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

Using 135 fb$^{-1}$ of proton-proton collision data at $\sqrt{s}=13.6$ TeV collected by the ATLAS detector, this paper presents a search for dark photons via semi-visible Higgs boson decays ($H \to \gamma \gamma_d$) in the gluon-gluon fusion channel, finding no significant excess over the Standard Model and setting an observed upper limit on the branching ratio of 1.4% (0.9% when combined with Run 2 results) at 95% confidence level.

The Gist

Imagine the Large Hadron Collider (LHC) as the world's most powerful particle smasher. Inside, protons zoom around at near-light speed and crash into each other, creating a shower of new particles. Usually, scientists look for the "standard" debris from these crashes, but this paper is about hunting for something much sneakier: a Dark Photon.

Here is the story of the hunt, explained simply:

The Mystery: The "Invisible" Partner

Think of the Higgs boson (the particle that gives other particles mass) as a celebrity. Usually, when this celebrity decays (breaks apart), it throws off recognizable items like electrons or photons (particles of light).

But in this theory, the Higgs might sometimes decay into a photon (a flash of light) and a dark photon.

• The Photon: This is the flash of light we can see.
• The Dark Photon: This is the "invisible partner." It doesn't interact with our detectors at all. It's like a ghost that slips right through the walls of the laboratory.

When the Higgs decays this way, the detector sees a single flash of light and a sudden "missing" amount of energy (because the dark photon ran away). Scientists call this a "semi-visible" decay because part of it is seen, and part is missing.

The Challenge: The "Needle in a Haystack" Problem

Finding this specific decay is incredibly hard for two reasons:

1. It's rare: The Higgs usually does other things. This specific "flash + ghost" event is very uncommon.
2. The "Haystack" is noisy: The LHC produces billions of collisions. Most of them create "fake" missing energy because of measurement errors or messy debris, which looks exactly like a dark photon running away.

In the past, the ATLAS detector (the giant camera taking pictures of these collisions) had a "security guard" (the trigger system) that was too strict. It would only let in events with very high-energy flashes. But the dark photon signal might be a "dimmer" flash. If the guard is too strict, the signal gets thrown out before scientists can even look at it.

The New Strategy: A Smarter Security Guard

This paper describes a new search using data from 2023 and 2024. The team upgraded their "security guard" (the trigger) to be more flexible.

• The Analogy: Imagine a bouncer at a club who used to only let in people wearing expensive suits (high energy). The new bouncer says, "Okay, if you have a cool jacket and you're carrying a specific type of bag, even if your suit isn't the most expensive, you can come in."
• The Result: This allowed them to catch events with lower energy thresholds (50 GeV for the photon, 70 GeV for the missing energy) that they would have missed before. This doubled their chances of catching the signal.

The Detective Work: Filtering the Noise

Once they let the events in, they had to separate the real signal from the background noise. They used several clever tricks:

• The "BDT" (Boosted Decision Tree): This is like a super-smart AI detective. It looks at the collision and asks, "Did we mess up the math on where the crash happened?" If the primary crash point was misidentified, the missing energy calculation is wrong. The AI filters out these messy events.
• The "Fake" Check: Sometimes, a jet of particles (a spray of debris) looks like a photon, or an electron gets mistaken for a photon. The team used "control rooms" (special data sets with known particles like muons) to estimate how often these mistakes happen, essentially creating a "noise map" to subtract from their results.

The Verdict: No Ghosts Found (Yet)

After analyzing 135 units of data (called "femtobarns," which is a massive amount of collision data), the team looked for an excess of events that didn't fit the Standard Model (the current rulebook of physics).

• The Result: They found no significant excess. The number of "flash + missing energy" events they saw matched exactly what they expected from known physics.
• The Limit: While they didn't find the dark photon, they set a very strict rule: If the Higgs does decay into a dark photon, it happens less than 1.4% of the time (and likely around 0.9% when combined with previous data).

The Takeaway

This paper is a story of technological improvement. By lowering the energy thresholds and using smarter algorithms to clean up the data, the ATLAS collaboration successfully searched a region of physics that was previously invisible to them. They didn't find the dark photon, but they proved that if it exists, it's hiding very well, and they have now mapped out exactly where it can't be hiding.

In short: They looked for a ghost in a crowded room using a better flashlight and a smarter filter. They didn't see a ghost, but they now know exactly how quiet the room has to be for one to be there.

Technical Summary

Technical Summary: Search for Dark Photons from Higgs Boson Decays in the Gluon–Gluon Fusion Channel

Problem and Motivation
This paper presents a search for semi-visible decays of the Standard Model (SM) Higgs boson ($H$) into a photon ($\gamma$) and a massless dark photon ($\gamma_d$), denoted as $H \to \gamma\gamma_d$. Such decays are motivated by various Beyond the Standard Model (BSM) scenarios, including dark sector models where the dark photon acts as a gauge boson for an additional $U(1)_D$ group. These models offer potential solutions to small-scale structure formation problems and mechanisms for Sommerfeld enhancement in dark matter annihilation. While previous searches by ATLAS and CMS have targeted vector boson fusion (VBF) and associated production ($ZH$) channels, the gluon–gluon fusion (ggF) channel remains the dominant production mode (approximately an order of magnitude larger cross-section) but has been historically unexplored for this specific signature due to trigger challenges. The $H \to \gamma\gamma_d$ signature in ggF results in a final state with a single photon and missing transverse momentum ($p_T^{miss}$), requiring low kinematic thresholds that were previously difficult to access without excessive trigger rates.

Methodology
The analysis utilizes 135 fb$^{-1}$ of proton–proton collision data at $\sqrt{s} = 13.6$ TeV collected by the ATLAS detector during 2023 and 2024 (Run 3).

• Trigger Strategy: A novel composite trigger deployed in 2023 is central to this analysis. It combines selections on photon transverse momentum ($p_T^\gamma$), missing transverse momentum ($p_T^{miss}$), and the transverse mass of the $\gamma$–$p_T^{miss}$ system ($m_T$). The thresholds were reduced to $p_T^\gamma > 50$ GeV, $p_T^{miss} > 70$ GeV, and $m_T > 80$ GeV. This configuration increases the acceptance for ggF events by a factor of two compared to previous single-photon or inclusive $p_T^{miss}$ triggers while maintaining rate control.
• Event Reconstruction and Selection: Events are required to have exactly one selected photon ($p_T^\gamma > 50$ GeV), $p_T^{miss} > 100$ GeV, and $m_T > 80$ GeV. A critical challenge in this signature is the misidentification of the hard-scatter primary vertex (PV) due to low track activity, which can lead to the incorrect removal of hard-scatter jets from the $p_T^{miss}$ calculation, thereby enhancing fake $p_T^{miss}$ backgrounds. To mitigate this, a Boosted Decision Tree (BDT) is trained to identify and reject events with misidentified PVs based on vertex properties, jet multiplicity, and angular separations.
• Background Estimation:
  • Genuine Photon Backgrounds ($Z\gamma, W\gamma$): Estimated using Monte Carlo (MC) simulations, normalized to data in control regions (CRs) enriched with one or two muons ($1\mu$CR, $2\mu$CR).
  • Electron Misidentification ($e \to \gamma$): Estimated using a data-driven "fake factor" method derived from $Z \to e^+e^-$ events, corrected for trigger efficiencies.
  • Jet Misidentification ($j \to \gamma$): Estimated using a data-driven method involving non-isolated photons in probe regions. The isolation distribution of jets misidentified as photons is extrapolated from a "Loose" selection to the "Tight" signal selection using an affine transformation validated with MC samples.
  • Fake $p_T^{miss}$: A dedicated low-$S_{p_T^{miss}}$ control region (where $2 < S_{p_T^{miss}} < 6$) is included in the fit to constrain the normalization of the $j \to \gamma$ background.
• Statistical Analysis: A binned maximum-likelihood fit is performed on the $m_T$ distribution across the Signal Region (SR) and multiple Control/Validation Regions. The branching ratio $\mathcal{B}_{H \to \gamma\gamma_d}$ is treated as the parameter of interest. Systematic uncertainties, including experimental, theoretical, and data-driven method uncertainties, are incorporated as nuisance parameters.

Key Contributions

• First ATLAS Search in ggF: This work represents the first ATLAS search for $H \to \gamma\gamma_d$ specifically optimized for the gluon–gluon fusion production channel.
• Novel Trigger Implementation: The analysis demonstrates the effectiveness of the new Run 3 composite trigger in accessing low-threshold kinematic regions previously inaccessible for this specific BSM signature.
• Advanced Background Mitigation: The implementation of a BDT to reject events with misidentified primary vertices significantly improves the discrimination between signal and backgrounds dominated by fake $p_T^{miss}$.
• Comprehensive Data-Driven Estimation: The use of orthogonal control regions and probe regions allows for robust, data-driven constraints on the dominant backgrounds arising from misidentified electrons and jets.

Results
The analysis observes no significant excess of events over the Standard Model expectation. The observed $p_0$-value is 0.37, indicating consistency with the background-only hypothesis.

• Branching Ratio Limits: An observed (expected) upper limit on the branching ratio $\mathcal{B}(H \to \gamma\gamma_d)$ of 1.4% (1.2%) is set at the 95% confidence level (CL).
• Combined Limits: When combined with previous ATLAS Run 2 searches in the $ZH$ and VBF channels, the observed (expected) upper limit improves to 0.9% (0.9%).

Significance
The paper claims that this result provides constraints on the $H \to \gamma\gamma_d$ decay comparable to the most stringent limits from previous Run 2 searches. By successfully targeting the dominant ggF production mode, the analysis extends the sensitivity to a previously unexplored kinematic region. The combination with Run 2 results yields the most stringent constraint to date on this specific decay mode, effectively probing sub-percent level branching ratios for Higgs boson decays into a photon and a massless dark photon. The work validates the utility of the new Run 3 trigger system for exotic Higgs searches and establishes a framework for future analyses in the low-threshold regime.