New Photon Results from CDFCostas Vellidis
Fermilab
DIS 2012, Marseilles, April 22
DIS 2013 – C. Vellidis 2
Photon analyses at CDF• Photon-related analyses have been hot topics at CDF• ~30 papers published using CDF Run II data on a
wide variety of photon-related topics.o Cross section measurementso Searches
4/24/13
Hγγ
Xgg
Inclusive-g
DIS 2013 – C. Vellidis 3
Diphoton cross sections
4/24/13
p
g
g
p_
DIS 2013 – C. Vellidis 4
Prompt gg production in hadron colliders
Born: a2
ggqq qgq gg
Compton+radiation asa2
Dg/q~a/as
Fragmentation: a2
Suppressed byisolation cut
“Box”: Dominantat the LHC
gggg
Hard QCD (“direct” gg production):
colinearsingularity
4/24/13
DIS 2013 – C. Vellidis 5
Prompt gg production in hadron colliders
Born: a2
ggqq qgq gg
Compton+radiation asa2
Dg/q~a/as
Fragmentation: a2
Suppressed byisolation cut
“Box”: Dominantat the LHC
gggg
Hard QCD (“direct” gg production):
Possible heavy resonance decays:
€
H →γγ
€
G∗ →γγ
Higgs boson
colinearsingularity
Extra dimensions4/24/13
DIS 2013 – C. Vellidis 6
• Identified the importance of resummation, qg fragmentation in the modeling of diphoton cross sections.
PRL 107 (2011) 102003PRD 84 (2011) 052006
5.4 fb-1
4/24/13
Previously published results – CDF
DIS 2013 – C. Vellidis 7
Previously published results – D0
• Sherpa describes data the best in the intermediate PT(gg) and low gg regions.
PT1(2)>18(17) GeV/c, |η1,2|<0.9R(g,g)>0.4, ET
iso<2.5 GeV
4/24/13
arXiv:1301.4536Full Run II data set
DIS 2013 – C. Vellidis 8
Previously published results – ATLASJHEP 1301 (2013) 086
PT1(2)>25(22) GeV/c,|η1,2|<2.37R(g,g)>0.4
4/24/13
DIS 2013 – C. Vellidis 9
Previously published results – CMS
DIPHOX discrepancy for PT(gg)>30 GeV and (g,g)<π/2
JHEP 1201 (2012) 133
4/24/13
DIS 2013 – C. Vellidis 10
Collinear diphoton production• Fragmentation – a higher-order effect
o The pQCD cross section is divergent when q and g are collinear logarithmic enhancement of the cross section
o Handled with a fragmentation function – MCFM, DIPHOXo Affects low m(gg), moderate PT(gg) and low gg regions
• Higher order subprocesses (23 at 1-loop and 24 at “tree” level) needed to describe the enhancement
4/24/13
DIS 2013 – C. Vellidis 11
Resummation• Remove singularities [PT(gg)->0] by adding initial gluon radiation
o RESBOS: Low-PT analytically resummed calculation (NNLL) matched to high-PT NLO
o PYTHIA and SHERPA: Use parton showering to add gluon radiation in a Monte Carlo simulation framework which effectively resums the cross section (LL)
o Affects low PT(gg) and gg = p regions
4/24/13
PRD 76, 013009 (2007)
€
a sn
PT2 γγ( )
lnm M 2 γγ( )PT
2 γγ( )
€
− α snδ
r P T γγ( )( )
€
n =1,...∞ m = 0,...,2n −1
or
Fixed-order calculation contains singular terms at and M(gg) ≠ 0 of the form
€
PT γγ( ) →0
DIS 2013 – C. Vellidis 12
Updated diphoton cross section measurements
• Use the full 9.5 fb-1 CDF run II dataset• Select isolated diphoton events
o Background subtraction using track isolation information• Pythia evaluation of efficiency/acceptance/unfolding• Compare results with new predictions
€
dσdX
=Nγγ
ε ⋅ A⋅ L⋅ Δ
4/24/13
DIS 2013 – C. Vellidis 13
The Tevatron and CDFTevatron:Proton-antiproton accelerator√s = 1.96 TeVDelivered ~12 fb-1
Recorded ~10 fb-1 for each experiment
CDFCollider Detector at FermilabTracking (large B field):
◦ Silicon tracking◦ Wire Chamber
Calorimetry:◦ Electromagnetic (EM)◦ Hadronic
Muon system
€
σ(E) / E =13.5% / E (GeV) ⊕1.5%4/24/13
A big thank you to Accelerator Division!
DIS 2013 – C. Vellidis 14
Photon identification and event selection
Used dedicated diphoton triggers with optimized efficiency
Photons were selected offline from EM clusters, reconstructed in a cone of radius R=0.4 in the – plane, and requiring:
• Fiducial to the central calorimeter: ||<1.1
• ET 17,15 GeV (gg events)
• Isolated in the calorimeter: Ical = Etot(R=0.4) - EEM(R=0.4) 2 GeV
• Low HAD fraction: EHAD/EEM 0.055 + 0.00045Etot/GeV
• At most one track in cluster with pTtrk 1 GeV/c + 0.005ET
g/c
• Shower profile consistent with predefined patterns: 2CES 20
• Only one high energy CES cluster: ET of 2nd CES cluster 2.4 GeV + 0.01 ET
γ
CP2: pre-shower CES: shower maximum profile
EM Cal HAD Cal
Isolation cone: R=0.4 rad
Imply thatR(g,g) or R(g,j) 0.4
4/24/13
DIS 2013 – C. Vellidis 15
Theoretical predictions• PYTHIA LO parton-shower calculation – including gg and gj with radiation [T. Sjöstrand et al., Comp. Phys. Comm. 135, 238 (2001)]
• SHERPA LO parton-shower calculation with improved matching between hard and soft physics [T. Gleisberg et al., JHEP 02, 007 (2009)]
• MCFM: Fixed-order NLO calculation including non-perturbative fragmentation at LO [J. M. Campbell et al., Phys. Rev. D 60, 113006 (1999)]
• DIPHOX: Fixed-order NLO calculation including non-perturbative fragmentation at NLO [T. Binoth et al., Phys. Rev. D 63, 114016 (2001)]
• RESBOS: Low-PT analytically resummed calculation matched to high-PT NLO [T. Balazs et al., Phys. Rev. D 76, 013008 (2007)]
• NNLO calculation with qT subtraction [L. Cieri et al., http://arxiv.org/abs/1110.2375 (2011)]
4/24/13
DIS 2013 – C. Vellidis 16
Theoretical predictions• PYTHIA LO parton-shower calculation – including gg and gj with radiation [T. Sjöstrand et al., Comp. Phys. Comm. 135, 238 (2001)]
• SHERPA LO parton-shower calculation with improved matching between hard and soft physics [T. Gleisberg et al., JHEP 02, 007 (2009)]
• MCFM: Fixed-order NLO calculation including non-perturbative fragmentation at LO [J. M. Campbell et al., Phys. Rev. D 60, 113006 (1999)]
• DIPHOX: Fixed-order NLO calculation including non-perturbative fragmentation at NLO [T. Binoth et al., Phys. Rev. D 63, 114016 (2001)]
• RESBOS: Low-PT analytically resummed calculation matched to high-PT NLO [T. Balazs et al., Phys. Rev. D 76, 013008 (2007)]
• NNLO calculation with qT subtraction [L. Cieri et al., http://arxiv.org/abs/1110.2375 (2011)]
4/24/13
Integrated cross section (pb)
Data (CDF) 12.3 ± 0.2stat ± 3.5syst
RESBOS 11.3
DIPHOX 10.6
MCFM 11.5
SHERPA 12.4
PYTHIA gg+gj 9.2
NNLO 11.8
DIS 2013 – C. Vellidis 17
m(gg)
• Good agreement between data and theory for Mgg>30 GeV/c2 except PYTHIA gg
4/24/13
DIS 2013 – C. Vellidis 18
PT(gg)
4/24/13
DIS 2013 – C. Vellidis 19
PT(gg) - ratios
• RESBOS agrees with low PT(gg) data the best
• SHERPA agrees with low PT(gg) data well
• NNLO and SHERPA describe the “shoulder” of the data at PT(gg) = 20 – 50 GeV/c (the “Guillet shoulder”)
NB: Vertical axis scales are not the same
DIPHOX
RESBOS
PYTHIA NNLO
MCFM
SHERPA
4/24/13
DIS 2013 – C. Vellidis 20
(gg)
4/24/13
DIS 2013 – C. Vellidis 21
(gg)- ratios
• RESBOS and SHERPA describe (gg) = p region• Fixed order calculations do not describe (gg) = p region• NNLO describes (gg) = 0 region
NB: Vertical axis scales are not the same
DIPHOX
RESBOS
PYTHIA NNLO
MCFM
SHERPA
4/24/13
DIS 2013 – C. Vellidis 22
Summary of diphoton cross sections• High precision gg cross sections are measured using the full CDF Run II dataset
• The data are compared with all state-of-the-art calculations
• The SHERPA calculation, overall, provides good description of the data, but still low in regions sensitive to nearly collinear gg emission (very low mass, very low Δϕ)
• The RESBOS calculation provides the best description of the data at low PT and large Δϕ, where resummation is important, but fails in regions sensitive to nearly collinear gg emission
• The NNLO calculation provides the best description of the data at low Δϕ, but still not very good at very low mass and at high PT
• More in PRL 110, 101801 (2013) (supplemental material online)
4/24/13
DIS 2013 – C. Vellidis 23
Photon+heavy flavor (b/c) cross sections
4/24/13
p
g
p_
b-jet
DIS 2013 – C. Vellidis 24
g+b/c+X production• Photon produced in association with heavy
quarks provides valuable information about heavy flavor excitation in hadron collisionso LO contribution: Compton scattering (QgQg)
dominates at low photon pT
o NLO contribution: annihilation (qqQQg) dominates at high photon pT
Compton scattering ~ aaS Annihilation ~ aaS2
Q
g
q
q
- -
4/24/13
-
DIS 2013 – C. Vellidis 25
Previous results – D0PRL 102, 192002 (2009) − 1 fb-1
• Good agreement for g+b+X• Discrepancy for g+c+X
PLB 714, 32 (2012) – 8.7 fb-1
g+b+X
PLB 719, 354 (2013) – 8.7 fb-1
Discrepancies in both channels.
g+c+X
4/24/13
DIS 2013 – C. Vellidis 26
Previous results – CDF
• Measure low pT cross section using a special trigger• g+b+X agrees with NLO up to 70 GeV
CDF: PRD 81, 052006 (2010) - 340 pb-1
4/24/13
DIS 2013 – C. Vellidis 27
Analysis overview• Measure g+b/c+X cross section using 9.1 fb-1 inclusive photon
data collected with CDF II detector• Use ANN (artificial neural network) to select photon
candidateso Fit ANN distribution to signal/background templates to get photon
fraction• Use SecVtx b-tag to select heavy-flavor jets
o Fit secondary vertex invariant mass to get light/c/b quark fractions• Use Sherpa MC to get efficiency/unfolding factor
o Photon ID efficiency, b-tagging efficiency, detector acceptance and smearing effects
• Cross sectiono
4/24/13
€
dσdET
γ =N⋅ fγ ⋅ fb / c
ε ⋅ A⋅ L⋅ ΔETγ
DIS 2013 – C. Vellidis 28
4 theoretical predictions• NLO – direct-photon subprocesses and fragmentation subprocesses
at O(aas2), CTEQ6.6M PDFs [T.P. Stavreva and J.F. Owens, PRD 79,
054017 (2009)]
• kT-factorization – off-shell amplitudes integrated over kT-dependent parton distributions, MSTW2008 PDFs [A.V. Lipatov et al., JHEP 05, 104 (2012)]
• Sherpa 1.4.1 – tree-level matrix element (ME) diagrams with one photon and up to three jets, merged with parton shower, CT10 PDFs [T. Gleisberg et al., JHEP 02, 007 (2009)]
• Pythia 6.216 – ME subprocesses: gQgQ, qqgg followed by gluon splitting: gQQ, CTEQ5L PDFs [T. Sjöstrand et al., JHEP 05, 026 (2006)]
__
4/24/13
DIS 2013 – C. Vellidis 29
g+b+X cross sections
• NLO fails to describe data at large photon Et – perhaps gluon splitting is treated at LO• kT-factorization and Sherpa agree with data reasonably well• Pythia with doubled gluon splitting rate to heavy flavor describes the shape
4/24/13
NB: Vertical axis scales are not the same
DIS 2013 – C. Vellidis 30
g+c+X cross sections
4/24/13
• NLO fails to describe data at large photon Et – perhaps gluon splitting is treated at LO• kT-factorization and Sherpa agree with data reasonably well• Pythia with doubled gluon splitting rate to heavy flavor describes the shape
NB: Vertical axis scales are not the same
DIS 2013 – C. Vellidis 31
Summary of photon+b/c cross sections• High precision g+b/c cross sections are measured using the full CDF Run II
dataset
• The data are compared with parton shower, fixed-order and kt-factorization calculations
• NLO does not reproduce data most likely because of its limitation in modeling gluon splitting rates.
• kT-factorization and Sherpa agree with data reasonably well
• Pythia with doubled gluon splitting rates to heavy flavor describes the data shape
4/24/13
DIS 2013 – C. Vellidis 32
Conclusions• The CDF experiment has produced a wealth of QCD physics
results and analysis techniques, which is a legacy for the current and future high energy physics experiments
• We have achieved an unprecedented level of precision for many photon-related observables
• Those results provide valuable information to the HEP community, e.g. the diphoton results can help the precision measurements of H boson in the gg channel.
• … and we are not done yet!!
4/24/13
DIS 2013 – C. Vellidis 334/24/13
DIS 2013 – C. Vellidis 34
Interesting kinematic variables•
o Search for resonances.•
o Sensitive to activity in the event.•
o Sensitive to production mechanism.
4/24/13
€
m(γγ) = (E1 + E2)2 − (p1 + p2)2
€
PT (γγ) = PT1 + PT 2
€
φ(γγ )
PT1
PT2
=0
g1
g2
p p_
DIS 2013 – C. Vellidis 35
Interesting kinematic variables•
o Search for resonances.•
o Sensitive to activity in the event.•
o Sensitive to production mechanism.
• Fragmentation/higher order diagramso Two g’s go almost collinearo Low m(gg), intermediate PT(gg), low (gg)
• Resummationo Low PT(gg), high (gg)
4/24/13
=0
g1
g2
p p_
€
m(γγ) = (E1 + E2)2 − (p1 + p2)2
€
PT (γγ) = PT1 + PT 2
€
φ(γγ )
Special case=0
g1
g2
p p_
PT1
PT2
DIS 2013 – C. Vellidis 36
Background subtraction using track isolation
• Sensitive only to underlying event and jet fragmentation (for fake g)
• Immune to multiple interactions (due to z-cut) and calorimeter leakage
• Good resolution in low-ET region, where background is most important
• Uses charged particles only
Signal: direct diphotonsBackground: jets misidentified as photons – jg, jj
trkT
cmzz
Rintrackstrk pI
trkvtx
5
4.0
Signal Probability (Itrk<1 GeV)
4/24/13
Background Probability (Itrk<1 GeV)
DIS 2013 – C. Vellidis 37
Background subtraction• For a single g, a weight can be defined to characterize it as signal or
background:o = 1 (0) if Itrk () 1 GeV/c
o s = signal probability for Itrk 1 GeV/c
o b = background probability for Itrk 1 GeV/c
• For gg, use the track isolation cut for each photon to compute a per-event weight under the different hypotheses (gg, g+jet and dijet):
b
b
s
w
€
w jj
w jγ
wγj
wγγ
⎛
⎝
⎜ ⎜ ⎜ ⎜ ⎜
⎞
⎠
⎟ ⎟ ⎟ ⎟ ⎟
= E −1 ×
N ff
N fp
N pf
N pp
⎛
⎝
⎜ ⎜ ⎜ ⎜ ⎜
⎞
⎠
⎟ ⎟ ⎟ ⎟ ⎟
Both photons fail
Leading fail, trailing passes
Leading passes, trailing fails
Both photons pass
€
N ff
N fp
N pf
N pp
⎛
⎝
⎜ ⎜ ⎜ ⎜ ⎜
⎞
⎠
⎟ ⎟ ⎟ ⎟ ⎟
=
0010
⎛
⎝
⎜ ⎜ ⎜ ⎜
⎞
⎠
⎟ ⎟ ⎟ ⎟
e.g. leading passes/trailing fails
4/24/13Transfer matrixFunction of s and b
DIS 2013 – C. Vellidis 38
• Average 40%• Better at high mass:
o 60-80% for m() 80-150 GeV/c2
o 80% for m( )>150 GeV/c2
• Better at high PT():o 70% for PT() >100 GeV/c
• 15-30% sys. errors4/24/13
Signal fractions
€
Signal fraction =Nγγ
Ndata
DIS 2013 – C. Vellidis 39
Efficiency×Acceptance• Estimated using detector-
and trigger-simulated and reconstructed PYTHIA events
• Procedure iterated to match PYTHIA kinematics to the data
Uncertainties in the efficiency estimation:• 3% from material uncertainty• 1.5% from the EM energy scale• 3% from trigger efficiency uncertainty• 6% (3% per photon) from underlying
event (UE) correction• Total systematic uncertainty: ~7-15%
4/24/13
DIS 2013 – C. Vellidis 40
Experimental systematic uncertainties
• Total systematic uncertainty 15-30%, smoothly varying with the kinematic variables considered
• Main source is background subtraction, followed by overall normalization (efficiencies: 7%; integrated luminosity: 6%; UE correction: 6%)
4/24/13
DIS 2013 – C. Vellidis 414/24/13
Integrated cross section (pb)
Data (CDF) 12.3 ± 0.2stat ± 3.5syst
RESBOS 11.3 ± 2.4
DIPHOX 10.6 ± 0.6
MCFM 11.5 ± 0.3
SHERPA 12.4 ± 4.4
PYTHIA gg+gj 9.2
NNLO 11.8 + 1.7 – 0.6
DIS 2013 – C. Vellidis 42
Comparison with D0
4/24/13 €
dσdX
=Nγγ
ε ⋅ A⋅ L⋅ Δ
DIS 2013 – C. Vellidis 43
A closer look at fragmentation: DIPHOX isolation study
iso < 2 GeViso < 2 GeViso < 2 GeV
Fragmentation strength is missing from the DIPHOX calculation possibly because of the approximate application of the isolation requirement at the parton level
4/24/13
DIS 2013 – C. Vellidis 44
TotalDirect1-frag2-frag
ETiso < 2 GeV
ETiso < 10 GeV
A closer look at fragmentation: DIPHOX isolation study
4/24/13
DIS 2013 – C. Vellidis 45
Event selection• Use inclusive photon trigger to select photon events
o Trigger efficiency is approximately 100% for g ET>30 GeV
• Interaction vertex in the fiducial region• Photon candidate must pass a neural-net based photon
IDo ANN>0.75o ||<1.05, 30<ET<300 GeV, divided into 8 ET bins
• Jets are reconstructed with JetClu cone size 0.4 and must be positively tagged. o ||<1.5, ET>20 GeV
• R(g,jet)>0.4
4/24/13
DIS 2013 – C. Vellidis 46
ANN photon ID• Trained with TMVA (Toolkit for Multivariate Data Analysis)• 7 input variables to take into account difference between g and p0/: isolation
(2), lateral shower shape (3), Had/Em, CES/CEM• ANN ID improves signal efficiency by 9% at the same background rejection
compared with the standard cut-based ID.• Use MC with full detector simulation to get templates
o Signal – prompt photonso Background – jets with prompt photons removed
prompt photons
p0,
4/24/13
DIS 2013 – C. Vellidis 47
True photon fraction• Fit data ANN distribution using signal and
background templates to get true photon fraction
4/24/13
DIS 2013 – C. Vellidis 48
True photon fraction (continued)
• Systematicso Photon energy scaleo Vary inputs to photon ID ANN according to their uncertaintieso Vary Photon ID ANN template binning to test sensitivity to shapeso 6% at low ET, 2% at high ET.
4/24/13
DIS 2013 – C. Vellidis 49
Standard b-jet identification
• B-hadrons are long-lived – search for displaced vertices
• Fit displaced tracks and cut on Lxy significance (σ ~ 200 mm)
• Charm hadrons have similar tag behavior but lower efficiency
• Use “tag mass” to deduce the flavor composition of a sample of tagged jetso Mass of the tracks forming the
secondary vertexo B-hadrons are heavy: will have
higher mtag spectrum than charm or light jet fakes
4/24/13
DIS 2013 – C. Vellidis 50
Light/c/b-jet fractions
• Fit data secondary vertex mass using MC templates• Shape of secondary vertex mass for event with fake photon is
taken from di-jet data4/24/13
DIS 2013 – C. Vellidis 51
Light/c/b-jet fractions (continued)
Results from fitter.4/24/13
DIS 2013 – C. Vellidis 52
Systematics on b/c-jet fractions
• Jet energy scale: affects acceptance• Uncertainty in tracking efficiency: scale secondary vertex
mass templates by ±3%o Dominant systematic effect
• Difference between single-quark and di-quark jets• Total systematic error is ~20%
4/24/13
DIS 2013 – C. Vellidis 53
Efficiency×Acceptance
• Use Sherpa MC to unfold photon ID efficiency, b-tagging efficiency, detector acceptance and smearing effects.
• Systematic effects evaluatedo photon energy scale and IDo jet energy scaleo b-tagging efficiencyo Generatoro PDF4/24/13
Top Related