ATLAS tau, jets, and missing-momentum squark and gluino search

ATLAS searches for pair-produced gluinos or first- and second-generation left-handed squarks in final states with hadronically decaying tau leptons, jets, and missing transverse momentum. The dataset combines $140{\mathrm{f}\mathrm{b}}^{-1}$ of $13\mathrm{T}\mathrm{e}\mathrm{V}$ proton-proton data from 2015–2018 with $51.8{\mathrm{f}\mathrm{b}}^{-1}$ at $13.6\mathrm{T}\mathrm{e}\mathrm{V}$ from 2022–2023. The three orthogonal channels require (1) exactly one hadronically decaying tau and no electron or muon, (2) exactly one hadronic tau and at least one electron or muon, or (3) at least two hadronic taus. Events are selected using missing-transverse-momentum triggers with thresholds between 70 and 115 GeV; their reported efficiencies are typically 80–90% and about 40% in the most compressed signal region.

The interpretation adopts $R$-parity-conserving simplified models, where $R=(-1{)}^{3(B-L)+2s}$ is a conserved $Z_2$ quantum number, $B$ and $L$ denote baryon and lepton number, $s$ denotes spin, Standard Model particles are even, and their superpartners are odd. The gluino may decay through a virtual squark; squark decays proceed through the lightest chargino or next-to-lightest neutralino. The tau-slepton and tau-sneutrino are co-next-to-lightest supersymmetric particles. The lightest neutralino is assumed bino-like, meaning predominantly the superpartner of the hypercharge gauge boson, and the lightest chargino and next-to-lightest neutralino are wino-like, meaning predominantly superpartners of weak gauge bosons. The source fixes $m({\tilde{\chi }}_1^{\pm })=m({\tilde{\chi }}_2^0)=\frac{m(\tilde{g}\text{ or }\tilde{q})+m({\tilde{\chi }}_1^0)}{2},m(\tilde{\tau })=m({\tilde{\nu }}_{\tau })=\frac{m({\tilde{\chi }}_1^{\pm })+m({\tilde{\chi }}_1^0)}{2}\mathrm{.}$ The chargino and next-to-lightest-neutralino branching fractions to tau-slepton or tau-sneutrino modes are set to $50\mathrm{\%}$, other supersymmetric couplings are set to zero, and the narrow-width approximation treats intermediate resonances as on-shell because their widths are assumed small relative to their masses. These choices define the reported simplified-model contours.

Measurement, operational observable, and calibration model

ATLAS tau search, model and dataset sections

The reported observables include missing transverse momentum, defined as the magnitude of the negative vector sum of reconstructed electrons, muons, hadronic taus, jets, and a track-based soft term, along with channel-specific kinematic variables. Jets are reconstructed with the anti-$k_t$ algorithm at radius $R=0.4$ and calibrated using simulation and in-situ measurements. Hadronic tau candidates are seeded by anti-$k_t$ jets, calibrated with a dedicated visible-tau energy-scale procedure, and selected with one or three tracks and a medium identification working point. Light leptons and jets have source-defined identification, overlap-removal, and kinematic selections. The analysis estimates jets misidentified as tau leptons with a data-driven fake-factor method and uses both cut-and-count and machine-learning strategies.

For each strategy the collaboration fits selected binned data with a likelihood containing Poisson event-count terms and constrained systematic and background nuisance parameters; source-stated correlations are retained. The statistic is a profile likelihood ratio, formed by maximizing over nuisance parameters at fixed signal-model parameters and comparing with the unrestricted fit. ATLAS uses the asymptotic approximation after validating it against simulated ensembles. It excludes a tested signal point when the source-defined $C{L}_s$ ratio falls below $0.05$; $C{L}_s$ is the signal-plus-background tail probability divided by the corresponding background tail probability under the same source-defined test ordering, not a posterior probability or exact coverage statement.

Likelihood functions for observed data

Profile likelihood ratios with nuisance parameters

Source-defined CLs exclusion ratio and its limits

ATLAS tau search, reconstruction, statistical model, and asymptotic validation

The collaboration reports no statistically significant overall deviation from the background-only expectation. Its most compatible machine-learning benchmark, labelled GG(1300,1200), has a local $p$-value $0.00058$, quoted as $3.2$ standard deviations; this is a local value at the tested benchmark, not an overall search significance. The observed $95\mathrm{\%}$ $C{L}_s$ machine-learning contours reach gluino masses up to $2.25\mathrm{T}\mathrm{e}\mathrm{V}$ and lightest-neutralino masses up to $1.35\mathrm{T}\mathrm{e}\mathrm{V}$ for gluino models; for squark models, the machine-learning lightest-neutralino reach is $0.85\mathrm{T}\mathrm{e}\mathrm{V}$. The cut-and-count contours reach gluino masses up to $2.3\mathrm{T}\mathrm{e}\mathrm{V}$ and squark masses up to $1.7\mathrm{T}\mathrm{e}\mathrm{V}$ for models with a $0.1\mathrm{T}\mathrm{e}\mathrm{V}$ lightest-neutralino.

Null-calibrated tail probability and its interpretation

CLs ratio used for the reported exclusions

ATLAS tau search, fit results and observed exclusion contours

Wasserman, statistical interpretation of p-values, full textbook §10.2

The fit includes the collaboration's statistical, detector, background, and theoretical nuisance constraints with source-reported correlations. The reported integrated-luminosity uncertainties are $0.83\mathrm{\%}$ for the 13 TeV sample and $2.0\mathrm{\%}$ for the 13.6 TeV sample. The theory cross-section uncertainty is displayed as a contour variation rather than profiled as an experimental nuisance. The observed contour uses an asymptotic approximation checked against simulated ensembles. This record does not reprocess collision events or independently reconstruct the likelihood, object calibrations, or nuisance covariance.

Confidence sets and their coverage property

ATLAS tau search, systematic uncertainties and validation

The local excess, $C{L}_s$ exclusions, and mass reaches are conditional on the three selected channels, the two analysis strategies, and the particular $R$-parity-conserving simplified spectra, midpoint mass relations, branching fractions, and narrow-width assumptions stated above. They do not rule out supersymmetry as a framework, establish a supersymmetry-breaking vacuum, or constrain spectra and decay patterns outside those parameterizations.

Supersymmetry-breaking and source-qualified collider interpretation

ATLAS tau search, simplified-model interpretation and conclusion