PID efficiency

[jackfrost373]

We should determine the ratio of efficiencies due to the PID selection between Xic and Lc. We do not expect a big difference, since the momenta are mostly the same. Due to the different lifetime, they could pass through a different region in RICH so we will have to bin in the right variables.

We should either 'resample' MC to match the data, or use a data-driven method such as LHCb's 'PIDCalib' tool to assess this.

[jackfrost373]

• [ ] Estimate PID efficiency (on top of trigger & selection) from MC (this is easy: cut on PID and compare yields to without cut. Then take the ratio of Lc and Xic).

• [x] Provide PID calibration samples (from D* tagged D0 -> K pi decays) from PIDCalib (@jackfrost373 )

• [ ] Determine efficiency of signal from calibration samples @SimonCalo

• [ ] Determine stat and syst error. [TBD - probably toy method & varying binning]

• [ ] Re-run data and MC to contain nTracks variable @jackfrost373 (partly done!)

[jackfrost373]

PID efficiency calibration histograms for Pions, Kaons and Protons, for our ProbNN selection, are now produced and can be found on stoomboot here: /dcache/bfys/jdevries/histograms/LcAnalysis/pidcalib/

These are TH3F (3D) histograms (in P, ETA and nTracks), where the bin content represents the efficiency for that P,ETA,nTrack bin, and the error on the bin represents the error on the efficiency.

Instructions:

• Total PID efficiencies for one particle PID cut can be obtained by binning the sWeighed signal of that particle into the same bins, and multiplying the histograms.
• The total efficiency (For the Lc, i.e. for all particles (p K pi)) can then be obtained by multiplying the three efficiencies.
• The total error can be obtained with error propagation - however, this does not take into account the kinematic correlations between the p, K and pi - you treat them as independent. To get around this, a common method is a 'toy method': Create 100 variations of the original PerfHists, by varying each bin value independently, within its error (i.e. assume for each bin a new gaussian distribution, with mean=value, and width=error, then pull a random number from that gaussian, and assign that value to that bin. Then repeat for the next bin, etc.). Then proceed as normal, but 100 times. In that way, you obtain 100 efficiencies for p, 100 for K, 100 for pi. And thus, 100 total (event) efficiencies. Take the mean of those 100 to be the real value for the total efficiency, and the standard deviation as the error. Then, you have taken the correlations into account.

[cpawley]

Following WG meeting, check error bars on efficiencies

[cpawley]

Probably due to low MC statistics

[jackfrost373]

The PID efficiency has been calculated thoroughly by @nurbonaite 's BTR - she should merge her code here.