Single-top s channel cross-section measurement with the ATLAS detector

The work reported in this thesis is aimed at measuring with the ATLAS detector the only mechanism of top quark electroweak production which has not yet been observed at the Large Hadron Collider: the s-channel. Its interest relies in the confirmation of the Standard Model predictions but, as well, in the possibility of constraining several new physics scenarios when comparing the s-channel cross section with the one of the other single top production modes (the Wt and t-channel). After a general introduction upon top physics and the experimental setup employed for the detection and the reconstruction of the physics objects, we present two analyses realized with proton-proton collisions collected by the ATLAS detector at a center of mass energy of 7 and 8 TeV. Since the s-channel production is characterized by a very low purity and, at the same time, is not favoured at the LHC because it proceeds via quark-antiquark annihilation, a multivariate approach is applied in both cases to discriminate the signal. For the 7 TeV study, based on an integrated luminosity of 4.7 fb^-1, two boosted decision trees classifiers are optimized against the main sources of background; a cut realized on one of such discriminating variables implements a tight event selection, after which the second output distribution is used for a maximum likelihood fit to the data. This allows to set an upper limit on the s-channel cross section of 21.5 pb (14.3 pb expected), corresponding to a significance of 0.6 standard deviations (0.8 expected). The 8 TeV analysis, based on an integrated luminosity of 20.3 fb^-1, has a simpler strategy: only one boosted decision trees classifier is trained to isolate signal after a preliminary cut-based selection; its output distribution is then fitted to the data with the same approach than at 7 TeV, taking into account the systematic and statistical uncertainty via the generation of pseudo-experiments. The significance of the s-channel measurement being lower than 3 σ (1.3 standard deviation observed, 1.4 expected), only a 95% C.L. limit can be set on the cross section; the result is however improved and corresponds to 14.6 pb (7.9 pb expected). (author)