Search for new particles decaying into top quark-antiquark pairs in proton-proton collisions at $\sqrt{s}$ = 13 TeV

Using 138 fb$^{-1}$ of proton-proton collision data at 13 TeV collected by the CMS detector, this study performs a comprehensive search for new particles decaying into top quark-antiquark pairs across all lepton final states, setting the most stringent limits to date on heavy Z' bosons, Kaluza-Klein gluons, dark-matter mediators, and two-Higgs-doublet model scalars.

The Gist

Imagine the Large Hadron Collider (LHC) at CERN as the world's most powerful microscope, but instead of looking at cells, it smashes tiny particles (protons) together at nearly the speed of light. When these particles collide, they create a chaotic explosion of energy that briefly turns into new, heavier particles before they instantly decay into familiar ones.

This paper is a report from the CMS Collaboration, one of the two giant teams of scientists watching these explosions. Their mission? To find a "ghost" in the machine: a new, heavy particle that decays into a pair of top quarks.

Here is the story of their search, explained simply.

1. The Target: The "Heavyweight Champion"

The top quark is the heaviest known elementary particle. It's like the heavyweight champion of the particle world. Because it's so heavy, it interacts strongly with the Higgs field (the field that gives particles mass).

• The Analogy: Think of the top quark as a very heavy bowling ball. If you see a bowling ball suddenly appear out of nowhere in a game of pinball, you know something strange is going on. Scientists suspect that new, undiscovered particles might be "hiding" by turning into these heavy top quarks.

2. The Search: Looking for a "Resonance"

The scientists are looking for a specific signature: a resonance.

• The Analogy: Imagine you are in a large, noisy room (the collision data) filled with people talking (standard background noise). You are listening for a specific song playing on a radio. If a new, heavy particle exists, it would appear as a sudden, loud spike in the volume of that specific song at a specific pitch (mass).
• The Goal: They want to find a "bump" in the data that doesn't fit the standard model of physics. If they find a bump, it means a new particle (like a heavy Z' boson or a Kaluza-Klein gluon) was created and then split into two top quarks.

3. The Challenge: The "Three Ways" to Catch a Top

Top quarks are tricky because they decay (break apart) almost instantly. They can break apart in three different ways, creating three different "channels" for the scientists to search:

1. The "All-Hadronic" Channel (0 Leptons): Both top quarks turn into jets of particles (like a shower of confetti). This is the hardest to spot because it looks like a messy pile of debris.
   • The Trick: They used a Deep Neural Network (AI) called "DEEPAK8." Think of this AI as a super-smart security guard who can look at a pile of confetti and instantly say, "Hey, this specific pile of confetti came from a top quark!"
2. The "Single-Lepton" Channel (1 Lepton): One top quark turns into a jet, and the other turns into a jet plus a single electron or muon (a "lepton").
   • The Trick: They used the same AI guard for the jet, plus a new AI to sort the "good" events from the "bad" background noise.
3. The "Dilepton" Channel (2 Leptons): Both top quarks turn into jets plus an electron or muon. This is cleaner but rarer.
   • The Trick: They looked at how far apart the particles were to figure out if they came from a heavy, fast-moving source.

4. The Method: The "ABCD" Game

How do you know if a "bump" is real and not just a fluke?

• The Analogy: Imagine you are trying to count how many people in a crowd are wearing red hats (the signal). But the crowd is huge, and you can't see everyone clearly.
  • You divide the crowd into four zones: A, B, C, and D.
  • You know that in zones A, B, and C, there are no red hats. You count the people there to learn what the "normal" crowd looks like.
  • You then use that information to predict how many people should be in zone D (the Signal Zone) if there were no red hats.
  • If you see more people in Zone D than your prediction, you might have found red hats!
• In the Paper: The scientists used a sophisticated version of this game (a 2D fit) to predict the background noise in their "Signal Zone" using data from the "Control Zones."

5. The Results: "No Ghosts Found (Yet)"

After analyzing 138 fb⁻¹ of data (which is like looking at 138 million billion collisions), the scientists found no evidence of these new heavy particles.

• The Outcome: The data matched the "Standard Model" (the current rulebook of physics) perfectly. There were no mysterious bumps.
• The Silver Lining: Even though they didn't find the new particle, they set strict limits.
  • The Analogy: It's like a detective searching a house for a thief. They didn't find the thief, but they can now say with 95% confidence: "The thief is not hiding in the living room, the kitchen, or the bedroom, and if they are in the attic, they must weigh less than 50 pounds."
• Specific Limits:
  • They ruled out heavy Z' bosons up to 7.4 TeV (depending on how wide they are).
  • They ruled out Kaluza-Klein gluons up to 5.5 TeV.
  • They ruled out Dark Matter mediators up to 4.2 TeV.

6. Why This Matters

Even though they didn't find the "new particle," this is a huge success.

• Ruling Out Worlds: By saying "these particles don't exist at these masses," they force theorists to rewrite their theories. If a theory predicts a particle at 3 TeV, and this experiment says "nope," that theory is now wrong or needs to be changed.
• The Future: They have pushed the search to higher energies and more complex scenarios than ever before. They are essentially saying, "If the new physics is hiding, it's hiding in a very small, very heavy corner that we haven't looked at yet."

Summary

The CMS team acted like cosmic detectives using a giant particle microscope and AI-powered security guards. They searched through millions of collisions for a specific "ghost" particle that turns into top quarks. They didn't find the ghost, but they successfully mapped out the "haunted house" and proved that the ghost isn't hiding in the rooms they checked. This narrows the search for the next big discovery in physics.

Technical Summary

1. Problem and Motivation

The top quark, being the heaviest known fermion, possesses a large coupling to the Higgs field, suggesting a potential special role in theories addressing the hierarchy between the electroweak and Planck scales. While the Standard Model (SM) predicts top quark pair ($t\bar{t}$) production, various Beyond the Standard Model (BSM) theories predict new heavy particles that decay into $t\bar{t}$ pairs. These could manifest as:

• Resonant production: Narrow or broad resonances (e.g., $Z'$ bosons, Kaluza-Klein gluons, heavy Higgs bosons).
• Non-resonant production: Modifications to the $t\bar{t}$ invariant mass ($m_{t\bar{t}}$) spectrum or angular distributions due to interference effects (e.g., spin-0 scalars).

Previous searches have focused primarily on resonant signatures. This analysis aims to provide a comprehensive search sensitive to both resonant and non-resonant BSM signatures across the full $t\bar{t}$ decay phase space, utilizing the full Run 2 dataset of the CMS experiment.

2. Methodology

Dataset and Channels

• Data: Proton-proton collisions at $\sqrt{s} = 13$ TeV collected between 2016 and 2018.
• Integrated Luminosity: $138 \text{ fb}^{-1}$.
• Channels: The analysis covers all $t\bar{t}$ decay modes, categorized by the number of charged leptons (electrons or muons):
  • 0$\ell$ (All-hadronic): Both top quarks decay hadronically.
  • 1$\ell$ (Single-lepton): One top decays hadronically, the other semileptonically.
  • 2$\ell$ (Dilepton): Both tops decay semileptonically.
  • Note: $\tau$ leptons are not explicitly vetoed; they contribute to the background or signal depending on the model.

Event Selection and Reconstruction

• 0$\ell$ Channel:
  • Utilizes large-radius jets ($R=0.8$, AK8) with $p_T > 400$ GeV.
  • Employs the DEEPAK8 deep neural network (DNN) for "top tagging" (identifying hadronically decaying top quarks within a single jet).
  • Uses a 2D sideband fit (control regions) on the leading jet mass ($m_t$) and $m_{t\bar{t}}$ to estimate the dominant QCD multijet background directly from data, avoiding reliance on simulation for this complex background.
  • Categorizes events into "central" and "forward" based on jet rapidity to exploit angular differences between SM and BSM production.
• 1$\ell$ Channel:
  • Optimized thresholds for lepton $p_T$ and missing transverse momentum ($p_T^{miss}$).
  • Uses DEEPAK8 for top tagging in the merged topology (highly boosted tops) and standard AK4 jets for resolved topologies.
  • Introduces a second DNN to discriminate between $t\bar{t}$, single-top, and $V$+jets backgrounds.
  • Spin-sensitive variable: Events are categorized based on $\cos(\theta^*)$ (angle between the top momentum in the $t\bar{t}$ rest frame and the $t\bar{t}$ system in the lab frame) to distinguish spin-0 signals (isotropic) from spin-1 signals and SM background.
• 2$\ell$ Channel:
  • Uses the scalar sum of transverse momenta ($S_T$) instead of $m_{t\bar{t}}$ due to the presence of two neutrinos making mass reconstruction difficult.
  • Categorizes events based on the angular distance sum ($\Delta R_{sum}$) between leptons and nearest jets to separate resolved and merged topologies.

Background Estimation

• SM $t\bar{t}$: Estimated using Monte Carlo (MC) simulations (POWHEG v2 + PYTHIA 8), normalized to NNLO QCD cross-sections.
• QCD Multijet (0$\ell$): Estimated via a data-driven ABCD method using a 2D transfer function derived from control regions in the $m_t$ and $m_{t\bar{t}}$ plane.
• Other Backgrounds (Single top, $W/Z$+jets, Diboson): Estimated using MC simulations, with normalizations constrained by control regions in the fit.

Statistical Analysis

• A simultaneous binned maximum likelihood fit is performed across all signal regions (SR) and control regions (CR) for all three channels.
• Nuisance parameters account for systematic uncertainties (jet energy scale, luminosity, PDFs, etc.).
• Limits are set using the CL$_s$ criterion with the profile likelihood ratio test statistic.

3. Key Contributions and Improvements

• Model-Agnostic Sensitivity: Unlike previous searches, this analysis is optimized for both resonant and non-resonant (interference) effects.
• Advanced Machine Learning:
  • Integration of DEEPAK8 for top tagging in both 0$\ell$ and 1$\ell$ channels, subsuming previous $b$-tagging categories into the algorithm.
  • Use of a dedicated DNN for background classification in the 1$\ell$ channel.
• Data-Driven Backgrounds: The 0$\ell$ channel utilizes a sophisticated 2D fit to estimate QCD backgrounds directly from data, significantly reducing reliance on simulation for high-mass regimes.
• Spin Correlation: The 1$\ell$ channel explicitly uses $\cos(\theta^*)$ to enhance sensitivity to spin-0 vs. spin-1 hypotheses.
• Extended Mass Reach: The analysis covers a mass range up to 9 TeV for $Z'$ bosons and 6 TeV for Kaluza-Klein gluons, with specific focus on the 0.5–1.0 TeV range for 2HDM scalars.

4. Results

Exclusion Limits (95% CL)

No significant excess over the SM prediction was observed. The results set the most stringent limits to date for the considered models:

• Spin-1 Resonances ($Z'$ bosons):
  • Excluded mass ranges: 0.4–4.8 TeV (1% width), 0.4–6.2 TeV (10% width), and 0.4–7.4 TeV (30% width).
• Kaluza-Klein Gluons ($g_{KK}$):
  • Excluded mass range: 0.5–5.5 TeV (Randall-Sundrum model).
• Dark Matter Mediators ($Z'_{DM}$):
  • Excluded mass range: 1.0–4.2 TeV.
• Spin-0 Resonances (2HDM - Scalar $H$ and Pseudoscalar $A$):
  • Upper limits set on the coupling strength modifier ($g_{\Phi tt}$) for masses 0.5–1.0 TeV with relative widths of 2.5%, 10%, and 25%.
  • Sensitivity is comparable to or better than previous CMS results for masses $> 800$ GeV.

Systematic Uncertainties

• The total uncertainty is dominated by statistical uncertainty (60%).
• Among systematic uncertainties, the largest contribution (22% of total variance) comes from top quark modeling (production rate, top tagging efficiency, and $p_T$ spectrum modeling).

5. Significance

This paper represents a major advancement in the search for new physics in the $t\bar{t}$ sector at the LHC:

1. Comprehensive Coverage: By combining all decay channels and utilizing advanced categorization (resolved/merged, spin-sensitive variables), it maximizes sensitivity across a wide mass spectrum.
2. Robust Background Handling: The data-driven estimation of QCD backgrounds in the all-hadronic channel reduces systematic biases associated with high-mass jet modeling.
3. Stringent Limits: The exclusion limits significantly constrain parameter spaces for popular BSM models, including Topcolor, Randall-Sundrum, and Two-Higgs-Doublet Models.
4. Future Outlook: The techniques developed (DEEPAK8, 2D fits, spin-sensitive categorization) set a benchmark for future Run 3 and High-Luminosity LHC analyses.

The results confirm the Standard Model description of $t\bar{t}$ production up to multi-TeV scales, placing tight constraints on the existence of heavy resonances or mediators decaying to top quarks.