MINOS Batch Processing, Monte Carlo Generation and Oscillation Appearance Analysis

1
MINOS Batch Processing, Monte Carlo Generation
and Oscillation Appearance
Analysis

• Alexandre Sousa
• Tufts University
• Tufts HEP Department Of Energy Review
• 10/26/2005

2
MINOS Batch Processing

• Work started in 2002 with J. Urheim (IU), H.
  Rubin (IIT) and AS (Tufts)
• Objective is to develop an infrastructure for a
  centralized, standardized MINOS Production Data
  Processing
• Runs on data files as they become available with
  minimal human intervention
• In parallel, handles other processing requests,
  Alignment, CalDet, MonteCarlo reconstruction,
  Mock Data challenge reconstruction
• Uses the Fermilab fixed target computer farm for
  data processing and Fermilabs ENSTORE tape
  archive for storage
• Since March 2005, AS responsibilities gradually
  transferred to Harvard group. AS continues to
  contribute in a consultant role
• The Tufts HEP Group main contributions included
• Batch Processing infrastructure design and
  original implementation
• MINOS Software installation, maintenance and
  testing at the farm
• Batch Farms MINOS database installation,
  configuration and administration
• Assembling of all the reconstruction scripts used
  in data processing
• Creation and maintenance of a Production
  software package making these scripts available
  to the collaboration at large.

3
MINOS Batch Processing
4
MINOS Batch Processing
5
MINOS Batch Processing
6
MINOS Batch Processing
7
MINOS MC Gen. (MDC)

• Mock Data Challenge
• Issued by the MINOS spokesmen in January 04
• Generate MC and Mock Data (MC with hidden truth
  information) samples
• Test the overall MINOS software framework and
  data handling
• Test the Near Detector and Far Detector
  reconstruction chains
• Test readiness of analysis groups for real beam
  data in early 2005
• In conjunction with its necessary batch
  processing reconstruction, was fundamental in
  identifying software bugs, reconstruction
  deficiencies and analysis shortcomings
• Challenge set
• Truth information removed
• FarDet 3yr at 2,4e13 pot/spill
• NearDet 25 x FarDet statistics in target region
• MC set
• Truth information retained
• FarDet 10 x challenge set statistics (100k
  events)
• Auxiliary flavor changed ne and nm files
• NearDet Same statistics as challenge set.

8
MINOS Monte Carlo Generation

• Originally motivated by the Mock Data Challenge
  (sample gen. would take 2 months)
• Configured and utilized the Tufts Linux Research
  Cluster for Monte Carlo file production in April
  2004. MDC sample generated in 9 days
• In January 2005, contributed to the successful
  inclusion of the College of William Mary
  cluster in the Off-Site MC Generation effort
• Participated in the development and testing of
  standard Monte Carlo generation scripts to be
  used by new contributing institutions (Rutherford
  Appleton Lab cluster coming up to speed)
• Generated Sets
• 75 of the total Mock Data Challenge sample
• 50 of the existing Near Detector Low Energy
  sample
• 100 of the Near Detector pseudo Medium Energy
  sample
• 100 of the Near Detector pseudo High Energy
  sample

9
Appearance Analysis

• One of the most relevant MINOS goals is to search
  for sub-dominant nm-gtne oscillations
• MINOS can potentially improve the limit on q13
  set by CHOOZ
• The Tufts HEP Group is a very active participant
  on the MINOS ne Appearance Analysis Group
• Developed an alternate shower reconstruction
  chain, based in the clustering and fitting of 3D
  Hits assembled from the strip information for
  each event
• Built a common analysis framework, NueAna, now
  part of the MINOS software, in collaboration with
  Harvard, UCL and Stanford MINOS groups
• Applied a Multivariate Discriminant Analysis
  (MDA) method to ne CC classification of MINOS
  events and completed the Mock Data Challenge.

10
Angular Clustering Algorithm

• Use seedless nearest-neighbor method to cluster
  hits in spherical coordinate event
  representation
• Calculate distance of each hit to all other hits
  omitting the reciprocal distances
• Aggregate all hits within some radius of a given
  hit discarding already used distances as we go
  (Caveat Hits assigned multiple times to
  overlapping aggregates)
• Histogram centroids of all aggregates and
  compute bounds of each high density region via a
  recursive algorithm
• From all the aggregates for which the centroid
  falls within these bounds, select the one with
  the most hits (cluster)
• Tunable parameters Radius, Noise.

11
Angular Clustering example (QE CC nue )
12
Angular Clustering example (QE CC nue )
Transverse Shower fitting
Longitudinal Shower fitting
13
NueAna Framework

• Originated as the Harvard/Tufts NueAnalysis
  framework, rewritten and expanded, in
  collaboration with UCL and Stanford, to become a
  MINOS Offline Software package.

14
Multivariate Discriminant Analysis

• Define a set of variables that
  appropriately describes the data sample
• Calculate the covariance matrix for each class
  
• Determine the Mahalanobis distance to each class
  for each event
• Compute the score for an event to belong to each
  class

15
Analysis Backgrounds
nm N ?nm pp0

• Neutral Current events Most significant
  contamination
• High-y CC nm Muon track is invisible
• Intrinsic Beam ne from kaon/muon decay, normally
  higher energy than signal peak
• nt CC tau decays into electron.

nm N ?m-n2p p-p0
ne N ?pe-
nt N ? pt-
16
Samples and Cuts

• Far samples
• Test sample composed of 19 nue, 19 numu and 19
  nutau MDC Far MC files processed with R1.12 (each
  file corresponds to 6.5x1020 POT)
• Training sample with identical size and
  proportions, no overlapping events
• Test on MDC Far Mock Data file also processed
  with R1.12 (7.4x1020 POT)
• Near samples for testing
• 229 MDC Near MC files (1.33x1016 POT each)
• 246 MDC Near Mock Data files (1.33x1016 POT)
• Sample cuts
• Fiducial and containment cuts
• Vertex contained in fiducial volume
• Full event containment in Far Det
• Full Z containment in Near Det
• Prong cut individual track or shower pulseheight
  gt 5000 sigcor (300 MeV)
• High energy Cut Total reconstructed event energy
  lt 150 MEU (6 GeV)
• Track Length Cut Events with track length lt 18
  planes.

17
Variable Selection

• Variable selection is performed using SAS
  Stepwise discriminant procedure
• Preliminary selection from 350 available
  variables
• 140 sorted by discriminating power
• Run classification method for the 19 variables
  with highest discriminating power.

18
MDA Classification Results

• Table summarizing FOMsig/sqrt(Sbg) for different
  oscillation parameters

19
Mock Data Challenge Results

• Using CC group best fit oscillation parameters
  sin2(2q23)0.925 Dm2322.175x10-3 eV2
• And Ue320.01
• FOMs at UE320.01, CC best fit, 7.4e20 POT
• MDA 0.532
• NN 0.510
• BDT 0.441
• NuMI-714 0.43

20
Mock Data Challenge Results

• MDC truth values
• sin2(2q13)0.151
• sin2(2q23)0.925 Dm2322.175x10-3 eV2
• Showing result summary plots for all three
  analyses with 90 and 99 confidence intervals
  for 7.4 and 22.2x1020 POT.

21
Conclusions

• The Tufts HEP Group continued to make fundamental
  contributions to MINOS batch processing and Monte
  Carlo sample generation
• Made significant contributions to the MINOS ne
  Appearance Analysis Group, in the form of an
  alternate shower reconstruction method
• Was a leading participant in the successful
  completion of the Mock Data Challenge. Results
  obtained compared favorably with other analysis
  methods
• Largest improvement over CHOOZ 90 CL sensitivity
• Inconsistent with sin2(2q13)0 at 90 CL
• True MDC value of sin2(2q13) well within the 90
  region around the best fit
• Future immediate work includes thesis defense and
  completion of knowledge transfer.