Tau trigger and identification at CMS in Run-2

Transcription

1 Tau trigger and identification at CMS in Run-2 Olivier DAVIGNON*, for the CMS Collaboration * Laboratoire Leprince-Ringuet (École Polytechnique / CNRS-IN2P3) The 14th International Workshop on Tau Lepton Physics IHEP Beijing China 21/9/16

2 CMS τhτh 1 eτh -2 Loose VBF tag -3 Obs / Bkg SM H!ττ (D. Zanzi, L. Dodd s talks) 1 Data µτe SM Higgs Z->ττ 6 +- Z->l l tt, t, t 5 CMS 2.3 fb (13 TeV) Preliminary τ 3µ Wγ LFV VBF Higgs (BR=1%) mod+ mh=126 GeV 2 mh=125 GeV expected H µ τ A/H bb (arxiv: ) A/H/h µ µ (arxiv: ) = 2 BR<5% 2 BR<% ma [GeV] /v mτ 4 4 µ collinear M(µτe) (GeV) 2 m 15 1 µ BR<1% 5 BR<.1% H WW/ZZ (arxiv: ) Yτ τ Y µ 3 M(µτe)col [GeV] -.5 CMS Z τ τ Preliminary 3 Electroweak tt QCD Background uncertainty h,h,a ττ Observation mmod+, h ma = GeV, tanβ=5-5 H± τν (arxiv: ) fb (13 TeV) τhτh b-tag Expected exclusion 95% CL A/H/h ττ (HIG4-29) mh=123 GeV mh=122 GeV observed 15 MW'[GeV] Observed exclusion 95% CL MSSM m h 8 CMS Preliminary 5.1 fb (7 TeV) fb (8 TeV) 4 τ µ γ 2 5 mh=124 GeV LFV GF Higgs (BR=1%) 3 Expected ± 1 σ Graviton, radion, Z, W 6 5 Misidentified leptons.5 Expected 1 1 VV 4 Observed 2 Post-fit background unc. µτe 2-Jet Yτ µ Events / 5 GeV 2.3 fb (13 TeV) 7 σ SSM W' LO Preliminary σ SSM W' NNLO PDF α s v Only (current) access to H-lepton coupling Provides best constraints on VBF CMS Bayesian 95% CL Exotic searches to τ s (Z. Mao s talk) mττ [GeV] v v Preliminary m(τh, τh, ET) [GeV] 3 tanβ 2 miss τ+et Expected ± 2 σ 2.3 fb (13 TeV) CMS Observed Z/ γ * ll QCD tt W+Jets VV x Z'(15) ττ dn/dmt,ττ (1/GeV) fb (13 TeV) CMS Preliminary 1 Yµ τ 1 LFV H!µτ (A. Nehrkorn s, K. De Bruyn s talks) Obs/Exp.8 2 σ B(W' τν) [fb] Events / GeV SM H(125 GeV) ττ Observed Z ττ Z ee W+jets tt QCD Bkg. uncertainty 1. Data-BG BG dn/dmττ [1/GeV] CMS, 19.7 fb at 8 TeV mt,ττ (GeV) MSSM h,h,a!ττ / H!hh!ττbb / A!Zh!llττ / H±!τν (L. Dodd s talks) v Provides best constraints on extended scalar sector 2

3 Challenges in τ h channels at CMS Signal: reconstruction of genuine hadronic taus (τ h ) [1/GeV] dn/dm ττ CMS, 19.7 fb at 8 TeV SM H(125 GeV) ττ Observed Z ττ Z ee W+jets tt QCD Bkg. uncertainty eτ h Loose VBF tag Backgrounds: Rejection of fake τ h from electrons / muons Rejection of fake τ h from QCD jets m ττ [GeV] + efficient trigger selection + rejection of pileup (LHC currently features up to ~4 collisions / bunch crossing) + reliable description of data using simulation 3

4 4 Overview 1 Run II τ h -trigger v Phase upgrade of L1 trigger: challenges & system overview v L1 τ h -algorithm: description and performance v τ h s at High Level Trigger (HLT) 2 Run II τ h -identification v Reconstruction based on Particle Flow algorithm v Jet!τ h discriminators v Anti-electron discriminator v Performance measured in Run-2 data

5 5 CMS trigger system v Select interesting processes that have cross sections orders of magnitude lower than proton-proton cross section v Cannot store all events (TB/s) v Fast and efficient selection, adapted to interesting physics processes with scales between O() GeV to several TeV à Rate of events that enter the system (~4 MHz), driven by LHC luminosity Implementation in two levels v Level (output rate khz): electronics-based, decision to keep/reject an event in few µs v High Level Trigger (output rate 1 khz): CPU-based, advanced reconstruction of the objects, decisicion to be taken in O(.2) s

6 6 CMS L1 trigger in 216 Complete replacement of L1 trigger system (hardware, software) Upgrade commissioned this past year Brand new architecture The key conceptual changes to the L1 calorimeter trigger are v Streaming data from single event into one FPGA v Dynamic clustering of energy deposits v Pile-up L1 The MP7 card on which runs the L1 calorimetric object algorithms (including τ h )

7 The L1 τ h -algorithm (1/2) a.u..12 preliminary Energy response and resolution CMS 12.9 fb (13 TeV) Barrel Endcaps L1 offline E T / p T Dynamic clustering of trigger towers (TT) ECAL + HCAL energies Designed to capture the τ-footprint Improved energy containment/ resolution Minimized effect of pileup Merging Merging neighboring clusters (~15%) Capture h ± π s (1-prong + π s) decays Energy calibration Improved energy response and resolution à sharp turn-ons Function of E T, pseudorapidity, merging and electromagnetic fraction Cluster shapes: Primary cluster Secondary cluster 7

8 8 The L1 τ h -algorithm (2/2) Isolation Isolation energy: E T (6x9 TT)-E T (τ-clusters) Sums of ECAL and HCAL energies Cut: isolation energy requirement function of p T, pileup and pseudorapidity η=-ln[tan(θ/2)] Isolation energy cut relaxed as function of p T Allows control of the rate, increases τ h -purity Enables pileup resilient selection of τ h at L1! Keep low p T thresholds to efficiently select physics events: H!ττ!τ h τ h H!ττ!eτ h, etc.

9 9 L1 τ h -trigger performance in data (1/3) Measurements based on ~ 12.9 fb of 216 data, using the Tag & Probe technique on µ+τ events selected by a single-µ trigger Resolution in position <.1 in both η and φ directions a.u CMS preliminary fb (13 TeV) Barrel Endcaps a.u. CMS.1 preliminary fb (13 TeV) Barrel Endcaps η L1 - η offline ϕ L1 - ϕ offline

10 L1 τ h -trigger performance in data (2/3) Measurements based on ~ 12.9 fb of 216 data, using the Tag & Probe technique on µ+τ events selected by a single-µ trigger Resolution in energy as function of p T and position in η (offl)> T RMS / <E T (L1)/p CMS preliminary fb (13 TeV) Barrel Endcaps (offl)> T RMS / <E T (L1)/p CMS preliminary fb (13 TeV) offline p [GeV] T v Resolution in energy improves with p T η offline v Resolution in energy ~ stable with η

11 L1 τ h -trigger performance in data (3/3) Measurements based on ~ 12.9 fb of 216 data, using the Tag & Probe technique on µ+τ events selected by a single-µ trigger Turn-ons: L1 τ-trigger efficiency w.r.t. offline selection as function of p T, for L1 E T thresholds of 28, 3 and 32 GeV 12.9 fb (13 TeV) L1 Efficiency CMS preliminary 216 Isolated seeds L1 T L1 T L1 T Isolated, E 28 GeV Isolated, E 3 GeV Isolated, E 32 GeV v In spite of increasing LHC lumi*, thresholds for L1 di-τ h -iso maintained in the GeV range throughout 216 à Large improvement over Run where we relied on jet triggers v Also: definition of µ+τ h, e+τ h, τ h +MET L1 à Possible to design many more algorithms thanks to the flexibility of the system offline p T [GeV] * Now inst. luminosity up to ~1.3 x 34 cm -2.s 11

12 12 HLT τ h -trigger: description Particle Flow v v v v Global event reconstruction Optimal combination of information from CMS subdetectors Tracking/magnetic field integral are key ingredients Separation between neutral and charged hadrons à List of stable particles (e/µ/γ/hadrons) à Combines them to build jets, τ s, MET HLT τ h -reconstruction: based on PF@HLT v τ h are isolated jets v Track-based isolation (PU resilient)

13 13 HLT τ h -trigger performance in Run-2 data HLT τ-trigger performance measured in 216 data with respect to an offline selection, and using the Tag & Probe technique in events triggered by a single-µ trigger CMSPreliminary 216 (13 TeV) HLT efficiency data High Level Trigger efficiency of the τ h - leg in µ+τ h events selected as in the H!ττ!µτ h analysis offline τ p T [GeV]

14 14 τ h reconstruction E τ h CMS: Hadron Plus Strips (HPS) algorithm v τ h build from combination of Particle Flow π ± and π!γγ candidates (strips) v Reconstruction of different decay modes ECAL strip Essentially same algorithm as Run New: dynamic strip size in 1-prong + π s reconstruction, function of expected e/γ p T 1-prong 1-prong + π s 3-prong

15 Jetàτh rejection: isolation Main handle in τh discrimination against QCD jets: isolation Cut based discriminant based on energies of PF particles in isolation cone MVA: ~5% better background rejection at same signal efficiency MVA discriminant w/ τh lifetime information CMS Simulation Preliminary Mis-ID probability v Charged and neutral energies in isolation cone v Reconstructed decay mode type v Transverse impact parameter + significance v Distance between PV and SV + significance v Shape variables v e/γ multiplicity τ p h > 2 GeV, ητ < 2.3 T h 13 TeV, 2 pileup at 25ns 2 cut-based MVA-based 3 Efficiency : H ττ MC Fake rate : QCD multi-jet MC (2 < p < GeV) T τh identification efficiency 15

16 eàτ h rejection Electrons can fake 1-prong + π if: v e track + deposit compatible w/ τ h s prong v Brem. γ(s) mimick π!γγ (strip) MVA-based discriminant v Fractions of energy in HCAL and ECAL v Photon multiplicity v Energy-weighted difference in η/φ position between leading track and strips v Visible mass of particles in signal cone Several working points defined efficiency Expected τ CMS Simulation Preliminary 1.4 MVA anti-electron discriminator very tight tight medium loose very loose Relative yield / bin -2-3 CMS Simulation Preliminary 1 τ h e E ECAL / (E +E HCAL ) ECAL Relative yield / bin -2-3 CMS Simulation Preliminary 1 τ h e Fake rates between and -4 fake rate Expected e τ CMS Simulation Preliminary MVA anti-electron discriminator very tight tight medium loose very loose in-sigcone m τh (GeV).2 * Z/γ ττ τ p h T (GeV) 4 5 * Z/γ ee e p T (GeV) 16

17 τ h -ID efficiency in Run-2 data Three data-driven methods used to measure τ-id efficiency 1 Tag & Probe on Z!µτ data 2 Measurement of the Z!ττ/Z!µµ ratio 3 Measurement in W*!τν events Complementary methods (different systematics) High p T Events/bin Obs./Exp CMS Observed 2 Preliminary Z τ µ τ h 18 6 Pass DY others tt+jets 16 5 Electroweak 14 QCD multijet 12 4 Uncertainty Pass 215, 2.3 fb (13 TeV) m vis (GeV) Events/bin Obs./Exp Fail CMS Preliminary Fail 215, 2.3 fb (13 TeV) Observed Z τ µ τ h DY others tt+jets Electroweak QCD multijet Uncertainty m vis (GeV) Discriminant Data/MC scale factor (m vis ) MVA Very Loose 1. ±.6 MVA Loose 1. ±.6 MVA Medium 1.1 ±.6 MVA Tight 1. ±.6 Cut-based Loose.98 ±.6 Cut-based Medium.98 ±.6 Cut-based Tight.97 ±.6 Simulation gives a good description of data (SFs are compatible within unity) The three methods give compatible results (SFs are compatible within uncertainties of ~ 6%) 17

18 18 jetàτ h fake rate in Run-2 data Measured in W!µν + jets data (dominated by jet!τ h fakes) v Dedicated selection for fakes v Real τ h contribution subtracted based on MC Fake rates at the level of a few % (low p T ) and less than 1% (high p T ) Fake rate higher in data than in MC! likely due to hadronization/ue tune in MC

19 τ h -energy scale in Run-2 data τ h -ES (= correction on energy scale for simulation) v Using Tag & Probe in Z!µτ h data v Data fitted using τ h -ES-shifted MC templates τ h -ES= -3% τ h -ES= % Fit variables: 1-prong: fit of the m vis (µτ h ) 1-prong + π s, 3-prong: m vis (τ h ) τ h -ES= +3% Measured for different τ h reconstructed decay modes Decay mode τ h -ES [%] 1-prong +. ± prong + π s +1. ±.4 3-prong -.1 ±.2 19

20 2 Summary Exciting developments at CMS in LHC Run-2 v L1 trigger upgrade à Substantial changes in architecture and algorithms à Maintains high efficiency to τ h thanks to isolation and PU subtraction v Improvements in τ h reconstruction and identification à Changes to better reconstruct 1-prong + π s decays à Re-training of anti-e/µ and MVA-isolation discriminants Full appraisal of τ h performance in Run-2 data LHC already delivered >31 fb this year! More improvements and nice results with τ h s to come in the next years

21 21 Backup

22 22 References L1 trigger, and in particular taus: v CMS-DP v JINST 11 (216) 2, C28 v CERN-LHCC , CMS-TDR2 (213) v CMS-DP v CMS-DP HLT tau trigger performance in 216 data v CMS-DP Tau reconstruction & ID in Run and Run-2 v Run: JINST 11 (216) P19 v Run-2: u CMS-PAS-TAU6-2 u Update of tau energy scale & muon fake rate: CMS DP6-4 u Preliminary performance: CMS-DP , CMS-DP

23 23 The CMS detector Superconducting magnet Field: 3.8 T à Compact detector

24 The CMS detector Inner Tracker Pixels & silicon strips à Charged tracks reconstruction with 5µm resolution t/x CMS Simulation 2.5 Support Tube TOB Pixel TEC TIB and TID Beam Pipe η 24

25 The CMS detector Electromagnetic Calorimeter homogeneous calorimeter, scintillating crystals (PbWO4) à Electron/photon reconstruction à Excellent energy and position resolution crystals 26 X ~ 23 cm (Rmoliere)2~ (22mm)2 25

26 26 The CMS detector Hadronic calorimeters plastic scintillators à Hadronic jets reconstruction (charged and neutral components) Muon chambers In the return yoke à Muon reconstruction

27 27 The CMS detector Compact and hermetic detector: measurement of the missing transverse energy E T Miss = -Σp T vis 2 trigger levels Level 1 (L1) 4 MHzàkHz High Level Trigger (HLT) khzà O(1 khz)

28 28 Topics not discussed (1/2) Tau-charge mis-identification: (.22 ±.5)% (CMS TAU6-2) Electron to tau fake rate v Measured for different WPs v Ref: CMS TAU6-2 Muon to tau fake rate v Measured for different WPs v Ref: CMS DP6-4

29 29 Topics not discussed (2/2) Tau identification in boosted topologies v Ref: CMS-DP Tau identification efficiency from W*!τν events v Ref: CMS TAU6-2 v Use highly virtual W* events to measure the tau ID efficiency at very high p T (> GeV) Events / GeV , 2.3 fb Observed W µν electroweak tt+single top (13 TeV) CMS Preliminary Events / bin MediumIso 215, 2.3 fb Observed W τν bkgd (fake taus) bkgd (true taus) (13 TeV) CMS Preliminary Obs./Exp m T (GeV) Obs./Exp m T (GeV)

30 Ttrigger time multiplexing Each card spans 4 out of 72 towers in φ and all of η. 18 cards, each receiving 6 links at between 5. Gb/s & 6.4 Gb/s of Calorimeter data Layer Cards CTP7 CTP7 CTP7 CTP7 CTP7 CTP7 CTP7 CTP7 CTP7 CTP7 CTP7 CTP7 x18 Layer 1 cards transmit 48 G MP7 Redundant Node 72 input links per Layer-2 node MP7 MP7 MP7 MP7 MP7 Nodes 3 to 9 De-multiplexing node: Separate card or firmware core in downstream system MP7 Node 2 MP7: Demux Layer-2 Cards 6 output links per MP Gb/s MP7 Node 1 x9 Flexible system: Simple to upgrade from 16 bit towers to 24 bit towers or provide extra logic resources. ugt

31 Details on Hadron Plus Strips (HPS) Reconstruction seeded by jets with pt>14 GeV built using anti-kt algorithm Clustering of the photons and electrons inside the jet into strips: v Minimum pt of e/γ that enter strips:.5 GeV v Construction of strips is an iterative process v Minimum pt of strips: 2.5 GeV v Size of strip dynamically enlarged in particular to account for magnetic drift of e+e- pairs τh = combination of strips and charged hadrons (pt) inside jet v Build τh kinematics from visible decay products v Test hypothesis: 1-prong, 1-prong + pi, 1-prong + 2pi, 3-prong, etc. v Check if pass visible mass criteria (specific to each decay mode) that depend on the τh pt, and if it passes charge requirements. Events CMS Preliminary, 19.7 fb at 8 TeV µτ observed h π± bkg. uncertainty Z ττ 3 π ± 1 π± + photons 1 π± no photons Z µ µ electroweak 8 tt ρ± QCD a m(τh) [GeV] 31