Estimation of backgrounds from jets misidentified as $τ$-leptons using the Universal Fake Factor method with the ATLAS detector

TL;DR

The paper introduces the Universal Fake Factor (UFF) method to estimate backgrounds from jets misidentified as $\tau$ leptons in ATLAS analyses. By substituting a single, data-driven combination of four FFs measured in dedicated fake-enriched regions, UFF obviates the need for many parallel determination regions and uses a template-fit to obtain the combination weights. The method is validated using Run 2 data, with background predictions agreeing with references within 15–35% depending on $p_T^{\tau}$ and tau decay prongness, and demonstrations in $W(\mu\nu)$ and $\mu 3j$ regions show robust performance. UFF offers a flexible, largely data-driven framework that can be validated in DRs and adapted to different tau-ID selections, enabling broader and faster background estimation for tau analyses in ATLAS.

Abstract

Processes with $τ$-leptons in the final state are important for Standard Model measurements and searches for physics beyond the Standard Model. The ATLAS experiment at the Large Hadron Collider observes $τ$-leptons produced in proton-proton collisions only through their decay products. Data analyses involving hadronically decaying $τ$-leptons face challenges due to backgrounds from jets misidentified as $τ$-leptons that are not modelled reliably by Monte Carlo simulations. Data-driven methods such as the fake-factor method allow such misidentified backgrounds to be predicted by measuring transfer factors, known as fake factors, in data from dedicated regions. This paper describes a refined technique for determining the fake factors, the Universal Fake Factor method. It evaluates the fake factors for a signal region by using fake factors from samples enriched in different sources of jets misidentified as $τ$-leptons (light-quark, gluon, $b$-quark, and pile-up jets). Each fake factor is calculated as a linear combination of fake factors measured in these different enriched samples. For the full Run 2 data set, the systematic uncertainty of the calculated fake factors, evaluated using $W(μν)$ enriched event sample, ranges from 15% to 35% depending on the $τ$-lepton's transverse momentum and charged-particle decay multiplicity.

Figures (15)

• Figure 1: The $q$/$g$/$b$/$p$ fake composition of the antiID $Z(\mu\mu)$, , $\mathrm{MJ\ hJVT}$, $\mathrm{MJ\ lJVT}$, $W(\mu\nu)$, and $\mu 3j$ subregions, with an inclusive selection in and prongness.
• Figure 2: FFs estimated in the $Z(\mu\mu)$, , $\mathrm{MJ\ hJVT}$ and $\mathrm{MJ\ lJVT}$ regions for (a) 1-prong (1p) fakes and (b) 3-prong (3p) fakes as a function of the fake's . The coloured bands display the statistical uncertainty of the FFs. The highest bin contains events with fake-$\Pgt_{\text{had-vis}}$ with of up to 300 $\text{Ge V}$.
• Figure 3: Templates of the q/g tagger score distributions in the 1-prong (1p) and $\pt \in (85, 100)~\text{Ge V}\xspace$ bin. They are determined in the $Z(\mu\mu)$, , $\mathrm{MJ\ hJVT}$ and $\mathrm{MJ\ lJVT}$ antiID subregions. For display purpose, the templates are normalized to unity. The bands display the statistical uncertainty of the four templates.
• Figure 4: (a) Results of the template fit in the 1-prong and $\pt \in (85, 100)~\text{Ge V}\xspace$ bin of the $W(\mu\nu)$ region. The black points represent the q/g tagger score distribution from the $W(\mu\nu)$ antiID subregion. The coloured histogram areas show the templates of the $Z(\mu\mu)$, , $\mathrm{MJ\ hJVT}$ and $\mathrm{MJ\ lJVT}$ antiID subregions and the real-$\tau$-lepton background in the antiID subregion of the SR. The four normalized templates are scaled by their corresponding post-fit normalization factors. The grey band displays the statistical uncertainty of the sum of the four templates and the real-$\tau$-lepton background. The post-fit normalization factors are $\mu_{Z(\mu\mu)\xspace} = 0.28 \pm 0.16$, $\mu_{\Pqt{}\Paqt} = 0.10 \pm 0.09$, $\mu_{\mathrm{MJ\ hJVT}\xspace} = 0.45 \pm 0.19$, $\mu_{\mathrm{MJ\ lJVT}\xspace} = 0.18 \pm 0.06$. (b) Breakdown plot of the post-fit $Z(\mu\mu)$, , $\mathrm{MJ\ hJVT}$, and $\mathrm{MJ\ lJVT}$ templates. Their sum is normalized to unity in each q/g tagger score bin so that each template contribution had the meaning of a fraction of the antiID $\Pgt_{\text{had-vis}}$ candidates in the bin.
• Figure 5: FFs for the (a) 1-prong (1p) $q$ fakes and (b) 3-prong (3p) $q$ fakes as a function of the fake-$\Pgt_{\text{had-vis}}$ in five regions of phase space, evaluated using the MC simulation and smoothed. The grey band displays the symmetrized envelope around the $Z(\mu\mu)$ FFs. The envelope is used to estimate the systematic uncertainty of the $Z(\mu\mu)$ FFs. The highest bin contains events with fake-$\Pgt_{\text{had-vis}}$ with of up to 300 $\text{Ge V}$.