Speaker
Description
The top quark is one of the fundamental fermions of the Standard Model, and is observed in the highest energy collisions. Our focus is the $t \bar t$ pair, which is produced through strong interaction in two cases: from gluon fusion ($gg$) or quark-antiquark annihilation ($q \bar q$). Different production channels lead to pairs with different characteristics: one example is the $t \bar t$ spin state, which near the production threshold presents higher correlations in the case of a $gg$ event. A study that proposes to study the entity of such correlations can thus benefit from a way to discriminate pairs on the basis of their production channels.
This work has therefore the purpose of obtaining, through the use of multivariate analysis methods, a tool to select events in such way. Multivariate algorithms are often used to separate a signal from a background that pollutes the sample; in this case for the signal we choose $gg$ events, while for the background we select to $q \bar q$ events. Such a problem is called a $\textit{classification problem}$.
We thus studied the performance of some classifiers, using the distributions of some variables associated to the $t \bar t$ production process. Then we selected the best performing algorithm evaluating its efficiency in selecting signal events and rejecting background ones. The chosen classifier turns out to be the $\textit{Boosted Decision Trees}$, which allows to obtain a sample of purity 0.92, starting from an initial purity of 0.81, at the cost of a reduced efficiency of 0.74.
| Title | Multivariate analysis to discriminate top quark pair production channels at LHC |
|---|
