Search for a resonance decaying into a scalar particle and a Higgs boson in the final state with two bottom quarks and two photons with 199 fb$^{-1}$ of data collected at $\sqrt{s}$=13 and 13.6 TeV with the ATLAS detector

Using 199 fb$^{-1}$ of proton-proton collision data collected by the ATLAS detector at 13 and 13.6 TeV, a search for a heavy scalar resonance decaying into a lighter scalar and a Higgs boson in the $b\bar{b}\gamma\gamma$ final state found no significant excess over the Standard Model background, leading to the setting of 95% confidence level upper limits on the production cross-section times branching fraction.

The Gist

Imagine the Large Hadron Collider (LHC) as a massive, high-speed particle racetrack where protons (tiny subatomic particles) are smashed together at nearly the speed of light. When they collide, they create a chaotic explosion of energy that briefly forms new, heavier particles before they instantly decay into lighter, more stable ones.

This paper is a report from the ATLAS Collaboration, a team of scientists using a giant detector (like a 3D camera) to watch these collisions. They are looking for a very specific, rare event: a "heavy parent" particle decaying into a "lighter child" particle and a famous "Higgs boson."

Here is the story of their search, explained simply:

The Mystery: A Heavy Parent and Two Children

The scientists are hunting for a hypothetical heavy particle they call $X$.

• The Theory: They believe $X$ might be a "parent" particle that doesn't last long. When it dies, it splits into two "children":
  1. A lighter scalar particle called $S$.
  2. The famous Higgs boson (the particle discovered in 2012 that gives other particles mass).
• The Decay Chain:
  • The Higgs boson immediately turns into two photons (particles of light).
  • The lighter particle $S$ immediately turns into two bottom quarks (which behave like jets of energy in the detector).
• The Goal: They want to find the "fingerprint" of this specific family tree: Two Photons + Two Bottom Quarks.

The Search Strategy: Finding a Needle in a Haystack

Imagine trying to find a specific, rare coin in a massive pile of dirt. The "dirt" is the background noise of billions of ordinary particle collisions happening every second. The "rare coin" is the signal they are looking for.

1. The Filter (Triggers): The detector is too busy to record every collision. It uses a "smart filter" to only save events where two high-energy flashes of light (photons) appear together.
2. The Identification (Tagging): Once they have a candidate event, they look for the "bottom quarks." They use a special algorithm (a type of AI called GN2) to identify jets of energy that likely came from bottom quarks. They look for events with either one or two of these "bottom-tagged" jets.
3. The Mass Check: They calculate the total weight (mass) of the particles.
   • The two photons should weigh about 125 GeV (the known weight of the Higgs).
   • The two bottom quarks should weigh whatever the lighter particle $S$ weighs.
   • The total weight of everything combined should reveal the weight of the heavy parent $X$.

The Improvements: A Sharper Lens

This paper is an update to a previous search. The team didn't just look at more data; they looked better.

• More Data: They combined data from two different eras of the LHC (Run 2 and the early part of Run 3), giving them a much larger "haystack" to search through.
• Better Tools: They upgraded their "AI" for spotting bottom quarks, making it more efficient at spotting the real thing and ignoring fake signals.
• Tighter Focus: They narrowed the window for the Higgs mass (the two photons), which helped cut out more of the background noise.

The Results: No New Particles Found

After analyzing 199 femtobarns of data (a massive amount of collision records), the team looked for a "bump" in the data—a sudden spike in the number of events that would indicate a new particle $X$ exists.

• The Outcome: They found no significant excess. The data looked exactly like what the Standard Model (our current best theory of physics) predicts for background noise.
• The "Ghost" Signal: In a previous search using older data, there was a small, intriguing "bump" at a specific mass (575 GeV) that looked like it might be a new particle. However, with this new, larger, and more precise dataset, that bump disappeared. It was likely just a statistical fluke or a misunderstanding of the background noise.

The Conclusion: Setting Limits

Even though they didn't find the new particle, the search wasn't a failure. In science, knowing what isn't there is just as important as knowing what is.

The team set strict limits on how heavy or how common this hypothetical particle $X$ could be. They essentially said:

"If this particle $X$ exists, it must be rarer than we can currently detect, or it must have a mass outside the range we tested."

They ruled out the existence of this particle for masses between 170 and 1000 GeV (for the heavy parent) and 15 and 500 GeV (for the lighter child), assuming it decays in this specific way.

In short: The ATLAS team used a super-powerful microscope to scan the universe's most energetic collisions for a specific, rare family of particles. They didn't find the family, but they successfully mapped out exactly where the family cannot be hiding, narrowing the search for future discoveries.

Technical Summary

Technical Summary: Search for a Resonance Decaying into a Scalar Particle and a Higgs Boson ($X \to S(\to b\bar{b})H(\to \gamma\gamma)$)

Problem and Motivation
While the Higgs boson ($H$) discovered in 2012 aligns well with Standard Model (SM) predictions, experimental data does not preclude the existence of additional scalar states predicted by various Beyond the Standard Model (BSM) theories, such as complex singlet extensions, Two-Higgs-Doublet Models (2HDM), and the Next-to-Minimal Supersymmetric Standard Model (NMSSM). This paper addresses the search for a heavy scalar resonance $X$ decaying into a lighter scalar $S$ and the SM Higgs boson ($X \to SH$), where $S$ decays to a pair of bottom quarks ($b\bar{b}$) and $H$ decays to a pair of photons ($\gamma\gamma$). This specific final state ($b\bar{b}\gamma\gamma$) offers unique sensitivity, particularly in the low-mass region for $S$, compared to other decay channels. Previous ATLAS and CMS analyses using Run-2 data have set limits but observed local deviations (e.g., ATLAS at $(m_X, m_S) = (575, 200)$ GeV and CMS at $(650, 90)$ GeV) that warrant further investigation with increased statistics and improved analysis techniques.

Methodology
The analysis utilizes proton-proton collision data collected by the ATLAS detector at the LHC, corresponding to an integrated luminosity of 140 fb$^{-1}$ at $\sqrt{s}=13$ TeV (Run 2) and 58.6 fb$^{-1}$ at $\sqrt{s}=13.6$ TeV (early Run 3).

• Event Selection: Events are selected requiring two photon candidates with an invariant mass ($m_{\gamma\gamma}$) between 105 and 160 GeV, and transverse energies satisfying $E_T > 0.35 \times m_{\gamma\gamma}$ and $E_T > 0.25 \times m_{\gamma\gamma}$. To suppress $t\bar{t}H$ backgrounds, events containing identified electrons or muons are rejected.
• Signal Regions (SR): The analysis defines a tight SR window around the Higgs mass ($122.5 < m_{\gamma\gamma} < 127.5$ GeV), narrowing the window from the previous analysis to reduce background by half while retaining 85–90% of the signal. Events are categorized into two exclusive regions based on the number of $b$-tagged jets: exactly one $b$-jet and exactly two $b$-jets. This categorization addresses kinematic variations where the $S$ boson is highly boosted (collimated $b$-jets merging into one) or has low momentum (one $b$-jet not reconstructed).
• Object Reconstruction: Photon identification uses "Tight" criteria. Jets are reconstructed using the anti-$k_t$ algorithm ($R=0.4$) with particle-flow inputs. A significant upgrade involves the use of the GN2 transformer neural network for $b$-tagging, achieving 85% efficiency (up from 77% in the previous DL1r algorithm) with comparable light-jet rejection.
• Multivariate Analysis (MVA): Parameterized Neural Networks (PNNs) are employed to discriminate signal from background. The PNNs take event kinematic features (e.g., invariant masses of $b$-jet systems, modified invariant masses $m^*_{bb\gamma\gamma}$ and $m^*_{b\gamma\gamma}$ designed to decorrelate from $m_{\gamma\gamma}$) and phase-space parameters ($m_X, m_S$) as inputs. This allows the training of a single discriminant that interpolates across the entire mass plane ($170 \le m_X \le 1000$ GeV, $15 \le m_S \le 500$ GeV).
• Background Estimation: The dominant non-resonant $\gamma\gamma$+jets background shape is modeled using Sherpa MC, while its normalization is derived from data in Control Regions (CR) defined outside the Higgs mass window. Other backgrounds (e.g., $t\bar{t}\gamma\gamma$, $Z\gamma\gamma$, SM Higgs production) are normalized using theoretical cross-sections. Systematic uncertainties, including those from luminosity, jet energy scale, $b$-tagging, and theoretical modeling, are treated as nuisance parameters in the fit.

Key Contributions and Improvements
This work represents an update to the previous Run-2 analysis [18] with several specific technical advancements:

1. Data Volume: Inclusion of early Run-3 data (13.6 TeV), increasing the total dataset by approximately 42%.
2. Algorithmic Upgrades: Implementation of the GN2 $b$-tagging algorithm, improving $b$-jet efficiency by 8% on average.
3. Refined Selection: A narrower $m_{\gamma\gamma}$ signal window (122.5–127.5 GeV) to significantly reduce background.
4. Calibration Updates: Revised $b$-jet energy corrections and updated photon identification/isolation criteria to account for Run-3 pile-up conditions.
5. Training Strategy: Re-training of the PNNs with updated inputs and a refined definition of signal regions.

Results
The analysis performs a simultaneous binned maximum-likelihood fit to the PNN output distributions in the SRs and CRs for both data-taking periods and $b$-tag categories.

• Observation: No significant excess over the SM background-only hypothesis is observed. The maximum local significance is 2.0 standard deviations, found at $(m_X, m_S) = (235, 60)$ GeV.
• Previous Anomalies: The local excess previously observed by ATLAS at $(575, 200)$ GeV (3.5$\sigma$) and the CMS deviation at $(650, 90)$ GeV are not confirmed in this analysis. The events contributing to the earlier ATLAS excess were found to fail current selection criteria due to updated photon/electron reconstruction and $b$-tagging, or falling outside the tighter $m_{\gamma\gamma}$ window.
• Limits: 95% Confidence Level (CL) upper limits are set on the product of the production cross-section and branching fraction ($\sigma \times \mathcal{B}$) for the process $X \to S(\to b\bar{b})H(\to \gamma\gamma)$ at 13 TeV. The limits range from 9.0 fb at $m_X=170$ GeV, $m_S=30$ GeV to 0.06 fb at $m_X=1000$ GeV, $m_S \in [175, 300]$ GeV.

Significance
The paper claims a significant improvement in sensitivity across the entire probed mass plane compared to the previous Run-2-only analysis. The expected limits on the cross-section are improved by 15% to 73%, with the most substantial gains (up to 73%) occurring in the low-mass region. This improvement is attributed to the combination of Run-3 data, the tighter $m_{\gamma\gamma}$ selection, the higher efficiency of the GN2 $b$-tagging algorithm, and the updated PNN training. The results provide the most stringent constraints to date on this specific BSM decay topology, effectively ruling out the previously observed local deviations as statistical fluctuations or artifacts of previous analysis configurations.