Search for decays of the Higgs boson into scalar particles decaying into four or six $b$-quarks using $pp$ collisions at $\sqrt{s}= 13\,\mathrm{TeV}$ with the ATLAS detector

Technical Summary: Search for Decays of the Higgs Boson into Scalar Particles Decaying into Four or Six $b$-Quarks

Problem and Motivation
Following the discovery of the Higgs boson ($H$) with a mass near 125 GeV, characterizing its properties remains a primary objective of the LHC program. While the Standard Model (SM) predicts specific decay modes, the Higgs boson's small total width ($\Gamma_H \approx 4$ MeV) makes it highly sensitive to new physics. Even small couplings to Beyond the Standard Model (BSM) particles can result in sizable branching fractions for exotic decays. This paper investigates scenarios where the Higgs boson decays into new light scalar or pseudoscalar particles, referred to as $a$-bosons. These scenarios arise in various theoretical frameworks, including extended Higgs sectors, dark matter models, and theories of neutral naturalness.

The analysis focuses on three specific decay topologies:

1. $H \to 2a \to 4b$: The Higgs decays into two identical scalars, each decaying into a $b\bar{b}$ pair.
2. $H \to a_1 a_2 \to 4b$: The Higgs decays into two scalars of different masses ($m_{a_1} < m_{a_2}$), both decaying into $b\bar{b}$.
3. $H \to a_1 a_2 \to 3a_1 \to 6b$: The heavier scalar ($a_2$) decays into two lighter scalars ($a_1$), resulting in a final state of six $b$-quarks.

The search targets the associated production of the Higgs boson with a $Z$ boson ($ZH$), where the $Z$ boson provides a trigger handle and suppresses top-quark backgrounds. The analysis covers the mass range $12 < m_a < 60$ GeV using proton-proton collision data at $\sqrt{s} = 13$ TeV.

Methodology
The analysis utilizes 140 fb$^{-1}$ of data collected by the ATLAS detector during Run 2 (2015–2018). To address the challenges posed by the low transverse momentum of $b$-quarks in these decays and the high multiplicity of the final state, the analysis employs several novel reconstruction techniques and categorizes events based on the $Z$ boson decay mode ($2\ell$ for charged leptons, $0\ell$ for neutrinos) and the $b$-quark multiplicity.

• Object Reconstruction and Heavy-Flavor Tagging:

  • Merged Regime ($m_a \lesssim 20$ GeV): For low-mass $a$-bosons, the decay products are highly collimated. A dedicated algorithm, DeXTer, is used to identify "double $b$-quark" jets ($B$-jets) by reconstructing secondary vertices within large-$R$ track-jets and utilizing deep-set neural networks to distinguish merged $b\bar{b}$ hadronization from single $b$-jets.
  • Resolved Regime ($m_a \gtrsim 25$ GeV): Standard $b$-tagging algorithms (DL1r) identify individual $b$-jets.
  • Soft Secondary Vertices: To enhance sensitivity to high-multiplicity final states, soft secondary vertices reconstructed outside of jets (soft-v) are utilized to capture $b$-hadron decays that do not form energetic jets.
  • $b$-Objects: A unified concept of "$b$-objects" is introduced, encompassing $B$-jets (yielding two $b$-objects), standard $b$-jets, and soft vertices.

• Event Selection and Categorization:

  • $2\ell$ Channel: Events with two same-flavor, opposite-charge leptons are selected. A quadruplet selection Neural Network (NN) is trained to identify the optimal combination of four $b$-objects (from $B$-jets, $b$-jets, or soft-v) consistent with the signal hypothesis. This NN is invariant under permutations of $b$-objects and $a$-bosons.
  • $0\ell$ Channel: Events with large missing transverse momentum ($E_T^{miss}$) are selected. This channel is used exclusively for the $H \to 2a \to 4b$ search. Events are categorized by the number of $B$-jets and $b$-jets.
  • $6b$ Search: For the $H \to a_1 a_2 \to 6b$ topology, full reconstruction is rare. Instead, events are categorized based on the multiplicity of reconstructed $b$-objects without attempting to identify all six.

• Background Estimation and Discrimination:

  • Main backgrounds include $Z$+jets and $t\bar{t}$ production. Data-driven corrections are applied to MC simulations to account for mismodeling of heavy-flavor jet multiplicities and kinematic properties (e.g., $H_T$ and $p_T^Z$).
  • For the $0\ell$ channel, the QCD multijet background is estimated using an ABCD method based on the azimuthal separation between $E_T^{miss}$ and $a$-candidates.
  • Signal discrimination is performed using a Boosted Decision Tree (BDT) trained on kinematic variables such as reduced masses ($m_{red}$), angular separations, and the quadruplet selection NN score.

Key Contributions

1. Novel Reconstruction Techniques: The application of the DeXTer algorithm for low-mass merged $B$-jets and the inclusion of soft secondary vertices significantly improve sensitivity in the low-mass regime ($m_a < 20$ GeV) and for high-multiplicity final states.
2. First $0\ell$ Search in Exotic Decays: This is the first search for exotic Higgs decays in the $ZH$ production mode utilizing the $0\ell$ (neutrino) channel, providing a complementary probe to the $2\ell$ channel.
3. Extended Topology Coverage: The analysis is the first to search for the $H \to a_1 a_2 \to 6b$ topology and the $H \to a_1 a_2 \to 4b$ topology with different scalar masses over the full mass range $12 < m_a < 60$ GeV.
4. Comprehensive Combination: The results combine the $0\ell$ and $2\ell$ channels, leveraging orthogonal signal regions and shared systematic uncertainty models to maximize sensitivity.

Results
No significant excess of events above the Standard Model background expectation was observed in any of the search channels. Consequently, 95% confidence level (CL) upper limits were set on the ratio of the production cross-section to the SM prediction times the branching ratio ($\sigma(ZH)/\sigma_{SM}(ZH) \times \mathcal{B}$):

• $H \to 2a \to 4b$: Limits range between 4% and 25% for $12 \le m_a \le 60$ GeV. The combination of the $0\ell$ and $2\ell$ channels improves the limit by approximately 15% at $m_a = 12$ and 16 GeV compared to the $2\ell$ channel alone.
• $H \to a_1 a_2 \to 4b$: Limits range between 24% and 38% for the tested mass hypotheses.
• $H \to a_1 a_2 \to 3a_1 \to 6b$: Limits range between 10% and 20% for the tested mass hypotheses.

The largest local deviations from the background-only hypothesis were observed in the $2\ell$ channel for the $H \to 2a \to 4b$ case at $m_a = 25$ GeV (local significance 2.83$\sigma$) and for the $H \to a_1 a_2 \to 4b$ case at $(m_{a_1}, m_{a_2}) = (50, 70)$ GeV (local significance 3.28$\sigma$). However, the global significances (2.04$\sigma$ and 2.57$\sigma$, respectively) remain below the threshold for discovery.

Significance
This work extends the sensitivity of previous ATLAS searches for $H \to 2a \to 4b$ by a factor of 5–15, depending on the mass hypothesis, by incorporating a larger dataset and advanced heavy-flavor reconstruction algorithms. By exploring the $6b$ final state and the $0\ell$ production mode, the analysis probes previously unexplored regions of the parameter space for Higgs boson decays into light scalars. The results provide stringent constraints on models predicting extended Higgs sectors and hidden-sector particles, contributing to the broader program of characterizing the Higgs boson and searching for new physics.