MET+multijet search beyond the PTDR

Transcription

1 MET+multijet search beyond the PTDR SUSY Meeting Christopher Rogan California Institute of Technology w/ VecBos + Jets group

2 Introduction All Hadronic + MET final state PTDR analysis in CMSSW Focus on specific elements of the analysis: Event electromagnetic and charge fraction clean-up variables Indirect lepton veto QCD cleanup variables 2

3 All hadronic MET + jets analysis path Baseline selection MET is uncorrected Analysis path: Jets uncorrected CaloJets, require to be counted Event clean-up cuts QCD rejection cuts Indirect lepton veto signal/background optimizaton Here: 3

4 softsusy LM point efficiencies in CMSSW (1_6_10) PTDR (LM1) cut efficiency (%) Inclusive efficiency (%) LM1 LM2 LM3 LM4 LM5 LM6 LM7 LM8 LM9 LM10 LM PTDR efficiencies are reproduced nicely for LM1 4

5 Event Charged Fraction Cosmic rays and other backgrounds do not originate from vertex - their calorimetric depositions do not correspond to tracks associated with the primary vertex Event Charged Fraction (ECHF) variable is used to remove these backgrounds see CMS IN 2006/010 Construction of this variable requires matching of tracks to CaloJets - see discussion in HCAL DPG talks (linked in extra slides) about jet position resolution improvement studies through CaloTower variations 5

6 Event Electromagnetic Fraction non-projective calorimetric depositions from cosmic bremsstrahlung often have high EM or hadronic component depending on calorimetric incidence All-hadronic beam halo events have low EMF close to 0 Event Electromagnetic Fraction (EEMF) variable is used to remove these backgrounds see CMS IN 2006/010 6

7 EEMF & ECHF LM2 Behavior of event clean-up variables consistent with PTDR study Need to optimize these variables/cuts w.r.t. the backgrounds they are intended to eliminate Cosmic data from global runs EEMF portion accessible w/ existing CRUZET data need tracker for ECHF Beam halo MC/real data when available PTDR cuts: Lepton/photon final states 7

8 Indirect Lepton Veto Goals: high signal efficiency large rejection of ILV 1 - jet EMF backgrounds ILV 2 - leading track isolation Look at lead track with Look at other tracks in cone of Around lead track - take sum p T Lead track is isolated (and event is vetoed) if: - indicates parameters that can be optimized Only tracks passing selection are considered: 8

9 ILV 1 - jet EMF LM1 - no MET requirement LM1 - MET > 300 GeV Predominantly hadronic final states result in blob topology shape is largely independent of MET magnitude 9

10 ILV 1 - jet EMF CSA07 pp->ele+x No MET requirement MET > 50 GeV Electrons from jets have same blob behavior - shift to higher EMF in both leading jets due to electron enhancement Shape is largely independent of MET magnitude 10

11 ILV 1 - jet EMF Z+1 parton pthat LM1 cut efficiency cut efficiency Points correspond to individual events - contours show cut efficiencies LM1 cut efficiency contours largely symmetric in 2 leading jets EMF Z->ee results in high EMF bands (circled) due to electrons being reconstructed as jets - efficiency contours asymmetric ( real jets from partons in addition to leptons) => opportunity for improvement w/ asymmetric cut w.r.t. 1st and 2nd leading jet 11

12 ILV 1 - jet EMF Assuming symmetric EMF cut on 2 leading jets Letpon final state content distinguishes LM points 12

13 CSA07 pp->ele+x ILV 2 - track isolation No MET requirement MET > 50 GeV Clear track structure in low ΔR due to other tracks in jet (same structure as pp->mu+x sample) Track topology largely independent of MET magnitude 13

14 ILV 2 - track isolation LM1 WW + 0 partons Again, clear track structure in low ΔR due to other tracks in jet for LM1 sample WW sample shows isolated leptons from W decays - appreciable fraction of events w/ low pt leading track (setting this parameter for ILV should be done in context of MET magnitude due to expected correlation) 14

15 ILV 2 - track isolation W+1 parton pthat WW+0 partons LM1 ILV efficiency ILV efficiency ILV efficiency The cut-off that classifies an isolated track and the parameter pull the efficiency contours in opposite directions the differences in the slopes of the contours between signal and background allow us to optimize the balance between these two parameters - this is potentially a large opportunity for improvement of the ILV 15

16 QCD clean-up variables In QCD multjet events, measured MET results from neutrinos from heavy flavor decays jet mis-measurements veto events where MET is aligned/anti-aligned with two leading jets PTDR selection: 16

17 QCD clean-up no MET requirement MET > 90 GeV MET > 210 GeV QCD pythia Here, we require: no MET requirement MET > 210 GeV MET > 450 GeV LM1 17

18 QCD clean-up QCD ( ) LM1 ( ) LM6 ( ) cut efficiency cut efficiency cut efficiency Looking at R1,R2 cut while requiring different magnitudes of inclusive MET yields interesting results QCD angular variable ARE correlated with MET magnitude (in non-trivial way) All LM points have inflection feature - inflection point highly correlated with 18

19 Variable CaloTower positioning Even in the absence of a displaced primary vertex (relative to (0,0,0)), jet position resolution can be improved by calculating the positions of the ECAL portion of CaloTowers event by event Here, the strategy is to use the granularity of the ECAL cells in the transverse plane w.r.t. the nominal interaction point The CaloTower η and φ are calculated for each event using an energy weighted average of the ECAL cell positions Only positive weights included (cells with less than 1.5% of the total EM energy are not included) in order to prevent bias in position reconstruction (see, for example, CMS DN 2007/001) Can use similar scheme to combine ECAL and HCAL positions, weighting them using their relative energy depositions (studies ongoing) 19

20 QCD clean-up QCD ( ) QCD ( ) LM1 no MET requ. MET > 200 GeV MET > 200 GeV Here, we require: Without MET requirement, primary vertex corrections and variable CaloTower positioning have negligible effect on QCD clean-up efficiencies for background In the tail of the MET distribution (signal region) -> larger effect (order %1) 20

21 Outlook PTDR all hadronic + MET search analysis reproduced for all soft SUSY LM signal points in CMSSW Optimization of ILV ongoing (very large opportunity for improvement) Understanding correlations related to QCD clean-up ongoing Current selection should be generalized to mult-jet topologies Event clean-up variable being examined in the context of CRUZET/global run data for realistic approach icsa08 samples Working from with HCAL DPG to make sure all improvements/variations are in CMSSW FW 21

22 Physics Focus Chris Rogan, Maurizio Pierini, Thiago Tomei, Joseph Lykken, Maria Spiropulu, Marco Zanetti, (HCAL, jetmet, commissioning, trigger, susy+exo jetmet searches) also Sezen Sekmen with G2 and SO(10) multijet hadronic modes and Massimiliano Chiorboli with trigger work Ilaria Segoni, Didar Dobur, Emanuele Di Marco, Chiara Rovelli, Paolo Meridiani (boson selection candles W/Z+j ratio) Michael Tytgat, Lukas Vanelderen (top, t/w) (Marcella Bona starting also with PFlow) 22

23 EXTRA SLIDES 23

24 HCAL DPG Talks Talks related to CaloTower studies from HCAL DPG: rialid=slides&confid= rialid=slides&confid= rialid=slides&confid=

25 ILV 1 - jet EMF 25

26 CaloTower Comparisons region of overlap between HB/HE Jet position resolution (η and φ) systematically improved by using weighted ECAL positioning (except in region of overlap between HE/HB - can revert to default scheme here, or adapt variable positioning scheme) Improvement by as much as a factor of 1.5 in position resolution relative to a fixed positioning scheme (improvement is even greater relative to jets uncorrected w.r.t. the primary vertex) 26

27 Position bias Low p T jets sample Displaced primary vertex not only degrades position resolution, but also significantly biases position reconstruction (and ultimately p T ) Corrections for primary vertex position significantly reduces these biases - using ECAL weighted positioning further decreases position bias 27

28 MET position resolution softsusy LM1 MET position resolution highly correlated with jet position resolution in the context of the corrections studied in this analysis Difficult to compare different CaloTower construction approaches with distributions like this - instead we can compare the distributions on a bin-by-bin basis, observing the migration of events from bin-to-bin All distributions shown here contain same number of events 28

29 MET position resolution For this plot: The cumulative distribution of each of the histograms is taken (integral of histogram from 0 to a given bin) The cumulative value in each bin of the uncorrected distribution is subtracted from the cumulative value in each bin of the corrected distributions softsusy LM1 the a posteriori jets with fixed positioning (red) are derived from the same CaloTower collection as the uncorrected jets (line at 0) a priori corrections have most events in the core of the distribution - variable positioning has best MET position resolution MET for each eta ring could be potentially used for a posteriori MET correction w.r.t. primary vertex 29

30 Jet/track matching softsusy LM1 Matching of tracks to corresponding jets depends directly on the position resolution of jets Here, tracks are those compatible with event primary vertex Variable positioning CaloTower scheme (both a priori and a posteriori primary vertex corrections) yields best matching - allows for more restrictive Δ R selection for Jet/track matching Iterative Cone R =

31 p T resolution It is difficult to quantify the effects of the studied corrections on the jet pt resolution (order of ~1% vs. many %) From plot below, we see the expected magnitude of corrections as a function of η (%) Toy MC One needs to look on an event-by-event basis in order to understand correction effects 31

32 Primary vertex geometry CaloTower/Jet = The distribution of the primary vertex z-position relative to (0,0,0), (0,L,D), is Gaussian - the corresponding change in E T is NOT Gaussian (in fact it is non-trivial, and can introduce bias in E T ) (0,0,0) (0,0,x) (0,0,D) beam-line y is Gaussian distributed 32

33 Toy MC Using a toy MC we can see the geometrical effects of primary vertex smearing on the E T bias (%) For low η we systematically over-estimate the E T of jets in the absence of a correction for the primary vertex Assuming L = 159 cm (depth = 6% btw front ECAL, back HO) Gaussian smearing of primary vertex in z Toy MC Here, the error bars show the error on the mean, not the spread in the distribution Toy MC 33

34 A closer look This term is only significant when is small (on the order the primary vertex spread) i.e. low η Only negative - i.e. corresponds to a correction that systematically shifts the mean correction below zero Fitting the distribution to the left with a Lorentzian, we see a shift in the mean of 0.033%, consistent with our toy MC prediction Only jets made with variable positioning have this feature relative to uncorrected jets 34

35 A closer look As (η) increases, this term dominates such that we have outlayers contribute to tails This causes the resulting PDF for ΔP T /P T to be Gaussian with σ scaled relative to the beamspread by a factor of q/(1+q 2 ) Checking this quantitatively: from the fit to the left σ = η = 1.6 => σ exp = Relatively good agreement 35

36 Difference between a priori and a posteriori PV corrections corr refers to jets clustered from CaloTowers with variable positioning and PV correction pthat > 170 GeV a priori and a posteriori primary vertex corrections result in identical jet position resolution - but jet construction is affected, yielding different numbers of jets in a fraction of events Variable positioning yields systematically less jets (when there is a difference) 36