Search for heavy scalar resonances decaying to Lorentz-boosted Higgs and Higgs-like bosons in the $\mathrm{b\bar{b}}$4q final state at $\sqrt{s}$ = 13 TeV

Using 138 fb$^{-1}$ of proton-proton collision data at 13 TeV collected by the CMS experiment, this study presents the first LHC search for heavy scalar resonances decaying into a Higgs boson and a Higgs-like scalar in the all-hadronic $\mathrm{b\bar{b}}$4q final state, employing advanced machine learning techniques to identify boosted jets and finding no significant excess over the Standard Model background.

The Gist

The Big Picture: Hunting for a "Heavy Parent" Particle

Imagine the universe is a giant, chaotic kitchen where particles are constantly being cooked up and smashed together. In this kitchen, the Standard Model is the "official recipe book" that explains how most ingredients behave. However, physicists suspect there are secret ingredients and hidden recipes that the book doesn't list yet.

This paper describes a search for a specific "secret ingredient": a heavy, invisible particle called X. The theory is that this heavy particle X is like a parent that splits into two children:

1. A known child: The Higgs boson (H), which we already know exists.
2. A mysterious child: A new, lighter particle called Y (which looks a lot like the Higgs but might be something new).

The goal of this experiment was to catch this heavy parent X in the act of splitting apart.

The Challenge: The "Fast-Moving Suitcase" Problem

The problem is that when these heavy particles are created in the Large Hadron Collider (LHC), they are moving incredibly fast—almost at the speed of light.

• The Analogy: Imagine a heavy suitcase (the particle X) moving so fast that when it opens, the clothes inside (the smaller particles) fly out so quickly that they all smash together into a single, messy lump before they can be separated.
• The Reality: Usually, scientists look for particles by seeing them fly apart into distinct tracks. But here, the "children" (the Higgs and the Y particle) are so fast that their decay products (the bits they turn into) get squashed together.
  • The Higgs turns into two bottom quarks, which look like one big, fuzzy blob.
  • The Y particle turns into four quarks (via W or Z bosons), which also look like one big, fuzzy blob (or sometimes two blobs, depending on how fast it is).

The scientists had to build special "detectors" (like high-tech cameras) to recognize these messy, merged blobs and figure out what was inside them.

The Tools: AI as the Detective

To find these needles in a haystack, the scientists used two main tools:

1. PARTICLENET: Think of this as a very smart AI that looks at the "fuzzy blob" and asks, "Does this look like a Higgs boson, or is it just random junk (background noise)?" It's trained to spot the specific pattern of a Higgs boson hiding inside a jet of particles.
2. The "Particle Transformer": This is a brand-new, cutting-edge AI tool (like a super-advanced pattern recognizer) specifically designed to look at the messy blob from the Y particle. It uses a technique called "attention" (similar to how humans focus on specific details in a crowd) to figure out if that blob contains four quarks, which would prove the existence of the Y particle.

The Search Strategy: The "Pass/Fail" Game

The scientists didn't just look at the data and hope for the best. They used a clever statistical game to separate the signal from the noise:

• The Signal Pass (SP): This is the "winning" zone. They looked for events where the AI was very confident that it saw a Higgs and a Y particle.
• The Signal Fail (SF): This is the "losing" zone. They looked at events where the AI said, "Nope, that's just random junk."
• The Trick: By studying the "losing" zone (where they know there are no real signals), they could mathematically predict how much "random junk" should be in the "winning" zone. If the "winning" zone had more junk than predicted, that would be a sign of a new particle.

They split the search into two categories based on how fast the particles were moving:

• Fully Merged: The Y particle is so fast that all its pieces are squashed into one single blob.
• Semi-Merged: The Y particle is fast, but its pieces are split into two distinct blobs.

The Results: No New Particles Found (Yet)

After analyzing a massive amount of data (equivalent to 138 "inverse femtobarns"—a huge number of collisions), the scientists compared what they saw against what the Standard Model predicted.

• The Outcome: The data matched the "random junk" prediction perfectly. There were no unexpected spikes or "excesses" that would indicate the heavy parent particle X or the mysterious child Y existed.
• The Conclusion: They didn't find the new particles. However, they didn't come up empty-handed. They set strict rules: "If these particles do exist, they cannot be heavier than X or lighter than Y within certain limits, and they cannot be produced more often than this specific rate."

Summary in One Sentence

The CMS team used advanced AI to look for a heavy, fast-moving particle that splits into a Higgs and a new "Higgs-like" particle, but after scanning a massive amount of collision data, they found no evidence of this new particle, only confirming that if it exists, it must be even rarer or heavier than previously thought.

Note: This paper is purely a search for new fundamental physics. It does not claim any immediate medical, technological, or practical applications. It is a fundamental investigation into the building blocks of the universe.

Technical Summary

Technical Summary: Search for Heavy Scalar Resonances Decaying to Lorentz-Boosted Higgs and Higgs-like Bosons in the $bb4q$ Final State

Problem and Motivation
The Standard Model (SM) of particle physics faces foundational inconsistencies, including the Higgs boson mass hierarchy problem and the observed baryon asymmetry of the universe. Various theoretical extensions, such as Two-Higgs-Doublet Models (2HDM), the Minimal Supersymmetric Standard Model (MSSM), the Next-to-MSSM, and Two-Real-Singlet models, predict a richer scalar sector. These models often allow for the decay of a heavy scalar resonance ($X$) into a pair of lighter scalars ($H$ and $Y$), where $H$ is the SM-like Higgs boson and $Y$ is a new Higgs-like scalar.

This paper addresses the search for the process $X \to HY$, where $H \to b\bar{b}$ and $Y \to VV \to 4q$ (with $V = W$ or $Z$). This specific all-hadronic final state ($bbVV$) presents significant challenges due to the overwhelming Quantum Chromodynamics (QCD) multijet background. Furthermore, the search targets the Lorentz-boosted regime, where the decay products of the heavy $X$ boson are highly collimated, requiring specialized reconstruction techniques to distinguish signal jets from background.

Methodology
The analysis utilizes proton-proton collision data collected by the CMS experiment at the CERN LHC at a center-of-mass energy of $\sqrt{s} = 13$ TeV, corresponding to an integrated luminosity of 138 fb$^{-1}$. The search is divided into two kinematic regimes based on the mass ratio of the resonances:

1. Fully Merged Regime ($m_Y \lesssim 0.1 m_X$): In this regime, the $Y$ boson is highly boosted, causing its decay products ($VV \to 4q$) to merge into a single large-area jet.
2. Semimerged Regime ($0.1 m_X \lesssim m_Y < m_X - m_H$): Here, the $Y$ boson is less boosted, and its decay products are reconstructed as two distinct large-area jets, each corresponding to a $V \to qq$ decay.

Key Technical Components:

• Jet Reconstruction: Events are reconstructed using the anti-$k_T$ algorithm with a distance parameter of 0.8 (AK8 jets). Jet mass is improved using a mass regression algorithm based on the PARTICLENET graph neural network architecture.
• Signal Identification:
  • $H \to b\bar{b}$: Identified using the PARTICLENET tagger ($T_{Xbb}$), which discriminates $b$-quark jets from QCD background. The tagger outputs are decorrelated from jet mass and transverse momentum ($p_T$).
  • $Y \to VV \to 4q$ (Fully Merged): A novel attention-based neural network, the "particle transformer" (PART), is employed to identify four-pronged jets originating from scalar resonances decaying to $WW$. A discriminant $T_{YWW}$ is derived from PART outputs.
  • $Y \to VV \to 4q$ (Semimerged): The two $V \to qq$ jets are identified using a discriminant $T_{Wqq}$ derived from the PARTICLENET model.
• Event Categorization: Events are classified into "Signal Pass" (SP), "Signal Fail" (SF), "Validation Pass" (VP), and "Validation Fail" (VF) regions based on tagger score thresholds and mass windows around the Higgs mass ($m_H$).
• Background Estimation: The dominant QCD multijet background is estimated using data-driven methods. The yield in the SP region is derived from the SF region (where tagger selections are inverted) multiplied by a transfer function $R(m_X, m_Y)$. These transfer functions are modeled using 2D Bernstein polynomials, fitted to data in validation regions to ensure the background model agrees with observations without biasing the signal search. Minor backgrounds ($t\bar{t}$, $V$+jets, SM Higgs) are estimated using Monte Carlo simulations.

Key Contributions

• First All-Hadronic Search: This work presents the first search at the LHC for scalar resonances in the all-hadronic $bbVV$ decay channel.
• Novel Machine Learning: The analysis introduces the application of the "particle transformer" (PART) architecture for identifying fully merged $Y \to 4q$ jets, a technique previously untested in this specific context at the LHC.
• Comprehensive Kinematic Coverage: The search covers a wide mass range, with $m_X$ between 900 and 4000 GeV and $m_Y$ between 60 and 2800 GeV, addressing both fully merged and semimerged topologies.

Results
No significant excess of events was observed above the Standard Model background expectation in either the fully merged or semimerged categories. The largest local deviations were observed at $m_X = 900$ GeV and $m_Y = 80$ GeV (fully merged, 3.3$\sigma$ local significance, $<1.0\sigma$ global) and $m_X = 1200$ GeV and $m_Y = 900$ GeV (semimerged, 1.7$\sigma$ local significance, $<1.0\sigma$ global), both consistent with background fluctuations.

Upper limits at the 95% confidence level (CL) were set on the product of the production cross section and branching fraction ($\sigma \times \mathcal{B}$) for $X \to HY \to bbVV$. The limits reach as low as 0.2 fb for various mass hypotheses.

Significance and Claims
The paper claims this is the first search for scalar resonances in the all-hadronic $bbVV$ final state at the LHC. While no new physics was discovered, the analysis successfully sets stringent exclusion limits on asymmetric cascade decays in a challenging kinematic regime. The results are interpreted within the context of a two-real-scalar-singlet extension of the SM. The authors note that while no mass exclusions can be made for the specific model parameters scanned, the observed limits are approximately a factor of 10 higher than the theoretical predictions for $m_X < 1.4$ TeV and $m_Y < 0.6$ TeV, motivating further investigation of this final state with additional datasets. The analysis demonstrates the viability of using advanced machine learning tools, such as the particle transformer, to reconstruct highly boosted, multi-pronged hadronic decays in high-pileup environments.