Uncovering the Origins of Mass: Discovery of the di-Higgs Process and Constraints on the Higgs Self-Coupling

Informazioni relative al progetto

HiggsSelfCoupling

ID dell’accordo di sovvenzione: 787331

DOI

10.3030/787331

Progetto chiuso

Data della firma CE 27 Giugno 2018

Data di avvio 1 Giugno 2019

Data di completamento 31 Maggio 2025

Finanziato da

• EXCELLENT SCIENCE - European Research Council (ERC)

Costo totale

€ 2 262 897,00

Contributo UE

€ 2 262 897,00

Coordinato da

HUMBOLDT-UNIVERSITAET ZU BERLIN
Germany

Periodic Reporting for period 4 - HiggsSelfCoupling (Uncovering the Origins of Mass: Discovery of the di-Higgs Process and Constraints on the Higgs Self-Coupling)

Periodo di rendicontazione: 2023-12-01 al 2025-05-31

Sintesi del contesto e degli obiettivi generali del progetto

The Higgs boson, discovered at CERN’s Large Hadron Collider in 2012, is central to our understanding of how elementary particles acquire mass. Yet, we still don’t fully grasp the mechanism behind this process—electroweak symmetry breaking—or why the Higgs potential has the shape it does. These are not just academic questions: they touch on the stability of the universe and the very foundations of particle physics.
One of the most direct ways to probe these mysteries is by measuring how the Higgs boson interacts with itself—its self-coupling. This measurement is uniquely sensitive to the shape of the Higgs potential and could reveal physics beyond the Standard Model. However, observing the self-coupling experimentally is extremely challenging. The most promising signature is the production of Higgs boson pairs, which often decay into four bottom quarks. But this process is rare, and extracting the signal from the overwhelming background requires cutting-edge techniques.
In this project, we aimed to solve the challenges of showing evidence for the Higgs self-coupling in the most common of the experimental signatures, the pair-production of Higgs bosons followed by the most common decay into a pair of bottom-quarks, generating a unique '4b' dataset that we study with the ATLAS experiment at CERN.
We developed a new analysis framework for the ATLAS ‘4b’ Higgs boson pair dataset. Using data from the ATLAS experiment, we developed new algorithms to improve the reconstruction of bottom quark decays, designed advanced machine learning tools to better identify Higgs pair events, and optimized the ATLAS trigger system to capture more relevant collisions. Our goal was to push the boundaries of what’s possible with the LHC’s Run 3 dataset, maximizing the scientific return.
As a large publicly-funded project, maximising the exploration of the LHC dataset is an essential goal and responsibility for the particle physics community. This project enhanced the value of the ongoing Run 3 data collection for di-Higgs searches by over 50% due to better signal acceptance, and the collective improvements to the Higgs self-coupling determination by the ATLAS collaboration make it possible to achieve in 2025 a precision that would otherwise only be reached beyond 2030.

Lavoro eseguito dall’inizio del progetto fino alla fine del periodo coperto dalla relazione e principali risultati finora ottenuti

The team was formed in 2020/2021 amidst the COVID-19 pandemic. In these unusual conditions, essential contributions were made to complete the search for Higgs pair-production with decays to four bottom quarks (HH→4b) using data from LHC Run 2. This analysis, published in 2023, performed beyond expectations, leading to the best contemporary sensitivity to HH production in combination with other ATLAS analyses. The methods behind this search were effectively applied to other searches for gravitons and other hypothetical new particles that decay to a pair of Higgs bosons.

In parallel, the team laid the groundwork for improved searches with the LHC Run 3 dataset, collected from 2022–2026 and twice the size of the preceding dataset. In several areas, the group led the development of new methods and systems. A concerted effort to apply modern deep learning and transformer models to identify bottom-quark jets, and to fully exploit detector information to reconstruct particle jets, culminated in a new strategy for selecting HH-like events from LHC collisions. This strategy is 50% more effective not only in the HH→4b search channel but also benefits the case where one of the Higgs bosons decays into a pair of tau leptons. Simultaneously, these methods were extended to identify special individual Higgs boson signals in the form of merged bottom-quark jet pairs that arise when Higgs bosons are generated with unusually high energies, opening up a new analysis channel.

Blazing a trail for the search with the Run 3 data, the project developed a flexible data analysis framework that was easily extensible to a wide variety of ATLAS analyses, featuring a user-friendly configuration scheme. This framework was adopted by the ATLAS DiHiggs group as the basis for efficient studies on the new dataset, as well as by dozens of individual analysis teams. Using this framework, the Run 3 HH→4b analysis has been developed extensively and is progressing toward publication based on the ATLAS dataset up to 2023. The boosted Higgs-jet tagger was used to develop a spinoff analysis to explore the production of one Higgs boson alongside an additional undiscovered scalar boson, demanding novel background estimation strategies.

The development of new triggers and bottom-quark taggers has been published in several papers and presented in various conference presentations. These results were also featured in the ATLAS reports to the LHC Council as highlights of developments within the collaboration.

Progressi oltre lo stato dell’arte e potenziale impatto previsto (incluso l’impatto socioeconomico e le implicazioni sociali più ampie del progetto fino ad ora)

By applying and optimizing modern deep-learning neural network architectures—such as graph networks and transformers—to the challenge of identifying b-hadron jets in the ATLAS detector, the HiggsSelfCoupling project has significantly improved the signal-to-noise ratio for high-profile analyses of di-Higgs production. In doing so, it achieved a paradigm shift, demonstrating that the multiple tiers of bespoke identification algorithms previously used for b-jet identification could be replaced by sophisticated machine learning. Crucially, these advances were not achieved by brute-force computing alone but by tuning model architectures, input datasets, and training procedures to extract substantially more information than a simple yes/no identification decision.

Similarly, discrimination between jets from the target hard-scatter proton collision and background jets from other “pile-up” collisions has been advanced through the development of more effective input observables. These observables better partition the available information about jet origins, unlocking further innovations that will be essential for the future High-Luminosity LHC era. At that stage, the levels of such backgrounds will be two to three times higher than current operating conditions.

This is an ATLAS data event recorded in 2018. It shows a diHiggs candidate decaying to 4 b-jets.