Title: A search for B ?? ? Recoiling Against B-?D0l-?X at babar
1A search for B??? Recoiling Against B-?D0l-?X
at babar
- Stephen Jacob Sekula
- The University of Wisconsin-Madison
- Ph.D. Thesis Defense
- February 27, 2004
2overview
- Theoretical Background and Motivation
- The PEP-II/BaBar B-Factory
- Tagging Events using Semi-Leptonic B-Decay
- Selecting B???
- Systematic Effects
- The Limit on the Branching Fraction
- Results and Conclusions
3The Standard Model
- Matter is described by the interactions between
quarks and leptons - Their interactions are described by a symmetry
group
electric charge
electric charge
LEPTONS
4quark interactions
- The relationship between weak and mass quark
eigenstates is given by the Cabibbo-Kobayashi-Mask
awa (CKM) matrix - Need to probe quark interactions cleanly
W
5The rare decay B?l?
l
Pure leptonic decay, free of hadronic
uncertainties
W
B
?
- The Standard Model branching fraction is
helicity suppression
6Theoretical Expectation
.
- StandardModelParameters
- Branching Fraction Expectations, by lepton flavor
7Existing measurements
- Only limits have been set on the branching
ratio - My work is conducted at the Stanford Linear
Accelerator Center's PEP-II/BaBar B-Factory.
8pep-II
- PEP-II is an electron-positron storage
ring/collider
PEP-II Specifications
Energy (HER/LER) 9.0/3.1 GeV Luminosity
(Instantaneous) 7.11033cm-2s-1 (250,000
BB/day)
9The ?(4S) decay
Electrons (9.0 GeV) and Positrons (3.1 GeV)
collide so that EcmM?(4S) (10.58 GeV)
The first B-meson decays
B1
B2
The second B-meson decays
10The babar detector
- Asymmetric Design
- Five Detector Sub-systems
- Silicon Vertex Tracker(SVT)
- Drift Chamber (DCH)
- DIRC (Cherenkov detector)
- Electromagnetic Calorimeter (EMC)
- Instrumented Flux Return (IFR)
11Silicon Vertex Tracker (SVT)
- Double-sided detector modules, arranged in five
layers - Single-hit resolution 10-40?m (incidence
angle-dependent) - important for high-efficiency track
reconstruction (97 efficient)
12Drift chamber (DCH)
- Detects ionization of gas by charged particles
- Important for high-efficiency track
reconstruction and vertexing - Hit resolution0.1mm
- Energy loss overdistance (dE/dx)needed for
particle ID (distinguishing kaons and pions by
4? up to 1 GeV/c)
13DIRC
- Detects Cherenkov radiation of incident charged
particles - velocity dependence of Cherenkov radiation allows
for particle species separation - gt90 kaon identi-fication efficiency (lt5 pion
misidentification) up to 4 GeV/c (gt4?
separation)
14Electromagnetic calorimeter (EMC)
- Cs-I crystals arranged in a quasi-projective
geometry - 30 MeV resolution for1 GeV photon
- Essential ingredientin electron
identification(90 electron efficiencywith 0.1
pion mis-identification rate) formomenta above
700 MeV
15instrumented flux return (IFR)
- Steel interwoven with resistive-plate chambers
(RPCs) - RPCs detect ionization by incident particles
(typically muons), or nuclear interactions
by/decay of long-lived hadrons (i.e. K0L) - Muon efficiency (? mis-id)2000 65 (5)2002
40 (8) - Muon efficiency is time-dependent due to loss
in RPC efficiency
16Data samples
Data used was collected between the summers of
1999 2002. NBB(88.9 1.0)106
Monte Carlo (MC) simulations are performed of the
major background species for this
analysis 3data for BB 1data for continuum
In figures and quoted yields, the MC is scaled to
the data luminosity
17analysis overview
- The events of interest to this search contain one
B meson decaying to ?? - This leaves one of the B-mesons unconstrained
potentially large source of contamination - Strategy
- First eliminate one of the B mesons by either
completely or partially reconstructing it - Constrain the remaining charged/neutral particles
to be consistent with B??? - Extract the signal-content in the data using this
technique
18A Unique search strategy
Reconstruct one of the B-mesons in a
semi-leptonic, open charm decay mode (B-?D0l-?X)
tag B meson
B-
l
?
B
recoil or signal B meson
?
Search the recoiling particles for B???
This is a new approach to searching for this rare
decay.
?
19Semi-leptonic B Decay
- Semi-Leptonic, Open Charm B-decays have
inherently high branching fractions - High-statistics tagged samplefrom which to start
the search for B??? - The neutrino cannot be reconstructed. Particles
other than the neutrino (X ?0,? from
higher-mass D meson decay) can also be missing - Higher levels of background than in
fully-reconstructed events
20B-?D0l-?X Event selection
.
- Global event restrictions
- at least one high-momentum lepton, zero event
charge, and uniform distribution of
charged/neutral particles in center-of-mass frame
(characteristic of B decays) - Selection of D0 mesons
- Decay modes D0?K-?, K-??0, K-??-?, KS0??-
(KS0???-) this is 28.2 of the D0 branching
fraction - D0 mass within 3? of the fitted mean
- Selection of Dl Candidates
- Vertex pairs of leptons and D0 mesons, constrain
them to be consistent with B decay
21Dl Reconstruction
- Vertex D0s and leptons. Also vertex D0s and ?s
to make Dl. These are collectively called Dl
candidates - Define the reconstructed angle between the Dl and
the B that would have created it (in ?(4S) rest
frame)(assumes only the neutrino is missing) - -2.5 lt cos?B,Dl lt 1.1
22D0l-?X Selection Efficiency
- The total tag efficiency in MC is
- factoring out the B-?D0l-?X and D0 branching
fractions, the method is 15.8 efficient at
selecting semi-leptonic decays (for the chosen D0
decay modes). - In 88.9 million BB events, the expected yield of
semi-leptonic tags is 395,000 - After selecting events with one B-?D0l-?X, the
tracks and neutrals remaining in each event (the
recoil or signal side) are constrained. - This sample is the basis for the search
23data/MC efficiency
.
- Efficiency is compared in data/MC using
double-tagged andsingle-tagged events - Tag Efficiency is related to yield by
24data/MC Efficiency (II)
.
- Assumption preselections affect both single and
double tags in the same way (?1?2) - The ratio N1/N2 cancels many terms
- The efficiency correction is then
25Signal Side Selection
- The search proceeds by isolating the decay
modes - One track (signal track) left over after
reconstructing B-?D0l-?X - Must be either a muon or an electron and NOT a
kaon - originate from close to the ee- interaction
point - low momentum (plt1.2 GeV/c), consistent with a
secondary decay - little energy remaining in the calorimeter
- After making all these selections, extract the
signal content from the data
26Extra Neutral energy
- Main discriminating variable
- The neutral energy remaining in the
calorimeterafter accounting for neutralenergy
from the tag B meson
Only signal and background MC
Signal BR 110-4
27Signal efficiency
- Total Monte Carlo signal efficiency
- factoring out the tag side efficiency and the
??l?? branching fraction, the efficiency is
30 - The Standard Model predicts
- Monte Carlo background expectation
- Background events are dominated by semi-leptonic
decays with missing charged particles or a K0L
28extracting the signal content from data
- The Eextra distribution has distinct shapes for
signal and background - Develop models for the neutral energy
- Combine the models into a likelihood function
- Fit the data to determine the signal and
background content
29Signal model
two gaussians
- Signal eventssubdivide intotwo classes
- The signal model isa combination of the models
for each subclass
Events with no physics
source of energy
30Signal Model validation
- Use double-tagged eventsto create a sample with
similarEextra properties
31BackGRound model
- A 3rd order polynomial is used to model Eextra
from background events - Other models aredeveloped for latercross-checks
of thefit - Data extrapolation of the polynomial from
0.5?0.0GeV in Eextra yields a background
estimate of 119.812.5 (compared to 123.96.9 in
MC)
32Maximum likelihood fit
- The signal and background models (PDFs) are
combined into a single maximum likelihood
function - ?s and ?b are the signal and background yields
to be fitted in the data - n is the total number of events in the data
- Fs(Eextra) and Fb(Eextra) are the signal and
background shapes for Eextra
33Likelihood fit checks
- Generate and fit toy MC with varying signal
hypotheses - Fit background-only MC samplesof size compatible
with the data
0 input signallt?fittedgt0 ?fitted5
GEANT4 background-only MC
5 input signallt?fittedgt5
34Data fit result
- The data is fit usingthe likelihood function
- Background yield (115.211.8) is in
goodagreement with estimatesfrom background
modelextrapolation (119.812.5) and MC
(123.96.9) - Signal yield (14.86.3) is 2? from zero
- 1.3? difference from Standard Model prediction
(42) - A limit will be set on the branching fraction
signalbackground
background
signal
35Efficiency systematicuncertainties
- Several sources of systematic uncertainty enter
into the efficiency estimate - Semi-leptonic tag B reconstruction (double-tags)
- encapsulates all particle identification,
tracking, etc. effects on the tag B meson - Particle identification of signal track
- Neutral particle simulation (Eextra cut
efficiency) - Tracking Efficiency (probability of
reconstructing signal track) - The systematic uncertainty will propagate into
the limit-setting procedure
36Particle id efficiency (signal track)
.
- Particle ID (PID) efficiency is not modeled
perfectly in MC - PID is studied in data/MC as a function of
species - PID Tables provide ?data/?MC as a function of
momentum and angle ? track-by-track weight to
correct efficiency - BaBar uses these PID Tables to correct efficiency
in the MC, but the tables do not take account of
correlations between particle selection criteria - I construct custom PID tables combining the
lepton acceptance and kaon veto used on the
signal track
37Particle id efficiency (II)
.
- I compare the correction from the custom PID
tables to those originally used to correct the
MC - Weighting the corrections by contribution to the
signal MC yields the total PID correction
MUONS
ELECTRONS
Low Momentum, Low Efficiency
38Total systematic uncertainty
- The corrected signal efficiency is
39Limit setting procedure
- The limit setting procedure is based on the CLs
Method (a modified frequentist approach) - Define an estimator, Q
- Determine the value of Q for the result (Qdata)
- Determine the valeue of Q for a given
signalbackground (background only) hypothesis
Qsb(Qb) - The confidence level (CL) of a signal hypothesis,
s, is defined as
40Limit setting (II)
- The fitted signal yield, ?sfitted, is the chosen
estimator - Determine the distribution of fitted signal
yields, D(?sfittedsb), from toy MC - Count the number of experiments below a given
point - The point ?s90 is the signal expectation that
corresponds to the 90 confidence level
(CLs1-0.9) for a given s
41Nominal sensitivity
- The sensitivity is the expected limit when no
signal is present - The branching fraction sensitivity is then
- This is the most sensitive search in the world
Null Signal Hypothesis
zero signal
data result(?s9024.2)
data result
42The impact of Shape systematic effects
- Several factors could affect the limit in data
- The background shape may be differerent
- Different true parameterization
- Unmodeled feature of the shape (i.e. a
signal-like feature) - The signal shape may be different
- Study the effect of such possibilities and choose
the most conservative option (i.e. use nominal
models, add signal-like features to the
background model, etc) - observe effects on the limit in the
background-only GEANT4 MC samples
43summary of shape study
Change the background parameterization No Effect
on Limit
Change the background shape according to data/MC
agreement in several control samples Nominal
Choice of Shape is most Conservative
Change signal shape within uncertainty No Effect
on Limit
The nominal PDF parameterizations are used to
fit the data and set the limit
44Including Efficiency Systematics in the limit
- Include the 9.9 uncertainty on the signal
efficiency and 1.1 uncertainty on the B-meson
count in the limit - Including systematics in the limit
calculation,I obtain the dotted curve - The data result ishigher than the
expectedresult when no signal is present
Nominal limit including systematics is 2.710-4
Nominal limit without systematics is 2.310-4
data result
45Data branching fraction limit
- The limit obtained from the fitted signal yield
in data (14.86.3) - After including efficiency systematic
uncertainty - The limit is worse than that expected if there's
no signal - The background-only assumption may not completely
apply because the Standard Model does predict
signal events (of order the fit uncertainty) - The signal yield is 2? from zero but 1.3? from
the Standard Model
46summary
- A search for B??? has been conducted at the
PEP-II/BaBar B-Factory - One B-meson is reconstructed as B-?D0l-?X
- Search the recoiling particles for B???
- A signal yield in data (14.86.3) is obtained
from a maximum likelihood fit - A limit is set on the branching fraction
47outlook
- The B-Factory program will collect 500 fb-1 by
2006 - the current BaBar dataset totals 170 fb-1 and
will exceed 200 fb-1 by summer - the KEK-B/Belle B-Factory has not reported a
result for B???, but its data sample will
exceed 250 fb-1 by summer - Currently, this analysis is sensitive to Standard
Model effects. By 2006, it will reach 5?
(statistical) discovery potential - A discrepancy with the Standard Model could
indicate new physics
current
(5?)
2006
48Backup slides
49The importance of B?l?
.
.
- Allowed by the Standard Model but not yet
observed - Pure leptonic decay no uncertainty from QCD
- In conjunction with measurements of Vub, offers
a chance to measure fB - fB is known only from lattice QCD and is a
critical ingredient in the theory of B meson
flavor mixing - Potentially sensitive to new physics that might
enhance or suppress the decay - charged Higgs exchange (SUSY)
50B-mixing
.
B mixing frequency
parameters and QCD terms
Use the Wolfenstein approximation to relate the
ratio to other constraints on the CKM unitarity
triangle
51supersymmetry
.
- The Minimal SuperSymmetric extension of the
Standard Model (MSSM) requires two Higgs doublets - one gives mass to up-type quarks, the other to
down-type - The introduction of a charged Higgs boson
corrects the branching fraction - A branching fraction measurement can set limits
on tan? and the charged Higgs mass
52B????
.
-
- Inherently not a significant background source
- Eextra cut is a powerful rejection factor in this
analysis
53Particle ID Efficiency by selector and year
.
ELECTRONS
MUONS
- The Particle Identification can change slightly
over time with changing detector conditions. - Muon identification is particularly affected by
the degrading performance of the RPCs in the IFR
54Global event selection
.
- Select events with at least one high-mometum
lepton - pl gt 1.0 GeV/c (l is either an electron or a
muon) - Require events to be well reconstructed
- Net event charge is zero
- Events are consistent with the target class
- Total number of charged particles 10
- Large missing mass (Mmissinggt1.0GeV/c2)
- Events are not collimated (R2alllt0.9) particles
BB decay tend to be uniformly distributed in the
?(4S) frame
55D0 Reconstruction
.
- D0 mesons are reconstructed in the modes
D0?K-?,K-??0, K-??-?, KS0??- (KS0???-) - Requirements
- daughters converge at a common point
(vertex) - reconstructed D0 has mass between
1.81-1.91 GeV/c2 reconstructed D0 has momentum
0.5 lt p lt 2.5 GeV/c
reconstructed D0 mass within 3? of the fitted
mean
K-??0
56treatment of multiple Dl candidates
- An event typically has multiple Dl candidates
- A best candidate is selected using the D0 mass
- mass probability density functions (PDFs) are
generated from reconstructed Dl candidates that
match real ones in simulation - If the best candidate isa Dl, the event is
vetoed
57Double tag method
.
- Use only the double-tagged events to compare
data/MC - assumptions
- the efficiency of selecting each semi-leptonic B
decay is the same - the double tags are dominated by BB- (2 from
B0B0 in MC) - Ratio of efficiencies in data and MC
- Compatible with single/double-tag method
58Signal Side Selection
.
- The search proceeds by isolating the decay
modes - One track (signal track) left over after
reconstructing B-?D0l-?X - The charged track must be either a muon or an
electron and NOT a kaon - The track must originate from close to the ee-
interaction point
59Continuum rejection
.
- Removing continuumbackground relies onevent
topology - Use angle between Dl thrust and signal track to
reject collimated events
Reject ee-???- using the minimum invariant
mass calculable from any three tracks in an
event
60Signal track momentum
.
- The signal track from B??? is the result of a
secondary decay (typically low momentum)
electron
muon
(Note for these plots, only events with lots of
extra energy in the calorimeter are retained.
This keeps background and removes signal)
61Energy and momentum
.
- Look at Eextra for high (gt1.2 GeV/c) and low
(lt1.2 GeV/c) momentum signal tracks
psignallt1.2 GeV/c
psignalgt1.2 GeV/c
- Notes on the disagreement
- not associated with any one (set of) MC
background species - Disagreement lives in psignalgt1.2 GeV/c (70)
- Signal MC contains only 30 of signal events
above 1.2 GeV/c
62Neutral simulation
63Tracking efficiency
- BaBar has also studied track reconstruction
efficiency as a function of momentum and angle - Weights relating efficiency in data/MC
- Applying these weights to the signal track yields
the efficiency correction - The correction has an uncertainty of 0.008 per
track
64Cls example 90 CL
This is repeated for many signal hypotheses to
determine the relationship between ?sfitted and
?s90
fitted
fitted
65Effect of a background estimate discrepancy
- Background Estimate(s)
- 119.812.5 (Data Extrapolation)
- 123.96.9 (Monte Carlo)
- Use Limit Curve on theGEANT4 experiments
Variation in the nominal upper limit is less than
one event Negligible effect
66Uncertainty on the background shape
- Potentially more serious concern
- Need a data control sample that mimics the Eextra
shape that expected from the background - Unable to isolate such a sample
- Do the next best thing develop samples related
to the signal topology and compare data/MC
agreement - Use bin-by-bin agreement as a correction function
to the nominal background PDF - Generate toy MC with the new PDF, fit the toy MC
with the original (nominal) PDF and observe the
effect on the CL curve
67background shape (II)
- Control samples for data/MC agreement study
-
-
High-statistics test of the Eextra agreement
Wrong-charge sample, close to signal sample, low
statistics
Neutral B-meson sample, wrong charge, very low
statistics
68background shape III
Eextra
Data/MC Comparison
MC Fit
ExampleCheck the effect by applying the
derived CL curve on the GEANT4 experiments
Data Fit
Data/MC Fit Ratio (Correction Function)
The nominal background model is more conservative
and is kept
69Uncertainty on the signal shape
- Change mean of peak gaussian at 0.2 GeV
- based off the uncertainty from the double-tag
model validation - shift the mean by 0.015 GeV
- No significant effect