Inclusive Flavour Tagging at LHCb

TL;DR

This paper introduces a deep neural network-based inclusive flavour tagging algorithm for neutral B mesons at LHCb, significantly improving tagging power and aiding precision measurements in particle physics.

Contribution

The paper presents a novel deep learning algorithm, DeepSets, for flavour tagging that outperforms existing methods by leveraging a comprehensive set of tracks in proton-proton collisions.

Findings

35% increase in tagging power for B^0 mesons

20% increase in tagging power for B_s^0 mesons

Enhanced precision for CP violation and mixing measurements

Abstract

A new algorithm based on a deep neural network, DeepSets, for tagging the production flavour of neutral $B^0$ and $B^0_s$ mesons in proton-proton collisions is presented. Exploiting a comprehensive set of tracks associated with the hadronization process, the algorithm is calibrated on data collected by the LHCb experiment at a centre-of-mass energy of $13$ TeV. This inclusive approach enhances the flavour tagging performance beyond the established same-side and opposite-side tagging methods. The observed gains in tagging power of $35\%$ for $B^0$ mesons and $20\%$ for $B_s^0$ mesons relative to the combined performance of the existing LHCb flavour-tagging algorithms offer significant benefits for precision measurements of $C\!P$ violation and mixing in the neutral $B$ meson systems.