⚛️ phenomenology

Measuring the trilinear Higgs self-coupling in Higgs boson pair production at multi-TeV muon colliders

This paper demonstrates that multi-TeV muon colliders, specifically at 3 TeV and 10 TeV energies, can achieve high-precision measurements of the trilinear Higgs self-coupling modifier $\kappa_3$ in Higgs boson pair production via vector boson fusion, utilizing advanced machine learning techniques to reach a few-percent precision at 10 TeV that significantly surpasses projections for the High-Luminosity LHC.

Original authors: Kingman Cheung, Jae Sik Lee, Soojin Lee, Chen Wang

Published 2026-08-04

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Kingman Cheung, Jae Sik Lee, Soojin Lee, Chen Wang

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 is like a giant, invisible trampoline. When you place a heavy bowling ball in the center, the fabric curves down, creating a dip. In the world of particle physics, this "dip" is called the Higgs field, and it's the reason why tiny particles like electrons and quarks have mass. Without this field, everything would zip around at the speed of light, and atoms (and you, and me) couldn't exist. Scientists discovered the particle that ripples this field—the Higgs boson—back in 2012, confirming the existence of the trampoline. But there's a mystery left: what does the trampoline actually look like? Is it a smooth, perfect bowl, or does it have a weird, bumpy shape? The answer lies in a specific property called the "trilinear self-coupling." Think of this as measuring how hard the trampoline pushes back when you try to squish it. If the shape is different than we expect, it could mean there are hidden rules of physics we haven't discovered yet, potentially explaining why the universe is the way it is.

Now, picture a team of physicists acting like detectives trying to solve this shape mystery. They know that to see the trampoline's true form, they need to smash two Higgs bosons together and watch how they bounce off each other. This is incredibly hard to do because the Higgs boson is shy and rarely shows up, and when it does, it's usually hiding in a crowd of billions of other particles. The paper you're about to read describes a simulation of a future experiment using a "muon collider." Imagine a muon as a heavy, point-like electron that doesn't waste energy spinning around; it's a clean, powerful bullet. The researchers simulated smashing these muons together at energies of 3 TeV and 10 TeV (which is like firing a cannonball with the energy of a speeding train, but at the subatomic scale). Their goal was to see if they could spot the rare event where two Higgs bosons are created and then immediately decay into four "bottom quarks" (which look like four jets of debris).

The challenge is that for every one of these rare Higgs events, there are millions of background events that look almost identical, like trying to find a specific needle in a haystack that is made of other needles. To solve this, the authors used a clever two-part strategy. First, they split the search into two zones: a "resolved" zone where the four debris jets are spread out and easy to count, and a "boosted" zone where the Higgs bosons are moving so fast that their debris squishes together into two giant, fat jets. Second, they trained artificial intelligence (AI) to be the ultimate detective. They built two different AI networks: one to tell the difference between the "needle" (signal) and the "haystack" (background), and another to measure the specific "shape" of the needle to see if it matches the standard theory or something new. They also used a mathematical tool called Topological Data Analysis, which is like looking at the global "shape" or "holes" in the energy flow of the event, rather than just counting particles.

The results of these simulations are promising. At the lower energy of 3 TeV, the team found they could distinguish the Higgs signal from the background with high confidence, narrowing down the possible value of the trilinear coupling to a range between 0.80 and 1.29. This means they could tell if the trampoline's shape is within about 30% of the standard prediction. However, the real magic happens at 10 TeV. With more energy and more data, the simulation shows they could pin down this value to a range between 0.96 and 1.05. This is a precision of just a few percent, which is a massive improvement over what the current Large Hadron Collider (LHC) is expected to achieve. The paper concludes that a 10 TeV muon collider would be a game-changer, capable of measuring the Higgs self-coupling with enough precision to finally reveal the true shape of the Higgs potential and the dynamics of how particles get their mass, surpassing the capabilities of any existing or planned hadron collider.

Technical Summary: Measuring the trilinear Higgs self-coupling in Higgs boson pair production at multi-TeV muon colliders

Problem and Motivation
The trilinear Higgs self-coupling ( $\lambda_3$ ) determines the shape of the Higgs potential and is a critical probe of the electroweak symmetry breaking (EWSB) sector. While the Standard Model (SM) predicts a specific value, many Beyond the Standard Model (BSM) scenarios, such as supersymmetry or composite Higgs models, predict significant deviations. Current constraints from the Large Hadron Collider (LHC) are loose ( $-0.71 < \kappa_3 < 6.1$ at 95% CL), and even the projected sensitivity of the High-Luminosity LHC (HL-LHC) remains at the $\mathcal{O}(50\%)$ level. This paper assesses the potential of multi-TeV muon colliders to measure the coupling modifier $\kappa_3 \equiv \lambda_3 / \lambda_3^{\text{SM}}$ with significantly higher precision. The study focuses on Higgs boson pair ($HH$) production via vector boson fusion (VBF) with the hadronic decay channel $HH \to b\bar{b}b\bar{b}$ , analyzing center-of-mass energies of $\sqrt{s} = 3$ TeV (with $1 \text{ ab}^{-1}$ ) and $10$ TeV (with $10 \text{ ab}^{-1}$ ).

Methodology
The analysis addresses two distinct statistical cases: (1) the observability of $HH$ production above the SM background, and (2) the discrimination of $\kappa_3 \neq 1$ from the SM point ( $\kappa_3 = 1$ ). To handle backgrounds orders of magnitude larger than the signal, the authors employ a multi-faceted strategy:

Kinematic Regions: Events are divided into a resolved region ( $p_T(H) < 200$ GeV), where the four $b$ -quarks appear as distinct jets, and a boosted region ( $p_T(H) > 200$ GeV), where the decay products of each Higgs merge into single large-radius jets.
Event Reconstruction:
- In the resolved region, a dedicated SPANet (Symmetry Preserving Attention Network) is used to optimally assign the four reconstructed jets to the two Higgs candidates, improving upon traditional kinematic pairing methods.
- In the boosted region, large-radius jets ( $R=1.0$ ) are used directly.
Machine Learning Architecture: Two complementary neural networks are trained for each region:
- $D_{HH}$ : A classifier to separate the $HH$ signal from the total SM background.
- $D_{\kappa_3}$ : A classifier trained to discriminate between specific $\kappa_3$ hypotheses (specifically $\kappa_3=0.4$ vs. $\kappa_3=1.6$ ) to extract shape information.
Feature Engineering: The models utilize standard kinematic variables, low-level particle cloud information, and Topological Data Analysis (TDA) descriptors. The TDA variables ( $H_0, H_1, S_0, S_1, LB_1$ ) summarize the global $(\eta, \phi)$ energy flow topology via persistent homology, providing frame-invariant geometric information complementary to standard observables.
Statistical Analysis: A two-dimensional binned maximum likelihood fit is performed on the outputs of the $D_{HH}$ and $D_{\kappa_3}$ classifiers. The analysis uses Asimov datasets to project expected sensitivities, assuming statistical uncertainties only and neglecting systematic uncertainties.

Key Contributions

Novel ML Strategy: The paper implements a two-network approach where one network isolates the signal from background and the other isolates the $\kappa_3$ shape information, combined via a 2D likelihood fit.
Integration of TDA: It demonstrates the utility of topological data analysis descriptors in collider phenomenology, showing that variables like $H_0$ (measuring the angular scale of energy cluster merging) are highly effective at separating the multi-pronged $HH \to 4b$ signal from backgrounds.
Jet Assignment: The application of SPANet for jet-to-Higgs pairing in the resolved region significantly improves the reconstruction of Higgs candidate masses compared to traditional kinematic minimization.
Comprehensive Kinematic Coverage: The study rigorously combines resolved and boosted regions, demonstrating their kinematic complementarity and the increasing importance of the boosted region at higher energies.

Results
The study projects the following sensitivities for the coupling modifier $\kappa_3$ :

Observability: The SM $HH$ signal is observable above the background-only hypothesis with an expected significance of 5.4 $\sigma$ at 3 TeV and 36 $\sigma$ at 10 TeV (combined channels).
$\kappa_3$ Sensitivity (68% CL):
- 3 TeV: $0.80 < \kappa_3 < 1.29$ (Combined).
- 10 TeV: $0.96 < \kappa_3 < 1.05$ (Combined).
$\kappa_3$ Sensitivity (95% CL):
- 3 TeV: $0.63 < \kappa_3 < \text{unbounded}$ (within the scanned range).
- 10 TeV: $0.92 < \kappa_3 < 1.10$ .

The 10 TeV result achieves a precision of a few percent, representing a substantial improvement over the 3 TeV projection and the projected HL-LHC sensitivity. The analysis notes that the 10 TeV precision is limited primarily by statistics in this projection.

Significance
The paper claims that a multi-TeV muon collider can constrain the Higgs trilinear self-coupling at the few-percent level, a precision that would directly probe the shape of the Higgs potential and the dynamics of electroweak symmetry breaking. The authors state that this precision "surpasses by a large amount the precision projected for the HL-LHC." Furthermore, the 10 TeV muon collider approach approaches the ultimate precision expected from a 100 TeV hadron collider but does so via a direct, tree-level measurement in vector boson fusion, offering a complementary probe to loop-induced measurements at $e^+e^-$ Higgs factories and gluon-fusion measurements at hadron colliders. The results suggest that such a machine is a critical tool for definitively testing the structure of the EWSB sector.

More like this