Tau Reconstruction at CMS Evan Friis
Outline Taus at hadron colliders Tau ID algorithms at CMS Tau ID measurements in 2010 Z ττ standard candle SVfit: τ pair mass reconstruction MSSM H ττ
Tau ID at hadron colliders proton - (anti)proton cross sections 10 9 10 9 10 8 10 8 tot Exciting new physics signals could appear in tau channels QCD Jets can fake taus QCD background is very large! Dominant background in many analyses is fake taus http://projects.hepforge.org/mstwpdf/plots/plots.html (nb) 10 7 10 6 10 5 10 4 10 3 10 2 10 1 10 0 10-1 10-2 10-3 10-4 10-5 10-6 10-7 jet (E T jet > s/20) jet (E T jet > 100 GeV) WJS2009 b W Z t jet (E T jet > s/4) Higgs (M H =120 GeV) 200 GeV 500 GeV Tevatron LHC 0.1 1 10 s (TeV) 10 7 10 6 10 5 10 4 10 3 10 2 10 1 10 0 10-1 10-2 10-3 10-4 10-5 10-6 10-7 events / sec for L = 10 33 cm -2 s -1
Particle Flow Algorithm taus in CMS are built using Particle Flow objects Clusters and links signals from all subdetectors Produces a list of particle candidates h h 0 e μ γ To the user looks just like Monte Carlo,*-*9-#$ 5/64 /"+7-*$7 0/64 /"+7-*$7 ;&$-%9"*<="#> 8$&9:7 F. Beaudette
Particle Flow Performance Jet Transverse Energy Resolution # 450 400 350 300 Particle Flow Particle Flow Jets CMS Preliminary SIM Particle Flow resolves overlaps in subdectors 250 200 150 100 50 50 GeV Et tau Calo Tower Jets Improves resolution 0-60 -50-40 -30-20 -10 0 10 20 30 40 Et Tau ET resolution
Traditional CMS Tau ID geometrically defined isolation define geometric region around tau candidate and require low detector activity relies on the fact that taus are more collimated than QCD CMS Physics Technical Design Report (PTDR) results use these algorithms see CMS PAS PFT-08-001
Decay Mode CMS Tau ID Particle Flow algorithm allows examination of meson content two new algorithms: Hadrons Plus Strips (HPS) algorithm Tau Neural Classifier (TaNC) algorithm GOAL: optimize tau identification for individual tau decay modes
Hadrons Plus Strips Algorithm build signal components combinatorially cluster gammas into π 0 candidates using η-φ strips build all possible taus that have a ʻtau-likeʼ multiplicity from the seed jet φ π 0 γ γ γ π 0 π + π + π 0 0.20 π + π + π - γ 0.05 η tau that is ʻmost isolatedʼ with compatible mvis is the final tau candidate associated to the seed jet
Tau Neural Classifier a neural network for each decay mode cluster gammas into π 0 candidates by combinatoric pairs compatible with mπ0 φ signal objects are defined using shrinking cone depending on decay mode π 0 π + π 0 γ γ γ π + π 0 π + π 0 π 0 π + π + π - γ π + π + π - π 0 a different neural network η is applied!
Tau Performance measuring the fake rate is easy (no lack of background at the LHC) fake rate is different for W+jets, multi-jet, heavy flavor independently measuring signal efficiency is hard in practice efficiency extracted in situ in Z and Higgs analyses
Fake Rate measurement TAU-11-001 open points = simulation red is W+jets closed points = data purple is heavy flavor CMS Preliminary 2010, -1 s=7 TeV, 36 pb CMS Preliminary 2010, -1 s=7 TeV, 36 pb fake rate from jets 10-2 W#µ" Data TANC TaNC loose! fake rate from jets 10-2 HPS HPS loose W#µ" Data W#µ" Simulation QCDj Data QCDj Simulation QCDµ Data QCDµ Simulation! W#µ" Simulation QCDj Data QCDj Simulation QCDµ Data -3 10 QCDµ Simulation -3 10 Data-Sim. Simulation 0.2 0-0.2 0 50 100 150 200 jet p T (GeV/c) Data-Sim. Simulation 0.2 0-0.2 0 50 100 150 200 jet p T (GeV/c)
sured in data are represented by solid symbols and compared to MC prediction y open symbols. TAU-11-001 Tau ID Performance CMS Preliminary 2010, -1 s=7 TeV, 36 pb fake rate from jets 10-2 PTDR, W+jet PTDR, QCDµ TaNC, W+jet TaNC, QCDµ HPS, W+jet HPS, QCDµ purple info box 3X improvement since PTDR measured! -3 10 0.1 0.2 0.3 0.4 0.5 0.6 expected! efficiency measured fake rate as a function of MC estimated efficiency for all working
Tag & Probe efficiency TAU-11-001 Preselect Z τ μτhad using loose isolation + Z analysis cuts Fit ττ M vis passing/failing spectra composition Method robust against Higgs contamination in Z peak CMS Preliminary 2010, -1 s=7 TeV, 36 pb CMS Preliminary 2010, -1 s=7 TeV, 36 pb number of events 70 60 50 40 30 20 Data * Z/# "! + QCD W + jets * Z/# " µ + tt + jets HPS loose -! - µ PASSED number of events 120 100 80 60 40 Data * Z/# "! + QCD W + jets * Z/# " µ + tt + jets HPS loose -! - µ FAILED 10 20 0 20 40 60 80 100 120 140 160 180 200 µ! jet visible mass (GeV/c 2 ) 0 20 40 60 80 100 120 140 160 180 200 µ! jet visible mass (GeV/c 2 ) Figure 2: The visible invariant mass of the µτ jet system for preselected events which passed (left) and failed (right) the HPS loose tau identification requirements compared to predictions of the MC simulation. result: DATA/MC 1.0 ± 25%
Zττ cross section EWK-10-013! Analysis goal: understand taus at CMS Four channels: μ-had, e-had, e-μ, μ-μ ~25% τ-id uncertainty limits utility as test of Standard model Instead: believe in lepton universality and use Z peak to measure τ-id Combined fit of all channels
Z ττ cross section final Zττ events selected EWK-10-013 Events / (10 GeV) 150 100 50-1 36 pb CMS preliminary at s = 7 TeV Z "!! " µ! had data Z "!! EWK+tt QCD Events / (10 GeV) 150 100 50-1 36 pb CMS preliminary at s = 7 TeV Z "!! " e! had data Z "!! EWK+tt QCD Events / (10 GeV) yields from fit 0 0 50 100 150 200 50 40 30 20 10 Visible Mass [GeV] -1 36 pb CMS preliminary s = 7 TeV data Z "!! EWK+tt QCD yields from fit 0 0 50 100 150 200 at Z "!! " eµ e-µ Invariant Mass [GeV] Events / (10 GeV) 3 10 2 10 yields from fit 0 0 50 100 150 200 10 1 Visible Mass [GeV] -1 36 pb CMS preliminary s = 7 TeV data Z "!! Z/#* " µµ QCD yields from fit 0 50 100 150 200 at Z "!! " µµ µ-µ Invariant Mass [GeV] Figure 2: Visible mass distributions of the τ µ τ had (top left), τ e τ had (top right), τ e τ µ (bottom left), and τ µ τ µ (bottom right) final states. non-tau backgrounds measured from data
Combined Zττ Fit simultaneously fit all channels for σz AND ετid e-μ and μ-μ channels drive σzττ central value EWK-10-013 Combined: σ (pp ZX) B Z τ + τ = 1.00 ± 0.05 (stat.) ± 0.08 (syst.) ± 0.04 (lumi.) nb, fitted ετid MC/DATA = 0.93±0.09 assume σzττ = σzee/μμ ετid MC/DATA = 0.96±0.07 Taus are well understood objects at CMS!
searches with taus
Tau pair mass reconstruction Searches for new resonances to taus Tau pair invariant mass is the most natural observable Tau decays have invisible component Recovery methods: Collinear approximation SVfit algorithm Missing Mass algorithm (see Alexeiʼs talk)
The Collinear Approximation Assume neutrinos collinear with visible products Approximates true ττ mass MET x = P ν 1 sin θ 1 cos φ 1 + P ν 2 sin θ 2 cos φ 2 τ vis 1 MET y = P ν 1 sin θ 1 sin φ 1 + P ν 2 sin θ 2 sin φ 2 Very sensitive to MET MET angular resolution! τ vis 2 ~45% of events thrown away due to unphysical negative energies
SVfit mass reconstruction parameterize the physics of tau decays maximize a likelihood function w.r.t. the tau decay parameters likelihood terms: decay phase-space reconstructed neutrinos and measured ME T p T-balance regularization compatibility with tracker information (optional)
Decay Parameterization a τ decay can be completely described by: θ φ minvis R Rest frame angle between boost and visible decay productions Azimuthal angle of visible decay products about boost direction Mass of invisible decay product system (minvis = 0 for τhad) Flight distance in lab frame
Decay Parameterization reconstructing the tau energy rest frame visible energy: E vis = m2 τ +m2 vis m2 νν 2m τ Lorentz invariant component of visible momentum sin θ LAB = p vis sin θ p LAB vis Solve Lorentz transformation along boost for γ p LAB vis cos θ LAB = γβe vis + γp vis cos θ E LAB τ = γm τ
Decay Parameterization reconstructing the tau direction sin θ LAB = p vis sin θ p LAB vis constrains τ direction to lie on cone with angle θ LAB about visible momentum uvis azimuthal fit parameter φ φ determines direction R flight path R determines θ L location of decay vertex (optional) PV
Phase Space Likelihood Two or three-body decay Hadronic decays trivial Leptonic decays depend on m νν dγ M 2 ((m2 τ (m νν + m vis ) 2 )(m 2 τ (m νν m vis ) 2 )) 1/2 2m τ m νν dm νν sin θdθ present implementation of the SVfit algorithm, the matrix element is assume Decays are assumed to be isotropic: M 2 = 1 Investigating extending term to account for polarization correlations between taus Possible handle to separate Higgses from Zs!
likelihood MET Compare SV-fitted neutrinos to measured missing transverse energy Resolution of ME T parameterized by ΣET of the event Resolution measured in Z μμ events Zμμ events
PT - balance likelihood to reduce background one must apply kinematic thresholds on visible decay products for Z/H μ-had channel: pt μ > 15 GeV, pt had > 20 GeV significantly biases the kinematic distributions pt balance term: probability for decay pt from mass M 0.1 τ μνν before P T cut after P T cut 0.12 0.1 τ τhν before P T cut after P T cut 0.08 H(130) 0.08 H(130) 0.06 0.06 0.04 0.04 0.02 0.02 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Visible energy fraction 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Visible energy fraction
PT - balance likelihood term which represents the probability P(pT Mττ) first assume isotropic decay of resonance M at rest each decay product has P(pT) P iso (p T ) p T 1 (2pT /M ) 2 smear with Gaussian and add Gamma to account for resolution and non-zero boost P total (p T ) P iso (p T ) exp( p2 T 2s 2 )+aγ(p T,k,θ) smear, Gamma fraction and shape are fitted as first order functions of Mττ
Tracking constraint Taus have non-negligible lifetime: cτ = 87µm uvis Use tracker information Constrain fitted SV SV φ to lie on track with error R Constrain flight time R Not currently used. θ L Working to understand alignment issues. track
SVfit Performance SV fit has better resolution than collinear approximation (left) SV fit has twice the acceptance as the collinear solution (right) a.u. 0.14 CMS Preliminary SV fit Z "!! Simulation s = 7 TeV Collinear approx. 0.12 Events 300 250 CMS Preliminary $(120)#!! s = 7 TeV tan"=30 SV fit Collinear approx. 0.1 200 0.08 150 0.06 0.04 100 0.02 50 0 0 50 100 150 200 250 2 [GeV/c ] M!! normalized to unity 0 0 50 100 150 200 250 2 [GeV/c ] M!! normalized to luminosity
SVfit Performance SVfit peaks at di-tau mass better relative resolution that visible mass a.u. 0.22 0.2 CMS Preliminary %(160)$## s = 7 TeV tan"=30 SV fit Visible mass Visible mass! 1.88 0.18 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0 50 100 150 200 250 300 2 mass [GeV/c ]
Extensions to SVfit currently under development Implement tracking likelihood Improve p T-balance parameterization Use polarization information Integration instead of maximization (similar to DLM, MMC) Extend code to fit N leptons Application to W τν, H ++, etc
MSSM H ττ Search Taus one of the best handles for MSSM μ + τ had e + τ had μ + e Benefit from tanβ enhancement σ and BR Search performed with three channels: Tau ID correction factor floats in final fit, constrained by Z peak & 25% tag-probe All backgrounds measured from data
MSSM H ττ Search HIG-10-002 results final Mττ distribution of all channels no bumps!"#$%&'()*+,#-'! ).!"#$%&'()*+,#-'! ). 1** 0** )** /** * ) /* /* purple info box =CD+E6#F575$26G 0H+49 I/ +++J+K#- 89"#: 7 A +B+)**+,#-'! ) ; ;!!,!@ <=>?+%% / * /** )** 0** 1** %23+4256+72&&+(,#-'! ).
MSSM Expected Limits HIG-10-002 Visible mass Expected Bayesian 95% CL limits on σh BR(H ττ) SVfit BR(h"!!) Expected limit on $ # 3 10 2 10 10 CMS Preliminary, s -1 = 7 TeV, L = 35 pb!! " µ! had!! " e! had!! " e µ Combined 1 100 200 300 400 500 m(a) [GeV/c 2 ] BR(h"!!) Expected limit on $ # 3 10 2 10 10 CMS Preliminary, s -1 = 7 TeV, L = 35 pb!! " µ! had!! " e! had!! " e µ Combined 1 100 200 300 400 500 m(a) [GeV/c 2 ] SVfit improves limit by 10%
MSSM Observed Limit HIG-10-002 observed limit on σh BR(H ττ) 900 >=?9 =++90"= 344 34 3 344,AB+C%$DE7E2F%G 6H+0" I3 +++J+K$; ()*+,-.-+/00$%+"1/2'#!"#$%&$' A$'EF2+$L0$MN$' 3 +$L0$MN$'+%F2O$ 5 +$L0$MN$'+%F2O$ 544 644 @44 )44 7 8 +++9:$;<! 5 = re 2: The expected one- and two-standard-deviation ranges (blue bands) an
ure 3: Region in the parameter space of tan versus m excluded at 95% C.L. Interpretation in tanβ HIG-10-002 /7!"#$%&'()*)+,&-$$$./$01 23 $$$4$5'6 :,+ ;7 97.7 87 37 7 *,H "##"$* < $$$$=>'+,&)?@$" #A#B $C$3$5'6 D;E$!FGF$'H>(IJ'J$&'K)?+= 377 3;7 877 8;7.77 * L $$$MN'6O! 8 P $!"#$'H>(IJ'J $Q3 $:<'?&- $!"#$'H0'>:'J GS%$'H>(IJ'J 5'R,:&?+$'H>(IJ'J
Summary Large improvements in τ-id since PTDR Tau fake rates & efficiency measured in real data Z ττ cross section measured with high precision Taus well understood at CMS SVfit method improves M ττ resolution CMS sets new limits on MSSM using taus
backup
Particle Flow Algorithm CMS Preliminary HCAL Entrance ECAL Entrance ECAL Entrance HCAL Entrance cluster linked to track unlinked cluster tracker hit see CMS PAS PFT-10-002
Particle Flow Performance Jet Transverse Energy Resolution # 450 CMS Preliminary invariant mass of PF photon pairs 400 350 300 250 200 150 Particle Flow Particle Flow Jets 50 GeV Et tau SIM Number of photon pairs 20000 15000 10000 fit 2 m! 0 = 135.2 ± 0.1 MeV/ c 2 " m! = 13.2 ± 0.1 MeV/ c 0-1 7-TeV Data, 0.1 nb CMS Preliminary 2010 100 50 Calo Tower Jets 0-60 -50-40 -30-20 -10 0 10 20 30 40 Et 5000 7 TeV Data, 0.1 nb -1 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 2 Mass (GeV/c ) Tau ET resolution Mass of photon pair [GeV/c 2 ] Figure 1: Photon-pair invariant-mass distribution the simulation (b).
Shrinking Cone Algorithm reduce QCD by applying isolation requirement require a leading candidate with pt > 5 within ΔR < 0.1 of jet axis signal objects are those with ΔR < ΔRsig of the lead candidate R sig =5.0/E jet T isolation objects are those in the region ΔRsig < ΔR < 0.5 about the lead candidate R = φ 2 + η 2
Zττ selections Trigger Events triggered by single Electron/Muon Triggers P T thresholds 9-15 GeV, depending on instantaneous Luminosity Lepton Selection Electrons Muons had.! Decays P T > 15 GeV P T > 15 GeV P T > 20 GeV! < 2.4! < 2.1! < 2.4 isolated isolated loose Tau id. Veto against e/µ Opposite Charge Lepton Pair Event Selection For µ +! had, e +! had and e + µ Channels, µ + µ Channel different (Backup) Transverse Mass e +! had, µ +! had : M T (l + MET) < 40 GeV e + µ: M T (e + MET ) < 50 GeV && M T (µ + MET) < 50 GeV Veto Events with additional isolated Leptons Christian Veelken Higgs and Z!! +! - in CMS 5
Zττ yields Event Yields CMS Data, 36 pb -1 @ 7 TeV Background Estimates quoted in Table obtained from Data-driven Methods > 600 Z!! +! - Signal Events selected in CMS Data Christian Veelken Higgs and Z!! +! - in CMS 7
Z recoil correction MET resolution better in Monte Carlo than data before correction after correction Zμμ events Zμμ events Project MET in directions parallel and perpendicular to Z and parametrize as function of Z pt
sub title Observed vs. expected Limits by Channel visible Mass full! +! - Mass 34
Higgs Search control plots Lepton P T and!: µ +! had Channel Christian Veelken Higgs and Z!! +! - in CMS 26
Higgs Search control plots Lepton P T and!: e +! had Channel Christian Veelken Higgs and Z!! +! - in CMS 27
Z!! +! - Systematic Uncertainties Largest Uncertainty: hadronic Tau Identification Efficiency!!" had Identification Efficiency constrained by Ratio of Event Yields in semi-leptonic/leptonic Channels!! Determine Z " " + " - Cross-section by simultaneous Fit of all four Channels Christian Veelken Higgs and Z " " + " - in CMS 25
Christian Veelken Higgs and Z!! +! - in CMS 28
Christian Veelken Higgs and Z!! +! - in CMS 29
Christian Veelken Higgs and Z!! +! - in CMS 30