Search for dark photons from Higgs boson decays in the gluon-gluon fusion channel in proton-proton collisions at sqrt s =13.6 TeV with the ATLAS detector

arXiv:2606.14362 · hep-ex · Submitted 2026-06-12 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: High Energy Physics: Experiment. The day in high-energy experiment: detectors, searches and limits.

Kai: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "Search for dark photons from Higgs boson decays in the gluon-gluon fusion channel in proton-proton collisions at sqrt s =13.6 TeV with the ATLAS detector".

Mira: Here's the gist of the paper. As a fastidious and diligent researcher, I will synthesize these provided text segments into a comprehensive, detailed summary of the paper, ensuring accuracy and thoroughness while adhering to your specified length constraint.

Kai: Detailed Summary of the Search for Higgs Boson Decays into a Photon and Missing Transverse Momentum (H to gamma gamma d) This research presents a dedicated search for semi-visible Higgs boson decays into a final state consisting of a photon (gamma) and significant missing transverse momentum (p T miss).

Mira: The analysis is specifically optimized for the gluon–gluon fusion (ggF) production channel in proton–proton collisions at a center-of-mass energy of sqrt s = 13.6 TeV, utilizing data collected by the ATLAS detector during Run 3 (2023 and 2024), accumulating a total integrated luminosity of 135 fb-1.

Kai: The core methodology hinges on exploiting a novel trigger deployed in 2023, which significantly lowered the selection thresholds for both photon transverse momentum (p gamma T) down to 50 GeV and missing transverse momentum (p T miss) down to 70 GeV.

Title and authors: Kai: So we’re looking at a paper called "Search for dark photons from Higgs boson decays in the gluon-gluon fusion channel in proton-proton collisions at sqrt s equals thirteen point six TeV with the ATLAS detector" <ref:2606.14362#pg1,Search for dark photons from Higgs boson decays in the gluon-gluon>. It sounds like they are hunting for something that isn't there, or maybe something new.

Mira: That’s right. The title tells us exactly what we’re looking for: dark photons coming from a Higgs boson decay, specifically when that Higgs is made in proton-proton collisions at thirteen point six TeV energy using the ATLAS detector <ref:2606.14362#pg1,in proton-proton collisions at>.

Lev: It sounds like a search for new physics, but you have to remember that the Higgs boson itself is already well understood in the Standard Model stuff we've been measuring.

Kai: Exactly, and what makes this specific search interesting is that they are focusing on a decay channel into a photon and missing transverse momentum. That’s not just any decay; it points toward physics beyond the Standard Model, like dark sector models.

Mira: And that fits perfectly because the paper mentions searching for a massless dark photon scenario where kinetic mixing can be tuned away through gauge field rotations, which means the interaction with ordinary matter is suppressed <ref:2606.14362#pg3>.

Lev: So, what does this mean practically? It means they are setting limits on how often this decay happens, which constrains the parameters of these dark sector models.

Kai: Right. The implication is that if we don't find a signal here, it puts strong constraints on those specific BSM theories we discussed earlier.

Mira: And the paper itself is using one hundred thirty-five femtobarns of Run three data from ATLAS in two thousand twenty-three and two thousand twenty-four to do this search <ref:2606.14362#pg1>.

Lev: The main point for us right now is that they’re using a very specific, somewhat novel trigger setup deployed in two thousand twenty-three to look at this region of phase space <ref:2606.14362#pg1>.

The paper's summary: Kai: Okay, so the paper summarizes their search strategy. They are looking for Higgs bosons decaying into a photon and missing transverse momentum, which they call H to gamma and dark photon d <ref:2606.14362#pg2>.

Mira: The core of their approach is exploiting a new trigger from two thousand twenty-three that lowered the thresholds significantly <ref:2606.14362#pg1>. They set the photon transverse momentum up to fifty GeV and the missing transverse momentum down to seventy GeV <ref:2606.14362#pg1>.

Lev: That’s a pretty low threshold for those variables, which is good for sensitivity but also means we have to be careful about the backgrounds in that region.

Kai: Right, and they refine this by requiring a high transverse mass of the photon plus missing transverse momentum system above eighty GeV <ref:2606.14362#pg1>. That kinematic selection is what really targets that unexplored region of the decay spectrum <ref:2606.14362#pg1>.

Mira: And for signal selection, they are very strict: exactly one selected photon with its transverse momentum above fifty GeV and no other baseline photons at all <ref:2606.14362#pg1>.

Lev: And they also need the missing transverse momentum to be above one hundred GeV to ensure they are looking in the right place, and that high transverse mass of over eighty GeV again <ref:2606.14362#pg1>.

Kai: To make sure those selections are robust against noise, they use a boosted decision tree discriminant to handle potential problems with misreconstructed primary vertices which can mess up the missing transverse momentum calculation <ref:2606.14362#pg1>.

Mira: And because they aren't just looking at the signal, they use data-driven methods to estimate backgrounds from jets and electrons that might be mistaken for photons <ref:2606.14362#pg1>.

The paper's improvements: Lev: The paper suggests a few ways to make this search better. First, they suggest using the boosted decision tree to filter events where the primary vertex might be misreconstructed, by applying a requirement on that vertex BDT score being greater than zero point one <ref:2606.14362#pg1>.

Kai: That makes sense for reducing background from those tricky vertex issues, because if the vertex reconstruction is bad, the missing transverse momentum calculation gets skewed <ref:2606.14362#pg1>.

Mira: They also propose using data-driven estimates for backgrounds by selecting samples with dielectron, single-electron, and single-muon triggers to model how often those misidentified jets actually happen <ref:2606.14362#pg1>.

Lev: That helps improve the background modeling accuracy, which is always a big deal when you’re pushing the limits on a new signal <ref:2606.14362#pg5>.

Kai: Then there's this idea about combining the trigger selections—putting together requirements on missing transverse momentum, photon transverse momentum, and that transverse mass above eighty GeV <ref:2606.14362#pg1>.

Mira: That composite trigger could increase the acceptance of those gluon-gluon fusion events by about a factor of two while keeping the trigger rate under control <ref:2606.14362#pg5>.

Lev: So, it's about optimizing sensitivity for the Higgs to dark photon decay in that specific production channel <ref:2606.14362#pg5>.

Conclusion: Kai: So, wrapping up the search described in "Search for dark photons from Higgs boson decays in the gluon-gluon fusion channel in proton-proton collisions at sqrt s equals thirteen point six TeV with the ATLAS detector" <ref:2606.14362#pg1,Search for dark photons from Higgs boson decays in the gluon-gluon>. The paper performed a binned maximum likelihood fit on the transverse mass distribution <ref:2606.14362#pg1>.

Mira: That fit was used to check if what they saw matched what we expected from the Standard Model <ref:2606.14362#pg4>. And the result is that there was no significant excess over the Standard Model prediction <ref:2606.14362#pg4>.

Lev: This leads them to set an observed upper limit on the branching ratio of H to dark photon d at one point, which is one point four percent or one point two percent at a ninety-five percent confidence level <ref:2606.14362#pg4>.

Kai: So what does this mean for us? It means they’ve tested a specific window in the parameter space for these dark photon models and found no evidence of this decay happening at that rate with the data they collected <ref:2606.14362#pg4>.

Mira: It sets a constraint on those BSM theories we talked about earlier, telling us how small their coupling needs to be if they want to survive these constraints <ref:2606.14362#pg3>.

Lev: For us building hardware, this means that if a dark photon exists with couplings that would lead to a one point four percent branching ratio, we’ve put a ceiling on how strong those interactions can be for this specific decay mode <ref:2606.14362#pg5>.

EUROPEAN ORGANISATION FOR NUCLEAR RESEARCH (CERN)

hep-ex

Submitted: 2026-06-12

Updated: 2026-11-03

Comments: 52 pages in total, author list starting page 35, 13 figures and 6 tables, published in JHEP. All figures including auxiliary figures are available at http://atlas.web.cern.ch/Atlas/GROUPS/PHYSICS/PAPERS/HMBS-2024-50

Journal ref: JHEP 10 (2026) 270

DOI: 10.1007/JHEP10(2026)270

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 76/100

The gist: As a fastidious and diligent researcher, I will synthesize these provided text segments into a comprehensive, detailed summary of the paper…

Key concepts

Gluon-Gluon Fusion (ggF)
This is a specific type of collision process where two gluons from protons collide to produce a Higgs boson. The study focuses on this channel because it is a primary production mechanism for the Higgs boson at high energies, providing the main source for searching for new decays.
Missing Transverse Momentum ($ ext{p}_{ ext{T}}^{ ext{miss}}$)
This refers to momentum that escapes detection in the detector, often indicating particles that are invisible or weakly interacting. In this search, it is a key signature used to isolate events where the Higgs decays into a photon and an undetected dark photon.
$H ightarrow \gamma\gamma_d$
This represents the hypothesized decay of a Higgs boson into one standard photon ($\gamma$) and one hypothetical particle known as a dark photon ($\gamma_d$). Finding this decay would provide evidence for physics beyond the Standard Model.
Boosted Decision Tree (BDT)
A BDT is a machine learning tool used to distinguish between signal events (like the hypothesized $H ightarrow \gamma\gamma_d$) and background events. It combines multiple variables, such as photon energy and missing momentum, to create a highly accurate selection criterion for the analysis.

Terminology

Summary

As a fastidious and diligent researcher, I will synthesize these provided text segments into a comprehensive, detailed summary of the paper, ensuring accuracy and thoroughness while adhering to your specified length constraint.


Detailed Summary of the Search for Higgs Boson Decays into a Photon and Missing Transverse Momentum (H to gamma gamma d)

This research presents a dedicated search for semi-visible Higgs boson decays into a final state consisting of a photon (gamma) and significant missing transverse momentum (p T miss). The analysis is specifically optimized for the gluon–gluon fusion (ggF) production channel in proton–proton collisions at a center-of-mass energy of sqrt s = 13. p. 0

Where the paper says it

EUROPEAN ORGANISATION FOR NUCLEAR RESEARCH (CERN) Submitted to: JHEP CERN-EP-2026-139 15th June 2026 Search for dark photons from Higgs boson decays in the gluon–gluon fusion channel in proton–proton collisions at √ 𝒔 = 13. 6 TeV with the ATLAS detector The ATLAS Collaboration This paper presents a search for semi-visible Higgs boson decays in final states with a photon (𝛾) and missing transverse momentum 𝑝 miss T .

6 TeV, utilizing data collected by the ATLAS detector during Run 3 (2023 and 2024), accumulating a total integrated luminosity of 135 fb-1. p. 4
Where the paper says it

6 TeV, collected in 2023 and 2024 by the ATLAS detector. Only data recorded during stable beam conditions and with all sub-detectors operational are considered [33], corresponding to a total integrated luminosity of 135 fb−1 , with an uncertainty of 1.

The core methodology hinges on exploiting a novel trigger deployed in 2023, which significantly lowered the selection thresholds for both photon transverse momentum (p gamma T) down to 50 GeV and missing transverse momentum (p T miss) down to 70 GeV. p. 0

Where the paper says it

6 TeV with the ATLAS detector The ATLAS Collaboration This paper presents a search for semi-visible Higgs boson decays in final states with a photon (𝛾) and missing transverse momentum 𝑝 miss T . The analysis is optimised for the gluon—gluon fusion production channel and exploits a novel trigger deployed in 2023, with reduced thresholds down to 50 GeV on the photon transverse momentum (𝑝 𝛾 T ) and 70 GeV on the 𝑝 miss T , in combination with a selection on the transverse mass of the 𝑝 miss T –𝛾 system (𝑚T) above 80 GeV.

This trigger combination is further refined by requiring a high transverse mass (m gamma gamma d T) of the gamma + p T miss system, specifically above 80 GeV. p. 3
Where the paper says it

However, the 𝐻 → 𝛾𝛾d decay of a Higgs boson with a mass of 125 GeV has not been specifically targeted, as the low thresholds on the photon transverse momentum (𝑝 𝛾 T ) and 𝑝 miss T required for this signature constitute a challenge for the trigger. Developments in the ATLAS trigger system between Run 2 and Run 3 led to the implementation of a composite trigger that combines selections on 𝑝 miss T , 𝑝 𝛾 T , and the transverse mass between the 𝛾 and the 𝑝 miss T , 𝑚T = √︃ 2𝑝 𝛾 T 𝑝 miss T (1 − cos(Δ𝜙)), 1 where Δ𝜙 is the azimuthal separation between the photon and the 𝑝 miss T .

This kinematic selection is crucial for targeting a previously unexplored region in the Higgs decay spectrum. p. 1
Where the paper says it

This paper presents a new strategy to search for exotic Higgs boson decays into a final state with a photon (𝛾) and missing transverse momentum (𝑝 miss T ), probing an unexplored region of the phase space with low kinematic thresholds. This semi-visible decay of the Higgs boson provides an interesting production channel for the dark photon (𝛾d), which arises as the gauge boson of an additional 𝑈(1)𝐷 gauge group in dark sector models.

The signal region (SR) selection criteria are stringent: requiring exactly one selected photon with p gamma T > 50 GeV, no additional baseline photons, p T miss > 100 GeV, and m gamma gamma d T > 80 GeV. p. 7

Where the paper says it

Events were selected using single-muon triggers and requiring one muon treated as invisible in the evaluation of 𝑝 miss T and 𝑚T. In addition, a subset of the selections used in the analysis is applied: 𝑝 miss T > 100 GeV, 𝑚T > 80 GeV, one photon with 𝑝 𝛾 T > 50 GeV and a requirement of |Δ𝜙( ®𝑝 miss T , 𝑝® miss,𝛾 T )| > 1.

Furthermore, the analysis employs a Boosted Decision Tree (BDT) discriminant to mitigate the impact of hard-scatter primary vertex misidentification caused by low track activity, ensuring robust event selection. p. 0
Where the paper says it

The analysis is performed using 135 fb−1 of Run 3 data collected by ATLAS in 2023 and 2024. A boosted decision tree is used to identify events with misreconstructed primary vertices, which can affect the 𝑝 miss T calculation.

The results are interpreted within the framework of searching for Higgs boson decays into a photon and a dark photon (H to gamma gamma d). p. 0
Where the paper says it

The analysis is optimised for the gluon—gluon fusion production channel and exploits a novel trigger deployed in 2023, with reduced thresholds down to 50 GeV on the photon transverse momentum (𝑝 𝛾 T ) and 70 GeV on the 𝑝 miss T , in combination with a selection on the transverse mass of the 𝑝 miss T –𝛾 system (𝑚T) above 80 GeV. The results are interpreted in terms of Higgs boson decays into a 𝛾 and a dark photon 𝛾d (𝐻 → 𝛾𝛾𝑑).

A binned maximum-likelihood fit is performed on the m gamma gamma d T distribution to rigorously evaluate the compatibility of the observed data with Standard Model (SM) expectations. p. 0

Where the paper says it

Data-driven methods are employed to estimate the backgrounds from jets and electrons misidentified as photons, while the backgrounds with genuine photons are taken from Monte Carlo simulations, normalised to data in control regions with muons. A binned maximum-likelihood fit to the 𝑚T distribution is performed to evaluate the compatibility of the observed data with the Standard Model expectation.

Crucially, no significant excess over the SM prediction was observed in this analysis. Consequently, an observed (expected) 95% confidence level upper limit on the branching ratio of H to gamma gamma d is established at 1. p. 0
Where the paper says it

No significant excess is observed over the Standard Model prediction, and an observed (expected) upper limit on the branching ratio of 𝐻 → 𝛾𝛾𝑑 of 1. 2%) is obtained at 95% confidence level.

4% (1.2%).

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Boosted decision tree implementation for primary vertex misidentification can be used to identify events with misreconstructed primary vertices, which can affect the p miss T calculation as described in Section 7.1, enhancing background rejection by applying a requirement on the misidentified vertex BDT score of BDTvertex > 0.1.

  2. Data-driven estimation of backgrounds from jets and electrons misidentified as photons can be employed to estimate the backgrounds from jets and electrons misidentified as photons using samples selected with dielectron, single-electron, and single-muon triggers, thereby improving background modeling accuracy for the signal region.

  3. A composite trigger combining selections on p miss T, p gamma T, and mT above 80 GeV can be used to increase the acceptance of ggF events by approximately a factor of two while keeping the trigger rate under control, optimizing sensitivity for the H -> gamma gamma d decay in the gluon–gluon fusion production channel.

  4. The use of a binned maximum-likelihood fit to the mT distribution allows for evaluating the compatibility of the observed data with the Standard Model expectation, providing a statistical tool to set limits on B gamma gamma d, such as obtaining an "observed (expected) upper limit on the branching ratio of H -> gamma gamma d of 1.4% (1.2%) at 95% confidence level."

  5. The background estimation for jets misidentified as photons can be refined by using particle flow constituents, combining tracks and calorimeter clusters, with a dedicated PU subtraction algorithm (‘pfopufit’) to compute p miss T, which allows samples to be selected with reconstructed muons in the final state and to treat them as invisible.

  6. The analysis can incorporate systematic uncertainty modeling for data-driven methods by treating the dependence of the f e gamma fake factor on the peak width and the definition of the sidebands as a systematic uncertainty, ensuring a more robust constraint on background estimates.

Abstract

This paper presents a search for semi-visible Higgs boson decays in final states with a photon (γ) and missing transverse momentum p T miss. The analysis is optimised for the gluon-gluon fusion production channel and exploits a novel trigger deployed in 2023, with reduced thresholds down to 50 GeV on the photon transverse momentum (p T γ) and 70 GeV on the p T miss, in combination with a selection on the transverse mass of the p T miss - γ system (m T) above 80 GeV. The results are interpreted in terms of Higgs boson decays into a γ and a dark photon γ d (H to γγ d). The analysis is performed using 135 fb-1 of Run 3 data collected by ATLAS in 2023 and 2024. A boosted decision tree is used to identify events with misreconstructed primary vertices, which can affect the p T miss calculation. Data-driven methods are employed to estimate the backgrounds from jets and electrons misidentified as photons, while the backgrounds with genuine photons are taken from Monte Carlo simulations, normalised to data in control regions with muons. A binned maximum-likelihood fit to the m T distribution is performed to evaluate the compatibility of the observed data with the Standard Model expectation. No significant excess is observed over the Standard Model prediction, and an observed (expected) upper limit on the branching ratio of H to γγ d of 1.4% (1.2%) is obtained at 95% confidence level. Combining this result with previous Run 2 searches improves the observed (expected) upper limit to 0.9% (0.9%).

Sources

Related papers