ATL

-PH

YS-

PUB-

2020

-005

19M

arch

2020

ATLAS PUB NoteATL-PHYS-PUB-2020-005

19th March 2020

Expected 1-tagging Performance with the upgradedATLAS Inner Tracker Detector at the

High-Luminosity LHC

The ATLAS Collaboration

In view of the operation of the ATLAS detector under High-Luminosity LHC conditions, thecentral tracking system will be upgraded to maintain high levels of performance with a peakinstantaneous luminosity of 7.5 × 1034 cm−2B−1, corresponding to an average of about 200inelastic proton-proton collisions per beam-crossing. The new Inner Tracker (ITk), with anextended pseudo-rapidity coverage up to |[ |=4, is planned to operate for more than ten years,during which time the LHC aims to deliver 4 ab−1 of integrated luminosity. The algorithmsused for the identification of the hadronic jets stemming from 1-quarks rely heavily on trackreconstruction, and their performance is therefore strongly dependent on the design of thisdetector. This document presents the 1-tagging performance expected with the latest ITkdesign, updated since the ITk Pixel Detector Technical Design Report.

© 2020 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.

1 Introduction

The scientific programme of the LHC spans the next 20 years and includes an ambitious series of upgrades,culminating in the High-Luminosity LHC (HL-LHC) phase [1], that will ultimately result in an accumulatedintegrated luminosity for proton-proton collisions of around 4 ab−1. This represents an order of magnitudemore data than what will be collected prior to the HL-LHC run. The foreseen increased luminosity forthis phase, up to 7.5 × 1034 cm−2s−1, with its associated data rate and accumulated radiation damage,would lead to unacceptable levels of track reconstruction performance loss with the current inner detector(ID). Therefore the ATLAS collaboration plans to replace the ID with a new all-silicon tracker to maintaintracking performance in these conditions. The upgraded detector installation (Phase-II) will take placeduring the so-called Long Shutdown 3, starting at the end of 2024.

The Inner Tracker (ITk) combines precision central tracking in the presence of 200 pile-up events, with theability to extend the tracking coverage to a pseudo-rapidity |[ | of 4 (from the current 2.5) with excellenttracking performance [2]. The ITk comprises two subsystems; a Strip Detector [3] surrounding a PixelDetector [4]. The Strip Detector has four barrel layers and six end-cap disks, covering |[ | < 2.7. It iscomplemented by the Pixel Detector with five barrel layers extending the coverage to |[ | < 4, and multiplering-shaped end-cap disks, the number of which depends on the pseudo-rapidity. Given the harsh radiationenvironment expected for HL-LHC, the two innermost layers of the Pixel Detector are replaceable.

The identification of jets arising from 1-hadrons, referred to as 1-tagging, is key to many physics analyses:top physics, the measurement of Higgs boson couplings to heavy quarks, probing the Higgs bosonself-coupling and the searches for new physics where new heavy particles tend to decay to heavy quarks.Therefore it is essential to improve or at least maintain the current Run 2 1-tagging performance for theITk, while extending it to larger rapidities. As such, the 1-tagging performance is a key benchmark for thedetector design. Results on this topic have been presented in the ATLAS Inner Tracker Pixel DetectorTechnical Design Report (TDR) [4]. Since then, the simulation of the ITk has been refined to take intoaccount, in particular, a reduction of the detector active volume required for the introduction of the proposedHigh-Granularity Timing Detector [5] and design choices regarding the local support structures for theinner Pixel system. This document focuses on the 1-tagging performance obtained with the so-called “ITklayout” introduced in [2].

This note is organized as follows. Section 2 gives details on the Monte Carlo (MC) samples and the objectand event selections used to estimate the 1-tagging performance. Section 3 briefly presents the 1-taggingalgorithms investigated, while their performance is presented in Section 4. The impact of the 1-taggingperformance on some key HL-LHC analyses is outlined in Section 5.

2 Monte Carlo Samples and Object Selections

The studies presented in this note are performed using a Monte Carlo sample of simulated CC̄ events, withinclusive top quark decays, produced in proton-proton collisions generated at a centre-of-mass energy√B = 14 TeV. This sample is generated with POWHEG [6] interfaced with PYTHIA 6 [7] for parton

shower and hadronization. An average of 200 pile-up interactions per event are simulated, in line withthe HL-LHC operating environment. All simulated event samples were processed with the GEANT4ATLAS full detector simulation [8, 9], including the ITk layout as described in [2], displayed in Fig. 1. The

2

corresponding number of radiation lengths encountered by particles going through the ITk is presented inFig. 2.

Figure 1: A schematic depiction of the ITk layout used in simulated samples considered in this document [2]. Theactive elements of the Strip Detector are shown in blue, and those of the Pixel Detector are shown in red. Thehorizontal axis is along the beam line with zero being the interaction point. The vertical axis is the radius measuredfrom the interaction region.

Some of the key inputs for 1-tagging are: the jet direction usually obtained from calorimeter jets, thetracks reconstructed in the ITk and the primary vertex of the hard-scattering collision of interest. Tracksare reconstructed from clusters of signals in the silicon pixel and strip sensors, collectively referred to ashits. For the studies shown in this document, the pixel hits are clustered using the analogue clusteringalgorithm [4], for which the center of the cluster is computed using a weighted sum based on the chargeinformation of the pixels belonging to the clusters. Minimal quality requirements are applied to thereconstructed tracks based on pseudorapidity, hit content and displacement with respect to the mean positionof the reconstructed beam spot in the transverse plane and along the beam axis. Those requirements aredetailed in Table 1. An average track reconstruction efficiency betwen 85% and 95% is associated withthose selections in CC̄ events [2], similar to the current Run-2 detector. The number of nuclear interactionlengths encountered by a track until enough detector layers have been crossed such that the reconstructionhit requirements are met is displayed in Fig. 3 and compared to the current ATLAS detector. A similaramount of material is traversed by successfully reconstructed particles in the barrel region, while it issignificantly lower in the end-cap and forward regions.

The 1-tagging performance is expected to be strongly correlated with the impact parameter resolutionsobtained with the ITk. The transverse impact parameter 30 used for 1-tagging algorithms is the distance ofclosest approach of the track to the primary vertex point, in the A-i projection, while in the longitudinaldirection, the parameter used is the difference between the I coordinates of the primary vertex position andof the track at this point of closest approach (i.e. the longitudinal impact parameter I0) multiplied by the

3

η0 1 2 3 4 5 6

] 0R

adia

tion

Leng

ths

[X

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2Dry NitrogenPatch Panel 1Electrical CablingTitanium Cooling PipesSupport StructureStrip ChipsStrip HybridsPixel ChipsPixel HybridsActive SensorsBeam Pipe

ATLASSimulationPreliminary

ITk Layout

Figure 2: Number of radiation lengths -0 versus the pseudorapidity |[ | for the ITk Layout [2]. The Patch Panel 1 islocated behind all of the ITk active layers and does not impact the tracking performance.

Requirement Pseudorapidity interval|[ | < 2.0 2.0 < |[ | < 2.6 2.6 < |[ | < 4.0

pixel+strip hits ≥ 9 ≥ 8 ≥ 7pixel hits ≥ 1 ≥ 1 ≥ 1holes ≤ 2 ≤ 2 ≤ 2

double holes ≤ 1 ≤ 1 ≤ 1?T [MeV] > 900 > 400 > 400|30 | ≤ 2 mm ≤ 2 mm ≤ 10 mm|I0 | ≤ 20 cm ≤ 20 cm ≤ 20 cm

Table 1: Set of selection requirements applied during the track reconstruction in different pseudorapidity intervals.Holes are counted if track candidates cross active sensors on which no hit was found, double holes are two consecutiveactive sensors crossed without a hit found. Layers before the first or after the last hit are excluded from the holecounting. The longitudinal and transverse impact parameters, I0 and 30, are defined with respect to the mean positionof the beam spot.

sine of the track polar angle \. The performance results presented in this document have been obtainedfor the two pixel pitches considered for the ITk pixel detector (50 × 50 `m2 and 25 × 100 `m2). Thecorresponding impact parameter resolutions, as well as the performance obtained with the current ATLASdetector with a pixel pitch of 50 × 250 `m2 for the innermost pixel layer (IBL [10]), are illustrated inFig. 4. For low transverse momenta, this version of the ITk geometry is not expected to match the impactparameter resolutions of the Run-2 ATLAS detector, due to the larger radius of the innermost pixel layer(39 mm for ITk compared to 33 mm for the IBL). An option of reducing the radius of the ITk innermost

4

η0 0.5 1 1.5 2 2.5 3 3.5 4

]0λ

Nucl

ear

inte

ract

ion le

ngth

s [

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

ATLAS SimulationPreliminary

Run-2

ITk Layout

0 0.5 1 1.5 2 2.5 3 3.5 4

Ru

n-2

ITk

La

yo

ut

0

0.5

1

η

Figure 3: Comparison of the number of nuclear interaction lengths seen by a particle as a function of pseudorapidityup to the position where enough detector layers have been crossed such that the reconstruction hit requirements ofTable 1 are met [2]. A comparison with the Run-2 detector is presented. For the Run-2 detector the hit requirement is7 hits.

pixel layer to reach similar level of performance is currently being considered. The performance for low-?Tmuons is not strongly affected by the ITk pixel pitch, as the resolutions are dominated by material effects.At high ?T, the impact parameter resolutions are instead dominated by the intrinsic pixel resolution directlycorrelated with the pixel size in the corresponding direction. The 30 resolution is therefore expected to betwo times better with 25 × 100 `m2 ITk pixels than with the 50 × 50 `m2 ITk pixels or the 50 × 250 `m2

IBL pixels. The I0 resolution is also expected to be significantly improved with ITk, thanks to the muchlower longitudinal pixel size with respect to IBL. The 50 × 50 `m2 pixels provide a better resolution thanthe 25 × 100 `m2 pixels for very central and very forward tracks, for which the cluster size is close to onepixel.

The primary and pile-up interaction vertices are reconstructed using the Adaptive Multi-Vertex Finder(AMVF) algorithm [11]. The use of this algorithm, instead of the Iterative Vertex Finder used duringRun-2 [12], was motivated by the increased pile-up. The inputs to the AMVF algorithm (beam spotproperties and reconstructed tracks), along with the procedure employed to reconstruct vertices, aredescribed in detail in the Pixel TDR [4]. The vertex reconstruction efficiency, defined as the fraction ofevents in which the vertex corresponding to the hard-scatter interaction is successfully reconstructed, isclose to 100% in CC̄ events. Around 100 vertices are reconstructed per event and the reconstructed vertexwith the largest sum of the squared transverse momenta of the associated tracks (

∑?2)) is chosen as the

signal primary vertex. Simulation studies indicate that the probability of choosing the correct signalprimary vertex, applying this criterion, is around 95% in CC̄ events.

As the aim of the studies presented here is to focus on the 1-tagging performance and factor out primary

5

|ηtrue track |0 0.5 1 1.5 2 2.5 3 3.5 4

m]

µ)

[0

(dσ

210

3102mµ50 ×Analogue Clustering, 50

2mµ100 ×Analogue Clustering, 25Analogue Clustering (Run-2)

ATLAS PreliminarySimulation

= 1 GeVT

Single muon, p

|ηtrue track |0 0.5 1 1.5 2 2.5 3 3.5 4

2mµ

100

×25

2mµ

50

×50

0.8

0.9

1

1.1

|ηtrue track |0 0.5 1 1.5 2 2.5 3 3.5 4

m]

µ)

[0

(dσ

10

2mµ50 ×Analogue Clustering, 502mµ100 ×Analogue Clustering, 25

Analogue Clustering (Run-2)

ATLAS PreliminarySimulation

= 100 GeVT

Single muon, p

|ηtrue track |0 0.5 1 1.5 2 2.5 3 3.5 4

2mµ

100

×25

2mµ

50

×50

1

3

0.3

(a) (b)

|ηtrue track |0 0.5 1 1.5 2 2.5 3 3.5 4

m]

µ)

[0

(zσ

210

310

410

2mµ50 ×Analogue Clustering, 502mµ100 ×Analogue Clustering, 25

Analogue Clustering (Run-2)

ATLAS PreliminarySimulation

= 1 GeVT

Single muon, p

|ηtrue track |0 0.5 1 1.5 2 2.5 3 3.5 4

2mµ

100

×25

2mµ

50

×50

0.8

0.9

1

1.1

|ηtrue track |0 0.5 1 1.5 2 2.5 3 3.5 4

m]

µ)

[0

(zσ

10

210

3102mµ50 ×Analogue Clustering, 50

2mµ100 ×Analogue Clustering, 25Analogue Clustering (Run-2)

ATLAS PreliminarySimulation

= 100 GeVT

Single muon, p

|ηtrue track |0 0.5 1 1.5 2 2.5 3 3.5 4

2mµ

100

×25

2mµ

50

×50

1

3

0.3

(c) (d)

Figure 4: Track parameter resolution in 30 (a-b) and I0 (c-d) as a function of [ for muons with ?T = 1 GeV and?T = 100 GeV without pile-up [2]. For comparison the results for the current Run-2 detector are shown as a blackline. The ratio in the lower part of the plots is defined as the results using 50 × 50 `m2 pixels over those obtainedusing 25 × 100 `m2 pixels.

vertex finding and selection performance, some events are discarded based on the true position of the hardscatter vertex. Events where the chosen reconstructed signal primary vertex is not within 0.1 mm of thetrue position of the hard scatter vertex along the beam axis make up around 7% of the CC̄ sample, and arerejected.

Jets are reconstructed by clustering energy deposits in the calorimeter with the anti-:C algorithm [13]with a radius parameter ' = 0.4, as implemented in the FastJet package [14]. Jet transverse momentaare further corrected to the corresponding particle-level jet ?T, based on the simulation [15]. Only jetswith ?T > 20 GeV and |[ | < 4.0 are considered. In order to decouple the 1-tagging performance fromthe ongoing optimisation of the pile-up jet tagging performance with the ITk detector, generator-levelselections are applied to discard pile-up jets. Selected jets are required to be matched within Δ' < 0.3 to atruth jet. Truth jets are built from the stable particles produced in the hard interaction, using the sameFastJet implementation of the anti-:C algorithm with ' = 0.4. Jets failing this criteria are labelled aspile-up jets and are not considered, unless stated explicitly. Therefore, for the studies shown in this note,light-flavour jets originate either from light quarks produced by the, from top decays or from QCD initial

6

or final state radiation but not from pile-up induced jets. However, inside the considered jets, the overlay ofenergy deposits and tracks coming from pile-up is present. In particular, 25% of the jets are associatedwith at least one pile-up track and are expected to display degraded 1-tagging performance with respect tojets associated with only hard-scatter tracks.

In order to label jets as being 1, 2- or light-flavour jets in the simulation, the jets are matched to the 1 and2-hadrons within a cone of radius Δ' = 0.3 around the jet axis. The transverse momenta of the 1- and 2-hadrons are required to be larger than 5 GeV. The labelling procedure is performed by first searching for1-hadrons within the cone centered on the axis of the jet. If no 1-hadron is found, the algorithm is repeatedfor 2-hadrons, and for g leptons. Ultimately the remaining jets which do not fall in the previous categoriesare considered as light-flavour jets. Results presented here differ from those presented in [2] or [4] due tothe implementation of an additional generator-level jet matching to quarks for light-jets, in order to removethe contamination from electrons produced in top quark decays. Reconstructed jets are thus required to bematched within Δ' < 0.3 with a generator-level light quark produced in the hard scattering interaction orthe top quark hadronic decays.

3 1-tagging Algorithms Studied

Algorithms typically used in ATLAS for 1-tagging fall into two broad categories. The so-called low-leveltaggers focus on specific features of basic jet constituents, such as large impact parameter tracks orsecondary vertices. The IP3D algorithm [16], presented in Section 3.1, exploits specifically the largeimpact parameters of tracks originating from 1-hadron decays, while the SV1 algorithm [17], presented inSection 3.2, attempts to reconstruct secondary vertices from the tracks and exploits kinematic variablesassociated to those objects. The performance of those low-level taggers are to a large extent uncorrelatedand they can therefore be combined to improve the 1-tagging performance into so-called high-level taggers,such as IP3D+SV1 or MV2 [12], detailed in Section 3.3 and 3.4, respectively. The algorithms have beenwidely used during Run 2 and have been reoptimised to take into account the new ITk geometry and theHL-LHC data-taking conditions.

3.1 IP3D Algorithm

Impact parameter (IP) based algorithms aim to identify 1-jets using the impact parameters of the charged-particle tracks from the 1-hadron decay products. The uncertainties associated to the transverse impactparameter 30 and the longitudinal impact parameter I0 multiplied by the sine of the track polar angle \I0 sin \ are denoted as f30 and fI0 sin \ . The tracks from the 1-hadron decay products tend to have largeimpact parameter significance. The IP3D tagger [16] is a low-level discriminant taking advantage ofthose properties and has been widely used for both the current 1-tagging algorithms and ATLAS upgradestudies.

IP3D is based on a log-likelihood ratio (LLR) method. For each track, the measurement of the impactparameter significances (30/f30 , I0 sin \/fI0 sin \ ) is compared to pre-determined simulation-based two-dimensional probability density functions (PDFs) obtained from simulation for both the 1- and light-flavour-jet hypotheses. As such, the IP3D discriminant is not specifically optimised to reject the background from2-jet but a 2-jet probability is still computed to be used as input of the MV2 discriminant described inSection 3.4. The ratio of probabilities defines the track weight. The jet weight is then defined as the sum

7

of the logarithms of the individual track weights. The LLR formalism allows exclusive track categoriesto be used by defining different dedicated PDFs for each category depending on the quality of the tracks.17 track categories are defined based on the hit content and expected impact parameter resolution. Theyhave been reoptimised for the Pixel TDR studies [4] to exploit the geometry of the new ITk detector. Onlytracks with ?) > 1 GeV are considered. The transverse and longitudinal impact parameters with respect tothe primary vertex must fulfil |30 | < 1 mm and |I0 | sin \ < 1.5 mm. Those selections are still inheritedfrom the Run 2 configuration and will benefit from a dedicated re-optimisation for the ITk in the future.The tracks are associated to jets following the same ?T-dependent Δ' criteria as those used in Run 2 [12].The projections of the PDF, associated with the impact parameter significances, in the category of centraltracks with |[ | < 1 corresponding to the best expected impact parameter resolutions, are presented in Fig. 5.The sign of the impact parameter significances is defined as positive if the track intersects the jet axis infront of the primary vertex along the flight direction, and as negative if the intersection lies behind theprimary vertex [16]. In the case of 1-hadrons, a positive sign is expected while for tracks originating fromthe primary vertex, the experimental resolution results in a symmetric distribution around zero.

20− 15− 10− 5− 0 5 10 15 20)

0(dσ/0d

4−10

3−10

2−10

1−10

1

a.u.

IP3D ITk layout2mµb-jets 50x50

2mµlight-jets 50x50 2mµb-jets 25x100

2mµlight-jets 25x100

ATLAS Simulation Preliminaryt>=200, tµ=14 TeV, <s

20− 15− 10− 5− 0 5 10 15 20

)0

(dσ/0d

2mµ

25x1

00 /

50x5

0

0.6

0.8

1

1.2

1.4 20− 15− 10− 5− 0 5 10 15 20)θsin

0(zσ/θsin0z

3−10

2−10

1−10

1a.

u.

IP3D ITk layout2mµb-jets 50x50

2mµlight-jets 50x50 2mµb-jets 25x100

2mµlight-jets 25x100

ATLAS Simulation Preliminaryt>=200, tµ=14 TeV, <s

20− 15− 10− 5− 0 5 10 15 20

)θsin0

(zσ/θsin0z

2mµ

25x1

00 /

50x5

0

0.6

0.8

1

1.2

1.4

(a) (b)

Figure 5: One-dimensional projections of the probability density function used for the IP3D algorithm in the categoryof central tracks with |[ | < 1 corresponding to the best expected impact parameter resolutions.

3.2 SV1 Algorithm

The SV1 algorithm [17] is based on the reconstruction of the secondary vertex formed by the decay productsof the 1-hadron. A single inclusive secondary vertex is reconstructed by the Single Secondary VertexFinder algorithm [17]. The transverse impact parameter of the tracks considered must fulfil |30 | < 3.5 mm.Vertices whose associated invariant mass is compatible with the kaon mass are discarded, as they wouldotherwise be associated with a significant mistagging rate for light flavour jets. Due to the complex ITkgeometry, no attempt to discard vertices close to the layers of the pixel detector has been implemented toreduce the impact of hadronic interactions with the detector material but it will be considered for futureimprovements. Subsequently, properties of that vertex are used via a likelihood ratio formalism similar to

8

the one used for the IP3D algorithm to define a discriminant. The maximum efficiency reachable with SV1is limited by the secondary vertex reconstruction efficiency in real 1-jets (∼ 75-85% for the consideredCC̄ sample, against ∼ 10% for light jets) but the discrimination power of SV1 provides additional handleswith respect to IP3D when combined into high-level taggers. The SV1 discriminant exploits three vertexproperties: the vertex mass (the invariant mass of all charged-particle tracks used to reconstruct the vertex,assuming that all tracks are pions), the energy fraction (the ratio of the sum of the energies of tracks usedin the vertex to the sum of the energies of all tracks in the jet), and the number of two-track vertices. Inaddition, the Δ' separation between the jet direction and the direction of the line joining the primaryvertex and the secondary vertex is used in the LLR. SV1 relies on a two-dimensional PDF of the first twovariables and on two one-dimensional PDFs of the latter variables. The projections of those PDFs arepresented in Fig. 6.

3.3 IP3D+SV1 Algorithm

The two previous algorithms provide low-level discriminants focusing on different properties of the jetsand can be combined in a straightforward manner by summing their respective weights, thus definingthe so-called IP3D+SV1 algorithm. This discriminant benefits from an improved mistag rate reductionwith respect to the individual IP3D and SV1 outputs, thanks to the limited correlation between those. Inparticular, for a 70% 1-tagging efficiency working point, the rejection obtained with the IP3D discriminantalone can be improved by ∼ 60% using the combined IP3D+SV1 discriminant.

3.4 MV2 Algorithm

MV2 [12] is a multivariate algorithm based on a Boosted Decision Tree (BDT) that combines inputs fromthe low-level impact parameter and secondary vertex algorithms described above, as well as from thedecay chain multi-vertex algorithm JetFitter [18], which attempts to reconstruct the toplogical structure ofcascade 1- and 2-hadron decays inside the jet and is used to define additional topological input variablesfor MV2. A dedicated reoptimisation of the JetFitter reconstruction parameters has not been performed forthose studies and may further improve the MV2 performance in the future. The version of the algorithmused for the upgrade studies includes a 10% 2-jet fraction in the background training sample.

The CC̄ sample used has been split in two independent subsamples, one being used for the training and theother for testing the performance of the algorithm. The BDT training parameters have been optimized toreflect the difference between the detectors and limited size of the training sample (500k as opposed to 5Mfor Run 2). In particular, the depth of the trees has been reduced from 30 to 12. While these parametershave been shown to ensure absence of over-training (by checking the difference in performance betweenthe training and testing samples), the algorithm performance is expected to improve when larger trainingsamples become available.

4 Expected Performance

In the following section, the expected performance of the 1-tagging algorithms introduced above arepresented, using the recent ITk layout simulation introduced in [2]. In all cases, the PDFs used in thelikelihood-based algorithms have been re-derived to optimize their discrimination power.

9

0 1 2 3 4 5 6 7 8SV mass [GeV]

2−10

1−10

1

a.u.

SV1 ITk layout2mµb-jets 50x50

2mµlight-jets 50x50 2mµb-jets 25x100

2mµlight-jets 25x100

ATLAS Simulation Preliminary

t>=200, tµ=14 TeV, <s

0 1 2 3 4 5 6 7 8

SV mass [GeV]

2mµ

25x1

00 /

50x5

0

0.6

0.8

1

1.2

1.4 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1SV energy fraction

3−10

2−10

1−10

1

10a.u.

SV1 ITk layout2mµb-jets 50x50

2mµlight-jets 50x50 2mµb-jets 25x100

2mµlight-jets 25x100

ATLAS Simulation Preliminaryt>=200, tµ=14 TeV, <s

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

SV energy fraction

2mµ

25x1

00 /

50x5

0

0.6

0.8

1

1.2

1.4

(a) (b)

2 4 6 8 10 12 14 16 18 20

# two-track vertices

3−10

2−10

1−10

1a.u.

SV1 ITk layout2mµb-jets 50x50

2mµlight-jets 50x50 2mµb-jets 25x100

2mµlight-jets 25x100

ATLAS Simulation Preliminaryt>=200, tµ=14 TeV, <s

2 4 6 8 10 12 14 16 18 20

# two-track vertices

2mµ

25x1

00 /

50x5

0

0.6

0.8

1

1.2

1.4 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45R(jet,PV-SV)∆

3−10

2−10

1−10

1

10a.u.

SV1 ITk layout2mµb-jets 50x50

2mµlight-jets 50x50 2mµb-jets 25x100

2mµlight-jets 25x100

ATLAS Simulation Preliminary

t>=200, tµ=14 TeV, <s

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45

R(jet,PV-SV)∆

2mµ

25x1

00 /

50x5

0

0.6

0.8

1

1.2

1.4

(c) (d)

Figure 6: One-dimensional projections of the probability density functions used for the SV1 algorithm, correspongingto the vertex mass (a), the energy fraction (b), the number of two-track vertices (c) and the Δ' separation betweenthe jet direction and the direction of the line joining the primary vertex and the secondary vertex (d). Distributionsare normalised to the same number of jets with a reconstructed secondary vertex.

The light-jet and 2-jet rejection, defined as the inverse of the corresponding background mis-identificationprobabilities, are shown as a function of the 1-jet efficiency for various [ and ?T ranges in Fig. 7- 10 forthe different 1-taggers investigated.

The 1-tagging performance is in general better in the central region of the detector than in the forwardregion, as it benefits from better track impact parameter resolutions due to the lower amount of material.Low-?) jets also suffer in general from the impact parameter resolutions of low-?) tracks. The worseimpact parameter resolutions for forward and low-?) tracks not only reduce the discrimination between

10

tracks associated with light jets and 1-jets but also induce a larger contamination of such pile-up tracks injets, which affects as well the 1-tagging performance. The performance for very high-?) jets is insteadlimited by the larger fraction of 1 hadrons flying beyond the innermost pixel layer, leading to a degradationof the impact parameter resolutions for the tracks associated to their decay products, as well as by the denseenvironment in the core of those jets, making the track reconstruction more challenging [19].

As shown, for all the investigated algorithms, even in the very forward region, a significant discriminationpower between 1-jets and light-jets can be achieved with the ITk. Moreover, the use of a pixel pitch of25 × 100 `m2 improves the light-jet rejection for all the algorithms, with an increase between 10 and 35%compared to 50 × 50 `m2, except in the region 2 < |[ | < 3. This improvement is expected, considering theimproved transverse impact parameter resolution achieved with this configuration, illustrated in Fig. 4. Theregion 2 < |[ | < 3 is associated to a larger amount of material as illustrated in Fig. 1, which counteractsthe improvement due to the intrinsic pixel resolution for the 25 × 100 `m2 pitch. The improvementobserved with the 25 × 100 `m2 pixel pitch in terms of light-jets allows flexibility, since operating pointswhich improve 2-jet rejection at a small cost to the light-jet performance (obtained via dedicated tuning ofhigh-level taggers) can be considered while still matching or improving on the current light-jet rejection.Regarding the choice of the pixel pitch to be used in the ITk pixel detector, the benefits of the 25× 100 `m2

pixel pitch for 1-tagging will also be balanced against the better expected performance of the pile-up jetrejection algorithms with the 50 × 50 `m2 pixel pitch, in particular in the forward region which suffersfrom a larger pile-up jet multiplicity. The 50 × 50 `m2 pixel pitch is the current ITk baseline and used inthe rest of this document. The use of a 25 × 100 `m2 pixel pitch is considered for a future evolution of theITk design.

The performance of the IP3D and MV2 taggers, retrained for the ITk detector in Phase-II conditions, withan average of 200 pile-up events, is also compared with the one obtained with the Run 2 algorithms [20],with an average of 30 pile-up events, in Fig. 11. As shown, the Phase-II performance within the acceptanceof the current detector is expected to be improved with respect to the Run 2 performance. However itshould be noted that the jet selections used to assess the Run 2 performance slightly differ from the onesused for those Phase-II studies. In particular, light jets also include contributions from gluon jets in the CC̄events considered, while this is not the case for the Phase 2 performance. Moreover the Run 2 b-taggingperformance is assessed after applying the Jet Vertex Tagger (JVT) algorithm, which is used in Run 2ATLAS analyses [21]. The performance of the JVT algorithm has been measured in collision data to havean efficiency of 92% for jets originating from the hard-scatter vertex and a mis-identification rate of pileupjets, which is approximately 5%. It is assumed here that jets originating from pileup interactions will berejected at a very high rate with the ITk in an HL-LHC scenario; they are therefore removed from the sampleof jets used to assess the b-tagging performance in these studies. Using pile-up jet rejection algorithmsoptimised for the ITk, the b-tagging performance is therefore not expected to be significantly worse thanin Run 2, in spite of the significantly larger pile-up expected for HL-LHC. MV2 performance is alsoexpected to be improved in the future thanks to the dedicated optimisation of the JetFitter reconstructionparameters.

For the purposes of Phase-II physics analyses, the MV2 tagging rates for 1-, 2-, light-, and pile-up jets areparameterized as functions of jet ?T and |[ |, which are then applied to the particle-level quantities [22].Figs. 12 and 13 show the efficiency parameterizations for the fixed Y1 = 70% operating point obtained withthe 50 × 50 `m2 and 25 × 100 `m2 pixel pitch, respectively. Some representative numbers in terms oflight-jet rejection are also highlighted in Table 2.

11

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

1

1.2

1.4

2mµ

25x1

00 /

50x5

0

1

10

210

310

410

Ligh

t-je

t rej

ectio

n

|<1η 0<|2mµ50x50

|<1η 0<|2mµ25x100

|<2η 1<|2mµ50x50

|<2η 1<|2mµ25x100

|<3η 2<|2mµ50x50

|<3η 2<|2mµ25x100

|<4η 3<|2mµ50x50

|<4η 3<|2mµ25x100

IP3D ITk Layout

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

1

1.1

2mµ

25x1

00 /

50x5

0

1

10

c-je

t rej

ectio

n

|<1η 0<|2mµ50x50

|<1η 0<|2mµ25x100

|<2η 1<|2mµ50x50

|<2η 1<|2mµ25x100

|<3η 2<|2mµ50x50

|<3η 2<|2mµ25x100

|<4η 3<|2mµ50x50

|<4η 3<|2mµ25x100

IP3D ITk Layout

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

(a) (b)

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

1

1.2

1.4

2mµ

25x1

00 /

50x5

0

1

10

210

310

410

Ligh

t-je

t rej

ectio

n

<100 GeVT

20<p2mµ50x50

<100 GeVT

20<p2mµ25x100

<300 GeVT

100<p2mµ50x50

<300 GeVT

100<p2mµ25x100

<600 GeVT

300<p2mµ50x50

<600 GeVT

300<p2mµ25x100

|<4.0ηIP3D ITk Layout |

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

0.9

1

1.1

2mµ

25x1

00 /

50x5

0

1

10

c-je

t rej

ectio

n

<100 GeVT

20<p2mµ50x50

<100 GeVT

20<p2mµ25x100

<300 GeVT

100<p2mµ50x50

<300 GeVT

100<p2mµ25x100

|<4.0ηIP3D ITk Layout |

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

(c) (d)

Figure 7: Light-jet and charm-jet rejections vs 1-tagging efficiency for the IP3D tagging algorithms, evaluated in CC̄events with 200 pile-up events, compared for a pixel pitch of 50 × 50 `m2 (full line) and 25 × 100 `m2 (dashed line).The results are compared in different [ regions (a-b) and ?) regions (c-d).

12

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

1

1.2

1.4

2mµ

25x1

00 /

50x5

0

1

10

210

310

410

Ligh

t-je

t rej

ectio

n

|<1η 0<|2mµ50x50

|<1η 0<|2mµ25x100

|<2η 1<|2mµ50x50

|<2η 1<|2mµ25x100

|<3η 2<|2mµ50x50

|<3η 2<|2mµ25x100

|<4η 3<|2mµ50x50

|<4η 3<|2mµ25x100

SV1 ITk Layout

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

1

1.1

2mµ

25x1

00 /

50x5

0

1

10

c-je

t rej

ectio

n

|<1η 0<|2mµ50x50

|<1η 0<|2mµ25x100

|<2η 1<|2mµ50x50

|<2η 1<|2mµ25x100

|<3η 2<|2mµ50x50

|<3η 2<|2mµ25x100

|<4η 3<|2mµ50x50

|<4η 3<|2mµ25x100

SV1 ITk Layout

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

(a) (b)

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

1

1.2

1.4

2mµ

25x1

00 /

50x5

0

1

10

210

310

410

Ligh

t-je

t rej

ectio

n

<100 GeVT

20<p2mµ50x50

<100 GeVT

20<p2mµ25x100

<300 GeVT

100<p2mµ50x50

<300 GeVT

100<p2mµ25x100

<600 GeVT

300<p2mµ50x50

<600 GeVT

300<p2mµ25x100

|<4.0ηSV1 ITk Layout |

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

0.9

1

1.1

2mµ

25x1

00 /

50x5

0

1

10

c-je

t rej

ectio

n

<100 GeVT

20<p2mµ50x50

<100 GeVT

20<p2mµ25x100

<300 GeVT

100<p2mµ50x50

<300 GeVT

100<p2mµ25x100

|<4.0ηSV1 ITk Layout |

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

(c) (d)

Figure 8: Light-jet and charm-jet rejections vs 1-tagging efficiency for the SV1 tagging algorithms, evaluated in CC̄events with 200 pile-up events, compared for a pixel pitch of 50 × 50 `m2 (full line) and 25 × 100 `m2 (dashed line).The results are compared in different [ regions (a-b) and ?) regions (c-d).

13

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

1

1.2

1.4

2mµ

25x1

00 /

50x5

0

1

10

210

310

410

Ligh

t-je

t rej

ectio

n

|<1η 0<|2mµ50x50

|<1η 0<|2mµ25x100

|<2η 1<|2mµ50x50

|<2η 1<|2mµ25x100

|<3η 2<|2mµ50x50

|<3η 2<|2mµ25x100

|<4η 3<|2mµ50x50

|<4η 3<|2mµ25x100

IP3D+SV1 ITk Layout

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

1

1.1

2mµ

25x1

00 /

50x5

0

1

10

c-je

t rej

ectio

n

|<1η 0<|2mµ50x50

|<1η 0<|2mµ25x100

|<2η 1<|2mµ50x50

|<2η 1<|2mµ25x100

|<3η 2<|2mµ50x50

|<3η 2<|2mµ25x100

|<4η 3<|2mµ50x50

|<4η 3<|2mµ25x100

IP3D+SV1 ITk Layout

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

(a) (b)

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

1

1.2

1.4

2mµ

25x1

00 /

50x5

0

1

10

210

310

410

Ligh

t-je

t rej

ectio

n

<100 GeVT

20<p2mµ50x50

<100 GeVT

20<p2mµ25x100

<300 GeVT

100<p2mµ50x50

<300 GeVT

100<p2mµ25x100

<600 GeVT

300<p2mµ50x50

<600 GeVT

300<p2mµ25x100

|<4.0ηIP3D+SV1 ITk Layout |

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

0.9

1

1.1

2mµ

25x1

00 /

50x5

0

1

10

c-je

t rej

ectio

n

<100 GeVT

20<p2mµ50x50

<100 GeVT

20<p2mµ25x100

<300 GeVT

100<p2mµ50x50

<300 GeVT

100<p2mµ25x100

|<4.0ηIP3D+SV1 ITk Layout |

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

(c) (d)

Figure 9: Light-jet and charm-jet rejections vs 1-tagging efficiency for the IP3D+SV1 tagging algorithms, evaluatedin CC̄ events with 200 pile-up events, compared for a pixel pitch of 50 × 50 `m2 (full line) and 25 × 100 `m2 (dashedline). The results are compared in different [ regions (a-b) and ?) regions (c-d).

14

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

0.8

1

1.2

1.4

2mµ

25x1

00 /

50x5

0

1

10

210

310

410

Ligh

t-je

t rej

ectio

n

|<1η 0<|2mµ50x50 |<1η 0<|2mµ25x100

|<2η 1<|2mµ50x50 |<2η 1<|2mµ25x100

|<3η 2<|2mµ50x50 |<3η 2<|2mµ25x100

|<4η 3<|2mµ50x50 |<4η 3<|2mµ25x100

MV2 ITk Layout

ATLAS Simulation Preliminaryt>=200, tµ = 14 TeV, <s

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

0.8

0.9

1

1.1

1.22mµ

25x1

00 /

50x5

0

1

10

210

c-je

t rej

ectio

n

|<1η 0<|2mµ50x50 |<1η 0<|2mµ25x100

|<2η 1<|2mµ50x50 |<2η 1<|2mµ25x100

|<3η 2<|2mµ50x50 |<3η 2<|2mµ25x100

|<4η 3<|2mµ50x50 |<4η 3<|2mµ25x100

MV2 ITk Layout

ATLAS Simulation Preliminaryt>=200, tµ = 14 TeV, <s

(a) (b)

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

0.8

1

1.2

1.4

2mµ

25x1

00 /

50x5

0

1

10

210

310

410

Ligh

t-je

t rej

ectio

n

<100 GeVT

20<p2mµ50x50 <100 GeV

T 20<p2mµ25x100

<300 GeVT

100<p2mµ50x50 <300 GeV

T 100<p2mµ25x100

<600 GeVT

300<p2mµ50x50 <600 GeV

T 300<p2mµ25x100

MV2 ITk Layout

ATLAS Simulation Preliminaryt>=200, tµ = 14 TeV, <s

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

0.95

1

1.05

2mµ

25x1

00 /

50x5

0

1

10

210

c-je

t rej

ectio

n

<100 GeVT

20<p2mµ50x50

<100 GeVT

20<p2mµ25x100 <300 GeV

T 100<p2mµ50x50

<300 GeVT

100<p2mµ25x100

MV2 ITk Layout

ATLAS Simulation Preliminaryt>=200, tµ = 14 TeV, <s

(c) (d)

Figure 10: Light-jet (a) and charm-jet rejections (b) vs 1-tagging efficiency for the MV2 tagging algorithm, evaluatedin CC̄ events with 200 pile-up events, compared for a pixel pitch of 50 × 50 `m2 (full line) and 25 × 100 `m2 (dashedline). The results are compared in different [ regions (a-b) and ?) regions (c-d).

15

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

1

1.2

1.4

1.6

|<2.

5)η

ITk

/ Run

2 (

|

1

10

210

310

410

Ligh

t-je

t rej

ectio

n

|<1η||<2η1<||<3η2<||<4η3<|

|<2.5η||<2.5)ηRun-2 (|

2mµIP3D ITk Layout 50x50

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

0.98

1

1.02

1.04

1.06

|<2.

5)η

ITk

/ Run

2 (

|

1

10

c-je

t rej

ectio

n

|<1η||<2η1<||<3η2<||<4η3<|

|<2.5η||<2.5)ηRun-2 (|

2mµIP3D ITk Layout 50x50

ATLAS Simulation Preliminary

t>=200, tµ = 14 TeV, <s

(a) (b)

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

1

1.5

2

|<2.

5)η

ITk

/ Run

2 (

|

1

10

210

310

410

Ligh

t-je

t rej

ectio

n

|<1η0<||<2η1<||<3η2<||<4η3<|

|<2.5η||<2.5)ηRun 2 (|

2mµMV2 ITk Layout 50x50

ATLAS Simulation Preliminaryt>=200, tµ = 14 TeV, <s

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

b-jet efficiency

1

1.1

1.2

1.3

1.4

|<2.

5)η

ITk

/ Run

2 (

|

1

10

210

c-je

t rej

ectio

n

|<1η0<||<2η1<||<3η2<||<4η3<|

|<2.5η||<2.5)ηRun 2 (|

2mµMV2 ITk Layout 50x50

ATLAS Simulation Preliminaryt>=200, tµ = 14 TeV, <s

(c) (d)

Figure 11: Light-jet and charm-jet rejections vs 1-tagging efficiency for the IP3D (a-b) and MV2 (c-d) taggingalgorithms, evaluated in CC̄ events with 200 pile-up events, obtained with the analogue clustering and a pixel pitch of50 × 50 `m2. The results are compared in different [ regions. For comparison, the performance obtained with thecurrent Run 2 Inner Detector with an average pile-up of 30 is also shown [20].

16

0 0.5 1 1.5 2 2.5 3 3.5 4

|ηJet |

50

100

150

200

250

300

[GeV

]T

Jet p

0

0.001

0.002

0.003

0.004

0.005

0.006

0.007

0.008

0.009

light

-fla

vour

tagg

ing

rate

Simulation PreliminaryATLASt>=200, tµ = 14 TeV, <s

2mµMV2(70) ITk Layout 50x50

0 0.5 1 1.5 2 2.5 3 3.5 4

|ηJet |

50

100

150

200

250

300

[GeV

]T

Jet p

0

0.02

0.04

0.06

0.08

0.1

0.12

c-je

t tag

ging

rat

e

Simulation PreliminaryATLASt>=200, tµ = 14 TeV, <s

2mµMV2(70) ITk Layout 50x50

(a) (b)

0 0.5 1 1.5 2 2.5 3 3.5 4

|ηJet |

50

100

150

200

250

300

[GeV

]T

Jet p

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

b-je

t tag

ging

rat

e

Simulation PreliminaryATLASt>=200, tµ = 14 TeV, <s

2mµMV2(70) ITk Layout 50x50

0 0.5 1 1.5 2 2.5 3 3.5 4

|ηJet |

20

30

40

50

60

70

80

90

100

[GeV

]T

Jet p

0

0.001

0.002

0.003

0.004

0.005

0.006

0.007

0.008

0.009

0.01

PU

-jet t

aggi

ng r

ate

Simulation PreliminaryATLASt>=200, tµ = 14 TeV, <s

2mµMV2(70) ITk Layout 50x50

(c) (d)

Figure 12: Light-jet (a), charm-jet (b), 1-jet (c), and pile-up jet (d) tagging rates for the MV2 tagging algorithm,parameterized as functions of jet ?T and |[ |. The efficiencies are evaluated in CC̄ events with 200 pile-up events,obtained with the analogue clustering and a pixel pitch of 50 × 50 `m2, for the fixed Y1 = 70% operating point.

20<?) <100 GeV ?) >100 GeV|[ | < 2.5 Y1 = 70% light-jet rej. = 381 Y1 = 70% light-jet rej. = 329

Y1 = 85% light-jet rej. = 41 Y1 = 85% light-jet rej. = 512.5 < |[ | < 4.0 Y1 = 70% light-jet rej. = 51 Y1 = 70% light-jet rej. = 59

Y1 = 85% light-jet rej. = 6.8 Y1 = 85% light-jet rej. = 10.9

Table 2: Light-jet rejection achieved with the MV2 tagging algorithm for jets in different [ regions and ?) regimesfor operating points corresponding to an average 1-jet efficiency of Y1 = 70% and 85% in the relevant [ region,obtained with the analogue clustering and a pixel pitch of 50 × 50 `m2.

17

0 0.5 1 1.5 2 2.5 3 3.5 4

|ηJet |

50

100

150

200

250

300

[GeV

]T

Jet p

0

0.001

0.002

0.003

0.004

0.005

0.006

0.007

light

-fla

vour

tagg

ing

rate

Simulation PreliminaryATLASt>=200, tµ = 14 TeV, <s

2mµMV2(70) ITk Layout 25x100

0 0.5 1 1.5 2 2.5 3 3.5 4

|ηJet |

50

100

150

200

250

300

[GeV

]T

Jet p

0

0.02

0.04

0.06

0.08

0.1

0.12

c-je

t tag

ging

rat

e

Simulation PreliminaryATLASt>=200, tµ = 14 TeV, <s

2mµMV2(70) ITk Layout 25x100

(a) (b)

0 0.5 1 1.5 2 2.5 3 3.5 4

|ηJet |

50

100

150

200

250

300

[GeV

]T

Jet p

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

b-je

t tag

ging

rat

e

Simulation PreliminaryATLASt>=200, tµ = 14 TeV, <s

2mµMV2(70) ITk Layout 25x100

0 0.5 1 1.5 2 2.5 3 3.5 4

|ηJet |

20

30

40

50

60

70

80

90

100

[GeV

]T

Jet p

0

0.001

0.002

0.003

0.004

0.005

0.006

0.007

PU

-jet t

aggi

ng r

ate

Simulation PreliminaryATLASt>=200, tµ = 14 TeV, <s

2mµMV2(70) ITk Layout 25x100

(c) (d)

Figure 13: Light-jet (a), charm-jet (b), 1-jet (c), and pile-up jet (d) tagging rates for the MV2 tagging algorithm,parameterized as functions of jet ?T and |[ |. The efficiencies are evaluated in CC̄ events with 200 pile-up events,obtained with the analogue clustering and a pixel pitch of 25 × 100 `m2, for the fixed Y1 = 70% operating point.

18

5 Impact on di-Higgs Analyses

As highlighted in the previous section, the 1-tagging performance is expected to be improved over the Run2 performance with the upgraded ATLAS detector, in spite of the more challenging running conditions atHL-LHC. This will benefit physics analyses involving final states with 1-quarks, such as the ones targetingthe production of a pair of Higgs bosons [23]. At a given light-jet rejection rate of about 300 in the centralregion of the detector |[ | < 2.5, the 1-jet efficiency with the MV2 tagger can be increased by around 2%with respect to Run 2, resulting in an increase in acceptance by 4% for final states with two 1-quarks,such as the �� → 11̄WW and 11̄g+g− final states and by up to 8% for final states with four 1-quarks,like in the �� → 11̄11̄ final state. Those numbers are still expected to be improved in the future butrepresent an evolved understanding of the upgraded detector with respect to Ref. [4]. Taking into accountthe performance results presented in this note, the expected sensitivity of those analyses with 3 ab−1 hasbeen re-assessed and are presented in Table 3. This in particular includes a more realistic description ofthe detector material, representing slightly less optimistic conditions than those considered in Ref. [23].The analysis strategy is unchanged with respect to the one presented in Ref. [23]. In particular, an MV2working point with a 1-jet efficiency around 70% is used to identify 1-jets in the region |[ | < 2.5. Thesensitivity of the �� → 11̄11̄ and 11̄g+g− analyses is extrapolated from the existing Run 2 analyses [24,25], while the sensitivity of the �� → 11̄WW is estimated using dedicated Phase-II simulations.

Channel Statistical-only Statistical + Systematic�� → 11̄11̄ 1.2 0.5�� → 11̄g+g− 2.3 2.0�� → 11̄WW 2.1 2.0Combined 3.3 2.9

Table 3: Standard Model significance of the individual �� → 11̄11̄, �� → 11̄g+g− and �� → 11̄WW channels aswell as their combination, taking into account the 1-tagging performance results presented in this note, correspondingto the 50 × 50 `m2 pixel pitch. The analysis strategy is unchanged with respect to the one presented in [23].

19

6 Conclusion

This note presents the 1-tagging performance expected with the ATLAS detector at HL-LHC, taking intoaccount the latest developments in the ITk simulation. All the 1-tagging algorithms investigated havebenefited from a dedicated reoptimisation to exploit the new layout of the ITk detector. Thanks to this, the1-tagging performance is not expected to be significantly worse than in Run 2, in spite of the significantlylarger pile-up expected for HL-LHC. This optimisation will be pursued in the future, opening the possibilityto further improve the 1-tagging performance. The study of the performance obtained with a pixel pitchof 25 × 100 `m2 has also been shown to improve the light jet rejection with respect to the 50 × 50 `m2

configuration by 10 to 35% in most of the phase space. The expected 1-tagging performance will stronglybenefit the search for the di-Higgs process, for which the combined sensitivity is expected to be close to3f by the end of the HL-LHC data-taking. Further improvements related to 1-tagging with the Phase-IIATLAS detector are also expected by then. First of all, the option to reduce the radius of the ITk innermostpixel layer is currently being studied and is expected to improve the 1-tagging performance significantly.In addition, the use of discriminants based on Deep Neural Networks, such as the DL1 tagger introducedin Run 2 [20], will become possible as larger simulated samples become available for their training andthe performance in the forward region is expected to improve thanks to the use of the timing informationbrought by the High-Granularity Timing Detector [5], which was not yet included in these studies.

References

[1] G. Apollinari et al., High-Luminosity Large Hadron Collider (HL-LHC): Technical Design ReportV. 0.1, 2017, url: https://cds.cern.ch/record/2284929 (cit. on p. 2).

[2] ATLAS Collaboration, Expected Tracking Performance of the ATLAS Inner Tracker at the HL-LHC,ATL-PHYS-PUB-2019-014, 2019, url: https://cds.cern.ch/record/2669540 (cit. onpp. 2–7, 9).

[3] ATLAS Collaboration, Technical Design Report for the ATLAS Inner Tracker Strip Detector,ATLAS-TDR-025, 2017, url: https://cds.cern.ch/record/2257755 (cit. on p. 2).

[4] ATLAS Collaboration, Technical Design Report for the ATLAS ITk Pixel Detector, ATLAS-TDR-30,2017, url: https://cds.cern.ch/record/2285585 (cit. on pp. 2, 3, 5, 7, 8, 19).

[5] ATLAS Collaboration, Technical Proposal: A High-Granularity Timing Detector for the ATLASPhase-II Upgrade, 2018, url: https://cds.cern.ch/record/2623663 (cit. on pp. 2, 20).

[6] S. Frixione, P. Nason and G. Ridolfi, A Positive-weight next-to-leading-order Monte Carlo for heavyflavour hadroproduction, JHEP 09 (2007) 126, arXiv: 0707.3088 [hep-ph] (cit. on p. 2).

[7] T. Sjöstrand, S. Mrenna and P. Z. Skands, PYTHIA 6.4 Physics and Manual, JHEP 05 (2006) 026,arXiv: hep-ph/0603175 [hep-ph] (cit. on p. 2).

[8] ATLAS Collaboration, The ATLAS Simulation Infrastructure, Eur. Phys. J. C 70 (2010) 823, arXiv:1005.4568 [physics.ins-det] (cit. on p. 2).

[9] S. Agostinelli et al., GEANT4: A Simulation toolkit, Nucl. Instrum. Meth. A506 (2003) 250 (cit. onp. 2).

[10] ATLAS Collaboration, ATLAS Insertable B-Layer Technical Design Report, ATLAS-TDR-19, 2010,url: https://cds.cern.ch/record/1291633 (cit. on p. 4), Addendum: ATLAS-TDR-19-ADD-1, 2012, url: https://cds.cern.ch/record/1451888.

20

[11] E. Bouhova-Thacker et al., ‘Vertex reconstruction in the ATLAS experiment at the LHC’, 2008IEEE Nuclear Science Symposium Conference Record, 2008 1720 (cit. on p. 5).

[12] ATLAS Collaboration, Optimisation and performance studies of the ATLAS 1-tagging algorithmsfor the 2017-18 LHC run, ATL-PHYS-PUB-2017-013, 2017, url: https://cds.cern.ch/record/2273281 (cit. on pp. 5, 7–9).

[13] M. Cacciari, G. P. Salam and G. Soyez, The anti-:C jet clustering algorithm, JHEP 04 (2008) 063,arXiv: 0802.1189 [hep-ph] (cit. on p. 6).

[14] M. Cacciari, G. P. Salam and G. Soyez, FastJet User Manual, Eur. Phys. J. C 72 (2012) 1896, arXiv:1111.6097 [hep-ph] (cit. on p. 6).

[15] ATLAS Collaboration, Jet energy scale measurements and their systematic uncertainties in proton–proton collisions at

√B = 13TeV with the ATLAS detector, Phys. Rev. D 96 (2017) 072002, arXiv:

1703.09665 [hep-ex] (cit. on p. 6).

[16] ATLAS Collaboration, Performance of 1-jet identification in the ATLAS experiment, JINST 11(2016) P04008, arXiv: 1512.01094 [hep-ex] (cit. on pp. 7, 8).

[17] ATLASCollaboration, Secondary vertex finding for jet flavour identification with the ATLAS detector,ATL-PHYS-PUB-2017-011, 2017, url: https://cds.cern.ch/record/2270366 (cit. on pp. 7,8).

[18] ATLAS Collaboration, Topological 1-hadron decay reconstruction and identification of 1-jets withthe JetFitter package in the ATLAS experiment at the LHC, ATL-PHYS-PUB-2018-025, 2018, url:https://cds.cern.ch/record/2645405 (cit. on p. 9).

[19] ATLAS Collaboration, Performance of the ATLAS track reconstruction algorithms in denseenvironments in LHC Run 2, Eur. Phys. J. C 77 (2017) 673, arXiv: 1704.07983 [hep-ex] (cit. onp. 11).

[20] ATLAS Collaboration, ATLAS 1-jet identification performance and efficiency measurement with CC̄events in ?? collisions at

√B = 13TeV, (2019), arXiv: 1907.05120 [hep-ex] (cit. on pp. 11, 16,

20).

[21] ATLAS Collaboration, Tagging and suppression of pileup jets with the ATLAS detector, ATLAS-CONF-2014-018, 2014, url: https://cds.cern.ch/record/1700870 (cit. on p. 11).

[22] ATLAS and CMS Collaborations, ‘Report on the Physics at the HL-LHC and Perspectives for the HE-LHC’, HL/HE-LHC Physics Workshop: final jamboree Geneva, CERN, 2019, arXiv: 1902.10229[hep-ex] (cit. on p. 11).

[23] ATLAS Collaboration, Measurement prospects of the pair production and self-coupling of theHiggs boson with the ATLAS experiment at the HL-LHC, ATL-PHYS-PUB-2018-053, 2018, url:https://cds.cern.ch/record/2652727 (cit. on p. 19).

[24] ATLAS Collaboration, A search for resonant and non-resonant Higgs boson pair production in the11̄g+g− decay channel in ?? collisions at

√B = 13TeV with the ATLAS detector, Phys. Rev. Lett.

121 (2018) 191801, arXiv: 1808.00336 [hep-ex], Erratum: Phys. Rev. Lett. 122 (2019) 089901(cit. on p. 19).

[25] ATLAS Collaboration, Search for pair production of Higgs bosons in the 11̄11̄ final state usingproton–proton collisions at

√B = 13TeV with the ATLAS detector, JHEP 01 (2019) 030, arXiv:

1804.06174 [hep-ex] (cit. on p. 19).

21