Jet LIfetime Probability (JLIP) b-tag Certification

Introduction:

Contents

1. How to use JLIP ?
2. Method description
3. SystemD macro
4. Event samples and selections
5. Calibration of the impact parameter resolution
6. V0 rejection
7. Track probability
8. Jet lifetime probability
9. Working points
10. Efficiency in Monte Carlo
11. Light quark tagging in Data (TRF_light)
12. b-tag efficiency in muon-in-jet Data
13. Data/Monte Carlo scale factor (SF_b)
14. b-tag (TRF_b) and c-tag (TRF_c) efficiencies in Data
15. Run dependence
16. JLIP performance and systematics

Instructions for using JLIP in your code:

For generic instructions see the d0root_example package. Specifically, instructions in the v00-01-07 tag are the most relevant.

• Generic Instructions (for all taggers, including JLIP). Includes tags for all packages you need.
• To use the tagger in your code you need to instantiate and initilize a JetLifeTimeTagger and a TrackHandler object. The following code gives an example:
  
  #include "d0root_analysis/algorithms/TrackHandler.h"
  #include "d0root_jlip/JetLifeTimeTagger.hpp"
  ...
  JetLifeTimeTagger _tagger = new JetLifeTimeTagger(); // ctor of JetLifeTimeTagger
  _tagger->SetCuts (0.5, 1.0, 10.0, 0.15, 0.4, 50.0, 0, 2); // Setting the parameters
  std::string data_type = "DATA"; // Use "MC" for Monte Carlo!!!!!
  std::string release = "p14.pass2";
  _tagger->Init(data_type, release); // Tagger initialisation
  TrackHandler _track_handler = new TrackHandler (); // TrackHandler ctor
  _track_handler->Init (data_type, "p14.03.00"); // TrackHandler initialisation
  
  
  
  • The data type can either be "DATA" or "MC". Note that the recommended procedure is to apply TRFs weight on MC instead of tagging.
  • The release should either be "p14.03" for pass1 data (JES 5.1 certification) or "p14.pass2" for pass2 data (JES 5.3 certification). Release in TrackHandler initialization should be "p14.03.00" in both cases
  You can find a complete example code that creates the JLIP tagger objects, calls them, and makes a cut on the returned probability within d0root_example
  • Code that shows you how to get the tracks, jets, and PV out of a ReadEvent object, and how to prepare them before calling JLIP (V0 removal, etc.).
  • The d0root_example package, which contains code to run on TMBs, TMBTrees, and top_trees -- and call all the taggers, including JLIP (above code references specific files in this package).

  NOTE: These links do not reference a specific tagged version of the code. If you want to look at v00-01-07, then look here (but it is slower for an interface).
  

Example of use within the d0root framework : here

Note that the recommended d0root_jlip package version is v00-00-15

and that the recommended d0root_analysis version is >= v00-09-61

JLIP tagging rate functions:

If you encounter any problems, please send a mail at : Benoit Clement ( ubik@fnal.gov ).

The Et-eta parametrizations (b-tagging efficiencies in data , light quark tagging efficiency and data to monte-carlo scale factor can be found in the TRF_JLIP_p14pass2.root included in the tar file, their name are self explanatory. Use the JLIP_TRF class to access them. More information on the use of these TRFs can be found here. Note that the JLIP_TRF class also add some systematics error that are not taken into account in the +/- 1 sigma parametrization provided in the root file. These parametrizations are now provided for 6 working points corresponding to ExtraLoose, SuperLoose, Loose, Medium, Tight and UltraTight cuts.
Jet flavour in MC is obtain by jet/hadron (not parton) matching within a dR<0.5 cone :

- A b-jet contains at least one b hadron.
- A c-jet contains at least one c hadron and no b hadron.
- A light jet contains neither c hadron nor b hadron.

These TRF's have been obtained on pass2 data and fixed MC, using TmbTrees produced with d0correct v8. Old pass 1 TRFs (3 working points only) are still available in the TRFs_old directory of d0root_jlip

NOTE : none taggability parametrization is provided since this later is analysis dependant: analysers have to compute it themselves.

Cross-checks plots:

Method description:

D0 Note xxxx gives details informations on JLIP for p14/pass2 data and simulations (preliminary certification note).

D0 Note 4159 describes the SystemD algorithm to estimate the b-tagging efficiency with data only.

D0 Note 4359 describes the calculation for the event weight using TRFs, for single and double tag.

SystemD macro:

Event samples and selections:

Event samples: Jes5.3 is used for Data and MC

• Real data from CSG pass 2 fixed and skimmed data, with >= 2 good jets (Et>15 GeV) and a good run quality (see the bid page ):
  • 21M BID skim muon-in-jets, with a loose muon, pt>4 GeV, inside a 0.5 cone jet (JCCB) for b-tag efficiency
  • 1.4M QCD skim jet triggers for mistag rate
  • 1.8M EM1TRK skim with one electron (Et>8 GeV) and missing Et<10 GeV for mistag rate
• MC (with fixing) (see the bid page ):
  • 700k qcd (pt within 40-80 and 80-160 GeV) for mistag corrections
  • 100k Z->bb->mu and 60k ttbar->b->mu ( ipanema ) for b-tag Data/MC scale factor
  • 100k Z->bb, 100k ttbar ( ipanema ) and 200k qcd->bb 40-80 and 80-160 for TRFb
  • 100k Z->cc and 200k qcd->cc 40-80 and 80-160 for TRFc

Primary vertex selection:

• use the 2-pass probabilistic primary vertex
• require a number of attached tracks N_PV >= 3 and |z_PV|<60cm

Jet selection and taggability:

• use 0.5 cone algorithm (JCCB) for calorimeter jets jet id and jes correction (without using the muon information)
• reject electrons
• match the calorimeter jet to a track jet (see SVT )
• require E_T(jet) > 15 GeV
• keep only events with >=2 taggable jets

Track selection for tagging:

• p_T > 1 GeV/c
• >= 1 SMT hit (including the inner layer if |eta|>1.6 and if the number of CFT hits is <= 6)
• track impact parameter w.r.t primary vertex |IP| < 0.15 cm in transverse plane, and |zIP| < 0.40 cm along beam axis
• the track is associated to a jet within Delta R < 0.5

Muon selection:

• use medium muons with nseg=3 (see here )
• if there are several muons in the same jet, that with the smallest global Chi2 is retained (comparing muons with p_T(muon)>2 GeV)
• keep the muon if p_T(muon) > 4 GeV/c
• require this muon to have a |zIP|<1cm
• define p_Trel = p_T(muon) / jet+muon axis

The following figure shows the taggability (ratio of the numbers of taggable jets over all reconstructed jets) in multi-jet Data and in qcd Monte Carlo.

Calibration of the impact parameter resolution:

Track categories: 29 track categories are considered, depending on the number of SMT hits, CFT hits, |eta|, p_T and Chi2 values of the tracks.

The IP significance (ratio of the IP value divided by its error) is fitted for each category in 16 pscat = p (sin theta)^{3/2} intervals, using the sum of a gaussian and an exponential function. In the following plots, the fitted IP significance is shown for tracks with 2.5 < pscat < 3 GeV/c:

The pull values (sigma of the previous gaussian) are used to correct the IP error:

Another correction is applied on the IP error, which depends on the number of tracks attached to the primary vertex:

Finally the experimental IP resolution is infered:

V0 rejection:

The CSIP V0 search algorithm is used to reject track candidates from K0s, /\ or photon conversions. The following plots show the K0s and /\ invariant mass distributions observed in jet trigger data:

Track probability:

The IP significance is given the same sign as the scalar product between the IP vector and the jet axis. The resolution functions are computed using tracks with negative significance in the jet trigger data. All track categories are used, giving a total of 29 parametrisations of the negative significances.

Some of these resolution functions, fitted with 4 Gaussians, are shown here:

The track probabilities are then computed, in jet trigger data or in the simulation:

This is the probability for a track to originate from the primary vertex. The background is flat by construction. Tracks from decays in flight have a probability close to zero.

Jet lifetime probability:

Assuming the individual track probabilities to be independent, a lifetime probability is finally computed for each selected jet.

Light quark jets tend to have a flat probability distribution. Jets from c and b quarks have a probability peaked at zero:

Working points:

6 working points are provided, corresponding to a JLIP probability value smaller than:

• 0.1% UltraTight
• 0.3% Tight
• 0.5% Medium
• 1.0% Loose
• 2.0% SuperLoose
• 4.0% ExtraLoose

Efficiency in Monte Carlo:

Jets are tagged by applying a cut on the jet lifetime probability. A quark-tag efficiency is simply the ratio of the number of tagged jets divided by the number of taggable jets for a given quark specy. Aplying a cut on the jet lifetime probability, the following efficiency plots are obtained for various MC samples and vs the jet E_T and eta (here for the Medium working point):

Light quark tagging in Data (TRF_light):

The light quark tagging efficiency is estimated from negative tag rates observed in jet trigger data and applying a correction from qcd MC. An illustration is provided here for the Medium cut:

and finally for all working points (in logarithmic scale):

b-tag efficiency in muon-in-jet Data:

The SystemD method is used to estimate the b-tagging efficiency in the muon-in-jet Data:

Data/Monte Carlo scale factor (SF_b):

The b-tag efficiency is higher in MC than in data.

Using muon-in-jets both in data and in MC, the ratio of b-tag efficiencies is found to depend on the jet lifetime probability cut, on the jet E_T and eta:

b-tag (TRF_b) and c-tag (TRF_c) efficiencies in Data:

The inclusive b-tag efficiency in Data is computed as the product of the inclusive b-tag efficiency in Monte-Carlo, times the previous Data/MC scale factor. The same procedure is applied for c-quarks. They are parametrized as a function of the jet E_T and eta, and depend on the working point:

Run dependence:

A run dependence has been observed for the b-tag and mistag efficiencies, corresponding to the increased Tevatron luminosity by the end of January 2004. For run numbers before and after 189000, the TRF's and SFb can be multiplied a correction factor (see here).

JLIP performance and systematics:

The SystemD method can be used to evaluate the b-tag efficiency in muon-in-jet Data. The light quark mistag rate is evaluated using multi-jet Data. Varying the JLIP cut for the 6 working points, the following performance curves are obtained in various jet Et and eta ranges (here with statistical errors only):

The systematic uncertainty on the b-tag efficiency is evaluated to be +_2.7% (with SystemD) for each working point.

The overall stat.+syst. uncertainties are illustrated in the following figure: