Otimização dos algoritmos univariados e bivariados aplicados à identificação de elétrons no experimento ATLAS

Abstract

The electron identification is of fundamental importance for the ATLAS physics groups due to the presence of these in the final process of interest particles decay. In particle physics environment, the occurrence probability of relevant electrons to the proposed studies are very low compared to particles considered background, requiring ATLAS performance groups algorithms with identification efficiency index and background rejection each time better. In this dissertation, the methods applied in the electron identification in ATLAS experiment will be reviewed and possible optimizations will be evaluated from the data produced by the ATLAS experiment. Concentrated in the offline context, the work reproduces the method based on Likelihood and proposes an improvement with the use of multivariate technique known as MKDE (Multivariate Kernel Density Estimation), capable of mitigate the error inserted in consideration of dependence between discriminating variables entered by the method Likelihood currently in use by ATLAS. Initially, this work proposes to implement the method Likelihood in use, which is based on univariate density used in the reconstruction of the joint density of the discriminant variables, and to study the possible impact of parameters related to the implementation of univariate densities estimation algorithm. This method is then compared with the ATLAS standard method known as e/γ. In a second step the implementation of the MKDE is inserted through a direct comparison to the univariate method.