Tau Reconstruction at CMS - University of California,...

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

• 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

0.1 1 1010-7

10-6

10-5

10-4

10-3

10-2

10-1

100

101

102

103

104

105

106

107

108

109

10-7

10-6

10-5

10-4

10-3

10-2

10-1

100

101

102

103

104

105

106

107

108

109

WJS2009

jet(ETjet > 100 GeV)

jet(ETjet > s/20)

jet(ETjet > s/4)

Higgs(MH=120 GeV)

200 GeV

LHCTevatron

eve

nts

/ sec

for L

= 1

033 c

m-2s-1

b

tot

proton - (anti)proton cross sections

W

Z

t

500 GeV

(nb

)

s (TeV)http://projects.hepforge.org/mstwpdf/plots/plots.html

http://projects.hepforge.org/mstwpdf/plots/plots.html

http://projects.hepforge.org/mstwpdf/plots/plots.html

Particle Flow Algorithm

• Clusters and links signals from all subdetectors

• Produces a list of particle candidates

• To the user looks just like Monte Carlo

taus in CMS are built using Particle Flow objects

h h0 e μ γ

2/21!"#$%&'()*&+,*--*(.(/0123441(

!"#$"%#&'()*#(+,$)%-.#(/.0&(,.10$%)*2!"#$"%#&'()*#(+,$)%-.#(/.0&(,.10$%)*2

5/64

/"+7-*$7

0/64

/"+7-*$7

8$&9:7

,*-*9-#$

;&$-%9"*<="#>

8?*("%7-(#=(%',%@%,+&"(;&$-%9"*7(%7(-?*'(+7*,(-#(A+%",(B*-7C(-#(

,*-*$D%'*(-?*(D%77%'E(-$&'7@*$7*(*'*$EFC(-#($*9#'7-$+9-(&',(

%,*'-%=F(-&+7(=$#D(-?*%$(,*9&F(;$#,+9-7C(-#(-&E(A(B*-7(G(

F. Beaudette

Particle Flow Performance

Et-60 -50 -40 -30 -20 -10 0 10 20 30 40

#

0

50

100

150

200

250

300

350

400

450

Particle Flow Jets

Calo Tower Jets

Jet Transverse Energy Resolution

50 GeV Et tau

CMS Preliminary

Particle Flow

Tau ET resolution

SIM Particle Flow resolvesoverlaps in subdectors

Improves resolution

Traditional CMS Tau ID

geometrically defined isolationdefine geometric region around tau

candidate and require low detector activity

see CMS PAS PFT-08-001

relies on the fact that taus are more collimated than QCD

CMS Physics Technical Design Report (PTDR) results use these algorithms

!"#$%"&"'()*"+,-$./0$1/0$23$%"425$6(7"8$

!"

#$%%"&'(")*+,"-./"

012.3&."4"!"!"(",+"5&"

678$94:19;"8$3<"19=<".>5"$7<?=")+@".$4:A"

"!".BC.81&.9=$D"E1;9$=?8."<2":$F8<914"'G>'H"I.4$J"("KG>KHL"MG>MHL"$)"#.%<9"

"!"N1&"=<"8.4<9%=8?4="MO"""KO"KPL"$)"""KO"KPL"$)"""KG"KH"KGL"$)"""KH"KG"KH"19"'GQ.="

I.4$J%"C8.F<&19$9=DJ"19=<":$F8<9%R"

Decay Mode CMS Tau IDParticle Flow algorithm allowsexamination 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 Algorithmbuild signal components combinatorially

cluster gammas into π0 candidates using η-φ strips

γ γ γ

γ

φ

η

π0

π0

π+

π+ π0

π+ π+ π-

build all possible tausthat have a ʻtau-likeʼ multiplicity

from the seed jet

tau that is ʻmost isolatedʼ with compatible mvis

is the final tau candidate associated to the seed jet0.05

0.20

Tau Neural Classifiera neural network for each decay mode

cluster gammas into π0 candidates by combinatoric pairs compatible with mπ0

γ γ γ

γ

φ

η

π0

π0π+

π+ π0

π+ π0 π0

π+ π+ π-

π+ π+ π- π0

signal objects are defined using shrinking cone

depending on decay mode

a different neural networkis applied!

Tau Performancemeasuring 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 measurement8 6 Fake–rate measurement

0 20 40 60 80 100 120 140 160180 200

fake

rate

from

jets

!

-310

-210 Data"µ#W

Simulation"µ#W

QCDj Data

QCDj Simulation

DataµQCD

SimulationµQCD

TaNC loose

-1=7 TeV, 36 pbsCMS Preliminary 2010,

(GeV/c)T

jet p0 50 100 150 200

Sim

ulat

ion

Dat

a-Si

m.

-0.20

0.2 0 20 40 60 80 100 120 140 160180 200

fake

rate

from

jets

!-310

-210

Data"µ#W

Simulation"µ#W

QCDj Data

QCDj Simulation

DataµQCD

SimulationµQCD

HPS loose


(GeV/c)T

jet p0 50 100 150 200

Sim

ulat

ion

Dat

a-Si

m.

-0.20

0.2

Figure 5: Probabilities of quark and gluon jets to pass “loose” working points of the TaNC (left)

and HPS (right) algorithms as a function of jet pT for QCD, QCD µ-enriched and W type events.

Fake-rate measured in data are represented by solid symbols and compared to MC prediction

represented by open symbols.

efficiency!expected 0.1 0.2 0.3 0.4 0.5 0.6

fake

rate

from

jets

!m

easu

red

-310

-210


fake

rate

from

jets

!m

easu

red

-310

-210

PTDR, W+jetµPTDR, QCD

TaNC, W+jetµTaNC, QCD

HPS, W+jetµHPS, QCD


Figure 6: The measured fake–rate as a function of MC estimated efficiency for all working

points for QCD µ-enriched and W data samples. The PTDR points represent results of the

fixed cone algorithm based on the PF taus.

open points = simulation closed points = data

TANC HPS

red is W+jets purple is heavy flavor

TAU-11-001

Tau ID Performance

purple info box

8 6 Fake–rate measurement

0 20 40 60 80 100 120 140 160180 200

fake

rate

from

jets

!

-310

-210 Data"µ#W

Simulation"µ#W

QCDj Data

QCDj Simulation

DataµQCD

SimulationµQCD

TaNC loose


(GeV/c)T

jet p0 50 100 150 200

Sim

ulat

ion

Dat

a-Si

m.

-0.20

0.2 0 20 40 60 80 100 120 140 160180 200

fake

rate

from

jets

!

-310

-210

Data"µ#W

Simulation"µ#W

QCDj Data

QCDj Simulation

DataµQCD

SimulationµQCD

HPS loose


(GeV/c)T

jet p0 50 100 150 200

Sim

ulat

ion

Dat

a-Si

m.

-0.20

0.2

Figure 5: Probabilities of quark and gluon jets to pass “loose” working points of the TaNC (left)

and HPS (right) algorithms as a function of jet pT for QCD, QCD µ-enriched and W type events.

Fake-rate measured in data are represented by solid symbols and compared to MC prediction

represented by open symbols.


fake

rate

from

jets

!m

easu

red

-310

-210


fake

rate

from

jets

!m

easu

red

-310

-210

PTDR, W+jetµPTDR, QCD

TaNC, W+jetµTaNC, QCD

HPS, W+jetµHPS, QCD


Figure 6: The measured fake–rate as a function of MC estimated efficiency for all working

points for QCD µ-enriched and W data samples. The PTDR points represent results of the

fixed cone algorithm based on the PF taus.

3X improvement since PTDR

TAU-11-001

Tag & Probe efficiency• Preselect Z→τμτhad using loose isolation + Z analysis cuts

• Fit ττ Mvis passing/failing spectra composition

• Method robust against Higgs contamination in Z peak

4 4 Efficiency of tau reconstruction and identification.

data driven background estimates instead of the MC templates and by varying invariant mass

ranges for the fit. All checks demonstrated consistent results within the uncertainties of the

method.

)2 visible mass (GeV/cjet!µ

20 40 60 80 100 120 140 160 180 200

num

ber o

f eve

nts

0

10

20

30

40

50

60

70Data

-! +! " *#Z/QCDW + jets

-µ +µ " *#Z/ + jetstt


)2 visible mass (GeV/cjet!µ

20 40 60 80 100 120 140 160 180 200

num

ber o

f eve

nts

0

20

40

60

80

100

120Data

-! +! " *#Z/QCDW + jets

-µ +µ " *#Z/ + jetstt


Figure 2: The visible invariant mass of the µτjet system for preselected events which passed

(left) and failed (right) the HPS “loose” tau identification requirements compared to predictions

of the MC simulation.

Algorithm Fit data Expected MC DATA/MC

TaNC “loose” 0.76 ± 0.20 0.72 1.06 ± 0.30

TaNC “medium” 0.63 ± 0.17 0.66 0.96 ± 0.27

TaNC “tight” 0.55 ± 0.15 0.55 1.00 ± 0.28

HPS “loose” 0.70 ± 0.15 0.70 1.00 ± 0.24

HPS “medium” 0.53 ± 0.13 0.53 1.01 ± 0.26

HPS “tight” 0.33 ± 0.08 0.36 0.93 ± 0.25

HPS “loose” combined fit [4] 0.94 ± 0.09

HPS “loose” ττ to µµ, ee fit [4] 0.96 ± 0.07

Table 1: Efficiency for hadronic tau decays to pass TaNC and HPS tau identification criteria

measured by fitting the Z → τ+τ−signal contribution in the samples of the “passed” and

“failed” preselected events. The errors of the fit represent statistical uncertainties. The last

column represents the data to MC correction factors and their full uncertainties including sta-

tistical and systematic components. Data to MC ratios for the tau reconstruction efficiency

measured using fits to the measured Z production cross sections as described in [4] are also

shown.

Results of the fits are summarized in Table 1. The values measured in data, “Fit data” are com-

pared with the expected values, “Expected MC”, obtained by repeating the fitting procedure

on simulated events. The performance of the tau algorithms on preselected events is expected

to be approximately 30% higher than for an inclusive sample of taus, without pre-selection.

In general the absolute value of the efficiency depends on pT and η requirements, which are

applied in each individual physics analysis, therefore the main goal of this measurement is to

perform the data to MC comparison and to determine data to MC correction factors and their

uncertainties. With the current data sample the statistical uncertainties of the fits are in the

result: DATA/MC ≅ 1.0 ± 25%

HPS loose HPS loose

PASSED FAILED

TAU-11-001

Zττ cross section

• 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 universalityand use Z peak to measure τ-ID

• Combined fit of all channels

•! N = Nobs – Nbgr

Background Contribution Nbgr estimated from Data

(using 1-3 complementary Methods, depending on Channel)

•! Acceptance taken from Monte Carlo

(POWHEG + TAUOLA, PYTHIA with CMS Z2 tune for Hadronization)

•! Efficiency factorized into independent Terms

Each Term either measured directly in Data or taken from Monte Carlo

and applying Data/MC Correction factor measured from Data

•! Branching Ratios for !+!- to decay into µ + !had, e + !had, e + µ, µ + µ

taken from PDG

•! Luminosity measured with Precision of 4%

Christian Veelken Higgs and Z ! !+!- in CMS 6

Cross-section Extraction

!

EWK-10-013

Z→ττ cross section6 7 Cross section measurement

Visible Mass [GeV]0 50 100 150 200

Even

ts /

(10

GeV

)

0

50

100

150

data!! " Z

t EWK+t QCD

yields from fit

CMS preliminary

= 7 TeVs at -136 pb

had!µ " !! "Z

Visible Mass [GeV]0 50 100 150 200

Even

ts /

(10

GeV

)

0

50

100

150

data!! " Z

t EWK+t QCD

yields from fit

CMS preliminary

= 7 TeVs at -136 pb

had! e" !! "Z

Invariant Mass [GeV]µe-0 50 100 150 200

Even

ts /

(10

GeV

)

0

10

20

30

40

50

data!! " Z

t EWK+t QCD

yields from fit

CMS preliminary

= 7 TeVs at -136 pbµ e" !! "Z

Invariant Mass [GeV]µ-µ0 50 100 150 200

Even

ts /

(10

GeV

)

1

10

210

310

data!! " Z

µµ "* # Z/ QCD

yields from fit

CMS preliminary

= 7 TeVs at -136 pbµµ " !! "Z

Figure 2: Visible mass distributions of the τµτhad (top left), τeτhad (top right), τeτµ (bottom left),and τµτµ (bottom right) final states.

τµτhad τeτhad τeτµ τµτµ

Acceptance: A 0.13 0.12 0.074 0.16Selection efficiency: � 0.40 0.25 0.55 0.17Mass window correction: fc 0.94 0.94 0.98 0.99

Table 3: Acceptance obtained with POWHEG for 60 < Mττ < 120 GeV and the selectionefficiency. The fraction of selected events outside the generator level mass window is alsoshown.

The measured values of the cross section for the four channels considered are:

σ (pp→ ZX)× B (Z→ ττ)µτ = 0.81± 0.07 (stat.)± 0.04 (syst.)± 0.09 (lumi.)± 0.19 (τ-ID) nb

σ (pp→ ZX)× B (Z→ ττ)eτ = 0.91± 0.11 (stat.)± 0.04 (syst.)± 0.10 (lumi.)± 0.21 (τ-ID) nbσ (pp→ ZX)× B (Z→ ττ)eµ = 0.99± 0.12 (stat.)± 0.06 (syst.)± 0.11 (lumi.) nb

σ (pp→ ZX)× B (Z→ ττ)µµ = 1.14± 0.25 (stat.)± 0.10 (syst.)± 0.12 (lumi.) nb.

The values are compatible with each other and with the next-to-next-to-leading-order (NNLO)175

final Zττ events selected

non-tau backgrounds measured from data

EWK-10-013

Combined Zττ Fitsimultaneously fit all channels for σz AND ετID

e-μ and μ-μ channels drive σZττ central value

Christian Veelken Higgs and Z ! !+!- in CMS

Measured Cross-Section in good Agreement with Theory Prediction (NNLO)

and CMS Measurement of Z/!* ! "+"-, l = e/µ Cross-section

Data/MC Correction factor for Tau Identification Efficiency compatible with 1.0

10

Simultaneous Fit Results

Combined:

" " BR(Z/!* ! !+!-) = 0.99 ± 0.06 (stat.) ± 0.08 (sys.) ± 0.04 (lumi.) nb

8 7 Cross Section Measurement

Table 4: The measured values of the cross section from the four final states considered. The

statistical, systematic and luminosity uncertainties are given. The uncertainty associated to the

τhad reconstruction and identification efficiency, τhad − ID, is shown separately.

Final state σ (pp → ZX)× B (Z → τ+τ−) nb stat. syst. lumi. τ ID

τµτhad 0.83 0.07 0.04 0.03 0.19

τeτhad 0.94 0.11 0.03 0.04 0.22

τeτµ 0.99 0.12 0.06 0.04

τµτµ 1.14 0.27 0.04 0.05

(left), where the likelihood contours for the best estimates of the cross section and the τhad effi-

ciency scale factor are shown. In addition to the one-standard-deviation contour, the contours

for which the likelihood L is reduced by 2∆ ln L = 2.30 and 6.18 compared to the maximum

vaue are also shown, corresponding to a coverage of 68% and 95% in the 2-parameter plane,

respectively. The value of the cross section extracted from the fit is

σ (pp → ZX)× B�Z → τ+τ−� = 1.00 ± 0.05 (stat.)± 0.08 (syst.)± 0.04 (lumi.) nb,

which is compared to the individual final state measurements in Fig. 3 (right). The value for218

the cross section is dominated by the dilepton final states which have smaller systematic un-219

certainties. In the simultaneous fit, the τhad-ID scale factor is measured to be 0.93 ± 0.09.220

A more precise value of the hadronic tau reconstruction efficiency can be obtained by perform-221

ing a fit of the τµτhad and τeτhad final states, where the cross section is fixed to the value mea-222

sured by CMS in the electron and muon decay channels, σ (pp → ZX)× B (Z → e+e−, µ+µ−)223

= 0.931 ± 0.026 (stat.)± 0.023 (syst.)± 0.102 (lumi.) nb [6]. The extracted value of the τhad-ID224

scale factor is 0.96± 0.07, which corresponds to a tau identification efficiency of (47.4± 3.3)% in225

data, for hadronically decaying tau leptons in the Z → τ+τ− sample with visible pT > 20 GeV226

within the detector acceptance.227

-ID Efficiency (data/sim)had!0.6 0.8 1 1.2 1.4

) [n

b]!!

" B

(Z

# Z

X)

"(p

p $

0.7

0.8

0.9

1

1.1

1.2

1.3

68% CL

95% CL

central value

$ 1 ±

CMS

= 7 TeVs at -136 pb

) [nb]!! " B(Z # ZX) "(pp $0.5 1 1.5 2 2.5

NNLO, FEWZ+MSTW08 [PDF4LHC 68% CL] (60-120 GeV)

µµ ee, "Z µ! + e!

µ! + µ!

had! + µ!

had! + e!

(combined) !! "Z

luminosity uncertainty not shown

CMS = 7 TeVs at -136 pb

Figure 3: Likelihood contours for the joint parameter estimation of the cross section and the

τ-identification (left). The fitted central values (solid line) and their estimated 1σ uncertainties

(dashed lines) are also shown. Summary of the measured Z → τ+τ− cross sections in the

τµτhad, τeτhad, τeτµ, and τµτµ final states, in the invariant mass range of 60 < Mτ+τ− < 120 GeV

(right). The inner error bar represents the statistical uncertainty. The extracted cross section

from the combined fit and the NNLO theoretical prediction are also shown.

assume σZττ = σZee/μμ →ετID MC/DATA = 0.96±0.07Taus are well understood objects at CMS!

fitted ετID MC/DATA = 0.93±0.09

EWK-10-013

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

τvis1

τvis2

MET

METx = P ν1 sin θ1 cos φ1 + P ν

2 sin θ2 cos φ2

METy = P ν1 sin θ1 sin φ1 + P ν

2 sin θ2 sin φ2

Very sensitive to MET angular resolution!

~45% of events thrown awaydue to unphysical negative energies

SVfit mass reconstructionparameterize the physics of tau decays

maximize a likelihood function w.r.t. the tau decay parameters

• decay phase-space

• reconstructed neutrinos and measured MET

• “pT-balance” regularization

• compatibility with tracker information (optional)

likelihood terms:

Decay Parameterizationa τ decay can be completely described by:

θ Rest frame angle between boost and visible decay productions

φ Azimuthal angle of visible decay products about boost direction

minvis Mass of invisible decay

product system (minvis = 0 for τhad)

R Flight distance in lab frame

Decay Parameterizationreconstructing the tau energy

Evis = m2τ+m2

vis−m2νν

2mτ

rest frame visible energy:

sin θLAB = pvis sin θpLAB

vis

Lorentz invariant component of visible momentum

Solve Lorentz transformation along boost for γ

pLABvis cos θLAB = γβEvis + γpvis cos θ

ELABτ = γmτ

Decay Parameterizationreconstructing the tau direction

sin θLAB = pvis sin θpLAB

vis

constrains τ direction to lie on cone with angle θLAB about visible momentum

azimuthal fit parameter φ determines direction

flight path R determineslocation of decay vertex

(optional) PV

θL

φ

R

uvis

Phase Space Likelihood• Two or three-body decay

• Hadronic decays trivial

• Leptonic decays depend on mνν

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!

3.2 Likelihood for tau decay 5

laboratory frame of the visible momentum component parallel to the tau direction:

P̄viscos θ̄ = γEvis + βγPvis

cos θ

⇒ γ =Evis[

�Evis�2 +

�P̄vis

cos θ̄�2 −

�Pvis

cos θ�2]1/2 − Pvis

cos θP̄viscos θ̄

(Evis)2 − (Pvis cos θ)2,

Eτ = γmτ

The energy of the tau lepton in the laboratory frame as function of the measured visible mo-155

mentum depends on two of the three parameters only - the rest frame opening angle θ and the156

invariant mass mνν of the neutrino system. The direction of the tau lepton momentum vector157

is not fully determined by θ and mνν, but is constrained to lie on the surface of a cone of open-158

ing angle θ̄ (given by equation 2), the axis of which is given by the visible momentum vector.159

The tau lepton four–vector is fully determined by the addition of the third parameter φ̄, which160

describes the azimuthal angle of the tau lepton with respect to the visible momentum vector.161

3.2 Likelihood for tau decay162

The probability density functions for the tau decay kinematics are taken from the kinematics

review of the PDG [12]. The likelihood is proportional to the phase–space volume for two–

body (τ → τhadν) and three–body (τ → eνν and τ → µνν) decays. For two–body decays the

likelihood depends on the decay angle θ only:

dΓ ∝ |M|2 sin θdθ

For three–body decays, the likelihood depends on the invariant mass of the neutrino system

also:

dΓ ∝ |M|2 ((m2τ − (mνν + mvis)2)(m2

τ − (mνν −mvis)2))1/2

2mτmννdmνν sin θdθ (3)

In the present implementation of the SVfit algorithm, the matrix element is assumed to be163

constant, so that the likelihood depends on the phase–space volume of the decay only1.164

3.2.1 Likelihood for reconstructed missing transverse momentum165

Momentum conservation in the plane perpendicular to the beam axis implies that the vectorial166

sum of the momenta of all neutrinos produced in the decay of the tau lepton pair matches the167

reconstructed missing transverse momentum. Differences are possible due to the experimental168

resolution and finite PT of particles escaping detection in beam direction at high |η|.169

The Emiss

Tresolution is measured in Z → µ+µ− events selected in the 7 TeV data collected by

CMS in 2010. Corrections are applied to Monte Carlo simulated events to match the resolution

measured in data. The momentum vectors of reconstructed Emiss

Tand neutrino momenta given

by the fit parameters are projected in direction parallel and perpendicular to the direction of the

τ+τ− momentum vector. For both components, a Gaussian probability function is assumed.

The width and mean values of the Gaussian in parallel (“�”) and perpendicular (“⊥”) direction

1The full matrix elements for tau decays may be added in the future, including terms for the polarization of the

tau lepton pair, which is different in Higgs and Z decays [19].

MET likelihood• Compare SV-fitted neutrinos to

measured missing transverse energy

• Resolution of MET parameterized by ΣET of the event

• Resolution measured in Z→μμ events

Zμμ events

PT - balance likelihoodto reduce background one must apply

kinematic thresholds on visible decay products

significantly biases the kinematic distributions

3.3 Secondary vertex information 7

]2 [GeV/c!!SV fit M0 50 100 150 200 250 300 3500

0.02

0.04

0.06

0.08

0.1

0.12

0.14 balanceTwith p balanceTwithout p

Figure 1: Distribution of di–tau invariant mass reconstructed by the SVfit algorithm in simu-

lated Higgs events with MA = 130 GeV /c2. The SVfit algorithm is run in two configurations,

with (blue) and without (red) the PT–balance likelihood term included in the fit.

Visible energy fraction0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.02

0.04

0.06

0.08

0.1 cutTbefore P cutTafter P

Visible energy fraction0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.02

0.04

0.06

0.08

0.1

0.12 cutTbefore P

cutTafter P

Figure 2: Normalized distributions of the fraction of total tau decay energy carried by the muon

(left) and hadronic constituents (right) in simulated Higgs events with MA = 130 GeV /c2. The

distribution is shown before (blue) and after (red) the requirement on the PT of the visible decay

products described in section 5.1.

The motivation to add the PT–balance likelihood to the SVfit is to add a “regularization” term172

which compensates for the effect of PT cuts applied on the visible decay products of the two tau173

leptons. In particular for tau lepton pairs produced in decays of resonances of low mass, the174

visible PT cuts significantly affect the distribution of the visible momentum fraction x = EvisEτ

.175

The effect is illustrated in figures 2 and 3. If no attempt would be made to compensate for176

this effect, equations 3.2, 3 would yield likelihood values that are too high at low x, resulting177

in the SVfit to underestimate the energy of visible decay products (overestimate the energy of178

neutrinos) produced in the tau decay, resulting in a significant tail of the reconstructed mass179

distribution in the high mass region. The τ+τ− invariant mass distribution reconstructed with180

and without the PT–balance likelihood term is shown in figure 1. A significant improvement181

in resolution and in particular a significant reduction of the non–Gaussian tail in the region of182

high masses is seen.183

3.3 Secondary vertex information184

The parametrization of the tau decay kinematics described in section 3.1 can be extended to185

describe the production and decay of the tau. As the flight direction of the tau is already fully186

for Z/H μ-had channel: pTμ > 15 GeV, pThad > 20 GeV

τ → μνν τ → τhν H(130)H(130)

pT balance term: probability for decay pT from mass M

PT - balance likelihoodterm which represents the probability P(pT | Mττ)

first assume isotropic decay of resonance M at rest

each decay product has P(pT)Piso(pT ) ∝ pT�

1− (2pT /M)2

smear with Gaussian and add Gamma to account for resolution and non-zero boost

Ptotal(pT ) ∝ Piso(pT )⊗ exp(− p2T

2s2) + aΓ(pT , k, θ)

smear, Gamma fraction and shape are fitted as first order functions of Mττ

Tracking constraint

• Use tracker information

• Constrain fitted SV to lie on track with error

• Constrain flight time R

• Not currently used.Working to understandalignment issues.

Taus have non-negligible lifetime: cτ = 87µm

θL

φ

R

uvis

track

SV

SVfit Performance

]2 [GeV/c!!M0 50 100 150 200 250

Even

ts

0

50

100

150

200

250

300SV fit

Collinear approx.

CMS Preliminary=30" tan!!#(120)$

= 7 TeVs

]2 [GeV/c!!M0 50 100 150 200 250

a.u.

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14 SV fit

Collinear approx.

CMS Preliminary Simulation! ! "Z

= 7 TeVs

normalized to unity normalized to luminosity

SV fit has better resolution than collinear approximation (left)SV fit has twice the acceptance as the collinear solution (right)

SVfit Performance

]2mass [GeV/c0 50 100 150 200 250 300

a.u.

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

0.22 SV fit

Visible mass

1.88!Visible mass

CMS Preliminary=30" tan##$(160)%

= 7 TeVs

SVfit peaks at di-tau massbetter relative resolution that visible mass

Extensions to SVfit

• Implement tracking likelihood

• Improve pT-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

currently under development

MSSM H→ττ Search• Taus one of the best handles for MSSM

• Benefit from tanβ enhancement σ and BR

• Search performed with three channels:

• μ + τhad

• e + τhad

• μ + e

• Tau ID correction factor floats in final fit,constrained by Z peak & 25% tag-probe

• All backgrounds measured from data

purple info box

MSSM H→ττ Search

4 5 Systematic Uncertainties

!"#$%&'()*+,#-'!) .

/

/*

)/*

/**

)**

0**

1**

!"#$%&'()*+,#-'!) .

%23+4256+72&&+(,#-'!).* /** )** 0** 1**

*

89&#6"#:

;

<=>?+%%; !!, !@

7A+B+)**+,#-'!)

=CD+E6#F575$26G0H+49I/+++J+K#-

Figure 1: The reconstructed tau pair invariant mass distribution on linear (above) and log (be-

low) scales, for the sum of the eτh, µτh, and eµ final states, comparing the observed distributions

(data points with error bars) to the sum of the expected backgrounds (shaded histograms). The

contribution from a Higgs boson signal (mA = 200 GeV/c2) is also shown, with normalization

corresponding to the 95% upper bound.

results

final Mττ distributionof all channels

no bumps

HIG-10-002

MSSM Expected Limits46 17 Results

]2m(A) [GeV/c100 200 300 400 500

)!!"

BR

(h#

$Ex

pect

ed li

mit

on

1

10

210

310-1 = 7 TeV, L = 35 pbsCMS Preliminary,

had! µ " !!

had! e " !! µ e " !!

Combined

]2m(A) [GeV/c100 200 300 400 500

)!!"

BR

(h#

$Ex

pect

ed li

mit

on

1

10

210


had! µ " !!

had! e " !! µ e " !!

Combined

Figure 33: The expected Bayesian limits on σ× BR(H → ττ) in pb obtained for τe-τhad, τµ-τhad

and τe-τµ final states separately and for their combination using the visible mass, Mvis in fit (leftplot) and fully reconstructed mass, Mττ in the fit

τe-τµ final states separately and for their combination using the visible mass, Mvis in fit (left690

plot) and fully reconstructed mass, Mττ in the fit.

]2m(A) [GeV/c100 200 300 400 500

)!!"

BR(

h#

$O

bser

ved

limit

on

1

10

210


had! µ " !!

had! e " !! µ e " !!

Combined

]2m(A) [GeV/c100 200 300 400 500

)!!"

BR(

h#

$O

bser

ved

limit

on

1

10

210


had! µ " !!

had! e " !! µ e " !!

Combined

Figure 34: The observed Bayesian limits on σ× BR(H → ττ) in pb obtained for τe-τhad, τµ-τhad

and τe-τµ final states separately and for their combination using the visible mass, Mvis in fit (leftplot) and fully reconstructed mass, Mττ in the fit

691

Figure 35 shows the expected and observed limits on σ × BR(H → ττ) in pb obtained from692

combination of τe-τhad, τµ-τhad and τe-τµ final states using the visible mass, Mvis in fit (left plot)693

and fully reconstructed mass, Mττ in the fit.694

Figure 36 shows the expected and observed limits on σ × BR(H → ττ) in pb obtained from695

combination of τe-τhad, τµ-τhad and τe-τµ final states using the visible mass, Mvis and fully re-696

constructed mass, Mττ in the fit.697

Figure 37 shows the expected limits obtained for τe-τhad, τµ-τhad and τe-τµ final states separately698

and for their combination using the visible mass, Mvis in fit (left plot) and fully reconstructed699

mass, Mττ in the fit.700

Visible mass SVfitExpected Bayesian 95% CL limits

on σH × BR(H→ττ)

SVfit improves limit by 10%

100 200 300 400 5001

10

210

310had! µ " !!

had! e " !! µ e " !!

Combined

HIG-10-002

MSSM Observed Limitobserved limit on σH × BR(H→ττ)

5

signal and the main background, Z → ττ, equally. This effectively allows the observed Z → ττ112

events to provide an in situ calibration of this efficiency, except for Higgs masses near that of113

the Z. Near the Z mass, the tau identification efficiency uncertainty dominates in the eτh and114

µτh channels and the eµ channel provides the greatest sensitivity.115

6 Results116

The mass spectra show no evidence for the presence of a Higgs boson signal, and we set 95%117

C.L. upper bounds on the Higgs boson cross section times the tau pair branching ratio (denoted118

σφ · Bττ) using a Bayesian method assuming a uniform prior in σφ · Bττ. The invariant mass119

spectrum in Fig. 1 shows the contribution of a Higgs boson signal corresponding to mA = 200120

GeV/c2

with σφ · Bττ = 8.71 pb, the value above which we exclude at 95% C.L.121

Figure 2 shows the observed 95% C.L. upper bound on σφ · Bττ as a function of mA, where we122

use as our signal acceptance model the combined spectra from the gg and bb̄ production pro-123

cesses for h, A, and H, assuming tan β = 30 [22]. The plot also shows the one- and two-standard124

deviation range of expected upper limits for various potential experimental outcomes. The125

observed limits are well within the expected range assuming no signal. The observed and126

expected upper limits are shown in Table 2.127

!"#$%&$'()*+,-.-+/00$%+"1/2'#

344

344

34

3544 644 )44

78+++9:$;<!5=

900

>=?9

=++90"=

@44

,AB+C%$DE7E2F%G6H+0"I3+++J+K$;

A$'EF2+$L0$MN$'3 +$L0$MN$'+%F2O$5 +$L0$MN$'+%F2O$

Figure 2: The expected one- and two-standard-deviation ranges (blue bands) and observed

(red line) 95% C.L. upper limits on σφ · Bττ as a function of mA.

We can interpret the upper limits on σφ · Bττ in the MSSM parameter space of tan β versus mA128

for an example scenario. We use here the mmax

h[23] benchmark scenario in which MSUSY = 1129

TeV/c2; XPQt = 2MSUSY; µ = 200 GeV/c

2; Mg̃ = 800 GeV/c

2; M2 = 200 GeV/c

2; and Ab = At,130

where MSUSY denotes the common soft-SUSY-breaking squark mass of the third generation;131

Xt = Atµ/ tan β the stop mixing parameter; At and Ab the stop and sbottom trilinear couplings,132

respectively; µ the Higgsino mass parameter; Mg̃ the gluino mass; and M2 the SU(2)-gaugino133

mass parameter. The value of M1 is fixed via the GUT relation M1 = (5/3)M2 sin θw/ cos θw.134

HIG-10-002

Interpretation in tanβ 7

!"#$%&'()*)+,&-$$$./$0123$$$4$5'6

7

87

97

/7

:,+

37

.7

;7

"##"$*<$$$$=>'+,&)?@$"#A#B$C$3$5'6D;E$!FGF$'H>(IJ'J$&'K)?+=

377 877 .77*L$$$MN'6O!8P

3;7 8;7

*,H

$Q3 $:<'?&-$!"#$'H>(IJ'J

5'R,:&?+$'H>(IJ'JGS%$'H>(IJ'J$!"#$'H0'>:'J

Figure 3: Region in the parameter space of tan β versus mA excluded at 95% C.L. in the context

of the MSSM mmaxh scenario, with the effect of ±1σ theoretical uncertainties shown. The other

shaded regions show the 95% C.L. excluded regions from the LEP and Tevatron experiments.

[4] G. Guralnik, C. Hagen, and T. Kibble Phys. Rev. Lett. 13 (1964) 585.156

[5] P. W. Higgs Phys. Rev. 145 (1966) 1156.157

[6] E. Witten Phys. Lett. B 105 (1981) 267.158

[7] S. P. Martin, “A Supersymmetry Primer”, arXiv:hep-ph/9709356 (1997). and references159

therein.160

[8] CDF and D0 Collaboration arXiv:hep-ex/1003.3363 (2010).161

[9] LEP Higgs Working Group Collaboration Eur. Phys. J. C47 (2006) 547–587.162

[10] CMS Collaboration, “The CMS experiment at the CERN LHC”, J. Instrum. 0803:S08004163

(2008).164

[11] CMS Collaboration, “Electron reconstruction and identification at sqrt(s) = 7 TeV”, CMS165

Physics Analysis Summary EGM-10-004 (2010).166

[12] CMS Collaboration, “Performance of muon identification in pp collisions at√

s = 7 TeV”,167

CMS Physics Analysis Summary MUO-10-002 (2010).168

[13] CMS Collaboration, “Particle-Flow Event Reconstruction in CMS and Performance for169

Jets, Taus, and Emiss”, CMS Physics Analysis Summary PFT-09-001 (2009).170

[14] CMS Collaboration, “Commissioning of the Particle-Flow Reconstruction in171

Minimum-Bias and Jet Events from pp Collisions at 7 TeV”, CMS Physics Analysis172

Summary PFT-10-002 (2010).173

HIG-10-002

Summary• Large improvements in τ-ID since PTDR

• Tau fake rates & efficiency measured in real data

• Z→ττ cross section measured withhigh precision

• Taus well understood at CMS

• SVfit method improves Mττ resolution

• CMS sets new limits on MSSM using taus

backup

Particle Flow AlgorithmCMS Preliminary

4 2 Commissioning of the link algorithm

(a) (b)

Figure 3: Charged-hadron tracks (lines with squares, representing the hits measured in the

tracker and the various extrapolation positions to the ECAL and HCAL) each linked to (a) one

or two ECAL clusters and (b) an HCAL cluster (dots). Each square represents a calorimeter

cell. The grey area is proportional to the logarithm of the energy measured in each cell. The

clusters represented by a star are linked to neither of the two tracks, and are therefore photon

candidates.

(a) (b)

Figure 4: Track-cluster link distance in (a) the ECAL and (b) the HCAL, for tracks with pT larger

than 1 GeV/c, in the data (dots with error bars) and in the simulation (histogram). The typical

cell size is 0.02 in the ECAL and 0.1 in the HCAL.

ECAL Entrance

ECAL Entrance

HCAL Entrance

HCAL Entrance

cluster linked to trackunlinked clustertracker hit

see CMS PAS PFT-10-002

Particle Flow Performance

invariant mass of PF photon pairs

2 2 Calorimeter response to photons and charged hadrons

The photon-pair invariant-mass distribution is shown in Fig. 1 along with a fit by a Gaussian

for the signal, added to an ad hoc function for the background. From this fit, an average re-

constructed mass of 135.2 ± 0.1 MeV/c2and a resolution of 13.2 ± 0.1 MeV/c2

are obtained, to

be compared with the prediction from the Geant-based simulation of 136.9 ± 0.2 MeV/c2for

the mass and 12.8 ± 0.2 MeV/c2for the resolution. The agreement with the world average π0

mass of 135 MeV/c2[8] is better than 1%, and the agreement between data and simulation is

at the same level. This agreement validates the photon energy scale used in the particle-flow

reconstruction algorithm.

)2Mass (GeV/c0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

Num

ber o

f pho

ton

pairs

0

5000

10000

15000

20000

)2Mass (GeV/c0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

Num

ber o

f pho

ton

pairs

0

5000

10000

15000

20000

CMS Preliminary 2010

2 0.1 MeV/ c± = 135.2 fit0!m

2 0.1 MeV/ c± = 13.2 0!m"

-17-TeV Data, 0.1 nb

(a)

)2Mass (GeV/c0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

Num

ber o

f pho

ton

pairs

0

500

1000

1500

2000

)2Mass (GeV/c0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

Num

ber o

f pho

ton

pairs

0

500

1000

1500

2000

CMS Preliminary 2010

2 0.2 MeV/ c± = 136.9 fit0!m

2 0.2 MeV/ c± = 12.8 0!m"

Simulation

(b)

Figure 1: Photon-pair invariant-mass distribution in the barrel (|η| < 1.0) for the data (a) and

the simulation (b).

2.2 Calorimeter response to hadrons

A charged or neutral hadron can deposit energy in the electromagnetic calorimeter and in the

hadron calorimeter. The hadron energy E is obtained with a simple linear calibration of the

resulting EECAL and EHCAL cluster energies [2]. This calibration is essential in the reconstruc-

tion of neutral hadrons, but also in the detection of neutral particles overlapping with charged

hadrons. The calibration is assumed to be identical for charged and neutral hadrons, and has so

far been obtained from single hadrons produced with the fast simulation of the CMS detector.

To verify the calibration procedure, a clean selection of charged hadrons was made in which

the charged-hadron candidate was required to have a track with pT > 1 GeV/c, p > 3 GeV/cand at least 15 hits, of which at least two in the pixel layers. The requirement on the number of

hits was tightened to up to 21 hits when the track points to the end-cap regions, which feature

more tracker layers. These stringent criteria are meant to select particles having high-quality

tracks with a robust momentum measurement and that reach the calorimeter without experi-

encing nuclear interactions with the tracker material. The hadron identification was ensured by

requiring the track to be linked to a cluster in the hadron calorimeter with EHCAL > 1 GeV. An

isolation requirement was applied by requesting that no other tracks be linked to this cluster,

which is loose enough to allow for a very large sample of tracks to be extracted.

Figures 2a and 2b display the average calibrated calorimeter response in the barrel (here de-

fined as |η| < 1.4) and in the end-caps (1.4 < |η| < 2.4) for the selected charged hadrons in

the data, with a comparison to the raw calorimeter response, as a function of the track mo-

mentum. The slope of the cluster calibration curve is close to unity up to the largest energies

available so far. Small deviations from unity at low momentum, however, can be observed in

7 TeV Data, 0.1 nb-1

Et-60 -50 -40 -30 -20 -10 0 10 20 30 40

#

0

50

100

150

200

250

300

350

400

450

Particle Flow Jets

Calo Tower Jets

Jet Transverse Energy Resolution

50 GeV Et tau

CMS Preliminary

Particle Flow

Tau ET resolution Mass of photon pair [GeV/c2]

SIM

Shrinking Cone Algorithmreduce 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

isolation objects are those in the region ΔRsig < ΔR < 0.5

about the lead candidate

∆R =�

∆φ2 + ∆η2

∆Rsig = 5.0/EjetT

Zττ selections

Trigger

Events triggered by single Electron/Muon Triggers

PT thresholds 9-15 GeV, depending on instantaneous Luminosity

Lepton Selection

Electrons Muons had. ! Decays

PT > 15 GeV PT > 15 GeV PT > 20 GeV

|!| < 2.4 |!| < 2.1 |!| < 2.4

isolated isolated “loose” Tau id.

Veto against e/µ

Opposite Charge Lepton Pair

Transverse Mass

e + !had, µ + !had: MT(l + MET) < 40 GeV

e + µ: MT(e + MET ) < 50 GeV && MT(µ + MET) < 50 GeV

Veto Events with additional isolated Leptons

For µ + !had, e + !had and e + µ Channels, µ + µ Channel different (Backup)


Event Selection

Zττ 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


Event Yields

Z recoil correctionbefore correction after correction

MET resolution better in Monte Carlo than data

Project MET in directions parallel and perpendicular to Z and parametrize as function of Z pT

Zμμ events Zμμ events

sub title

“full” !+!- Mass visible Mass

34

Observed vs. expected Limits by Channel

control plotsHiggs Search Lepton PT and !: µ + !had Channel


control plotsHiggs Search Lepton PT and !: e + !had Channel


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

Z ! !+!- Systematic Uncertainties

Tau Reconstruction at CMS - University of California,...

Documents

Transcript of Tau Reconstruction at CMS - University of California,...