⚛️ high-energy experiments

Improved analysis of non-resonant Higgs boson pair production in the $b\bar{b}\tau^+\tau^-$ final state with $196$ fb $^{-1}$ of data collected at $\sqrt{s}$ = 13 TeV and 13.6 TeV with the ATLAS detector

Using an updated dataset of 196 fb⁻¹ collected at 13 TeV and 13.6 TeV, the ATLAS collaboration reports a non-resonant Higgs boson pair production cross-section consistent with Standard Model expectations and derives new constraints on the Higgs self-coupling modifier, while also validating the analysis strategy through measurements of ZH and ZZ processes.

Original authors: ATLAS Collaboration

Published 2026-07-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: ATLAS Collaboration

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Great Higgs Hunt: Chasing Ghosts in a Particle Storm

Imagine the universe as a giant, invisible ocean. For decades, physicists have known that this ocean exists, but they couldn't quite see the waves. Then, in 2012, they found a splash: the Higgs boson. Think of the Higgs boson as a celebrity at a party. When it walks through a crowd of other particles, people stop to talk to it, slowing it down and giving it "mass." Without this interaction, particles would zip around at the speed of light, and atoms (and you, and me) couldn't exist.

But finding the celebrity was only step one. The real mystery is how the celebrity talks to itself. Does the Higgs boson have a "self-coupling"? In the Standard Model (our best rulebook for how the universe works), the Higgs should be able to interact with another Higgs, creating a pair. This is like the celebrity inviting a twin to the party. If we can catch two Higgs bosons hanging out together, we can measure exactly how they interact. This interaction holds the key to understanding why the universe is stable and how it began. However, catching a pair of Higgs bosons is incredibly hard. They are rare, they vanish instantly, and they leave behind a messy trail of other particles that look very similar to the background noise of the universe.

The ATLAS Team's New Detective Work

In this paper, the ATLAS Collaboration at CERN's Large Hadron Collider (LHC) acts like a team of high-tech detectives trying to spot that rare Higgs pair. They didn't just look at old clues; they combined a massive amount of new data. They analyzed 196 fb⁻¹ of data (a unit of how much particle collision data they collected), split between two energy levels: 13 TeV (from 2015–2018) and 13.6 TeV (from 2022–2023).

To find the Higgs pair, the team looked for a specific "fingerprint": a final state where one Higgs decays into two bottom quarks (which turn into jets of particles) and the other decays into two tau leptons (heavy cousins of electrons). This specific combination, written as 𝑏¯𝑏𝝉+𝝉−, is like looking for a specific type of shoe print in a muddy field. It's not the most common print, but it's one of the clearest ways to tell a Higgs pair apart from the millions of other particle collisions happening every second.

The New Tools
The team didn't just use more data; they upgraded their magnifying glass. They introduced a new "multivariate analysis" strategy based on transformer architectures. If the old method was like a human detective looking at a single clue at a time, this new AI-based system is like a super-intelligent detective that looks at the entire crime scene at once, understanding how every piece of evidence relates to every other piece. They also used a newer, smarter algorithm to identify "flavor" (distinguishing bottom quarks from other particles) and updated their trigger systems to catch more events.

The Findings: A Glimmer, Not a Shout
After crunching the numbers, the team found something interesting, but not a slam-dunk discovery.

The Signal: They observed a signal that is 2.6 standard deviations above the "background-only" hypothesis (the idea that there are no Higgs pairs at all). In the world of particle physics, a "5-sigma" result is needed to claim a formal discovery. So, while this is a very exciting hint—like seeing a shadow that looks exactly like the celebrity—it isn't quite loud enough to say, "We found them!" for sure.
The Strength: They measured the "signal strength" (how often these pairs appear compared to what the Standard Model predicts) to be 𝜇𝐻𝐻 = 2.6+1.4−1.0. This number is higher than the expected value of 1.0, suggesting there might be more Higgs pairs than the Standard Model predicts, but the uncertainty is large enough that it could still be a statistical fluke.
The Self-Coupling: Because they didn't find a definitive signal, they set limits on the "Higgs self-coupling modifier" (𝜅𝜆). This number tells us how strongly the Higgs talks to itself. The paper rules out certain values, stating that with 95% confidence, the true value lies in the range 𝜅𝜆 ∈ [−3.4, 1.6] ∪ [5.5, 10.1]. Essentially, they narrowed the search, but the answer is still hiding in the gaps.

Validating the Method
To prove their new detective tools were working correctly, the team also looked for two other known processes: 𝑍𝐻 (a Z boson and a Higgs) and 𝑍𝑍 (two Z bosons).

They found evidence for the 𝑍𝐻 process with a significance of 3.5 standard deviations. This is a strong hint (over 3 sigma) that their method is solid.
They saw a 1.8 standard deviation hint for the 𝑍𝑍 process.
These results act as a "sanity check," confirming that their new AI and analysis techniques are correctly identifying particle interactions.

The Bottom Line
This paper doesn't claim to have solved the mystery of the Higgs self-coupling. Instead, it represents a massive step forward in the hunt. By combining more data, smarter AI, and better detectors, the ATLAS team has tightened the net around the Higgs boson. They haven't caught the fish yet, but they've definitely made the water much clearer, and they've found a few splashes that suggest the fish might be right there, just waiting to be reeled in. The results are consistent with the Standard Model, but the slight excess in the data keeps the hope alive that new physics might be lurking just around the corner.

Technical Summary: Improved Analysis of Non-Resonant Higgs Boson Pair Production in the $b\bar{b}\tau^+\tau^-$ Final State

Problem and Motivation
The measurement of the Higgs boson self-coupling is a critical objective in particle physics, as it probes the shape of the Higgs potential and the mechanism of electroweak symmetry breaking. While the Higgs boson was discovered in 2012, its trilinear self-coupling ( $\lambda_{HHH}$ ) remains weakly constrained. The most direct method to access this coupling at the Large Hadron Collider (LHC) is through the measurement of non-resonant Higgs boson pair ($HH$) production. The $b\bar{b}\tau^+\tau^-$ final state is one of the most sensitive channels for this search, offering a favorable compromise between signal yield and purity. This paper presents an updated search for non-resonant $HH$ production in this final state, utilizing the full Run 2 dataset and the early Run 3 dataset collected by the ATLAS detector.

Methodology and Analysis Strategy
The analysis utilizes $140\,\text{fb}^{-1}$ of proton-proton collision data at $\sqrt{s}=13\,\text{TeV}$ (Run 2, 2015–2018) and $56\,\text{fb}^{-1}$ at $\sqrt{s}=13.6\,\text{TeV}$ (Run 3, 2022–2023). The study focuses on two decay channels based on the $\tau$ -lepton decay modes: $\tau_{\text{lep}}\tau_{\text{had}}$ (one leptonic, one hadronic) and $\tau_{\text{had}}\tau_{\text{had}}$ (both hadronic).

Key methodological improvements over previous ATLAS results include:

Multivariate Analysis (MVA): A novel transformer-based architecture is employed for event categorization and signal-background discrimination. This replaces the boosted decision trees used in prior analyses. The transformer model processes both per-object features (jets, $\tau$ -candidates, leptons, missing transverse momentum) and global event-level features to capture complex correlations. The network is trained with four output heads: a multiclass classifier (distinguishing background, SM ggF, SM VBF, and BSM ggF with $\kappa_\lambda=5$ ), a binary classifier for the high-mass region ( $m_{HH} > 350\,\text{GeV}$ ), and two auxiliary tasks for top-quark and Higgs boson parentage.
Event Categorization: Events are assigned to three mutually exclusive signal regions (SRs) based on the transformer outputs and the reconstructed $HH$ invariant mass ( $m_{HH}$ $m_{H H}$ ):
- SR VBF: Targets Vector Boson Fusion topology ( $S_{\text{cat}} < 0.1$ ).
- SR Hi: Targets high-mass ggF events ( $m_{HH} > 350\,\text{GeV}$ ) collected via main-stream triggers.
- SR Lo: Contains remaining events, sensitive to anomalous $\kappa_\lambda$ values.
Object Reconstruction and Selection: The analysis incorporates state-of-the-art GN2 transformer-based flavour-tagging algorithms for $b$ -jet identification. Hadronic $\tau$ candidates are reconstructed using a recurrent neural network (RNN). The Missing Mass Calculator (MMC) is used to reconstruct the di- $\tau$ invariant mass, significantly improving mass resolution.
Background Estimation: Major backgrounds ( $t\bar{t}$ , $Z$ +jets, dibosons, single Higgs) are estimated using Monte Carlo simulations corrected by data-driven methods where necessary. Specifically, fake- $\tau_{\text{had}}$ backgrounds (from jets misidentified as $\tau$ s) are estimated using data-driven fake-factor methods in dedicated control regions. The $t\bar{t}$ background normalization is constrained by a dedicated $t\bar{t}$ control region.
Validation: The analysis strategy is validated by measuring $ZZ$ and $ZH$ production in the same final state using the same event selection and MVA strategy.

Key Results
The analysis yields the following primary results:

Signal Strength: The measured signal strength for $HH$ production is $\mu_{HH} = 2.6^{+1.4}_{-1.0}$ , where the uncertainty includes both statistical and systematic components. This result is consistent with the Standard Model (SM) prediction ( $\mu_{HH}=1$ ) within 1.65 standard deviations.
Significance: Relative to the background-only hypothesis, the observed significance is 2.6 standard deviations, compared to an expected significance of 1.2 standard deviations.
Coupling Constraints: Assuming all other Higgs couplings are SM-like, the observed 95% confidence level (CL) interval for the trilinear self-coupling modifier is $\kappa_\lambda \in [-3.4, 1.6] \cup [5.5, 10.1]$ . The expected interval is $\kappa_\lambda \in [-1.7, 8.5]$ . The quartic coupling modifier is constrained to $\kappa_{2V} \in [-0.2, 2.4]$ (observed) and $[0.0, 2.1]$ (expected).
Upper Limits: The observed 95% CL upper limit on the signal strength is $\mu_{HH} < 5.3$ , compared to an expected limit of 2.4 under the background-only hypothesis.
Validation Measurements:
- $ZH$ Production: Evidence for $ZH \to b\bar{b}\tau^+\tau^-$ is observed with a significance of 3.5 standard deviations (expected 2.4), with a measured signal strength of $\mu_{ZH} = 1.52^{+0.52}_{-0.48}$ .
- $ZZ$ Production: The $ZZ \to b\bar{b}\tau^+\tau^-$ process is measured with an observed significance of 1.8 standard deviations (expected 2.7), with $\mu_{ZZ} = 0.64^{+0.39}_{-0.36}$ . Both measurements are consistent with SM predictions.

Significance and Claims
The paper claims that this analysis represents a significant improvement in sensitivity over the previous ATLAS search in the same final state, with expected significance improvements ranging from 15% to 65% depending on the parameter of interest. The authors attribute these gains to:

The inclusion of the Run 3 dataset ( $56\,\text{fb}^{-1}$ at 13.6 TeV).
The implementation of the transformer-based MVA, which provides a 15–20% sensitivity boost.
Upgrades to the flavour-tagging algorithm (GN2), enhancing sensitivity by approximately 10%.
The removal of the electron veto requirement in the $\tau_{\text{had}}\tau_{\text{had}}$ channel, yielding a ~5% gain.
Improved trigger strategies in Run 3.

The observation of $ZH$ production in the $b\bar{b}\tau^+\tau^-$ channel with a significance exceeding 3 standard deviations is highlighted as a first-time observation in this specific final state, serving as a robust validation of the analysis methodology. The results provide the most stringent constraints to date on the Higgs self-coupling from the $b\bar{b}\tau^+\tau^-$ channel alone, contributing to the global picture of Higgs boson properties.

The Great Higgs Hunt: Chasing Ghosts in a Particle Storm

The ATLAS Team's New Detective Work

More like this