Number 
Figures 
Description 
1 

In the Light Neutralino and Gravitino GMSB models, all but the
neutralino and gravitino can be inaccessible at Tevatron energies. In
this case, sparticles are produced through the production and decay of
a Higgs boson. These models suggest exclusive γ γ + MET if the
neutralino has a short lifetime or delayed γ + MET if the lifetime is
long, on the order of 5 ns.

2 


a) Prompt photons have an arrivaltime consistent with
traveling from the collision point to the calorimeter at the speed of
light. However, multiple collisions can occur in each event, and sometimes the
right vertex is not reconstructed. The selection of the wrong vertex can significantly affect the timing distribution for promptly produced photons.
b) Photons coming from the decay of heavy, longlived neutralinos
come from a point displaced in space and time from the collision point so photons tend to arrive late relative to expectations from prompt
photons.
b) A cartoon showing the expected timing distribution divided
into components. Standard model backgrounds contribute right and
wrongvertex Gaussians and cosmics contribute a flact background.
Signal from a longlived neutralino would look approximately like a
falling exponential smeared by the detector resolution for positive values of tcorr.


3 


a) Hadronic energy cut for cosmic ray removal.
b) CES energy cut for cosmic ray removal.

4 


a) Rightvertex distribution for e + MET in data.
b) Wrongvertex distribution for e + MET in data.
c) The tcorr distribution for e + MET events where the electron track is ignored in the vertexing and the highest SumPt vertex is selected as is done for γ+MET events. The red and blue
showthe bestfit result from the doubleGaussian fit even though we do not know whether it was a right vertex or a wrong vertex on an eventbyevent basis. Note
that the timing distribution is welldescribed by the twoGaussian
model.


5 

The tcorr distribution for MC W > eν > e + MET events
with the wrongvertex distribution mean fixed to 0.0 ns and the background rates determined from the fit to the data in the region [7 ns, 2 ns]. 
6 


a) Both the Et and the tcorr are
mismeasured by choosing a wrong vertex. There is a high degree of
correlation between the mismeasurements because picking a wrong vertex
causes the apparent path length (TOF^WV) to increase relative to the
true path length (TOF^RV), causing the measured corrected time and the
measured Et to both decrease. If picking a wrong vertex causes the
apparent path length to decrease, the measured t_corr and the measured
Et both increase.
b) This shows the Et distribution as measured around the true
collision point for MC W > eν > γ+MET Monte Carlo with Et^WV
greater than 25 GeV (white) and Et^WV greater than 45 GeV (green) as
measured around the selected vertex. The events to the left of the line
at 45 GeV are those promoting over threshold. The difference between
the white and the green above 45 GeV are those demoting below
threshold. Both effects conspire to cause a net positive shift in the
wrongvertex distribution.

7 

The true collision position for a MC sample of γ + jet
events that pass the final γ + MET requirements. We note
that a largerthanaverage number of these events are produced with z > 60 cm because events that are produced at large z have a higher probability that the orientation of the jet is directed outside of the detector than for events produced near the center of the detector. Because the primary vertex selection only considers SpaceTime vertices with z < 60 cm, all events with z > 60 cm have the wrong vertex selected and have large corrected values of (TOF^RV  TOF^WV). We reject events with evidence that there was a collision with z > 60 cm. 
8 


A comparison of the average path length for W > eν events where the electrons are reconstructed as electrons, and where they are reconstructed as photons. In (a) we show the full path length and in (b) we show the
difference of the time of flight from the center of the detector which is a fixed point in space and effectively averages over all WV positions. In both cases, the path length is bigger for photons than fake electrons.

9 


Some plots showing the new e > γ_fake rejection techniques
used in this analysis.
a) Δφ vs. Δη (between the reconstructed photon and the closest track, for both variables) for the MC W > eν
> γ_fake + MET sample. The oval indicates the DeltaRpull > 5 cut.
b) The closest trackphoton distributions in ΔRpull for
the control sample and the sample of fakephoton events. Making a cut
on ΔRpull provides a better MC Zγ > ννγ > γ+MET efficiency
and e > γfake rejection power than cutting on ΔR . Note that
both samples are set to the same normalization.
c) The efficiency and rejection power of our cut as a function
of ΔRcut pull. Note that a cut at ΔRpull > 5 (red dashed
line) results in approximately 95% efficiency of MC Zγ > ννγ >
γ + MET and 73% rejection of e > γfake candidates.


10 


Double Gaussian fits to the MC background and data e + MET
samples.
a) MC W > eν > γ_fake + MET
b) MC γ + jet
c) MC Z γ
d) data e + MET
e) RMS vs. Mean of the wrong vertex distribution




11 


(a) WrongVertex sideband region and signal region.
(b) The ratio of N(signal region) to N(wrongvertex sideband
region) vs. the fitted wrongvertex mean. We find that all samples
agree well with the prediction from the double Gaussian approximation
for a wide range of wrongvertex means. (MC and two electron datasets)

12 


For a number of MC datasets as well as two electron datasets
from data, we isolate wrongvertex (using generator quantities or the
electron track) and novertex events.
(a) If the characteristics of the true collisions are similar for wrong and novertex events, on average, the novertex and wrongvertex times dier only by a small geometrical factor in their timesoflight.
b) The novertex mean is as predictive of the N(signal
region)/N(wrongvertex sideband) ratio as the wrongvertex mean.
c)The fitted novertex RMS vs. the fitted novertex mean. We find that the novertex RMS is consistent with the assumption of 1.6 ns for all samples, regardless of the fitted novertex mean. (MC & 2 electron datasets)
d) The novertex mean is an excellent predictor of the N(signal region)/N(wrongvertex sideband) ratio as the wrong as the wrongvertex mean. Note that the line is not a fit, but rather the integrated ratio from the Gaussian assumption using the mean from the xaxis. 


13 


a) This shows the estimation of the wrongvertex mean using
the novertex distribution.
b) Exclusive γ + MET events across the full tcorr range,
signal region blinded.

14 


The likelihood methods predict the number of events in the
signal region for our pseudoexperiments as well as their uncertainty.
a) This figure shows the pull distribution for
pseudoexperiemnts generated using perfect Gaussians for right, wrong,
and novertex distributions and uniform distributions for the cosmics
contribution in the good and novertex samples. The pull distribution
has a mean very close to zero, which indicates a lack of bias. It has
an RMS very close to 1, which indicates that the fit uncertainty is
well estimated.
b) Uncertainty on background estimation.
c) For generated wrongvertex means from 0.0 to 0.8 ns, the
fit remains unbiased and the uncertainties remain well estimated.


15 


a) The final tcorr distribution for our data sample along with
the best fit values of the backgrounds from timing distributions for
right vertex, wrong vertex, and cosmics as a function of tcorr.
b) Results with expected backgrounds subtracted.
