Zero Suppression Study with a fast MC

1
Zero Suppression Studywith a fast MC

• John Krane
• At large member of DØ

Calorimeter Task Force December, 2002
2
Goals

• Run 1k events in a few seconds(actual is
  currently 15k jets/min)
• Simulate jet energy profiles
• Longitudinal
• Ncells in each layer
• Simulate noise
• Simulate arbitrary ZSP scheme

3
Method of Fast MC

• Sample (random number thrown against histo)
  Layer_E
• While total jet layer E lt Layer_E
• Sample fractional cell E distribution
• Add noise by sampling actual calibration
  histograms (R. Zitoun)
• Result fluctuates around the generated
  Layer_Ethis is OK
• Add noise to any remaining cells in jet cone
• Suppress cells with 1.5s, 2.5s, or different s
  for different layerswhatever you liketowers

4
Method for Inputs

• Study jets in Monte Carlo for initial profiles
• Free of noise pathology and zero-suppression
• Low, med, high-pT 15-20, 100, 300 GeV
• Get histo of E, layer by layer
• Get histo of Cell E fraction per layer
• Currently using p10.14 with noise and zsp, want
  Jet MC with no noise and no zsp

Forced positive!
5
Input Histograms Total Energy by Layer
Central lyr 1
Central lyr 2
Central lyr 15
Low pt lyr 15 is noise High pt lyr 15 populated
Central lyr 15 for jets gt20 GeV
6
Input Histograms fractional Energy per cell
Central lyr 15
Central lyr 1
7
Current Results (pending better inputs!)

• Have result histos for 9 different ZSP schemes
• Central, ICR, Forward
• Low, med, high jet pT
• Also, 0 GeV Jets, i.e. how noise can fake jetsI
  allowed double-wide jets
• Looking at positive-E towers vs. /- towers

8
Definition of a tower in Fast MC
etc
Cell array 154

• Store cell energies per layer in an array
• Random number of cells rqd to reach the generated
  Layer_E
• A tower is each vertical sum through these
  arrays
• Ignoring merged jets noise jets

Lyr 12
Lyr 11
Lyr 7
Lyr 6
Lyr 5
Lyr 4
Lyr 3
Lyr 2
Lyr 1
9
Aside on Noise jets

• Many noise jets are merged and split DOZENS of
  times (as many as 99 times!)
• Multiply that by number of jet algo
• Reco CPU needs increase
• Event has many many cells so disk needs increase
• My numbers on noise jets provide info on one ZSP
  scheme relative to anotherabsolute numbers not
  so valid because Im not simulating merge/split
  behavior

10
ZSP schemes
1 EM1
2 EM2
3,4,5,6 EM3
7 EM4
8
9
10
11 FH1
12 FH2
13 FH3
14
15 CH
16
17

• 1.5 s zsp
• 2.5 s zsp
• 1.5 in low-noise layers (2,3,4,5,6,7)and 2.5
  elsewhere
• 2.5 in EM and 3.5 elsewhere

11
Key
12
Noise, central region
13
Noise, ICR
14
Noise, forward region
15
Jet Energy
Central
Central
Forward
Forward
16
Jet Energy Loss, Resolution (central)
Still Gaussian
17
Jet Energy Loss, Resolution (ICR)
18
Jet Energy Loss, Resolution (forward)
19
Resolution Calculation

• Previous plots were raw resolution
• Want to assume an Energy Scale Correction that
  performs as well as it can
• Perfect noise correction on average will not
  affect width of resolution gaussian
• Perfect response correction on average
  resolution gaussian width scalles with
  correction

20
Corrected Resolution (Central)
Circ /- tow Star pos-only tow Line is Run I
reso Note odd slope forextreme zsp
21
Corrected Resolution (ICR)
Circ /- tow Star pos-only tow
22
Corrected Resolution (Forward)
Circ /- tow Star pos-only tow I dont trust
this one!
23
Summary

• Fast ZSP Monte Carlo is working
• Post-analysis of JES-corrected resolution
• Need better MC for input histograms