Unbiased Universal Curve Fits

1
Unbiased Universal Curve Fits
Matthew Jones
April 20, 2004

• Motivation
• Prior belief functions for TOF t0 calculation
• Combine with particle ID from TOF SVX
• Method
• Use TOF to identify particles in 1/?2 region
• Assume the universal curve is universal
• Simultaneous likelihood fit with TOFdE/dx

2
Universal curve with XFT bias

• Tuned to K/? with pT gt 2 GeV/c

Momentum
3
Time-of-Flight analysis

• Calculate ti t TOFi for mass i
• ? ti ti t0 is Gaussian with zero mean for
  correct mass hypothesis
• Likelihood function
• Imbelishments
• Double Gaussian resolution function
• Background parameterization

4
Time-of-Flight Analysis

• Resolution function
• Core resolution scale factor, s
• Wide fraction, fw
• Wide scale factor, sw
• Background function
• Two background components
• Mis-measured time, ti ? positive bias
• Mis-measured t0 ? negative bias
• Parameterize as exponentials convoluted with
  resolution function

5
Example Time-of-Flight fit
Pions
Kaons
Protons
Background
Deuterons
6
dE/dx parameterization

• Apply macroscopic corrections (note 6932)
• Standard CDF parameterization
• Resolution parameterization

7
Fit results
Positive tracks from data taken in March, 2002.
Momentum

• Resolution

8
Example dE/dx projection
Pions
Kaons
Electrons
Protons
Deuterons
9
Particle Fractions in hbhd0c data

• Used for (sample specific) prior belief functions
  in TOF reconstruction

10
Other TOF fit parameters
11
Resolution History

• Resolution might not be completely stable with
  time.
• Not all data had dE/dx stored in CdfTrack

12
Observations

1. TOF resolution function might be better
   constrained using ?s from D decay
2. Should fit positive and negative particle
   fractions separately and go to higher momenta
3. Probably should bin fractions in pT instead of p

13
Observations

• Need to cross check with pure samples
• Use leptonD sample, avoiding XFT tracks
• Use ? from D strip
• Use conversions to check electron dE/dx
• Could probably get a better description of dE/dx
  resolution function, ?dE/dx
• Parameters are still probably good enough for
  development code release