Search for B → K(*) Neutrino Anti-Neutrino Decays Using a New Probabilistic Full Reconstruction Method [Elektronische Ressource] / Sebastian Neubauer. Betreuer: M. Feindt

Sebastian Neubauer

Sujets

• Physik

Informations

• Publié par: karlsruher_institut_fur_technologie
• Publié le: 01 janvier 2011
• Nombre de lectures: 26
• Poids de l'ouvrage: 3 Mo

Extrait

(∗)
Search for B → K νν¯ Decays
Using a New Probabilistic Full
Reconstruction Method
Zur Erlangung des akademischen Grades eines
Doktors der Naturwissenschaften
von der Fakult¨ at fur¨ Physik des
Karlsruher Institut fur¨ Technologie (KIT)
genehmigte
Dissertation
von
Dipl.-Phys. Sebastian Neubauer
aus Schw¨abisch Gmund¨
Tag der mundlic¨ hen Prufung:¨ 18. November 2011
Referent: Prof. Dr. M. Feindt
Korreferent: Prof. Dr. G. Quast2Contents
I. Introduction 1
1. Introduction 3
II. Theoretical Foundations 5
2. The Standard Model of Particle Physics 7
2.1.StructureoftheStandardModel........................... 7
2.1.1.TheLeptons.................................. 8
2.1.2.TheQuarks................................... 9
2.2.ForcesoftheStandardModel............................. 9
2.2.1.TheElectroweakForce 9
2.2.2.TheStrongForce................................ 1
2.3.BeyondtheStandardModel.............................. 13
2.4.RareBDecays..................................... 14
3. Parameter Estimation 15
3.1.CountingExperiments................................. 15
3.1.1.FrequentistApproach 17
3.1.2.BayesianApproach............................... 20
3.1.3.Markov-ChainMonteCarloMethod..................... 21
3.1.4.LimitEstimationwithSystematicUncertainties .............. 21
4. Multivariate Analysis Algorithms 23
4.1.Introduction....................................... 23
4.2.ArtificialNeuralNetworks 23
4.3. Preprocessing in NeuroBayes ............................. 24
III. The Experimental Setup 27
5. The Belle Experiment 29
5.1. The KEKB Accelerator . . . 29
5.2.TheBeleDetector................................... 30
6. Particle Identification 35
iContents
IV. The Full Reconstruction Method 37
7. Introduction 39
7.1.Overview........................................ 39
8. Functional Requirements 41
8.1.Reconstruction..................................... 41
8.2.EfficiencyMaximization................................ 43
8.3. Adjustable Purity.................................... 43
8.4. Throughput Constraint 4
9. The Hierarchical Approach 45
9.1.ExploitingtheNaturalHierarchicalStructure.................... 45
9.2. Estimate Posterior Probabilities............................ 45
10.The Framework Structure 49
10.1.The DecayChannel and DigitalPhysicist Classes................. 49
10.2.The ParticleClas.................................. 49
10.3.The VariableClas 50
10.4.The NeuroBayesInterface............................... 51
10.5.TheFourStages.................................... 51
10.6.TheDecayChannels 51
10.7.TheNeuroBayesTrainings 52
11.Performance Optimizations 55
11.1.The Best B CandidateSelection.......................... 5Tag
¯12.Suppression of non BB Background 57
12.1.TopologicalVariables.................................. 57
12.1.1.Fox-WolframMoments............................. 57
12.1.2.ThrustAngle 57
12.1.3. Momentum Direction of the B candidate................. 57Tag
12.2.Continuum Suppression Module............................ 57
13.Resulting Performance 59
13.1.Fits to the M Distribution.............................. 59bc
13.1.1.WithoutNewChannels 60
13.2.EfficiencyEstimationUsingaSignal-SideAnalysis................. 60
14.The ekpturbo Module 63
15.Conclusion 65
V. The Analysis 67
16.Introduction 69
16.1.Overview........................................ 69
16.2.OutlineoftheAnalysis................................. 71
iiContents
17.Candidate Reconstruction 73
17.1.UsedDataSamples................................... 73
17.2.TrackSelectionCriteria................................ 74
17.3.CandidateSelectionCriteria.............................. 74
18.Cut-based Analysis 77
18.1.Signal-SideSelectionCriteria............................. 77
18.2.ComparisonBetwenMonteCarloandData..................... 78
18.3.BackgroundComposition............................... 81
18.4.BackgroundEstimationandSignalEfficiency.................... 86
18.5.SystematicUncertainties 91
18.5.1.SystematicUncertaintiesontheSignalNormalization........... 91
18.5.2.SystematicUncertaintiesontheBackgroundLevel............. 93
18.6.LimitEstimation.................................... 96
19.Neural-Network-based Analysis 103
19.1.SelectionCriteria....................................103
19.2.TheSetofVariables..................................103
19.3.TheTrainingResults109
19.4.OptimizationoftheNeuroBayesCut.........................112
19.5.Background Estimation and Signal Efficiency for the NeuroBayes Selection . . . 112
19.6.SystematicUncertaintiesfortheNeuroBayesSelection...............14
19.6.1.SystematicUncertaintiesontheSignalNormalization...........14
19.6.2.SystematicUncertaintiesontheBackgroundLevel.............14
19.7.LimitEstimation15
20.Partial Unblinding 121
20.1.Cut-basedSelection..................................121
20.2.NeuroBayesSelection..................................121
20.3.FurtherChecks.....................................123
21.Further Improvements 125
VI. Conclusion 127
VII.Appendix 131
22.Appendix for the Full Reconstruction 133
22.1.VariablesUsedfortheNeuroBayesTrainings....................133
22.1.1.Stage1Trainings................................133
2.1.2.Stage2Trainings137
2.1.3.Stage3Trainings138
2.1.4.Stage4Trainings138
(∗)23.Appendix for the Analysis B → K νν¯ 139
23.1.N-1Plots........................................139
23.2.VariablesUsedfortheNeuroBayestraining.....................145
iiiContents
List of figures 176
Bibliography 181
ivPart I.
Introduction
11. Introduction
There is an intrinsic dilemma in high energy physics. On the one hand, the ultimate goal is to
find a model which describes all observable processes in nature and gives conclusive answers to
all remaining questions. On the other hand, such a model would of course question the right
to exist for this field of physics. As of today, the situation is even worse. After an explosion
of surprising new findings from the 50’s to the 70’s a visionary called Standard Model was able
to explain all observations up to that time and was established in the mid 70’s. Furthermore,
all predictions made by this model are found to be correct inside the errors until today. But
for all that we know that this model is not the theory of everything as it does not answers all
the questions we have. For this reason we are obliged to find experimental deviations from the
Standard Model predictions in order to gain information about the next steps towards our final
goal.
Inparticlephysics, therearetwofrontierswheretosearchforsuchadeviation. Inthehighenergy
frontier one tries to find new heavy particles produced in high energetic particle collisions. In the
precision frontier on tries to find quantitative deviations from the Standard Model predictions.
Hints for physics beyond the Standard Model will show up indirectly in tiny deviations from the
Standard Model predictions. The Belle experiment, and also this thesis, is part of the latter. In
the field of B-physics one tries to find deviations in the B meson sector by precisely measuring
decay rates, CP-asymmetries and other parameters.
The special setup of the Belle experiment enables a unique method to measure otherwise inac-
cessible decay channels, called the full reconstruction method. In this thesis such a method was
developed introducing a new hierarchical r approach with extensive usage of multi-
variate analysis algorithms. With this new method we could double the efficiency compared to an
existing cut-based method. For this we reconstruct B mesons in 1104 exclusive decay channels,
employing nearly hundred neural networks in a hierarchical structure based on each other, in
order to be able to reduce the processing time by many orders of magnitude. This new method
is now officially used by the Belle collaboration for many interesting analyses and increases the
sensitivity for the search for tiny deviations from the Standard Model predictions.
In the second part of my thesis I performed an analysis using the new full reconstruction method
(∗)as a hadronic tag. I searched for the decays B → K νν¯. These decays are very interesting
because they are predicted by the Standard Model to be highly suppressed, but the theoretical
error on the branching ratio is very small. As we know that contributions from physics beyond
the Standard Model are small, these rare decays are interesting, because here, contributions from
the S Model and from new physics can be in the same order of magnitude. Depending on
the new physics model, the branching ratio of such a rare decay can even be increased by several
(∗)orders of magnitude. Because of the two neutrinos in the final state of the decays B → K νν¯
it is not possible to measure the quantities of these decays with conventional reconstruction
methods. It is only possible using the full reconstruction method, where the missing information
from the escaping neutrinos is compensated by fully reconstructing the entire event. The tag-
side is reconstructed using the full reconstruction tool and the rest of the event is used as the
(∗)signal-side B → K νν¯ decays in several modes.
It turned out that there is a significant improvement on the expected upper limit