Deep learning approaches to top FCNC couplings to photons at the LHC

TL;DR

It is shown that at the high-luminosity LHC, rare top branching ratios can be probed down to values as low as 10−6, highlighting the significant potential of attention-based architectures for improving the sensitivity to new physics signatures in top quark processes at colliders.

Abstract

We investigate the sensitivity of the LHC to flavour-changing neutral current interactions involving the top quark and a photon using a model-independent effective field theory framework, focusing on two complementary processes: single top production via $qg \to tγ$ and the rare decay $t \to qγ$ in top pair events. To enhance signal discrimination, we employ a range of deep learning classifiers, including multi-layer perceptrons, graph attention networks and transformers, and compare them against a traditional cut-based analysis. Our results demonstrate that attention-based architectures, in particular transformer networks, significantly outperform other strategies, yielding up to a factor of five improvement in the expected exclusion limits. In particular, we show that at the high-luminosity LHC, rare top branching ratios can be probed down to values as low as $10^{-6}$. Our results thus highlight the significant potential of attention-based architectures for improving the sensitivity to new physics signatures in top quark processes at colliders.

Figures (11)

• Figure 1: Representative tree-level Feynman diagrams for the FCNC production of a single SM-decaying top quark in association with a photon (left), and for the SM production of a top-antitop pair where one the tops decays in the SM way and the other through an FCNC process (right).
• Figure 2: Leading-order cross sections for single top production in association with a photon via $cg$ (or equivalently $\bar{c}g$) fusion (left), $\bar{u}g$ fusion (middle) and $ug$ fusion (right). The bands indicate theoretical uncertainties from scale and PDF variations summed in quadrature.
• Figure 3: Signal rates $\sigma \times \text{BR}$ for single top + photon production (left) and top pair production with an FCNC decay (right). The dotted, dashed and solid contours correspond to constant cross section values of 0.1, 1 and 10 fb respectively.
• Figure 4: Differential cross sections for representative kinematic variables used in the cut-based and MLP analyses of this study. Background contributions are stacked in ascending order of total cross section, while signal processes are overlaid as solid lines: single top production (grey) and top-pair production (red) for the benchmark coupling values $f_{tq}^\gamma = h_{tq}^\gamma = 3 \times 10^{-2}$. The top row displays the missing transverse energy (left), the transverse momentum of the signal photon (centre) and the transverse momentum of the signal lepton (right). The middle row shows the invariant mass of the jet-photon system (left), the transverse mass of the lepton-missing transverse momentum system (centre) and the scalar sum of jet transverse momenta (right). The bottom row presents azimuthal angle separations between the photon and the leading $b$-tagged jet (left), the photon and the missing transverse momentum (centre) and the lepton and the missing transverse momentum (right).
• Figure 5: Feature importance for the top 15 ranked input variables used in the MLP network. The $y$-axis shows the increase in model loss when each variable is randomly shuffled, averaged over 10 repetitions.