Two New Developments on the Statistical Treatment of Flavour Tagging Uncertainties in ATLAS

The document introduces two new methods on the implementation of flavour tagging uncertainties in ATLAS physics analyses. In order to reduce the number of flavor-tagging calibration uncertainties, the physics analyses use an eigenvector decomposition approach. However, the resulting flavour tagging eigenvectors are in general not the same across flavour tagging selections, so the uncertainties can not be directly correlated in combination analyses. A new method, called eigenvector recomposition, has been designed to overcome this problem. This proceeding describes the method and gives practical examples about its usage in physics analyses, focusing on the VH, H → bb analysis. The second development involves the flavour tagging uncertainties in analyses with high-transverse momentum jets. The in-situ calibration of the flavour tagging uncertainties is computed using events with jet-p_T spectra up to 140-250 GeV and used through all the jet-p_T spectrum. Therefore, at higher transverse momenta the calibration needs dedicated extrapolation uncertainties in order to account for possible deviations from the central value. The second part of the document describes the method used to extract these extrapolation uncertainties starting from Z’ simulated events.