Missing Energy Look-alikes at the LHC

Transcription

1 Missing Energy Look-alikes at the LHC Jay M. Hubisz Syracuse University West Coast LHC Meeting /2/28 arxiv: [hep-ph] with Joseph Lykken, Maurizio Pierini, and Maria Spiropulu

2 Twenty LHC We begin running the LHC with a very large (pseudo-infinite) number of possible extensions of the SM Each data release gives another step in the game; want order /2 (or more) possibilities eliminated each time Each step (set of analyses) should be designed according to the previous ones, and gain increasing focus We want to be smart when we design our analyses

3 Goal: Create a strategy for the moment of discovery (perhaps first pb - of data) Do the most that we can (find powerful model discriminators) taking into account the limitations of what we ll have available with this first small data sample poorly understood systematics immature detector simulation primitive (at best) flavor tagging poor understanding of jets

4 Consider this a presentation of a dry run of how to manipulate the first signal data Things will be slightly different once there s real data to tune det. sim. and Monte Carlo, but we want a strategy ready-inhand By keeping as realistic as we can, we bolster our confidence that the strategies we develop will carry over well to the real thing

5 First understood data Won t be well understood will have some handle on detector response to jets from earliest studies ( TeV run?) won t have sophisticated jet corrections (partonic jets), just raw/uncorrected jets primitive flavor tagging (enrichment) observables that are available will be strongly correlated by both physics and systematics want to keep these errors to a minimum bad time for a global analysis

6 MET + LHC WHY? Dark matter experimental evidence Theoretical prior that hierarchy problem is solved near the TeV scale (top quark is involved) The necessity of conserving certain global symmetries custodial SU(2), baryon # Compelling case for LHC pair production of strongly coupled exotica decaying to jets + X + ET(miss)

7 ET(miss) + LHC NP j X / ET New Physics + Parity SUSY: R-parity (baryon #) / ET Little Higgs: T-parity (cust. SU(2)) j2 X2 Universal Extra Dim.: KK-parity Parities keep protons from decaying, prevent gross violation of flavor constraints, keep MW consistent with exp., and (if exact) provide dark matter

8 Starting point There is a 5 sigma excess in a MET+jets search with pb - (N signal events) We don t utilize any other potential search channels (i.e. that don t trigger on MET)

9 Models SUSY Little Higgs produce squarks + gluinos produce heavy T-odd quarks decay to lightest R-odd particle (neutralino) decay to lightest T-odd particle (neutral vector boson) Hierarchy problem saved by spin statistics and SUSY coupling relations Hierarchy problem saved by global symmetries cancellations with opp. spin cancellations with same spin

10 Group and Group 2 We perform two sets of analyses - in each, model plays the role of data and the rest are candidate theories or Look-alikes Look-alike: model that gives same # of events in ET(miss) analysis path Group Group 2 Data is SUSY Data is Little Higgs SUSIC look-alikes SUSIC look-alikes Pass all models through a software chain

11 What will be in early data sets? (CMSPTDR) Study of SUSY benchmark scenarios series of cleanup/analysis/bkgd rej. cuts on ET trigger sample up to 25% eff. on signal for σ5pb, > 5σ discovery in pb -! we adopt very similar analysis path (bkgds are done) New: we go beyond the benchmarks (even non-susy) and refine/develop the analysis for efficiency in model discrimination

12 Backgrounds for / ET ttbar, single top, W+jets, Z+jets, dibosons most have hard lepton in association with neutrino (gives MET) generated by CMS with AlpGen and passed through full det. sim. QCD beam halo, cosmics, detector noise

13 CMSPTDR MET analysis path Cut/Sample Signal t t Z( ν ν)+ jets EWK + jets All (%) Trigger E miss T > 2 GeV PV N j η j d EEMF ECHF QCD angular Iso lead trk = EMF (j), EM F (j2) E T, > 8 GeV, E T,2 > GeV H T > 5 GeV focuses on 2 WIMP final states (Njets 2) QCD pileup, radiation often gives a third jet efficient for signal, strong reduction of background events remaining per pb

14 Backgrounds 245 evt./fb - in ET(miss) analysis path after cuts Error: how good is det. sim., how good was CMSPTDR SM cocktail, how well was QCD simulated, PDF s In models we studied, can triple SM background count after cuts, add 5% systematics on signal and still have discovery in pb -

15 Early LHC poor jet resolution fakes missing ET muons contribute (don t deposit all energy) plagued by instrumental + spurious bkgds (cosmics, scattering off beam halo) - primary cleanup use raw (uncorrected) jets in early running not much like generator level (partonic) missing ET

16 Our Toolkit Models (Theory) Hard Process 2 2 (N) (MG/ME) on shell decays (BRI/DGE) Showering and hadronization (PYTHIA) Detector Simulation (PGSCMS) Madgraph/MadEvent Great tool for BSM easy to put in a new model keeps spin correlations Ours Ours Analysis binning, histograms plots (ExRootAnalysis)

17 PGSCMS + analysis needed fast (parametrized) detector simulation PGS (pretty good simulation) tuned for CMS detector tuned to LM study in CMSPTDR add pileup, z-vertex, some B-field effects Cut/Software Full Fast Trigger and ET miss > 2 GeV 53.9% 54.5% N j 3 72.% 7.6% η j d.7 88.% 9.% QCD angular 75.6% 77.6% Iso lead trk = 85.3% 85.5% E T, > 8 GeV, E T,2 > GeV 63.% 63.% H T > 5 GeV 92.8% 93.9% Total efficiency 2.9% 3.8% agreement good!

18 Software Chain Software/Models Group models Group 2 models Spectrum generator Isajet v7.69 [3] or private little Higgs or SUSY-HIT v. [3] or SUSY-HIT v. Matrix element calculator Pythia v6.4 [32] MadGraph v4 [33] Event generator Pythia v6.4 MadEvent v4 [34] with BRIDGE [35] Showering and hadronization Pythia v6.4 Pythia v6.4 Detector simulation PGSCMS v.2.5 PGSCMS v.2.5 plus parameterized plus parameterized corrections corrections Analysis: Specifically designed - ExRootAnalysis (includes jet energy corrections) and dependent rescaling of jet energies 5 GeV PGS jet = 3 GeV uncorrected (raw) jet

19 Some flaws Drop primary cleanup and parts of ILV can t simulate these ET(miss) not quite right - need full sim. no electrons - need full sim future study

20 How does PGSCMS do? Cut/Software Full Fast Trigger and ET miss > 2 GeV 53.9% 54.5% N j 3 72.% 7.6% η j d.7 88.% 9.% QCD angular 75.6% 77.6% Iso lead trk = 85.3% 85.5% E T, > 8 GeV, E T,2 > GeV 63.% 63.% H T > 5 GeV 92.8% 93.9% Total efficiency 2.9% 3.8% Biggest disagreement is on QCD angular cuts (we don t accurately reproduce jet mismeasurement effects) Level of agreement - we expect fast sim. look-alikes to remain look-alikes (or at least be close) with full det. sim.

21 Observables Want to focus on robust objects shapes, distributions, too sensitive to systematics, poor simulation, etc. not good for moment of discovery Large bins bring this problem under better control Boxes Ratios of counts in diff. boxes lower systematics since many are common to all boxes - cancel out

22 Observables: Ratios systematics cancel effectively from about 2% down to about 5% luminosity uncertainty - completely pdf uncertainty - partially higher order corrections - partially

23 Boxes Simplest thing: Bin events in analysis path according to other trigger samples (mock-ups of CMS triggers) MET (%) Muon2 DiJet TriJet Trigger efficiency.8 Trigger efficiency.8 Trigger efficiency.8 Trigger efficiency (GeV) miss E T p (GeV/c) T E T (GeV) E T (GeV) Trigger efficiencies - simulate turn-on

24 Simplest Boxes MET TriJet Muon2 DiJet

25 Meff. M eff = E miss T + i=,4 p j T Create one new box for events with large Meff. N(Meff4) Stransverse mass (MT2) m 2 T 2(m dm ) p () T min +p(2) T =pmiss T [ max { m 2 T 2(m dm ; p () T ), m2 T 2(m dm ; p (2) T ) }] Choose mdm to be mass in candidate theory

26 Stransverse mass Stransverse mass boxes: N(mT2-3) the number of events after selection with m T 2 > 3 GeV/c 2, N(mT2-4) the number of events after selection with m T 2 > 4 GeV/c 2, N(mT2-5) the number of events after selection with m T 2 > 5 GeV/c 2. N(mT2-6) the number of events after selection with m T 2 > 6 GeV/c 2.

27 Hemispheres and Cones Event shape: extract info about scattering subprocess and resulting decay chains Our Signal: pair production of exotics - each event has natural separation into halves (hemispheres) Each hemisphere (ideally) has wimp and visible decay products of parent exotic (+ neutrinos)

28 Flavor Enrichment Very low level tagging tau enrichment for each jet,.375 cone, count tracks > 2 GeV, if only one, and > 5 GeV, call tau b enrichment if muon within.2 of jet axis, call b LM2p LM5 LM8 CS4d CS6 τ jets per fb tags per fb correct tags per fb efficiency 2% 8% 2% 3% 6% purity 55% 23% 7% 4% 8% LM2p LM5 LM8 CS4d CS6 b jets per fb tags per fb correct tags per fb efficiency 5% 5% 5% 5% 5% purity 72% 72% 75% 7% 39%

29 Our Ratios r(nj)(3j), with n=4,5 r(met32) r(met42) r(met52) r(ht9) r(meff4) r(m4) r(m8) r(hemj) with j=,2,3 r(2µ-nj)(µ-nj) with n=3,4 r(τ-tag) r(b-tag) r(mt2-3) with the theory LSP mass r(mt2-4) with the theory LSP mass r(mt2-5) with the theory LSP mass r(mt2-6) with the theory LSP mass r(dijet) r(trijet) r(muon2) r(mt2-4/3) with the theory LSP mass r(mt2-5/3) with the theory LSP mass r(mt2-6/3) with the theory LSP mass r(nt-cα) for n=,2,3,4 and α = 3,45, 6 75, 9 r(ntdiff-cα) for for n=,2,3,4 and α = 3, 45 6, 75, 9

30 Group spectra Data is LM2p Low mass points and compact SUSY models are candidates LM2p LM5 LM8 CS4d CS6 qq χ χ 57% 6% 34% 38% 98% qqq χ χ 2% 9% 3% 4% 79% qqqq χ χ % % % % 77% τ ν τ q χ χ 39% % - - % ττ q χ χ 25% % - - % b q χ χ 3% 25% 33% 69% 9% b t W q χ χ % 9% 3% 67% - W q χ χ 25% 52% 56% 93% - h q χ χ 3% 2% tt q χ χ 9% 4% 4% % 2% Z q χ χ % 8% 35% % - Z W q χ χ 2% 6% 23% 6% - bb tt W W χ χ - - 2% 8% - ] 2 mass [GeV/c g u L d R!" l R LM2p LM5 LM8 CS4d CS6 ± d L u R t! " #" 2! " g u L d R!" l R! " ± d L u R t! " 2 #" d L u R g l R!"! " ± u L dr t #"! " 2 d L g d R R #" l R ±! "!"! " u L u 2 t! " l R! " 2 g!" ± #"

31 Results (Group ) LM2p LM5 LM8 CS4d CS6 LM2p r(5j)(3j).6σ r(5j)(3j) 4.4σ r(met52) 4.σ r(mt2-6/3).4σ r(mt2-3).4σ r(met52) 3.7σ r(ht9) 3.6σ r(met52).6σ r(τ-tag).2σ r(t-c45) 2.9σ r(meff4) 3.σ r(ht9) 6.8σ r(τ-tag) 3.σ r(met52) 8.2σ r(met52) 9.4σ r(mt2-6/3) 33.σ r(5j)(3j) 2.8σ r(mt2-5) 6.7σ r(ht9) 6.4σ r(met52) 26.6σ r(mt2-4) 2.6σ r(5j)(3j) 6.5σ r(mt2-6) 6.σ r(ht9) 4.6σ LM5 r(5j(3j).8σ r(5j)(3j) 2.9σ r(ht9) 3.6σ r(mt2-6/3).6σ r(mt2-3).5σ r(met52) 2.7σ r(meff4) 3.2σ r(met52) 9.2σ r(t-c3).4σ r(muon2) 2.5σ r(met52) 3.σ r(ht9) 6.8σ r(5j)(4j) 3.4σ r(met52) 6.σ r(met52) 7.σ r(mt2-6/3) 33.7σ r(τ-tag) 2.7σ r(muon2) 4.9σ r(ht9) 6.4σ r(met52) 22.9σ r(mt2-4) 2.6σ r(5j)(3j) 4.3σ r(mt2-6/4) 6.σ r(ht9) 4.6σ LM8 r(5j)(3j) 5.5σ r(5j)(3j) 3.3σ r(5j)(3j) 3.σ r(muon2).σ r(t-c3) 3.7σ r(muon2) 3.σ r(mt2-4) 2.2σ r(mt25/3) 5.2σ r(muon2) 3.6σ r(met52) 2.4σ r(2t-c45) 2.σ r(hem3) 4.σ r(5j)(3j).σ r(muon2) 7.2σ r(5j)(3j) 5.4σ r(muon2) 25.8σ r(muon2) 8.σ r(hem3) 5.7σ r(hem3) 5.3σ r(mt2-6/3) 2.σ r(hem3) 7.3σ r(5j)(3j) 5.6σ r(muon2) 4.σ r(hem3) 4.2σ CS4d r(met52) 3.5σ r(ht9) 3.σ r(5j)(3j) 2.8σ r(muon2) 6.8σ r(ht9) 3.2σ r(met52) 2.7σ r(mt2-3) 2.σ r(met42) 5.5σ r(meff4) 2.6σ r(meff4) 2.6σ r(t-c3).9σ r(mt2-5/3) 5.2σ r(met52) 6.5σ r(met52) 5.σ r(5j)(3j) 4.2σ r(muon2) 7.3σ r(mt2-6) 5.3σ r(mt2-6/4) 4.8σ r(tdiff-c3) 3.6σ r(mt2-5) 2.8σ r(ht9) 5.2σ r(ht9) 4.5σ r(hem3) 3.6σ r(met52).5σ CS6 r(met42) 7.σ r(met42) 6.σ r(b-tag) 6.5σ r(met42) 4.3σ r(mt2-5/3) 5.σ r(mt2-5/3) 4.6σ r(muon2) 5.2σ r(muon2) 4.σ r(ht9) 4.8σ r(ht9) 4.5σ r(met42) 4.σ r(mt2-5/3) 2.9σ r(met52).5σ r(b-tag).σ r(b-tag) 5.6σ r(b-tag) 4.9σ r(b-tag).2σ r(met52).3σ r(muon2).2σ r(muon2) 8.4σ r(mt2-5).2σ r(mt2-5) 9.2σ r(met52) 7.6σ r(met42) 7.6σ

32 Group 2 spectra LH2 - little Higgs model littlest Higgs w/ T-parity ] 2 mass [GeV/c 9 8 LH2 NM6 NM4 CS7 NM6 - LH2 s susic sister NM4 - LH2 s susic little sister (look-alike) uh i Z H i dh W H d L u R u L! " 2 d R!" ± l R #"! " 2 d L u u R ±!" d L R! " 2 l R g!" ± #" CS7 - No relation (look-alike) 3 AH is LH lightest neutralino 2 A H! "! "! "

33 Group 2 Analysis LH2 (6.5 pb LO x-sec) before cuts after cuts Q i Qi 55% 64% u i Hd i H, u i Hu i H, d i Hd i H 4% 6% d i HW + H, ui HW H 2% 7% u i HZ H, u i HA H, d i HZ H, d i HA H 9% 5% Q i Qj, i j 3% 3% other 7% 5% No gluino in LH models qq A H A H 64% W qq A H A H 39% h qq A H A H 22% bb A H A H 4% W W qq A H A H 8% hh bb A H A H 4% hh qq A H A H 3% tt A H A H 3% SUSY: NM6 NM4 CS7 qq χ χ 84% 83% % qqq χ χ 8% 6% % qqqq χ χ % bb q χ χ 2% 5% % W qq χ χ 26% 35% - h q χ χ 4% 9% - tt q χ χ % % % Z q χ χ 4% 5% - W W q χ χ 4% 9% - NM6 NM4 CS7 LO cross section (pb) before after before after before after cuts cuts cuts cuts cuts cuts q i q i 3% 29% 34% 26% - - ũ d, ũũ, d d 32% 28% 29 23% - - squark gluino 3% % 5% 23% 4% 8% gluino gluino % 9% squark chargino 2% 2% 3% % - - squark neutralino 4% % 4% q i q j, i j 5% 7% 7% 4% - - other 3% 3% 8% 3% - -

34 Results (Group 2) LH2 NM4 CS7 LH2 r(mt2-5) 4.9σ r(mt2-5) 6.7σ r(meff4) 3.σ r(met42) 6.5σ r(m4) 2.7σ r(4j)(3j) 4.σ r(mt2-5) 4.σ r(mt2-5) 8.9σ r(mt2-3) [TriJet].σ r(met42) 6.7σ r(mt2-4) [DiJjet] 7.9σ r(mt2-5) [TriJet] 8.8σ r(meff4) 7.2σ r(4j)(3j) [DiJet] 7.3σ r(m4) 6.6σ r(mt2-3) [DiJet] 6.7σ NM4 r(meff4) 4.2σ r(meff4) 4.3σ r(m4) 4.σ r(dijet) 4.σ r(mt2-4) 3.8σ r(met42) 4.σ r(meff4).8σ r(meff4).2σ r(trijet).4σ r(met52).6σ r(m4) 9.8σ r(dijet).6σ r(dijet 8.2σ r(ht9) 9.σ r(ht9) 8.σ r(4j)(3j) 6.σ CS7 r(met42) 4.9σ r(4j)(3j) 4.4σ r(4j)(3j) 4.6σ r(met42) 3.3σ r(mt2-4) 4.σ r(hem) 3.2σ r(5j)(3j) [DiJet] 6.8σ r(4j)(3j) 9.4σ r(trijet).4σ r(5j)(3j) [DiJet] 7.4σ r(met42) 9.6σ r(meff4) 7.4σ r(4j)(3j) 9.5σ r(dijet) 6.9σ r(mt2-5) 8.3σ r(ht9) 6.2σ Huh? I thought you couldn t do spin at the LHC...

35 Spin statistics Little Higgs vs SUSY Boson cancels boson Fermion cancels fermion Boson cancels fermion and v.v. Ancient wisdom - models differing only by spin have very different cross sections Lesser Known: different pt spectra!

36 About spins Heavy quarks (red solid) vs. colored scalars (blue dashed) (e.g. T-odd quarks) (e.g. squarks) ( GeV/c) - - T gg init ( GeV/c) T qqbar init dlog!/dp 2 dlog!/dp (GeV/c) p T (GeV/c) p T Normalized pt distributions

37 Actual Models LH2 (red) vs susy model NM6 (blue) number of events (GeV/c) p T Normalized pt distributions

38 Distinguish spins: ) Start with initial sample (analysis path) 2) Create box with higher pt events ratio of counts in high pt box to total # in path should be a discriminating variable LH2 6.5pb 4% NM4.3pb 9.4% Lookalikes, but already discriminated at pb -!

39 Spin done in first pb - LH2 vs. NM4 [ pb ] Variable LH2 NM4 Separation MET r(mt2-5) r(mt2-4) r(mt2-3) r(meff4) r(mt2-5/3) r(m4) r(mt2-4/3) r(ht9) r(met42) r(mt2-5/4) We tried hard to construct a SUSY model that was a closer lookalike, but did not succeed... MT2 at high pt strong discriminator If this holds up under closer examination, will overturn a lot of conventional wisdom Most techniques rely on detailed shapes of complicated distributions (lots of data)

40 Future Study This was a dry-run /exploration of what the best things are to do with the first pb - of data given a signal in the ET(miss) analysis path Next: primitive sim, left out a few details, but stayed realistic enough to flesh out a grand master plan Full sim, including electrons, better ET(miss), more lookalikes (UED? More LH models?), NLO Spin discrimination in model comparisons what do we learn about dark matter? other analysis paths?

41 Conclusions We will hopefully have 5 sigma discovery by the end of the pb - era we ve developed techniques to discriminate models efficiently with small amounts of data set of robust observables (ratios of inclusive counts) realistic in that we minimize systematics, and stick to things that are (or should be) achievable in first year Compelling evidence for spin discrimination at moment of discovery