1. 1 First observation of a New b-baryon  b at D0: Celebrating 30 Years of Beauty @ Fermilab Eduard De La Cruz Burelo University of Michigan for the DØ Collaboration

2. 2 The quest … a long journey Everything begins at … Fermilab Everything begins at … Fermilab The quest for B hadrons The quest for B hadrons The quest for … the  b The quest for … the  b  b signal  b signal Mass measurement Mass measurement Relative production ratio Relative production ratio Summary Summary

3. 3 The quest begins 30 years ago at… Fermilab's giant accelerator reveals another new sub- nuclear particle Summer 1977 !! EXTRA!! Fermilab Experiment Discovers New Particle "UPSILON" B Physics, a whole field, was born on June 30, 1977, here at Fermilab.

4. 4 Since then … Exclusive B decays CLEO (1983) Bc by CDF (1998) Main assumption to look for these particles: ½ of the mass of the upsilon!. Last B meson in the ground 0 - state to be observed

5. 5 Recently … October 2006 CDF announced a preliminary result using 1.1 fb -1 of the  b ’s almost 8 months ago.

6. 6 Status: Mesons: Mesons: B +, B 0, B s, B c + ( B +, B 0, B s, B c + (established) B* (established), B d **(submitted to PRL DØ, Preliminary CDF) B s ** (Preliminary DØ & CDF) Baryons: Baryons:  b (established)  b (established)  b +, and  b *+ (preliminary CDF)  b +, and  b *+ (preliminary CDF)

7. 7 The quest for b baryons Plus there is a J = 3/2 baryon multiplet

8. 8 Data we use … In this analysis we use 1.3 fb -1 of data collected by DØ detector (RunIIa data). Thanks to the Fermilab Accelerator division for doing wonderful work. Keep delivering, and we will keep collecting data and analyzing. The entire D0 detector is important, but the muon and central tracker subdetectors are particularly important in this analysis

9. 9 Motivation Spectroscopy: Spectroscopy: One of the best ways to test our understanding of QCD and potential models One of the best ways to test our understanding of QCD and potential models Production and Fragmentation: Production and Fragmentation: Major source of uncertainty in many measurements Major source of uncertainty in many measurements Discovery: Discovery: Practice techniques for BSM searches by finding undiscovered SM particles. Practice techniques for BSM searches by finding undiscovered SM particles.

10. 10 Understanding these: Helps us understand these: D s ** B d ** bb X(3872) D sJ   BELLE BABAR JLAB

11. 11 Theoretical prediction of the masses E. Jenkins, PRD 55, R10-R12, (1997). Predicted mass hierarchy: M(Λ b )< M(  b ) < M(  b ) Just today, first citation … Karliner et al., arXiv:0706.2163

12. 12 LEP measurements: LEP experiments only deduce the presence of the  b - indirectly; they look for an excess of events in the right-sign combinations of  - (  -) l -  - mass Wrong sign combination events They measure the lifetime of this excess of events. 1.45 +0.55/−0.43(stat.) ± 0.13(syst.) ps. Eur. Phys. J. C 44, 299–309 (2005)

13. 13 What do we know about the  b - ? Predicted mass: 5805.7 ± 8.1 MeV Predicted mass: 5805.7 ± 8.1 MeV Predicted to follow the mass hierarchy Predicted to follow the mass hierarchy M(Λ b )< M(  b - ) < M(  b ) By using preliminary  b mass measurement from CDF and predicted mass hierarchy: By using preliminary  b mass measurement from CDF and predicted mass hierarchy: 5.624 GeV < M(  b - ) < 5.808 GeV  b - lifetime by LEP: 1.42 +0.28/-0.24 ps.*  b - lifetime by LEP: 1.42 +0.28/-0.24 ps.* * This is the world average (ALEPH+DELPHI). HFAG: arXiv:0704.3575 [hep-ex]

14. 14 Our knowledge about b-baryons D0 has experience with the  b : 3 results on the  b lifetime. D0 has experience with the  b : 3 results on the  b lifetime.  Λ πp Beam line

15. 15  πp Beam line π   Searching for  b in  b - →J/  +  -

16. 16 Impact parameter cut … a killer   p  When tracks are reconstructed, a maximum impact parameter is required to increase the reconstruction speed and lower the rate of fake tracks. But for particles like the  b -, this requirement could result in missing the  and proton tracks from the  and  - decays  

17. 17 What did we do to solve this problem? We need to open up the IP at reconstruction We need to open up the IP at reconstruction To reprocesses all DØ data with a wider IP for track reconstruction is a very difficult task. But … To reprocesses all DØ data with a wider IP for track reconstruction is a very difficult task. But … Thanks to our muon detector (and the guys from the muon team), J/  →  +  - is a golden channel.. Although B -> J/  X is fairly rare, it is very clean channel for us and easy to trigger on. Thanks to our muon detector (and the guys from the muon team), J/  →  +  - is a golden channel.. Although B -> J/  X is fairly rare, it is very clean channel for us and easy to trigger on. We therefore reprocessed DØ RunIIa data for events containing a J/ , which is ~35 million events. We therefore reprocessed DØ RunIIa data for events containing a J/ , which is ~35 million events.

18. 18 Mass distribution for K 0,Λ 0 and  - signals for the “standard” (bottom histograms) and “extended” (opening up IP) tracks reconstruction. K0S→+-K0S→+- →p-→p-

19. 19 Mass distribution for  - signal for the “standard” (bottom histograms) and “extended” (opening up IP) tracks reconstruction.

20. 20 Reconstruction strategy for  b Reconstruct J/  →  +  - Reconstruct J/  →  +  - Reconstruct  → p  Reconstruct  → p  Reconstruct  →  +  Reconstruct  →  +  Combine J/  +  Combine J/  +  Improve mass resolution by using an event-by-event mass difference correction. Improve mass resolution by using an event-by-event mass difference correction. The guides: The guides: The sister:  b →J/  decays in data The impostor: J/  +  (fake from  (p  - )  + ) The clone: Monte Carlo simulation of  b - →J/  +  -

21. 21 Natural constraints in  b - →J/  +  - Natural constraints in  b - →J/  +  - Three daughter signal particles need to be reconstructed: Three daughter signal particles need to be reconstructed:  →p+   →p+   →  +   →  +  J/  →  +  - J/  →  +  - The final state particles (p,  -,  - ) have significant impact parameter with respect to the interaction point. The final state particles (p,  -,  - ) have significant impact parameter with respect to the interaction point. Charge correlation: both pions must have the same charge Charge correlation: both pions must have the same charge ~5 cm c  =7.89 cm ~5 cm c  =4.91 cm

22. 22 More features in  b - →J/  +  -  b has a decay length of few hundred microns, PV separation  - has a decay length of few centimeters.  has a decay length of few centimeters ~5 cm ~0.1 cm

23. 23 Reconstructing the daughters Background events from wrong-sign combinations ( Λ(p  - )  + ) J/  →  +  -  → 

24. 24 What background do we expect? Prompt background: Prompt background: ~80% of the J/  are directly produced at the collision. ~80% of the J/  are directly produced at the collision. Real B’s: Real B’s: The remaining ~20% of J/  come from B decays The remaining ~20% of J/  come from B decays Combinatoric background: Combinatoric background: Real J/  plus fake  - Real J/  plus fake  - Fake J/  plus fake  - Fake J/  plus fake  - Fake J/  plus real  - Fake J/  plus real  - Real J/  plus real  -, but not from  b - Real J/  plus real  -, but not from  b - Our wrong-sign combination events have these.

25. 25 Determination of Selection Criteria To retain efficiency, try to keep cuts loose To retain efficiency, try to keep cuts loose We use independent samples: We use independent samples:  b →J/  decays from data  b →J/  decays from data Background from wrong-sign combination Background from wrong-sign combination Background from J/  sideband events Background from J/  sideband events Background from  - sideband events Background from  - sideband events Use  b - signal MC events only when no choice (e.g., pion from Xi) Use  b - signal MC events only when no choice (e.g., pion from Xi)

26. 26 Example 1: pT(  - ) from  Λ b →J/  (μ+μ-)  (p  )

27. 27 Example 2: pT(  - ) from  - Monte Carlo of  b - →J/  +  - Background events from wrong sign combination (Λ(p  - )  + )

28. 28 Example 3: topology cut  Λ  Collinearity in XY: Cosine(  ) pT(  ) Monte Carlo of  b - →J/  +  - Background events from wrong sign combination (Λ(p  - )  + ) Cos(  )>0.99 100% efficiency

29. 29 Finally we have:  b Selection  →p  decays:  →p  decays: pT(p)>0.7 GeV pT(p)>0.7 GeV pT(  )>0.3 GeV pT(  )>0.3 GeV  - →  decays:  - →  decays: pT(  )>0.2 GeV pT(  )>0.2 GeV Transverse decay length>0.5 cm Transverse decay length>0.5 cm Collinearity>0.99 Collinearity>0.99  - b particle:  - b particle: Lifetime significance>2. ( Lifetime divided by its error ) Lifetime significance>2. ( Lifetime divided by its error )

30. 30 So now … let’s first look at the background control samples after all cuts We have three independent background samples: We have three independent background samples: Wrong sign combination (fake  - ’s Wrong sign combination (fake  - ’s from  (p  - )  + ) J/  sideband events  - sideband events.

31. 31 J/   (p  - )  + Background: Wrong sign combinations No peaking structure observed in this background control sample

32. 32 Background: J/  sideband events J/   (p  - )  + No peaking structure observed in this background control sample

33. 33 Background:  - sideband events J/   (p  - )  - No peaking structure observed in this background control sample

34. 34 Now let’s look at background MC We investigated with high MC statistics, B decay channels such as: We investigated with high MC statistics, B decay channels such as: No peaking structure observed any these B decays MC samples

35. 35 What we expect: signal MC Mean of the Gaussian: 5.839 ± 0.003 GeV Width of the Gaussian: 0.035 ± 0.003 GeV MC events of  b-→J/  +  - Input mass at generation level = 5.840 GeV

36. 36 Looking at data Clear excess of events just below 5.8 GeV

37. 37 Y Z Event scan of event in the signal peak

38. 38 Y Z Event scan of event in the signal peak

39. 39 Mass measurement Fit: Unbinned extended log-likelihood fit Gaussian signal, flat background Number of background/signal events are floating parameters Number of signal events: 15.2 ± 4.4 Mean of the Gaussian: 5.774 ± 0.011(stat) GeV Width of the Gaussian: 0.037 ± 0.008 GeV Compare to width measured in MC: 0.035 ± 0.003 GeV

40. 40 Nothing in the background samples: Bkg from  sidebands Bkg from J/  sidebands Bkg from wrong-sign

41. 41 Significance of the peak Two likelihood fits are perform: Two likelihood fits are perform: 1. Signal + background hypothesis (L S+B ) 2. Only background hypothesis (L B ) We evaluate the significance: We evaluate the significance: Significance of the observed signal: 5.5  Significance of the observed signal: 5.5 

42. 42 Alternative significance In the mass region of 2.5 times the fitted width centered on the fitted mass, 19 candidate events are observed while 14.8 ± 4.3 (stat.)+1.9/−0.4 (syst.) signal and 3.6 ± 0.6 (stat.)+0.4/−1.9 (syst.) background events are estimated from the fit. The probability of backgrounds fluctuating to 19 or more events is 2.2 ×10 −7, equivalent to a Gaussian significance of 5.2σ

43. 43 Consistency checks Decay length distribution Decay length distribution

44. 44 Intermediate Resonances Signal visible in all intermediate resonances J/     m(  ) m(p  ) m(  )

45. 45 A second analysis approach would be … A different approach is to rely heavily on Monte Carlo simulation. A different approach is to rely heavily on Monte Carlo simulation. There are many multivariate techniques in the market: There are many multivariate techniques in the market: Artificial Neural Networks, Artificial Neural Networks, Boosted Decision Trees, Boosted Decision Trees, Likelihood ratio, Likelihood ratio, etc. etc. As a cross check we used Decision Trees As a cross check we used Decision Trees

46. 46 Example: Decision Trees (BDT) In order to apply the same BDT to  b in data, we use only J/  and  variables as input to the BDT. In order to apply the same BDT to  b in data, we use only J/  and  variables as input to the BDT. Minimum overlap between BDT and cut-based variables

47. 47 Decision Trees We observe a signal consistent with that observed with cut-based analysis. We observe a signal consistent with that observed with cut-based analysis. Only ~50% overlap between selected events and the cut- based analysis. Only ~50% overlap between selected events and the cut- based analysis. Width consistent with MC Width consistent with MC Background shape consistent with wrong-sign combination shape. Background shape consistent with wrong-sign combination shape.  b - using BDT A multivariate technique with a simple set of input variables, not including  - variables, also results in a  b - signal.

48. 48 Combining: cuts+BDT After we remove duplicate events, we observe 22.8 ± 5.8 events. After we remove duplicate events, we observe 22.8 ± 5.8 events. Significance: Significance: Sqrt(-2  L) = 5.9

49. 49 Systematic Uncertainties on Mass Fitting models Two Gaussians instead of one for the peak. Negligible. First order polynomial background instead of flat. Negligible. Momentum scale correction: Fit to the  b mass peak in data, < 1 MeV. Fit to B 0 signal peak. Negligible effect < 1 MeV. Study of dE/dx corrections to the momentum of tracks finds a maximum deviation of 2 MeV from the measured mass. Event selection: From the mass shift observed between the cut-based and BDT analysis, once removing the statistical correlation, a 15 MeV variation is estimated.

50. 50 Discovery!

51. 51 Production ratio In addition to the observation, we also measure: This provides a measurement to allow other experiments to compare their production rate with this result. f(b→X) : fraction of times b quark hadronizes to X

52. 52 Production ratio We use Monte Carlo samples of: We use Monte Carlo samples of:  b - →J/  +  -  b - →J/  +  -  b →J/  +   b →J/  +  MC passed through D0 detector simulation MC passed through D0 detector simulation Same reconstruction and selection criteria as used on data is applied to Monte Carlo. Same reconstruction and selection criteria as used on data is applied to Monte Carlo. Monte Carlo distributions need to be reweighted due to the Data/MC pT spectrum differences and to account for trigger effects. Monte Carlo distributions need to be reweighted due to the Data/MC pT spectrum differences and to account for trigger effects. From comparison of  b kinematic distributions in data and MC, determine further weighting factor, then apply to  b - From comparison of  b kinematic distributions in data and MC, determine further weighting factor, then apply to  b -

53. 53 Systematic uncertainties in the relative production ratio SourceUncertainty (%)  b /  b hadronization models Negligible MC stat. on  b /  b 10 pT(  ) reconstruction 7 Effect of mass difference between data and MC 5  b /  b MC reweighting 27 Syst. uncertainties on the number of  b in data +13, -3 Conservatively take difference between reweighting result and no reweighting.

54. 54 Production ratio Ignoring the ratio of Br’s, from ratio of hadronization fractions of B s to B d, expect ~1/4 or less

55. 55 Last Tuesday June 12, DØ submitted a PRL article announcing the discovery a new b baryon:  b -

56. 56 Production ratio We measure the relative production ratio to be: We measure the relative production ratio to be: Allows comparison between experiments

57. 57 Submitted to PRL 6/12/07 arXiv:0706.1690, Fermilab-Pub-07/196-E

58. 58 www.fnal.gov 6/13/07 Thanks Kurt and Judy!

59. 59 Collaboration The quest begins … and continues @ Fermilab Celebrating 30 years of the b quark discovery @ Fermilab A New b baryon: an anniversary gift