1. Trigger studies for GMSB with photons (Internal note for approval) Shilei Zang Bernadette Heyburn, Uriel Nauenberg University of Colorado, Boulder TSG Meeting, 19th Mar. 2008

2. 2 Outline GMSB with photons Signal and background samples Efficiency and rate for default triggers A new method to optimize triggers Optimization of 3 triggers Results Compare rate with other studies Summary

3. 3   p p q q q q … … jet GMSB with photons Gauge Mediated Supersymmetry Breaking models NLSP (neutralino)  LSP (gravitino) + photon Prompt decay (ctau=0) high pT photons large MET due to gravitinos multi-jets  Experimental signature

4. 4 GMSB parameters Λ: scale of the SUSY beraking M: messenger mass scale tanβ: the ratio of the Higgs vev N 5 : number of messengers sign(μ): the sign of Higgsino mass term C grav : sets NLSP lifetime

5. 5 CSA07 samples of GMSB photons to estimate the signal efficiency. GEN-SIM: 1_4_X; DIGI-RAW: 1_6_7. 100k for each point: GM1b, GM1c, GM1e, GM1f, GM1g 1_6_0-PreCSA07 (or 1_6_7-CSA07) samples are used to estimate the background rates, which include: Photon jets (all pt bin) (CMSSW_160-PreCSA07) QCD jets (all pt bin) (CMSSW_160-PreCSA07) W  enu, Z  ee (CMSSW_167-CSA07) Totally processed about15 million events to minimize the error of rate.  GMSB signal samples:  Background samples:  The study have been done under 1_6_0, and optimized for start-up luminosity of 10 32 cm -2 s -1.  HLTriggerOffiline/Egamma package is used for the study.

6. 6 6 paths with thresholds Single Photon Relaxed Single Photon Double Photon Relaxed Double Photon EMHighEt (modified-HighEt) EMVery HighEt Et (GeV) >30>40>20 >80>200 Iso-Ecal (GeV) <1.5 <2.5 <5. Iso-Hcal (Barrel) <6. <8. H(ΔR<0.15)/Et<5% && H(0.15<ΔR<0.3)<8. (modified: IHcal<12.) Iso-Hcal (Endcaps) <4. <6. Iso-track<1 <3 <4 HLT paths for photons

7. 7 QCD Jets N NPhoton Jets NBkg N 0_15 295,613_47088,0860_15500,000Wenu205,707 15_201,255,976_60055,00015_20509,825Zee162,219 20_302,513,934_80021,97420_30606,680 30_502,416,441_100033,33030_50510,094 50_802,451,439_1400 5,29950_80169,741 _1201,161,823_1800_120164,000 _170 499,389_2200_170 69,993 _230 428,888_2600_300 24,993 _300 172,619_500 15,554 _380 82,998_7000 6,666 Number of bkg events processed for rate estimation

8. 8 Efficiency and rate for default triggers RDRD RSRS D S H VHVH S S, RS S, RS, D S, RS, D, RD S, RS, D, RD, H all

9. 9 Three triggers for GMSB photons: Optimize the two triggers: EMHighEt (H), and EMVeryHighEt (VH). Choose another one from: D, RD, RS, MET, or Jets Strategy:  How to optimize the triggers?

10. 10 Problem: How to optimize the trigger thresholds with figures of Efficiency vs. Rate in an objective way ? Optimize trigger thresholds Usually the cuts are determined by eye to give reasonable values of efficiency and rate. Threshold, how to set?

11. 11 Physics Analysis Selection criteria are optimized to maximize statistics ( Optimize relative error of BR; Significance; 90% CL limit, etc ) Selection criteria are optimized to minimize the mass uncertainty in mass measurement (e.g. top mass measurement) Artificially reduced the error of physical result! Not Really Blind !!

12. 12 N events, the amount of information : log 2 N. N is number of messengers; physical results are the meaning of information taken by such N messengers. For BR, number of messengers is the meaning of info.; For width, mass, …, meaning of info. is taken by the messengers; depends on the kinematics (not just on the number of events). Good property: log (xy)= log(x) + log(y). Information theory

13. 13 Amount of information: log(N S ), log(N B ) Signal efficiency ε and background efficiency b After the cut: log(N S ε), log(N B b) Reductions of information: -log(ε), -log(b) Ratio of the reductions: log(ε)/ log(b) the smaller log(ε)/ log(b), the better log(ε)/ log(b) b a (0< ε, b, a ≤1).  We can use statistics log(ε)/ log(b) to optimize trigger thresholds!  Good property: Blind Analysis! log(ε)/ log(b) depends on the amount of information; does not depend on the meaning of information.

14. log(ε)/ log(b) b a. a=1.0 a=0.7 a=0.5 a=0.3 a=0.2 a=0.1 a=0.05 a=0.02 a=0.01 ε = b a (1.,1.) (0.,0.) Trigger Study MVA b ε ε ε b 1-b K ID ε b

15. 15 How to deal such a problem in Physics Analysis?  Solution: log(ε)/ log(b) to optimize selections with final ε and b after the kinematics cut.  Our method will give worse physical results, but they are blind analysis and can be trusted.

16. 16 log(ε)/ log(b) vs. Cuts (default EMHighEt) 0.035 0.129 Min=0.129 Min=0.101 Min=0.0047 0.017 0.102 Et>80; Iecal<5; Ihcal<12; Itrack<4 Itrack is better than Iecal and Ihcal. Each figure is plotted with other cuts applied.

17. 17 log(ε)/ log(b) vs. Cuts (proposed EMHighEt) Et>60; Itrack<2 Itrack is better only when the Itrack cut point <5 0.022 Min=0.174 Min=0.190 0.068

18. 18 Et>40GeV Relaxed Single Photon candidates

19. 19 Propose two new triggers: p-EMHighEt (pH): Et>60GeV, Itrack<2 p-EMVeryHighEt (pVH): Et>120GeV Propose to use: pH, pVH, D for our physics. I.Isolation is useful at low Et region to suppress bkg, but bad in high Et region for our signal. II.Track isolation (cut position <6) is better than other isolaitons III.GMSB points with small Lambda parameter (GM1b) have more events with two signal photons at generator level, so the Double trigger is helpful for them.

20. 20 TriggerHpHVHpVHH, VHpH, pVH Rate (Hz)0.632.010.130.970.7532.888 EfficiencyGM1bGM1cGM1eGM1fGM1gRate (Hz) RS, H, VH81.3185.6889.5890.5291.103.510 pH, pVH, D87.7091.0293.2793.7893.963.139 pH, pVH, RS88.7292.2894.4494.7194.725.226 2.14 Hz

21. 21 efficiencyGM1bGM1cGM1eGM1fGM1gRate (Hz) pH, pVH, D87.7091.0293.2793.7893.963.139 pH, pVH, mD89.7792.2493.8194.1494.223.762 D: Et>20; Iecal<2.5; Ihcal<8 or 6; Itrack<3 (0.26 Hz) mD: Et>20; Itrack<3 (0.90 Hz) 0.64 Hz Further possible improvement It’s easy to reach 92% or 93% efficiency with 3.5 Hz, but difficult to reach 95% efficiency within 5.5 Hz! 98.6% events have signal photons at generator level; after SusyAnalyzer, only 93.5% events have reconstructed photons.

22. 22 Efficiency and Rates for each group of triggers 0: H, VH 1: H, VH, RS 2: H, VH, RD 3: H, VH, D 10: pH, pVH 11: pH, pVH, RS 12: pH, pVH, RD 13: pH, pVH, D 15: pH, pVH, mD pH, pVH, D pH, pVH, mD

23. 23

24. 24 Compare the rates with other studies in 13X and 16X Default Trigger Path SRSDRDHighEtVery HighEt Our Rate (160) (Hz) All bkg 8.96 ±0.24 2.93 ±0.09 0.26 ±0.05 1.90 ±0.14 0.63±0.01 (m-H) 0.13 ±0.00 13X exercise (Hz) All bkg 8.4 ±0.7 2.8 ±0.2 0.6 ±0.4 1.8 ±0.5 0.5 ±0.0 0.1 ±0.0 16X (Hz) (Matthias Mozer) w/o PhotonJets. 0.80.14 16X (Hz) (Aram Avetisyan) w/o PhotonJets; Fewer events. 10.06 ±4.79 5.82 ±3.39 0.04 ±0 6.69 ±4.61 0.43 ±0.13 0.07 ±0.01 Agree well Agree with error

25. 25 We give the rates of all HLT paths on photons, which are comparable with 13X exercise and other studies in 16X. Our rates have small errors. We propose to modify EM1HighEt and EM1VeryHighEt tirggers (basically loosen the isolation variables and Et). We propose to use 3 triggers for GMSB photons, which can reach ~93% efficiency with 3.2 Hz. We find a new method for trigger study, and physics analysis. Summary Thank you!  The draft of the note: http://www-hep.colorado.edu/~slzang/cmsnote_gmsb_trigger_v3.ps