Metodi di ripesamento e applicazioni all’esperimento LHCb - Reweighting methods and applications at the LHCb experiment

The first application of a novel approach to Monte Carlo reweighting at the LHCb experiment is reported. The typical approach to reweighting is also used. The novel approach consists in a machine learning algorithm. The performances of the two methods are evaluated in two case of interest. The first application herein reported is the reweighting of the BDT’s input variables for the B0s —> Ds-*Pi+ decay channel. A comparison of the variations of the reduced chi-squared between the unweighted and weighted distributions is employed to evaluate the performances of the two reweighting methods. Results show that the novel approach to reweighting generally increases the agreement between Monte Carlo and data distributions with respect to the typical approach. It is however found that some distributions are unaffected by the reweighing procedures. These results have been considered by the LHCb official analysis and the variables corresponding to those distributions have been discarded from the next training iteration of the BDT. In order to further improve the agreement between Monte Carlo and Data, the official analysis increased the cut on the Ds* photon radiative decay transverse momentum. The second application herein reported is the reweighting of distributions that are then considered for the flavor tagging of the B mesons produced at the LHCb detector. Since flavour tagging algorithms are calibrated on a channel other than the one considered in this work, this analysis checks whether the taggers are portable and if reweighting has any impact on the quality of the portability. This is the first time that this procedure is applied in the B0s —> Ds*-Pi+ decay channel at LHCb. Results show that reweighting has little impact on the portability of the taggers and that the novel approach shows a slight edge over the typical one.