Approval of the ATLAS combined Z’ ττrreece/docs/talks/2012-08-15_R... · Approval of the ATLAS...
Transcript of Approval of the ATLAS combined Z’ ττrreece/docs/talks/2012-08-15_R... · Approval of the ATLAS...
Ryan Reece
University of Pennsylvania
August 15, 2012ATLAS Exotics meeting, CERN
Approval of the
ATLAS combined Z’ → ττ
search with the 2011 data
on behalf of the Z’ → ττ analysis team:Will Davey (Bonn), Jochen Dingfelder (Bonn), Andres Florez (York), Julian Glatzer (Bonn), Gabriel
Palacino (York), Ryan Reece (Penn), Alex Tuna (Penn), Peter Wagner (Penn), Brig Williams (Penn)
and its editorial board:Yann Coadou (CNRS), Jean-Baptiste De Vive De Regie (Paris-Sud), Gideon Bella (Tel Aviv), Ashutosh
Kotwal (Duke)
Ryan Reece (Penn)
Introduction
2
• 2011 analysis Z’ → ττ →τhτh, µτh, eµ channels were
combined to exclude SSM of 1.3 TeV and were shown
at ICHEP 2012.
• eτh has now caught up and improved the combined
limit.
• minor updates to the τhτh and µτh channels.
Ryan Reece (Penn)
) [GeV]missTE, hτ, e(
totTm
0 500 1000 1500
Events
/ 5
0 G
eV
-210
-110
1
10
210
310
410ATLAS Preliminary
-1dt L = 4.6 fb∫ = 7 TeVs
(c)Data 2011
ττ→*γ/Z
+jetsZ/W
Multijet
ee→Z
tt
Dibosonsingle top
ττ→(1000)Z’
Z’ → ττ → eτh overview
3
Event selection
Dominant systematics
MT =
miss
miss
miss
Ryan Reece (Penn)
New systematics in plots• e →τh fake rate
uncert.≈ 50% Z→ee
• jet→τh fake factor
uncert.≈ 30% W+jet
• τh ID for Z→ττ ≈ 6%
• 100% multijet uncert.
• stat. uncert.
• hashing shows these
errors in quadrature
(nearly all the uncert.)4
Ele
ctr
ons /
(5 G
eV
)
0
50
100
data 2011
! ! "Z W/Z+jets
multijet
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
(e) [GeV]T
p
0 20 40 60 80 100 120 140obs.
/ exp.
0
1
2
• Systematics are “stacked”
in quadrature.
• Error bars are Poisson
Ryan Reece (Penn)
∆ϕ(τh ,e)
5
/30)
πE
vents
/ (
0
500
1000
1500
2000 data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1
dt L = 4.6 fb∫
, e)hτ(φ∆
0 0.5 1 1.5 2 2.5 3obs.
/ exp.
0
1
2
/30)
πE
vents
/ (
-110
1
10
210
310
410
510data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1
dt L = 4.6 fb∫
, e)hτ(φ∆
0 0.5 1 1.5 2 2.5 3obs.
/ exp.
0
1
2
Figure 19: The distribution of the absolute difference in φ between the selected electron and hadronic
tau. These plots include the requirements of: exactly one selected electron, no additional
preselected electrons or muons, and exactly one selected 1-prong tau.
Ryan Reece (Penn)
ET
6
Events
/ (
5 G
eV
)
0
200
400
600
data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
[GeV]missTE
0 20 40 60 80 100 120 140obs.
/ exp.
0
1
2
Events
/ (
20 G
eV
)
-110
1
10
210
310
410
510 data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
[GeV]missTE
0 50 100 150 200 250 300350 400 450 500obs.
/ exp.
0
1
2
Figure 21: The distribution of transverse missing energy. These plots include the requirements of: ex-
actly one selected electron, no additional preselected electrons or muons, exactly one selected
1-prong tau, |∆φ(e, τh)| > 2.7, and opposite sign e and τh.
miss
miss
Ryan Reece (Penn)
Opposite sign
7
Events
/ (
10 G
eV
)
0
50
100
150 data 2011
! ! "Z W/Z+jets
multijet
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
) [GeV]miss
T, E
h!(e, TM
0 50 100 150 200 250 300 350 400obs.
/ exp.
0
1
2
Events
/ (
10 G
eV
)
0
10
20
30data 2011
! ! "Z W/Z+jets
multijet
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
) [GeV]miss
T, E
h!(e, TM
0 50 100 150 200 250 300 350 400obs.
/ exp.
0
1
2
Opposite sign Same sign
Ryan Reece (Penn)
mT(e, ET )
8
Events
/ (
10 G
eV
)
0
50
100
data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
) [GeV]miss
T(e, ETm
0 20 40 60 80 100 120140 160 180 200obs.
/ exp.
0
1
2
Events
/ (
20 G
eV
)
-110
1
10
210
310
410 data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
) [GeV]miss
T(e, ETm
0 100200 300 400 500 600700 800 9001000obs.
/ exp.
0
1
2
Figure 22: The distribution of transverse missing energy. These plots include the requirements of: ex-
actly one selected electron, no additional preselected electrons or muons, exactly one selected
1-prong tau, |∆φ(e, τh)| > 2.7, opposite sign e and τh, and EmissT> 30 GeV.
miss
miss
Ryan Reece (Penn)
Events
/ (
10 G
eV
)
0
20
40
60
data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
) [GeV]miss
T, E
hτ(e, TM
0 50 100 150 200 250 300 350 400obs.
/ exp.
0
1
2
Events
/ (
50 G
eV
)
-110
1
10
210
310
410 data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
) [GeV]miss
T, E
hτ(e, TM
0 200 400 600 800 1000 1200 1400obs.
/ exp.
0
1
2
Figure 26: The distribution of the total transverse mass of the four-vector sum of the selected electron,
selected hadronic tau, and the transverse missing energy. These plots include the require-
ments of: exactly one selected electron, no additional preselected electrons or muons, ex-
actly one selected 1-prong tau, |∆φ(e, τh)| > 2.7, opposite sign e and τh, EmissT> 30 GeV, and
mT(e, EmissT) < 50 GeV.
mT(e, τh, ET )
9
misstot
misstot
Ryan Reece (Penn)
(e) [GeV]T
p
0 50 100 150 200 250 300 350 400
: m
ultije
te-iso
f
0
0.2
0.4
0.6
0.8
1
1.2
1.4Inclusive
Barrel
Endcap
ATLAS Internal
Multijet background estimation
10
Multijet control region
• In the control region, divide leptons
into pass and fail isolation.
• Define fake factor:
• Predict the number of QCD events:
fe–iso(pT, η) ≡Ne iso(pT, η)
Ne anti–iso(pT, η)
∣
∣
∣
∣
∣
∣
multijet–CR
Nmultijet(pT, η, x) = fe–iso(pT, η) · Ne anti–isomultijet (pT, η, x) .
We correct the sample of anti-isolated electrons in the data by subtracting the expected contamination ofx) = fe–iso(pT, η) ·(
Ne anti–isodata
(pT, η, x) − Ne anti–isoMC (pT, η, x)
)
.
Ryan Reece (Penn)) [GeV]
h!(
Tp
0 50 100 150 200 250
)h!
)+je
ts,
OS
(e,
" e
#
: W
(!
f
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
Inclusive
Barrel (inner)
Barrel (outer)
Endcap
ATLAS Internal
Events
/ (
10 G
eV
)
0
500
1000
data 2011
! ! "Z W/Z+jets
multijet
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
) [GeV]miss
T(e, ETm
0 20 40 60 80 100 120140 160 180 200obs.
/ exp.
0
1
2
W+jet background estimation
11
W+jet control region
• In a W+jet control region, divide tau candidates into
pass and fail identification.
• Define fake factor:
• Predict the number of W/Z+jet events:
fτ(pT, η) ≡Npass τ−ID(pT, η)
Nfail τ−ID(pT, η)
∣
∣
∣
∣
∣
∣
W–CR
NW/Z+jet(pT, η, x) = fτ(pT, η) · Nfail τ−IDW/Z+jet (pT, η, x) ,
x) = fτ(pT, η) ·(
Nfail τ−IDdata (pT, η, x) − Nfail τ−IDmultijet (pT, η, x) − N
fail τ−IDMC (pT, η, x)
)
.
Ryan Reece (Penn)
Double fake factor procedure
12
!"#$%&
!"#$%&#'()*+(
,-.#&%$.&#$-.
!"#$%&#'
/01'#"
,-.#&%2
$.&#$-.
!"#$%&#'()*+(
,-.#&%$.&#$-.
'(#&"#")&(*+,-(./*01$.%2"$%."1$
'(#&"#")&(*+,-(./*01$.%2"$%."1$'1$3$(#&"#")&(*456*01$.%2"$%."1$
7%//*.%8*96
7%//*&(:.1$*"/1
!"#$%&"'%()
7%//*&(:.1$*"/1
7%//*.%8*96
!"#$%$*+&,-%#.,
!"#$%&"'%()
!"#$%$*+&,-%#.,
• The multijet contamination is estimated from the rate of non-isolated leptons, in
both the signal region that passes tau ID, and the sample that fails.
• Then, the corrected number of tau candidates failing ID are weighted to predicted
the W+jet background.
• This way, the corrections are small at each step.
Ryan Reece (Penn)
Events
/ (
10 G
eV
)
0
500
1000
data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
) [GeV]miss
T(e, ETm
0 20 40 60 80 100 120140 160 180 200obs.
/ exp.
0
1
2
Ele
ctr
ons /
(0.0
2 m
m)
-110
10
310
510
710data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
(e) [mm]0
d
-1 -0.8-0.6-0.4-0.2 0 0.2 0.4 0.6 0.8 1obs.
/ exp.
0
1
2
Figure 28: (left) The distribution of the transverse mass of the combination of the selected electron and
the EmissT, mT(e, E
missT). in events with exactly one selected electron, no additional preselected
electrons or muons, and exactly one selected 1-prong tau. (right) The distribution of the
electron impact parameter, d0, in events with exactly one selected electron, no additional
preselected electrons or muons, and exactly one 1-prong tau candidate (without ID).
Control plots
13
Ryan Reece (Penn)
Cross-check: single fake factor method
14
!"#$%&
!"#$%&#'()*+(
,-.#&%$.&#$-.
!"#$%&#'
/01'#"
,-.#&%2
$.&#$-.
!"#$%&#'()*+(
,-.#&%$.&#$-.
'(#&"#")&(*+,-(./*01$.%2"$%."1$
'(#&"#")&(*+,-(./*01$.%2"$%."1$'1$3$(#&"#")&(*456*01$.%2"$%."1$
7%//*.%8*96
7%//*&(:.1$*"/1
!"#$%&"'%()
7%//*&(:.1$*"/1
7%//*.%8*96
!"#$%$*+&,-%#.,
!"#$%&"'%()
!"#$%$*+&,-%#.,
• Avoid issues of subtracting multijet from W/Z+jets background
• H→ττ uses a fake factor method covering all tau fakes from
W/Z+jets and multijet
• instead of estimating multijet independently with isolation fake factors
+
Ryan Reece (Penn)
Fake factors more quark-like at high-pT
15
1600 50 100 200 400 800 1600
Q
G
0%
100%
80%
60%
40%
20%
Z/W+1jet
pT Cut on All Jets (GeV)50 100 200 400 800 1600
G
Q
0%
100%
80%
60%
40%
20%
b+1jet
pT Cut on All Jets (GeV)
J. Gallicchio, M. Schwartz. “Pure Samples of Quark and Gluon Jets at the LHC”. arXiv:1104.1175
) [GeV]hτ(
Tp
0 50 100 150 200 250
)+je
ts
ν µ
→:
W(
τ f
0
0.02
0.04
0.06
0.08
0.1
Inclusive
)hτ, µOS(
)hτ, µSS(
ATLAS Internal
) [GeV]hτ(
Tp
0 50 100 150 200 250
: m
ultije
tτ
f
0
0.02
0.04
0.06
0.08
0.1
Inclusive
)hτ, µOS(
)hτ, µSS(
ATLAS Internal
our support note: ATLAS-CONF-2012-067
Ryan Reece (Penn)
Cross-check: single fake factor method
16
Events
/ (
10 G
eV
)
0
20
40
60
data 2011
! ! "Z W/Z+jets
multijet
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
Blin
ded
) [GeV]miss
T, E
h!(e, TM
0 50 100 150 200 250 300 350 400obs.
/ exp.
0
1
2
Events
/ (
10 G
eV
)
0
20
40
60
80 data 2011
! ! "Z
h!fake
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
Blin
ded
) [GeV]miss
T, E
h!(e, TM
0 50 100 150 200 250 300 350 400obs.
/ exp.
0
1
2
double fake factor single fake factor
• as expected, the single fake factor method over-
estimates in regions where the multijet contamination
is large
Ryan Reece (Penn)
Cross-check: single fake factor method
17
Events
/ (
20 G
eV
)
0
5
10
15
20 data 2011
! ! "Z
h!fake
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
Blin
ded
) [GeV]miss
T, E
h!(e, TM
200 250 300 350 400 450 500 550 600obs.
/ exp.
0
1
2
Events
/ (
20 G
eV
)
0
5
10
15
data 2011
! ! "Z W/Z+jets
multijet
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
Blin
ded
) [GeV]miss
T, E
h!(e, TM
200 250 300 350 400 450 500 550 600obs.
/ exp.
0
1
2
double fake factor single fake factor
double fake factor single fake factor
W/Z+jets multijet total fake τh
MT > 400 GeV 0.8(6) 0.3(3) 1.1(4) 1.3(4)
MT > 500 GeV 0.8(4) < 0.1 0.8(4) 0.9(4)
Ryan Reece (Penn)
Cross-check: single fake factor method
18
Events
/ (
50 G
eV
)
-110
1
10
210
310
410 data 2011
! ! "Z
h!fake
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
Blin
ded
) [GeV]miss
T, E
h!(e, TM
0 200 400 600 800 1000 1200 1400obs.
/ exp.
0
1
2
Events
/ (
50 G
eV
)
-110
1
10
210
310
410 data 2011
! ! "Z W/Z+jets
multijet
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
Blin
ded
) [GeV]miss
T, E
h!(e, TM
0 200 400 600 800 1000 1200 1400obs.
/ exp.
0
1
2
double fake factor single fake factor
double fake factor single fake factor
W/Z+jets multijet total fake τh
MT > 400 GeV 0.8(6) 0.3(3) 1.1(4) 1.3(4)
MT > 500 GeV 0.8(4) < 0.1 0.8(4) 0.9(4)
Ryan Reece (Penn)
eτh cut-flow
19
eτh µτh
data total SM Z → ττ W/Z+jets multijet Z → ee t t Wγ diboson single top Z′(1000)
one e, one τh 7295 7270(51) 1340(20) 3251(21) 766(22) 1280(35) 402(4) 68(6) 115(2) 49(2) 15.6(2)
|∆φ(e, τh)| > 2.7 3929 3933(43) 918(17) 1242(15) 556(18) 1020(32) 106(2) 28(4) 48(1) 15(1) 14.9(2)
opposite sign e and τh 3183 3322(40) 902(17) 1004(14) 279(13) 958(31) 99.9(2) 22(4) 44(1) 13.3(9) 14.6(2)
EmissT> 30 GeV 832 817(15) 158(6) 388(8) 39(6) 101(10) 85(2) 9(2) 26.8(9) 10.7(9) 12.8(2)
mT(e, EmissT
) < 50 GeV 263 298(10) 113(5) 101(5) 22(4) 41(6) 15.1(7) 1.3(9) 3.1(2) 1.9(3) 8.7(1)
MT > 200 GeV 46 59(4) 15.6(7) 31(2) 4(2) 0.03(2) 5.7(5) 0.6(6) 1.1(1) 0.8(2) 8.5(1)
MT > 300 GeV 14 14(2) 4.4(3) 6(1) 2(1) 0.02(2) 1.4(2) < 0.1 0.29(7) 0.11(8) 7.7(1)
MT > 400 GeV 4 3.0(8) 1.42(4) 0.8(6) 0.3(3) < 0.01 0.4(1) 0.14(5) < 0.1 6.5(1)
MT > 500 GeV 0 1.6(4) 0.57(2) 0.8(4) < 0.1 0.13(7) 0.06(3) 5.0(1)
MT > 600 GeV 0 0.5(2) 0.23(2) 0.2(2) 0.04(4) 0.02(2) 3.67(9)
Ryan Reece (Penn)
W+jets Z → ττ t t diboson Z′(1000)
expected events 0.8 0.6 0.1 0.1 5.0
total. uncert. 52 19 72 55 10
stat. uncert. 43 4 54 50 2
syst. uncert. 30 19 48 23 10
e efficiency - 1 1 1 1
e energy scale - 0 0 0 0
e energy resolution - 0 0 0 0
τh efficiency - 6 5 6 8
jet→ τh fake rate - 0 21 0 0
e→ τh fake rate - 0 23 17 0
jet energy scale - 14 24 11 6
jet energy resolution - 1 25 6 0
cluster energy scale - 0 5 0 1
luminosity - 2 2 2 2
theo. cross section - 11 10 7 -
τh fake factor 30 - - - -
eτh systematics
20
• statistics
• e→τh scale factors
• fake factor on
W+jets
• JES
(uncertainties in %)
Ryan Reece (Penn)
[GeV]Z’
m
500 1000 1500
) [p
b]
ττ
→Z’
(BR
×
) Z’
→pp
(σ
-310
-210
-110
1
ATLAS Preliminary-1
dt L = 4.6 fb∫ = 7 TeVs
channelhτµ
Expected limit
σ 1±Expected
σ 2±Expected
Observed limit
SSMZ’
[GeV]Z’
m
500 1000 1500
) [p
b]
ττ
→Z’
(BR
×
) Z’
→pp
(σ
-310
-210
-110
1
ATLAS Preliminary-1
dt L = 4.6 fb∫ = 7 TeVs
channelhτe
Expected limit
σ 1±Expected
σ 2±Expected
Observed limit
SSMZ’
µτh and eτh limits
21
• µτh: 1.06 (1.05) TeV
• eτh: 1.10 (1.02) TeV(BAT Bayesian limit)
Z’ SSM Exclusions: observed (expected) @ 95% CL
eτh channel has comparable
sensitivity to the µτh channel
eτh channelµτh channel
Ryan Reece (Penn)
eτh summary
22
• Adding the MET > 30 GeV helped decouple the
multijet contamination subtracted from the W+jet
estimate and further suppressed the Z→ee
background.
• Adding mT(e, MET) < 50 GeV suppressed the W+jet
background.
• Cross-checking with a single fake factor gave
consistent results and a conservative estimate of the
combined fake tau backgrounds.
µτh updates
Ryan Reece (Penn)
) [GeV]missTE, hτ, µ(tot
Tm
0 500 1000 1500
Events
/ 5
0 G
eV
-210
-110
1
10
210
310
410ATLAS Preliminary
-1dt L = 4.6 fb∫ = 7 TeVs
(b)Data 2011
ττ→*γ/Z
+jetsW
Multijet
µµ→Z
tt
Dibosonsingle top
ττ→(1000)Z’
Z’ → ττ → µτh overview
24
Event selection
Dominant systematics
miss
Ryan Reece (Penn)
µτh updates
25
• A few things have updated for the mu-had channel since the progress we made in the eτh channel.
• RootCore-00-00-33
• JetUncertainties-00-05-03
• |tau track eta| > 0.05 to update fiducial recommendation -- drops 1 event in the signal region
• systematics
• now using the new lumi uncertainty of 1.8% from ATLAS-CONF-2012-080
• The systematics were updated for µτh following the same strategy as for eτh of using a lower mass cut
(MT > 400 GeV) to evaluate the object-level systematic shifts, which gave more reasonable measure that
are statistically significant to a few percent instead of the previous statistically limit shifts. The main
changes were to things like JES and JER for diboson, which are now 2% instead of an insignificant 0.
Please note that the statistical errors still dwarf these object-level systematics.
• updated all the plots to now have systematics bands in the ratios
• In recent drafts the plots and tables were not yet updated for the change in luminosity; only the
final limit config was shifted. Plots and tables have been updated in
ZprimeTauTau2011COM-03-00-04
• no significant changes in background estimates.
Ryan Reece (Penn)
µτh updated limit
26
[GeV]Z’
m
500 1000 1500
[pb]
Bσ
-310
-210
-110
1
ATLAS Preliminary-1
dt L = 4.7 fb∫ = 7 TeVs
channelhτµ
Expected limit
σ 1±Expected
σ 2±Expected
Observed limit
SSMZ’
Old (CONF) New (Paper)
[GeV]Z’
m
500 1000 1500
) [p
b]
ττ
→Z’
(BR
×
) Z’
→pp
(σ
-310
-210
-110
1
ATLAS Preliminary-1
dt L = 4.6 fb∫ = 7 TeVs
channelhτµ
Expected limit
σ 1±Expected
σ 2±Expected
Observed limit
SSMZ’
• old: 1.00 (1.05) TeV
• new: 1.06 (1.05) TeV
Z’ SSM Exclusions: observed (expected) @ 95% CLMore toys giving higher resolution sampling of the bands in new limits, but expected and observed values do not change.
τhτh updatesWill Davey
Changes to τhτh analysis
• The only changes specific to the Z’→τhτh analysis involve the multijet estimate:
1. Included shape systematic uncertainty
2. Increased minimum of fit range to better avoid threshold effects
• The lumi central value and uncertainty was also updated (as for all channels)
• These changes have negligible effect on the final limit
28
Recap of method
29
Norm
aliz
ed e
vent
rate
/ 1
0 G
eV
0
0.05
0.1
0.15
0.2
0.25
0.3
ATLAS Preliminary-1
dt L = 4.7 fb! = 7 TeVs
Opposite Charge (OC)
Same Charge (SC)
) [GeV]missTE, h", h"(TM
200 250 300 350
OC
/ S
C
0.51
1.5
Fit same-sign (SS) data with dijet function:
Events
/ 4
0 G
eV
-110
1
10
210
310
410ATLAS Preliminary
-1dt L = 4.7 fb! = 7 TeVs
Data 2011Multijet
""#*$/Z
%"#W
Others""#(1250)Z’
) [GeV]missTE, h", h"(TM
500 1000 1500obs.
/ exp.
0.51
1.5
• OS/SS shapes agree well
• normalize in OS sideband with 160 < MT < 200 GeV
Changes
30
• raise the fit minimum to 200 GeV
• add additional shape systematic by comparing the fit result with a crystal ball function
[GeV]TM
0 200 400 600 800 100010!
Events
/ 2
0 G
eV
-210
-110
1
10
210
310
410
ATLAS Internal
QCD
Chi2/NDF: 7.60 / 11
700) = 0!(DataN
700) = 0.2!(FitN
old fit min too close to turn-on
region(not modelled by fitting functions)
new fit min
old fit min
[GeV]TM
0 200 400 600 800 1000 1200 1400 1600 1800 200010×
Events
-410
-310
-210
-110
1
10
210
310dijet function
crystal ball
Fit to SS region using nominal and alternate fit functions
high-mass counting region
Summary of fitting changes
• For Z’ signal masses of 500 and 625 GeV:
• nothing changed
• For signal masses mZ’ ≥ 750 GeV:
• Changed minimum of fit range from 160 → 200
• Changed side-band from [160,200] → [200,250]
31
) [GeV]missTE, h
!, h!(tot
Tm
200 400 600 800 1000
Events
/ 2
0 G
eV
-210
-110
1
10
210
310 ATLAS Preliminary
-1dt L = 4.7 fb" = 7 TeVs
/ NDF: 9.43 / 132#
700) = 0$TM(DataN
700) = 0.3$TM(FitN
) [GeV]missTE, h
!, h!(tot
Tm
200 400 600 800 1000
Events
/ 2
0 G
eV
-210
-110
1
10
210
310 ATLAS Preliminary
-1dt L = 4.6 fb" = 7 TeVs
/ NDF: 7.60 / 112#
700) = 0$TM(DataN
700) = 0.2$TM(FitN
Old New
Change in final estimates
32
Mass Point [GeV] 500 625 750-875 1000-1125 ≥1250
estimate 28.5 12.7 3.2 0.59 0.36
shape stat. uncertainty+3.1 -3.4
+2.2-2.5
+0.65-0.75
+0.16-0.20
+0.11-0.13
Mass Point [GeV] 500 625 750-875 1000-1125 ≥1250
estimate 28.5 12.7 2.7 0.42 0.24
shape stat. uncertainty +3.1-3.4
+2.2-2.5
+0.96-0.72
+0.30-0.16
+0.20-0.10
shape sys. uncertainty 1.1 0.9 0.02 0.02 0.02
norm. stat. uncertainty 0.7 0.6 0.13 0.02 0.01
Old multijet estimate
New multijet estimate
New multijet estimate has a slightly lower central value and slightly larger absolute uncertainty
[GeV]Z’
m
500 1000 1500
) [p
b]
ττ
→Z’
(BR
×
) Z’
→pp
(σ
-310
-210
-110
1
ATLAS Preliminary-1
dt L = 4.6 fb∫ = 7 TeVs
channelhτhτ
Expected limit
σ 1±Expected
σ 2±Expected
Observed limit
SSMZ’
τhτh updated limits
Changes do not significantly effect the limit
33
[GeV]Z’
m
500 1000 1500
[pb]
B!
-310
-210
-110
1
ATLAS Preliminary-1
dt L = 4.7 fb" = 7 TeVs
channelh#h#
Expected limit
! 1!Expected
! 2!Expected
Observed limit
SSMZ’
Old (CONF) New (Paper)
More toys giving higher resolution sampling of the bands in new limits, but expected and observed values do not change.
1.25 (1.35) TeV
Ryan Reece (Penn)
[GeV]Z’
m
500 1000 1500
) [p
b]
ττ
→Z’
(BR
×
) Z’
→pp
(σ
-310
-210
-110
1
ATLAS Preliminary-1
dt L = 4.6 fb∫ = 7 TeVs
(a)
µe
hτe /hτµ
hτhτ
comb.
Observed limitsExpected limits
SSMZ’
[GeV]Z’
m
500 1000 1500
) [p
b]
ττ
→Z’
(BR
×
) Z’
→pp
(σ
-310
-210
-110
1
ATLAS Preliminary-1
dt L = 4.6 fb∫ = 7 TeVs
Combined
(b) Expected limit
σ 1±Expected
σ 2±Expected
Observed limit
SSMZ’
Combined limit
34
• τhτh: 1.25 (1.35) TeV
• µτh: 1.06 (1.05) TeV
• eτh: 1.10 (1.02) TeV
• eµ: 0.73 (0.79) TeV(BAT Bayesian limit)
Z’ SSM Exclusions: observed (expected) @ 95% CL
Ryan Reece (Penn)
Final mass plots
35
) [GeV]missTE, hτ, hτ(
totTm
0 500 1000 1500
Events
/ 5
0 G
eV
-210
-110
1
10
210
310
410
ATLAS Preliminary-1
dt L = 4.6 fb∫ = 7 TeVs
(a)Data 2011
Multijetττ→*γ/Z
ντ→W
Others
ττ→(1250)Z’
) [GeV]missTE, hτ, µ(tot
Tm
0 500 1000 1500
Events
/ 5
0 G
eV
-210
-110
1
10
210
310
410ATLAS Preliminary
-1dt L = 4.6 fb∫ = 7 TeVs
(b)Data 2011
ττ→*γ/Z
+jetsW
Multijet
µµ→Z
tt
Dibosonsingle top
ττ→(1000)Z’
) [GeV]missTE, hτ, e(
totTm
0 500 1000 1500
Events
/ 5
0 G
eV
-210
-110
1
10
210
310
410ATLAS Preliminary
-1dt L = 4.6 fb∫ = 7 TeVs
(c)Data 2011
ττ→*γ/Z
+jetsZ/W
Multijet
ee→Z
tt
Dibosonsingle top
ττ→(1000)Z’
) [GeV]missTE, µ, e(tot
Tm
0 500 1000 1500
Events
/ 5
0 G
eV
-210
-110
1
10
210
310
410ATLAS Preliminary
-1dt L = 4.6 fb∫ = 7 TeVs
(d)Data 2011
Dibosonττ→*γ/Z
tt+jetsW
µµ→Zττ→(750)Z’
Ryan Reece (Penn)
Conclusions
36
• The analysis of the eτh channel has been completed
and combined with the previous three channels.
• The other three channels have only had minor updates.
• The combined limit on SSM Z’ has increased from 1.3
(1.4) to 1.4 (1.4) TeV, observed (expected).
• We are competitive with CMS, who published the limit
of 1.4 (1.1) TeV
http://arxiv.org/abs/1206.1725
• We have a publication first draft under the review of
our ed-board.
Ryan Reece (Penn)
Documents
• Supporting document
ATL-COM-PHYS-2012-394
https://cdsweb.cern.ch/record/1439018
• ICHEP 2012 Conference note
ATLAS-CONF-2012-067
https://cdsweb.cern.ch/record/1460263
• Publication draft for PLB to be reviewed by ATLAS
ATL-COM-PHYS-2012-1216
https://cdsweb.cern.ch/record/1472945/
37
Back up
Ryan Reece (Penn)
Summary tables
39
Uncertainty [%] Signal Background(τhτh / µτh / eτh / eµ)
Stat. uncertainty 1 / 2 / 2 / 3 5 / 20 / 23 / 7Eff. and fake rate 16 / 10 / 8 / 1 12 / 16 / 4 / 3Energy scale and res. 5 / 7 / 6 / 2 +22
−11 / 3 / 8 / 5Theory cross section – / – / – / – 9 / 4 / 4 / 5Luminosity 2 / 2 / 2 / 2 1 / 1 / 1 / 2Data-driven methods – / – / – / – +21
−11 / 6 / 16 / –
Table 2: Uncertainties on the estimated signal and total back-
ground contributions in percent for each channel. The following sig-
nal masses are used: 1250 GeV for τhτh; 1000 GeV for µτh and eτh;
and 750 GeV for eµ. A dash denotes that the uncertainty is not ap-
plicable. The statistical uncertainty corresponds to the uncertainty
due to limited sample size of the MC and control regions.
Ryan Reece (Penn)
Summary tables
40
τhτh µτh eτh eµZ/γ∗
→ ττ 0.73±0.23 0.36±0.06 0.6±0.1 0.55±0.07W+jets 0 0.3±0.2 0.8±0.4 0.33±0.10Z(→ ℓℓ)+jets 0 0 0 0.06±0.02tt̄ 0 0.33±0.15 0.13±0.09 1.0±0.2Diboson 0 0.23±0.07 0.06±0.03 1.7±0.2Single top 0 0.2±0.2 0 0Multijet 0.24±0.15 0 0 0Total expected background 1.0±0.3 1.4±0.4 1.6±0.5 3.6±0.4Events observed 2 1 0 5Expected signal events 6.3±1.1 5.5±0.7 5.0±0.5 6.7±0.3Signal efficiency (%) 4.3 1.1 1.0 0.4
Table 3: Number of expected and observed events in selected signal regions for each analysis channel. The expected contribution from the
signal in each channel is calculated for the mass point closest to the Z′
SSMexclusion limit: 1250 GeV for τhτh; 1000 GeV for µτh and eτh;
and 750 GeV for eµ. The signal regions corresponding to these mass points are used. The total uncertainties on each estimated contribution
are shown. The signal efficiency denotes the expected number of signal events divided by the product of the Z′
SSMcross section and the
integrated luminosity.
Ryan Reece (Penn)
Log-x binning
41
Events
-110
1
10
210
310
410
510 data 2011
τ τ →Z W+jets
multijet
µ µ →Z
tt dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
) [GeV]miss
T, E
hτ, µ(TM
2103
10
obs.
/ exp.
0
1
2
Events
-110
1
10
210
310
410 data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
) [GeV]miss
T, E
hτ(e, TM
2103
10
obs.
/ exp.
0
1
2
eτh channelµτh channel
Ryan Reece (Penn)
pT(e)
42
Ele
ctr
ons /
(5 G
eV
)
0
50
100
data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
(e) [GeV]T
p
0 20 40 60 80 100 120 140obs.
/ exp.
0
1
2
Events
/ (
20 G
eV
)
-110
1
10
210
310
410data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
(e) [GeV]T
p
0 100 200 300 400 500 600obs.
/ exp.
0
1
2
Figure 23: The distribution of the transverse momentum of the selected electron. These plots include the
requirements of: exactly one selected electron, no additional preselected electrons or muons,
exactly one selected 1-prong tau, |∆φ(e, τh)| > 2.7, opposite sign e and τh, EmissT> 30 GeV,
and mT(e, EmissT) < 50 GeV.
Ryan Reece (Penn)
pT(τh)
43
Ta
u C
an
did
ate
s / (
5 G
eV
)
0
20
40
60
80 data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
) [GeV]hτ(
Tp
0 20 40 60 80 100 120 140obs.
/ exp.
0
1
2
Events
/ (
20 G
eV
)
-110
1
10
210
310
410data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
syst.⊕stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
) [GeV]hτ(
Tp
0 100 200 300 400 500 600obs.
/ exp.
0
1
2
Figure 24: The distribution of the transverse momentum of the selected hadronic tau. These plots
include the requirements of: exactly one selected electron, no additional preselected elec-
trons or muons, exactly one selected 1-prong tau, |∆φ(e, τh)| > 2.7, opposite sign e and τh,
EmissT> 30 GeV, and mT(e, E
missT) < 50 GeV.
Ryan Reece (Penn)
Fake factors more quark-like at high-pT
44
Tau C
andid
ate
s / (
5 G
eV
)
0
20
40
60
80 data 2011
! ! "Z
h!fake
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
) [GeV]h!(
Tp
0 20 40 60 80 100 120 140obs.
/ exp.
0
1
2
Events
/ (
10 G
eV
)
0
10
20
30 data 2011
! ! "Z
h!fake
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
Blin
ded
) [GeV]h!(
Tp
100 150 200 250 300obs.
/ exp.
0
1
2
• Because the quark-fraction increases with pT, the single
fake factor method agrees better at high-pT.
Ryan Reece (Penn)
Events
/ (
10 G
eV
)
0
50
100
data 2011
! ! "Z
h!fake
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
) [GeV]miss
T(e, ETm
0 20 40 60 80 100 120140 160 180 200obs.
/ exp.
0
1
2
Cross-check: single fake factor method
45
Events
/ (
10 G
eV
)
0
50
100
data 2011
! ! "Z W/Z+jets
multijet
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
) [GeV]miss
T(e, ETm
0 20 40 60 80 100 120140 160 180 200obs.
/ exp.
0
1
2
• as expected, the single fake factor method over-
estimates in regions where the multijet contamination
is large
double fake factor single fake factor
Ryan Reece (Penn)
Object preselection
46
Muons
StoreGate key: StacoMuonCollection
Tau D3PD prefix: mu staco *
pT > 10 GeV
|η| < 2.5
mu staco loose == 1
Require a B-layer hit if expected (expectBLayerHit == 0 or nBLHits > 0)
N(pixel hits) + N(pixel dead) ≥ 2
N(SCT hits) + N(SCT dead) ≥ 6
N(pixel holes) + N(SCT holes) ≤ 2
TRT quality cuts:if abs(eta) < 1.9:
if not ( (nTRTHits + nTRTOutliers > 5) and \
(nTRTOutliers < 0.9*(nTRTHits + nTRTOutliers))):
return False
elif (nTRTHits + nTRTOutliers > 5):
if not (nTRTOutliers < 0.9*(nTRTHits + nTRTOutliers)):
return False
return True
Electrons
Tau D3PD prefix: el *
pT > 15 GeV
(|η| < 1.37) or (1.52 < |η| < 2.47)
el author in (1, 3)
el mediumPP == 1
Require a B-layer hit if expected (expectBLayerHit == 0 or nBLHits > 0)
Taus
Ryan Reece (Penn)
Object preselection
47
Require a B-layer hit if expected ( )
Taus
StoreGate key: TauRecContainer
Tau D3PD prefix: tau *
pT > 25 GeV
(|η| < 1.37) or (1.52 < |η| < 2.47)
tau author in (1, 3)
tau numTrack > 0
Remove candidates overlapping with preselected electrons or muons within ∆R < 0.2
Jets
StoreGate key: AntiKt4LCTopoJets
Tau D3PD prefix: jet *
pT > 25 GeV
|η| < 4.5
|JVF| > 0.75 for jets with |η| < 2.4
Remove candidates overlapping with preselected electrons or selected taus within ∆R < 0.2
Ryan Reece (Penn)
Fake factors
48
Electron isolation Tau identification
(e) [GeV]T
p
0 50 100 150 200 250 300 350 400
: m
ultije
te-iso
f
0
0.2
0.4
0.6
0.8
1
1.2
1.4Inclusive
Barrel
Endcap
ATLAS Internal
) [GeV]h!(
Tp
0 50 100 150 200 250
)h!
)+je
ts,
OS
(e,
" e
#
: W
(!
f
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
Inclusive
Barrel (inner)
Barrel (outer)
Endcap
ATLAS Internal
Ryan Reece (Penn)
Electron isolation
49
•
corrected etcone20/pT < 5% if pT < 100 GeV
corrected etcone20 < 5 GeV if pT > 100 GeV,
• ptcone40/pT < 5%,
•
corrected etcone20/pT > 8% if pT < 100 GeV
corrected etcone20 > 8 GeV if pT > 100 GeV,
• or ptcone40/pT > 6%.
isolated anti-isolated
track isolation calorimeter isolation
Gap between isolated and anti-isolated makes
data-driven multijet more pure (less corrected)
Tptcone40 / p
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Ele
ctr
ons /
0.0
1
-110
1
10
210
310
410
510
610
710data 2011
! ! "Z
# e "W
# ! "W
e e"Z
tt
$W
diboson
single top
syst.%stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal -1dt L = 4.6 fb&
Tcorrected Etcone20 / p
-0.1 0 0.1 0.2 0.3 0.4 0.5
Ele
ctr
ons /
0.0
1
-110
1
10
210
310
410
510
610 data 2011
! ! "Z
# e "W
# ! "W
e e"Z
tt
$W
diboson
single top
syst.%stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal -1dt L = 4.6 fb&
Ryan Reece (Penn)
Electron veto
50
• above are the scale factors
and errors for the
EleBDTMedium e-veto
• These SF have huge ≈100%
uncertainties, and we’ve also
previously shown that this
veto is inefficient at high pT.
) [GeV]hτ(truth
Tp
0 100 200 300 400 500 600
effic
ien
cy
0
0.2
0.4
0.6
0.8
1
EleBDTLoose
EleBDTMedium
EleBDTTight
MuonVeto
ATLAS Internal
|η| < 1.37 1.37 < |η| < 1.52 1.52 < |η| < 2.0 2.0 < |η|
BDT medium e-veto 1.64(0.81) 1.0(1.0) 0.71(0.63) 2.90(1.42)
BDT loose e-veto 1.21(0.30) 0.96(0.46) 0.59(0.21) 1.76(0.55)
Ryan Reece (Penn)
Multijets background
51
• Loosened tau ID requirement
in the anti-isolated lepton +
tau data sample used to
model the multijet
background.
• Sample with tau ID is already
multijet dominated.
Loosening tau ID improves
stats.
• Shapes are statistically
consistent. Inclusive shape
scaled to the prediction with
medium tau ID to give the
estimate in the tail.
) [GeV]miss
T, Ehτ(e, TM
0 200 400 600 800 1000 1200 1400
Events
/ (
50 G
eV
)-310
-210
-110
1
10
210
310
410
510
610
JetBDTSigMedium
JetBDTSigLoose
Inclusive
SM background
multijetATLAS Internal -1dt L = 4.6 fb∫
Ryan Reece (Penn)
Z → ee + jets background
52
• Categorize Monte Carlo events by electron or jet faking tau.
• Loosen electron veto in Monte Carlo sample matched to electron fakes.
• Shapes are consistent, and only driven by the Z → ee kinematics.
• Z → ee with e-faking tau is negligible
• Z → ee + jet-fake covered with the data-driven W/Z+jet tau fake factor method.
) [GeV]miss
T, Ehτ(e, TM
0 200 400 600 800 1000 1200 1400
Events
/ (
20 G
eV
)
-310
-210
-110
1
10
210
310
410
510
610
e e-fake→Z
e e + jet-fake→Z
SM background
e e→Z ATLAS Internal -1dt L = 4.6 fb∫
) [GeV]miss
T, Ehτ(e, TM
0 200 400 600 800 1000 1200 1400
Events
/ (
20 G
eV
)
-310
-210
-110
1
10
210
310
410
510
610
EleBDTMedium
EleBDTLoose
BDTEleScore > 0.3
SM background
e e→Z ATLAS Internal -1dt L = 4.6 fb∫
Ryan Reece (Penn)
High-pT tau efficiency systematic
53
) [GeV]τ(probe-T
p
0 200 400 600 800 1000
Scale
Facto
r
0
0.2
0.4
0.6
0.8
1
1.2
1.4 / ndf 2χ 4.2 / 6
p0 0.05± 0.81
p1 0.00016± 0.00015
/ ndf 2χ 4.2 / 6
p0 0.05± 0.81
p1 0.00016± 0.00015
/ ndf 2χ 4.2 / 6
p0 0.05± 0.81
p1 0.00016± 0.00015
BDT60• The dominant systematic uncertainty for
the Z’ signal and the Z→ττ background.
• Low pT uncertainty of 4% taken from the
Tau WG blessed Z→ττ tag-and-probe.
• No high-pT control sample of true taus.
• Assume mis-modeling comes from either:
1. tau kinematics (TAUOLA)
2. detector response to high-pT pions←dominant
• Instead of using true taus, measure the trend in the scale
factor for fakes from dijet events.
• pT ≤ 100 GeV: ∆ε = 4% (taken from the Z → ττ measurement)
• pT > 100 GeV: ∆ε = 4+ 0.016(pT − 100)%, with pT in GeV (taken from the dijets measurement).
Ryan Reece (Penn)
Summary of µτh/eτh channel differences
54
Events
/ (
5 G
eV
)
0
500
1000data 2011
! ! "Z W/Z+jets
multijet
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
[GeV]missTE
0 20 40 60 80 100 120 140obs.
/ exp.
0
1
2
Events
/ (
10 G
eV
)
0
50
100
data 2011
! ! "Z W/Z+jets
multijet
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
) [GeV]miss
T(e, ETm
0 20 40 60 80 100 120140 160 180 200obs.
/ exp.
0
1
2
after MET > 30 GeV and OS
• electron isolation:
• etcone20 / pt < 5% if pt < 100 GeV else etcone20 < 5 GeV
• ptcone40 < 5%
• MET > 30 GeV
• mT(e,MET) < 50 GeV
To further suppress multijet
and Z→ee backgrounds.
Ryan Reece (Penn)
µτh cut-flow
55
data total SM Z → ττ W+jets multijet Z → µµ t t diboson single top Z′(1000)
one µ, one τh 11605 11653(47) 3867(35) 4913(21) 583(5) 897(21) 963(7) 323(4) 108(3) 18.7(2)
|∆φ(µ, τh)| > 2.7 6061 5941(38) 2829(31) 1811(14) 403(4) 533(17) 221(3) 117(2) 27(1) 17.9(2)
opposite sign µ and τh 5320 5242(37) 2791(31) 1446(13) 213(3) 452(16) 208(3) 107(2) 24(1) 17.6(2)
MT > 200 GeV 229 263(5) 30.0(5) 114(4) 5.5(4) 5(1) 76(2) 22.3(8) 9.4(9) 16.5(2)
MT > 300 GeV 31 53(2) 7.3(2) 18(2) 0.6(1) 0.3(2) 17.6(9) 6.7(6) 3.0(5) 14.1(2)
MT > 400 GeV 13 15(1) 2.31(5) 5.0(8) 0.15(7) 0.1(1) 4.5(4) 1.7(2) 0.9(3) 11.0(2)
MT > 500 GeV 1 4.5(5) 0.82(3) 1.6(5) 0.02(2) < 0.1 1.2(2) 0.6(1) 0.3(2) 8.0(1)
MT > 600 GeV 1 1.4(3) 0.36(2) 0.3(2) < 0.01 0.3(1) 0.23(7) 0.2(1) 5.5(1)
MT > 700 GeV 1 0.5(1) 0.17(1) 0.07(7) 0.03(3) 0.13(5) 0.1(1) 3.38(9)
Ryan Reece (Penn)
µτh systematics
56
W+jets Z → ττ t t diboson single top Z′(1000)
expected events 0.3 0.4 0.3 0.2 0.2 5.5
total. uncert. 77 18 46 33 95 13
stat. uncert. 71 5 35 29 74 2
syst. uncert. 30 17 30 15 59 12
µ efficiency - 0 0 0 0 2
µ pT resolution ID - 0 0 2 0 2
µ pT resolution MS - 0 0 1 0 2
τh efficiency - 6 5 5 0 10
jet→ τh fake rate - 0 11 0 0 0
e→ τh fake rate - 0 25 11 58 0
jet energy scale - 11 2 2 0 6
jet energy resolution - 1 0 2 0 2
cluster energy scale - 0 1 2 0 2
luminosity - 2 2 2 2 2
theo. cross section - 11 10 7 13 -
τh fake factor 30 - - - - -
• statistics
• e→τh SF
• fake factor on
W+jets
(uncertainties in %)
[GeV]Z’
m
500 1000 1500
) [p
b]
!!
"Z’
(BR
!
) Z’
"pp
(#
-310
-210
-110
1
ATLAS Preliminary-1
dt L = 4.6 fb$ = 7 TeVs
channelh!h!
Expected limit
# 1"Expected
# 2"Expected
Observed limit
SSMZ’
[GeV]Z’
m
500 1000 1500
) [p
b]
!!
"Z’
(BR
!
) Z’
"pp
(#
-310
-210
-110
1
ATLAS Preliminary-1
dt L = 4.6 fb$ = 7 TeVs
channelh!h!
Expected limit
# 1"Expected
# 2"Expected
Observed limit
SSMZ’
95% mu upper limit0 2 4 6 8 10 12 14 16 18 20
0
2
4
6
8
10
12
14
16
18
MuUpper95CLhlow_mu_UL
Entries 1000Mean 4.946
RMS 1.588
MuUpper95CL
95% mu upper limit0 2 4 6 8 10 12 14 16 18 20
0
100
200
300
400
500
600
MuUpper95CLhmed_mu_UL
Entries 5000Mean 4.985
RMS 1.59
MuUpper95CL
Running bands at high precision• Band Calculation:
• Generate pseudo-data by randomly sampling ‘N observed’ from background only model
• For each ‘N observed’ calculate the upper limit on the signal strength
• The quantiles of the resulting distribution for the mu upper limit give you the median and ±1σ ±2σ bands
• Until now, was running mu upper limit calculation at low precision for the bands to save time, which artificially broadens the peaks in the output distribution
• Running at high precision can cause the -1σ and -2σ bands to overlap for low expected bkg.
57
pseudo N Observed0 1 2 3 4 5 6 7 8 9 10
95%
mu u
pper
limit
0
2
4
6
8
10
12
14
16
18
20
MuUpper95CL:NObshlow_mu_UL_nobs
Entries 1000Mean x 1.039Mean y 4.946
RMS x 1.047RMS y 1.588
MuUpper95CL:NObs
pseudo N Observed0 1 2 3 4 5 6 7 8 9 10
95%
mu u
pper
limit
0
2
4
6
8
10
12
14
16
18
20
MuUpper95CL:NObshmed_mu_UL_nobs
Entries 5000Mean x 1.064Mean y 4.985
RMS x 1.066RMS y 1.59
MuUpper95CL:NObs
Low Precision High Precision
mZ’=1750 GeVmZ’=1750 GeV
mZ’=1750 GeVmZ’=1750 GeV
[GeV]Z’
m
500 1000 1500
) [p
b]
!!
"Z’
(BR
!
) Z’
"pp
(#
-310
-210
-110
1
ATLAS Preliminary-1
dt L = 4.6 fb$ = 7 TeVs
channelh!h!
Expected limit
# 1"Expected
# 2"Expected
Observed limit
SSMZ’
[GeV]Z’
m
500 1000 1500
) [p
b]
!!
"Z’
(BR
!
) Z’
"pp
(#
-310
-210
-110
1
ATLAS Preliminary-1
dt L = 4.6 fb$ = 7 TeVs
channelh!h!
Expected limit
# 1"Expected
# 2"Expected
Observed limit
SSMZ’
95% mu upper limit0 2 4 6 8 10 12 14 16 18 20
0
2
4
6
8
10
12
14
16
18
MuUpper95CLhlow_mu_UL
Entries 1000Mean 4.946
RMS 1.588
MuUpper95CL
95% mu upper limit0 2 4 6 8 10 12 14 16 18 20
0
100
200
300
400
500
600
MuUpper95CLhmed_mu_UL
Entries 5000Mean 4.985
RMS 1.59
MuUpper95CL
Running bands at high precision• Band Calculation:
• Generate pseudo-data by randomly sampling ‘N observed’ from background only model
• For each ‘N observed’ calculate the upper limit on the signal strength
• The quantiles of the resulting distribution for the mu upper limit give you the median and ±1σ ±2σ bands
• Until now, was running mu upper limit calculation at low precision for the bands to save time, which artificially broadens the peaks in the output distribution
• Running at high precision can cause the -1σ and -2σ bands to overlap for low expected bkg.
58
pseudo N Observed0 1 2 3 4 5 6 7 8 9 10
95%
mu u
pper
limit
0
2
4
6
8
10
12
14
16
18
20
MuUpper95CL:NObshlow_mu_UL_nobs
Entries 1000Mean x 1.039Mean y 4.946
RMS x 1.047RMS y 1.588
MuUpper95CL:NObs
pseudo N Observed0 1 2 3 4 5 6 7 8 9 10
95%
mu u
pper
limit
0
2
4
6
8
10
12
14
16
18
20
MuUpper95CL:NObshmed_mu_UL_nobs
Entries 5000Mean x 1.064Mean y 4.985
RMS x 1.066RMS y 1.59
MuUpper95CL:NObs
Low Precision High Precision
-2σ-1σ+1σ
µ +2σ-2σ-1σ +1σ
µ
+2σ
mZ’=1750 GeVmZ’=1750 GeV
Recap of method1. Model multijet MT shape using same-sign
events:
• 99.2% multijet purity in SS events
• OS/SS shapes agree well (top figure)
• Fit with ‘dijet’ function (bottom figure):
• good χ2/NDF
• tail integral matches observed SS events well for any choice of MT threshold
• shape of high-mass tail verified using dijet MC enriched in high-mass events
• statistical uncertainty 10-40% depending on MT threshold used in signal region, determined using pseudo-experiments
• only model down to 160 GeV to avoid threshold effects (will come back to this)
59
Norm
aliz
ed e
vent
rate
/ 1
0 G
eV
0
0.05
0.1
0.15
0.2
0.25
0.3
ATLAS Preliminary-1
dt L = 4.7 fb! = 7 TeVs
Opposite Charge (OC)
Same Charge (SC)
) [GeV]missTE, h", h"(TM
200 250 300 350
OC
/ S
C
0.51
1.5
) [GeV]missTE, h
!, h!(TM
200 400 600 800 1000
Events
/ 2
0 G
eV
-210
-110
1
10
210
310 ATLAS Preliminary
-1dt L = 4.7 fb" = 7 TeVs
/ NDF: 9.43 / 132#
650) = 1$TM(DataN
650) = 0.5$TM(FitN
OS/SS shapes agree well
Recap of method
2. Normalise shape to data in low-mass side-band:
• 160 < MT < 200 GeV (first bin of MT plot)
• 6% contamination from other backgrounds subtracted
• negligible signal contamination for all signal mass hypotheses
• negligible statistical uncertainty
60
Events
/ 4
0 G
eV
-110
1
10
210
310
410ATLAS Preliminary
-1dt L = 4.7 fb! = 7 TeVs
Data 2011Multijet
""#*$/Z
%"#W
Others""#(1250)Z’
) [GeV]missTE, h", h"(TM
500 1000 1500obs.
/ exp.
0.51
1.5
Issue that was addressed for paper
• For the CONF there was no estimate of the systematic uncertainty due to the choice of the fitting function.
• This was a request during the CONF final approval, but there was not enough time to address it at that time
• The issue has been addressed for the paper.
61
Change 1
• Include a new systematic uncertainty that accounts for the choice of the fitting function
• Method: Find another function that models multijet events well, fit it to the SS data, and take the deviation of the estimate w.r.t the nominal fitting function as an additional uncertainty
• Crystal Ball chosen, as it was shown to model multijets well (see backup slides)
• Tail fractions in signal region 36% higher when fitting with CB function
62
[GeV]TM
0 200 400 600 800 1000 1200 1400 1600 1800 200010×
Events
-410
-310
-210
-110
1
10
210
310dijet function
crystal ball
Fit to SS region using nominal and alternate fit functions
high-mass counting region
Change 2
• Raise the minimum of the fitting range:
• The large discrepancy between the nominal and CB fit was unexpected because CB was shown to be an unbiased estimator of the tail-fraction using pseudo-experiments (see backup)
• After further investigation it turns out that the MT>160 GeV range used for the fit was not adequately avoiding the turn on region (see backup)
• Solution:
• raise the fit minimum to 200 GeV
63
[GeV]TM
0 200 400 600 800 100010!
Events
/ 2
0 G
eV
-210
-110
1
10
210
310
410
ATLAS Internal
QCD
Chi2/NDF: 7.60 / 11
700) = 0!(DataN
700) = 0.2!(FitN
old fit min too close to turn-on
region(not modelled by fitting functions)
new fit min
old fit min
Consequences of raising fit min
• Raising the fit minimum from 160 to 200 GeV has the following effects:
• reduces bias in the multijet estimate
• increases the statistical uncertainty of the fit
• the side-band must be shifted from [160,200] to [200,250], which means:
• increased statistical uncertainty from normalisation (now non-negligible)
• increased signal contamination for very low mass-points:
• as a result, the old fit range is still used for the two low-mass signal points because for these points there is non-negligible signal contamination in the 200-250 side-band
64
τhτh channel summary
• Lower bound of the fit range for the high-mass points was increased which gives a more unbiased central value for the multijet estimate.
• Included new systematic uncertainty to account for shape systematics.
• Despite taking further systematic effects into account, the multijet estimation doesn't vary much and it doesn't affect the limit.
65
Determining alternate fitting functions
Key: We only want to accept functions that accurately model dijets over the full MT range:
1. Fit dijet function to high-stat dijet MC
2. Generate high-stat pseudo-data from this model
3. Check which functions can model the pseudo-data over the full mass range
Only other suitable function: Crystal Ball
66
[GeV]TM
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Events
/ 1
0 G
eV
-410
-310
-210
-110
1
10
210
310 10±a = 7
0.38±b = -0.548
ATLAS Internal
Simulation
Chi2/NDF = 379.1 / 157
MT
400 600 800 1000 1200 1400 1600 1800 200010!
Events
/ (
16000 )
-110
1
10
210
310
410
510 0.0059"acb = -0.26035
8021"mcb = 196780
0.070"ncb = 9.137
229"scb = 10093
sig frac: 0.00255
/NDF: 1.752!
ATLAS Internal
MT
400 600 800 1000 1200 1400 1600 1800 200010!
Events
/ (
16000 )
-110
1
10
210
310
410
510 77"ml = 418075
24"sl = 16763
sig frac: 0.023
/NDF: 2.21e+032!
ATLAS Internal
MT
400 600 800 1000 1200 1400 1600 1800 200010!
Events
/ (
16000 )
-110
1
10
210
310
410
510 0.0055"a = -6.76913
sig frac: 0.00489
/NDF: 52.42!
ATLAS Internal
Step 1
Step 2 and 3
✔ ✖ ✖
Note: fit range of 400-2000 used to avoid region effected by pT thresholds
Fit range
many other models were also tried....
Low stats bias (Yann’s test)• Method:
• Generate pseudo-data from the di-jet function fitted to MC (top right figure previous slide)
• Refit it with di-jet function and Crystal Ball
• Perform pseudo-experiments in batches with decreasing numbers of events per trial to see if either di-jet or CB become biased with low stats in the tail
• Result: Both produce unbiased tail fractions
67
Number of events per trial
310 410
510
610
Ave
rag
e f
it f
ractio
n in
sig
na
l re
gio
n
0.001
0.0015
0.002
0.0025
0.003Nominal Tail Fraction
Number of events per trial
310 410
510
610
Ave
rag
e f
it f
ractio
n in
sig
na
l re
gio
n
0.0015
0.002
0.0025
0.003
0.0035
0.004 Nominal Tail Fraction
di-jet function Crystal Ball
Determining the best fitting range
• Increasing the minimum of the fit range has 2 main effects:
• estimates from nominal (dijet) function asymptote
• estimates from dijet and crystal ball converge
• These indicate that a choice of 200-250 would give the most unbiased estimates
• At the same time, the statistical uncertainty from the fit, and the uncertainty from the side-band normalisation also increases when increasing the minimum of the fit range, so 200 is the best choice
68
[GeV]TM
0 200 400 600 800 1000 1200 1400 1600 1800 200010×
Events
-410
-310
-210
-110
1
10
210
310dijet function
crystal ball
[GeV]TM
0 200 400 600 800 1000 1200 1400 1600 1800 200010×
Events
-410
-310
-210
-110
1
10
210
310dijet function
crystal ball
[GeV]TM
0 200 400 600 800 1000 1200 1400 1600 1800 200010×
Events
-410
-310
-210
-110
1
10
210
310dijet function
crystal ball
Min fit range dijet function crystal ball
MT>160 0.36 0.49
MT>180 0.32 0.40
MT>200 0.24 0.26
MT>250 0.23 0.23
Multijet estimates in MT>700 GeV signal region
Fitmin: 160 GeV
Fitmin: 200 GeV
Fitmin: 250 GeV
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
69
(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:
Ryan Reece (Penn)
Jet width for quark/gluons
70
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
71
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
72
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
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)
Ryan Reece (Penn)
W+jets fake factor systematic
73
) [GeV]hτ(
Tp
0 50 100 150 200 250
τ f
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
)+jetν l →W(
)+jetµµ →Z(
ee)+jet→Z(
) [GeV]hτ(
Tp
0 50 100 150 200 250
τ f
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
)hτ)+jet, OS(l, ν l →W(
)+jetµµ →Z(
ee)+jet→Z(
Figure 44: (left) Tau identification fake factors taken from regions of data enriched in W(→ ℓν)+jets,
Z(→ µµ)+jets, and Z(→ ee)+jets events. The fake factors are consistent within statistical
errors among samples. (right) Tau identification fake factors taken from regions of data
enriched inW(→ ℓν)+jets, Z(→ µµ)+jets, and Z(→ ee)+jets events, where the lepton and jet
are restricted to have opposite signs in the W-rich regions. The fake factors differ by ≈ 20%
at low hadronic tau transverse momentum.
Ryan Reece (Penn)
Cross-check: single fake factor method
74
!"#$%%&'()%!*+!% ,-./.(0%1''#234% 5'2#0%603070!!!!!!"!
&)&%8'-.9#%03:%-;-#'10#2<-%
3-prong
ATLAS work in progress
0jet bin
Fake !"#$%&'("
comes from
quark.
Systematic uncertainty - Conservative way
!Up : F.F in W+0jet events.
!Down: F.F in multi-jets events.
MMC
+50%
-50%
Assign !50% to bkg. Normalization!
Ryan Reece (Penn)
Reminder of previous eτh issues
75
Ele
ctr
ons /
(5 G
eV
)
0
200
400
600
800data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
Blin
ded
(e) [GeV]T
p
0 20 40 60 80 100 120 140obs.
/ exp.
0
1
2 data stat. uncert.
model stat. uncert.
Ele
ctr
ons /
(10 G
eV
)
-110
1
10
210
310
410
510data 2011
τ τ →Z W/Z+jets
multijet
e e→Z
tt γW
dibosonsingle top
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb∫
Blin
ded
(e) [GeV]T
p
0 100 200 300 400 500 600obs.
/ exp.
0
1
2 data stat. uncert.
model stat. uncert.
• Reminder: our e-had channel has been delayed because of the following
issues with the modeling:
• High pT(e) over-estimated; W+jet over-estimated; ruins signal region
• General lack of confidence in the Z → ee normalization; no real effect on signal region
• Last few weeks our general strategy has been to illustrate the systematics to try to cover
the discrepancy, varying electron isolation, and re-binning both the fake factors and
kinematic distributions.
Ryan Reece (Penn)
) [GeV]miss
T, Eh!(e, TM
210 310
Events
/ (
10 G
eV
)
-110
1
10
210
310
410data 2011
! ! "Z
multijet
e e"Z
tt
#W
diboson
single top
syst.$stat.
ATLAS Internal -1dt L = 4.6 fb%
(e) [GeV]T
p
210
Events
/ (
10 G
eV
)
-110
1
10
210
310
410data 2011
! ! "Z
multijet
e e"Z
tt
#W
diboson
single top
syst.$stat.
ATLAS Internal -1dt L = 4.6 fb%
Failing tau ID control region
• Our understanding of what was wrong with the modeling is that
the multijet contamination is underestimated in the failing tau ID
control region, therefore causing the W+jet to be overestimated.
• Subtraction in the highest unblinded pT bin is ≈ 30%!
76
(used to estimate W+jet after subtraction)
Ryan Reece (Penn)
[GeV]missTE
0 20 40 60 80 100 120 140
Events
/ (
5 G
eV
)
0
50
100
150
200
250
300data 2011
! ! "Z
multijet
e e"Z
tt
#W
diboson
single top
syst.$stat.
ATLAS Internal -1dt L = 4.6 fb%
[GeV]missTE
210
Events
/ (
10 G
eV
)
-110
1
10
210
310
410data 2011
! ! "Z
multijet
e e"Z
tt
#W
diboson
single top
syst.$stat.
ATLAS Internal -1dt L = 4.6 fb%
Failing tau ID control region
• To help decouple the multijet and W+jet estimates, we added a
MET cut back to the event selection.
• The new/current selection requires MET > 30 GeV.
• This lessens the multijet contamination that needs to be
subtracted, improving the precision of the W+jet estimate.77
(used to estimate W+jet after subtraction)
Ryan Reece (Penn)
Events
/ (
10 G
eV
)
0
50
100
data 2011
! ! "Z W/Z+jets
multijet
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
) [GeV]miss
T(e, ETm
0 20 40 60 80 100 120140 160 180 200obs.
/ exp.
0
1
2
Events
/ (
20 G
eV
)
-110
1
10
210
310
410 data 2011
! ! "Z W/Z+jets
multijet
e e"Z
tt #W
dibosonsingle top
syst.$stat.
Z’(750)
Z’(1000)
Z’(1250)
ATLAS Internal-1dt L = 4.6 fb%
) [GeV]miss
T(e, ETm
0 100200 300 400 500 600700 800 9001000obs.
/ exp.
0
1
2
Adding mT(e,MET) cut
• Many months ago we debated about using mT(e,MET) or ∆!(e,MET) cuts for rejecting
W+jets, but decided instead to use a simpler / more signal inclusive selection to not
reject signal where the MET is aligned with the hadronic tau.
• We propose adding a mT(e,MET) < 50 GeV cut, to suppress W+jet.
• This will make the W+jets modeling less important (and suppress ttbar), increasing the
purity of our irreducible Z→ττ background and simplifying our background composition.
78