Heavy stop direct decay searches in final states with two leptons using TMVA

1. Heavy stop directdecaysearches in finalstateswithtwoleptonsusing TMVA M. Bianco, E. Gorini, L. Longo, M. Primavera, M. Reale, A. Ventura on behalfof the analysisgroupsof Lecce, Milano, Pavia, Texas, Udine

2. Updates wrt December 2011 • https://indico.cern.ch/getFile.py/access?contribId=7&resId=1&materialId=slides&confId=165089 • Simulation: from p621 to p832 • More signal points considered in grid • Updated cross sections • Trainings based on channels (ee,mumu,emu) and not on streams (Egamma,Muons) • Different set of input variables (from 6 to 12) • Relative systematic uncertainty on NB in sensitivity brought from 0.6 to 1.0

3. MC samples and preliminarycuts • In allplots the signalpointusedis: mstop= 300 GeV, mchi10 = 50 GeV(127849_p832) • MC background sourcesconsidered: • Z+jets, T1, diboson, Drell-Yan, Wt, W+jets • Stillmissing: {Z,T1,DY}+bb+jets, T1+{Z,W,WW} • Allresultsnormalizedto 4.71 fb-1 • Preselectioncuts: • Exactly 2 OS leptonswithusualpTcuts and mll> 20 GeV • At least 2 jetswithpT > 50, 20 GeV

4. Variable distributions, ee 1-6/12 ETmiss Meff MT2 mll bJet1 bJet2

5. Variable distributions, ee 7-12/12 (l1,l2) (l1,l2) (l1,ETmiss) (l2,ETmiss) (l2,j1) (l1,j1)

6. Additional requirements • Two possible sets of additional requirements are applied to improve TMVA performance: • “cut11”: strong cut against all SM: ETmiss > 50 GeV and Meff > 300 GeV or, alternatively, • “cut12”: Anti-Z cut on SF dileptons only: |mll – 91| < 20 GeV.

7. Unnormalizeddistributions, ee, cut12 Signal mstop=300 GeV, mchi10=50 GeV SM background 1-6/12 Similar plots for the case mumu

8. Unnormalizeddistributions, ee, cut12 Signal mstop=300 GeV, mchi10=50 GeV SM background 7-12/12 Similar plots for the case mumu

9. Other variables (not included) • Wehavealsocheckedmanyothervariables on SF and OFchannels, and foundthemnoteffectiveforsignal-backgroundseparation: • ETmiss/Meff, pTl2 , pTl1+pTl2 , pTj2 , pTj1+pTj2 , MCT , (j1,j2) , (j1,j2) , (j1,ETmiss) , (j2,ETmiss) , (l1,j2) , (l2,j2) , (l1,j1) , (l1,j2) , (l2,j1) , (l2,j2) . • In the previous set ofresults(December 2011) wewereconsidering 6 variables: • ETmiss, meff, pTj2, pTl2, mT2 , mll

10. Correlations among the 12 variables, ee • Meff and ETmissobviouslycorrelated • MT2hascorrelationwithlepton-basedvariables • (Anti)correlations on phi angles.

11. Unnormalizeddistributions, emu, cut12 Signal mstop=300 GeV, mchi10=50 GeV SM background 1-6/12

12. Unnormalizeddistributions, emu, cut12 Signal mstop=300 GeV, mchi10=50 GeV SM background 7-12/12

13. Correlations among the 12 variables, emu • Correlations are similar to what observed in the SF channels (a bit more evident in phi angles).

14. Some training details • Wetriedmanydifferentmethods in TMVA (Fisher, Likelihood, PDEFoam, HMatrix, BDT, MLP, SVM …). • As observedalso in manyother ATLAS analyses, BDTGlooks the best method: • Weoptimizeditsusewithparameters: NTrees = 1000, nCuts=20, NNodesMax=5. • Foreach set ofpreselectioncuts (cut11 and cut12), wehaveseparatelytrainedee, mumu and emu. • Weused the chosensignalpointagainst the full SM sample. By default, eventswith negative weightswereignored in the training (butwereusedby TMVA). • The “BDTG value” rangesfrom –1 (ideallyfor background only) to +1 (ideallyforsignalonly)

15. Proposed cuts for BDTG • Tochoose the cut foreachconfiguration/channel, wetook the BDTG valueforsignal and SM background MC sampleswhichmaximisessensitivity (in the chosensignalpoint), definedas: • Togetstable and reliablevalues, the maximizedquantityisactuallyS – S. • NBisaskedtobeobtainedfrom2 unweighted SM events. • The uncertainty on the cut valueisfoundtobeoforder0.1.

16. BDTG value distributions 0.130 -0.184 0.482 cut11, mumu cut11, ee cut11, emu 0.564 0.744 0.616 cut12, mumu cut12, ee cut12, emu

17. Sensitivity for cut11, SF and OF • Comparisonwithsignalsignificance in cut-basedanalysisisnowpossibleforbothSF and OF • cut11favours OF

18. # of events for cut11, SF and OF NB = 0.7611  0.1367 NB = 0.1091  0.0545 • SFhaslowersignificancethanOF owingto more surviving background events

19. Sensitivityfor cut12, SF and OF • Comparison with signal significance in cut-based analysis is now possible for both SF and OF • Results for SF and OF are comparable

20. # ofeventsfor cut12, SF and OF • Signal and background have comparable number of surviving events in both SF and OF NB = 0.1208  0.0570 NB = 0.1131  0.0566

21. Eventsselectedafter BDTG cut • Main SM backgrounds are ttbar and dibosons • Other minor backgrounds are still missing • Only statistical errors shown here

22. Summary and outlook • We have a reasonable set of variables and of preliminary cuts for a multi-variate analysis to obtain expected sensitivities higher than those in the cut-based analysis. • “cut11” performs better in OF dileptons, “cut12” in SF. • We plan to get results also by training 2-3 other relevant signal points in the grid. • We need to define adequate CRs for the main known backgrounds. • We need to estimate fake leptons. • We are testing consistency and stability of results, especially for systematic purposes. • A draft of an INT note is being prepared. • We plan to take a look at the chargino-b grid.

23. Backup slides

24. Sensitivity for cut11, mumu and ee • Combined SF = mumu + ee improves sensitivity almost everywhere in the grid

25. # ofeventsfor cut11, mumu and ee NB = 0.3641  0.0940 NB = 0.3970  0.0993

26. Sensitivity for cut12, mumu and ee • CombinedSF = mumu + eeimprovessensitivityalmosteverywhere in the grid

27. # ofeventsfor cut12, mumu and ee NB = 0.0955  0.0551 NB = 0.0253  0.0146

28. # ofsignaleventsfor cut11 and cut12

29. # ofsignaleventsfor cut11 and cut12

30. # ofsignaleventsfor cut11 and cut12

31. # ofsignaleventsfor cut11 and cut12