Latest CDF Higgs Results and Combined Higgs Limits. Nils Krumnack (Baylor University)

Transcription

1 Latest CDF Higgs Results and Combined Higgs Limits

2 Introduction this seminar falls into two parts in the first part I will give an overview of CDF Standard Model Higgs result in the second part I will explain the details of our Bayesian limit calculation not covering: details of the experiments and CDF accelerator beyond standard model Higgs results

3 Higgs Production 3 SM Higgs production gg! h TeV II 5! [fb] SM Higgs production LHC! [fb] qq! Wh 4 qq! qqh gg! h qq! qqh 3 bb! h qq! Wh qq! Zh bb! h gg,qq! tth gg,qq! tth qb! qth TeV4LHC Higgs working group m h [GeV] TeV4LHC Higgs working group qq! Zh m h [GeV] Production modes very similar between Tevatron and LHC ( one exception: Tevatron has more associated Higgs particularly clean event signature used in most Tevatron searches 3

4 Higgs Decay SM Higgs branching ratios (HDECAY) bb _ WW Higgs has many choices to decay main decay modes: bb and WW gg H WW: sensitive channel. ττ - -3 gg ZZ cc _ γγ Zγ (GeV/c ) gg H bb: huge backgrounds can use associated production (WH/ZH) for better s/b in bb alternatives: ττ and γγ have lower backgrounds, but also lower branching ratio 4

5 common techniques lepton id: leptons are a part of most Higgs final states leptons are easy to identify and trigger on only few backgrounds have leptons most channels work on looser lepton id b-tagging: identifies jets from b quarks essential for low mass channels uses displaced vertices (and soft leptons from b decays) divide sample into several tag categories (different s/b) somewhat difficult to calibrate/model 5

6 common techniques missing transverse energy: part of many Higgs final states only few backgrounds, but many fakes from multi-jet events can be triggered on (at the Tevatron) neural networks (NN) and boosted decision trees (BDT) trained on Monte Carlo samples for signal and background provides event discriminant for every single event pulls apart high from low s/b regions matrix element calculates event discriminant from LO ME and detector resolution functions calculates differential rate for the topology of each event for signal and major backgrounds output normally fed into a NN or BDT 6

7 H WW lνlν the main Higgs search channel most sensitive channel down to mh=5 branching ratio at 5 is only % Events CDF Run II Preliminary Region: Base (6) " data WW WZ ZZ L dt = 3. fb tt W! W+jets DY Syst. Uncertainty CL = 36.% KS CL = 3.4% 5 accepts events from all the production modes including gg H possible through clean WW signature innovations: looser lepton ID accepting leptons in many categories Events e e e µ µ µ e trk µ trk CDF Run II Preliminary Region: Base (6) " data WW WZ ZZ L dt = 3. fb tt W! W+jets DY Syst. Uncertainty CL = 6.9% KS CL = 4.5% TCE TCE lepton categories TCE PHX PHX PHX TCE_CMUP TCE_CMX _CMIOCES _CMIOPES PHX_CMUP PHX_CMX _CMIOCES _CMIOPES MUP CMUP CMUP CMX P CMIOCES P CMIOPES CMX CMX X CMIOCES X CMIOPES TCE Trk PHX Trk CMUP Trk CMX Trk 7

8 H WW lνlν Events / 5. GeV/c separate analysis for, & jets different signal and background allows for ZH, WH and VBF events uses ME and NN for the event discriminant CDF Run II Preliminary L dt = 3. fb 9 Region: Base data tt (6) WW W! 8 7 WZ ZZ W+jets DY Syst. Uncertainty " CL = 6.% KS CL = 4.% Overflow = 6 Events /. 4 Region: Base data tt (6) WW W" WZ ZZ W+jets DY CDF Run II Preliminary # L dt = 3. fb Syst. Uncertainty CL = 6.% KS CL = 9.7% Events Events / 5. GeV 7 Region: Base data tt (6) WW W! 6 WZ ZZ W+jets DY CDF Run II Preliminary L dt = 3. fb Syst. Uncertainty CL = 68.6% KS CL = 84.4% CDF Run II Preliminary " njet N L dt = 3. fbjets 6 Region: Base data tt (6) WW W! 4 WZ ZZ W+jets DY Syst. Uncertainty " CL = 6.% KS CL = 39.8% Overflow = Dilepton Mass [GeV/c ] !R leptons E T [GeV] 8

9 H WW lνlν 3 train a separate neural network for each mass and each number of jets extract limit from NN output distributions most sensitive around mh=65 GeV observed (expected) limit at 65 GeV:.7 (.6) CDF Run II Preliminary = 65 GeV/c All Jets HWW M H " L = 3. fb Total Wj W! tt WZ ZZ DY WW HWW Data 95% C.L./! SM CDF Run II Preliminary " L = 3. fb HWW Expected HWW Observed HWW ±! HWW ±! NN Output Standard Model Higgs Mass [GeV/c ] 9

10 WH WWW lνlν(l)ν the other high mass analysis more than times less sensitive than the WW result requires two same sign leptons, vetoes on third lepton uses BDT for signal discrimination currently optimized for fermiophobic Higgs, working on update with better sensitivity for SM Higgs Events LS CDF Run-II Preliminary: Data Fake leptons Residual conversions WZ H T (GeV) ZZ Uncertainty of the total BG Wh GeV/c (FH) 5 Wh 6 GeV/c (SM) 5.7 fb Events 9 Data LS 8 Fake leptons Residual conversions 7 WZ CDF Run-II Preliminary: BDT Output ZZ Uncertainty of the total BG Wh 6 GeV/c (FH) 5 Wh 6 GeV/c (SM) 5.7 fb 95% C.L. limit/sm 3 CDF Run-II Preliminary: P.E. of 95% P.E. of 68% Expected Limit (95% CL) Observed Limit (95% CL) SM higgs.7 fb Higgs Mass (GeV/c )

11 WH lνbb most sensitive channel for low mass identifies W through a lepton and missing transverse energy identifies H as dijet pair with b-tags divides data into lepton and b-tag categories adds lepton acceptance with very loose, stub-less muons these are not triggered directly use MET+dijet trigger instead muons look like MET in trigger QCD background surprisingly low trigger non-trigger

12 WH lνbb Candidate Events CDF Run II Preliminary, L=.7 fb train separate discriminants one with ME+BDT and one with NN combine both into a superdiscriminant using an evolutionary NN obs (exp) 5.6 (4.8) Events BDT WH(5) W+light W+charm W+bottom Non-W Diboson Z+jets Single_top Top_pair Data ME+BDT tag WH(5) Normalized to Prediction 95% CL Limit/SM Events CDF Run II Preliminary, L =.7 fb - Limits for WH Combination CDF Run II Preliminary.7 fb 95% of PE 68% of PE ST+JP Median Expect WH Combination Observed WH Combination WH (m = 5 Gev) ( ) W+bb W+cc, W+c NN Output W+jj Top Other Data Higgs Mass (GeV/c ) Normalized to Prediction

13 ZH ννbb also a sensitive channel for low mass identifies H as two b-tagged jets identifies Z as missing transverse energy also picks up WH with missing lepton allows third jet for this main background are multi-jet events no real missing energy, but a lot through jet missmeasurement cleaned up by comparing tracking and calorimeter missing energy correlated for true missing energy Dijet Invariant Mass, Signal Region, ST+JP CDF Run II Preliminary,. fb M (GeV) TrackMET NN, Signal Region, ST+JP 5 5 Data Multijet Top Pair Single Top Diboson Z/W+H.F. Background Error VH5 5 CDF Run II Preliminary,. fb 3 Data jj Multijet Top Pair Single Top Diboson Z/W+H.F. Background Error VH trackmet 3

14 ZH ννbb train a NN to distinguish signal from multi-jet background cut on NN value train new NN to distinguish from other backgrounds extract limit from that NN output distribution obs (exp) 6.9 (5.6) QCD Rejection NN, Signal Region, ST+JP 4 CDF Run II Preliminary,. fb 6 Data Multijet Top Pair Single Top Diboson Z/W+H.F. Background Error VH5 5 NNoutput, Signal Region, ST+JP CDF Run II Preliminary,. fb Data Multijet Top Pair Single Top Diboson Z/W+H.F. Background Error VH5 5 95% C.L.limit/SM CDF Run II Preliminary,. fb VH! E T + bb Expected (68%CL) VH! E T + bb Expected (95%CL) Expected 95% C.L. limit Observed 95% C.L. limit QCD Rejection NN NNoutput Higgs Mass (GeV/c ) 4

15 ZH llbb third most sensitive channel at low mass identifies H through two b-tagged jets identifies Z through two leptons with invariant mass at Z peak extends lepton acceptance through extra loose id events contain no true missing energy all missing energy from jet missmeasurements improving jet resolution by correcting jets for observed missing energy CDF Run II Preliminary (.7 fb ) Number of Events double T tag (high) data 8 ZH 6 WZ WW ZZ tt fakes Z "!! Z+cc Z+bb Z+jets mistags uncertainty Number of Events CDF Run II Preliminary (.7 fb double T tag (high) data ZH WZ WW ZZ tt fakes Z "!! Z+cc Z+bb Z+jets mistags ) uncertainty M jj (GeV/c ) Missing E (GeV) t 5

16 ZH llbb Number of Events CDF Run II Preliminary (.7 fb ) double T tag (high) ZH 5 = GeV/c M H data ZH WZ WW ZZ tt fakes Z "!! Z+cc Z+bb Z+jets mistags uncertainty Number of Events CDF Run II Preliminary (.7 fb 7 data double T tag (high) ZH WZ WW ZZ tt fakes Z "!! Z+cc Z+bb Z+jets mistags ) uncertainty % Slice Along the Z+Jets vs. ZH Axis % Slice Along the ZH vs. tt Axis train a d NN to discriminate against backgrounds limit extracted from d distribution in several tag and lepton categories obs (exp) 7. (9.9) M H (GeV/c ) 95% CL Upper Limit/SM 3 CDF Run II Preliminary (.7 fb ) + - ZH " l l bb Expected Observed ±! ±! 6

17 H ττ only search for SM Higgs in the ττ decay mode at a hadron collider contributes % to the final limit looks for one hadronic and one leptonic tau requires two additional jets accepts events from all the production modes including gg H and VBF 7

18 H ττ KS prob =.6% CDF Run II Preliminary KS prob = 6.% CDF Run II Preliminary KS prob = 65.% CDF Run II Preliminary Events 3 ) Observed (L =. fb Jet "! (QCD+Wjets) Top Diboson/Z " ll Z "!! + jets Higgs(M =) 3 H Events 4 3 ) Observed (L =. fb Jet "! (QCD+Wjets) Top Diboson/Z " ll Z "!! + jets Higgs(M =) 3 H Events 3 ) Observed (L =. fb Jet "! (QCD+Wjets) Top Diboson/Z " ll Z "!! + jets Higgs(M =) 3 H NN Score : MixedSig vs Top NN Score : MixedSig vs QCD train 3 NN to distinguish signal from top, QCD and Z background use minimum of NN discriminants as event discriminant observed (expected) limits at 5: 3.5 (4.8) Events #(95% Limit)/(#(SM) BR(H"!!)) KS prob = 68.4% NN Score : MixedSig vs Z"!! Min(NN(Sig Z),NN(Sig vs CDF Run II Preliminary ) Observed (L =. fb Jet "! (QCD+Wjets) Top Diboson/Z " ll Z "!! + jets Higgs(M =) 3 H Top),NN(Sig vs QCD)) vs CDF Run II Preliminary (L =. fb ) Observed Limit Expected Limit P.E.: ± # Band Standard Model (NLO) Higgs Mass [GeV/c ] 8

19 VH qqbb heroic attempt to search for Higgs using only jets not in current combination, hopefully in next identify H as two b-tagged jets identify W/Z as two more jets main problem: background modeling! /! SM 3 CDF Run II preliminary median 68% of experiments L =. fb using ME discriminant obs (exp) 37.5 (36.8) 95% of experiments observed (GeV/c ) m h 9

20 Channel Plot CDF Run II Preliminary, L=.-3. fb 95% CL Limit/SM LEP Excl. fb Obs fb Exp H!"". fb Obs H!"". fb Exp ZH!llbb.7 fb Obs ZH!llbb.7 fb Exp WH+ZH!bbMET. fb Obs WH+ZH!bbMET. fb Exp WH!l#bb.7 fb Obs WH!l#bb.7 fb Exp H!WW 3. fb Obs H!WW 3. fb Exp Combined Obs Combined Exp SM (GeV/c ) January 6, 9

21 Combination Plot CDF Run II Preliminary, L=.-3. fb 95% CL Limit/SM LEP Limit Expected Observed ±σ ±σ SM m (GeV/c ) H January 5, 9

22 mh=6 GeV CDF Run II Preliminary, L=.7-3. fb =6 Mean.5956 RMS.49 Pseudoexperiments CDF Run II Preliminary, L=.7-3. fb Mean.869 RMS.648! / ndf / 67 Prob.639 p 3836 ± 5. p.5946 ±.3 p.85 ±.73 p3 3.7 ±.369 =6 Events 3 CDF Run II Preliminary, L=.7-3. fb =6 GeV/c CDF Data Signal Background !*BR/SM most sensitive mass point overall really just H WW observed limit: σ=.56 * SM expected limit: σ=.75 * SM "*BR/SM Expected Limit/SM xcdf Preliminary Projection, =6 GeV Summer 4 Summer 5 Summer 7 January 8 December 8 With Improvements log (s/b) Integrated Luminosity/Experiment (fb )

23 mh=6 GeV CDF Run II Preliminary, L=.7-3. fb =6 Mean.5956 RMS.49 Pseudoexperiments CDF Run II Preliminary, L=.7-3. fb Mean.869 RMS.648! / ndf / 67 Prob.639 p 3836 ± 5. p.5946 ±.3 p.85 ±.73 p3 3.7 ±.369 =6 Events 3 CDF Run II Preliminary, L=.7-3. fb =6 GeV/c CDF Data Signal Background !*BR/SM most sensitive mass point overall these plots will be really just H WW observed limit: σ=.56 * SM expected limit: σ=.75 * SM explained on future slides "*BR/SM Expected Limit/SM xcdf Preliminary Projection, =6 GeV Summer 4 Summer 5 Summer 7 January 8 December 8 With Improvements log (s/b) Integrated Luminosity/Experiment (fb )

24 mh=6 GeV CDF Run II Preliminary, L=.7-3. fb =6 Mean.5956 RMS this plot compares data to 4 signal and background 35 shapes 3 combines bins of similar s/b 5 Pseudoexperiments CDF Run II Preliminary, L=.7-3. fb 5 Mean.869 RMS.648! / ndf / 67 Prob.639 p 3836 ± 5. p.5946 ±.3 p.85 ±.73 p3 3.7 ±.369 =6 Events 3 CDF Run II Preliminary, L=.7-3. fb =6 GeV/c CDF Data Signal Background !*BR/SM most sensitive mass point overall really just H WW observed limit: σ=.56 * SM expected limit: σ=.75 * SM "*BR/SM Expected Limit/SM xcdf Preliminary Projection, =6 GeV Summer 4 Summer 5 Summer 7 January 8 December 8 With Improvements log (s/b) Integrated Luminosity/Experiment (fb )

25 mh=6 GeV Mean.869 CDF Run II Preliminary, L=.7-3. fb Mean.5956 CDF Run II Preliminary, L=.7-3. fb RMS.648 compares past and present expected! / ndf / 67 m H =6 limits GeV/c RMS Prob p 3836 ± 5. scaled by 4 for Tevatron p.5946 expectation ±.3.8 p.85 ± p3 3.7 ± =6 most sensitive mass point overall really just H WW observed limit: σ=.56 * SM expected limit: σ=.75 * SM Pseudoexperiments 3 =6 5 uses / L scaling with luminosity 5. - band anticipates 5 future improvements !*BR/SM "*BR/SM Expected Limit/SM Events CDF Run II Preliminary, L=.7-3. fb CDF Data Signal Background xcdf Preliminary Projection, =6 GeV Summer 4 Summer 5 Summer 7 January 8 December 8 With Improvements log (s/b) Integrated Luminosity/Experiment (fb )

26 mh=6 GeV CDF Run II Preliminary, L=.7-3. fb =6 Mean.5956 RMS.49 Pseudoexperiments CDF Run II Preliminary, L=.7-3. fb Mean.869 RMS.648! / ndf / 67 Prob.639 p 3836 ± 5. p.5946 ±.3 p.85 ±.73 p3 3.7 ±.369 =6 Events 3 CDF Run II Preliminary, L=.7-3. fb =6 GeV/c CDF Data Signal Background !*BR/SM most sensitive mass point overall really just H WW observed limit: σ=.56 * SM expected limit: σ=.75 * SM "*BR/SM Expected Limit/SM xcdf Preliminary Projection, =6 GeV Summer 4 Summer 5 Summer 7 January 8 December 8 With Improvements log (s/b) Integrated Luminosity/Experiment (fb )

27 mh=5 GeV CDF Run II Preliminary, L=.-3. fb.8.6 =5 Mean.57 RMS.47 Pseudoexperiments CDF Run II Preliminary, L=.-3. fb Mean 3.4 RMS.4! / ndf 58.4 / 64 Prob.6739 p ± 7. p.999 ±.833 p.834 ±.79 p3.6 ±.93 =5 Events 3 CDF Run II Preliminary, L=.-3. fb =5 GeV/c CDF Data Signal Background !*BR/SM most sensitive low mass point above LEP exclusion observed limit: σ=3.76 * SM expected limit: σ=3.7 * SM "*BR/SM Expected Limit/SM xcdf Preliminary Projection, =5 GeV Summer 5 Summer 6 Summer 7 January 8 December 8 With Improvements log (s/b) Integrated Luminosity/Experiment (fb ) 3

28 Limit Calculation the limit calculation is done in a purely bayesian framework for the Tevatron combination we compared it with a frequentist calculation and found agreement within % for the combined limit we take all results back to the stage of data histograms and signal and background templates the limit is calculated from this in a single step on the following pages, I will explain how: first for a single bin counting experiment then for a combination of bins finally how to incorporate systematic error 4

29 Bayesian vs. Frequentist what is the difference between Bayesian and Frequentist? simple example: what does mtop=7.4±. mean? Bayesian answer: we believe the mass of the top to be between 7. and 73.6 with 68% probability (credibility) Frequentist answer: No, No, you can t do/say that! either the mass is in that range or it is not you can t assign a probability just because you don t know we say that our experiment set the mass range we know that if we repeated the experiment many times it would include the right value 68% of the time (coverage) 5

30 Counting Experiments a counting experiment consists of simply counting events equivalent to a single histogram bin need some definitions now: n: the observed number of events b: the expected number of background s: the assumed expected number of signal events true value unknown L(n;s): the likelihood (probability) of observing n events giving an expected signal of s for counting experiments we use poisson statistics: L(n;s)=pois(n,s+b)=c (s+b) n e -(s+b) 6

31 Bayesian Posterior try to calculate the (posterior) probability p(s;n) of having a true expected signal of s, when observing n events exploit Bayes theorem: p(s;n)=l(n;s) Ps(s)/Pn(n) we are missing the functions Ps and Pn: Ps is unknown and interpreted as the (prior) probability of an expected signal s before the measurement arbitrarily choose it to be flat ( for s, for s < ) Pn is unknown, but constant for any n only affects normalization: p(s;n)=cn L(n;s) normalize p(s;n) as probability: p(s;n)ds= 7

32 Observed Limits a 95% confidence interval R has the property: R p(s;n)ds=.95 can choose R arbitrarily CDF Run II Preliminary, L=.7-3. fb.8.6 =6 Mean.5956 RMS.49.4 can choose to cover the maximum values of p(s;n) can choose to leave.5% uncovered on each end for upper limits choose a range starting at : R=[,x] x is then the 95% limit !*BR/SM x depends on the choice of Ps e.g.: choosing Ps flat in ln(s) would give a different result 8

33 Expected Limits expected limits show the sensitivity of the experiment they do not take the observed events into account at all based on pseudo-experiments (PE) with signal run complete limit calculation on each PE plot the obtained limits for a large number of PEs take the median as the expected limit take quantiles %, 6%, 84% and 98% 5 as borders for & σ bands bands not to be mistaken for errors Pseudoexperiments CDF Run II Preliminary, L=.-3. fb 4 3 Mean 3.4 RMS.4! / ndf 58.4 / 64 Prob.6739 p ± 7. p.999 ±.833 p.834 ±.79 p3.6 ±.93 =5 excess of observed limit (red) would be sign of Higgs slice through limit vs. mass plot "*BR/SM 9

34 Multi-Bin Limit we rarely do counting experiments anymore in reality use histograms for data and signal and background expectation every histogram bin interpreted as a counting experiment individual likelihood per bin: Li(ni;s)=pois(ni, s si + bi) ni and bi correspond to the n and b for single bin si is the relative signal acceptance for each bin overall likelihood is defined as L(n,n, ;s)=l(n;s) L(n;s) from there everything goes as before PEs consist of the numbers ni, generated independently also works when combining several histograms 3

35 Nuisance Parameters I have not yet described the incorporation of systematic uncertainties ignoring them is unrealistic simple example: assume we have a % error on b handle it by introducing a nuisance parameter x redefine likelihood: L(n;s,x)=pois(n, s + b (+. x)) however: likelihood is not complete somebody measured the % error on b his likelihood is: Lx(n;s,x)=gaus(x) overall combined likelihood: L (n;s,x)=l(n;s,x) Lx(n;s,x) 3

36 Prior Probabilities recall: p(s,x;n)=c L (n;s,x) P(s) P(s) is the prior probability, assumed flat now we can restructure: p(s,x;n)=c L (n;s,x) P(s)=c L(n;s,x) gaus(x) P(s) define: P (s,x)=gaus(x) P(s) p(s,x;n)=c L(n;s,x) P (s,x) P (s,x) is the new prior probability now incorporates all our prior knowledge on x can be easily extended to a large number of systematics 3

37 Limits With Systematics we are stuck with a d posterior now: p (s,x;n) need to get back to d for calculating the limit could calculate limit as function of x (like we do for mh) however: we are doing Bayesian calculations as such, we can combine different hypothesises can just add up all the different hypothesises for a given s do that by integrating over x: p (s;n)= p (s,x;n)dx for pseudo-experiments pick nuisance parameters randomly according to prior probabilities means that every pseudo-experiment happens at a different point in nuisance parameter space 33

38 Putting It Together expected entries in each bin: Sij(x,x, )=sij Πk(+yijk xk) Bij(x,x, )=bij Πk(+zijk xk) yijk and zijk are the fractional errors in bin i for systematic k and template j likelihood per bin: Li(ni;s,x,x, )=pois(ni, s ΠSik + ΠBik) prior probability: P(s,x,x, )=Πkgaus(xk) posterior probability: p(s,x,x, ;n,n, )=P(s,x,x, )ΠLi(ni;s,x,x, ) p (s;n,n, )= p(s,x,x, ;n,n, )dxdx calculate 95% limit A as: A p (s;n,n, )=.95 34

39 other stuff templates derived from data events or Monte Carlo limited statistics lead to statistical errors on templates assign a statistical error to each sij and bij integrate over the statistical errors with extra nuisance parameters for each bin can also use posteriors to perform measurements the mean and RMS correspond to the measured value and its error similar to fitting, but more stable (and slower) only thing left: calculating the integral can be done by Monte Carlo integration in a few hundred/thousand dimensions 35

40 Markov Chain MC Monte Carlo integration method improvement on scattershot MC faster and more stable results problem: most integration points contribute little would like to sample only the peaks of the distribution problem: we throw nuisance parameters, not likelihood solution: we slowly walk along the peaks taking care not to step into the valley Markov Chain Monte Carlo 36

41 Markov Chain MC commonly used for Bayesian integrals in high dimensions creates a sequence of integration points relative frequency is according to relative weights individual points can be repeated starts from a random point and run until it reaches the stationary distribution using Metropolis-Hastings algorithm at each point use a (gaussian) proposal function to select a nearby point x compare weights: new/old > : move to new point new/old=p < : move to new point with probability p 37

42 Summary and Outlook presented an overview of CDF Higgs results showed the latest CDF Higgs combination provided an introduction into Bayesian limit calculation did not explain all the details that go into preparing the inputs for the combination (the main part of the work) expect that we combine the CDF results with D results for the Winter conferences should hopefully show a nice exclusion region at high mass also working on combining MSSM Higgs limits 38

43 backup slides

44 Signal Cross Sections 4

45 Limit Values 4

46 s/b integrated Cumulative Events CDF Run II Preliminary, L=.-3. fb =5 GeV Signal+Background Background CDF Data Cumulative Events CDF Run II Preliminary, L=.7-3. fb =6 GeV Signal+Background Background CDF Data Integrated Expected Signal Integrated Expected Signal corresponds to sliding a cut on the s/b for a counting experiment note that the data points are correlated 4

47 Tevatron Potential Probability of! Excess LEP Exclusion xcdf Preliminary Projection Analyzed L= fb With Improvements Analyzed L=5 fb With Improvements January 6, (GeV/c ) solely based on extrapolation of expected limits not taking data collected so far into account Probability of 3! Evidence LEP Exclusion xcdf Preliminary Projection Analyzed L= fb With Improvements Analyzed L=5 fb With Improvements January 6, 9 (GeV/c ) 43

48 p values used if you actually want to discover something run background only PEs p value is the fraction of PEs are more signal like than the observation low p value corresponds to discovery could use observed limit here however there are better observables best observable: ratio of likelihood of signal+background and background-only hypothesis 44