October Exercise: Measurement of the Drell-Yan Cross Section

Stoyan Stoynev, Andy Kubik, Michael Schmitt (Northwestern University, USA)

contact person: Stoyan Stoynev

Goals and Motivation

The October exercise was intended as a kind of dress rehearsal for physics analysis with real data. The primary objective was to find problems in the workflow, especially regarding data transfers, production and distribution of skims, etc. In the electroweak group, four analyses were chosen as official exercises. In addition, volunteers were invited to participate to simulate the random nature of physics analysis within a large collaboration.

Our goal was to attempt the measurement of the Drell-Yan cross section in the di-muon final state, as one of the so-called random analyses.

Previous presentations of the details of this analysis are available:

• EWK Muon Meeting 21-May-2009
• EWK Muon Meeting 4-June-2009
• EWK Muon Meeting 2-July-2009

Work Flow

We made use of the secondary data sets for muons, and produced our own PATs:

1. identify relevant secondary data sets (skims)
   • SD_Mu9
   • four sources: Drell-Yan, top pairs, W+jets and inclusive muon
     1. /Zmumu/Summer09-MC_31X_V3_SD_Mu9-v1/GEN-SIM-RECO
     2. /TTbar/Summer09-MC_31X_V3_SD_Mu9-v1/GEN-SIM-RECO
     3. /Wmunu/Summer09-MC_31X_V3_SD_Mu9-v2/GEN-SIM-RECO
     4. /InclusiveMu15/Summer09-MC_31X_V3_SD_Mu9-v1/GEN-SIM-RECO
2. produce inclusive PAT files (no additional skimming)
   • use the GRID
   • standard PAT event content + some RECO for specialized studies
   • PAT files resident on disk at FNAL (/store/user/ at T3_US_FNALLPC)
3. publish the PAT files in local DBS (cms_dbs_ph_analysis_02)
4. run our analysis code at the LPC

We experienced essentially no technical problems during any of these steps.

We executed this work flow for the 10 TeV samples only.

Elements of the Analysis

Definition

• we are measuring (1/σ(Z)) (dσ/dM)
• we accept di-muon events with all invariant masses above 20 GeV
• we use 14 unequal mass bins
• we unsmear the observed spectrum

Event Selection

• demand two global muons, pT > 12 and pT > 8, opposite sign
• trigger requirement: HLT_Mu9
• apply cuts against muons from meson decays, etc.
  • isolation: no more than three tracks in a cone R < 0.5
  • calorimeter compatibility
  • matching compatibility

Acceptance and Efficiency

• acceptance calculated from simulations (Summer09 MC)
• trigger efficiency calculated based on 10 pb-1 of Summer09 MC (necessary SDs not available)
• muon reconstruction efficiency calculated using tag-and-probe

Background Estimates

• top quark pairs and W+jets estimates from simulations directly
• QCD estimated using the quadrant method

Unfolding the Spectrum

• after subtracting all backgrounds, the spectrum must be unfolded
• we use currently two rather independent techniques
  1. traditional matrix approach
  2. weighting approach
• they have been shown to give consistent results with different systematics

Results

Five main results from this exercise:

1. Event Yields nEvents_Minv.png
   • four processes are shown: Drell-Yan signal, top pairs, W+jets and inclusive muons from jets
   • yields correspond to the October exercise, roughly 10 pb-1
   • at low masses, multi-jet backgrounds are most important. At high backgrounds, top pairs are.
2. Acceptance Times Efficiency acc_Minv.png
   • this plot lumps together the acceptance and all efficiency factors
3. Trigger Efficiency trig_eff__Mu9_RedIsJet30Based_secondSample.png
   • for the October exercise, there is no jet-based skim of the simulated Drell-Yan data set, so we cannot exercise our method for measuring the trigger efficiency.
   • we used the Summer09 MC samples to produce the figure shown (about 500 pb-1)
   • black histogram gives the true efficiencies, while the red points give the data-driven estimates
   • in the table below, we scaled the errors to 10 pb-1
4. Background Subtraction superimposed_PredTrue_PtISO_20.png
   • use a quadrant method based on the muon impact parameter and pT in a cone around the muon
   • the solid histogram gives the true number of signal events, and the red points show the predicted number
5. Measurement DY_crossSect_October_allErrors.png
   • the background-subtracted spectrum is unfolded and corrected for acceptance and efficiencies
   • the black open squares are the true MC values.
   • The red and blue points with errors show the results obtained using the two different unfolding techniques.

The following table summarize the information about the predicted number of signal events NS and the main uncertainties in terms of number of events - statistical from combined samples(stat.), background estimation (BG) and Trigger efficiency uncertainties (trig.) for each of the 14 di-muon mass bins:

The error bars shown on the final measurement plot include the total errors in this table.

-- StoyanStoynev - 15-Oct-2009

• Acceptance and reco efficiency:
  

• Normalized DY spectrum:
  

• Signal and background di-muon mass spectra:
  

• Background subtracted sample:
  

• HLT_Mu9 efficiency: