Difference between revisions of "PID study proposal"
From GlueXWiki
(→PID Studies) |
(→Systematic Studies) |
||
Line 20: | Line 20: | ||
* efficiency of the cut is N(pass) / [N(pass) + N(fail)] | * efficiency of the cut is N(pass) / [N(pass) + N(fail)] | ||
* compare this efficiency between data and MC to determine how well it is modeled | * compare this efficiency between data and MC to determine how well it is modeled | ||
+ | ** Note: need to check for the presence of peaking backgrounds, good to look at bggen MC | ||
Note this probably only works for the timing PID right now. Will need a separate set of files to test CDC dE/dx, but for now just look at the distributions. The biggest contributor here is probably the rate of events without enough hits to properly calculate dE/dx | Note this probably only works for the timing PID right now. Will need a separate set of files to test CDC dE/dx, but for now just look at the distributions. The biggest contributor here is probably the rate of events without enough hits to properly calculate dE/dx |
Revision as of 17:27, 22 May 2020
PID Studies
For each final state
- Determine p/theta range of each final state particle
- Compare PID variable distributions between data and MC
- First stage: 1D distributions integrated over all kinematics
- Optional: !D distributions from different p/theta bins
- Do this for each run period under investigation
- Determine selection criteria which are 99% and 95% efficient
Systematic Studies
How to determine systematic uncertainty in efficiency due to PID cuts (assumes you have a final state with some clean peak: rho, phi, pi0, eta, eta'...):
- make tight PID cuts on all particles except the one you are testing the efficiency of - call this particle P
- make two sets of invariant mass distributions for whatever peak you have
- masses for events in which P satisfies the standard PID requirements
- masses for events in which P fails the standard PID requirements
- Fit each mass distribution to get the yields: N(pass) and N(fail)
- efficiency of the cut is N(pass) / [N(pass) + N(fail)]
- compare this efficiency between data and MC to determine how well it is modeled
- Note: need to check for the presence of peaking backgrounds, good to look at bggen MC
Note this probably only works for the timing PID right now. Will need a separate set of files to test CDC dE/dx, but for now just look at the distributions. The biggest contributor here is probably the rate of events without enough hits to properly calculate dE/dx