Measurement of the singly Cabibbo-suppressed decay $Λ_c^+\to pη'$ with Deep Learning

TL;DR

This work reports the first precise measurement of the singly Cabibbo-suppressed decay $\Lambda_c^+ \to p\eta'$ using a single-tag approach with a Transformer-based deep-learning classifier to suppress backgrounds in BESIII data ($4.5\,\mathrm{fb}^{-1}$ at $\sqrt{s}=4.600$–$4.699$ GeV). The analysis yields a relative branching fraction $\mathcal{B}(\Lambda_c^+ \to p\eta')/\mathcal{B}(\Lambda_c^+ \to p\omega)=0.55\pm0.22_{stat.}\pm0.05_{syst.}$ and observes the $\Lambda_c^+ \to p\eta'$ signal with $3.4\sigma$ significance, achieved through a simultaneous fit across seven energy points and a DL-based background suppression that retains ~40% of signal. With the world-average $\mathcal{B}(\Lambda_c^+ \to p\omega)$, the absolute $\mathcal{B}(\Lambda_c^+ \to p\eta')$ is determined as $(6.15\pm2.45_{stat.}\pm0.53_{syst.}\pm1.16_{ref.})\times10^{-4}$, consistent with Belle and prior BESIII results. The measurement provides valuable input for SU(3)-flavour and topological-model descriptions of charmed-baryon decays and demonstrates the efficacy of Transformer-based DL in high-energy physics analyses.

Abstract

Using $4.5$ fb$^{-1}$ of $e^+e^-$ collision data collected with the BESIII detector at center-of-mass energies from 4.600 to 4.699 GeV, we report a measurement of the singly Cabibbo-suppressed decay $Λ_c^+ \to pη'$ with the single-tag method. To effectively distinguish the signal from the large backgrounds, we exploit a deep-learning classifier built on a Transformer-based neural network. Extensive validation and uncertainty quantification are carried out. The $Λ^+_c\to pη'$ signal is observed with a statistical significance of $3.4 σ$. The ratio of branching fractions of $\mathcal{B}{Λ^+_c\to pη'}/\mathcal{B}{Λ^+_c\to pω}$= $0.55\pm 0.22_{\rm{stat.}} \pm 0.05_{\rm{syst.}}$ is obtained, where the first uncertainty is statistical and the second systematic.

Figures (4)

• Figure 1: Topological diagrams of the $\Lambda_c^+\to p \eta'$ decay.
• Figure 2: Distributions of $M_{\rm{BC}}$ for (a) $\Lambda_c^+\to p \eta'$ and (b) $\Lambda_c^+\to p \omega$ after preselection before implementing deep learning. The black points with error bars represent data and the pink shaded histograms represent the signal components. The purple shaded histograms denote the hadronic backgrounds, while the blue shaded histograms correspond to the $\Lambda_c^+\bar{\Lambda}_c^-$ non-signal backgrounds. In (b), the magenta shaded histograms denote the $\Lambda_c^+\bar{\Lambda}_c^-$ non-$\omega$ backgrounds, while the blue shaded histograms show the remaining $\Lambda_c^+\bar{\Lambda}_{c}^{-}$ contributions.
• Figure 3: Simultaneous fit results for (a) $\Lambda_c^+\to p \eta'$ and (b) $\Lambda_c^+\to p \omega$ after the DNN-classifier. The points with error bars denote data, the blue solid lines represent the total fit results, and the purple dashed lines represent the hadronic background components. The red dashed areas represent the signal components. In (a), the magenta dashed line represents the $\Lambda_c^+\bar{\Lambda}_{c}^{-}$ background components. In (b), the magenta dashed line denotes the non-$\omega$ component, and the blue region represents the residual backgrounds from $\Lambda_c^+\bar{\Lambda}_{c}^{-}$.
• Figure 4: Comparison of our $\mathcal{B}(\Lambda_c^+\to p \eta')$ result (red point with error bar, obtained using the world average $\mathcal{B}(\Lambda_c^+\to p \omega)$ from the PDG as input) with the previous theoretical predictions (black points with error bars) and experimental measurements (blue points with error bars) in time order. Note that the previous BESIII results were obtained using a DT method.