Search for New Phenomena in Two-Body Invariant Mass Distributions Using Unsupervised Machine Learning for Anomaly Detection at $\sqrt{s}=13\mathrm{TeV}$ with the ATLAS Detector

Searches for new resonances are performed using an unsupervised anomaly-detection technique. Events with at least one electron or muon are selected from $140{\mathrm{fb}}^{-1}$ of $pp$ collisions at $\sqrt{s}=13\mathrm{TeV}$ recorded by ATLAS at the Large Hadron Collider. The approach involves training an autoencoder on data, and subsequently defining anomalous regions based on the reconstruction loss of the decoder. Studies focus on nine invariant mass spectra that contain pairs of objects consisting of one light jet or $b$ jet and either one lepton $(e,\mu )$, photon, or second light jet or $b$ jet in the anomalous regions. No significant deviations from the background hypotheses are observed. Limits on contributions from generic Gaussian signals with various widths of the resonance mass are obtained for nine invariant masses in the anomalous regions.

Figure 1

Distributions of the anomaly score from the AE for data and five benchmark BSM models. Their legends, from top to bottom, are: (1) charged Higgs boson production in association with a top quark, $tb{H}^{+}$ with $H^{+}\to t\overline{b}$; (2) a Kaluza-Klein gauge boson, $W_{\mathrm{KK}}$, with the SM $W$ boson and a radion $\varphi$; (3) a $Z^{\prime }$ boson decaying to a composite lepton $E$ and $\ell$, with $E\to Z\ell$ with a mass of 0.5 TeV; (4) the SSM $W^{\prime }\to W{Z}^{\prime }\to \ell \nu q\overline{q}$; (5) a simplified dark-matter model with an axial-vector mediator $Z^{\prime }\to q\overline{q}$, where one of the quarks radiates a $W$ boson decaying to $\ell \nu$. The BSM predictions represent the expected number of events from $140{\mathrm{fb}}^{-1}$ of data for heavy particle ($H^{+}$, $W_{\mathrm{KK}}$, $Z^{\prime }$, $W^{\prime }$, and $Z^{\prime }$, respectively) masses around 2 TeV. The distributions for the BSM models are smoothed to remove fluctuations due to low MC event counts. The vertical lines indicate the start of the three anomaly regions (ARs). The labels of the three ARs indicate the visible cross section for hypothetical processes yielding the same number of events as observed in the $140{\mathrm{fb}}^{-1}$ dataset. The AE is applied to preselected events without any requirements on invariant mass distributions.

Figure 2

Invariant mass distributions of $\mathrm{jet}+Y$ for $m_{jY}>0.3\mathrm{TeV}$ in the 10 pb AR along with the fit of Eq. (1). The fits are represented by the lines, while the associated statistical uncertainties are indicated by the shaded bands. The lower panels show the bin-by-bin significances of deviations from the fit, calculated as $(d_i-f_i)/{\delta }_i$, where $d_i$ is the data yield, $f_i$ is the fit value, and ${\delta }_i$ is the data uncertainty in the $i$th bin.

Figure 3

Values of $\mathrm{\Delta }Z$ for the discovery sensitivity, as defined in the text, as a function of the invariant mass $m$. The nine invariant mass distributions are calculated in the 10 pb AR. Positive percentages indicate improvements in sensitivity. Horizontal dashed lines are drawn at 100% and 200% to guide the eye. The five benchmark BSM models are (1) charged Higgs boson production in association with a top quark, $tb{H}^{+}$ with $H^{+}\to t\overline{b}$; (2) a Kaluza-Klein gauge boson, $W_{\mathrm{KK}}$, with the SM $W$ boson and a radion $\varphi$; (3) a $Z^{\prime }$ boson decaying to a composite lepton $E$ and $\ell$, with $E\to Z\ell$; (4) the SSM $W^{\prime }\to W{Z}^{\prime }\to \ell \nu q\overline{q}$; (5) a simplified dark-matter model with an axial-vector mediator $Z^{\prime }\to q\overline{q}$, where one of the quarks radiates a $W$ boson decaying to $\ell \nu$. The multiple markers shown for the composite-lepton model at the same invariant mass values correspond to different composite lepton ($E$) masses between 0.25 and 3.5 TeV. The center positions of the markers are set to the masses of the corresponding heavy particles.

Figure 4

The 95% C.L. upper limits on the cross section times acceptance ($A$), efficiency ($\epsilon$), and branching ratio ($B$) for Gaussian-shaped signals with different signal widths. The limits are calculated for events with at least one lepton with $p_{\mathrm{T}}^{\ell }>60\mathrm{GeV}$ in the 10 pb AR. Two width hypotheses are shown: ${\sigma }_X/m_X=0$ and ${\sigma }_X/m_X=0.15$. In both cases, the detector resolution for jets is included in the simulation of signal samples. The $\pm 1\sigma$ and $\pm 2\sigma$ bands around the expected limit are shown for ${\sigma }_X/m_X=0$ signals. Mass points are spaced 5% apart, relative to the preceding point, starting at 0.3 TeV.