First two sided limit on BR(B s µ + µ - ) Matthew Herndon, University of Wisconsin Madison SUSY 2011 M. Herndon, SUSY 2011 1
B s(d) µ + µ - Beyond the SM! Indirect searches for new physics! Look at processes that are suppressed in the SM! Excellent place to spot small non SM contributions! B s(d) µ + µ -! SM:! No tree level decay! GIM, CKM and helicity suppressed! BF(B s µ + µ - ) = 3.2x10-9! New Physics:! Loop: MSSM: msugra, Higgs Doublet! Rate tan 6 β/(m A ) 4! Large enhancement possible M. Herndon, SUSY 2011 2
Tevatron and CDF! Tevatron: 2TeV pp collider! CDF properties! Silicon Tracker EXCELLENT TRACKING Results in this talk uses 6.9fb -1! η <2, 90cm long, r L00 =1.3-1.6cm! Drift Chamber(COT)! 96 layers between 44 and 132cm! Muon coverage! Triggered to η <1.0 TRIGGERED AT 1.5 GeV/c B s(d) µ + µ - large integrated lumi of the Tevatron and the excellent tracking of the CDF detector M. Herndon, SUSY 2011 3
! Apply minimal selection! Ensure sample of well measured dimuon events! Final discrimination with NN! Relative normalization search B s(d) µ + µ - Method! Apply same sample selection criteria for the B + J/ψK +, J/ψ µ + µ - sampe! Measure relative decay rate of B s(d) µ + µ - BF(B s " µ + µ # ) = (N # N ) cand bg $ B +% B + $ Bs % Bs N B + f u f s BR(B + " J /&K + ) BR(J /& " µ + µ # ) 5 X 10 8 B s events M. Herndon, SUSY 2011 4
! Estimate all relative selection acceptances and efficiencies.! Identify variables that discriminate signal and background! Validate modelling using B + B s(d) µ + µ - Method! Design multivariate discriminant, NN, for background rejection! Unbiased optimization based on Pythia signal MC and part of mass sidebands. Excludes masses just below B mass B µ + µ - X! Validate performance on B + data! Estimate combinatoric background level from sidebands! Separately estimate B hh! Validate background prediction method in control regions designed to be enhanced in expected backgrounds! Check low significance signal regions before highest significance region M. Herndon, SUSY 2011 5
! Need to discriminate signal from background! Reduce background by a factor of ~ 10000! Signal characteristics! Final state fully reconstructed Signal vs. Background µ+ L3D P(µµ) 55K Events L3D di-muon vertex µ-! B s is long lived (cτ = 438 µm)! B fragmentation is hard: few additional tracks! Background contributions and characteristics primary vertex µ+ L3D! Sequential semi-leptonic decay: b cµ - X µ + µ - X! Double semileptonic decay: bb µ + µ - X! Continuum µ + µ -! µ + fake, fake+fake! Partially reconstructed, lower p T, short lived, doesn t point to the primary vertex, and has additional tracks - L3D P(µµ) µ- di-muon vertex primary vertex Cut on mass, lifetime, p T, how well p points to the vertex and isolation M. Herndon, SUSY 2011 6
Discriminating Variables! 14 discrimating variables +! Mass m µµ 2.5σ window: σ = 24MeV/c 2! First 5 variables are the most powerful! Combine in NN! Unbiased optimization based on simulated signal and data sidebands:! Determine background by extrapolating mass sidebands: Extensively tested for mass bias! Perform search in two detector regions, 8 NN bins and 5 mass bins M. Herndon, SUSY 2011 7
Combinatoric background! Combinatoric background estimated by extrapolating mass sidebands! Shape (slope) with mass determined using all events NN>0.7! Systematic uncertainties from statistical power of sidebands samples and variation of the fit functions M. Herndon, SUSY 2011 8
B hh background! Peaking background dominated by B hh, h = pi K! Rates of hadron misidentification as muons estimated from D* + D 0 π + K - π + π + decays! Large sample allows determination of misidentification rate vs. pt and luminosity and study of run conditions dependence B s backgrounds vs NN: order 1% X1% all but B s ππ shifted down in mass M. Herndon, SUSY 2011 9
Control Regions! Use independent data samples to test background estimates! OS-: opposite sign muons, negative lifetime (signal sample is OS+) µ+ µ! SS+ and SS-: same sign muons, positive and negative lifetime. No trigger matching! ** OS-, SS: Opposite side B hadrons! FM: fake µ enhanced, one µ fails the muon Id cuts, loose vertex cuts. Has a significant B->hh contribution! ** FM: False muon backgrounds! Compare predicted vs. observed # of bg. events: For multiple NN cuts primary vertex Fake di-muon vertex M. Herndon, SUSY 2011 10
Control Regions! Background predictions and observed background in control regions B hh contribution to FM control region! 64 Independent checks of the background estimation method! B hh seen with expected mass shape and is well estimated M. Herndon, SUSY 2011 11
Expected Sensitivity! Efficiencies and acceptances Have reached single event sensitivity to the SM rate of B s(d) µ + µ - M. Herndon, SUSY 2011 12
Expected Sensitivity M. Herndon, SUSY 2011 13
Results! Error are combined systematic and statistical Poisson interval! Significant excess observed in B s CC channel M. Herndon, SUSY 2011 14
Limits! In the assumption that the observed events are from background processes! Limits using the CLs technique! Incorporates all systematic uncertainties BF(B s µ + µ - ) < 4.0x10-8 at 95% CL BF(B d µ + µ - ) < 6.0x10-9 at 95% CL M. Herndon, SUSY 2011 15
P Values! P values in the background only and background + SM signal hypothesis! Found by comparing an ensemble of pseudo experiments for each hypothesis to the observed data! Systematic uncertainties are included in the pseudo experiments P Value B 0 : 23.3% P Value B s : 0.27% P Value B s : 0.66% Only bins with significant signal expectation P Value B s (with SM signal): 1.92% P Value B s (with SM signal): 4.14% Only bins with significant signal expectation M. Herndon, SUSY 2011 16
Signal Hypothesis! BF(B s(d) µ + µ - ) in the hypothesis that the observed events have a significant contribution from either a SM or a new physics source of B s(d) µ + µ -! BF(B s(d) µ + µ - ) and 90% C.L. interval using a simple likelihood fit BF(B s µ + µ - ) = 1.8 +1.1 x10-8 -0.9 4.6x10-8 < BF(B s µ + µ - ) < 3.9x10-8 at 90% CL M. Herndon, SUSY 2011 17
All Experiments! LHCb: BF(B s µ + µ - ) < 1.5x10-8 at 95% CL! CMS: BF(B s µ + µ - ) < 4.0x10-8 at 95% CL! LHC combination: BF(B s µ + µ - ) < 1.08x10-8 at 95% CL! CDF: 4.6x10-8 < BF(B s µ + µ - ) < 3.9x10-8 at 90% CL BF(B s µ + µ - ) = 1.8 +1.1 x10-8 -0.9! LHC 95% C.L. combination limit compatible within the 1σ CDF error! Both experiments see low significance excesses compatible with SM (naïvely subtracting the backgrounds gives ~2x SM excesses)! Mild disagreement between the experiments. More data needed M. Herndon, SUSY 2011 18
B s B µ + - s µ + : µ Physics - Conclusion Reach! First significant excess observed in a B s µ + µ - search! Bounds set at 90C.L. 4.6x10-8 < BF(B s µ + µ - ) < 3.9x10-8 Example of physics reach Green area of m 0 vs. m 1/2 CMSSM plane favoured by B s µ + µ - result Would indicate reasonably detectable super partner masses and high tan(β) Could be exciting times ahead! M. Herndon, SUSY 2011 19
Backup Material M. Herndon, SUSY 2011 20
Discriminating Variables! BSignal simulated with Pythia! Reweighed to match observed B s p T and isolation distributions M. Herndon, SUSY 2011 21
Discriminating Variables M. Herndon, SUSY 2011 22
Variables: Validation! B + J/ψK +, J/ψ µ + µ - sample used to validate kinematic, NN variables, and NN performance! Simulated using Pythia as with signal same! Reweighed to match observed B + p T and isolation distributions M. Herndon, SUSY 2011 23
Variables: Validation M. Herndon, SUSY 2011 24
Mass Bias! Search method requires there be no mass dependence in the NN! Conbinatoric background estimated by extrapolating! Shape (slope) with mass determined using all events NN>0.7! Check for NN correlation with mass in sideband and control sample! Test if NN training can distinguish mass M. Herndon, SUSY 2011 25
Mass Bias M. Herndon, SUSY 2011 26
Result M. Herndon, SUSY 2011 27
Result B 0 M. Herndon, SUSY 2011 28
Results B s M. Herndon, SUSY 2011 29
Results B s M. Herndon, SUSY 2011 30
Test Statistic and Interpretation! -2ln(Q), Q=L(s+b data)/l(b data)! L is a likelihood obtained by multiplying all 80 Poison probabilities of hypothesis s+b or b given the observed data. The likelihood is minimized with respect to nuisance parameters that model the systematic uncertainties and a freely floating parameter for the signal.! P values compare the values of 2ln(Q) observed in data to a ensemble of pseudo experiments in a given hypothesis.! The most likely value of the signal is fit using the same likelihood and the confidence intervals are determined by using the change in chi2 M. Herndon, SUSY 2011 31