March 25, 2009 1 / 15

Duplicate Cluster Studies using J/ψ and Z decays

Andree Robichaud-Veronneau, Gabriella Pasztor (Universite de Geneve)

Outline

March 25, 2009 2 / 15

• Examples of duplicate events from J/ψ→ ee events in 14.5.0

• Cluster Correction tool for low energy clusters

• Sliding Window duplicate deletion

• Properties of duplicates

• Duplicate clusters statistics

• Possible solutions

Example event #1

March 25, 2009 3 / 15

• From J/ψ→ ee events private reconstruction in 14.5.0 (using RDOsamplemc08.105751.Pythia directJpsie3e3.digit.RDO.e347 s462 d126,from F. Derue)

• In event #59617, we see:

◦ 2 duplicate clusters with 3x7 energies 4.5 GeV and 2.8 GeV

◦ Cluster seeds separation barely above 2 cells in η(0.0500304636 > 0.05)

◦ True electron energy is 4.8 GeV

◦ True electron bremed, loosing 32.1 % of its energy

◦ One cluster strongly off-centered

• This leads to one of the clusters to have a drop in its energy goingfrom 5x5 to 3x7 (4.6 GeV to 2.8 GeV)

Cluster Correction tool

March 25, 2009 4 / 15

Recall: How is the 3x7 CaloCluster built

• From the 5x5 cluster, after track-matching is done, a 3x7 cluster isbuilt (for the electron object in Barrel)

• The 3x7 cluster has the same seed position (η0,φ0) as the 5x5cluster

• Several cluster correction tools are called to build its energy

• In CaloFillRectangularCluster, the initial energy is computed perlayer

◦ The Middle is computed first using (η0,φ0), and a newbarycenter (η2,φ2) is defined

◦ (η2,φ2) is used for the Front layer and a new barycenter(η1,φ1) is defined

◦ The position in η1 is refined around the hottest cell

◦ Then the PS layer is calculated using (η1,φ2) and the Backlayer, using (η2,φ2)

Cluster Correction tool for off-centered clusters in η

March 25, 2009 5 / 15

• In our case, problem originates from the Middle layer

◦ As the hottest cells are 2 cells away in η from the (η0,φ0)position, they are not included in the 3x7 cluster

◦ As the cluster correction continues, the position of the clustercenter changes but the energy of the hottest cells (from theMiddle Layer) is never recovered

◦ See layout of cells from 5x5 to 3x7 next slide

• One possible solution would be to do a second iteration of correction(i.e. rebuilding the cells list) using the last computed position as astarting point (η1,φ1) only for these off-centered clusters:

◦ When |η0 - η1| > 0.025 (i.e. about 2 cells apart)

◦ Or when e55

e37> 1.2

◦ or some more optimized value - TBD

Front Layer Cells: 5x5 vs 3x7

March 25, 2009 6 / 15

Cluster 1: e55=4562 MeV, e37=4542 MeV

-20 33 (33) 672 (672) 45 (45) 0-25 -7 (-7) 995 (995) 94 (94) -29

Cluster 2: e55=4554 MeV, e37=2770 MeV

-15 39 672 (672) 64 (45) 1 (0)-5 -2 992 (995) 81 (94) -33 (-29)

• Front cells (strips) are grouped here in bunches of 8

• Cluster 2 5x5 is shifted by one strip with respect to Cluster 1

• Cluster 2 3x7 is NOT shifted with respect to Cluster 1

Middle Layer Cells: 5x5 vs 3x7

March 25, 2009 7 / 15

Cluster 1: e55=4562 MeV, e37=4542 MeV

2 10 610 3 613 228 -11-9 40 666 6 -511 18 27 35 -1316 43 329 17 51-22 -7 0 -11 -25

8 -32 -45Cluster 2: e55=4554 MeV, e37=2770 MeV

6 -13 46613 228 -11 2 40666 6 -5 0 -1727 35 -13 -91 46

329 17 51 4 410 -11 -25 41 -21

-45 31 8

η0=0.5625, φ0=-2.2915η2=0.5642, φ2=-2.2689η1=0.5693, φ1=-2.2700Final η1=0.5689|η0 − η1|=0.0064

PS E=421 MeVFront E=1834 MeVMiddle E=1953 MeVBack E=-132 MeV

η0=0.6125, φ0=-2.2948η2=0.5909, φ2=-2.2514η1=0.5689, φ1=-2.2705Final η1=0.5689|η0 − η1|=0.0436

PS E=421 MeVFront E=1780 MeVMiddle E=260 MeVBack E=-36 MeV

Energy ratio between 5x5 and 3x7 clusters

March 25, 2009 8 / 15

e55/e370 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

1

10

210

310

410

-Ratio distribution for truthmatched e E5x5_E3x7_ratioEntries 67694Mean 0.9952RMS 0.09897

-Ratio distribution for truthmatched e

e55/e370 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

1

10

210

-Ratio distribution for duplicate e E5x5_E3x7_ratio_dupEntries 3255Mean 1.081RMS 0.2498

-Ratio distribution for duplicate e

e55/e370 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

1

10

210

310

410

> 5 GeVT with E-Ratio distribution for truthmatched e E5x5_E3x7_ratioEntries 31749Mean 0.9802RMS 0.06731

> 5 GeVT with E-Ratio distribution for truthmatched e

e55/e370 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

1

10

210

> 5 GeVT with E-Ratio distribution for duplicate e E5x5_E3x7_ratio_dupEntries 482Mean 0.9575RMS 0.133

> 5 GeVT with E-Ratio distribution for duplicate e

3.0 % 27.1 %

0.3 % 3.1 %

Example event #2

March 25, 2009 9 / 15

• From J/ψ→ ee events private reconstruction in 14.5.0 (using RDOsamplemc08.105751.Pythia directJpsie3e3.digit.RDO.e347 s462 d126)

• In event #6522, we see:

◦ 2 duplicate clusters

◦ Cluster seeds separation is only 1 cell in η and φ

◦ True electron energy is 4.4 GeV

◦ 3x7 clusters have similar energies: 4.4 and 4.5 GeV

• Shouldn’t SW delete duplicates when ∆(η0, φ0) < 2x2?

Middle Layer Cells: 5x5

March 25, 2009 10 / 15

Cluster 1 5x5 and Cluster 2 5x5

40 56 26 -14 -17-25 -44 -27 60 19 5473 42 189 289 73 -60 -3 118 246 -12 14

32 -21 68 80 25 -88 21 27 23 12

Seed separation for J/ψ clusters

March 25, 2009 11 / 15

The seed positions ∆ for all cluster pairs (% shown in 0.05x0.05 window)

0φ ∆

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

0η ∆

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

-Seed separation for truthmatched e Delta_eta0_phi0Entries 37194Mean x 0.03571Mean y 0.04171RMS x 0.0488RMS y 0.04201

-Seed separation for truthmatched e

0φ ∆

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

0η ∆

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

-Seed separation for duplicate e Delta_eta0_phi0_dupEntries 1711Mean x 0.01873Mean y 0.04357RMS x 0.02787RMS y 0.02259

-Seed separation for duplicate e

0φ ∆

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

0η ∆

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2 > 5 GeVT with E-Seed separation for truthmatched e Delta_eta0_phi0

Entries 19646Mean x 0.05522Mean y 0.04871RMS x 0.0598RMS y 0.05595

> 5 GeVT with E-Seed separation for truthmatched e

0φ ∆

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

0η ∆

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2 > 5 GeVT with E-Seed separation for duplicate e Delta_eta0_phi0_dup

Entries 248Mean x 0.03159Mean y 0.03129RMS x 0.03949RMS y 0.02884

> 5 GeVT with E-Seed separation for duplicate e

4.1 % 43.5 %

3.2 % 60.9 %

Correlation between seed separation and Energy ratio

March 25, 2009 12 / 15

2 entries on X axis and 1 entry on Y axis per pair of clusters

e55/e370 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0η ∆

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

-Correlation for truthmatched e Delta_eta0_vs_ratioEntries 74388Mean x 1.006Mean y 0.08539RMS x 0.1238RMS y 0.06028

-Correlation for truthmatched e

e55/e370 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0η ∆

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

-Correlation for duplicate e Delta_eta0_vs_ratio_dup

Entries 3422Mean x 1.079Mean y 0.04302RMS x 0.2506RMS y 0.02436

-Correlation for duplicate e

e55/e370 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0η ∆

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2 > 5 GeVT with E-Correlation for truthmatched e Delta_eta0_vs_ratio

Entries 39292Mean x 0.9921Mean y 0.08867RMS x 0.09693RMS y 0.06117

> 5 GeVT with E-Correlation for truthmatched e

e55/e370 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0η ∆

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2 > 5 GeVT with E-Correlation for duplicate e Delta_eta0_vs_ratio_dup

Entries 496Mean x 0.9568Mean y 0.03153RMS x 0.1423RMS y 0.0294

> 5 GeVT with E-Correlation for duplicate e

Duplicate clusters statistics

March 25, 2009 13 / 15

• We used Egamma and Overlap electrons only (Author==1 || Author==3)

• We obtain 2.4 % of duplicates for J/ψ (86 452 events)

◦ 95 % of all duplicates share the same track

◦ For ET > 5 GeV, we get 7.6 h duplicates with 85 % share the same track

• We obtain 0.8 h of duplicates for Zee validation sample in 14.5.1.2 (10 000events)

◦ 13 pair of duplicates for the whole sample

◦ 12/26 clusters share the same track

◦ 0 duplicate when requiring ET > 20 GeV

◦ 24/26 duplicate clusters have φ < 0

Possible improvements

March 25, 2009 14 / 15

• Solutions were suggested by Guillaume Unal to improve the situation:

◦ Use η0-φ0 (or η-φ) distance to remove duplicates

• Quick and easy check was to put back CaloClusterRemoveDuplicatesTool (from M. Cooke) for Cluster Correction

• Was tried on J/ψ sample with encouraging results, still need to bevalidated on other e/gamma samples

• After running CaloClusterRemoveDuplicates, we obtain 1,7 h ofduplicates for Jpsi sample in 14.5.0 (99 952 events)

◦ 34 % of all duplicates share the same track◦ For ET > 5 GeV, we get 1,2 h duplicates with 32 % share the

same track

◦ Try using different cluster size as seeds (instead of 5x5)

• Using 3x7 seeds for 3x7 clusters?• Using 3x5 seeds for all clusters?• Lydia Iconomidou-Fayard agreed to look into this.• This could solve the problem of energy miscalculation in η

Conclusions

March 25, 2009 15 / 15

• Energy miscalculation going from 5x5 to 3x7 for duplicate clusters

◦ Only a problem for low energy clusters (ET < 5 GeV)

◦ Could be solved be a new iteration of initial correction

• Good fraction of duplicate clusters are less than 2 cells apart

◦ Explained since SW duplicate deletion is not done using η0,φ0

◦ First look at CaloClusterRemoveDuplicates shows that this toolcan be used successfully to remove those duplicates(preliminarily, it removed mainly the off-centered clusters)

• Still some work to be done

◦ Need to check which duplicates were removed byCaloClusterRemoveDuplicates

◦ Verify if we can use this tool for standard reco

◦ Look at the effect of modifying cluster sizes