Efficiency corrections for a tagger for boosted $H\rightarrow b\bar{b}$ decays in $pp$ collisions at $\sqrt{s}=13$ TeV with the ATLAS detector

The identification of massive particles decaying into bottom-antibottom quark pairs at high transverse momenta in ATLAS is performed using a multivariate algorithm, denoted an $X \rightarrow b\bar{b}$ tagger. The hadronic decays are identified by reconstructing a large-$R$ ($R=1.0$) jet, associating variable radius subjets to the large-$R$ jet, and then assigning a flavour discriminant to each of the subjets using a flavour tagging algorithm. The flavour discriminants of up to three subjets are used as input to a feed-forward neural network. The $X \rightarrow b\bar{b}$ tagger algorithm is trained using Monte Carlo simulated events, thus the efficiency of the tagger to identify $X \rightarrow b\bar{b}$ events might differ between data and simulation. This document describes the performance of the $X \rightarrow b\bar{b}$ tagger and focuses on the techniques used to calibrate the response of the tagger in data, presenting calibration results using proton$-$proton collision data corresponding to 139 fb$^{-1}$ collected by the ATLAS detector at the Large Hadron Collider, operating at a centre-of-mass energy of $\sqrt{s} = 13$~TeV. The signal efficiency is calibrated using $Z (\rightarrow b\bar{b}) + jets$ and $Z(\rightarrow b\bar{b})\gamma$ events. The background efficiency is calibrated using $t \bar{t}$ events. Data-to-Monte-Carlo efficiency scale factors are measured as a function of the large-$R$ jet transverse momentum. In addition, the modelling of large-$R$ jet kinematics in Monte Carlo simulation is checked after the application of the $X \rightarrow b\bar{b}$ tagger using multijet events enriched in $g \rightarrow b\bar{b}$ splitting.