Hadronic decay branching ratio measurements of the Higgs boson at future colliders using the Holistic Approach

This paper demonstrates that the holistic approach, which utilizes inclusive information from all reconstructed particles, significantly enhances the precision of Higgs boson hadronic decay branching ratio measurements at the CEPC by a factor of two to four compared to previous methods, with projected statistical uncertainties ranging from 0.16% to 5.21%.

The Gist

Imagine you are a detective trying to solve a very specific crime in a massive, chaotic city. The city is a particle collider (like the CEPC), and the "crime" is the Higgs boson decaying into different types of particles.

Most of the time, the Higgs boson breaks apart into "hadrons" (a messy pile of particles like quarks and gluons). This happens about 80% of the time, making it the most common way the Higgs disappears. However, in the past, trying to identify exactly which type of hadrons it broke into was like trying to find a specific needle in a haystack while wearing blindfolded gloves. The "haystack" is the background noise of other particle collisions, and the "gloves" are the limitations of old computer methods.

This paper introduces a new, super-smart detective method called the "Holistic Approach."

Here is how it works, broken down into simple concepts:

1. The Old Way vs. The New Way

• The Old Way (Cut-and-Paste): Imagine you are sorting a pile of mixed-up toys. The old method would say, "If a toy is red, put it in the red bin. If it's blue, put it in the blue bin." It looks at one feature at a time. In physics, this meant looking at individual particle tracks or energies one by one. It missed the big picture and got confused easily.
• The Holistic Way (The Big Picture): The new method looks at the entire pile of toys at once. It doesn't just ask, "Is this red?" It asks, "How does this red toy sit next to that blue one? What is the shape of the whole pile? How heavy is the group?"
  • In the paper, this is done using Artificial Intelligence (AI). Instead of looking at single particles, the AI looks at the "cloud" of every single particle created in a collision simultaneously. It treats the whole event as a single, complex puzzle.

2. The Two Crime Scenes (Channels)

The researchers tested this method in two different "neighborhoods" of the collider:

• The "Muon" Neighborhood ($\mu^+\mu^-H$): This is a cleaner area. The background noise is low. It's like a quiet library. Here, the AI can easily spot the Higgs breaking apart into different types of particles (like bottom quarks, charm quarks, or gluons) with incredible precision.
• The "Neutrino" Neighborhood ($\nu\bar{\nu}H$): This is a noisy, crowded market. The background noise is massive (billions of events). It's like trying to hear a whisper in a rock concert.
  • The Trick: To handle the noise, the researchers used a "pre-selection" filter. They threw out the loudest, most obvious noise first (like asking everyone in the crowd to leave except those wearing a specific hat). Then, they let the AI work on the remaining, quieter crowd. This made the AI's job much easier.

3. The "Training Gym" (Scaling Behavior)

One of the coolest findings in the paper is about how the AI gets smarter.

• Imagine the AI is a student studying for a test.
• Phase 1 (The Confusion): If you give the student only 10 practice questions, they will guess randomly. They don't know the rules yet.
• Phase 2 (The "Aha!" Moment): As you give them 1,000 or 10,000 questions, their grades shoot up rapidly. They start seeing patterns.
• Phase 3 (The Plateau): Eventually, you give them a million questions. They get really good, but they hit a ceiling. They can't get any better because the test questions themselves are too similar (the physics is just too hard to distinguish perfectly).
• The Paper's Discovery: The authors found that for the most common Higgs decays, the AI is almost at that perfect ceiling. It is learning as fast as physics allows.

4. The Results: A Massive Leap Forward

The paper compares this new "Holistic" method to the old "Snowmass" estimates (the previous best guess).

• The Improvement: The new method is 2 to 4 times more precise.
• The Analogy: If the old method could tell you the weight of a package within 10 grams, the new method can tell you within 2 or 3 grams.
• Why it matters: Because the Higgs boson is so rare and its signals are so faint, getting that extra precision is the difference between just "seeing" the Higgs and truly understanding its secrets. It allows scientists to check if the Higgs behaves exactly as the Standard Model predicts or if it's hiding "New Physics."

5. The "Glitch" in the Simulation

The authors also checked if their AI was just memorizing the training data (like a student memorizing answers instead of learning the subject).

• They trained the AI on one type of computer simulation (Pythia) and tested it on a different one (Herwig).
• The Result: For simple decays (like breaking into two pieces), the AI worked perfectly on both. But for complex, messy decays (breaking into four pieces), the AI struggled a bit more when the simulation changed. This tells scientists that while the AI is amazing, they still need to be careful about how they simulate the "messy" parts of the universe.

Summary

This paper says: "Stop looking at the trees; look at the forest."

By using advanced AI to look at the entire collision event all at once, rather than piece-by-piece, scientists at future colliders (like the CEPC) can measure the Higgs boson's behavior with unprecedented accuracy. They are getting so close to the theoretical limit of what is possible that they are essentially squeezing every last drop of information out of the data. This brings us one step closer to discovering what lies beyond our current understanding of the universe.

Technical Summary

Here is a detailed technical summary of the paper "Hadronic decay branching ratio measurements of the Higgs boson at future colliders using the Holistic Approach."

1. Problem Statement

The precise measurement of Higgs boson properties, particularly its hadronic decay modes ($H \to b\bar{b}, c\bar{c}, gg, WW^*, ZZ^*$), is a primary objective for future high-energy colliders. These modes account for ~80% of Higgs decays but are notoriously difficult to measure due to overwhelming Quantum Chromodynamics (QCD) backgrounds and severe pile-up conditions at hadron colliders (like the LHC).

While future electron-positron "Higgs factories" (such as the Circular Electron–Positron Collider, CEPC) offer a cleaner environment, traditional analysis methods relying on jet clustering and sequential cuts often fail to fully exploit the available information. These methods struggle with:

• High-dimensional data complexity.
• Misidentification between similar topologies (e.g., $H \to WW^* \to 4q$ vs. $H \to ZZ^* \to 4q$).
• Limited statistical precision for rare or complex decay channels.

The paper addresses the need for a more powerful analysis framework to maximize the physics potential of the CEPC, specifically targeting a relative statistical uncertainty significantly lower than current projections.

2. Methodology: The "Holistic Approach"

The authors propose and implement the Holistic Approach, an end-to-end deep learning strategy that treats every collision event as an ensemble of reconstructed particles rather than pre-clustered jets.

• Detector Simulation: The study utilizes the AURORA detector model for the CEPC (operating at $\sqrt{s} = 240$ GeV). AURORA features high-granularity calorimetry (ECAL/HCAL) and a 5D particle-flow capability, achieving a Boson Mass Resolution (BMR) of ~2.7% and near-universal particle identification (PID).
• Data Generation: Signal samples ($Z(\mu^+\mu^-)H$ and $Z(\nu\bar{\nu})H$ with five hadronic decay modes) and dominant Standard Model backgrounds ($ZZ, WW, q\bar{q}$) were generated using MadGraph5_aMC@NLO and Pythia8 (Monash 2013 tune).
• Algorithm Architecture: The approach uses ParticleNet (PN), a deep learning framework based on EdgeConv operations.
  • Input: The network ingests inclusive, particle-level information for every reconstructed particle in an event. Features include geometric coordinates ($\Delta\eta, \Delta\phi$), kinematics ($p_T, E$), impact parameters ($d_0, d_z$), and one-hot encoded particle species ($\mu, e, \gamma, \pi, K, p, n_{had}$).
  • Processing: The EdgeConv layers aggregate features from each particle and its $k$-nearest neighbors, learning local and global correlations directly from the "particle cloud."
  • Output: A set of event-level classification scores (softmax probabilities) for all signal and background categories simultaneously.
• Training Strategy: Models were trained on datasets ranging from $10^4$to$10^6$events. A specific categorization strategy was developed for the$H \to ZZ^*$ channel to handle its complex topology and dominant backgrounds.

3. Key Contributions

1. Implementation of the Holistic Approach: Demonstrating that end-to-end particle-level classification significantly outperforms traditional cut-based or jet-based methods for Higgs hadronic decays at lepton colliders.
2. Scaling Behavior Analysis: The paper introduces and analyzes the scaling behavior of the AI model. It observes that performance follows an S-curve relative to training dataset size, characterized by:
   • A trivial phase (random classification).
   • A rapid improvement phase.
   • An asymptotic phase approaching a statistical limit.
   • This behavior is used to optimize analysis tactics and monitor model robustness.
3. Robustness Testing: The study evaluates the model's dependence on hadronization models (comparing Pythia8 vs. Herwig7). It finds that while two-body decays ($b\bar{b}, c\bar{c}, gg$) converge across models with sufficient data, four-body vector boson decays ($WW^*, ZZ^*$) exhibit divergent trends, highlighting specific systematic uncertainties.
4. Optimization of $H \to ZZ^*$: A novel categorization strategy based on 2D score distributions of background probabilities was developed, improving the measurement precision for the challenging $H \to ZZ^*$ channel by ~20%.

4. Key Results

The study projects the relative statistical uncertainties for Higgs hadronic decays at the CEPC with an integrated luminosity of 21.6 ab$^{-1}$.

Performance Comparison (Holistic vs. CEPC Snowmass Baseline):
The Holistic Approach improves measurement precision by a factor of 2 to 4 compared to previous CEPC Snowmass results.

Decay Mode	Channel	Projected Uncertainty (Holistic)	Statistical Limit	Improvement Factor
$H \to b\bar{b}$	$Z(\mu^+\mu^-)H$	0.36%	0.35%	~2x
	$Z(\nu\bar{\nu})H$	0.16%	N/A	~2x
$H \to c\bar{c}$	$Z(\mu^+\mu^-)H$	1.99%	1.55%	~2x
	$Z(\nu\bar{\nu})H$	0.86%	N/A	~2x
$H \to gg$	$Z(\mu^+\mu^-)H$	1.08%	0.90%	~2x
	$Z(\nu\bar{\nu})H$	0.43%	N/A	~2x
$H \to WW^*$	$Z(\mu^+\mu^-)H$	0.96%	0.85%	~2x
	$Z(\nu\bar{\nu})H$	0.39%	N/A	~2x
$H \to ZZ^*$	$Z(\mu^+\mu^-)H$	5.21%	2.33%	~2x
	$Z(\nu\bar{\nu})H$	2.52%	N/A	~2x

• Asymptotic Limits: For $H \to b\bar{b}$, the precision is approaching the statistical limit (0.35%). For $H \to ZZ^*$, the precision remains roughly twice the statistical limit due to complex topology and irreducible backgrounds, though the categorization strategy mitigates this.
• Scaling Insights: Increasing the training dataset size from $10^4$to$10^6$events yields significant gains, particularly for vector boson decays ($WW^, ZZ^$), which show steeper improvement slopes than heavy-flavor modes.
• Pre-selection Impact: While kinematic pre-selection suppresses backgrounds effectively, the Holistic Approach without pre-selection can eventually surpass it with sufficient data, though pre-selection remains optimal for current data volumes.

5. Significance

• Physics Potential: The results demonstrate that the CEPC can achieve sub-percent level precision for major Higgs hadronic decay modes, enabling stringent tests of the Standard Model and sensitivity to new physics via the Higgs portal.
• Methodological Shift: The paper validates the shift from "jet-centric" to "particle-centric" analysis in high-energy physics. By leveraging high-dimensional particle-level data, the Holistic Approach extracts information previously lost in jet clustering.
• AI Integration: It establishes a framework for using scaling laws (similar to those in Large Language Models) to diagnose AI performance, optimize training data requirements, and quantify systematic uncertainties arising from data-MC discrepancies.
• Systematic Control: The study highlights that as statistical uncertainties drop to the per-mille level, systematic uncertainties (luminosity, detector calibration, hadronization modeling) become the dominant limitation. The scaling behavior analysis provides a tool to monitor and control these systematics.

In conclusion, the Holistic Approach represents a paradigm shift in collider data analysis, offering a robust, scalable, and highly precise method for measuring Higgs properties at future lepton colliders.