At the Tevatron, protons and antiprotons collide with a center-of-mass energy
of 1.8 TeV. In this energy range the dominant source of top quarks is the 
production of $t\bar{t}$ pairs via quark-antiquark annhilation and gluon-gluon 
fusion. We present our analyses to determine the 
mass of the top quark reconstructed through the ``lepton+jets'' decay channel
in the 106 $\pm$ 4.1 pb$^{-1}$ of data collected by the Collider Detector at 
Fermilab (CDF) from 1992-1996. 

In the past, the top mass was obtained by comparing the observed 
kinematic features of top events to those predicted for different top quark 
masses. These distributions are known as $templates$. While 
any kinematic variable, which exhibits sensitivity to the top mass can be used 
to calculate the mass of the top quark, the lowest statistical uncertainty is 
achieved by reconstructing the top mass from the decay products of the 
$t\bar{t}$ pair. It was also observed that it is possible to obtain a better 
estimate of the top mass by fitting the different templates to a smooth 
function. Previously, we have used a combination of a Gaussian and a Gamma 
function to fit the distributions. In this analysis, we find that using a
Neural Network (NN) to fit the distribution gives slightly better results, and 
that the NN fitting technique is applicable to any kinematic variable.  
Following this recipe the top mass is measured to be 
$177.9 \pm 4.7(stat.) \pm 4.6(syst.)$ GeV/c$^2$ when using the reconstructed 
mass, $M_{rec}$. When we use the total transverse energy of the events, $\Ht$, 
the mass of the top quark is found to be 
$204.4 \pm 9.2(stat.) \pm  9.2(syst.)$ GeV/c$^2$. 

As noted, there are different kinematic variables that can be used to 
calculate the top mass. A Neural Network provides a simple and elegant way 
of combining all of these variables which have mass information. The idea is 
that combining the information from more than one kinematic variable would
result in a more accurate measurement of the top mass. Therefore, this NN based 
technique uses a combination of $M_{rec}$, $H_T$, the invariant mass of the 
$t\bar{t}$ system, $M_{t\bar{t}}$, and the sum of the $P_T$'s of the two 
leading jets, $P_T(1) + P_{T}(2)$. These variables were chosen because they 
exhibit the greatest mass dependence. The Neural Networks attempt to classify 
the events as $t\bar{t}$ signal or background. For each event the NN provides 
a set of probabilities that it has come from any of the top masses used in this 
analysis, as well as background. Using this information we construct a 
discrete likelihood function from which the top mass is calculated to be 
$181.9 \pm 5.1 (stat.) \pm 5.2(syst.)$ GeV/c$^2$. 
 



Posted to /cdf/pub/thesis/cdf6216_topMass_NN_thesis.ps
    http://www-cdf.fnal.gov/thesis/cdf6216_topMass_NN_thesis.ps

Posted by csanchez@cdfsga