Search for Beyond the Standard Model physics with anomaly detection in multilepton final states in $pp$ collisions at $\sqrt{s}=13$ TeV with the ATLAS detector

Abstract

A model-agnostic search for Beyond the Standard Model physics is presented, targeting final states with at least four light leptons (electrons or muons). The search regions are separated by event topology and unsupervised machine learning is used to identify anomalous events in the full 140 fb$^{-1}$ of proton-proton collision data collected with the ATLAS detector during Run 2. No significant excess above the Standard Model background expectation is observed. Model-agnostic limits are presented in each topology, along with limits on several benchmark models including vector-like leptons, wino-like charginos and neutralinos, or smuons. Limits are set on the flavourful vector-like lepton model for the first time.

Figures (11)

• Figure 1: Example Feynman diagram for the pair production of VLL, denoted by $\psi_i$, in the flavourful model, with the VLLs produced via a (a, b) $W^*$ boson and (c) a $Z$ boson, with the decay of the VLLs into SM leptons and (a) two SM vector bosons, and (b, c) two scalars $S_{ji}$. Considering the decays of (a) $W^+ \to q \bar{q'}$ and $Z \to \ell_j^+ \ell_j^-$, and (b, c) $S^{(*)}_{ji} \to \ell_j^{-(+)} \ell_i^{+(-)}$, these processes result in multilepton final states. The subscripts $i$ and $j$ refer to the lepton generation, with allowed values $i=1,2$ and $j=1,2,3$.
• Figure 2: Comparison of background shapes for (a) $m_{\mathrm{T}} (4\ell, E_{\text{T}}^{\text{miss}}\xspace)$ in the 2Z 0b region and (b) $H_{\mathrm{T}}^{\mathrm{lep}}$ in the 1Z 0b 2SFOS region, considering various requirements on the anomaly score corresponding to the specified background rejection cuts. The last bin contains the overflow.
• Figure 3: Illustrative sketch of the control regions and (left) discovery or (right) benchmark regions in the (left) model-independent or (right) model-dependent search. For the model-independent scenario, the discovery regions correspond to low-anomaly score regions fitted together with a high-anomaly score discovery signal bin at a time. Each layer corresponds to the different fits with the same control regions and low-anomaly score regions, and varying signal bins. The <90% (<50%) label corresponds to the requirement on the anomaly score to be below the 90% (50)% background rejection point. For the model-dependent scenario, all benchmark regions are used simultaneously together with the control regions in a single profile likelihood fit to data, illustrated with one single layer.
• Figure 4: Comparison between data and the background prediction for the (a) $m_{\mathrm{T}} (4\ell, E_{\text{T}}^{\text{miss}}\xspace)$, (b) $m^{\mathrm{high}}(3\ell)$, (c) $m(Z)$, (d) $E_{\text{T}}^{\text{miss}}$, (e) $p_{\mathrm{T}} (Z)$, and (f) $N_\mathrm{jets}$ distribution in the (a, d) 2Z 0b, (b, e) 1Z 1b 2SFOS, and (c, f) 0Z 2SFOS region, after requiring the anomaly score to be below the 90% background rejection point. The background contributions after the likelihood fit to data ('post-fit') for the background-only hypothesis are shown as filled histograms. The 'tt+X' background component includes the $t\bar{t}Z$, and $t\bar{t}H$ processes. The 'HF $\ell$' ('LF $\ell$') background component refers to processes containing one non-prompt light lepton from heavy-flavour (light-flavour) hadron decays. The ratio of the data to the background prediction ('Bkg.') is shown in the lower panel. The 'Other' contribution is dominated by the $tWZ$ production. The size of the combined statistical and systematic uncertainty in the background prediction is indicated by the blue hatched band. The upward-pointing blue arrows indicate points for which the data-to-background ('Data/Bkg.’) ratio exceeds the vertical range of the figure. The last bin contains the overflow.
• Figure 5: Comparison between data and the background prediction for the (a) $N_\mathrm{jets}$, (b) $E_{\text{T}}^{\text{miss}}$, (c) $m(3\ell)$, and (d) $H_{\mathrm{T}}^{\mathrm{lep}}$ distribution in the (a) HFe CR, (b) HF$\mu$ CR, (c) LFe CR, and (d) the W$\gamma^*$/$t\bar{t}\gamma^*$ validation region. The background contributions after the likelihood fit to data ('post-fit') for the background-only hypothesis are shown as filled histograms. The ratio of the data to the background prediction ('Bkg.') is shown in the lower panel. The 'tt+X' background component includes the $t\bar{t}W$, $t\bar{t}Z$, and $t\bar{t}H$ processes. The 'Other' contribution in the $2\ell SS$ regions is dominated by the $t\bar{t}\xspace$ and $t\bar{t}W$ production, and in the $3\ell$ regions is dominated by the $ZZ$ + LF production. The size of the combined statistical and systematic uncertainty in the background prediction is indicated by the blue hatched band. The upward-pointing blue arrows indicate points for which the data-to-background ('Data/Bkg.’) ratio exceeds the vertical range of the figure. The last bin contains the overflow.