT),
based on the concurrent-processing version of Gaudi, named GaudiHive [27] which is itself based on
Intel’s Threading Building Blocks (TBB) library [28]. HLT requirements were taken into account during
this transition [29]. AthenaMT allows greater memory sharing across compute cores than is possible with
AthenaMP and, consequently, greater flexibility and efficiency when running on hardware with limited
memory per core. It is also a prerequisite for the eventual use of compute accelerators such as GPUs in
online and offline data processing. All data processing steps (except event generation) will use AthenaMT
in Run 4.
Three types of parallelism are included: inter-event parallelism (multiple events are processed in parallel),
intra-event parallelism (multiple algorithms can run in parallel for a single event) and in-algorithm
parallelism (algorithms can internally utilise multi-threading and vectorisation). The execution order of the
job’s algorithms is determined by the input data required, and output data produced, by each algorithm.
These data dependencies must be exposed by the developer to the AthenaMT scheduler via “data handles”,
which are used by the scheduler to appropriately order the algorithms for execution. The execution depends
on the configured number of threads (made available to execute algorithms) and event slots (made available

13
to concurrently process events). Asynchronous or time-varying data (conditions data) whose lifetime
can be longer than an event are handled by the framework with special conditions algorithms, which are
scheduled appropriately by the framework so that the required conditions data are available when needed.
To take advantage of the major changes in AthenaMT, the ATLAS HLT framework was to a large extent
redesigned and rewritten. While in the Run-2 HLT framework all trigger algorithms were implemented
using an HLT-specific interface (with offline algorithms requiring a wrapper), the Run-3 framework requires
no HLT-specific interfaces and takes full advantage of the AthenaMT scheduler and other AthenaMT
components to control event execution in the HLT. These modifications eased online integration of the offline
reconstruction developments due to the seamless integration between the offline and HLT frameworks.
The development of the AthenaMT framework was a common project of the offline and HLT groups from
the start [29], incorporating the key principles of the HLT event selection in ATLAS that remain unchanged
from Runs 1 and 2:
1. Where possible, trigger selections make use of partial-event data which are processed step-wise
(from low to high granularities) using regional reconstruction inside RoIs corresponding to a part (or
the whole) detector.
2. Event processing is terminated as soon as the event is known to have failed all active trigger selections.
These two key principles require some extensions to AthenaMT, namely the concept of EventViews to
accommodate the regional reconstruction as well as the introduction of the control flow to allow for early
rejection. EventViews allow for an algorithm to be executed multiple times in a single event on either partial
or full event data, making use of already existing framework functionality to execute the reconstruction steps.
Their execution is prepared by Input Maker algorithms that are scheduled to run before the reconstruction
algorithms. Subsequent combinatorial hypothesis algorithms allow for topological selections to be defined
between sets of candidate physics objects. HLT-specific filter algorithms reside between the steps of the
step-wise HLT processing. They provide the possibility of early rejection by gating which steps are allowed
to be executed on any given event.
The scheduling of the execution of algorithms in the HLT is augmented by a control-flow logic determined
when the HLT configuration is prepared, applying an additional layer of steering logic to the event
processing. This layer is defined on nested lists of algorithms and is in addition to the basic execution
steering based on each algorithm’s input and output data. Figure 7 shows an example of a data flow graph.
The control flow in conjunction with the data dependencies ensure that for a given step, its components are
executed in a fixed order: Filter →
− Input Maker → − reconstruction → − hypothesis algorithms. The additional
control logic is required because each processing step can have multiple Filters: for example, the step 1
electron and muon paths in Figure 7 each have two feeding Filters (electron, electron+muon) and (muon,
electron+muon), respectively. These paths may activate in a given event due to one or both of these Filters
passing. The control flow creates a graph representation of all possible execution paths at configuration
time. The graph is built from a list of all physics selections configured to be executed and does not change
during run time. However, each trigger chain corresponds to one path through the graph, and these chains
can be individually enabled or disabled during run time or executed on only a fraction of events.
The HLT software configuration is stored in the trigger-specific Oracle database in JSON [30] format with
blobs of entire JSON job configurations being uploaded to the database. This allows for a considerable
simplification of the database schema compared to Run 2 [20] allowing further improvements to associated
tools and metadata. Several tools allow developers easy inspection, modification or downloading of
JSON files of database configuration data for debugging of problems and reproducing setups. The trigger

14
 Initial processing

Filter step 1 Filter step 1 Filter step 1

Data dependencies
Muon Muon+Electron Electron/Photon define how algorithms are scheduled

Trigger chains
Input maker Input maker correspond to different paths through the fixed control flow diagram
Muon RoIs EM RoIs
Filter algorithms
run at the start of each step and implement the early rejection
Reconstruction Reconstruction
Standalone muons Calorimeter clustering
Input maker algorithms
restrict the following reconstruction to a region of interest
Hypo step 1 Hypo step 1 Hypo step 1
Muon Muon+Electron Electron/Photon Reconstruction algorithms
process detector data to extract features

Filter step 2 Filter step 2 Filter step 2 Filter step 2

Muon Muon+Electron Electron Photon
Hypothesis algorithms
execute hypothesis testing (e.g. pT > 10 GeV) for all active chains

Input maker
Track RoIs

Reconstruction Input maker

Fast ID tracking Photon RoIs

Reconstruction Reconstruction Reconstruction

Combined muons Electrons Photons

Hypo step 2 Hypo step 2 Hypo step 2 Hypo step 2

Muon Muon+Electron Electron Photon

Filter step 3 Filter step 3 Filter step 3 Filter step 3

Muon Muon+Electron Electron Photon

Input maker Input maker

Muon RoIs EM cluster RoIs

Figure 7: Example of control flow for menu processing. The control flow graph is created during initialisation and
the steps are executed based on the available data. If a filter passes, processing continues through the next steps until
all filters fail where the processing of the following steps is stopped. If the last step is reached with a chain passing all
of its steps the event will be accepted.

configuration service in AthenaMT can be either run from the trigger database, from JSON files or from
standard ATLAS python-based configurations.
In 2022 the HLT software was used in a mixed multi-processing / multi-threaded configuration with
two event slots per worker during commissioning and low-luminosity data-taking periods. For the MC
production the configuration was pure MT with typically eight event slots per grid job [31]. The HLT
software performance in 2022 is further detailed in Section 8.

3.5 DAQ system

The DAQ system, as shown in Figure 1, supports the operation of the two trigger levels, beginning with
detector-specific on and off-detector electronics which perform a variety of data processing and monitoring
features before passing events either to the L1 trigger or to the downstream systems. In Runs 1 and
2 the off-detector stage was performed in detector-specific custom hardware modules called Read-Out
Drivers (RODs), typically sitting in VME crates. On receipt of a L1 accept, these RODs read out their
data to the first common stage of the DAQ system, the Read-Out System (ROS). The ROS consists of
a set of commodity server machines hosting custom-built I/O cards. These receive data from detector
front-end electronics and store them in internal memory buffers, called Read-Out Buffers (ROBs). The

15
Data Collection Manager (DCM) orchestrates the data flow from the ROS through the HLT Multi-Process
Processing Units (HLTMPPUs). During processing the HLT requests data from the ROS as needed before
either accepting an event for permanent storage or requesting its deletion from the ROS buffers. In 2022 a
dedicated HLT computing farm of around 56k (112k with hyperthreading) cores was running up to 90k
HLTMPPU selection applications. Every processing node hosts one DCM and one or a few HLTMPPU
selection applications, depending on the chosen multi-process and multi-threading configuration. Individual
events are assigned to processing nodes in the computing farm by a HLT SuperVisor (HLTSV) application
according to available free event nodes on each machine.
New in Run 3 is the introduction of the Front-End Link eXchange (FELIX) read-out system and software,
running on commodity servers (SW ROD). These new DAQ systems are integrated into the read-out
path for those detector systems with new or upgraded front-end electronics (LAr digital trigger, L1Calo,
L1Topo, NSW). The interface to the HLT is identical between ROS and SW ROD, with event data routed
on-demand to HLT processing nodes in both cases. While the ATLAS data flow architecture remains
unchanged for Run 3, the change to a multi-threaded event processing environment required changes in
the interface between the HLTMPPU and the DCM, and in the event assignment from the HLTSV. Once
HLT processing has been completed, accepted events are sent to a dedicated cluster of servers (known for
historical reasons as Sub-Farm Outputs (SFOs)) for packing, compression, and transfer to offline storage.
To accommodate the increased average rate of physics data to be written out, the Run-3 physics bandwidth
to permanent storage has been doubled to allow for up to 8 GB/s.

3.6 Validation strategy

The ATLAS software is being continuously developed to adapt to changing requirements and conditions, fix
defects, improve resource usage and in the particular case of the software used for recording data, to further
optimise the performance of the selection algorithms. As the software is evolving, its validation both in
terms of the signature reconstruction performance (see Sections 5 and 6) and the trigger menu functionality
and execution speed (see Section 4) is of great importance to ensure the reliability and predictability of its
performance.
Validation jobs are run nightly using the ATLAS Release Tester (ART) system [32] to monitor the status
of the trigger software. To facilitate the monitoring and tracking of the ART test results, an automated
system has been developed. This system analyzes any log files produced by the tests, looking for errors
and warnings. It is able to catch any newly introduced changes on a daily basis, including changes to the
effectiveness of the software system with respect to time constraints and allocation of resources. Any
newly introduced failures are flagged and followed up with the corresponding developer. By tracking the
history of the number of events accepted by a given trigger, it is possible to also monitor the evolution of
the trigger algorithm performance and detect unexpected degradation or changes immediately.
In addition to the daily validation, the trigger software is periodically validated using high statistics
simulated samples (e.g. 𝑡 𝑡¯ as well as other more specific physics signal processes) and collision data.
For both types of validation, sets of histograms representing the performance of the HLT and signature
reconstruction are produced and checked against a reference. The main purpose of validating with simulated
samples is to check the reconstruction performance, including new developments and to validate new trigger
chains that are foreseen to be used online in terms of selection efficiency. Before a new software version is
deployed on the HLT farm, it is used to process an enhanced-bias data set [33]. The enhanced-bias data set
consists of roughly one million events, and is collected with filters that ensure an even representation of

16
events with different topologies and rarities. This data reprocessing is, due to its large statistics, executed on
the LHC computing grid using the ATLAS production system [34]. Not only is the performance checked in
terms of reconstruction efficiency and rates of trigger chains, but also the performance in terms of resource
usage (such as run time, memory consumption, etc.) is checked to ensure smooth data taking. This is
important in order to achieve an overall high data quality, and efficient data taking. Further information
about the assessment of the data quality can be found in Ref. [20].

4 Trigger menu

Events are selected to be recorded if they satisfy the conditions of one or more trigger chains. The list of
trigger chains used for data taking is known as a trigger menu, which also includes prescales for each trigger
chain4 . The trigger menu consists of Physics triggers, detailed in Section 6, and Auxiliary triggers, detailed
in Section 7. Triggers which use information from more than one object type are called Combined.
Physics triggers are used for physics analyses and can be subdivided into the following categories:
• Primary triggers, which cover all signatures relevant to the ATLAS physics programme and are
typically unprescaled.
• Support triggers, which are used for efficiency and performance measurements, background estimates,
or for monitoring. These are typically operated at a small rate (of the order of 0.5-1.0 Hz each) using
prescale factors. About 15% of the HLT bandwidth in Run 2 was dedicated to support triggers.
• Alternative triggers, using alternative (sometimes experimental or new) reconstruction algorithms
complementary to the primary or support selections, and often heavily overlapping with the primary
triggers. They are often used, for example, as part of the commissioning process for future primary
triggers.
• Backup triggers, with tighter selections and lower processing or output rate. They can replace the
relevant primary triggers if their CPU usage or output rate becomes too high. These triggers require
almost no additional computing resources or output rate as they select a subset of the primary
triggered events.
Auxiliary triggers can in turn be subdivided into the following categories:
• Calibration triggers, used for detector calibrations.
• Cosmic ray triggers.
• Beam-induced background triggers, which are recorded in LHC bunches with single beam or no
beam present.
• Noise triggers, which are collected by a random trigger at L15 .

4 To control the rate of accepted events and to manage CPU consumption at the HLT, a prescale value, or simply prescale, can be
applied. For a prescale value of 𝑛, a trigger chain has a probability of 1/𝑛 to be activated in the event. By default they are
randomly generated for every individual trigger. However, a mechanism of coherent prescale sets exists for defining groups of
triggers whose prescales are correlated. Individual prescale factors can be given to each chain at L1 or at the HLT, and can be
any value greater than or equal to one. The value -1 is used to disable triggers.
5 The random trigger item at L1 corresponds to the read-out from the detector of events chosen at random. They are always
prescaled. L1 random triggers on filled bunches can be used to seed specific triggers to overcome potential inefficiencies at L1,
while L1 random triggers on the unfilled LHC bunches are typically used for noise and background studies.

17
 • Other dedicated triggers and algorithms.
To facilitate further processing and analysis, accepted events are recorded into different data sets, called
streams, which are designed to have minimal overlap. The trigger menu defines the streams to which an
event is written, depending on the trigger chains that accepted the event. The five different types of data
streams considered in the recording rate budget available at the HLT during nominal 𝑝 𝑝 data taking are:
• Physics streams: collision events of interest for physics studies. The events contain full detector
information and dominate in terms of processing, bandwidth and storage requirements. There
are three physics streams for the 𝑝 𝑝 data taking: the Main stream, the 𝐵-physics and light states
(BLS) stream and the Hadronic stream. Events in the Main stream are promptly reconstructed after
completion of the first-pass calibration and data quality assessment, as described in Ref. [35], while
events in the other two, Delayed, streams are reconstructed when resources allow it.
• Express stream: a very small subset of the physics stream events for prompt monitoring, detector
calibration, and first-pass data quality checks. It is fully reconstructed offline within a day of having
been recorded.
• Background streams: background events of interest for physics and detector performance studies.
• Debug streams: events for which no trigger decision could be made are written to this stream. Typical
reasons are crashes, timeouts in the HLT processing, and HLT data payloads exceeding set thresholds.
These events need to be analysed and recovered separately to identify and fix possible problems in
the TDAQ system.
• Calibration streams: events containing only partial detector information for calibration of specific
subdetectors.
• Trigger-Level Analysis (TLA) streams: events sent to this stream contain only specific physics objects
reconstructed by the HLT, and optionally only partial detector information. These data are used
directly in the corresponding physics analysis (e.g. Ref. [36]). The average TLA event size is 4.5 kB
in 2022, which is approximately 0.3% of the size of a full ATLAS event.
• Monitoring streams: events to be sent to dedicated monitoring nodes for online analysis for, e.g.,
detector monitoring, but not recorded.
For special data-taking configurations, it is possible to introduce additional streams, such as, for example,
the Enhanced bias physics data stream, which is used to record events for trigger rate predictions [20, 33].
With the exception of the debug streams, the streaming model is inclusive, which means that an event can
be written to multiple streams.
The trigger menu composition and trigger thresholds are optimised for several luminosity ranges in order
to maximise the physics output of the experiment and to fit within the rate and bandwidth constraints of the
ATLAS detector, TDAQ system and offline computing. The effect of such optimisation on the HLT stream
output for a typical run is shown in Figure 8. For Run 3, the most relevant constraints are the maximum L1
rate of 100 kHz (unchanged with respect to Run 2) as defined by the ATLAS detector read-out capability, a
target average HLT physics output rate of 3 kHz (1.2 kHz in Run 2) and CPU resources of the HLT farm,
as detailed in Section 8. Substantial increases in the HLT output rate were enabled by the expanded SFO
capacity, providing improved coverage for physics, while also allowing space for sustained operation with
very long periods (up to ten hours) of constant instantaneous luminosity close to its peak value, which
can be seen in Figure 8 before 6:30. Such periods make the average Run-3 HLT rate almost equal to the
peak rate, in contrast to Run 2 for which the average was about 2/3 of the peak. To ensure an optimal

18
 × ×
14
ATLAS TLA ATLAS Main
6
pp data, Nov 2022, = 13.6 TeV Calibration pp data, Nov 2022, = 13.6 TeV Hadronic
12 Main 2 BLS 2
HLT Global Rate
Luminosity Hadronic Other

HLT output bandwidth [GB/s]

5
BLS Express
10 Other Calibration
HLT event rate [kHz]

]
Express 1.5 TLA 1.5
4
Luminosity
8

Luminosity [

Luminosity [
3
6 1 1

2
4
0.5 0.5
2 1

0 0 0 0
01:00 03:00 05:00 07:00 09:00 11:00 13:00 15:00 01:00 03:00 05:00 07:00 09:00 11:00 13:00 15:00

Time [h:m] Time [h:m]

Figure 8: An example of (left) the rate and (right) bandwidth output by the HLT streams in a typical 2022 𝑝 𝑝 run.
The total HLT event rate is lower than the sum of the stream rates, because the same events may be written to multiple
streams. The peak bandwidth of the TLA stream is approximately 25 MB/s, and consequently barely visible in the
right hand figure. The event size per stream can vary up to a factor of 300.

trigger menu within the rate constraints for a given LHC luminosity, prescale factors can be applied to L1
and HLT triggers and these are changed during data taking in such a way that triggers may be disabled
entirely or only executed for a certain fraction of events. Supporting triggers usually run at a constant
rate. The small event size of triggers in the TLA stream allows it to be recorded at rates of the order of
1-10 kHz while taking up a minor fraction of the total HLT bandwidth. This strategy is effective in avoiding
high prescales at the HLT for low 𝑝 T TLA triggers. Some triggers are enabled only later in an LHC fill
when the luminosity and pile-up have reduced and the required HLT farm resources are available. The
effects of prescale changes towards the end of the run on the HLT output rate and bandwidth can be seen
in Figure 8 around 8:30, 11:00 and 14:30. Further flexibility is provided by bunch groups, which allow
triggers to include specific requirements on the proton bunches in the LHC. These requirements include
paired (colliding) bunch-crossings for physics triggers, empty or unpaired crossings for background studies
or searches for long-lived particle decays, and dedicated bunch groups for detector calibration.

4.1 Baseline physics trigger menu for proton–proton collisions

The primary 𝑝 𝑝 menu triggers cover all signatures relevant to the ATLAS physics programme including
electrons, photons, muons, taus, jets and 𝐸 Tmiss which are used for Standard Model (SM) precision
measurements including decays of the Higgs, 𝑊 and 𝑍 bosons, and searches for physics beyond the SM
such as heavy particles, supersymmetry or exotic particles. A set of low-𝑝 T di-muon and di-electron
triggers is used to collect 𝐵-meson decays, which are essential for the 𝐵-physics programme of ATLAS.
The Run-3 trigger menu aims to maximise the physics impact of the Run-3 data set by exploiting the newly
implemented detector features, more performant HLT hardware, and algorithmic advancements, while
simultaneously maintaining a level of consistency with the Run-2 trigger menu to allow for combined
analyses on both data sets. Trigger thresholds at L1 and HLT were generally kept the same as during
Run 2, benefiting from improvements to reduce trigger rate. The trigger menu strategy remains focused on
assigning the majority of the rate to inclusive triggers rather than analysis-specific triggers. In particular,
the Run-3 trigger menu maintains the unprescaled isolated single-electron and single-muon trigger 𝑝 T

19
thresholds around 25 GeV, as described in Sections 6.1.2 and 6.2.1. Dedicated triggers are added for
specific analyses that are not covered by inclusive triggers. The additional available HLT rate compared
to Run 2 is dedicated to expanding the physics menu, in the physics and TLA streams, lowering trigger
thresholds and including new triggers for previously unexplored phase space. The increase in the SFO
capacity enables a larger recording bandwidth which is exploited with the delayed streams. The breakdown
of the approximate rates feeding these streams grouped by trigger signature, as detailed in Section 6, is
shown in Table 1.
In 2022 the baseline physics trigger menu was based on the L1Calo and L1Topo legacy systems. This
led to some limitations; in particular, HLT chains seeded from L1Muon topological triggers could not be
run until the new L1Topo and L1Muon were commissioned. This mostly affected triggers for the ATLAS
𝐵-physics programme.

Table 1: Example breakdown

√ of approximate total rates for physics triggers grouped by signature at luminosity of
1.8 × 1034 cm−2 s−1 and 𝑠 = 13.6 TeV. Rates are quoted for the Main, Delayed and TLA streams subtracting off the
contributions from the less inclusive streams.

Signature Rate per stream [Hz]

Main Delayed TLA
Electron 270
Photon 120
Muon 290
Tau 160
Missing transverse momentum 140
Unconventional Tracking 40
𝐵-physics and light states 240
Jet 490 460 5000
Jet with 𝑏-hadrons 190 160
Combined 240 50 830

3500
Average rate per run [Hz]

B-physics and light states

Hadronic ATLAS
3000 Main pp data 2022, s = 13.6 TeV
Average total rate (1.7 kHz)
Average rate Physics Main (1.2 kHz)
2500

2000

1500

1000

500

0
July August October November
Date of run

Figure 9: The average recording rate of physics data streams for the ATLAS 𝑝 𝑝 physics runs taken in 2022. The
average of all runs for these three streams is indicated as a dash-dotted line, and the average of the Main stream is
indicated as a dashed line.

20
The evolution of the average recording rates of the physics data streams in 2022 is shown in Figure 9. The
average Run-3 rates of the Main, Hadronic and BLS streams are designed to be 1.6 kHz, 1.1 kHz and
0.5 kHz, respectively. However, in 2022 they used only part of this allocation, which in the case of BLS
was dictated by the commissioning schedule.
The triggers with the largest rate contributing to the delayed streams are summarised in Table 2. The
𝐵-physics triggers are strongly limited by the rate of L1 triggers. In order to maximise the acceptance, L1
triggers with lower thresholds are enabled at the end of the fills, once the luminosity is significantly below
its peak value. The rate of the BLS stream is kept approximately constant and thus the fraction of the BLS
events in the recorded physics events increases significantly at lower luminosities.

Table 2: Summary of selected triggers in the delayed streams. The VBF di-jet trigger selects events characteristic of
Vector Boson Fusion, requiring the presence of a pair of jets satisfying kinematic correlations and an invariant mass
√ jets are within |𝜂| < 2.5 unless otherwise specified. Rates are given at luminosity of
greater than 1000 GeV. All
1.8 × 1034 cm−2 s−1 and 𝑠 = 13.6 TeV.

Trigger 𝑝 T threshold [GeV] Rate [Hz]

VBF di-jet 70 (|𝜂| < 3.2), 50 (|𝜂| < 4.9) 270
Two jets, two 𝑏-jets (𝜖 = 77%) 80, 55, 28, 20 160
Six jets 6 × 35 140
Five jets, one 𝑏-jet 5 × 35, 25 50
𝐵-physics di-muon 11, 6 40
𝐵 → 𝐾 ∗ 𝑒𝑒 5, 5 170

The Run-3 TLA stream was expanded to include photon, muon and b-jet triggers, in addition to the jet
triggers, that were used in Run 2 [36]. It is also now possible to record multiple trigger object collections at
the same time, permitting the analysis of more complex final states. The content of each individual event in
the TLA stream is determined by specialised physics chains targeting these objects and their combinations,
as detailed in Sections 6.1.4, 6.4.4 and 6.5.3.
The improvements to existing triggers and their Run-3 performance are detailed in Sections 5 and 6. In
particular, the tracking improvements [37] allowed for the introduction of triggers based on unconventional
tracking which are listed in Table 3 and detailed in Section 6.9. The shown rates are not unique and can
have overlap with other physics triggers, as is the case for the displaced tau trigger, for example.

4.2 Baseline physics trigger menu for heavy–ion collisions

The characteristics of HI collisions of Pb+Pb are largely different from 𝑝 𝑝 collisions. Apart from any
hard scattering of interest, each HI collision is composed of multiple simultaneous nucleon–nucleon
interactions, which generate a sizable underlying event (UE) contribution dominated by soft particle
production. As a result of this large UE contribution, the particle multiplicities and the total energy
deposited into calorimeters in HI collisions are on average much larger than those in 𝑝 𝑝 collisions, and they
also vary significantly from event to event. The event-by-event variations correspond to the variations in
the size and geometry of the overlap region of the two colliding Pb nuclei. The variations are characterised
by the centrality of Pb+Pb collisions, which is strongly correlated to total transverse energy measured in
the forward calorimeter, Σ𝐸 TFCal [38]. Centrality classes are defined by dividing the Σ𝐸 TFCal distribution of

21
Table
√ 3: Summary of selected unconventional tracking trigger rates at luminosity of 1.8 × 1034 cm−2 s−1 and
𝑠 = 13.6 TeV.

Trigger Rate [Hz]

Isolated high 𝑝 T track 1
Large dE/dx 5
Disappearing track 4
Hit-based displaced vertex 𝐸 Tmiss -seeded 1
Hit-based displaced vertex jet-seeded 1
Emerging jet (jet-seeded) 10
Emerging jet (photon-seeded) 10
Displaced single-jet 15
Displaced di-jet 7
Displaced electron 13
Displaced muon 4

minimum-bias inelastic Pb+Pb collisions into percentiles. The 0–10% interval includes events with the
largest Σ𝐸 TFCal , corresponding to the most central collisions (largest geometric overlap), while the 90–100%
interval includes the most peripheral collisions (smallest geometric overlap). In the trigger, centrality can
hence be mapped to the total energy measured in the full calorimeter system or in its forward part, both of
which are accessible at L1.
Besides the variations of particle multiplicity from event to event, there is also an azimuthal anisotropy of
particle production present in each HI event. This is a result of the initial-state spatial anisotropy of the
overlap region leading to sizable anisotropies of particle momenta in the final state.
The trigger menu for HI collisions must be designed to handle the large event-by-event variations and
azimuthal anisotropy in each event. On the other hand, HI collisions are operated at a lower centre-of-mass
energy per nucleon pair and a lower luminosity compared to 𝑝 𝑝 collisions, so the hard scattering rate
is lower in HI collisions. The main goal of the Run-3 HI trigger menu design is to keep 𝑝 T thresholds
for unprescaled triggers of different signatures as low as possible while minimizing the sensitivity of all
triggers to UE contributions. The unprescaled single electron and photon 𝑝 T thresholds are set at 15 GeV,
while the unprescaled single muon trigger has a 𝑝 T threshold around 8 GeV. This strategy ensures the
collection of the majority of events with leptonic 𝑊 and 𝑍 boson decays, which are the main source of
events for the study of electroweak processes. The lower 𝑝 T threshold for the single muon trigger allows
for the selection of events with semi-leptonic decays of heavy-flavour hadrons. The unprescaled single jet
𝑝 T threshold is set to 85 GeV to collect events with jets for the study of QCD processes. In addition to
single physics signature triggers, dedicated multi-object triggers are added: an unprescaled di-muon trigger
with 𝑝 T thresholds of 4 GeV for both muons to collect events with di-muon decays of quarkonium states;
an unprescaled muon and jet trigger with a muon 𝑝 T threshold of 4 GeV and a jet 𝑝 T threshold of 60 GeV
to collect events with 𝑏-jets based on soft muons from 𝐵-hadron decays.
To reduce their sensitivity to UE contributions, calorimeter-based triggers – such as electron, photon and
jet triggers – include a correction for the average energy contributed by the UE. The average UE energy is
evaluated per calorimeter layer and cell following the iterative procedure used in offline reconstruction [39,
40]. For each interval of width Δ𝜂 = 0.1, the UE energy is averaged over 𝜙 taking into account the

22
azimuthal anisotropy of deposited energy due to the collective motion of particles created in HI collisions.
This estimated average UE energy is then subtracted from the cluster constituent cells. As a result, the
cells in the RoI – which are used as inputs to clustering and identification algorithms – contain a reduced
systematic bias due to the UE.
In cases where two nuclei do not interact hadronically due to a lack of geometric overlap, they can still
interact electromagnetically, producing so-called ultraperipheral collisions (UPC). The large electric
charge of Pb nuclei (𝑍 = 82) results in a photon flux that is enhanced by a factor of 𝑍 2 compared to 𝑝 𝑝
collisions. Therefore, cross-sections for electromagnetic (photon–photon and photon–nucleus) interactions
are significantly increased. The events produced in these processes are typically accompanied by small
multiplicities of produced particles, as well as forward rapidity gaps defined by very low activity in the
forward calorimeters.
The three physics streams for Pb+Pb data taking differ from those in 𝑝 𝑝 collisions and aim to target different
classes of events:
• physics_HardProbes: events produced by hard processes in inelastic Pb+Pb collisions and triggered
with high-𝑝 T electrons, photons, muons, jets and 𝑏-jets.
• physics_UPC: events produced by various processes in ultraperipheral Pb+Pb collisions. Triggers in
this stream select events with low-𝑝 T electrons, photons, muons and jets as well as specific track
multiplicities, in combination with vetos on the total calorimeter energy, and/or deposits in the ZDC.
• physics_PC/physics_CC: minimally trigger-biased (minimum bias) inelastic Pb+Pb collisions
separated into peripheral (PC) and central (CC) events. Triggers for these streams are seeded off L1
total energy triggers.
The calibration and performance streams largely remain the same.
The HI triggers with muons rely on standard reconstruction described in Sections 6.2 and 6.7, and are not
discussed further. More details on other signature-specific HI configurations can be found in Sections 6.1.5,
6.4.5, 6.5.2 and 6.8.

5 High-level trigger reconstruction

Once an event is accepted at L1, it is processed by the HLT by making use of finer-granularity calorimeter
information, precision measurements from the MS and tracking information from the ID. The following
sections describe the main HLT algorithms used for ID tracking, calorimeter clustering and muon
reconstruction.

5.1 Inner detector tracking

The sequence of track reconstruction steps is briefly summarised below. More detailed information can be
found in Ref. [41]. The track reconstruction starts with the data preparation in the ID, which reconstructs
clusters and space points using information from the pixel and SCT data providers which fetch the raw
detector data from the pixel and SCT read-out systems. This step makes use of the RoI mechanism which
allows the HLT to request only the data from those silicon modules inside the RoI. The data preparation
is then followed by the fast-tracking step, which runs a custom pattern recognition and a fast-track fit.

23
Following this, a precision tracking step takes the tracks reconstructed by the fast tracking and refits
them using the offline track reconstruction code. It does not use the slower offline pattern recognition
code. Additionally, these track candidates are extended into the TRT: using hits at larger radii leads to
an improved track momentum resolution. The TRT data preparation is performed only for the precision
tracking following the extension of the tracks into the TRT.
The Run-3 tracking software is enhanced with respect to Run 2 with several new developments:
• Restriction of the RoI width along the beamline for the track reconstruction for the muon trigger
isolation requirement (Section 5.1.2);
• Implementation of the bremsstrahlung recovery for the electron triggers (Section 5.1.3);
• Track reconstruction in the entire ID volume (full scan) for the jet- and 𝐸 Tmiss -based triggers, followed
by the vertex finding with these tracks (Section 5.1.4);
• Running of a preselection tracking stage in the 𝑏-jet trigger to allow the use of fast 𝑏-tagging
algorithms prior to execution of the full scan tracking (Section 5.1.5);
• Implementation of large radius tracking (LRT) [42] (Section 5.1.6).
The HLT minimum-bias triggers (Section 6.8) run the full offline pattern recognition and track reconstruc-
tion [43].
Except where stated, the standard selection6 of offline tracks [43] and objects [44–48] is used for performance
studies in this subsection. Efficiencies are measured by taking the correlated ratio of the number of offline
reference objects that have a matched trigger track to the total number of offline objects passing the selection.
The methodology for tracking performance studies is described in detail in Ref. [41].

5.1.1 Tau lepton tracking

The tau lepton trigger tracking runs as a two-stage process, essentially unchanged since Run 2 and described
in more detail in Section 6.3 and in Ref. [41]. First, the fast tracking runs in a core RoI of 0.2 × 0.2 in
𝜂 − 𝜙 space, fully extended along the beamline in the range |𝑧| < 225 mm, to identify the leading 𝑝 T tracks
from the tau decay. This is followed by running both the fast tracking and the precision tracking in a wider
RoI (0.8 × 0.8 in 𝜂 − 𝜙 space) to determine whether the tau candidates are isolated, and to accommodate
the wider opening angle for tau candidates with three tracks associated with them (3-prong). This RoI is
centred on the 𝑧 position of the leading 𝑝 T track identified by the first stage and limited to |𝑧| < 10 mm
relative to this leading track. The efficiency of the tau lepton tracking is above 99%.

5.1.2 Muon tracking

The muon triggers use the standard tracking sequence, which is essentially unchanged since Run 2 [41]
and has an efficiency better than 99% for tracks originating from the beamline with an almost negligible
dependence of the efficiency on the pile-up multiplicity. Those triggers in which a muon track is required
to be isolated with respect to other tracks from the interaction run the muon isolation (muonIso) tracking
stage, after the standard muon fast and precision tracking, using a wider RoI (0.7 × 0.7) in 𝜂 − 𝜙 space. For
6 With the addition that for the standard tight offline requirement on the number of holes (e.g. missing hits from active layers) [43],
the tracks are required to contain at least one pixel hit.

24
 1.02 1.02
Efficiency

Efficiency
1 1

0.98 0.98

0.96 0.96
ATLAS ATLAS
0.94 Data 2022, s = 13.6 TeV 0.94 Data 2022, s = 13.6 TeV
24 GeV muon trigger 24 GeV muon trigger
Offline tracks p > 1.2 GeV Offline tracks p > 1.2 GeV
0.92 T
Muon isolation precision tracking
0.92 T
Muon isolation precision tracking
Muon isolation fast tracking Muon isolation fast tracking
0.9 0.9
1 2 3 4 5 6 7 10 20 30 102 10 20 30 40 50 60
Offline track p [GeV] Pile-up <m>
T

Figure 10: The tracking efficiency for non-muon tracks reconstructed in the muonIso RoI centred on the trigger muon
selected by the 24 GeV muon performance trigger described in Section 7.3, but without selection on the trigger tracks
from the muonIso reconstruction. Efficiencies are shown for both the fast and precision tracking algorithms as a
function of (left) the offline muon transverse momentum and (right) the average pile-up. Only statistical uncertainties
are shown.

Run 3, this wider RoI has a restricted full 𝑧-width of 20 mm centred on the 𝑧 position at the beamline of the
muon candidate identified in the first round of precision tracking and after the muon reconstruction. The
restriction of the 𝑧-width reduces the execution time of the muonIso tracking, and is discussed in more
detail in Section 5.1.7.
The tracking efficiency in the widened RoI used for the muon isolation is shown in Figure 10. It is evaluated
with respect to offline tracks within the wider RoI once the muon track candidate is removed. The efficiency
is greater than 99% across the full 𝑝 T range shown. There is a small dependence (less than 0.5%) of the
efficiency on pile-up.

5.1.3 Electron tracking

The electron triggers run both the fast, and precision tracking in the same RoI, produced following the fast
calorimeter reconstruction, as detailed in Section 6.1. The dimensions of the RoI used for the tracking are
0.1 × 0.2 in 𝜂 − 𝜙 space, with the 𝜂 width being reduced to half of that used in Run 2, with no reduction of
the physics performance.
The track reconstruction for electrons can be affected by bremsstrahlung which causes tracks, in particular
at lower 𝑝 T , to deviate from the expected helical path through the tracking detectors. To improve the
resolution of the electron track reconstruction in Run 3 the offline Gaussian sum filter (GSF) algorithm [49]
is added to the electron track reconstruction sequence in the HLT.
Figure 11 shows the efficiency of the electron track reconstruction as a function of the transverse energy
and the pile-up for offline electron candidates with 𝐸 T greater than 14 GeV. To minimise the contribution
from non-electron tracks, the tag-and-probe method is used for this study [41]. The slow increase of the
efficiency at low 𝐸 T is due to the removal of some seed candidates that share hits with other seeds in the
fast-tracking stage, reducing the efficiency for bremsstrahlung candidates where the seeds may not lie
so closely on a single helix. By construction, the GSF efficiency can not be higher than the precision
tracking efficiency since only tracks from the precision tracking are used for the refit for the GSF tracks.

25
 1.02 1.02
Efficiency

Efficiency
1 1

0.98 0.98

0.96 0.96
ATLAS ATLAS
0.94 Data 2022, s = 13.6 TeV 0.94 Data 2022, s = 13.6 TeV
14 GeV electron trigger, Offline electrons E > 14 GeV 14 GeV electron trigger, Offline electrons E > 14 GeV
T T
GSF tracking GSF tracking
0.92 Precision tracking
0.92 Precision tracking
Fast tracking Fast tracking
0.9 0.9
20 30 40 102 2× 102 103 10 20 30 40 50 60 70
Offline electron ET [GeV] Pile-up <m>

Figure 11: The ID tracking efficiency for electron candidates selected by the 14 GeV electron performance trigger
described in Section 7.3 which does not use ID tracking information. The efficiency with respect to offline electrons
is shown as a function of (left) the offline electron transverse energy and (right) the average pile-up. Only statistical
uncertainties are shown.

The small apparent excess in the GSF tracking efficiency in Figure 11 (left) arises since the resolutions
of the GSF tracks are better, and as such fewer tracks are excluded by failing the matching criteria. In
addition, different performance triggers are used for the GSF and precision tracking studies, such that the
event samples used for the calculation of efficiency are not exactly identical.
The effect of bremsstrahlung on the reconstruction of electron candidates is shown in Figure 12: as the
electron candidates are required to have 𝐸 T > 14 GeV, the track candidates with 𝑝 T below this value would
have undergone bremsstrahlung. Because of this both the precision and fast tracking overestimate the 1/𝑝 T
with a very long tail to positive values shown in Figure 12 (top left). This bias becomes progressively
smaller at higher 𝑝 T as shown in Figure 12 (top right). For the GSF tracking however, the distribution is
more symmetric, and the bias is close to zero over the entire 𝑝 T range.
The resolution with respect to offline of the trigger electron 1/𝑝 T is also seen in Figure 12 (bottom) and
clearly shows the resolution improvement from the GSF tracking with respect to the precision tracking. For
reliable estimates of higher offline track transverse momentum, only candidates with 𝐸 T /𝑝 T > 0.8 are used
for the determination of the resolutions. The GSF tracking improves the 1/𝑝 T resolution by nearly a factor
of two. Similar improvement is observed for the resolutions of the azimuthal angle and transverse impact
parameter, but not in the track pseudorapidity and 𝑧 at the beamline as the latter variables are less sensitive
to the bending of the track in the magnetic field.

5.1.4 Full scan tracking and vertex finding

To improve the trigger reconstruction of hadronic signatures (jets, 𝑏-jets, 𝐸 Tmiss , etc.) with respect to the
offline reconstruction, tracks from the entire ID volume are combined with the calorimeter topological
clusters, described in Section 5.2, to form particle flow objects (PFO) [50]. This approach improves the jet
energy resolution at lower transverse momenta and better separates the hard interaction from pile-up.
The full scan tracking is executed only for the hadronic signatures. These algorithms are activated at a rate
close to 14 kHz at 2 × 1034 cm−2 s−1 after the preselections discussed in Sections 6.4-6.6. Even with the
reduced input rate after the preselections and with all improvements described below, the full scan fast

26
 mean ∆ 1/p [GeV-1]
Normalised entries

104 ATLAS 0.1 ATLAS

Data 2022, s = 13.6 TeV Data 2022, s = 13.6 TeV

T
103 14 GeV electron trigger, Offline electrons E > 14 GeV
T
0.08 14 GeV electron trigger, Offline electrons E > 14 GeV
T
GSF tracking GSF tracking
Precision tracking Precision tracking
102 0.06
Fast tracking Fast tracking

10 0.04

1 0.02

10−1 0

−2
10− −0.02
0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 3 4 5 6 7 10 20 30 102 2×102
-1
∆ 1/p [GeV ] Offline electron track p [GeV]
T
T

10
1/p resolution [GeV-1]

1/p resolution [GeV-1]

ATLAS 10 ATLAS
Data 2022, s = 13.6 TeV Data 2022, s = 13.6 TeV
1 14 GeV electron trigger, Offline electrons E > 14 GeV
T
14 GeV electron trigger, Offline electrons E > 14 GeV
T
GSF tracking 1 GSF tracking
Precision tracking Precision tracking
Fast tracking Fast tracking
10−1
10−1
T

10−2 10−2

10−3 10−3

−3 −2 −1 0 1 2 3 3 4 5 6 7 10 20 30 102 2×102
Offline electron track η Offline electron track p [GeV]
T

Figure 12: (top left) The ID trigger residual in 1/𝑝 T with respect to offline and (top right) the mean of the residual as
a function of the offline track 𝑝 T . The resolution in the inverse transverse momentum as a function of offline track
(bottom left) 𝜂 and (bottom right) 𝑝 T . Distributions are shown for the fast, precision and GSF tracking algorithms for
electron candidates selected with electron performance trigger with 𝐸 T > 14 GeV and passing the offline electron
𝐸 T /𝑝 T > 0.8 requirement. Only statistical uncertainties are shown.

tracking accounts for 26% of the total event processing time of the HLT, as described later in Section 8. An
optimisation of the processing time at the cost of a reduced tracking efficiency was thus crucial for the
implementation of the full scan tracking for Run 3.
One of the most time-consuming aspects of the full scan tracking is the seed making. In this case the
standard seed making [41] is reconfigured to use space-point triplets consisting of pixel-only or SCT-only
hits. Additionally, machine learning techniques [51, 52] for the seed selection are adopted to further
reduce the processing time. The assignment of SCT hits to the tracks is only performed during the track
extrapolation into the SCT with a reduced window to search for these hits in the subsequent layers. The
full scan tracking for the hadronic triggers is executed only once per event for the full ID volume, with the
resulting tracks and vertices used by each trigger that requires them.
The full scan tracking efficiency for events recorded with a 45 GeV jet trigger is shown in Figure 13. The
efficiency is approximately 94% at 1 GeV, the threshold used in the pattern recognition, and reaches a
plateau at approximately 𝑝 T > 5 GeV. A slight asymmetry of the efficiency between negative and positive
𝜂 is due to the mean beam-spot position not being at 𝑧 = 0. Since 𝜂 for the tracks is always defined
with respect to the 𝑧 0 position of the point of closest approach of the track to the beamline and not with

27
 1.02 1.02
Efficiency

Efficiency
1 1

0.98 0.98

0.96
0.96
0.94
0.94
ATLAS ATLAS
Data 2022, s = 13.6 TeV
0.92 Data 2022, s = 13.6 TeV
0.92 Offline tracks p > 1 GeV Offline tracks p > 1 GeV
T T
45 GeV jet trigger full scan tracking 0.9 45 GeV jet trigger full scan tracking
0.9
1 2 3 4 5 6 7 10 20 30 40 102 −3 −2 −1 0 1 2 3
Offline track p [GeV] Offline track η
T

1.02
Efficiency

0.98

0.96

0.94
ATLAS
Data 2022, s = 13.6 TeV
0.92 Offline tracks p > 1 GeV
T
45 GeV jet trigger full scan tracking
0.9
10 20 30 40 50 60
Pile-up <m>

Figure 13: The full scan track finding efficiency with respect to offline tracks versus the offline track (top left) 𝑝 T ,
(top right) 𝜂, and (bottom) the average pile-up. The efficiency is evaluated for the 45 GeV jet trigger. Only statistical
uncertainties are shown.

respect to 𝑧 = 0, two tracks with identical 𝜂 but different 𝑧 0 pass through different parts of the detector.
The overall efficiency is approximately 98.5%, but with a large pseudorapidity dependence: it is only
90% for 2.4 < |𝜂| < 2.5. There is also a dependence on the pile-up, falling from 98.5% at ⟨𝜇⟩ = 20 to
approximately 97.5% for ⟨𝜇⟩ = 54.
The offline vertex algorithm is also used for the vertex reconstruction in the trigger. It uses tracks from the
full scan tracking to identify the likely primary interaction as well as any additional interactions. The vertex
with the largest sum of the squared transverse momenta for the tracks assigned to it is chosen as the primary
interaction vertex, for both offline and in the trigger. Only the primary offline vertex is considered as a
reference for the trigger vertex efficiency study. The efficiency for reconstructing the primary interaction
vertex in the trigger is shown in Figure 14. For offline vertices with more than six constituent tracks, the
trigger vertex efficiency is better than 99.5%. Similarly, the vertex finding efficiency is close to 100% for
all 𝑧 values and all pile-up multiplicities.
The resolution of the reconstructed 𝑧 position of the vertex is shown in Figure 15, illustrating the dependence
on the number of tracks from which the vertex is constructed. The resolution is approximately 1 mm for
very low track multiplicities and can have values as low as 40 𝜇m at higher multiplicities. The resolution is
between 40 and 50 𝜇m over the full 𝑧 range.

28
 1.02
Vertex finding efficiency

Vertex finding efficiency

1.05

1 1

0.95 0.98
0.9
0.96
0.85
0.94
0.8 ATLAS ATLAS
Data 2022, s = 13.6 TeV Data 2022, s = 13.6 TeV
0.75 Offline primary vertex 0.92 Offline primary vertex
Vertex finder Vertex finder

0.7 0.9
0 10 20 30 40 50 60 70 80 −150 −100 −50 0 50 100 150
Number of tracks Offline vertex z [mm]
1.02
Vertex finding efficiency

0.98

0.96

0.94
ATLAS
Data 2022, s = 13.6 TeV
0.92 Offline primary vertex
Vertex finder

0.9
20 30 40 50 60
Pile-up <m>

Figure 14: The trigger efficiency for finding the primary offline vertex, with respect to (top left) track multiplicity,
(top right) 𝑧 position, and (bottom) the average pile-up. The efficiency is evaluated for the 45 GeV jet trigger. Only
statistical uncertainties are shown.

20
Primary vertex z resolution [mm]

Primary vertex z resolution [mm]

10 ATLAS 1 ATLAS
Data 2022, s = 13.6 TeV Data 2022, s = 13.6 TeV
Offline primary vertex Offline primary vertex
2 Vertex finder Vertex finder
1 2×10−1
10−1
2×10−1
10−1
2×10−2
2×10−2 10−2
10−2

3×10−3 0 20 40 60 80 100 3×10−3 −150 −100 −50 0 50 100 150

Offline track multiplicity Offline vertex z [mm]

Figure 15: The resolution of the primary trigger vertex with respect to the primary offline vertex (left) track multiplicity
and (right) 𝑧 position. The efficiency is evaluated for the 45 GeV jet trigger. Only statistical uncertainties are shown.

29
 1.02 1.02
Efficiency

Efficiency
1 1

0.98 0.98

0.96 0.96
ATLAS ATLAS
0.94 Data 2022, s = 13.6 TeV 0.94 Data 2022, s = 13.6 TeV
Offline tracks p > 1 GeV Offline tracks p > 1 GeV
T T
3 jet 75 GeV trigger : b-jet preselection tracking 3 jet 75 GeV trigger : b-jet preselection tracking
0.92 30 GeV jet trigger : precision tracking
0.92 30 GeV jet trigger : precision tracking
30 GeV jet trigger : fast tracking 30 GeV jet trigger : fast tracking
0.9 0.9
1 2 3 4 5 6 7 10 20 30 40 102 10 20 30 40 50 60
Offline track p [GeV] Pile-up <m>
T

Figure 16: The performance of the tracking for the 𝑏-jet signature versus (left) the offline track 𝑝 T and (right) the
average pile-up. Efficiencies with respect to offline tracks are shown for 𝑏-jet preselection stage as well as for the
𝑏-jet RoI fast and precision tracking stages. Only statistical uncertainties are shown.

5.1.5 Tracking in the jets containing 𝒃-hadrons

To reduce the rate of full scan tracking for 𝑏-jet triggers in Run 3, a calorimeter-only jet-finding preselection
step is implemented, followed by a 𝑏-tagging preselection stage running the fast tracking. This instance of
the fast tracking uses the same optimisations as the full scan tracking but executes in a single composite
RoI (super-RoI), which is constructed from all regions centred around |Δ𝜂| < 0.4, |Δ𝜙| < 0.4 with respect
to the jet axes for all jets with 𝐸 T > 20 GeV identified in the jet preselection stage. This RoI is extended
along the beamline by |𝑧| < 150 mm. A fast 𝑏-tagging algorithm is then applied to the reconstructed tracks,
as detailed in Section 6.5 and Ref. [53]. Shown in Figure 16, for the 𝑏-tagging preselection stage the
efficiency is only 95% at the 1 GeV threshold and does not reach the plateau until approximately 5 GeV.
The preselection tracking efficiency also shows a slight reduction with increasing pile-up.
Following this preselection, after the rejection of events where no 𝑏-tag is found, the full scan tracking is
executed in the full ID volume to determine the primary vertex position and reconstruct the tracks used
for the particle flow (PFlow) jet reconstruction. The need to reduce the processing time for the full scan
tracking, described in the previous section, necessarily compromises the efficiency to some degree. While
this may be acceptable for the PFlow reconstruction, for which it is predominantly intended, this would
not be desirable for the final 𝑏-tagging where the best possible efficiency for individual tracks is required.
Consequently, as in Run 2, the standard RoI-based fast and precision track reconstruction is executed in a
separate RoI for each PFlow jet, centred on the jet direction and with the RoI 𝑧-position determined from
the primary vertex information. These precision tracks are then used for the full 𝑏-tagging, described in
Section 6.5. For Run 3, the track 𝑝 T requirement used in the pattern recognition for the 𝑏-jet tracking
stage is reduced to 0.8 GeV from the 1 GeV used in Run 2. As shown in Figure 16, the RoI-based fast and
precision tracking results in higher efficiencies for the tracks from 𝑏-jets. The efficiency for the precision
𝑏-jet tracking is better than 99% over the full phase space, an improvement on the 84% in the 1.0 − 1.2 GeV
range seen in Run 2 [41].

30
5.1.6 Large radius tracking

LRT is a new feature introduced in the HLT for Run 3. It uses the same algorithms as the previously
described standard tracking, but with modified configurations in order to reconstruct tracks at large radii or
impact parameter (𝑑0 ). These configurations are based on the offline LRT reconstruction [42], which was
improved for Run 3 to reduce the number of fake tracks and processing time. As with the standard tracking,
LRT is split into fast and precision track reconstruction steps, which can be performed either inside RoIs or
in the entire ID volume (full scan LRT). In the case of RoI-based LRT, e.g. for leptons, the LRT is run by
itself. For the full scan LRT, the tracking uses the remaining hits after a standard tracking pass, as is done
for the offline LRT.
The standard seeding step [41] is modified as follows: only hits from the SCT are used and the ordering of
seeds by impact parameter is removed. Tracking is expanded to cover |𝑧 0 | < 500 mm and |𝑑0 | < 300 mm,
but with generally stricter requirements on track quality to reduce the number of fake tracks, such as
requiring at least eight hits on the track. This limits tracks to originate before the first SCT layer in the
barrel region, at a radius of approximately 300 mm. The same momentum threshold of 𝑝 T > 1 GeV as
standard tracking is used. For LRT in electron and muon RoIs, the size of the RoI is expanded in both 𝜂
and 𝜙, compared to the tracking for prompt leptons, in order to accommodate tracks that do not point to the
beamline. The size of the RoI in 𝜙 is a limiting factor in the efficiency to reconstruct tracks at large 𝑑0 ,
and a trade-off has to be made between tracking acceptance and computing cost. In order to reduce the
processing time for full scan LRT, tracking is restricted to |𝑑0 | > 2 mm, tracks are not extended into the
TRT, and track candidates are required to have 𝑝 T > 1 GeV (on top of the 𝑝 T requirement on the seeds).
Figure 17 shows the performance of LRT in electron and muon trigger RoIs with respect to offline electrons
and muons, respectively. No additional identification requirements are applied beyond the reconstruction.
Data were collected during 2022 using a trigger that does not apply any selection based on the tracking,
described in Section 7.3. The offline lepton tracks are required to have at least eight silicon hits, |𝑑0 | > 2 mm,
and 𝑝 T > 10 GeV. Leptons using both the standard offline track reconstruction and large radius offline
track reconstruction are used as the reference. There is a small overlap where standard and LRT tracks
matched to the same calorimeter cluster or muon spectrometer segment that is not removed here. Since
there are very few SM processes that produce displaced tracks, the tracks in these plots originate from
a combination of photon conversions, heavy-flavour hadron decays, long-lived neutral kaon decays, and
combinations of hits resulting in fake tracks, which are not necessarily representative of high momentum
leptons.
The efficiencies in data are similar to that of tracks in 𝑡 𝑡¯ MC simulation, while the efficiency for leptons in
long-lived particle (LLP) signal MC simulation is much higher. Figures 18 and 19 overlay the efficiency of
electron GSF- and muon precision-tracking with respect to offline electrons or muons for these samples.
Tracking in the trigger for signal-like tracks is much more efficient out to large impact parameter values
and less dependent on the amount of pile-up.
Data events collected in one run of 2022 were reprocessed to run the full scan trigger tracking configuration
of LRT for every event. A collection of offline 𝐾𝑆0 candidate vertices is produced by selecting both standard
and LRT offline tracks with 𝑝 T > 1 GeV and opposite-charge that form a vertex with a mass within 25 MeV
of 497 MeV [54]. The offline tracks associated to the 𝐾𝑆0 candidates are then matched to online standard
full scan tracks if they fall within Δ𝑅 < 0.05 and Δ𝑑0 < 2.5 mm of each other. Remaining offline tracks
are used in the denominator to compute the online LRT efficiency. Figure 20 shows the efficiency with
respect to these remaining offline tracks versus the 𝑝 T and 𝑑0 of the offline track. The tracks from 𝐾𝑆0

31
 1.2
Efficiency

Efficiency
1.2 ATLAS ATLAS
Data 2022, s=13.6 TeV 1 Data 2022, s=13.6 TeV
1.1
30 GeV LRT Electron Trigger 30 GeV LRT Electron Trigger
1 Offline Electron p >10 GeV, d0>2 mm Offline Electron p >10 GeV, d0>2 mm
T T
Fast Tracking 0.8 Fast Tracking
0.9 Precision Tracking Precision Tracking
GSF Tracking GSF Tracking
0.8 0.6

0.7 0.4
0.6
0.2
0.5

0.4 0
7 10 20 30 102 2×102 103 −200 −150 −100 −50 0 50 100 150 200
Offline electron track p [GeV] Offline electron track d0 [mm]
T

0.8 1.2
Efficiency

Efficiency
ATLAS ATLAS
0.7 Data 2022, s=13.6 TeV 1 Data 2022, s=13.6 TeV
20 GeV LRT Muons Trigger 20 GeV LRT Muons Trigger
0.6 Offline Muon p >10 GeV, d0>2 mm
T
Offline Muon p >10 GeV, d0>2 mm
T
Fast Tracking 0.8 Fast Tracking
0.5 Precision Tracking Precision Tracking
0.6
0.4
0.4
0.3

0.2 0.2

0.1 0
7 10 20 30 102 2×102 103 −200 −150 −100 −50 0 50 100 150 200
Offline muon track p [GeV] Offline muon track d0 [mm]
T

Figure 17: Fast and precision LRT efficiencies for (top) electron and (bottom) muon triggers as a function of offline
electron and muon (left) track 𝑝 T and (right) 𝑑0 . The corresponding merged collections of standard and large radius
offline electron and muon tracks are used. The efficiency for the additional GSF step for electrons is also included.
Only statistical uncertainties are shown.

decays tend to have low 𝑝 T and the number of offline tracks falls off rapidly after a 𝑝 T of 5 GeV and a |𝑑0 |
of 80 mm. Due to the steeply falling 𝑝 T spectrum of the 𝐾𝑆0 decay products, low momentum particles
with poor 𝑝 T resolution make up a significant proportion of the high 𝑝 T offline reference sample. Many
of these are not reconstructed by the trigger leading to an apparent decrease in trigger efficiency at large
𝑝 T which is not representative of the actual efficiency for higher 𝑝 T particles. The bottom of Figure 20
shows the efficiency of matching both tracks of the offline vertices to online tracks, including standard and
LRT, versus the reconstructed radius of the 𝐾𝑆0 vertex. The efficiency drops at radii corresponding to barrel
layers of the Pixel detector, e.g. 88.5 mm and 122.5 mm.

5.1.7 Execution time

The ID track reconstruction in the HLT comprises 59% of the total event processing time in the trigger as
detailed further in Table 5 of Section 8.2. The fast tracking is typically the most time-consuming algorithm
due to the combinatorial nature of the pattern recognition stage which has a non-linear dependence on
pile-up. The precision tracking timing has a much smaller dependence on pile-up since it depends only on
the number of tracks passed to it from the fast tracking. The precision tracking also includes hits from the

32
 1.8 1.8
Efficiency

Efficiency
1.6 ATLAS 1.6 ATLAS
Data 2022, s = 13.6 TeV Data 2022, s = 13.6 TeV
1.4 LRT electron trigger, Offline electron p >10 GeV, d0>2 mm
T
1.4 LRT electron trigger, Offline electron p >10 GeV, d0>2 mm
T
MC di-selectron, mass = 100 GeV, lifetime = 1 ns MC di-selectron, mass = 100 GeV, lifetime = 1 ns
1.2 MC ttbar 1 lepton 1.2 MC ttbar 1 lepton
Data Data
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
10 102 −150 −100 −50 0 50 100 150 200
Offline electron track p [GeV] Offline electron track d0 [mm]
T

1.8 1.8
Efficiency

Efficiency
1.6 ATLAS 1.6 ATLAS
Data 2022, s = 13.6 TeV Data 2022, s = 13.6 TeV
1.4 LRT electron trigger, Offline electron p >10 GeV, d0>2 mm
T
1.4 LRT electron trigger, Offline electron p >10 GeV, d0>2 mm
T
MC di-selectron, mass = 100 GeV, lifetime = 1 ns MC di-selectron, mass = 100 GeV, lifetime = 1 ns
1.2 MC ttbar 1 lepton 1.2 MC ttbar 1 lepton
Data Data
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
−2 −1 0 1 2 20 25 30 35 40 45 50 55 60
Offline electron track η Pile-up <µ>

Figure 18: The GSF tracking efficiency for the LRT electron trigger with respect to the merged standard and large
radius offline electron track collections versus (top left) 𝑝 T , (top right) 𝑑0 , (bottom left) 𝜂, and (bottom right) the
average pile-up for data, semi-leptonic 𝑡 𝑡¯ MC events, and simulated pair production of selectrons with a 1 ns lifetime
and a mass of 100 GeV. Only statistical uncertainties are shown.

TRT, which are obtained by the extension of the tracks into the TRT before the final fit is performed.
The timing measurements presented here were recorded from the HLT farm, during a typical physics run
from October 2022. Figure 21 shows the mean execution times for the ID trigger-related algorithms for the
muon reconstruction as a function of the pile-up interaction multiplicity. Shown are the data preparation
times for the silicon detectors prior to the fast tracking, and the execution times for the fast, and precision
tracking themselves together with the combined silicon data preparation time and the total time for the
full TRT extension used for the precision tracking. The combined pixel and SCT data preparation, at
12 ms at high pile-up is slightly faster than the full TRT extension, at approximately 13 ms. The slowest
component from the silicon data preparation is the pixel data provider which fetches the pixel data from the
pixel read-out. The slowest component overall, taking approximately 40 ms per RoI at high pile-up is the
fast tracking, which exhibits a non-linear dependence on the pile-up multiplicity due to the combinatorial
nature of the pattern recognition. The precision tracking takes only approximately 6 ms at high pile-up in
the muon RoI.
The timing of the muonIso tracking, shown in Figure 22, is significantly improved compared to Run 2 [41].
With the reduced 𝑧 width, the muon isolation tracking algorithms take less than 25 ms in total per RoI at a
pile-up of 52. The total combined data preparation for the pixel and SCT takes less than 5 ms, with the

33
 1.8 1.8
Efficiency

Efficiency
1.6 ATLAS 1.6 ATLAS
Data 2022, s = 13.6 TeV Data 2022, s = 13.6 TeV
1.4 LRT muon trigger, Offline muon p >10 GeV, d0>2 mm
T
1.4 LRT muon trigger, Offline muon p >10 GeV, d0>2 mm
T
MC di-smuon, mass = 100 GeV, lifetime = 1 ns MC di-smuon, mass = 100 GeV, lifetime = 1 ns
1.2 MC ttbar 1 lepton 1.2 MC ttbar 1 lepton
Data Data
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
10 102 103 −150 −100 −50 0 50 100 150 200
Offline muon p [GeV] Offline muon d0 [mm]
T

1.8 1.8
Efficiency

Efficiency
1.6 ATLAS 1.6 ATLAS
Data 2022, s = 13.6 TeV Data 2022, s = 13.6 TeV
1.4 LRT muon trigger, Offline muon p >10 GeV, d0>2 mm
T
1.4 LRT muon trigger, Offline muon p >10 GeV, d0>2 mm
T
MC di-smuon, mass = 100 GeV, lifetime = 1 ns MC di-smuon, mass = 100 GeV, lifetime = 1 ns
1.2 MC ttbar 1 lepton 1.2 MC ttbar 1 lepton
Data Data
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
−2 −1 0 1 2 20 25 30 35 40 45 50 55
Offline muon η Pile-up <µ>

Figure 19: The precision tracking efficiency for the LRT muon trigger with respect to the merged standard and large
radius offline muon track collections versus (top left) 𝑝 T , (top right) 𝑑0 , (bottom left) 𝜂, and (bottom right) the
average pile-up for data, semi-leptonic 𝑡 𝑡¯ MC events, and simulated pair production of smuons with a 1 ns lifetime
and a mass of 100 GeV. Only statistical uncertainties are shown.

longest components being the pixel and SCT data providers, which fetch the data across the network, with
the pixel and SCT clustering being somewhat faster. The TRT extension has a combined execution time of
approximately 20 ms, in this case, longer that the other algorithms. In contrast, the standard muon fast and
precision tracking with |𝑧| < 225 mm takes around 40 ms per RoI with approximately 15 ms for the silicon
data preparation. The fast tracking alone for the muon isolation in Run 2 typically exceeded 115 ms per
RoI at similar pile-up [41].
The execution times for the electron-based algorithms are shown in Figure 23. The small size of the electron
RoI (0.1 × 0.2 in 𝜂–𝜙 space) results in an execution time for the electron pixel and SCT data preparation of
less than 16 ms at a pile-up of 52. The electron tracking takes an additional 20 ms in total for the fast and
precision tracking together. The new GSF tracking takes approximately 11 ms which is nearly three times
slower than the precision tracking due to the more complicated nature of the GSF reconstruction. Since
the GSF tracking is seeded by the tracks from the precision tracking, it has only a weak dependence on
the pile-up. The extension of the electron tracks into the TRT and subsequent TRT data preparation and
processing result in an additional 11 ms per RoI, as shown in Figure 23 (left). The dominant contribution
to the TRT processing is 4 ms at high pile-up from the TRT data provider, which includes the fetching of
the TRT data from the ROS system.

34
 Efficiency

Efficiency
1.4 ATLAS Ks candidate vertices, 472 < mv < 522 MeV 1.4 ATLAS Ks candidate vertices, 472 < mv < 522 MeV
Track selection: p > 1 GeV, |d0| > 5mm Track selection: p > 1 GeV
T T
s = 13.6 TeV Track matching: s = 13.6 TeV Track matching:
1.2 Data 2022 |∆d0| < 2.5mm, ∆Rtrk,trk < 0.05
trk,trk
1.2 Data 2022 |∆d0| < 2.5mm, ∆Rtrk,trk < 0.05
trk,trk
Enhanced bias events LRT online Enhanced bias events LRT online
1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 2 4 6 8 10 12 14 16 18 20 0 20 40 60 80 100 120 140 160
Offline track p [GeV] Offline track |d | [mm]
T 0
Efficiency

1.4 ATLAS Ks candidate vertices, 472 < mv < 522 MeV

Vertex selection: Two tracks p > 1 GeV
T
s = 13.6 TeV Track matching:
1.2 Data 2022 |Δd0| < 2.5mm, ΔRtrk,trk < 0.05
trk,trk
Enhanced bias events Both tracks matched to online
1

0.8

0.6

0.4 IBL PIX0 PIX1 PIX2

0.2

0
0 50 100 150 200 250 300
Rxy [mm]

Figure 20: Efficiency of online full scan LRT with respect to offline tracks associated with 𝐾𝑆0 candidates that have
not been matched to standard online full scan tracks. This efficiency is shown as a function of (left) the 𝑝 T and (right)
𝑑0 of the offline tracks. An additional requirement of |𝑑0 | > 5 mm is used when computing the efficiency versus 𝑝 T
to remove tracks with low displacement, which are expected to be reconstructed by the standard tracking. The bottom
plot shows the efficiency, combining both standard and LRT, to reconstruct both tracks of the offline vertex versus the
radius of the vertex. Only statistical uncertainties are shown.
Mean execution time per call [ms]

Mean execution time per call [ms]

ATLAS muon 45 ATLAS muon

10 Data s = 13.6 TeV, October 2022 Data s = 13.6 TeV, October 2022
Pixel data provider 40 Fast tracking
Pixel clustering Precision tracking
8 SCT data provider
35 Total Si data preparation
SCT clustering 30 Total TRT extension
Space point finder
6 25
20
4
15

2 10
5
0 0
25 30 35 40 45 50 55 25 30 35 40 45 50 55
Pile-up <m> Pile-up <m>

Figure 21: Execution times for the muon track reconstruction in the trigger. Shown are the times for (left) the
pixel and SCT data preparation, and (right) the muon fast, and precision tracking, together with the combined data
preparation for pixel and SCT, and the total time spent in the TRT extension, all as a function of the average pile-up.
Solid lines show ad hoc fits to data to guide the eye. Only statistical uncertainties are shown.

35
 5 30
Mean execution time per call [ms]

Mean execution time per call [ms]

ATLAS muonIso ATLAS muonIso
4.5
Data s = 13.6 TeV, October 2022 Data s = 13.6 TeV, October 2022
Pixel data provider
25 Fast tracking
4
Pixel clustering Precision tracking
3.5 SCT data provider Total Si data preparation
20
3 SCT clustering Total TRT extension
Space point finder
2.5 15
2
10
1.5
1
5
0.5
0 0
25 30 35 40 45 50 55 25 30 35 40 45 50 55
Pile-up <m> Pile-up <m>

Figure 22: Execution times for the muon isolation track reconstruction shown as a function of the average pile-up.
(left) Detailed breakdowns for the Pixel and SCT data preparation components. (right) The muon isolation tracking,
the combined silicon data preparation and TRT extension times. Solid lines show ad hoc fits to data to guide the eye.
Only statistical uncertainties are shown.

5
Mean execution time per call [ms]

Mean execution time per call [ms]

ATLAS electron 20 ATLAS electron

4.5
Data s = 13.6 TeV, October 2022 18 Data s = 13.6 TeV, October 2022
4 Track extension Fast tracking
Track extension processor 16 Precision tracking
3.5 TRT drift circle maker GSF reconstruction
14
3 TRT data provider Total Si data preparation
12 Total TRT extension
2.5 10
2 8
1.5 6
1 4
0.5 2
0 0
25 30 35 40 45 50 55 25 30 35 40 45 50 55
Pile-up <m> Pile-up <m>

Figure 23: Execution times for the (left) TRT extension and (right) tracking and data preparation for the electron
tracking, shown as a function of the average pile-up. Solid lines show ad hoc fits to data to guide the eye. Only
statistical uncertainties are shown.

The execution times for the full scan trigger track and vertex reconstruction are shown in Figure 24. The
combined data preparation for pixel and SCT is around 250 ms. Even with the optimisation to reduce the
processing time, the full scan tracking still takes longer than 1.1 s per event at pile-up of 52, while the
vertex finding with the full scan tracks takes only approximately 14 ms.
The execution times for the various tracking stages of 𝑏-jet triggers are shown in Figure 25. The super-RoI
tracking for the 𝑏-jet preselection stage takes 140 ms at a pile-up of 52, with the data preparation taking
approximately 80 ms in total. The precision tracking is not executed for the preselection. The data
preparation for the final RoI-based 𝑏-tagging stage is extremely fast (less than 4 ms) since only the data for
those parts of the detector not already processed by the preselection tracking need to be reconstructed.
The fast and precision tracking in the RoI for this final 𝑏-tagging stage takes less than 35 ms per jet RoI at
high pile-up. As with the other signatures the fast tracking shows the largest dependence on the pile-up,
although the longest contribution is the TRT extension at low pile-up, which shows a similar trend to the
precision tracking. It is slightly exceeded by the fast tracking at higher pile-up.

36
 120
Mean execution time per call [ms]

Mean execution time per call [ms]

ATLAS full scan 1200 ATLAS full scan
Data s = 13.6 TeV, October 2022 Data s = 13.6 TeV, October 2022
100 Pixel data provider Fast tracking
Pixel clustering
1000 Total Si data preparation
80 SCT data provider
SCT clustering 800
Space point finder
60
600

40 400

20 200

0 0
25 30 35 40 45 50 55 25 30 35 40 45 50 55
Pile-up <m> Pile-up <m>
Mean execution time per call [ms]

16 ATLAS full scan

Data s = 13.6 TeV, October 2022
14 Vertexing

12
10
8
6
4
2
0
25 30 35 40 45 50 55
Pile-up <m>

Figure 24: Execution times for (top left) the pixel and SCT data preparation, (top right) fast tracking and total silicon
data preparation time for the full scan tracking in jet triggers, and (bottom) for the vertex finding with the full scan
tracks, all shown as a function of the average pile-up. Solid lines show ad hoc fits to data to guide the eye. Only
statistical uncertainties are shown.

Figures 26 and 27 show the execution times in the HLT for the muon and electron LRT, respectively. For
muon LRT compared to standard muon tracking, the data preparation steps take approximately 1.5–3
times as long as standard tracking, except for the pixel data provider which is similar. The muon LRT fast
tracking is 2–3 times faster than standard tracking, while the precision tracking is similar. Differences
in the processing times of the LRT tracking with respect to the standard tracking are due to the special
configuration, such as using larger RoIs or seeding using only the SCT hits. For electron LRT compared
to standard electron tracking, the data preparation steps are a few milliseconds slower, the fast tracking
is approximately 1.5–2 times faster, while the precision and GSF tracking are 1.5–2 times slower. The
execution time of the track extension step is similar to standard tracking. While the other steps to include
the TRT are 1.5–2 times slower.

5.2 Calorimeter reconstruction

The HLT Calorimeter (HLTCalo) software performs the translation of the raw data read out from the
LAr and Tile calorimeters into the final software objects, the CaloCells. There are 187,652 CaloCells for
the entire detector. CaloCells associate the geometrical information (𝜂, 𝜙 and the longitudinal layer) of
calorimeter cells with their energy, pulse peaking time, hardware gain and quality factor as detailed in
Ref. [55]. The calorimeter reconstruction algorithms use the CaloCells information to reconstruct clusters

37
 Mean execution time per call [ms]

Mean execution time per call [ms]

30 ATLAS b-jet preselection
160 ATLAS b-jet preselection
Data s = 13.6 TeV, October 2022 Data s = 13.6 TeV, October 2022
Pixel data provider 140 Fast tracking
25 Pixel clustering Total Si data preparation
SCT data provider 120
20 SCT clustering
Space point finder 100
15 80
60
10
40
5
20
0 0
25 30 35 40 45 50 55 25 30 35 40 45 50 55
Pile-up <m> Pile-up <m>

4 35
Mean execution time per call [ms]

Mean execution time per call [ms]

ATLAS b-jet ATLAS b-jet
3.5 Data s = 13.6 TeV, October 2022 30 Data s = 13.6 TeV, October 2022
Pixel data provider Fast tracking
3 Pixel clustering Precision tracking
SCT data provider
25 Total Si data preparation
2.5 SCT clustering Total TRT extension
Space point finder 20
2
15
1.5
10
1

0.5 5

0 0
25 30 35 40 45 50 55 25 30 35 40 45 50 55
Pile-up <m> Pile-up <m>

Figure 25: Execution times for data preparation, tracking, and TRT extension for (top) the 𝑏-jet preselection tracking
and (bottom) 𝑏-jet RoI as a function of the average pile-up. The detailed breakdowns of the silicon data preparation
components are shown on the left. Solid lines show ad hoc fits to data to guide the eye. Only statistical uncertainties
are shown.

30
Mean execution time per call [ms]

Mean execution time per call [ms]

9 ATLAS muon LRT ATLAS muon LRT

8 Data s = 13.6 TeV, October 2022 Data s = 13.6 TeV, October 2022
Pixel data provider
25 Fast tracking
7 Pixel clustering Precision tracking
SCT data provider 20 Total Si data preparation
6 SCT clustering Total TRT extension
Space point finder
5 15
4
3 10

2
5
1
0 0
25 30 35 40 45 50 55 25 30 35 40 45 50 55
Pile-up <m> Pile-up <m>

Figure 26: The data preparation, TRT track extension, and tracking times for the muon large radius tracking algorithms
as a function of the mean pile-up. The detailed breakdown of the pixel and SCT data preparation stages are shown on
the left, and the tracking times and combined silicon data preparation time and total TRT extension on the right.
Solid lines show ad hoc fits to data to guide the eye. Only statistical uncertainties are shown.

38
 30
Mean execution time per call [ms]

Mean execution time per call [ms]

8 ATLAS electron LRT ATLAS electron LRT
Data s = 13.6 TeV, October 2022 Data s = 13.6 TeV, October 2022
7 Track extension
25 Fast tracking
Track extension processor Precision tracking
6 TRT drift circle maker GSF reconstruction
20
TRT data provider Total Si data preparation
5 Total TRT extension
15
4
3 10
2
5
1
0 0
25 30 35 40 45 50 55 25 30 35 40 45 50 55
Pile-up <m> Pile-up <m>

Figure 27: The TRT track extension, data preparation and tracking times for the electron large radius tracking
algorithms as a function of the mean pile-up. The detailed breakdown of the TRT extension and data preparation
stages are shown on the left, and the tracking times and combined silicon data preparation time and total TRT
extension on the right. Solid lines show ad hoc fits to data to guide the eye. Only statistical uncertainties are shown.

of energy for candidate electrons, photons, taus and jet objects as well as shower shape variables useful for
particle identification. The HLTCalo software handles both the high rate of regional data requests, tens
of kHz in multiple RoIs, and a similar rate of data processing requests which performs reconstruction
in the entire volume of the calorimeters for jets and missing transverse energy determination (full scan).
The HLTCalo software was adapted for the Run-3 AthenaMT framework, though still not optimally as
discussed in Section 8.1.
Two different clustering algorithms are used to reconstruct the clusters of energy deposited in the calorimeter:
a sliding-window algorithm [2] used for the fast electron and photon reconstruction step and a topological-
clustering (topo-cluster) algorithm [56]. The topo-cluster algorithm begins with a seed-cell search and
iteratively adds neighbouring cells to the cluster if their energies7 (𝐸 cell ) are above a given energy threshold
that is a function of the root-mean-square of the expected electronics and pile-up noise (𝜎). The seed cells
are first identified as those cells that have |𝐸 cell |/𝜎 > 4. All neighbouring cells with |𝐸 cell |/𝜎 > 2 are
then added to the cluster and, finally, all the remaining neighbours to these cells with |𝐸 cell |/𝜎 > 0 are
also added. These 𝜎 thresholds (4,2,0) can, in principle, be adapted but the numbers above are the same as
used in the offline version of the topo-cluster algorithm. The clusters are grown by energy occupancy with
no predefined shape. Given the number of searches (via look-up table helpers) of the cell neighbours and
calculations of energy ratios, this algorithm is one of the most resource-consuming parts of the HLTCalo
reconstruction.
One of the necessary conditions to calculate the energy in a given cell is related to pile-up-dependent effects.
There is a small and measurable bunch-by-bunch energy shift which depends on the LHC luminosity profile
(i.e. the structure of empty and filled bunches as well as their relative intensities, which is given by the
profile of ⟨𝜇⟩ as a function of bunch-crossing identification). The details of the offline correction procedure
are provided in Ref. [57]. The only difference for the online application is that this profile is updated only
if the pile-up value ⟨𝜇⟩ changes by more than 5%, which means that a few tens of LBs can pass before
the update. This correction improves the resolution of many variables used for particle identification and
energy estimation. The calculation of this correction was adapted to the new AthenaMT environment of
Run 3 by separating it into a run-long component and an event-by-event component.

7 The LAr calorimeter electronics are designed such that signals from pile-up in earlier bunch crossings appear as negative energy.

39
 Arbitrary units
ATLAS Full scan topo-cluster
Topo-cluster for tau
10−1 Topo-cluster for e/ γ
Data 2022
Fast reconstruction
s= 13.6 TeV

10−2

10−3

−40 −30 −20 −10 0 10 20 30 40

(E - ET )/E [%]
Toffline online Toffline

Figure 28: Transverse energy resolution in percent for HLTCalo clusters obtained with respect to calorimeter clusters
reconstructed offline for 2022 data for the following algorithms: full scan topo-cluster for jets and 𝐸 Tmiss , topo-cluster
in RoI for taus, topo-cluster in RoIs for electrons (e) and photons (𝛾), fast sliding-window reconstruction for e/𝛾.

0.7 70
Arbitrary units

Mean execution time per call [ms]

ATLAS Topo-cluster for e/γ ATLAS

0.6 Topo-cluster for tau 60
Data 2022
s= 13.6 TeV Fast Reconstruction Data 2023, s= 13.6 TeV
0.5 50
Full scan topo-cluster
0.4 40
0.3
30
0.2 Full scan topo-cluster processing
20
0.1
10
0
0
10−1 1 10 102 103 25 30 35 40 45 50 55 60
Processing Time per Call [ms] Pile-up <µ>

Figure 29: (left) Processing time per call for the following algorithms: topo-cluster in RoIs for e/𝛾, topo-cluster in
RoIs for taus, fast sliding-window reconstruction for e/𝛾, full scan topo-cluster for jets and 𝐸 Tmiss . (right) Processing
time as a function of the average pile-up for the full scan topo-cluster algorithm. A solid line shows an ad hoc fit to
data to guide the eye.

40
The measurements presented here were done in the HLT farm during a typical physics run. The 𝐸 T
resolution of HLTCalo clusters obtained with different reconstruction algorithms are shown in Figure 28.
The peaks of all four distributions are consistent with zero within 1%. The residual differences are
due to different reconstruction and calibration algorithms. The fast calorimeter reconstruction relies
on a sliding window cluster algorithm, which allows for a good electron energy estimate, despite the
very short time available (< 5 ms). It has an energy resolution of 4.3% and a longer low-energy tail
than the topo-cluster reconstruction algorithms, which are shapeless and pick up much more low-energy
activity. The topo-clusters used for electrons, photons and taus are reconstructed in an RoI, as discussed in
Sections 6.1 and 6.3, and those used for jets and 𝐸 Tmiss are reconstructed in the full calorimeter volume
(details in Sections 6.4 and 6.6). Topo-clusters can be calibrated at the EM scale, for which the energy of an
isolated topo-cluster is the sum of its constituent cell energies. Alternatively, local hadronic calibration [58]
can be applied as a cell-level correction to improve the mean response to hadronic showers, and is primarily
used for tau reconstruction, detailed further in Section 6.3, as well as for large-radius jet triggers, detailed
in Section 6.4. The 𝐸 T resolutions are 4.1%, 3.7% and 2.8% for e/𝛾, tau and full scan topo-clusters,
respectively. The processing time of the main HLTCalo reconstruction algorithms per call is shown in
Figure 29 (left). It varies from a few milli-seconds running at 22 (14) kHz for topo-cluster reconstruction in
e/𝛾 (tau) RoIs to a few tens of milliseconds at 23 kHz for full scan topo-clustering. The latter also exhibits a
known pile-up dependence as shown in Figure 29 (right), unlike the other algorithms. The fast calorimeter
reconstruction runs at a total call rate of 132 kHz8 and its timing distribution has a double-peaked structure
seen in Figure 29 (left). The lower peak in the processing time is due to a fraction of calls which do
not require time-consuming data requests as data was cached once by the first RoI request and only the
reconstruction is run.

5.3 Tracking in the muon spectrometer

The HLT muon reconstruction [22] consists of two steps: the first is fast and trigger specific, while
the second is based on precision reconstruction. The precision reconstruction makes use of the same
software as the offline muon reconstruction with some adaptations for online running. Muon candidates
are reconstructed from combined tracks in the MS and the ID subdetectors. Most of the muon triggers are
based on combined muon candidates. MS-only candidates are used to trigger particular topologies, such as
LLPs where there might be no corresponding ID track.
In the fast reconstruction stage, each L1 muon candidate is refined by including the precision data
from the MDT chambers in the RoI defined by the L1 candidate. A track fit is performed using the
MDT drift times and positions as well as the sTGC and MM chambers in the endcap regions, and a 𝑝 T
measurement is assigned using lookup tables, creating MS-only muon candidates. The MS-only muon
track is back-extrapolated to the interaction point and combined with tracks reconstructed in the ID to form
a combined muon candidate with a refined track parameter resolution.
In order to recover the efficiency in low 𝑝 T di-muon topologies, e.g. 𝐵-meson decays where multiple
muons arrive close together in the MS, a new inside-out algorithm was developed for Run 3. The MS-only
back-extrapolated tracks are used as an RoI to reconstruct the ID tracks, which are then extrapolated to the
MS and used as seeds for the fast MS-only muon reconstruction. Collimated di-muon trigger candidates
can hence be distinguished.

8 It runs at L1 EM rate of 46 kHz for about 2.9 RoIs per event on average.

41
The precision stage follows the same strategy, combining tracks in the MS and the ID subdetectors to
reconstruct the muon candidates. It starts from the refined RoIs identified in the fast reconstruction step
to form muon candidates using information from the MS detectors. They are then extrapolated to the
interaction point and combined with ID tracks. If the MS track cannot be matched to an ID track, combined
muon candidates are searched for by extrapolating ID tracks to the MS.
The full scan mode is used to find additional muons that are not found by the RoI-based method mainly
due to L1 inefficiencies (‘noL1’ triggers). In the full scan mode, muon candidates are first sought in all
muon detectors. Then, RoIs are constructed around the found MS tracks and ID tracks are reconstructed
within these RoIs. The same combination procedure as for the RoI-based method is used to construct
combined full scan muons. Given the high CPU demand of the full scan reconstruction, it is only executed
in multi-object triggers with at least one of the trigger objects found by an RoI-based algorithm.
The performance of the muon tracks at the HLT level compared to the offline muon reconstruction is
illustrated in Figure 30. The 𝑝 T resolution of the trigger muon candidates is measured with 𝑍 → 𝜇𝜇
candidate events using the tag-and-probe method [22]. Events are required to contain a pair of muons with
opposite charge and an invariant mass within 10 GeV of the Z pole mass and the offline reconstructed muons
are required to pass the medium identification requirements. The distributions of the relative residuals
between online and offline track parameters are constructed in bins of offline muon 𝑝 T and the width of the
distribution in each bin is obtained by means of a Gaussian fit. The resolution for combined trigger muons
is better than the resolution for MS-only muons as expected, thanks to the higher precision of the ID track
measurements, especially at low transverse momentum. The relative 𝑝 T resolution is about 3–4% (8%) and
1–2% (2–4%) in the barrel (endcap) region for precision MS-only and combined muons, respectively, and
tends generally to degrade towards higher-𝑝 T . The MS-only resolution in the endcap region is expected to
improve after the NSW commissioning is completed.

0.1 0.1
σ(pOffline - pOnline)/pOffline

σ(pOffline - pOnline)/pOffline

0.09 ATLAS 0.09 ATLAS

T

0.08 Data 2022, s= 13.6 TeV, |η|<1.05 0.08

0.07 0.07
T

Precision MS-only muons

Precision MS-only muons
0.06 0.06
Precision combined muons Precision combined muons
0.05 0.05
T

0.04 0.04
0.03 0.03
0.02 0.02
0.01 0.01 Data 2022, s= 13.6 TeV, 1.05<|η|<2.5
0 0
20 40 60 80 100 120 140 20 40 60 80 100 120 140
Offline muon p [GeV] Offline muon p [GeV]
T T

Figure 30: Width of the residuals for 𝑝 T as a function of the offline muon 𝑝 T for the precision MS-only and combined
algorithms in (left) the barrel and (right) endcaps. Only statistical uncertainties are shown.

The timing measurements presented in were recorded from the HLT farm, during a typical physics run
from November 2022. The processing times per RoI are shown in Figure 31 for the fast MS-only and fast
combined algorithms. The call frequencies are 37 kHz and 23 kHz, respectively. Figure 32 shows the
most CPU consuming steps of the precision reconstruction: segment-finding algorithm, MS-only track
building, MS and combined muon candidate building and inside-out recovery. The call frequency of the
segment-finding, track-building and MS muon candidate building algorithms is 11 kHz, while they are
2.7 kHz and 0.34 kHz for the combined muon candidate building and inside-out recovery, respectively.

42
The large tails of the distribution of the fast combined and precision combined algorithms are related to
the algorithms not being fully optimised for Run-3 conditions in 2022. The performance has since been
optimised.

109

Entries
ATLAS
p Fast MS-Only
Data s =13.6 TeV, <time>: 1.4 ms
108 November 2022 Fast Combined
<time>: 42 ms

107

106

105

104

103
0 10 20 30 40 50 60
Processing time per Call [ms]

Figure 31: Processing times per RoI for the fast MS-only and fast combined algorithms. The mean time of each
algorithm is indicated in the legend. The last bin of the distribution includes the overflow events.

108 108
Entries

Entries

ATLAS
p Precision MS Track Building ATLAS
p Precision MS Candidate Building
Data s =13.6 TeV, <time>: 21 ms Data s =13.6 TeV, <time>: 32 ms
107 November 2022 107 November 2022
Precision Segment Making Precision Combined Muon Building
<time>: 5.7 ms <time>: 168 ms
Inside Out Recovery
106 106 <time>: 21 ms

105 105

104 104

103 103

102 102
0 100 200 300 400 500 600 700 800 0 100 200 300 400 500 600 700 800
Processing time per Call [ms] Processing time per Call [ms]

Figure 32: Processing times per RoI for the most CPU consuming precision steps: segment-finding and track-building
algorithm (left), muon candidate building and inside-out recovery (right). The mean time of each algorithm is
indicated in the legend. The last bin of the distributions includes the overflow events.

43
6 Physics triggers

The final event selection is based on trigger signatures, such as leptons, hadrons, and global event quantities
such as missing transverse momentum. They are formed by placing different selection criteria on the
various reconstructed objects. The selection criteria and performance of the various trigger signatures is
described in this section, highlighting the differences with respect to the Run-2 selection and performance
of primary triggers during 2022.

6.1 Electrons and photons

6.1.1 Electron and photon trigger reconstruction and selection

Electron and photon reconstruction at the HLT is performed on each EM RoI provided by L1, which
satisfies the 𝐸 T requirement, and any other L1 selection requirements specified by the trigger menu (e.g.
isolation) of at least one active HLT chain. In the HLT, fast calorimeter algorithms are executed first,
allowing precision algorithms to run at a reduced rate later in the trigger sequence. The reconstruction of
candidate electrons and photons uses the sliding-window algorithm with rectangular clustering windows
of size Δ𝜂 × Δ𝜙 = 0.075 × 0.175 in the barrel and 0.125 × 0.125 in the endcaps. The fast calorimeter
selection step has three implementations [5]. The default fast calorimeter selection step for electrons uses
a neural-network-based Ringer algorithm [59], which uses as input energy sums of all the cells in 100
concentric rings centred around the most energetic cell in each calorimeter sampling layer. In Run 2 it was
used only for triggering electrons with 𝐸 T ≥ 15 GeV, but it is applied from 𝐸 T ≥ 5 GeV in Run 3. The
Ringer algorithm is optimised in two regions of 𝐸 T , between 5–15 GeV with 𝐽/𝜓 → 𝑒𝑒 MC samples and
for 𝐸 T > 15 GeV with 𝑍 → 𝑒𝑒 MC samples. For electrons, it is optionally possible to use fast calorimeter
selections which take as an input either only the cluster 𝐸 T (𝐸 T -based) or the cluster 𝐸 T with three shower
shape parameters (cut-based) [5]. The electron candidates are then required to have tracks within the RoI
(obtained from the fast track reconstruction) matching the corresponding clusters [5]. In contrast, photon
candidates are reconstructed using the calorimeter information only, with cut-based selection criteria that
are the same as in Run 2 [5].
For Run 3, the HLT electron and photon precision reconstruction becomes closer to the offline reconstruc-
tion [44], due to the implementation of a new super-clustering algorithm, described below, and the use of
the GSF tracking algorithm for electrons, described in Section 5.1.3.
Electron and photon super-clusters are reconstructed in two stages. The first stage consists of finding the
seed topo-cluster candidates in the same RoI as used for the tracking: 0.1 × 0.2 in 𝜂 − 𝜙 space. These
seed topo-cluster candidates form the basis of super-clusters. The second stage is the identification of
topo-clusters near the seed candidates which are identified as satellite cluster candidates. They may emerge
from bremsstrahlung radiation or topo-cluster splitting. For a cluster to become an electron super-cluster
seed, it is required to have a minimum 𝐸 T of 1 GeV and match to a track with at least four hits in the silicon
tracking detectors. For photon reconstruction, a cluster must have 𝐸 T greater than 1.5 GeV to qualify as a
super-cluster seed. If a cluster meets the seed cluster requirements, the algorithm attempts to find satellite
clusters, which represent secondary EM showers originating from the same initial electron or photon. A
cluster is considered a satellite if it falls within a window of Δ𝜂 × Δ𝜙 = 0.075 × 0.125 around the seed
cluster barycentre. The energy of the clusters is calibrated using a multivariate technique such that the
response of the calorimeter layers is corrected in data and simulation [44].

44
At the precision step, the electron identification relies on a multivariate technique using a likelihood
discriminant, while the photon identification is cut-based. These identifications, as well as isolation
requirements which are applied to some triggers, remain unchanged with respect to their Run-2 configurations
detailed in Ref. [5].

6.1.2 Electron and photon trigger menu

Data taken with electron and photon triggers are used in a wide range of ATLAS physics analyses, from
SM precision physics to searches for new physics. The various triggers cover the energy range between a
few GeV and several TeV. The 2022 electron and photon trigger thresholds remain the same as in 2018 [5].
The minimum 𝐸 T thresholds for the isolated single electron and non-isolated photon triggers are 26 and
140 GeV with rates of 186 Hz and 46 Hz at luminosity of 1.8 × 1034 cm−2 s−1 , respectively. To increase the
single electron trigger efficiency at 𝐸 T of 60 GeV and 140 GeV, triggers with no isolation requirements and
looser identification are present at rates of 20 Hz and 2 Hz, respectively. Additional single electron triggers
with 𝐸 T thresholds at 140 (300) GeV and the cut-based (𝐸 T -based) fast calorimeter selection, instead of
the default Ringer algorithm, run at 3 (6) Hz improving sensitivity to merged electrons coming from decays
of boosted dibosons [5]. Triggering on low 𝐸 T electrons and photons is very challenging because of the
high rates at low trigger thresholds. However, this can be mitigated by requiring the presence of multiple
electrons or photons in the event, which helps to reduce the trigger thresholds with respect to single electron
or photon triggers. The primary di-photon trigger is mainly designed for the efficient selection of events
with Higgs boson candidates in the di-photon decay channel. It has a rate of 18 Hz for medium photon
identification and trigger 𝐸 T thresholds of 35 and 25 GeV for the leading and subleading photons. A second
symmetric di-photon trigger with loose identification and the 𝐸 T requirement of 50 GeV has a rate of 7 Hz.
A di-electron trigger at lower 𝐸 T requirement of 17 GeV each and loose identification has a rate of 12 Hz.

6.1.3 Electron and photon trigger performance

The efficiency of the lowest unprescaled single-electron trigger with respect to offline electron candidates
is shown in Figure 33 for each step of the online reconstruction. This and the next measurements are
performed using the tag-and-probe method in 𝑍 → 𝑒𝑒 events [5]. The HLT inefficiency is due to differences
in the online and offline electron identification and isolation [5] and a 1–2% lower efficiency of the trigger
precision tracks, shown in Figure 11. The efficiency of the combination of all primary single-electron
triggers is shown in Figure 34. Scale factors derived from the observed data/MC simulation differences are
used to correct other MC simulation samples used in data analyses.
Figure 35 shows the efficiencies of the primary di-photon triggers extracted with the bootstrap method [5].
Photon trigger efficiencies are always very high, but while the 25 GeV trigger is fully efficient at 5 GeV
above its threshold, higher 𝐸 T triggers are not, as can be seen in Figure 35 (right). As for the electron
triggers, scale factors are used to correct for the observed data/MC simulation differences.

6.1.4 Trigger level analysis with photons

Hadronically decaying low-mass resonances around the electroweak scale can be probed through composite
trigger chains targeting both the final-state di-jet pair and an additional feature, such as an initial-state
radiation photon. The Run-3 TLA workflow is upgraded to allow for the recording of HLT-reconstructed

45
 1.2 1.2
Trigger efficiency

Trigger efficiency
ATLAS Data 2022, s = 13.6 TeV ATLAS Data 2022, s = 13.6 TeV
1 1

0.8 0.8
Z→ ee tag-and-probe, -2.47 < η < 2.47
Z→ ee tag-and-probe, E >27 GeV
0.6 Single isolated electron trigger of E >26 GeV
T
0.6 T
Single isolated electron trigger of E >26 GeV
T
L1Calo L1Calo
0.4 FastCalo 0.4 FastCalo
FastElectron FastElectron
PrecisionCalo PrecisionCalo
HLT HLT
0.2 0.2

0 0
20 30 40 50 60 70 80 90 100 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2
Offline electron ET [GeV] Offline electron η

Figure 33: Efficiency of the primary isolated electron trigger with 𝐸 T >26 GeV for all steps of the HLT electron
reconstruction as a function of the offline electron (left) 𝐸 T and (right) 𝜂. Efficiency is given with respect to the
offline electrons which satisfy the tight identification and isolation criteria. Only statistical uncertainties are shown.
Trigger efficiency

Trigger efficiency

1.2 ATLAS 1.2 ATLAS

s = 13.6 TeV s = 13.6 TeV
p p

1.0
Z ee tag-and-probe, -2.47 < < 2.47 1.0
Z ee tag-and-probe, ET > 27 GeV

0.8 0.8
0.6 0.6
0.4 0.4
Single electron trigger combination Single electron trigger combination
0.2 Data 2022 0.2 Data 2022
Z ee MC Z ee MC
0.0 0.0
Data/MC

Data/MC

1.0 1.0

0.8 0.8
0 50 100 150 200 250 2 1 0 1 2
Offline electron ET [GeV] Offline electron

Figure 34: Combined efficiency of three primary single electron triggers as a function of the offline electron (left) 𝐸 T
and (right) 𝜂 for both MC simulation and 2022 data. Efficiency is given with respect to the offline electrons which
satisfy the tight identification and isolation criteria. Only statistical uncertainties are shown.

46
 1.2 1.2
Trigger efficiency

Trigger efficiency
ATLAS s = 13.6 TeV ATLAS s = 13.6 TeV
1 1

0.8 0.8
HLT photon triggers: HLT photon triggers:
ET>25 GeV, medium Data 2022 Single γ MC ET>25 GeV, medium Data 2022 Single γ MC
0.6 ET>35 GeV, medium Data 2022 Single γ MC 0.6 ET>35 GeV, medium Data 2022 Single γ MC
ET>50 GeV, loose Data 2022 Single γ MC ET>50 GeV, loose Data 2022 Single γ MC
0.4 0.4

0.2 0.2
Data/MC

Data/MC
1.05 1.05
1 Offline photon E [GeV]
T
1 Offline Photon η
0.95 0.95
0.9 0.9
10 20 30 40 50 60 70 80 90 100 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
Offline photon E [GeV]
T
Offline photon η

Figure 35: Efficiency of the 25, 35 and 50 GeV legs of di-photon triggers as a function of the offline photon (left)
𝐸 T and (right) 𝜂 for both MC simulation and 2022 data. The ratios of data to MC simulation efficiencies are also
shown. Efficiency is given with respect to the offline photons reconstructed with the tight offline identification
and isolation requirements; candidates in the calorimeter transition region 1.37 < |𝜂| < 1.52 are not considered.
10 20
Legacy non-isolated 30 L140
EM 50 60 with
trigger 70 𝐸 T80> 20 100defined in−Ref.
90GeV 2.5 −[5] −1.5
2 is used−1as−a0.5seed.
0 For
0.5 (right),
1 1.5 only2offline
2.5
candidates with 𝐸 T values 5 GeV above the corresponding trigger threshold are used. Only statistical uncertainties
are shown.

photons, either standalone or in association with other TLA-compatible signatures to enable such searches
at rates higher than possible with the standard triggers. A new TLA trigger records all jets in events with
at least one tight photon [5] with 𝐸 T > 35 GeV and three additional jets with 𝑝 T > 25 GeV reconstructed
with the PFlow algorithm, described in Section 6.4. The photon will, in most events, be reconstructed as a
jet as well, making this effectively a trigger for a signature with a photon and two jets. A nominal HLT rate
√
of such trigger is about 700 Hz at luminosity of 1.8 × 1034 cm−2 s−1 at 𝑠 = 13.6 TeV. This TLA approach
allows extended sensitivity of hadronic searches to resonance masses potentially as low as 100 GeV.
HLT photons saved to the TLA stream are decorated with additional variables computed within the HLT
precision photon reconstruction and calorimeter isolation trigger sequences. Such variables are intended to
enable the derivation of a custom residual photon energy calibration, as well as to constrain the contribution
from non-prompt, non-isolated photons in the TLA data set.
Figure 36 shows the energy response of HLT photons spatially matched to offline photons in the 𝐸𝑇 range
of interest for Run-3 TLA searches targeting photons. HLT photons calibrated with the default 2022
calibration sequence are shown to be already within 2% of offline-calibrated photons across the probed 𝐸𝑇
range.

6.1.5 Electron and photon triggers for heavy-ion collisions

Due to the compactness of electron and photon showers in the detector, the standard L1 EM RoIs can still
be used in HI collisions to select electron and photon candidates with reasonable rate and purity. Typical
𝐸 T thresholds for L1 EM RoIs are 12 and 16 GeV. However, the large UE contribution present in HI
collisions leads to significant distortions of the shower shapes and subsequent inefficiency associated with
the electron and photon identification requirements at HLT. To improve the electron and photon trigger
performance, the UE subtraction procedure is applied at the HLT.

47
 1.10

HLT/Offline ET response
ATLAS
1.08 Data 2022, s = 13.6 TeV
1.06 Tight single photon trigger E >25 GeV (prescaled)
T

1.04
1.02
1.00
0.98
0.96
0.94
0.92
0.90
0 50 100 150 200 250 300
offline photon E [GeV]
T

Figure 36: Energy response in TLA photons spatially matched to the leading offline photons. HLT-level photons
are calibrated with the default sequence employed within the HLT reconstruction for the 2022 data taking, while
reference offline photons are calibrated according to the Run-2 recommendations [44]. Events are required to satisfy
a 25 GeV tight single-photon prescaled support trigger. Only statistical uncertainties are shown.

Two electron HLT sequences are designed as primary physics triggers in the HI menu: a cut-based trigger
with an 𝐸 T > 20 GeV requirement and a likelihood-based trigger with an 𝐸 T > 15 GeV requirement.
For both triggers the UE subtraction is performed at the HLT as described in Section 4.2. For the
likelihood-based electron identification [5], the standard 𝑝 𝑝 probability density functions are evaluated
using the UE-corrected variables. An advantage of the likelihood-based approach is the significant
reduction in the output rate in comparison to the cut-based trigger for a given 𝑝 T threshold. The likelihood
trigger has significantly better purity at the cost of a loss in trigger efficiency.
The primary photon HLT sequence in the HI trigger menu uses a cut-based photon identification with an
𝐸 T > 20 GeV requirement including the subtraction of the UE contribution. After the UE subtraction
procedure, the photon trigger efficiency remains high across the full range of centralities.

6.2 Muons

The reconstruction of muon candidates at the HLT is described in detail in Section 5.3. The criteria for
selecting muon candidates are dependent on the algorithm used for their reconstruction. The MS-only
chains select on the 𝑝 T of the muon candidate measured solely by the muon spectrometer, while the
combined muon chains apply requirements on the matching between the ID and MS tracks and their
combined 𝑝 T .

6.2.1 Muon trigger menu

Muon triggers cover a wide momenta range between a few GeV, for 𝐵-meson-decay studies, up to several
TeV for new phenomena searches. The primary triggers in the muon trigger menu include single-muon
triggers with and without isolation requirements, symmetric and asymmetric di-muon and multi-muon

48
triggers. The Run-3 muon trigger menu is similar to that used in Run 2 [22], accounting for the refinement
of L1Muon thresholds, discussed in Section 3.2.
The improved suppression of fake muons and the 2022 running conditions allowed for a lowered HLT
threshold of the lowest-unprescaled isolated single-muon trigger by 2 GeV to 𝑝 T > 24 GeV maintaining
its rate at about 200 Hz at a luminosity of 1.8 × 1034 cm−2 s−1 . As in Run 2, combined muon candidates
must fulfil the following track isolation requirement: the scalar sum of the 𝑝 T values of tracks within a
variable-size cone around the muon (excluding its own track) must be less than 7% of the muon 𝑝 T . The
track isolation cone size for muons, Δ𝑅, is given by the smaller of Δ𝑅 = 10 GeV/𝑝 T and Δ𝑅 = 0.3. A
non-isolated trigger with 𝑝 T > 50 GeV helps to increase the efficiency for high-𝑝 T muons. Additionally, a
trigger that selects only muons in the barrel region (|𝜂| < 1.05) reconstructed using MS-only information
is available at 𝑝 T > 60 GeV.
There were no changes to the multi-muon thresholds in 2022 with respect to Run 2. Two combined muon
candidates are required with a 𝑝 T threshold of 14 GeV each at the HLT with a rate of 24 Hz at a luminosity
of 1.8 × 1034 cm−2 s−1 . To avoid an efficiency loss due to the limited acceptance of L1, di-muon and
tri-muon triggers seeded by the single L1 muon trigger are also present in the menu. These sub-leading
muons, with 𝑝 T greater than 8 GeV and (4 GeV, 4 GeV) for di-muon and tri-muon triggers, respectively,
are reconstructed by the muon full scan algorithm described in Section 5.3. Low-𝑝 T di-muon triggers are
further discussed in Section 6.7.

6.2.2 Muon trigger performance

The muon efficiencies are determined using the tag-and-probe method with 𝑍 → 𝜇𝜇 events. The efficiencies
of the combination of the two lowest-unprescaled single-muon triggers are shown in Figures 37 and 38.
The HLT efficiency relative to L1 is close to 100% both in the barrel and in the endcaps. The L1 muon
trigger efficiency is about 60% in the barrel and 80% in the endcap regions for offline medium muons [45]
with 𝑝 T > 25 GeV.
The measured trigger efficiency in Figure 37 is lower than that of the expected efficiency in MC. This is
due to the L1 muon trigger in the barrel region being simulated with an optimistic lower-bound on chamber
efficiency to allow for a reasonable MC efficiency for chambers whose efficiencies are later recovered
during data taking. Scale factors derived from these observed differences are used to correct other MC
simulation samples used in data analyses.

6.3 Taus

6.3.1 Tau trigger reconstruction and selection

The HLT tau trigger targets only hadronic decays of tau lepton candidates (𝜏had−vis ). Events with taus
decaying leptonically are recorded by electron and muon triggers described in Sections 6.1 and 6.2,
respectively. The HLT tau trigger reconstruction is subdivided in two steps: a calorimeter-only preselection
and track reconstruction with an offline-like selection.
At the first step, the 𝜏had−vis candidate is reconstructed purely from calorimeter information. Calorimeter
cells inside the RoI identified at L1 are retrieved and the topo-clustering algorithm, described in Section 5.2,
is executed. Thanks to the full detector granularity and the bunch-by-bunch pile-up corrections, the energies

49
 Trigger efficiency

Trigger efficiency
ATLAS ATLAS
s = 13.6 TeV, single muon trigger combination s = 13.6 TeV, single muon trigger combination
1 HLT: 24 GeV isolated or 50 GeV L1: 14 GeV 1 HLT: 24 GeV isolated or 50 GeV L1: 14 GeV

0.5 0.5
Data 2022 Data 2022
MC MC
Z→µµ tag-and-probe, |η|<1.05 Z→µµ tag-and-probe, |η|<1.05
0 0
Data / MC

Data / MC
1 1
0.8 0.8
0.6
0.6
20 30 40 50 60 70 80 90 100 −3 −2 −1 0 1 2 3
Offline muon p [GeV]
T
Offline muon φ [rad]

p [GeV]
Figure 37: Muon trigger efficiencies as a function
T
of (left) 𝑝 T and (right) 𝜙 for 𝑝 T > 25 GeV ofOffline muon φmedium
the offline [rad]
muon for the combination of two single-muon triggers in the barrel region. The ratios of data to MC simulation
efficiencies are also shown. Only statistical uncertainties are shown.
Trigger efficiency

Trigger efficiency

ATLAS ATLAS
s = 13.6 TeV, single muon trigger combination s = 13.6 TeV, single muon trigger combination
1 HLT: 24 GeV isolated or 50 GeV L1: 14 GeV 1 HLT: 24 GeV isolated or 50 GeV L1: 14 GeV

0.5 0.5
Data Data 2022
MC MC
Z→µµ tag-and-probe, 1.05<|η|<2.5 Z→µµ tag-and-probe, 1.05<|η|<2.5
0 0
Data / MC

Data / MC

1.02 1.02
1 1
0.98 0.98
20 30 40 50 60 70 80 90 100 −3 −2 −1 0 1 2 3
Offline muon p [GeV]
T
Offline muon φ [rad]

p [GeV]
Figure 38: Muon trigger efficiencies as a function
T
of (left) 𝑝 T and (right) 𝜙 for 𝑝 T > 25 GeV ofOffline muon φmedium
the offline [rad]
muon for the combination of two single-muon triggers in the endcap region. The ratios of data to MC simulation
efficiencies are also shown. Only statistical uncertainties are shown.

50
of these reconstructed topo-clusters are very close to the offline ones. These clusters are calibrated with the
local hadron calibration (LC) [58] and their vectorial sum is used as a ‘jet seed’ for the reconstruction of
the 𝜏had−vis candidate. The energy of the 𝜏had−vis candidate is calculated from the LC clusters in a cone
of Δ𝑅 < 0.2 around the barycentre of the jet seed. A dedicated 𝜏had−vis energy calibration is applied to
improve the precision of the energy measurement and follows the offline procedure [60]. Then, a selection
on the minimum 𝑝 T of the 𝜏had−vis candidate is applied, and only the remaining candidates pass to the next
step.
The second step of the HLT tau trigger first runs a fast-tracking algorithm, followed by a precise measurement
of the tracks associated with the 𝜏had−vis candidate and a final 𝜏had−vis identification based on a Recurrent-
Neural-Network (RNN) algorithm. The fast tracking algorithm is a trigger-specific pattern recognition
algorithm that runs in two stages. In the first stage, the leading track is sought in a narrow Δ𝜂 × Δ𝜙 around
the 𝜏had−vis candidate along the entire beamline. In the second stage, additional tracks associated with
the 𝜏had−vis candidate are sought in a larger Δ𝜂 × Δ𝜙 region but in a narrow range around the leading
track along the beamline. This strategy is CPU-efficient as it minimises the volume in which the pattern
recognition algorithm is executed, as discussed in Section 5.1.1.
A precision-tracking algorithm similar to the offline one is run using the tracks identified by the second
step of the fast tracking as seeds to measure their properties more precisely. Using these tracks as well
as the calorimeter information, the input variables for the 𝜏had−vis identification are computed. Three
sets of variables for zero-prong, one-prong and multi-prong 𝜏had−vis candidates are used depending on
whether the number of precision tracks associated to the 𝜏had−vis in Δ𝑅 < 0.2 is zero, one, or more than
one, respectively. The architecture implementation of the RNN follows closely its offline counterpart, as
described in Ref. [60]. Finally, tau candidates are required to have up to three tracks within Δ𝑅 < 0.2 and
up to one track within 0.2 < Δ𝑅 < 0.4 around the 𝜏had−vis . In addition, they are required to be identified by
the RNN. Given the small increase in event rate and potential efficiency gain, the identification working
point is loosened for the 𝜏had−vis 𝑝 T > 280 GeV, and both criteria (identification and number of tracks)
are completely dropped when the 𝜏had−vis 𝑝 T > 440 GeV. A different RNN is used to trigger on LLP and
trained for zero-, one- and multi-prong taus using a MC sample of LLP.

6.3.2 Tau trigger menu

The tau trigger menu selects a wide spectrum of final states that involve hadronically decaying tau leptons.
There are four main categories of triggers: single-tau, di-tau, tau+X (X=light leptons, 𝐸 Tmiss ) and events
for tag-and-probe performance studies. For each of these categories, the definition of the triggers varies
based on (i) the identification requirement, (ii) the 𝑝 T threshold applied and, for the multi-object final
states, (iii) the eventual presence of a topological cut on the angular distance between the two objects
selected in the final state. Dedicated triggers for identifying tau leptons originating from the decay of LLP
are included. The lowest threshold of the single-tau unprescaled triggers is 160 GeV and 180 GeV for the
standard and LLP taus. The individual rates are about 40 Hz and 50 Hz, respectively, at a luminosity of
1.8 × 1034 cm−2 s−1 . The overlap fraction of the higher threshold LLP trigger with the 160 GeV standard
trigger is 90%. Lower threshold single-tau triggers with or without looser requirements are prescaled at
L1 and/or the HLT and are used to support the primary triggers. There are two primary di-tau triggers,
which run at rates of 25 Hz and 10 Hz, respectively. The first, main, trigger requires two tau candidates
of 𝑝 T > 35 and 25 GeV and the second, LLP, trigger requires 𝑝 T > 80 and 60 GeV. LLP tau triggers are
discussed further in Section 6.9.

51
6.3.3 Tau trigger performance

The tau trigger efficiency is determined using a tag-and-probe analysis, which selects events in two
complementary signal regions [2], one enriched in 𝑍 → 𝜏𝜏 events and another one enriched in dileptonic
𝑡 𝑡¯ → 𝑊 (𝑒𝜈)𝑏𝑊 (𝜏𝜈)𝑏 decay events. The events for the efficiency measurement are selected using single
light lepton triggers as well as dedicated tag-and-probe trigger chains that use as “tag” a single light
lepton trigger other than the tau. The combination of the two signal regions allow for the measurement
of the tau trigger performance with sufficient statistics both at the low and high 𝑝 T regimes up to a tau
𝑝 T of about 200 GeV. The background events in these signal regions mostly come from events where a
jet has been misidentified as a hadronic 𝜏, and these are estimated using either data-driven methods or
simulated events. In the 𝑍 → 𝜏𝜏 enriched region the background from misidentified jets come from QCD
or 𝑊+jets processes, while in the 𝑡 𝑡¯ enriched region the background comes from QCD or semi-leptonic 𝑡 𝑡¯
decay events. Other minor background sources are due to diboson or 𝑍 → 𝑒𝑒 events where an electron is
mis-reconstructed as a hadronic 𝜏. Figure 39 shows the comparison between data and MC simulation of
the tau 𝑝 T in the one-prong case for the 𝑍 → 𝜏𝜏 and 𝑡 𝑡¯ selections obtained with the electron tag and tau
probe trigger.

×106 ×103
Events / 5 GeV

Events / 5 GeV

0.12
ATLAS Data ATLAS Data
Z → ττ 25 tt
Data 2022, s = 13.6 TeV Other background Data 2022, s = 13.6 TeV Other background
0.1
Z→τ τ SR Misidentified tau tt SR Misidentified tau
1 prong offline medium tau Uncertainty 1 prong offline medium tau Uncertainty
20
0.08

15
0.06

0.04 10

0.02 5

0 0
Data / Pred.

Data / Pred.

1.2 1.2
1.1 1.1
1 1
0.9 0.9
0.8 0.8
20 30 40 50 60 20 40 60 80 100 120
Offline tau p [GeV] Offline tau p [GeV]
T T

Figure 39: Distributions for data and MC simulation of the 𝑝 T of one-prong taus in (left) the 𝑍 → 𝜏𝜏 and (right) the
𝑡 𝑡¯ signal regions (SRs). The data set was recorded with the 𝑒 − 𝜏 tag-and-probe trigger. The bottom panels show the
ratio between data and the signal-plus-background prediction. Only statistical uncertainties are shown.

The tau trigger efficiency is calculated with respect to the reconstructed offline tau candidates and is
separated between one-prong and three-prong tau cases. Figure 40 shows the resulting efficiency as a
function of the offline tau 𝑝 T for one-prong and three-prong tau candidates using 2022 data recorded with a
combined electron tag and tau probe trigger.
In the one-prong selection, the resulting efficiency of the tau trigger reaches a plateau slightly below 95%
for tau candidates with 𝑝 T > 40 GeV, while it is about 90% for the three-prong case. The small differences
between the data and MC signal prediction observed in Figure 40 are used to derive scale factors to correct
other MC simulation samples used in data analyses.

52
 Trigger efficiency

Trigger efficiency
ATLAS Z→ττ + tt SRs ATLAS Z→ττ + tt SRs
1.5 Data 2022, s = 13.6 TeV 1.5 Data 2022, s = 13.6 TeV
1 prong offline medium tau 3 prong offline medium tau
HLT tau p >25 GeV, medium HLT tau p >25 GeV, medium
T T
1 1

0.5 Data - Bkg 0.5 Data - Bkg

Signal MC Signal MC

0 0
1.2 1.2
(Data-Bkg)/MC

(Data-Bkg)/MC
1.1 1.1
1 1
0.9 0.9
0.8 0.8
50 100 150 200 50 100 150 200
Offline tau p [GeV] Offline tau p [GeV]
T T

Figure 40: Tau trigger efficiencies as a function of offline tau 𝑝 T for (left) the one-prong and (right) three-prong
tau candidates. The legacy isolated L1 tau trigger with 𝐸 T > 12 GeV defined in Ref. [2] is used as the seed. The
measurement is performed using a combination of 𝑍 → 𝜏𝜏 and 𝑡 𝑡¯ selections [2] in events recorded with the 𝑒 − 𝜏
tag-and-probe trigger. The efficiency from the sum of the 𝑍 → 𝜏𝜏 and 𝑡 𝑡¯ signal processes estimated with MC (data
after subtracting the background) is shown with open (filled) markers. The bottom panels show the ratio of the
efficiencies. Only statistical uncertainties are shown.

6.4 Jets

6.4.1 Jet trigger reconstruction

A detailed description of jet triggers in Run 2 can be found in Ref. [2]. At the HLT, small- and large-radius
jets are reconstructed using the anti-𝑘 𝑡 algorithm [61] with a radius parameter of 𝑅 = 0.4 and 𝑅 = 1.0,
respectively. While jet triggers in Run 2 employed only calorimeter topo-clusters as input, Run-3 triggers
are able to rely on PFO thanks to the extended tracking capabilities offered by the HLT in Run 3.
Despite the large reductions in tracking time at the HLT, the CPU time usage of jet chains is still high due
to full scan tracking, as discussed in Section 5.1.4. For high-𝑝 T jet chains such as single- and multi-jet
chains, the large gap between L1 and HLT thresholds allows for an early reduction in event rate via the
application of a calorimeter-only jet selection (calorimeter preselection) before any kind of tracking is
executed. It is based on topo-cluster jets, described in Section 5.2. Minimal HLT jet 𝑝 T requirements
are adjusted to obtain the maximal CPU-time reduction which does not impact trigger efficiencies. The
preselection stage is found to reduce the output rates of HLT jet trigger chains by less than 2%. Table 4
shows the impact of different preselections on the event rates of various jet triggers.
Even though this strategy works well for baseline jet triggers, it is not optimal for triggers whose HLT jet
thresholds are very close to the L1 thresholds. One example of such triggers is the 𝑏-jet triggers, where a
more complex preselection step detailed in Sections 5.1.5 and 6.5.1 is included in the HLT reconstruction.
After the preselection step, fast full scan tracking is executed as detailed in Section 5.1.4 and the resulting
tracks are extrapolated to the calorimeter. The PFO formation starts by matching full scan tracks and
topo-clusters taking into account the extrapolated track position and topo-cluster location. The topo-clusters

53
Table 4: Impact of calorimeter preselections on the event rates of jet triggers in which tracking reconstruction is
performed. The percentage of events passing the preselection step is estimated with respect to the original L1 trigger
rate from a reprocessing of Run-2 Enhanced-bias data [33]. The HLT selection column shows the lowest jet trigger
threshold which this preselection is applied for.

L1 selection Preselection HLT selection Events passing preselection

1 jet, 𝑝 T > 100 GeV 1 jet, 𝑝 T > 225 GeV 1 jet, 𝑝 T > 275 GeV 23%
3 jets, 𝑝 T > 50 GeV 4 jets, 𝑝 T > 85 GeV 4 jets, 𝑝 T > 115 GeV 15%
4 jets, 𝑝 T > 15 GeV 5 jets, 𝑝 T > 50 GeV 5 jets, 𝑝 T > 70 GeV 7%
4 jets, 𝑝 T > 15 GeV 6 jets, 𝑝 T > 40 GeV 6 jets, 𝑝 T > 55 GeV 6%

which are not matched to any track are referred to as neutral PFOs. The matched topo-clusters have their
energy removed according to the expected calorimeter energy deposited by the matched track. The tracks
are considered as charged PFOs. Any topo-cluster surviving the energy subtraction procedure becomes
a neutral PFO. Charged PFOs not matched to the primary vertex are discarded, which is the dominant
means of pile-up suppression. Outside the geometrical acceptance of the tracker, only the calorimeter
information is available. Hence, in the forward region the topo-clusters alone are used as inputs to the
PFlow jet reconstruction.
After reconstruction, small-𝑅 jets are calibrated through a procedure similar to that used offline [46]. A
correction accounting for pile-up contamination is applied on an event-by-event basis to the jets. This
is followed by another correction, compensating for the energy response of the calorimeter. Finally, a
sequence of calibrations is applied to PFlow triggers to correct for the residual discrepancies between
reconstructed and simulated jets and accounts, for example, for energy differences resulting from the
different showering of quarks and gluons.
For Run 3, large-𝑅 jet triggers are also extended to use PFOs as inputs. In addition, the Constituent
Subtraction [62] and SoftKiller [63] algorithms are applied to neutral PFOs to subtract energy contamination
originating from pile-up interactions. After this step, Soft Drop grooming [64] with parameters 𝛽 = 1.0 and
𝑧 cut = 0.1 is applied to mitigate the contamination from initial state radiation and underlying event. Finally,
a calibration similar to small-𝑅 jets is used to correct for the energy response of the ATLAS calorimeter
and adjust the reconstructed mass of the large-𝑅 jet.

6.4.2 Jet trigger menu

Jet triggers are used for a wide set of measurements, ranging from precision physics measurements to
detector performance studies. The inclusive jet, di-jet and multi-jet production measurements rely on
the events selected by small-𝑅 single- and multi-jet triggers. Events selected by these triggers are also
employed for the calibrations of the jet energy scale and resolution, as well as for new physics searches
such as supersymmetry. Another important class of jet triggers is represented by large-𝑅 triggers, selecting
final states with boosted weak vector bosons (W/Z) or Higgs bosons. These triggers are generally employed
by analyses searching for heavy resonances predicted, for example, by theories of extra-dimensions and the
two-Higgs-doublet model.
The jet trigger chains are initiated by L1 algorithms selecting single jets, multi-jets, or, for very low-𝑝 T HLT
thresholds, a random trigger at L1. The primary, unprescaled jet chain applies a threshold of 420 GeV and
has a rate of 42 Hz (at 1.8 × 1034 cm−2 s−1 ). Small-𝑅 single jet chains with 𝑝 T thresholds below 420 GeV

54
 Per-event L1 efficiency

Per-event L1 efficiency
L1 jet |η|<3.2 (ref. HLT PFLow jet |η|<3.2) L1 jet 3.2 < |η| < 4.8 (ref. HLT PFLow jet 3.2 < |η| < 4.8)
ATLAS p >15 GeV ( p >0 GeV) p >30 GeV ( p >0 GeV)
ATLAS p >15 GeV ( p >0 GeV) p >20 GeV ( p >0 GeV)
1.2 Data 2022, s = 13.6 TeV T T
p >50 GeV ( p >45 GeV)
T T
T T
p >75 GeV ( p >60 GeV)
T T
1.2 Data 2022, s = 13.6 TeV T T
p >30 GeV ( p >15 GeV)
T T
T T
p >50 GeV ( p >45 GeV)
T T
Offline leading jet |η| < 2.8 p >100 Gev (p >110 GeV) Offline leading jet 3.6 < |η| < 4.5 p >75 GeV ( p >60 GeV)
T T T T

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
50 100 150 200 250 20 40 60 80 100 120 140 160 180 200
Leading offline particle flow jet p [GeV] Leading offline particle flow jet p [GeV]
T T

Figure 41: Efficiency of the legacy L1 single jet triggers in the (left) central and (right) forward regions. The efficiency
is computed using the bootstrap method with respect to events taken by an independent trigger, shown in brackets,
that is 100% efficient in the relevant region. Only statistical uncertainties are shown.
Trigger efficiency

Trigger efficiency
HLT PFlow jet |η| < 3.2 (ref. HLT PFlow jet |η| < 3.2) HLT PFlow jet 3.2 < |η| < 4.8 (ref. HLT PFlow jet 3.2 < |η| < 4.8)
ATLAS p >35 GeV ( p >0 GeV) p >110 GeV ( p >60 GeV)
ATLAS p >25 GeV ( p >0 GeV) p >45 GeV ( p >25 GeV)
1.2 Data 2022, s = 13.6 TeV T T
p >175 GeV ( p >110 GeV)
T T
T T
p >260 GeV ( p >175 GeV)
T T
1.2 Data 2022, s = 13.6 TeV T T
p >85 GeV ( p >45 GeV)
T T
T T
p >110 GeV ( p >60 GeV)
T T
Offline leading jet |η| < 2.8 p >360 GeV ( p >260 GeV) p >420 GeV ( p >360 GeV) Offline leading jet 3.6 < |η| < 4.5 p >175 GeV ( p >110 GeV) p >220 GeV ( p >175 GeV)
T T T T T T T T

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
100 200 300 400 500 50 100 150 200 250 300
Leading offline particle flow jet p [GeV] Leading offline particle flow jet p [GeV]
T T

Figure 42: Efficiency of single jet trigger selection by the HLT in the (left) central and (right) forward regions. Central
PFlow jet triggers exploit the particle flow reconstruction with calorimeter clusters and tracks, while forward PFlow
jet triggers rely on the topo-cluster-based reconstruction to trigger events outside the ID acceptance (|𝜂| < 2.5). The
efficiency is computed using the bootstrap method with respect to events taken by an independent trigger, shown in
brackets, that is 100% efficient in the relevant region. Only statistical uncertainties are shown.

are prescaled to provide a complete jet 𝑝 T spectrum, with a constant rate per chain ranging from 1 to
4 Hz; large-𝑅 single jets with thresholds between 110 GeV and 460 GeV are also collected using prescaled
triggers at a rate range between 2 and 3 Hz per chain. Trigger chains, which select events based on the
scalar sum of the transverse momentum of all jets (𝐻T ), contribute with a rate of 34 Hz. Topo-cluster-based
jet triggers are also available in Run 3 as a backup for PFlow jet triggers.

6.4.3 Jet trigger performance

The efficiency of jet triggers is primarily a function of the 𝑝 T of the leading jet in the event, and is measured
using the bootstrap method. In addition to the selection criteria specified in the following figures, the
offline jets are required to satisfy a timing cut of less than 12.5 ns to minimise the contamination from
out-of-time pile-up jets.

55
 1.4 1.4
Per-event L1 efficiency

Per-event HLT efficiency

L1 small-R jet |η|<3.2 (HLT PFlow small-R jet |η|<3.2) HLT PFlow small-R jet |η|<3.2 (HLT PFlow 1 jet p >110 GeV)
ATLAS ATLAS T

Data 2022, s = 13.6 TeV 3 L1 jets p > 50 GeV (HLT jet p > 45 GeV) s = 13.6 TeV 3 jets p > 200 GeV
1.2 th
T T 1.2 th
T Data 2022 Simulation
Offline N leading jet |η| < 2.8 4 L1 jets p > 15 GeV (HLT jet p > 25 GeV) Offline N leading jet |η| < 2.8 4 jets p > 115 GeV
T T
T Data 2022 Simulation
1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 20 40 60 80 100 120 140 160 180 200 0 50 100 150 200 250 300
th th
N Leading offline particle flow jet p [GeV] N Leading offline particle flow jet p [GeV]
T T

Figure 43: Multi-jet trigger efficiency as a function of the offline reconstructed 𝑁 th jet 𝑝 T (left) at L1 and (right) in
the HLT. The efficiency is computed using the bootstrap method with respect to events taken by an independent single
jet trigger, shown in brackets, that is 100% efficient in the relevant region. Only statistical uncertainties are shown.
Per-event trigger efficiency

Per-event trigger efficiency

1.4 ATLAS L1/HLT jet |η| < 3.2 (ref. HLT PFlow large-R jet |η| < 3.2)
small-R L1 jet, p > 100 GeV ( p > 110 GeV) 1.4 ATLAS HLT PFlow large-R jet |η| < 3.2
(ref. HLT PFlow large-R jet |η| < 3.2)
T T
p > 420 GeV, m > 35 GeV, small-R jet L1
Data 2022, s = 13.6 TeV large-R L1 jet, p > 111 GeV ( p > 110 GeV)
T T Data 2022, s = 13.6 TeV T
HLT, p > 360 GeV ( p > 260 GeV) ( p > 260 GeV)
1.2 Offline selection: T T
Offline selection: T
p > 420 GeV, m > 35 GeV, large-R jet L1
1 large-R jet, |η| < 2.2
HLT, p > 460 GeV, small-R jet L1 (p > 260 GeV)
T T
HLT, p > 460 GeV, large-R jet L1 (p > 260 GeV)
1.2 1 large-R jet, |η| < 2.2, p > 470 GeV
T
( p > 260 GeV)
T T T T

1
1
0.8
0.8
0.6
0.6
0.4

0.2 0.4

0 0.2
100 150 200 250 300 350 400 450 500 550 600 0 50 100 150 200 250 300
Leading offline large-R jet p [GeV] Highest mass offline large-R jet mass [GeV]
T

1.6
Per-event trigger efficiency
Per-event trigger efficiency

1.4 ATLAS HLT PFlow large-R jet |η| < 3.2

(ref. HLT PFlow large-R jet |η| < 3.2)
ATLAS L1 HT > 190 GeV
HLT PFlow small-R jet Data 2022 Simulation
Data 2022, s = 13.6 TeV p > 420 GeV, m > 35 GeV, small-R jet L1
T
1.4 s = 13.6 TeV p > 45 GeV, |η|<3.2
( p > 260 GeV)
1.2
T
Offline selection: T Offline selection:
p > 420 GeV, m > 35 GeV, large-R jet L1
1 large-R jet, |η| < 2.2, m > 70 GeV
T |η| < 2.8, p > 30 GeV HLT PFlow HT > 1000 GeV
( p > 260 GeV)
T
1.2 T
Data 2022 Simulation
(Isolated 1 electron p > 26 GeV)
1 T

1
0.8
0.8
0.6 0.6
0.4 0.4
0.2 0.2

0 0
300 350 400 450 500 550 0 200 400 600 800 1000 1200 1400 1600
Leading offline large-R jet p [GeV] HT [GeV]
T

Figure 44: L1 and HLT efficiencies are shown for single large-𝑅 jet triggers as a function of (top left) the leading
offline large-𝑅 jet 𝑝 T and (top right) highest mass offline large-𝑅 jet mass (m), (bottom left) large-𝑅 single jet triggers
with a mass cut as a function of leading offline large-𝑅 jet 𝑝 T and (bottom right) 𝐻T triggers as a function of offline
𝐻T . The efficiency of triggers is computed using the bootstrap method with respect to events taken by an independent
trigger, shown in brackets, that is 100% efficient in the relevant region. Only statistical uncertainties are shown.

56
The efficiencies of L1 and HLT single jet triggers as a function of offline reconstructed jet 𝑝 T are shown in
Figures 41 and 42, respectively. The reference trigger selections reported on the legend have a looser L1
and HLT selection than the probed triggers to ensure an unbiased reference dataset. It must be noted that
the jet-energy-scale correction is applied at the HLT but not at L1, meaning that the L1-jet thresholds are
effectively 50-100% higher when considered at the jet-energy scale of the HLT. A steep rise in efficiency
near the nominal thresholds reflects the compatibility of energy scales for jets reconstructed in the trigger
and offline.
The efficiency of a multi-jet trigger requiring 𝑁 jets depends mostly on the 𝑝 T of the 𝑁 th 𝑝 T -ranked jet. The
relevant efficiency measurements are shown in Figure 43 as a function of the offline reconstructed 𝑁 th jet
𝑝 T for the primary L1 and HLT multi-jet triggers. Lower efficiency at high 𝑝 T in the L1 four-jet trigger is
due to the merging of the close-by jets at L1, which should be improved in the upgraded L1Calo system.
Efficiencies of the large-𝑅 jets as a function of the leading offline jet 𝑝 T are shown in Figure 44 (top left),
for both HLT and L1 trigger chains. Figures 44 (top right) and (bottom left) present the efficiencies as a
function of the offline jet mass and 𝑝 T , respectively, for two main HLT chains with mass cuts. The L1 seed
is shown not to have an impact on the efficiency of these triggers due to the large gap between L1 and
HLT thresholds. The triggers shown are fully efficient at high jet 𝑝 T and mass values, and the steep rise in
efficiency near the nominal thresholds reflects a good energy resolution.
The 𝐻𝑇 trigger efficiency as a function of offline 𝐻T is presented in Figure 44 (bottom right) for both L1
and HLT triggers, which also rises sharply. The observed data/MC differences will be corrected later-on in
physics analyses by dedicated scale factors applied to MC simulation samples.

6.4.4 Trigger level analysis with jets

HLT topo-cluster-based and PFlow jets with transverse momentum larger than 20 GeV are saved to the
TLA stream described in Section 4. To remain within the CPU constraints of the HLT farm, full scan
tracking and PFO reconstruction are executed only if the event passes the calorimeter preselection step.
In 2022, jet chains populating the TLA stream were seeded by two L1Calo legacy triggers, L1 jet
𝑝 T > 100 GeV (which reaches 50% efficiency at around 160 GeV) and 𝐻T > 190 GeV, with a total rate
of approximately 4 kHz. This allows for the recording of orders of magnitude more events containing
lower-𝑝 T jets with respect to standard ATLAS jet triggers, extending the sensitivity of hadronic searches to
resonance masses as low as 400 GeV [36].
As shown in Figure 45, the response of trigger jets with the default calibration is already within 2% of
offline jets across the momentum range of interest to TLA studies. For Run 3, a custom residual energy
scale calibration is to be derived and applied at the analysis level in order to further improve the response
and resolution of TLA jets. In order to derive such a calibration, an extensive set of jet variables, beyond the
jet four-momentum, is saved in the TLA stream. This includes the energy fraction deposited in the EM and
hadronic endcap calorimeters, the momentum density of soft radiation in the event, the number of primary
vertices reconstructed at the HLT, the jet active area [46], and the number of constituents encompassing
90% of the jet transverse energy.

57
 ATLAS
1.25

Entries
HLT/Offline pT response
Single jet triggers:
1.20
pT > 15 GeV
pT > 25 GeV 109
1.15
pT > 35 GeV
1.10 1 histogram
quantile
1.05
108
1.00

0.95

0.90
p
Data s = 13.6 TeV, October 2022 107
0.85 HLT and offline anti-kt R = 0.4 PFlow jets
| | < 2.4
0.80
25 30 40 50 60 70 80 90 100 200
Offline R-matched jet pT [GeV]

Figure 45: The response curve for HLT-level PFlow jets in the momentum range of interest for Run-3 TLA. The
online-to-offline response is evaluated at the final energy scale in the calibration sequence of the jets. HLT-level jets
are calibrated with the default sequence used at the HLT for the 2022 data taking. Offline jets are corrected with
the calibration available during 2022 data taking. Events are required to pass the single jet 𝑝 T > 15 GeV (circle
markers), 𝑝 T > 25 GeV (triangle markers) or 𝑝 T > 35 GeV (square markers) random-seeded triggers depending on
the 𝑝 T of the leading reference jet. For each event, only HLT jets matched (Δ𝑅 < 0.3) to the leading and sub-leading
offline jets reconstructed within |𝜂| < 2.4 are considered. The overlaid heat map refers to the number of events in the
analyzed sample for each configuration of the reference jet momentum and response.

6.4.5 Jet triggers for heavy-ion collisions

Jets are direct probes of the quark–gluon plasma produced in HI collisions, and studying how they are
modified in such collisions compared to 𝑝 𝑝 collisions provides insight into their interactions with this
QCD medium. The main challenge for triggering on jets in HI collisions is the presence of a large amount
of UE. The HLT jet algorithm uses the HI UE subtraction procedure described in Section 4.2 and selects
events containing jets with transverse energies exceeding a threshold ranging from 60 to 85 GeV. Jets are
reconstructed at the HLT across the entire calorimeter using the anti-𝑘 𝑡 algorithm with distance parameter
𝑅 = 0.4, from projective towers of size Δ𝜂 × Δ𝜙 = 0.1 × 0.1 formed from the summation of calorimeter
cell energies with UE subtraction applied.

6.5 Jets containing 𝒃-hadrons

6.5.1 𝒃-jet trigger reconstruction and selection

Triggers based on jets containing 𝑏-hadrons, so-called 𝑏-jet triggers, are designed to allow for highly
efficient recording of fully hadronic events with predominantly heavy flavour content jets. The detailed
description of the 𝑏-jet triggers in Run 2 can be found in Ref. [65], the changes for Run 3 are discussed
here. The basic inputs to 𝑏-tagging are reconstructed jets, reconstructed tracks and the position of the
primary vertex. The jet reconstruction is described in Section 6.4. The primary vertex finding and 𝑏-jet
track reconstruction are discussed in Sections 5.1.4 and 5.1.5, respectively.

58
The main 𝑏-jet identification steps are as follows:
• A fast 𝑏-tagging algorithm is run on super-RoIs and uses the tracks inside them to provide the first
𝑏-jet preselection. To reduce the rate of the full scan tracking for 𝑏-jet triggers, the 𝑏-jet preselection
stage is run with lower track 𝑝 T thresholds and with a super-RoI composed of wider regions around
jets than were used for the vertex tracking in Run 2 [41, 65]. The details of the fast 𝑏-tagging
algorithm used at this stage are provided in Ref. [53].
• The final 𝑏-tagging algorithm, detailed below, uses precision tracking and primary vertex information.
The 𝑏-jet identification relies on the properties of 𝑏-hadrons: long lifetime (about 1.5 ps), hard fragmentation,
a relatively large mass of about 5 GeV, and a displaced (secondary) vertex (SV) formed a few millimeters
away from the primary vertex. Tracks associated with 𝑏-jets are characterised by larger transverse (𝑑0 ) and
longitudinal (𝑧 0 ) impact parameters. In addition, 𝑏-hadrons decay semileptonically, either promptly, or via
a subsequent 𝑐-hadron decay, to electrons or muons. The branching ratio of these semileptonic decays is
about 20% each and results in the presence of a low-𝑝 T lepton close to the 𝑏-jet.
The flavour tagging identification is done in two steps [48]. Low-level algorithms reconstruct characteristic
features of the heavy flavour jets based on track properties, such as impact parameters, or combine those
tracks to reconstruct the SV. The outputs of these low-level algorithms are then combined into high-level
algorithms, usually using some multivariate technique.
Low-level taggers used at the HLT are the IP2D algorithm, which utilises the signed transverse impact
parameter significance of tracks, and the IP3D algorithm, which uses, in addition, the longitudinal impact
parameter significance [48].
The secondary vertex algorithm, SV1, uses tracks associated with jets (after rejecting those compatible
with 𝐾𝑠 or Λ, photon conversions or interactions with detector material) to assign decay products from 𝑏-
or 𝑐-hadrons to a single common SV. Several discriminating variables associated with the SV are then used
as inputs to the high-level tagger. Finally, the JetFitter algorithm exploits the topology of weak 𝑏- and
subsequent 𝑐-hadron decays inside a jet to reconstruct the full decay chain, recreating the approximate
𝑏-hadron path from the primary vertex via bottom and charm vertices.
Another low-level algorithm is the Deep Impact Parameter Sets (DIPS) [66], based on the Deep Sets
architecture. It uses impact parameter information, accounting for correlations between the track features,
among other variables. DIPS considers tracks in the jet as an unordered and variable-sized set, which is
physically better motivated than the algorithm based on recurrent neural network [65] used in Run 2, given
that the 𝑏-hadron decay products do not exhibit any intrinsic sequential ordering. The performance of the
DIPS tagger is investigated using 𝑡 𝑡¯ MC events and 2022 data collected with a calibration trigger with one
electron, one muon and two jets. The final DIPS discriminant distribution is shown in Figure 46 (left). In
the case of SM 𝐻𝐻 → 𝑏 𝑏𝑏 ¯ 𝑏,
¯ a key signature relying on 𝑏-jet triggers, the 𝑏-jet preselection step lowers
the input rate to the remaining HLT by a factor of five at the cost of reducing the overall signal efficiency by
roughly 2%. More details about DIPS, its training and usage in 𝑏-jet triggers are given in Ref. [53].
The newly developed algorithm based on deep feed-forward Neural Networks, the so-called DL1 series [48],
replaced boosted-decision-tree based taggers utilised in Run 2 [65]. The particular instance of the algorithm
used in 2022 is called DL1d and takes as inputs the kinematic variables (𝑝 T and 𝜂) of the jet as well as final
discriminants from lower-level taggers (IPxD, SV1, JetFitter and DIPS).
From 2023 the 𝑏-jet trigger relies on a novel algorithm GN1, which is based on Graph Neural Networks
(GNNs) [67]. Unlike the DIPS and DL1d, the GN1 utilises a single neural network taking the tracks and

59
jet information directly as inputs and is thus independent of other flavour tagging algorithms. The GN1
combines a GNN with the two auxiliary training objectives: the grouping of tracks originating from a
common vertex and the prediction of the underlying physics process from which each track originated.
This approach leads to a better understanding of the jet’s internal structure and, thus, a better algorithm
performance.
The expected trigger rate as a function of the 𝑏-tagging efficiency using the DL1d and GN1 algorithms is
shown in Figure 46 (right). The expected light jet rejection as a function of the 𝑏-tagging efficiency for
various 𝑏-taggers as well as their operating points are shown in Figure 47. The GN1 tagger performance
exceeds that of DL1d, which was the main tagger in 2022.

40

High Level Trigger Rate [Hz]

ATLAS
Fraction of jets

35
p
ATLAS Run 3 Enhanced Bias Data, s = 13.6 TeV
Data 2022
Predicted Luminosity = 2.0 × 1034cm 2s 1
s =13.6 TeV, October 2022 L1: 1 jet pT > 45 GeV, | | <2.1
1 30
jet p ≥ 20 GeV, jet |η| ≤ 2.5 2 jets pT > 15 GeV, | | < 2.5
T MC t t HLT: 4 jets pT > 80 GeV, 55 GeV, 28 GeV,
25 20 GeV, | | < 2.4, Ntag 3
10−1
20

10−2 15

10
10−3
5 GN1
DL1d

0
−4 −2
Ratio to DL1d

0 2 4 6 8 10
Data/MC

fast DIPS Discriminant

1.4
1.2 1.0
1
0.8
0.6
0.5
−4 −2 0 2 4 6 8 10 0.650 0.675 0.700 0.725 0.750 0.775 0.800 0.825
fast DIPS Discriminant b-Tagging Efficiency

Figure 46: (left) Distribution of the 𝑏-jet discriminant score for the DIPS algorithm shown in data from selected
runs and 𝑡 𝑡¯ MC events collected with a calibration trigger with one electron, one muon and two jets. Only statistical
uncertainties are shown. (right) Expected trigger rates as a function of 𝑏-tagging efficiency using the DL1d and
GN1 algorithms, while requiring at least four HLT PFlow jets, three of which are required to be above the 𝑏-tagging
threshold. Rates are estimated with Run-3 Enhanced-Bias data, and the 𝑏-jet efficiencies are estimated for 𝑡 𝑡¯ samples
using PFlow jet. The relative errors on trigger rates are between 8(7)% and 18(20)% for DL1d (GN1).

6.5.2 𝒃-jet trigger menu

𝑏-jet triggers are crucial for several precision measurements and searches for new particles such as
resonant and non-resonant 𝐻𝐻 → 𝑏 𝑏𝑏 ¯ 𝑏¯ [68, 69], VBF 𝐻 → 𝑏 𝑏¯ [70, 71], 𝑡 𝑡¯𝐻 → 𝑡 𝑡¯𝑏 𝑏¯ with both top
quarks decaying hadronically [72], 𝑏𝜙 → 𝑏𝑏 𝑏¯ [73], third generation squarks [74], low mass di-𝑏-jet
resonances [75], etc.
The 𝑏-jet trigger menu consists of the following physics triggers: single-𝑏-jet triggers, multi-𝑏-jet triggers
and specialised 𝑏-jet triggers designed for specific physics processes which often occur in combination
with other signatures. Single-𝑏-jet triggers and multi-𝑏-jet triggers are seeded from dedicated single-jet or
multi-jet L1 objects. At the HLT various combinations of the 𝑝 𝑇 thresholds and 𝑏-tagging operating points
are employed. In multi-𝑏-jet triggers, between one and four jets are tagged, while other jets are not tagged.
An asymmetric chain requiring four jets with 𝑝 T greater than 80 GeV, 55 GeV, 28 GeV, and 20 GeV of
which two are 𝑏-tagged (with efficiency of 77%) has a rate around 150 Hz, the lowest unprescaled single

60
 Light jets rejection
DL1d
105
ATLAS Simulation GN1
√s =13.6 TeV, tt̄ sample fastDIPS | PFlow jets
fastDIPS | EMTopo jets
60%
104
70%

77%
3
10
85%

102

101

100
Ratio

2
1.5
1
0.5

0.5 0.6 0.7 0.8 0.9 1.0

b-jet efficiency

Figure 47: The light jet rejection as a function of the 𝑏-tagging efficiency (ROC curve) for the DIPS algorithm used
on topo-cluster jets for preselection stage (dash-dotted line), the DIPS algorithm used on PFlow jets for TLA triggers
(dashed line), the DL1d algorithm used as the final tagger in 2022 (solid line) and the GN1 algorithm used as the
final tagger from 2023 (dotted line) obtained on a 𝑡 𝑡¯ MC sample. Statistical uncertainties for each ROC curve are
represented with shaded regions around the curves. The vertical dashed lines represent the operating points for
𝑏-tagging used at the HLT. The bottom panel displays the ratio of all the ROC curves with respect to the DL1d
performance.

𝑏-jet trigger (𝑝 T > 225 GeV) around 50 Hz and one jet of 𝑝 T > 150 GeV plus two 𝑏-jets with 𝑝 T > 55 GeV
and 𝑏-tagging efficiency of 70% around 10 Hz at luminosity of 1.8 × 1034 cm−2 s−1 .
Flavour-tagging calibration triggers [65] include a single lepton (electron or muon) with one or two jets
with 𝑝 T > 20 GeV, electron-plus-muon with one or two jets with 𝑝 T > 20 GeV, and muon-jet matched
triggers, that are also used for physics studies in HI collisions. These latter triggers were seeded from
either a single-muon or a muon-plus-jet trigger item at L1 in 2022. New chains seeded from L1 objects
which use topological information, such as Δ𝑅, are planned to be introduced later in Run 3. At the HLT, a
muon is required to be matched to a jet by fulfilling a requirement of Δ𝑅(𝜇, jet) < 0.5, with an additional
requirement for 𝑝 𝑝 collisions Δ𝑧(𝜇, jet) < 2 mm, where the 𝑧-position of the jet is taken to be the primary
vertex 𝑧-position.

6.5.3 Trigger level analysis with 𝒃-jets

The Run-3 TLA HLT infrastructure allows for the full outcome of flavour tagging algorithms to be written
out in events triggered by TLA chains targeting 𝑏-jet signatures. In every TLA event, each HLT-level jet
(for which flavour tagging is performed) is linked to the corresponding 𝑏-tagging object in which the values
of low and high-level taggers are stored. Furthermore, the value of the DIPS tagger is added to every

61
PFlow jet saved to the TLA stream, regardless of whether the triggering chain is configured to perform
dedicated flavour tagging or not. This setup allows for the recording of orders of magnitude more events
containing lower-𝑝 T 𝑏-jets with respect to standard 𝑏-jet triggers, extending the sensitivity of low mass
di-𝑏-jet resonance searches to resonance masses down to 100 GeV.

6.6 Missing transverse momentum

6.6.1 𝑬 Tmiss trigger reconstruction

The missing transverse momentum (𝐸 Tmiss ) trigger is used to identify events with particles that do not
interact with the ID or calorimeters because of the absence of strong or electromagnetic interactions and
that have lifetimes large enough to leave the detector without decaying into detectable particles. Examples
of such processes include SM Z boson decays to neutrinos [76], as well as searches for beyond SM decays
involving dark matter [77] or supersymmetric [78] particles.
The 𝐸 Tmiss can be computed from the magnitude of the vector sum of constituents, c:

∑︁
𝐸 Tmiss = 𝑝®𝑇 ,𝑐 .
𝑐

In addition to the 𝐸 Tmiss , the scalar sum of the constituents is also computed. The main variations in how
𝐸 Tmiss is computed in the trigger involve details of which constituents to utilise in the sum in the equation
above.
The following HLT 𝐸 Tmiss algorithms are available in Run 3:
• cell: The 𝐸 Tmiss is formed from a sum over all calorimeter cells passing the selection 𝐸 𝑖 > −5𝜎𝑖 ,
|𝐸 𝑖 | > 2𝜎𝑖 , where 𝜎𝑖 is the estimated noise in that cell. This is to protect against spurious large
negative energy signals, discussed in Section 5.2. The cell energy magnitude must be greater than
2𝜎𝑖 to reduce the effect of noise from electronics and pile-up.
• tcpufit: The 𝐸 Tmiss is formed using topo-clusters as inputs, which are combined in 𝜂 − 𝜙 patches
with dimensions of approximately 0.7 × 0.7 (roughly the size of an 𝑅 = 0.5 jet). The energy
contribution from pile-up to these patches is estimated by a fit over them, requiring that pile-up events
have no true 𝐸 Tmiss and are approximately evenly distributed over the calorimeter [79]. The estimated
pile-up contribution to each patch is subtracted, and the remaining patch transverse momenta are
summed to obtain the 𝐸 Tmiss . The full algorithm is described in Appendix A of Ref. [79].
• trkmht: The 𝐸 Tmiss is formed over all jets passing a Jet Vertex Tagger (JVT) selection [80] where
applicable. The vector sum of the 𝑝 T of tracks from the primary vertex that are not associated with a
passing jet defines a track soft term which is added into the 𝐸 Tmiss calculation.
• pfopufit: The pfopufit algorithm uses the same techniques as tcpufit, but the input topo-
clusters are modified to use PFO tracks [50]. As well as the improved momentum resolution of the
PFOs, the vertex information provided by the charged PFOs is used to improve the categorisation of
deposits into hard-scatter and pile-up.

62
 • mhtpufit: The 𝐸 Tmiss is formed from a sum over jets passing a JVT selection [80]. A similar
technique to tcpufit is used to correct these jets for the impact of pile-up. Two variants are
used: mhtpufit_pf uses PFOs and jets formed from them to estimate the pile-up contributions,
mhtpufit_em uses jets formed from EM scale topo-clusters and the hadronic scale topo-clusters to
estimate the pile-up contributions.
• pfsum: The 𝐸 Tmiss is formed from a sum over the PFOs in the event. Two variants exist: in
pfsum_vssk the PFOs have their energies reduced according to their Voronoi areas [81], whereas in
pfsum_cssk the constituent subtraction method [62] is used instead. In both cases, the SoftKiller
algorithm [63] is used to remove PFOs from low-energy areas of the calorimeter.
In 2022, the primary L1 item for 𝐸 Tmiss was the L1Calo legacy trigger with 𝐸 Tmiss > 50 GeV [21, 79]. Before
any kind of tracking is executed, an early reduction of rate at HLT is achieved by a calorimeter preselection
requirement that the cell-based 𝐸 Tmiss is greater than 65 GeV. The choice of default Run-3 algorithm is
made based on the performance in terms of background rejection versus signal efficiency and is discussed
in the following paragraphs.

6.6.2 𝑬 Tmiss trigger performance and menu

Since muons are treated as invisible by the 𝐸 Tmiss trigger algorithms described above, events with a
boosted 𝑍 → 𝜇𝜇 can be used for the 𝐸 Tmiss performance studies with 𝑝 𝑇 (𝜇𝜇) serving as a proxy for
𝐸 Tmiss . A background acceptance vs. trigger efficiency curve is shown in Figure 48 (left) comparing the
performance of a selection of 𝐸 Tmiss algorithms considered for Run 3. The efficiencies are calculated as the
fraction of events passing a given 𝐸 Tmiss requirement for data events with an actual number of interactions
per bunch crossing of at least 38, and a 𝑍 → 𝜇𝜇 event selection passing a single muon trigger, with
76 < 𝑀 (𝜇𝜇) < 106 GeV and 𝑝 𝑇 (𝜇𝜇) > 175 GeV. The background rate is obtained from an offline trigger
reprocessing of the 2022 data collected with zero-bias triggers described in section 7.2. The new Run-3
default 𝐸 Tmiss trigger is based on pfopufit algorithm, which shows the best ability to retain signal, whilst
rejecting background, of all the algorithms listed in Section 6.6.1. The Run-2 default tcpufit trigger is
maintained as backup and for analyses where the primary vertex may not match the online primary vertex.
These triggers had unprescaled HLT thresholds of 90 GeV and 115 GeV in 2022 and rates of about 70 Hz
and 30 Hz at luminosity of 1.8 × 1034 cm−2 s−1 , respectively. Requiring the presence of additional objects
in the event, allows for decreased thresholds for 𝐸 Tmiss triggers, as discussed in Section 6.9.
A signal efficiency curve is shown in Figure 48 (right) comparing the performance of the primary L1
𝐸 Tmiss trigger alone to the default HLT 𝐸 Tmiss trigger chains used in Run 3 and Run 2. The efficiencies are
calculated as the fraction of events passing a given 𝑝 𝑇 (𝜇𝜇) requirement for data events with a 𝑍 → 𝜇𝜇
event selection passing a single muon trigger, with 76 < 𝑀 (𝜇𝜇) < 106 GeV. The 2018 L1 efficiency
matches that of 2022 within a few percent, so it is not shown. The improved efficiency seen for the
Run-3 pfopufit-based chain compared to the Run-2 chain reflects the pfopufit algorithm’s enhanced
momentum resolution and hard-scatter/pile-up categorisation.

6.7 𝑩-physics and light states

The trigger selection of events for 𝐵-physics analyses is primarily based on the identification of 𝑏-hadrons
through decays with a muon pair in the final state. However, one or more muons or electrons could also
be present in specific selections. Examples are decays with charmonium, 𝐵 → (𝐽/𝜓 or 𝜓 ′ ) 𝑋 → 𝜇𝜇𝑋,

63
 1
Background acceptance

Trigger efficiency
Data 2022, s = 13.6 TeV
ATLAS 1.2 miss
L1 ET > 50 GeV
miss miss
ET,cell> 65 GeV, ET,pfopufit> 90 GeV
Data 2022, s = 13.6 TeV
Data 2018, s=13 TeV
Z → µµ candidates miss miss
miss ET,cell > 65 GeV, ET,tcpufit > 110 GeV
L1 ET > 50 GeV
µ > 38, p (µµ) > 175 GeV 1
−1
T

10 cell
tcpufit
mhtpufit_em
0.8
mhtpufit_pf
pfopufit
0.6
10−2
0.4
ATLAS
0.2
10−3 Z → µµ candidates

0.75 0.8 0.85 0.9 0.95 1 0 50 100 150 200 250 300
Trigger efficiency pT(µµ) [GeV]

Figure 48: (left) Background acceptance vs. trigger efficiency curves for a selection of Run-3 𝐸 Tmiss trigger algorithms.
(right) Trigger efficiency as a function of 𝑝 𝑇 (𝜇𝜇) for the primary L1 𝐸 Tmiss trigger and the full trigger chain, with
specific thresholds listed. The trigger rates for the HLT chains shown are approximately the same rate (to within
10%). The data are presented in comparison to 2018 efficiency. Only statistical uncertainties are shown.

rare decays 𝐵0(𝑠) → 𝜇𝜇, and 𝐵 → 𝜇𝜇𝑋 decays. Decays of prompt charmonium and bottomonium are also
identified through their di-muon decays, and are therefore similar to 𝐵-meson decays, apart from the lack of
measurable displacement from the 𝑝 𝑝 interaction point. As the BLS topologies are significantly different
from the majority of the ATLAS physics triggers, they are recorded to a separate BLS stream, as discussed
in Section 4.1. In 2022 this stream collected data with an average rate of approximately 240 Hz.
Entries / 10 MeV

ATLAS √
Data 2022, s = 13.6 TeV, 30.7 fb−1
low-M ffi J= B Υ(nS) pmin min
T ( —1 ) p T ( —2 )

106 11 GeV 6 GeV

105

104

2 4 6 8 10 12
M(—+, —−) [GeV]

Figure 49: Invariant mass distribution of offline-selected di-muon candidates passing the lowest thresholds of di-muon
𝐵-physics triggers. Triggers targeting different invariant mass ranges are illustrated with different colours, and the
differing thresholds are shown with different shadings. No accounting for overlaps between triggers is made, and the
distributions are shown overlaid, and not stacked.

As 𝐵-mesons are light, the resulting muon momenta are rather soft. To control the input rate to the HLT,
most of the 𝐵-physics triggers require two muons at L1. Their rate is substantially reduced compared

64
to single-muon L1 triggers. Depending on the mass of the resonance this can result in the muons being
within a single RoI or in separate RoIs. At the HLT, muons are reconstructed using the same algorithms as
described in Section 5.3 with the additional requirement that the combined muons have opposite charges
and form a good vertex (where a fit is performed using the ID track parameters) within a certain invariant
mass window. For example, the mass range of the bJpsimumu is between 2.5 and 4.3 GeV while bUpsi is
between 8 and 12 GeV. Examples of these mass spectra can be seen in Figure 49.
Di-muon trigger rate restrictions at L1 define the lowest muon transverse momentum thresholds for primary
𝐵-physics triggers. HLT triggers based on two L1 muons passing a 3 GeV 𝑝 T threshold need to be prescaled
for most data-taking luminosities. The prescales are adjusted to maximise the recorded number of low-𝑝 T
di-muon events while remaining within operational constraints of the ATLAS TDAQ system as discussed
in Section 4.1. Higher threshold HLT triggers seeded from two muons passing 5 GeV or 8 GeV thresholds
at L1 are unprescaled when the L1 and HLT bandwidths allow it.
Additional primary and supporting triggers are also implemented. Triggers requiring three muons at
L1 help to maintain the lowest muon 𝑝 T thresholds for certain event signatures with a likely presence
of a third muon. For semileptonic decays, such as 𝐵0 → 𝜇𝜇𝐾 ∗0 (→ 𝐾 + 𝜋 − ), searches for additional ID
tracks and a combined vertex fit are performed, assuming a few exclusive decay hypotheses. For the
start of Run 3 a so-called bBmuX selection is implemented to perform the partial reconstruction of the
B-hadron decay final state 𝐷 ∗+ 𝜇 − 𝑋 with the 𝐷 ∗+ reconstructed through the cascade hadronic decay chain
𝐷 ∗+ → 𝜋 + 𝐷 0 (→ 𝐾 − 𝜋 + ).
Entries / 6 MeV

Entries / 6 MeV

ATLAS 2500 ATLAS

7000 Data 2022, s = 13.6 TeV, 6.8 fb-1 Data 2022, s = 13.6 TeV, 6.8 fb-1
Full BphysDelayed Stream Full BphysDelayed Stream
6000 2000
pT (µ 1) > 11, pT (µ 2) > 6 GeV pT (µ 1) > 11, pT (µ 2) > 6 GeV
+ Dimuon selections + Dimuon selections
5000
p (µ ) > 11, p (µ ) > 6 GeV 1500 p (µ ) > 11, p (µ ) > 6 GeV
T 1 T 2 T 1 T 2
+ other Bd selections + other Bs selections
4000
pT (µ 1), pT (µ 2) > 10 GeV pT (µ 1), pT (µ 2) > 10 GeV

3000 + Dimuon selections 1000 + Dimuon selections

2000
500
1000

0 0
5200 5300 5400 5500 5600 5700 5200 5300 5400 5500 5600 5700
M(µ µ K π) [MeV] M(µ µ K K) [MeV]

Figure 50: To validate the selection of the bBmumux the mass spectra of the selected (left) 𝐵 𝑑 → 𝐽/𝜓𝐾 ∗ and (right)
𝐵𝑠 → 𝐽/𝜓𝜙 events from the BLS Delayed stream are plotted along with the spectra for a variety of bJpsimumu and
bBmumux trigger decisions. Only statistical uncertainties are shown.

To evaluate the 𝐵-meson selection at the HLT, supporting triggers without vertex or charge requirements
are used. The mass spectra of various bBmumux trigger decisions are plotted and fitted using events from
the BphysDelayed stream. The consistency of the mass peaks in Figure 50 with PDG averages and previous
measurements [82] demonstrate the validity of the HLT reconstruction and selection.

6.8 Minimum-bias and forward signatures

To study diffraction, soft-QCD and similar topics, highly efficient triggers are necessary to select 𝑝 𝑝
inelastic, diffractive and scattering interactions with the least possible trigger bias. These data also serve as

65
references for heavy-ion measurements. The minimum-bias triggers discussed below are enabled only
during dedicated low-𝜇 runs.

6.8.1 Minimum-bias triggers

Depending on the number of collisions per bunch crossing, 𝜇, the minimum-bias triggers play various roles.
At very low values of 𝜇 (𝜇 ≪ 1) the triggers require one or two signals on any side of the Minimum Bias
Trigger Scintilator (MBTS) detector to select actual collision events and to ensure that the data sample is
not dominated by empty events with no 𝑝 𝑝 interactions. The MBTS detector was replaced for Run 3, as
discussed in Ref. [3]. It is a two-armed large-area plastic scintillator with a very high light yield covering
2.0 < |𝜂| < 4.0. On each side, the scintillator is divided into two concentric rings, each consisting of
8 octants covering the full azimuth. Altogether, the MBTS deliver 32 signals to the CTP, which can be
used independently or combined. Further event selection is possible at the HLT by requiring matching
timing between MBTS signals or the reconstruction of tracks in the ID. As a side benefit, reconstructing the
MBTS signals at the HLT allows for the monitoring of the timing and energies of individual channels.
At 𝜇 values close to unity the need for MBTS as a source for the hardware trigger is less relevant and the
minimum-bias sample can be collected by a random trigger at L1 followed by the requirement of a track
reconstructed at the HLT.
At 𝜇 ∼ 1 − 3, every bunch crossing contains an inelastic collision and the L1 random trigger is sufficient for
minimum-bias triggering. In these conditions, the focus shifts to collecting events with high multiplicity
tracks (HMT) or events with a high momentum track for analyses like Bose-Einstein correlations or
azimuthal correlations similar to what is done in HI physics [83]. The possibility of using a trigger with
the total (transverse) energy deposit in the calorimeter is also planned.
At the start of 2022 data taking no track selection was applied at the HLT. This allows for the comparison
in performance of the minimum-bias tracking selection in offline versus the trigger. In Run 3 the track
reconstruction for the minimum-bias trigger is based on offline algorithms. The minimum-bias online
tracks are required to satisfy the minimum-bias track selection: 𝑝 T > 0.1 GeV and |𝜂| < 2.5. In addition to
the online selection above, the offline tracks are required to satisfy the standard minimum-bias analysis
selection [84]: |𝑑0 | < 1 mm, |𝑧 0 sin 𝜃| < 1.5 mm, a measurement in the IBL if it is expected, at least
one pixel measurement and the number of measurements in the SCT above 2, 4, and 6 for tracks of
𝑝 T above 0.1, 0.3, and 0.4 GeV, respectively. Furthermore, the online and offline conditions are not
exactly identical, e.g. the map of dead pixel and SCT modules is known only after data have been
taken. A comparison of transverse momentum, 𝑝 T , and 𝜂 for online tracks and offline tracks undergoing
corresponding minimum-bias selections is shown in Figure 51. Despite the looser online track selection,
the purity of minimum-bias triggers is quite high.
Triggers with track reconstruction include a preselection step based on the count of pixel clusters and SCT
space-points (coinciding pairs of hits on both sides of an SCT module). In particular, the single-track
trigger preselection requires two-pixel clusters and three SCT space points. This preselection step reduces
the input rate prior to execution of the tracking algorithms without any efficiency loss. Random-seeded
triggers with only the preselection step and no further HLT requirements are used to collect unbiased
samples for performance studies. Figure 52 shows the efficiency of HLT tracking selection and the L1
MBTS trigger as a function of the number of tracks reconstructed offline passing minimum-bias selection,
as obtained with a data set collected with the random-seeded trigger at L1. The HLT tracking selection is
nearly fully efficient if at least two tracks in the event are present (a required minimum multiplicity for the

66
 4 1

Tracks per event / 0.1

Tracks per event / 0.05 GeV

3.5 ATLAS 0.9 ATLAS

Data 2022, s=13.6 TeV 0.8 Data 2022, s=13.6 TeV
3 Low-µ run Low-µ run
0.7
2.5 online
0.6
offline
2 0.5

1.5 0.4
0.3
1 online
0.2
0.5 offline
0.1
0 0
0 0.2 0.4 0.6 0.8 1 −2 −1 0 1 2
p [GeV]
T
η

Figure 51: Comparison of (left) 𝑝 T spectra and (right) 𝜂 of tracks reconstructed offline and passing minimum-bias
quality selection and tracks reconstructed by the HLT online during low pile-up run in 2022.

1.05 1.05
Trigger efficiency

Trigger efficiency

1 1

0.95 ATLAS 0.95

Data 2022, s=13.6 TeV Single track
0.9 0.9 MBTS 1
Low-µ run
MBTS 2
0.85 0.85

0.8 0.8 ATLAS

Single track Data 2022, s=13.6 TeV
0.75 0.75
MBTS 1 Low-µ run
0.7 MBTS 2 0.7

2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20
Ntrk (p > 0.1 GeV) Ntrk (p > 0.5 GeV)
T T

Figure 52: (left) Efficiency of triggering for events with at least two tracks of 𝑝 T > 0.1 GeV and passing minimum-bias
quality track criteria (see text) by single track trigger and MBTS triggers requiring single (MBTS 1) or two hits
(MBTS 2) in the scintillator. (right) Efficiency of triggering for events with at least one track of 𝑝 T > 0.5 GeV passing
minimum-bias track quality criteria by the same triggers as shown on the left panel. The efficiency is measured as a
function of the number of offline tracks, 𝑁trk , with respect to the fully efficient trigger requiring only two clusters in
the pixels detector and three space-points in SCT seeded from random bunch crossing at L1. Efficiencies of the
MBTS single and two hit triggers are measured with respect to the single track trigger. Only statistical uncertainties
are shown. These triggers were active during low pile-up runs in 2022.

67
 Trigger efficiency
1

0.9 ATLAS
Data 2022, s=13.6 TeV
0.8 Low-µ run

40 tracks

0.7 40 tracks PU SUPP

60 tracks

60 tracks PU SUPP

0.6 80 tracks

80 tracks PU SUPP

0.5
40 60 80 100 120 140
Ntrk

Figure 53: Efficiencies of high multiplicity triggers targeting selection of events with 40, 60, and 80 tracks as a
function of number of tracks passing minimum-bias selection quality, 𝑁trk . For each requirement the performance of
the pile-up suppression (PU SUPP) variant of the trigger, designed to be robust in higher values of pile-up, is shown.
Only statistical uncertainties are shown. These triggers were active during low pile-up runs in 2022.

𝑝 T > 0.1 GeV working point). The inefficiency of the MBTS triggers is attributed to disabled modules
(one on each side) and only a partial geometric overlap between the MBTS and the inner tracking volume.
For track 𝑝 T above 0.5 GeV, the track trigger is again fully efficient and the MBTS trigger performance
mostly recovered. This is because the presence of at least one track of this momentum is usually correlated
with additional activity resulting in signals in a few MBTS counters and thus corresponding triggers.
For HMT triggers the requirement on pixel clusters is removed and only the SCT space-points are counted
and subject to a threshold requirement, optimised to have full efficiency at a given number of tracks.
For HMT triggers at moderate values of 𝜇, an additional pile-up mitigation strategy is required. This is
because the counting of all tracks in the event instead of only tracks belonging to the highest multiplicity
vertex impacts the purity of this type of trigger. In addition, performing tracking in the full volume of
the ID becomes more time-consuming with rising 𝜇, and so any preselection is beneficial. An algorithm
to approximate the vertex position along 𝑧 and the count of tracks originating from it was developed and
optimised for expected data-taking conditions. It uses the triplets of clusters from the Pixel detector that are
used to construct a linear extrapolation to the luminous region. Extrapolated positions are histogrammed
along 𝑧 with the binning optimised so that the coincidental combinations form a negligible background and
real vertices form a well-pronounced peak. The threshold cut is applied on the count in the peak in the step
preceding the tracking step. Because of the availability of the vertex 𝑧 position, the tracks in HMT triggers
with pile-up suppression are counted only if they are within 10 mm along 𝑧 from the approximate vertex
position. The impact of the additional pile-up suppression procedure on the efficiency of HMT triggers
is shown in Figure 53: here again the offline tracks are required to pass a minimum-bias selection and
the triggers are seeded from the random L1 trigger. At moderate 𝜇 values, these chains are planned to be
seeded by a trigger that sums up all energy in the calorimeters, which was not yet commissioned at the
time of the 2022 low-𝜇 runs.

68
6.8.2 Forward triggers

In addition to the minimum-bias triggers above, a diverse set of triggers aims to record events with elastic,
diffractive or central-exclusive interactions. Their usage depends on the data-taking conditions such as
pile-up or LHC beam optics [85]. The latter is usually defined by the value of the betatron function at the
collision point, 𝛽∗ [85]. During the high-𝛽∗ runs the focus is on elastic scattering events triggered solely
by the ALFA detector, although there are also some triggers combining ALFA detector information with
that of other ATLAS subdetectors which target the soft diffractive events. A detailed description of ALFA
triggers can be found in Ref. [12]. During low-𝛽∗ (‘standard’) runs triggers are based on signals from
various stations of the AFP detectors with or without coincidence with standard signatures from the central
detector. A detailed description of AFP can be found in Refs. [3, 11]. The AFP can deliver trigger signals
from two of its detector systems: the Silicon Trackers (SiT) and the Time of Flight (ToF) detectors. SiT
trigger signals are expected to be more efficient (98-99% per station) than the signals from the ToF system
(about 80% per side), but are also known to suffer from a 400 ns dead time after each hit [86], causing the
efficiency to significantly decrease for the later bunches in a bunch train when the pile-up exceeds one. As
the SiT trigger dead time depends on the beam intensity and train structure, the following logic is used:
• During high-𝜇 runs, when the presence of protons in AFP is expected to be in every second bunch
crossing9 , the trigger items are based on ToF,
• during low-𝜇 runs, when the probability of observing a proton in consecutive bunch crossings is
small, the trigger items are based on SiT.
During high-𝜇 data taking, the physics programme using proton tagging is focused on the measurements of
exclusive and two-photon exchange processes, typically in conjunction with high-𝑝 𝑇 object(s) produced in
the ATLAS detector to keep the rate low. An example of such a process is exclusive jet production when
un-prescaled jets with 𝑝 𝑇 of about 150 GeV are required. In order to trigger such events, the presence
of a jet with a minimum threshold of 50 GeV and a proton(s) in AFP (ToF trigger) is required at L1. At
the HLT, matching is required between the kinematics of the centrally reconstructed di-jet system and the
scattered protons as reconstructed using the SiT. Depending on the settings of the algorithm selection
criteria, the efficiency is expected to be 60 − 85%. In addition, the match between the di-jet vertex and the
vertex 𝑧 location reconstructed using the ToF data can be applied. Such trigger chains may obtain a rate
reduction of a factor of 100 when compared to a nominal jet trigger with a similar 𝑝 𝑇 threshold.
The composition of the trigger menu for low-𝜇 runs depends on the exact data-taking conditions. The
number of colliding bunches, pile-up and beam optics [85] play a key role in the expected event rate.
Studies of soft diffractive processes are usually realised using the least biased triggers. For AFP, this
translates into the requirement of a proton to be detected in both stations on the same side of ATLAS
(single tagged events) or all four AFP stations (both ATLAS sides; double-tagged events). AFP triggers
can be combined with other ATLAS trigger objects. For example, for diffractive charm meson production,
in addition to the presence of protons in AFP, a track with a certain transverse momentum is additionally
required.

9 In the first approximation this depends on pile-up. E.g. for 𝜇 = 50 and a probability of registering pile-up proton originating
from a single 𝑝 𝑝 interaction of 2%, the chance of observing a proton in AFP is 1 − (1 − 0.02) 50 ≈ 64%.

69
6.9 Unconventional tracking signatures

The search for LLPs is an important part of the Run-3 physics programme, as they appear in many motivated
scenarios of phenomena beyond the SM. LLPs which decay within the ID volume or pass entirely through
it result in a variety of unconventional tracking signatures in the detector. Several new triggers presented
below were developed to target such unique signatures: long-lived charged particles, displaced jets, jets
with displaced tracks and displaced leptons. They make use of both standard tracking and LRT, described
in Section 5.1. The use of tracking for these signatures leads to lower background rates, which allows for
lower particle momentum or 𝐸 Tmiss requirements in the trigger, resulting in large gains in sensitivity for
Run-3 searches. Rates for the triggers discussed below are given in Table 3.

6.9.1 Long-lived charged particles that partially or fully traverse the inner detector

Three new triggers are developed targeting long-lived charged particles that partially or fully traverse the ID.
They make use of standard prompt full scan tracking executed after a calorimeter 𝐸 Tmiss or jet preselection
is applied to reduce the full scan tracking rate, as described in Section 5.1.4. In Run 2, these searches
generally relied on the 𝐸 Tmiss trigger [87–89] to select events, which resulted in a low acceptance for models
not producing large 𝐸 Tmiss .
Isolated high 𝒑 T track
Heavy charged particles (such as charginos, sleptons, and 𝑅-hadrons) with sufficiently long lifetimes can
leave an isolated, high-𝑝 T track in the ID. A new isolated high-𝑝 T track trigger is introduced for Run 3 to
increase sensitivity to LLP signatures with low 𝐸 Tmiss in the event. It requires a tcpufit 𝐸 Tmiss threshold
of 80 GeV, 30 GeV below the lowest unprescaled 𝐸 Tmiss trigger, to reduce the amount of CPU-time spent
on full scan tracking. The trigger uses tracks from the fast tracking step to select events with at least one
isolated track with 𝑝 T > 120 GeV that also passes additional track quality requirements. For the track to be
isolated, the scalar sum of the track 𝑝 T within Δ𝑅 < 0.3 must be less than 10 GeV. Figure 54 shows the
expected performance of the isolated high-𝑝 T track trigger with respect to the HLT 𝐸 Tmiss calculated with
cell and tcpufit algorithms, compared to the “or” of the lowest unprescaled 𝐸 Tmiss triggers based on
tcpufit and pfopufit algorithms. All of these algorithms are described in Section 6.6. SUSY di-stau
LLP MC simulation for a stau mass of 600 GeV and a lifetime of 10 ns is used to evaluate the efficiency.
The new trigger increases the acceptance to signal-like events at lower values of 𝐸 Tmiss .
Large 𝒅𝑬/𝒅𝒙 triggers
New heavy, charged particles, mentioned above, may also leave large energy deposits in the ID silicon
layers compared to what is expected from a minimally ionizing particle. The measurement of these
large ionisation energy losses per unit pathlength, 𝑑𝐸/𝑑𝑥, in the Pixel detector is a handle to identify
tracks as signal candidates. A new trigger targeting long-lived, heavy, charged particles [88] uses the
𝑑𝐸/𝑑𝑥 measurement capabilities of the ID for trigger decisions. The same 𝐸 Tmiss triggers, as used for
the isolated track trigger above, are required before running the full scan tracking. The 𝑑𝐸/𝑑𝑥 trigger
selects events with at least one track with 𝑝 T > 50 GeV, an average 𝑑𝐸/𝑑𝑥 > 1.7 MeV /cm, at least two
hits with 𝑑𝐸/𝑑𝑥 > 1.7 MeV /cm, track |𝑑0 | < 2.5 mm, and |𝜂| < 2.5. Figure 55 shows the online 𝑑𝐸/𝑑𝑥
distribution of tracks with 𝑝 T > 10 GeV in data collected during 2022 with 𝐸 Tmiss , single-jet, and multi-jet
triggers where full scan tracking was run. It also shows the correlation of the online 𝑑𝐸/𝑑𝑥 measurement
to the offline measurement.

70
 1.8 1.8
Trigger efficiency

Trigger efficiency

(tcpufit) [Fraction of Events]

ATLAS Simulation s=13.6 TeV ATLAS Simulation s=13.6 TeV
1.6 miss
ET (pfopufit OR tcpufit) OR isolated high-pT track 1.6 miss
ET (pfopufit OR tcpufit) OR isolated high-pT track

(cell) [Fraction of Events]

miss
ET (pfopufit OR tcpufit)
miss
ET (pfopufit OR tcpufit)
0.45
1.4 miss
0.25 1.4 miss
ET (cell) distribution ET (tcpufit) distribution 0.4
long-lived ∼τ∼τ, m(∼τ) = 600 GeV, τ(∼τ) = 10 ns long-lived ∼τ∼τ, m(∼τ) = 600 GeV, τ(∼τ) = 10 ns
1.2 ≥1 offline track with p > 100 GeV, 0.2 1.2 ≥1 offline track with p > 100 GeV, 0.35
T T
|η| < 2.5, d < 5, 2 pixel hits |η| < 2.5, d < 5, 2 pixel hits
1 0
1 0 0.3
0.15 0.25
0.8 0.8
0.2
0.6 0.1 0.6
0.15
0.4 0.4

miss
HLT Emiss
0.05 0.1

HLT ET
T
0.2 0.2 0.05
0 0 0 0
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
HLT Emiss
T (cell) [GeV] HLT Emiss
T (tcpufit) [GeV]

Figure 54: Isolated high-𝑝 T track trigger expected efficiency using Run-3 MC simulation of SUSY di-stau events
with a mass of 600 GeV and a lifetime of 10 ns vs (left) the cell and (right) the tcpufit algorithm 𝐸 Tmiss . The
efficiency of the new trigger (circles) is compared with that of a logical “or” of the lowest unprescaled 𝐸 Tmiss triggers
based on tcpufit and pfopufit algorithms described in Section 6.6 (squares). Both of these are overlaid on the
signal sample 𝐸 Tmiss distribution of fraction of events per bin (gray histogram), and include the L1 trigger efficiency.
Only statistical uncertainties are shown.

×103
Number of entries

Number of entries
Trigger track dE/dx [MeV cm2/g]

5 50
ATLAS ATLAS
Data 2022, s = 13.6 TeV 4.5
Data 2022, s = 13.6 TeV
106 40
4
Trigger track p > 10 GeV,
T
dE/dx > 1.25 MeV cm2/g 3.5 Offline track
p > 12 GeV
30
5 T
10 3 dE/dx > 1.25 MeV cm2/g

2.5 20
4
Trigger track
10 2 p > 10 GeV
T
dE/dx > 1.25 MeV cm2/g
10
1.5

1 0
1 1.5 2 2.5 3 3.5 4 4.5 5 1 1.5 2 2.5 3 3.5 4 4.5 5
2 2
Trigger track dE/dx [MeV cm /g] Offline track dE/dx [MeV cm /g]

Figure 55: (left) Distribution of 𝑑𝐸/𝑑𝑥 for tracks reconstructed in the trigger used for monitoring the dE/dx-based
trigger. (right) Correlation of online vs offline 𝑑𝐸/𝑑𝑥 for offline tracks with 𝑝 T > 12 GeV.

71
Disappearing track triggers
Charged particles with slightly shorter lifetimes than the signatures above can decay part way through
the ID and leave a short tracklet of a few hits. These tracklets are referred to as disappearing tracks
when the charged particle decays into invisible and low-𝑝 T particles that are not reconstructed. As in the
previous two cases, the disappearing track trigger makes use of full scan tracking executed after the same
𝐸 Tmiss trigger requirements, which are lower than the lowest unprescaled 𝐸 Tmiss trigger. The fast tracking
algorithm is modified to save tracklets with four hits in the inner layers of the ID that fail to become tracks.
In order to reduce the large background to this signature, a Boosted Decision Tree (BDT) based on the
track parameters, quality of fit, and number of hits in the Pixel and SCT detectors is used to separate
signal-like tracklets from background. The disappearing track trigger selects events with at least one
tracklet with 𝑝 T > 20 GeV that passes a stringent requirement on the BDT score. Figure 56 shows the
expected performance of the trigger compared to that of the 𝐸 Tmiss trigger with a 110 GeV threshold, which
was used in Run 2, for a model with long-lived charginos [90]. The acceptance of the trigger to events with
chargino momentum below 150 GeV is greatly improved.
Efficiency

0.8 ATLAS Simulation

High Level Trigger for Disappearing Track

0.6 E miss
T >110 GeV

E miss
T >110 GeV or
(E miss >80 GeV & Disappearing Track)
0.4 T

mχ∼± = 91-1000 GeV,τχ∼± = 1 ns

1 1
∼
0.2 at least 1 χ± decaying at 13 < r < 30 cm
1

|η|<1.8

0
100 150 200 250 300 350 400
∼ ∼
p of χ±-χ± system [GeV]
T

Figure 56: Efficiency as a function of the chargino-pair system transverse momentum for the HLT 𝐸 Tmiss trigger
(dotted line) and a logical OR with the disappearing track trigger (solid line) with respect to the L1 𝐸 Tmiss > 50 GeV
trigger. Both triggers are based on tcpufit 𝐸 Tmiss algorithm. MC simulation with Run-2 conditions is used. The
events are taken from models with a chargino lifetime of 1 ns and the chargino mass of either 91 GeV or in the range
200–1000 GeV with 100 GeV spacing. Only events with 𝐸 Tmiss > 60 GeV and at least one chargino decaying between
the pixel and SCT detectors (13–30 cm) in the central region, |𝜂| < 1.8, are considered.

6.9.2 Long-lived particle decays into jets

A second set of three new triggers have been developed targeting long-lived particle decays into jets, which
may be displaced themselves or contain displaced tracks. The triggers make use of full scan tracking run
after a 𝐸 Tmiss or jet requirement and in some cases use LRT as an additional handle to select events.
Hit-based displaced vertex triggers
Neutral LLPs may travel some distance into the detector before decaying, resulting in a displaced vertex
(DV) or jet. Previous searches relied on a variety of triggers looking for other objects in the final state.
Two hit-based DV triggers make use of hits not associated to tracks after the standard full scan tracking
is performed in events passing the L1 jet and 𝐸 Tmiss triggers, respectively. Jets with a large number of
remaining hits on the outer layers of the ID, and few on the inner layers, are indicative of DVs with

72
displaced tracks that are not reconstructed. Using this as an extra requirement allows for lower 𝐸 Tmiss and
jet thresholds compared to the lowest unprescaled triggers. A BDT trained on the fraction of hits in the
layers of the ID is used to implement this selection on jets. The first trigger seeded by the same 80 GeV
𝐸 Tmiss threshold as above selects events containing jets with 𝑝 T > 200 GeV and |𝜂| < 1. Figure 57 shows
its expected performance versus the number of pile-up interactions using MC simulation of a heavy Higgs
boson decaying into two long-lived scalars, each subsequently decaying into two 𝑏-jets compared to a
background process of 𝑡 𝑡¯ with an all-hadronic final state. The signal efficiency is around 70%, compared to
a background efficiency of less than 5%, for high numbers of additional interactions. The algorithm is
tuned such that there is no strong dependence of the signal efficiency on pile-up. The second trigger seeded
by a L1 jet with 𝑝 T > 100 GeV selects events containing jets with 𝑝 T > 260 GeV and |𝜂| < 1. It requires a
jet with 𝑝 T > 180 GeV at the HLT calorimeter preselection step before running the tracking.
Another hit-based trigger under development is seeded by a L1 trigger with 𝐸 Tmiss > 50 GeV and runs only
the fast tracking step of LRT in a composite RoI around the jets. Tracks with 𝑝 T > 2 GeV are then used to
build vertices using a modified version of the offline secondary-vertexing algorithm [91]. The algorithm is
optimised to be faster by requiring track pairs to be consistent with an approximate vertex position and by
reducing the combinatorics of clustering by binning the vertex positions.
Trigger efficiency

1
ATLAS Simulation s =13 TeV

0.8

0.6

0.4 Hit-based displaced vertex trigger

Signal H → 2S → 4b
Background (tt) mH = 1 TeV
0.2 mS = 50 GeV, cτ = 9 mm

0
10 20 30 40 50 60 70 80
Pile-up <µ>

Figure 57: Expected performance of the hit-based displaced vertex trigger seeded by a tcpufit 𝐸 Tmiss threshold of
80 GeV versus the average pile-up. MC simulation is used for signal and background processes. The signal process
is a heavy Higgs boson (𝑚 𝐻 = 1 TeV) decaying into two scalars (𝑚 𝑆 = 50 GeV) with a proper decay length of 9 mm,
which subsequently decay into two 𝑏-jets each. The background process is 𝑡 𝑡¯ production with an all-hadronic final
state. Only statistical uncertainties are shown.

Emerging jet triggers

In the models of new phenomena with a dark sector, decays of particles into hadronically interacting
particles may result in emerging jets with a large fraction of events possessing a DV and tracks with large
𝑑0 . The fraction of momentum associated to prompt tracks in the jet relative to the total jet momentum is
used to identify those with a large fraction of displaced tracks, i.e. a low momentum fraction associated to
prompt tracks. This trigger is run in events that contain a L1 jet with 𝑝 T > 100 GeV and an HLT jet with
𝑝 T > 200 GeV in order to reduce the rate of full scan tracking. The 𝑝 T of tracks within Δ𝑅 < 1.2 of the
leading large-R jet axis are summed. Jets are only considered if within |𝜂| < 1.8 and the tracks must have
𝑝 T > 2 GeV, |𝜂| < 2.4, |𝑑0 | < 8 mm, and |𝑑0 | < 2.5 · 𝜎𝑑0 mm, where the 𝑑0 resolution (𝜎𝑑0 ) is computed
from 𝑍+jets events in data. The ratio of the sum of track 𝑝 T over the jet 𝑝 T is computed, and events are
selected if this ratio, PTF, is less than 0.08. This extra requirement allows for a reduction of more than a

73
factor of two of the jet 𝑝 T threshold as compared to that of the lowest unprescaled trigger (420 GeV in
2022).
A second trigger of this type is seeded by a 45 GeV photon and selects events with two large-R jets with
|𝜂| < 2.0, 𝑝 T > 55 GeV, and PTF < 0.1. Figure 58 (left) shows that the emerging jet trigger is efficient
down to much lower jet 𝑝 T than the single large-R jet trigger. A model of a 1.5 TeV 𝑍 ′ decaying into two
20 GeV dark pions with a proper decay length of 50 mm is used [92]. The overall efficiency depends on the
acceptance of the PTF requirement.

Trigger efficiency
Trigger efficiency

1.4 ATLAS Simulation 1.4 ATLAS Simulation s = 13 TeV

s = 13 TeV Emerging jet p > 200 GeV
T
Large-R PFlow jet p > 460 GeV 1.2 H→aa, ma = 55 GeV, cτ = 100 mm
1.2 Offline selection: T
large-R jet p > 200 GeV, |η |<1.8 Combination (single OR di-jet)
T
1 1 Displaced di-jet trigger p > 50 GeV
T
Displaced single jet trigger p > 140 GeV
0.8 0.8 T

0.6 0.6
mZ' = 1500 GeV
0.4 0.4
mπdark = 20 GeV
cτ = 50 mm
0.2 0.2

0 0
200 300 400 500 600 700 800 180 200 220 240 260 280 300 320 340 360
Leading offline large-R jet p [GeV] Leading offline topo-cluster jet p [GeV]
T T

Figure 58: (left) Efficiency of the new emerging jet trigger compared to a single large-radius jet trigger for a 𝑍 ′
decaying into two dark pions with a proper decay length (c𝜏) of 50 mm. The efficiency is calculated as a function
of the 𝑝 T of the leading offline anti-𝑘 𝑡 R=1.0 jet, reclustered from R=0.4 topocluster jets. (right) Efficiency of the
LRT-based displaced jets triggers plotted against the leading offline anti-𝑘 𝑡 R=0.4 jet 𝑝 T for a model of exotic decays
of the Higgs boson into long-lived pseudoscalars 𝑎 with mass 55 GeV and proper decay length of 100 mm. Both
studies use MC simulation with Run-2 conditions. Only statistical uncertainties are shown.

Displaced jet triggers

Similar to the previous case, neutral LLPs decaying into quarks or gluons may result in displaced jets. Two
new LRT-based triggers select such events with a jet from initial-state radiation and at least one or two
displaced jets, but with a lower 𝑝 T threshold than the lowest unprescaled single-jet trigger at 420 GeV.
Tracks with 𝑝 T > 1 GeV and |𝜂| < 2.4 within Δ𝑅 < 0.4 of the jet are counted as prompt (𝑛trk prompt ) or
displaced (𝑛disp ) if they have |𝑑0 | < or > 3 mm, respectively. After the standard tracking is run for the
trk

events that contain a L1 jet with 𝑝 T > 100 GeV and an HLT jet with 𝑝 T > 180 GeV, jets are preselected by
prompt ≤ 2. LRT is then run on the remaining hits in RoIs around the three leading jets passing
requiring 𝑛trk
the preselection. In addition to the leading jet 𝑝 T > 180 GeV requirement, the displaced single-jet trigger
selects events with a 𝑝 T > 140 GeV jet with 𝑛trk disp ≥ 3 and 𝑛prompt ≤ 1. Similarly, the displaced di-jet
trk

trigger selects events with two 𝑝 T > 50 GeV jets where at least one must satisfy 𝑛trkdisp ≥ 3 and 𝑛prompt ≤ 2,
trk

disp ≥ 0 if 𝑛prompt ≤ 1. Figure 58 (right) shows the expected performance

and the other may satisfy only 𝑛trk trk

of the displaced jets triggers based on MC simulation of a model with exotic Higgs decay into long-lived
pseudoscalars 𝑎 in the jet 𝑝 T region below the primary single jet trigger. Details of the model can be found
in Ref. [93].

74
6.9.3 Long-lived particle decays into SM leptons

LLPs may also decay into SM leptons resulting in displaced electrons, muons, and taus. Three new sets of
triggers target these decays using LRT, and also standard tracking in the case of taus.
To select such events in Run 2, searches generally relied on triggers without tracking information such as
photon triggers to select displaced electrons and MS-only triggers to select muons [94]. These triggers had
high 𝑝 T thresholds of 50–120 GeV, requiring two objects for the 50 GeV threshold, and restrictions in 𝜂 in
the case of muons. New triggers for displaced electrons and muons directly trigger on these signatures
allowing for lower thresholds with respect to those used in Run 2. The displaced electron trigger runs
LRT in RoIs in events passing the same L1 threshold as the primary prompt electron chain, described in
Section 6.1. It selects events that have an electron with 𝑝 T > 30 GeV and |𝑑0 | > 3 mm that passes a loose
likelihood electron identification [5] without the use of 𝑑0 or requirements on the number of hits in the
Pixel detector. The displaced muon trigger runs LRT in RoIs in events passing the same L1 threshold
as the primary prompt muon chain, described in Section 6.2. It selects events that have a muon with
𝑝 T > 20 GeV and |𝑑0 | > 2 mm. Figures 59 and 60 show the expected efficiency of the displaced electron
and muon HLT triggers with respect to their L1 seeds in terms of the offline reconstructed lepton transverse
impact parameter (𝑑0 ) and the production radius of offline reconstructed electrons and muons, respectively,
using MC simulation of LLP di-stau production. MC simulation of pair production of staus with a 1 ns
lifetime is used, and stau masses of 100–500 GeV in 100 GeV steps are merged. The offline electrons
are required to pass the same loose likelihood identification as used in the trigger and have |𝑑0 | > 3 mm.
The offline muons are required to pass the medium working point described in Ref. [45] without a cut on
the number of pixel hits and have |𝑑0 | > 2 mm. The acceptance of the LRT-based triggers extends out
to the first layer of the SCT at 300 mm, where the layout of the detector no longer allows for eight hits
on the track in most regions. The standard tracking runs out to a |𝑑0 | of 5 and 10 mm for electrons and
muons respectively, limiting the acceptance of the standard prompt lepton triggers. The larger value for
muon tracking increases the acceptance of 𝐵-meson decays. Combining the standard and LRT light lepton
triggers provides continuous acceptance from small to large displacements.
∼τ∼τ (100-500 GeV, 1 ns) ∼τ∼τ (100-500 GeV, 1 ns)
1.4 1.4
HLT efficiency

HLT efficiency

ATLAS Simulation ATLAS Simulation

s=13.6 TeV LRT p >30 GeV, |d0|>3 mm s=13.6 TeV LRT p >20 GeV, |d0|>2 mm
1.2 T 1.2 T

RoI Electrons: p > 30 GeV Standard p >26 GeV

T
RoI Muons: p > 20 GeV Standard p >24 GeV
T
T T
(Standard HLT tracking: |d | < 5 mm) (Standard HLT tracking: |d | < 10 mm)
1 0 1 0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
−200 −150 −100 −50 0 50 100 150 200 −150 −100 −50 0 50 100 150
Offline electron d0 [mm] Offline muon d0 [mm]

Figure 59: Displaced (left) electron and (right) muon trigger efficiencies with respect to their L1 seeds versus the
offline lepton 𝑑0 (LRT, open triangles). The efficiencies for the isolated primary single electron and muon triggers
(Standard), described in Sections 6.1.2 and 6.2.1, are shown as inverted triangles. MC simulation samples with
Run-3 conditions are used. Only statistical uncertainties are shown.

75
 ∼τ∼τ (100-500 GeV, 1 ns) ∼τ∼τ (100-500 GeV, 1 ns)
HLT efficiency

HLT efficiency
1.2 ATLAS Simulation 1.2 ATLAS Simulation
s=13.6 TeV Combination (OR) s=13.6 TeV Combination (OR)
1 LRT p >30 GeV, |d0|>3 mm 1 LRT p >20 GeV, |d0|>2 mm
T T
Standard p >26 GeV Standard p >24 GeV
T T

0.8 0.8

0.6 0.6

0.4 RoI Electrons: p > 30 GeV, |d | > 3 mm 0.4 RoI Muons: p > 20 GeV, |d | > 2 mm
T 0 T 0
(Standard HLT tracking: |d | < 5 mm) (Standard HLT tracking: |d | < 10 mm)
0 0
0.2 0.2

0 0
0 50 100 150 200 250 300 0 50 100 150 200 250 300
Offline electron transverse production radius [mm] Offline muon transverse production radius [mm]

Figure 60: Displaced (left) electron and (right) muon trigger efficiencies with respect to their L1 seeds versus the
offline reconstructed lepton production radius (LRT, open triangles). The efficiencies for the isolated primary single
electron and muon triggers (Standard), described in Sections 6.1.2 and 6.2.1, are shown as inverted triangles. A
logical OR between the LRT and standard triggers (Combination) is marked with open circles. MC simulation
samples with Run-3 conditions are used. Only statistical uncertainties are shown.

Two dedicated triggers for displaced hadronically decaying taus are in development for Run 3. One is based
on standard tracking and the other on LRT. A search during Run 2 with this signature was not performed,
but would have needed to use jet or 𝐸 Tmiss triggers with high thresholds. The standard tau identification
RNN [95] is retrained to use standard tracking to identify displaced taus. Standard input samples are
replaced by representative signal samples with displaced tau content, which additionally include association
of large radius tracks in the tau reconstruction. MC samples with lifetimes of 0.01–100 ns are combined
for the training. The first tau LLP trigger selects events containing a tau with 𝑝 T > 200 GeV passing the
medium working point of this displaced tau identification. This identification is also used in multi-object
triggers looking for a tau+X, which allows for lower thresholds on the tau object. Primary L1 di-tau, single
muon, single isolated electron and 𝐸 Tmiss triggers are used to seed them. The second tau LLP trigger runs
LRT in an RoI around the calorimeter tau seed and uses the same RNN-based tau identification as trained
for the previous trigger. Events are selected if they contain at least one tau passing the displaced tau
identification. The trigger is expected to have a 𝑝 T requirement slightly lower than the lowest unprescaled
single tau trigger, 160 GeV. Figure 61 shows the expected performance of these triggers compared to the
prompt tau trigger based on an MC simulation of 100 GeV, 1 ns staus. The efficiency is computed with
respect to standard and large radius offline tau tracks that are truth-matched to signal tau decays. Hence,
multiple tracks from the same generated tau may be included in the computation. New LLP and LLP LRT
tau triggers significantly improve sensitivity to the displaced tau signals.

7 Auxiliary triggers

7.1 Non-collision background

Non-collision backgrounds comprise detector signals which do not originate from the collisions of paired
bunches at the ATLAS interaction point. They are categorised into beam-induced and cosmic-induced

76
 ∼τ∼τ (100 GeV, 1ns) ∼τ∼τ (100 GeV, 1ns)
1.4 1.4
Trigger efficiency

Trigger efficiency
Ntruth taus

Ntruth taus
Signal distribution ATLAS Simulation Signal distribution ATLAS Simulation 6000
Combination (OR) 2500
1.2 HLT tau p >25 GeV, medium, LLP, LRT s = 13.6 TeV
1.2 Combination (OR) s= 13.6 TeV
T
HLT tau p >25 GeV, medium, LLP HLT tau p >25 GeV, medium, LLP, LRT
T
5000
T
1 HLT tau p >25 GeV, medium 1 HLT tau p >25 GeV, medium, LLP
T
2000 T
HLT tau p >25 GeV, medium
T 4000
0.8 0.8
1500
3000
0.6 0.6
1000 2000
0.4 0.4
Efficiency wrt. offline tracks
500 Efficiency wrt. offline tracks
0.2 no d0 selection 0.2 1000
no d0 selection

0 0 0 0
0 50 100 150 200 250 0 50 100 150 200 250 300
Truth tau p [GeV] Truth tau Rdecay [mm]
T

Figure 61: Displaced tau trigger efficiency as a function of truth signal tau (left) 𝑝 T and (right) decay radius of the tau.
MC simulation with Run-3 conditions of staus with a mass of 100 GeV and a 1 ns lifetime is used. The efficiencies
for the prompt tau trigger (inverted triangle), LLP tau trigger using standard tracking (square), and LLP tau trigger
using LRT (triangle) are overlaid by the logical OR of all three (circle). The gray histogram shows the distribution of
truth taus for this model. Only statistical uncertainties are shown.

backgrounds. Both beam-induced and cosmic-induced backgrounds are non-negligible background sources
in searches for new phenomena targeting delayed or displaced detector signatures, like those discussed in
Section 6.9. A number of dedicated triggers are used to select non-collision backgrounds for study.

7.1.1 Cosmic ray triggers

The cosmic-induced background results from the impact of extremely energetic cosmic muons on the
detector, which induce signals in the muon spectrometers or energy deposits in the calorimeters. These
events are collected either during LHC data taking (with L1 triggers in the empty bunch crossings which
do not have 𝑝 𝑝 collisions) or during periods with no LHC data taking (e.g. for detector commissioning as
described in Section 3.2). The cosmic-muon triggers during data taking do not run any muon reconstruction
at the HLT and they only run for dedicated cosmic data taking when there is no beam. Further details on
studies of the ATLAS detector performance with cosmic-ray muons can be found in Ref. [96].
The IDCosmic stream is designed to have a total rate of about 20 Hz. It consists of two single L1 muon
triggers with 𝑝 T thresholds of 3 GeV and 8 GeV at approximate rates of 8 Hz and 5 Hz respectively, the
latter being unprescaled. In the same stream, a dedicated TRT-based L1 trigger for cosmics data taking uses
a fast read-out path for groups of channels in the detector [97], running at a typical rate of around 10 Hz.
The CosmicCalo stream aims for a target rate of 5 Hz and includes legacy L1Calo EM triggers with
𝐸 T > 3 GeV and 7 GeV, a tau lepton trigger with 𝐸 T > 8 GeV and jet triggers with 𝐸 T > 12 GeV and
30 GeV for |𝜂| < 3.1 and 𝐸 T > 30 GeV for 3.1 < |𝜂| < 4.9.

7.1.2 Beam-induced background triggers

Beam-induced background (BIB) [98] originates from (1) the inelastic interactions of protons with residual
gas molecules upstream and nearby the detector, producing showers of secondary particles (beam-gas
background), (2) protons with high transverse amplitude, or (3) from protons deflected in beam-gas

77
scattering hitting the tertiary collimators resulting in background/secondary particles entering the detector
(beam halo). The online monitoring of BIB is essential to track live information about the beam conditions,
which is also provided to the LHC. Further studies of BIB are essential to understand its origin and
composition, and maintain and develop adequate monitoring.
Two single jet triggers with a prescaled L1 threshold of 12 GeV and an unprescaled L1 threshold of 50 GeV
are used to record BIB events in bunch crossings where a proton bunch is present in only one or in neither
of the beams. In addition, triggers based on hits in the Beam Conditions Monitor (BCM) [99] are used
to record BIB events. The BCM consists of two stations of detectors located symmetrically around the
interaction point at 𝑧 = ±184 cm and 𝑟 = 55 mm (𝜂 ≈ 4.2). Each station has four modules of two diamond
sensors read out in parallel. The implementation of the BCM hit-based trigger is based on a coincidence of
an early hit on one side (A or C side) and one hit in-time with the bunch crossing on the other side (C or
A side) in unpaired bunch crossings, where only one of the two beams is filled with a proton bunch. In
order to be more independent of the presence of unpaired bunch crossings and to improve the purity of the
selected events, a new logic is implemented for Run 3 which relies on two early hits on the same side (2A
or 2C). This allows for the triggering on paired bunch crossings while maintaining a similar rate to the
AC/CA counterparts, especially when triggering on the first colliding pair of bunches in a train which helps
to mitigate against the impact of afterglow (increased cavern backgrounds following 𝑝 𝑝 collisions at high
𝜇). Furthermore, this allows for a measurement of the composition and fraction of BIB in paired bunch
crossings, which was found to be different compared to unpaired bunch crossings. The contribution of
ghost collisions10 is eliminated and thus the purity of the recorded BIB events is increased. A new bunch
group with first paired bunch crossings in a train, containing all paired bunch crossings following a gap of
at least 29 empty bunch crossings is defined. In order to evaluate the performance of the new triggers, they
are defined in different flavours, triggering on empty, unpaired, and the aforementioned first paired bunch
crossings in a train. Figure 62 shows the L1 rate of the old and new BCM triggers as a function of a bunch
crossing identification (BCID) value. For each trigger the per-BCID rate is shown for five different bunch
groups, paired colliding bunches, the first colliding bunches in a train, empty bunch crossings, unpaired
bunch crossings with either beam 1 or beam 2 filled with a proton bunch. As expected the rate of the
2A/2C variants is much lower than the AC/CA variants in BCIDs with paired colliding bunches. Further,
the rate in unpaired BCIDs is similar for both variations.

7.2 Zero-bias trigger

The zero-bias data are collected using a dedicated trigger, which fires one LHC turn after a L1 EM trigger
with 𝐸 T > 15 GeV fires. This approach allows to collect data which is unbiased with respect to the activity
in the event and at the same time proportional to the luminosity in each bunch crossing, which can not be
achieved with random triggers. Such triggers are used for detector studies as well as for the MC simulation
overlay method [100], which relies on the zero-bias data to account for the pile-up background, cavern
background, and detector noise. The zero-bias trigger, prescaled to a constant rate of about 10 Hz, was
based on the legacy L1Calo EM trigger in 2022. A new Run-3 zero-bias algorithm is implemented in the
new L1Topo system to be used following decommissioning of the L1Calo legacy system, because the data
connections between the new L1Calo system and CTP prevent the use of the legacy zero-bias trigger.

10 Ghost collisions occur between protons from a filled bunch, which typically has > 1011 protons, and protons in an unfilled
(empty) bunch, which has < 108 protons, due to diffusion from filled bunches which takes place at the interaction point.

78
 106 106
L1 rate [Hz/BCID]

L1 rate [Hz/BCID]
ATLAS BCM AC ATLAS BCM 2A
105 105
s = 13.6 TeV, November 2022 s = 13.6 TeV, November 2022
104 Paired Unpaired beam 1 104 Paired
FirstInTrain FirstInTrain
103 Empty Unpaired beam 2 103 Empty
Unpaired beam 1
102 102 Unpaired beam 2

10 10

1 1
−1 −1
10 10
−2
10 0 50 100 150 200 250 300 350 10−2 0 50 100 150 200 250 300 350
BCID BCID
6 6
10 10
L1 rate [Hz/BCID]

L1 rate [Hz/BCID]
5
ATLAS BCM CA ATLAS BCM 2C
10 105
s = 13.6 TeV, November 2022 s = 13.6 TeV, November 2022
104 Paired Unpaired beam 1 104 Paired
FirstInTrain FirstInTrain
10 3 Empty Unpaired beam 2 10 3 Empty
Unpaired beam 1
2
10 102 Unpaired beam 2

10 10

1 1
−1 −1
10 10
−2
10 0 50 100 150 200 250 300 350 10−2 0 50 100 150 200 250 300 350
BCID BCID

Figure 62: Per-BCID L1 rate of BCM hit-based triggers based on coincidence of (top left) AC and (top right) 2A,
and (bottom left) CA and (bottom right) 2C. The per-BCID rates are shown for paired colliding bunches, the first
colliding bunches in a train, empty bunch crossings (orange), unpaired bunch crossings with either beam 1 or beam 2
filled with a proton bunch. The BCID range shown is restricted to the first 350 BCIDs.

7.3 Triggers for tracking performance studies

There are dedicated triggers in the trigger menu to study lepton tracking performance, described in
Section 5.1. These idperf chains run the same selection as the corresponding triggers under study but do
not apply any requirements on tracking. This means that they are unbiased with respect to tracking and can
be used for efficiency measurements. For instance, if an electron candidate is formed from a track and
a cluster, the idperf electron chain does not make any selection on the electron candidate, as this would
include requirements on track-cluster matching to define the candidate, as described in Section 6.1.
Two types of idperf triggers are employed in the trigger menu: di-lepton triggers and single lepton triggers.
The di-lepton idperf triggers target 𝑍 boson or 𝐽/𝜓 decays to leptons in order to study the performance
with the tag-and-probe method. These chains are based on primary single or di-object electron and muon
triggers, described in Sections 6.1 and 6.2, and can run unprescaled at about 15-30 Hz not adding any
unique rate. These idperf triggers do not apply any requirements on tracking on the subleading trigger
leg which can be an electron, muon or tau candidate. The single object idperf triggers have thresholds of
5, 14, 20, 26 and 30 GeV for electrons, 6, 20, 24 and 40 GeV for muons and 25, 35, 80 and 160 GeV for
taus. These are heavily prescaled to record events at 0.5 – 1 Hz.

79
There is at most one reference object (one track) per RoI from the leptonic triggers. The purity of the muon
triggers is quite high. For the tau trigger, it is very low, and for the electron trigger, lower still. There
are very few real electrons in the sample accumulated by single electron idperf triggers, so the dilepton
idperf triggers are crucial to evaluate electron tracking performance.

7.4 Triggers for detector performance studies (calibration, noise)

Triggers used for detector performance studies often run at very high rates. To compensate for this, only
partial detector information is recorded through a strategy called Partial Event Building (PEB), which has
the potential to significantly reduce the event size and thus overall bandwidth. These triggers operate either
with a fixed subset of ROBs which are to be included in events recorded to the PEB stream, or a list of
ROBs can be derived dynamically on an event-by-event basis based on a set of RoIs. A combination of
both static and dynamic ROB lists is also possible. These triggers are used for the calibration of muon
subdetectors, low-𝑝 T offline muon calibration and LAr calorimeter performance studies.
For example, a dedicated calibration trigger, called a laser trigger, is active during collisions runs to
monitor the stability of the Tile calorimeter channels. The LHC abort gap of about 3 𝜇s is used to send
and register laser pulses that are arbitrated, timed and controlled with respect to the LHC signals [101].
Laser pulses are sent to the Tile calorimeter at the increased rate of 12 Hz for Run 3 and recorded for an
analysis. Laser events are used to track possible fast gain changes in the photomultiplier tubes and monitor
the timing calibration.
Random events with PEB information for Pixel and SCT detectors are recorded in the unfilled LHC bunches
into dedicated streams to identify noisy channels which need to be masked11 in these subdetectors. These
triggers typically run at about 10 Hz. A pixel with noise hit occupancy above 5 × 10−4 is classified as a
noisy channel and masked in DAQ.

7.4.1 LArNoiseBurst algorithm

The LAr calorimeter has a small but luminosity-dependent probability of generating large noise signals
involving a considerable number of cells, as discussed in detail in Ref. [102]. While this happens with
milli-Hertz frequency, high spurious values of energy are provided by these cells and it is therefore
important to veto affected events from the data quality monitoring and downstream of this from potentially
affected physics analyses. Veto windows occur across a much shorter timescale than that of a single
luminosity block. The detection and registration of LAr noise bursts is done via a LArNoiseBurst algorithm
which runs in the HLT.
Noise bursts generate electrical pulses which do not correspond to physics objects and thus these have a bad
quality of fit when energy reconstruction is attempted. Offline tools examine the quality factor of the cell
energy reconstruction procedure, counting the number of cells with a bad quality factor in a given event.
Given that many other factors could interfere with the cell quality factor (e.g. intense pile-up in a particular
event), a minimum of two events that are strongly correlated in time is required to declare a veto interval.
In the HLT, the offline algorithm is run for every accepted event that is written to certain streams, including
the Main stream, the TLA, delayed streams and streams recording events with cosmic signatures in the
calorimeter. Events that are considered bad by the offline tool have their time stamp published online
11 Not used for the track reconstruction.

80
via an information messaging system. A dedicated application then picks the time stamps, performs a
coincidence operation between them (a maximum time difference of 250 𝜇s is required) and declares the
interval to a database used by the offline reconstruction. The veto is then applied to reject events in the
post-run analysis. The computing resources taken by this algorithm are negligible due to very low rate of
the Noise Bursts and large overlap with other calorimeter triggers.

7.5 Beam-spot algorithms

A dedicated system allows the HLT to continuously measure the position, size and orientation of the
luminous region (also known as beam spot) at the ATLAS interaction point [103, 104]. Beam-spot
parameters can change from fill to fill, and some, in particular the transverse position and size, show
significant variation over the course of a fill as well. Knowledge of the current beam-spot parameters
is crucial for several HLT algorithms, most notably for the selection of events with 𝑏-jets, and for HLT
tracking itself. The parameters are also transmitted online to the LHC status display [105].
The beam-spot parameters are continuously monitored and archived, determined bunch-by-bunch as well
as fill average, and sampled over different intervals from one to many minutes depending on the required
statistics. One dedicated set that is used by the HLT is only updated whenever significant deviations from
the currently used values are detected. The process of obtaining the parameters, updating them and feeding
them to the algorithms as condition parameters is referred to as beam-spot calibration.
The calibration and update process is not trivial as it consists of multiple steps involving several cooperating
sub-systems to synchronise the beam-spot parameters across the distributed HLT farm at the same time.
This requires orchestrating a series of steps, primarily through the CTP:

• HLT algorithms extract tracking and vertexing information and publish their distributions in the form
of histograms,
• histograms from all individual HLT instances are aggregated by a monitoring infrastructure referred
to as the Gatherer,
• on each new luminosity block the merged histograms are processed and new estimates of the
beam-spot parameters are calculated by an external application called the BeamspotTool,
• when the BeamspotTool determines that the new estimate constitutes a significant change with
respect to the previous one, it sends these new parameters to the CTP process,
• the CTP process writes new beam-spot parameters to the conditions database with a validity interval
starting with the next luminosity block and then notifies the HLT processes of the pending update
via its event fragment,
• individual HLT processes read the new beam-spot parameters from the conditions database when
they receive events from the next luminosity block, for which the CoralServer and CoralProxy [106]
infrastructure provides scalable access to the conditions database from the HLT farm.

The online beam-spot calibration received significant improvements in preparation for Run 3.
One long-standing issue with the online beam-spot calibration during Run 2 was a small (5 – 15 𝜇𝑚)
systematic difference observed in the transverse beam position with respect to the offline calibration.
During LS2 this was tracked down to an incorrect transverse position at the coordinate system origin being

81
used for track clustering, which caused a systematic bias towards the origin. The issue was resolved by
using the current estimate of the transverse beam location for track clustering. Reprocessing of the Run-2
data with the fixed vertex finder showed the online calibration matching the offline measurement with
excellent precision. In 2022, there was still an observable difference in transverse position between offline
and online due to different alignment constants, which makes a direct comparison difficult.
The original calibration used for Run 1 and Run 2 is based on optimised vertex finding and fitting algorithms.
The Run-3 calibration introduces an additional algorithm based only on track information that does not
involve vertexing. The luminous region is narrower than the typical vertexing uncertainty, and an accurate
estimation of the transverse widths requires precise knowledge of the vertexing resolution. The resolution
itself depends on detector and trigger conditions, and is therefore evaluated in real time through a split-vertex
method that was introduced in Run 1. However, the smaller beam-spot sizes of Run 2 required larger
resolution corrections and the vertex method has shown limitations in case of low statistics. To study
alternatives, a new method was developed for Run 3. This method utilises information from reconstructed
tracks only and determines beam-spot parameters by a likelihood fit to the observed set of 𝑑0 and 𝜙
parameters of tracks. In the ATLAS online environment the tracking information is local to each of the
HLT processing nodes, and it is not possible to collect all track data at one location to perform fitting. To
support fitting of all available tracking data at a single location, the likelihood function is approximated to a
sufficiently small set of additive terms. These terms are calculated from the local set of tracks on each HLT
node and merged for the final fit by utilizing the Gatherer infrastructure. The new method demonstrated
reasonable performance with the reprocessed Run-2 data and a shorter ramp-up period between the start
of data collection and calibration availability compared to the vertex-based method. For Run 3 both
vertex-based and track-based algorithms are in use to further study their performance. During the initial
2022 data-taking period the track-based algorithm was used preferentially due to its better robustness,
lower demand on statistics and earlier availability of the beam spot. For high-luminosity running the vertex
algorithm is still the preferred method for its accuracy over a wider range in beam sizes.
The multi-step calibration process outlined above introduces a delay in availability of the calibrated
beam-spot position for HLT algorithms, with a typical latency of several luminosity blocks. This is not an
issue for the HLT during a run since the variations are small on that timescale. However, at the start of
each new fill this requires a short bootstrap period, during which certain HLT algorithms, such as those
used for 𝑏-tagging, have to be held off until a first measurement of the beam spot has succeeded. This
results in a loss of a few luminosity blocks of data for the individual triggers: on average 4.5 luminosity
blocks were missing beam-spot calibration at the start of each data-taking run in Run 2, which corresponds
to approximately one percent of a duration of a typical run.
Improvements in the Gatherer infrastructure helped to reduce the calibration bootstrap delay. The calibration
depends on histograms produced by the HLT and merged by the Gatherer. Improvements in handling of
histograms for each of the luminosity blocks in the Gatherer have reduced propagation delays for those types
of histograms. This results in a faster availability of those histograms for the BeamspotTool application
and helps in shortening the inherent ramp-up delay, which has reduced to an average of 2.5 luminosity
blocks in Run 3.
With the higher bunch intensities achieved by the injector upgrades, the LHC had for the first time begun
𝛽∗ -levelling in order to limit pile-up in 2022. This involves a large variation in spot sizes, in turn demanding
more frequent updates of the HLT during the levelling phase between 𝛽∗ of 60 cm to 30 cm. In 2023 the
levelling range is extended to be between 120 cm and 30 cm.

82
Another development begun in 2022 that can be applied in 2023 is to constrain HLT tracking to a 3𝜎
window around the longitudinal beam position which is projected to save a significant fraction of CPU
time in the HLT. This will involve yet another bootstrap procedure to settle on the beam spot at the start of
each run.
Figure 63 shows the time evolution of the vertical beam-spot position as measured, and subsequently
applied, by the HLT over the course of one LHC fill during November 2022 data taking. An update of
all parameters is performed whenever the position changes by more than 10% of the width, or the width
changes by more than 10%, or the uncertainty on any of the parameters decreases by more than half.
Y beam-spot position [m m]

−360

ATLAS
−365 pp data, Nov 2022, s = 13.6 TeV Online beam spot

HLT beam spot

−370
± 10% of width

−375

−380

−385

−390
20:00 23:00 02:00 05:00 08:00 11:00 14:00

CET Time

Figure 63: Time evolution of the vertical beam-spot position over the course of one LHC fill. The dots represent the
monitored beam-spot position measured in luminosity blocks. The line represents the position currently used by the
HLT algorithms which is measured as an average over a sliding window of a series of luminosity blocks. The band
indicates a range corresponding to 10% of the vertical beam-spot width that was measured at the same time.

8 Trigger software performance

8.1 Trigger software performance scaling

For Run 3 the same process forking architecture of the HLTMPPU was kept as in Run 2. However, due
to the multithreaded event selection, each worker process can now contain multiple event slots, which
are processed in parallel. In Run 2, each worker process handled one event, and the selection progressed
sequentially. This allowed the HLTSV to assign new events directly to idle worker processes. In Run 3,
the AthenaMT scheduler decides when an event slot is freed and when a new event can be read from the
input source. The HLTSV, therefore, assigns events to an event queue for each HLTMPPU from which new
events are pulled by the AthenaMT scheduler for each worker process.
In the Run-3 system, the number of forked worker processes per HLTMPPU and the number of parallel
event slots in each worker process are freely configurable. This allows for an adiabatic transition of a pure
multi-process-based event selection like in Run 2 (i.e. many worker processes with one event slot each) to a
pure multithreaded configuration with one worker process and many parallel event slots. Event throughput
measurements are used to decide what configuration will be used.

83
 55 6400 140
50 ATLAS ATLAS ATLAS
Application throughput [events / s]

Data 2022, 54 5600 Data 2022, 54 120 Data 2022, 54

45 Multi-processing Multi-processing Multi-processing
40 Multi-threading 4800 Multi-threading Multi-threading
100

Memory usage [GB]

Hybrid, 4 threads Hybrid, 4 threads Hybrid, 4 threads

CPU usage [%]

35 Hybrid, 8 threads 4000 Hybrid, 8 threads Hybrid, 8 threads
30 Hardware limit = 6400% 80 Hardware limit = 128 GB
3200
25 60
20 2400
15 40
1600
10
800 20
5
00 8 16 24 32 40 48 56 00 8 16 24 32 40 48 56 00 8 16 24 32 40 48 56
Number of events processed in parallel Number of events processed in parallel Number of events processed in parallel

Figure 64: (left) Application throughput in events/s, (middle) CPU usage (CPU time divided by wall time) in
percent and (right) memory usage in GB as a function of the number of events processed in parallel for AthenaMT
executing trigger selection algorithms. Blue squares represent a multi-processing approach. Pink circles represent a
multi-threading approach with a single process using a number of threads equal to the number of events requested
to process in parallel. Threads are not bound to events; instead, a pool of a number of threads (𝑁thread ) is used to
process an equivalent sized pool of events (𝑁event = 𝑁thread ). Green diamonds and orange triangles represent a hybrid
approach where a number of processes, 𝑁process = 𝑁event /𝑁thread , forked after initialisation use a fixed number of
threads each (𝑁thread = 4 and 8, respectively) to process an equivalent number of events in parallel. Differences
between these approaches are discussed in Section 3.4.

Figure 64 shows the trigger software performance scaling in terms of application throughput in events/s
(left), CPU usage (middle), and memory usage (right) as a function of the number of events processed
in parallel for AthenaMT executing trigger selection algorithms. The measurements were performed in
a standalone local environment using a machine identical to those used in the ATLAS HLT computing
farm during data taking. It is a dual processor machine with 128 GB RAM using a NUMA memory
architecture and two AMD EPYC 7302 CPUs, where each CPU has 16 real cores with two hyper-threads
per core, giving the total number of 64 threads. The data sample contains a mix of events representative
of the real HLT input data and trigger selection configuration identical to one used during data taking.
Four ways of achieving the parallelism are presented. Data were taken in 2022 using a pure multi-process
configuration with 48 forks as event throughput is the most critical metric for the HLT. Different constraints
apply, however, to Grid [31] processing of MC simulated events. Here, available memory per core is
significantly more restricted than on the dedicated HLT machines, and these tasks execute in Run 3 in a
pure multi-threaded configuration, the trigger simulation included. A typical Grid site in 2022 ran the
multi-threaded configuration with eight event slots. As ATLAS’ transition to multi-threading was new for
Run 3, a number of components still make use of mutex locking to provide safe shared access to certain
common resources. In aggregate, these bottlenecks severely limit event throughput at very high levels of
multi-threaded execution, as seen in Figure 64. Work continues to refactor these components to minimise
blocking behaviour in future software releases for both Run 3 and Run 4. In addition, the hybrid modes of
operation continue to be studied as significant memory savings are made even in hybrid modes which use a
small number of multi-threading event slots, hence minimising losses due to resource contention.
Figure 65 shows the calorimeter processing time dependency on the number of threads obtained offline by
concurrently processing various numbers of data events with different thread configurations. In the Run-3
AthenaMT framework, the same block of cells cannot be requested by two different algorithms (e.g. two
overlapping RoIs or a full scan request at the same time as any RoI) as it may cause simultaneous reading
and writing to a given memory area. To avoid this, locks are added to the HLTCalo data preparation service,

84
leading to a non-linear dependency between the processing time and the number of parallel processing
threads allocated for intra-event processing as shown in Figure 65. Unpacking of the calorimeter cells
and topological clustering algorithms do not show any dependence on the number of threads, contrary to
fast reconstruction, which exhibits turn-on-like dependence on the number of threads the shape of which
depends on the number of concurrently processed events. Optimisations are planned to either employ a
more fine-grained locking or through an updated processing model. The time of the full scan calorimeter
cell unpacking per call is slightly extended (mostly due to locks associated to RoIs) with the increased
number of threads, but more linear scaling performance is achieved through increasing the number of
inter-event parallel processing slots. This does not affect the HLTCalo algorithm’s functional performance,
such as cell and cluster parameters or reconstruction efficiency.

4.5 50
Mean execution time per call [ms]

Mean execution time per call [ms]

ATLAS Data 2022, s= 13.6 TeV
4
Unpacking of the full scan calorimeter cells
3.5 45
Events : 1 Events : 2 Events : 4 Events : 8
3 Events : 12 Events : 16 Events : 24 Events : 32
40
2.5
ATLAS
2 Data 2022, s= 13.6 TeV 35
1.5 Fast reconstruction
1 Events : 1 Events : 2 Events : 4 Events : 8 30
0.5 Events : 12 Events : 16 Events : 24 Events : 32
0 25
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Number of threads Number of threads

80
Mean execution time per call [ms]

ATLAS Data 2022, s= 13.6 TeV

78
Topological Clustering Algorithm
76
74
72
70
68
66
64 Events : 1 Events : 2 Events : 4 Events : 8
62 Events : 12 Events : 16 Events : 24 Events : 32
60
0 5 10 15 20 25 30
Number of threads

Figure 65: Mean execution time per call of (top left) the fast calorimeter reconstruction, (top right) the unpacking of
the full scan calorimeter cells and (bottom) the full scan topo-clustering algorithm as a function of the number of
threads for various numbers of concurrently processed events.

8.2 HLT processing time

The overall utilisation of the HLT is dependent on the mean time taken to process each event and the
incoming event rate from L1. The mean time to process an event is a product of the trigger menu, its
associated luminosity-dependent prescale sets, and of the collisions being supplied by the LHC. Here the
processing time is most heavily dependent on the mean number of pile-up interactions. An example of
the mean HLT processing time per event at different instantaneous luminosity values throughout a run is
presented in Figure 66 (left). The mean HLT processing time decreases with decreasing average pile-up due

85
to a reduction in event complexity. The menu is adjusted based on the current rate of events in order to fully
use available resources while retaining a constant 5000 slot safety margin (about 6% of the final 2022 farm)
to protect against losing slots due to power glitches, technical problems, etc. The slope of the distribution
in Figure 66 (left) becomes less steep at lower 𝜇 values where additional trigger selections are enabled.
Additional event processing restrictions came from the delayed delivery of replacement ROS servers. In
2022 the ROS (like in Run 2) was able to supply full detector information at half of the maximum L1 rate
(50 kHz). However, it is possible to reach the L1 rate of 100 kHz for a subset of detectors by re-configuring
the read-out, for example by increasing the number of ROS servers assigned to it.
In addition to the online monitoring of high-level HLT processing statistics, offline tools, such as Cost
Monitoring [33], are available to investigate the detailed performance of the HLT (e.g. the average
processing times of individual algorithms or whole trigger chains). Data required for such studies is saved
to a dedicated calibration stream which enables recording of this data for all events, not just those accepted.
This stream contains information on algorithms’ execution times and on any data requests they make to the
ROSes. By default, only the first 250 LBs are monitored in this way during physics data taking. After the
run has finished, the data from the stream undergo additional post-processing and the performance details
are available on a dedicated website.
The Cost Monitoring used during Runs 1 and 2 had to be adapted to the multi-threaded framework for
Run 3. While the following figures and tables present data taken using the MP configuration in 2022, some
changes and challenges which become more relevant in more MT-like configuration are pointed out in the
following. Upgraded monitoring includes a new MT-compatible algorithm gathering the data, a redesigned
post-processing framework, and additional monitors (including thread monitoring). Given that the new
HLT runs under the AthenaMT scheduler, the analysis of the results is no longer as simple as in a sequential
framework, therefore multiple metrics were prepared to provide an overview of the event processing time,
for example, an event wall time or the sum of the execution times of all algorithms. The wall time includes
the framework operations/delays between scheduled algorithms. A comparison between those values can
be seen in Figure 66 (right), which shows the wall time which includes both algorithm execution time and
time spent on framework operations (including algorithms scheduling, data traffic) as well as the total time
of just the algorithms. These data were recorded in a MP-mode, where the scheduler was only tasked with
processing a single event at a time using a single CPU core.
In the figure, three peaks can be identified, representing fast (approximately 30 ms), medium (approximately
300 ms), and slow (approximately 2s) events. The last type of the event is the rarest due to the early
rejection mechanism. For faster events, when only fast algorithms are executed and most of the trigger
chains are rejected, the impact of steering operations from the underlying framework is observable as an
overhead of about 10 ms. For events with execution times much longer than 100 ms, when time-consuming
algorithms, for example, tracking algorithms, are executed, it is negligible. Table 5 shows an example of
collected Cost Monitoring metrics for a given period of time during a run, broken down into the main types
of object reconstruction and the framework. The Cost Monitoring data are used in an iterative process to
identify areas where optimisation of code and strategy can yield the greatest impact with zero or negligible
physics impact. The difference between the wall time and the total time of all algorithms per event can
further diverge in different ways when running in MT mode. For the MP case, the total event wall time is
always greater than the algorithm time, but with sufficient intra-event parallelism in an MT configuration,
the total event wall time can be smaller than the algorithm time, as some of the event processing may occur
simultaneously on multiple cores.
At the peak instantaneous luminosity during the run, the HLT processing time of one event is approximately
600 ms (590 ms algorithm time), compared to 500 ms achieved during Run 2. Out of this number about

86
 ×106
HLT Processing Time [ms]

Events
14
600 ATLAS
ATLAS 12 pp data, Nov 2022
500 pp data, Nov 2022 s=13.6 TeV
10 Total event time [ms]
s=13.6 TeV
400 Algorithm time [ms]
8
300 6

200 4

100 2

0 0
25 30 35 40 45 50 55 10−1 1 10 102 103 104 105 106
Average pile-up <µ> HLT Processing Time [ms]

Figure 66: (left) Mean HLT wall time as a function of the average pile-up throughout a run. A vertical line marks the
instantaneous luminosity at which additional trigger selections are enabled. Error bars denote the Gaussian width of
the underlying per event measurements. (right) HLT processing time distribution per event for an instantaneous
luminosity of 1.8 × 1034 cm−2 s−1 at pile-up of 51. Shown are the wall time spent per event as well as the sum of the
algorithms only processing time. The difference between these two distributions is due to the time spent in framework
operations.

Table 5: Example of collected Cost Monitoring metrics for 50 LBs for an instantaneous luminosity of 1.8 ×
1034 cm−2 s−1 at pile-up of 51, showing how the total time is distributed between the main types of object
reconstruction and the framework software. The fractional time consumption of algorithms is calculated based on the
sum of the execution times of all algorithms.

Total time [%]

ID reconstruction 59
Muon reconstruction 14
Calorimeter reconstruction 11
Combined reconstruction and hypothesis algorithms 8
Trigger infrastructure 2
Other 6

59% of the total event time is spent on the ID tracking, despite the improvements discussed in Section 5.1.
This is due to the expanded use of the full scan tracking for the hadronic signatures in Run 3. The HLT
farm computing capacity has been increased for Run 3 by routine replacement of old servers, benefiting
from the sustained industry trend of increasing processor performance. This has enabled the additional
HLT processing for Run 3 that is described in this paper to support the physics goals of the experiment.
During Run 2, the HLT farm consisted of processors with a performance of 22.8 HS06/core [107] (0.8
MHS06 total farm performance). For Run 3, 60% of the farm was upgraded to 36.2 HS06/core (1.7 MHS06
total farm performance) for early 2022, rising to 100% by November 2022 (2.0 MHS06 available for the
remainder of Run 3 starting from 2023).

87
9 Conclusion

A large number of trigger upgrades and developments for the ATLAS experiment were made during the
second long shutdown of the LHC in preparation for the Run-3 data taking. A summary of the various
updates as well as the first Run-3 performance studies can be found in this paper.
Many changes in the L1 trigger system improve both its rejection of background events and acceptance
for interesting physics processes. Upgrades in the L1 calorimeter trigger increased the granularity of
information used by the trigger and enable it to run more sophisticated algorithms to identify physics
objects, and to calculate missing transverse momentum in the event with higher precision. The L1 muon
system was enhanced through the addition of New Small Wheels, new resistive plate chambers in the
barrel/endcap transition region as well as upgraded electronics. The topological trigger, installed during
Run 2, has also undergone a hardware upgrade. Its refined kinematic measurements of muons and other
objects as well as new flexibility to define multiplicity triggers allow for more sophisticated event selections
at L1.
The underlying framework of the HLT was completely rewritten in order to execute trigger algorithms
within the multi-threaded software framework AthenaMT, sharing common features between the trigger
software and the software used for offline reconstruction. These changes help to optimise the use of the HLT
computing resources, both in terms of computing power and memory consumption. New-and-improved
selection algorithms and strategies further improve the reconstruction of objects at the trigger level, in the
case of some of the hadronic signatures, at the expense of higher CPU time needs due to the expanded use
of full scan tracking.
While maintaining a level of consistency with the Run-2 trigger menu, the Run-3 trigger menu sets out
to exploit the newly implemented detector features, more performant HLT hardware, and algorithmic
advancements. Trigger thresholds at L1 and HLT were generally kept the same as during Run 2, benefiting
from improvements to reduce trigger rate. The additional available HLT rate compared to Run 2 is dedicated
to expanding the physics menu in the physics and TLA streams, lowering trigger thresholds and including
new triggers for previously unexplored phase space, which make extensive use of the various inner detector
tracking algorithms in the HLT.
The ATLAS trigger system was successfully (re-)commissioned with the first data acquired at 13.6 TeV,
with some final commissioning steps for L1Calo and L1Muon to be completed during Run 3. First
performance studies of the different trigger signatures and trigger efficiencies with respect to the offline
quantities are presented using the 13.6 TeV proton-proton collision data collected during 2022.

Acknowledgements

We thank CERN for the very successful operation of the LHC and its injectors, as well as the support staff
at CERN and at our institutions worldwide without whom ATLAS could not be operated efficiently.
The crucial computing support from all WLCG partners is acknowledged gratefully, in particular from
CERN, the ATLAS Tier-1 facilities at TRIUMF/SFU (Canada), NDGF (Denmark, Norway, Sweden),
CC-IN2P3 (France), KIT/GridKA (Germany), INFN-CNAF (Italy), NL-T1 (Netherlands), PIC (Spain),
RAL (UK) and BNL (USA), the Tier-2 facilities worldwide and large non-WLCG resource providers.
Major contributors of computing resources are listed in Ref. [108].

88
We gratefully acknowledge the support of ANPCyT, Argentina; YerPhI, Armenia; ARC, Australia; BMWFW
and FWF, Austria; ANAS, Azerbaijan; CNPq and FAPESP, Brazil; NSERC, NRC and CFI, Canada; CERN;
ANID, Chile; CAS, MOST and NSFC, China; Minciencias, Colombia; MEYS CR, Czech Republic; DNRF
and DNSRC, Denmark; IN2P3-CNRS and CEA-DRF/IRFU, France; SRNSFG, Georgia; BMBF, HGF and
MPG, Germany; GSRI, Greece; RGC and Hong Kong SAR, China; ISF and Benoziyo Center, Israel; INFN,
Italy; MEXT and JSPS, Japan; CNRST, Morocco; NWO, Netherlands; RCN, Norway; MEiN, Poland; FCT,
Portugal; MNE/IFA, Romania; MESTD, Serbia; MSSR, Slovakia; ARRS and MIZŠ, Slovenia; DSI/NRF,
South Africa; MICINN, Spain; SRC and Wallenberg Foundation, Sweden; SERI, SNSF and Cantons of
Bern and Geneva, Switzerland; MOST, Taipei; TENMAK, Türkiye; STFC, United Kingdom; DOE and
NSF, United States of America.
Individual groups and members have received support from BCKDF, CANARIE, CRC and DRAC, Canada;
CERN-CZ, PRIMUS 21/SCI/017 and UNCE SCI/013, Czech Republic; COST, ERC, ERDF, Horizon 2020,
ICSC-NextGenerationEU and Marie Skłodowska-Curie Actions, European Union; Investissements d’Avenir
Labex, Investissements d’Avenir Idex and ANR, France; DFG and AvH Foundation, Germany; Herakleitos,
Thales and Aristeia programmes co-financed by EU-ESF and the Greek NSRF, Greece; BSF-NSF and
MINERVA, Israel; Norwegian Financial Mechanism 2014-2021, Norway; NCN and NAWA, Poland; La
Caixa Banking Foundation, CERCA Programme Generalitat de Catalunya and PROMETEO and GenT
Programmes Generalitat Valenciana, Spain; Göran Gustafssons Stiftelse, Sweden; The Royal Society and
Leverhulme Trust, United Kingdom.
In addition, individual members wish to acknowledge support from CERN: European Organization for
Nuclear Research (CERN PJAS); Chile: Agencia Nacional de Investigación y Desarrollo (FONDECYT
1190886, FONDECYT 1210400, FONDECYT 1230812, FONDECYT 1230987); China: National Natural
Science Foundation of China (NSFC - 12175119, NSFC 12275265, NSFC-12075060); Czech Republic:
PRIMUS Research Programme (PRIMUS/21/SCI/017); EU: H2020 European Research Council (ERC -
101002463); European Union: European Research Council (ERC - 948254), Horizon 2020 Framework
Programme (MUCCA - CHIST-ERA-19-XAI-00), European Union, Future Artificial Intelligence Research
(FAIR-NextGenerationEU PE00000013), Italian Center for High Performance Computing, Big Data
and Quantum Computing (ICSC, NextGenerationEU), Marie Sklodowska-Curie Actions (EU H2020
MSC IF GRANT NO 101033496); France: Agence Nationale de la Recherche (ANR-20-CE31-0013,
ANR-21-CE31-0013, ANR-21-CE31-0022), Investissements d’Avenir Idex (ANR-11-LABX-0012), In-
vestissements d’Avenir Labex (ANR-11-LABX-0012); Germany: Baden-Württemberg Stiftung (BW
Stiftung-Postdoc Eliteprogramme), Deutsche Forschungsgemeinschaft (DFG - 469666862, DFG - CR
312/5-1); Italy: Istituto Nazionale di Fisica Nucleare (FELLINI G.A. n. 754496, ICSC, NextGener-
ationEU); Japan: Japan Society for the Promotion of Science (JSPS KAKENHI JP21H05085, JSPS
KAKENHI JP22H01227, JSPS KAKENHI JP22H04944, JSPS KAKENHI JP22KK0227); Netherlands:
Netherlands Organisation for Scientific Research (NWO Veni 2020 - VI.Veni.202.179); Norway: Re-
search Council of Norway (RCN-314472); Poland: Polish National Agency for Academic Exchange
(PPN/PPO/2020/1/00002/U/00001), Polish National Science Centre (NCN 2021/42/E/ST2/00350, NCN
OPUS nr 2022/47/B/ST2/03059, NCN UMO-2019/34/E/ST2/00393, UMO-2020/37/B/ST2/01043, UMO-
2021/40/C/ST2/00187); Slovenia: Slovenian Research Agency (ARIS grant J1-3010); Spain: BBVA
Foundation (LEO22-1-603), Generalitat Valenciana (Artemisa, FEDER, IDIFEDER/2018/048), La Caixa
Banking Foundation (LCF/BQ/PI20/11760025), Ministry of Science and Innovation (MCIN & NextGenEU
PCI2022-135018-2, MICIN & FEDER PID2021-125273NB, RYC2019-028510-I, RYC2020-030254-I,
RYC2021-031273-I, RYC2022-038164-I), PROMETEO and GenT Programmes Generalitat Valenciana
(CIDEGENT/2019/023, CIDEGENT/2019/027); Sweden: Swedish Research Council (VR 2018-00482,
VR 2022-03845, VR 2022-04683, VR grant 2021-03651), Knut and Alice Wallenberg Foundation (KAW

89
2017.0100, KAW 2018.0157, KAW 2018.0458, KAW 2019.0447); Switzerland: Swiss National Science
Foundation (SNSF - PCEFP2_194658); United Kingdom: Leverhulme Trust (Leverhulme Trust RPG-
2020-004); United States of America: U.S. Department of Energy (ECA DE-AC02-76SF00515), Neubauer
Family Foundation.

References
[1] ATLAS Collaboration, Performance of the ATLAS Trigger System in 2010,
Eur. Phys. J. C 72 (2012) 1849, arXiv: 1110.1530 [hep-ex].
[2] ATLAS Collaboration, Performance of the ATLAS trigger system in 2015,
Eur. Phys. J. C 77 (2017) 317, arXiv: 1611.09661 [hep-ex].
[3] ATLAS Collaboration, The ATLAS Experiment at the CERN Large Hadron Collider: A
Description of the Detector Configuration for Run 3, (2023),
arXiv: 2305.16623 [physics.ins-det].
[4] ATLAS Collaboration,
Preliminary analysis of the luminosity calibration of the ATLAS 13.6 TeV data recorded in 2022,
ATL-DAPR-PUB-2023-001, 2023, url: https://cds.cern.ch/record/2853525.
[5] ATLAS Collaboration, Performance of electron and photon triggers in ATLAS during LHC Run 2,
Eur. Phys. J. C 80 (2020) 47, arXiv: 1909.00761 [hep-ex].
[6] ATLAS Collaboration, The ATLAS Experiment at the CERN Large Hadron Collider,
JINST 3 (2008) S08003.
[7] B. Abbott et al., Production and integration of the ATLAS Insertable B-Layer,
JINST 13 (2018) T05008, arXiv: 1803.00844 [physics.ins-det].
[8] G. Aad et al.,
The Phase-I trigger readout electronics upgrade of the ATLAS Liquid Argon calorimeters,
JINST 17 (2022) P05024, arXiv: 2202.07384 [physics.ins-det].
[9] A. Armbruster et al.,
The ATLAS Muon to Central Trigger Processor Interface Upgrade for the Run 3 of the LHC,
2017 IEEE Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC) (2017) 1.
[10] A. Sidoti, Minimum Bias Trigger Scintillators in ATLAS Run II, JINST 9 (2014) C10020.
[11] L. Adamczyk et al., Technical Design Report for the ATLAS Forward Proton Detector,
CERN-LHCC-2015-009, ATLAS-TDR-024, 2015,
url: https://cds.cern.ch/record/2017378.
[12] S. Abdel Khalek et al., The ALFA Roman Pot detectors of ATLAS, JINST 11 (2016) P11013.
[13] G. Avoni et al., The new LUCID-2 detector for luminosity measurement and monitoring in ATLAS,
JINST 13 (2018) P07017.
[14] ATLAS Collaboration,
√
Luminosity determination in 𝑝 𝑝 collisions at 𝑠 = 13 TeV using the ATLAS detector at the LHC,
Eur. Phys. J. C 83 (2023) 982, arXiv: 2212.09379 [hep-ex].
[15] ATLAS Collaboration, Zero degree calorimeters for ATLAS, CERN-LHCC-2007-01, 2007,
url: https://cds.cern.ch/record/1009649/.

90
[16] H. Bertelsen et al.,
Operation of the upgraded ATLAS Central Trigger Processor during the LHC Run 2,
JINST 11 (2016) C02020.
[17] C. Leggett et al., AthenaMT: upgrading the ATLAS software framework for the many-core world
with multi-threading, Journal of Physics: Conference Series 898 (2017) 042009.
[18] ATLAS TDAQ Collaboration, The ATLAS Data Acquisition and High Level Trigger system,
JINST 11 (2016) P06008.
[19] ATLAS Collaboration, The ATLAS Collaboration Software and Firmware,
ATL-SOFT-PUB-2021-001, 2021, url: https://cds.cern.ch/record/2767187.
[20] ATLAS Collaboration, Operation of the ATLAS trigger system in Run 2, JINST 15 (2020) P10004,
arXiv: 2007.12539 [hep-ex].
[21] ATLAS Collaboration,
Performance of the upgraded PreProcessor of the ATLAS Level-1 Calorimeter Trigger,
JINST 15 (2020) P11016, arXiv: 2005.04179 [hep-ex].
[22] ATLAS Collaboration, Performance of the ATLAS muon triggers in Run 2,
JINST 15 (2020) P09015, arXiv: 2004.13447 [physics.ins-det].
[23] ATLAS Collaboration, Performance of the ATLAS Level-1 topological trigger in Run 2,
Eur. Phys. J. C 82 (2022) 7, arXiv: 2105.01416 [hep-ex].
[24] ATLAS Collaboration, ATLAS TDAQ System Phase-I Upgrade: Technical Design Report,
ATLAS-TDR-023; CERN-LHCC-2013-018, 2013,
url: https://cds.cern.ch/record/1602235.
[25] ATLAS Collaboration, Athena, version 22.0.1, 2019,
url: https://doi.org/10.5281/zenodo.2641997.
[26] G. Barrand et al.,
GAUDI — A software architecture and framework for building HEP data processing applications,
Comput. Phys. Commun. 140 (2001) 45.
[27] M. Clemencic, B. Hegner, P. Mato and D. Piparo,
Introducing concurrency in the Gaudi data processing framework,
J. Phys. Conf. Ser. 513 (2014) 022013, ed. by D. L. Groep and D. Bonacorsi.
[28] Intel Threading Building Blocks, url: https://github.com/oneapi-src/oneTBB.
[29] ATLAS Collaboration, ATLAS Future Framework Requirements Group Report,
ATL-SOFT-PUB-2016-001, 2016, url: https://cds.cern.ch/record/2196967.
[30] F. Pezoa, J. L. Reutter, F. Suarez, M. Ugarte and D. Vrgoč, Foundations of JSON schema,
Proceedings of the 25th International Conference on World Wide Web (2016) 263.
[31] K. Bos et al., LHC computing Grid: Technical Design Report, Technical design report. LCG,
Geneva: CERN, 2005, url: http://cds.cern.ch/record/840543.
[32] T. Cuhadar Donszelmann, W. Lampl and G. A. Stewart, ART ATLAS Release Tester using the Grid,
EPJ Web Conf. 245 (2020) 05015, ed. by C. Doglioni et al.
[33] ATLAS Collaboration, Trigger monitoring and rate predictions using Enhanced Bias data from the
ATLAS Detector at the LHC, ATL-DAQ-PUB-2016-002, 2016,
url: https://cds.cern.ch/record/2223498.

91
[34] M. Borodin et al.,
Scaling up ATLAS production system for the LHC Run 2 and beyond: project ProdSys2,
J. Phys. Conf. Ser. 664 (2015) 062005.
[35] ATLAS Collaboration,
ATLAS data quality operations and performance for 2015–2018 data-taking,
JINST 15 (2020) P04003, arXiv: 1911.04632 [physics.ins-det].
[36] ATLAS Collaboration, Search for Low-Mass Dijet Resonances Using Trigger-Level Jets with the
√
ATLAS Detector in 𝑝 𝑝 Collisions at 𝑠 = 13 TeV, Phys. Rev. Lett. 121 (2018) 081801,
arXiv: 1804.03496 [hep-ex].
[37] ATLAS Collaboration, Software Performance of the ATLAS Track Reconstruction for LHC Run 3,
ATL-PHYS-PUB-2021-012, 2021, url: https://cds.cern.ch/record/2766886.
[38] ATLAS Collaboration, Measurement of longitudinal flow decorrelations in Pb+Pb collisions at
√
𝑠NN = 2.76 and 5.02 TeV with the ATLAS detector, Eur. Phys. J. C 78 (2018) 142,
arXiv: 1709.02301 [hep-ex].
[39] ATLAS Collaboration, Measurements of azimuthal anisotropies of jet production in Pb+Pb
√
collisions at 𝑠NN = 5.02 TeV with the ATLAS detector, Phys. Rev. C 105 (2022) 064903,
arXiv: 2111.06606 [nucl-ex].
[40] ATLAS Collaboration, Measurement of the nuclear modification factor for inclusive jets in Pb+Pb
√
collisions at 𝑠NN = 5.02 TeV with the ATLAS detector, Phys. Lett. B 790 (2019) 108,
arXiv: 1805.05635 [hep-ex].
[41] ATLAS Collaboration,
The ATLAS Inner Detector Trigger performance in 𝑝 𝑝 collisions at 13 TeV during LHC Run 2,
Eur. Phys. J. C 82 (2022) 206, arXiv: 2107.02485 [hep-ex].
[42] ATLAS Collaboration,
Performance of the reconstruction of large impact parameter tracks in the inner detector of ATLAS,
Eur. Phys. J. C 83 (2023) 1081, arXiv: 2304.12867 [hep-ex].
[43] ATLAS Collaboration,
Performance of the ATLAS track reconstruction algorithms in dense environments in LHC Run 2,
Eur. Phys. J. C 77 (2017) 673, arXiv: 1704.07983 [hep-ex].
[44] ATLAS Collaboration, Electron and photon performance measurements with the ATLAS detector
using the 2015–2017 LHC proton–proton collision data, JINST 14 (2019) P12006,
arXiv: 1908.00005 [hep-ex].
[45] ATLAS Collaboration, Muon reconstruction and identification efficiency in ATLAS using the full
√
Run 2 𝑝 𝑝 collision data set at 𝑠 = 13 TeV, Eur. Phys. J. C 81 (2021) 578,
arXiv: 2012.00578 [hep-ex].
[46] ATLAS Collaboration, Jet energy scale and resolution measured in proton–proton collisions at
√
𝑠 = 13 TeV with the ATLAS detector, Eur. Phys. J. C 81 (2021) 689,
arXiv: 2007.02645 [hep-ex].
[47] ATLAS Collaboration,
𝐸 Tmiss performance in the ATLAS detector using 2015–2016 LHC 𝑝 𝑝 collisions,
ATLAS-CONF-2018-023, 2018, url: https://cds.cern.ch/record/2625233.
[48] ATLAS Collaboration, ATLAS flavour-tagging algorithms for the LHC Run 2 𝑝 𝑝 collision dataset,
Eur. Phys. J. C 83 (2023) 681, arXiv: 2211.16345 [physics.data-an].

92
[49] ATLAS Collaboration, Improved electron reconstruction in ATLAS using the Gaussian Sum
Filter-based model for bremsstrahlung, ATLAS-CONF-2012-047, 2012,
url: https://cds.cern.ch/record/1449796.
[50] ATLAS Collaboration,
Jet reconstruction and performance using particle flow with the ATLAS Detector,
Eur. Phys. J. C 77 (2017) 466, arXiv: 1703.10485 [hep-ex].
[51] N. Lad and on behalf of the ATLAS Collaboration, Machine learning predictor for
’measurement-to-track’ association for the ATLAS Inner Detector trigger,
Journal of Physics: Conference Series 2438 (2023) 012101.
[52] J. Long, The ATLAS Inner Detector tracking trigger at 13 TeV in LHC Run-2 and new
developments on standard and unconventional tracking signatures for the upcoming LHC Run-3,
ATL-DAQ-PROC-2022-005, 2022, url: https://cds.cern.ch/record/2813981.
[53] ATLAS Collaboration,
Fast 𝑏-tagging at the high-level trigger of the ATLAS experiment in LHC Run 3,
JINST 18 (2023) P11006, arXiv: 2306.09738 [hep-ex].
√
[54] ATLAS Collaboration, 𝐾𝑆0 and Λ production in 𝑝 𝑝 interactions at 𝑠 = 0.9 and 7 TeV measured
with the ATLAS detector at the LHC, Phys. Rev. D 85 (2012) 012001,
arXiv: 1111.1297 [hep-ex].
[55] D. Oliveira Damazio,
ATLAS High Level Calorimeter Trigger Software Performance for Cosmic Ray Events,
ATL-COM-DAQ-2009-045, 2009, url: https://cds.cern.ch/record/1179859.
[56] ATLAS Collaboration,
Topological cell clustering in the ATLAS calorimeters and its performance in LHC Run 1,
Eur. Phys. J. C 77 (2017) 490, arXiv: 1603.02934 [hep-ex].
[57] ATLAS Collaboration, Electron and photon energy calibration with the ATLAS detector using
2015–2016 LHC proton–proton collision data, JINST 14 (2019) P03017,
arXiv: 1812.03848 [hep-ex].
[58] T. Barillari et al., Local Hadronic Calibration, ATL-COM-LARG-2008-006, 2008,
url: http://cds.cern.ch/record/1112035.
[59] J. Seixas, L. Caloba, M. Souza, A. Braga and A. Rodrigues,
Neural second-level trigger system based on calorimetry, Comput. Phys. Commun. 95 (1996) 143.
[60] ATLAS Collaboration, Reconstruction, Identification, and Calibration of hadronically decaying
tau leptons with the ATLAS detector for the LHC Run 3 and reprocessed Run 2 data,
ATL-PHYS-PUB-2022-044, 2022, url: https://cds.cern.ch/record/2827111.
[61] M. Cacciari, G. P. Salam and G. Soyez, The anti-𝑘 𝑡 jet clustering algorithm, JHEP 04 (2008) 063,
arXiv: 0802.1189 [hep-ph].
[62] P. Berta, L. Masetti, D. Miller and M. Spousta,
Pileup and underlying event mitigation with iterative constituent subtraction, JHEP 08 (2019) 175.
[63] M. Cacciari, G. P. Salam and G. Soyez, SoftKiller, a particle-level pileup removal method,
Eur. Phys. J. C 75 (2015).
[64] A. J. Larkoski, S. Marzani, G. Soyez and J. Thaler, Soft drop, JHEP 05 (2014) 146.

93
[65] ATLAS Collaboration, Configuration and performance of the ATLAS 𝑏-jet triggers in Run 2,
Eur. Phys. J. C 81 (2021) 1087, arXiv: 2106.03584 [hep-ex].
[66] ATLAS Collaboration,
Deep Sets based Neural Networks for Impact Parameter Flavour Tagging in ATLAS,
ATL-PHYS-PUB-2020-014, 2020, url: https://cds.cern.ch/record/2718948.
[67] ATLAS Collaboration, Graph Neural Network Jet Flavour Tagging with the ATLAS Detector,
ATL-PHYS-PUB-2022-027, 2022, url: https://cds.cern.ch/record/2811135.
[68] ATLAS Collaboration, Search for resonant pair production of Higgs bosons in the 𝑏 𝑏𝑏
¯ 𝑏¯ final state
√
using 𝑝 𝑝 collisions at 𝑠 = 13 TeV with the ATLAS detector, Phys. Rev. D 105 (2022) 092002,
arXiv: 2202.07288 [hep-ex].
[69] ATLAS Collaboration, Search for nonresonant pair production of Higgs bosons in the 𝑏 𝑏𝑏 ¯ 𝑏¯ final
√
state in 𝑝 𝑝 collisions at 𝑠 = 13 TeV with the ATLAS detector, Phys. Rev. D 108 (2023) 052003,
arXiv: 2301.03212 [hep-ex].
[70] ATLAS Collaboration, Measurements of Higgs Bosons Decaying to Bottom Quarks from Vector
√
Boson Fusion Production with the ATLAS Experiment at 𝑠 = 13 TeV,
Eur. Phys. J. C 81 (2021) 537, arXiv: 2011.08280 [hep-ex].
[71] ATLAS Collaboration,
Search for Higgs boson production in association with a high-energy photon via vector-boson
√
fusion with decay into bottom quark pairs at 𝑠 = 13 TeV with the ATLAS detector,
JHEP 03 (2021) 268, arXiv: 2010.13651 [hep-ex].
[72] ATLAS Collaboration,
Search for the Standard Model Higgs boson decaying into 𝑏 𝑏¯ produced in association with top
√
quarks decaying hadronically in 𝑝 𝑝 collisions at 𝑠 = 8 TeV with the ATLAS detector,
JHEP 05 (2016) 160, arXiv: 1604.03812 [hep-ex].
[73] ATLAS Collaboration, Search for heavy neutral Higgs bosons produced in association with
√
𝑏-quarks and decaying into 𝑏-quarks at 𝑠 = 13 TeV with the ATLAS detector,
Phys. Rev. D 102 (2020) 032004, arXiv: 1907.02749 [hep-ex].
[74] ATLAS Collaboration, Search for single vector-like 𝐵 quark production and decay via
¯ in 𝑝 𝑝 collisions at √𝑠 = 13 TeV with the ATLAS detector, JHEP 11 (2023) 168,
𝐵 → 𝑏𝐻 (𝑏 𝑏)
arXiv: 2308.02595 [hep-ex].
[75] ATLAS Collaboration, Search for resonances in the mass distribution of jet pairs with one or two
√
jets identified as 𝑏-jets in proton–proton collisions at 𝑠 = 13 TeV with the ATLAS detector,
Phys. Rev. D 98 (2018) 032016, arXiv: 1805.09299 [hep-ex].
[76] ATLAS Collaboration,
Observation of 𝐻 → 𝑏 𝑏¯ decays and 𝑉 𝐻 production with the ATLAS detector,
Phys. Lett. B 786 (2018) 59, arXiv: 1808.08238 [hep-ex].
[77] ATLAS Collaboration, Search for invisible Higgs-boson decays in events with vector-boson fusion
signatures using 139 fb−1 of proton–proton data recorded by the ATLAS experiment,
JHEP 08 (2022) 104, arXiv: 2202.07953 [hep-ex].
[78] ATLAS Collaboration, Search for direct production of electroweakinos in final states with one
lepton, missing transverse momentum and a Higgs boson decaying into two 𝑏-jets in 𝑝 𝑝 collisions
√
at 𝑠 = 13 TeV with the ATLAS detector, Eur. Phys. J. C 80 (2020) 691,
arXiv: 1909.09226 [hep-ex].

94
[79] ATLAS Collaboration, Performance of the missing transverse momentum triggers for the ATLAS
detector during Run-2 data taking, JHEP 08 (2020) 080, arXiv: 2005.09554 [hep-ex].
[80] ATLAS Collaboration, Performance of pile-up mitigation techniques for jets in 𝑝 𝑝 collisions at
√
𝑠 = 8 TeV using the ATLAS detector, Eur. Phys. J. C 76 (2016) 581,
arXiv: 1510.03823 [hep-ex].
[81] ATLAS Collaboration, Constituent-level pile-up mitigation techniques in ATLAS,
ATLAS-CONF-2017-065, 2017, url: https://cds.cern.ch/record/2281055.
[82] Particle Data Group, Review of Particle Physics, PTEP 8 (2022) 083C01.
[83] ATLAS Collaboration,
Measurement of long-range multiparticle azimuthal correlations with the subevent cumulant
method in 𝑝 𝑝 and 𝑝+Pb collisions with the ATLAS detector at the CERN Large Hadron Collider,
Phys. Rev. C 97 (2018) 024904, arXiv: 1708.03559 [hep-ex].
√
[84] ATLAS Collaboration, Charged-particle distributions in 𝑠 = 13 TeV 𝑝 𝑝 interactions measured
with the ATLAS detector at the LHC, Phys. Lett. B 758 (2016) 67, arXiv: 1602.01633 [hep-ex].
[85] M. Trzebiński, Machine Optics Studies for the LHC Measurements,
Proc. SPIE Int. Soc. Opt. Eng. 9290 (2014) 929026, ed. by R. S. Romaniuk,
arXiv: 1408.1836 [physics.acc-ph].
[86] J. Lange et al., Beam tests of an integrated prototype of the ATLAS Forward Proton detector,
JINST 11 (2016) P09005.
[87] ATLAS Collaboration, Search for heavy charged long-lived particles in the ATLAS detector in
√
36.1 fb−1 of proton–proton collision data at 𝑠 = 13 TeV, Phys. Rev. D 99 (2019) 092007,
arXiv: 1902.01636 [hep-ex].
[88] ATLAS Collaboration, Search for heavy, long-lived, charged particles with large ionisation energy
√
loss in 𝑝 𝑝 collisions at 𝑠 = 13TeV using the ATLAS experiment and the full Run 2 dataset,
JHEP 06 (2023) 158, arXiv: 2205.06013 [hep-ex].
[89] ATLAS Collaboration, Search for long-lived charginos based on a disappearing-track signature in
√
𝑝 𝑝 collisions at 𝑠 = 13 TeV with the ATLAS detector, JHEP 06 (2018) 022,
arXiv: 1712.02118 [hep-ex].
[90] ATLAS Collaboration, Search for long-lived charginos based on a disappearing-track signature
√
using 136 fb−1 of 𝑝 𝑝 collisions at 𝑠 = 13 TeV with the ATLAS detector,
Eur. Phys. J. C 82 (2022) 606, arXiv: 2201.02472 [hep-ex].
[91] ATLAS Collaboration, Performance of vertex reconstruction algorithms for detection of new
long-lived particle decays within the ATLAS inner detector, ATL-PHYS-PUB-2019-013, 2019,
url: https://cds.cern.ch/record/2669425.
[92] D. Linthorne and D. Stolarski, Triggering on emerging jets, Phys. Rev. D 104 (2021) 035019,
arXiv: 2103.08620 [hep-ph].
[93] ATLAS Collaboration, Search for exotic decays of the Higgs boson into long-lived particles in 𝑝 𝑝
√
collisions at 𝑠 = 13 TeV using displaced vertices in the ATLAS inner detector,
JHEP 11 (2021) 229, arXiv: 2107.06092 [hep-ex].
[94] ATLAS Collaboration,
√
Search for displaced leptons in 𝑠 = 13 TeV 𝑝 𝑝 collisions with the ATLAS detector,
Phys. Rev. Lett. 127 (2021) 051802, arXiv: 2011.07812 [hep-ex].

95
 [95] ATLAS Collaboration,
Identification of hadronic tau lepton decays using neural networks in the ATLAS experiment,
ATL-PHYS-PUB-2019-033, 2019, url: https://cds.cern.ch/record/2688062.
[96] ATLAS Collaboration, Studies of the performance of the ATLAS detector using cosmic-ray muons,
Eur. Phys. J. C 71 (2011) 1593, arXiv: 1011.6665 [hep-ex].
[97] S. Fratina et al., The TRT Fast-OR Trigger, ATL-INDET-PUB-2009-002, 2009,
url: https://cds.cern.ch/record/1229213.
[98] ATLAS Collaboration, Beam-induced and cosmic-ray backgrounds observed in the ATLAS
detector during the LHC 2012 proton–proton running period, JINST 11 (2016) P05013,
arXiv: 1603.09202 [hep-ex].
[99] V. Cindro et al., The ATLAS beam conditions monitor, JINST 3 (2008) P02004.
[100] A. Haas, ATLAS Simulation using Real Data: Embedding and Overlay,
Journal of Physics: Conference Series 898 (2017) 042004.
[101] J. Abdallah et al., The Laser calibration of the Atlas Tile Calorimeter during the LHC run 1,
JINST 11 (2016) T10005, arXiv: 1608.02791 [physics.ins-det].
[102] ATLAS Collaboration,
Monitoring and data quality assessment of the ATLAS liquid argon calorimeter,
JINST 9 (2014) P07024, arXiv: 1405.3768 [hep-ex].
[103] R. Bartoldus,
Online Determination of the LHC Luminous Region with the ATLAS High-Level Trigger,
Phys. Procedia 37 (2012) 2080, ed. by T. Liu.
[104] F. Winklmeier, R. Bartoldus, J. Cogan, A. Salnikov and E. Strauss,
A system for monitoring and tracking the LHC beam spot within the ATLAS High Level Trigger,
2012 18th IEEE-NPSS Real Time Conference (2012) 1.
[105] url: https://op-webtools.web.cern.ch/vistar/vistars.php?usr=LHCCONFIG.
[106] A. Valassi, R. Bartoldus, A. Kalkhof, A. Salnikov and M. Wache, CORAL Server and CORAL
Server Proxy: Scalable Access to Relational Databases from CORAL Applications,
Journal of Physics: Conference Series 331 (2011) 042025.
[107] M. Michelotto et al.,
A comparison of HEP code with SPEC1 benchmarks on multi-core worker nodes,
Journal of Physics: Conference Series 219 (2010) 052009.
[108] ATLAS Collaboration, ATLAS Computing Acknowledgements, ATL-SOFT-PUB-2023-001, 2023,
url: https://cds.cern.ch/record/2869272.

96
The ATLAS Collaboration

G. Aad 102 , E. Aakvaag 16 , B. Abbott 120 , K. Abeling 55 , N.J. Abicht 49 , S.H. Abidi 29 ,
A. Aboulhorma 35e , H. Abramowicz 151 , H. Abreu 150 , Y. Abulaiti 117 , B.S. Acharya 69a,69b,m ,
C. Adam Bourdarios 4 , L. Adamczyk 86a , S.V. Addepalli 26 , M.J. Addison 101 , J. Adelman 115 ,
A. Adiguzel 21c , T. Adye 134 , A.A. Affolder 136 , Y. Afik 39 , M.N. Agaras 13 , J. Agarwala 73a,73b ,
A. Aggarwal 100 , C. Agheorghiesei 27c , A. Ahmad 36 , F. Ahmadov 38,z , W.S. Ahmed 104 ,
S. Ahuja 95 , X. Ai 62e , G. Aielli 76a,76b , A. Aikot 163 , M. Ait Tamlihat 35e , B. Aitbenchikh 35a ,
I. Aizenberg 169 , M. Akbiyik 100 , T.P.A. Åkesson 98 , A.V. Akimov 37 , D. Akiyama 168 ,
N.N. Akolkar 24 , S. Aktas 21a , K. Al Khoury 41 , G.L. Alberghi 23b , J. Albert 165 ,
P. Albicocco 53 , G.L. Albouy 60 , S. Alderweireldt 52 , Z.L. Alegria 121 , M. Aleksa 36 ,
I.N. Aleksandrov 38 , C. Alexa 27b , T. Alexopoulos 10 , F. Alfonsi 23b , M. Algren 56 ,
M. Alhroob 120 , B. Ali 132 , H.M.J. Ali 91 , S. Ali 148 , S.W. Alibocus 92 , M. Aliev 33c ,
G. Alimonti 71a , W. Alkakhi 55 , C. Allaire 66 , B.M.M. Allbrooke 146 , J.F. Allen 52 ,
C.A. Allendes Flores 137f , P.P. Allport 20 , A. Aloisio 72a,72b , F. Alonso 90 , C. Alpigiani 138 ,
M. Alvarez Estevez 99 , A. Alvarez Fernandez 100 , M. Alves Cardoso 56 , M.G. Alviggi 72a,72b ,
M. Aly 101 , Y. Amaral Coutinho 83b , A. Ambler 104 , C. Amelung36 , M. Amerl 101 , C.G. Ames 109 ,
D. Amidei 106 , S.P. Amor Dos Santos 130a , K.R. Amos 163 , V. Ananiev 125 , C. Anastopoulos 139 ,
T. Andeen 11 , J.K. Anders 36 , S.Y. Andrean 47a,47b , A. Andreazza 71a,71b , S. Angelidakis 9 ,
A. Angerami 41,ac , A.V. Anisenkov 37 , A. Annovi 74a , C. Antel 56 , M.T. Anthony 139 ,
E. Antipov 145 , M. Antonelli 53 , F. Anulli 75a , M. Aoki 84 , T. Aoki 153 , J.A. Aparisi Pozo 163 ,
M.A. Aparo 146 , L. Aperio Bella 48 , C. Appelt 18 , A. Apyan 26 , S.J. Arbiol Val 87 ,
C. Arcangeletti 53 , A.T.H. Arce 51 , E. Arena 92 , J-F. Arguin 108 , S. Argyropoulos 54 ,
J.-H. Arling 48 , O. Arnaez 4 , H. Arnold 114 , G. Artoni 75a,75b , H. Asada 111 , K. Asai 118 ,
S. Asai 153 , N.A. Asbah 61 , K. Assamagan 29 , R. Astalos 28a , S. Atashi 159 , R.J. Atkin 33a ,
M. Atkinson162 , H. Atmani35f , P.A. Atmasiddha 128 , K. Augsten 132 , S. Auricchio 72a,72b ,
A.D. Auriol 20 , V.A. Austrup 101 , G. Avolio 36 , K. Axiotis 56 , G. Azuelos 108,ag , D. Babal 28b ,
H. Bachacou 135 , K. Bachas 152,q , A. Bachiu 34 , F. Backman 47a,47b , A. Badea 39 , T.M. Baer 106 ,
P. Bagnaia 75a,75b , M. Bahmani 18 , D. Bahner 54 , A.J. Bailey 163 , V.R. Bailey 162 ,
J.T. Baines 134 , L. Baines 94 , O.K. Baker 172 , E. Bakos 15 , D. Bakshi Gupta 8 ,
V. Balakrishnan 120 , R. Balasubramanian 114 , E.M. Baldin 37 , P. Balek 86a , E. Ballabene 23b,23a ,
F. Balli 135 , L.M. Baltes 63a , W.K. Balunas 32 , J. Balz 100 , E. Banas 87 , M. Bandieramonte 129 ,
A. Bandyopadhyay 24 , S. Bansal 24 , L. Barak 151 , M. Barakat 48 , E.L. Barberio 105 ,
D. Barberis 57b,57a , M. Barbero 102 , M.Z. Barel 114 , K.N. Barends 33a , T. Barillari 110 ,
M-S. Barisits 36 , T. Barklow 143 , P. Baron 122 , D.A. Baron Moreno 101 , A. Baroncelli 62a ,
G. Barone 29 , A.J. Barr 126 , J.D. Barr 96 , F. Barreiro 99 , J. Barreiro Guimarães da Costa 14a ,
U. Barron 151 , M.G. Barros Teixeira 130a , S. Barsov 37 , F. Bartels 63a , R. Bartoldus 143 ,
A.E. Barton 91 , P. Bartos 28a , A. Basan 100 , M. Baselga 49 , A. Bassalat 66,b , M.J. Basso 156a ,
C.R. Basson 101 , R.L. Bates 59 , S. Batlamous35e , J.R. Batley 32 , B. Batool 141 , M. Battaglia 136 ,
D. Battulga 18 , M. Bauce 75a,75b , M. Bauer 36 , P. Bauer 24 , L.T. Bazzano Hurrell 30 ,
J.B. Beacham 51 , T. Beau 127 , J.Y. Beaucamp 90 , P.H. Beauchemin 158 , P. Bechtle 24 ,
H.P. Beck 19,p , K. Becker 167 , A.J. Beddall 82 , V.A. Bednyakov 38 , C.P. Bee 145 ,
L.J. Beemster 15 , T.A. Beermann 36 , M. Begalli 83d , M. Begel 29 , A. Behera 145 , J.K. Behr 48 ,
J.F. Beirer 36 , F. Beisiegel 24 , M. Belfkir 116b , G. Bella 151 , L. Bellagamba 23b , A. Bellerive 34 ,
P. Bellos 20 , K. Beloborodov 37 , D. Benchekroun 35a , F. Bendebba 35a , Y. Benhammou 151 ,
K.C. Benkendorfer 61 , L. Beresford 48 , M. Beretta 53 , E. Bergeaas Kuutmann 161 , N. Berger 4 ,

97
B. Bergmann 132 , J. Beringer 17a , G. Bernardi 5 , C. Bernius 143 , F.U. Bernlochner 24 ,
F. Bernon 36,102 , A. Berrocal Guardia 13 , T. Berry 95 , P. Berta 133 , A. Berthold 50 ,
I.A. Bertram 91 , S. Bethke 110 , A. Betti 75a,75b , A.J. Bevan 94 , N.K. Bhalla 54 , M. Bhamjee 33c ,
S. Bhatta 145 , D.S. Bhattacharya 166 , P. Bhattarai 143 , K.D. Bhide 54 , V.S. Bhopatkar 121 ,
R.M. Bianchi 129 , G. Bianco 23b,23a , O. Biebel 109 , R. Bielski 123 , M. Biglietti 77a ,
C.S. Billingsley44 , M. Bindi 55 , A. Bingul 21b , C. Bini 75a,75b , A. Biondini 92 ,
C.J. Birch-sykes 101 , G.A. Bird 32 , M. Birman 169 , M. Biros 133 , S. Biryukov 146 , T. Bisanz 49 ,
E. Bisceglie 43b,43a , J.P. Biswal 134 , D. Biswas 141 , K. Bjørke 125 , I. Bloch 48 , A. Blue 59 ,
U. Blumenschein 94 , J. Blumenthal 100 , V.S. Bobrovnikov 37 , M. Boehler 54 , B. Boehm 166 ,
D. Bogavac 36 , A.G. Bogdanchikov 37 , C. Bohm 47a , V. Boisvert 95 , P. Bokan 36 , T. Bold 86a ,
M. Bomben 5 , M. Bona 94 , M. Boonekamp 135 , C.D. Booth 95 , A.G. Borbély 59 ,
I.S. Bordulev 37 , H.M. Borecka-Bielska 108 , G. Borissov 91 , D. Bortoletto 126 , D. Boscherini 23b ,
M. Bosman 13 , J.D. Bossio Sola 36 , K. Bouaouda 35a , N. Bouchhar 163 , J. Boudreau 129 ,
E.V. Bouhova-Thacker 91 , D. Boumediene 40 , R. Bouquet 165 , A. Boveia 119 , J. Boyd 36 ,
D. Boye 29 , I.R. Boyko 38 , J. Bracinik 20 , N. Brahimi 62d , E.D. Brandani 129 , G. Brandt 171 ,
O. Brandt 32 , F. Braren 48 , B. Brau 103 , J.E. Brau 123 , R. Brener 169 , L. Brenner 114 ,
R. Brenner 161 , S. Bressler 169 , D. Britton 59 , D. Britzger 110 , I. Brock 24 , G. Brooijmans 41 ,
E. Brost 29 , L.M. Brown 165 , L.E. Bruce 61 , T.L. Bruckler 126 , P.A. Bruckman de Renstrom 87 ,
B. Brüers 48 , A. Bruni 23b , G. Bruni 23b , M. Bruschi 23b , N. Bruscino 75a,75b , T. Buanes 16 ,
Q. Buat 138 , D. Buchin 110 , A.G. Buckley 59 , O. Bulekov 37 , B.A. Bullard 143 , S. Burdin 92 ,
C.D. Burgard 49 , A.M. Burger 36 , B. Burghgrave 8 , O. Burlayenko 54 , J.T.P. Burr 32 ,
C.D. Burton 11 , J.C. Burzynski 142 , E.L. Busch 41 , V. Büscher 100 , P.J. Bussey 59 ,
J.M. Butler 25 , C.M. Buttar 59 , J.M. Butterworth 96 , W. Buttinger 134 , C.J. Buxo Vazquez 107 ,
A.R. Buzykaev 37 , S. Cabrera Urbán 163 , L. Cadamuro 66 , D. Caforio 58 , H. Cai 129 ,
Y. Cai 14a,14e , Y. Cai 14c , V.M.M. Cairo 36 , O. Cakir 3a , N. Calace 36 , P. Calafiura 17a ,
G. Calderini 127 , P. Calfayan 68 , G. Callea 59 , L.P. Caloba83b , D. Calvet 40 , S. Calvet 40 ,
M. Calvetti 74a,74b , R. Camacho Toro 127 , S. Camarda 36 , D. Camarero Munoz 26 ,
P. Camarri 76a,76b , M.T. Camerlingo 72a,72b , D. Cameron 36 , C. Camincher 165 , M. Campanelli 96 ,
A. Camplani 42 , V. Canale 72a,72b , J. Cantero 163 , Y. Cao 162 , F. Capocasa 26 , M. Capua 43b,43a ,
A. Carbone 71a,71b , R. Cardarelli 76a , J.C.J. Cardenas 8 , F. Cardillo 163 , G. Carducci 43b,43a ,
T. Carli 36 , G. Carlino 72a , J.I. Carlotto 13 , B.T. Carlson 129,r , E.M. Carlson 165,156a ,
L. Carminati 71a,71b , A. Carnelli 135 , M. Carnesale 75a,75b , S. Caron 113 , E. Carquin 137f ,
S. Carrá 71a , G. Carratta 23b,23a , A.M. Carroll 123 , T.M. Carter 52 , M.P. Casado 13,i ,
M. Caspar 48 , F.L. Castillo 4 , L. Castillo Garcia 13 , V. Castillo Gimenez 163 , N.F. Castro 130a,130e ,
A. Catinaccio 36 , J.R. Catmore 125 , T. Cavaliere 4 , V. Cavaliere 29 , N. Cavalli 23b,23a ,
Y.C. Cekmecelioglu 48 , E. Celebi 21a , S. Cella 36 , F. Celli 126 , M.S. Centonze 70a,70b ,
V. Cepaitis 56 , K. Cerny 122 , A.S. Cerqueira 83a , A. Cerri 146 , L. Cerrito 76a,76b , F. Cerutti 17a ,
B. Cervato 141 , A. Cervelli 23b , G. Cesarini 53 , S.A. Cetin 82 , D. Chakraborty 115 , J. Chan 17a ,
W.Y. Chan 153 , J.D. Chapman 32 , E. Chapon 135 , B. Chargeishvili 149b , D.G. Charlton 20 ,
M. Chatterjee 19 , C. Chauhan 133 , Y. Che 14c , S. Chekanov 6 , S.V. Chekulaev 156a ,
G.A. Chelkov 38,a , A. Chen 106 , B. Chen 151 , B. Chen 165 , H. Chen 14c , H. Chen 29 ,
J. Chen 62c , J. Chen 142 , M. Chen 126 , S. Chen 153 , S.J. Chen 14c , X. Chen 62c,135 ,
X. Chen 14b,af , Y. Chen 62a , C.L. Cheng 170 , H.C. Cheng 64a , S. Cheong 143 , A. Cheplakov 38 ,
E. Cheremushkina 48 , E. Cherepanova 114 , R. Cherkaoui El Moursli 35e , E. Cheu 7 , K. Cheung 65 ,
L. Chevalier 135 , V. Chiarella 53 , G. Chiarelli 74a , N. Chiedde 102 , G. Chiodini 70a ,
A.S. Chisholm 20 , A. Chitan 27b , M. Chitishvili 163 , M.V. Chizhov 38 , K. Choi 11 , Y. Chou 138 ,
E.Y.S. Chow 113 , K.L. Chu 169 , M.C. Chu 64a , X. Chu 14a,14e , J. Chudoba 131 ,

98
J.J. Chwastowski 87 , D. Cieri 110 , K.M. Ciesla 86a , V. Cindro 93 , A. Ciocio 17a , F. Cirotto 72a,72b ,
Z.H. Citron 169,k , M. Citterio 71a , D.A. Ciubotaru27b , A. Clark 56 , P.J. Clark 52 , C. Clarry 155 ,
J.M. Clavijo Columbie 48 , S.E. Clawson 48 , C. Clement 47a,47b , J. Clercx 48 , Y. Coadou 102 ,
M. Cobal 69a,69c , A. Coccaro 57b , R.F. Coelho Barrue 130a , R. Coelho Lopes De Sa 103 ,
S. Coelli 71a , B. Cole 41 , J. Collot 60 , P. Conde Muiño 130a,130g , M.P. Connell 33c ,
S.H. Connell 33c , I.A. Connelly 59 , E.I. Conroy 126 , F. Conventi 72a,ah , H.G. Cooke 20 ,
A.M. Cooper-Sarkar 126 , A. Cordeiro Oudot Choi 127 , L.D. Corpe 40 , M. Corradi 75a,75b ,
F. Corriveau 104,x , A. Cortes-Gonzalez 18 , M.J. Costa 163 , F. Costanza 4 , D. Costanzo 139 ,
B.M. Cote 119 , G. Cowan 95 , K. Cranmer 170 , D. Cremonini 23b,23a , S. Crépé-Renaudin 60 ,
F. Crescioli 127 , M. Cristinziani 141 , M. Cristoforetti 78a,78b , V. Croft 114 , J.E. Crosby 121 ,
G. Crosetti 43b,43a , A. Cueto 99 , T. Cuhadar Donszelmann 159 , H. Cui 14a,14e , Z. Cui 7 ,
W.R. Cunningham 59 , F. Curcio 43b,43a , P. Czodrowski 36 , M.M. Czurylo 63b ,
M.J. Da Cunha Sargedas De Sousa 57b,57a , J.V. Da Fonseca Pinto 83b , C. Da Via 101 ,
W. Dabrowski 86a , T. Dado 49 , S. Dahbi 33g , T. Dai 106 , D. Dal Santo 19 , C. Dallapiccola 103 ,
M. Dam 42 , G. D’amen 29 , V. D’Amico 109 , J. Damp 100 , J.R. Dandoy 34 , M. Danninger 142 ,
V. Dao 36 , G. Darbo 57b , S. Darmora 6 , S.J. Das 29,ai , S. D’Auria 71a,71b , C. David 33a ,
T. Davidek 133 , B. Davis-Purcell 34 , I. Dawson 94 , H.A. Day-hall 132 , K. De 8 ,
R. De Asmundis 72a , N. De Biase 48 , S. De Castro 23b,23a , N. De Groot 113 , P. de Jong 114 ,
H. De la Torre 115 , A. De Maria 14c , A. De Salvo 75a , U. De Sanctis 76a,76b , F. De Santis 70a,70b ,
A. De Santo 146 , J.B. De Vivie De Regie 60 , D.V. Dedovich38 , J. Degens 114 , A.M. Deiana 44 ,
F. Del Corso 23b,23a , J. Del Peso 99 , F. Del Rio 63a , L. Delagrange 127 , F. Deliot 135 ,
C.M. Delitzsch 49 , M. Della Pietra 72a,72b , D. Della Volpe 56 , A. Dell’Acqua 36 ,
L. Dell’Asta 71a,71b , M. Delmastro 4 , P.A. Delsart 60 , S. Demers 172 , M. Demichev 38 ,
S.P. Denisov 37 , L. D’Eramo 40 , D. Derendarz 87 , F. Derue 127 , P. Dervan 92 , K. Desch 24 ,
C. Deutsch 24 , F.A. Di Bello 57b,57a , A. Di Ciaccio 76a,76b , L. Di Ciaccio 4 ,
A. Di Domenico 75a,75b , C. Di Donato 72a,72b , A. Di Girolamo 36 , G. Di Gregorio 36 ,
A. Di Luca 78a,78b , B. Di Micco 77a,77b , R. Di Nardo 77a,77b , M. Diamantopoulou 34 , F.A. Dias 114 ,
T. Dias Do Vale 142 , M.A. Diaz 137a,137b , F.G. Diaz Capriles 24 , M. Didenko 163 , E.B. Diehl 106 ,
L. Diehl 54 , S. Díez Cornell 48 , C. Diez Pardos 141 , C. Dimitriadi 161,24 , A. Dimitrievska 17a ,
J. Dingfelder 24 , I-M. Dinu 27b , S.J. Dittmeier 63b , F. Dittus 36 , F. Djama 102 , T. Djobava 149b ,
C. Doglioni 101,98 , A. Dohnalova 28a , J. Dolejsi 133 , Z. Dolezal 133 , K.M. Dona 39 ,
M. Donadelli 83c , B. Dong 107 , J. Donini 40 , A. D’Onofrio 72a,72b , M. D’Onofrio 92 ,
J. Dopke 134 , A. Doria 72a , N. Dos Santos Fernandes 130a , P. Dougan 101 , M.T. Dova 90 ,
A.T. Doyle 59 , M.A. Draguet 126 , E. Dreyer 169 , I. Drivas-koulouris 10 , M. Drnevich 117 ,
M. Drozdova 56 , D. Du 62a , T.A. du Pree 114 , F. Dubinin 37 , M. Dubovsky 28a , E. Duchovni 169 ,
G. Duckeck 109 , O.A. Ducu 27b , D. Duda 52 , A. Dudarev 36 , E.R. Duden 26 , M. D’uffizi 101 ,
L. Duflot 66 , M. Dührssen 36 , A.E. Dumitriu 27b , M. Dunford 63a , S. Dungs 49 , K. Dunne 47a,47b ,
A. Duperrin 102 , H. Duran Yildiz 3a , M. Düren 58 , A. Durglishvili 149b , B.L. Dwyer 115 ,
G.I. Dyckes 17a , M. Dyndal 86a , B.S. Dziedzic 87 , Z.O. Earnshaw 146 , G.H. Eberwein 126 ,
B. Eckerova 28a , S. Eggebrecht 55 , E. Egidio Purcino De Souza 127 , L.F. Ehrke 56 , G. Eigen 16 ,
K. Einsweiler 17a , T. Ekelof 161 , P.A. Ekman 98 , S. El Farkh 35b , Y. El Ghazali 35b ,
H. El Jarrari 36 , A. El Moussaouy 108 , V. Ellajosyula 161 , M. Ellert 161 , F. Ellinghaus 171 ,
N. Ellis 36 , J. Elmsheuser 29 , M. Elsing 36 , D. Emeliyanov 134 , Y. Enari 153 , I. Ene 17a ,
S. Epari 13 , P.A. Erland 87 , M. Errenst 171 , M. Escalier 66 , C. Escobar 163 , E. Etzion 151 ,
G. Evans 130a , H. Evans 68 , L.S. Evans 95 , M.O. Evans 146 , A. Ezhilov 37 , S. Ezzarqtouni 35a ,
F. Fabbri 59 , L. Fabbri 23b,23a , G. Facini 96 , V. Fadeyev 136 , R.M. Fakhrutdinov 37 ,
D. Fakoudis 100 , S. Falciano 75a , L.F. Falda Ulhoa Coelho 36 , P.J. Falke 24 , J. Faltova 133 ,

99
C. Fan 162 , Y. Fan 14a , Y. Fang 14a,14e , M. Fanti 71a,71b , M. Faraj 69a,69b , Z. Farazpay 97 ,
A. Farbin 8 , A. Farilla 77a , T. Farooque 107 , S.M. Farrington 52 , F. Fassi 35e , D. Fassouliotis 9 ,
M. Faucci Giannelli 76a,76b , W.J. Fawcett 32 , L. Fayard 66 , P. Federic 133 , P. Federicova 131 ,
O.L. Fedin 37,a , M. Feickert 170 , L. Feligioni 102 , D.E. Fellers 123 , C. Feng 62b , M. Feng 14b ,
Z. Feng 114 , M.J. Fenton 159 , L. Ferencz 48 , R.A.M. Ferguson 91 , S.I. Fernandez Luengo 137f ,
P. Fernandez Martinez 13 , M.J.V. Fernoux 102 , J. Ferrando 91 , A. Ferrari 161 , P. Ferrari 114,113 ,
R. Ferrari 73a , D. Ferrere 56 , C. Ferretti 106 , F. Fiedler 100 , P. Fiedler 132 , A. Filipčič 93 ,
E.K. Filmer 1 , F. Filthaut 113 , M.C.N. Fiolhais 130a,130c,c , L. Fiorini 163 , W.C. Fisher 107 ,
T. Fitschen 101 , P.M. Fitzhugh135 , I. Fleck 141 , P. Fleischmann 106 , T. Flick 171 , M. Flores 33d,ad ,
L.R. Flores Castillo 64a , L. Flores Sanz De Acedo 36 , F.M. Follega 78a,78b , N. Fomin 16 ,
J.H. Foo 155 , A. Formica 135 , A.C. Forti 101 , E. Fortin 36 , A.W. Fortman 17a , M.G. Foti 17a ,
L. Fountas 9,j , D. Fournier 66 , H. Fox 91 , P. Francavilla 74a,74b , S. Francescato 61 ,
S. Franchellucci 56 , M. Franchini 23b,23a , S. Franchino 63a , D. Francis36 , L. Franco 113 ,
V. Franco Lima 36 , L. Franconi 48 , M. Franklin 61 , G. Frattari 26 , A.C. Freegard 94 ,
W.S. Freund 83b , Y.Y. Frid 151 , J. Friend 59 , N. Fritzsche 50 , A. Froch 54 , D. Froidevaux 36 ,
J.A. Frost 126 , Y. Fu 62a , S. Fuenzalida Garrido 137f , M. Fujimoto 102 , K.Y. Fung 64a ,
E. Furtado De Simas Filho 83b , M. Furukawa 153 , J. Fuster 163 , A. Gabrielli 23b,23a ,
A. Gabrielli 155 , P. Gadow 36 , G. Gagliardi 57b,57a , L.G. Gagnon 17a , S. Galantzan 151 ,
E.J. Gallas 126 , B.J. Gallop 134 , K.K. Gan 119 , S. Ganguly 153 , Y. Gao 52 ,
F.M. Garay Walls 137a,137b , B. Garcia29 , C. García 163 , A. Garcia Alonso 114 ,
A.G. Garcia Caffaro 172 , J.E. García Navarro 163 , M. Garcia-Sciveres 17a , G.L. Gardner 128 ,
R.W. Gardner 39 , N. Garelli 158 , D. Garg 80 , R.B. Garg 143,n , J.M. Gargan52 , C.A. Garner155 ,
C.M. Garvey 33a , P. Gaspar 83b , V.K. Gassmann158 , G. Gaudio 73a , V. Gautam13 , P. Gauzzi 75a,75b ,
I.L. Gavrilenko 37 , A. Gavrilyuk 37 , C. Gay 164 , G. Gaycken 48 , E.N. Gazis 10 , A.A. Geanta 27b ,
C.M. Gee 136 , A. Gekow119 , C. Gemme 57b , M.H. Genest 60 , A.D. Gentry 112 , S. George 95 ,
W.F. George 20 , T. Geralis 46 , P. Gessinger-Befurt 36 , M.E. Geyik 171 , M. Ghani 167 ,
M. Ghneimat 141 , K. Ghorbanian 94 , A. Ghosal 141 , A. Ghosh 159 , A. Ghosh 7 , B. Giacobbe 23b ,
S. Giagu 75a,75b , T. Giani 114 , P. Giannetti 74a , A. Giannini 62a , S.M. Gibson 95 , M. Gignac 136 ,
D.T. Gil 86b , A.K. Gilbert 86a , B.J. Gilbert 41 , D. Gillberg 34 , G. Gilles 114 , L. Ginabat 127 ,
D.M. Gingrich 2,ag , M.P. Giordani 69a,69c , P.F. Giraud 135 , G. Giugliarelli 69a,69c , D. Giugni 71a ,
F. Giuli 36 , I. Gkialas 9,j , L.K. Gladilin 37 , C. Glasman 99 , G.R. Gledhill 123 , G. Glemža 48 ,
M. Glisic123 , I. Gnesi 43b,f , Y. Go 29 , M. Goblirsch-Kolb 36 , B. Gocke 49 , D. Godin108 ,
B. Gokturk 21a , S. Goldfarb 105 , T. Golling 56 , M.G.D. Gololo 33g , D. Golubkov 37 ,
J.P. Gombas 107 , A. Gomes 130a,130b , G. Gomes Da Silva 141 , A.J. Gomez Delegido 163 ,
R. Gonçalo 130a,130c , L. Gonella 20 , A. Gongadze 149c , F. Gonnella 20 , J.L. Gonski 41 ,
R.Y. González Andana 52 , S. González de la Hoz 163 , R. Gonzalez Lopez 92 ,
C. Gonzalez Renteria 17a , M.V. Gonzalez Rodrigues 48 , R. Gonzalez Suarez 161 ,
S. Gonzalez-Sevilla 56 , G.R. Gonzalvo Rodriguez 163 , L. Goossens 36 , B. Gorini 36 ,
E. Gorini 70a,70b , A. Gorišek 93 , T.C. Gosart 128 , A.T. Goshaw 51 , M.I. Gostkin 38 ,
S. Goswami 121 , C.A. Gottardo 36 , S.A. Gotz 109 , M. Gouighri 35b , V. Goumarre 48 ,
A.G. Goussiou 138 , N. Govender 33c , I. Grabowska-Bold 86a , K. Graham 34 , E. Gramstad 125 ,
S. Grancagnolo 70a,70b , C.M. Grant1,135 , P.M. Gravila 27f , F.G. Gravili 70a,70b , H.M. Gray 17a ,
M. Greco 70a,70b , C. Grefe 24 , I.M. Gregor 48 , P. Grenier 143 , S.G. Grewe110 , C. Grieco 13 ,
A.A. Grillo 136 , K. Grimm 31 , S. Grinstein 13,t , J.-F. Grivaz 66 , E. Gross 169 ,
J. Grosse-Knetter 55 , J.C. Grundy 126 , L. Guan 106 , W. Guan 29 , C. Gubbels 164 ,
J.G.R. Guerrero Rojas 163 , G. Guerrieri 69a,69c , F. Guescini 110 , D. Guest 18 , R. Gugel 100 ,
J.A.M. Guhit 106 , A. Guida 18 , E. Guilloton 167,134 , S. Guindon 36 , F. Guo 14a,14e , J. Guo 62c ,

100
L. Guo 48 , Y. Guo 106 , R. Gupta 48 , R. Gupta 129 , S. Gurbuz 24 , S.S. Gurdasani 54 ,
G. Gustavino 36 , M. Guth 56 , P. Gutierrez 120 , L.F. Gutierrez Zagazeta 128 , M. Gutsche 50 ,
C. Gutschow 96 , C. Gwenlan 126 , C.B. Gwilliam 92 , E.S. Haaland 125 , A. Haas 117 ,
M. Habedank 48 , C. Haber 17a , H.K. Hadavand 8 , A. Hadef 50 , S. Hadzic 110 , A.I. Hagan 91 ,
J.J. Hahn 141 , E.H. Haines 96 , M. Haleem 166 , J. Haley 121 , J.J. Hall 139 , G.D. Hallewell 102 ,
L. Halser 19 , K. Hamano 165 , M. Hamer 24 , G.N. Hamity 52 , E.J. Hampshire 95 , J. Han 62b ,
K. Han 62a , L. Han 14c , L. Han 62a , S. Han 17a , Y.F. Han 155 , K. Hanagaki 84 , M. Hance 136 ,
D.A. Hangal 41 , H. Hanif 142 , M.D. Hank 128 , J.B. Hansen 42 , P.H. Hansen 42 , K. Hara 157 ,
D. Harada 56 , T. Harenberg 171 , S. Harkusha 37 , M.L. Harris 103 , Y.T. Harris 126 , J. Harrison 13 ,
N.M. Harrison 119 , P.F. Harrison167 , N.M. Hartman 110 , N.M. Hartmann 109 , Y. Hasegawa 140 ,
R. Hauser 107 , C.M. Hawkes 20 , R.J. Hawkings 36 , Y. Hayashi 153 , S. Hayashida 111 ,
D. Hayden 107 , C. Hayes 106 , R.L. Hayes 114 , C.P. Hays 126 , J.M. Hays 94 , H.S. Hayward 92 ,
F. He 62a , M. He 14a,14e , Y. He 154 , Y. He 48 , Y. He 96 , N.B. Heatley 94 , V. Hedberg 98 ,
A.L. Heggelund 125 , N.D. Hehir 94,* , C. Heidegger 54 , K.K. Heidegger 54 , W.D. Heidorn 81 ,
J. Heilman 34 , S. Heim 48 , T. Heim 17a , J.G. Heinlein 128 , J.J. Heinrich 123 , L. Heinrich 110,ae ,
J. Hejbal 131 , A. Held 170 , S. Hellesund 16 , C.M. Helling 164 , S. Hellman 47a,47b ,
R.C.W. Henderson91 , L. Henkelmann 32 , A.M. Henriques Correia36 , H. Herde 98 ,
Y. Hernández Jiménez 145 , L.M. Herrmann 24 , T. Herrmann 50 , G. Herten 54 , R. Hertenberger 109 ,
L. Hervas 36 , M.E. Hesping 100 , N.P. Hessey 156a , E. Hill 155 , S.J. Hillier 20 , J.R. Hinds 107 ,
F. Hinterkeuser 24 , M. Hirose 124 , S. Hirose 157 , D. Hirschbuehl 171 , T.G. Hitchings 101 ,
B. Hiti 93 , J. Hobbs 145 , R. Hobincu 27e , N. Hod 169 , M.C. Hodgkinson 139 , B.H. Hodkinson 32 ,
A. Hoecker 36 , D.D. Hofer 106 , J. Hofer 48 , T. Holm 24 , M. Holzbock 110 ,
L.B.A.H. Hommels 32 , B.P. Honan 101 , J. Hong 62c , T.M. Hong 129 , B.H. Hooberman 162 ,
W.H. Hopkins 6 , Y. Horii 111 , S. Hou 148 , A.S. Howard 93 , J. Howarth 59 , J. Hoya 6 ,
M. Hrabovsky 122 , A. Hrynevich 48 , T. Hryn’ova 4 , P.J. Hsu 65 , S.-C. Hsu 138 , Q. Hu 62a ,
Y.F. Hu 14a,14e , S. Huang 64b , X. Huang 14c , X. Huang 14a,14e , Y. Huang 139 , Y. Huang 14a ,
Z. Huang 101 , Z. Hubacek 132 , M. Huebner 24 , F. Huegging 24 , T.B. Huffman 126 , C.A. Hugli 48 ,
M. Huhtinen 36 , S.K. Huiberts 16 , R. Hulsken 104 , N. Huseynov 12 , J. Huston 107 , J. Huth 61 ,
R. Hyneman 143 , G. Iacobucci 56 , G. Iakovidis 29 , I. Ibragimov 141 , L. Iconomidou-Fayard 66 ,
J.P. Iddon 36 , P. Iengo 72a,72b , R. Iguchi 153 , T. Iizawa 126 , Y. Ikegami 84 , N. Ilic 155 ,
H. Imam 35a , M. Ince Lezki 56 , T. Ingebretsen Carlson 47a,47b , G. Introzzi 73a,73b , M. Iodice 77a ,
V. Ippolito 75a,75b , R.K. Irwin 92 , M. Ishino 153 , W. Islam 170 , C. Issever 18,48 , S. Istin 21a,ak ,
H. Ito 168 , R. Iuppa 78a,78b , A. Ivina 169 , J.M. Izen 45 , V. Izzo 72a , P. Jacka 131,132 , P. Jackson 1 ,
B.P. Jaeger 142 , C.S. Jagfeld 109 , G. Jain 156a , P. Jain 54 , K. Jakobs 54 , T. Jakoubek 169 ,
J. Jamieson 59 , K.W. Janas 86a , M. Javurkova 103 , L. Jeanty 123 , J. Jejelava 149a,aa , P. Jenni 54,g ,
C.E. Jessiman 34 , C. Jia62b , J. Jia 145 , X. Jia 61 , X. Jia 14a,14e , Z. Jia 14c , S. Jiggins 48 ,
J. Jimenez Pena 13 , S. Jin 14c , A. Jinaru 27b , O. Jinnouchi 154 , P. Johansson 139 , K.A. Johns 7 ,
J.W. Johnson 136 , D.M. Jones 32 , E. Jones 48 , P. Jones 32 , R.W.L. Jones 91 , T.J. Jones 92 ,
H.L. Joos 55,36 , R. Joshi 119 , J. Jovicevic 15 , X. Ju 17a , J.J. Junggeburth 103 , T. Junkermann 63a ,
A. Juste Rozas 13,t , M.K. Juzek 87 , S. Kabana 137e , A. Kaczmarska 87 , M. Kado 110 ,
H. Kagan 119 , M. Kagan 143 , A. Kahn41 , A. Kahn 128 , C. Kahra 100 , T. Kaji 153 ,
E. Kajomovitz 150 , N. Kakati 169 , I. Kalaitzidou 54 , C.W. Kalderon 29 , A. Kamenshchikov 155 ,
N.J. Kang 136 , D. Kar 33g , K. Karava 126 , M.J. Kareem 156b , E. Karentzos 54 , I. Karkanias 152 ,
O. Karkout 114 , S.N. Karpov 38 , Z.M. Karpova 38 , V. Kartvelishvili 91 , A.N. Karyukhin 37 ,
E. Kasimi 152 , J. Katzy 48 , S. Kaur 34 , K. Kawade 140 , M.P. Kawale 120 , C. Kawamoto 88 ,
T. Kawamoto 62a , E.F. Kay 36 , F.I. Kaya 158 , S. Kazakos 107 , V.F. Kazanin 37 , Y. Ke 145 ,
J.M. Keaveney 33a , R. Keeler 165 , G.V. Kehris 61 , J.S. Keller 34 , A.S. Kelly96 , J.J. Kempster 146 ,

101
P.D. Kennedy 100 , O. Kepka 131 , B.P. Kerridge 134 , S. Kersten 171 , B.P. Kerševan 93 ,
S. Keshri 66 , L. Keszeghova 28a , S. Ketabchi Haghighat 155 , R.A. Khan 129 , A. Khanov 121 ,
A.G. Kharlamov 37 , T. Kharlamova 37 , E.E. Khoda 138 , M. Kholodenko 37 , T.J. Khoo 18 ,
G. Khoriauli 166 , J. Khubua 149b , Y.A.R. Khwaira 66 , B. Kibirige33g , A. Kilgallon 123 ,
D.W. Kim 47a,47b , Y.K. Kim 39 , N. Kimura 96 , M.K. Kingston 55 , A. Kirchhoff 55 , C. Kirfel 24 ,
F. Kirfel 24 , J. Kirk 134 , A.E. Kiryunin 110 , C. Kitsaki 10 , O. Kivernyk 24 , M. Klassen 63a ,
C. Klein 34 , L. Klein 166 , M.H. Klein 44 , S.B. Klein 56 , U. Klein 92 , P. Klimek 36 ,
A. Klimentov 29 , T. Klioutchnikova 36 , P. Kluit 114 , S. Kluth 110 , E. Kneringer 79 ,
T.M. Knight 155 , A. Knue 49 , R. Kobayashi 88 , D. Kobylianskii 169 , S.F. Koch 126 ,
M. Kocian 143 , P. Kodyš 133 , D.M. Koeck 123 , P.T. Koenig 24 , T. Koffas 34 , O. Kolay 50 ,
I. Koletsou 4 , T. Komarek 122 , K. Köneke 54 , A.X.Y. Kong 1 , T. Kono 118 , N. Konstantinidis 96 ,
P. Kontaxakis 56 , B. Konya 98 , R. Kopeliansky 68 , S. Koperny 86a , K. Korcyl 87 , K. Kordas 152,e ,
A. Korn 96 , S. Korn 55 , I. Korolkov 13 , N. Korotkova 37 , B. Kortman 114 , O. Kortner 110 ,
S. Kortner 110 , W.H. Kostecka 115 , V.V. Kostyukhin 141 , A. Kotsokechagia 135 , A. Kotwal 51 ,
A. Koulouris 36 , A. Kourkoumeli-Charalampidi 73a,73b , C. Kourkoumelis 9 , E. Kourlitis 110,ae ,
O. Kovanda 123 , R. Kowalewski 165 , W. Kozanecki 135 , A.S. Kozhin 37 , V.A. Kramarenko 37 ,
G. Kramberger 93 , P. Kramer 100 , M.W. Krasny 127 , A. Krasznahorkay 36 , J.W. Kraus 171 ,
J.A. Kremer 48 , T. Kresse 50 , J. Kretzschmar 92 , K. Kreul 18 , P. Krieger 155 ,
S. Krishnamurthy 103 , M. Krivos 133 , K. Krizka 20 , K. Kroeninger 49 , H. Kroha 110 , J. Kroll 131 ,
J. Kroll 128 , K.S. Krowpman 107 , U. Kruchonak 38 , H. Krüger 24 , N. Krumnack81 , M.C. Kruse 51 ,
O. Kuchinskaia 37 , S. Kuday 3a , S. Kuehn 36 , R. Kuesters 54 , T. Kuhl 48 , V. Kukhtin 38 ,
Y. Kulchitsky 37,a , S. Kuleshov 137d,137b , M. Kumar 33g , N. Kumari 48 , P. Kumari 156b ,
A. Kupco 131 , T. Kupfer49 , A. Kupich 37 , O. Kuprash 54 , H. Kurashige 85 , L.L. Kurchaninov 156a ,
O. Kurdysh 66 , Y.A. Kurochkin 37 , A. Kurova 37 , M. Kuze 154 , A.K. Kvam 103 , J. Kvita 122 ,
T. Kwan 104 , N.G. Kyriacou 106 , L.A.O. Laatu 102 , C. Lacasta 163 , F. Lacava 75a,75b ,
H. Lacker 18 , D. Lacour 127 , N.N. Lad 96 , E. Ladygin 38 , B. Laforge 127 , T. Lagouri 27b ,
F.Z. Lahbabi 35a , S. Lai 55 , I.K. Lakomiec 86a , N. Lalloue 60 , J.E. Lambert 165 , S. Lammers 68 ,
W. Lampl 7 , C. Lampoudis 152,e , A.N. Lancaster 115 , E. Lançon 29 , U. Landgraf 54 ,
M.P.J. Landon 94 , V.S. Lang 54 , O.K.B. Langrekken 125 , A.J. Lankford 159 , F. Lanni 36 ,
K. Lantzsch 24 , A. Lanza 73a , A. Lapertosa 57b,57a , J.F. Laporte 135 , T. Lari 71a ,
F. Lasagni Manghi 23b , M. Lassnig 36 , V. Latonova 131 , A. Laudrain 100 , A. Laurier 150 ,
S.D. Lawlor 139 , Z. Lawrence 101 , R. Lazaridou167 , M. Lazzaroni 71a,71b , B. Le101 ,
E.M. Le Boulicaut 51 , B. Leban 93 , A. Lebedev 81 , M. LeBlanc 101 , F. Ledroit-Guillon 60 ,
A.C.A. Lee96 , S.C. Lee 148 , S. Lee 47a,47b , T.F. Lee 92 , L.L. Leeuw 33c , H.P. Lefebvre 95 ,
M. Lefebvre 165 , C. Leggett 17a , G. Lehmann Miotto 36 , M. Leigh 56 , W.A. Leight 103 ,
W. Leinonen 113 , A. Leisos 152,s , M.A.L. Leite 83c , C.E. Leitgeb 18 , R. Leitner 133 ,
K.J.C. Leney 44 , T. Lenz 24 , S. Leone 74a , C. Leonidopoulos 52 , A. Leopold 144 , C. Leroy 108 ,
R. Les 107 , C.G. Lester 32 , M. Levchenko 37 , J. Levêque 4 , L.J. Levinson 169 , G. Levrini 23b,23a ,
M.P. Lewicki 87 , D.J. Lewis 4 , A. Li 5 , B. Li 62b , C. Li62a , C-Q. Li 110 , H. Li 62a , H. Li 62b ,
H. Li 14c , H. Li 14b , H. Li 62b , J. Li 62c , K. Li 138 , L. Li 62c , M. Li 14a,14e , Q.Y. Li 62a ,
S. Li 14a,14e , S. Li 62d,62c,d , T. Li 5 , X. Li 104 , Z. Li 126 , Z. Li 104 , Z. Li 14a,14e , S. Liang14a,14e ,
Z. Liang 14a , M. Liberatore 135 , B. Liberti 76a , K. Lie 64c , J. Lieber Marin 83b , H. Lien 68 ,
K. Lin 107 , R.E. Lindley 7 , J.H. Lindon 2 , E. Lipeles 128 , A. Lipniacka 16 , A. Lister 164 ,
J.D. Little 4 , B. Liu 14a , B.X. Liu 142 , D. Liu 62d,62c , J.B. Liu 62a , J.K.K. Liu 32 , K. Liu 62d,62c ,
M. Liu 62a , M.Y. Liu 62a , P. Liu 14a , Q. Liu 62d,138,62c , X. Liu 62a , X. Liu 62b , Y. Liu 14d,14e ,
Y.L. Liu 62b , Y.W. Liu 62a , J. Llorente Merino 142 , S.L. Lloyd 94 , E.M. Lobodzinska 48 ,
P. Loch 7 , T. Lohse 18 , K. Lohwasser 139 , E. Loiacono 48 , M. Lokajicek 131,* , J.D. Lomas 20 ,

102
J.D. Long 162 , I. Longarini 159 , L. Longo 70a,70b , R. Longo 162 , I. Lopez Paz 67 ,
A. Lopez Solis 48 , N. Lorenzo Martinez 4 , A.M. Lory 109 , G. Löschcke Centeno 146 , O. Loseva 37 ,
X. Lou 47a,47b , X. Lou 14a,14e , A. Lounis 66 , P.A. Love 91 , G. Lu 14a,14e , M. Lu 80 , S. Lu 128 ,
Y.J. Lu 65 , H.J. Lubatti 138 , C. Luci 75a,75b , F.L. Lucio Alves 14c , F. Luehring 68 , I. Luise 145 ,
O. Lukianchuk 66 , O. Lundberg 144 , B. Lund-Jensen 144,* , N.A. Luongo 6 , M.S. Lutz 36 ,
A.B. Lux 25 , D. Lynn 29 , R. Lysak 131 , E. Lytken 98 , V. Lyubushkin 38 , T. Lyubushkina 38 ,
M.M. Lyukova 145 , H. Ma 29 , K. Ma 62a , L.L. Ma 62b , W. Ma 62a , Y. Ma 121 ,
D.M. Mac Donell 165 , G. Maccarrone 53 , J.C. MacDonald 100 , P.C. Machado De Abreu Farias 83b ,
R. Madar 40 , W.F. Mader 50 , T. Madula 96 , J. Maeda 85 , T. Maeno 29 , H. Maguire 139 ,
V. Maiboroda 135 , A. Maio 130a,130b,130d , K. Maj 86a , O. Majersky 48 , S. Majewski 123 ,
N. Makovec 66 , V. Maksimovic 15 , B. Malaescu 127 , Pa. Malecki 87 , V.P. Maleev 37 ,
F. Malek 60,o , M. Mali 93 , D. Malito 95 , U. Mallik 80 , S. Maltezos10 , S. Malyukov38 ,
J. Mamuzic 13 , G. Mancini 53 , M.N. Mancini 26 , G. Manco 73a,73b , J.P. Mandalia 94 ,
I. Mandić 93 , L. Manhaes de Andrade Filho 83a , I.M. Maniatis 169 , J. Manjarres Ramos 102,ab ,
D.C. Mankad 169 , A. Mann 109 , S. Manzoni 36 , L. Mao 62c , X. Mapekula 33c , A. Marantis 152,s ,
G. Marchiori 5 , M. Marcisovsky 131 , C. Marcon 71a , M. Marinescu 20 , S. Marium 48 ,
M. Marjanovic 120 , E.J. Marshall 91 , Z. Marshall 17a , S. Marti-Garcia 163 , T.A. Martin 167 ,
V.J. Martin 52 , B. Martin dit Latour 16 , L. Martinelli 75a,75b , M. Martinez 13,t ,
P. Martinez Agullo 163 , V.I. Martinez Outschoorn 103 , P. Martinez Suarez 13 , S. Martin-Haugh 134 ,
V.S. Martoiu 27b , A.C. Martyniuk 96 , A. Marzin 36 , D. Mascione 78a,78b , L. Masetti 100 ,
T. Mashimo 153 , J. Masik 101 , A.L. Maslennikov 37 , P. Massarotti 72a,72b , P. Mastrandrea 74a,74b ,
A. Mastroberardino 43b,43a , T. Masubuchi 153 , T. Mathisen 161 , J. Matousek 133 , N. Matsuzawa153 ,
J. Maurer 27b , B. Maček 93 , D.A. Maximov 37 , R. Mazini 148 , I. Maznas 115 , M. Mazza 107 ,
S.M. Mazza 136 , E. Mazzeo 71a,71b , C. Mc Ginn 29 , J.P. Mc Gowan 104 , S.P. Mc Kee 106 ,
C.C. McCracken 164 , E.F. McDonald 105 , A.E. McDougall 114 , J.A. Mcfayden 146 ,
R.P. McGovern 128 , G. Mchedlidze 149b , R.P. Mckenzie 33g , T.C. Mclachlan 48 ,
D.J. Mclaughlin 96 , S.J. McMahon 134 , C.M. Mcpartland 92 , R.A. McPherson 165,x ,
S. Mehlhase 109 , A. Mehta 92 , D. Melini 163 , B.R. Mellado Garcia 33g , A.H. Melo 55 ,
F. Meloni 48 , A.M. Mendes Jacques Da Costa 101 , H.Y. Meng 155 , L. Meng 91 , S. Menke 110 ,
M. Mentink 36 , E. Meoni 43b,43a , G. Mercado 115 , C. Merlassino 69a,69c , L. Merola 72a,72b ,
C. Meroni 71a,71b , J. Metcalfe 6 , A.S. Mete 6 , C. Meyer 68 , J-P. Meyer 135 , R.P. Middleton 134 ,
L. Mijović 52 , G. Mikenberg 169 , M. Mikestikova 131 , M. Mikuž 93 , H. Mildner 100 , A. Milic 36 ,
D.W. Miller 39 , E.H. Miller 143 , L.S. Miller 34 , A. Milov 169 , D.A. Milstead47a,47b , T. Min14c ,
A.A. Minaenko 37 , I.A. Minashvili 149b , L. Mince 59 , A.I. Mincer 117 , B. Mindur 86a ,
M. Mineev 38 , Y. Mino 88 , L.M. Mir 13 , M. Miralles Lopez 59 , M. Mironova 17a , A. Mishima153 ,
M.C. Missio 113 , A. Mitra 167 , V.A. Mitsou 163 , Y. Mitsumori 111 , O. Miu 155 ,
P.S. Miyagawa 94 , T. Mkrtchyan 63a , M. Mlinarevic 96 , T. Mlinarevic 96 , M. Mlynarikova 36 ,
S. Mobius 19 , P. Mogg 109 , M.H. Mohamed Farook 112 , A.F. Mohammed 14a,14e , S. Mohapatra 41 ,
G. Mokgatitswane 33g , L. Moleri 169 , B. Mondal 141 , S. Mondal 132 , K. Mönig 48 ,
E. Monnier 102 , L. Monsonis Romero163 , J. Montejo Berlingen 13 , M. Montella 119 ,
F. Montereali 77a,77b , F. Monticelli 90 , S. Monzani 69a,69c , N. Morange 66 ,
A.L. Moreira De Carvalho 130a , M. Moreno Llácer 163 , C. Moreno Martinez 56 , P. Morettini 57b ,
S. Morgenstern 36 , M. Morii 61 , M. Morinaga 153 , F. Morodei 75a,75b , L. Morvaj 36 ,
P. Moschovakos 36 , B. Moser 36 , M. Mosidze 149b , T. Moskalets 54 , P. Moskvitina 113 ,
J. Moss 31,l , E.J.W. Moyse 103 , O. Mtintsilana 33g , S. Muanza 102 , J. Mueller 129 ,
D. Muenstermann 91 , R. Müller 19 , G.A. Mullier 161 , A.J. Mullin32 , J.J. Mullin128 , D.P. Mungo 155 ,
D. Munoz Perez 163 , F.J. Munoz Sanchez 101 , M. Murin 101 , W.J. Murray 167,134 ,

103
M. Muškinja 17a , C. Mwewa 29 , A.G. Myagkov 37,a , A.J. Myers 8 , G. Myers 68 , M. Myska 132 ,
B.P. Nachman 17a , O. Nackenhorst 49 , K. Nagai 126 , K. Nagano 84 , J.L. Nagle 29,ai , E. Nagy 102 ,
A.M. Nairz 36 , Y. Nakahama 84 , K. Nakamura 84 , T. Nakamura 85 , K. Nakkalil 5 , H. Nanjo 124 ,
R. Narayan 44 , E.A. Narayanan 112 , I. Naryshkin 37 , M. Naseri 34 , S. Nasri 116b , C. Nass 24 ,
G. Navarro 22a , J. Navarro-Gonzalez 163 , R. Nayak 151 , A. Nayaz 18 , P.Y. Nechaeva 37 ,
F. Nechansky 48 , L. Nedic 126 , T.J. Neep 20 , A. Negri 73a,73b , M. Negrini 23b , C. Nellist 114 ,
C. Nelson 104 , K. Nelson 106 , S. Nemecek 131 , M. Nessi 36,h , M.S. Neubauer 162 ,
F. Neuhaus 100 , J. Neundorf 48 , R. Newhouse 164 , P.R. Newman 20 , C.W. Ng 129 , Y.W.Y. Ng 48 ,
B. Ngair 116a , H.D.N. Nguyen 108 , R.B. Nickerson 126 , R. Nicolaidou 135 , J. Nielsen 136 ,
M. Niemeyer 55 , J. Niermann 55,36 , N. Nikiforou 36 , V. Nikolaenko 37,a , I. Nikolic-Audit 127 ,
K. Nikolopoulos 20 , P. Nilsson 29 , I. Ninca 48 , H.R. Nindhito 56 , G. Ninio 151 , A. Nisati 75a ,
N. Nishu 2 , R. Nisius 110 , J-E. Nitschke 50 , E.K. Nkadimeng 33g , T. Nobe 153 , D.L. Noel 32 ,
T. Nommensen 147 , M.B. Norfolk 139 , R.R.B. Norisam 96 , B.J. Norman 34 , M. Noury 35a ,
J. Novak 93 , T. Novak 48 , L. Novotny 132 , R. Novotny 112 , L. Nozka 122 , K. Ntekas 159 ,
N.M.J. Nunes De Moura Junior 83b , E. Nurse96 , J. Ocariz 127 , A. Ochi 85 , I. Ochoa 130a ,
S. Oerdek 48,u , J.T. Offermann 39 , A. Ogrodnik 133 , A. Oh 101 , C.C. Ohm 144 , H. Oide 84 ,
R. Oishi 153 , M.L. Ojeda 48 , Y. Okumura 153 , L.F. Oleiro Seabra 130a , S.A. Olivares Pino 137d ,
D. Oliveira Damazio 29 , D. Oliveira Goncalves 83a , J.L. Oliver 159 , Ö.O. Öncel 54 ,
A.P. O’Neill 19 , A. Onofre 130a,130e , P.U.E. Onyisi 11 , M.J. Oreglia 39 , G.E. Orellana 90 ,
D. Orestano 77a,77b , N. Orlando 13 , R.S. Orr 155 , V. O’Shea 59 , L.M. Osojnak 128 ,
R. Ospanov 62a , G. Otero y Garzon 30 , H. Otono 89 , P.S. Ott 63a , G.J. Ottino 17a , M. Ouchrif 35d ,
F. Ould-Saada 125 , M. Owen 59 , R.E. Owen 134 , K.Y. Oyulmaz 21a , V.E. Ozcan 21a ,
F. Ozturk 87 , N. Ozturk 8 , S. Ozturk 82 , H.A. Pacey 126 , A. Pacheco Pages 13 ,
C. Padilla Aranda 13 , G. Padovano 75a,75b , S. Pagan Griso 17a , G. Palacino 68 , A. Palazzo 70a,70b ,
J. Pampel 24 , J. Pan 172 , T. Pan 64a , D.K. Panchal 11 , C.E. Pandini 114 , J.G. Panduro Vazquez 95 ,
H.D. Pandya 1 , H. Pang 14b , P. Pani 48 , G. Panizzo 69a,69c , L. Paolozzi 56 , S. Parajuli 162 ,
A. Paramonov 6 , C. Paraskevopoulos 53 , D. Paredes Hernandez 64b , A. Pareti 73a,73b , K.R. Park 41 ,
T.H. Park 155 , M.A. Parker 32 , F. Parodi 57b,57a , E.W. Parrish 115 , V.A. Parrish 52 ,
J.A. Parsons 41 , U. Parzefall 54 , B. Pascual Dias 108 , L. Pascual Dominguez 151 ,
E. Pasqualucci 75a , S. Passaggio 57b , F. Pastore 95 , P. Patel 87 , U.M. Patel 51 , J.R. Pater 101 ,
T. Pauly 36 , C.I. Pazos 158 , J. Pearkes 143 , M. Pedersen 125 , R. Pedro 130a , S.V. Peleganchuk 37 ,
O. Penc 36 , E.A. Pender 52 , G.D. Penn 172 , K.E. Penski 109 , M. Penzin 37 , B.S. Peralva 83d ,
A.P. Pereira Peixoto 60 , L. Pereira Sanchez 47a,47b , D.V. Perepelitsa 29,ai , E. Perez Codina 156a ,
M. Perganti 10 , H. Pernegger 36 , O. Perrin 40 , K. Peters 48 , R.F.Y. Peters 101 , B.A. Petersen 36 ,
T.C. Petersen 42 , E. Petit 102 , V. Petousis 132 , C. Petridou 152,e , A. Petrukhin 141 , M. Pettee 17a ,
N.E. Pettersson 36 , A. Petukhov 37 , K. Petukhova 133 , R. Pezoa 137f , L. Pezzotti 36 ,
G. Pezzullo 172 , T.M. Pham 170 , T. Pham 105 , P.W. Phillips 134 , G. Piacquadio 145 ,
E. Pianori 17a , F. Piazza 123 , R. Piegaia 30 , D. Pietreanu 27b , A.D. Pilkington 101 ,
M. Pinamonti 69a,69c , J.L. Pinfold 2 , B.C. Pinheiro Pereira 130a , A.E. Pinto Pinoargote 100,135 ,
L. Pintucci 69a,69c , K.M. Piper 146 , A. Pirttikoski 56 , D.A. Pizzi 34 , L. Pizzimento 64b ,
A. Pizzini 114 , M.-A. Pleier 29 , V. Plesanovs54 , V. Pleskot 133 , E. Plotnikova38 , G. Poddar 4 ,
R. Poettgen 98 , L. Poggioli 127 , I. Pokharel 55 , S. Polacek 133 , G. Polesello 73a , A. Poley 142,156a ,
A. Polini 23b , C.S. Pollard 167 , Z.B. Pollock 119 , E. Pompa Pacchi 75a,75b , D. Ponomarenko 113 ,
L. Pontecorvo 36 , S. Popa 27a , G.A. Popeneciu 27d , A. Poreba 36 , D.M. Portillo Quintero 156a ,
S. Pospisil 132 , M.A. Postill 139 , P. Postolache 27c , K. Potamianos 167 , P.A. Potepa 86a ,
I.N. Potrap 38 , C.J. Potter 32 , H. Potti 1 , T. Poulsen 48 , J. Poveda 163 , M.E. Pozo Astigarraga 36 ,
A. Prades Ibanez 163 , J. Pretel 54 , D. Price 101 , M. Primavera 70a , M.A. Principe Martin 99 ,

104
R. Privara 122 , T. Procter 59 , M.L. Proffitt 138 , N. Proklova 128 , K. Prokofiev 64c , G. Proto 110 ,
J. Proudfoot 6 , M. Przybycien 86a , W.W. Przygoda 86b , A. Psallidas 46 , J.E. Puddefoot 139 ,
D. Pudzha 37 , D. Pyatiizbyantseva 37 , J. Qian 106 , D. Qichen 101 , Y. Qin 101 , T. Qiu 52 ,
A. Quadt 55 , M. Queitsch-Maitland 101 , G. Quetant 56 , R.P. Quinn 164 , G. Rabanal Bolanos 61 ,
D. Rafanoharana 54 , F. Ragusa 71a,71b , J.L. Rainbolt 39 , J.A. Raine 56 , S. Rajagopalan 29 ,
E. Ramakoti 37 , I.A. Ramirez-Berend 34 , K. Ran 48,14e , N.P. Rapheeha 33g , H. Rasheed 27b ,
V. Raskina 127 , D.F. Rassloff 63a , A. Rastogi 17a , S. Rave 100 , B. Ravina 55 , I. Ravinovich 169 ,
M. Raymond 36 , A.L. Read 125 , N.P. Readioff 139 , D.M. Rebuzzi 73a,73b , G. Redlinger 29 ,
A.S. Reed 110 , K. Reeves 26 , J.A. Reidelsturz 171 , D. Reikher 151 , A. Rej 49 , C. Rembser 36 ,
M. Renda 27b , M.B. Rendel110 , F. Renner 48 , A.G. Rennie 159 , A.L. Rescia 48 , S. Resconi 71a ,
M. Ressegotti 57b,57a , S. Rettie 36 , J.G. Reyes Rivera 107 , E. Reynolds 17a , O.L. Rezanova 37 ,
P. Reznicek 133 , H. Riani 35d , N. Ribaric 91 , E. Ricci 78a,78b , R. Richter 110 , S. Richter 47a,47b ,
E. Richter-Was 86b , M. Ridel 127 , S. Ridouani 35d , P. Rieck 117 , P. Riedler 36 , E.M. Riefel 47a,47b ,
J.O. Rieger 114 , M. Rijssenbeek 145 , M. Rimoldi 36 , L. Rinaldi 23b,23a , T.T. Rinn 29 ,
M.P. Rinnagel 109 , G. Ripellino 161 , I. Riu 13 , J.C. Rivera Vergara 165 , F. Rizatdinova 121 ,
E. Rizvi 94 , B.A. Roberts 167 , B.R. Roberts 17a , S.H. Robertson 104,x , D. Robinson 32 ,
C.M. Robles Gajardo137f , M. Robles Manzano 100 , A. Robson 59 , A. Rocchi 76a,76b , C. Roda 74a,74b ,
S. Rodriguez Bosca 63a , Y. Rodriguez Garcia 22a , A. Rodriguez Rodriguez 54 ,
A.M. Rodríguez Vera 156b , S. Roe36 , J.T. Roemer 159 , A.R. Roepe-Gier 136 , J. Roggel 171 ,
O. Røhne 125 , R.A. Rojas 103 , C.P.A. Roland 127 , J. Roloff 29 , A. Romaniouk 37 ,
E. Romano 73a,73b , M. Romano 23b , A.C. Romero Hernandez 162 , N. Rompotis 92 , L. Roos 127 ,
S. Rosati 75a , B.J. Rosser 39 , E. Rossi 126 , E. Rossi 72a,72b , L.P. Rossi 57b , L. Rossini 54 ,
R. Rosten 119 , M. Rotaru 27b , B. Rottler 54 , C. Rougier 102,ab , D. Rousseau 66 , D. Rousso 32 ,
A. Roy 162 , S. Roy-Garand 155 , A. Rozanov 102 , Z.M.A. Rozario 59 , Y. Rozen 150 ,
A. Rubio Jimenez 163 , A.J. Ruby 92 , V.H. Ruelas Rivera 18 , T.A. Ruggeri 1 , A. Ruggiero 126 ,
A. Ruiz-Martinez 163 , A. Rummler 36 , Z. Rurikova 54 , N.A. Rusakovich 38 , H.L. Russell 165 ,
G. Russo 75a,75b , J.P. Rutherfoord 7 , S. Rutherford Colmenares 32 , K. Rybacki91 , M. Rybar 133 ,
E.B. Rye 125 , A. Ryzhov 44 , J.A. Sabater Iglesias 56 , P. Sabatini 163 , H.F-W. Sadrozinski 136 ,
F. Safai Tehrani 75a , B. Safarzadeh Samani 134 , M. Safdari 143 , S. Saha 165 , M. Sahinsoy 110 ,
A. Saibel 163 , M. Saimpert 135 , M. Saito 153 , T. Saito 153 , D. Salamani 36 , A. Salnikov 143 ,
J. Salt 163 , A. Salvador Salas 151 , D. Salvatore 43b,43a , F. Salvatore 146 , A. Salzburger 36 ,
D. Sammel 54 , D. Sampsonidis 152,e , D. Sampsonidou 123 , J. Sánchez 163 ,
V. Sanchez Sebastian 163 , H. Sandaker 125 , C.O. Sander 48 , J.A. Sandesara 103 , M. Sandhoff 171 ,
C. Sandoval 22b , D.P.C. Sankey 134 , T. Sano 88 , A. Sansoni 53 , L. Santi 75a,75b , C. Santoni 40 ,
H. Santos 130a,130b , A. Santra 169 , K.A. Saoucha 160 , J.G. Saraiva 130a,130d , J. Sardain 7 ,
A. Sarkar 56 , O. Sasaki 84 , K. Sato 157 , C. Sauer63b , F. Sauerburger 54 , E. Sauvan 4 ,
P. Savard 155,ag , R. Sawada 153 , C. Sawyer 134 , L. Sawyer 97 , I. Sayago Galvan163 , C. Sbarra 23b ,
A. Sbrizzi 23b,23a , T. Scanlon 96 , J. Schaarschmidt 138 , U. Schäfer 100 , A.C. Schaffer 66,44 ,
D. Schaile 109 , R.D. Schamberger 145 , C. Scharf 18 , M.M. Schefer 19 , V.A. Schegelsky 37 ,
D. Scheirich 133 , F. Schenck 18 , M. Schernau 159 , C. Scheulen 55 , C. Schiavi 57b,57a ,
E.J. Schioppa 70a,70b , M. Schioppa 43b,43a , B. Schlag 143,n , K.E. Schleicher 54 , S. Schlenker 36 ,
J. Schmeing 171 , M.A. Schmidt 171 , K. Schmieden 100 , C. Schmitt 100 , N. Schmitt 100 ,
S. Schmitt 48 , L. Schoeffel 135 , A. Schoening 63b , P.G. Scholer 54 , E. Schopf 126 , M. Schott 100 ,
J. Schovancova 36 , S. Schramm 56 , T. Schroer 56 , H-C. Schultz-Coulon 63a , M. Schumacher 54 ,
B.A. Schumm 136 , Ph. Schune 135 , A.J. Schuy 138 , H.R. Schwartz 136 , A. Schwartzman 143 ,
T.A. Schwarz 106 , Ph. Schwemling 135 , R. Schwienhorst 107 , A. Sciandra 136 , G. Sciolla 26 ,
F. Scuri 74a , C.D. Sebastiani 92 , K. Sedlaczek 115 , P. Seema 18 , S.C. Seidel 112 , A. Seiden 136 ,

105
B.D. Seidlitz 41 , C. Seitz 48 , J.M. Seixas 83b , G. Sekhniaidze 72a , L. Selem 60 ,
N. Semprini-Cesari 23b,23a , D. Sengupta 56 , V. Senthilkumar 163 , L. Serin 66 , L. Serkin 69a,69b ,
M. Sessa 76a,76b , H. Severini 120 , F. Sforza 57b,57a , A. Sfyrla 56 , Q. Sha 14a , E. Shabalina 55 ,
R. Shaheen 144 , J.D. Shahinian 128 , D. Shaked Renous 169 , L.Y. Shan 14a , M. Shapiro 17a ,
A. Sharma 36 , A.S. Sharma 164 , P. Sharma 80 , P.B. Shatalov 37 , K. Shaw 146 , S.M. Shaw 101 ,
A. Shcherbakova 37 , Q. Shen 62c,5 , D.J. Sheppard 142 , P. Sherwood 96 , L. Shi 96 , X. Shi 14a ,
C.O. Shimmin 172 , J.D. Shinner 95 , I.P.J. Shipsey 126 , S. Shirabe 89 , M. Shiyakova 38,v ,
J. Shlomi 169 , M.J. Shochet 39 , J. Shojaii 105 , D.R. Shope 125 , B. Shrestha 120 , S. Shrestha 119,aj ,
E.M. Shrif 33g , M.J. Shroff 165 , P. Sicho 131 , A.M. Sickles 162 , E. Sideras Haddad 33g ,
A. Sidoti 23b , F. Siegert 50 , Dj. Sijacki 15 , F. Sili 90 , J.M. Silva 20 , M.V. Silva Oliveira 29 ,
S.B. Silverstein 47a , S. Simion66 , R. Simoniello 36 , E.L. Simpson 59 , H. Simpson 146 ,
L.R. Simpson 106 , N.D. Simpson98 , S. Simsek 82 , S. Sindhu 55 , P. Sinervo 155 , S. Singh 155 ,
S. Sinha 48 , S. Sinha 101 , M. Sioli 23b,23a , I. Siral 36 , E. Sitnikova 48 , J. Sjölin 47a,47b ,
A. Skaf 55 , E. Skorda 20 , P. Skubic 120 , M. Slawinska 87 , V. Smakhtin169 , B.H. Smart 134 ,
S.Yu. Smirnov 37 , Y. Smirnov 37 , L.N. Smirnova 37,a , O. Smirnova 98 , A.C. Smith 41 ,
E.A. Smith 39 , H.A. Smith 126 , J.L. Smith 92 , R. Smith143 , M. Smizanska 91 , K. Smolek 132 ,
A.A. Snesarev 37 , S.R. Snider 155 , H.L. Snoek 114 , S. Snyder 29 , R. Sobie 165,x , A. Soffer 151 ,
C.A. Solans Sanchez 36 , E.Yu. Soldatov 37 , U. Soldevila 163 , A.A. Solodkov 37 , S. Solomon 26 ,
A. Soloshenko 38 , K. Solovieva 54 , O.V. Solovyanov 40 , V. Solovyev 37 , P. Sommer 36 ,
A. Sonay 13 , W.Y. Song 156b , A. Sopczak 132 , A.L. Sopio 96 , F. Sopkova 28b , J.D. Sorenson 112 ,
I.R. Sotarriva Alvarez 154 , V. Sothilingam63a , O.J. Soto Sandoval 137c,137b , S. Sottocornola 68 ,
R. Soualah 160 , Z. Soumaimi 35e , D. South 48 , N. Soybelman 169 , S. Spagnolo 70a,70b ,
M. Spalla 110 , D. Sperlich 54 , G. Spigo 36 , S. Spinali 91 , D.P. Spiteri 59 , M. Spousta 133 ,
E.J. Staats 34 , R. Stamen 63a , A. Stampekis 20 , M. Standke 24 , E. Stanecka 87 , M.V. Stange 50 ,
B. Stanislaus 17a , M.M. Stanitzki 48 , B. Stapf 48 , E.A. Starchenko 37 , G.H. Stark 136 ,
J. Stark 102,ab , P. Staroba 131 , P. Starovoitov 63a , S. Stärz 104 , R. Staszewski 87 ,
G. Stavropoulos 46 , J. Steentoft 161 , P. Steinberg 29 , B. Stelzer 142,156a , H.J. Stelzer 129 ,
O. Stelzer-Chilton 156a , H. Stenzel 58 , T.J. Stevenson 146 , G.A. Stewart 36 , J.R. Stewart 121 ,
M.C. Stockton 36 , G. Stoicea 27b , M. Stolarski 130a , S. Stonjek 110 , A. Straessner 50 ,
J. Strandberg 144 , S. Strandberg 47a,47b , M. Stratmann 171 , M. Strauss 120 , T. Strebler 102 ,
P. Strizenec 28b , R. Ströhmer 166 , D.M. Strom 123 , R. Stroynowski 44 , A. Strubig 47a,47b ,
S.A. Stucci 29 , B. Stugu 16 , J. Stupak 120 , N.A. Styles 48 , D. Su 143 , S. Su 62a , W. Su 62d ,
X. Su 62a,66 , K. Sugizaki 153 , V.V. Sulin 37 , M.J. Sullivan 92 , D.M.S. Sultan 78a,78b ,
L. Sultanaliyeva 37 , S. Sultansoy 3b , T. Sumida 88 , S. Sun 106 , S. Sun 170 ,
O. Sunneborn Gudnadottir 161 , N. Sur 102 , M.R. Sutton 146 , H. Suzuki 157 , M. Svatos 131 ,
M. Swiatlowski 156a , T. Swirski 166 , I. Sykora 28a , M. Sykora 133 , T. Sykora 133 , D. Ta 100 ,
K. Tackmann 48,u , A. Taffard 159 , R. Tafirout 156a , J.S. Tafoya Vargas 66 , Y. Takubo 84 ,
M. Talby 102 , A.A. Talyshev 37 , K.C. Tam 64b , N.M. Tamir151 , A. Tanaka 153 , J. Tanaka 153 ,
R. Tanaka 66 , M. Tanasini 57b,57a , Z. Tao 164 , S. Tapia Araya 137f , S. Tapprogge 100 ,
A. Tarek Abouelfadl Mohamed 107 , S. Tarem 150 , K. Tariq 14a , G. Tarna 102,27b , G.F. Tartarelli 71a ,
P. Tas 133 , M. Tasevsky 131 , E. Tassi 43b,43a , A.C. Tate 162 , G. Tateno 153 , Y. Tayalati 35e,w ,
G.N. Taylor 105 , W. Taylor 156b , A.S. Tee 170 , R. Teixeira De Lima 143 , P. Teixeira-Dias 95 ,
J.J. Teoh 155 , K. Terashi 153 , J. Terron 99 , S. Terzo 13 , M. Testa 53 , R.J. Teuscher 155,x ,
A. Thaler 79 , O. Theiner 56 , N. Themistokleous 52 , T. Theveneaux-Pelzer 102 , O. Thielmann 171 ,
D.W. Thomas95 , J.P. Thomas 20 , E.A. Thompson 17a , P.D. Thompson 20 , E. Thomson 128 ,
Y. Tian 55 , V. Tikhomirov 37,a , Yu.A. Tikhonov 37 , S. Timoshenko37 , D. Timoshyn 133 ,
E.X.L. Ting 1 , P. Tipton 172 , S.H. Tlou 33g , A. Tnourji 40 , K. Todome 154 , S. Todorova-Nova 133 ,

106
S. Todt50 , M. Togawa 84 , J. Tojo 89 , S. Tokár 28a , K. Tokushuku 84 , O. Toldaiev 68 , R. Tombs 32 ,
M. Tomoto 84,111 , L. Tompkins 143,n , K.W. Topolnicki 86b , E. Torrence 123 , H. Torres 102,ab ,
E. Torró Pastor 163 , M. Toscani 30 , C. Tosciri 39 , M. Tost 11 , D.R. Tovey 139 , A. Traeet16 ,
I.S. Trandafir 27b , T. Trefzger 166 , A. Tricoli 29 , I.M. Trigger 156a , S. Trincaz-Duvoid 127 ,
D.A. Trischuk 26 , B. Trocmé 60 , L. Truong 33c , M. Trzebinski 87 , A. Trzupek 87 , F. Tsai 145 ,
M. Tsai 106 , A. Tsiamis 152,e , P.V. Tsiareshka37 , S. Tsigaridas 156a , A. Tsirigotis 152,s ,
V. Tsiskaridze 155 , E.G. Tskhadadze 149a , M. Tsopoulou 152,e , Y. Tsujikawa 88 , I.I. Tsukerman 37 ,
V. Tsulaia 17a , S. Tsuno 84 , K. Tsuri 118 , D. Tsybychev 145 , Y. Tu 64b , A. Tudorache 27b ,
V. Tudorache 27b , A.N. Tuna 61 , S. Turchikhin 57b,57a , I. Turk Cakir 3a , R. Turra 71a ,
T. Turtuvshin 38,y , P.M. Tuts 41 , S. Tzamarias 152,e , P. Tzanis 10 , E. Tzovara 100 , F. Ukegawa 157 ,
P.A. Ulloa Poblete 137c,137b , E.N. Umaka 29 , G. Unal 36 , M. Unal 11 , A. Undrus 29 , G. Unel 159 ,
J. Urban 28b , P. Urquijo 105 , P. Urrejola 137a , G. Usai 8 , R. Ushioda 154 , M. Usman 108 ,
Z. Uysal 82 , V. Vacek 132 , B. Vachon 104 , K.O.H. Vadla 125 , T. Vafeiadis 36 , A. Vaitkus 96 ,
C. Valderanis 109 , E. Valdes Santurio 47a,47b , M. Valente 156a , S. Valentinetti 23b,23a , A. Valero 163 ,
E. Valiente Moreno 163 , A. Vallier 102,ab , J.A. Valls Ferrer 163 , D.R. Van Arneman 114 ,
T.R. Van Daalen 138 , A. Van Der Graaf 49 , P. Van Gemmeren 6 , M. Van Rijnbach 125,36 ,
S. Van Stroud 96 , I. Van Vulpen 114 , M. Vanadia 76a,76b , W. Vandelli 36 , E.R. Vandewall 121 ,
D. Vannicola 151 , L. Vannoli 57b,57a , R. Vari 75a , E.W. Varnes 7 , C. Varni 17b , T. Varol 148 ,
D. Varouchas 66 , L. Varriale 163 , K.E. Varvell 147 , M.E. Vasile 27b , L. Vaslin84 , G.A. Vasquez 165 ,
A. Vasyukov 38 , R. Vavricka100 , F. Vazeille 40 , T. Vazquez Schroeder 36 , J. Veatch 31 ,
V. Vecchio 101 , M.J. Veen 103 , I. Veliscek 126 , L.M. Veloce 155 , F. Veloso 130a,130c ,
S. Veneziano 75a , A. Ventura 70a,70b , S. Ventura Gonzalez 135 , A. Verbytskyi 110 ,
M. Verducci 74a,74b , C. Vergis 24 , M. Verissimo De Araujo 83b , W. Verkerke 114 ,
J.C. Vermeulen 114 , C. Vernieri 143 , M. Vessella 103 , M.C. Vetterli 142,ag , A. Vgenopoulos 152,e ,
N. Viaux Maira 137f , T. Vickey 139 , O.E. Vickey Boeriu 139 , G.H.A. Viehhauser 126 , L. Vigani 63b ,
M. Villa 23b,23a , M. Villaplana Perez 163 , E.M. Villhauer52 , E. Vilucchi 53 , M.G. Vincter 34 ,
G.S. Virdee 20 , A. Vishwakarma 52 , A. Visibile114 , C. Vittori 36 , I. Vivarelli 146 ,
E. Voevodina 110 , F. Vogel 109 , J.C. Voigt 50 , P. Vokac 132 , Yu. Volkotrub 86a , J. Von Ahnen 48 ,
E. Von Toerne 24 , B. Vormwald 36 , V. Vorobel 133 , K. Vorobev 37 , M. Vos 163 , K. Voss 141 ,
M. Vozak 114 , L. Vozdecky 94 , N. Vranjes 15 , M. Vranjes Milosavljevic 15 , M. Vreeswijk 114 ,
N.K. Vu 62d,62c , R. Vuillermet 36 , O. Vujinovic 100 , I. Vukotic 39 , S. Wada 157 , C. Wagner103 ,
J.M. Wagner 17a , W. Wagner 171 , S. Wahdan 171 , H. Wahlberg 90 , M. Wakida 111 , J. Walder 134 ,
R. Walker 109 , W. Walkowiak 141 , A. Wall 128 , T. Wamorkar 6 , A.Z. Wang 136 , C. Wang 100 ,
C. Wang 11 , H. Wang 17a , J. Wang 64c , R.-J. Wang 100 , R. Wang 61 , R. Wang 6 ,
S.M. Wang 148 , S. Wang 62b , T. Wang 62a , W.T. Wang 80 , W. Wang 14a , X. Wang 14c ,
X. Wang 162 , X. Wang 62c , Y. Wang 62d , Y. Wang 14c , Z. Wang 106 , Z. Wang 62d,51,62c ,
Z. Wang 106 , A. Warburton 104 , R.J. Ward 20 , N. Warrack 59 , S. Waterhouse 95 , A.T. Watson 20 ,
H. Watson 59 , M.F. Watson 20 , E. Watton 59,134 , G. Watts 138 , B.M. Waugh 96 , C. Weber 29 ,
H.A. Weber 18 , M.S. Weber 19 , S.M. Weber 63a , C. Wei 62a , Y. Wei 126 , A.R. Weidberg 126 ,
E.J. Weik 117 , J. Weingarten 49 , M. Weirich 100 , C. Weiser 54 , C.J. Wells 48 , T. Wenaus 29 ,
B. Wendland 49 , T. Wengler 36 , N.S. Wenke110 , N. Wermes 24 , M. Wessels 63a , A.M. Wharton 91 ,
A.S. White 61 , A. White 8 , M.J. White 1 , D. Whiteson 159 , L. Wickremasinghe 124 ,
W. Wiedenmann 170 , M. Wielers 134 , C. Wiglesworth 42 , D.J. Wilbern120 , H.G. Wilkens 36 ,
D.M. Williams 41 , H.H. Williams128 , S. Williams 32 , S. Willocq 103 , B.J. Wilson 101 ,
P.J. Windischhofer 39 , F.I. Winkel 30 , F. Winklmeier 123 , B.T. Winter 54 , J.K. Winter 101 ,
M. Wittgen143 , M. Wobisch 97 , Z. Wolffs 114 , J. Wollrath159 , M.W. Wolter 87 , H. Wolters 130a,130c ,
E.L. Woodward 41 , S.D. Worm 48 , B.K. Wosiek 87 , K.W. Woźniak 87 , S. Wozniewski 55 ,

107
K. Wraight 59 , C. Wu 20 , J. Wu 14a,14e , M. Wu 64a , M. Wu 113 , S.L. Wu 170 , X. Wu 56 ,
Y. Wu 62a , Z. Wu 135 , J. Wuerzinger 110,ae , T.R. Wyatt 101 , B.M. Wynne 52 , S. Xella 42 ,
L. Xia 14c , M. Xia 14b , J. Xiang 64c , M. Xie 62a , X. Xie 62a , S. Xin 14a,14e , A. Xiong 123 ,
J. Xiong 17a , D. Xu 14a , H. Xu 62a , L. Xu 62a , R. Xu 128 , T. Xu 106 , Y. Xu 14b , Z. Xu 52 ,
Z. Xu14c , B. Yabsley 147 , S. Yacoob 33a , Y. Yamaguchi 154 , E. Yamashita 153 , T. Yamashita 85 ,
H. Yamauchi 157 , T. Yamazaki 17a , Y. Yamazaki 85 , J. Yan62c , S. Yan 126 , Z. Yan 25 ,
H.J. Yang 62c,62d , H.T. Yang 62a , S. Yang 62a , T. Yang 64c , X. Yang 36 , X. Yang 14a , Y. Yang 44 ,
Y. Yang62a , Z. Yang 62a , W-M. Yao 17a , H. Ye 14c , H. Ye 55 , J. Ye 14a , S. Ye 29 , X. Ye 62a ,
Y. Yeh 96 , I. Yeletskikh 38 , B.K. Yeo 17b , M.R. Yexley 96 , P. Yin 41 , K. Yorita 168 ,
S. Younas 27b , C.J.S. Young 36 , C. Young 143 , C. Yu 14a,14e , Y. Yu 62a , M. Yuan 106 ,
R. Yuan 62b , L. Yue 96 , M. Zaazoua 62a , B. Zabinski 87 , E. Zaid52 , Z.K. Zak 87 ,
T. Zakareishvili 163 , N. Zakharchuk 34 , S. Zambito 56 , J.A. Zamora Saa 137d,137b , J. Zang 153 ,
D. Zanzi 54 , O. Zaplatilek 132 , C. Zeitnitz 171 , H. Zeng 14a , J.C. Zeng 162 , D.T. Zenger Jr 26 ,
O. Zenin 37 , T. Ženiš 28a , S. Zenz 94 , S. Zerradi 35a , D. Zerwas 66 , M. Zhai 14a,14e ,
D.F. Zhang 139 , J. Zhang 62b , J. Zhang 6 , K. Zhang 14a,14e , L. Zhang 14c , P. Zhang 14a,14e ,
R. Zhang 170 , S. Zhang 106 , S. Zhang 44 , T. Zhang 153 , X. Zhang 62c , X. Zhang 62b ,
Y. Zhang 62c,5 , Y. Zhang 96 , Y. Zhang 14c , Z. Zhang 17a , Z. Zhang 66 , H. Zhao 138 , T. Zhao 62b ,
Y. Zhao 136 , Z. Zhao 62a , A. Zhemchugov 38 , J. Zheng 14c , K. Zheng 162 , X. Zheng 62a ,
Z. Zheng 143 , D. Zhong 162 , B. Zhou 106 , H. Zhou 7 , N. Zhou 62c , Y. Zhou 14c , Y. Zhou7 ,
C.G. Zhu 62b , J. Zhu 106 , Y. Zhu 62c , Y. Zhu 62a , X. Zhuang 14a , K. Zhukov 37 , N.I. Zimine 38 ,
J. Zinsser 63b , M. Ziolkowski 141 , L. Živković 15 , A. Zoccoli 23b,23a , K. Zoch 61 ,
T.G. Zorbas 139 , O. Zormpa 46 , W. Zou 41 , L. Zwalinski 36 .
1 Department of Physics, University of Adelaide, Adelaide; Australia.
2 Department of Physics, University of Alberta, Edmonton AB; Canada.
3 (𝑎) Department of Physics, Ankara University, Ankara; (𝑏) Division of Physics, TOBB University of

Economics and Technology, Ankara; Türkiye.

4 LAPP, Université Savoie Mont Blanc, CNRS/IN2P3, Annecy; France.
5 APC, Université Paris Cité, CNRS/IN2P3, Paris; France.
6 High Energy Physics Division, Argonne National Laboratory, Argonne IL; United States of America.
7 Department of Physics, University of Arizona, Tucson AZ; United States of America.
8 Department of Physics, University of Texas at Arlington, Arlington TX; United States of America.
9 Physics Department, National and Kapodistrian University of Athens, Athens; Greece.
10 Physics Department, National Technical University of Athens, Zografou; Greece.
11 Department of Physics, University of Texas at Austin, Austin TX; United States of America.
12 Institute of Physics, Azerbaijan Academy of Sciences, Baku; Azerbaijan.
13 Institut de Física d’Altes Energies (IFAE), Barcelona Institute of Science and Technology, Barcelona;

Spain.
14 (𝑎) Institute of High Energy Physics, Chinese Academy of Sciences, Beijing; (𝑏) Physics Department,

Tsinghua University, Beijing; (𝑐) Department of Physics, Nanjing University, Nanjing; (𝑑) School of Science,
Shenzhen Campus of Sun Yat-sen University; (𝑒) University of Chinese Academy of Science (UCAS),
Beijing; China.
15 Institute of Physics, University of Belgrade, Belgrade; Serbia.
16 Department for Physics and Technology, University of Bergen, Bergen; Norway.
17 (𝑎) Physics Division, Lawrence Berkeley National Laboratory, Berkeley CA; (𝑏) University of California,

Berkeley CA; United States of America.

18 Institut für Physik, Humboldt Universität zu Berlin, Berlin; Germany.

108
19 Albert Einstein Center for Fundamental Physics and Laboratory for High Energy Physics, University of
Bern, Bern; Switzerland.
20 School of Physics and Astronomy, University of Birmingham, Birmingham; United Kingdom.
21 (𝑎) Department of Physics, Bogazici University, Istanbul; (𝑏) Department of Physics Engineering,

Gaziantep University, Gaziantep; (𝑐) Department of Physics, Istanbul University, Istanbul; Türkiye.
22 (𝑎) Facultad de Ciencias y Centro de Investigaciónes, Universidad Antonio Nariño,

Bogotá; (𝑏) Departamento de Física, Universidad Nacional de Colombia, Bogotá; Colombia.

23 (𝑎) Dipartimento di Fisica e Astronomia A. Righi, Università di Bologna, Bologna; (𝑏) INFN Sezione di

Bologna; Italy.
24 Physikalisches Institut, Universität Bonn, Bonn; Germany.
25 Department of Physics, Boston University, Boston MA; United States of America.
26 Department of Physics, Brandeis University, Waltham MA; United States of America.
27 (𝑎) Transilvania University of Brasov, Brasov; (𝑏) Horia Hulubei National Institute of Physics and Nuclear

Engineering, Bucharest; (𝑐) Department of Physics, Alexandru Ioan Cuza University of Iasi, Iasi; (𝑑) National
Institute for Research and Development of Isotopic and Molecular Technologies, Physics Department,
Cluj-Napoca; (𝑒) National University of Science and Technology Politechnica, Bucharest; ( 𝑓 ) West
University in Timisoara, Timisoara; (𝑔) Faculty of Physics, University of Bucharest, Bucharest; Romania.
28 (𝑎) Faculty of Mathematics, Physics and Informatics, Comenius University, Bratislava; (𝑏) Department of

Subnuclear Physics, Institute of Experimental Physics of the Slovak Academy of Sciences, Kosice; Slovak
Republic.
29 Physics Department, Brookhaven National Laboratory, Upton NY; United States of America.
30 Universidad de Buenos Aires, Facultad de Ciencias Exactas y Naturales, Departamento de Física, y

CONICET, Instituto de Física de Buenos Aires (IFIBA), Buenos Aires; Argentina.

31 California State University, CA; United States of America.
32 Cavendish Laboratory, University of Cambridge, Cambridge; United Kingdom.
33 (𝑎) Department of Physics, University of Cape Town, Cape Town; (𝑏) iThemba Labs, Western

Cape; (𝑐) Department of Mechanical Engineering Science, University of Johannesburg,

Johannesburg; (𝑑) National Institute of Physics, University of the Philippines Diliman
(Philippines); (𝑒) University of South Africa, Department of Physics, Pretoria; ( 𝑓 ) University of Zululand,
KwaDlangezwa; (𝑔) School of Physics, University of the Witwatersrand, Johannesburg; South Africa.
34 Department of Physics, Carleton University, Ottawa ON; Canada.
35 (𝑎) Faculté des Sciences Ain Chock, Réseau Universitaire de Physique des Hautes Energies - Université

Hassan II, Casablanca; (𝑏) Faculté des Sciences, Université Ibn-Tofail, Kénitra; (𝑐) Faculté des Sciences
Semlalia, Université Cadi Ayyad, LPHEA-Marrakech; (𝑑) LPMR, Faculté des Sciences, Université
Mohamed Premier, Oujda; (𝑒) Faculté des sciences, Université Mohammed V, Rabat; ( 𝑓 ) Institute of Applied
Physics, Mohammed VI Polytechnic University, Ben Guerir; Morocco.
36 CERN, Geneva; Switzerland.
37 Affiliated with an institute covered by a cooperation agreement with CERN.
38 Affiliated with an international laboratory covered by a cooperation agreement with CERN.
39 Enrico Fermi Institute, University of Chicago, Chicago IL; United States of America.
40 LPC, Université Clermont Auvergne, CNRS/IN2P3, Clermont-Ferrand; France.
41 Nevis Laboratory, Columbia University, Irvington NY; United States of America.
42 Niels Bohr Institute, University of Copenhagen, Copenhagen; Denmark.
43 (𝑎) Dipartimento di Fisica, Università della Calabria, Rende; (𝑏) INFN Gruppo Collegato di Cosenza,

Laboratori Nazionali di Frascati; Italy.

44 Physics Department, Southern Methodist University, Dallas TX; United States of America.
45 Physics Department, University of Texas at Dallas, Richardson TX; United States of America.

109
46 National Centre for Scientific Research "Demokritos", Agia Paraskevi; Greece.
47 (𝑎) Department of Physics, Stockholm University; (𝑏) Oskar Klein Centre, Stockholm; Sweden.
48 Deutsches Elektronen-Synchrotron DESY, Hamburg and Zeuthen; Germany.
49 Fakultät Physik , Technische Universität Dortmund, Dortmund; Germany.
50 Institut für Kern- und Teilchenphysik, Technische Universität Dresden, Dresden; Germany.
51 Department of Physics, Duke University, Durham NC; United States of America.
52 SUPA - School of Physics and Astronomy, University of Edinburgh, Edinburgh; United Kingdom.
53 INFN e Laboratori Nazionali di Frascati, Frascati; Italy.
54 Physikalisches Institut, Albert-Ludwigs-Universität Freiburg, Freiburg; Germany.
55 II. Physikalisches Institut, Georg-August-Universität Göttingen, Göttingen; Germany.
56 Département de Physique Nucléaire et Corpusculaire, Université de Genève, Genève; Switzerland.
57 (𝑎) Dipartimento di Fisica, Università di Genova, Genova; (𝑏) INFN Sezione di Genova; Italy.
58 II. Physikalisches Institut, Justus-Liebig-Universität Giessen, Giessen; Germany.
59 SUPA - School of Physics and Astronomy, University of Glasgow, Glasgow; United Kingdom.
60 LPSC, Université Grenoble Alpes, CNRS/IN2P3, Grenoble INP, Grenoble; France.
61 Laboratory for Particle Physics and Cosmology, Harvard University, Cambridge MA; United States of

America.
62 (𝑎) Department of Modern Physics and State Key Laboratory of Particle Detection and Electronics,

University of Science and Technology of China, Hefei; (𝑏) Institute of Frontier and Interdisciplinary
Science and Key Laboratory of Particle Physics and Particle Irradiation (MOE), Shandong University,
Qingdao; (𝑐) School of Physics and Astronomy, Shanghai Jiao Tong University, Key Laboratory for Particle
Astrophysics and Cosmology (MOE), SKLPPC, Shanghai; (𝑑) Tsung-Dao Lee Institute, Shanghai; (𝑒) School
of Physics and Microelectronics, Zhengzhou University; China.
63 (𝑎) Kirchhoff-Institut für Physik, Ruprecht-Karls-Universität Heidelberg, Heidelberg; (𝑏) Physikalisches

Institut, Ruprecht-Karls-Universität Heidelberg, Heidelberg; Germany.

64 (𝑎) Department of Physics, Chinese University of Hong Kong, Shatin, N.T., Hong Kong; (𝑏) Department

of Physics, University of Hong Kong, Hong Kong; (𝑐) Department of Physics and Institute for Advanced
Study, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong; China.
65 Department of Physics, National Tsing Hua University, Hsinchu; Taiwan.
66 IJCLab, Université Paris-Saclay, CNRS/IN2P3, 91405, Orsay; France.
67 Centro Nacional de Microelectrónica (IMB-CNM-CSIC), Barcelona; Spain.
68 Department of Physics, Indiana University, Bloomington IN; United States of America.
69 (𝑎) INFN Gruppo Collegato di Udine, Sezione di Trieste, Udine; (𝑏) ICTP, Trieste; (𝑐) Dipartimento

Politecnico di Ingegneria e Architettura, Università di Udine, Udine; Italy.

70 (𝑎) INFN Sezione di Lecce; (𝑏) Dipartimento di Matematica e Fisica, Università del Salento, Lecce; Italy.
71 (𝑎) INFN Sezione di Milano; (𝑏) Dipartimento di Fisica, Università di Milano, Milano; Italy.
72 (𝑎) INFN Sezione di Napoli; (𝑏) Dipartimento di Fisica, Università di Napoli, Napoli; Italy.
73 (𝑎) INFN Sezione di Pavia; (𝑏) Dipartimento di Fisica, Università di Pavia, Pavia; Italy.
74 (𝑎) INFN Sezione di Pisa; (𝑏) Dipartimento di Fisica E. Fermi, Università di Pisa, Pisa; Italy.
75 (𝑎) INFN Sezione di Roma; (𝑏) Dipartimento di Fisica, Sapienza Università di Roma, Roma; Italy.
76 (𝑎) INFN Sezione di Roma Tor Vergata; (𝑏) Dipartimento di Fisica, Università di Roma Tor Vergata,

Roma; Italy.
77 (𝑎) INFN Sezione di Roma Tre; (𝑏) Dipartimento di Matematica e Fisica, Università Roma Tre, Roma;

Italy.
78 (𝑎) INFN-TIFPA; (𝑏) Università degli Studi di Trento, Trento; Italy.
79 Universität Innsbruck, Department of Astro and Particle Physics, Innsbruck; Austria.
80 University of Iowa, Iowa City IA; United States of America.

110
81 Department of Physics and Astronomy, Iowa State University, Ames IA; United States of America.
82 Istinye University, Sariyer, Istanbul; Türkiye.
83 (𝑎) Departamento de Engenharia Elétrica, Universidade Federal de Juiz de Fora (UFJF), Juiz de

Fora; (𝑏) Universidade Federal do Rio De Janeiro COPPE/EE/IF, Rio de Janeiro; (𝑐) Instituto de Física,
Universidade de São Paulo, São Paulo; (𝑑) Rio de Janeiro State University, Rio de Janeiro; Brazil.
84 KEK, High Energy Accelerator Research Organization, Tsukuba; Japan.
85 Graduate School of Science, Kobe University, Kobe; Japan.
86 (𝑎) AGH University of Krakow, Faculty of Physics and Applied Computer Science, Krakow; (𝑏) Marian

Smoluchowski Institute of Physics, Jagiellonian University, Krakow; Poland.

87 Institute of Nuclear Physics Polish Academy of Sciences, Krakow; Poland.
88 Faculty of Science, Kyoto University, Kyoto; Japan.
89 Research Center for Advanced Particle Physics and Department of Physics, Kyushu University, Fukuoka ;

Japan.
90 Instituto de Física La Plata, Universidad Nacional de La Plata and CONICET, La Plata; Argentina.
91 Physics Department, Lancaster University, Lancaster; United Kingdom.
92 Oliver Lodge Laboratory, University of Liverpool, Liverpool; United Kingdom.
93 Department of Experimental Particle Physics, Jožef Stefan Institute and Department of Physics,

University of Ljubljana, Ljubljana; Slovenia.

94 School of Physics and Astronomy, Queen Mary University of London, London; United Kingdom.
95 Department of Physics, Royal Holloway University of London, Egham; United Kingdom.
96 Department of Physics and Astronomy, University College London, London; United Kingdom.
97 Louisiana Tech University, Ruston LA; United States of America.
98 Fysiska institutionen, Lunds universitet, Lund; Sweden.
99 Departamento de Física Teorica C-15 and CIAFF, Universidad Autónoma de Madrid, Madrid; Spain.
100 Institut für Physik, Universität Mainz, Mainz; Germany.
101 School of Physics and Astronomy, University of Manchester, Manchester; United Kingdom.
102 CPPM, Aix-Marseille Université, CNRS/IN2P3, Marseille; France.
103 Department of Physics, University of Massachusetts, Amherst MA; United States of America.
104 Department of Physics, McGill University, Montreal QC; Canada.
105 School of Physics, University of Melbourne, Victoria; Australia.
106 Department of Physics, University of Michigan, Ann Arbor MI; United States of America.
107 Department of Physics and Astronomy, Michigan State University, East Lansing MI; United States of

America.
108 Group of Particle Physics, University of Montreal, Montreal QC; Canada.
109 Fakultät für Physik, Ludwig-Maximilians-Universität München, München; Germany.
110 Max-Planck-Institut für Physik (Werner-Heisenberg-Institut), München; Germany.
111 Graduate School of Science and Kobayashi-Maskawa Institute, Nagoya University, Nagoya; Japan.
112 Department of Physics and Astronomy, University of New Mexico, Albuquerque NM; United States of

America.
113 Institute for Mathematics, Astrophysics and Particle Physics, Radboud University/Nikhef, Nijmegen;

Netherlands.
114 Nikhef National Institute for Subatomic Physics and University of Amsterdam, Amsterdam;

Netherlands.
115 Department of Physics, Northern Illinois University, DeKalb IL; United States of America.
116 (𝑎) New York University Abu Dhabi, Abu Dhabi; (𝑏) United Arab Emirates University, Al Ain; United

Arab Emirates.
117 Department of Physics, New York University, New York NY; United States of America.

111
118 Ochanomizu University, Otsuka, Bunkyo-ku, Tokyo; Japan.
119 Ohio State University, Columbus OH; United States of America.
120 Homer L. Dodge Department of Physics and Astronomy, University of Oklahoma, Norman OK; United

States of America.
121 Department of Physics, Oklahoma State University, Stillwater OK; United States of America.
122 Palacký University, Joint Laboratory of Optics, Olomouc; Czech Republic.
123 Institute for Fundamental Science, University of Oregon, Eugene, OR; United States of America.
124 Graduate School of Science, Osaka University, Osaka; Japan.
125 Department of Physics, University of Oslo, Oslo; Norway.
126 Department of Physics, Oxford University, Oxford; United Kingdom.
127 LPNHE, Sorbonne Université, Université Paris Cité, CNRS/IN2P3, Paris; France.
128 Department of Physics, University of Pennsylvania, Philadelphia PA; United States of America.
129 Department of Physics and Astronomy, University of Pittsburgh, Pittsburgh PA; United States of

America.
130 (𝑎) Laboratório de Instrumentação e Física Experimental de Partículas - LIP, Lisboa; (𝑏) Departamento

de Física, Faculdade de Ciências, Universidade de Lisboa, Lisboa; (𝑐) Departamento de Física,

Universidade de Coimbra, Coimbra; (𝑑) Centro de Física Nuclear da Universidade de Lisboa,
Lisboa; (𝑒) Departamento de Física, Universidade do Minho, Braga; ( 𝑓 ) Departamento de Física Teórica y
del Cosmos, Universidad de Granada, Granada (Spain); (𝑔) Departamento de Física, Instituto Superior
Técnico, Universidade de Lisboa, Lisboa; Portugal.
131 Institute of Physics of the Czech Academy of Sciences, Prague; Czech Republic.
132 Czech Technical University in Prague, Prague; Czech Republic.
133 Charles University, Faculty of Mathematics and Physics, Prague; Czech Republic.
134 Particle Physics Department, Rutherford Appleton Laboratory, Didcot; United Kingdom.
135 IRFU, CEA, Université Paris-Saclay, Gif-sur-Yvette; France.
136 Santa Cruz Institute for Particle Physics, University of California Santa Cruz, Santa Cruz CA; United

States of America.
137 (𝑎) Departamento de Física, Pontificia Universidad Católica de Chile, Santiago; (𝑏) Millennium Institute

for Subatomic physics at high energy frontier (SAPHIR), Santiago; (𝑐) Instituto de Investigación
Multidisciplinario en Ciencia y Tecnología, y Departamento de Física, Universidad de La
Serena; (𝑑) Universidad Andres Bello, Department of Physics, Santiago; (𝑒) Instituto de Alta Investigación,
Universidad de Tarapacá, Arica; ( 𝑓 ) Departamento de Física, Universidad Técnica Federico Santa María,
Valparaíso; Chile.
138 Department of Physics, University of Washington, Seattle WA; United States of America.
139 Department of Physics and Astronomy, University of Sheffield, Sheffield; United Kingdom.
140 Department of Physics, Shinshu University, Nagano; Japan.
141 Department Physik, Universität Siegen, Siegen; Germany.
142 Department of Physics, Simon Fraser University, Burnaby BC; Canada.
143 SLAC National Accelerator Laboratory, Stanford CA; United States of America.
144 Department of Physics, Royal Institute of Technology, Stockholm; Sweden.
145 Departments of Physics and Astronomy, Stony Brook University, Stony Brook NY; United States of

America.
146 Department of Physics and Astronomy, University of Sussex, Brighton; United Kingdom.
147 School of Physics, University of Sydney, Sydney; Australia.
148 Institute of Physics, Academia Sinica, Taipei; Taiwan.
149 (𝑎) E. Andronikashvili Institute of Physics, Iv. Javakhishvili Tbilisi State University, Tbilisi; (𝑏) High

Energy Physics Institute, Tbilisi State University, Tbilisi; (𝑐) University of Georgia, Tbilisi; Georgia.

112
150 Department of Physics, Technion, Israel Institute of Technology, Haifa; Israel.
151 Raymond and Beverly Sackler School of Physics and Astronomy, Tel Aviv University, Tel Aviv; Israel.
152 Department of Physics, Aristotle University of Thessaloniki, Thessaloniki; Greece.
153 International Center for Elementary Particle Physics and Department of Physics, University of Tokyo,

Tokyo; Japan.
154 Department of Physics, Tokyo Institute of Technology, Tokyo; Japan.
155 Department of Physics, University of Toronto, Toronto ON; Canada.
156 (𝑎) TRIUMF, Vancouver BC; (𝑏) Department of Physics and Astronomy, York University, Toronto ON;

Canada.
157 Division of Physics and Tomonaga Center for the History of the Universe, Faculty of Pure and Applied

Sciences, University of Tsukuba, Tsukuba; Japan.

158 Department of Physics and Astronomy, Tufts University, Medford MA; United States of America.
159 Department of Physics and Astronomy, University of California Irvine, Irvine CA; United States of

America.
160 University of Sharjah, Sharjah; United Arab Emirates.
161 Department of Physics and Astronomy, University of Uppsala, Uppsala; Sweden.
162 Department of Physics, University of Illinois, Urbana IL; United States of America.
163 Instituto de Física Corpuscular (IFIC), Centro Mixto Universidad de Valencia - CSIC, Valencia; Spain.
164 Department of Physics, University of British Columbia, Vancouver BC; Canada.
165 Department of Physics and Astronomy, University of Victoria, Victoria BC; Canada.
166 Fakultät für Physik und Astronomie, Julius-Maximilians-Universität Würzburg, Würzburg; Germany.
167 Department of Physics, University of Warwick, Coventry; United Kingdom.
168 Waseda University, Tokyo; Japan.
169 Department of Particle Physics and Astrophysics, Weizmann Institute of Science, Rehovot; Israel.
170 Department of Physics, University of Wisconsin, Madison WI; United States of America.
171 Fakultät für Mathematik und Naturwissenschaften, Fachgruppe Physik, Bergische Universität

Wuppertal, Wuppertal; Germany.

172 Department of Physics, Yale University, New Haven CT; United States of America.
𝑎 Also Affiliated with an institute covered by a cooperation agreement with CERN.
𝑏 Also at An-Najah National University, Nablus; Palestine.
𝑐 Also at Borough of Manhattan Community College, City University of New York, New York NY; United

States of America.
𝑑 Also at Center for High Energy Physics, Peking University; China.
𝑒 Also at Center for Interdisciplinary Research and Innovation (CIRI-AUTH), Thessaloniki; Greece.
𝑓 Also at Centro Studi e Ricerche Enrico Fermi; Italy.
𝑔 Also at CERN, Geneva; Switzerland.
ℎ Also at Département de Physique Nucléaire et Corpusculaire, Université de Genève, Genève;

Switzerland.
𝑖 Also at Departament de Fisica de la Universitat Autonoma de Barcelona, Barcelona; Spain.
𝑗 Also at Department of Financial and Management Engineering, University of the Aegean, Chios; Greece.
𝑘 Also at Department of Physics, Ben Gurion University of the Negev, Beer Sheva; Israel.
𝑙 Also at Department of Physics, California State University, Sacramento; United States of America.
𝑚 Also at Department of Physics, King’s College London, London; United Kingdom.
𝑛 Also at Department of Physics, Stanford University, Stanford CA; United States of America.
𝑜 Also at Department of Physics, Stellenbosch University; South Africa.
𝑝 Also at Department of Physics, University of Fribourg, Fribourg; Switzerland.
𝑞 Also at Department of Physics, University of Thessaly; Greece.

113
𝑟 Also at Department of Physics, Westmont College, Santa Barbara; United States of America.
𝑠 Also at Hellenic Open University, Patras; Greece.
𝑡 Also at Institucio Catalana de Recerca i Estudis Avancats, ICREA, Barcelona; Spain.
𝑢 Also at Institut für Experimentalphysik, Universität Hamburg, Hamburg; Germany.
𝑣 Also at Institute for Nuclear Research and Nuclear Energy (INRNE) of the Bulgarian Academy of

Sciences, Sofia; Bulgaria.

𝑤 Also at Institute of Applied Physics, Mohammed VI Polytechnic University, Ben Guerir; Morocco.
𝑥 Also at Institute of Particle Physics (IPP); Canada.
𝑦 Also at Institute of Physics and Technology, Mongolian Academy of Sciences, Ulaanbaatar; Mongolia.
𝑧 Also at Institute of Physics, Azerbaijan Academy of Sciences, Baku; Azerbaijan.
𝑎𝑎 Also at Institute of Theoretical Physics, Ilia State University, Tbilisi; Georgia.
𝑎𝑏 Also at L2IT, Université de Toulouse, CNRS/IN2P3, UPS, Toulouse; France.
𝑎𝑐 Also at Lawrence Livermore National Laboratory, Livermore; United States of America.
𝑎𝑑 Also at National Institute of Physics, University of the Philippines Diliman (Philippines); Philippines.
𝑎𝑒 Also at Technical University of Munich, Munich; Germany.
𝑎 𝑓 Also at The Collaborative Innovation Center of Quantum Matter (CICQM), Beijing; China.
𝑎𝑔 Also at TRIUMF, Vancouver BC; Canada.
𝑎ℎ Also at Università di Napoli Parthenope, Napoli; Italy.
𝑎𝑖 Also at University of Colorado Boulder, Department of Physics, Colorado; United States of America.
𝑎 𝑗 Also at Washington College, Chestertown, MD; United States of America.
𝑎𝑘 Also at Yeditepe University, Physics Department, Istanbul; Türkiye.
∗ Deceased

114