Evidence of Higgs boson inclusive production at high transverse momentum decaying to a pair of $b$-quarks with the ATLAS detector

Using 301 fb$^{-1}$ of proton-proton collision data at 13 and 13.6 TeV collected by the ATLAS detector, this study reports the first evidence of Higgs boson inclusive production at high transverse momentum ($p_{\text{T}} > 450$ GeV) decaying to a $b\bar{b}$ pair, achieving a 3.8$\sigma$ observed significance through advanced jet identification and regression techniques.

The Gist

The Big Picture: Catching a Ghost at High Speed

Imagine the Large Hadron Collider (LHC) as the world's most powerful particle accelerator. It smashes protons together at nearly the speed of light, creating a chaotic explosion of debris. Among this debris, physicists are looking for a specific, very shy particle: the Higgs boson.

The Higgs boson is famous for giving other particles mass, but it's incredibly unstable. It lives for a split second and then instantly decays (falls apart) into other particles. In this specific study, the ATLAS team is looking for the Higgs boson decaying into a pair of bottom quarks (let's call them "b-quarks").

The Challenge:
Finding a Higgs boson is like trying to find a specific needle in a haystack. But this isn't just any haystack; it's a haystack made of billions of other needles that look exactly the same. The "haystack" here is the background noise of the universe—trillions of other particle collisions that happen constantly.

The New Strategy: The "High-Speed Chase"

In the past, scientists mostly looked for Higgs bosons moving slowly. But this paper focuses on high transverse momentum ($p_T$).

The Analogy:
Imagine you are trying to identify a specific type of car (the Higgs) in a busy city.

• Low Speed: If the car is driving slowly through traffic, it's hard to spot because it looks like every other car.
• High Speed: If that same car is speeding down the highway at 200 mph, it stands out. It leaves a distinct trail, and its behavior is different from the slow cars.

The scientists hypothesized that if the Higgs boson is created with massive energy (moving very fast), it might reveal new secrets about the laws of physics. However, catching these "speeding" Higgs bosons is incredibly hard because they are rare, and the background noise is even louder.

The Toolkit: AI and Better Glasses

To solve this, the ATLAS team used two major upgrades, which are the stars of this paper:

1. The "Transformer" AI (The Smart Detective):
   In the past, identifying the Higgs boson's decay products (the b-quarks) was like trying to identify a suspect in a crowd using a blurry photo. The team introduced a new Transformer-based algorithm (a type of advanced AI, similar to the tech behind modern chatbots).

   • How it works: Instead of just looking at the shape of the particle spray, the AI looks at the "topology" (the arrangement) of the particles inside the jet. It's like the detective realizing, "This suspect isn't just wearing a red hat; they are holding a specific type of umbrella and walking with a limp."
   • Result: This AI is so good at spotting the Higgs that it filters out 99.8% of the background noise, leaving a much cleaner signal.

2. The "Regression" Model (The Sharper Glasses):
   When particles fly out of the collision, they lose a bit of energy or get smeared out by the detector. The team used a new mathematical model to "sharpen" the image.

   • Analogy: Imagine looking at a blurry photo of a speeding car. This new model acts like a high-end photo editor that takes the blurry image and reconstructs the exact speed and mass of the car with incredible precision. This allowed them to measure the Higgs boson's mass much more accurately than before.

The Result: Evidence, Not Proof (Yet)

The team analyzed data from 2015 to 2024 (a massive amount of data, equivalent to 301 "inverse femtobarns"—a unit of collision volume).

• The Finding: They found a "bump" in the data at the mass of the Higgs boson (around 125 GeV) specifically in the high-speed category.
• The Significance: In physics, we need a "5-sigma" result to claim a discovery (like finding a new planet). This result is 3.8 sigma.
  • The Analogy: If you flip a coin 10 times and get 10 heads, it's suspicious. If you get 100 heads, it's proof the coin is rigged. A 3.8 sigma result is like getting 95 heads out of 100 flips. It's strong evidence that the coin is rigged, but not quite absolute proof yet. It's a very loud "Aha!" moment, but scientists are waiting for a bit more data to shout "Eureka!"

Why Does This Matter?

Why bother looking at Higgs bosons moving at 1,000 GeV (which is incredibly fast)?

• The "New Physics" Hunt: The Standard Model (our current rulebook of physics) predicts exactly how often these fast Higgs bosons should appear. If the number we find is different from the prediction, it means the rulebook is missing a chapter. It could hint at "New Physics"—like Supersymmetry or extra dimensions.
• The Current Status: The number they found (1.53 times the predicted amount) is slightly higher than expected, but the error bars are still large enough that it could just be a statistical fluke. However, it is compatible with the Standard Model, meaning our current rules are still holding up, even at these extreme speeds.

Summary

The ATLAS collaboration used super-smart AI and sharper measurement tools to hunt for Higgs bosons moving at extreme speeds. They found strong evidence (3.8 sigma) that these fast Higgs bosons exist and decay into bottom quarks, just as the Standard Model predicts.

It's like finding a rare, fast-moving bird in a storm. You haven't caught it in a cage yet (5 sigma), but you have a very clear photo and a strong hunch that it's real. This success paves the way for future experiments to either confirm the Standard Model is perfect or finally crack the code on what lies beyond it.

Technical Summary

1. Problem and Motivation

• Physics Context: While the Standard Model (SM) has been extensively validated, deviations in the Higgs boson's couplings at high energy scales could reveal New Physics (BSM). High transverse momentum ($p_T$) Higgs production is particularly sensitive to the loop structure of gluon-gluon fusion (ggF) and potential contributions from heavy new particles.
• The Challenge: The dominant Higgs decay channel is $H \to b\bar{b}$ (branching fraction $\approx 58\%$). However, reconstructing this channel at high $p_T$ ($>450$ GeV) is extremely difficult due to:
  • Huge Backgrounds: The QCD multijet background is orders of magnitude larger than the signal.
  • Boosted Topology: At high $p_T$, the two $b$-quarks from the Higgs decay become highly collimated, merging into a single large-radius jet.
  • Resolution: Distinguishing the Higgs mass peak ($\approx 125$ GeV) from the falling QCD background requires excellent jet mass resolution and precise $b$-tagging.
• Goal: To provide the first evidence of inclusive Higgs production in the $H \to b\bar{b}$ channel at high $p_T$ ($p_T > 450$ GeV) using the full Run 2 and early Run 3 datasets.

2. Methodology

The analysis utilizes proton-proton collision data recorded by the ATLAS detector at center-of-mass energies of $\sqrt{s} = 13$ TeV (Run 2) and $13.6$ TeV (Run 3), corresponding to an integrated luminosity of 301 fb$^{-1}$.

A. Event Selection and Reconstruction

• Trigger: Events are selected using high-level triggers requiring a large-radius jet with $p_T > 420$ GeV (online) and offline requirements of $p_T > 450$ GeV and jet mass $m_J > 45$ GeV.
• Jet Reconstruction: Jets are reconstructed using the anti-$k_t$ algorithm with radius parameter $R=1.0$ from Unified Flow Objects (UFO). They are groomed using Soft-Drop to mitigate pile-up and soft radiation effects.
• Signal Regions (SR): Events are categorized based on the tagging status of the leading and subleading jets:
  • SRL (Signal Region Leading): Leading jet passes the $b\bar{b}$ tagger.
  • SRS (Signal Region Subleading): Leading jet fails, but subleading jet passes.
  • Control Regions (CR) & Validation Regions (VR): Jets failing the tagger are used to constrain the shape of the multijet background.

B. Key Technological Innovations

The analysis achieves a sensitivity improvement of 7–10 times over previous Run 2 studies through two primary advancements:

1. Transformer-based $b\bar{b}$ Tagger (GN2X):
   • A deep learning model (based on Transformer architecture) trained on simulated samples to identify jets containing two $b$-hadrons.
   • It exploits kinematic and topological properties of jet constituents.
   • Performance: Suppresses multijet background by a factor of $\sim 500$ while maintaining $\sim 45\%$ efficiency for signal jets. It effectively removes $V \to q\bar{q}'$ and $t\bar{t}$ contributions from the signal region.
2. Transformer-based Regression (bJR):
   • A regression model used to correct the jet mass and $p_T$ response.
   • Performance: Improves jet mass resolution by $\sim 30\%$ and $p_T$ resolution by $\sim 20\%$, significantly sharpening the signal peak.

C. Statistical Analysis

• Fit Strategy: A simultaneous profile likelihood fit is performed on the jet mass ($m_J$) distributions across 12 regions (SRL, SRS, CRL, CRS) and three $p_T$ bins (450–650, 650–1000, >1000 GeV).
• Background Modeling: The dominant QCD multijet background is modeled using an exponentiated polynomial function ($f_N$) for the CR and a transfer function ($t_M$) to map CR shapes to SR shapes. This allows the background shape to be determined entirely from data.
• Calibration: The $Z \to b\bar{b}$ peak is used in situ to calibrate the $b\bar{b}$ tagging efficiency ($\kappa_{GN2X}^{b\bar{b}}$), jet mass scale (JMS), and mass resolution (JMR).
• Signal Extraction: The signal strength modifier $\mu_H$ (ratio of observed yield to SM prediction) is extracted simultaneously with nuisance parameters for systematic uncertainties.

3. Key Results

• Overall Significance:
  • For $p_T > 450$ GeV, the observed significance is 3.8$\sigma$ (expected 2.5$\sigma$), constituting the first evidence of inclusive Higgs production in this kinematic regime.
  • The measured signal strength is:
    $$\mu_H = 1.53 \pm 0.27 \text{ (stat.)} ^{+0.33}_{-0.27} \text{ (syst.)} \pm 0.17 \text{ (theo.)}$$
  • This result is compatible with the Standard Model prediction ($\mu_H = 1$).
• Differential Measurements:
  • Results are provided in three $p_T$ intervals:
    • 450–650 GeV: $\mu_H = 1.98 \pm 0.36 \text{ (stat.)} \dots$ (Observed significance: 3.9$\sigma$).
    • 650–1000 GeV: $\mu_H = 0.23 \pm 0.67 \dots$ (Observed significance: 0.3$\sigma$).
    • >1000 GeV: $\mu_H = 2.3 \pm 1.4 \dots$ (Observed significance: 1.3$\sigma$).
  • All intervals are consistent with SM predictions within uncertainties.
• Cross-Section Measurements:
  • Separate cross-section measurements for Run 2 (13 TeV) and Run 3 (13.6 TeV) were performed, yielding values consistent with SM expectations, though with large uncertainties in the highest $p_T$ bins.

4. Significance and Impact

• First Evidence: This is the first observation of evidence for inclusive $H \to b\bar{b}$ production at high $p_T$. Previous studies were limited to associated production ($VH$) or had lower sensitivity.
• Technological Milestone: The paper demonstrates the successful application of Transformer-based neural networks in high-energy physics for both object identification (tagging) and regression (calibration). This marks a shift from traditional boosted decision trees (BDTs) to more advanced deep learning architectures in ATLAS analyses.
• Background Suppression: The ability to suppress the QCD multijet background by a factor of 500 while retaining signal efficiency is a critical breakthrough, enabling the study of rare Higgs processes in the TeV scale.
• BSM Sensitivity: By measuring the Higgs $p_T$ spectrum up to and beyond 1 TeV, this result provides stringent constraints on Effective Field Theory (EFT) operators and models of New Physics that would alter the Higgs-gluon coupling at high energies.
• Future Prospects: The methodology and the improved resolution set the stage for future Run 3 and Run 4 analyses, where increased luminosity will allow for precision measurements of the Higgs $p_T$ spectrum and potentially the first observation (5$\sigma$) of this channel.

In conclusion, this paper represents a major step forward in Higgs physics, combining advanced machine learning techniques with large datasets to probe the Higgs boson in a previously inaccessible kinematic regime, confirming Standard Model predictions while opening new windows for discovering physics beyond the Standard Model.