Light flavour mistag calibration for ATLAS $b$-jet identification algorithms

Many analyses in ATLAS rely on the identification of jets containing $b$-hadrons ($b$-jets) with high efficiency while rejecting more than 99\% of non-$b$-jets. Identification algorithms, called $b$-taggers, exploit $b$-hadron properties like their long lifetime. Recently developed ATLAS $b$-taggers using neural networks outperform previous $b$-taggers by a factor of two in terms of light jet rejection. Nevertheless, contributions from light jet mistags can be non-negligible in certain analyses phase spaces and a precise measurement of the light jet mistag rate in data and simulation to correct the rate in simulation is important. Due to the high light jet rejection of the $b$-taggers, the mistag rate cannot be measured directly but rather by means of a modified tagger, designed to decrease the $b$-jet efficiency while leaving the light jet response unchanged. This so-called "negative tag method" has been improved recently: uncertainties are reduced by constraining non-light flavour contribution with a data-driven method and the dominant systematic uncertainty has been reduced from 10-60\% to 5-20\% due to improved inner detector modeling. The method and a selection of results released recently to the ATLAS collaboration using $pp$ collisions at $\sqrt{s}=$ 13 TeV are presented.