Jet Energy Corrections in CMS

1
Jet Energy Corrections in CMS

• Daniele del Re
• Universita di Roma La Sapienza and INFN Roma

2
Outline

• Summary of effects to be corrected in jet
  reconstruction
• CMS proposal factorization of corrections
• data driven corrections
• Strategy to extract each correction factor from
  data
• Perspectives for early data
• Priorities, expected precisions, statistics
  needed
• Note results and plots in the following are
  preliminary and not for public use yet

3
CMS Detector Calorimetry
gt75k lead tungstate crystals
crystal lenght 23cm Front face 22x22mm2
PbWO4 30g/MeV X00.89cm
HO
Had Barrel HB brass Absorber and Had
Endcaps HE scintillating tilesWLS Had
Forward HF scintillator catcher. Had Outer
HO iron and quartz fibers
HB
HE
HF
4
Jet reconstruction and calibration

• Calorimeter jets are reconstructed using towers
• Barrel un-weighted sum of energy deposits in one
  or more HCAL cells and 5x5 ECAL crystals
• Forward more complex HCAL-ECAL association
• In CMS we use 4 algorithms iterative cone,
  midpoint cone, SIScone and kT
• will give no details on algorithms, focusing on
  corrections
• Role of calibration
• correct calorimeter jets back either to particle
  or to parton jets (see picture)

5
Parton level vs particle level corrections

• In CMS
• Calojets are jets reconstructed from calorimeter
  energy deposits with a given jet algorithm
• Genjets are jets reconstructed from MC particles
  with the same jet algorithm
• Two options
• convert energy measured in jets back to partons
  (parton level)
• convert energy measured in jets back to particles
  present in jet (particle level)
• Idea is to correct back to particle level
  (Genjets)
• Parton level corrections are extra and can be
  applied afterwards

6
Causes of bias in jet reconstruction

• jet reconstruction algorithm
• Jet energy only partly reconstructed
• non-compensating calorimeter
• non-linear response of calorimeter
• detectors segmentation
• presence of material in front of calorimeters and
  magnetic field
• electronic noise
• noise due to physics
• Pileup and UE
• flavor of original quark or gluon

7
Dependence of bias

• vs pT of jet
• Non-compensating calorimeter
• low pT tracks in jet
• vs segmentation
• large effect vs pseudorapidity h (large detector
  variations)
• small effect vs f (except for noisy or dead cal
  towers)
• vs electromagnetic energy fraction
• non-compensating calorimeter
• vs flavor
• vs machine and detector conditions
• vs physics process
• e.g. UE depends on hard interaction

8
Dependence of bias vs causes
Jet algorithm Non-compensating Segmentation Material in front of cal. Electronic noise Physics noise Original quark/gluon
vs pT
vs h
vs em fraction
vs flavor
vs conditions
vs process
Complicated grid better to estimate dependences
from data than study each single effect
9
Factorization of corrections

• correction decomposed into (semi)independent
  factors applied in a fixed sequence
• choice also guided by experience from previous
  experiments
• many advantages in this approach
• each level is individually determined, understood
  and refined
• factors can evolve independently on different
  timescales
• systematic uncertainties determined independently
• Prioritization facilitated determine most
  important corrections first (early data taking),
  leave minor effects for later
• better collaborative work
• prior work not lost (while monolithic corrections
  are either kept or lost)

10
Levels of corrections

• Offset removal of pile-up and residual
  electronic noise.
• Relative (h) variations in jet response with h
  relative to control region.
• Absolute (pT) correction to particle level
  versus jet pT in control region.
• EM fraction correct for energy deposit fraction
  in em calorimeter
• Flavor correction to particle level for
  different types of jet (b, t, etc.)
• Underlying Event luminosity independent
  spectator energy in jet
• Parton correction to parton level

L2 Relh
L1 Offset
L3 AbspT
L4 EMF
L5 Flavor
L1 UE
L1 Parton
Reco Jet
Calib Jet
Required
Optional
11
Level 1 Offset

• Goal correct for two effects 1) electronic noise
  2) physics noise
• 1) noise in the calorimeter readouts
• 2a) multiple pp interactions (pile-up)
• 2b) (underlying events, see later)
• additional complication energy thresholds
  applied to reduce data size
• selective readout (SR) in em calorimeter (ECAL)
• zero suppression (ZS) in had calorimeter (HCAL)
• with SR-ZS, noise effect depends on energy
  deposit
• need to properly take into account SR-ZS effect
  before subtracting noise

12
Level 1 Correction
Evaluate effect of red blobs without ZS in data
taking

• 1) take runs without SR-ZS triggered with jets
• perform pedestal subtraction
• evaluate the effect of SR-ZS vs pT
• Apply ZS offline and calculate multiplicative
  term
• 2) take min-bias triggers without SR-ZS
• run jets algorithms and determine noise
  contribution (constant term)
• 3) correct for SR-ZS and subtract noise

no pileup and noise
with pileup and noise
Under threshold removed by ZS
Now over threshold not removed
13
Level 2 h dependence

• Goal flatten relative response vs h
• extract relative jet response with respect to
  barrel
• barrel has larger statistics
• better absolute scale
• small dep. vs h
• extract
• h correction in bins of pT (fully
• uncorrelated with the next
• L3 correction)

Relative Response
Before
1
After
1
3
2
4
Jet h
14
Level 2 data driven with pT balance

• use of 2?2 di-jet process
• main selection based on
• back-to-back jets (x-y)
• events with 3 jets removed
• di-jet balance with quantity
• response is extracted with

Probe Jet other jet
y
Trigger Jet ?lt1.0
z
Probe Jet other jet
y
x
Trigger Jet ?lt1.0
15
Level 2 Missing Projection Function

• MPF pT balance of the full event
• in principle independent on jet algo
• purely instrumental effects
• less sensitive to radiation (physics modeling) in
  the event
• ... but depends on good understanding of missing
  ET
• need to understand whole calorimeter before it
  can be used
• Response ratio extracted as

16
Level 3 pT dependence

• Goal flatten absolute response variation vs pT
• Balance on transverse plane (similar to L2 case),
  two methods
• g jet
• mainly qg-gtqy
• large cross section
• not very clean at low pT
• Z jet
• relatively small cross
• cleanest
• response is
• rescale to parton level, extra MC correction
  needed from parton to particle
• also MPF method (as for L2 case)

17
Level 3 gjet example

• main bkg QCD events (di-jet)
• selection based on
• g isolation from tracks, other em and had.
  deposits
• per event selection reject events with multiple
  jets, g and jet back-to-back in x-y plane
• 1 fb-1 enough for decent
• statistical error over pT range
• but for low pT large contamination
• from QCD (use of Zjet there)

pT(jet)/pT(g)
18
Level 4 electromagnetic energy fraction

• Goal correct response dependence vs relative
  energy deposit in the two different calorimeters
  (em and had)
• detector response is different for em particles
  and hadrons
• electrons fully contained in em calorimeter
• fraction of energy deposited by hadrons in em
  calorimeter varies and change response
• independent from other
• corrections (h, pT)
• introducing em fraction correction
• improves resolution

19
Level 4 extract corrections

• start with MC corrections
• idea is to use large gjet samples (not for
  early data)
• also possible with di-jet
• in principle used to improve resolution, no
  effect on bias. Less crucial to have data driven
  methods.

20
Level 5 flavor

• Goal correct jet pT for specific parton flavor
• L3 correction is for QCD mixture of quarks and
  gluons
• Other input objects have different jet
  corrections
• quarks differ from gluons
• jet shape and content depend on quark flavors
• heavy quark very
• different from light
• for instance b in 20 of
• cases decays
• semileptonically

21
Level 5 data driven extraction

• correction is optional
• many analyses cannot identify jet flavors, or
  want special corrections
• correction desired for specialized analysis (top,
  h g bb, h g t t, etc.)
• corrections from
• tt events tt?Wb?qqb
• leptonic hadronic W decay in event, tag 2b
  jets,
• remaining are light quark
• constraints on t and W masses used
• to get corrections
• gjets, using b tagging
• pp?bbZ, with Z?ll

22
Level 6 UE

• Goal remove effect of underlying event
• UE event depends on details of hard scatter
• ? dedicated studies for each process
• ? in general this correction may be not
  theoretically sound since UE is part of
  interaction
• plan (for large accumulated stats) is to use same
  approach as L1 correction but only for events
  with one reconstructed vertex

23
Level 7 parton

• Goal correct jet back to originating parton
• MC based corrections compare
• Calojets after all previous corrections
• with partons in bins of pT
• dependent on MC generators
• (parton shower models, PDF, ...)

24
Sanity checks

• given
• number of corrections
• possible correlation between corrections
• not infinite statistics in calculating
  corrections
• smoothing in extracting corrections
• sanity checks are needed
• after corrections, re-run gjet balance and check
  that distribution is flat
• cross-checks between methods should give same
  answer
• e.g. extract corrections from tt and check them
  on gjet sample

25
Plan for early data taking

• day 1 corrections from MC, including lessons
  from cosmics runs and testbeams
• datalt1fb-1 use of high cross-section data driven
  methods. Tune MC
• longer term run full list of corrections
  described so far

Integrated luminosity Minimum time Systematic uncertaintiy
10 pb-1 gt1 month 10
100 pb-1 gt6 months 7
1 fb-1 gt1 year 5
10 fb-1 gt3 years 3

• numbers do not take into account
• low pT low resolution, larger backgrounds
• larger uncertainties
• 2) large pT control samples have low
• cross section
• ? larger stat. needed

26
Conclusions

• CMS proposes a fixed sequence of factorized
  corrections
• experience from previous experiments guided this
  plan
• first three levels noise-pileup, vs h and vs pT
  sub-corrections represent minimum correction for
  most analyses
• priority in determining from data
• EM fraction correction improves resolution
• last three corrections flavor, UE and parton are
  optional and analyses dependent
• jet energy scale depends on understanding of
  detector
• very first data will be not enough to extract
  corrections (rely on MC)
• 1fb-1 should allow to have 5 statsyst error
  on jet energy scale