Study of Higgs boson pair production in the $HH \rightarrow b \overline{b} γγ$ final state with 308 fb$^{-1}$ of data collected at $\sqrt{s} =$ 13 TeV and 13.6 TeV by the ATLAS experiment

The ATLAS collaboration performed a search for Higgs boson pair production in the $b\bar{b}\gamma\gamma$ final state using 308 fb$^{-1}$ of proton-proton collision data at $\sqrt{s} = 13$ and 13.6 TeV, finding no significant excess over the Standard Model prediction and setting a 95% confidence-level upper limit on the production cross-section of $\mu_{HH} < 3.7$ while constraining the Higgs self-coupling modifier to the range $-1.6 < \kappa_\lambda < 6.6$.

The Gist

Imagine the universe as a giant, complex machine. For decades, physicists have been trying to understand how this machine works, specifically how it gives mass to particles. In 2012, they found a crucial part of the machine: the Higgs boson. Think of the Higgs boson as a "master switch" that turns on the property of mass for everything around us.

But there's a mystery left unsolved: How does the Higgs boson talk to itself?

In physics, particles usually interact with each other. But the Higgs is special because it can also interact with another Higgs boson. This is called "Higgs self-coupling." If we can measure how strong this conversation is, we might discover that the universe is built on a different foundation than we thought, potentially revealing new laws of physics or even why the universe exists at all.

This paper is a report from the ATLAS experiment at CERN (the Large Hadron Collider in Switzerland). It's like a team of master detectives trying to find a very specific, very rare crime scene to solve this mystery.

The Big Hunt: Finding a "Double Higgs"

The team is looking for a moment where two Higgs bosons are created at the same time and then immediately decay (break apart) into a specific set of particles: two bottom quarks (heavy particles) and two photons (particles of light).

Why is this hard?

• It's incredibly rare: Creating two Higgs bosons is like trying to win the lottery twice in a row. It happens about 1,000 times less often than creating just one.
• It's buried in noise: The collider smashes protons together billions of times a second, creating a chaotic mess of particles. Finding this specific "double Higgs" signal is like trying to hear a single whisper in the middle of a roaring stadium crowd.

The New Tools: Sharper Eyes and Better Brains

This paper describes an upgrade to their search compared to previous attempts. They used a massive amount of data (308 "inverse femtobarns"—a unit of data volume that is hard to visualize, but imagine it as a library containing billions of collision records).

To find the needle in the haystack, they used two major upgrades:

1. The "Smart Filter" (AI): They used a new type of Artificial Intelligence (based on "transformer neural networks," the same tech behind modern chatbots) to identify the bottom quarks. Think of this as upgrading from a human security guard checking IDs to a super-fast AI camera that can instantly recognize a specific face in a crowd of millions.
2. The "Kinematic Fit" (Reconstructing the Crime Scene): When particles break apart, they don't always leave a perfect trail. The team used a mathematical "kinematic fit" to reconstruct the event. Imagine a detective looking at shattered glass on the floor and using physics to perfectly reconstruct the shape of the original window. This made their measurements much sharper.

The Results: "No New Physics... Yet"

After analyzing all the data, here is what they found:

• The Signal: They found a tiny hint of double Higgs production, but it wasn't statistically significant enough to claim a discovery. It's like hearing a whisper that might be the person you're looking for, but it's too quiet to be sure.
• The Measurement: They measured the strength of the Higgs self-coupling. The result is compatible with the Standard Model (our current best theory of physics). In simple terms, the Higgs boson seems to be talking to itself exactly the way we predicted it would.
• The Limits: While they didn't find "new physics," they set a very strict boundary. They can now say with 95% confidence that the Higgs self-coupling is not wildly different from the prediction. They have narrowed the search area significantly.

Why Does This Matter?

Think of the Standard Model as a map of the known world. For a long time, we've been exploring the edges of this map. Finding the double Higgs interaction is like finding a new continent.

• If they had found a huge difference: It would have been a massive earthquake in physics, proving our map was wrong and pointing toward "New Physics" (like Dark Matter or extra dimensions).
• Since they found it matches the map: It confirms our current understanding is very robust. However, it also means the "New Physics" we are hoping for is hiding even deeper, or perhaps it doesn't exist in the way we thought.

The Bottom Line

The ATLAS team has built a better microscope and looked harder than ever before. They found that the Higgs boson behaves exactly as the "Standard Model" predicts. They haven't found the "smoking gun" for new physics yet, but they have cleared the fog, making it much easier for future experiments (like the next generation of colliders) to find the truth.

It's a bit like searching for a specific star in the night sky. They didn't find a new star this time, but they proved that the sky is exactly as clear and predictable as the old maps said it was, which is a victory in itself for science.

Technical Summary

1. Problem and Physics Motivation

The primary objective of the Large Hadron Collider (LHC) program following the 2012 discovery of the Higgs boson ($H$) is to probe the shape of the Higgs potential, which governs electroweak symmetry breaking. In the Standard Model (SM), this potential is determined by the Higgs mass ($m_H \approx 125$ GeV) and the vacuum expectation value ($v \approx 246$ GeV).

• The Trilinear Coupling: The potential includes a trilinear term ($\lambda_3 H^3$) and a quartic term ($\lambda_4 H^4$). The trilinear coupling ($\lambda_3$) is accessible via Higgs boson pair ($HH$) production. Deviations from the SM prediction would indicate New Physics (NP) with profound implications for cosmology and the stability of the vacuum.
• The Challenge: The $HH$ production cross-section is extremely small ($\sim 30$ fb at 13 TeV), roughly three orders of magnitude smaller than single Higgs production.
• The Channel: This analysis focuses on the $HH \to b\bar{b}\gamma\gamma$ final state. While this channel has a low branching fraction ($\sim 0.26\%$), it offers the best sensitivity near the production threshold due to the excellent mass resolution of the diphoton system ($\gamma\gamma$) and the ability to reconstruct the $b\bar{b}$ system.

2. Methodology and Experimental Setup

Data and Simulation

• Dataset: The analysis utilizes the full ATLAS Run 2 dataset (2015–2018, $\sqrt{s}=13$ TeV, 140 fb$^{-1}$) and the first part of Run 3 (2022–2024, $\sqrt{s}=13.6$ TeV, 168 fb$^{-1}$), totaling 308 fb$^{-1}$.
• Signal Generation:
  • Gluon-Gluon Fusion (ggF): Generated using Powheg Box v2 with PDF4LHC21 PDFs. Samples include SM ($\kappa_\lambda=1$) and Beyond-Standard-Model (BSM) scenarios ($\kappa_\lambda = -1, 0, 2.5, 5, 10$).
  • Vector-Boson Fusion (VBF): Generated using MadGraph5_aMC@NLO with NNPDF3.0nlo. Includes variations in trilinear ($\kappa_\lambda$), quartic VBF ($\kappa_{2V}$), and VBF vertex ($\kappa_V$) couplings.
• Backgrounds: Modeled via simulation (resonant $H \to \gamma\gamma$, $Z(\to \ell\ell)\gamma$, etc.) and data-driven techniques for the dominant continuum non-resonant diphoton background ($\gamma\gamma$+jets).

Event Selection and Reconstruction

• Pre-selection: Events require exactly two isolated photons and at least two $b$-tagged jets.
  • Photons: $p_T > 35$ GeV (leading) and $25$ GeV (sub-leading), with invariant mass $105 < m_{\gamma\gamma} < 160$ GeV.
  • Jets: $p_T > 25$ GeV, $|y| < 4.5$.
• $b$-tagging: A major upgrade in this analysis is the use of the GN2 algorithm, a transformer-based neural network, achieving 85% efficiency for $b$-jets with a 0.01 misidentification rate for light jets.
• Kinematic Fit (KF): A two-step kinematic fit is applied to improve mass resolution:
  1. Corrects jet energies using transverse momentum conservation.
  2. Constrains the $b\bar{b}$ invariant mass to the Higgs mass (125 GeV).
  • Impact: This improves the $m_{b\bar{b}\gamma\gamma}$ resolution by 46% and the $m_{b\bar{b}}$ resolution by up to 27%.

Event Categorization

Events are classified into 14 exclusive categories to maximize sensitivity:

1. Mass Regions: Split into a Low-Mass region ($m^*_{b\bar{b}\gamma\gamma} < 350$ GeV) and a High-Mass region ($m^*_{b\bar{b}\gamma\gamma} \ge 350$ GeV). The variable $m^*$ corrects for detector resolution effects.
2. Signal Purity: Within each mass region, a Boosted Decision Tree (BDT) (XGBoost) discriminates signal from background. The BDT is trained separately for Run 2 and Run 3 data but combined for optimization.
   • High-Mass BDT: Trained on a mix of SM and BSM VBF samples to enhance sensitivity to $\kappa_{2V}$.
   • Low-Mass BDT: Trained on a mix of SM and BSM ggF samples to enhance sensitivity to $\kappa_\lambda$.
3. Run Period: Categories are further split by Run 2 and Run 3 data-taking periods.

Statistical Analysis

• A simultaneous profile-likelihood fit is performed on the diphoton invariant mass ($m_{\gamma\gamma}$) distribution across all 14 categories.
• Signal Model: Double-sided Crystal Ball functions.
• Background Model: Analytic functions (exponentials) for non-resonant backgrounds; resonant backgrounds are fixed to SM expectations.
• Systematics: Major uncertainties include photon energy scale/resolution, $b$-tagging efficiency, heavy-flavor content in single Higgs production, and theoretical cross-section uncertainties (QCD scales, PDFs).

3. Key Contributions and Improvements

Compared to the previous ATLAS analysis (based on 13 TeV data only, Ref. [29]), this study introduces significant technical advancements:

1. GN2 $b$-tagging: The implementation of the transformer-based GN2 algorithm significantly improved $b$-jet identification efficiency and purity.
2. Kinematic Fit: The introduction of a kinematic fit to constrain the $b\bar{b}$ mass to 125 GeV drastically improved the reconstruction of the $HH$ invariant mass.
3. Run 3 Data: Inclusion of 168 fb$^{-1}$ at $\sqrt{s}=13.6$ TeV, increasing the dataset size by more than double.
4. Correlated Background Modeling: Treating background shapes in Run 2 and Run 3 as correlated allowed for tighter selection criteria, improving signal significance.
5. Performance: These improvements led to an approximate 20% improvement in significance from the new algorithms and 60% from the additional data compared to the previous iteration.

4. Results

Signal Strength ($\mu_{HH}$)

The signal strength modifier is defined as $\mu_{HH} = \sigma_{obs} / \sigma_{SM}$.

• Observed: $\mu_{HH} = 0.9^{+1.4}_{-1.1}$ (stat. + syst.).
• Expected: $1.0^{+1.3}_{-1.0}$.
• Significance: The observed significance for $HH$ production is 0.78$\sigma$ (expected 1.01$\sigma$).
• Upper Limit: The 95% Confidence Level (CL) upper limit is $\mu_{HH} < 3.7$.
  • Comparison: This is a significant improvement over the previous limit of $\mu_{HH} < 5.0$ (expected 2.6) from Run 2 alone.

Coupling Modifiers

The analysis constrains the trilinear Higgs coupling modifier ($\kappa_\lambda$) and the quartic VBF coupling modifier ($\kappa_{2V}$).

• Trilinear Coupling ($\kappa_\lambda$):
  • Observed 95% CL interval: $-1.6 < \kappa_\lambda < 6.6$.
  • Expected interval: $-1.8 < \kappa_\lambda < 6.9$.
• Quartic VBF Coupling ($\kappa_{2V}$):
  • Observed 95% CL interval: $-0.5 < \kappa_{2V} < 2.6$.
  • Expected interval: $-0.4 < \kappa_{2V} < 2.6$.
• 2D Fit: A simultaneous fit yields $\kappa_\lambda = 2.8^{+2.2}_{-3.0}$ and $\kappa_{2V} = 1.6^{+0.6}_{-1.8}$.

5. Significance

This paper represents the most stringent constraint to date on Higgs boson pair production in the $b\bar{b}\gamma\gamma$ channel using LHC data.

• Validation of SM: The results are fully compatible with the Standard Model prediction ($\mu_{HH} = 1$), confirming the current understanding of the Higgs potential within the experimental precision.
• New Physics Constraints: The constraints on $\kappa_\lambda$ and $\kappa_{2V}$ significantly narrow the parameter space for BSM theories that predict anomalous Higgs self-couplings or modified VBF interactions.
• Technological Milestone: The successful integration of transformer-based neural networks (GN2) and advanced kinematic fitting techniques demonstrates the maturity of ATLAS analysis methods for the High-Luminosity LHC (HL-LHC) era, where even greater precision will be required to detect subtle deviations in the Higgs potential.