Improved results on Higgs boson pair production in the 4b final state

Using proton-proton collision data at 13 and 13.6 TeV collected by the CMS experiment, this paper presents improved measurements of Higgs boson pair production in the 4b final state, achieving the most stringent constraints to date on the signal strength and Higgs self-couplings through advanced analysis techniques in both resolved and merged jet topologies.

The Gist

The Great Higgs Double-Double Hunt: A CMS Collaboration Story

Imagine the Large Hadron Collider (LHC) as the world's most powerful particle smasher. Scientists at CERN's CMS experiment are like detectives trying to find a very specific, incredibly rare crime scene: the moment two Higgs bosons are created at the same time.

Why is this important? The Higgs boson is the particle that gives other particles their mass. But physicists want to know how Higgs bosons interact with each other. Do they hug? Do they push each other away? The answer lies in a number called the "trilinear coupling" (think of it as the strength of the Higgs handshake). Measuring this helps us understand the fundamental shape of the universe's energy landscape.

However, finding two Higgs bosons is like finding two specific needles in a haystack the size of a planet. The "haystack" is a massive amount of background noise from other particle collisions.

The Challenge: The "Four-Bottom" Mystery

When a Higgs boson decays, it often turns into a pair of "bottom quarks" (heavy particles that quickly turn into sprays of other particles called jets). So, the scientists are looking for two Higgs bosons, which means they are looking for four bottom quarks (or "4b" in physics shorthand).

The problem? The universe loves making four bottom quarks all the time through ordinary, boring processes. It's like trying to hear a whisper at a rock concert. The "whisper" is the signal (the two Higgs bosons), and the "rock concert" is the background noise (ordinary particle collisions).

The Strategy: Two Ways to Listen

Because the Higgs bosons can be moving at different speeds, the scientists had to look for them in two different "topologies" (ways they appear):

1. The "Resolved" Topology (The Slow Walkers):
   Imagine two Higgs bosons moving relatively slowly. Their decay products (the four bottom quarks) spread out enough to be seen as four separate, distinct jets.

   • The Analogy: It's like seeing four distinct people walking in a crowd. You can count them easily, but it's hard to tell which two belong to the same group because there are so many other people around.

2. The "Merged" Topology (The Speedsters):
   Imagine two Higgs bosons moving incredibly fast. Their decay products are squished together so tightly that they merge into two giant, single jets.

   • The Analogy: It's like two people running so fast they blur into a single streak. You can't see them as individuals, but you can see the giant streak they leave behind.

The New Tools: Sharper Eyes and Faster Triggers

The paper describes a major upgrade in how the CMS experiment "sees" these events. They introduced new tools to filter out the noise:

• The "Smart Trigger" (The Bouncer):
  The LHC produces millions of collisions per second. The computer system (the trigger) has to decide in a microsecond which ones to save. In the past, the bouncer was too strict and let many interesting events slip away.

  • The Upgrade: They installed a new, AI-powered bouncer (called PNET@HLT) that is much better at spotting the specific "footprints" of bottom quarks. It's like upgrading from a bouncer who just looks at shoes to one who recognizes the specific gait of the VIPs. This allowed them to save twice as many potential Higgs events.

• The "Jet Regressor" (The GPS):
  When particles fly out, they lose some energy (like a car losing speed on a hill). The scientists used a new machine-learning algorithm to predict exactly how fast the original particles were moving, correcting for the energy lost.

  • The Analogy: It's like a GPS that doesn't just tell you where you are, but calculates exactly how fast you were driving before you hit a pothole, giving you a much clearer picture of the journey.

• The "Mass Minder" (The Scale):
  They also improved how they measure the "weight" (mass) of the giant merged jets. They used a new algorithm called GLOPART that acts like a super-precise scale, distinguishing between a heavy Higgs boson and a lighter, ordinary particle that just happens to look similar.

The Results: Finding the Limit

The scientists analyzed data from 2022–2023 (Run 3) and combined it with older data from 2015–2018 (Run 2).

• Did they find the Higgs Double-Double?
  Not yet. They didn't see a clear "bump" in the data that would prove the signal is there. The data looks mostly like the background noise.
• But they set a new "Speed Limit":
  Even without finding the event, they can say: "If this event happens, it cannot happen more than 4.4 times as often as the Standard Model predicts."
  • The Analogy: Imagine you are looking for a rare bird. You don't see it, but you can confidently say, "If this bird exists here, there are no more than 4 of them in this forest."

The Improvements:

• Compared to previous results, their ability to set this limit has improved by more than a factor of two in the "resolved" (slow walker) category.
• They also improved the "merged" (speedster) category.
• By combining the new data with old data, they set the strictest limit yet on how the Higgs boson interacts with itself.

The Bottom Line

The paper concludes that while they haven't discovered a new physics phenomenon yet, they have built the most sensitive "net" ever created to catch the Higgs boson pair. They have narrowed down the possible behaviors of the Higgs boson more tightly than ever before.

If the Higgs boson behaves exactly as the Standard Model predicts, their current data is consistent with that. If it behaves differently (which would be a huge discovery), their new, sharper tools are now ready to catch it in the next round of data taking. For now, they have successfully ruled out many "wild" possibilities, bringing us one step closer to understanding the true nature of the universe's mass-giving mechanism.

Technical Summary

1. Problem Statement

The primary goal of the High-Luminosity LHC physics program is to measure the Higgs boson self-coupling ($\lambda$), which determines the shape of the Higgs potential and is crucial for understanding electroweak symmetry breaking and the stability of the universe. This is achieved by measuring the production of Higgs boson pairs ($HH$).

• The Challenge: The $HH \to b\bar{b}b\bar{b}$ (4b) decay channel has the largest branching fraction (~33%) but is extremely difficult to isolate due to the overwhelming Quantum Chromodynamics (QCD) multijet background, which is orders of magnitude larger than the signal.
• The Context: Previous Run 2 (13 TeV) results provided constraints but were limited by trigger inefficiencies, background modeling uncertainties, and lower integrated luminosity. This paper presents updated results using Run 3 data (13.6 TeV, 62 fb$^{-1}$) and an improved analysis of Run 2 data (138 fb$^{-1}$), aiming to significantly tighten constraints on the signal strength ($\mu_{HH}$) and coupling modifiers ($\kappa_\lambda$ and $\kappa_{2V}$).

2. Methodology

The analysis targets two distinct event topologies based on the transverse momentum ($p_T$) of the Higgs bosons:

1. Resolved Topology: Both Higgs bosons decay into two separated small-radius jets (AK4).
2. Merged Topology: Both Higgs bosons are highly boosted, and their decay products merge into single large-radius jets (AK8).

Key Technical Innovations

• Trigger Strategy (Run 3): A major bottleneck in Run 2 was the trigger efficiency. For Run 3, CMS introduced PNET@HLT, a Graph Neural Network (GNN) based algorithm running at the High-Level Trigger (HLT).
  • It performs $b$-tagging (for AK4) and $bb$-tagging (for AK8) directly at the trigger level.
  • This increased the absolute signal efficiency by more than a factor of two compared to Run 2, allowing for lower $p_T$ thresholds and the collection of larger control samples.
• Jet Reconstruction:
  • Jet $p_T$ Regression: A novel approach using PNET to simultaneously perform flavor identification and $p_T$ regression. This improved the $H \to b\bar{b}$ mass resolution by ~20% compared to previous methods.
  • GLOPART Algorithm: A global particle transformer used for AK8 jet identification and mass regression, improving the discrimination between $H \to b\bar{b}$ and QCD jets.
• Background Estimation:
  • Data-Driven Reweighting: Instead of relying on Monte Carlo (MC) simulations for the dominant QCD background, the analysis uses multidimensional Deep Neural Networks (DNNs) to reweight data events from Control Regions (CRs) to Signal Regions (SRs).
  • Two Approaches: Two independent analysis strategies were developed for each topology to cross-validate results:
    1. Transformer-based (Resolved): Uses a transformer architecture (FEYNNET) for event classification and background modeling, including validation in $ZZ/ZH$ regions.
    2. Parametric/GNN-based: Uses parametric extrapolation (alphabet method) for the merged analysis and GNNs for signal/background separation.
• Signal Extraction: Signal is extracted via binned maximum likelihood fits to classifier scores (SvsB) or reconstructed mass distributions ($m_{reg}$), treating systematic uncertainties as nuisance parameters.

3. Key Contributions

• First Run 3 Results in 4b: This is the first comprehensive measurement of $HH \to 4b$ using Run 3 data (13.6 TeV).
• Trigger Revolution: The implementation of PNET@HLT fundamentally changed the feasibility of the 4b search, doubling signal acceptance and enabling new background estimation techniques.
• Dual-Topology Combination: The paper provides a rigorous combination of Resolved and Merged topologies, carefully handling event overlaps to maximize sensitivity.
• Run 2 Update: An updated analysis of Run 2 data using the new Run 3 techniques (e.g., improved background modeling and $p_T$ regression) superseded previous Run 2 results, showing a 25% improvement in expected limits.
• Comprehensive Uncertainty Treatment: The analysis explicitly separates "variance" (statistical fluctuations in training data) from "bias" (systematic limitations of the method) in background predictions, providing a robust framework for future projections.

4. Results

The results are presented as 95% Confidence Level (CL) upper limits on the signal strength $\mu_{HH} = \sigma_{obs}/\sigma_{SM}$ and constraints on coupling modifiers.

• Run 3 (13.6 TeV, 62 fb$^{-1}$):
  • Resolved + Merged Combination: Observed (Expected) upper limit on $\mu_{HH}$ is 4.4 (4.4).
  • Coupling Constraints:
    • $\kappa_\lambda$ (trilinear coupling): Observed range $[-3.3, 9.7]$.
    • $\kappa_{2V}$ (quartic coupling): Observed range $[0.63, 1.43]$.
• Combined Run 2 + Run 3 (138 + 62 fb$^{-1}$):
  • Combined Upper Limit: Observed (Expected) upper limit on $\mu_{HH}$ is 4.7 (2.8).
  • Coupling Constraints:
    • $\kappa_\lambda$: Observed range $[-4.1, 9.6]$.
    • $\kappa_{2V}$: Observed range $[0.77, 1.27]$.
• Comparison to Previous Results:
  • The expected limit for the resolved topology with equivalent luminosity improved by more than a factor of two compared to previous Run 2 results.
  • The merged topology also showed significant improvement.
  • These are the most stringent constraints on $HH \to 4b$ production to date.

5. Significance

• Physics Impact: These results represent the most precise constraints on the Higgs self-coupling in the 4b channel to date. While the data is consistent with the Standard Model (no significant excess observed), the improved sensitivity narrows the window for Beyond-the-Standard-Model (BSM) physics scenarios that predict deviations in the Higgs potential.
• Methodological Benchmark: The successful deployment of GNNs (PNET, GLOPART) at the trigger level and for offline analysis demonstrates a new paradigm for handling high-background, low-signal searches at the LHC. The data-driven background estimation techniques reduce reliance on theoretical modeling of QCD, a major source of uncertainty.
• Future Outlook: The paper establishes a robust framework for the High-Luminosity LHC (HL-LHC) era. The separation of statistical and systematic uncertainties suggests that with the anticipated increase in luminosity, the precision of $\mu_{HH}$ and $\kappa_\lambda$ measurements will improve substantially, potentially reaching the sensitivity required to observe the Higgs self-coupling directly.