Hadronic decay branching ratio measurements of the Higgs boson at future colliders using the Holistic Approach

TL;DR

The study demonstrates that a holistic, particle-cloud approach using ParticleNet can substantially improve the precision of Higgs hadronic decay measurements at future $e^+e^-$ colliders, with CEPC as a benchmark at $L=20~\mathrm{ab}^{-1}$. By treating each event as a rich, inclusive feature set and applying channel-specific pre-selection where needed, the authors achieve relative uncertainties approaching the statistical limits for $H\to b\bar{b}, c\bar{c}, gg,$ and $WW^{*}$, while $H\to ZZ^{*}$ remains more challenging. The two primary $ZH$ channels, $Z(\mu^{+}\mu^{-})H$ and $Z(\nu\bar{\nu})H$, show complementary strengths, and their combination yields projected uncertainties as low as $0.11\%$ for $H\to b\bar{b}$ and $0.56\%$ for $H\to c\bar{c}$, representing a factor of $2$–$3$ improvement over Snowmass baselines. A scaling analysis reveals favorable data-size dependence with near-statistical-limit performance for several decays, though some channels exhibit irreducible or generator-dependent limitations, underscoring the importance of systematic control and cross-generator robustness. Overall, the holistic approach offers a powerful path to precision Higgs physics at future Higgs factories and can be extended to rare decays and BSM signatures.

Abstract

Accurately measuring the properties of the Higgs boson is a primary objective of the high-energy frontier. Using a holistic approach that incorporates inclusive information from reconstructed particles, we estimate the relative statistical uncertainty for the Higgs decay modes $H \to b\bar{b}$, $c\bar{c}$, $gg$, $WW^*$, and $ZZ^*$ at the Circular Electron-Positron Collider (CEPC) operating as a Higgs factory with an integrated luminosity of 20 ab$^{-1}$. In the $Z(μ^+μ^-)H$ and $Z(ν\barν)H$ channels, the relative statistical uncertainties for these decay modes are projected to range from 0.38% to 6.56% and 0.17% to 2.55%, respectively. Notably, the projected precisions for $H \to b\bar{b}$, $c\bar{c}$, $gg$, and $WW^*$ closely approach the statistical limits, while the $H \to ZZ^*$ channel remains approximately a factor of two to four from the statistical limits. Compared to the CEPC Snowmass results, this holistic approach improves measurement precision by a factor of two to three. The scaling behavior, specifically the dependence of the anticipated accuracy on the training dataset size, is also analyzed.

Figures (10)

• Figure 1: Overview of the AURORA detector performance. Left: The reconstructed invariant mass distributions for hadronic systems in $WW \to \mu\nu q\bar{q}$, $ZZ \to \nu\nu q\bar{q}$, and $ZH \to \nu\nu gg$ processes at $\sqrt{s}=240$ GeV. Right: Confusion matrix for particle identification (PID) reconstruction, illustrating the identification efficiency for charged particles, photons, and neutral hadrons. The detector concept achieves near-universal particle identification performance.
• Figure 2: Confusion matrix for the $Z(\mu^{+}\mu^{-})H$ classification task using the holistic approach, trained on 1M events. Matrix elements are normalized to a unit in each row.
• Figure 3: Score distributions and relative statistical uncertainties for $H \to b\bar{b}$, $H \to c\bar{c}$, $H\to gg$, $H\to WW^{*}$, and $H \to ZZ^{*}$ processes in the $Z(\mu^{+}\mu^{-}) H$ analysis using holistic approach trained on 1M events. The purple curve denotes the signal, while the colored histograms show the backgrounds. The red solid curve gives the statistical uncertainty versus the score cut, and the vertical red dashed line marks the optimal score cut.
• Figure 4: Confusion matrix in the $Z(\nu\bar{\nu}) H$ channel using the holistic approach, trained on 1M events. Matrix elements are normalized to a unit in each row.
• Figure 5: Score distributions and relative statistical uncertainties for $H \to b\bar{b}$, $H \to c\bar{c}$, $H\to gg$, $H\to WW^{*}$, and $H \to ZZ^{*}$ processes in the $Z(\nu\bar{\nu}) H$ analysis using holistic approach trained on 1M events. The purple curve denotes the signal, while the colored histograms show the backgrounds. The red solid curve gives the statistical uncertainty versus the score cut, and the vertical red dashed line marks the optimal score cut.