Presentation is loading. Please wait.

Presentation is loading. Please wait.

Higgs Detection Sensitivity from GGF H  WW Hai-Jun Yang University of Michigan, Ann Arbor ATLAS Higgs Meeting October 3, 2008.

Similar presentations


Presentation on theme: "Higgs Detection Sensitivity from GGF H  WW Hai-Jun Yang University of Michigan, Ann Arbor ATLAS Higgs Meeting October 3, 2008."— Presentation transcript:

1 Higgs Detection Sensitivity from GGF H  WW Hai-Jun Yang University of Michigan, Ann Arbor ATLAS Higgs Meeting October 3, 2008

2 10/03/2008H. Yang - GGF H->WW2 2 Outline Introduction Monte Carlo Samples for H  WW study Cut-based analysis Boosted Decision Trees analysis H  WW detection sensitivity Summary

3 10/03/2008H. Yang - GGF H->WW3 Introduction This study is part of the effort at the University of Michigan to contribute to the HG4 CSC note. Our studies have used both cut-based analysis and Boosted Decision Trees technique. Major results are summarized in [H. Yang et.al., ATL-COM-PHYS-2008-023] This talk will focus on studies using the H  WW events from gluon-gluon-fusion process. UM contributors for this work Hai-Jun Yang, Tiesheng Dai, Dan Levin, Xuefei Li, Alan Wilson, Zhengguo Zhao, Bing Zhou H. Yang - GGF H->WW3

4 10/03/2008H. Yang - GGF H->WW4 4 MC Higgs Signal Used in Study (ATLAS software rel. v12) Pythia Generator (Gluon-Gluon Fusion) H  WW  e e, , e  There is no official PYTHIA ggF H  WW sample with v12.0.6.4 up Above Higgs samples were produced at UM using jobOptions similar to official jobOption DS5320 (with diff. M H and separate the ggF and VBF production) UM Pythia Higgs samples were compared to Higgs dataset 5320 by separating the ggF and the VBF events, they are in good agreement. UM samples are available at BNL Tier-1 center. GGF H  WWDataset #MC Events  × BR (fb) M H = 150 GeV301097400767 M H = 165 GeV302596200866 M H = 170 GeV5329167200825 M H = 175 GeV3035193450770 M H = 180 GeV304096250716

5 10/03/2008H. Yang - GGF H->WW5 5 MC Backgrounds Used in Study (SM samples were used for ATLAS diboson CSC note) BackgroundsDataset #MC Events  × BR (fb) qq  WW gg  WW 2821 – 2829 5921 – 5929 210 K 370 K 12503 648 ttbar5200529 K4.6E5 WZ5941, 5971281 K688 W + X: W  ln W+Jets(E>80) 5250 – 5255 4288, 4289 5.25 M 595 K 5.75E7 5.62E7 1.3E6 Z + X: ZZ Drell-Yan Z+Jets(E>80) Zbb 6356, 5980 4295 - 4297 4293, 4294 5175 – 5177 181 K 10.5 M 597 K 200 K 6.9E6 84 6.8E6 52800 48720

6 10/03/2008H. Yang - GGF H->WW6 6 Event Pre-selection for H  WW  l l Two leptons with opposite charges; each lepton with P T > 10 GeV Missing E T > 15 GeV Events must pass one of lepton trigger requirements: 2E10, 2MU6, E25I, MU20 Physics objects: –Electron ID based on likelihood ratio > 0.6 –Muon ID based on Staco algorithm –Jet class: C4TopoJet (E T > 20 GeV)

7 10/03/2008H. Yang - GGF H->WW7 7 Detection Sensitivity Studies Based on Pre-selected Events Cut-based analysis –Optimize the straight cuts for better sensitivity Analysis based on Boosted Decision Trees (BDT) Consider two leptons with 0-jet and 1-jet events Results from cut-based and BDT analyses

8 10/03/2008H. Yang - GGF H->WW8 8 Select H  WW  l l with Straight Cuts Pt (l) > 20 GeV; Max (Pt(l1),Pt(l2)) > 25 GeV Lepton Isolation –In R=0.4 cone,  Pt(  ) < 5 GeV –In R=0.4 cone,  Pt(e) < 8 GeV MET > 50 GeV N jet (Et>20 GeV) = 0 or 1  l1,l2) < 1.0 12 < M(l1,l2) < 50 GeV

9 10/03/2008H. Yang - GGF H->WW9 9 Some Variable Distributions After Pre-selection

10 10/03/2008H. Yang - GGF H->WW10H. Yang - GGF H->WW10 Some Variable Distributions After Pre-selection

11 10/03/2008H. Yang - GGF H->WW11H. Yang - GGF H->WW11 Invariant Mass of two leptons (applied all cuts except M ll cut)

12 10/03/2008H. Yang - GGF H->WW12 Results from Cut-based Analysis (1/fb) H  WW  l l Events / fb M H =150 GeV M H =165 GeV M H =170 GeV M H =175 GeV M H =180 GeV Bkgd Cuts (e  + 0 jet) 18.833.328.524.919.764.2 Cuts (e  + 1 jet) 12.425.220.317.814.976.8 Cuts (e  ) 31.258.548.842.734.6141.0 Cuts (ee + 0 jet) 6.311.39.98.16.880.6 Cuts (ee + 1 jet) 4.39.07.96.45.338.7 Cuts (ee) 10.620.317.814.412.1119.3 Cuts (  + 0 jet) 10.118.515.713.310.333.3 Cuts (  + 1 jet) 7.013.311.210.48.758.4 Cuts (  ) 17.131.826.923.719.091.7 Cuts (ee+  +e  ) 58.9110.693.580.865.7352.0

13 10/03/2008H. Yang - GGF H->WW13H. Yang - GGF H->WW13 BDT Analysis (H. Yang et.al., ATL-COM-PHYS-2008-023) Signal for Training: PYTHIA Gluon-Gluon fusion H  WW Backgrounds for Training: WW, ttbar, WZ, W+X and Z+X Input variables for training: E T > 20 GeV BDT Ref: H. Yang et.al. NIM A555 (2005)370

14 10/03/2008H. Yang - GGF H->WW14H. Yang - GGF H->WW14 BDT Discriminator BDT discriminator is the total score of the BDT output as shown in left plot. Event Selection: 1)For 0-jet events: BDT >=200 2)For 1-jet events: BDT >=220 Detection sensitivity is defined as Significance = N S /√N B (With or without systematic error)

15 10/03/2008H. Yang - GGF H->WW15 Results (1/fb): Straight Cuts vs BDT Cut-basedBDT-based H. Yang - GGF H->WW

16 10/03/2008H. Yang - GGF H->WW16H. Yang - GGF H->WW16 Results (1/fb): Straight Cuts vs BDT Cut-basedBDT-based

17 10/03/2008H. Yang - GGF H->WW17H. Yang - GGF H->WW17 Results (1/fb): Straight Cuts vs BDT Cut-basedBDT-based

18 10/03/2008H. Yang - GGF H->WW18H. Yang - GGF H->WW18 BDT Results: H  WW  e  (1/fb) H  WW  e  Events / fb M H =150 GeV M H =165 GeV M H =170 GeV M H =175 GeV M H =180 GeV Bkgd  pre-sel (fb) 169.6210.1196.4194.0180.038143 BDT ( 0 jet) 22.545.141.036.629.453.6 BDT ( 1 jet) 9.321.819.216.413.316.3 BDT ( 0 jet+1 jet) 31.867.060.253.042.769.8 Cuts (0 jet) 18.833.328.524.919.764.2 Cuts (1 jet) 12.425.220.317.814.976.8 Cuts (0 jet+1 jet) 31.258.548.842.734.6141.0

19 10/03/2008H. Yang - GGF H->WW19H. Yang - GGF H->WW19 H  WW   1/fb  H  WW   Events / fb M H =150 GeV M H =165 GeV M H =170 GeV M H =175 GeV M H =180 GeV Bkgd  pre-sel (fb) 94.6117.0103.496.486.844359 BDT ( 0 jet) 13.225.322.820.617.139.1 BDT ( 1 jet) 7.916.313.111.48.419.3 BDT ( 0 jet+1 jet) 21.141.635.932.025.558.4 Cuts (0 jet) 10.118.515.713.310.333.3 Cuts (1 jet) 7.013.311.210.48.758.4 Cuts (0 jet+1 jet) 17.131.826.923.719.091.7

20 10/03/2008H. Yang - GGF H->WW20H. Yang - GGF H->WW20 H  WW  e e  1/fb  H  WW  e e Events / fb M H =150 GeV M H =165 GeV M H =170 GeV M H =175 GeV M H =180 GeV Bkgd  pre-sel (fb) 58.171.084.464.462.4150156 BDT ( 0 jet) 11.217.816.715.114.256.8 BDT ( 1 jet) 6.312.811.09.27.833.2 BDT ( 0 jet+1 jet) 17.530.627.724.322.090.0 Cuts (0 jet) 6.311.39.98.16.880.6 Cuts (1 jet) 4.39.07.96.45.338.7 Cuts (0 jet+1 jet) 10.620.317.814.412.1119.3

21 10/03/2008H. Yang - GGF H->WW21H. Yang - GGF H->WW21 H  WW  l l Selection Statistical Sensitivity (1/fb) GGF H  WW N s /  N b (1/fb) M H =150 GeV M H =165 GeV M H =170 GeV M H =175 GeV M H =180 GeV Cuts (e  ) 2.64.94.13.62.9 Cuts (  ) 1.83.32.82.52.0 Cuts (ee)1.01.91.61.31.1 BDT (e  ) 3.88.07.26.35.1 BDT (  ) 2.85.44.74.23.3 BDT (ee)1.83.22.92.62.3

22 10/03/2008H. Yang - GGF H->WW22H. Yang - GGF H->WW22 H  WW  l l Selection Statistical Sensitivity (1/fb) GGF H  WW Events / fb M H =150 GeV M H =165 GeV M H =170 GeV M H =175 GeV M H =180 GeV Bkgd Cuts (ee+  +e  ) Efficiency 58.9 7.7% 110.6 12.8% 93.5 11.3% 80.8 10.5% 65.7 9.2% 352.0 N s /  N b (no syst) Cuts (ee+  +e  ) 3.15.95.04.33.5N/A BDT (ee+  +e  ) Efficiency 70.4 9.2% 139.2 16.1% 123.8 15.0% 109.3 14.2% 90.2 12.6% 218.2 N s /  N b (no syst) BDT (ee+  +e  ) 4.89.48.47.46.1N/A

23 10/03/2008H. Yang - GGF H->WW23 Systematic Uncertainties 6.5% Luminosity uncertainty (ref. Tevatron) 5% Parton Density Function uncertainty 3% Lepton identification acceptance uncertainty 5% Energy scale uncertainty (3% on lepton energy and 10% on hadronic energy) 6% BDT training uncertainty due to energy scale uncertainty and imperfect MC cross section estimation of major backgrounds) 15% background estimation uncertainty due to limited MC data sample statistics (W/Z+X)  The total systematic uncertainty from above sources is 19%. We use conservative systematic error 20% for Higgs detection significance estimation.

24 10/03/2008H. Yang - GGF H->WW24 Efficiency Change due to Uncertainties of Background Cross Sections To estimate systematic uncertainty caused by BDT training with imperfect MC background cross sections estimation, cross sections of main backgrounds (ww, tt) are changed by ±20% for BDT training. The relative change of background with fixed signal efficiency are listed in the table. Relative change of background H  WW (e  H  WW (  H  WW (e e   WW +20% 4.6%2.0%2.3%  WW - 20% 6.8% 8.4%  ttbar +20% 2.4%4.0%3.1%  ttbar - 20% 5.7%1.1%1.2%

25 10/03/2008H. Yang - GGF H->WW25 Uncertainty from lepton and Jet Energy Scale and Resolution To estimate the systematic uncertainty, all energy-dependent variables in testing samples are modified by adding additional energy uncertainty, 3% for lepton and 10% for jets. The relative changes of signal and background efficiencies are calculated by using same BDT cut. Relative change H  WW (e  H  WW (  H  WW (e e  Signal (resolution) <0.1%0.1%<0.1% Signal (Scale) 1.1%1.7%2.6% Background (resolution) 0.4%0.9%0.4% Background (Scale) 3.1%2.0%5.6%

26 10/03/2008H. Yang - GGF H->WW26H. Yang - GGF H->WW26 H  WW Detection Sensitivity (1/fb, with 20% systematic error) GGF H  WW Events / fb M H =150 GeV M H =165 GeV M H =170 GeV M H =175 GeV M H =180 GeV Bkgd Cuts (ee+  +e  ) 58.9110.693.580.865.7352.0 N s /  N b +(0.2*N b ) 2 Cuts (ee+  +e  ) 0.81.51.31.10.9N/A N s /  N b +(0.2*N b ) 2 Cuts (e  ) 1.01.91.61.41.1N/A BDT (ee+  +e  ) 70.4139.2123.8109.390.2218.2 N s /  N b +(0.2*N b ) 2 BDT (ee+  +e  ) 1.53.02.72.42.0N/A N s /  N b +(0.2*N b ) 2 BDT (e  ) 2.04.13.73.32.6N/A

27 10/03/2008H. Yang - GGF H->WW27H. Yang - GGF H->WW27 Summary Gluon-gluon fusion H  WW with three leptonic decay final states produced by Pythia MC generator are studied using large background samples. The BDT is a very useful analysis tool to improve the Higgs detection sensitivity. H  WW channel could be a promising discovery channel in early LHC runs. It is crucial to control systematic uncertainties for a ‘counting’ experiment. Studies on VBF H  WW analysis with 1 or 2 jets in events will be performed.

28 10/03/2008H. Yang - GGF H->WW28H. Yang - GGF H->WW28 Backup slides

29 10/03/2008H. Yang - GGF H->WW29H. Yang - GGF H->WW29 Results (1/fb): Straight Cuts vs BDT

30 10/03/2008H. Yang - GGF H->WW30H. Yang - GGF H->WW30 Results (1/fb): Straight Cuts vs BDT

31 10/03/2008H. Yang - GGF H->WW31H. Yang - GGF H->WW31 H  WW  e e

32 10/03/2008H. Yang - GGF H->WW32H. Yang - GGF H->WW32 H  WW  

33 10/03/2008H. Yang - GGF H->WW33H. Yang - GGF H->WW33 H  WW  e 

34 Detection Sensitivity Comparison using PYTHIA and MC@NLO Signal efficiencies with PYTHIA are higher than that with MC@NLO by more than a factor of 2 in both straight-cut and BDT analysis!

35 More Comparisons Using Similar HG4 CSC note cuts: Pt (l) > 15 GeV, |  l | < 2.5 12 < M(l1,l2) < 300 GeV  l1,l2) < 1.575 MET > 30 GeV N jet (Et>20 GeV) = 0 In R=0.4 cone,  Pt(  ) < 5 GeV In R=0.4 cone,  Pt(e) < 8 GeV

36 Using HG4 Cuts, Normalize to 1/fb integrated luminosity M Higgs (GeV) Pythia N precut Pythia N s Pythia Eff s Pythia N s /N bg Pythia N s /  N bg 150169.642.124.8%0.071.73 165210.161.629.3%0.112.54 170196.454.927.9%0.092.26 175194.050.826.2%0.092.10 180180.043.924.4%0.081.81 M Higgs (GeV) MC@NLO N precut MC@NLO N s MC@NLO Eff s MC@NLO N s /N bg MC@NLO N s /  N bg 150168.232.719.5%0.061.35 165210.347.622.6%0.081.96 170160.930.418.9%0.051.26 175190.038.320.1%0.071.58 180178.334.319.3%0.061.42 BG total WWTtbarWZW+XZ+X 39802.51203.614097.7126.311702.412696.8 588.5142.010.513.5401.121.4

37 10/03/2008H. Yang - GGF H->WW37H. Yang - GGF H->WW37 “A procedure that combines many weak classifiers to form a powerful committee” Boosted Decision Trees H. Yang et.al. NIM A555 (2005)370, NIM A543 (2005)577, NIM A574(2007) 342  Relatively new in HEP – MiniBooNE, BaBar, D0(single top discovery), ATLAS  Advantages: robust, understand ‘powerful’ variables, relatively transparent, … BDT Training Process Split data recursively based on input variables until a stopping criterion is reached (e.g. purity, too few events) Every event ends up in a “signal” or a “background” leaf Misclassified events will be given larger weight in the next decision tree (boosting)

38 10/03/2008H. Yang - GGF H->WW38H. Yang - GGF H->WW38 A set of decision trees can be developed, each re-weighting the events to enhance identification of backgrounds misidentified by earlier trees (“boosting”) For each tree, the data event is assigned +1 if it is identified as signal, - 1 if it is identified as background. The total for all trees is combined into a “score” negativepositiveBackground-likesignal-like BDT discriminator


Download ppt "Higgs Detection Sensitivity from GGF H  WW Hai-Jun Yang University of Michigan, Ann Arbor ATLAS Higgs Meeting October 3, 2008."

Similar presentations


Ads by Google