Photon ID MVA Update for 2012 Guoming Chen

Photon. ID MVA Update for 2012 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 H→γγ meeting 23. 05. 2012 1

Outlines Ø Update from 21 May: bkg samples Ø Performance: inclusive EB/EE Ø Separate MVAs: SSV mva, ISO mva Ø Summary 2

Photon. ID MVA: preselection and samples Ø New pf-iso based preselection: charged PF ISO(0. 02 -0. 2)< 4 Ge. V Ø Samples Signal: all prompt photon from 123 Ge. V Higgs for training and 124 Ge. V for test (gluon fusion Summer 12_S 7) Summer 12_S 7 Background: Non prompt “photon” from double EM enriched photon jets (Summer 12_S 7, Pthat 40 AND Pthat 20 to 40, weighted by XS) XS ; not passing above matching for signal. Split model “Alternate”: odd for training and even for test Ø Photon candidates Ø Method: BDT same as 2011 configuration (NTrees=400, n. Events. Min=400, …) Ø Only Eta reweighting for the signal training candidates. 3

Photon Id inputs for 2012 3 pf ISO: pfchargedisogood 03 pfchargedisobad 03 pfphotoniso 03 6 common shower shape: sieie sieip etawidth phiwidth r 9 s 4 ratio + rho + SCeta -> EB (11) + ES -> EE (12) 4

Inclusive Performances: new EE+ES variable EB Bkg rej eff when keeping 0. 90 sig eff. Red: retrained with 2011 variable set Bkg rej eff when keeping 0. 90 sig eff. EE wo ES variable ~2. 8%/ ~3. 4% improvement in EB/EE, and ~1% improvement in EE without ES variable, with the new preselections. 5

MVA outputs EB: training and test EB: new and 2011 retrained EE: training and test EE: new and 2011 retrained Red/green: retrained with 2011 variable set (12) 6

MVA outputs validation: 2012 Zee Data/MC Zee samples, ~370/pb MC PU reweighting, normalized to Ndata No PT reweighting 2011 rescaling factors for shower shape Satisfied agreement 7

Separate MVA validations: 2012 Zee ISO: Barrel ISO: Endcap Zee Data/MC, ~370/pb 4 Variables: pfchargedisogood 03 pfchargedisobad 03 SSV: Barrel 6 Shower shape: sieie sieip etawidth phiwidth r 9 s 4 ratio + SCeta -> EB (7) + SCeta + ES -> EE (8) pfphotoniso 03 rho SSV: Endcap 8

Efficiency studies: Nvtx, PT, Eta inclusive EB/EE 9

Efficiency: preselection+MVA vs Ci. C 4 (2011) Signal: single photon from gg. H 120 samples Background: non-prompt photon from Gjet_40 Efficiency: Nphoton pass cut/Nphoton within acceptance Compare the bkg efficiency with Ci. C 4 when keeping the same signal efficiency EB: new EE: new (mva> 0. 005) Pre. Sel+MV A eff. (%) Ci. C 4 eff. Pre. Sel+MVA (%) VA/Ci. C (%) /Pre. Sel (%) (mva> 0. 036) Pre. Sel+MV A eff. (%) Ci. C 4 eff. (%) Pre. Sel+MV A/Ci. C (%) Pre. S+MVA /Pre. S (%) Sig 79. 33 79. 34 ~100 82. 8 Sig 55. 63 55. 68 ~100 59. 53 Bkg 3. 21 4. 63 69. 29 12. 68(rej. 87. 3) Bkg 1. 80 2. 85 62. 98 5. 83 (rej. 94. 2) EB: 2011 default (weight files) EE: 2011 default (mva> -0. 03) Pre. Sel+MV A eff. (%) Ci. C 4 eff. Pre. Sel+MVA (%) VA/Ci. C (%) /Pre. Sel (%) (mva> 0. 008) Pre. Sel+MV A eff. (%) Ci. C 4 eff. (%) Pre. Sel+MV A/Ci. C (%) Pre. S+MVA /Pre. S (%) Sig 79. 32 79. 34 ~100 82. 8 Sig 55. 67 55. 68 ~100 59. 57 Bkg 3. 92 4. 63 84. 55 15. 47 (rej. 84. 5) Bkg 1. 95 2. 85 68. 18 6. 31 (rej. 93. 7) Preselection eff. of sig. : 95. 80%/93. 45% for EB/EE Preselection eff. of bkg. : 25. 31%/30. 87% for EB/EE Comparable to ROCs 10

Efficiency with Nvtx: EB Signal Bkg. Black: Preselections efficiency Red: Ci. C 4 efficiency Blue: Preselections + mva cut (2011) Green: Preselections + mva cut (new) Same signal efficiency as Ci. C 4 from 120 Ge. V gg. H 11

Efficiency with Nvtx: EE Signal Bkg. Black: Preselections efficiecny Red: Ci. C 4 efficiency Blue: Preselections + mva cut (2011) Green: Preselections + mva cut (new) Same signal efficiency as Ci. C 4 from 120 Ge. V gg. H 12

Efficiency with PT: EB Signal Bkg. Black: Preselections efficiency Red: Ci. C 4 efficiency Blue: Preselections + mva cut (2011) Green: Preselections + mva cut (new) Same signal efficiency as Ci. C 4 from 120 Ge. V gg. H 13

Efficiency with PT: EE Signal Bkg. Black: Preselections efficiency Red: Ci. C 4 efficiency Blue: Preselections + mva cut (2011) Green: Preselections + mva cut (new) Same signal efficiency as Ci. C 4 from 120 Ge. V gg. H 14

Efficiency with Eta: EB Signal Bkg. Black: Preselections efficiency Red: Ci. C 4 efficiency Blue: Preselections + mva cut (2011) Green: Preselections + mva cut (new) Same signal efficiency as Ci. C 4 from 120 Ge. V gg. H 15

Efficiency with Eta: EE Signal Bkg. Black: Preselections efficiency Red: Ci. C 4 efficiency Blue: Preselections + mva cut (2011) Green: Preselections + mva cut (new) Same signal efficiency as Ci. C 4 from 120 Ge. V gg. H 16

Summary Ø Photon ID MVA for 2012 with the new PF-based preselections: PF Iso (3), more cluster shape variables (6); in total 11(EB)/12(EE) Some improvements than 2011 variables Weight files have been used privately in Globe, will update with new MC rescaling factors Validation with Zee, satisfied agreement between Data/MC Ø Separate MVAs, SSV+ISO. 17

Backup slides 18

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 Redone with loose preselections: R 9 *=1. 0035 (EB/EE) S 4 Ratio *= 1. 0055(EB)/1. 0085 (EE) Lambdaratio *= 0. 99(EB) Etawidth *= 0. 98(EB) Phiwidth *= 0. 99(EB/EE) ESEff. Sig. RR *= 1. 04(EE) likelihood ratio method, see from H. Brun, Weekly Offline Meetings , 27 February 2012 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&material. Id=slides&co nf. Id=156142 Many with satisfactory agreement 19

Inclusive Performances: 21. 05. 2012 EE+ES variable EB Bkg rej eff when keeping 0. 90 sig eff. Red: retrained with 12 MIT-variables Bkg rej eff when keeping 0. 90 sig eff. EE wo ES variable ~2. 9%/ ~1. 3% improvement in EB/EE, and <~1% improvement in EE without ES variable, with the new preselections. 20

PT EB EE Summer 12: 20 to 40, 40 toinf Fall 11: 20 toinf Different preselections 21

Ranking and “N+1” inclusive EB: because of our rule , Ieliminated "SCeta", pfchargedisobad 03 pfchargedisogood 03 sieie sieip etawidth pfphotoniso 03 phiwidth r 9 ratio. S 4 rho inclusive EE: because of out rule , eliminated "phiwidth", "sieip", sieie pfchargedisobad 03 pfchargedisogood 0 3 etawidth SCeta pfphotoniso 03 r 9 s 4 ratio ESEff. Sigma. RR rho Barrel Endcap 22