Search for anomalies in vector-boson fusion production of the Higgs boson in $H(\rightarrow \gamma\gamma) jj$ events using 164 fb$^{-1}$ of $pp$ collision data collected at $\sqrt{s}=13.6$ TeV with the ATLAS detector

Using 164 fb⁻¹ of 13.6 TeV proton-proton collision data collected by the ATLAS detector, this study employs matrix element-based observables and angular distributions to constrain CP-odd couplings and probe Higgs boson interactions with polarized vector bosons in vector-boson fusion events, confirming Standard Model predictions while achieving improved sensitivity through a new neural network classification algorithm.

The Gist

Imagine the universe as a giant, high-stakes dance floor. For decades, physicists have been trying to figure out the steps of the most important dancer on the floor: the Higgs boson. This particle is special because it gives mass to everything else, like a cosmic glue holding the universe together.

In 2012, scientists at the Large Hadron Collider (LHC) finally found this dancer. But now, they aren't just looking to see if it's there; they are watching how it moves to see if it's following the choreography written in the "Standard Model" (the universe's rulebook) or if it's improvising with new, unknown steps.

This paper from the ATLAS collaboration is like a high-definition security camera recording of that dance floor, specifically focusing on a very rare and tricky dance move called Vector Boson Fusion (VBF).

Here is the breakdown of what they did, explained simply:

1. The Setup: Catching a Ghost in a Storm

The Higgs boson is shy. It appears for a split second and then disappears, usually turning into other particles. In this study, the scientists were looking for a specific scenario:

• Two protons smash together.
• They exchange force-carrying particles (like $W$ or $Z$ bosons) that fuse together to create a Higgs.
• The Higgs immediately decays into two photons (particles of light) and two jets (sprays of particles).

The Analogy: Imagine trying to spot a specific firefly in a stadium full of fireworks. The "two jets" are like the smoke trails left behind by the fireworks, and the "two photons" are the bright flash of the firefly itself. The scientists had to filter out billions of other fireworks to find that one specific firefly. They analyzed 164 "inverse femtobarns" of data—that's a massive amount of collision data, equivalent to watching the stadium for several years straight.

2. The Two Investigations

The team ran two different "tests" on this data to see if the Higgs was behaving normally.

Test A: The Mirror Test (CP Violation)

• The Question: Does the Higgs look the same in a mirror?
• The Concept: In physics, there's a rule called CP symmetry. If you swap particles with their anti-particles (Charge) and look at them in a mirror (Parity), the laws of physics should stay the same.
• The Metaphor: Imagine a dancer spinning. If you watch them in a mirror, they should spin the same way. If the Higgs is "CP-odd," it's like a dancer who spins clockwise in real life but counter-clockwise in the mirror. This would be a huge deal because it could explain why the universe is made of matter instead of antimatter.
• The Tool: They used a "Neural Network" (a type of AI) to act as a super-sleuth. This AI looked at the angles and speeds of the particles and calculated an "Optimal Observable." Think of this as a "suspicion meter." If the meter reads zero, the Higgs is behaving normally. If it reads positive or negative, the Higgs might be breaking the mirror rule.
• The Result: The meter read zero. The Higgs is a good mirror-image dancer. No new physics found here, but it's a very precise confirmation of the rules.

Test B: The Spin Test (Polarization)

• The Question: How does the Higgs grab onto its dance partners?
• The Concept: The particles the Higgs fuses with ($W$ and $Z$ bosons) can vibrate in different ways, called polarizations. They can vibrate "longitudinally" (like a slinky stretching and compressing) or "transversely" (like a rope shaking side-to-side).
• The Metaphor: Imagine the Higgs is a handshake. Does it shake hands firmly with a partner who is stretching out (longitudinal) or one who is shaking side-to-side (transverse)? The Standard Model predicts a very specific ratio of these handshakes.
• The Tool: They looked at the angle between the two jets (the smoke trails). This angle changes depending on how the Higgs grabbed its partners.
• The Result: The handshake ratio matched the prediction perfectly. The Higgs is shaking hands exactly as the rulebook says it should.

3. The Secret Weapon: The AI and the "Fast Forward"

This paper is special for two technical reasons that made the results so sharp:

1. The AI Classifier: In the past, scientists used simple rules to separate the "good" events from the "bad" background noise. Here, they used a Neural Network (a deep learning AI) trained on millions of simulated events. It's like upgrading from a human security guard with a checklist to a facial recognition system that can spot a suspect in a crowd of a million people instantly. This made the signal much clearer.
2. The Fast Simulation (AF3): Usually, simulating how particles hit the detector takes a long time (like rendering a movie frame by frame). This team used a new tool called AtlFast3, which uses AI to "guess" the outcome of the particle collisions with 99% accuracy but 150 times faster. It's like using a high-speed drone to map a forest instead of walking every tree. This allowed them to process way more data than ever before.

4. The Grand Conclusion

After combining this new data with older data from previous years, the scientists found:

• The Higgs boson is exactly what we thought it was. It follows the Standard Model rules perfectly.
• It doesn't break the mirror rule (no CP violation).
• It shakes hands with the right partners in the right way (correct polarization).

Why does this matter if they found nothing new?
In science, proving that the "known" is still true is just as important as finding the "unknown." By tightening the rules and showing that the Higgs behaves exactly as predicted, they have closed the door on many wild theories about what the universe could be. They have drawn a very tight circle around the Standard Model.

If there is "New Physics" hiding out there, it's going to be very good at hiding, because the Higgs isn't giving up any secrets in this dance. The search continues, but now we know the Higgs is a very disciplined dancer indeed.

Technical Summary

1. Problem and Motivation

The Standard Model (SM) of particle physics successfully predicts the properties of the Higgs boson, yet it leaves fundamental questions unanswered, such as the origin of the matter-antimatter asymmetry in the universe. This motivates the search for physics Beyond the Standard Model (BSM).

• CP Violation: The known sources of CP violation in the SM are insufficient to explain the observed baryon asymmetry. The Higgs boson's coupling to vector bosons ($HVV$) is a sensitive probe for new CP-violating interactions.
• Polarization Dependence: In the SM, the Higgs boson couples to longitudinally polarized vector bosons to unitarize scattering amplitudes at high energies. Deviations in the coupling strengths to longitudinal ($L$) versus transverse ($T$) polarizations could indicate that the Higgs is a composite particle or that new physics exists.
• Channel Selection: The $H \to \gamma\gamma$ decay channel offers a clean experimental signature with excellent mass resolution. While the branching ratio is small, the Vector Boson Fusion (VBF) production mode provides distinct kinematic features (two forward jets) that allow for effective background suppression and access to the $HVV$ vertex structure.

2. Methodology and Analysis Strategy

Data and Simulation

• Dataset: The analysis uses proton-proton collision data collected by the ATLAS detector at $\sqrt{s} = 13.6$ TeV between 2022 and 2024, corresponding to an integrated luminosity of 164 fb$^{-1}$.
• Simulation: A significant methodological advancement is the exclusive use of AtlFast3 (AF3), a fast simulation tool combining parameterized approaches with machine learning, for the calorimeter response in all Monte Carlo (MC) samples. This replaces the traditional Geant4 simulation for the bulk of the samples, offering a speedup of ~150x while maintaining precision.
• Signal Modeling: VBF $H \to \gamma\gamma$ events are simulated using MadGraph5_aMC@NLO.
  • CP Study: Uses the SMEFT framework (Warsaw basis) to parameterize CP-odd couplings via the Wilson coefficient $c_{H\tilde{W}}$.
  • Polarization Study: Uses coupling-strength scale factors $a_L$ and $a_T$ for longitudinal and transverse polarizations.
• Background Modeling:
  • Resonant: Other Higgs production modes (ggF, $VH$, $t\bar{t}H$) modeled with Powheg/Herwig.
  • Non-resonant: Continuum $\gamma\gamma$+jets and misidentified jets ($\gamma j$, $jj$). These are estimated using a novel data-driven flow-matching method (based on Normalizing Flows) trained on sideband data ($m_{\gamma\gamma} \le 120$ GeV or $> 130$ GeV) to generate high-statistics background samples that accurately reproduce high-dimensional correlations.

Event Selection and Classification

• Reconstruction: Events require two high-$p_T$ photons ($E_T > 25$ GeV) and at least two jets ($p_T > 30$ GeV).
• VBF Topology: Selected events must have a large pseudorapidity gap between the two leading jets ($|\Delta\eta_{jj}| > 2.0$) and the diphoton system must be centrally located (Zeppenfeld variable).
• Neural Network (NN) Classifier: A multi-class neural network (PyTorch) is employed to separate VBF signal from dominant backgrounds (ggF Higgs and continuum $\gamma\gamma$).
  • Inputs: 16 features including jet kinematics, diphoton kinematics, and angular correlations.
  • Architecture: Three hidden layers (128, 64, 32 nodes) with dropout and layer normalization.
  • Training: Uses a flow-matching generated sample for the non-resonant background to ensure robust training. A two-fold cross-validation scheme prevents bias.
  • Output: Events are categorized into "tight," "medium," and "loose" signal regions based on the NN score ($D_{NN}$).

Observables

• CP Violation: Uses an Optimal Observable (OO) constructed from the interference term between SM and BSM amplitudes. The distribution of the OO is symmetric for CP-even processes but shifts for CP-odd contributions.
• Polarization: Uses the absolute azimuthal angular difference between the two VBF-tagged jets ($\Delta\Phi_{jj}$), which is highly sensitive to the polarization state of the intermediate vector bosons.

3. Key Contributions

1. First ATLAS Results with AtlFast3: This is the first ATLAS analysis to rely exclusively on the AF3 fast simulation for all signal and background MC samples, validating its viability for precision Higgs physics.
2. Flow-Matching for Backgrounds: Introduction of a flow-matching algorithm to generate high-statistics, data-driven non-resonant background samples, overcoming limitations of standard MC in modeling complex topologies.
3. First Polarization Measurement in $H \to \gamma\gamma$ VBF: This study provides the first determination of polarisation-dependent coupling scale factors ($a_L, a_T$) in the diphoton VBF channel, a channel previously limited by background modeling in other decay modes (e.g., $H \to WW$).
4. Run-3 Data: Utilization of the first full Run-3 dataset at 13.6 TeV, providing increased statistics and improved kinematic reach compared to Run-2.

4. Results

CP Violation Constraints

• The analysis constrains the Wilson coefficient $c_{H\tilde{W}}$ (assuming a new physics scale $\Lambda = 1$ TeV).
• Run-3 Result: The observed 95% Confidence Level (CL) interval is $[-0.35, 0.88]$.
• Combination: Combining Run-3 with the previous Run-2 ($\sqrt{s}=13$ TeV, 140 fb$^{-1}$) analysis yields a 95% CL interval of $[-0.23, 0.75]$.
• Improvement: The combined result represents a 50% improvement in sensitivity over the Run-2 result alone. The Run-3 analysis alone shows a 38% improvement, with 29% attributed to the new NN-based classification and flow-matching techniques.
• All results are consistent with the SM prediction of a CP-even Higgs boson ($c_{H\tilde{W}} = 0$).

Polarization-Dependent Couplings

• Constraints are placed on the scale factors $a_L$ (longitudinal) and $a_T$ (transverse).
• Shape-only fits: $a_L \in [0.84, 2.02]$ and $a_T \in [0.50, 1.20]$ (95% CL).
• Shape + Rate fits: Constraints tighten significantly to $a_L \in [0.93, 1.17]$ and $a_T \in [0.70, 1.14]$ (95% CL).
• Significance: These measurements represent a major improvement over previous $H \to WW^*$ measurements, reducing the width of the 95% CL intervals by more than a factor of three. The results are consistent with the SM expectation of $a_L = a_T = 1$.

5. Significance

This paper marks a significant milestone in Higgs boson characterization at the LHC:

• Validation of Fast Simulation: It demonstrates that fast simulation tools (AF3) can be used for precision measurements without sacrificing accuracy, which is crucial for the high-luminosity phase of the LHC where full Geant4 simulation becomes computationally prohibitive.
• Advanced Machine Learning: The successful integration of flow-matching for background generation and deep neural networks for signal classification sets a new standard for handling complex backgrounds in rare decay channels.
• BSM Sensitivity: The results provide some of the most stringent constraints to date on CP-violating $HVV$ couplings and polarisation-dependent couplings in the VBF channel, effectively ruling out large deviations from the SM in these specific sectors.
• Future Prospects: The methodology developed here is directly applicable to future Run-3 and High-Luminosity LHC analyses, enabling more precise tests of the Higgs sector as a portal to new physics.