N. Saoulidou, Fermilab, Wilson Fellowship Seminar 170106

1
Minos Status Results
N. Saoulidou, Fermilab, Wilson Fellowship
Seminar 17-01-06
2
Outline

• Introduction
• Hardware Involvement Work
• MINOS QIE Electronics
• MINOS 9 Plane Prototype
• MINOS Near Detector Installation, Commisioning
  Maintenance
• Software Reconstruction
• Development of Time Slicing Techniques for the
  Near Detector.

3
Outline contd

• Data Analysis
• Near Detector Performance.
• Selection of first Far Detector events.
• Artificial Neural Network techniques for Charged
  Current Neutral Current event Selection in Near
  and Far detectors.
• Description of technique.
• Performance on Near Detector data.
• 3D MC fits for estimation of dm2, sin2theta, and
  sterile neutrino fraction.
• Comparison with exiting PDF selection method.
• Near detector detailed comparison of Data with MC
  ( Systematics uncertainties included)
• Automated Selection of Low Energy Far detector
  events.
• Near Far extrapolation fitting technique for
  obtaining oscillation measurement.

4
Outline contd

• Minos related future Research Interests
• Fermilab Neutrino Program related Research
  Interests

5
3-Flavor Oscillation Formalism

• If neutrinos oscillate, then the interaction
  eigenstates (what we observe) can be expressed in
  terms of the mass eigenstates as follows

Atmospheric
Solar
Cross Mixing
U
Majorana phases
0nbb decays
6
2-Flavor Neutrino Mixing

• In certain experimental situations only one q
  contributes, in which case one can write the
  oscillation probability as

Physics
Experiment

• Different neutrino experiments , depending on
  what components of the mixing matrix they want to
  measure involve
• Different baselines
• Different neutrino energies
• Different neutrino flavors
• When the region of parameter space (Dm2, sin2
  (2q)) is known then Dm2 determines the L/E
  ratio for which the oscillation phenomenon will
  be maximum and therefore easier to observe (in
  reverse, L/E determines the experiment
  sensitivity).

7
SuperK , Atmospheric neutrinos

• Study muon and electron neutrinos produced
  in the upper atmosphere.
• Observation fewer muon neutrinos than
  expected
• as many electron neutrinos as expected
• as many NC
  interactions as expected

Observed / Expected interactions
68,90,99 C.L.
sin2(2q23)gt0.92 0.0015ltDm 232lt0.0035
8
K2K the 1st Long-Baseline Accelerator-based
Experiment

• Goal was to confirm SK result with
  accelerator muon neutrinos

L250Km
107 Observed / 149.7 Expected
99 CL
90 CL
68 CL
0.89 x 1020 p.o.t.
Plots courtesy C. Walter
9
Minos Timeline

• 1998 Approved
• 1998 Project Baseline
• gtAugust 2003 (I Join MINOS)
• August 2004 Completion of ND installation
• December 2004 First Beam to hadron absorber
• January 2005 First vs at ND
• March 2005 First vs at FD
• April 2005 Target Leak
• May 2005 PME, PHE Running
• June 2005 Start nominal LE Running (
  Blind our FD data)
• December 2005 MINOS gets 1E20 POTs
• January 2006 Collaboration decides to open
  the box in February 2006
• First preliminary results for Neutrino06
• First MINOS publication to follow

10
MINOS Experiment

• MINOS (Main Injector Neutrino Oscillation Search)
  is a two detector long baseline neutrino
  oscillation experiment.
• Its goal is to study the region of parameter
  space indicated by atmospheric neutrino
  experiments and make precise measurement of the
  oscillation parameters Dm2 sin2(2q)

Near Detector
Far Detector
735 km
Comparison between Near/Far measurements will
establish the oscillation signal and
characteristics
11
MINOS Collaboration
MINOS Near Detector Surface Building
32 institutions 175 physicists
Argonne Athens Benedictine Brookhaven
Caltech Cambridge Campinas Fermilab
College de France Harvard IIT Indiana
ITEP-Moscow Lebedev LivermoreMinnesota-Twin
Cities Minnesota-Duluth Oxford Pittsburgh
Protvino Rutherford Sao Paulo South Carolina
Stanford Sussex Texas AM Texas-Austin
Tufts UCL Western Washington William Mary
Wisconsin
12
MINOS Physics Goals

• The study and comparison of nµ CC interactions
  between the NEAR and FAR detector will allow us
  to
• Confirm oscillation hypothesis with accelerator
  muon neutrinos (nm disappearing from the beam)
• - Obtain precise measurements of the oscillation
  parameters, (?m23 lt10) and sin22?23

13
MINOS Physics Goals

• What can we do in a few months where we will
  have 1 x 1020 POTs...
• Check that we are running with the right beam
  energy!!

MC
14
NuMI Neutrino Beam

207m

• 120 GeV protons strike the graphite target
• Initial intensity 1.5
  x 1013 ppp every 2-4 sec
• Current intensity 2.5 x
  1013 ppp every 2.2 sec
• Goal for 2006 is to run stably at 2.5 x
  1013 ppp every 2 sec.
• (2008-9) expected rate
  3.4 x 1020 protons/year

15

NuMI Target Horns

• Fully optimized spectra for each energy are
  obtained by moving the target and the 2nd horn
  (provision is made for three different 2nd horn
  positions).

16
NuMI Neutrino Beam configurations
pME beam
pHE beam
Running in the LE configuration we expected 1300
nm CC events for 2.5x1020/year in the 5.4kt FAR
detector (in the absence of oscillations).

• One can also obtain different neutrino spectra by
  just moving the target (fast, have taken data
  already for three different energy
  configurations).
• LE, pME and pHE data used to perform systematic
  studies in the Near Detector and tune our Monte
  Carlos (more about this later).

17
The MINOS Detectors
NEAR 0.98 kt
FAR 5.4kt

• Basic Idea Two detectors identical in all
  their important features.
• Both detectors are tracking calorimeters
  composed of interleaved planes of steel and
  scintillator
• 2.54 cm thick steel planes
• 1 cm thick 4.1 cm wide scintillator strips
• 1.5 T toroidal magnetic field.
• Multi-Anode Hamamatsu PMTs (M16 Far M64 Near)
  
• Energy resolution 55/?E for hadrons, 23/?E
  for electrons (measured with Calibration detector
  at Cern)
• Muon momentum resolution 6 from range (
  12 from curvature )

18
The MINOS Detectors contd
Identical but yet different
NEAR
FAR

• 5.4 kton magnetised tracking calorimeter, B 1.5T
• Multi-pixel (M16) PMTs read out with VA
  electronics
• 8-fold optical multiplexing
• chips individually digitised, sparsified read
  out when dynode above a threshold
• excellent time resolution 1.56ns timestamps
• Continuous untriggered readout of whole detector
• Interspersed light injection (LI) for calibration
  
• Software triggering in DAQ PCs (independent of
  ND)
• highly flexible plane, energy, LI triggers in
  use
• spill times from FNAL to FD trigger farm under
  dev.
• GPS time-stamping to synch FD data to ND/Beam

• 1 kton (total mass) magnetised tracking
  calorimeter
• Same basic design as Far Detector
• High instantaneous ? rate, 20ev/spill in LE
  beam
• No multiplexing except in spectrometer region
  (4x)
• Fast QIE electronics
• continuous digitisation on all channels during
  spill
• (19ns time-slicing). Mode enabled by spill
  signal.
• dynode triggered digitisation out of spill
  (cosmics)
• GPS time-stamping / Software triggering in DAQ
• all in spill hits written out by DAQ
• standard cosmics triggers out of spill

19
Neutrino Events per Spill
(rock muons contained neutrino events)
LE (3)
ME (5)
Number of spills
HE (7)
Number of reconstructed neutrino events per spill

• For the same beam intensity the number of
  neutrino events scales with neutrino energy
  (scaling factor as expected from MC).
• Very reassuring as far as detector performance
  is concerned.

20
Near Detector Timing
Detector activity
One Spill
Starting time of Events with respect to the Spill
Gate
usec

• 8 10 usec spill of 5-6 batches each of
  1.6 usec length.
• Events recorded within a 18 micro sec window.
• Many neutrino interactions per spill , time and
  space used to slice separate individual
  events (timing resolution 18.9 nsec)

21
Near Detector Electronics QIE chips
October December 2003
Response (ADC counts)
Qin

• QIE characteristics Large Dynamic Range
• Dead time-less
• Capable of
  coping with large Near Detector Rates
• Calibration Performance and Monitoring of
  QIE response important for Near Detector data
  quality.

22
Near Detector Electronics Define Pathologies
October December 2003

• Rms
• Definition of bad channels rms greater than
  maximum in at least 8 of the 37 different DAC
  values.
• Definition of suspicious channels rms greater
  than maximum in any of the 37 different DAC
  values.
• Linearity
• Definition of bad channels Chi-square of the
  fit (mean vs dacval ) gt 10.
• Entries
• Definition of bad channels Missing entries for
  each QIE calibration point (256 in total 64 for
  each CAPID).
• Calibration Points
• Definition of bad channels Missing some of
  the 37 different calibration points
• Mean vs Time
• Mean fluctuating more than 3RMS, which
  corresponds to changes of the mean of 3

23
Near Detector Electronics Example of
pathological channels
October December 2003
Mean instability
Large RMS

• All these pathological cases are flagged
  with developed code which is now part of MINOS
  Online Monitoring.
• The code developed is also used by
  Electronics Maintenance Experts (me along with B.
  Rebel and A. Marino) to check and monitor ND QIE
  response.

Calibration QIE problem
24
MINOS 9 Plane Prototype Goals
November 2003 February 2004

• Discover potential problems before actual Near
  Detector Installation.
• Exercise and time the whole procedure, in order
  to better organize things for the commissioning
  and installation phase.
• Provide detailed documentation for future
  shifters (no mine crew available this time!).

25
MINOS 9 Plane Prototype Integration Setup
November 2003 February 2004

• Last 4 Spectrometer planes

Minder Master Racks
Pulser Box Rack
Light Leak Checker
HV Rack
26
MINOS 9 Plane Prototype Timing Documentation
November 2003 February 2004

• We timed the plane instrumentation and light leak
  check to help in the determination of the final
  installation and commissioning schedule.
• Timed the whole procedure for one plane
• Connect Optical LI cables to the plane 130
  hour
• Connect Optical cables to Alner Boxes LI cables
  to the Pulser Box
  30 minutes
• Light Leak Check
  30 minutes
• Total 230 hours

This determined the aggressive (and finally
successful) schedule of Near Detector
Installation.

• The whole procedure was documented in detail and
  was available to the whole collaboration
  home.fnal.gov/niki/nd_readout.html so that the
  installation commissioning phase would proceed
  as smoothly and efficiently as possible .

27
March 2004 August 2004
MINOS ND installation Commisioning Detector
schematic
MINDER Crate
M64 PMT
connectors
Charge injection
Clear fiber cables
PMT-gt Minder Cables
Scintillator, green fiber
HV
LI
MinderAux,Minder, Menu Keeper, MTM
Detector
Electronics
Plot courtesy G. Rameika
Non-accessible After installation
Accessible After installation
28
MINOS ND installation Commisioning
Definition of Procedures
March 2004 August 2004

• Procedure for plane instrumentation
• Planes Cabled.
• Planes Light leak checked.
• Alner box end is light leak tested
• The LI end is light leak tested as well.
• At the end of the day we take
• Pedestals runs.
• QIE Calibration runs.
• Null trigger runs with lights ON and OFF.
• At the end of the week it would be very useful to
  take Null Trigger Runs with
• HV ON for several hours (PMT response
  stabilized).
• Either dynode threshold or sparcification
  threshold high in order to write out and study
  mostly muons.

29
MINOS ND installation Commisioning Example
of Pathologies found corrected
March 2004 August 2004
Strip Profiles
Strip Profiles
OK
PROBLEM with LI SYSTEM
NOT OK

• Identified Problems along with their fixes.
• Documented the problems procedures followed
  for MINOS ND shifters

Strip Profiles
BAD PMT
30
MINOS ND installation Commisioning Example
of Pathologies found corrected
March 2004 August 2004
PMT Response
OK
NOT OK

• Identified Problems along with their fixes.
• Documented the problems procedures followed
  for MINOS ND shifters

• A normal PMT should have
• MEAN (for singles) 100 ADC counts
• SIGMA 50 ADC counts
• RATE gtgt 1KHz (0.8-1.4KHz)

31
Summary of hardware involvement

• Developed code for the checking the performance
  of the ND QIE Electronics, used in Online
  Monitoring and by the ND electronics.
• Successfully competed MINOS 9 Plane prototype
  test (G. Rameika was one of the few along with me
  people pushing for it). Without this test the ND
  installation would not be as fast, smooth and
  successful.
• Heavily involved in the ND installation and
  Commisioning (one of the few people being there
  nearly all the time) by establishing,
  coordinating and overseeing daily installation
  procedures related with detector assembly and
  performance and by training new shifters (The
  installation and commisioning was very fast,
  smooth and successful)
• Hardware wise I am now one of the ND
  electronics, HV, and PMT maintenance experts.
  

32
First ND Event !

• January 20-21(software in place to go, show it
  in my laptop screen 15 minutes after it
  occurred )

33
June 2004 August 2004
Software Slicing Techniques for Near Detector

• Slicing of Near Detector Events in Time
  critical for Neutrino Event reconstruction.
• ND Slicing potentially sensitive to any MC
  Data differences (noise, detector effects e.t.c)
• Need more than one Slicing Method in order to
  be able to perform cross-checks and systematic
  studies
• Simple Clustering based on time.
• Clustering based on Graph Theory (time space)

34
Slicing Techniques for Near Detector
-Minimal Spanning Trees Basics-
Graph

• An edge weighted Linear Graph is composed of a
  set of points called nodes and a set of node
  pairs called edges with a number called weight
  assigned to each edge.

• A path in a graph is a sequence of edges
  joining two nodes. A circuit is a closed path.

Spanning Tree

• A spanning tree is a connected graph with no
  circuits which contains all nodes.

• A minimal spanning tree is the spanning tree
  whose weight ( sum of weights of its
  constituent edges) is minimum among all spanning
  trees in this set of nodes.

Minimal Spanning Tree
June 2004 August 2004
35
Slicing Techniques for Near Detector
- MST Theorem 1 -

• Theorem 1 If S denotes the nodes of G and C
  is a nonempty subset of S with the property that
  p (P,Q) lt p (C,S-C) for all partitions (P,Q) of
  C then the restriction of any MST to the nodes of
  C forms a connected subtree of the MST . The
  significance of this theorem for cluster
  detection can be illustrated if the following
  figure which depicts the MST for a point set
  containing two clusters C and S-C

• This theorem assures us that the subgraph of an
  MST does not break up the real clusters in S, but
  on the other hand neither does it force breaks
  where real gaps exist in the geometry of the
  point set.
• A spanning tree is forced by its very nature to
  span all the points but at least the MST jumps
  across the smaller gaps first.

June 2004 August 2004
36
Slicing Techniques for Near Detector
- MST Theorem 2-

• Theorem 2 If T is an MST for graph G and X,Y
  are two nodes of G, then the unique path in T
  from X to Y is a minimax path from X to Y.1

• Cost maximum edge weight of the path e.g the
  path (CBADE) has a cost of 8.
• Minimax path The path between a pair of nodes
  that has the least cost e.g there are four
  minimax paths from C to F all of cost 8.
• The minimax path each of whose subpaths are also
  minimax lies within the MST and that is not a
  coincidence as shown in the previous theorem.
• So the preference of minimax paths in the MST
  forces it to connect two nodes X and Y belonging
  to a tight cluster without straying outside the
  cluster.

June 2004 August 2004
37
Slicing Techniques for Near Detector
- MST properties-

• The MST is deterministic. It does not depend on
  random choices in the algorithm or on the order
  in which nodes and edges are selected and
  examined but only on the given set of nodes.
• The MST is invariant under similarity
  transformations, that is under all
  transformations that preserve the monotony of the
  metric (rotations, translations changes of the
  scale and even some nonlinear distortions) .
• The metric for the weight assignment can be
  defined in many ways and does not have to be the
  Euclidean Distance between 2 nodes .

June 2004 August 2004
38
Slicing Techniques for Near Detector
-MST Cluster Analysis-

• Main Idea After forming the MST of a set of
  points group the points into disjoint sets by
  joining all edges of weights Di or less. Each set
  is then said to form a cluster at level Di.Thus
  all segments joining two clusters defined at
  level Di will have lengths greater than Di.

June 2004 August 2004
39
MST in ND (Clustering in time space)

• I have used a hybrid metric for constructing the
  length of the MST branch connecting two strips
  
• time difference x c a x length difference (in
  z) and I using a semi-random initial cut that is
  going to be tuned.
• This method has just ONE parameter that needs to
  be tuned
• Ok I lied it has 3
• Minimum number of strips to form a slice
• Minimum pulse height to consider the strip
• Cut on the length of my TREE.

June 2004 August 2004
40
June 2004 August 2004
Slicing Techniques for Near Detector 1.

• MST gives same (or slightly better) results with
  respect to Event Completeness Event Purity.
• MST already implemented in Standard Minos offline
  code.
• MST slightly slower (speed) that existing
  slicing.
• MST potentially more powerful for higher event
  intensities (since it uses space along with
  time).
• It is going to be considered as becoming the
  standard in the major 2006 MINOS software
  upgrade.

41
June 2004 August 2004
Slicing Techniques for Near Detector 2.

• Developed second very very simple (and therefore
  easy to tune and debug) slicing method called
  ASAP (AsSimpleAsPossible).
• Method gives slightly better results with
  respect to event Completeness and Purity.
• Method already implemented and used in Standard
  Minos offline code.
• Standard Slicer failed quite seriously in the
  first ND data, so we used the ASAP for our first
  results which also helped in founding and fixing
  the bug in the standard code.

42
March 2005 August 2005
ND Data Analysis Detector Performance
Event Pulse Height (ADC counts)

• First analysis on March (on March data) showed a
  problem of an excess of Low Pulse Height events
  in the ND.
• That excess affected nearly all low and high
  level reconstructed quantities.

43
March 2005 August 2005
ND Data Analysis Detector Performance Low
Pulse Height Noise Source of Problem
Problematic Events Events (PHlt10000 ADC counts)
Healthy Events (PHgt10000 ADC counts)

• Source of problem was identified to be PMT
  response.

44
March 2005 August 2005
ND Data Analysis Detector
Performance Low Pulse Height Noise Solution of
Problem
True Event 1st CandStrips
Shadow Event 2nd CandStrips
ADC 1st 10491 2nd 114
ADC 1st 4489 2nd 81
ADC 1st 6827 2nd 52
ADC 1st 13031 2nd 300
ADC 1st 4779 2nd 61
Difference of first occurrence of strip from
subsequent ones
Frequency of multiple Strips DATA (LE-10) 10
MC (LE-10) 4 Frequency of multiple Strips
DATA (LE-10) 45 MC (LE-10) 6 in
events with Total PH lt10000 Frequency of
multiple Strips DATA (LE-10) 6 MC
(LE-10) 4 in events with Total PH gt10000

• Solution was time window cut after strip
  first seen in event (Work done together with
  Elizabeth Barnes, a Fermilab summer student
  during the Summer).

45
March 2005 August 2005
ND Data Analysis Detector
Performance Low PH Noise Removal Results
46
March 2005 August 2005
ND Data Analysis Detector Performance Low
PH Noise Removal Results (CC-like events)
NO LOW PH REMOVAL
LOW PH REMOVAL
47
Far Detector Data
May 2005
Typical events

• In the Far detector we record events that
  satisfy either of the following trigger
  conditions
• 4/5 consecutive planes
• OR
• Sum of ADC gt1500 or 6 hits in any 4
  consecutive plane window
• 0R
• Events within /-50 usec from a beam spill (beam
  data) (along with some random fake spill data
  in-between beam spills)
• Mostly we record cosmic ray muons _at_ a rate of
  0.5 Hz

48
First Far Detector Beam Data Analysis
May 2005

• I am going to show events from the High Energy
  running (Mean energy at 10 GeV).
• In this one week of data 150000 spills were
  recorded in the FAR detector and less than 100
  had detector activity (that survived basic cuts).
  
• Selected FAR detector CC-like event candidates
  based on timing (Far Spill Events) and requiring
  that they contain a track.
• Then visually scanned all candidates and
  categorized them and found
• Contained CC-like Events (21)
• Rock muons (9)
• Cosmics
  (6) (expect 7)

49
May 2005
First Far Detector Beam Data Analysis
COSMICS
Y angle (vert.)
Neutrino Candidates
X angle (horz.)
Track angle with respect to the Beam direction
Preliminary
Time difference (in sec) between neutrino
candidates and far spill signal in the /-50 usec
window. Beam neutrino candidates are within a 10
usec time interval, as expected for the 10usec
width of NuMI beam.
50
May 2005
First Far Detector Beam Data Analysis
Z vertex (m)
X-Y vertex (m)

• Timing and topological characteristics of beam
  neutrino event candidates in agreement with
  expectations, event rate in the right range.

51
Slicing, Low PH noise First Far Near
Detector Events

• Developed powerful slicing techniques that the
  Software Czar (J.Musser) likes and wants to
  consider in the next major MINOS software
  release.
• First close and detailed look at Near Detector
  data were I found and solved a major
  detector-hardware related problem (Low pulse
  height noise from PMT afterpulsing). Without this
  fix in place no NC analysis in the ND would be
  possible and high level CC related quantities
  were affected as well.
• First systematic and detailed analysis on Far
  Detector data that established the Far detector
  beam event selection and validation procedure.

52
Selection of CC-NC neutrino eventsMethod
Artificial Neural Networks
September 2004 - December 2004

• ANN can be trained by MC generated events
• A trained ANN provides multidimensional cuts
  for data that are difficult to deduce in the
  usual manner from 1-d or 2-d histogram plots.
• ANN has been used in HEP
• HEP Packages
• JETNET
• SNNS
• MLP fit

53
-ANN BASICS-
September 2004 - December 2004
X

• Event sample characterized by two variables X
  and Y (left figure)
• A linear combination of cuts can separate
  signal from background (right fig.)
• Define step function
• Separate signal from background with the
  following function

Signal (x, y) OUT
Signal (x, y) IN
54
-ANN BASICS-
September 2004 - December 2004
Visualization of function C(x,y)

• The diagram resembles a feed forward neural
  network with two input neurons, three neurons in
  the first hidden layer and one output neuron.
• Threshold produces the desired offset.
• Constants ai, bi are the weights wi,j (i and j
  are the neuron indices).

Y
X
Thres.
b1
a3
b3
a1
a2
b2
1
1
1
-2
c2
c1
c3
Output
55
September 2004 - December 2004
-ANN basics Schematic-
HIDDEN LAYER
Biological Neuron
INPUT LAYER
X1
WEIGHTS
. . .
OUTPUT LAYER
neuron k
. . .
Bayesian Probability
wik
. . .
Xi
wkj
neuron i
neuron j
INPUT PARAMETERS
Bias
56
-ANN BASICS-
September 2004 - December 2004

• Output of tj each neuron in the first hidden
  layer
• Transfer function is the sigmoid function
• For the standard backpropagation training
  procedure of neural networks, the derivative of
  the neuron transfer functions must exist in order
  to be able to minimize the network error (cost)
  function E.
• Theorem 1 Any continuous function of any number
  of variables on a compact set can be approximated
  to any accuracy by a linear combination of
  sigmoids
• Theorem 2 Trained with desired output 1 for
  signal and 0 for background the neural network
  function (output function tj) approximates the
  Bayesian Probability of an event being a signal.

57
September 2004 - December 2004
-ANN Probability (review)-
ANN analysis Minimization of an Error
(Cost) Function
The ANN output is the Bayes a posteriori
probability in the proof no special assumption
has been made on the a priori P(S) and P(B)
probabilities (absolute normalization)..TRUE BUT
THEIR VALUES DO MATTER (They should be what
nature gave us)
58
September 2004 - December 2004
-ANN probability (review)-

• Bayesian a posteriori probability
• ANN output P(S/x)
• ANN training examples P(x/S) P(x/B)
• ANN number of Signal Training Examples P(S)
• ANN number of Background Training Examples
  P(B)
• The MLP (ann) analysis
• and the Maximum Likelihood
• Method ( Bayes Classifier )
• are equivalent.
• (c11 c22 cost for making the
• correct decision
• c12 c21 cost for making the
• wrong decision )

59
- CC-like Selection ANN Variables MC -
CC- NC Events normalized to the same area

• Thirteen variables to construct ANN based
  Particle Identification Function.
• The ANN is trained for all short (lengthlt40
  planes) events with or without a track. Event
  with length gt 40 planes are pre-classified as
  CC-like.
• Here the ANN is applied for events that have a
  track with trkfitpass1.

September 2004 - December 2004
60
- CC-like Selection ANN Variables MC -
CC- NC Events normalized to the same area

• Thirteen variables to construct ANN based
  Particle Identification Function.
• The ANN is trained for all events with or without
  a track. Here is it applied for events that have
  a track with trkfitpass1.

September 2004 - December 2004
61
- CC-like Selection Importance of ANN
Variables MC -
Relative weight () ANN Variable
10.565750 Total Pulse Height 10.446102
Total of Strips 9.2708178 Event Length
8.7206430 Number of Tracks 8.6607571 Track
Pulse Height per Plane 8.5564222 Pulse
height per Plane 8.1546698 Shower
Pulse Height per Digit 7.4450355 Pulse
height per Strip 6.6567850 Difference of
Track-Shower Length (V view) 6.4418235
Pulse height Fraction in first 3 planes
5.7088947 Pulse height Fraction in planes 3-6
5.1340508 Difference of Track-Shower Length
(V view) 4.2382522 Pulse height Fraction in
planes 6-last.

• All ANN variables are important.

September 2004 - December 2004
62
ANN Results NEAR - FAR (lengthlt40)
September 2004 - December 2004
NEAR (RED CC, BLACK NC)
FAR (RED CC, BLACK NC)
ANN Probability
ANN Probability

• The ANN shows good discrimination power between
  CC and NC.
• The ANN performs better for the Far detector than
  the Near (no event overlapping).

63
ANN Results NEAR cont
September 2004 - December 2004
NEAR CC selection
FAR CC selection
Efficiency (red) and purity (magenta) as a
function of cut in the ANN output function for
the signal (CC events)

• If we set the cut _at_ 0.25 (i.e ) we have an
  efficiency of 92 and a purity of 53.

64
ANN Results NEAR cont
September 2004 - December 2004
NEAR NC selection
FAR NC selection
Efficiency (red) and purity (magenta) as a
function of cut in the ANN output function for
the signal (CC events)

• If we set the cut _at_ 0.25 (i.e ) we have an
  efficiency of 85 and a purity of 95.

65
CC event classified as NC-like 1
September 2004 - December 2004

• Blue squares slice hits
• Red triangles track hits
• Magenta triangles shower hits

TRUTH DISPLAY Etrue 4.5 Emu 0.7 Y0.8
No obvious reco track, resembling NC
66
CC event classified as NC-like 2
September 2004 - December 2004

• Blue squares slice hits
• Red triangles track hits
• Magenta triangles shower hits

TRUTH DISPLAY Etrue 2.1 Emu 0.3 Y0.9
No obvious and reco rack, resembling NC
67
ANN Miss-classifications
September 2004 - December 2004
FAR NC like spectrum (blue NC contamination)
FAR CC like spectrum (blue NC contamination)
Miss-classifications as a function of Y

• Miss-classifications come from high Y CC
  events were the muon is a short track with
  respect to the shower length.

68
- CC-like Selection ANN Variables Data/MC -
Data and MC Events normalized to the same area

• The shapes of the distributions of the ANN
  variables between data and MC agree quite well.

October 2005 - Present
69
- CC-like Selection ANN Variables Data/MC -
Data and MC Events normalized to the same area

• Pulse height fractions indicate that there is
  more EM fraction in the data showers than the MC.
• Most of the variables used in the ANN
  (especially variables related with pulse height
  that is lower in the data than the MC) would
  result in the events being slightly more CC-like.

October 2005 - Present
70
- CC-like Selection ANN Function Data/MC -
Data MC normalized to the same area

• ANN PID agrees well for CC-Like events and shows
  a discrepancy (in shape) between data and MC on
  the region populated by NC-like events.

October 2005 - Present
71
- CC-like Selection ANN Function Data/MC -
Data MC normalized to the same area
Events with lengthgt12 planes OR (lengthlt12 AND
Eshwlt0.5GeV)
Events with lengthlt12 AND Eshwgt0.5GeV

• This discrepancy seems to be coming from short
  events with relatively large showers (the events
  tend to be characterized as more CC-like)
• The cut _at_ 0.2 selects events above the region
  where the discrepancy is present

October 2005 - Present
72
- CC-like Selection PDF selection (D.Petyt)
Variables MC -
CC- NC Events normalized to the same area

• Use three variables to construct PDF based
  Particle Identification Function for
  reconstructed events with a track and with
  trkfitpass1.

73
- CC-like Selection PDF Function MC -
Efficiency Purity vs PDF Cut
PDF PID for CC NC MC events
CC- NC Events normalized to MC priors

• The PDF PID function shows good discriminating
  power.
• Cut _at_ -0.2 Efficiency 85 (65) Purity 96
  
• Efficiency for selecting CC events from total
  sample (including the ones that do not
• have a reconstructed track with trkfitpass1
  is 64)

74
ANN vs PDF Selection CC Far fits
Courtesy Plot D. Petyt
December 2004 April 2005

• ANN selection gives better results for all
  different dm2s when no NC subtraction is used in
  the fit.
• ANN selection gives better results for dm2s
  greater than 0.002 when NC subtraction is used
  in the fit.
• The ANN method is going to be used along with the
  PDF for oscillation fits.

FAR CC like spectrum (blue ANN, Red PDF)
75
ANN vs Cut based Selection NC Far fits
dm2
dm2
fsterile
sin2theta
Delta Chi-square
dm2
sin2theta
fsterile

• dm20.0022 sin2theta230.88 fsterile0.1
• ANN selection gives best results for ND 3D fits
  as well, due to higher efficiency and purity.

December 2004 April 2005
76
October 2005 Present
ND Data Analysis Detector Performance
Data/MC comparisons

• Studies performed and presented (event selection
  is going to be shown later) on
• Comparisons of Lower Level Quantities (LLQ)
• Comparisons of Higher Level Quantities (HLQ)
• Systematic uncertainties
• En , Pm, Eshw, Y with different selections (PDF,
  ANN, cut based)
• Comparisons of corrected En , Pm, Eshw, Y for
  different selections
• Responsible (author) for ND Performance and
  Data/MC comparisons position paper, that
  along with
• Far detector Data position paper
• Systematic position paper
• Beam, Calibration Cross Section position papers
• Flux position paper
• are the key analyses for the MINOS first
  1E20 preliminary oscillation measurement.

77
October 2005 Present

• Comparisons of Low Level Quantities (LLQ) -

Purpose of the Analysis The goal of this
analysis is to examine (cc-like fiducial
region selected with ANN discussed later)
If there differences, and if so what is their
magnitude, in low level quantities that are
mostly related with possible detector effects and
modeling of the magnetic field.
78
October 2005 Present

• Comparisons of (LLQ) Event Related ones 1. -

• Vertex distributions agree very well between data
  and MC.
• ltPulse Height Per Planegt agrees very well
  between data and MC.

79
October 2005 Present

• Comparisons of (LLQ) Track related ones 3. -

• Muon momentum as a function of Phi angle (angle
  with respect to X axis in the XY plane) agrees
  well for momentum from range.
• Muon momentum as a function of Phi angle (angle
  with respect to X axis in the XY plane) shows a
  5 discrepancy for momentum from curvature
  around 180 degrees).
• These events (momentum taken from curvature with
  emission angle gt 160 degrees) is 6 of the total
  sample.

80
October 2005 Present

• Comparisons of (LLQ) Track related ones 1. -

• Track end Y positions agree well between data and
  MC.
• Track end X positions agree better than before
  (bug in DetSim related with coil hole modeled in
  different places fixed) but there is still a
  discrepancy for tracks approaching the coil
  hole.
• These discrepancy can receive contributions form
  a) either truly different muon momenta
  distributions and/or b) remaining modeling issues
  related with modeling of the magnetic field in
  the coil hole,

81
October 2005 Present

• Comparisons of (LLQ) Track related ones 2. -

• Track emission angles in X, Y and Z direction
  agree well between data and MC

82
October 2005 Present

• Comparisons of (LLQ) Track related ones 4. -

SA Fitter
SR Fitter

• Muon momentum from curvature compared to momentum
  from range agree within 2 between data and MC
  for the SR fitter. (Behavior of SR fitter between
  data and MC similar.)
• Muon momentum from curvature compared to momentum
  from range for data only agrees well using the SA
  fitter (S. Avvakumov). (Accurate modeling of the
  magnetic field).

83
October 2005 Present

• Comparisons of (LLQ) Track related ones 5. -

• Slopes of the muon momentum as a function of X, Y
  and Z position agree within errors between data
  and MC.

84
October 2005 Present

• Comparisons of (LLQ) Shower related ones 1.
  -

• Behavior of shower energy as a function of X and
  Y position is the same for data and MC.
• Slope of shower energy vs Z position in MC higher
  than in data (1 sigma effect) due to larger
  showers in the MC than the data.

85
October 2005 Present

• Comparisons of High Level Quantities (HLQ) -

Purpose of the Analysis The goal of this
analysis is to Determine level of
agreement between data and MC for higher level
quantities, En, Pm, Eshw, Y. If systematic
uncertainties from calibration, cross section and
beam modeling can account for the observed
differences.
86

• Systematic Uncertainties -

Flux
p
to far Detector
(stiff)
target
qf
p
qn
(soft)
Decay Pipe
ND

• Spread due to models
• 8 (peak)
• 15 (tail)

Slide Courtesy S. Kopp
87

• Systematic Uncertainties -

Cross Sections
MINOS ND CC SPECTRUM (up to 6 GeV) 19 QE
25 RES 56 DIS
88
October 2005 Present

• Comparisons of (HLQ) Systematic uncertainties
  -

• Value of muon momentum uncertainty based on
  comparisons of momentum from range and curvature
  between data and MC.
• Shower energy uncertainty based on 6 absolute
  calibration estimated by the Calibration Group
  and an additional 8 (conservative guesstimate)
  to account for intranuclear scattering effects.
• Cross section uncertainties based on
  recommendation from ND group (cross section
  experts)
• Beam uncertainties based on recommendation from
  Beam Group.

89
October 2005 Present

• Systematic Uncertainties Effect on Near Spectra
  -

Ratios of Weighted to Nominal Values
Cross sections QE fraction
Cross sections RES fraction
Cross sections DIS fraction
Flux (red) all combined (black)
Muon momentum -2
Shower Energy-10

• Most dominant is Flux (hadron production)
  uncertainty.
• NOTE Different scale in plots

90

• Comparisons of (HLQ) ANN selection (cut _at_ 0.2)

Normalization to the same area

• Data and MC agree (within 1.8 sigma for muon
  momentum) assuming the previously discussed
  systematic uncertainties.

October 2005 Present
91

• Comparisons of (HLQ) ANN selection (cut _at_
  0.2)Ratios

Normalization to the same area

• Data and MC agree (within 1.8 sigma for muon
  momentum) assuming the previously discussed
  systematic uncertainties.

October 2005 Present
92

• Comparisons of (HLQ) ANN selection varying the
  cut

Normalization to the same area

• Varying the ANN cut, the corrected for
  efficiency and purity, data distributions agree
  with each other within errors and with the MC
  within 1.8 sigma (first bin of muon momentum).

October 2005 Present
93

• Comparisons of (HLQ) ANN selection (cut _at_ 0.2)
  PME

Normalization to the same area

• Data/MC differences can be largely accommodated
  assuming previous systematic uncertainties.

October 2005 Present
94

• Comparisons of (HLQ) ANN selection (cut _at_ 0.2)
  PMERatios

Normalization to the same area

• Data and MC agree within 1.3 sigma given
  current knowledge of systematic uncertainties.

October 2005 Present
95

• Comparisons of (HLQ) ANN selection (cut _at_ 0.2)
  PHE

Normalization to the same area

• Data/MC differences can be largely accommodated
  assuming previous systematic uncertainties.

October 2005 Present
96

• Comparisons of (HLQ) ANN selection (cut _at_ 0.2)
  PHERatios

Normalization to the same area

• Data and MC agree within 1.3 sigma given
  current knowledge of systematic uncertainties.

October 2005 Present
97
Summary of ND data analysis

• The Near Detector is perfrorming as expected. No
  detector pathologies observed.
• Near Detector differences in high level
  quantities between data and MC accommodated by
  systematic uncertainties
• This of course is work in progress

98
June 2005 - Present
Low Energy Far Detector Beam Data Analysis
Neutrino Event Selection
Developed automated method, using topological
cuts to select Far Detector Beam events.
EFFICIENCY (Monte Carlo Events) 91 of
neutrino interactions (89 of CC and 80 of
NC) BACKGROUND REJECTION (Real fake spills)
From 921 events in FAKE spills (identical
triggering and reconstruction as real spills but
without neutrinos) 7 survived my cuts. (
0.8-0.3) This method has the highest
selection efficiency among all automated
selection methods.
99
June 2005 - Present
Low Energy Far Detector Beam Data Analysis
Neutrino Event Selection contd
Analyzed all Far data June 2005 December 7th
2005 Selected 335 (75) neutrino events (rock
muons) using visual scanning. 311 out of 335
neutrino events pass my automated topological
cuts Expected efficiency 91.0
Measured efficiency
92.5 Selected 351 events with automated
selection out of which 311 are neutrinos
Expected background
26-9.8 events Observed (identified by visual
scanning) background 40
100
Low Energy Far Detector Beam Data Analysis
Neutrino Events / POT vs Time
June 2005 - Present
Events/ X POT
Month
Number of recorded (and proccesed) neutrino
events per POT as a function of month is flat (no
pathologies-problems in recorded event rate).
101
June 2005 Present
Low Energy Far Detector Beam Data Analysis
COSMICS
Y angle (vert.)
Neutrino Candidates
X angle (horz.)
Track angle with respect to the Beam direction
Time difference (in sec) between neutrino
candidates and far spill signal in the /-50 usec
window. Beam neutrino candidates are within a 10
usec time interval, as expected for the 10usec
width of NuMI beam.
102
June 2005 Present
Low Energy Far Detector Beam Data Analysis
cont
X Angle (degrees)
Mean Track Y angle at 87 degrees (beam is
pointing up!) Mean Track X angle at 90
degrees
103
June 2005 Present
Low Energy Far Detector Beam Data Analysis
cont

• Timing and topological characteristics of beam
  neutrino event candidates in agreement with
  expectations.

104
Low Energy Far Detector Beam Neutrino Events
105
Low Energy Far Detector Beam Neutrino Events
106
Far oscillation Fitting techniques
October 2005 Present

• There are two approaches
• Fit simultaneously the near and far spectra with
  aim to determine cross section beam parameters
  that will be used to tune the MC in order to
  obtain predicted far un-oscillated spectrum. lt
  Method more sensitive to modeling of the Flux and
  the Cross sections. Fits the ND data with the ND
  MC and the FD data with the FD MC, which means
  smaller probability of extrapolating ND
  detector effect in the Far.
• Use Near Detector Data spectrum to extrapolate in
  the Far Detector and obtained far predicted
  un-oscillated spectrum. lt Method quite robust
  to modeling of the Flux and the Cross sections.
  Uses ND data spectrum to extrapolate so which
  means higher probability of extrapolating ND
  detector effect in the Far.
• Both are going to be used and results are going
  to be compared.

107
Brief Description of the Extrapolation Method

• The basic steps are the following
• NEAR RECO SPECTRUM CC-like gt NEAR TRUE SPECTRUM
  CC
• For that step we
• a) Correct Near reco CC-like spectrum for
  purity (Bkg subtraction)
• b) Convert corrected Near reco CC selected
  spectrum to Near CC selected true spectrum
  using the MC NxN Reco gt True matrix.
• c) Correct the Near true CC selected spectrum
  for efficiency to obtain Near true CC spectrum

October 2005 Present
108
Brief Description of the Method cont

• The basic steps are the following
• (2) NEAR TRUE CC SPECTRUM gt FAR TRUE CC
  SPECTRUM
• For that step
• We use the BEAM (GNUMI) NxN Matrix that converts
  Near true Neutrino Spectrum to Far True Neutrino
  Spectrum.
• (3) FAR TRUE CC SPECTRUM gt FAR RECO
  CC-selected SPECTRUM
• For that step
• a) Convert FAR TRUE CC spectrum to FAR RECO CC
  selected spectrum using the MC Far NxN True
  gt Reco matrix and the selection efficiency.

October 2005 Present
109
October 2005 Present
Consistency Check STEP 1

• Used a FAKE Data sample which consists of
  selected CC-like events in the Near Detector.

Black points Obtained CC spectrum Black
histogram Nominal CC spectrum

• Near Nominal CC Spectrum and Near
  Obtained CC spectrum doing all the steps (a),
  (b) and (c) of Step 1 agree perfectly with each
  other.

110
October 2005 Present
Beam Matrix
Beam matrix (GNUMI) that coverts Near True to
Far True spectrum
x 10-6

• Beam matrix basically reflects pion 2 body decay
  kinematics and gives the probability of a
  neutrino from a certain pion decay (of energy E
  and with decay angle theta) to reach the ND and
  the FD.
• Beam matrix largely invariant in different
  beam energies.

111
October 2005 Present
Consistency Check STEP 2

• We have used the LE-10 Beam Matrix and the
  GMINOS Near True Spectrum of step 1 to obtain the
  Far Extrapolated true Spectrum and compare it
  with the Far GMINOS one (Nominal MC)

Black points Obtained CC spectrum Black
histogram Nominal CC spectrum

• The agreement is very close and small
  differences are coming from statistical
  fluctuations since we are not using the exact
  same events.

112
October 2005 Present
Using Different Beam Parameters

• To examine differences, along with their
  magnitude, that could be introduced in the
  Obtained Far spectrum by different beam
  parameters, we use instead of the LE-10 Beam
  Matrix the LE-10 Beam Matrix with different horn
  current (8-10 change in true spectrum)and
  compare the results.

• There is no significant differences observed
  between the two Obtained Far True CC spectra.

113
October 2005 Present
Varying the x-sections ma_qe,ma_res -5
dis_factors -20

• In order to study the effect (systematic
  uncertainties) introduced by the x-sections
  used, we varied (for both Near and Far MC) the
  x-section and repeated the whole procedure (steps
  1 2 and 3)

• There are no significant differences observed
  between the two Obtained Far True CC spectra.
  Cross sections cancel out between two detectors.

114
October 2005 Present
Relative Energy Calibration of 7

• In order to study the effect (systematic
  uncertainties) introduced by a relative
  calibration of 7, we introduced this shift in
  the ND MC only and repeated the whole procedure
  (steps 1 2 and 3)

• 7 Relative energy calibration is OVER
  conservative given that the current relative
  calibration is 4.
• There are differences introduced (as expected)
  which are of the order of 5 and smaller than the
  statistical errors (these statistics correspond
  to 3.4 E20 POTs).

115
October 2005 Present
Relative Muon Momentum of 5

• In order to study the effect (systematic
  uncertainties) introduced by a relative muon
  momentum difference of 5

• There are differences introduced (as expected)
  which are of the order of 5 and smaller than the
  statistical errors (these statistics correspond
  to 3.4 E20 POTs).

116
October 2005 Present
Example MC fits with the previous uncertainties
1E 20 POTs 7 Relative shower calibration vs
Nominal
1E 20 POTs Different Beam matrix vs Nominal

• The true oscillation values are obtained in
  both cases.
• The contours are slightly bigger than the
  nominal (perfect knowledge) value but not
  significantly for the measurement we want to
  obtain now (1E20 POTs)

117
Summary of FD data analysis

• Developed Automated method with 91 efficiency
  for selecting far detector beam events
• ANN CC event selection (discussed previously)
  gives better results (smaller errors) and is
  going to be used for final oscillation fits.
• Proposed method for Near- Far extrapolation
  invariant in cross section (and to a large extend
  beam) uncertainties. Method is going to be used
  for 1E20 (and subsequent ) oscillation fit.

118
MINOS Future Research Plans

• Conclude extrapolation method studies (study
  fit biases e.t.c) in order to perform final fit
  for the 1E20 MINOS result.
• Improve speed of MST slicing and do final
  studies (related with CC and NC slicing and
  reconstruction efficiencies) in order to be
  adopted in the 2006 MINOS major software update.
• Continue studies on sterile fitting and
  sensitivities using the ANN selection in order to
  be ready for the NC opening of the box.
• Start working on Beam systematics which seems
  to be the dominant source of uncertainty for the
  MINOS experiment and also needs to be better
  understood for future Fermilab neutrino
  experiments that use the NUMI beam.

119
Future neutrino related research plans

• Main Goal Get actively involved in the study
  for future Fermilab Neutrino Initiatives (very
  long baseline neutrino beam to Homestake, Liquid
  Argon electron neutrino detectors e.t.c) . I view
  this as something that is extremely interesting
  to me (I am a quite passionate neutrino
  physicist) and also vital for the future of the
  Lab.
• NoVa Interested in working on
  simulation and analysis (using MC) related with
  possible improvement on the efficiency for
  electron neutrino event selection in the NoVa
  detector
• Miverva Interested,given the expertise gained
  from the MINOS ND installation, to contribute in
  the installation phase (hardware project) and
  also maybe participating in QE data analysis (as
  a data related project).

120
(No Transcript)
121
- CC-like Selection PDF 3. Variables Data/MC -

• Data have higher track pulse height fraction
  (difference in shape as well), lower track pulse
  height per plane (by 2) and an excess on smaller
  track lengths. We expect these difference to
  also be present in the PDF PID distribution.
• The first two distributions would tend to make
  the events more CC-like and the third one, more
  NC-like. Since event length is the strongest
  variable we expect to see an excess in NC-like
  events.

Data and MC Events normalized to the same area
122
- CC-like Selection PDF 4a. Function Data/MC
-
Data MC normalized to the same area

• PDF PID agrees well for CC-Like events and shows
  a discrepancy (in shape) between data and MC on
  the region populated by NC-like events.

123
- CC-like Selection PDF 4b. Function Data/MC
-
Data MC normalized to the same area
Events with lengthgt12 planes OR (lengthlt12 AND
Eshwlt0.5GeV)
Events with lengthlt12 AND Eshwgt0.5GeV

• This discrepancy seems to be coming from short
  events with relatively large showers.
• That indicates that although track
  characteristics are modeled properly, shower
  characteristics are different between data and
  MC.
• The cut _at_ -0.2 selects events above the region
  where the discrepancy is present (next we will
  show what regions of reconstructed neutrino
  energy these events populate)

124
Mock Results (no significant changes than initial
results)
Black ANN, Red Tom-Tobi
125
October 2005 Present

• Studies on Event Overlapping in the ND -

Purpose of the Analysis The goal of this
analysis is to examine a) The effect of time
slicing in event reconstruction. For that we
reconstructed the same Data and MC sample with
the SR and the ASAP slicers and compared
results on event by event basis. b) The
effect of beam intensity in event reconstruction.
For that we used JUNE 2005 data sample were data
with different beam intensities were recorded,
and examined various quantities as a function of
POTs.
October 2005 Present
126
October 2005 Present

• Different Slicing methods used Matching
  Criteria -

DATA
MC

• Data 95 of CC-like events are found in both
  samples.
• MC 93 of CC-like events are found in both
  samples
• 75 of the unmatched events (5) are the same
  with either timing or vertexing pathologies.

127
October 2005 Present

• Different Slicing methods used Ratios of En,
  Pm, Eshw -

DATA ASAP/SR
MC ASAP/SR

• Ratios of reconstructed muon momentum, shower
  energy and neutrino energy are centered to unity
  and are flat to within 1 for both Data and MC.
• Different slicing techniques do not introduce
  differences in event reconstruction.

128
October 2005 Present

• Reconstructed Event Quantities vs beam
  intensity 1. -

Total Event length and Total pulse height remain
flat as a function of beam intensity . (Event
length shows a 1.3 sigma effect, having
non- zero slopes)
129
October 2005 Present

• Reconstructed Event Quantities vs beam
  intensity 2. -

En, Pm, Eshw and Y remain flat as a function of
beam intensity. Muon momenta and shower
energies show an 1.3 sigma effect, h