Run 9, 200 GeV, BEMC Prompt Photon Analysis




Data Selection

 The photon analysis utilizes data taken during the 2009 200 GeV running period.  In particular, the data set begins with the st_gamma production stream that took all events with L2BGamma and L2EGamma triggers (technically these are stored in the File Catalog as two separate streams, st_gamma and st_gamma_adc).

After selecting the production2009_200GeV_Single were  runs, I required custom QA criteria using the same offline run QA that Brian had used to determine the first and second priority lists.  This QA was convolved with Grant's QA and lastly a few runs were removed due to poor bunch crossing reconstruction



The parameters of the filtered photon simulation produced at the MIT Tier2 facility are documented in the initial request.





The following analyses has been conveniently summarized in a (slightly out-of-date) graphical flowchart.


Event Selection

Events are first required to have a StGammaCandidate, a 3x3 BEMC tower cluster with seed energy greater than 5.25 GeV and total transverse energy greater than 7 GeV.  In addition to the implied L2BGamma trigger from the stream selection, the event must also have a simulated trigger accept.  Lastly, only events with a reconstructed vertex with positive rank and | vz | < 60 cm are accepted.


Variable Selection


Data/Simulation Comparison

After reconstructing StGammaCandidates in both simulation and data, the simulation is weighted such that the vertex distribution matches that of data.  Various characteristics of the clusters (and the towers and tracks in the surrounding neighborhood) are compared, and those used in the final analysis have been compared.  

Note that, while no variable is a great discriminator of photons and the QCD background, many are weak discriminants and the spectral comparisons are sensitive to the input prompt photon cross section in Pythia.  Because this is known to be incorrect (in general the Pythia cross section underestimates NLO and other measurements) the agreement in these variables is not expected to be perfect, especially in regions where photons peak.  The final check of consistency between and simulation must wait until the signal extraction.


Gaussian Process Regression

The variables selected are mapping into a single, one-dimensional discriminant using Gaussian process regression that tunes the mapping to optimize the separation between signal and background events.


Signal Extraction

Although the mapping given by the Gaussian process is not sufficient to full separate signal and background, it does induce a mapping to a one-dimensional space where the full spectrum can be fit to a sum of signal and background shapes inferred from simulation.

Separating the full simulation into two admits a test of the consistency of this approach.  Treating one sample as simulation and the other as data, the fitting procedure extracts the correct signal fraction across the full range of photon kinematics (the deviation in the 8 GeV bin is an artifact of the splitting procedure but still consistent with the given uncertainty).


One particularly powerful feature of this approach is that the residuals of the fit are directly sensitive to systematic errors in the simulation.  Instead of studying in simulation parameter is isolation, then, the systematic error due to data/simulation discrepancy can be immediately constrained by studying non-statistical behavior in the fit residuals in each bin.


The signal extracted above is binned in reconstructed energy; the measured spectrum is corrected the true energy spectrum with Bayesian deconvolution.

In order to test the validity of the deconvolution, the signal extracted from the split simulation above is deconvolved and compared to the true spectrum.  Indeed, the algorithm performs well.

Cross Section


Trigger, acceptance, and reconstruction efficiencies are calculated together using the full simulation.


Preliminary measurements of the BBC and ZDC cross sections in the 2009 running have been measured by CAD but not yet released to the collaboration.  In the meantime, the nominal BBC cross section of 26.1 mb was used (with a conservative 10% uncertainty) to compute a L2BemcGamma luminosity of ~13.5 pb-1.