Photon ID MVA inclusive shower shapeisolation MVA and

Photon ID MVA: inclusive (shower shape+isolation) MVA and separated MVAs Guoming Chen, Jiawei Fan, Yuqiao Shen, Junquan Tao, Hong Xiao IHEP Beijing O. Bondu, H. Brun, S. Gascon-Shotkin, M. Lethuillier, L. Sgandurra IPN-Lyon Nicolas Chanon ETH Zuerich 23. 04. 2012 1

Outline Ø Update of our MVA analysis in Globe J. Tao, S. Gascon-Shotkin , H gg meeting 26 Mar 12, https: //indico. cern. ch/get. File. py/access? contrib. Id=35&session. Id=3&res. Id=1&material. Id=slides&conf. Id=183504 Debug on the limit calculation; New inclusive photon. ID mva with additional shower shape variables; Diphoton MVA with new photon. ID inputs Ø Separated Photon ID MVA: Pileup shower shape variables (SSV) mva and isolation variables mva Ø Summary 2

Repeat H→γγ MVA in globe: Limit Agreement! Repeat limit: same as Nick’s Data: still Nov 30 Rereco Nick’s limit Previous limit with option: --r. Min=0. 2 --r. Max=4 How did this affect the result? From Nick, “Synchronization Report “, Hgg meeting, 20 February 2012 https: //indico. cern. ch/get. File. py/access? contrib. Id=27&session. Id=5&r es. Id=2&material. Id=slides&conf. Id=178898 3

Photon. ID MVA: preselection and samples Ø Preselection (same as MIT in Globe) cuts are never looser than the trigger cuts and than the MC “EM-enrichement”filter. Ø Samples Signal: prompt photon from 121 Ge. V Higgs (gluon fusion Fall 11 -PU_S 6) Background: Non prompt “photon” from double EM enriched photon jets (Fall 11 PU_S 6) ; Matching failed Ø Method: BDT with same configuration as MIT (from the default weight files) Ø PU reweighted MVA training ( without SCeta reweighting, AN 2012 -048 ) half training and half test 4

Photon. ID MVA comparisons EB CASE: 1) MIT default ( 5 shower + 5 iso +nvtx+SCeta) 2) 8 shower shape variables (NN) Emax/ESCraw, E 2 x 2/E 5 x 5, l-/l+, CEP, Etawidth, ’Brem’, l-/CEE “AND” R 19 (instead of r 9, etawidth, phiwidth in MIT) + ���� η +H/E + 5 iso+ nvtx+ Sceta (17) EB 2): 80. 7% 1): 77. 7% (MIT) ~3% improvement (4% rel)in EB at 90% sig efficiency, compared (2) with (1) EE CASE: 1) MIT default (12) 2) 8 shower shape variables same as EB (instead of r 9, etawidth, phiwidth in MIT) + ���� η +H/E + 5 iso+ nvtx+ Sceta (17) + ESEff. Sig. RR (next slide) (18) EE 2): 76. 5% w ES 2): 75. 3 wo ES 1): 74. 7% (MIT) ~<1% /2%(ES) improvement (1%/2% rel) in EE at 90% sig efficiency, compared (2) with (1) 5

ES variables: Effective σRR Distributions From Chia-Ming Kuo, Yu-Wei Chang et al. Yu-Wei Chang , Hgg weekly meeting, Aug 8, 2011, https: //indico. cern. ch/get. File. py/access? contrib. Id=11&session. Id=6&res. Id=0&material. Id=slides&conf. Id=150032 v Based on expected ES strip (extrapolated from the position of EE super cluster) v σXX : standard deviation of the ES Rec. Hit distribution. v Eff. Shower Shape (x plane) v ES Shower Shape � For EE region with | |>1. 566 ES only with | |>1. 65 Initial value 0. 0 6

MC rescaling of ‘new’ Photon. ID MVA input variables Usual G 4 simulation problem with Fall 11 MC samples EE after rescaling EB after rescaling S 4 Ratio *= 1. 0055(EB)/1. 0085 (EE) Lambdaratio *= 0. 99(EB) Etawidth *= 0. 98(EB) ESEff. Sig. RR *= 1. 04(EE) H. Brun, https: //indico. cern. ch/get. File. py/access? contrib. Id=1&session. Id=4&res Id=0&material. Id=slides&conf. Id=175788 J. Fan, EGM working meeting 22 sep 11, data-MC comparisons and proposal of a possible workaround fix to sim. Prob. https: //indico. cern. ch/get. File. py/access? contrib. Id=3&res. Id=0&materi al. Id=slides&conf. Id=156142 Many with satisfactory agreement 7

Diphoton Preselection and samples Ø Diphoton Preselection cuts are never looser than the trigger cuts and than the MC “EM-enrichement”filter. AND both Photo. ID MVA (MIT) > -0. 3 Ø Samples Signal: 123 Ge. V Higgs (gluon fusion Fall 11 -PU_S 6) Background: Fall 2011 MC backgrounds Ø Variables (10): photon ID MVA 1(2) event vertex probability Ø Method: BDTG with same configuration as MIT (from the default weight files) Ø MC signal reweighted MVA training 8

Diphoton MVA Signal: gg. H 121 Ge. V Higgs Background: MC background Keeping 90% signal efficiency: Black with our Photon. ID: 54. 5% Red with MIT Photon. ID: 52. 8% ~1. 7% improvement (~3. 2% relative) at 90% sig efficiency Diphoton Preselection, Pt 1>mass/3, Pt 2>mass/4, 100<mass<180 9

Diphoton MVA Data/MC MIT previous weight file (Nov 30) Red: MIT Black: Ours Data/MC Our new weight file Keeping the same Nevt of MC signal : Nmcsig (gg. H 125) Data MIT mva > 0. 05 42850 20413 Our mva > -0. 06 42851 20132 ( -1. 4% ) 10

Outline Ø Update of our MVA analysis in Globe Debug on the limit calculation; New inclusive photon. ID mva with additional shower shape variables; Diphoton MVA with new photon. ID inputs Ø Separated Photon ID MVA: Pileup shower shape variables (SSV) mva and isolation variables mva Ø Summary 11

Separated MVA– Shower shape variables MVA && isolation MVA : pileup Ø Shower shape variables: little effect from pileup Data/MC Emax/Escraw in EB Same area Profile with Nvtx Ø Isolation variables: affected a lot from pileup Data/MC Rel. trk iso in EB after trigger and preselection Profile with Nvtx 12

MVA inputs Ø Shower shape variables MVA (SSV): SSV § (MIT Default SSV): ���� η , R 9, SC η width, SC ϕ width (H/E affected by pileup) Old H/E § additional variables (NN-8): Emax/ESCraw, E 2 x 2/E 5 x 5, l-/l+, CEP, Etawidth, ’brem’, l-/CEE, R 19 § combined: η�� η , R 9 + Smaj combined NN-8 + ���� + ESEff. Sigma. RR(EE) EB-11/EE-12 variables Ø Isolation MVA (try) § EG isolation -- Absolute CIC Iso(without factor 50/PT): combined Iso (selected vertex), combined Iso (worst), track Iso (selected vertex) + PUCorr Hcal. Iso=Hcal. Iso(ΔR<0. 4)-0. 17*Rho + PUCorr Ecal. Iso=Ecal. Iso(ΔR<0. 3)-0. 17*Rho; +H/E § PF isolation: sumiso (selected vertex, ΔR<0. 3 ) , sumiso (worst, 0. 3), chargediso (pv, 0. 3), chargediso (worst, 0. 3), neutraliso (0. 3), photoniso D. Franci, S. Rahatlou, and D. del Re, CMS Note 2008/075 (2008) 13

Photon. ID: Combined SSV MVA inputs in EB (11) 14

Photon. ID: Combined SSV MVA inputs in EE (12) ES 15

SSV Photon. ID mva: inputs linear correlations EB EE 16

Photon. ID: SSV MVA EB case: keeping 90% sig EE case: keeping 90% sig Combined : NN SS (8) : Default (4): Combined +ES: 69. 1% Combined (wo ES): 67. 8% +2% NN SS (8): 66. 3% (+3%rel) Default (4): 67. 1% (Sig. Ie) 73. 9% 72. 1% 69. 5% +4. 4% (+6. 3%rel) With MIT preselection 17

SSV Photon. ID: combined MVA Data/MC Leading EBEB Subleading EE EE 18

Photon. ID: ISO MVA EB: EG EE: PF EG PF EB case: keeping 90% sig EE case: keeping 90% sig PF iso : EG iso: 50. 7% 48. 7% With MIT preselection 51. 2% 50. 7% 19

Photon. ID: ISO EG MVA Data/MC Leading EB Subleading EE 20

SSV Photon. ID MVA: Pt, Eta and Nvtx --EB Flat EB case: much flat with pileup (Nvtx) 21

SSV Photon. ID MVA: Pt, Eta and Nvtx --EE EE case: not so much effect from pileup 22

EG-ISO Photon. ID MVA: Pt, Eta and Nvtx --EB EB case: affect from pileup 23

EG-ISO Photon. ID MVA: Pt, Eta and Nvtx --EE EE case: also affect from pileup 24

Photon. ID MVA Summary Ø Inclusive Photon. ID MVA v ~3% (4% rel)/ 2%(2% rel) improvement in EB/EE, when keeping 90% prompt photon signal efficiency. v ~1. 7% improvement (~3. 2% relative) at 90% sig efficiency for Diphoton MVA, using the new inclusive photon. ID instead of MIT photon. ID. v The data (Bkg) was reduced by ~1. 4% when keeping the same signal (MC) efficiency, for the final selected events. Ø Separated Photon ID MVA 1) SSV mva: § Combined variables can give ~4. 4% (6. 3%rel)/2% (3%rel) improvement in EB/EE, compared to the 4 MIT SSVs. § Less affected from pileup 2) ISO mva: § Tried sub-detector based EG iso and PF iso: almost identical results § Affected a lot from pileup 25 § PF iso: will apply the rho-correction

Backup slide 26

Inclusive Photon. ID MVA inputs correlations - EB 27

Preselection+MVA VS Ci. C: EB Signal: single photon from Fall 11 gg. H 121 samples Background: non-prompt photon from GJet (Fall 11) Efficiency: Nphoton pass cut/Nphoton with acceptance Compare with the same selections and same signal efficiency: redo with MIT previous weight file MIT(mva Pre. Sel+M Ci. C eff. >-0. 0403) VA eff. (%) Pre. Sel+ MVA/Ci. C Ours(mva Pre. Sel+MV Ci. C eff. >-0. 0331) A eff. (%) Pre. Sel+M VA/Ci. C Sig 83. 61 ~100 Sig 83. 62 83. 61 ~100 Bkg 4. 94 5. 65 87. 4 Bkg 4. 20 5. 65 74. 4 gg. H 121 same signal efficiency 28

Preselection+MVA VS Ci. C: EE Compare with the same selections MIT(mva Pre. Sel+M Ci. C eff. >0. 0135) VA eff. (%) Pre. Sel+ MVA/Ci. C Ours(mva Pre. Sel+MV Ci. C eff. >0. 0078) A eff. (%) Pre. Sel+M VA/Ci. C Sig 61. 94 61. 93 ~100 Sig 61. 95 61. 93 ~100 Bkg 2. 60 3. 86 67. 3 Bkg 2. 12 3. 86 54. 9 gg. H 121 29

Compare to Mingming’s result Different selection from Mingming Yang ( Feb 9, 2012, Egamma meeting) 30

Separated Photon. ID: SSV MVA & ISO MVA TO DO LIST Ø Inputs for Diphoton MVA : SSV + ISO + other variables OTHER CONSIDERATION Ø SSV mva: xstal (seed) timing Ø ISO mva: Frixione isolation Theory 31

Photon variables versus Pt, Supercluster Eta and number of vertexes in EB 32

Shower shape variables: R 9 in EB 33

Shower shape variables: R 19 in EB 34

Shower shape variables: Emax /ESCraw in EB 35

Shower shape variables: S 4 ratio in EB 36

Shower shape variables: i i in EB Trigger: 0. 014 37

Shower shape variables: Cep in EB 38

Shower shape variables: φwidth in EB 39

Shower shape variables: width in EB 40

Shower shape variables: brem in EB 41

Shower shape variables: λratio in EB 42

Shower shape variables: λ-/Cee in EB 43

Shower shape variables: Smaj in EB 44

Isolation variables: H/E in EB 45

Isolation variables: Sum/ET (pv) in EB 46

Isolation variables: Sum/ET (worst) in EB 47

Isolation variables: Track. PT/ET (pv) in EB Rel. trk iso in EB after trigger and preselection 48

Isolation variables: Track. PT (pv, 0. 3) in EB 49

Isolation variables: ECAL (0. 3) in EB 50

Isolation variables: HCAL (0. 3) in EB 51

Isolation variables: pf sum. ET (0. 3, pv) in EB 52

Isolation variables: pf sum. ET (0. 3, worst) in EB 53

Isolation variables: pf neutral (0. 3) in EB 54

Isolation variables: pf gamma (0. 02 -0. 3) in EB 55

Isolation variables: pf charged (0. 3, pv) in EB 56

Isolation variables: pf charged (0. 3, worst) in EB 57