Zelimir Djurcic

1
Search for Oscillation Signal at MiniBooNE

• Zelimir Djurcic
• Columbia University

6th Rencontres du Vietnam, Hanoi, Vietnam, 2006
2
Before MiniBooNE The LSND Experiment
LSND took data from 1993-98 - 49,000 Coulombs
of protons - L 30m and 20 lt Enlt 53 MeV
Saw an excess of??e 87.9 22.4 6.0
events. With an oscillation probability of
(0.264 0.067 0.045). 3.8 s significance for
excess.
Oscillations?
Signal p ? e n n p ? d ?(2.2MeV)
Need definitive study of ????e at high ?m2
MiniBooNE
3
MiniBooNE
(Booster Neutrino Experiment)
4
Search for ?e appearance in ?? beam
Use protons from the 8 GeV booster ? Neutrino
Beam ltE?gt 1 GeV
FNAL 8 GeV Beamline
50 m decay pipe
MiniBooNE Detector 12m diameter sphere 950000
liters of oil (CH2) 1280 inner PMTs 240 veto PMTs
decay region ? ? ??? ,  K ? ???
little muon counters measure
K flux in-situ
magnetic horn meson focusing
?? ??e?
absorber stops undecayed mesons
magnetic focusing horn
????e ???
5
Energy Calibration
We have calibration sources spanning wide range
of energies and all event types !
Michel electrons from ? decay provide E
calibration at low energy (52.8 MeV), good
monitor of light transmission, electron PID
12 E res at 52.8 MeV
?0 mass peak energy scale resolution at medium
energy (135 MeV), reconstruction
cosmic ray ? tracker cubes energy
scale resolution at high energy (100-800 MeV),
cross-checks track reconstruction
PRELIMINARY
provides ? tracks of known length ? E?
6
Particle Identification
Cerenkov rings provide primary means of
identifying products of ? interactions in
the detector
beam m candidate
nm n ? m- p
Michel e- candidate
ne n ? e- p
beam p0 candidate
nm p ? nm p p0
n n
p0 ? gg
7
Particle Identification II
Angular distributions of PMT hits relative to
track direction
muon
PRELIMINARY
Search for oscillation ne n ? e- p events is by
detection of single electron like-rings, based on
Cerenkov ring profile.
electron
8
Signal Separation from Background
Search for O(102) ?e oscillation events in O(105)
?? unoscillated events
Backgrounds
Reducible NC ?0 (1 or 2 e-like rings) ??N? decay
(1 e-like ring) Single ring ? events
Irreducible Intrinsic ?e events in beam from K/?
decay
p0?g g
Signal
??N?
9
Background Rejection and Blind Analysis
Two complementary approaches for reducible
background
Simple cutsLikelihood easy to understand
Boosted decision trees maximize sensitivity
MiniBooNE is performing a blind analysis

• We do not look into the data region where the
  oscillation candidates
• are expected (closed box).
• We are allowed to use
• Some of the info in all of the data
• All of the info in some of the data
• (But NOT all of the info in all of the data)

10
Boosting PID Algorithm
Boosted decision trees

• Go through all PID variables and find best
• variable and value to split events.
• For each of the two subsets repeat
• the process
• Proceeding in this way a tree is built.
• Ending nodes are called leaves.
• After the tree is built, additional trees
• are built with the leaves re-weighted.
• The process is repeated until best S/B
• separation is achieved.
• PID output is a sum of event scores from
• all trees (score1 for S leaf, -1 for B
  leaf).

Reference NIM A 543 (2005) 577.
Boosting Decision Tree
Boosted Decision Trees at MiniBooNE Use about
200 input variables to train the trees -target
specific backgrounds -target all backgrounds
generically
PRELIMINARY
Muons
Electrons
11
Likelihood Approach
Compare observed light distribution to fit
prediction Does the track actually look like an
electron?
Apply likelihood fits to three hypotheses -single
electron track -single muon track -two
electron-like rings (?0 event hypothesis )
Form likelihood differences using minimized
logL quantities log(Le/L?) and log(Le/L?)
log(Le/L?)
log(Le/L?)lt0 ?-like events
log(Le/L?)gt0 e-like events
PRELIMINARY
12
log(Le/L?) Current ?0 Studies

• Ntank gt 200, Nveto lt 6, Fid.Vol.
• No Michel electron
• 2-ring fit on all events

Reconstructed ?0 mass
Translate reconstructed ?0 events into the
spectrum of mis-identified events!
PRELIMINARY
Not looked into this region expect osc.
candidates (blindness)
The data is used to test likelihood based e/?0
separation.
PRELIMINARY
Good data/MC agreement demonstrates robust ?0
reconstruction
13
Appearance Signal and Backgrounds
Full data sample 5.3 x 1020 POT

• Oscillation ?e
• Example oscillation
• signal
• ?m2 1 eV2
• sin22? 0.004
• Fit for excess as
• function of
• reconstructed ?e
• energy

14
Appearance Signal and Backgrounds

• MisID ??
• of these
• 83 ?0
• Only 1 of ?0s are misIDed
• Determined by clean ?0 measurement
• 7 ? ? decay
• Use clean ?0 measurement to estimate ? production
• 10 other
• Use ?? CCQE rate to normalize and MC for shape

15
Appearance Signal and Backgrounds

• ?e from ?
• Measured with ?? CCQE sample
• Same parent ? kinematics
• Most important low E background
• Very highly constrained (a few percent)

16
Appearance Signal and Backgrounds

• ?e from K
• Use High energy ?e and ?? to normalize
• Use kaon production data for shape

17
Appearance Signal and Backgrounds

• High energy ?e
• data
• Events below 2.0 GeV still in closed box (blind
  analysis)

18
Important Cross-check
comes from NuMI events detected in MiniBooNE
detector!
We get ?e , ?? , ?0 , ?/- , ? ,etc. events from
NuMI in MiniBooNE detector, all mixed together
Use them to check our ?e reconstruction
and PID separation!
Remember that MiniBooNE conducts a blind data
analysis! We do not look in MiniBooNE data
region where the osc. ?e are expected
The beam at MiniBooNE from NuMI is significantly
enhanced in ?e from K decay because of the
off-axis position.
MiniBooNE
Decay Pipe
Beam Absorber
NuMI events cover whole energy region relevant to
?e osc. analysis at MiniBooNE.
19
Events from NuMI beam
Boosted Decision Tree
Likelihood Ratios
e/?
PRELIMINARY
PRELIMINARY
e/?
Data/MC agree through background and signal
regions
20
MiniBooNE Oscillation Sensitivity
MiniBooNE aims to cover LSND region. Almost
there, with final work on systematic error
determination
?
LSND best fit sin22? 0.003 ?m2 1.2 eV2
21
Recent MiniBooNE Progress
Total accumulated dataset 7.5 x 1020 POT,
worlds largest dataset in this energy
range. Jan 2006 Started running with
antineutrinos. Detected NuMI neutrinos using
in analysis. Oscillation Analysis progress
results are expected soon.
22
Backup Slides
23
More ?0 Studies
24
MiniBooNE CC? Cross-Section
Obtained by multiplying measured CC ?/QE ratio
by QE ? prediction (?QE with MA1.03 GeV, BBA
non-dipole vector form factors)

• Efficiency
• corrected
• CC ?/QE ?
• Ratio
• measuremet
• on CH2

current systematics estimate - light
propagation in oil 20 - ? cross sections
15 - energy scale 10 - statistics 5
25 lower than prediction, but within errors
25
PID Inputs
Calibration Sample
Signal-like Events
Primary Background
Mean 1.80, RMS 1.47 Mean 1.19, RMS 0.76
Mean 20.83, RMS 25.59 Mean 3.48, RMS 3.17
Mean 16.02, RMS 25.90 Mean 3.24, RMS 2.94