Particle-level transformers for 95 GeV Higgs boson searches at future $e^+e^-$ Higgs factories

TL;DR

The paper targets a potential light Higgs near 95 GeV within the flipped N2HDM (N2HDM-F) at future $e^+e^-$ Higgs factories, focusing on Higgsstrahlung $e^+e^-\to ZS$ with $Z\to\mu^+\mu^-$ and $S\to\tau^+\tau^-$ or $b\bar b$. It combines a detailed MC study at CEPC-like conditions with a parameter-space scan of N2HDM-F, and introduces particle-level transformer architectures (ParT and MIParT) to exploit full event information. The ML-based analysis markedly improves the precision on the signal strength compared to a cut-based approach, achieving roughly $2.2$–$2.4\times$ better precision in $\tau^+\tau^-$ and about $1.4\times$ in $b\bar b$, and enabling 5$\sigma$ discovery and sub-percent measurements over large regions of parameter space (with specific thresholds $\mu_{\tau\tau}^{ZS}>1.6\times10^{-2}$, $\mu_{bb}^{ZS}>4.9\times10^{-3}$ for discovery and $\mu_{\tau\tau}^{ZS}>0.96$, $\mu_{bb}^{ZS}>0.14$ for 1% precision). The results indicate that particle-level transformers substantially enhance light-Higgs searches at future lepton colliders, and the gains are expected to extend to FCC-ee and ILC with similar methodology and assumptions.

Abstract

Motivated by several mild excesses around 95 GeV, we investigate the prospects for a light scalar $S$ produced via Higgsstrahlung, $e^+e^- \to Z(μ^+μ^-)S$, at future $e^+e^-$ Higgs factories. We take the CEPC as a benchmark, with a center-of-mass energy of $\sqrt{s}=240$ GeV and an integrated luminosity of $L=20~\mathrm{ab}^{-1}$. We focus on the decay modes $S\toτ^+τ^-$ and $S\to b\bar b$. To maximize sensitivity, we employ the particle-level transformer networks Particle Transformer (ParT) and its more-interactive variant MIParT, which exploit the features of all reconstructed objects and their correlations. For a representative signal benchmark, this approach improves the expected statistical precision on the signal strength by factors of 2.4 in the $τ^+τ^-$ channel and 1.4 in the $b\bar b$ channel compared to a cut-based analysis. Within the flipped Next-to-Two-Higgs-Doublet Model (N2HDM-F), the CEPC can measure the signal strength with a statistical precision down to 1.0% in the $τ^+τ^-$ channel and 0.68% in the $b\bar b$ channel using MIParT. It can achieve a $5σ$ discovery for $μ_{ττ}^{ZS}>1.6\times10^{-2}$ or $μ_{bb}^{ZS}>4.9\times10^{-3}$, and reach 1% precision for $μ_{ττ}^{ZS}>0.96$ or $μ_{bb}^{ZS}>0.14$. These gains are expected to qualitatively carry over to other future lepton colliders such as FCC-ee and the ILC. Our results demonstrate the potential of particle-level machine-learning techniques to strengthen light Higgs searches at future $e^+e^-$ Higgs factories.

Figures (13)

• Figure 1: surviving samples under the above constraints in the $\mu^{ggS}_{\tau\tau}$ versus $\mu^{ggS}_{\gamma\gamma}$ plane (left panel), the $\mu^{ggS}_{\gamma\gamma}$ versus $\mu^{ZS}_{bb}$ plane (middle panel), and $\mu^{ggS}_{\tau\tau}$ versus $\mu^{ZS}_{bb}$ plane (right panel) with colors indicating the fit result of $\chi^2_S$ and the gray samples having $\chi^2_S>7.82$.
• Figure 2: The branching ratios of the light Higgs $S$ to $b\bar{b}$, $c\bar{c}$, $\gamma\gamma$, $\tau^+\tau^-$, and $gg$ in the N2HDM-F as a function of $\alpha_1$, for the benchmark points with $\cos\alpha_2 = \sqrt{2}/2$ and $\tan\beta = 2$.
• Figure 3: The dominant Feynman diagrams of the signal and irreducible backgrounds in the $\tau^+\tau^-$ decay channel, where panel (a) corresponds to the signal, and panels (b), (c), and (d) represent the irreducible backgrounds.
• Figure 4: Surviving samples in the tan$\beta$ versus $\alpha _1$ plane with the colors indicating the square of the reduced coupling of $S$ to $Z$ boson (left panel), the cross section of the $b\bar{b}$ channel (middle panel), and the cross section of the $\tau^+\tau^-$ channel (right panel).
• Figure 5: Normalized event distributions for the signal and background in the recoil mass $M_{\mathrm{recoil}}$ (upper left panel) and the $\tau$-jet number N [$\tau$-jet] (upper right panel), along with the signal and background cross sections under different selections on these variables (lower left panel), as well as the corresponding measurement precision of the signal strength after applying these selections (lower right panel) in the $\tau^+\tau^-$ channel.