Page 1
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
Page 2
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:
Page 4
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...)
Page 5
Ryan Reece (Penn) 5tt̄ → bb̄(µν)(τhν) candidate
Page 6
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?
-
Page 7
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.
Page 9
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
Page 10
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
Page 11
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
Page 13
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)
[ATLAS-CONF-2011-152]
0.40.2
pile-up
tau underlying
event
calculate
REM, Rtrack
in cone
count
# tracks
in cone
∆R
jets
true τhadpile-up
Page 14
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.
[ATLAS-CONF-2011-152]
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
Page 15
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
Page 16
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.
Page 17
TriggeringATLAS
dt L = 1.0 fb
EF_tau20_medium1
Page 18
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
Page 19
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:
https://twiki.cern.ch/twiki/bin/view/AtlasPublic/TauTriggerPublicResults
Page 20
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)
Page 21
Systematic
Uncertainties 0.5
1
1.5
Page 22
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
[ATLAS-CONF-2012-142]
• Scale factor ≈ 1, known to a few percent, 2-3% (1-prong),
5-6% multi-prong.
Page 23
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
https://twiki.cern.ch/twiki/bin/view/AtlasPublic/TauTriggerPublicResults
• 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.
Page 24
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
Page 25
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
[ATLAS-CONF-2012-054]
[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
Page 26
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%
Page 27
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.
Page 29
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%
Page 30
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
Page 32
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.
Page 33
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
[ATL-COM-PHYS-2009-229]
• 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
,
Page 34
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
[ATL-COM-PHYS-2009-229]
• Monte Carlo based substructure studies
• Cell-based shower-shape subtraction π0 reconstruction.
• Still unvalidated with data.
Page 35
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”
[ATLAS-CONF-2010-059]
“Tau Reconstruction and Identification Performance in ATLAS”
[ATLAS-CONF-2010-086]
• 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
Page 36
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
dijet Monte Carlo-1dt L = 23 pb∫2010 dijet data
<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
dijet Monte Carlo-1dt L = 23 pb∫2010 dijet data
<60 GeVT
3 prongs 15 GeV<p
ATLAS Preliminary
ATLAS Work in progress
Page 37
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
Page 38
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]
Page 39
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]
Page 40
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]
Page 41
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
Page 42
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.
Page 43
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:
Page 44
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)
Page 46
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
Page 47
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.
Page 48
Ryan Reece (Penn)
CMS: Hadron Plus Strip (HPS)
48
Discrimination with calorimeter based isolation ∆R < 0.5.
[CMS PAS TAU-11-001]
Page 49
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%
[CMS PAS TAU-11-001]
Page 50
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,
[CMS PAS TAU-11-001]
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
-1 = 7 TeV, 36 pbsCMS,
(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.
Page 51
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
-1 = 7 TeV, 36 pbsCMS,
[CMS PAS TAU-11-001]
Page 52
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
Page 53
The LHC,
ATLAS, and CMS
Page 54
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
Page 55
Ryan Reece (Penn) 55
Page 56
Ryan Reece (Penn) 56
Page 58
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
Page 59
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
Page 60
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
Page 61
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
Page 62
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