1
1 T2K Experiment
The Tokai to Kamioka long baseline neutrino oscillation experiment, or T2K [?],
is an experiment in Japan to more finely measure the parameters of the lepton
mixing matrix. Long baseline experiments, like T2K, are designed to measure
the masses and mixing angles of the unitary mixing matrix by sending a beam
with well understood energy from a proton accelerator to a large particle detector
positioned to maximize the oscillation probability.
The T2K experiment is composed of a neutrino factory, a near detector, and
a far detector. The neutrino source is a fast extracted proton beam from the
J-PARC proton synchrotron in Tokai, Ibaraki [?], which produces 30 GeV pro-
tons. The far detector, located 295 kilometers away, is the world’s largest water
Cherenkov detector, Super-Kamiokande [?], where neutrino oscillations in atmo-
spheric neutrinos were discovered in 1998 [?]. The near detector is composed of
multiple sub-detectors, placed 280 meters from the beam source in order to mea-
sure the background of the various T2K measurements and help in estimating the
initial flavor content of the beam. T2K is a collaboration of hundreds of physi-
cists from twelve countries. The scale of the effort reflects both the difficulty and
importance of the measurements.
2
1.1 Physics Goals
T2K is designed for several specific physics goals. First is an improved mea-
surement of the oscillation parameters sin(θ23) and ∆m223 . This is found by
measuring the νµ content of the beam at the accelerator, and then measuring the
remaining unoscillated νµ at the far detector. The probability in the three-flavor
approximation for νµ disappearance is [?]
1− P (νµ → νµ)
= sin2 2θ sin2(∆m2
31L
4E) + 4s213s
223(s
223 − c223)
× sin2(∆m2
31L
4E)− c212 sin
2 2θ23(∆m2
21L
4E) sin(
∆m231L
2E)
(1.1)
Several experiments have already measured these oscillation parameters, with
recent measurements by MINOS [?] of sin2(2θ23) >0.90 (90% C.L.) and ∆m223
= (2.32+0.12−0.08) x 10−3 eV2 [?]. T2K hopes to improve the accuracy with which the
parameters are known, especially θ23, and investigate the oscillation on a different
energy scale than the previous long baseline experiments.
The second and most important goal of the T2K experiment is the search for
νµ → νe oscillation by measuring electron neutrino appearance at the far detector.
This particular oscillation is important for several reasons. First, it was designed
to make the first measurement of θ13 from electron neutrino appearance in this en-
ergy range. Secondly, while competing experiments, such as CHOOZ [?] [?], Daya
Bay [?] [?] and RENO [?] have measured disappearance of electron antineutrinos,
with a Daya Bay measurement of sin2 2θ13<0.092±0.016(stat)±0.005(syst), T2K
has the advantage of detecting oscillated electron neutrinos, providing important
constraints. Third, as shown in Eq. 1.1, the νµ disappearance is dependent on the
size and uncertainty on θ13, and having both a νµ disappearance and a νe appear-
ance measurement at the same detector combined with precise θ13 measurements
from reactor experiments will give the most precise θ23 measurements.
3
Third and somewhat more interesting, large values of all the mixing angles are
required to measure the imaginary phase δ, as shown in the mixing matrix [?] [?]
U =
1 0 0
0 c23 s23
0 −s23 c23
X
c13 0 s13e−iδ
0 1 0
−s13eiδ 0 c13
X
c12 s12 0
−s12 c12 0
0 0 1
.
A large value of θ13, as has been seen in Daya Bay, means that long baseline
experiments like T2K and NOvA [?] should be able to make measurements of
the δ parameter. The imaginary phase of the mixing matrix, if non-zero, can be
used to determine the strength of the CP violation in the neutrino sector. This
is fundamentally interesting in its own right and might contribute to a deeper
understanding of the matter-antimatter asymmetry of the universe as discussed in
Chapter ??. Reducing the uncertainty on the parameter θ23 is also very important
in order to precisely constrain δ, as it currently contributes one of the largest
uncertainties [?].
1.2 The T2K Beam
The T2K neutrino beam is created from the collision of protons that have been
accelerated to a kinetic energy of 30 GeV [?]. This acceleration is done in three
accelerators, a linear accelerator (LINAC), a rapid cycling synchrotron (RCS)
and the main ring (MR). First, hydrogen (H−) is accelerated to 400 MeV in the
LINAC, and the electrons are stripped by charge stripping foil at the injection
point of the RCS.
In the RCS, the beam of protons is accelerated up to 3 GeV. Acceleration
in synchrotrons is done by radio frequency (RF) cavities. The nature of the
acceleration is such that the protons naturally form into separate bunches. The
4
Figure 1.1: a)Particles entering an RF cavity at the right moment are accelerated
b)Particles that arrive slightly too early or slightly too late will not receive the necessary
acceleration, and will be pushed in the direction that aligns them with the Vref [?].
electric field accelerates particles maximally at the wave peak and minimally at
the wave trough. Particles arriving just before the wave peak will be accelerated
so that they arrive later in the next cycle, while particles arriving just before
the wave trough will be slowed as shown in Fig.1.1. This results in the particles
clustering in the middle, separated in time by an amount equal to the period of
the field [?].
For T2K, bunches created by the RCS are injected into the MR. In the MR
they are accelerated to their full kinetic energy. Various parameters of the main
ring are described in Table 1.1. In fast extraction mode, eight bunches per spill
are extracted within a single turn by a set of five kicker magnets. This beam is
pointed toward the far detector by the primary beamline, and then interacts with
a stationary graphite target in the secondary beamline.
1.2.1 Beam Modeling
Particle production and interactions for the beam are predicted by a chain of
Monte Carlo simulations as shown in Fig. 1.2 [?]. First, 30 GeV protons are
simulated upstream of the baffle using a Monte Carlo particle generator called
5
Circumference 1567 m
Beam Power ∼750 kW
Beam Kinetic Energy 30 GeV
Beam Intensity ∼3 x 1014 p/spill
Spill Cycle ∼0.5 Hz
Number of Bunches 8/spill
RF Frequency 1.67-1.72 MHz
Spill Width ∼5 µs
Table 1.1: Machine design parameters for the JPARC MR for the data in this thesis
FLUKA. FLUKA was chosen because it shows better agreement with data from
the NA61 [?] experiment compared to other particle generators. Position and
momentum information of secondary particles exiting from the target and from
their parents and decay chains is then used by the T2K simulation package
JNUBEAM [?].
JNUBEAM uses the GEANT3 framework to track particles. Precise simula-
tion of the particle interactions in beam components including the baffle, target,
three horn magnets, helium vessel, decay volume, beam dump and muon monitor,
which will be discussed in the next section, are necessary to accurately predict
the trajectories and energy of the different particles,. JNUBEAM information is
used to predict the neutrino flux at the near and far detectors.
The flux simulations are tuned using information from the NA61 experiment.
Located at CERN, on of the main goals is to look at the interaction of 30 GeV
protons on a graphite target. Currently proton interaction data from a T2K
replica target and a thinner graphite target have been taken. Figure 1.3 shows
the experimental setup. The primary goal of these studies was to measure the
rate of secondary pion and kaon production from these interactions.
6
Figure 1.2: Various components of simulation that are used to make flux predictions
for T2K [?].
Figure 1.3: The NA61 detector. NA61 uses 30 geV protons on a graphite target like
T2K to give precise predictions of particle production [?].
A comparison of pion production data in NA61 with FLUKA Monte Carlo is
shown in Fig. 1.4. Data from NA61 covers about 90% of the pion phase space,
and about 60% of the kaon phase space as shown in Fig. 1.5. Systematics from
7
Figure 1.4: A comparison of pion production data from NA61 with FLUKA predic-
tions [?].
Figure 1.5: Phase space of the π+ (left) and K+ (right) that produce neutrinos in the
beam. The color represents the size of the contribution to the neutrino flux. The phase
space covered by NA61 for π+ and for NA61, Eichten et al. [?] and Allaby et al.[?] for
K+ are highlighted [?].
the beam simulation uncertainties will be discussed in further chapters.
1.2.2 Beam Components
The primary beamline is composed of a preparation section, an arc section and a
final focusing section as shown in Fig. 1.6. The preparation section uses normal
conducting magnets to align the beam so that it can be turned in the arc section.
8
Figure 1.6: Components of the T2K Beamline [?].
Superconducting magnets in the arc section are used to tune the beam direction
such that Super-Kamiokande is between 2 and 2.5 degrees away. Normal con-
ducting magnets in the final focusing section then align the beam again before it
enters the secondary beamline [?].
The primary beamline is also equipped with many devices for measuring prop-
erties of the beam. Preventing beam loss and assuring that the composition of
the beam is well understood are important for producing a stable neutrino beam.
Current transformers, electrostatic monitors, segmented secondary emission mon-
itors and beam loss monitors are used to measure the intensity, profile, position
and loss of the proton beam. Fig. 1.7 shows pictures of the monitors and their
location in the primary beamline.
Current transformers (CT) measure the intensity of the beam with a 50 turn
toroid wrapped around an iron core. To calibrate the CT a second coil is wound
around the core, and an applied current shaped to emulate the beam passing
through the coil is measured. The CT can measure the intensity of the beam to
9
Figure 1.7: Left: Photographs of the primary beamline monitors. Upper left: CT.
Upper right: ESM. Lower left: SSEM. Lower right: BLM. Right: Location of the
primary beamline monitors [?].
within 2%, the relative fluctuations of the intensity to within a 0.5% precision,
and the beam timing to within 10 ns.
Electrostatic monitors (ESM) are composed of four cylindrical segmented elec-
trodes arranged at equal distances around the beam direction. The current that
is measured by each electrode is compared, without interfering destructively with
the beam. This gives the beam distance from each electrode, which is expected
to be equal if the beam were located in the center of the monitor, which is the
desired goal. From the ESM, the beam position is known to within 450 µm, which
is comfortably within the design requirement of 500 µm.
During beam tuning, segmented secondary emission monitors (SSEM) are used
to measure the beam width and divergence. Protons from the beam hit the vertical
and horizontal strips, composed of two titanium foils surrounding an HV foil, that
make up the SSEM. These interactions produce secondary electrons in proportion
to the number of incident protons, which induce currents on the strips that can
then be measured. Since the beam interacts with the foil, there is a beam loss
10
of about 0.005% due to irradiation of nearby components, so the SSEM can not
be used during data taking. The systematic uncertainty on the beam width from
this measurement is 200 µm compared to the requirement of 700 µm.
Beam loss monitors (BLM) are proportional counters with a mixture of argon
and carbon dioxide gas. These are installed near the beam pipe, which means any
beam passing through them represents a “loss”. If more than a certain amount
of beam is lost during a spill, a beam abort interlock signal is fired, stopping the
beam.
The secondary beamline contains the target station, the decay volume and
the beam dump. Protons enter the target station through the beam window,
which separates the vacuum of the primary beamline from the helium gas of the
secondary beamline. A large block of graphite called a baffle collimates the beam
and protects the surrounding devices from the high energy protons.
The proton beam profile in the secondary beamline is then imaged using a
device called an optical transition radiation monitor. In this device a thin tita-
nium, aluminum, or fluorescent ceramic foil, depending on the intensity of the
beam, is placed at a 45 degree angle with respect to the beam direction. Tran-
sition radiation produces light at 90 degrees to the beam direction. This light is
then collected after reflection through a series of mirrors in a region with lower
radiation levels, where a charge injection camera can then be used to produce an
image of the beam.
The beam then interacts with a graphite target, 1.9 interaction lengths long,
surrounded by a graphite tube and a titanium case. Helium gas flowing into the
gaps between the different components cools the target. Pions are produced from
the collision which are then focused along the beam direction by three magnetic
horns.
The horns are made of two coaxial aluminum alloy conductors encompassing
a closed volume. The first horn is primarily cylindrical, as shown in Fig. 1.8, and
11
Figure 1.8: A schematic showing a cross section of the target and the first magnetic
horn [?].
has the target inserted into its downstream end to increase pion containtment.
The second and third horns are parabolic in shape, and look like Fig. 1.9 [?].
As shown in Fig. 1.9, current moving in loops in the horn produces a magnetic
field. This magnetic field goes as ∼ 1
r, where r is the distance from the horn axis.
Secondary pions traversing the horn toward the axis are either focused or deflected,
depending on the direction of the current and their charge [?]. Having three horns
allows pions in a wide range of energy and angle to be focused, collecting them
with the first horn and focusing with the other two. The neutrino flux in the peak
energy region, 0.6 GeV, was increased by about a factor of 16 through the use of
these horns.
The majority of the produced pions decay in the decay volume to produce
muons and muon neutrinos. Other decays and contributions from kaons produced
in the target interaction are discussed in the following chapters. The decay vol-
ume is a 96 m long steel tunnel ending in the beam dump. The helium used
in the decay volume minimizes pion absorption and eliminates the production of
unwanted compounds such as tritium and NOx that would certainly result from
12
Figure 1.9: A cartoon showing the function of a parabolic horn. When current flows
in one direction, π− are deflected and π+ are focused. Reversing the current will reverse
which pion is focused and which is deflected [?].
beam interactions in air.
All hadrons and muons produced from pion decay with energy less than 5 GeV
are absorbed by the beam dump. The beam dump is composed of an inner core,
3.7 m deep, of graphite surrounded by fifteen iron plates comprising a total depth
of 2.4 m. The remaining muons are then measured by a succession of detectors in
order to characterize the neutrino beam.
1.3 Beam Monitors
Monitoring the beam is done by a series of detectors, starting with the muon
monitor, followed by an emulsion tracker, and finally by the INGRID detector
280 m from the target region. The muon monitor measures the direction and the
flux of high energy muons produced by secondary pion decay. It is necessary for
the experiment that the neutrino beam direction be known to within 0.25 mRad
and the intensity within 3%. For this purpose, two detector arrays are used, one
of PIN photodiodes 118.7 m from the target and one of ionization chambers 117.5
13
Figure 1.10: The muon monitor, with the photodiode detector on the right and the
ionization chamber on the left [?].
m from the target as shown in Fig. 1.10 [?].
The seven ionization chambers in the muon monitor are each made up of
seven sensors with active volume of 75x75x3 mm3 fitted inside aluminum tubes.
Depending on the beam intensity either argon with 2% nitrogen or helium with 1%
nitrogen are fed into the tubes, which are kept at constant temperature, pressure
and oxygen contamination.
For the PIN photodiodes, the active area is 10x10 mm2 with a thickness of 300
µm. The silicon layer is fully depleted by applying 80V. The distribution of charge
in the two detectors is used to reconstruct a 2D image of the muon distribution.
The resolution of the beam intensity from the muon monitor measurements is
0.1% and for the profile center is 0.3 cm.
The emulsion tracker is designed in a similar fashion to that used by the
OPERA experiment. Two separate modules are used. The first module is com-
posed of eight emulsion films, and measures the muon flux with a systematic
14
Figure 1.11: Left: The INGRID near Detector Right: A typical neutrino event in an
INGRID module [?].
uncertainty of 2%. For the other module 25 emulsion films are interleaved with
1 mm lead plates. For 2 GeV/c muons, this module can measure the momentum
with a precision of 28%. Interactions in the modules are analyzed by scanning
microscopes.
INGRID sits on the axis of the beam and is used to measure the direction, pro-
file, intensity and if possible, the energy of the neutrino beam [?]. It is composed
of 14 identical modules arranged in a cross shaped configuration as in Fig. 1.11.
Each of these modules is made up of 11 tracking scintillator planes, composed of
24 scintillator bars in the x and y direction read out by MPPCs, and nine 6.5 cm
iron plates.
These modules are surrounded on all sides by veto planes. These planes are
made of 22 scintillator bars segmented in the beam direction. Adjacent modules
can share a veto plane so that every module has either three or four associated
15
veto planes. A typical neutrino interaction event in an INGRID module is shown
in Fig. 1.11, where a particle, most likely a muon, interacts and traverse the
planes, leaving energy deposits whose magnitude is represented by the size of the
circle in the image. From these events INGRID can measure the beam center to
10 cm, which implies a beam angle precision of 0.4 mRad given its location with
respect to the beam.
Besides the 14 standard INGRID modules, two off-axis and one proton module
are also used. The off-axis modules are identical to the standard modules and are
used to check the axial symmetry of the beam. Also, the standard INGRID
modules cannot resolve the proton in charged current interactions. Therefore the
special proton module is constructed entirely of very fine grain scintillator, and
placed adjacent to the central modules in the INGRID cross. Then quasi-elastic
events, which consist of a lepton and a proton, can be detected. The quasi-elastic
cross section is important for comparing different models of beam neutrino Monte
Carlo, and this interaction will be discussed in detail in Chapter 4.
1.4 ND280
The off-axis near detector consists of the the pi-zero detector (also known as the
P0D), three time projection chambers (TPC), two fine grain detectors (FGD), the
electromagnetic calorimeter (ECal), and the side muon range detector (SMRD)
(See Fig. 1.12). The P0D, TPC, FGD and ECal sit inside a support struc-
ture known as the basket and are surrounded by a magnet with inner volume
of 7x3.5x3.6 m3. The SMRD is contained within the magnet yokes, and is used
primarily as a veto for the other detectors [?].
The purpose of ND280 is to provide information to quantify the neutrino flux
at Super-Kamiokande and to study neutrino interactions. Each detector is built
to make sure that the important parameters, such as the νµ and νe components of
16
Figure 1.12: The T2K near Detector [?].
the flux at the near detector and the rate of π0 production, are well understood.
Various neutrino event types such as charged and neutral current quasi-elastic,
resonant and inelastic reactions are all measured in the off-axis near detectors.
These detectors predominately sit at an angle of 2.5 degrees off the axis of
the beam, aligned with the direction to the far detector at Super-Kamiokande.
The T2K experiment measures neutrino oscillations at an off-axis angle based on
a design that was first proposed at Brookhaven National Laboratory in 1995 [?].
As previously discussed, neutrinos in the beam primarily come from the π+ →µ+ + νµ reaction. The kinematics of this reaction create neutrinos with energy
proportional to the pion energy in the direction of the beam.
However, the distribution of the neutrino energies at angles off-axis from the
beam are not proportional to the pion energy. As shown in Fig. 1.13, as the angle
increases, pions at a wide variety of energies begin to produce neutrinos that all
have the same momentum. Therefore, using an off-axis beam will both increase
17
Figure 1.13: Neutrino momentum from the decay of pions at various energies, with
each pion energy represented by a circle. Distributions for angles of 1.5 and 3 degrees
are marked. As the angle increases, all pions regardless of energy start to produce
neutrinos with the same momentum [?].
the flux of neutrinos at a given energy and produce a narrower neutrino energy
spectrum.
For the T2K experiment, the 2.5 degree off-axis angle was chosen to reduce
the high energy tail in the neutrino beam spectrum, reducing the size of the
expected backgrounds from neutral current processes. The distribution of events
in neutrino energy with various off-axis angles is shown in Fig. 1.14. At this angle
the neutrino spectrum peaks at an energy of about 650 MeV and this produces a
theoretical oscillation maximum at the far detector. [?].
18
E (GeV)v
v F
lux (
au)
μ
Figure 1.14: Energy spectra from an off-axis beam at: red line (2 degrees), green line
(2.5 degrees), and blue line (3 degrees) [?].
1.4.1 Tracker
The TPC and FGD make up the Tracker region of the near detector. Each TPC
uses low diffusion gas for momentum resolution of better than 10 percent for
particles with energy less than 1 GeV/c. It is designed to distinguish electron and
muon tracks with various momenta by measurement of dE/dx over a minimum
track length of 60 cm [?]. This is important for making precision measurements
of the νe component of the beam [?] [?].
The TPC have an inner box filled primarily with argon and an outer box filled
with carbon dioxide as shown in Fig. 1.15. Particle interactions ionize the gas
in the inner box and, due to the magnetic field, these ions drift toward readout
planes located on the side of the TPC. The pattern and arrival time of the signal
19
Figure 1.15: An image of the main components of the TPC [?].
can be used to produce a 3D image of the particle trajectory.
The FGD is composed of two targets. One of the two FGD targets is composed
of 30 layers of rectangular scintillator bars arranged in alternating x and y con-
figurations each containing 192 bars. A drawing of an FGD scintillator module is
shown in Fig. 1.16. The scintillator provides target mass for neutrino interactions,
and each is instrumented with MPPCs that measure the energy deposited by any
interacting particle. Fine segmentation is necessary to track low energy protons
in order to distinguish charged current quasi-elastic and non-elastic events.
The second FGD target contains water in order to provide a comparison of
the neutrino cross sections with Super-Kamiokande, a water Cherenkov detector.
Six 2.5 cm thick water targets alternate with seven modules of scintillator, where
each module consists of an x and a y layer. The water targets are made of thin
polycarbonate sealed with polyurethane. A vacuum pump system maintains the
pressure of the water targets, preventing water from spilling into the FGD.
20
Figure 1.16: The FGD with the front cover removed [?].
1.4.2 ECal and SMRD
Both the SMRD and the ECal are used to measure particles which either originate
outside the detector, or that escape the detector before they can be classified. In
its completed form, the ECal surrounds the P0D and Tracker and is used to
measure photons, primarily from π0 decay. It also helps differentiate electrons
and muons [?].
A total of 13 ECal modules of three different types are used. Barrel ECal
surrounds the tracker, P0D ECal surrounds the P0D and a single module called
the DSECal is located downstream of the tracker. Much of the data used in the
analysis described here was taken before the P0D and Barrel ECal were installed.
ECal modules are composed of scintillator and read out by MPPC, similar to
the FGD and INGRID. All of the modules contain scintillator layers combined
21
Figure 1.17: An ECal module [?].
with a lead layer. In the Barrel and DSECal the scintillator layers alternate
between two views separated by 90 and have lead layers that are 1.75 mm thick
to optimize detection of π0. In the P0D ECal the scintillator bars are all aligned
in the same direction and the lead layers are 4 mm thick because the P0D itself is
used for detection of π0. For the P0D ECal the main purpose is to track photons
or particles that leave the P0D and to act as a veto for interactions outside the
detector. An external view of an ECal module is shown in Fig. 1.17.
SMRD is used to detect muons which are at large angle with respect to the
beam which can not be measured in the TPCs and can also veto cosmic rays and
beam interactions in the sand. The magnet has 16 yokes, and in each of these
yokes there are 15 1.7 cm air gaps. SMRD scintillator modules are placed in these
gaps, with three layers of scintillator modules on the top and bottom for all yokes,
and both sides instrumented with three layers for yokes 1 through 5, four layers
for yoke 6 and six layers for yokes 7 and 8.
Each scintillator is read out by MPPCs as in the other scintillator detectors.
The SMRD has a position resolution for tracks, measured from relative timing
information, of 7 cm. It is able to provide a measurement of energy from detected
muons with a resolution less than 10 percent.
22
1.4.3 P0D
The P0D is the most upstream detector in ND280, and the detector measuring
neutrino cross sections with the greatest fiducial volume [?]. Its purpose is to
detect the π0 in neutral current νµ reactions that make up the largest reducible
background to detecting electron neutrinos at the far detector. The detector
dimensions are 2103 mm width, 2239 mm height and 2400 mm length and the
mass of the detector is 15,800 kg with water and 12,900 kg without.
The P0D contains alternating vertical and horizontal (Y and X) layers of
scintillator read out by MPPC much like the FGD, except that the scintillator
bars are co-extruded triangular polystyrene instead of rectangular [?]. Each bar
has a height of 17 mm and a width of 33 mm with a hole in the center with 2.6 mm
diameter which holds the fiber. The fibers are wavelength shifting fibers made by
Kuraray, doped with Y-11 and with a diameter of about 1 mm. Fig. 1.18 shows a
setup with several layers of the triangular scintillator instrumented with a similar
fiber.
The P0D, unlike previous scintillator based neutrino detectors, doesn’t use
photomultiplier tubes to detect light from the scintillator. Instead it uses Multi-
pixel photon counters, or MPPCs. MPPCs are ”avalanche photo-diodes with
metal resistor semiconductor layer like structure operating in the geiger mode” [?].
Basically this means MPPCs are detectors able to measure single photons using
special diodes with an avalanching property on a silicon substrate. Avalanche
photodiodes allow for the detection of single photons by multiplying the photon
into a many electron signal through a process called avalanching, effectively in-
creasing the gain. These have an advantage over photomultiplier tubes in that
they are much smaller and are not affected by magnetic fields.
The avalanching effect is caused by electrons in the semiconductor valence
band which, when reverse biased, are released into the conduction band due to
23
Figure 1.18: A photograph of a test setup for the scintillator. The fiber shown is only
used for the test, and is not the fiber used for T2K, but the holes for holding the fiber
are identical. However, the arrangement of the scintillator into overlapping triangles is
the same configuration used for T2K.
Figure 1.19: Left: A Multi-pixel photon counter. Right: The grid structure which is
composed of multiple photodiodes. Each photodiode section of the grid accepts photons
and then goes into an on state. Total photoelectron count is gotten by summing the
number of activated grid sites.
their thermal energy. If the voltage is high enough, these electrons have the ability
to knock out other electrons and form a high electric current in an effect called
impact ionization. In avalanching photodiodes, absorption of a photon causes the
original release of the electron which then causes the impact ionization.
Each pixel on an MPPC operates in a separate avalanche photodiode and resis-
tor combination [?]. They act as independent photon counters with a gain of the
24
Figure 1.20: The components of the P0D
same order as a photomultiplier. A pixel is activated when a photon causes impact
ionization, and the result is that a single photoelectron is registered. (See Fig.
1.19) These pixels are arranged on a grid, and the total number of photoelectrons
is determined by the total number of pixels activated in the grid.
The signal from an MPPC is read out by TriP-T based front end boards, or
TFBs. Each MPPC is read out by two channels on the TFB, in order to get both
a low and a high gain response to the same input. Therefore each 128 channel
TFB can be used to read out 64 MPPCs.
Charge is integrated in each channel by 23 capacitors. The capacitors integrate
consecutively, with each capacitor having a 480 ns integration window followed by
a 100 ns reset period. These integration windows are set to correspond with the
width of each beam bunch, which is 580 ns, such that charge from interactions
25
Figure 1.21: Read out chain for the P0D electronics. Information from MPPCs is
collected by TFBs, which are read out by RMMs. Data is sent from the RMMs to the
DAQ, and triggers and sychronization signals are sent from the clock modules to the
TFBs [?].
from each bunch are integrated by separate capacitors.
There are four main regions of the P0D, the downstream ECal, the downstream
target, the upstream target and the upstream ECal as shown in Fig. 1.20. The
ECal is composed of alternating podules with 4 mm lead radiator, to increase the
photon conversion probability. In the target region, the radiators are made of 1.3
mm brass and the podule and radiator also alternate with a 3 cm water target,
which is used to compare cross sections with the water target at the far detector.
In total the P0D contains 40 podules with 134 vertical (Y ) and 126 horizontal
(X) bars in each.
Unlike the FGD the water target contains HDPE water bladders which can be
filled or emptied at any time. Drip pans beneath the detector were installed to
catch any water that might leak out of the bags. The depth of the water in the
bags is also monitored with water depth sensors to in case of catastrophic leaks
and to provide information about the water mass.
The detector uses a total of 174 TFBs, with every 29 connected to the data
aquisition (DAQ) system by a read-out merger modules (RMM). Each of the
26
ECal regions is instrumented with 29 TFBs and each of the water target regions
is instrumented with 58 TFBs, so a total of 6 RMMs are used. The DAQ transfers
control commands, “clock” and “trigger” signals to the TFBs and the TFBs then
transfer data and cosmic triggers back to the DAQ through the RMMs as shown
in Fig. 1.21 [?].
“Clock” and ”trigger” signals are produced by the slave (SCM) and master
(MCM) clock modules. Information about the timing of the beam spills, the
“trigger” signals, and synchronization signals for the clocks of all the RMMs and
all the detectors, the “clock” signals, are received by the MCM for use by all of
ND280. The SCM then transfers these signals to the RMMs, keeping the P0D
timing separate from the other ND280 detectors. Other triggers, such as light
injection and pedestal taking triggers, can also be produced by the MCM and
transferred by the SCM to the RMMs.
The TFBs use trigger primitives based on coincidences of MPPCs, which could
indicate the interaction of cosmic rays, to create “cosmic” triggers which are then
combined with the other detectors through the cosmic trigger monitor, or CTM.
Front-end processing nodes (FPN) control and read out data from the TFBs.
Configuration and read out information comes from the read-out task (RXT), and
the data is converted into a readable format by the data processing task (DPT).
Finally, the MIDAS framework, developed by TRIUMF, provides an interface
between the electronics output and the users of the data. More details about the
P0D components, calibration and construction will be given in Chapter 2.
1.5 Super-Kamiokande
The detector at Kamioka has been running since 1996 [?]. It consists of large
cylindrical water Cherenkov detector with an inner (ID) and outer detector (OD)
region as shown in Fig. 1.22 [?]. Since the detector has been running for more
27
Figure 1.22: The far detector at Kamioka. It is a water Cherenkov detector located 1
km deep in Mount Ikenoyama [?].
than a decade, the behavior is very well understood. Calibration of the energy
and uncertainties in the modeling for the detector are both known to the percent
level [?].
In the ID, a 33.8 m diameter and 36.2 m tall cylinder, 11,129 inward-facing
50 cm diameter PMTs are used to detect Cherenkov radiation from interacting
particles. Cherenkov light from particles moving faster than the speed of light in
water form rings that are detected by the PMTs. Example interactions are shown
in Fig. 1.23.
The 2 m thick OD, which surrounds the ID, is used primarily to veto back-
ground events and contains 1885 20 cm PMT. A highly reflective material, Tyvek,
lines the OD and is used to reflect light off the surface of the OD walls and into
the PMT. For cosmic ray backgrounds the OD gives almost 100% rejection. Using
beam timing information, neutrino beam events that produce light in the OD can
be separated from other backgrounds.
Number of rings, ring thickness, energy deposited, timing, the OD veto and
other factors are used to discriminate between particle interaction types. For
28
Figure 1.23: An example of a muon (left) and an electron (right) interacting in the
Super-Kamiokande volume [?].
example, neutral pions are rejected by forcing two rings in the event, then using
information from the rings to calculate an invariant mass. Any event with an
invariant mass calculated to be above 105 MeV/c2 using this method is rejected
as shown in Fig. 1.24.
The T2K νµ → νe oscillation result has six cuts. It requires that the event be
fully contained in the inner detector fiducial volume, have only a single Cherenkov
ring, that the ring be electron-like, that no Michel electron be detected, that the
invariant mass previously described be below 105 MeV/c2 and that the total
energy deposited by less than 1250 GeV [?]. After all the cuts the current T2K
oscillation result finds a total of 11 events as shown in Fig. 1.25.
29
)2Invariant mass (MeV/c
0 100 200 300
Num
ber
of e
vent
s
0
5
10
RUN1-3 data)POT2010×(3.010
CCeνOsc. CCµν+µν CCeν+eν
NC=0.1)13θ22(MC w/ sin
Figure 1.24: In order to reject neutral pions two rings are forced and events with
invariant mass greater than 105 MeV/c2 are rejected
energy (MeV)νReconstructed
0 1000 2000 3000
Num
ber
of e
vent
s
0
2
4
6
RUN1-3 data)POT2010×(3.010
CCeνOsc. CCµν+µν CCeν+eν
NC=0.1)13θ22(MC w/ sin
Figure 1.25: Remaining events after all cuts have been applied in the most recent
T2K oscillation analysis.
30
2 P0D Construction and
Calibration
In the previous chapter, the components of the P0D detector were described. Each
of these components needs to be well understood to measure particle interactions.
The efficiency and time delay of photon transmission in the fiber and scintillator,
the measurement of these photons by MPPCs, and then the chain of electronics
that finally lead to a signal were all calibrated before the P0D was constructed.
Measurements were made of the dimensions and composition of all components
to increase the accuracy of the simulation.
In late 2009 the P0D was installed in the ND280 detector basket. After its
final construction, monitoring devices are used to keep constant updates on the
status of these processes and to monitor the water target, which has the highest
likelihood to change its composition over time. Finally online monitoring is used
to detect problems during data taking runs and to prevent damage to the detector.
2.1 The P0D Modules: P0Dule Calibration
Calibration of the P0Dules predominately involves calibration of the scintillator
and the fiber. For fiber, factors contributing to uncertainty are the time-walk,
photon loss, efficiency of photon detection by wavelength, mirror reflectivity and
31
Figure 2.1: Fiber time walk correction for the P0D and ECal
attenuation. The scintillator time-walk is negligible because of the short decay
times of the fluors, POPOP and PPO, that are responsible for light production in
the scintillator. However, other effects in scintillator are not negligible. Calibra-
tion for detection efficiency, and variation in light output with time is necessary [?].
In order to produce a “time stamp”, a measurement of the time the event
occurred, between 2.5-3.5 photoelectrons have to be detected by the TFB, where
a photoelectron is the energy needed to eject an electron by fluorescence in the
material. The fluors produce light through an exponential decay, and therefore
there is some time between the initial emission of light and detection by the
TFB. This “time walk” is probabilistic and depends on the decay constant, the
discriminator threshold and the total number of photoelectrons produced in the
interaction. Figure 2.1 shows the time walk correction for the P0D with corrections
for the effects of MPPC after pulsing.
Sometimes photons do not transmit fully through the fiber. Interactions with
particles in the fiber result in energy loss compared to the initial interaction.
Therefore the measured signal will be less than the actual amount of deposited
32
Figure 2.2: Left:Transmission loss in the Y-11 fiber versus wavelength according to
Kuraray for 3 lengths: 100, 150 and 200 mm
energy. This loss of photons is dependent on the length of the fiber, the wavelength
of the transmitted light and the radius of any bending in the fiber [?]. The
expected photon loss for the Y11 fiber is shown in Figure 2.2 and these values are
used to calibrate the measured photoelectrons.
MPPCs are very sensitive to the wavelength of the photons that are measured.
Some wavelengths are not detected as efficiently as others. Interactions in the
fiber alter the distribution of photon wavelengths. The spectrum of wavelengths
that are emitted by the Y11 is therefore dependent on its length, with longer
fibers having broader emission spectra with fewer photons at the most efficient
wavelength.
Therefore the total light measured at the MPPC in the preferred wavelength is
reduced as distance from the interaction increases, altering the nominal “photon
detection efficiency” or PDE of the MPPC. Patrick Masliah and Antonin Vacheret
measured the PDE for fibers of various lengths and the results are shown in Fig.
2.3. It can easily be seen that although there are differences in the PDE for various
lengths, the shift in wavelength is not large enough to make a very big difference.
Because the light is only read out on one end of the fiber, the other end is
polished and coated with sputtered aluminum to reflect light back to the sensor
33
Figure 2.3: Photon Detection Efficiency vs. Fiber Length measured by ND280 Cali-
bration Group and fitted with a line
end. The mirror is not perfectly reflective, and does absorb some of the light. In
order to determine what fraction of the light is reflected, a comparison was made
between the mirrored fibers and fibers that were cleaved at 45 degrees and painted
black [?]. The reflectivity was found to be 86% with a root mean square of 6%
measured over all fibers.
Attenuation is loss of light intensity over distance. The effect is modeled as
the sum of two exponentials as follows
a(x) = fe−x
Ll + (1− f)e−x
Ls , (2.1)
where x is the position of the signal, Ll is the long attenuation length, Ls is the
short attenuation length and f is the fraction of light in the long mode. The light
that is read out from the MPPC is then the sum of the attenuated signals from
the direct and reflected light
34
x(mm)
response
Figure 2.4: Examples of scanner response measurements as a function of scanner
position with pedestal subtracted. Every color represents a different scintillator bar.
r(x) = a(x) +Ra([l − x] + l), (2.2)
where R is the mirror reflectivity and l is the length of the bar. The parameters
Ls, Ll and f are not known so the attenuation in all bars has to be measured,
and then the truncated means are fitted to determine the unknowns.
One method that was used to determine the attenuation in each bar was
P0Dule scanning. A 60Co source was used to produce a signal at 11 different
positions in the X and Y directions [?]. A sample of the response of channels at
different positions is shown in Fig. 2.4. This method allowed all the planes to be
scanned for any problems before they were shipped to Japan, and gives a good
constraint on the attenuation.
Further attenuation data was taken in situ using “through going muons”.
Through going refers to the fact that they pass through the detector, but muons
for this study are produced specifically from atmospheric neutrino interactions
35
Distance to Sensor (mm) Distance to Sensor (mm)
Nom
alized R
esponse
Nom
alized R
esponse
Figure 2.5: Truncated mean responses (solid lines) for x and y bars, along with fits
(dashed lines). The left-hand plot compares the fits to the scanner data with the scanner
data itself, while the right-hand plot compares the scanner data with the fit function
from the through going muon data, with good agreement.
and interact in the detector uniformly with well understood properties. Measur-
ing the response of the muons in different parts of the detector, it is possible to
get information about the response at many more than 11 positions in each ori-
entation. It is also possible to take data from these muons at any time, and the
data that was used to calculate the attenuation parameters was taken after the
detector was installed. A comparison between the scanner data and the muon
data is shown in Fig 2.5, and the fits using the parameters calculated from the
muon data fit both results well.
After the detector had been installed for slightly over a year, it was discovered
that a variation with position still existed in the data. This difference is shown in
Fig. 2.6. It was determined that the difference was independent of the fit parame-
ters. In order to correct the difference, the detector was divided into sections and
each section was calibrated separately. The final result gives a consistent response
in all areas of the detector.
Because of the triangular shape of the scintillator bars, it is most likely that
36
Mea
n N
od
e E
nerg
y (
ph
oto
ele
ctr
on
s)
Figure 2.6: After attenuation correction, some variation is still seen in the response
as a function of distance from the MPPC. The detector is then split into sections, and
each section is calibrated seperately to a normalized response.
a through going muon will go through two bars in a layer as shown on the right
side of Fig. 2.7. In this case, a particle with very low energy may have its energy
deposit split between the two bars such that neither bar will pass minimum energy
cuts. It is also possible that it can go through the point, and therefore only travel
in one bar, as shown on the left side of Fig. 2.7.
To insure that the calibration was done properly it is important to understand
the effect of these different types of interactions on the muon detection effeciency.
This efficiency was studied, and the result is in Fig. 2.8. It was found that except
for some events that originated in the front of the P0D, all of the tracks were
detected in either one or two of the bars at every position in the detector.
The possibility that the transmission of light in the scintillator might become
less efficient over time was taken into account. About twenty different samples
were taken at equally spaced times using a radioactive source counting setup to
measure the light output with a reference piece of scintillator from the MINERvA
37
Figure 2.7: A cartoon of a minimum ionizing particle(MIP) interacting in the trian-
gular scintillator bars. The distribution of energy will be different depending on if the
particle interacts in one or two bars at a time.
production. No evidence was observed for variation in light output beyond the
uncertainties in the monitoring measurement, roughly 5%.
2.2 Construction
The scintillator for the P0D was the first component to be constructed. A fac-
tory at Fermilab was used, which was also used to produce scintillator for the
MINERvA experiment [?]. The fiber was delivered to FNAL in precut “canes”,
unspooled fiber 67 mm longer than the bar length, to avoid “memory effects”,
bending in the fiber caused by being wound on a spool [?].
The exterior coat of TiO2 is extruded simultaneously with the polystyrene
38
Figure 2.8: Detection efficiency in the P0D. The plots show the proportion of tracks
with 0 (green circles), 1 (blue triangles) and 2 or more (red inverted triangles) hits in a
layer, for both x and y.
scintillator bars, and then the end of the bars are painted with white paint. The
fiber readout ends were coated in epoxy, polished by a diamond and then the
mirror ends were coated in the Thin Films laboratory at FNAL. Each readout
end was then fitted with a ferrule sitting within 1 mm from the fiber end. These
ferrules were specifically designed for mounting into the MPPC housing.
Construction of the P0D began in the Fall of 2008 with P0Dule assembly.
After scintillator bars were glued together to form layers, these and the active
material layers were glued inside of a large 1.8 x 2.1 m P0Dule box made of
black polystyrene. The finalized P0Dule has a rigidity similar to a solid mass of
polystyrene with the same thickness. A PVC frame covers the edge of each P0Dule
and provides a structure to hold MPPCs that have been attached to ferrule fitted
fibers. These covers also provide support for the P0Dule weight when it is installed
in the detector basket.
The MPPCs were subjected to quality control tests at various values of over-
voltage. Those that passed inspection were tagged electronically and then inserted
39
into each scintillator bar by hand. Once all the fibers were installed, each P0Dule
was moved by crane into position in a super-P0Dule, which is the colloquial term
for the grouping of P0Dules into ECAL and target segments.
The water targets are constructed of HPDE bladders, based on designs used
in the Pierre Auger Observatory Cherenkov detector [?]. These bladders are
sandwiched between P0Dules which provide support in the water target super-
P0Dule. Two bladders are used in each water layer and separated by an HPDE
strut, in order to prevent deflections due to unbalanced pressure.
P0D installation began in September of 2009, and involved several steps. The
first step was lowering the super-P0Dules into the pit. Each super-P0Dule was
prefabricated with holes to allow installation with a very low tolerance of .5mm
on internal dimensions and 1mm on the thickness, and have dimensions of 2212
mm (x) by 2348 mm (y) 102 by 38.75 mm (z). Precise measurement of P0Dules
and super-P0Dules was required to fit the final P0D into its allocated space in the
basket. Small errors in positioning of the installed super-P0Dules were fixed by
sanding the aluminum support so that screws could be inserted.
The second step was installing the water cable used in cooling the P0D and
filling the water targets. The cables have to be run from the main water reservoir
at the south end of the pit to the bottom of the detector, where they are connected
to water pipes entering the detector from the bottom. The water is filtered tap
water and is mixed with a 0.25% solution of chlorine bleach to prevent contamina-
tion with living organisms. A pump rack was installed outside the magnet, which
contains 50 self-contained pumps, one to fill each of the water bags.
The third step was installing the power for the P0D. A cable tray had to be
build to hold the massive copper cables which are used to power the P0D. Then
these cables are connected at the P0D to four power boxes, which then connect
directly to the TFBs, and two floors below the location of the P0D are connected
to the central power station for the detector. The TFBs are thermally mounted
40
on plates attached to six rails on the sides of the detector.
Finally the RMMs (Readout Merger Modules) and all of the data taking cables
had to be installed. Each RMM has to be placed in the center of the P0D at the
top. Then, the TFBs are connected individually to each RMM. Finally, cables
are run from the central data processing unit two floors below the P0D up to the
top of the P0D using orange tubing as a protective covering.
2.3 MPPC and Electronics Calibration
Once the detector was built, the electronics were calibrated. The MPPC, as
described in Chapter 1, have 1.3 mm by 1.3 mm grids made up of photon counting
pixels. Each pixel is always in either an on or off state.
Therefore, when multiple photons activate the same pixel, only a single photon
is measured. This leads to a lower number of measured compared to actual pho-
toelectrons. This effect as well as noise caused by crosstalk (called avalanching)
create a need for studying and monitoring the output of the MPPCs. Calibration
efforts included measuring the pedestal, calibrating the gain using bias voltage,
monitoring pedestal and gain changes over time and corrections for saturation of
the MPPC pixels and the electronics.
The first calibration which was made was in measuring pedestal data. The
“pedestal” is a noise peak which exists even in the case that there are no excited
electrons in the photodiode. Therefore the location of the “pedestal peak” de-
termines the value of “zero” for each capacitor. A simple diagram of the MPPC
output, including the “pedestal” peak, is shown in Fig. 2.9.
Pedestal graphs are made up of the measurements taken with the electronics
activated, but without any cosmic rays or beam. In the analysis, some cosmic rays
do appear, but the pedestal events far outweigh the actual events. The MPPCs
have both high and low gain pedestals and each TFB has 64 MPPC channels. Each
41
Figure 2.9: A simple cartoon that shows the “pedestal”, one photoelectron and two
photoelectron peaks. The distance between the pedestal and the one photoelectron
peak gives the ADC count to one photoelectron conversion. The distance between the
one and two photoelectron peaks should be the same, and provides a cross check on the
ratio.
Integrated Charge (ADC Counts)
Res
po
nse
Integrated Charge (ADC Counts)
Res
po
nse
Figure 2.10: Left: One high gain channel of a TFB. Right: Low gain channel analyzed
in the same fashion as high gain, but only the pedestal is visible
channel has 23 capacitors, and each of these capacitors is calibrated separately.
The pedestal graph for high gain is typically a large zero peak, followed by a
smaller one photoelectron peak, and a much smaller two photoelectron peak. For
the low gain pedestal, usually only the pedestal peak is visible as shown in Fig.
2.10.
42
Integrated Charge (ADC Counts)
Res
pon
se
Figure 2.11: All the pedestal peak locations for all twenty nine TFBs
Figure 2.12: After subtracting the pedestal, the MPPC signal is a gaussian distributed
around a central value of zero, with higher order peaks from noise.
Once pedestal graphs have been made for all channels and all events, the data
can be analyzed. First, a graph is made of all of the pedestal peak locations for
all of the channels. This gives a good idea of how much the pedestal differs from
channel to channel as shown in Fig. 2.11.
In order to get the ADC count of a signal, it is necessary to subtract the
pedestal value from each capacitor. When the pedestal is subtracted from runs
43
Integrated Charge (ADC Counts)
Res
po
nse
Integrated Charge (ADC Counts)
Res
po
nse
Figure 2.13: Left: Distance from pedestal peak to one photoelectron peak. Right:
Distance from pedestal peak to two photoelectron peak
with no data being taken, the result is a gaussian distributed around a central
value of zero as shown in Fig. 2.12. Secondary peaks in the tail of the distribution
are caused by the higher order noise peaks.
The distance from the pedestal peak to the one photoelectron peak gives the
photoelectron to ADC (analog to digital conversion) gain. By the same logic, the
distance from the pedestal to the two photoelectron peak is the two photoelectron
to ADC gain. If the measurement is good, the two photoelectron gain should be
about twice as large as the one photoelectron gain. In general, this was seen to be
true. Most of the measured one photoelectron to ADC gains were between 12-15
ADC counts/photoelectrons as shown in Fig. 2.13.
The data can be used to give an estimate of the crosstalk. Assuming the
photoelectron peaks obey a poisson distribution, the number of events in the
pedestal as a fraction of the total number of events will give you an estimate of
the number of one and two photoelectron events that are expected. The poisson
distribution gives
f =µke−µ
k!, (2.3)
44
Res
po
nse
Expectation Value (photoelectrons/response)
Res
po
nse
Expectation Value (photoelectrons/response)
Figure 2.14: Left: Value of one photoelectron events over pedestal events gives ex-
pected photoelectrons per response. Right: Value of two photoelectron events over
pedestal events gives half square of expected photoelectron per response.
where µ is the photoelectron expectation value, k is the number of photoelectrons
seen, and f is the probability, or the fraction of events with that number of photo-
electrons in the total sample. It can be shown that the photoelectron expectation
value is equal to the natural log of the ratio of pedestal events to total events, and
also equal to the number of one photoelectron events divided by the number of
pedestal events. This value is around 0.15-0.18 when operating in pedestal mode.
The value that is measured is higher than the expectation value predicted by
the Poisson distribution. This is because of crosstalk. Crosstalk in the MPPCs is
caused by excess thermal electrons, excited either in adjacent pixels or along the
path of the thermal electron in the initial pixel, by the passage of thermal electrons
that were excited by the initial photon interaction. Some thermal electrons are
also produced continuously by energy from the environment, even when no photon
interaction occurs.
The integrated charge from these additional electrons then produces excesses
in the higher order peaks. Since the predicted amount represents what you expect
to see with no crosstalk, measuring the excess photoelectrons will give you a good
approximation of the crosstalk noise contribution. Knowing this value makes it
45
Run Number
Inte
gra
ted
Ch
arg
e (A
DC
Co
un
ts)
Run Number
Inte
gra
ted
Ch
arg
e (A
DC
Co
un
ts)
Figure 2.15: Left: Change in pedestal over time. X axis is run number, where runs
are spaced an hour apart. Y axis is change in pedestal from run 122 in ADC counts.
Right: Change in photoelectron/ADC gain over time.
possible to correct all measured photoelectron values by subtracting the measured
crosstalk. (See Fig. 2.14)
Another useful aspect of pedestal to study is how it changes over time. Factors
like temperature and overvoltage lead to fluctuation over the course of as little
as a few hours. Several data runs were taken to try and see what if any changes
occur in the pedestal data. These studies showed RMS of about 0.3 ADC over
the course of several days, or about 2 percent of the one photoelectron to ADC
gain.
Since the pedestals vary as previously described, corrections are made over
time to account for the shifts. When beam data is taken, pedestals are also
taken. Individual capacitor gains cannot be calculated during beam running as
only averages over all the channels are taken to reduce the necessary computer
memory. In order to calibrate each capacitor individually, data from previous
measurements of the channel and capacitor pedestals are used as follows
pcapt = pcapt0 + (pchant − pchant0), (2.4)
where pt is the new pedestal measurement and pt0 is the previous pedestal mea-
46
Comparator Level (DAC) Comparator Level (DAC)
Inte
gra
ted
Ch
arg
e (A
DC
Cou
nts
)
Inte
gra
ted
Ch
arg
e (A
DC
Cou
nts
)
Figure 2.16: Change in gain for high gain channels (left) and low gain channels (right)
between the January and November runs of 2010
surement, and the subscripts cap and chan stand for the capacitor and the channel
respectively.
Change in one photoelectron gain over time was also monitored. The one
photoelectron gain seems to change related to the change in the pedestal. Change
in two photoelectron gain is also monitored, but is about the same as change in
one photoelectron gain. (See Fig. 2.15) The variation in the one photoelectron
gain was also compared between the two data taking runs in 2010. The result of
this comparison is in Fig. 2.16.
Both the low and the high gain channels on the TFBs will saturate if enough
charge is injected. In Fig. 2.17 it shows the ADC count measurement compared
to the injected charge for both high and low gain. The high gain saturates at
much lower values of injected charge, so the low gain channel is only used above
that threshold, since the high gain channel is otherwise more accurate.
Since the response is non-linear, it is fitted with two cubic polynomials su-
perimposed using error functions. We consider the channels to be adequately
calibrated when the channel response is modeled with less than a 5% residual.
Currently P0D calibration has gotten a fit that has around a 2% residual.
47
Figure 2.17: A comparison of the ADC signal to the injected charge for the high and
low gain channels. At high levels of injected charge, the high gain channels saturate
and the low gain channel value is used.
A basic model of light entering the MPPC assumes that the light is all absorbed
at the same time, that the beam is larger than the MPPC, that the light is evenly
distributed on the surface of the MPPC and that the MPPC response is Poisson
distributed. A more realistic model needs to correct for the fact that none of these
assumptions are strictly true. The reponse and saturation of the MPPC can be
described as a function of the photon detection efficiency, PDE, and the number
of effective pixels with parameters p0 and p1 as follows
〈Npe〉 = p1 ×
1− e
−〈Nγ〉 × p0
p1
, (2.5)
48
Figure 2.18: Saturation curves for over-voltages of 0.95 V, 1.14 V, 1.34 V, 1.53 V,
1.73 V and 1.92 V fitted with a 2 parameter fit
where 〈Npe〉 is the number of photoelectrons read out by the detector and 〈Nγ〉is the number of incident photons. The result of fitting the data at various over-
voltages is shown in Fig. 2.18. The residual of these fits is mostly under 5% of
the fit value.
The response in photoelectrons in each bar varies due to individual properties
of the fiber, scintillator and MPPC readout. To account for this, the response
in each bar from through going muons was measured. The most probable value,
MPV, of 38.15 photoelectrons was used to normalize the response of the bars to
give a consistent response to particle interactions.
Through going muons were also used to calibrate the timing across TFBs on
and RMM and RMMs over the full P0D. Timing responses for the first and last
TFB are shown in Fig. 2.19. These gaussian distributions are fitted for all the
TFBs in an RMM, and the timing is normalized across the RMM by subtracting
the average value from the value from the individual TFB. Similarly, the RMMs
are offset from their measured value by the average of the TFB averages from
49
Time(ns) Time(ns)
Figure 2.19: Examples from TFB 0 and TFB 268 of timing distributions relative to
the trigger.
Figure 2.20: Timing distribution relative to trigger for all TFBs before and after
calibration.
all the other RMMs. TFB average times are stable to within 2.5 ns, but RMM
averages can vary by as much as ±10 ns, so the average of multiple measurements
is used to determine the final calibration constants. The result of this calibration
is shown in Fig. 2.20.
Finally, like the fiber, the electronics also have a time walk. The electronics
time walk is caused by the fact that the capacitor will charge more quickly and
hit the threshold for creating a time stamp if the signal is larger. The time it
takes for the threshold to be reached is dependent upon the decay constant of the
circuit τ , the threshold charge Qth and the signal charge Q0 as follows
50
tth = −τ × ln(1− Qth
Q0), (2.6)
Data taken by Imperial college [?] shows that this equation accurately models
the time walk behavior with residuals compared to the fit of less than 10% (apart
from a single outlier). In the P0D comparison between multiple measurements
has shown good agreement with the fit with variations on the order of 15%.
2.4 Light Injection
To monitor the P0Dule response over time, it is useful to have a system that can
be used to produce a well understood response. In the P0D light injection is used
for this purpose. The light injection system consists of 80 pairs of 400 nm LEDs
installed in 5 mm cavities in the P0Dule support [?].
The LEDs have the ability to expose the MPPCs to light intensities with
multiple orders of magnitude. This is necessary to simulate the expected range
of scintillator responses. These intensities are achieved by varying the height and
the width of the LED pulse, which in turn is caused by varying the current pulse
applied to the LED.
The electronics that are used to control the light injection were originally
created for the MINOS experiment. They consist of four pulser boxes, a control
box, a distribution box and a power supply. Special mounting was created to
affix the pulser boxes and the control box to the P0D itself, while the distribution
box and the power supply are on a crate outside the basket. Each LED is then
connected to the pulser boxes by a 60 cm long cable. The signal is transmitted
from the control box to the pulser box via ethernet link.
When beam data is not being taken, light injection runs can be taken. Pro-
grams with specific pulse height and width patterns are used to make specific
51
Figure 2.21: Light injection can be used to monitor photosensor variation. Here the
average signal in all photosensors from light injection measurements is shown over a
period of several weeks.
calibrations. These calibrations can be used to check the status of the electronics,
such as stability of the photosensors as in Fig. 2.21 and also to measure important
properties like timing variation between different bars.
2.5 Water Target Calibration
The water target needs to be continuously monitored because changes in the vol-
ume of the water or the configuration of the water bags can change the probability
of a neutrino interaction. Any time the water targets are filled or drained the
amount of water changes. Also, besides scheduled fills and emptying, sometimes
water is added to assure that the fiducial volume was completely filled, and also
water can leaks out of the detector for whatever reason. Because the detector is
not perfectly rigid, the bags can also settle over time, creating uncertainty in the
thickness[?].
52
Figure 2.22: Variation over the course of 40 minutes in pressure sensor readings on
full water targets.
Pressure sensors and wet-dry level sensors were used in both of the data taking
runs. In Run 1 a dipstick was used to check the levels of several of the bags, and the
results did not agree with the pressure sensors. Cross checks were also done with a
flow meter and by draining a few individual bags and measuring the final volume,
and in all cases the pressure sensor readings were different. For these reasons
pressure sensor data from Run 1 is considered unreliable, and the pressure sensors
were completely replaced for Run 2.
The water was completely drained and the volume measured after Run 1.
Even though the pressure sensors were faulty, the wet-dry level sensors continued
to be accurate within 50 mm for the entire run, and therefore it is known that the
fiducial region was always completely filled with water. To calculate the fiducial
volume the drained volume is used with the average of the levels measured by
the working pressure sensors just before the drain [?]. The drained volume was
found from the measurement of the depth sensor in the main water tank, which
53
measured a 2293±10 mm depth increase after the P0D was drained. For the 37
functioning pressure sensors the average depth for a water bag was 1880 ± 45 mm
compared to the total P0D height of 2239 mm.
In Run 2 more reliable pressure sensors were used. The variation of these
sensors over time is shown in Fig. 2.22. The residual is taken as ±1 mm with
a ±15 mm calibration offset which will be reduced by wet-dry sensor calibration
references in situ [?].
Since the water bags are made of thin plastic, it is possible for the water
to bulge. This bulging can lead to variations in the water levels as the bags
settle into their new bulging configuration, and therefore creates uncertainty in
measurements of the fiducial mass. In order to understand the magnitude of the
change in the levels, the water bags were monitored for one week after they were
filled. The result is shown in Fig. 2.23. This shows a variation of 2 cm for the
upstream bag and about 30 mm for the central water bag.
The average variation for all of the water bags over the course of a week
was 50 mm. In Fig. 2.24 the variation in the levels over the course of 19
days is shown. Bags on the upstream and the downstream end have the most
variation. Downstream water bags are affected especially badly because they are
not supported by the basket. More on the effects of the water level uncertainty
will be discussed in Chapter 6.
2.6 Online Monitoring
An important tool for P0D calibration as changes occur over time is online mon-
itoring. Some calibrations can only be taken when beam is down, like light in-
jection. Other things like pedestal data, electronics response and water target
measurements can be taken continuously.
54
Figure 2.23: Variation in water levels over 1 week due to settling for a)Upstream water
bag and b)Central water bag. Upstream bags are not supported as much by surrounding
P0Dules and therefore experience more settling.
The majority of the online monitoring that is done for the TriP-T detectors is
of the activity recorded from the beam events. Total response over the course of
the run from each of the capacitors is monitored globally. Also, a 3D display of
each event allows continuous visual checks. For example, if a MIP travels through
the detector but does not create activity in a P0D layer or a TPC chamber this
is cause for immediate concern[?].
Alarms are also set in case of major failures. The water targets will set off an
alarm if the water levels drop too low. Low water levels could mean that a major
leak has occurred, which has the potential to do severe damage to the detector
55
Figure 2.24: Total variation in water levels for all bags over the course of 19 day
observation period. Downstream bags show the largest variation because they are not
supported by the basket.
Time (ns) Time (ns)
Figure 2.25: Left: Hit Time with relation to Trigger for Pedestal Right: Hit time
relative to Trigger for Event data
electronics.
Before the current online monitoring for the P0D was designed, simple online
monitoring was used for testing the detector functioning. This was before the full
detector was installed, and therefore only about half the channels are included on
the plots. This simple online monitoring for the P0D originally consisted of an
56
126*(P0dule #) + Bar # 134*(P0dule #) + Bar #
23*(Chan #) + Cap #1920*(RMM#) + 64*(TFB #) + Chan#
Figure 2.26: Simple online monitoring tools used before the full detector was installed.
Top: A hit map by P0Dule and bar number in X and Y. Bottom: A hit map by RMM
number, TFB number and channel
electronics hit map, a detector hit map, and hit time for each event.
The electronics hit map was a map of the hits in each channel in a TFB in
order to monitor for the possible existence of empty channels. An empty or low
channel would mean that the electronics were never reading out, which indicates
a malfunction. For the detector hit map, each channel was converted to a location
in the P0D in x and y, so that it was possible to see exactly where each event was
happening in the detector (See Fig. 2.26). The hit time monitored the timing
of the event, to see if it is assigning time correctly, and to look at the timing
distribution (See Fig. 2.25).
Through going muon data from the beam is also taken. As described in pre-
vious sections, these interactions can be used to calibrate and recalibrate many
57
different components of the detector. In the next chapter, the implementation of
these calibrations in the software is described.
58
3 Software
In order to understand the data that is collected by the detector and compare it
with generated Monte Carlo, multiple steps have to be taken. First the raw data
or Monte Carlo is converted into “hits”, which have been corrected for electronics
effects. Once it is converted, methods are used to extract various signals. Once
“tracks” and large energy clusters have been differentiated from the hits, different
properties of each event can be investigated. This section includes only those parts
of the electronics simulation and reconstruction work that affect the analysis that
is detailed in the final section. The software chain for data and Monte Carlo is
illustrated in Fig. 3.1.
3.1 Electronics Simulation
To create simulated events first GEANT4 [?] objects are created from Monte
Carlo generated by NEUT, which is described in detail in in Chapter 4. These
generated events are then interpreted using the “nd280mc” package. The purpose
of the “nd280mc” package is to create a data structure, called a “digit”, with
the necessary information to perform calibration, such as information about the
charge deposited by the interacting particle.
59
Figure 3.1: Two different paths in software to process data and Monte Carlo for
reconstruction. Monte Carlo is generated by NEUT, becomes GEANT hits, is processed
by ND280MC, and gets effects from the electonics added by ElecSim. Data only needs
to be calibrated using OACalib before it enters reconstruction.
The ElecSim package then uses information from these “digits” to simulate
the effects of the scintillator, MPPC and electronics read-out [?]. Some effects,
like noise, are not simulated by the NEUT even generator and are therefore added
by the ElecSim package. After the “digit” is calibrated, the new structure that is
created is called a “hit”, and can be used for particle reconstruction,.
Effects in the MPPC described in Chapter 2, such as crosstalk, avalanching
and noise, are not simulated by the generator. In order to accurately compare
data and Monte Carlo, these noise events must be simulated by ElecSim. Noise
in the TriP-T electronics is similarly simulated by the ElecSim package.
MPPC noise is simulated uniformly in position and time. Light from the
generated event is either assumed to distribute uniformly across the face of the
MPPC or to have a Gaussian distribution as shown in Fig. 3.2. Information from
the noise hits are added to the generated event hits and the combination is sorted
in order of time.
60
Figure 3.2: Two different Gaussian light distribution models. The value of sigma in
the distribution dramatically affects the distribution of light.
For the purpose of the simulation, noise hits always cause afterpulsing, but
generated events will only create a response a certain fraction of the time, de-
pending on the photon detection efficiency. Pixel firing in the MPPC is simulated
in a binary fashion, with 1 representing a photon interaction or “on” state, and
0 representing the “off” state. In the case of signal photon interaction, the pixel
charge is given as 1 at the full voltage, smeared by the photoelectron resolution.
If the voltage is less than full, the response is reduced according to the measured
calibration as given in the Chapter 2. When no signal is provided, the pixels
default to their “off” setting of 0.
The probability of crosstalk and afterpulsing are determined by the instanta-
neous charge of the pixel. For both noise and generated events avalanching effects
are modeled probabilistically with a double exponential decay. Crosstalk is simu-
lated to occur at the measured rate. The position in the MPPC of the crosstalk
response is calculated as a function of the original pixel position and the time of
the event is set immediately after the initial pixel response.
For the TriP-T electronics noise is generated using a timestamp probability
method. Noise is generated only above the TriP-T TDC photoelectron threshold
value. Simulating noise below this threshold is unnecessary because noise de-
creases linearly as a function of photoelectrons on a log scale. The computational
61
Figure 3.3: Monte Carlo is calibrated to the data in ElecSim using data from through
going muons. The muon data is fitted with a Landau distribution convolved with a
Gaussian and corrected for photon detection efficiency based on overvoltage.
power required to simulate this noise is much larger than the benefit gained, so it
is ignored.
The response of the data and the Monte Carlo is normalized using information
from through going muon data and simulated muon events. A Landau distribution
convoluted with a Gaussian is fitted to the responses, and the most probable
value is extracted as in Fig. 3.3. The measured ratio of Monte Carlo to data was
1.34, and this parameter was used to calibrate the number of photons per MeV.
Unfortunately x and y bar difference caused by photon detection efficiency effects
were not taken into account, so the section by section calibration mentioned in
Chapter 2 was necessary to normalize the response of the two views.
Monte Carlo hits produced by ElecSim can be used for reconstruction with
after linearity corrections are applied. Raw data needs to be calibrated before it
can be used. In Chapter 2 the details of these calibrations were outlined. The
package created for this purpose is OACalib. After OACalib is run on raw data,
62
it can also be used in reconstruction.
3.2 P0D Reconstruction
Reconstruction takes the calibrated hits from the previously described stage and
uses pattern recognition algorithms to create basic structures for analyses. The
final products of this reconstruction chain are “track”, “vertex” and “node” ob-
jects that are used in particle identification. Unlike previous stages, it is important
to consistently treat data and Monte Carlo information equally in this stage to
prevent bias. A brief description of the methods is given, followed by a detailed
description broken into sections.
The first step in reconstruction is to separate data by time stamp into 23
cycles. These 23 cycles correspond to the 23 capacitors in each channel, and are
a result of the method of charge integration that is used. Each charge cycle is
about 480 ns and cycles are separated by 100ns.
Between six and and eight of the cycles will contain hits in data, but usually
only one cycle contains the interaction in Monte Carlo. In data the period where
interactions take place is dependent on the bunch structure of the beam as detailed
in Chapter 1. So, though it is simple to specify the cycle containing the interaction
in Monte Carlo, it is necessary to try to reconstruct events in every cycle in order
to treat Monte Carlo and data equally.
Then noise from the electronics and MPPCs, which make up the majority of
hits, must be removed as best as possible. Reconstructing the correct total energy
of a particle, or having 100% “energy completeness”, is the ideal goal. However,
including too many noise events will reduce the efficiency of the reconstruction. In
order to maximize the energy completeness of the event it is important to balance
noise reduction with keeping as many true event hits as possible.
63
After that, a “road following” algorithm is used to create “track” objects.
“Road following” describes the basic method of using a line segment found from
a simple algorithm to add other hits that are likely to belong to the same object.
Thetrack objects created by “road following” may or may not be minimum ionizing
particles (MIPs). To increase the chance of discriminating between MIP and
shower events, hits outside the identified track are added. Also, showers can
produce high energy hits separated by enough distance to be identified as separate
tracks, so track objects are added to other tracks based on proximity to produce
a final “track”.
Because of the geometry of the P0D, these final “tracks” are created sepa-
rately in the X and Y views and then combined. The final 3D object is fitted
to determine a likely “vertex” where the event originated, under the assumption
that the particle traveled from upstream to downstream. Multiple vertices can be
identified in each event.
These vertices can then be looped over in an analysis. Properties of the vertex
or track can be used to identify the event as a shower or a MIP, and to reduce
background processes in an analysis. The cuts that were used in the νe analysis
work will be described in a later chapter.
3.2.1 Noise Cleaning
The first step after hits are separated into cycles is “cleaning” the hits. Cleaning is
the process of trying to eliminate noise hits using charge, timing and proximity to
other hits. Any hit with a charge greater than 15 photoelectrons is automatically
a “clean” hit.
In order to determine whether a hit with a charge less than 15 photoelectrons
is “clean”, the hits are sorted using Delaunay triangulation [?]. Delaunay triangu-
lation is an algorithm which creates triangles from the position of all of the hits.
64
Figure 3.4: Delaunay triangulation groups hits into triangles. Triangles are then
surrounded by circles, and if more than three points exist in the circle, new triangles
are formed until all points are formed into circles with no more than three points.
In each iteration of the algorithm, circles are drawn around the current triangles.
If any of these circles contains more than three points, new smaller triangles are
formed and the process is repeated until all the hits are contained in circles with
no more than three points as in Fig. 3.4. Any pair of points in the triangle is
called an “edge”.
The difference in position and time between the two points in every edge is
determined. If the charge of the hit is greater than 7 photoelectrons and its edge
partner is within 30 ns in time and 10cm, about three podules or bars, in space
then it is considered a likely particle interaction product. Any particle with less
than 7 photoelectrons must have an edge partner within 30 ns in time and 3.5
mm in space to be considered. All other hits are eliminated as noise. The hit
distribution before and after the cleaning process is shown in Fig. 3.5. If a cycle
does not contain at least five hits after cleaning, no further reconstruction is done.
65
Figure 3.5: Distribution of hits before (blue) and after (red) cleaning. All hits with
charge greater than 15 photoelectrons are kept, and hits below that charge are kept
only if they meet minimum criteria.
3.2.2 Creating Track Seeds
To differentiate particle interactions the first step is to look for “tracks”, which
appear as energy deposited following a line. Fig. 3.6 shows the basic “track”
construction steps. In the P0D this is done separately for the X and Y views.
These “track” candidates are expanded in the next step to allow for showering
events which do not deposit energy following a line.
A linear pattern can be detected using an algorithm called a Hough trans-
form [?]. This algorithm assumes a line with an equation ρ = xcosθ + ysinθ with
ρ and θ as shown in Fig. 3.7. Every “clean” hit has an X or Y and Z position. In
the ρθ space these points form curves. Points in ρθ space where the most curves
intersect describe likely lines in the XZ or Y Z coordinate space.
The curves may not precisely intersect, so it is important to define the size
of the ρ and θ bins. For the P0D the angular bins are 1.8 and the radial bins
are 25mm. Also at least four curves must intersect in order to be considered a
66
Figure 3.6: Steps in creating a track seed. Energy deposits in the detector (a) are
fitted with a simple line using a Hough Transform(b). Hits included in the Hough line
are then refitted using PCA to determine the track direction and the endpoints and hits
near the end of the newly fitted line are added(c).
Figure 3.7: Points in XY space (left) are transformed into curves in ρθ space (right).
When curves on the right intersect, the ρθ coordinate of the intersection describes a line
in the XY space.
candidate for a line. Linear “tracks” found using this method are called “seeds”.
After the seeds are created, a “road following” algorithm is used to add any
remaining hits. In this particular version, hits are added to nearby seeds starting
with the seed that has the largest number of hits and then cycling over all other
seeds in order of size. This is based on the hypothesis that the largest seed is
more likely to be the origin of nearby hits than other nearby smaller seeds.
67
Figure 3.8: Two basis vectors with the greatest variance for the given distribution of
points. Principal component analysis transforms hits from the original basis into these
new bases for any data set.
Before adding hits, information about the track seed is found using principal
component analysis, or PCA. PCA is a mathematical method of transforming
hits from the original basis into a new basis that is a linear combination of the
original basis vectors. Minimizing redundancy and maximizing the variance in
the new basis produces principal components, or vectors aligned with the natural
coordinate system of the distribution, as shown in Fig. 3.8. To determine the
basis with the maximal variance the matrix of eigenvectors, P , of the covariance
matrix, C = XXT , is used, where X is the vector of the hit positions with the
mean hit position subtracted [?]. The positions Y in the new basis, are calculated
as Y = PX.
The information from PCA that is used in the road following are the end point
positions and the direction of the seed. Hits are first added to the upstream end
of the seed and then to the downstream end. Only one hit is added at a time,
with the added hit being the hit that is determined to be closest to the end point
and still passes the necessary criteria.
68
Figure 3.9: A graphic of the Kalman filter process.
To focus on adding hits in the forward going direction, the dot product of the
distance between the end position and the candidate hit with the PCA direction
is required to be less than 30 mm. Also, hits further than 8cm in the Z or 60mm
in the X or Y direction if the hit is in the same layer as the end point, are not
added. Finally, any hit within the road width but outside an opening angle of
0.15 rad is not considered.
Hits are added to the track until no hits are found that meet the necessary
criteria. After hits are added, any nearby hits in the same layer are also added,
up to a maximum of three additional hits.
Seeds are not allowed to share any hits added in the previous steps. However,
in the next step hits near the vertex are allowed to be shared between multiple
seeds. At the vertex it is very difficult to determine what hits belong with what
seed, so we allow them to be in any nearby seed. Hits added in this way must
we within 40 mm in the X or Y directions, and a maximum of four hits can be
added to a seed with this method.
The final “track” candidates are then fitted using a Kalman filter. Kalman
69
filters can provide fits even for distributions that change stochastically along a
path. Predictions from Bayesian probability of subsequent points on the path
from previous predictions compared with actual measurements produce precise
fits [?]. It is able to do this by using information about the error covariance P ,
and assuming the process, for some value of A and B, approximately obeys
xk = Axk−1 +Buk−1, (3.1)
where x is an estimated position and u is an optional control input. If this can be
assumed then measurements, denoted as z, can be compared at each point using
the Kalman gain K and the projection matrix H , which must be determined
independently for each data set, as shown in Fig. 3.9. Updated information for
P and K from the comparison with measurements are then used to continue the
process of prediction until all points are fitted.
For many systems the errors can be taken initially to be normally distributed,
with process error covariance Q and measurement error covariance R. In the
case of MIP tracks process noise is determined from multiple scattering. Multiple
scattering can be calculated from the momentum of the particle p, radiation length
l, and the width of the region w as
Cms =
(
13.6
p
)2
× w
l(1 + 0.038 ∗ lnw
l)2. (3.2)
The filter uses a fit to a line as the initial assumption. It is run first in
reverse, from downstream to upstream, and then forward. This creates a fit that
is accurate at both ends of the track candidate. A collection of hits in a P0Dule
layer fitted in this way is called a “node”.
70
Angle(rad) Angle(rad)
Figure 3.10: Angle of secondary tracks with respect to most upstream track for Left:
Electrons and Right: All other particles. Secondary tracks are found primarily within
0.2 rad in the electron case, but over a wider angle otherwise.
3.2.3 Seed Merging and Expansion
The previously described algorithm works well for tracks that are approximately
linear, such as MIPs. For showering tracks, like electrons or gammas, multiple
track seeds are often found in a single particle interaction. It is useful for the
purpose of particle identification to group as many associated seeds into a single
track object as possible, and also to gather nearby hits that would be noise in a
MIP event but could shower deposits.
It is important to only merge seeds that are likely to come from the same par-
ticle interaction. For example, the decay of a neutral pion produces two photons.
These two photons are most likely to interact in different areas of the detector,
and the products of their interaction will be separated from each other in both
distance and angle. In contrast, when an electron interacts, the products will all
be in the same direction as the initial interaction. A simple drawing of the two
cases is shown in Fig. 3.11.
The angle between the most upstream seed direction and the direction of
any downstream seeds, where direction is found from the Kalman fit of the node,
71
Figure 3.11: A cartoon showing the general energy deposit of a)electron and b)neutral
pion tracks. It is important for particle identification to avoid merging the two photon
tracks from neutral pion decay into a single track like the electron.
discriminates between seeds belonging to electron tracks, which were usually found
within 0.2 radians from the upstream track, and all other particle interactions,
which are more widely distributed, as shown in Fig. 3.10. Also it was found that
requiring that the merging candidate seed be within 100 mm in the XY direction
from the upstream seed minimized incorrect merging.
Track merging is done iteratively. Seeds are first merged with the most up-
stream seed, which is considered to be the likely location of the interaction vertex,
if more than 50% of their nodes are within the necessary angle. Once the upstream
seed is expanded and refit, any unmerged seeds are rerun through the merging
algorithm and added if they meet the necessary criteria with respect to the up-
stream seed. Fig. 3.12 shows a set of seeds from a shower event before and after
the initial merging procedure. A second iteration of the algorithm will add the
green seed shown to the red upstream seed.
Once the seed merging procedure is finished unassociated hits are added. To
add as little noise as possible the allowed XY separation between the seed and
the hit is required to be less than 40 mm, or slightly more than two bar widths.
Hits may also be added even if the hit is downstream of the track end as long as
72
Y Position (mm)
Z P
osi
tio
n (
mm
)
Z P
osi
tion
(m
m)
Y Position (mm)
Figure 3.12: Left: Before and Right: After initial track merging. Some tracks frag-
ments like the one shown in green may be rejected initially for being too far from the
upstream track, but will be added in subsequent iterations
Y Position (mm) 500
Z P
osi
tion (
mm
)5
00
Figure 3.13: A cartoon to represent the method of adding hits to an existing track
object. Hits in black are from the original track seed, hits in red are added, and hits in
purple are either too far to the side or downstream to be added.
the XY separation is less than 40 mm and the Z separation between the hit and
the track end is no greater than 200 mm. This is done to allow the possibility of
small secondary showers which are not large enough or energetic enough to form
seeds. This process is illustrated in Fig. 3.13.
73
3.2.4 Track Matching and Vertexing
Because the scintillator in each P0Dule is divided into two views, ”XZ” and
”Y Z”, the previous algorithms are all done independently in each 2D view and
then combined. The method of matching 2D XZ and Y Z tracks into a single 3D
object is fairly simple. Tracks are matched as long as they overlap with another
track in enough Z layers.
A numerical score is calculated for each possible combination, where an overlap
increases the score by one and a non-overlapping point decreases the score by
one. If a pair of tracks does not share any overlaps, the pair is penalized with
a large negative score. The ratio of the charge separation to the total charge of
overlapping nodes is subtracted from the score to penalize overlapping nodes with
widely varying charge deposits.
The pairs with the highest scores are kept as the final 3D “track” candidates.
Any 2D tracks that do not achieve a high enough total score are left unmatched.
Any new 3D tracks are refitted with the Kalman filter to make 3D nodes. Also,
attenuation corrections done in situ are applied at this stage.
Once the 3D “track” candidates are constructed, all the 2D and 3D objects are
passed to an algorithm which determines the interaction vertices. Every “track”
object, 2D or 3D, is paired with every other track object, including itself, except
for the case of two 2D tracks in different views. Vertices are determined from
each pair by either extrapolating backwards from the track direction to the most
upstream point for a track with itself, or looking for the likely intersection point
for two separate tracks. These constructed vertices are kept as long as the time
difference between the two tracks is less than 40 ns, the X , Y or Z position
uncertainty is greater than 50 cm.
Once vertices have been constructed for every possible combination, the ver-
tices are clustered as long as they are within a 40 ns time window and are within
74
Z Position (mm)
X P
osi
tion
(m
m)
Figure 3.14: Result of vertex finding algorithm. The vertices are marked with crosses.
Often multiple vertices are identified (blue and yellow) even if the majority of the “track”
objects are associated with one (red).
20 cm in distance. This process is iterated until all vertices are clustered with
their neighbors, reducing the number of vertices as much as possible. The combi-
nation of tracks into vertices is shown in Fig. 3.14, with a few leftover unclustered
vertices at the downstream end of the detector.
3.3 Electron Energy Calibration
The primary signature of CCQE νe events is an electron shower. An analysis of
simulated data was performed in 2009 to develop an algorithm for electron energy
reconstruction. The first step in this was to construct a Monte Carlo program to
find the relationship between photoelectrons in the detector and energy deposited
by showering objects in MeV.
Since the P0D is made up of several sections that have very different charac-
teristics, it was necessary to construct several extra geometries containing only
these sections. Currently we have a P0D geometry which contains all ECAL, one
that contains all water targets, one containing all water targets with empty water
bags, and a geometry with only scintillator.
75
Figure 3.15: Left: Ratio of PE/MeV in Water Target between 100-165 MeV Right:
Same ratio in ECAL target between 165-270 MeV
The Monte Carlo particle gun generated electrons in an energy range from 0
to 1000 MeV with semi-logarithmic binning. Then, the energy deposited in the
detector, and the photoelectrons in the detector were compared for each energy
as seen in Fig. 3.15
The ratio of PE/MeV created a Gaussian shape with a left-side tail. This
shape was fitted with a Gaussian with a varying width using the equation
Cex−µ2
2(σmax(1,1+a(x−µ)))2 . (3.3)
where µ is the mean of the Gaussian and σ is the width. The width of the Gaussian
is related to the error in measuring photoelectrons. Statistically this was modelled
as [?]
B√E
+ C, (3.4)
where B√Ecomes from photostatistics where
√
Nphot∼=
√E ∗ constant, with Nphot
is the number of photons, E is the energy and C is a constant term from sampling.
Using this equation to fit the width of the Gaussian at various energies produced
results in agreement with what was seen in the integrated energy distribution as
seen in Fig. 3.16.
76
Figure 3.16: Left: Width of Energy Ratio Gaussian for Water Target Right: Width
for ECAL Target
Electron Energy(MeV)0 200 400 600 800 1000 1200
(pe
/MeV
)tr
ueQ
/ E
3.4
3.6
3.8
4
4.2
4.4
4.6
4.8
5
Electron Energy(MeV)0 200 400 600 800 1000 1200
(pe
/MeV
)tr
ueQ
/ E
2.6
2.8
3
3.2
3.4
3.6
3.8
4
4.2
4.4
4.6
4.8
Figure 3.17: Left: Fitted Ratio of PE/MeV in Water Target for various energies with
uncertainty from plot RMS Right: Fitted Ratio of PE/MEV in ECAL target
We expect that the value of the mean of the Gaussian fit of the PE/MeV
ratio for all energies ought to be approximately the same, and give the conversion
from deposited photoelectron to actual energy in MeV. Therefore, the mean vs.
weighted average energy was plotted for each bin. The result is fairly linear for
the scintillator only, ECAL and water target geometries as shown in Fig. 3.17.
Here the errors are given by the RMS of the width of the Gaussian, and not the
fitted value.
The average value of the mean of the Gaussian for each geometry was then
found. Using this, the photoelectron to MeV conversion was found by plotting the
reconstructed MeV to true MeV ratio in the pure geometries. This is illustrated
77
Figure 3.18: Left: Normalization for Water Target Right: Normalization for ECAL
Target
Figure 3.19: Normalization for Full Detector
in Fig 3.18 where the calbrated reconstructed energy over Monte Carlo energy are
distributed as Gaussian random variables with a mean of 1. This conversion works
very well for all energies in the water target, ECAL and scintillator geometries.
The normalization data was used to reconstruct the energy in the full detector
geometry. The various constants are applied in their respective geometry, and
then the total deposited energy was calculated. Figure 3.19 shows the result,
which is a Gaussian nicely centered around a mean of around 1.
Fluctuations over time in the electronics performance and improvements in
MC geometry caused a mostly linear change in the photoelectron to MeV ratio
for shower deposits [?]. Since the change is mostly linear, it can be corrected for
78
Constant 700.2
Mean 1.07
Sigma 0.09896
Reconstructed/True Energy0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 20
100
200
300
400
500
600
700
Constant 700.2
Mean 1.07
Sigma 0.09896
Constant 495.7
Mean 1.053
Sigma 0.07098
Reconstructed/True Energy0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 20
100
200
300
400
500
Constant 495.7
Mean 1.053
Sigma 0.07098
Constant 550.7Mean 1.057Sigma 0.05418
Reconstructed/True Energy0.6 0.8 1 1.2 1.4 1.60
100
200
300
400
500
600
Constant 550.7Mean 1.057Sigma 0.05418
Figure 3.20: Ereco/Etrue for single electron particle gun events. The particles are
produced at the upstream end of the water target region, and represent three energies
relevant to this analysis; 1, 2 and 4 GeV.
without redoing the previously discussed studies. Instead in 2011, Monte Carlo
electrons were produced with energies of 1,2 and 4 GeV. The reconstructed and
true energy were compared, and the result is shown in Fig. 3.20. The average
ratio of reconstructed energy to true energy was 1.06, so this scaling factor is
applied to all reconstructed tracks in the νe analysis.
79
4 Electron Neutrinos at T2K
and Simulation
4.1 Intrinsic Electron Neutrinos
There are several backgrounds to the signal processes which have to be well under-
stood in order for T2K to make precision measurements. Neutral current single
π0 events are a background for both the electron neutrino measurement and the
sterile neutrino measurement. Since T2K is searching for νe appearance at the
far detector, it is also very important to understand the intrinsic νe of the beam,
which is estimated to make up about 0.4 percent at peak energy as can be seen in
the neutrino flux calculations shown in Fig. 4.1. The most recent T2K analysis
found 11 signal events with an estimated contribution of 1.7 events from electron
neutrinos produced by the beam. Being able to calculate the actual electron neu-
trinos in the beam, as well as producing a Monte Carlo approximation for the
electron neutrino interactions at the near detector are both necessary steps to
reduce the uncertainty on this background.
A major focus of study is searching for charged current quasi-elastic inter-
actions of the beam electron neutrinos in the pi-zero detector(P0D), where the
Feynman diagram is shown in Fig. 4.2. Intrinsic νe in the beam comes predom-
inately from three sources: pion, kaon (K+, KL), and µ decay as shown in Fig.
80
(GeV)νE0 1 2 3 4 5 6 7 8 9 10
PO
T)
21/5
0MeV
/10
2F
lux
(/cm
710
810
910
1010
1110
1210 at ND280µν at ND280µν at ND280eν at ND280eν
Figure 4.1: Electron Neutrino flux at off-axis near detector from the various neutrino
types. The parent particles that dominantly contribute to the nuµ and νe flux are shown
in Fig. 4.3.
Figure 4.2: An electron neutrino charged current quasi-elastic event
4.3. It is important to measure the electron neutrinos explicitly because, although
the K+ contribution is well understood, KL and µ contributions are not. The pro-
portion of KL and KS that makes up the total K0 production is not known, and
low energy muons can decay before they reach the muon monitor, and therefore
cannot be measured.
81
(GeV)νE0 1 2 3 4 5 6 7 8 9 10
PO
T)
21/5
0MeV
/10
2F
lux
(/cm
510
610
710
810
910
1010 allpion parents
kaon parentsmuon parents
(GeV)νE0 1 2 3 4 5 6 7 8 9 10
PO
T)
21/5
0MeV
/10
2F
lux
(/cm
610
710
810
910
1010
1110
1210 allpion parents
kaon parentsmuon parents
Figure 4.3: Left: Electron Neutrino flux from various parents at off-axis near detector
(ND280) Right: Muon Neutrino flux from various parents at off-axis near detector
(ND280).
4.2 Electron Neutrino CCQE Cross Section
In order to constrain backgrounds to the oscillation results, electron neutrino
interactions are measured in the near detector. The expectation is that any de-
viation between the simulation and the data will be caused by backgrounds from
the beam that are not well understood. It is also possible that there could be
some uncertainty in the cross section for the electron neutrino interaction itself,
due to various factors.
To calculate a cross section Monte Carlo simulations based on models and data
from other experiments are used. The T2K experiment officially uses the NEUT
generator [?] to simulate particle interactions. This generator uses the Llewellyn-
Smith approximation as the basis of its simulation. However, this approximation
is based on many simplistic assumptions which do not hold for all values of energy
and four-momentum transfer. Several attempts have been made to parameterize
the difference, but still there are many questions.
Also this approximation is only valid for neutrino interactions with a single
nucleon, not with complex nuclei like those of the P0D. To compensate for this
fact, T2K uses a modified form of the Fermi gas model to estimate the effect of
82
the complex nuclei. The Fermi gas model gives a description of the arrangement
of the nucleons inside the atom.
Then, final state interactions within the nuclei must be considered. Final state
interactions refer to interactions between particles in the nuclei with particles
that interact with or are created by the incident neutrino. These interactions can
produce various experimental signals and must be well understood to distinguish
between quasi-elastic events and their backgrounds. T2K models these effects
with a cascade model and various studies on the production of final state particles
in resonance and deep inelastic scattering reactions.
4.2.1 Llewellyn-Smith Approximation
The Llewellyn-Smith cross section is determined from the well known lepton cur-
rent and a hadron current that can be written in terms of a vector and an axial
component as follows
JH = JV + JA. (4.1)
JV contains three terms related to the vector form factors F 1V , F
2V , and F 3
V
and JA respectively contains three related to the axial form factors FA, F3A and
Fp. A description the the bilinear covariant structure of the currents is given by
several authors [?] [?], and the total sum is given in Llewellyn-Smith’s paper as
Eq. 3.13[?].
Using this hadron current, Llewellyn-Smith calculates the quasi-elastic cross
section to be
dσ
dq2(νn→l−pνp→l+n) =
M2G2cos2θc8πE2
ν
[
A(q2)∓B(q2)s− u
M2+ C(q2)
(s− u)2
M4
]
, (4.2)
83
where q is the invariant four-momentum transfer, M is the mass of the nucleon, G
is the Fermi coupling constant, θc is the Cabibbo angle, Eν is the neutrino energy,
and
s− u = 4MEν + q2 −m2, (4.3)
where m is the mass of the lepton. The functions A(q2), B(q2) and C(q2) are
written in terms of the form factors and ξ, the difference between the anomalous
magnetic moment of the proton and the neutron, as follows
A(q2) =m2 − q2
4M2
[(
4− q2
M2
)
|F 2A| −
(
4 +q2
M2
)
|F 1V |2 −
q2
M2ξ|F 2
V |2(
1 +q2
4M2
)
−4q2ReF 1∗V ξF 2
V
M2+
q2
M2
(
4− q2
M2
)
|F 3A|2 −
m2
M2
(
|F 1V + ξF 2
V |2 + |FA + 2Fp|2
+
(
q2
M2− 4
)
(
|F 3V |2 + |Fp|2
)
)]
,
(4.4)
B(q2) =−q2
M2ReF ∗
A
(
F 1V + ξF 2
V
)
− m2
M2Re
[(
F 1V +
q2
4M2ξF 2
V
)∗F 3V
−(
FA +q2Fp
2M2
)∗F 3A
]
and
(4.5)
C(q2) =1
4
(
|FA|2 + |F 1V |2 −
q2
M2
∣
∣
∣
∣
ξF 2V
2
∣
∣
∣
∣
2
− q2
M2|F 3
A|2)
. (4.6)
F 1V and F 2
V are the vector and FA and Fp the axial form factors of the first class
currents. First class currents have the property of conserving both time and
charge symmetry as well as signs following those of the leading term in their
current under a transformation from neutron to proton. The terms associated
with F 1V and FA are considered the leading terms in the hadron current since
they have no dependence on the four-momentum transfer besides that of the form
factors.
From electron scattering experiments F 1V and F 2
V have been measured, and the
best representation of the result is the BBBA07 parametrization [?]. However,
84
early experiments showed that for low values of four-momentum transfer, the form
factors are approximately dipoles with the following form
1
(1 + Q2
m2V
)2. (4.7)
Since the form of FA was initially unknown, it was considered reasonable to esti-
mate that it would also be approximately a dipole. The value mA, the axial mass
corresponding to the mV variable in the vector dipole approximation, has been
measured by many neutrino scattering experiments at various energies and various
nuclei. None of these experiments have currently tried to make any corrections to
the dipole form of FA similar to the corrections done for the vector form factors
at values of high four momentum transfer.
F 3V and F 3
A are form factors associated with the second class current (SCC).
The existence of such currents requires charge or time symmetry violation, and
current measurements show the size of these violations to be small. Additionally
a nonzero F 3V term would violate conservation of the vector current (CVC). Both
F 3V (0) and F 3
A(0) can be limited experimentally in studies of beta decay. Almost
all current analyses of neutrino data assume that the SCCs are zero. The vector
SCCs only enter the cross-section in terms suppressed by m2/M2, but there are
unsuppressed terms involving the axial SCC form factor.
In many analyses, the term Fp is ignored because it is only involved in terms
∼ m2
M2 , and therefore contributes little to the cross section. If this term is included,
the form factor is determined from PCAC which states (if the residual operator
is considered to be small)
δµJA = Cφ, (4.8)
where φ is the renormalized field operator that creates the π+ and the value of C
can be calculated [?] at q2 = 0. Using this value and the Goldberger-Treiman
85
relation [?], which is given as
gπNNFπ = FAMN , (4.9)
where gπNN is the pion strong coupling constant and Fπ is the pion decay constant.
PCAC gives the following relation between Fp and the pion nucleon form factor
gπnn
Fp(q2) =
2M2FA(0)
q2(−FA(q
2)
FA(0)+
gπnn(q2)
gπnn(0)
1
(1− q2
M2π)). (4.10)
Under the assumption that the Goldberger-Treiman relation holds for all values
of q2, then Fp is given as
Fp(q2) =
2M2FA(q2)
M2π − q2
, (4.11)
where Mπ is the pion mass. This is the relationship that is used in all modern
neutrino generators.
4.3 Quasi-Elastic Cross Section Comparison
The theory of lepton universality would mean that the structure of the cross
section should be the same as in the muon case, and therefore that the Llewellyn-
Smith approximation could also be used. Both kinematic limits and lepton mass
dependent terms in the approximation contribute to the difference in the cross
section. Here we look at the differences in terms of the neutrino energy and the
square of the four-momentum transfer Q2.
As metrics, we define the fractional differences between the muon and electron
86
neutrino CCQE cross-sections
δ(Eν , Q2) ≡
dσµ
dQ2 − dσe
dQ2
∫
dQ2 dσe
dQ2
(4.12)
∆(Eν) ≡∫
dQ2 dσµ
dQ2 −∫
dQ2 dσe
dQ2
∫
dQ2 dσe
dQ2
. (4.13)
The integrals over Q2 in Eqs. 4.12 and 4.13 are taken within the kinematic limits
of each process, and those limits depend on lepton mass as discussed in the next
section.
Another useful metric is the difference between a cross-section in a model with
a varied assumption from that of a reference model. Our reference model derives
F 1V and F 2
V from the electric and magnetic vector Sachs form factors which follow
the dipole form of Eq. 4.7 with C = c2M2V = (0.84) (GeV/c)2, and it assumes
FA is a dipole with C = c2M2A = (1.1) (GeV/c)2. The reference model uses the
derived FP from Eq. 4.11, and assumes that F 3V = F 3
A = 0 at all Q2. We then
define
∆ℓ(Eν) ≡∫
dQ2 dσℓ
dQ2 −∫
dQ2 dσrefℓ
dQ2
∫
dQ2 dσrefℓ
dQ2
, (4.14)
where σrefℓ is the reference model for νℓn → ℓ−p or its anti-neutrino analogue and
σℓ is the model to be compared to the reference.
Both the neutrino and anti-neutrino cross sections are restricted by a maximum
and minimum allowed region in four-momentum transfer. These limits are written
in terms of the center of mass energy El and momentum pl of the lepton and the
neutrino energy Eν as
Q2min = −m2
l + 2Eν (El − |pl|) and (4.15)
Q2max = −m2
l + 2Eν (El + |pl|) . (4.16)
87
Energy(GeV)0.5 1 1.5 2 2.5 3
(N
o K
Lim
s)∆-
-310
-210
-110 Minimum2 Qν
Maximum2 Qν
Energy(GeV)0.5 1 1.5 2 2.5 3
(N
o K
Lim
s)∆-
-310
-210
-110 Minimum2 Qν
Maximum2 Qν
Figure 4.4: The total charged-current quasi-elastic cross-section difference for neutri-
nos (top) and anti-neutrinos (bottom) due to the kinematic limits in Q2. This difference
is −∆ defined in Eq. 4.13, meaning that the electron neutrino cross-section is larger than
the muon neutrino cross-section.
88
Energy(GeV)0.5 1 1.5 2 2.5 3
)2=
1.1
GeV
/cA
(m∆(m
odifi
ed))
-
A(m∆
-0.03
-0.02
-0.01
0
0.01
0.02
0.03=0.9 A mν=0.9 A mν=1.4 A mν=1.4 A mν
Figure 4.5: The change in the fractional difference of muon CCQE cross-section and
electron CCQE when mA is changed from a reference value of 1.1 GeV/c2 in a range
generously consistent with current experimental data.
Figure 4.4 shows the effect of the kinematic limits. Not surprisingly, the effect
is very large near the threshold for the muon neutrino and anti-neutrino reaction.
These effects are accounted for in the description of the quasi-elastic process in
all commonly used neutrino generators. However, it is worth noting that the
difference in Q2 spanned by the two reactions can lead to large effects in varying
form factors that significantly affect either the small or large Q2 parts of the
cross-section.
As previously stated, the differences between the muon and the electron neu-
trino cross section that do not come from kinematic limits come from the depen-
dence on the lepton mass in the form factor approximation. Since the term C(q2)
does not contain a lepton mass term, it has no effect on the difference.
In the anti-neutrino case, the B(q2) term is modified. Going from the neutrino
to the anti-neutrino case is equivalent to exchanging s and u. Since the A(q2) and
C(q2) terms are multiplied by even powers of (s − u) they are not affected. The
effect on the B(q2) term is then to change sign to -B(q2).
As noted above, the vector form factors F 1V and F 2
V are precisely measured
in charged lepton scattering [?]; however, the axial form factor is still uncertain
because neutrino experiments that measure it do not agree amongst themselves or
89
with determinations in pion electroproduction as discussed above. Therefore the
axial form factor will dominate any differences in the electron and muon cross-
sections due to uncertainties in leading form factors.
Figure 4.5 illustrates the change in the fractional difference of muon and elec-
tron neutrino CCQE cross-sections when the axial form factor is varied by chang-
ing the assumed dipole mass in a range consistent with experimental measure-
ments. The size of the effect is of order 1% at very low energy and drops with
increasing energy. This difference in cross-section may be accounted for in varia-
tions of the axial form factor within the analysis of an experiment using a modern
neutrino interaction generator.
At low Q2, the pseudoscalar form factor does have a significant contribution
to the muon neutrino CCQE cross-section, of nearly the same order of the leading
terms. However, Eq. 4.11 shows that the contribution will be suppressed for Q2 >∼M2
π , and all terms involving FP are suppressed by m/M and so the contribution
to the cross-section is negligible for electron neutrinos. At low neutrino energies,
the pseudoscalar form factor effect on the cross-section difference, ∆(Eν) is nearly
as large as that of the kinematic limits. The effect of the form factor as a function
of neutrino energy and Q2 is different for neutrinos and anti-neutrinos.
Current neutrino interaction generators [?; ?; ?; ?] include the effect of FP
shown in Eq. 4.11 under the assumptions of PCAC and that the Goldberger-
Treiman relation holds for all Q2. Experimental tests of the Goldberger-Treiman
relation have identified small discrepancies which imply that the left hand side of
Eq. 4.9 is between 1% and 6% less than the right-hand side [?; ?]. Guidance from
models suggests that this effect is likely to disappear at high Q2 [?]. We examine
the effect of varying FP (0) by 3% of itself as a reasonable approximation to the
possible difference due to this effect. A more significant difference may arise due
to violations of PCAC. This has been directly checked in pion electroproduction
studies [?] which can directly measure FP (Q2) in the range of 0.05 to 0.2 GeV/c2.
90
Uncertainties in this data limit the reasonable range of pole masses in Eq. 4.10
to be between 0.6Mπ and 1.5Mπ. Effects due to these possible deviations from
PCAC and the Goldberger-Treiman relation are shown in Fig. 4.6 along with the
effect of assuming FP = 0 for comparison.
As noted previously, non-zero second class currents violate a number of sym-
metries and hypotheses, and are therefore normally assumed to be zero in analysis
of neutrino reaction data and in neutrino interaction generators. We take a data
driven approach and look at the effect of the largest possible second-class current
form factors, F 3V and F 3
A that do not violate constraints from this data.
Vector second-class currents enter the cross-sections for neutrino quasi-elastic
scattering always suppressed by m/M and therefore only appear practically in
muon neutrino scattering cross-sections. Both vector and axial vector form factors
give large contributions to the B(Q2) term given in Eqs. 4.2 and 4.5, and therefore
typically have very different effects, often even different in sign, for neutrino and
anti-neutrino scattering.
The vector second-class currents are difficult to detect in most weak processes
involving electrons because the process is generally suppressed by powers ofme/M .
Therefore even very precise beta decay measurements have difficulty limiting the
size of F 3V (0) to less than several times the magnitude of the regular vector form
factors [?]. The best limits from beta decays currently limit F 3V (0)/F
1V (0) to be
(0.0011±0.0013)mN
me≈ 2.0±2.4 [?]. Studies of muon capture on nuclei can provide
modestly better limits, but at the expense of assuming there are no axial second
class currents [?]. An analysis of anti-muon neutrino quasi-elastic scattering has
been used to place limits of similar strength, but again under the assumption of
no axial second class currents and with an assumed Q2 dependence, F 3V (Q
2) =
F 3V (0)/(Q
2 +M23V )
2 with a fixed M3V of 1.0 GeV/c2 [?]. From the preponderance
of the data, we choose to parameterize the maximum size of the allowed vector
second class current as F 3V (Q
2) = 4.4F 1V (Q
2), which is not excluded by the results
91
Energy(GeV)0.5 1 1.5 2 2.5 3
mod
ified
|∆-no
min
al|∆
-510
-410
-310
-210
-110π>0 0.6mPF
π>0 0.6mP Fν
G-T Violations
G-T Violationsν
Energy(GeV)0.5 1 1.5 2 2.5 3
nom
inal
|∆-m
odifi
ed|∆
-510
-410
-310
-210
-110=0pF
=0P Fν
π>0 1.5mPF
π>0 1.5mP Fν
Figure 4.6: The effect of variations of FP from the reference model which assumes
PCAC and the Goldberger-Treiman relation. The plots illustrate the change cross-
section difference, ∆(Eν), between a varied model and the reference model. Possible
violations of the G-T relation produce a negligibly small effect, even at low energy. The
range of violations from PCAC allowed by current data would allow significantly larger
changes. The effect of setting FP to zero is shown for comparison.
92
2Q0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
δ
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
V 200 MeV SCCν 200 MeV No SCC ν 200 MeV No SCC ν
V 200 MeV SCCν
2Q0.1 0.2 0.3 0.4 0.5 0.6
δ
-0.4
-0.3
-0.2
-0.1
0
0.1
V 600 MeV SCCν 600 MeV No SCC ν 600 MeV No SCC ν
V 600 MeV SCCν
2Q0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
δ
-0.07
-0.06
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
V 1500 MeV SCCν 1500 MeV No SCC ν 1500 MeV No SCC ν
V 1500 MeV SCCν
Figure 4.7: δ(Eν , Q2), defined in Eq. 4.12, as a function of Q2 for several selected Eν .
The difference between including and not including the maximum allowed second class
vector current (“SCCV”), F3V (Q
2) = 4.4F 1V (Q
2), is shown.
93
Energy(GeV)0.5 1 1.5 2 2.5 3
=0)
V3(F∆
>0)
-V3
(F∆
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
V SCCν
V SCCν
Figure 4.8: Changes in the difference between the muon and electron neutrino cross
sections due to including F 3V .
of any of the above studies. The effect of this is significant, particularly at low
neutrino energies and is shown in Figs. 4.7 and 4.8. Recall that the effect on the
electron neutrino cross-section from F 3V is negligible, so this effect occurs almost
entirely in the muon neutrino cross-section.
In order to examine the effects of the second class current F 3A first we need an
approximation for its four-momentum dependence. It seems reasonable to assume
that, like F 1V , F
2V and FA, F
3A should also be approximately a dipole as in Eq. 4.7.
Since it is an axial current form factor, it seems reasonable to assume that it would
also depend on the same mA that FA depends on. The full form of the form factor
F 3A using the dipole approximation is
F 3A(0)
(1 + Q2
m2a)2, (4.17)
where F 3A(0) is dependent on the constraint that is chosen, and with our default
choice being the Wilkinson calculation.
To fully constrain the behavior, it is necessary to understand the behavior at
Q2=0. Our method is to use constraints on the form factor imposed by the KDR
parameters of Kubodera et al [?]. Several experiments [?] [?] [?] [?] have been
done to constrain the value of the KDR parameters. Using Wilkinson’s method
94
Constraint1 2 3 4 5 6
(0)
A(0
) / F
A3F
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
Figure 4.9: Constraints on the ratio F 3A/FA with errors from 1)Wilkinson 2)Wilkin-
son with Short Range Effects 3)Wilkinson calculation using A=20 KDR parameters
4)Wilkinson calculation using A=20 KDR parameters with Short Range Effects 5)A=12
data 6)A=20 data. The value used in simulation is the maximum magnitude of the re-
sult combined with the error. A red dashed line marks the value that is used as the
base model for the second class currents.
of calculating the magnitude of the form factor from the KDR parameters, F 3A is
constrained to be less than 10% of the value of F 2V at at Q2=0. Figure 4.9 shows
these experimental constraints and the effect we allow in this study.
Figure 4.10 shows the effect of including this allowed axial second class current
on both the difference of electron and muon neutrino cross-sections and on the
muon neutrino cross-section itself. It is significantly smaller than the effect of the
vector second class current because the limits on these currents are more stringent.
4.3.1 Radiative Corrections
To calculate the effect of radiative corrections on the total quasi-elastic cross-
section, we follow the approximate approach of calculating the leading log correc-
tion to order logQ/m, where Q is the energy scale of the interaction process [?].
This approach has a calculational advantage in investigating the differences due to
the lepton mass, m because the lepton leg leading log only involves sub-processes
where photons attach to leptons. The key result from this approach is that the
95
Energy(GeV)0.5 1 1.5 2 2.5 3
=0)
A3(F∆
>0)
-A3
(F∆
-0.006
-0.004
-0.002
0
0.002
0.004
0.006
A SCCν
A SCCν
A SCCν
A SCCν
Energy(GeV)0.5 1 1.5 2 2.5 3
µ∆
-0.006
-0.004
-0.002
0
0.002
0.004
0.006
A SCCµν
A SCCµν
Figure 4.10: Top: Changes in the difference between the muon and electron neutrino
cross sections due to including F 3A; Bottom: the change in muon neutrino cross-sections
due to including F 3A.
96
Energy(GeV)0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
(No
Rad
Cor
r)∆
(Rad
Cor
r)-
∆
-0.22
-0.2
-0.18
-0.16
-0.14
-0.12
-0.1
-0.08
ν
ν
ν
ν
ν
ν
ν
ν
Figure 4.11: Our estimate of the fractional difference between the electron and muon
neutrino total charged-current quasi-elastic cross-sections, ∆ as defined in Eq. 4.13, as
a function of neutrino energy. The negative difference means that the electron neutrino
cross-section is larger than the muon neutrino cross-section.
97
cross-section which allows for the presence of radiated photons, σLLL is related to
the Born level cross-section, σB, by
dσLLL
dEℓdΩ≈ dσB
dEℓdΩ+
αEM
2πlog
4E∗ℓ
m2
∫ 1
0
dz1 + z2
1− z
×(
1
z
dσB
dEℓdΩ
∣
∣
∣Eℓ=Eℓ/z− dσB
dEℓdΩ
)
, (4.18)
where E∗ℓ is the center-of-mass frame lepton energy.
In the case of elastic scattering, the relationship in σB between Eℓ and the
scattering angle, θℓ simplifies the calculation because there is at most one z in the
integrand for which the cross-section does not vanish for a particular lepton angle
z =[
2Eℓ (M + Eν)(
m2 + 2MEν
)
− 2 cos2 θℓEℓEν
×√
m4 + 4E2ν
(
M2 −m2 sin2 θℓ)
− 4m2M2 − 4m2MEν
]
/[
m4 + 4Eν
(
Eν
(
m2 cos2 θℓ +M2)
+m2M)]
. (4.19)
We then obtain the remaining cross-section by integrating Eq. 4.18 over the final
state lepton energy. Note that this procedure only gives a prescription for eval-
uating dσ(Eν,true)/dQ2true; however, the radiation of real photons means that the
relationship between lepton energy and angle and Eν and Q2 in elastic scattering
will no longer be valid. The effect of this distortion of the elastic kinematics will
depend on the details of the experimental reconstruction and the neutrino flux
seen by the experiment, so the effect must be evaluated in the context of a neu-
trino interaction generator and full simulation of the reconstruction for a given
experiment.
The difference of the effect on the total cross-sections as a function of neutrino
energy is shown in Fig. 4.11. We estimate a difference of approximately 10% over
the energies of interest in oscillation experiments. The magnitude of the lepton leg
correction to the muon neutrino total cross-section is smaller, roughly 0.4 times
this difference, so the larger effect is on the electron neutrino cross-section.
98
Our estimation of the effect is surprisingly large at the relevant energies for
oscillation experiments. Although the leading log correction in the lepton mass
may accurately estimate the difference between corrections to electron and muon
neutrino induced reactions, this is only an approximate treatment which should
be confirmed in a full calculation implemented inside a generator. This difference
in cross-sections due to radiative corrections is not included in the commonly used
neutrino interaction generators [?; ?; ?; ?].
4.3.2 Summary of Quasi-Elastic Cross Section Effects
Large differences between the electron and muon neutrino quasi-elastic cross-
sections exist at low neutrino energies from the presence of different kinematic
limits due to the final state lepton mass and due to the presence of the pseu-
doscalar form factor, FP , derived from PCAC and the Goldberger-Treiman rela-
tion. These differences are typically accounted for in modern neutrino interaction
generators.
There are also significant differences due to radiative corrections, particularly
in diagrams that involve photon radiation attached to the outgoing lepton leg
which are proportional to logQ/m. These differences are calculable, but are
typically not included in neutrino interaction generators employed by neutrino
oscillation experiments. If our estimate of these differences, of order 10%, is
confirmed by more complete analyses, then this is a correction that needs to be
included as it is comparable to the size of current systematic uncertainties at
accelerator experiments [?; ?].
Modifications of the assumed FP from PCAC and the Goldberger-Treiman
relation and the effect of the form factors F 3V and F 3
A corresponding to second class
vector and axial currents, respectively, are not included in neutrino interaction
generators. A summary of the possible size of these effects, as we have estimated
99
Energy(GeV)0.5 1 1.5 2 2.5 3
|no
min
al∆-
mod
ified
∆|
-410
-310
-210
-110 Non-StandardPF Non-ZeroV
3F Non-ZeroA
3F
ν
Energy(GeV)0.5 1 1.5 2 2.5 3
|no
min
al∆-
mod
ified
∆|
-410
-310
-210
-110 Non-StandardPF Non-ZeroV
3F Non-ZeroA
3F
ν
Energy(GeV)0.5 1 1.5 2 2.5 3
| ν)no
min
al∆-
mod
ified
∆-( ν)
nom
inal
∆-m
odifi
ed∆|(
-410
-310
-210
-110 Non-StandardPF
Non-ZeroV3F
Non-ZeroA3F
Non-StandardPF
Non-ZeroV3F
Non-ZeroA3F
ν-ν
Figure 4.12: Top and Middle: For the form factors not well constrained and not
accounted for in neutrino generators, a summary of the magnitude of the fractional size
of differences in the total charged-current quasi-elastic cross-sections between electron
and muon neutrinos and anti-neutrinos as a function of neutrino energy. For FP the
average of the magnitude of the PCAC violating effects are summed linearly with the
magnitude of the Goldberger-Treiman violation effect. Bottom: The magnitude of the
difference between ν and ν of the fractional differences which illustrates the size of
apparent CP violating asymmetries in oscillation experiments.
100
them, is shown in Fig. 4.12.
These differences, particularly from the second class vector currents, may be
significant for current [?; ?; ?] and future[?] neutrino oscillation experiments
which seek precision measurements of νµ → νe and its anti-neutrino counterpart
at low neutrino energies. Previous work [?] has demonstrated sensitivity to these
second class currents in neutrino and anti-neutrino quasi-elastic muon neutrino
scattering, and future work with more recent data [?; ?] and newly analyzed
data[?] may help to further limit uncertainties on possible second class currents.
4.3.3 Fermi Gas Model
In atoms with multiple nucleons, the momentum distribution of the nucleons
within the nucleus has a large effect on the cross section for neutrino interactions.
One way of describing this is the Fermi Gas model. This model describes the
distribution of a collection of weakly interacting fermions. Current nuclear theory
says that protons and neutrons move semi-freely inside the nucleus [?], and so the
model can be applied for this situation.
This model describes fermions as existing in rectangular wells, with neutrons
and protons in separate wells. At a temperature of zero, the number of states
allowed for the particle can be found in terms of the particle momentum from the
relation
dn =4πp2V dp
(2πh)3. (4.20)
For a nucleus in its ground state, the lowest states will be filled up to a limiting
momentum pf , called the Fermi momentum. Therefore, the total number of states
can be found by integrating from zero to pf . From this you can get a value for
the quantization volume of an atom with a certain number of nucleons. The value
101
Figure 4.13: Protons and neutrons in their rectangular wells according to the Fermi
Gas Model. The energy Ef is calculated from the Fermi momentum pf and the mass
of the nucleon. Particles inside the well are part of the “Fermi sea”.
of the Fermi momentum is different for the proton and the neutron, so you get
respectively,
Vi =3π2h3Ni
p3fi, (4.21)
where Vi, Ni and pfi are the volume, number and Fermi momentum of the re-
spective nucleon. In NEUT the value of pf is taken to be 225 MeV/c. These set
of assumptions are considered the “pure” or “relativistic” form of the Fermi Gas
model.
NEUT uses an implementation of the “relativistic” Fermi Gas model created
by Smith and Moniz [?]. In order to calculate the total hadron current tensor
component of the cross section Wµν , they sum over all the nucleon momentum
states
Wµν =
∫
d~kf(~k, ~q, ω)Tµν. (4.22)
The tensor terms Tµν are the hadron current tensors calculated for different values
of nucleon energy and momentum. Smith and Moniz give the form of the Tµν terms
102
Figure 4.14: A fit to data of the Smith and Moniz Fermi Gas model [?]. Note poor
fit at high energy transfer.
for the Llewellyn-Smith hadron currents in the 1971 paper. The function f(~k, ~q, ω)
is used to describe the distribution of the nucleons and is given as [?]
f(~k, ~q, ω) =mTΩ
2π3
δ(ǫk − ǫk−q + ω)
ǫkǫk−qni(|~k|)(1− nf (|~k − ~q|). (4.23)
Ω is the quantization volume, like that calculated previously, mT is the target
mass, and ω is the energy transfer from the neutrino. The energy ǫk and ǫk−q and
the number of states as a function of momentum ni(|~k|) and nf |~k − ~q) are the
values for the target before and after the interaction. In the case of the relativistic
model, the n(|~k|) are equivalent to θ(pf − |~k − ~q|).
(1-nf(|~k−~q|)) is the Pauli Blocking parameter. This is included to insure that
the recoil nucleon is outside of the Fermi sea, which means that the nucleon state
is not occupied and therefore forbidden by Pauli exclusion. If the state is already
occupied, the interaction cannot occur and the interaction probability is zero.
After Wµν is calculated, the cross section can be determined in the same way
as described previously. This model works well for lower values of energy loss
103
ω but does not account for the tail at higher values [?]. Some generators make
attempts to correct the tail region, but NEUT does not [?].
4.3.4 Final State Interactions
When a neutrino interaction occurs in complex nuclei, there is a large probability
that there will be further interactions inside the nucleus. The simplest case is when
the struck nucleon then interacts with other nucleons. There is also a possibility
that particles like deltas, kaons, etas or pions are created in the collision and then
go on to interact inside the nucleus as well. The kinds of secondary interactions
inside the nucleus are called final state interactions.
The basic method for simulating the final state particle interactions is called
the cascade model. First, the location of the neutrino-nucleon interaction is de-
termined. To do this, it is necessary to simulate the density of the nucleus.
For atoms with mass number less than seven, the density is considered to be
approximately Gaussian. Since oxygen is the primary target in the T2K exper-
iment and has a mass number of sixteen, this approximation cannot be used.
Instead, NEUT models the nuclear density using a Woods-Saxon potential [?],
which has the following form
ρ(~r) =Z
AρA
1
1 + e|~r|−c
a
, (4.24)
where ρA is the average nuclear density, Z is the atomic number, A is the mass
number, ~r is the radial distance, c describes the size and a describes the width of
the nuclear surface.
Once the initial position is determined, the final state particles are moved a
unit length from the origin and the probability of an interaction is calculated.
If the particle interacts, the products of that interaction, if any exist, are then
followed. If there is no interaction, the original particle is then followed for another
104
unit length and the probability of interaction is calculated again. This process
continues until either the particle is absorbed or is calculated to have left the
nucleus.
Therefore, understanding the probability of interaction inside the nucleus for
different particles is very important. Each of the particles can also interact in
a variety of ways, and the probability of each of these modes must be known.
In NEUT, well tested models and information from various nuclear scattering
experiments is used to predict the interaction probabilities.
Pions can interact in three ways, inelastic scattering, charge exchange and ab-
sorption. High energy pions in these interactions can produce additional particles
which will then be followed separately. For these three modes, NEUT separates
the interaction probability into two categories, pions with momentum greater and
less than 500 MeV/c.
For pions with momentum less than 500 MeV/c a model created by L.L Sal-
cedo et al. is used [?]. This model takes into account many effects including
Fermi motion and Pauli blocking (like in the Fermi Gas model), but also includes
effects like the ∆ width and renormalization corrections. Delta pole dominance
is assumed, but Kroll Ruderman, pion pole, and direct and crossed nucleon pole
loop diagrams are also considered.
It uses a method for determining particle interaction properties called partial
wave scattering. This is a method of describing a plane wave as a series of spherical
waves. Scattering is considered to cause a phase shift in these waves, δ, which can
then be measured experimentally. In the absorption mode, both two body s-wave
absorption and three body p-wave absorption are considered.
Absorption of pions is a major problem in analyses. If a pion is produced and
absorbed, pion production events will look like other signals. In the case of a
charged current quasi-elastic reaction, ν+p → l−+p+π− or ν+n → l−+n+π+,
the signal is a proton and electron in the final state. If a π− is produced in the
105
Figure 4.15: Data for various nuclei of pions produced in photon-nucleon interactions
compared with prediction by L.L. Salcedo et al. [?] with (solid line) and without (dashed
line) pions with energy ≤ 40 MeV in order to compare to data.
nucleus and absorbed, a proton and an electron are the only particles in the final
state, exactly the same signal of the charged current quasi-elastic reaction ν+n →l− + p, but the true reaction is pion production. These kinds of events are called
“CCQE-like” and are a major background for a CCQE cross section measurement.
Therefore simulating absorption correctly is one of the most important aspects of
a final state interaction model.
The Fermi momentum in this model is described as a function of the nuclear
density as follows
pf (~r) =
[
3
2π2ρ(~r)
]−13
, (4.25)
106
Figure 4.16: Probability calculated from model and pion-nucleon scattering data for
inelastic scattering (white), charge exchange (blue) and absorption (green). Black re-
gions represent probability of no interaction.
where ρ(~r) is given by the Woods-Saxon potential as before. A comparison of the
model predictions with data from photon induced pion production is shown in
Fig. 4.15.
Since the density dependence causes a dependence on the pion position and
momentum, these must be modeled precisely. Direction and momentum of the
pion are determined using the average of the phase shifts from previous pion nu-
cleon scattering experiments and uncertainties in phase shift from data and Monte
Carlo comparison of models of the effective nuclear density for pion interactions
[?] [?].
For pions with momentum greater than 500 MeV/c, the interaction probability
is only dependent on the momentum of the pion. NEUT uses pion-nucleon scat-
tering information at these momentums to determine the interaction probability.
The calculated probability values for the three modes are shown in Fig. 4.16.
Kaon nucleon interactions are handled similarly to high momentum pions.
107
Figure 4.17: Interaction probabilities for a nucleon in 16O as a function of nucleon
momentum [?]. The solid curve, the dashed curve, the dotted curve and the dash-dotted
curve correspond to no interaction, elastic scattering, single pion production and double
pion production respectively.
Information about the cross section at various energies is stored from multiple
kaon scattering experiments [?] [?] [?]. The kinematics of the particles produced
in the kaon interactions are also stored and used to predict the momentum and
angle of particles created in the collision.
η absorption through the ηN → N∗ → π(π)N channel is simulated [?]. This
process is considered to be likely to contribute to nucleon decay into η. The cross
section is estimated by the following equation
σ =π
k2(J +
1
2)
ΓηNΓπ(π)N
(W −M∗)2 + Γtot
4
, (4.26)
where k is the momentum of the η in the center of mass system, J is the spin of
the resonance, Γ is the width of the resonance, W is the invariant mass of the ηN
system and M∗ is the mass of the resonance.
In the case of a nucleon interacting with another nucleon, the quasi-elastic
scattering, single and double pion production cases are considered. If a nucleon
scatters quasi-elastically then the particle is followed until it either produces a
pion or exits the nucleus. The possibility of nucleon absorption is not considered.
108
Measurements from nucleon-nucleon scattering experiments are used to predict
the probability of pion production. Like in the high energy pion and kaon cases,
this information is taken from averages of multiple experiments and not from
any particular model. NEUT does consider the case of pions created with and
without a delta resonance separately. Without a delta resonance, it uses the same
information as GCALOR [?], and with a delta resonance it uses a model by
Lindenbaum et Al [?]. The resulting probabilities of each interaction are shown
in figure Fig. 4.17.
4.4 Modeling Backgrounds
When making cross section measurements from data, it is important to understand
all the processes that are contributing. Even if the signal process is well modeled,
poor modeling of the background could lead to data and Monte Carlo differences
in the final result. NEUT models background interactions in the quasi-elastic,
resonance and deep inelastic scattering energy regions.
4.4.1 Resonance
The resonance energy region is dominated by interactions where there is a baryon
resonance that decays to another particle. NEUT considers the possibility of decay
into a single pion, single photon, single kaon and single η (and a nucleon). The
model of Rein and Sehgal [?] is used to predict the cross section for the eighteen
known partial wave resonances with invariant mass (W) less than 2 GeV/c2.
For single pion resonance production, Rein and Sehgal consider fourteen dif-
ferent pion final states. The basic description of these interactions is
ν +N → l +N∗ → l + π +N ′, (4.27)
109
where ν is the neutrino, N and N ′ are the nucleons, π are the pions, l is the final
state lepton and N∗ is the baryon resonance. Rein and Sehgal do not consider
the photon, kaon or η final states, but NEUT uses their method to calculate their
decay amplitudes.
In order to calculate the amplitude for producing each resonance, a modified
Feynman-Kislinger-Ravendal (FKR) model is used. This is a harmonic oscillator
quark model that can be used to calculate the charged and neutral current induced
transition matrix elements from ground state nucleons to the excited resonance
state. The matrix elements are completely specified by the SU6 multiplet, which
describes the wave function of the constituent quarks, and by the radial excitation
mode.
To calculate the amplitude to decay into a certain final state, three things are
needed: the probability distribution, the branching ratio of the state, and the
sign of the decay. The probability distribution is determined using a normalized
Breit-Wigner factor ηνBW (W ), which can be calculated from the width of the
resonance and the kinematics of the interaction. An “elasticity” factor xνE is used
to modify the amplitude based on the branching ratio of the single pion final state
for the decay. The value of the branching ratio is determined from experimental
measurements. The sign of the decay sgn(N∗ν) is calculated, with the requirement
that the isospin Clebsch-Gordon coefficient of the P33 resonance be positive, from
the FKR model previously described. These three factors are combined to form
the decay amplitude as follows
ην =√
xνE · ηνBW (W ) · sgn(N∗ν). (4.28)
The total cross section for each of the final states is then calculated by multiplying
the decay and the resonance amplitudes. The Rein Sehgal model also considers
the affect of interference of the resonances in this region. This is done using a
density matrix which is a sum of all the amplitudes of all the interfering resonances
110
Figure 4.18: Cross-sections of (a) νµp → µ−pπ+ (b) νµn → µ−pπ0 (c) νµn → µ−nπ+
. Solid lines show the calculated cross-sections.
over their spins, which can then be used to describe the angular distribution of
decay products caused by the interference.
NEUT uses this method in the case of the P33 resonance, but not the other
seventeen resonances. For all other resonances, NEUT generates the final momen-
tum of the decay products isotropically in the resonance rest frame. This method
ignores the effects of the interference completely, but still gives reasonable agree-
ment with the data (Fig. 4.18).
4.4.2 Deep Inelastic Scattering
At higher energies, the dominant process is deep inelastic scattering. Deep in-
elastic scattering is different from lower energy scattering processes in that the
neutrino begins to interact with the individual quarks instead of the nucleons.
SLAC in the late 1960s was the first experiment to see this interaction and con-
firm the existence of the quarks, which had previously been only a mathematical
concept.
111
Constructing the cross section for deep inelastic scattering is similar to con-
structing the cross section for quasi-elastic scattering. At the higher energies, some
of the terms can be ignored. NEUT eliminates non-dominant structure functions,
those with a Ml/MN prefactor (with Ml the lepton mass and MN the nucleon
mass), in its method. Also, it is possible to approximately relate the structure
functions F1 and F2 in the following way, in a method called the Callan-Gross
approximation [?]
F2 = 2xF1, (4.29)
where x is −q2/2Mν and ν is (Eν − El), with EN and El the energy in the
laboratory frame of the nucleon and the lepton respectively. This approximation
is only valid for leading order terms in QCD [?]. Higher order corrections exist
but are not used in NEUT [?] [?].
After making all of these approximations, the final form of the cross section
used for W ≥ 1.3GeV/c2 is
d2σ
dxdy=
G2fMNEν
π
(
(1− y +1
2y2 + C1)F2(x, q
2)± y(1− 1
2y + C2)xF3(x, q
2))
)
,
(4.30)
C1 =yM2
l
4MNEνx− xyMN
2Eν
− M2l
4E2ν
− M2l
2MNEνxand (4.31)
C2 = − M2l
4MNEνx, (4.32)
where y = ν/Eν .
To model the form factors F2 and F3 a parton distribution function is used.
These functions cannot be computed by using perturbative QCD, and so they
are constructed from data. NEUT uses information from the GRV98 [?] parton
112
distribution function with corrections to improve fits at lower q2 by Bodek and
Yang [?].
In the region where W ≥ 2GeV/c2 PYTHIA/JetSet is used to generate the
deep inelastic scattering events. For W < 2GeV/c2 PYTHIA does not work as
well, and the single pion production is already being simulated from the resonance
production. For this reason, a separate method is used to generate events in the
low W region. This method uses KNO scaling [?], which relates the probability
P(n) for a certain particle multiplicity n to the average pion multiplicity for that
value of energy in the laboratory frame 〈n〉 as follows
P (n) =1
〈n〉φ(n
〈n〉). (4.33)
Both 〈n〉 and the function φ( n〈n〉) are dependent on the interaction and must be
determined from data. For the pion multiplicity, NEUT uses the value 〈nπ〉 =
0.09 ± 1.83 lnW 2, which is derived from a hydrogen bubble chamber experiment
at Fermilab [?]. The function φ( n〈n〉) is derived from data from BEBC [?]. Using
this combination, all pion multiplicities can be generated in the low W region.
Pion production does also have a forward backward asymmetry which is ac-
counted for by using an empirical fit to data from the BEBC experiment [?]
〈nπf 〉〈nπb〉
=0.35± 0.41lnW 2
0.5± 0.09lnW 2. (4.34)
Finally the location of the deep inelastic scattering interaction is determined by
a formation length. This formation length is the distance required for hadroniza-
tion to occur. The formation length L is the ratio of the momentum of the hadron
and a parameter µ2 which is determined from experiment. In NEUT µ is set to
0.08 GeV 2 based on results from the SKAT experiment.
113
4.4.3 Coherent Pion Production
The final background that is considered is a poorly understood process called co-
herent pion production. This is caused by the neutrino interacting with a complex
nucleus, such as oxygen, instead of one of the constituent nuclei. Current fits are
based on very limited amounts of data, and therefore the uncertainty for this type
of background is very high.
The usual model used for this process is a PCAC approximation by Rein and
Sehgal [?]. The cross section for pion-nucleon interactions has been studied in
detail and has been shown to be fairly well modeled by the PCAC approach.
Therefore, in order to predict the cross section for the coherent pion production,
it is necessary to have a relationship between the pion-nucleon interaction and the
coherent pion interaction. Rein and Sehgal predict the following relationship
dσ(π0ℵ → π0ℵ)|dt| = A2|Fℵ|2
dσ(π0N → π0N)
|dt| |t=0, (4.35)
where ℵ is the complex nucleus, A is the number of nucleons in the nucleus, N
is a single nucleon, π0 is a neutral pion and Fℵ is the nuclear form factor. Fℵ
is broken into a piece representing the absorption of the pion Fabs, and a factor
related to the square of the four-momentum transferred to the nucleus t, like so
|Fℵ|2 = Fabse−b|t|, (4.36)
where b is one third of the square of the nuclear radius, which in the case of oxygen
is 80 GeV −2. With these relationships, the cross section can be determined from
PCAC methods to be
dσ
dxdyd|t| =G2MNf
2πA
2Eν
32π3(1− y)
[
σπ0Ntot
]2
(1 + r2)m2
A
m2A +Q2
Fabse−b|t|, (4.37)
114
where G is the weak coupling constant, r = Re(fπN )/Im(fπN), mA is the axial
vector mass, fπ = .93 mπ, MN is the mass of the nucleon, and y = (Eν − El)/Eν
with Eν and El the energy of the neutrino and the lepton respectively. Fabs is
given by
Fabs = e−〈x〉ρσπNinel , (4.38)
where 〈x〉 is the mean path length, ρ is the nuclear density, and σπNinel is the average
inelastic pion-nucleon cross section.
Like in the quasi-elastic case, there are limitations on the four-momentum
transfer because the lepton in the interaction has mass. NEUT uses a correction
suggested by Adler and implemented by Rein and Sehgal [?]. This correction
factor is multiplied by the previously calculated cross section, and has the form
C =
(
1− Q2min
2(Q2 +m2π)
)2
+y
4
Q2min(Q
2 −Q2min)
(Q2 +m2π)
2, (4.39)
where C is the correction factor, and Q2min = m2
ly
1−y.
115
5 Selection Cuts
5.1 Cut Descriptions
To measure the intrinsic νe content of the beam, νe interactions in the detector
have to be separated from the various background interactions. First data is
reconstructed into track objects. Then algorithms extract properties from the
track objects that are used for discriminating between various particle types and
interactions.
Many methods of discriminating based on different observables are used, such
as log likelihoods and neural net algorithms. In this analysis, a simple cut method
is used. For each discriminating observable, a single value is chosen. Events on
the wrong side of that value are characterized as background candidates.
These cuts were based on many considerations. Where the reconstruction was
known to have problems, such as at large angles with respect to the beam direction,
cuts were made to minimize the uncertainty from these problems. When possible
cuts were made in order to maximize the purity and efficiency of extracting νe
events.
116
5.1.1 Fiducial and Single 3D Track Cuts
Most activity in the detector comes from particles that are created outside the
detector. Since the P0D is not very deep underground, particles created by cosmic
ray interactions in the atmosphere can sometimes be detected. Also, neutrinos
in the beam can interact in the rock and sand that surround the detector. This
latter category provides the majority of the background from the environment.
To limit these backgrounds, only vertices created inside the detector volume
are considered in the analysis. The region inside the detector where vertices
are considered valid candidates is called the fiducial volume. Fiducial volume
dimensions in the P0D are given in Table 5.1.
Min (mm) Max (mm)
x −829.0 771.0
y −846.0 894.0
z −2997.5 −1272.2
Table 5.1: Boundaries of the P0D νe fiducial volume, in terms of ND280 coordinates.
The origin is the center of the basket, as defined in Chapter 1; the center of the P0D is
offset in x and y from the origin. This is the reason for the asymmetric values of the x
and y cut.
For the upstream and downstream definition the edge of the water target is
used. Interactions in the water region can be compared directly to interactions
at the Super-Kamiokande water Cherenkov detector, where ECal regions are pre-
dominately intended for veto and energy containment. Requiring that the track be
within 250 mm of the sides was also determined to provide optimal containment
and background reduction. These boundaries are shown schematically relative to
the different detector components in Fig. 5.1.
A study was done using a simple cut requiring a single 3D track, exiting the
117
Figure 5.1: A schematic showing the approximate location of the fiducial cut in the
P0D. Interaction is restricted to within the water target region and 250 mm from the
side of the detector.
P0D through the back, not the sides. This cut produced a sample that is more
than 90% muons. From this sample the distribution of vertices in data and Monte
Carlo, normalized by POT, can be compared as in Fig. 5.2. This shows that the
current fiducial definition contains the area where the event rate is approximately
linearly distributed in x,y and z. This means there are no obvious excesses to
suggest events are leaking in, or regions where the rate of event reconstruction is
anomalously low.
Particles that exit out the side of the detector, determined by having energy
deposited in the two outermost bars, are automatically vetoed in the νe analysis.
Such tracks were shown to often have poor energy resolution, and sometimes could
not be fit with a Kalman filter. Tracks that exit out the back of the detector how-
ever generally have good energy containment however, because of the downstream
ECal, and are not rejected.
This analysis considers every vertex found in the reconstruction step as a
candidate νe event. However, vertice must have an X , Y and Z position in order
to properly apply the fiducial cut, and therefore must be 3D. Therefore the fiducial
118
(m)vtxReconstructed x-1000 -500 0 500 1000
# of
Eve
nts
0
100
200
300
400
500
600
700
(m)vtx
Reconstructed y-1000 -500 0 500 1000
# of
Eve
nts
0
100
200
300
400
500
600
700
(m)vtxReconstructed z-3000 -2500 -2000 -1500 -1000
# of
Eve
nts
0
200
400
600
800
1000
1200
1400
1600
1800
2000
Figure 5.2: Distribution of vertices for single track vertices selected from muon events
in x, y and z. MC is in red and is scaled by pot, data events are represented by black
circles. The blue lines represent the location of the fiducial volume for the νe analysis.
The distribution in z is truncated, not showing a massive data excess in the second bin;
this excess is from sand muon events entering theP0D, which are not simulated in the
MC.
119
Number of 3D tracks0 1 2 3 4 5 6 7 8 9 10
0
10
20
30
40
50
310×
Signal0π w/ µ0π no µ0π, µno
0π no µno Out of P0DData
Figure 5.3: Number of 3D tracks for vertices passing the fiducial volume cut. Vertices
with exactly one 3D track pass the cut, as shown by the dashed blue lines.
cut removes vertices that do not contain any 3D tracks.
Due to the effectiveness of the track merging algorithm, electron showers are
likely to be reconstructed as a single track. To reduce background from particles
that produce multiple tracks, like pions, a requirement is made that the vertex
contain only a single 3D track. A comparison of data and Monte Carlo for this
cut is shown in Fig 5.3 and shows reasonable agreement.
5.1.2 Median Width Cut
The beam is primarily composed of νµ, and therefore the most common particle
produced is a muon. Muons are minimally ionizing particles, and the tracks they
produce tend to be narrow. In the P0D, these particles predominately deposit
energy entirely in one bar per layer, or split their energy over two adjacent bars.
Fig. 5.4 shows the typical MIP interactions in comparison with a typical shower
interaction.
The P0D contains layers of high Z material, brass in the water region and
120
Figure 5.4: A cartoon showing the typical energy deposit mode of MIPs compared
to electrons. The dashed line shows the initial angle of the particle that produced
the interaction (parent particle), and the solid lines show the path of the subsequent
interactions in the scintillator caused by the parent particle. To reduce the calculated
width of MIP interactions, charge in adjacent bars is merged.
lead in the ECal, which create the electron “showers”, or electron-positron pairs
created from bremsstrahlung radiation. Therefore, a useful variable for separating
muon and electron events in the P0D is the width of the track. The algorithm
that is used to calculate the width calculates the median of the energy weighted
standard deviation of all the node positions in the track.
To calculate the width, first the bars in each node are ordered by the amount
of charge deposited in each. If the two bars with the highest charge in the node
are adjacent, the position from the two bars is averaged, and treated as a single
bar. This is done in order to reduce the width that is reported in the case a MIP
deposits energy in two triangular scintillator bars. Otherwise information from
each hit in a node is used to calculate the median width.
Then the median width for each node, wi is calculated as follows
wi =
√
∑
j Ej(xj − µ′)2∑
j Ej, (5.1)
where Ej is the energy of one of the j hits in the node, xj is the position of the
121
Median width (mm)0 5 10 15 20 25 30
# of
Eve
nts
0
5
10
15
20
25
30
35
40Signal
0π w/ µ0π no µ0π, µno
0π no µno Out of P0DData
Figure 5.5: Distribution of median width with all other cuts applied. Signal and
background are shown stacked, and the data is shown as points with statistical errors.
The dashed line shows the cut position, and the arrow indicates the region passing the
cut. The highest bin is an overflow.
hit, and µ′ is the average of all the positions in the node. Using this method, the
width of MIP tracks will be predominately below 1 mm. The decision to cut at
1 mm was also the only cut in the analysis chosen specifically to maximize the
purity and the efficiency.
The distribution of widths for different interactions, after all other cuts are
applied to maximize discrimination, is shown in Fig. 5.5. It can be seen that
backgrounds with pions are predominately above 20 mm. So, it can be seen that
computing the width in this way is a strong method for eliminating backgrounds.
5.1.3 Kinematic Cuts
Previously the methods for determining the energy and angle of a track were
discussed. Cuts based on this information are performed predominately to reduce
uncertainty due to these methods. For instance, tracks at high angles cannot be
properly fitted with a Kalman filter. This combined with the fact that νe electrons
122
zθcos 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
# of
Eve
nts
0
10
20
30
40
50
60
70Signal
0π w/ µ0π no µ0π, µno
0π no µno Out of P0DData
Figure 5.6: Distribution of cos θz, where θz is angle between the track and the z axis,
after all cuts have been applied. Signal and background are shown stacked, and the
data is shown as points with statistical errors. The dashed line shows the cut position,
and the arrow indicates the region passing the cut. Other cuts, including requiring a
single track and high Eν , as well as the high CCQE content after those cuts, cause the
distribution to be very forward going.
tend to be forward going, motivate a cut at 45% from the beam direction. The
result of this cut is shown in Fig. 5.6.
Because the analysis requires a single 3D track in the vertex, quasi-elastic νe
events are favored over other νe reactions. For this reason, neutrino energy is
determined from a CCQE approximation as follows
Eν =m′
nEe +(
mp2 −m′
n2 −me
2)
/2
m′n −Ee + pe cos θ
, (5.2)
where Eν and Ee are the neutrino and electron energies, m′n is the neutron mass
modified by the binding energy, 34 MeV, mp, and me are the masses proton and
electron, θ is the track angle and pe is the electron momentum.
The deviation of the true energy from the reconstructed energy for νe events
was studied and is shown in Fig. 5.7. At higher energy where the reaction is less
likely to be CCQE the approximation is more likely to underestimate the neutrino
123
True Neutrino Energy(GeV)1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
), T
ruth
ν)/
(E, T
ruth
ν -
E, R
eco
ν(E
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
0.22
), Truthν
)/(E, Truthν - E, Recoν
(E-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
a.u
.
0
2
4
6
8
101 - 2 GeV
2 - 6 GeV
Figure 5.7: Fractional uncertainty in neutrino energy reconstruction for νe events
taken from νe sample Monte Carlo after requiring exactly one 3D track in the fiducial
volume. Each column is normalized to 1, to best illustrate the spread. No energy cut
is applied. The second plot is a profile, split between the 1-2 GeV and 2-6 GeV energy
ranges (without the column normalization used in the left plot).
energy. This shift is shown clearly in the associated profile plot, which shows the
region from 1-2 GeV and 2-6 GeV separately. Other possible causes of energy loss
are energy from the electron leaking out the back of the detector and failure to
reconstruct all the deposited energy into a single track object.
The current analysis does not have a good ability to discriminate νe from back-
grounds at low energies. In Fig. 5.8 it can be seen that events with neutral pions
and low energy muons dominate. For this reason, and also to look predominately
at νe events from kaon decays, a cut was made in energy at 1.5 GeV. Future
analyses will work to improve the discrimination at the lower energies.
5.1.4 Secondary Vertex Cut
For the analysis, every vertex in a cycle is considered as a possible νe candidate
separately. Information from secondary vertices, those vertices in the cycle that
are not the vertex currently being considered, can help to reduce potential back-
124
(GeV)νReconstructed E0 1 2 3 4 5 6
# of
Eve
nts
0
10
20
30
40
50 Signal0π w/ µ 0π no µ0π, µno
0π no µno Out of P0DData
Figure 5.8: Reconstructed quasi-elastic neutrino energy distribution of events passing
all cuts with no energy cut applied, but with weighting factors that are discussed in
Chapter 6. The last bin is an overflow.
ground contamination. Currently the energy and angle of secondary vertices is
used for discrimination purposes, in a method similar to calculating the transverse
momentum, or pT . For this reason it is referred to as the “pT” cut.
Figure 5.9 shows a cartoon of an electron event with a secondary vertex. Re-
construction most often creates secondary vertices in electron events from lower
energy hits near the far end of the track, but within a small angle from the track
direction. Backgrounds, like neutral pions, create secondary vertices that are at
a higher angle from the track direction and deposit more energy on average.
For this cut, all events with only a single reconstructed vertex pass automat-
ically. Electron events are much more likely to have only a single reconstructed
vertex in a cycle than the backgrounds. For events with multiple vertices in a
cycle, for each node of the secondary vertex the angle θi from the direction of the
candidate vertex and energy Ei are summed as follows
pT =∑
i
Eiθi. (5.3)
125
Figure 5.9: A cartoon showing the likely location of secondary vertices (red) for an
electron shower. Pion events are more likely to create secondary vertices from large
energy deposits at higher angles.
This creates a pT distribution as shown in Fig. 5.10. Signal was found to have
a pT value less than 100 MeV rad for 3D secondary vertices and 30 MeV rad for
2D secondary vertices. Because the electron signal is more likely to reconstruct
only a single vertex in a cycle, this cut has a very high efficiency for retaining νe
events.
126
(MeV rad)T
Maximum 3D p0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
310×
# of
Eve
nts
0
20
40
60
80
100
120
140
160 Signal0π w/ µ0π no µ0π, µno
0π no µno Out of P0DData
Figure 5.10: Distribution of the maximum pT in the cycle, after all other cuts (in-
cluding Eν > 1.5 GeV) have been applied. Signal and background are shown stacked,
and the data is shown as points with statistical errors. The dashed line shows the cut
position, and the arrow indicates the region passing the cut. Vertices from cycles in
which there is only one vertex appear in the zero bin, and the highest bin is an overflow.
127
6 Systematic Uncertainties
In previous chapters the methods for simulating the detector, reconstructing
events, modeling cross sections, and finally selecting an electron neutrino charged
current quasi-elastic sample have been described. It is important to understand
the differences between the models that are used and what is actually seen in
data. This chapter deals with trying to understand and quantify these systematic
uncertainties.
6.1 Beam Systematics
Previously the T2K beam line was described in detail. Understanding the beam
is important for understanding the neutrino flux, and then subsequently for pre-
dicting an expected number of neutrino interactions in the detector.
Systematic uncertainty from the neutrino beam can be broken into two basic
categories. The first category are uncertainties in the production of particles
in the initial collision between the proton and the target and any subsequent
interactions that produce particles which contribute to the neutrino flux. To
estimate the uncertainty in this category, interaction models are compared to
data from outside experiments like NA61, as described in Chapter 1.
128
Figure 6.1: Ratio of NA61 to FLUKA for pion multiplicity in bins of pion momentum
p and proton direction θ.
The second category are uncertainties in the properties of the beam, like the
position, angle, focusing and others. Information from the beam monitors is used
to determine the changes in these properties. Information about the stability of
the beam over time is also measured using the muon monitor and INGRID.
6.1.1 Particle Production Uncertainty
Uncertainty in particle production comes from multiple sources. Officially the
pion multiplicity, kaon multiplicity, secondary nucleon multiplicity, tertiary pion
scaling, and beam cross section uncertainty are quantified. Wherever NA61 data
is used it is from the most recent finalized result [?].
Predictions from FLUKA and results from NA61 for pion multiplicity are
compared in bins of pion momentum p and proton direction θ for π+ and π−
separately (Fig. 6.1). The uncertainty is calculated for each bin, and is generally
between 5-10%. In regions where no data is available from NA61, an uncertainty
of 50% was assigned.
Uncertainty for charged kaon multiplicity is primarily calculated using data
from T. Eichten et al [?]. This experiment does not use a T2K replica target and
129
Figure 6.2: Ratio of Eichten data to FLUKA for charged kaon multiplicity in bins of
transverse momentum and xF .
uses protons with 24 GeV/c instead of 30 GeV. The differences caused by using a
Be target instead of a C target are taken into account by doing A scaling, which
accounts for the differences in the kaon absorption in different materials [?]. This
scaling has the approximate form
Ed3σ
dp3= Aα, (6.1)
where E is the proton energy, p is the kaon momentum, A is the atomic number
and α is a function of the scaling variable, found by fitting to data. Comparisons
between Be that is scaled to Al and actual data from the same experiment by
Eichten et al. on Al give about a 10% uncertainty from this scaling method.
Differences caused by the different proton momenta are accounted for by using
Feynman scaling. This type of scaling is used to parameterize the interaction
of particles using the center of mass energy s, the energy, longitudinal and the
transverse momentum of the kaon. An important parameter for this scaling is xF ,
or Feynman x, which is 2pL/√s.
FLUKA at 24 GeV/c and the Eichten data are compared, and then the differ-
ence is converted into an uncertainty on the kaon multiplicity using the Feynman
130
Figure 6.3: Comparison of FLUKA and NA61 for kaon cross section at different
momenta.
scaling. These uncertainties are not in terms of Eν but in terms of the scaling
variables xF and transverse momentum as in Fig. 6.2. The total uncertainty on
the kaon multiplicity is the sum in quadrature of the A scaling uncertainty, the
difference between the FLUKA and the Eichten predictions, and the systematic
uncertainty of the Eichten data itself.
No data exists to compare neutral kaon production for T2K, so the information
from the Eichten study is used with a simple quark parton model assumption to
make a prediction. With isospin symmetry the expected number of valence and
sea quarks in the proton is n = uv/dv = 2, us = us = ds = ds and ds = ds. From
this assumption, a relation between the number of K0 and the number of K± can
be constructed
N(K0L) = N(K0
S) =N(K+) + (2n− 1)N(K−)
2n=
N(K+) + 3N(K−)
4. (6.2)
This model has been shown to agree within 15% with data in the range of xF for
the T2K beam. The relation between the Eichten data and the FLUKA model are
then scaled the same as the charged kaon case (Fig. 6.4). The total uncertainty
131
Figure 6.4: Ratio of data to FLUKA for neutral kaon multiplicity in bins of transverse
momentum and xF .
Figure 6.5: Ratio of NA61 to FLUKA for second nucleon production in bins of trans-
verse momentum and xF .
for neutral kaons is then the sum in quadrature of the simple quark parton model
uncertainty, the Eichten data uncertainty and the scaling uncertainty.
Secondary nucleon multiplicity is studied by comparisons to data from Eichten,
as in the kaon multiplicity case, and Allaby et al. [?].These studies include data
132
from secondary protons and secondary neutrons, but the data is split into regions
of phase space by the Feynman scaling factor. For protons the regions with xF ≥.9
and xF ≤.9 are treated separately and for neutrons only the region with xF ≤.9
is covered.
In the xF ≤.9 region, the production uncertainty is only dependent on the
ratio of the FLUKA result to the data converted by the Feynman scaling. The
result is shown in Fig. 6.5. Outside the region covered by the data, uncertainties
are determined by the nearest comparable data point. In the region with with
.99 ≥ xF ≥.9 production is scaled by an arbitrary factor of 2 and in the region
xF ≥.99 a factor of .944 is used.
Tertiary pion scaling uncertainty is the uncertainty on the “tertiary pion”
production, pions produced from secondary protons and neutrons. Tertiary pion
production uncertainty is calculated using NA61 data compared to FLUKAMonte
Carlo using the Feynman scaling [?]. Because of questions about the accuracy of
this method, a second scaling variable, xR = E ∗ /E∗max, where E∗ is the energy
of the produced pion and E∗max is the maximum available center of mass energy,
is also used and the difference between the two results is compared [?]. The
difference between these two methods gives about a 1% uncertainty on the νµ
flux, which is then used as the tertiary pion production uncertainty.
Finally, the uncertainty on the inelastic production cross sections are calcu-
lated. Data from proton interactions in carbon in multiple experiments is used [?].
There is a large discrepancy between the measured cross section in the Denisov
result [?] compared to the other data sets that are used to calculate these cross
sections as shown in Fig. 6.6. This discrepancy is equal to the proton-nucleon
quasi-elastic cross section, and the belief is that unlike the other data sets, the
quasi-elastic component has not been subtracted.
The difference between the Denisov result and the other data sets dominates
the uncertainty, and cannot be ignored because the theory that the quasi-elastic
133
Figure 6.6: Measured inelastic nucleon production cross sections from several experi-
ments. Results from Denisov [?] are clearly anomalous, and the discrepancy is though
to be from not subtracting the quasi-elastic cross section [?].
cross section is not subtracted is only speculation. Therefore the quasi-elastic cross
section for each of the produced particles is taken to be the uncertainty on the
production cross section for that particle. To see the effect of these uncertainties on
the flux, Monte Carlo is run for multiple “throws”. A “throw” involves assigning
a rate which is chosen randomly from a Gaussian distribution around the mean
with a standard deviation equal to the uncertainty. For each of multiple throws,
with each throw having a different rate, the flux predictions will be different. The
variation in the flux from these throws is then used as the uncertainty from the
production cross section.
6.1.2 Uncertainty in Beam Properties
Beam monitors as described in Chapter 1 provide an estimate for the beam center
and angle of the beam. Alignment uncertainty of the beam monitors and for
the beamline, as well as systematic uncertainties in the monitor measurements
contribute to the total uncertainty on these values. Effects such as uncertainty in
134
Figure 6.7: Variations in the beam position from the muon monitor and INGRID
the beam monitor width measurements, effects of the momentum dispersion and
uncertainties in the quadrapole magnet model also contribute.
Uncertainty in the beam direction is calculated using information from IN-
GRID and the muon monitors. The variation in beam direction for various runs
is shown in Fig 6.7. The total uncertainty on the beam direction is the sum of
the systematic uncertainty of the INGRID and muon monitor measurements, the
actual measured deviation of the beam position, and the alignment uncertainties
for the near detector and Super-Kamiokande. For both the beam center and angle
and the beam direction cases, the uncertainty on the flux is calculated by rerun-
ning the Monte Carlo using new values of these parameters that are modified by
the magnitude of the uncertainty.
Since the three horns focus the beam, horn misalignment can also have an
effect on the neutrino flux. Studies were done to determine the effect, and it was
concluded from this study that only a misalignment in the Y, and not in the X,
directions significantly affected the νµ flux. Horn 1 was aligned with the beam
position monitors, and therefore the uncertainty in its alignment is considered as
part of the beam position monitor uncertainty. For Horn 2 and 3 the effects of
moving the horn in the positive and negative Y direction by 5 mm was investigated,
and the difference in the flux that was caused was used as the uncertainty from
135
Figure 6.8: Ratio of νe flux at near detector with and without horn alignment changes
the horn alignment. The calculated effect of these changes on the νe flux at the
near detector can be seen in Fig 6.8.
The alignment of the target is another important consideration. Beam Monte
Carlo samples were generated with the target rotated by 1.3 mRad in the hor-
izontal direction and 0.1 mRad in the vertical direction. The changes in the νe
flux at the near detector that were caused by this modification are shown in Fig.
6.9.
Finally, the uncertainty on the horn field asymmetry and current are mea-
sured. The current is measured using Rogowski coils, which are toroidal coils of
wire. The manufacturer of the horns gave the uncertainty for the coil calibration
and setting. During the run, the current was continually measured, and the max-
imum difference in the value over that time was taken as the monitor stability
uncertainty. Radial dependence of the magnetic field varies approximately as 1/r.
However measurements show a difference from the prediction, and the difference
is taken as the current-field relation uncertainty. The changes in the νe flux at the
136
Figure 6.9: Ratio of νe flux at near detector with and without target alignment changes
Figure 6.10: Ratio of νe flux at near detector with and without horn current changes
near detector that were caused by horn current uncertainties are shown in Fig.
6.10.
137
The field asymmetry is thought to be caused by a difference in the path length
of the striplines that provide current to the upper and lower parts of the horn. To
test the affect of this asymmetry, a magnetic field that is modified azimuthally is
simulated. Measurements were then taken in both Horn 2 and 3 of the actual field
asymmetry, and the deviation from the simulation were taken as the uncertainty
from the field asymmetry.
6.2 Cross Section Uncertainty
The Monte Carlo simulation program NEUT is based on models constructed to
fit the available data as accurately as possible as described in Chapter 4. However
some of the assumptions used in NEUT are overly simplistic or outdated. Com-
parisons between NEUT predictions and newer data sets and models give an idea
of the uncertainty in the Monte Carlo predictions [?].
Table 6.1shows the systematic uncertainties that were calculated for the dif-
ferent simulated interactions in NEUT.
Process Systematic uncertainty (comment)
CCQE energy dependent (< 1% at 2 GeV)
CC 1π 30% (Eν < 2 GeV) – 20% (Eν > 2 GeV)
CC coherent π± 100%
CC other 30% (Eν < 2 GeV) – 25% (Eν > 2 GeV)
NC 1π0 30% (Eν < 1 GeV) – 20% (Eν > 1 GeV)
NC coherent π 30%
NC other π 30%
Table 6.1: Summary of systematic uncertainties for the relative rate of different
charged-current (CC) and neutral-current (NC) reactions to the rate for CCQE based
on NIWG guideline.
138
Figure 6.11: Results of MiniBooNE CCQE study [?]. This favors an abnormally high
value of the axial mass MA and a large normalization factor of 10%. Uncertainties
from fits to this data are large.
The CCQE interaction study by MiniBooNE is shown in Fig. 6.11. The large
discrepancies at low Q2 with axial mass MA=1.03 GeV, or near the current world
average measurement for that parameter, have been explained by many theorists
as problems with the Relativistic Fermi Gas (RFG) model [?] [?]. Therefore,
systematic uncertainties in the CCQE cross section from nuclear modeling are
given by the difference between the RFG predictions and the prediction of spectral
function models used by the NuWro [?] generator.
For the charged current single pion reaction, NEUT with unmodified cross
sections is compared to MiniBooNE data with FSI effects taken into account.
This is done by comparing the ratio of the CC1π cross section to the CCQE cross
section to limit experiment dependent effects such as uncertainties in the beam
flux. The result is shown in Fig. 6.12. The uncertainty on the data is 20% and
the difference between the NEUT prediction and the MiniBooNE result is 10%
so the overall uncertainty is set to be 30% below 2 GeV.
139
E [GeV]vμ
Fra
ctio
nal
Unce
rtai
nti
es
Figure 6.12: Comparison of the NEUT νµ CC1π+/CCQE cross section ratio to
the FSI corrected MiniBooNE data. Fractional uncertainty in the data (black) and
NEUT/data discrepancy (red) [?]. Error bars include statistical and systematic uncer-
tainties.
For the neutral current single pi-zero measurement MiniBooNE data is com-
pared to the NEUT neutral currents that produce the same final state, a single
pion with no visible charged particles, without requiring that the true reaction be
NC1π0 [?]. The cross section for the reaction was then measured in bins of pion
angle and momentum, and for neutrino energy of 808 MeV. The total uncertainty
for the NC1π0 reaction is taken as the sum of the systematic uncertainty of the
data and the difference between the NEUT and MiniBooNE values as shown in
Fig. 6.2.
Measuring the charged current coherent cross section is challenging. SciBooNE
has several results [?] [?] that set upper bounds, as shown in Fig. 6.13. NEUT
clearly overestimates the CC coherent cross section, but due to the nature of the
current experimental limits a 100% uncertainty is assigned for this interaction.
140
σ [×10−40 cm2/nucleon] Frac. Err. or Diff. [%]
MiniBooNE data 4.56 ± 0.72 ± 15.8
NEUT MC 4.25 -6.8
Table 6.2: Total MiniBooNE flux-averaged cross section for inclusive NC1π0 produc-
tion, including fractional uncertainty in the data (first row) and NEUT/data difference
(second row)[?].
Figure 6.13: Left: Comparison of the NEUT bare νµ CC (red) and NC (blue) coherent
π cross sections compared to the SciBooNE ratios, normalized to the NEUT total CC
cross section. The CC points are 90% CL upper limits. Right: Uncertainty in NC
Coherent cross section from M cohA parameter variation
The neutral current coherent result is a comparison between the measured bare
cross section normalized to the NEUT total CC cross section. The data here has
a 20% uncertainty and the difference between the SciBooNE data and the NEUT
prediction is 15%. This along with the effects in the parameter M cohA as shown in
Fig. 6.13 give the 30% uncertainty for this interaction.
141
6.3 Detector Systematics
As discussed in Chapters 1 and 2, the P0D is made up of multiple components,
each with their own models and calibrations. Uncertainties in these can affect
the event selection by shifting the vertex location, energy, interaction rate, or cut
efficiency. Also, the difference between the assumption that the majority of the
energy in the reconstructed νe events came from the electron and the measured
energy contributions give some systematic uncertainty.
6.3.1 Fiducial Definition
Previously a simple selection requiring a single 3D track exiting the P0D through
the back of the detector was described. The distribution of the vertices for these
tracks, which are predominately muons, show good agreement with data. There-
fore, to see the effect of varying the fiducial volume, the number of selected events
in this sample was compared for different x, y and z boundaries.
These variations are done separately for the +z and -z directions. Interactions
can be produced in either the active material or the scintillator. The reconstruc-
tion however does not distinguish between vertices in brass and vertices in water,
which are adjacent layers, and instead locates all non-scintillator vertices in the
center of the water target. This misplacement causes a discrepancy in the vertex
distribution in the upstream and downstream parts of the detector. This problem
is illustrated in Fig. 6.14, where the difference between the true and reconstructed
vertex position for νe events is plotted for x,y and z. Systematically moving the
vertices in the +z direction results in a higher uncertainty in the vertex position
in the +z than in the −z direction.
To calculate the systematic, the boundaries are varied by the νe vertex res-
olution as determined from the plots in 6.14. For x, y and z this resolution is
approximately 30 mm, or two bar widths. The distributions have long tails, but
142
Entries 126428Mean -0.424RMS 29.91
Reconstructed-True Position(mm)-100 -80 -60 -40 -20 0 20 40 60 80 1000
500
1000
1500
2000
2500
3000
3500
Entries 126428Mean -0.424RMS 29.91
Entries 126428Mean 0.06147RMS 29.59
Reconstructed-True Position(mm)-100 -80 -60 -40 -20 0 20 40 60 80 1000
500
1000
1500
2000
2500
3000Entries 126428Mean 0.06147RMS 29.59
Entries 126428Mean -12.54RMS 32.41
Reconstructed-True Position(mm)-100 -80 -60 -40 -20 0 20 40 60 80 1000
500
1000
1500
2000
2500
Entries 126428Mean -12.54RMS 32.41
Recon P0Dule - True P0Dule-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Fra
ctio
n of
Ver
tices
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8 P0Dule difference between true and recon’d vertex
Figure 6.14: Difference between true and reconstructed vertex position in x, y, and
z, for true νe Monte Carlo events.
these were determined to be caused primarily by the reconstruction of incorrect
secondary vertices, which are predominately eliminated by the pT cut. We choose
to ignore the contribution from these tails to the resolution, since the purpose of
using the resolution is just to get a rough measure of likely variations in vertex
position from reconstruction uncertainties.
The fractional difference between data and Monte Carlo is calculated with the
following equation
V =D′/M ′
D/M, (6.3)
where D and M are the selected data and Monte Carlo events in the simple cut
sample with the original fiducial volume and D′ and M ′ are the data and Monte
143
−2σ −1σ +1σ +2σ
xy 0.995 ± 0.004 0.994 ± 0.003 0.998 ± 0.003 1.002 ± 0.004
−z 0.999 ± 0.001 0.999 ± 0.001 1.002 ± 0.001 1.004 ± 0.002
+z 0.979 ± 0.004 0.979 ± 0.004 1.000 ± 0.001 1.006 ± 0.004
Table 6.3: Data/MC percent difference in event rate per volume ratio. The value of
σ is determined from the vertex resolution for each direction.
Uncertainty (%)
xy 0.6
−z 0.2
+z 2.1
Total 2.2
Table 6.4: Systematic uncertainty in event rate from fiducial volume choice. The total
shown is the sum in quadrature of the three individual uncertainties.
Carlo using the modified boundaries. These fractions are given in Table 6.3.
The systematic uncertainty is taken to be the largest variation in the fractional
difference from one, U , or mathematically U = |V − 1|, in each direction. These
maximum variations and their sum in quadrature, the total systematic uncertainty
for the fiducial volume cut, are given in Table 6.4.
6.3.2 Energy Scale
Uncertainty in the energy scale can come from electronics, MPPC or detector
material modeling. Each of these components was investigated individually. Many
of these parameters are shown to have no significant variation between data and
Monte Carlo, and therefore are ignored in the systematic uncertainty calculation.
For the electronics, known uncertainties exist from variations in bar response,
144
Node E (path length corrected)0 10 20 30 40 50 60 70 80 90 100
Are
a no
rmal
ized
0
10
20
30
40
50
60
70
-310×
No smearing10% smearing15% smearing20% smearingData
Node E (path length corrected)0 10 20 30 40 50 60 70 80 90 100
Rat
io M
C/D
ata
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
No smearing10% smearing15% smearing20% smearing
Figure 6.15: Left: Node energy spectrum in through-going muons for data (dashed
black line) and Monte Carlo with different levels of channel-to-channel response varia-
tion. The node energies are corrected for the path length in the scintillator by a factor
cos θz. Right: Data/MC ratio for each value of the response variation, showing that
σ = 15% best reproduces the data.
and changes in electronics response over time. Variations in the measured response
from the electronics between different scintillator bars is calibrated as discussed
in Chapter 2. However, some residual variation remains. Also, variations in light
production within each individual bar is not accounted for in calibrations.
The effects of random variations between various bars are investigated by vary-
ing the response of each bar. This variation is chosen randomly from a Gaussian
distribution with a mean of zero deviation and various widths. The value of the
variance is determined by comparing the node energy distribution from through
going muon data to models with smearing of 10,15 and 20 %. In Fig. 6.15 the
effect of the bar to bar variation and the ratio of data to Monte Carlo is shown.
The Monte Carlo used in the analysis was then varied randomly using the same
Gaussian with both a width of 0.15/√NHits, where NHits is the number of hits
in the track, and with the full 0.15.
Changes in the electronics response over time are due to changes in over-
voltage from tunings and changes in temperature. Many of these changes have
been studied and corrected for, but still large variations can be seen. The response
145
MPV32 33 34 35 36 37 38 39 40
PO
T-w
eigh
ted
tota
l
0
0.5
1
1.5
2
2.5
3
1810× Region AEntries 22Mean 36.44RMS 0.2368
Region AEntries 22Mean 36.44RMS 0.2368
MPV32 33 34 35 36 37 38 39 40
PO
T-w
eigh
ted
tota
l
0
0.5
1
1.5
2
2.5
3
1810× Region AEntries 22Mean 36.44RMS 0.2368
MPV32 33 34 35 36 37 38 39 40
PO
T-w
eigh
ted
tota
l
0
2
4
6
8
10
12
14
16
1810× Region AEntries 22Mean 36.44RMS 0.2368
Region BEntries 32Mean 34.96RMS 0.1225
Region BEntries 32Mean 34.96RMS 0.1225
MPV32 33 34 35 36 37 38 39 40
PO
T-w
eigh
ted
tota
l
0
2
4
6
8
10
12
14
16
1810× Region AEntries 22Mean 36.44RMS 0.2368
Region BEntries 32Mean 34.96RMS 0.1225
Region CEntries 43Mean 34.06RMS 0.09862
Region CEntries 43Mean 34.06RMS 0.09862
MPV32 33 34 35 36 37 38 39 40
PO
T-w
eigh
ted
tota
l
0
2
4
6
8
10
12
14
16
1810× Region AEntries 22Mean 36.44RMS 0.2368
Region BEntries 32Mean 34.96RMS 0.1225
Region CEntries 43Mean 34.06RMS 0.09862
Region DEntries 33Mean 33.34RMS 0.1048
Region DEntries 33Mean 33.34RMS 0.1048
MPV32 33 34 35 36 37 38 39 40
PO
T-w
eigh
ted
tota
l
0
2
4
6
8
10
12
14
16
1810× Region AEntries 22Mean 36.44RMS 0.2368
Region BEntries 32Mean 34.96RMS 0.1225
Region CEntries 43Mean 34.06RMS 0.09862
Region DEntries 33Mean 33.34RMS 0.1048
Region DEntries 33Mean 33.34RMS 0.1048
Region CEntries 43Mean 34.06RMS 0.09862
Region DEntries 33Mean 33.34RMS 0.1048
Region BEntries 32Mean 34.96RMS 0.1225
Region CEntries 43Mean 34.06RMS 0.09862
Region DEntries 33Mean 33.34RMS 0.1048
Region BEntries 32Mean 34.96RMS 0.1225
Region CEntries 43Mean 34.06RMS 0.09862
Region DEntries 33Mean 33.34RMS 0.1048
Figure 6.16: Response histograms for four time periods in which known variations
occurred. Each entry in the histogram is weighted by the p.o.t. for the contributing
runs. The mean of each histogram is used to derive the calibration constant for the
time period.
of through going muons at different times in the data taking run is fitted with
a Landau distribution convolved with a Gaussian. The results for four different
periods are shown in Fig. 6.16. The variation in response over time from these
studies is 10% and since deviation in time averaged over space can be expected to
be approximately the same as the deviation in space averaged over time, it gives
good agreement with the 15% smearing that was calculated.
The peaks found from these plots are uncorrected by the currently used calibra-
tion constants. After the calibration constants have been applied, the distribution
of MIP peaks forms a narrower peak as shown in Fig. 6.17. The width of this
distribution is taken as the systematic uncertainty, and is approximately 3%.
For MPPCs, the noise rate is known to be significantly off. In Monte Carlo
a sample of 1 GeV electrons was created. In Fig. 6.18 the reconstructed energy
for the sample for different noise levels is compared. No large change is observed,
and the systematic uncertainty is taken to be zero.
Another known uncertainty in the MPPCs is the saturation. As described in
146
MPV35 35.5 36 36.5 37 37.5 38 38.5 39
PO
T-w
eigh
ted
0
5
10
15
20
25
1810×
Figure 6.17: Through-going sand muon responses after correction by the calibration
factors, with each entry weighted by p.o.t.
Reconstructed Energy (MeV)0 200 400 600 800 1000 1200 1400
0
50
100
150
200
250
300
350
400 Noise Level
MC Default
5×
10×
20×
Figure 6.18: The reconstructed energy of 1 GeV electrons. Increasing the simulated
noise rate has no discernible effect on the energy scale.
Chapter 2, saturation is simulated using data from muon interactions, and the
saturation model assumes a uniform light distribution across the fiber. For this
study, the photon detection efficiency and the light distribution were varied, and
the effect on the reconstructed energy for a Monte Carlo distribution of electrons
is taken as the systematic uncertainty.
The electrons are produced at 500 MeV, 1,2 and 4 GeV, and are distributed in
147
Figure 6.19: A measurement of PDE for various trigger times in the integration
window with an overvoltage of 1.34 V. This overvoltage is higher than the overvoltage
used in the P0D data taking, but the figure clearly shows that an error of almost 50%
on the PDE is clearly too large [?].
the upstream portion of the detector. Since modifications in the photon detection
efficiency and light distribution effect the MIP peak, it is also necessary to produce
a sample of muons for each configuration. Once the MIP peak is measured from
the muons, it and the recalibrated parameters are propagated through the entire
reconstruction and calibration chain.
In the UK saturation study, a Gaussian distribution with sigma varying be-
tween 0.15 and 0.3 were considered, with a value of 0.25 giving an improvement
in data and Monte Carlo comparison. We vary this parameter by ± 0.5 and look
at the variation in the reconstructed energy in the electron distribution.
Photon detection efficiency variation was also measured. The fit result was
0.26 ± 0.13, compared to the value used in the analysis which is 0.29. From [?] it
is obvious that an uncertainty of 0.13 on this parameter is unreasonably large, as
can be seen in Fig. 6.19. Therefore the difference between the measured value of
0.26 and the analysis value of 0.29 is taken as a reasonable variation. The results
148
500 MeV 1 GeV 2 GeV 4 GeV
% % % %
Light Model
σ = 0.20mm −3.91 ± 0.18 −5.59 ± 0.15 −7.36 ± 0.14 −10.97 ± 0.19
σ = 0.25mm −2.16 ± 0.17 −2.83 ± 0.16 −4.07 ± 0.13 −6.52 ± 0.17
σ = 0.30mm −0.52 ± 0.18 −1.14 ± 0.15 −1.79 ± 0.13 −3.08 ± 0.18
PDE
0.26 0.88 ± 0.18 0.80 ± 0.15 0.86 ± 0.13 0.47 ± 0.18
Table 6.5: The percentage change in the reconstructed energy of mono-energetic elec-
trons, for different MPPC sensor configurations. The 1 GeV values are used to estimate
the potential variation in the energy scale, as these electrons are closest in energy to
those produced by 1.5 GeV neutrinos, the energy of the kinematic cut.
of these variations is shown in Table 6.5.
One source of possible variation between data and Monte Carlo from detector
material is in the bar attenuation. As discussed in Chapter 2, the attenuation was
measured with through going muons and the P0Dule scanner to within 2% of the
fit prediction. The attenuation function that is derived from these measurements
agrees in data and Monte Carlo, so no systematic uncertainty is assigned from
this source.
The density and thickness of the detector material also introduces some pos-
sible data and Monte Carlo differences. The properties of the lead and the steel
in the detector have uncertainties in their densities, but the default thickness is
correct. Varying the density by ±1σ, from the maximum to the minimum value
shown in Table 6.6, the uncertainties from the density are calculated.
For the brass and the Ti02 known uncertainties in the thickness exist. In both
cases the mean value of the thickness is modified to the correct value and then
the ±1σ variation is taken around that value. For the brass, the total uncertainty
149
is calculated by modifying the Monte Carlo values of the density, as shown in ??,
and thickness separately. The thickness of the brass in the P0D was measured to
be 1.28 with an uncertainty of 0.03 mm [?]. The total uncertainty for brass is then
the sum in quadrature of the two measured variations. For Ti02 the uncertainty
comes from the thickness alone.
Material Default Minimum Maximum
Brass 8.50 g/cm3 8.13 g/cm3 8.87 g/cm3
Steel 8.00 g/cm3 7.72 g/cm3 8.28 g/cm3
Lead 11.35 g/cm3 11.30 g/cm3 11.40 g/cm3
TiO2 0.10 mm 0.17 mm 0.24 mm
Table 6.6: Variation of geometry parameters, to study the effect on the energy scale.
The ‘default’ values were all taken directly from the Monte Carlo.
To see the effect on the energy reconstruction, the same sample of electrons
used to calculate the MPPC saturation effect is used. The values in the Monte
Carlo, given as default in Table 6.6, are changed to the ±σ values, and the change
for each energy is found. For the purpose of the systematic, the variation of the
1 GeV electrons is used, since it is closest to the energy cut value of 1.5 GeV.
6.3.3 Rate and Mass Normalization Weightings
The Monte Carlo is normalized to the data based on several measurements. For
the inclusive νe analysis, an overall rate normalization is done in the same manner
as the Run I-II νe appearance paper [?], based on a mostly charged current quasi-
elastic sample of muon neutrino interactions in the tracker. The ratio of data to
Monte Carlo that they found was
Rµ,DataND /Rµ,MC
ND = 1.04± 0.028(stat)+0.044−0.037(det.syst.) ± 0.038(phys.syst.).
150
Material Variation 500 MeV 1 GeV 2 GeV 4 GeV
% % % %
Brass −1 σ 0.68 ± 0.50 1.10 ± 0.44 0.65 ± 0.48 1.30 ± 0.85
+1 σ −0.52 ± 0.52 −0.57 ± 0.43 0.12 ± 0.47 0.50 ± 0.86
Steel −1 σ 0.86 ± 0.51 0.31 ± 0.44 0.55 ± 0.49 1.09 ± 0.84
+1 σ 0.44 ± 0.51 0.00 ± 0.45 0.50 ± 0.48 1.12 ± 0.83
Lead −1 σ 1.13 ± 0.50 0.86 ± 0.43 1.07 ± 0.47 0.61 ± 0.92
+1 σ 1.26 ± 0.50 0.74 ± 0.43 0.80 ± 0.47 2.08 ± 0.82
TiO2 −1 σ 0.10 ± 0.51 −0.53 ± 0.46 0.34 ± 0.49 0.04 ± 0.86
+1 σ 0.42 ± 0.50 0.31 ± 0.42 0.63 ± 0.49 1.25 ± 0.84
Table 6.7: The percentage change in the reconstructed energy of mono-energetic elec-
trons, as the detector geometry is varied. In each case, the geometry is changed to
the current best estimate, and then an individual parameter is varied by 1σ. The 1
GeV values are used to estimate the potential variation in the energy scale, as these
electrons are closest in energy to those produced by 1.5 GeV neutrinos, the energy of
the kinematic cut.
This result implies a 6% uncertainty from the rate normalization.
For the CCQE-like study, the rate is normalized by 1 and the uncertainty on
this normalization is calculated from the previously mentioned beam systematics.
These systematics are combined into a 160×160 bin covariance matrix as shown in
Fig 6.20. The entries are separated by bins of energy, neutrino type, and detector
to account for the correlations between the different parameters. For instance, the
νµ and νe fluxes are correlated since both are dependent on the rate of secondary
particle production in the beam.
To get the uncertainties from the covariance matrix, Cholesky decomposition
is used. The portion of the n × n matrix corresponding to the parent neutrino
151
Figure 6.20: 160×160 bin covariance matrix of beam flux uncertainties. The first 80
bins are the near detector and the last 80 bins are the far detector, ordered by neutrino
type and energy.
and detector for the event is selected for decomposition. A variation in the flux is
generated for each of the n bins by multiplying the decomposed covariance matrix
by a randomly sampled value from a Gaussian with mean of zero and a width of
one.
Ten thousand of these variations form a Gaussian distribution as shown on
the left side of Fig. 6.21. The width for this distribution is then taken as the
systematic uncertainty for that bin. For each bin, the ratio of the width and the
mean in each bin is plotted. The average uncertainty from all bins of this plot is
10.8%.
Another normalization that is used is the mass uncertainty normalization.
Measurements were made of the P0D fiducial mass for Run I and Run II [?]. The
definition of the fiducial region is an area with an X width of 1600 mm and a Y
width of 1740 mm, with the center at the center of the P0D and within the water
target region.
Currently there is no known change in the detector mass between Run I and
Run II, and malfunctions in the water depth sensors in Run I, as described in
152
/ ndf 2χ 130.5 / 131Constant 2.4± 195.2 Mean 0.02± 18.91 Sigma 0.015± 2.019
10 15 20 25 300
20
40
60
80
100
120
140
160
180
200
220
/ ndf 2χ 130.5 / 131Constant 2.4± 195.2 Mean 0.02± 18.91 Sigma 0.015± 2.019
Fractional Variation0.08 0.09 0.1 0.11 0.12 0.13 0.140
2
4
6
8
10
12
14
Figure 6.21: Left: An example set of ten thousand throws. The result is a Gaussian
with a sigma equal to the systematic uncertainty from the flux measurement for that
bin. Right: Sigma of the fitted gaussion as a fraction of the mean.
Chapter 2, lead to a higher uncertainty in the water mass measurement. For this
reason, we use the Run II measurement, which is known with greater confidence.
The uncertainties calculated in this measurement take into account the total un-
certainty from the water mass measurements and other detector materials such
as brass, lead, scintillator, and even the p0dule covers and epoxy that hold the
detector together.
The Run II measurement for the period when the detector was filled with
water was 5484 ± 44 kg compared to the Monte Carlo value of 5799 kg. The
deviation from the measured mass is 4.3 ± 0.8 % [?]. This uncertainty of 0.8 %
is taken as a systematic uncertainty on the Monte Carlo predicted rates.
6.3.4 Muon Rejection
The ability of the width cut to reject muons depends on the agreement between
data and Monte Carlo of the reconstructed width. In order to do this comparison,
“sand muons”, or muons produced by neutrino interactions in the sand surround-
ing the detector hall, were selected from data. This selection required that there
be a single track in an interaction cycle that passes through the upstream ECal
153
Track length (m)0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
3.5
310×
Particle gun MC
Data
Figure 6.22: Length of tracks in sand muon particle gun MC (red) and data (black),
used as a proxy for muon energy. The agreement is generally good, except for short
tracks. The MC is normalized to the same number of events as the data.
of the P0D and stop inside the water target or downstream ECal, with no associ-
ated reconstructed shower. Since no Monte Carlo predictions exist for the exact
distribution of sand muon interactions, a sample was created with the above re-
quirements, and distributed such that the angle and energy of the tracks closely
matched that of the selected data events, as shown for track length in Fig 6.22.
The width is calculated using the same algorithm as the width cut, as discussed
in Chapter 5, but only for nodes that are downstream of the upstream ECal, since
that region is excluded in the analysis. A comparison of the results for data
and Monte Carlo is shown in Fig. 6.23. In order to calculate the systematic
uncertainty, the ratio of data to Monte Carlo above the cut width, as given in
Table 6.8, is used to scale up the background that is composed of muons with no
associated neutral pions.
The large disparity between data and Monte Carlo is believed to be caused by
the factor of four difference in the noise rate and a mis-modelling of the charge
distribution of the noise. In Fig. 6.24 it can be seen that the large discrepancy
just above the cut width disappears if the contributing hits are required to have
a charge of no less than 10 photoelectrons. The mechanism of the width increase
154
Median width (mm)
0 1 2 3 4 5 6 7 8 9 10
Are
a n
orm
aliz
ed
-410
-310
-210
-110
1
Particle gun MC
Data
Figure 6.23: Median width distribution of stopping sand muons in data (black) and
particle gun Monte Carlo (red) with median width cut shown(blue). The selection
criteria are described in the text.
Selected fraction (%)
Data 1.4
MC 1.2
Ratio Data/MC 1.18
Table 6.8: Fractions of stopping muons selected by the width cut in data and Monte
Carlo, and their ratio. The ratio is used in the analysis to scale up selected MC events
without a π0, and 100% of the correction is included as a systematic on this effect.
from the excess of high charge noise is that noise hits are added to nodes that
would otherwise contain only a single hit, or two hits that would be merged based
on the width algorithm criteria. The separation between the noise hit and the true
MIP hits is then recorded as an anomalously high width, creating the discrepancy.
6.3.5 Angular Resolution
Reconstruction of the track angle is used to calculate the kinematics of the recon-
structed tracks. Variations between the reconstructed angle in data and Monte
155
Node width (mm)0 5 10 15 20 25 30
1
10
210
310
Node width (mm)0 5 10 15 20 25 30
1
10
210
310
Figure 6.24: Distribution of node widths in data (black) and particle gun sand muon
MC (red), with the width cut also shown(blue). Left: with default reconstruction
parameters. Right: with all hits below 10 photoelectrons removed before reconstruction.
In each case, the MC histogram is scaled to the same number of events as the data.
Carlo could therefore affect the reconstructed energy and Q2 distributions. To
simulate the effect of an incorrect angular resolution in MC on an event-by-event
basis, the unit vectors representing the true electron direction t and the recon-
structed track direction r are used to construct a new track direction r’ as follows
r′ =t+ α(r− t)∣
∣t+ α(r− t)∣
∣
. (6.4)
This has the approximate effect of increasing the angular resolution by a factor of
α if r is close to t. After the new direction r’ is calculated, the result is modified
and the difference between the original result and the new result with the modified
angle is taken as the systematic uncertainty. For a range of values of α between
0.5 and 1.5, the fractional change in number of selected events is less than 0.5%.
This uncertainty is small and therefore is not included in the total systematic
uncertainty.
156
Charge Deposit (p.e.)0 100 200 300 400 500 600
# of
Eve
nts
-110
1
10
210Side DepositSide Deposit
Figure 6.25: The side charge deposit of events passing all cuts other than the side
exiting cut.
6.4 Additional Checks
6.4.1 Side Exiting Events
As previously noted, the noise rate in the P0D is about four times the simulated
value. This might cause more events to be reconstructed as having energy deposits
in the side of the P0D in data than in Monte Carlo. These events would then be
classified as “side exiting” and be removed by the analysis cuts.
Noise hits are known to predominately have energies in the 0 to 10 photoelec-
tron range as discussed in Chapter 2. In Fig. 6.25 the side deposit charge for all
events that pass all cuts except the side exiting cut are shown. For clarity the
plot has been shifted down by a small fraction of a photoelectron and all of the
events which pass the side exiting cut, which make up the majority of the events,
are moved into the -20 to 0 bin. It can be seen that the 6 remaining data events
which are rejected by the side exiting cut all have higher energy than the expected
noise. Therefore, no systematic uncertainty is deemed necessary for this cut.
157
MC Data
Scanner Ntot Nmerged Fraction (%) Ntot Nmerged Fraction (%)
1 64 8 12± 5 119 19 16± 3
2 63 5 8± 3 119 6 5± 2
3 64 4 6± 3 119 10 8± 2
Table 6.9: Results of hand scan of selected events. The total number of events scanned
Ntot, and the fraction classified by the scanner as “merged” Nmerged, are shown, along
with the fraction including statistical uncertainty. )
6.4.2 Hand Scan
The track merging stage of the P0D reconstruction can cause multiple MIP-like
tracks to be merged, forming one track with high width. These events are then
likely to pass the width cut despite being composed entirely of MIP-like tracks.
Since it is difficult to determine for certain if components of merged tracks in data
are actually MIPs, one way to compare the rate of merging in data and Monte
Carlo was by using human event scanning. Three scanners looked at all data
events events passing all cuts, and a common set of MC events passing the same
selection, noting events which they believed contained a set of merged tracks.
The results are shown in Table 6.9. For each scanner, there is no statistically-
significant difference in the fraction identified as merged between data and Monte
Carlo, so this systematic is taken to be negligible.
6.4.3 Proton Reconstruction
The most likely particle in a νe interaction to be reconstructed as a single track
is the electron. In some cases however the proton is identified instead. Protons
misidentified as signal can also come from background interactions.
The P0D reconstruction does not have a dedicated particle identification method
158
Figure 6.26: Left: Efficiency of reconstructing a track, for single particle gun protons,
gener- ated with uniform energy and isotropic, but downstream, angular distributions.
The dominant effect is the number of hits passing the cleaning thresholds, leading to
a strong energy dependence. Right: Kinetic energy distribution of protons in events
passing all selection criteria.
for protons. Therefore the efficiency for selecting protons with the current selec-
tion methods was studied using Monte Carlo particle gun protons. In Fig. 6.26
the distribution of energies for protons selected by the analysis and the efficiency
for selecting them at each energy from the particle gun study are plotted. Below
150 MeV few protons are reconstructed because the energy deposit is so low in
the detector that it is vetoed as a noise signal.
The effect of incorrect proton identification is greatest near the reconstruction
threshold, at 150 MeV, and in the high energy tail. For the energy threshold, a
20% variation in the proton detection efficiency was considered plausible due to
the high discrepancy between the noise rate in data and Monte Carlo. In the high
energy tail 20% was considered plausible due to consensus from NIWG studies of
the CCQE cross section as a function of Q2. As was previously discussed, there
currently exist discrepancies between data and Monte Carlo that are large enough
to justify the 20% uncertainty.
159
7 Sidebands, Final Analysis and
Results
All analyses and sidebands that will be discussed use the full Run I and Run II
data in which the P0D was filled with water. Only data passing quality checks in
the P0D are used, ultimately amounting to 8.6×1019 p.o.t. divided between Run I
and Run II as shown in Table 7.1. In Table 7.2 the subrun and run identification
for the data is given. All data comes from P0D Production 4B with corrections
specified previously for known saturation mis-modelling.
p.o.t. / 1019
Run I Run II
Data 2.9 5.7
MC 49.9 109.9
Table 7.1: p.o.t. used in data and Monte Carlo, by run period. The Run I and Run II
MCs differ by detector geometry (the latter includes with the full ECal installed between
the P0D and the magnet), and by simulated beam intensity and number of bunches.
Each MC sample is scaled to the relevant data sample, and summed after scaling.
Before looking at the effects of the full set of cuts on data and Monte Carlo,
sidebands are created from events failing the width and pT cut. These sidebands
are used to estimate the effectiveness of the cuts and the validity of the systemat-
160
ND280 Run Number Date
Start End Start End
Run I 4165 5115 2010-03-18 2010-06-25
Run II 6462 7663 2010-11-18 2011-02-14
Table 7.2: ND280 run numbers and dates for the data sample used for this analysis.
ics. Because these cuts predominately eliminate background from specific particle
interactions, muons in the case of the width and π0 for pT , they can be used to
constrain the backgrounds in the νe analyses.
Two separate νe analyses were performed. The first is done as a crosscheck
on the νe flux to assure that the beam modeling was reasonable for the 2011
oscillation analysis. In the second, the νe CCQE-like signal is compared for data
and Monte Carlo as a function of Q2.
The two analyses use two different sets of flux weights. Flux weights were
described in Chapter 6, and the relevant weights for these analyses are called the
“11av1” and “11av2” weights. The primary difference between the two is that the
11av1 weights do not include kaon tuning, which was added in 11av2.
The analysis that uses the 11av1 weights was done as a crosscheck for the
νe oscillation analysis and uses the flux weights from the TPC measurement of
the νµ flux. Therefore the TPC measurement, the oscillation measurement and
the νe crosscheck are all consistent. In the second analysis the 11av2 weights are
used, since it is a stand alone analysis and the kaon tuning decreases the flux
uncertainties.
For a better understanding of the relative size of the systematic effects, the
fractional uncertainty in the number of selected signal and background MC events
for the rate analysis are shown in Table 7.3. From the table it is clear that the
largest systematic errors come from the cross section and energy scale uncertain-
161
Systematic Uncertainty Signal (%) Background (%)
EM Scale 7 16
Cross Sections 0 25
Rate Normalization 6 6
Fiducial Volume 2 2
Muon Rejection 0 4
MC statistics 3 4
Mass uncertainty 1 1
Total 10 31
Table 7.3: The fractional uncertainty in total number of selected signal and background
MC events for each systematic. The total is a sum in quadrature of all errors. Errors
with the same value, and the EM scale, are 100% correlated. (This is taken into account
in the final result.)
ties. The uncertainty on the background is especially useful for understanding the
size of reasonable variations in the sideband studies.
7.1 Sidebands
The largest backgrounds for the νe analysis are from events with muons and
neutral pions. Because the beam is primarily νµ, muons are produced far more
frequently than the electrons that are the expected signal. Even if 99% of the
muons are eliminated, the number of muon events remaining in the sample will
still far outnumber the νe events. The muon background is dramatically reduced
by the existing cuts, but is most dramatically targeted by the width cut.
Therefore an “anti-width” cut is done as a sideband. In Fig. 7.1 it is shown
that events that fail the width cut are predominately muon events with no as-
sociated neutral pion. Flux weights, along with the FV mass and tracker result
162
(GeV)νReconstructed E1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
# of
Eve
nts
1
10
210Signal
0π w/ µ0π no µ0π, µno
0π no µno Out of P0DData
Figure 7.1: Data and Monte Carlo comparison for events that fail the width cut but
pass all other cuts (the range covers events below the Eν threshold, but this is marked
in blue at 1.5 GeV).
MC scaling factors are applied, but muon rejection scaling is not applied, since
it applies only to events passing the width cut, while this sample contains events
which fail the cut. An example of an event that fails the width cut, likely a muon,
is shown in Fig. 7.2.
In the analysis region, to the right of the blue line in the figure, the data and
Monte Carlo agree well. Below this line the agreement is visibly worse. Above 1.5
GeV the Monte Carlo predicts 26 events predicted and data predicts 23 events for
a 0.6σ difference (with statistical errors only). Between 1 and 1.5GeV the most
significant discrepancy occurs, where there are 330 data events with a prediction
of 365 events. This still gives a difference of 10%, which is within the background
systematic errors of 25%. Overall there are 13844 data events with a prediction
of 14340, an agreement at the 4% level.
Neutral pions are not produced at as high a rate as muons. However the
two photons that are produced from neutral pion decay produce electromagnetic
showers similar to those of the electron. If one of the photons is too low energy
163
-300 -250 -200 -150 -100
-100
-50
0
50
100
Top (X-Z)
-300 -250 -200 -150 -100
-100
-50
0
50
100
Side (Y-Z)
Figure 7.2: An example of an event (Run 4542, event 44384) failing the “Median
Width” cut. Boxes indicate detector hits, with size proportional to hit charge. Hits
not associated with any reconstruction object are drawn in black, and those associated
with a track are drawn in red. The red line indicates the reconstructed track position
at each layer as found by the Kalman filter, and the cross indicates the reconstructed
vertex position with an ellipse indicating its uncertainty.
-300 -250 -200 -150 -100
-100
-50
0
50
100
Top (X-Z)
-300 -250 -200 -150 -100
-100
-50
0
50
100
Side (Y-Z)
Figure 7.3: An example of an event (Run 4408, event 18642) failing the pT cut. See
Figure 7.2 for details of the display. Each track is drawn in a different color.
to be detected or if the two photons overlap then the problem is especially bad.
Ideally the neutral pion background would be measured in the P0D and provide
164
(GeV)νReconstructed E1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
# of
Eve
nts
0
20
40
60
80
100
Signal0π w/ µ0π no µ0π, µno
0π no µno Out of P0DData
Figure 7.4: Data and Monte Carlo comparison for events that fail the pT cut but pass
all other cuts (the range covers events below the Eν threshold, but this is marked in
blue at 1.5 GeV). Flux weights and the MC scaling factors are applied. The highest bin
is an overflow.
some constraints for this effect. However, when the analyses were done neutral
pion production in the P0D had not been measured. The “anti-pT” sideband
predominately selects events containing a neutral pion, either with or without a
muon. An example of an event rejected by this cut is shown in Fig. 7.3. In Fig.
7.4 it is shown that there is good agreement between data and Monte Carlo for
this sideband. Above 1.5 GeV 363 events are predicted by the MC compared with
376 events in data, for a difference of 0.7σ [?].
7.2 Final Analyses and Results
7.2.1 Rate Analysis
The purpose of the first analysis is to constrain the νe background in the T2K
neutrino beam above 1.5 GeV. Cutting above 1.5 GeV serves two purposes. The
first is that the current P0D reconstruction and cut methods produce the cleanest
165
True neutrino energy (GeV)0 1 2 3 4 5 6
Effi
cie
ncy
(%
)
0
5
10
15
20
25
30
35
νAll Reco E
>1.5 GeVνReco E
Reconstructed neutrino energy (GeV)0 1 2 3 4 5 6
Pur
ity (
%)
0
10
20
30
40
50
60
70
80
90
Figure 7.5: Efficiency (left) and purity (right) as a function of neutrino energy. The er-
ror bars shown are from MC statistics. Efficiency is shown as a function of true neutrino
energy (since unreconstructed events included in the denominator do not have a recon-
structed energy) both with and without the Eν > 1.5 GeV cut applied in reconstructed
neutrino energy.
sample of electron neutrino events in this energy range.
This is also the region where νe from kaon decay are most likely to be found.
Kaon decay was a major source of uncertainty in the 2010a oscillation analysis
because kaon data from NA61 had not yet been used to tune the beam Monte
Carlo.
Applying all of the previously described cuts in Monte Carlo gives Nsel =
114.9 ± 2.7 (stat) selected events. N selνe = 75.0 ± 2.2 (stat) of those events are
Signal (CC νe). The total number of predicted CC νe interactions in the P0D
fiducial volume with Eν > 1.5 GeV in MC is Nνe = 621 ± 8. From this the
efficiency, ǫ, and purity, p, is calculated as follows
ǫ =N sel
νe
Nνe
= (12.1± 0.4 (stat))% (7.1)
p =N sel
νe
Nsel
= (65± 2 (stat))%, (7.2)
where the error shown is statistical.
166
The efficiency and purity as functions of energy are shown in Figure 7.5. At
lower energies the efficiency is lower because the width cut eliminates many elec-
tron tracks that do not have sufficient energy to produce wide showers. This
reinforces the previous statement that the efficiency of this analysis is best for
events with energy greater than 1.5 GeV.
Event Type NEUT Reaction Codes Background (%) Signal (%)
CC NC
Elastic ±1,51,52 0.8 0.2 60.8
1 π± ±11,13,33,34 4.6 1.2 23.1
1 π0 ±12,31,32 2.7 8.7 5.0
Coherent π ±16,36 1.6 10.9 3.4
Multi π ±21,41 5.1 3.6 6.2
Other Resonant ±22,23,42,43,43,45 0.7 1.0 0.6
DIS ±26,46 30.0 27.0 0.8
Outside P0D n/a 0.6 1.4 0.0
Table 7.4: Signal and background breakdown by interaction type, as predicted
by the NEUT MC (See http://www.hep.lancs.ac.uk/nd280Doc/nd280mc/v4r42/-
classND280NeutKinematicsGenerator.html for a complete list of reaction codes.).
The background percentages are relative to the total predicted background, and the
signal percentages relative to the total predicted signal.
The dominant background after all cuts (75%) are events that contain a π0.
Table 7.4 shows the selected backgrounds broken down by interaction type, with
the main contributions coming from νµ-induced deep inelastic scattering events
and from νµ-induced NC coherent and resonant π0 production.
About 61% of the selected signal CC νe was found to be quasi-elastic. This is
taken as confirmation of the reasonable use of quasi-elastic kinematics in recon-
structing the neutrino energy. The comparison between true neutrino energy and
167
(GeV)νTrue E0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
(G
eV)
νR
econ
stru
cted
E
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Figure 7.6: Comparison between true and reconstructed neutrino energy.
-300 -250 -200 -150 -100
-100
-50
0
50
100
Top (X-Z)
-300 -250 -200 -150 -100
-100
-50
0
50
100
Side (Y-Z)
Figure 7.7: One of the 129 signal events (Run 4477, event 2453) selected from the
Run I and Run II data. See Figure 7.2 for details of the display.
the quasi-elastic approximation is shown in Fig. 7.6. Each column is normalized
to one, so the figure shows the smearing due to the reconstruction. All cuts have
been applied, except for the reconstructed neutrino energy cut, which is marked
in blue. Note that compared to the energy completeness from events with a single
3D track in Chapter 5, the tail at low reconstructed energy has been largely re-
moved by the pT cut removing events with secondary vertices far from the primary
vertex containing large amounts of energy. The red line marks Ereco = Etrue. The
last row and column are overflows.
168
Systematic shift R′ −Rnom
+ 1σ − 1σ
EM Scale −0.13 0.21
Cross Sections −0.13 0.13
Rate Normalization −0.10 0.11
Fiducial Volume −0.03 0.04
Muon Rejection −0.03 0.02
Monte Carlo Statistics −0.04 0.04
Proton Reconstruction −0.07 0.07
Total 0.26
Table 7.5: Contribution to the systematic uncertainty on R in the first analysis from
each source. R′ is the value of R calculated with the ±1σ systematic shift applied, and
Rnom is the nominal value of R. The method used to compute the total systematic shift
is described in the text.
169
19Accumulated pot / 100 1 2 3 4 5 6 7 8
Sel
ecte
d ev
ents
0
20
40
60
80
100
120
Figure 7.8: Cumulative number of selected events as a function of total p.o.t. collected.
The data distribution is shown in black, with the expected distribution for a constant
rate of events/p.o.t. in red.
(GeV)νReconstructed E1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
# of
Eve
nts
0
5
10
15
20
25
30
35
40
45 Signal0π w/ µ 0π no µ0π, µno
0π no µno Out of P0DData
Figure 7.9: Reconstructed neutrino energy distribution of events passing all cuts. All
MC scaling factors are applied, along with the flux weights. The last bin is an overflow.
In data, after all selection cuts are applied 129 events were selected. An
example of a selected event is shown in Fig. 7.7. Fig. 7.8 shows the event rate
compared to the number of protons on target in data. Using a Kolmogorov-
Smirnov test returns a distance of 0.05, for a p-value of 0.86, indicating that the
rate of events is constant as is expected. Finally, the general agreement between
170
Data and Monte Carlo within statistical errors is shown in Fig. 7.9.
We form a ratio, R, that should be one if data and Monte Carlo simulation
agree. R is defined as
R =D − B
S= 1.19± 0.15(stat)± 0.26(syst), (7.3)
where D is the number of data events, and S and B are the numbers of signal
and background events predicted by the Monte Carlo, respectively. Within the
statistical and systematic errors this result agrees with one. The total effects
of the previously described systematics on R are given in Table 7.5. The errors
are assumed to be uncorrelated and therefore summed in quadrature. With no
significant excess no evidence is found from this measurement of mis-modelling of
the kaon decay produced νe background.
7.2.2 Q2 Analysis
The purpose of the Q2 analysis is to use the improved beam systematics to make
an estimate of the CCQE-like component of the selected events as a function of
Q2. CCQE cross sections are often used to normalize other reactions, and are
therefore very important to understand. For muon neutrinos the CCQE cross
section is well measured but for electron neutrinos currently it is not.
The CCQE-like sample contains all events that have the same final state as
a CCQE event. From particle gun Monte Carlo it was shown that charged pions
with energy less than 50 MeV cannot be identified in the P0D. Therefore the
definition of CCQE-like is taken to be a CC νe event with no mesons except for
charged pions with energy less than 50 MeV, and no baryons besides protons and
neutrons.
Utilizing the updated flux tuning alters the composition of the signal and back-
ground reactions. Changing the definition of signal to be CCQE-like also increases
171
the background in categories like single charged pion, since νe interactions con-
taining a pion are no longer considered signal. The composition of the selected
events in Monte Carlo after all these changes is shown in Table 7.6.
Event Type NEUT Reaction Codes Background (%) Signal (%)
CC NC
Elastic ±1,51,52 1.9 0.1 76.3
1 π± ±11,13,33,34 17.1 0.9 17.5
1 π0 ±12,31,32 4.8 6.4 3.3
Coherent ±16,36 5.4 8.1 0.3
Multi π ±21,41 9.1 2.5 2.3
Other Resonant ±22,23,42,43,43,45 1.1 1.0 0.3
DIS ±26,46 21.9 18.7 0.0
Outside P0D NA 0.2 0.1 0.0
Table 7.6: Signal and background breakdown by interaction type for the second anal-
ysis. The conventions are the same as Table 7.4.
With this new definition of signal and all of the previously described cuts, as
well as the effects of using the updated flux tuning, Nsel = 130.7 ± 2.9 (stat)
selected events are found with N selνe = 66.4 ± 2.0 (stat) of those identified as
CCQE-like signal. The total number of selected events in data is unchanged by
the change in the flux tuning or the signal definition. Therefore, the new ratio
RQE is found to be
RQE =D − B
S= 0.97± 0.17(stat)± 0.40(syst), (7.4)
This ratio agrees with 1 within the uncertainties. The calculated value of the
systematic error on RQE from all sources is shown in Table 7.7. In Fig. 7.10
the Q2 distribution of signal and background events in data and Monte Carlo is
shown.
172
Systematic shift R′QE −RQEnom
+ 1σ − 1σ
EM Scale −0.14 0.25
Cross Sections −0.24 0.24
Rate Normalization −0.19 0.24
Fiducial Volume −0.04 0.04
Muon Rejection −0.04 0.02
Monte Carlo Statistics −0.04 0.04
Proton Reconstruction −0.09 0.09
Total 0.40
Table 7.7: Contribution to the systematic uncertainty on RQE in the second analysis
from each source. All conventions agree with Table 7.5.
)2 (GeV2Reconstructed Q0 0.2 0.4 0.6 0.8 1 1.2 1.4
# of
Eve
nts
0
5
10
15
20
25
30 SignalCCNuE0π w/ µ
0π no µ0π, µno
0π no µno Out of P0DData
Figure 7.10: Reconstructed neutrino distribution in Q2 of events passing all cuts.
Another distribution that is interesting to consider is the signal compared with
the background subtracted data as shown in Fig. 7.11. Previous analyses have
found discrepancies in this distribution for muon neutrinos [?]. This discrepancy
was attributed to possible mis-modeling in the Fermi Gas model. Visually a similar
discrepancy seems to exist, but within the current large errors no conclusions can
173
)2 (GeV2Reconstructed Q0 0.2 0.4 0.6 0.8 1 1.2 1.4
# of
Eve
nts
0
2
4
6
8
10
12
14
16
18
Signal
Data
Figure 7.11: Reconstructed distribution of CCQE-like events passing all cuts compared
to background subtracted data. The last bin is an overflow.
be drawn.