Christof Roland Massachusetts Institute of Technology for the CMS Collaboration

Christof Roland / MITTrento 2008 1

Christof RolandChristof RolandMassachusetts Institute of TechnologyMassachusetts Institute of Technology

for the CMS Collaborationfor the CMS Collaboration

ECT Workshop on Parton Fragmentation Processes in the Vacuum and in the Medium

Trento, Italy, 2008

Extracting jet fragmentation functionsExtracting jet fragmentation functionsusing photon-tagged jet events using photon-tagged jet events

in Pb+Pb collisionsin Pb+Pb collisionswith the CMS experiment with the CMS experiment


q,g

Hadrons

Jet

q,g

q,g

Photon

Jet axis provides parton direction

Photon energy tagsparton energy ET

Charged hadron tracks used to calculate z = pT(track)/ET

Multiplicity and flow measurements characterize density, path length

In-medium fragmentation functions

Measure dN/dlog(1/z)

Ingredients:• Event/Centrality selection• Reaction plane determination• Vertex + track finding • Jet finding• Photon detection

All results based on GEANT-4 simulations using full reconstruction algorithms on 1-year statistics


Compton

Annihilation

Bremsstrahlung

Fragmentation

Signal processes

Use photon to tag parton energyGoal: Best correlation of photon and parton energyIdeal: Use leading order photon

In practice: Isolated photons to select events with good correlation of photon and parton energy


Compton Annihilation

(*) excluding neutrinos and muons

MC signal definition

Photon Isolation:• calculate total ET (*) in cone of R=0.5, ET

tot

• Require (ETtot – ET)< (5GeV + 0.05 ET) and

• find hadron with highest pT, pTmax

• pTmax < (4.5GeV + 0.025 ET)

Photon-Jet pair selection• Events where an isolated photon is emitted back-to-

back with a jet are our signal events


generator-level

σ~13%

172o

generator-level

172o

FF using photon energy tag

• Using isolated photons + “back-to-back” cut on azimuthal opening angle between the photon and the jet to suppress NLO and background events

• Photon energy well correlated with parton energy (RMS ~ 13%)

• Fragmentation functions using parton energy and isolated photon energy agree (better than 10%)


dN/dξ

ξ=log(ET/p

T)

Adapted from hep-ph/0506218

z→1 z→0

Quenching reduceshigh-p

T particles

Quenching increaseslow-p

T particles

jet

hadrons(p

T→0)(p

T→E

T)

10GeV/c

1GeV/c

In-medium modified frag. functions


PYQUEN v1.2: Eur. Phys. J. C 45 (2006) 211HYDJET v1.2: hep-ph/0312204

Event generation

• Study two scenarios• No quenching: PYTHIA signal and QCD background

(p+p) events mixed with central unquenched Pb+Pb HYDJET events

• No high-pT particle suppression• Leads to high background rates

• Quenching: PYQUEN signal and QCD background (p+p) events mixed with central quenched Pb+Pb HYDJET events

• Suppression of high-pT particles• Energy loss radiated out of jet cone• Challenging for jet finder


• Study for one nominal LHC Pb+Pb run “year”

• 106 sec, 0.5nb-1, 3.9 x 109 events• Use 0-10% most central Pb+Pb

• dN/dη|η=0

~2400

• Simulate signal and background QCD (p+p) events

• Mix into simulated Pb+Pb events (~1000 events)

generator-level

Signal statistics


Background statistics

• Select all PYTHIA (PYQUEN) events with potential to create high ET ECAL supercluster• i.e. is there a hadron, charged lepton or photon with ET > 70

GeV

• For each such event, check if corresponding simulated event has a supercluster with ET > 60 GeV

• If yes, mix PYTHIA event with simulated central Pb+Pb background event (at digi level)• We used 1000 Pb+Pb background events

• Run through reconstruction


• Large (mid-rapidity) acceptance (tracker and calorimetry)

• DAQ+HLT will inspect every single Pb+Pb event

• Large statistics for rare probes

2

= 5 for Si tracker

= 10 for Calorimetry

Capabilities

High-precision tracking over < 2.5Muon identification over < 2.5

High resolution calorimetry over < 5Forward coverage

Large bandwidth: DAQ + Trigger

J.Phys.G34:2307-2455,2007

Very well suited for HI environment

CMS Detector


Jet finding: Eur. Phys. J. 50 (2007) 117

Tracking: NIM A566 (2006) 123

Reconstruction• Tracking

• Low pT cutoff at 1GeV/c• Efficiency (algorithmic + geometric) ~ 50-60%• Fake rate ~ few %

• Jet finding• Iterative cone algorithm with underlying event

subtraction (R=0.5)• Performance studies on away-side jet finding (see

later)

• Photon ID • Reconstruction of high-ET isolated photons • New for this analysis (see next slides)


p+p all hits (selective readout)

p+p after seed threshold (0.5/0.18 GeV)

Pb+Pb all hits

dN/dη~2400

Pb+Pb after seed threshold (0.5/0.18 GeV)

ECAL reconstruction chain used with standard p+p settingsNB: The two p+p (QCD) events are not the same.

ECAL response in p+p and Pb+Pb


ECAL

HCAL

Prompt photon

ECAL

HCAL

• Identification• 10 cluster shape

variables• based on ECAL

• 10 isolation variables

• based on ECAL/HCAL

• Track-based cut• Selection

• Total of 21 variables grouped into 3 sets

• Linear discriminant analysis (Fisher) andcut optimization using TMVA

TMVA: http://tmva.sourceforge.net

*) Maximum set to 10

*

Photon ID: Isolation and cluster shape cuts


WP

p+p

Pb+Pb

Photon identification performance

• Set working point to 60% signal efficiency

• Leads to 96.5% background rejection

• Training is done on unquenched samples only


Before cuts: After cuts:S/B=4.5S/B=0.3

WP=60% Seff

Quenched Pb+Pb

Photon isolation and shape cuts improve S/B by factor ~15

Photon identification performance


• Select away-side jet with (,jet) > 1720, ||< 2 and ET > 30 GeV• The energy cut reduces the false rate to 10% level

• Analysis does not use jet energy otherwise

• Jet finding efficiency rises sharply between 30-100 GeV MC jet ET

• Main source of systematic uncertainty in reconstructed FFs

Quenched Pb+Pb

Δφap

>172o, |η|<2

Aw

ay-s

ide

jet f

indi

ng e

ffici

ency

Fra

ctio

n of

fals

ely

mat

ched

jets

Jet finding (away-side)


Jet finding in PYQUEN

• Quenching mechanism in PYQUEN moves energy out of R=0.5 jet cone

• This lowers jet finding efficiency for given initial parton ET


• Obtain dN/dξ using tracks in R=0.5 cone around jet axis• For ξ>3 (~p

T<4GeV/c) dN/dξ dominated by underlying

Pb+Pb event• Estimate background using R=0.5 cone rotated in φ by 90º

relative to jet• Sum event-by-event backgrounds and subtract

Unquenched Quenched

Underlying event

Underlying event

Fragmentation functions


• Major contributions to systematic uncertainty (added in quadrature)– Photon selection and background contamination (15%)

– Track finding efficiency correction (10%)

– Wrong/fake jet matches (10%)

– Jet finder bias (largest contribution in quenched case, see next slide)

QuenchedUnquenched

No or small dependence

Reconstructed frag. functions


Up to ~30%

Quenched

MC truth for found reco jets MC truth for all jets

Jet finder bias

Jet finder bias• Systematic uncertainties

• Up to 30% in quenched case

• 10% for unquenched case• It has two contributions

• FFs and jet finding efficiency depend on parton ET

• Can be corrected with known turn-on curve (not done here)

• For a given parton ET, jet finding probability depends on parton fragmentation pattern

• The jet finder is more likely to find a jet with few high pT particles than jets with many soft particles

• MC based correction might be possible (not done here)

• MC studies suggest that 2) dominates

Up to ~10%

Unquenched


• Medium modification of fragmentation functions can be measured

• High significance for 0.2 < ξ < 5 for both, ET0GeV and

ET100GeV

ET70GeV E

T100GeV

Reco quenched Pb+Pb / MC unquenched p+p

Fragmentation function ratio

Shaded bands show systematic uncertainties


Summary• Complete study of in-medium fragmentation functions

using photon-tagged jet events• Two scenarios: Unquenched and quenched cases

• Pythia and Pyquen (+ Hydjet)

• Key features of the study• Full statistics expected for nominal one-year CMS Pb+Pb run

at LHC• Full detector simulations of signal and background• Complete reconstruction chain

• Track finding • Jet finding• Photon isolation• Underlying event subtraction

• Analysis of systematic errors• Uncertainty dominated by jet finder bias

• Measurement of expected strong medium modification of fragmentation functions can be done reliably in central Pb+Pb


BackUp Slides


Rate Estimate

• HYDJET + Pythia• Using HLT-Note/Jet Note parameters:

• L = 0.5 nb-1, 7.8 b inelastic PbPb x-section• => 3.9 x 109 events• 0-10% central• HYDJET default data cards used for Pythia configuration • Ncoll from HYDJETNN = 58mb• Run Pythia/Pyquen and scale with:

Ncoll x 1/ NN x 3.9*108 (0-10% central) -Jet rate (pT>100 GeV/c, ||<2):

1230 (0-10%), 3680 (mb)


• Use the energy content in cone around candidate direction in ECAL and HCAL

• ECAL:

• HCAL:

Candidate

R2 = + 2

R

Photon isolation: Cone energy


Subtract HI background

ECAL:

HCAL:

Use avg outside of cone(normalized with R/4)

Photon isolation: Background subtraction


Based on cone variables form

Combine to

to be determined coefficients

0-10% quenched Hydjet

Photon isolation: Combined calorimeters


dRxy: R of (y+1st) nearest track with pT >

(0.4*x + 0.2) GeV/c

We use

dR10 for p+p

dR41 for Pb+Pb

R

Photon isolation: Tracks


Based on ECAL shape variables form

Combine to

to be determined coefficients

0-10% quenched Hydjet

Photon ID: Shape variables


After cuts:S/B=4.5Before cuts:S/B=0.3

ET70GeV

Quenched Pb+Pb

• Performance for ET70GeV

• Efficiency > 60%• Fake rate < 20%• Transverse energy resolution: 2-5%

Isolated photon reconstruction


Event Selection Summary• Pb+Pb background events

• 0-10% HYDJET v1.2, 1000 events, dN/d ~ 2400• PYTHIA (v6.411)/PYQUEN (v1.2) events

• pT > 70 GeV potential trigger particle• ET > 60 GeV reconstructed supercluster

• Tracks• pT > 1 GeV/c, > 8 hits, prob > 0.01, DCA/DCA_Err < 3.0

• Reconstructed events• Isolated photon with ET > 70 (100) GeV, | < 2• Jet with ET > 30 GeV, | < 2, (,jet) > 3

• Fragmentation function• Cone-size around jet axis: 0.5


Jet Finder Bias (Quenched Jets)

Significant bias (up to 30%)in quenched sample with ET, > 70GeV

Bias about 50% smallerfor ET, > 100GeV

N.b.: Bias is present in p+p, i.e. can be studied independent of our measurement in Pb+Pb


Reconstructed FF agrees with MC FF within expected uncertainty

Largest deviation at small large z)

QuenchedUnquenched

QuenchedUnquenched

Fragmentation function results (ET, > 70 GeV)


Reconstructed FF agrees with MC FF within expected uncertainty

70 vs 100 GeV: Trade-off between statistical and systematic uncertainties

QuenchedUnquenched

QuenchedUnquenched

Fragmentation functions (ET, > 100 GeV)


ET, > 70GeV

ET, > 100GeV

QuenchedUnquenched

QuenchedUnquenched

Fragmentation functions (vs z)

Documents

Christof Roland Massachusetts Institute of Technology for the CMS Collaboration