Point Source Localization: Optimizing the CTBCORE cut

1
Point Source LocalizationOptimizing the CTBCORE
cut

• Toby Burnett University of Washington

2
Recall the Simple PSF parameterization

• LATDOC AM-04355 We define a PDF in the angular
  deviation ? with two parameters ? a scale ?
  the power law

An example fit using all-gamma
Corner defined by ?
Tail defined by ?
3
The Classification Tree Analysis

• Recall that there are four trees, one for front
  and back, and whether a single track or vertex
  predicts a better PSF. They are combined in a
  single variable CTBCORE.
• The DC2 event class class A is defined by
  CTBCOREgt0.35.
• This cut was not optimized while increasing it
  reduces the effective area, it also dramatically
  improves the PSF, as Bill has shown
• Can we improve the pointing resolution by
  restricting CTBCORE?

All gamma, class A
Class A cut
4
Localization Analysis

• Presented last week http//d0.phys.washington.edu
  /burnett/GLAST/AnaGroup/DC2_2_localization.ppt
• Based on UW simple likelihood function the
  simple PSF uniform background, one parameter
  describing the signal fraction in each of 8
  energy bands.
• Localization done by maximizing the total
  likelihood (or TS) with respect to position.
• Output from /users/burnett/tools two stages
  spectrum, position

Initial position for signal fraction fits
Spectrum of source MRF0141_v1r2 at ra,
dec111.597000, -27.990000level events  sig
fraction  TS   6   515  0.00 /- 0.02    -0   7
  131  0.00 /- 0.05    -0   8    41  0.27 /-
0.11    11   9    18  0.93 /- 0.13    37  10
   19  1.00 /- 0.14    42  11    10  0.97 /-
0.18    21  12    20  1.00 /- 0.18    49  13
   32  1.00 /- 0.23    98                        
    258 Gradient   delta     ra        dec
      error     Ts  11166  0.0019    111.5970
 -27.9900  0.0031    257.9995    685  0.0001
   111.5988  -27.9890  0.0031    258.3586   
good fit
Total TS just add TS for each band
Projected error (deg) from likelihood curvature
Maximize total TS by varying position (fractions
fixed)
5
Expected position Resolution

• See the LATDOC projected resolution varies with
  ?, ?, and the efficiency ? asNote that for the
  Gaussian case, ???, it becomes ?/?N, where N is
  the total number of events(This ignores the
  effect of background, which degrades the
  resolution)
• So, as we increase CTBCORE, we expect ? to
  increase, but ? to decrease is there an optimum?

6
Measurements from AllGamma

• The following plot was derived from PSF fits to
  class A front, integrating over all angles for
  each of the 8 energy bands, weighed for a E-2
  signal.
• The effect is spectacular!
• The variation of ? is flat.
• So, what about the counting factor ??

7
The bottom Line
CTBCOREgt0.35 looks just fine super job, Bill !