Probing Heavy Neutral Higgs Bosons via Single Vector-Like Bottom Quark Production at the HL-LHC

This paper demonstrates that a multivariate XGBoost analysis significantly enhances the discovery potential for singly produced vector-like bottom quarks decaying via the exotic $B \to \phi b$ ($\phi \to t\bar{t}$) channel at the High-Luminosity LHC, extending the mass reach up to 1.6 TeV with 3 ab$^{-1}$ of data even in the presence of substantial systematic uncertainties.

The Gist

Imagine the universe is a giant, high-stakes game of "Where's Waldo?" played inside a microscopic, super-fast particle accelerator called the Large Hadron Collider (LHC). Physicists are the detectives, and they are looking for "Waldo"—a new, heavy particle that doesn't exist in our current rulebook (the Standard Model).

This paper is a detective's guide on how to find a specific type of "Waldo" called a Vector-Like Bottom Quark (let's call him "Vector-B") at the future, super-powered version of the collider, the HL-LHC.

Here is the story of their investigation, broken down into simple concepts:

1. The Suspect: Vector-B

In our current understanding of physics, particles have "handedness" (left or right). But the theorists suspect there might be a new kind of particle, Vector-B, that doesn't care about handedness. It's like a chameleon that can blend in perfectly with the standard particles but is actually much heavier and stranger.

2. The Secret Escape Route

Usually, when a heavy particle like Vector-B is created, it falls apart (decays) into familiar, boring particles like a Z-boson or a W-boson. The LHC has been looking for these boring decay patterns for years and hasn't found Vector-B yet.

But this paper suggests: What if Vector-B is taking a secret exit?

Instead of the boring route, the authors propose that Vector-B might decay into a heavy, invisible Higgs boson (a new kind of Higgs, let's call it "Super-Higgs") and a regular bottom quark.

• The Chain Reaction: Vector-B $\rightarrow$ Super-Higgs + Bottom Quark.
• The Follow-up: The Super-Higgs is unstable and immediately splits into a pair of top quarks (the heaviest known particles).

This creates a very messy, chaotic final scene: One electron/muon (a charged lepton), missing energy (from a neutrino), and a pile of "b-jets" (jets of particles from bottom quarks).

3. The Problem: The "Needle in a Haystack"

The problem is that the Standard Model (the current rulebook) creates millions of events that look exactly like this messy pile of debris. It's like trying to find a specific, slightly different-looking grain of sand on a beach during a storm.

The authors tried two methods to find the signal:

Method A: The "Cut-Based" Approach (The Rigid Filter)

Imagine trying to find your friend in a crowd by saying, "I only want people wearing a red hat, taller than 6 feet, and holding a blue umbrella."

• How it worked: They set strict rules (cuts) on the energy and angles of the particles.
• The Result: It worked okay, but it was like using a sieve with holes that were too big. You lost a lot of your "friend" (the signal) along with the crowd. To find the particle, they would need to wait for the collider to run for a very long time (collecting huge amounts of data).

Method B: The "XGBoost" Approach (The AI Detective)

This is where the paper gets exciting. Instead of rigid rules, they used a machine learning algorithm called XGBoost.

• The Analogy: Imagine a seasoned detective who has seen thousands of crime scenes. Instead of checking a checklist, the detective looks at the whole picture: the way the suspect stands, the angle of the umbrella, the texture of the hat, and the wind direction all at once.
• How it worked: The AI was trained on millions of simulated events. It learned the subtle, complex patterns that distinguish the "Vector-B signal" from the "Standard Model background." It didn't just look at one thing; it looked at how everything fit together.
• The Result: The AI was incredibly good at spotting the difference. It could find the signal much earlier and with much less data than the rigid filter.

4. The Big Discovery

The paper concludes that if the universe is hiding this "Vector-B" particle, the old way of looking (rigid cuts) might miss it entirely. But the new way (AI/Machine Learning) gives us a very high chance of finding it.

• The Reach: With the AI, they could potentially discover a Vector-B particle weighing up to 1.6 TeV (which is about 1,700 times heavier than a proton) using the data the HL-LHC will collect in the coming years.
• The Robustness: Even if the detectors aren't perfect and have some "noise" or errors (systematic uncertainties), the AI is smart enough to ignore the noise and still find the signal.

Summary

Think of this paper as a manual for upgrading the LHC's search strategy.

• Old Strategy: "Look for X, Y, and Z." (Too simple, misses the target).
• New Strategy: "Use a super-smart AI to recognize the unique fingerprint of a new particle, even when it's hiding in a crowd of look-alikes."

The authors are essentially saying: "Don't just look harder; look smarter. If we use machine learning, we can find these heavy, exotic particles that have been hiding from us all this time."

Technical Summary

1. Problem Statement

The Standard Model (SM) faces theoretical challenges such as the hierarchy problem, motivating extensions like the Two-Higgs-Doublet Model (2HDM) and the inclusion of Vector-Like Quarks (VLQs).

• The Gap: Experimental searches for VLQs (specifically vector-like bottom quarks, $B$) at the LHC have traditionally focused on "conventional" decay modes into SM particles ($B \to Wt, Zb, hb$).
• The Issue: In extensions combining 2HDM and VLQs, "exotic" decay channels mediated by heavy neutral Higgs bosons ($H$ or $A$) can dominate. Specifically, the decay chain $B \to \phi b$ (where $\phi = H, A$) followed by $\phi \to t\bar{t}$ can have branching ratios (BR) as high as $\sim 50\%$.
• Consequence: If these exotic modes dominate, standard search strategies assuming SM decays lose sensitivity, potentially allowing heavy VLQs to evade detection. There is a need to quantify the discovery potential of these specific non-standard channels at the High-Luminosity LHC (HL-LHC).

2. Theoretical Framework

The study is based on a Type-II 2HDM extended by an $SU(2)_L$ vector-like $(T, B)$ doublet.

• Model Parameters: The analysis considers the alignment limit ($\sin(\beta-\alpha) \to 1$), where the light CP-even Higgs $h$ mimics the SM Higgs. The heavy neutral scalars are $H$ (CP-even) and $A$ (CP-odd).
• Mixing: The vector-like bottom quark $B$ mixes with the SM bottom quark. The mixing is parametrized by angles $\theta_{L,R}^d$. The study assumes a hierarchy where the right-handed mixing with the down-type sector ($s_R^d$) is significantly larger than the up-type mixing ($s_R^u$).
• Decay Dynamics:
  • The coupling for the exotic decay $B \to \phi b$ scales as $\propto \tan\beta \cdot s_R^d c_R^d$.
  • The coupling for conventional decays ($B \to Zb, hb$) scales as $\propto s_R^d c_R^d$ (no $\tan\beta$ enhancement).
  • Result: For large $\tan\beta$ and appropriate mixing, the exotic decays dominate, with $BR(B \to Hb) \approx BR(B \to Ab) \approx 50\%$.

3. Methodology

The authors performed a full collider simulation at $\sqrt{s} = 14$ TeV to evaluate discovery prospects.

A. Signal and Backgrounds

• Signal Process: Single production of a vector-like bottom quark via $Z$-boson fusion ($pp \to B\bar{b}j$), followed by the decay chain:
  $$B \to \phi b \to (t\bar{t}) b \to (W^+ b \bar{t}) b \to (\ell^+ \nu b \bar{t}) b$$
  Final State: One charged lepton ($\ell$), missing transverse energy ($E_T^{miss}$), and multiple $b$-jets (specifically 3 $b$-jets from the top and $B$ decays, plus the recoil jet).
• Backgrounds: Dominant SM backgrounds include $t\bar{t}b\bar{b}$, $t\bar{t}V$ ($V=W, Z, h$), $t\bar{t}jj$, and $b\bar{b}WW$.
• Simulation Tools:
  • Event Generation: MadGraph5_aMC@NLO (v2.9.14).
  • Showering/Hadronization: Pythia 8.30.
  • Detector Simulation: Delphes 3.5.0 (CMS card).
  • Analysis Framework: MadAnalysis 5.

B. Analysis Strategies

Two distinct analysis approaches were compared:

1. Cut-Based Analysis: A traditional approach using sequential kinematic cuts on observables like $E_T^{miss}$, $E_T$, and $p_T$ of leptons and jets.
2. Machine Learning (ML) Analysis: A multivariate approach using XGBoost (Extreme Gradient Boosting).
   • Input Features: 13 kinematic variables including $p_T$ of jets/leptons/b-jets, invariant masses, angular separations ($\Delta R$), and event counts ($N_b, N_j$).
   • Training: The classifier was trained to distinguish signal from background, with hyperparameters optimized (e.g., n_estimators=800, max_depth=4).
   • Validation: Kolmogorov-Smirnov (KS) tests confirmed no significant overfitting.

4. Key Results

A. Branching Ratios and Cross Sections

• In the benchmark scenarios (BP1, BP2, BP3), the exotic decays $B \to Hb$ and $B \to Ab$ dominate with branching ratios of ~30% to 48%.
• The production cross section for $pp \to B\bar{b}j$ drops from $\sim 0.04$ pb at $m_B = 1$ TeV to $\sim 1.5 \times 10^{-3}$ pb at $m_B = 2$ TeV.

B. Performance Comparison: Cut-Based vs. XGBoost

• Cut-Based: Achieved a $5\sigma$ discovery significance only at very high integrated luminosities ($L \approx 3 \text{ ab}^{-1}$) and for lower masses ($m_B \lesssim 1.3$ TeV). Sensitivity dropped rapidly with increasing mass and systematic uncertainties.
• XGBoost: Demonstrated superior discrimination power.
  • AUC (Area Under Curve): Achieved values between 0.963 and 0.980, indicating excellent separation between signal and background.
  • Feature Importance: The transverse momentum of the leading jet ($p_T(j)$) and the dijet invariant mass were the most discriminating variables.

C. Discovery Reach at HL-LHC

The study evaluated the $5\sigma$ discovery reach for different integrated luminosities ($L$) and systematic uncertainties ($\delta$):

Integrated Luminosity	Discovery Reach ($m_B$)	Conditions
600 fb$^{-1}$	$\sim 1.3$ TeV	Valid for $\delta = 5\%, 10\%, 15\%$
1.5 ab$^{-1}$	$\sim 1.5$ TeV	Valid for $\delta = 5\%$
3.0 ab$^{-1}$	$\sim 1.6$ TeV	Valid for $\delta = 5\%$; $\sim 1.5$ TeV for $\delta = 10-15\%$

• Significance: For $L = 3 \text{ ab}^{-1}$ and $m_B = 1.2$ TeV, the XGBoost analysis yields a significance of $\sim 18.5\sigma$, compared to a much lower value for the cut-based approach.
• Robustness: The ML approach maintains $5\sigma$ sensitivity even with systematic uncertainties as large as 15%.

5. Significance and Conclusion

• Phenomenological Impact: The paper demonstrates that assuming only SM decay modes for VLQs is insufficient. In 2HDM+VLQ scenarios, exotic decays can dominate, making them the primary discovery channel.
• Methodological Advancement: The study highlights the critical role of multivariate analysis (XGBoost) in high-energy physics. It significantly outperforms traditional cut-based methods, extending the mass reach by hundreds of GeV and providing robustness against systematic uncertainties.
• Future Outlook: The proposed channel ($B \to \phi b \to t\bar{t}b$) is a promising target for the HL-LHC. It offers a unique probe of both the heavy vector-like quark sector and the extended Higgs sector, potentially uncovering new physics that would remain hidden in conventional searches.

In summary, this work provides a comprehensive roadmap for detecting heavy vector-like bottom quarks decaying into heavy Higgs bosons, proving that advanced machine learning techniques are essential to maximize the discovery potential of the HL-LHC in this specific BSM scenario.