PaperlayerView exact revision on arXiv

Source-linked AI summary

Performance of reconstruction and identification of $\tau$ leptons decaying to hadrons and $\nu_\tau$ in pp collisions at $\sqrt{s}=$ 13 TeV

CMS Collaboration

arXiv:1809.02816v3hep-ex

TL;DR

CMS improves HPS reconstruction and identification of hadronic τ decays, addressing backgrounds from jets and misidentified leptons. The updated algorithm is evaluated with 2016 collision data at 13 TeV; efficiencies are similar to simulation, while jet misidentification probabilities differ from simulation by 10–20%.

  • Problem

    τh reconstruction must distinguish hadronic τ decays from abundant quark and gluon jets, while electron or muon misidentification can also be a sizeable background.

    Source: §1 · p. 4

  • Method

    The improved HPS algorithm uses dynamic strip reconstruction and enhanced multivariate discriminants, with extensions for boosted τ pairs and the high-level trigger.

    Source: §1 · p. 4, §14 · p. 45

  • Results

    τh identification efficiencies are similar to simulation, jet misidentification probabilities differ from simulation by 10–20%, and electron and muon misidentification probabilities are smaller than for the previous algorithm under the same conditions.

    Source: §14 · p. 46

  • Takeaways & Limitations

    A specialized τh reconstruction and identification algorithm is used in the high-level trigger.

    Source: §14 · p. 46

    Jet-to-τh misidentification probabilities are indicative mainly for W →µν+jet events, which contain a large fraction of light-quark jets and have relatively high misidentification probability.

    Source: §10.1 · p. 34

Abstract

from arXiv · show

The algorithm developed by the CMS Collaboration to reconstruct and identify $\tau$ leptons produced in proton-proton collisions at $\sqrt{s}=$ 7 and 8 TeV, via their decays to hadrons and a neutrino, has been significantly improved. The changes include a revised reconstruction of $\pi^0$ candidates, and improvements in multivariate discriminants to separate $\tau$ leptons from jets and electrons. The algorithm is extended to reconstruct $\tau$ leptons in highly Lorentz-boosted pair production, and in the high-level trigger. The performance of the algorithm is studied using proton-proton collisions recorded during 2016 at $\sqrt{s}=$ 13 TeV, corresponding to an integrated luminosity of 35.9 fb$^{-1}$. The performance is evaluated in terms of the efficiency for a genuine $\tau$ lepton to pass the identification criteria and of the probabilities for jets, electrons, and muons to be misidentified as $\tau$ leptons. The results are found to be very close to those expected from Monte Carlo simulation.

1 Introduction

Improved τh reconstruction and identification supports LHC searches where τ leptons provide sensitive Higgs-coupling probes and help distinguish new-physics signatures. The updated HPS algorithm revises π0 reconstruction and strengthens discrimination against jets and electrons.

1 Introduction

  • Improved τ reconstruction benefits searches for Higgs bosons, leptoquarks, supersymmetric particles, and gauge bosons.The Higgs-to-ττ channel is especially sensitive to Higgs couplings to fermions; the SM Higgs decay to τ pairs was reported in [1] [2].Source: §1 · p. 4
  • The HPS algorithm reconstructs τh by combining charged hadrons with π0 candidates built from photon and conversion-electron candidates in η×φ strips.Its main identification challenge is separating τh candidates from abundant quark and gluon jets; electron and muon misidentification can also be sizeable backgrounds.Source: §1 · p. 4
  • The updated τh algorithm adds dynamic strip reconstruction and improved multivariate discriminants against jets and electrons.Dynamic strips collect π0 decay products more effectively; the jet discriminant combines isolation, τ lifetime, and shower-energy information.Source: §1 · p. 4
  • The paper evaluates the updated algorithm, its boosted-pair extension, and its specialized high-level-trigger version, then measures efficiency and misidentification performance.The paper also covers event selection, systematic uncertainties, and the τh energy scale.Source: §1 · p. 5

2 The CMS detector

The CMS detector combines tracking, calorimetry, and muon systems around a superconducting solenoid, with a two-tier trigger selecting events for storage.

2 The CMS detector

  • A 6 m superconducting solenoid provides a 3.8 T field containing the silicon tracker and barrel-and-endcap ECAL and HCAL, while forward calorimetry extends coverage.Muon detectors are embedded in the steel flux-return yoke outside the solenoid.Source: §2 · p. 5
  • The silicon tracker covers |η| < 2.5, and its material causes many photons from π0 decays to convert into e−e+ pairs.The tracker contains 1 440 pixel and 15 148 strip modules; its material ranges from at least 0.4 X0 centrally to about 2.0 X0 near |η| ≈ 1.4.Source: §2 · p. 5
  • The PbWO4 ECAL covers |η| < 3.0 and is more than 25 X0 thick, while the brass-and-scintillator HCAL extends to |η| = 3.0.The ECAL comprises a barrel and two endcaps; the HCAL thickness is 7–11 interaction lengths depending on η.Source: §2 · p. 5
  • Four planes of gas-ionization detectors measure muons, using drift tubes in the barrel and cathode strip chambers in the endcaps, with RPCs used only for triggering.Source: §2 · p. 5
  • The trigger reduces the bunch-crossing rate of up to 40 MHz to about 100 kHz at L1 and about 1 kHz at HLT before storage.L1 uses custom hardware and has a fixed interval below 4 µs; HLT runs optimized full-event reconstruction software.Source: §2 · pp. 5–6

3 Data and simulated events

The performance study uses 2016 CMS data at 13 TeV and simulated signal and background samples spanning the processes and τh momenta relevant to evaluation.

3 Data and simulated events

  • Performance is evaluated with 35.9 fb−1 of 2016 pp data at √s = 13 TeV and simulated signal samples including H →ττ, Z′ →ℓℓ, W′ →ℓν, and Z/γ∗→ℓℓ.Signal samples, including MSSM H →ττ, cover τh identification over a wide pT range; simulated matching uses a cone of ∆R < 0.3.Source: §3 · p. 6
  • Background simulations include W+jets, Z/γ∗→ℓℓ, top-quark, diboson, and QCD multijet production, with process-specific event generators and cross-section normalizations.The samples use PYTHIA for parton showering and hadronization, and NNPDF3.0 parton distributions.Source: §3 · p. 6
  • Pileup interactions are overlaid on simulated events according to the analyzed data’s luminosity profile, followed by GEANT4 detector simulation and CMS event reconstruction.Source: §3 · p. 6

4 Event reconstruction

CMS reconstructs particles with particle flow and combines them into higher-level objects, including τh candidates, jets, and missing transverse momentum.

4 Event reconstruction

  • The particle-flow algorithm combines CMS subdetector information to reconstruct particles, which are grouped into jets, missing transverse momentum, τh candidates, and lepton-isolation inputs.Source: §4 · p. 7
  • Tracks come from charged-particle hits in the silicon tracker; electrons use tracker trajectories and ECAL clusters, while muons combine inner-tracker and outer-spectrometer tracks.Electron identification uses shower and tracker/calorimeter observables, including bremsstrahlung-sensitive categories; muon identification uses track quality and minimal calorimeter energy deposition.Source: §4 · p. 7
  • Electron and muon isolation sums charged and neutral hadrons and photons within ∆R < 0.3 and 0.4, respectively, around the lepton direction.The primary interaction vertex is the reconstructed vertex with the largest summed jet pT², with jets formed from its assigned tracks.Source: §4 · p. 7
  • Charged hadrons not originating from the primary vertex are excluded to suppress pileup, and the neutral pileup contribution is estimated separately for electrons and muons.For muons, the charged-hadron pT sum from non-primary vertices is scaled by 0.5 to estimate the neutral contribution.Source: §4 · p. 7
  • Jets use anti-kT clustering with distance parameter 0.4, while missing transverse momentum is the negative vector sum of reconstructed-particle momenta projected perpendicular to the beams.Jet momentum corrections are propagated to missing transverse momentum.Source: §4 · pp. 7–8

5 Reconstruction and identification of τh

Section 5 describes the HPS algorithm for reconstructing hadronic τ candidates and the discriminants used to distinguish them from jets, electrons, and muons. It covers the algorithm’s decay-mode reconstruction, dynamic strip clustering, and mass constraints.

5 Reconstruction and identification of τh

  • The 13 TeV HPS algorithm retains its basic earlier features while improving π0 reconstruction and describing discriminants against jets, electrons, and muons.The section also covers reconstruction of τh pairs in highly boosted topologies and a trigger-specific version.Source: §5 · p. 8

5.1.1 Dynamic strip reconstruction

  • Fixed-size strips can leave τ-decay electrons and photons outside the window, reducing isolation efficiency for genuine τh candidates.Nuclear interactions and photon conversions can also produce lower-pT particles outside the strip; for high-pT τh, smaller strips can reduce background contributions while retaining decay products.Source: §5.1.1 · p. 9
  • The highest-pT unclustered e/γ seeds each strip, whose pT-weighted center must lie inside the signal cone even when part of the strip extends outside it.The strip position is recomputed from its e/γ constituents, and clustering continues until no other e/γ candidate lies within the window.Source: §5.1.1 · pp. 9–10
  • Strip dimensions vary with e/γ pT, fitting the 95% envelope of τh-to-e/γ separations and imposing upper and lower limits in η and φ.The fitted dimensions are bounded between 0.05 and 0.15 in Δη and 0.05 and 0.3 in Δφ.Source: §5.1.1 · p. 10, Figure 1 · p. 11

5.1 The hadrons-plus-strips algorithm

  • HPS combines charged particles and strips into decay hypotheses, applying visible-mass compatibility with ρ(770) or a1(1260) resonances for specified modes.Table 1 lists τ decay modes, branching fractions, and intermediate resonances; the algorithm does not consider h−h+h−π0 because of its greater jet contamination.Source: Table 1 · p. 8, §5.1 · p. 9, §5.1.2 · p. 10

5.1.2 Mass constraints on decay modes

  • Visible-mass windows around the ρ(770) and a1(1260) are optimized by decay mode to maximize τh reconstruction efficiency relative to jet-to-τh misidentification probability.For h±π0 and h±π0π0, the upper mass-window limits are constrained to 1.3–4.2 GeV and 1.2–4.0 GeV, respectively; the h±π0 window enlarges at high pT.Source: §5.1.2 · p. 10
  • The mass change Δmτh accounts for the addition of e/γ candidates to a strip and is calculated using the τh and strip four-momenta.The displayed calculation uses derivatives with respect to the strip position in η and φ.Source: §5.1.2 · p. 11

5.2 Discrimination of τh candidates against jets

The jet-rejection strategy combines reoptimized isolation requirements with MVA discriminants, improving rejection while retaining τh efficiency. Its isolation-sum and MVA-based versions address different conditions, including high-object-multiplicity and high-pT events.

5.2 Discrimination of τh candidates against jets

  • A smaller isolation cone of ΔR = 0.3 improves τh efficiency in high-object-multiplicity ttH events, where nearby objects affect isolation.The standard cone was ΔR = 0.5.Source: §5.2 · p. 11

5.2.1 Isolation sum discriminants

  • The isolation-sum method estimates charged and neutral activity around τh candidates, suppressing pileup charged particles by vertex compatibility and estimating neutral pileup from nonvertex charged activity.Source: §5.2.1 · pp. 11–12
  • Requiring the scalar pT sum of strip e/γ candidates outside the signal cone reduces jet→τh misidentification by about 20% at similar efficiency.Dynamic strips can include photons outside the signal cone, lowering Iτh for misidentified candidates; the additional sum provides another rejection handle.Source: §5.2.1 · p. 12
  • At high efficiency, the current isolation-sum method’s jet misidentification probability rises faster than efficiency, as relaxing Iτh is constrained by the pstrip, outer requirement.Source: §5.2.1 · pp. 12–13

5.2.2 MVA-based discriminants

  • MVA discriminants combine isolation with τ-lifetime information; isolation sums and the Δβ correction are most powerful, while track impact parameters and flight length also contribute.Photon/electron multiplicity contributes at levels similar to lifetime variables for candidates decaying to three charged hadrons.Source: §5.2.2 · pp. 13–14
  • MVA-based discriminants reduce the expected jet→τh misidentification probability by a factor of 2 at efficiencies similar to isolation-sum discriminants.The two MVA training sets follow the same performance curve despite shifted working points, indicating stable performance.Source: §5.2.2 · p. 14
  • For the tight 2016 MVA working point, expected low-to-medium-pT τh efficiency is 49% with 0.21% jet→τh misidentification; at high pT, misidentification is 0.07% with efficiency constant.Source: §5.2.2 · p. 14

5.3 Discrimination of τ leptons against electrons

The electron discriminator is refined to reduce electron-to-τh misidentification while maintaining genuine τh selection efficiency. Added photon-related inputs exploit differences in photon content and distribution.

5.3 Discrimination of τ leptons against electrons

  • The electron MVA discriminator is improved to reduce electron→τh misidentification while maintaining high genuine-τh selection efficiency over a wide pT range.Electrons can mimic h± or h±π0 decays, particularly when tracker material causes bremsstrahlung photons to resemble neutral pions.Source: §5.3 · p. 14
  • Added classifier inputs include strip photon count, photon spread around the leading track, and the fraction of τh energy carried by photons.These quantities are computed separately for photons inside and outside the signal cone.Source: §5.3 · pp. 14, 16
  • Figure 5 evaluates genuine-τh efficiency versus reconstructed-candidate pT and electron→τh misidentification versus generated-electron pT across five MVA working points.The efficiency sample requires the loose isolation-sum working point; the electron sample uses pT > 20 GeV and |η| < 2.3, excluding 1.46 < |η| < 1.56.Source: Figure 5 · p. 16

5.4 Discrimination of τ leptons against muons

The muon discriminator vetoes τh candidates using nearby muon-system signals and calorimeter response. Its selection retains over 99% of genuine τh candidates while rejecting most muons misidentified as τh.

5.4 Discrimination of τ leptons against muons

  • The loose veto rejects candidates with track segments in at least two muon-detector planes within ΔR = 0.3, or with leading-hadron calorimeter energy below 20% of track momentum.Source: §5.4 · p. 17
  • The tight veto adds rejection when a hit appears in an outer muon station within ΔR = 0.3, after the loose-condition checks.Source: §5.4 · p. 17
  • Genuine τh selection efficiency exceeds 99%; muon→τh misidentification is approximately 3.5 × 10^-3 for loose and 1.4 × 10^-3 for tight working points.Source: §5.4 · p. 17

6 Reconstruction of highly boosted τ lepton pairs

A dedicated HPS variant reconstructs close, high-momentum τ-lepton pairs inside large-radius jets, where standard HPS is inefficient. Above about 0.5 TeV, it provides considerably higher efficiency than standard HPS, with misidentification probability of order 10^-4.

6 Reconstruction of highly boosted τ lepton pairs

  • For boosted boson decays, nearby τ decay products form one jet, so the standard algorithm designed for one τh per jet performs poorly.The dedicated variant uses Cambridge–Aachen large-radius jets with distance parameter 0.8; above 100 GeV jet pT, both τ decays are expected inside one such jet.Source: §6 · p. 17
  • The algorithm repeatedly splits each CA8 jet into subjets, requiring each subjet pT above 10 GeV and the heavier-subjet mass below two-thirds of the jet mass.If a split fails, it retries by splitting the heaviest subjet; if no acceptable pair is found, that CA8 jet yields no τh reconstruction.Source: §6 · p. 17
  • Accepted subjets seed HPS, after which relaxed decay-mode criteria recover three-prong τ decays when a track is lost in the dense high-pT environment.Isolation and decay-mode requirements use PF candidates belonging to the seed subjet; nearby identified electrons or muons flag semileptonic pair decays, while fully leptonic pairs are excluded.Source: §6 · pp. 17–18
  • Above ≈0.5 TeV, the boosted algorithm has considerably higher τ-pair efficiency than standard HPS, while misidentification remains of order 10^-4.The passage says the expected increase in misidentification probability is compatible with use in this high-pT regime because backgrounds are highly suppressed.Source: §6 · p. 18

7 Identification of τh candidates in the high-level trigger

The trigger uses dedicated τh identification alongside offline reconstruction, with L1 upgrades permitting more sophisticated algorithms and HLT reconstruction adapted for fast processing. Its HLT path proceeds through calorimeter, pixel-track, and full-track/PF stages, while L3 identification uses inclusive reconstruction and isolation criteria.

7 Identification of τh candidates in the high-level trigger

  • HLT τh reconstruction uses specialized fast or regional algorithms and multistep selection to meet an average 150 ms decision time, about 100 times faster than offline reconstruction.The HLT uses full-granularity information from CMS subdetectors, but its reconstruction differs slightly from offline processing.Source: §7 · p. 18
  • The three HLT stages progress from calorimeter-only L2 jets with a pT threshold, through pixel-track isolation at L2.5, to full tracking and PF reconstruction at L3.At L2, anti-kT jets use distance parameter 0.2 within ΔR < 0.8 around L1 τh objects; L3 uses pixel and strip detectors in regions of Δη×Δφ = 0.5×0.5.Source: §7 · pp. 18, 20
  • L3 candidates combine up to three leading charged hadrons and available π0 candidates inside a variable signal cone, with nearby neutral hadrons included to recover tracking inefficiencies.The signal cone is limited to ΔRL3sig = 0.08–0.12; criteria are deliberately inclusive, with larger signal and smaller isolation cones than offline reconstruction.Source: §7 · p. 20
  • For 2016 L3 τh candidates, charged isolation sums other charged hadrons with dz < 0.2 cm in the isolation cone, with loose, medium, and tight thresholds of 3.0, 2.0, and 1.5 GeV, respectively.Combined isolation includes photons in the annulus between the signal and isolation cones that do not belong to signal strips, plus a pileup-neutral contribution estimated using the jet-area method.Source: §7 · p. 20
  • Isolation cutoffs may be relaxed by a few percent as τh pT rises, increasing genuine-candidate efficiency while the misidentified-candidate rate decreases with pT and trigger rates remain controlled.The photon-strip sum outside the signal cone was unused in 2016 τh triggers but included in triggers during 2017 data taking.Source: §7 · pp. 20–21

8 Event selection and systematic uncertainties

The section defines data samples for measuring τh reconstruction and identification performance and describes common systematic uncertainties. The selections use control and tag-and-probe samples tailored to genuine τh efficiency, lepton-to-τh misidentification, and background normalization.

8 Event selection and systematic uncertainties

  • Simulated events are reweighted for trigger, identification, isolation, and pileup differences between data and simulation; the section also sets out shared measurement uncertainties.The uncertainty overview covers luminosity, lepton efficiencies, production normalizations, τh energy scale, simulated-event statistics, boson and top-quark spectrum weights, and pileup.Source: §8 · p. 21, §8.9 · pp. 23–24

8.1 The Z/γ*→ττ events

  • Z/γ*→ττ tag-and-probe samples require an isolated electron or muon tag and a loosely preselected τh probe, with kinematic, charge, separation, and background-suppression selections.The eτh and μτh selections use different single-lepton trigger and offline thresholds; transverse-mass and Dζ requirements reduce W+jets contamination.Source: §8.1 · p. 21

8.2–8.3 Control samples

  • The tt→μτh+jets sample follows the μτh selection except that mT and Dζ cuts are omitted, and it requires a b-tagged jet.A separate high-purity Z/γ*→μμ sample constrains the Drell–Yan normalization used in the τh-efficiency tag-and-probe measurement.Source: §8.2 · pp. 21–22, §8.3 · p. 22

8.4–8.5 Off-shell W samples

  • Off-shell W→τν events are selected in a high-pT, back-to-back topology with no additional hard jet or lepton; off-shell W→μν events constrain the W→τν normalization.The τh sample requires pT > 100 GeV and missing transverse momentum above 120 GeV; the muon control sample requires pT > 120 GeV.Source: §8.4 · p. 22, §8.5 · p. 22

8.6–8.7 Background-control samples

  • W→μν+jet events require a high-mT muon, exactly one central jet balanced against the W, while tt eμ events require an isolated electron and muon.The W+jets selection uses mT > 60 GeV and a jet-to-W pT ratio between 0.7 and 1.3.Source: §8.6 · p. 22, §8.7 · p. 23

8.8 Lepton-to-τh misidentification samples

  • High-purity Z→ee and Z→μμ tag-and-probe samples measure electron- and muon-to-τh misidentification probabilities using tight τh isolation and mass-window requirements.The selected τh must also satisfy decay-mode, charge, separation, and against-muon criteria as applicable; mT ≤ 30 GeV suppresses W+jets and tt backgrounds.Source: §8.8 · p. 23

8.9 Systematic uncertainties affecting all studied final states

  • Common uncertainties include a 2.5% integrated-luminosity uncertainty, 1.2–3% τh energy-scale uncertainty, and normalization uncertainties for modeled processes.Table 2 summarizes the sources of uncertainty and whether each affects the measured distributions.Source: §8.9 · p. 23, Table 2 · p. 24

9 Measurement of the τh identification efficiency

This section measures τh identification and MVA-based isolation efficiencies and data-to-simulation scale factors across momentum ranges, using Z/γ*→ττ, tt, and off-shell W→τν events. It also presents a tag-and-probe measurement in highly boosted τ pairs, including the dependence on separation between the τh candidate and lepton.

9.1 Using the tag-and-probe method in Z/γ∗events

  • Isolation scale factors range from 0.92 to 0.99 with uncertainties of about 5%; electron- and muon-rejection scale factors are compatible with unity within 1–3%.For the isolation results, 3.9% is added in quadrature to the fit uncertainty for track-reconstruction efficiency.Source: §9.1 · pp. 25, 27, Table 3 · p. 26

9.2 Using tt events

  • In tt events, scale factors are slightly lower than in Z/γ*→ττ events, attributed to a higher simulated jet-to-τh misidentification probability; no significant differences are observed across decay modes.The tt measurements also cover different isolation working points and pτh regions and feed the high-pT extrapolation.Source: §9.2 · p. 27

9.3 Using off-shell W →τν events

  • For τh above 100 GeV, an off-shell W sample with mW > 200 GeV is fit jointly with a W→μν control sample to extract the efficiency scale factor and W normalization.The fit uses transverse-mass distributions and is designed to reduce uncertainty from the W-event normalization.Source: §9.3 · p. 29

9.5 Using the tag-and-probe method in Z/γ∗events for highly boosted τ lepton pairs

  • For highly boosted τ pairs, single isolated-lepton triggers enable a tag-and-probe measurement, but τh candidates within ΔR < 0.4 of the triggering lepton are inaccessible.The method selects high-pT Z-boson decays with one leptonically decaying τ and one hadronic τ.Source: §9.5 · p. 31

9.4 Extrapolation of the τh identification efficiency to large pτh

  • The scale factor is compatible with being constant versus pT: the fitted linear slope differs from zero by only about one standard deviation.A constant scale factor is used for the high-pT extrapolation, with an asymmetric uncertainty that includes the fit difference between constant and linear forms and track-reconstruction uncertainty.Source: §9.4 · p. 31, Figure 10 · p. 32

9.5 Using the tag-and-probe method in Z/γ∗events for highly boosted τ lepton pairs

  • DY backgrounds are modeled with simulation and split into matched signal and misidentified Z-decay contributions, while W+jets and tt shapes use simulation with normalizations from dedicated control samples.The W+jets control sample inverts the mT requirement; the tt control sample requires at least one b-tagged jet.Source: §9.5 · p. 32
  • QCD multijet background is estimated with a signal-like sample having inverted pmiss_T below 75 GeV, then extrapolated using control-sample ratios at lepton–τh ΔR between 0.8 and 2.0.Other-process contributions are subtracted using simulation in the inverted-requirement sample and the two control regions.Source: §9.5 · p. 32
  • Background-estimation uncertainties include limited control-sample statistics, and the tt and W+jets normalization uncertainties are 3% and 13%, respectively.The procedure also accounts for systematic uncertainties discussed in Section 8.9.Source: §9.5 · p. 32
  • Scale factors for all six MVA-based isolation working points are compatible with unity and measurements in Sections 9.1–9.3; no significant dependence on ΔR between τh and the lepton is observed.Passing and failing events are defined by whether τh passes or fails the given working point; fitted medium-working-point distributions are shown in Figure 11.Source: §9.5 · p. 33, Table 6 · p. 33, Figure 11 · p. 33

10 Measurement of the jet 7→τh misidentification probability

Jet→τh misidentification probabilities are measured in W→μν+jet and eμ+jets events and compared with simulation. The rates vary with jet kinematics and sample composition, so the W+jet results are mainly indicative for light-quark-rich events.

10.1 Using W →µν+jet events

  • The probability is the fraction of jets with pT > 20 GeV and |η| < 2.3 identified as τh; W+jet rates are mainly indicative for light-quark-rich samples.Jet flavor and charge can also affect misidentification probabilities.Source: §10.1 · pp. 33–34
  • W+jet fake probabilities are nearly constant versus ηjet and decrease with jet pT from about 40 GeV as jet isolation increases.Figure 12 shows the loose, medium, and tight MVA-isolation working points as a function of jet pT.Source: §10.1 · p. 34, Figure 12 · p. 35
  • In W+jet events, loose-WP misidentification probabilities range from 2.0 to 0.1% and tight-WP probabilities from 1.0 to less than 0.1%; observed probabilities differ from MC expectations by 10–20%, within shower-model and underlying-event variations.Source: §10.1 · p. 34, Figure 12 · p. 35

10.2 Using eµ+jets events

  • In eµ+jets, observed fake rates differ from simulation by 10–20%, reaching 50% in a few ηjet bins, and are lower than in W+jet events.The eµ+jets sample is dominated by top-quark events, with misidentified τh candidates mainly from b-quark jets.Source: §10.2 · p. 36
  • The lower eµ+jets rates are attributed to its larger b-quark-jet fraction: these jets are less collimated and less likely to pass τh isolation.Source: §10.2 · p. 36

11 The e/µ 7→τh misidentification probability

The electron and muon misidentification probabilities are measured with tag-and-probe methods in Z/γ* events. In both measurements, rates in data exceed simulation, with the probabilities depending on working point and detector region.

11.1 Measurement of the e 7→τh probability

  • The electron misidentification probability is measured with a tag-and-probe method in Z/γ*→ee events, fitting passing and failing categories simultaneously.The passing-category mvis distribution spans 60 < mvis < 120 GeV; the failing-category yield constrains the Z/γ*→ee normalization.Source: §11.1 · p. 36

11.2 Measurement of the µ 7→τh probability

  • The muon misidentification probability is measured in Z/γ*→µµ events by fitting passing and failing probe categories simultaneously in five |η| regions.The tag-and-probe pair mass fit uses 70 < mvis < 120 GeV.Source: §11.2 · p. 38

11.1 Measurement of the e 7→τh probability

  • In the barrel, electron misidentification probabilities range from ≈5% at the very loose working point to less than 0.1% at the very tight working point; endcap probabilities range from 0.1 to 10%.Data probabilities exceed simulation, with the difference increasing for tight and very tight working points in both regions.Source: §11.1 · pp. 36, 38, Figure 15 · p. 39

11.2 Measurement of the µ 7→τh probability

  • Muon misidentification probabilities are 0.1–0.5% for the loose working point and 0.03–0.40% for the tight working point, peaking at |η| between 0.8 and 1.2.Data rates exceed predictions, with the difference possibly increasing at larger |η| and appearing more significant for the tight working point.Source: Figure 17 · p. 40, §11.2 · p. 40

12 Measurement of the τh energy scale

The τh energy scale is measured by fitting mass distributions in eτh and µτh samples across three decay modes. The fits from the two final states are compatible, and the measurement is limited by systematic rather than statistical uncertainties.

12 Measurement of the τh energy scale

  • The correction is defined from the difference between average reconstructed τh energy and generator-level visible decay-product energy, with fits performed separately for h±, h±π0, and h±h∓h± decays.The data-to-simulation correction is extracted from mτh and mvis distributions in Z/γ* eτh and µτh samples.Source: §12 · p. 40
  • For h±π0 and h±h∓h± decays, mτh is more sensitive to the energy scale, while the h± mode uses mvis alone.Templates are made by shifting reconstructed τh energy between −6% and +6%; unmatched background templates are not shifted.Source: §12 · pp. 40–41
  • The measurement is limited by systematic rather than statistical uncertainties; stability checks motivate an additional 1.0% uncertainty, yielding a total uncertainty below 1.2%.The checks vary simulated-event samples, selection criteria, binning, and fit range.Source: §12 · p. 42
  • Fits to mτh and mvis distributions in eτh and µτh final states are compatible, and their combined τh energy-scale correction is reported in Table 7.The table gives the correction relative to the reconstructed energy from simulation, in percent.Source: Table 7 · p. 42, §12 · p. 42

13 Performance of τh identification in the high-level trigger

The 2016 τh triggers use staged reconstruction and selections tailored to cross-triggers, τhτh, and other final states, with thresholds adjusted to control rates. Measured τh trigger efficiencies agree well with simulation, and their turn-on reflects both isolation and energy-response differences between trigger and offline candidates.

13 Performance of τh identification in the high-level trigger

  • Cross-triggers use lepton or missing-momentum L1 seeds before full L3 τh reconstruction, while τh-only triggers first apply L2 and L2.5 filters, each with efficiency above 95% per τh candidate.The τh-only class also restricts HLT reconstruction to regions around L1 τh candidates to reduce processing time.Source: §13 · pp. 42–43
  • The µτh trigger requires an L1 muon above 18 GeV and an L3 τh above 20 GeV, and its rate is about 20 Hz at 1.4 × 10^34 cm−2 s−1.Two eτh trigger variants accommodate changing luminosity; their rates remain below 15 Hz in their designed operating ranges.Source: §13 · p. 43
  • The τhτh trigger dynamically adjusts its L1 pT threshold across 28–36 GeV to maintain a constant L1 rate, and its HLT rate is below 60 Hz at 1.4 × 10^34 cm−2 s−1.At L3, both τh candidates must exceed 35 GeV and pass medium charged isolation, later replaced by combined isolation at higher luminosity.Source: §13 · p. 43
  • The high-pT single-τh trigger covered phase space not covered by cross-triggers, requiring an isolated L1 τh with pT > 120 GeV; about 30 Hz was allocated to this trigger.Source: §13 · p. 44
  • Trigger efficiencies are measured with tag-and-probe data enriched in Z/γ*→ττ→µτh events, using a τh sample with purity above 95%.For the τhτh measurement, only the 2016 data-taking period using combined isolation is included.Source: §13 · pp. 44–45
  • The τh parts of the µτh and τhτh triggers agree well between data and simulation; nominal τh pT thresholds correspond to 50% efficiency.The slow turn-on reflects relaxed HLT isolation at higher pT and an asymmetric HLT-to-offline energy response just above threshold.Source: §13 · p. 45, Figure 19 · p. 45

14 Summary

The CMS HPS algorithm is improved for 13 TeV data, including dynamic strip reconstruction, boosted τ-pair reconstruction, and changes to discriminants for rejecting jets and electrons. Measurements find genuine-τh efficiency broadly similar to simulation, while jet misidentification remains discrepant and electron and muon misidentification probabilities are reduced relative to the previous algorithm.

14 Summary

  • The improved algorithm adds dynamic strip reconstruction, boosted τ-pair reconstruction, and multivariate-discriminant variables to reject jets and electrons.The isolation discriminants were also optimized for the large pileup in 13 TeV collisions.Source: §14 · pp. 45–46
  • Genuine-τh efficiency at low, intermediate, and high pT and for highly boosted τ pairs is similar to simulation, while jet→τh misidentification probabilities differ by 10–20%.The study uses 35.9 fb−1 of 2016 data at √s = 13 TeV.Source: §14 · p. 46
  • Electron and muon misidentification probabilities are lower than with the previous algorithm under the same running conditions while genuine-τh efficiency remains high.The τh energy-scale response relative to simulation is close to unity, and performance of the specialized HLT algorithm is presented.Source: §14 · p. 46

Related papers

Loading 1809.02816v3…