Observation of tWZ production at the CMS experiment

✨

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

The Cosmic "Rare Bird" Hunt: How CMS Found the tWZ

Imagine the Large Hadron Collider (LHC) at CERN as the world's most powerful, high-speed particle smasher. It fires two beams of protons (tiny building blocks of matter) at each other at nearly the speed of light. When they collide, it's like smashing two Swiss watches together at 300 mph and hoping to find a specific, rare gear that fell out.

For decades, physicists have been looking for a very specific, incredibly rare "gear" combination: a top quark (the heaviest known particle) appearing alongside a W boson and a Z boson (particles that carry the weak nuclear force). This event is called tWZ production.

Here is the story of how the CMS experiment finally spotted this rare bird, explained simply.

1. The Challenge: Finding a Needle in a Haystack

The problem is that tWZ is exceedingly rare.

The Analogy: Imagine you are trying to find a specific, unique snowflake in a blizzard. But the blizzard is actually a massive storm of "common" snowflakes (background noise) that look almost exactly like the one you want.
The Reality: Every time protons collide, they produce millions of "boring" events. The tWZ event happens so rarely that for every one tWZ event, there are thousands of other events (like top quarks paired with Z bosons, or W and Z bosons with extra jets) that look very similar.

In 2024, the CMS team found evidence of this event, but it wasn't quite enough to say, "We definitely saw it." They needed more data and a sharper eye.

2. The New Tools: Machine Learning as a Super-Filter

For this new paper (published in 2026), the team didn't just look at the data; they upgraded their "eyes."

The Old Way: They used standard rules to filter out the noise. It was like using a metal detector that beeps for any metal.
The New Way: They used Advanced Machine Learning (ML) algorithms.
- The Analogy: Think of the old method as a security guard checking IDs. The new method is like a super-intelligent AI detective that has studied millions of crime scenes. It doesn't just check the ID; it looks at the person's gait, their nervousness, the way they hold their bag, and the subtle patterns in the background.
- This AI (called a "Particle Transformer") could distinguish the tiny differences between a "fake" tWZ event and a "real" one with unprecedented precision.

3. The Hunt: 200 "Years" of Data

The team analyzed data from two different energy levels (13 TeV and 13.6 TeV).

The Analogy: If one collision is like flipping a coin, they flipped the coin 200 billion times (200 inverse femtobarns of data).
They looked for events where three or four charged particles (electrons or muons) flew out of the collision. These are the "footprints" left behind when the heavy top quark and the W/Z bosons decay.

4. The Result: The "5.8 Sigma" Moment

After feeding all this data into their super-AI, the results came in.

The Discovery: They found the tWZ signal with a statistical significance of 5.8 standard deviations (sigma).
The Analogy: In science, "5 sigma" is the gold standard for a "discovery." It's the difference between guessing you heard a ghost and actually seeing it clearly.
- If you flip a coin 100 times and get 100 heads, that's suspicious (maybe 3 sigma).
- Getting 5.8 sigma is like flipping a coin and getting heads 20 times in a row by pure chance. The odds of this being a fluke are about 1 in 10 million.
The Verdict: They officially observed tWZ production for the first time.

5. Why Does This Matter?

You might ask, "So what? We found a rare particle collision."

The Standard Model: This is the "Rulebook" of the universe. It predicts how particles should behave.
The Test: The tWZ process is a direct test of the "Electroweak" rules of this book. It's like checking if the laws of gravity work exactly the same way on a mountain top as they do in a valley.
The Future: If the rate of tWZ production was slightly different from what the Rulebook predicted, it would mean New Physics exists—perhaps dark matter or new forces we don't know about yet.
- The Result: The team measured the rate, and it matched the Standard Model's prediction very closely (within a small margin of error). This confirms our current understanding of the universe is solid, but it also sets a new, ultra-precise baseline. If we see a deviation in the future, we will know exactly where to look.

Summary

The CMS collaboration used a massive dataset and a "super-smart" AI to filter through a mountain of cosmic noise. They successfully isolated a signal that happens only once in a trillion collisions, proving that the universe behaves exactly as our best theories predict. It's a victory for the Standard Model, but more importantly, it proves we have the tools to hunt down the rarest phenomena in the cosmos.

1. Problem and Motivation

The Standard Model (SM) of particle physics is highly successful but incomplete, failing to explain phenomena such as dark matter, neutrino masses, and the matter-antimatter asymmetry. Consequently, searching for deviations from SM predictions is a primary goal of high-energy physics.

The specific process investigated is the associated production of a single top quark ( $t$ ) with a $W$ and a $Z$ boson ( $pp \to tWZ$ ).

Rarity: This is an exceptionally rare process with a predicted next-to-leading order (NLO) cross-section of approximately 136 fb at $\sqrt{s} = 13$ TeV and 148 fb at 13.6 TeV.
Physics Potential: The $tWZ$ process provides direct access to the electroweak (EW) couplings of the top quark. It serves as a sensitive probe for:
- Testing the gauge structure of the SM.
- Constraining anomalous couplings induced by new heavy particles (Beyond the Standard Model or BSM physics).
Experimental Challenge: The signal is difficult to isolate due to its small cross-section compared to dominant backgrounds with similar signatures, specifically $t\bar{t}Z$ (top-antitop plus Z) and diboson production ($WZ$, $ZZ$) with additional jets. Previous CMS analyses (2024) achieved only "evidence" (3.4 standard deviations) using 13 TeV data.

2. Methodology

Data and Dataset

Detector: CMS experiment at the CERN Large Hadron Collider (LHC).
Data: Proton-proton collisions at center-of-mass energies of 13 TeV (2016–2018) and 13.6 TeV (2022–2023).
Integrated Luminosity: A total of 200 fb $^{-1}$ (62 fb $^{-1}$ at 13.6 TeV).

Event Selection and Reconstruction

Trigger: Events selected via triggers requiring one, two, or three reconstructed leptons (electrons or muons).
Object Definition:
- Leptons: $p_T > 10$ GeV (13 TeV) or $15$ GeV (13.6 TeV); $|\eta| < 2.5$ (e) or $2.4 $($ \mu$). Advanced multivariate discriminants reduce non-prompt lepton backgrounds.
- Jets: Reconstructed using the anti- $k_T$ algorithm ( $R=0.4$ ), $p_T > 25$ GeV, $|\eta| < 2.5$ .
- b-tagging: DeepJet algorithm (13 TeV) and an improved transformer-based algorithm (13.6 TeV) with ~80% efficiency for true $b$ -jets.
- Missing Transverse Momentum ( $p_T^{miss}$ ): Calculated from Particle-Flow candidates.
Signal Regions (SR):
- SR3 $\ell$ : 3 leptons, 1 Opposite-Sign Same-Flavor (OSSF) pair within the $Z$ mass window ( $m_Z \pm 15$ GeV), $\ge 2$ jets (at least 1 $b$ -tagged).
- SR4 $\ell$ : 4 leptons, exactly 1 OSSF pair within the $Z$ mass window, $\ge 1$ $b$ -tagged jet.
Control Regions (CR): Defined to constrain backgrounds ($ZZ$, $WZ$, Drell-Yan, $t\bar{t}$ ) and validate modeling of non-prompt leptons and jet multiplicities.

Simulation and Background Modeling

Signal: Generated with MADGRAPH5_aMC@NLO at NLO accuracy in the 5-flavor scheme. Overlap with $t\bar{t}Z$ is handled using the MADSTR plugin with Diagram Removal (DR) techniques (nominal DR1 scheme).
Backgrounds:
- $t\bar{t}Z$ and $tZq$: MADGRAPH5_aMC@NLO (NLO).
- $WZ$: POWHEG v2 (NLO).
- Other processes (VV, X $\gamma$ , $tX$): Various generators (MG, POWHEG, MCFM).
Non-prompt Leptons: Estimated via a data-driven fit to the trailing lepton $p_T$ in control regions, treating misidentification rates as free parameters.

Machine Learning (ML) Analysis

A key innovation is the use of Particle Transformer (PART) algorithms, a transformer-based architecture designed for jet tagging and event classification.

Architecture: Inputs include particle four-momenta, $\eta$ , $\phi$ , $b$ -tagging scores, and pairwise interaction features ( $\Delta$ , $k_T$ , $z$ , $m^2$ ).
Training:
- SR3 $\ell$ : Four-output node classifier distinguishing $tWZ$, $t\bar{t}Z$ , $WZ$, and other backgrounds. Events are binned based on the maximum output node.
- SR4 $\ell$ : Binary classifier ($tWZ$ vs. background) due to lower statistics.
Strategy: Separate models trained for 13 TeV and 13.6 TeV data to account for detector performance differences.

Statistical Analysis

Method: A binned maximum likelihood fit using the COMBINE tool.
Inputs: Distributions of classifier output nodes in SRs and control variables (jet multiplicity in CRs, trailing lepton $p_T$ ) in CRs.
Systematics: Includes experimental uncertainties (jet energy scale, lepton ID, $b$ -tagging, luminosity) and theoretical uncertainties (PDFs, scales, PS matching, background normalizations). $t\bar{t}Z$ normalization is treated as a free parameter in a simultaneous fit.

3. Key Contributions

First Observation: This paper reports the first observation of $tWZ$ production, surpassing the "evidence" threshold.
Advanced ML Application: The successful deployment of Transformer-based architectures (PART) significantly improved signal-background discrimination compared to previous analyses.
Combined Dataset: The analysis combines Run 2 (13 TeV) and Run 3 (13.6 TeV) data, utilizing improved reconstruction algorithms for the 13.6 TeV dataset (e.g., better non-prompt lepton rejection).
Simultaneous Fit: The measurement of $tWZ$ and $t\bar{t}Z$ signal strengths simultaneously without external constraints on the $t\bar{t}Z$ cross-section, reducing correlations between the two measurements.

4. Results

Statistical Significance:
- Observed significance: 5.8 standard deviations (s.d.).
- Expected significance (background-only hypothesis): 3.5 s.d.
- This exceeds the 5.0 s.d. threshold required for a formal "observation."
Signal Strength ( $\mu$ ):
- Combined $\mu_{tWZ} = 1.77 \pm 0.23 \text{ (stat)} \pm 0.22 \text{ (syst)}$ .
- This is consistent with the SM prediction within 2.3 s.d. (the central value is slightly higher than expected, but not statistically significant).
Cross Sections:
- At $\sqrt{s} = 13$ TeV: 248 $\pm$ 52 fb.
- At $\sqrt{s} = 13.6$ TeV: 242 $\pm$ 77 fb.
- (SM NLO predictions: ~136 fb and ~148 fb, respectively).
Background Constraints:
- The simultaneous fit yielded a $t\bar{t}Z$ signal strength of $\mu_{t\bar{t}Z} = 0.91 \pm 0.07$ , consistent with SM predictions.
- The correlation between $\mu_{tWZ}$ and $\mu_{t\bar{t}Z}$ was significantly reduced compared to previous CMS analyses due to the improved ML separation.

5. Significance

Validation of SM: The observation confirms the SM prediction for this rare electroweak process, validating the theoretical calculations of top quark interactions with $W$ and $Z$ bosons.
BSM Sensitivity: By establishing a baseline measurement with high precision, this result sets stringent limits on anomalous top-quark couplings. Any future deviation in the measured cross-section or kinematic distributions could indicate new physics (e.g., heavy resonances or effective field theory operators).
Methodological Benchmark: The success of the Particle Transformer algorithm in isolating rare signals from complex backgrounds demonstrates the power of modern deep learning techniques in high-energy physics, paving the way for future analyses of even rarer processes at the High-Luminosity LHC.
Future Prospects: This measurement opens the door for differential cross-section studies and detailed investigations of the top quark's electroweak sector, crucial for understanding the mechanism of electroweak symmetry breaking and potential new physics scales.