1. Preliminaries

The following comparisons address the Run 9 200 GeV running period.  Specifically, the data is culled from the st_gamma production stream with a custom QA designed for a BEMC only analysis and the simulation is a photon-oriented filtered production.  StGammaCandidates, 3x3 tower clusters with seed and total energy thresholds, are constructed and the SMD response from strips associated with the cluster are considered.

All simulated spectra will be presented in red, data spectra in black

2. SMD Calibration Tables

The BSMD, like all BEMC detector systems, features two calibration coefficients,

        E_reco = (ADC - Pedestal) * Calib * Gain,

where Calib converts ADC to reconstructed energy and Gain was originally reserved for perturbations such as time dependencies.  Note that, because only the Calib tables are used when modeling the ADC response in simulation,

        ADC_simu = E_deposit * (Sampling Fraction)^-1 * (Calib)^-1 + Pedestal,

the Gain tables are irrelevant for data/simulation comparisons.  Consequently, the Gain tables will not be used in the following analysis.

The Calib tables are determined in two steps.  Initial values are selected to equalize the relative response of each strip and then an overall scale is applied.  In 2009 this scale was determined by comparing the total SMD response to isolated electrons in data and single electron simulations at relatively low energies (2-4 GeV).

3. Nominal Results

The BEMC slow simulator was initially tuned with my own educated guesses of the BSMDE cross talk and ADC thresholds.

    StEmcSimulatorMaker* emcSim = new StEmcSimulatorMaker();

    emcSim->setMaxCrossTalkPercentage(kBarrelSmdEtaStripId, 0.5);

    emcSim->setMaximumAdc(kBarrelSmdEtaStripId, 900);

    emcSim->setMaximumAdc(kBarrelSmdPhiStripId, 900);

3.2 Nominal BSMDP Spectra

4. Tuned Results

Tweaking the Calib tables does not seem to provide an immediate solution.  Naively, tuning to optimize the agreement of the maximum strip energy spectra would only spoil the agreement of the summed energy spectra.  The subtlety here, however, is the presence of the ADC saturation thresholds and their effect on simulation.

As the Calib values are decreased, the ADC for a given energy deposition increases and draws closer to the saturation threshold and spectral information in the higher energy tail is lost to the saturation peak.  Increasing the values, on the other hand, pulls the spectra away from the threshold and more shape information survives.

This behavior is evident in the following spectra of the maximum energy Phi strip (eMaxPhiStrip) verses the detector eta of the parent StGammaCandidate.  With larger Calib values the simulated ADCs remain below the saturation threshold and the full eta dependence of the energy spectrum survives.  At smaller values the simulated ADCs fall into the saturation peak and the "wings" of the energy spectrum are clipped.

By first setting the ADC thresholds to appropriate values, an appropriate tuning of the Calib values (and the BSMDE cross talk) should be able to offer agreement in all SMD spectra.  The entire procedure suggests the slow simulator parameters,

    StEmcSimulatorMaker* emcSim = new StEmcSimulatorMaker();

    emcSim->setMaxCrossTalkPercentage(kBarrelSmdEtaStripId, 4.0);

    emcSim->setMaximumAdc(kBarrelSmdEtaStripId, 913);

    emcSim->setMaximumAdcSpread(kBarrelSmdEtaStripId, 18);

    emcSim->setMaximumAdc(kBarrelSmdPhiStripId, 895);

    emcSim->setMaximumAdcSpread(kBarrelSmdPhiStripId, 18);

The tuned cross talk conveniently agrees with the studies of the original UCLA bench tests (see attached note).

With these changes the spectra are in excellent agreement, especially in the ratio of maximum energy to summed energy which is most sensitive to the shape of the SMD response.  Note also the agreement in the shape of the saturation peak evidence in the ADC spectra of the highest energy strip (adcMaxEtaStrip and adcMaxPhiStrip).

Small discrepancies do remain at small energies, but these are almost entirely limited to events with only a single non-zero SMD strip.  Such events are sensitive to effects, such as heavy-tailed pedestal spectra and low energy hadronic energy depositions, that are not expected to be well simulated in the first place.

4.1 Tuned BSMDE Spectra

4.2 Tuned BSMDP Spectra

4.3 Tuned Energy Dependence

If the SMD response is linear within these kinematics, then the above agreement should hold for all incident particle energies.  

The one giant caveat, however, is that the Pythia modeling of the prompt photon cross section is known to be flawed.  In particular, Pythia underestimates the photon cross section and that underestimation grows at higher energies.  At higher energies where the QCD background falls and the photon abundance increases, then, variables sensitive to the differences between photons and the QCD background (such as the max strip energy-summed strip energy ratio) will not be perfectly modeled at high energies.  Attempting to calibrate such a descrepancy away will only bias subsequent analyses.  

That said, below are the energy dependencies of the max strip energy-summed strip energy ratio.  Simulation is on the left, data on the right.

4.4 Tuned Eta Dependence

Here the energy of the maximum strip (first for the eta plane and then for the phi plane) is binned as a function of detector eta (dEta).  As above, simulation is on the left and data on the right.

Note the flatness of the distribution for the phi plane, a consequence of the saturation clipping the "wings" at larger eta as discussed above.