a review of the tau object reconstructed and supported by ...rreece.github.io/talks/pdf/2012-11-13-RReece-ATLAS-tau...2012/11/13 · 2011 pp s = 7 TeV 2012 pp s = 8 TeV ATLAS Online

Ryan Reece

University of Pennsylvania

[email protected]

November 13, 2012Chicago 2012 Workshop on LHC Physics, The University of Chicago

Hadronic tau decays in ATLAS

a review of the tau object reconstructed and

supported by ATLAS and its current status

Ryan Reece (Penn)

Outline

1. Introductionmotivation, dataset, pile-up

2. Reconstructionseeding, vertex choice, track selection

3. Identificationjet and electron discriminants, performance

4. Triggeringtau chain, VBF triggers

5. Systematic uncertaintiesefficiency and energy scale measurements

2

Hadronic tau decays in ATLAS...

• Focus on current issues

• Adapting to luminosity

increases

• Pile-up robustness

• Improving systematics with

additional data

Themes:

Introduction

Ryan Reece (Penn)

What’s a tau?

4

• Only lepton massive enough to decay hadronically.

• Decay in beam pipe: cτ ≈ 87 µm

• 65% hadronic 50% 1-prong, 15% 3-prong.

• Signature: narrow jet with 1 or 3 tracks, possibly additional EM clusters.

• Challenge: large multijet background at hadron colliders.

• Importance: often preferred coupling to new physics (SM H→ττ, H+→τ+ν, Z’→ττ, high-tanβ SUSY...)

Ryan Reece (Penn) 5tt̄ → bb̄(µν)(τhν) candidate

Ryan Reece (Penn)

Month in YearJan Apr Jul

Oct

]-1

Deliv

ere

d L

um

inosity [

fb

0

5

10

15

20

25

30

= 7 TeVs2010 pp

= 7 TeVs2011 pp

= 8 TeVs2012 pp

ATLAS Online Luminosity

2012 (>20 fb-1)

2011 (5 fb-1)

2010 (36 pb-1)

Timeline of taus at ATLAS

6

• Nov 2010: Obs. of W→τν

• Feb 2011: Obs. of Z→ττ

• July 2011: W→τν and Z→ττ cross section measurements

• Feb 2012: Z→ττ cross section with 1.5 fb-1.

• June 2012: SM H→ττ excluded 3-4×SMat mH≈125 GeV[arXiv:1206.5971]

• Several other analyses: MSSM H→ττ, tt with τ, H+→τν, Z’→ττ,

SUSY τ+MET, ...• Now eagerly waiting to see if H→ττ will be excluded at 1×SM this year?

-

Ryan Reece (Penn)

Mean Number of Interactions per Crossing

0 5 10 15 20 25 30 35 40

/0.1

]-1

Record

ed L

um

inosity [

pb

0

20

40

60

80

100

120Online LuminosityATLAS

> = 20.0!, <-1Ldt = 14.0 fb! = 8 TeV, s

> = 9.1!, <-1Ldt = 5.2 fb! = 7 TeV, s

7

Pile-up

[https://twiki.cern.ch/twiki/bin/view/AtlasPublic/LuminosityPublicResults]

• 1-40 pile-up interactions / crossing

• The additional tracks and clusters

from pile-up are especially challenging for tau identification, which

discriminates hadronic tau decays from jets with isolation-related

track and calorimeter quantities.

• Efforts in 2011→2012 involved re-defining or adding corrections

to identification variables to be more robust against the increasing

pile-up.

Reconstruction

Ryan Reece (Penn)

Reconstruction

9

1. Seeded by anti-kt jets (R=0.4) of

3-D topological calorimeter clusters.

2. Define the four-momentum as the

jet-axis with a tau-specific calibration.

3. Associate tracks with the jet that are

consistent with the chosen vertex.

4. Calculate discriminating variables

from the combined calorimeter and

tracking information, later used to

identify hadronic tau decays with BDT

and likelihood based discriminants. 0.40.2

pile-up

tau underlying

event

calculate

REM, Rtrack

in cone

count

# tracks

in cone

∆R

Ryan Reece (Penn)

∑PV

pT(track)∑

allpT(track)

Fraction of track pT

from the primary

vertex.

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

!#,)-.!'()"*)+

Tau vertex association

10

• pT > 1 GeV,

• Number of pixel hits ≥ 2,

• Number of pixel hits + number of SCT hits ≥ 7,

• |d0| < 1.0 mm,

• |z0 sin θ| < 1.5 mm,

Tau track selection

JVF(jet, vertex)

∑{

tracks matched

to jet

}

pT(track)

∑

{

tracks matched

to jet and vertex

}

pT(track)

=

“Jet Vertex Fraction”

( )( )beamline

Ryan Reece (Penn)

Track selection efficiency

11

• In 2011, the track selection for tau candidates cut on the d0 and

z0 with respect to the vertex with the highest ∑pT2.

• Selecting the vertex with the highest JVF recovers efficiency in

high pile-up (Tau Jet Vertex Association).

Number of tracks

0 1 2 3 4

1-p

rong

τN

um

ber

of

sele

cte

d

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

=0µ=20µ=20 with TJVAµ

SimulationPreliminary ATLAS

[ATLAS-CONF-2012-142]

µ

0 5 10 15 20 25 30 35 40

Tra

ck s

ele

ction e

ffic

iency

0.5

0.55

0.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1

TJVA

Default

SimulationPreliminary ATLAS

2011(2012)

true 1-prong taus

Identification

Ryan Reece (Penn)

Identification and pile-up

13N(vertex)

ATLAS Preliminary

2 4 6 8 10 12 14

⟩EM

R⟨

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

0.22

• Important offline variable in 2010-2011:

EM radius - “width of jet in calorimeter”

• Strong pile-up dependence due to

using calorimeter deposits in the wide

cone: ∆R < 0.4.

REM =

∑{∆R<0.4} E

EM

T(cell)∆R(cell, jet)

∑{∆R<0.4} E

EM

T(cell)


0.40.2

pile-up

tau underlying

event

calculate

REM, Rtrack

in cone

count

# tracks

in cone

∆R

jets

true τhadpile-up

Ryan Reece (Penn) coref

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Arb

itra

ry U

nits

0

0.02

0.04

0.06

0.08

0.1

0.12ττ→+Zντ→W

dijet Monte Carlo-1dt L = 23 pb∫2010 dijet data

<60 GeVT

3 prongs 15 GeV<p

ATLAS Preliminary

Pile-up robust variables

14

• Beginning in 2012, the core energy

fraction is used instead of REM, which

has less pile-up dependence by using the

ratio of energies in smaller ∆R cones of

0.1 and 0.2.


fcore =

∑{∆R<0.1} E

EM

T(cell)

∑{∆R<0.2} E

EM

T(cell)

0.40.2

pile-up

tau underlying

event

calculate

REM, Rtrack

in cone

count

# tracks

in cone

∆R

fcore

0.1

Ryan Reece (Penn)VtxN

0 5 10 15 20 25 30

Sig

nal E

ffic

iency

0

0.2

0.4

0.6

0.8

1

1.2

BDT loose

BDT medium

BDT tight

1 Prong

| < 2.3! > 20 GeV, |T

p

ATLAS Preliminary 2012

Pile-up corrections

15

• Also beginning in 2012, the variables with the largest pile-up dependence

(fcore and ftrack) are corrected with terms that are linear in the number of

reconstructed vertices.

• Tight/Medium/Loose working points of the BDT and LLH are defined

(≈40%, 60%, 70% efficient), optimized as function of pT and in separate

N(vertex) categories.

fcore =

∑{∆R<0.1} E

EM

T(cell)

∑{∆R<0.2} E

EM

T(cell)

+ (0.3%/vertex) × N(vertex)

[https://twiki.cern.ch/twiki/bin/view/AtlasPublic/TauPublicCollisionResults]

Signal Efficiency0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Inve

rse

Ba

ckgro

un

d E

ffic

ien

cy

1

10

210

310

410

510

BDT

Likelihood

1 Prong| < 2.3! > 20 GeV, |

Tp

ATLAS Preliminary

-1 dt L = 370 pb"

2012

Ryan Reece (Penn)

Electron veto

16

HTf

ATLAS Preliminary

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Sam

ple

Fra

ction /

0.0

2

0

0.05

0.1

0.15

0.2

0.25 ττ→Z

ee→Z>20 GeV

Tp

Signal Efficiency

0.5 0.6 0.7 0.8 0.9 1

Inve

rse

Ba

ckgro

un

d E

ffic

ien

cy

1

10

210

310

ATLAS Preliminary Simulation

| < 2.0η> 20 GeV, |T

1-prong, p

BDT-based electron veto

[ATLAS-CONF-2011-152, ATLAS-CONF-2012-142]

• Electrons provide a track and

calorimeter deposit that can fake

hadronic tau decay identification.

• ATLAS provides a BDT to

discriminate electrons from tau

candidates, even after removing

overlaps with selected electrons.

• Tight/Medium/Loose working

points are defined (≈75%, 85%,

95% efficient).

• In 2012, the BDT is being re-

optimized to have better efficiency

at high-pT.

TriggeringATLAS

dt L = 1.0 fb

EF_tau20_medium1

Ryan Reece (Penn)

Tau triggering

18

Vertical sumsΣ

Σ Horizontal sums

Σ Σ

Σ

Σ

Electromagneticisolation ring

Hadronic inner coreand isolation ring

Electromagneticcalorimeter

Hadroniccalorimeter

Trigger towers (∆η × ∆φ = 0.1 × 0.1)

Local maximum/Region-of-interest

1. Level 1: (latency 2.5 µs)Coarse EM+Had calorimeter trigger towers

∆η×Δ! = 0.1×0.1. Candidate passing

thresholds on the sum of energies:

1. highest 2×1 towers2. surrounding 4×4 isolation ring

2. Level 2: (latency 40 ms)Fast tracking. Region-of-interest (RoI)

calculation of track- and calorimeter-based

ID variables. Similar selection to offline cut-

based ID.

3. Event Filter: (latency 4 s)Beginning in 2012, started using the offline

BDT algorithm at the EF trigger.

[GeV]T

p!Offline

0 10 20 30 40 50 60 70 80 90 100

Effic

ien

cy w

.r.t

. M

ediu

m B

DT

0

0.2

0.4

0.6

0.8

1

L1

L1+L2

L1+L2+EF

ATLAS Preliminary

-1 L dt = 2.8 fb"Data 2012,

tau20_medium1

https://twiki.cern.ch/twiki/bin/view/AtlasPublic/TauTriggerPublicResults

Ryan Reece (Penn)

Number of vertices

0 2 4 6 8 10 12 14 16 18

Effic

ien

cy w

.r.t

. M

ediu

m B

DT

0

0.2

0.4

0.6

0.8

1

L1

L1+L2

L1+L2+EF

ATLAS Preliminary

-1 L dt = 2.8 fb!Data 2012,

tau20_medium1

L2 pile-up robustness

19

• Smaller ∆R cone for calculating EM radius 0.4→0.2

• Select tracks within ∆z < 2 mm of the highest-pT track within the RoI

(cannot vertex at L2).

Number of vertices

0 2 4 6 8 10 12 14

Effic

ien

cy w

.r.t

Me

diu

m B

DT

0

0.2

0.4

0.6

0.8

1

L1

L2

L2+EF

ATLAS Preliminary

Data 2011, -1dt L = 2.5 fb!

tau20_medium

Example improvements to variable definitions to lessen

sensitivity to pile-up:


Ryan Reece (Penn)

VBF triggers

20

τhade/μ

• New VBF triggers relax tau identification required at L2 and the EF by adding requirements for forward jets.

• This increases the control sample of tau candidates that will fail identification, used to estimate the fake contribution.

• Being evaluated for the H→ττ→ lep+τhad search.

τhad+µ τhad+etau20_medium1_mu15

tau20Ti_medium1_e18vh_medium1

mu15_vbf_L1TAU8_MU10 e18vh_medium1_vbf_2L1TAU11I_EM14VH

• 2 L2 jets pT > 15 GeV, |∆η| > 2.5

• 2 EF jets pT > 25 GeV, |∆η| > 2.8,

Mjj > 400 GeV

|∆η| > 2.8

New as of periods G1, H2

Trigger menu

VBF reqs:relax ID

/Z

/Z

(jet)

(jet)

Systematic

Uncertainties 0.5

1

1.5

Ryan Reece (Penn)

15202530354045505560IDε

0.4

0.6

0.8

1

1.2

1.4

1.6Data 2011

ττ→Z

MC Stat. + Measur. Syst. Uncert.

ATLAS Preliminary

1-Prong, BDT Medium

=7 TeVs, -1

dt L = 3.6 fb∫

) [GeV]had-visτ(

Tp

15 20 25 30 35 40 45 50 55 60M

CIDε/

Data

IDε 0.5

1

1.5

) [GeV]had-visτ,µm(

0 20 40 60 80 100 120 140 160 180 200

Events

/ 5

GeV

0

500

1000

1500

2000

2500

3000

3500

4000

4500ATLAS Preliminary

-1dt L = 1.8 fb∫ = 7 TeVs

Before Tau ID

Inclusive

Data early 2011

ττ→Zνµ→W

Multi-jet

µµ→Z

ντ→W

(mis-ID)ττ→Z

tt

Identification efficiency

22

• Tag-and-probe: selecting a sample of a known composition

without some ID, so one can probe its efficiency.

• For the case of tau ID, select Z→ττ→μτh3ν by triggering on the

muon and selecting events with muon + tau candidate.

Before After Tau ID


• Scale factor ≈ 1, known to a few percent, 2-3% (1-prong),

5-6% multi-prong.

Ryan Reece (Penn)

Trigger efficiency

23

Effic

ien

cy

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

ATLAS Preliminary

, -1

dt L = 1.0 fb! = 7 TeVs

EF_tau20_medium1

Data

""#Z

) [GeV]"(T

p

0 10 20 30 40 50 60 70 80 90 100

MC

$/D

ata

$

0.8

1

1.2


• The same Z→ττ→μτh3ν tag-and-probe sample is used to

measure the efficiency of the tau triggers.

• Known to O(5%) in the turn-on.

• Improving with

statistics in 2012.

Ryan Reece (Penn)

Electron veto fake rate

24[ATLAS-CONF-2012-142]

) [GeV]had-visτm(e,

50 60 70 80 90 100 110 120 130 140 150

Entr

ies /

GeV

0

10

20

30

40

50

60

70

310×

Data 2011

ee→Z

-1dt L = 4.3 fb∫= 7 TeV, s

ATLAS Preliminary

) [GeV]had-visτm(e,

50 60 70 80 90 100 110 120 130 140 150

Entr

ies /

GeV

0

20

40

60

80

100

120

140

160

180

Data 2011

ee→Z

-1dt L = 4.3 fb∫= 7 TeV, s

ATLAS Preliminary

loose BDT tau ID

medium BDT electron ID

• Tag e + tau candidates

• Probe the e-veto efficiency after removing overlap with selected electrons.

• Statistically limited by the sample that pass the veto, giving uncertainties ≈ 50-100%.

• Improving with the data added in 2012.

from Z→ee tag-and-probe with 2.6/fb from 2011

electron BDT veto |ηtrk| < 1.37 1.37 < |ηtrk| < 1.52 1.52 < |ηtrk| < 2.00 |ηtrk| > 2.00

loose 0.96±0.22 0.8±0.3 0.47±0.14 1.7 ±0.4

medium 1.3 ±0.5 - 0.5 ±0.4 2.8 ±1.3

data/MC scale factor and uncertainty

Ryan Reece (Penn)

Energy scale

25

[GeV]τrecoE

20 30 40 210 210×2

response

0.95

1

1.05

1.1

1.15

1.2

1-prong| < 0.3η |≤0 | < 0.6η |≤0.3 | < 1η |≤0.6 | < 1.3η |≤1 | < 1.6η |≤1.3 |η |≤1.6

ATLAS Preliminary

Simulation

= 7 TeVs


[GeV]τTP

20 30 40 50 60 70 80 100 200

Fra

ctional uncert

ain

ty

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

|<0.3η|

1 prong decaysSingle particle resp. Material modeling

Underlying event Non-closure

Pile-Up Total uncertainty

2011 Data + Simulation PreliminaryATLAS

= 7 TeV s

• Tau candidates are first brought from the EM to the Jet Energy Scale with LC calibration of the clusters within ∆R < 0.2 (from 0.4 to be pile-up robust).

• Then response functions are calibrated with tau Monte Carlo to make final corrections of a few percent.

• Uncertainties are determined by smearing the Monte Carlo truth according the tau decays true composition, using uncertainties constrained by single particle response measurements (CTB, E/p, Z→ee/π0-resp.)

Response functions Scale uncertainties

Ryan Reece (Penn)

| | −

) [GeV]hadτ, l(

vism

0 20 40 60 80 100 120 140 160 180 200

Entr

ies/5

GeV

0

200

400

600

800

1000

1200

1400

1600

1800

2000

tt τ →W µ µ →Z µ →W

Multijetτ τ →Z

Data

ATLAS Preliminary

= 7 TeVs

Data 2011-1

L dt = 4.26 fb∫

Energy scale cross check

26[ATLAS-CONF-2012-054]

• Tau energy scale ismanually shifted in the modeling.

• Median of the visible masspeak is used to decide whichscale matches the data.

• Toy experiments are usedto estimate the uncertainty.

• Scale consistent with 1 within single-particle-response uncertainties ≈ 3%.

• May become primary method with more data.

scale shifted -10%

| | −

) [GeV]hadτ, l(

vism

0 20 40 60 80 100 120 140 160 180 200

Entr

ies/5

GeV

0

200

400

600

800

1000

1200

1400

1600

1800

2000

tt τ →W µ µ →Z µ →W

Multijetτ τ →Z

Data

ATLAS Preliminary

= 7 TeVs

Data 2011-1

L dt = 4.26 fb∫

best fit +1.5%|η| best scale uncert.

0.0-0.8 -1.5% 3.3%

0.8-2.5 +1.5% 2.8%

Ryan Reece (Penn)

Conclusions

27

• The rise of pile-up in 2011 challenged the performance

of tau identification and triggering.

• Efforts in multiple areas (identification, triggering,

energy calibration) have mitigated the effects of pile-

up with better design choices or corrections.

• The future will bring opportunities to further shrink

our scale factor uncertainties with additionally

analyzed 2012 data.

• It is an exciting time to analyze tau final states at

ATLAS.

Back up

Ryan Reece (Penn)

Phenomenology of tau decays

29

τ−

→ e− ν̄e ντ

µ−

ν̄µ ντ

17.8%17.4%

}

leptonic 35.2%

π−

π0ντ

π−

ντ

π− 2π0

ντ

K− (Nπ0) (NK0) ντ

π− 3π0

ντ

25.5%10.9%9.3%1.5%1.0%

1 prong 49.5%

π−

π−

π+

ντ

π−

π−

π+

π0ντ

9.0%4.6%

}

3 prong 15.2%

Ryan Reece (Penn)

1. Core energy fraction*

2. Leading track momentum fraction*

3. Track radius

4. Number of isolation tracks

5. Leading track impact parameter significance

6. Transverse flight path significance

7. Mass of track system

8. Maximum ∆R between jet-axis and core tracks

*has pile-up correction term linear in N(vertex)

Current tau identification variables

30

Rtrack =

∑{∆R<0.4} pT(track)∆R(track, jet)

∑{∆R<0.4} pT(track)

fcore =

∑{∆R<0.1} E

EM

T(cell)

∑{∆R<0.2} E

EM

T(cell)

Number of tracks in isolation annulus (N0.2<∆R<0.4trk

):

S lead track =d0

σd0

is the distance of closest approach of the track to the reconsSflight

T=

Lflight

T

σLflight

T

tauRec

Ryan Reece (Penn)

of tau candidate (GeV)TE0 10 20 30 40 50 60 70 80 90 100

Fra

ction o

f candid

ate

s-510

-410

-310

-210

calo-seeded only

both seeds

track-seeded only

ATLAS

Seeds of reconstruction

32

“Performance of the tau reconstruction and identification algorithm with 14.2.20 and mc08”

[ATL-COM-PHYS-2009-229]

1. tauRec - seeded by

pT > 10 GeV anti-kT

0.4 topo-jets.

“calo-seeded”

2. tau1p3p - seeded by

pT > 6 GeV inner

detector tracks.

“track-seeded”

Once upon a time, there were two tau reconstruction algorithms.

Since virtually all candidates have a calo-seed, we effectively merged the

variable calculation of both algorithms, using only calo-seeds.

Ryan Reece (Penn)

Early MV identification

33

IsEle(%) IsEle eg(%)

Candidate Overall 1P 3P Overall 1P 3P

τ from W→ τν 93.2 92.7 95.3 99.8 99.8 99.8

τ form A→ ττ 93.3 92.5 96.3 99.9 98.8 99.5

Electron form W→ eν 2.8 2.4 0.1 14.8 13.4 0.3

Electron form A→ ττ 5.9 4.5 0.5 18.0 15.8 0.8

Efficiency0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Reje

ction

1

10

210

310

410

510

610

= 10 - 30 GeVTE

= 30 - 60 GeVTE

= 60 - 100 GeVTE

ATLAS

1 prong

all calo seeds

Efficiency0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Reje

ction

1

10

210

310

410

510

610

= 10 - 30 GeVTE

= 30 - 60 GeVTE

= 60 - 100 GeVTE

ATLAS

3 prong

all calo seeds

Figure 13: Rejection vs efficiency for single or three prong candidates, for events where -candidates


• Jet-tau discrimination

Prefers narrow calorimeter jets, likelihood-based discriminant.

• Electron-tau discrimination

REM =

∑∆Ri<0.4i

EEM

T,i∆Ri

∑∆Ri<0.4i

EEM

T,i

,

Ryan Reece (Penn)

Early sub-structure studies

34

0πnumber of reconstructed

0 2 4 6 8 10

Fra

ction o

f candid

ate

s

0

0.1

0.2

0.3ν π → τ

ν ρ → τ

ν) π0π(21

a→ τ

ATLAS

Invariant mass (GeV)0 0.5 1 1.5 2 2.5

Fra

ction o

f candid

ate

s

0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

ν π → τν ρ → τ

ν) π0π(21 a→ τ

ATLAS


• Monte Carlo based substructure studies

• Cell-based shower-shape subtraction π0 reconstruction.

• Still unvalidated with data.

Ryan Reece (Penn)

[GeV] Tp

0 10 20 30 40 50 60 70 80 90 100

bkgd

ε

-210

-110

1

-1Integrated Luminosity 244 nb / Loose Cuts (Data/MC) / Medium Cuts (Data/MC) / Tight Cuts Data(MC)

PreliminaryATLAS PreliminaryATLAS

EMR

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35

candid

ate

s / 0

.01

τN

um

ber

of

0

20

40

60

80

100

120

140

160

180

200

220

310×

ATLAS Preliminary = 7 TeV )sData 2010 (

Pythia QCD JetsττPythia Z->

-1Integrated Luminosity 15.6 nb

First data

35

“Reconstruction of hadronic tau candidates in QCD events at ATLAS with 7 TeV pp collisions”


“Tau Reconstruction and Identification Performance in ATLAS”


• First comparisons of background distributions and the QCD

fake-rate between data and Monte Carlo.

• Already see that MC over-estimates the jet fake-rate. ⇒kW ≈ 0.5

Ryan Reece (Penn)

Tau discriminants

36

CutspT-parametrized cuts on REM and

Rtrack, and a cut on ftrack.

Projective likelihood

d = ln

(

LS

LB

)

=∑N

i=1 ln

(

pSi(xi)

pBi(xi)

)

Boosted decision trees (BDT)

[GeV]T

p

20 30 40 50 60 70 80 90 100

⟩tr

ack

R⟨

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

Likelihood Score

-20 -15 -10 -5 0 5 10 15 20

Arb

itra

ry U

nits

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18ττ→+Zντ→W


<60 GeVT

3 prongs 15 GeV<p

ATLAS Preliminary

BDT Score

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1A

rbitra

ry U

nits

0

0.02

0.04

0.06

0.08

0.1

0.12 ττ→+Zντ→W


<60 GeVT

3 prongs 15 GeV<p

ATLAS Preliminary

ATLAS Work in progress

Ryan Reece (Penn)

Maturing of discriminants

37

Signal Efficiency

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Inve

rse

Backgro

und E

ffic

ien

cy

1

10

210

310

Cuts

BDT

Likelihood

>20GeVT

1-Prong p

ATLAS Preliminary

[GeV]T

p

20 30 40 50 60 70 80 90 100

⟩E

MR⟨

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

(a) R cut curves for 1-prong

• Cuts are pt-parametrized to account for the Lorentz collimation

of boosted taus.

• Experience grows with LLH and BDT discriminants, which

become the preferred discriminants in 2011.

“Reconstruction, Energy Calibration, and Identification of Hadronically Decaying Tau Leptons

in the ATLAS Experiment” [ATLAS-CONF-2011-077, ATL-PHYS-INT-2011-068]

ATLAS Work in progress

Ryan Reece (Penn)

Seeing first hadronic taus

38

Number of tracks

0 1 2 3 4 5 6 7 8 9 10

Num

ber

of

events

0

500

1000

1500

2000

2500

0 1 2 3 4 5 6 7 8 9 100

500

1000

1500

2000

2500

0 1 2 3 4 5 6 7 8 9 100

500

1000

1500

2000

2500

0 1 2 3 4 5 6 7 8 9 100

500

1000

1500

2000

2500

0 1 2 3 4 5 6 7 8 9 100

500

1000

1500

2000

2500

0 1 2 3 4 5 6 7 8 9 100

500

1000

1500

2000

2500

0 1 2 3 4 5 6 7 8 9 100

500

1000

1500

2000

2500

0 1 2 3 4 5 6 7 8 9 100

500

1000

1500

2000

2500 = 7 TeV)sData 2010 (

τν hτ →W

EW background

QCD background (B)

ATLAS Preliminary

-1 L dt = 34 pb∫

EMR

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14

Num

ber

of

events

/ 0

.01

0

100

200

300

400

500

600

700

0 0.02 0.04 0.06 0.08 0.1 0.12 0.140

100

200

300

400

500

600

700

0 0.02 0.04 0.06 0.08 0.1 0.12 0.140

100

200

300

400

500

600

700

0 0.02 0.04 0.06 0.08 0.1 0.12 0.140

100

200

300

400

500

600

700

0 0.02 0.04 0.06 0.08 0.1 0.12 0.140

100

200

300

400

500

600

700

0 0.02 0.04 0.06 0.08 0.1 0.12 0.140

100

200

300

400

500

600

700

0 0.02 0.04 0.06 0.08 0.1 0.12 0.140

100

200

300

400

500

600

700

0 0.02 0.04 0.06 0.08 0.1 0.12 0.140

100

200

300

400

500

600

700 = 7 TeV)sData 2010 (

τν hτ →W

EW background

QCD background (BD)

ATLAS Preliminary

-1 L dt = 34 pb∫

• Nov 2010: Observation of W→τhν [ATLAS-CONF-2010-097]

• Feb 2011: Observation of Z→τhτl [ATLAS-CONF-2011-010]

Ryan Reece (Penn)

W→τν cross section

39

σ(W → τν) = 11.1 ± 0.3(stat.)± 1.7(sys.)± 0.4(lumi.) nb

σtheory = 10.46 ± 0.52 nb at NNLO

) [nb]lν l →(W σ

6 7 8 9 10 11 12 13 14 15 16

τν τ →W ATLAS

eν e →W ATLAS

µν µ →W ATLAS

= 7 TeV)sData 2010 (

Stat uncertainty

Stat ⊕Sys

Lumi⊕ Stat ⊕Sys

Prediction (NNLO)

Theory uncertainty

= 7 TeV)sData 2010 (

Stat uncertainty

Stat ⊕Sys

Lumi⊕ Stat ⊕Sys

Prediction (NNLO)

Theory uncertainty

ATLAS Preliminary

) [nb]lν l →(W σ

6 7 8 9 10 11 12 13 14 15 16

τν τ →W ATLAS

eν e →W ATLAS

µν µ →W ATLAS

Dominant systematics

τh efficiency 10.3%

τh energy scale 8.0%

τh + MET trigger

efficiency 7.0%

luminosity 3.4%

acceptance 2.3%

“Measurement of the W→τν cross section in pp collisions at sqrt(s)= 7 TeV with the ATLAS experiment”

[arXiv:1108.4101]

Ryan Reece (Penn)

Z→ττ cross section

40

σcombined = 0.97 ± 0.07(stat.)± 0.07(sys.)± 0.03(lumi.) nb

σtheory = 0.96 ± 0.05 nb at NNLO

<116 GeV) [nb]inv

ll, 66<m→(Z σ0.6 0.8 1 1.2 1.4 1.6

-136pb

combinedττ →Z

-133-36pb

µµ ee/→Z

hτ µτ

hτ eτ

µτ eτ

µτ µτ

Stat

Stat ⊕Syst

Lumi⊕ Stat ⊕Syst

Theory (NNLO)

Stat

Stat ⊕Syst

Lumi⊕ Stat ⊕Syst

Theory (NNLO)

ATLAS Preliminary

<116 GeV) [nb]inv

ll, 66<m→(Z σ0.6 0.8 1 1.2 1.4 1.6

-136pb

combinedττ →Z

-133-36pb

µµ ee/→Z

hτ µτ

hτ eτ

µτ eτ

µτ µτ

Dominant systematics

τh energy scale 11%

τh efficiency 8.6%

µ efficiency 8.6%

e efficiency 3-10%

acceptance 3%

luminosity 3.4%

“Measurement of the Z→ττ cross section in pp collisions at sqrt(s)= 7 TeV with the ATLAS detector”

[arXiv:1108.2016]

Ryan Reece (Penn)

pass/f

ail

0

0.05

0.1

W CR: Inclusive, 1p

W CR: OS, 1p

W CR: SS, 1p

Tp

0 50 100 150 200 250

ratio

0

1

2

pass/f

ail

0

0.05

0.1

QCD CR: Inclusive, 1p

QCD CR: OS, 1p

QCD CR: SS, 1p

Tp

0 50 100 150 200 250

ratio

0

1

2

BDT Medium BDT Medium

Observed variance in fake-rates

41

(BDTMedium)

1. Why do quarks and gluons have different tau fake-rates?

2. How does the quark/gluon fraction vary among samples?

• Hypothesis: quarks vs gluons

• Divide the issue into two questions:

ATLAS work in progress ATLAS work in progress

Ryan Reece (Penn)

Jet width for quark/gluons

42

J. Gallicchio, M. Schwartz. “Quark and Gluon Tagging at the LHC”. arXiv:1106.3076.

• !(r) = fraction of jet

energy within ∆R < r.

• Quark jets are more

narrow than gluon jets

of the same energy.

• Tau identification prefers

narrow candidates.

• This is consistent with samples of quark-enriched jets, like

W+jet, having higher fake-rates.

Ryan Reece (Penn)

OS vs SS W+jet

43

q

Wg

q′

(a)

q W

g q′

(b)

q

q̄′W

g

(c)

• The charge of the quark should correlate with the

reconstructed charge of the tau candidate, therefore (a) and

(b) preferably produce opposite sign W+jet events.

• OS and SS will have different quark/gluon fractions.

Leading order W+jet production:

Ryan Reece (Penn)

Madgraph predicted Quark/Gluon

44

50 100 200 400 800 1600

Q

G

0%

100%

80%

60%

40%

20%

pT Cut on All Jets (GeV)50 100 200 400 800 1600

Q

G

0%

100%

80%

60%

40%

20%

pT Cut on All Jets (GeV)

J. Gallicchio, M. Schwartz. “Pure Samples of Quark and Gluon Jets at the LHC”. arXiv:1104.1175

50 100 200 400 800 16000%

100%

80%

60%

40%

20%

GG

QG

QQ

pT Cut on All Jets (GeV)50 100 200 400 800 1600

0%

100%

80%

60%

40%

20%

GGG

QGG

QQG

QQQ

pT Cut on All Jets (GeV)

CMS Tau ID

Ryan Reece (Penn)

High-pT τh reconstruction

46

Isolation

annulus

pile-up

tau

UE

Count # tracks

in core cone

!"#$

"%&'()*'

+"#$

"%&'()*'

,*-./'

0)().(1*

2-*(3.%)45%16

• τh reco seeded by calorimeter jets

• associate tracks in ∆R < 0.2, select 1 or 3

• combine calorimeter and tracking

information in a BDT or likelihood

discriminant, preferring narrow clustering,

hadronic activity[ATLAS-CONF-2011-152, CMS PAS TAU-11-001]

• particle-flow reconstructs constituent 4-vectors

• τh reco seeded by particle-flow hadrons

• Hadron Plus Strip (HPS) algorithm for

counting π0s

• isolation cone for rejecting QCD jets

•Hadronic decays dominantly to 1 or 3 π± and possibly a few additional π0s

•Decay in beam-pipe: cτ ≈ 87 µm

Ryan Reece (Penn)

CMS Particle Flow

47

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

!"#$%!&'()&*+

!"#$%!&'()&*+

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

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

.(/01'&$()+2'3$14$1&,.2

/+()+2'3$.(/01'&

1&,.2'&$5)1

0''$!"#$%6#$%789:;9;;<

• Matches track to clusters to form charged and neutral PF objects.

• PF objects are used as input for all CMS tau reconstruction.

Ryan Reece (Penn)

CMS: Hadron Plus Strip (HPS)

48

Discrimination with calorimeter based isolation ∆R < 0.5.

[CMS PAS TAU-11-001]

Ryan Reece (Penn)

CMS: Tau Neural Classifier (TaNC)

49

• Uses a shrinking core-cone:

• ∆R(photons) < 0.15 for photons

• ∆R(charged) < (5 GeV)/ET for charged hadrons

• ∆R(charged) < ∆R(isolation) < 0.5

• Immediately discarded if the candidate doesn’t match

an expected tau decay mode.

• Dedicated Neural-net classifier for each decay mode

Decay mode Resonance Mass (MeV/c2) Branching fraction (%)

τ− → h−ντ 11.6%

τ− → h−π0ντ ρ− 770 26.0%

τ− → h−π0π0ντ a

−

1 1200 9.5%

τ− → h−

h+

h−ντ a

−

1 1200 9.8%

τ− → h−

h+

h−π0ντ 4.8%


Ryan Reece (Penn)

CMS Performance

50

(GeV/c)hτ

Tgenerated p

0 50 100

effic

ien

cy

τexpecte

d

0

0.2

0.4

0.6

0.8

1

HPS loose

HPS mediumHPS tight

= 7 TeVsCMS Simulation,

(GeV/c)hτ

Tgenerated p

0 50 100

effic

ien

cy

τexpecte

d

0

0.2

0.4

0.6

0.8

1

TaNC loose

TaNC mediumTaNC tight

= 7 TeVsCMS Simulation,


0 50 100 150 200

mis

identification r

ate

for

jets

τ

-310

-210

Dataνµ→W

Simulationνµ→W

QCD Data

QCD Simulation

DataµQCD

SimulationµQCD

HPS loose

-1 = 7 TeV, 36 pbsCMS,

(GeV/c)T

jet p0 50 100 150 200

Sim

ula

tion

Data

-Sim

.

-0.2

0

0.20 50 100 150 200

mis

identification r

ate

for

jets

τ

-310

-210

Dataνµ→W

Simulationνµ→W

QCD Data

QCD Simulation

DataµQCD

SimulationµQCD

TaNC loose


(GeV/c)T

jet p0 50 100 150 200

Sim

ula

tion

Data

-Sim

.

-0.2

0

0.2

• Not trivial to

compare ATLAS and

CMS tau

performance because

we bin fake-rates in

N(track) instead of

categorizing the

decay mode.

Ryan Reece (Penn)

CMS decay mode ID

51

0.85 0.16 0.05

0.13 0.83 0.04

0.02 0.01 0.91

decay modeτgenerated

π )0π(0ππ πππ

decay m

ode

τre

constr

ucte

d

π

0ππ

πππ

= 7 TeV sCMS Simulation,

decay modeτreconstructed

π 0ππ πππ

rela

tive y

ield

0

0.2

0.4

0.6

0.8

Dataτ τ →Z

W+jets

/ewkt tQCD



Ryan Reece (Penn)

Calorimeter granularity

52

• B = 3.8 T

• ∆η ×∆! = 0.0174×0.0174

• R = 0.5 anti-kT PF-jets

• B = 2.0 T

• ∆η ×∆! = 0.025×0.0245

• R = 0.4 anti-kT topo-jets

∆ϕ = 0.0245

∆η = 0.02537.5mm/8 = 4.69 mm ∆η = 0.0031

∆ϕ=0.0245x4 36.8mmx4 =147.3mm

Trigger Tower

TriggerTower∆ϕ = 0.0982

∆η = 0.1

16X0

4.3X0

2X0

1500

mm

470

mm

η

ϕ

η = 0

Strip cells in Layer 1

Square cells in Layer 2

1.7X0

Cells in Layer 3 ∆ϕ× ∆η = 0.0245× 0.05

ATLAS

CMS

ATLAS Barrel EM Calorimeter

Granularity could fundamentally limit our capacity to

reconstruct sub-structure / π0s.

0.0250.0174

The LHC,

ATLAS, and CMS

Ryan Reece (Penn)

Month in YearJan Apr Jul

Oct

]-1

Deliv

ere

d L

um

inosity [

fb

0

5

10

15

20

25

30

= 7 TeVs2010 pp

= 7 TeVs2011 pp

= 8 TeVs2012 pp

ATLAS Online Luminosity

Mean Number of Interactions per Crossing

0 5 10 15 20 25 30 35 40

/0.1

]-1

Record

ed L

um

inosity [

pb

0

20

40

60

80

100

120Online LuminosityATLAS

> = 20.0!, <-1Ldt = 14.0 fb! = 8 TeV, s

> = 9.1!, <-1Ldt = 5.2 fb! = 7 TeV, s

54

Datasets

[https://twiki.cern.ch/twiki/bin/view/AtlasPublic/LuminosityPublicResults]

Integrated luminosity by year

1-40 pile-up interactions / crossing

Ryan Reece (Penn) 55

Ryan Reece (Penn) 56

References

Ryan Reece (Penn)

Charged Higgs

• ATLAS

• “Search for charged Higgs bosons decaying via H± → τ±ν in tt ̄events using pp

collision data at √s = 7 TeV with the ATLAS detector” [arxiv:1204.2760]

• CMS

• “Search for a light charged Higgs boson in top quark decays in pp collisions at

√s = 7 TeV” [arvix:1205.5736]

58

Ryan Reece (Penn)

SUSY

• ATLAS

• “Search for events with large missing transverse momentum, jets, and at least two

tau leptons in 7 TeV proton-proton collision data with the ATLAS detector” [arxiv:

1203.6580]

• “Search for supersymmetry with jets, missing transverse momentum and at least one

hadronically decaying τ lepton in proton-proton collisions at √s = 7 TeV with the

ATLAS detector” [arxiv:1204.3852]

• CMS

• “Search for anomalous production of multilepton events in pp collisions at √s = 7

TeV” [arvix:1204.5341]

• “Search for new physics with same-sign isolated dilepton events with jets and

missing transverse energy” [arxiv:1205.6615]

• “Search for new physics in events with opposite-sign leptons, jets, and missing

transverse energy in pp collisions at √s = 7 TeV” [arxiv:1206.3949]

59

Ryan Reece (Penn)

Exotics

• ATLAS

• “A search for high-mass resonances decaying to τ+τ− in pp collisions at √s = 7

TeV with the ATLAS detector” [ATLAS-CONF-2012-067]

• CMS

• “Search for high-mass resonances decaying into τ-lepton pairs in pp collisions

at √s = 7 TeV” [arvix:1206.1725]

• “Search for pair production of third generation leptoquarks and stops that

decay to a tau and a b quark” [CMS PAS EXO-12-002]

60

Ryan Reece (Penn)

Tau performance

• ATLAS

• “Reconstruction, Energy Calibration, and Identification of Hadronically

Decaying Tau Leptons” [ATLAS-CONF-2011-077]

• “Performance of the Reconstruction and Identification of Hadronic Tau Decays

with ATLAS” [ATLAS-CONF-2011-152]

• “Z → ττ cross section measurement in proton-proton collisions at 7 TeV with

the ATLAS experiment” [ATLAS-CONF-2012-006]

• https://twiki.cern.ch/twiki/bin/view/AtlasPublic/TauPublicCollisionResults

• https://twiki.cern.ch/twiki/bin/view/AtlasPublic/TauTriggerPublicResults

61

Ryan Reece (Penn)

Tau performance• CMS

• “Performance of τ-lepton reconstruction and identification in CMS”

[arvix:1109.6034, CMS PAS TAU-11-001]

• “CMS Strategies for tau reconstruction and identification using particle-flow

techniques” [CMS PAS PFT-08-001]

• “Particle–Flow Event Reconstruction in CMS and Performance for Jets, Taus,

and ETmiss” [CMS PAS PFT-09-001]

• “Commissioning of the Particle-Flow Reconstruction in Minimum-Bias and Jet

Events from pp Collisions at 7 TeV” [CMS PAS PFT-10-002]

• “Commissioning of the particle-flow event reconstruction with leptons from

J/Psi and W decays at 7 TeV” [CMS PAS PFT-10-003]

• “Study of tau reconstruction algorithms using pp collisions data collected at

√s = 7 TeV” [CMS PAS PFT-10-004]

62

a review of the tau object reconstructed and supported by ...rreece.github.io/talks/pdf/2012-11-13-RReece-ATLAS-tau...2012/11/13 · 2011 pp s = 7 TeV 2012 pp s = 8 TeV ATLAS Online

Documents