Combination of searches for Higgs boson decays into a photon and a massless dark photon using pp collisions at √s = 13 TeV with the ATLAS detector

The paper addresses the possibility that the Higgs boson couples to a dark sector via decays into a photon and an undetectable (massless or ultra-light) dark photon, H → γγd. Motivated by astrophysical evidence for dark matter and existing limits on undetected Higgs decays (~10%), such a decay would manifest as a high-energy isolated photon accompanied by missing transverse momentum from the invisible dark photon. The study probes both the Standard Model-like Higgs (mH = 125 GeV) and additional heavy Higgs bosons predicted by BSM scenarios, produced via ggF, VBF, and ZH mechanisms. The research goal is to improve sensitivity to H → γγd by statistically combining complementary ATLAS Run 2 searches targeting γ + ETmiss signatures in different production topologies, thereby setting the most stringent limits to date and interpreting the results within a minimal simplified dark-photon model.

Previous searches have targeted identical final states. CMS performed searches for H → γγd in ZH production with 137 fb−1 and in VBF with 130 fb−1, including a combination. ATLAS and CMS also conducted earlier 8 TeV γ + ETmiss searches. These works provide benchmarks and motivate combined analyses exploiting multiple production modes to enhance sensitivity.

Data: 139 fb−1 of pp collisions at √s = 13 TeV collected with ATLAS in Run 2. Detector and software frameworks follow standard ATLAS configurations. Signal modelling: MC samples for ggF (NNLO), VBF (NLO), and ZH (qq→ZH at NLO, gg→ZH at LO) generated with POWHEG V2 interfaced to PYTHIA 8 (AZNLO tune). Cross-sections normalized to NNLO (VBF, ZH) or N3LO (ggF) QCD with NLO EW corrections. SM Higgs mass set to 125 GeV; BSM Higgs mass grid: 400–1000 GeV (200 GeV steps) and 1000–3000 GeV (500 GeV steps). Full GEANT4 detector simulation for SM; fast calorimeter parametrization for BSM. Pile-up overlaid with PYTHIA 8 A3 tune and reweighted to data. Input analyses and selections: - VBF channel (targets H125 and HBSM): Trigger: ETmiss. Event requirements: ETmiss > 150 GeV; 2 (or 3) jets with pTj1 > 60 GeV, pTj2 > 50 GeV, |Δηjj| > 3, ηj1·ηj2 < 0, mjj > 250 GeV, Δφjj < 2; at most one additional central jet (pT > 25 GeV) with Cj < 0.7; no leptons (e, μ). Photon: isolated, |ηγ| < 2.37 excluding 1.37–1.52; 15 GeV < ETγ < max(110 GeV, 0.733×mT), Cγ > 0.4; transverse mass mT(γ,ETmiss) = √(2 ETγ ETmiss [1−cos(φγ−φETmiss)]). Ten SRs based on mjj and mT; four CRs constrain W→ℓνγ+jets, Z→ννγ+jets, and fakes (e→γ, j→γ). Original analysis did not include ggF in SRs; for this combination ggF signals are added via RECAST (contribution up to 30% at low masses, ~1% at highest masses). Simultaneous profile-likelihood fit over SRs/CRs. - ZH channel (targets H125): Triggers: single- and di-lepton. Event requirements: exactly two SFOC leptons with 76 < mℓℓ < 116 GeV; ETmiss > 60 GeV; one isolated photon with ET > 25 GeV; Δφ(ETmiss, pT) > 2.4; mTγ > 100 GeV; veto on third lepton (pT > 10 GeV) and b-tagged jets (77% WP). A BDT discriminant separates signal from backgrounds. Backgrounds: ETmiss mismeasurement in Zγ+jets and Z+jets (data-driven), e→γ mis-ID in dibosons (data-driven), irreducible Vγ (constrained in a dedicated CR), minor top/Higgs from MC. Simultaneous fit to BDT distributions and Vγ CR. ggF/VBF Higgs production contributions to ZH topology are negligible and ignored. - ggF channel (targets HBSM): Reinterpretation of ATLAS mono-γ search via RECAST. Trigger: single-photon ETγ > 140 GeV. Event selection: ETmiss > 200 GeV; leading photon ETγ > 150 GeV, |Δzγ| < 250 mm, Δφ(γ,ETmiss) > 0.4; at most one jet (pT > 30 GeV, |η| < 4.5) with Δφ(j,ETmiss) > 0.4; lepton veto (e, μ, hadronic τ). Four SRs in ETmiss bins; four CRs constrain Z→ννγ, γ+jets, W→ℓνγ, Z→ℓℓγ; fake-γ from e/j estimated with data-driven methods. Both ggF and VBF signals included. Higher HBSM masses populate higher ETmiss SRs (∼30% of events in highest bin for mH ≥ 1 TeV). Combination and statistics: Likelihood L(μ,θ⃗) built as product of channel likelihoods across SR/CR categories; μ is the POI (branching ratio or σ×B), θ⃗ are nuisance parameters. Upper limits derived with profile-likelihood ratio and CLs using asymptotic formulae. Systematic uncertainties: common sources (luminosity, pile-up) correlated; object-related systematics correlated where appropriate except when reduced-uncertainty schemes differ or where strong constraints in one channel should not be transferred (treated uncorrelated; correlation impact ≤3%). Background-modelling uncertainties uncorrelated between channels due to different compositions/phase spaces; signal-modelling uncertainties (PDF/QCD) minor and treated uncorrelated. Neglecting inter-channel correlations changes limits by ≤2%. Dominant uncertainty impacts: For H125 → γγd, data statistics ~66% and systematics ~75% relative to total uncertainty; main contributors are background modelling (47%), jet and ETmiss calibration (40%), MC statistical (36%), and fake-background estimate (35%). For HBSM → γγd, statistical fraction rises to 86% at high masses; dominant systematics from fake-background estimate (52%→29% with mass) and background modelling (27–38%), followed by jet/ETmiss, lepton, and MC size (~20% each). Fit regions among input analyses are orthogonal or have negligible overlap and are treated as statistically independent.

- H125 → γγd branching ratio: Observed (expected) 95% CL upper limit B(H125 → γγd) < 1.3% (1.5%). Sensitivity improves by 29% (14%) relative to the most stringent single-channel (VBF) result. - HBSM → γγd cross-section limits: Observed (expected) 95% CL upper limits on σ(ggF+VBF) × B(HBSM → γγd) range from 16 fb (26 fb) at mH = 400 GeV to 1.0 fb (1.5 fb) at mH = 3000 GeV. - Exclusions with model assumptions: Assuming B(HBSM → γγd) = 5% and theoretical σ from the LHC Higgs Cross-Section Handbook, HBSM masses below about 1600 GeV (expected 1500 GeV) are excluded. Combination with VBF improves σ×B sensitivity by 33% (14%) at mH = 1500 GeV relative to ggF-only. - Minimal simplified dark-photon model interpretation (messenger sector parameters aα and ξ, interference sign χ): Using the VBF+ZH combination for H125 → γγd together with the ATLAS H → invisible reinterpretation, regions down to approximately ξ ≈ 0.7 at aα = 1 are excluded for χ = +1. The H → γγ measurement further disfavors low-aα regions; no constraints are set for χ = −1.

The combined analysis directly targets the hypothesized Higgs-to-dark-photon decay by exploiting complementary production topologies with distinct kinematics and backgrounds. The absence of excesses across VBF, ZH, and ggF-like signatures enables stringent upper limits on B(H125 → γγd) and on σ×B for heavy Higgs states, thereby significantly constraining Higgs portal scenarios to dark sectors with massless or ultra-light dark photons. The achieved limits improve upon the best single-channel sensitivities through a rigorous statistical combination and careful treatment of systematics and control regions. In the context of a minimal messenger model linking SM and dark U(1) sectors, the results, together with independent ATLAS constraints on invisible Higgs decays and the precise H → γγ rate measurement, carve out previously allowed portions of the (aα, ξ) parameter space for constructive interference (χ = +1). Overall, the findings substantially sharpen the experimental landscape for monophoton plus missing energy Higgs decays and guide theory development of dark-sector couplings to the Higgs.

Using the full ATLAS Run 2 data set at √s = 13 TeV (139 fb−1), ATLAS performed a combined search for H → γγd across VBF, ZH, and ggF-like final states, covering both mH = 125 GeV and heavy Higgs masses up to 3 TeV. The combination sets the most stringent limits to date: B(H125 → γγd) < 1.3% (expected 1.5%) and σ(ggF+VBF)×B(HBSM → γγd) between 16 fb (26 fb) at 400 GeV and 1.0 fb (1.5 fb) at 3 TeV, excluding HBSM masses up to about 1.6 TeV assuming B = 5% and standard cross-sections. Interpretation within a minimal simplified dark-photon model yields complementary exclusions in the (aα, ξ) plane for χ = +1. Future work could include incorporating additional production modes, improving modelling and control of fake backgrounds, extending mass coverage below 400 GeV in ggF-like selections, and combining with other monophoton and invisible Higgs decay searches to further tighten constraints.

- Sensitivity at lower heavy-Higgs masses in the ggF-like channel is limited, leading to the combined HBSM search starting at mH = 400 GeV. - Assumptions include SM-like production cross-sections for H125 and the narrow width approximation for HBSM states; deviations could alter interpretations. - Some systematic uncertainty correlations between channels are neglected or treated as uncorrelated to avoid over-constraining; estimated impact on limits is ≤ 2–3%. - Dominant systematics arise from background modelling, jet/ETmiss calibration, fake-photon estimates, and limited MC statistics, especially at high masses where statistics dominate. - The ZH analysis assumes negligible contributions from ggF/VBF to the selected topology and relies on BDT modelling and data-driven background estimates, which carry associated uncertainties.