⚛️ phenomenology

High energy probes of Higgs self-coupling via $W$ boson fusion at future lepton colliders

This paper demonstrates that a graph neural network applied to $W$ boson fusion di-Higgs production at the 3 TeV CLIC collider can achieve a signal significance of approximately 20 $\sigma$ , offering superior sensitivity to the Higgs self-coupling and electroweak symmetry breaking mechanisms compared to the High-Luminosity LHC.

Original authors: Amir Subba, Hrishikesh Deka, Subhaditya Bhattacharya, Abhik Sarkar

Published 2026-08-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Amir Subba, Hrishikesh Deka, Subhaditya Bhattacharya, Abhik Sarkar

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

Imagine the universe as a giant, invisible ocean. For decades, physicists have been trying to figure out what happens when you drop a stone into this ocean. They know there's a "Higgs field" filling all of space, and when particles swim through it, they get heavy. This is the "Higgs mechanism," the reason why some things have mass and others don't. But there's a deeper mystery: how does this field talk to itself? If you push the field hard enough, does it push back in a predictable way, or does it squirm and twist in a way that hints at hidden, new physics? This question is like trying to understand the rules of a game by watching how the players bump into each other. If the players follow the standard rulebook perfectly, we know the universe is boringly safe. But if they bump in a weird way, it means there's a secret rulebook we haven't found yet.

To find these secret rules, scientists need to smash particles together with incredible force to create pairs of Higgs bosons (the "stones" in our ocean). The problem is, these pairs are rare, and the background noise from other particle crashes is deafeningly loud. It's like trying to hear a single whisper in a stadium full of cheering fans. The paper you are about to read tackles this challenge by looking ahead to a future machine called the Compact Linear Collider (CLIC), a giant particle accelerator that will smash electrons and positrons together at energies of 3 TeV. The authors ask: if we build this machine and use a super-smart computer brain to listen for the whisper, can we finally hear how the Higgs field talks to itself?

The Paper's Mission: Tuning the Radio in a Storm

The authors of this paper are essentially building a super-powered radio to listen for a specific signal in a very noisy room. They are focusing on a process called "W boson fusion," where two invisible W particles (carriers of the weak force) crash together to create a pair of Higgs bosons. At the high energies of the future CLIC machine, this becomes the most dominant way to make Higgs pairs. However, the signal is tiny, and the "noise" (background events from other particle interactions) is massive.

To solve this, the team didn't just rely on old-school filters. Instead, they developed a "Graph Neural Network" (GNN). Think of a particle collision as a messy party where everyone is dancing. A traditional filter might just count how many people are wearing red hats. A GNN, however, looks at the whole party: it sees who is standing near whom, how they are moving relative to each other, and the shape of the crowd. It treats every particle jet as a "node" and the space between them as "edges," creating a map of the event. This allows the computer to spot the subtle, unique dance moves of the Higgs pair that the background noise simply doesn't do.

What They Found: A Clear Whisper

Using simulations of what would happen at the CLIC with an energy of 3 TeV and a massive amount of data (5 inverse attobarns, or $5 \text{ ab}^{-1}$ ), the authors found that their GNN is incredibly effective. In their simulations, the classifier achieved a "signal significance" of about $Z \approx 20 \sigma$ . To put that in perspective, in the world of particle physics, a $5 \sigma$ result is the gold standard for claiming a discovery. A $20 \sigma$ result is like hearing that whisper so clearly that you could recite the whole speech without a single mistake. This suggests that with this machine and this smart computer, the Higgs self-coupling (how the Higgs talks to itself) could be measured with extreme precision.

The paper also explored two different "rulebooks" for how the universe might work.

The "Linear" Rulebook (SMEFT): Here, the way the Higgs talks to the W bosons and the way it talks to itself are tightly linked, like two gears in a clock. If you turn one, the other must turn.
The "Non-Linear" Rulebook (HEFT): Here, those gears are disconnected. The Higgs could talk to the W bosons one way and to itself in a completely different, independent way.

The authors found that their method is sensitive enough to disentangle these two scenarios by observing how the measurement precision changes under each assumption. If the universe follows the "Linear" rulebook, they can measure the Higgs self-coupling modifier ( $\kappa_\lambda$ ) to be between 0.76 and 1.31 (at 95% confidence). If the universe follows the "Non-Linear" rulebook, the range widens slightly, but the distinct sensitivities in the two scenarios allow physicists to distinguish between them. This is crucial because it means future colliders won't just measure a number; they could actually reveal the fundamental structure of how the universe breaks its own symmetry.

The Verdict

The paper concludes that high-energy lepton colliders, when paired with advanced graph-based machine learning, offer a "powerful probe" into the electroweak sector. They demonstrate that the CLIC, operating at 3 TeV, would be able to disentangle the different ways the Higgs field behaves, potentially revealing new physics that the current Large Hadron Collider (LHC) might miss. The results are based on detailed simulations and statistical analysis, showing that the combination of high energy and smart algorithms could turn the "whisper" of the Higgs self-coupling into a shout, finally letting us hear the rules of the universe's most fundamental field.

=== DRAFT ===

Technical Summary: High Energy Probes of Higgs Self-Coupling via W Boson Fusion at Future Lepton Colliders

Problem Statement
The precise measurement of the Higgs self-coupling ( $\lambda$ ) is central to validating the Standard Model (SM) scalar sector and understanding the mechanism of electroweak symmetry breaking (EWSB). While the discovery of the Higgs boson at the LHC confirmed the existence of the field, a direct determination of $\lambda$ requires di-Higgs production, a process with a small cross-section that is difficult to disentangle from large SM backgrounds. Current constraints from the LHC and projections for the High-Luminosity LHC (HL-LHC) face significant challenges in signal-to-background separation. Furthermore, deviations in Higgs couplings may arise from different theoretical frameworks: the Standard Model Effective Field Theory (SMEFT), which assumes a linearly realized EWSB, or the Higgs Effective Field Theory (HEFT), which assumes a non-linear realization. Distinguishing between these scenarios requires precise measurements of both the trilinear Higgs self-coupling modifier ( $\kappa_\lambda$ ) and the Higgs-gauge boson coupling modifiers ( $\kappa_V$ and $\kappa_{2V}$ ).

Methodology
This study investigates the sensitivity of the Compact Linear Collider (CLIC) operating at a center-of-mass energy of $\sqrt{s} = 3$ TeV to the Higgs self-coupling via W boson fusion (WBF) di-Higgs production ( $e^+e^- \to hh\nu\bar{\nu}$ ). At this energy scale, WBF becomes the dominant di-Higgs production mechanism, surpassing Z-associated production.

The analysis employs a two-stage strategy to discriminate the signal from overwhelming SM backgrounds (primarily single-Higgs associated production, multi-b production, and top-quark processes):

Optimized Selection Cuts: Initial cuts suppress reducible backgrounds while preserving signal efficiency. Events are required to have at least four jets ( $N_j \geq 4$ ), at least two b-tagged jets ( $N_b \geq 2$ ), and missing transverse energy ( $\not{E}_T \geq 20$ GeV). A Higgs mass-window requirement ( $m_{jj} \in [100, 150]$ GeV) is applied to di-jet pairs to ensure the reconstruction of Higgs bosons.
Graph Neural Network (GNN) Classification: A message-passing Graph Neural Network (specifically a GATv2 architecture) is developed as the primary discriminant. The collider events are represented as heterogeneous directed graphs where reconstructed jets serve as nodes and angular proximity defines edges. The network utilizes 14-dimensional feature vectors for jets (including kinematics, particle multiplicity, and b-tagging information) and a separate node for missing transverse energy. The classifier is trained to distinguish the $hh\nu\bar{\nu}$ signal from backgrounds ( $hb\bar{b}\nu\bar{\nu}$ and $t\bar{t}X$ ).

Theoretical sensitivity is quantified using a binned likelihood analysis based on kinematic observables ( $m_{hh}$ , $\Delta\eta_{hh}$ , $\Delta\phi_{hh}$ ) reconstructed from the four leading jets. The study evaluates two benchmark scenarios for the coupling modifiers:

Case 1: $\kappa_V = 1$ , with $\kappa_\lambda$ and $\kappa_{2V}$ varied independently (motivated by HEFT).
Case 2: $\kappa_V = \kappa_{2V}$ , with $\kappa_\lambda$ and $\kappa_{2V}$ as free parameters (motivated by SMEFT).

Key Results

Signal Significance: The GNN-based classifier achieves a micro-averaged Area Under the Curve (AUC) of 0.95, with near-perfect separation ($AUC=0.999$) for top-quark backgrounds. For the SM benchmark ( $\kappa_\lambda = \kappa_{2V} = \kappa_V = 1$ ), the analysis yields a signal significance of $Z \approx 8\sigma$ , $15\sigma$ , and $20\sigma$ at integrated luminosities of 1, 3, and 5 ab $^{-1}$ , respectively. This substantially exceeds projected HL-LHC sensitivities.
Coupling Constraints: At 5 ab $^{-1}$ $^{- 1}$ , the study projects the following 95% Confidence Level (C.L.) constraints:
- Case 1 ( $\kappa_V=1$ ): $\kappa_\lambda \in [0.76, 1.31]$ and $\kappa_{2V} \in [0.95, 1.05]$ .
- Case 2 ( $\kappa_V=\kappa_{2V}$ ): $\kappa_\lambda \in [0.76, 1.31]$ and $\kappa_{2V} \in [0.90, 1.10]$ .
  The sensitivity to $\kappa_{2V}$ is notably stronger in Case 1 due to the constructive/destructive interference patterns between amplitudes proportional to $\kappa_{2V}$ and $\kappa_V^2$ .
Systematic Uncertainties: The inclusion of systematic uncertainties (up to 10%) degrades the sensitivity gradually, particularly for $\kappa_\lambda$ , but the projected reach remains robust. The constraints on $\kappa_{2V}$ remain relatively stable even with systematics.

Significance and Claims
The authors claim that high-energy lepton colliders, specifically the 3 TeV CLIC, combined with graph-based machine learning techniques, provide excellent sensitivity to the Higgs self-coupling. The study demonstrates that WBF di-Higgs production serves not only as a direct probe of the Higgs self-interaction but also as an indirect discriminator between linearly (SMEFT) and non-linearly (HEFT) realized electroweak symmetry breaking. The distinct sensitivity contours in the $\kappa_\lambda$ - $\kappa_{2V}$ plane for the two scenarios highlight the potential of future lepton colliders to disentangle the underlying structure of new physics in the electroweak sector. The results underscore the unique capability of high-energy lepton colliders to perform precision Higgs measurements that are difficult to achieve at hadron colliders due to QCD backgrounds.

Technical Summary: High Energy Probes of Higgs Self-Coupling via W Boson Fusion at Future Lepton Colliders

More like this