Data Cleanup by ert554898


									Late Activity Cuts Without Bias

   Thomas H. Osiecki
   University of Texas at Austin

               Oct. Coll Meet. 2005   1
 Well known excess at low energies for both slices and events

  Red = Data
 Black = MC

This huge excess
Exists for both slices
And Events

Normalized by Number
Of Events

                           Oct. Coll Meet. 2005                 2
 Data Set
 Clues about origin
 New/Old MC Differences
   Effects change previous results
 Results from application of different
  late activity cuts
 Proposal
 Conclusion
                 Oct. Coll Meet. 2005     3
Data Set
 5.96e18 POT August LE-10 Data
 2.35e18 POT New LE-10 MC
 All plots Normalized to 1.0 / POT unless
  stated otherwise
 Cut on Horn Current to be nominal, there
  was high and low current running and it
  makes a difference
 Did not use July because of toroid
 All Events are subject to fiducial volume cut

                  Oct. Coll Meet. 2005        4
            Detector Clues
Time (ns)

                                Time (ns)

               No Cut                          Strip PH > 2.0 pe
                        Oct. Coll Meet. 2005                       5
Low PH Correlation
 For all lowph slices, found slice with
  closest first plane

                   Oct. Coll Meet. 2005    6
Time Correlation
   If I look at events on that correlate with beginning plane, one finds a
    long time distribution, a.k.a. late activity

                             How to get rid of them?

                              Oct. Coll Meet. 2005                            7
Exp Tail in Batch Structure
 Tail indicates late-activity, can be studied using LI
  in an sgate – See Rustem Ospanov’s Talk

                                            Long Exponential
                                            Tail of Activity

                     Oct. Coll Meet. 2005                 8
New MC
 LE-10
                                      Normalized to POT
 Inter-nuclear scattering
  turned on
 B-field Map 159
 Better estimate of
  cosmic rays, ala Robert

                       Oct. Coll Meet. 2005               9
MC Data difference I observe

            Oct. Coll Meet. 2005   10
Different Late-activity Cuts
 Timing Cut and Strip Removal (Niki)
 Will focus on the cuts that I have
  explored (Peter S. Suggestion)
   Rho – Fraction of event with early
   Exponentially Weighted Rho
   Rho in different time regimes

                 Oct. Coll Meet. 2005    11
Plan of Attack
 For each rho cut I look at:
   Spectra of Data/MC before/after cut
       Can one get them to agree?
       How much statistics does one lose?
    Effect of Cut at different beam intensities
       If there exists no bias, then the event spectrum should
        be the same after a cut for different beam intensities
       Use of Kolmogorov-Smirnov Test and Chi2 Test
       Keep in mind that statistics lower at lower intensities
    Single Event Spectrum
       Ideally would like infinite single-event sample, but will
        use this just for comparison

                         Oct. Coll Meet. 2005                       12
Event PH at Different Intensities
  Event Spectrum Shouldn’t change (at least for LE)

                   Oct. Coll Meet. 2005          13
‘Single’ Event Spectrum
 Take the first event from every snarl and
  plot this as a kind of ‘single’ event
  spectrum – throws out any notion of late
 I’m selecting one event per snarl, so I can’t
  just scale by POT.
 Need to scale using number of events
 Since this is to study bias, need to scale
  according to where I KNOW they agree, i.e.
  the HE tail.
 Keep in mind this is approximate, since it
  includes NO late activity
                  Oct. Coll Meet. 2005        14

 Cut based on previous hypothesis
 Since these junk slices correlate in time
  with a previous event, why not make a cut
  depending on how much previous activity
  occurred in the channels for said slice?

                 Oct. Coll Meet. 2005         15
Rho vs Energy
           Not in MC

           Oct. Coll Meet. 2005   16
Effect of Rho Cut

            Oct. Coll Meet. 2005   17
Zoom of effect of Rho

            Oct. Coll Meet. 2005   18
Rho Cut at Different Intensities

             Oct. Coll Meet. 2005   19
Bias from Rho

           Oct. Coll Meet. 2005   20
Weighted Rho

 Tau is approximately the characteristic time for
  later hits to be considered late-activity
 By weighting each strip hit by an exponential factor
  will increase w dramatically depending on how ‘late’
  the activity is
 If all an events hits are less than the ‘late’ activity
  one expects for ‘good’ events for rho to be small
  and for ‘bad’ event, rho is large

                       Oct. Coll Meet. 2005            21
Weighted Rho vs Energy

           Oct. Coll Meet. 2005   22
Effect of Weighted Rho

            Oct. Coll Meet. 2005   23
Zoom on Effect of Weighted Rho

             Oct. Coll Meet. 2005   24
Weighted Rho at diff. Intensities

              Oct. Coll Meet. 2005   25
Bias from Weighted Rho

           Oct. Coll Meet. 2005   26
3 Different Rhos
 In addition to the first rho I define
   Rho1 = Rho between[0,200] ns
   Rho2 = Rho between[200,1000] ns
   Rho3 = Rho between[1000,infinity] ns
 Hope is that since we observe
  different time scales for late activity
  that splitting rho up will give us
  greater cleanup power

                 Oct. Coll Meet. 2005       27
   3 Rhos vs Energy



        rho3        rho2              rho1
               Oct. Coll Meet. 2005          28
Effect of 3 rho cut

             Oct. Coll Meet. 2005   29
Zoom on Effect of 3 rhos

            Oct. Coll Meet. 2005   30
3 Rho’s vs Intensity

            Oct. Coll Meet. 2005   31
Bias from 3 different rhos

             Oct. Coll Meet. 2005   32
Final Data/MC with cuts

            Oct. Coll Meet. 2005   33
   So we need to clean up our data
       Essential to understand for NC analysis, not as big an issue for
   How are we making sure we do not bias?
       See how cuts affect spectra at different intensities
           Issue – Low statistics at lower intensities
       Use a ‘single’ event spectrum
           Not a real single event spectrum
   Proposal
       For a batch every 3 seconds, running 20 hours a day, one
        would get 24000 spills. I suggest 1 to 2 days of running at
       About 1 neu in the far every 4 hours. -> Would lose about 6-
        12. Is this acceptable?
       The only way to truly know if we’re biasing is to get as close to
        a single event spectra as we can.
       Comments?

                              Oct. Coll Meet. 2005                     34
Plots for Proposal

             Oct. Coll Meet. 2005   35
 All 3 do a comparable job of cleaning up the data
 Original rho seems to match data/mc the best
 Weighted rho seems to cause the least bias –
  especially to the lower side of the main peak (minor
 Still this minor deficit in data on lower side of peak
 I like the original rho because it matches data better,
  and slight bias is almost neglible compared to
  weighted rho
   Last NC meeting I concluded that the 3 rho’s is
       better, but that was before new MC.

                       Oct. Coll Meet. 2005             36

To top