Search for dijet resonances with data scouting in proton-proton collisions at $\sqrt{s}$ = 13 TeV

This paper presents a search for narrow dijet resonances in the 0.6–1.8 TeV mass range using 117 fb$^{-1}$ of 13 TeV proton-proton collision data collected via the "data scouting" technique, finding no evidence of new particles and setting model-independent upper limits on production cross sections and dark matter mediator couplings.

The Gist

The Big Picture: Hunting for Invisible Ghosts in a Storm

Imagine the Large Hadron Collider (LHC) at CERN as a massive, high-speed train station where two trains of protons crash into each other at nearly the speed of light. When they crash, they create a chaotic explosion of debris—particles flying everywhere.

Most of the time, this debris is just "background noise," like static on a radio. But physicists are looking for something specific: a new, heavy particle that shouldn't exist according to our current rulebook (the Standard Model). If this new particle exists, it would be like a ghost that appears for a split second and then instantly splits into two jets of debris.

This paper is a report from the CMS experiment (one of the detectors at the LHC) saying: "We looked very hard for these ghosts, but we didn't find any."

The Challenge: The "Data Scouting" Trick

Usually, when these collisions happen, the data is so massive that the computer system has to be very picky. It only saves the "most interesting" crashes (usually the ones with the highest energy), throwing away the rest to save space.

However, the new particles the scientists were looking for might be lighter than what the standard filters catch. To find them, the CMS team used a clever trick called "Data Scouting."

• The Analogy: Imagine a security guard at a concert who usually only writes down the names of VIPs (high-energy events). But for this search, the guard decided to write down a quick, abbreviated note about everyone who walked through the door, even if they looked like regular fans.
• The Result: By using this "abbreviated note" method, they could lower the threshold and catch collisions that usually get ignored. This allowed them to look for particles with masses between 0.6 and 1.8 TeV (a range that was previously hard to explore with full data).

The Search: Looking for a Spike in the Noise

The scientists analyzed 117 "inverse femtobarns" of data (a fancy way of saying they looked at a huge number of collisions collected between 2016 and 2018).

They looked at the "dijet mass spectrum."

• The Analogy: Imagine you are listening to a crowd of people talking. The background noise (QCD events) sounds like a smooth, steady hum that gets quieter as the volume gets higher.
• The Goal: They were looking for a sudden, sharp spike or a "bump" in that smooth hum. A spike would mean a new particle was created and decayed into two jets.

The Findings: Smooth Sailing, No Ghosts

After crunching the numbers, the result was clear:

1. No Spikes Found: The data looked exactly like the smooth, predictable background hum. There were no sudden bumps.
2. The "Ghost" is Still a Ghost: They did not find evidence of new particles like heavy versions of the Z boson, axigluons, or dark matter mediators in the mass range they searched.
3. Setting the Rules: Even though they didn't find the particles, they set a "speed limit." They can now say with 95% confidence that if these particles do exist, they must be either much heavier than 1.8 TeV or interact with normal matter so weakly that this experiment couldn't see them.

A Special Note on Dark Matter

The paper also looked specifically at Dark Matter mediators. These are hypothetical particles that act as a bridge between normal matter (quarks) and invisible Dark Matter.

• The Result: They found that if these mediators exist, their "handshake" (coupling strength) with normal matter must be incredibly weak (less than 0.04).
• The Surprise: The sensitivity of this search was better than expected. Usually, if you double your data, you only improve your sensitivity by a small amount (the square root of two). But because they used a smarter statistical method (using fewer "knobs" to tune their background model), they got a much bigger boost in sensitivity than the data volume alone would suggest.

The Conclusion

The CMS team successfully used a "data scouting" technique to scan a specific mass range for new particles. They found the background noise was perfectly smooth, meaning no new narrow resonances were discovered in this range.

However, the search wasn't a failure. By ruling out these particles in this specific mass range, they have narrowed the map for future explorers, telling them: "Don't look here; the treasure isn't buried in this spot." They also proved that their new statistical method is a powerful tool for finding subtle signals in the future.

Technical Summary

1. Problem Statement

The search for new physics beyond the Standard Model (SM) often involves looking for narrow resonances decaying into pairs of partons (quarks and gluons), which manifest as dijet resonances in the invariant mass spectrum ($m_{jj}$).

• The Challenge: Standard high-mass searches (above 2 TeV) rely on offline reconstruction of jets, requiring high trigger thresholds ($H_T > 1050$ GeV) to manage data bandwidth. This limits sensitivity to lower mass resonances (0.5–2 TeV) where the QCD background is overwhelming.
• The Gap: While low-mass searches exist (using initial-state radiation triggers), the "medium-mass" region (0.6–1.8 TeV) has historically been difficult to probe with high sensitivity due to trigger bandwidth limitations and the steeply falling QCD background.
• Specific Goal: To search for narrow resonances in the mass range 0.6 to 1.8 TeV decaying to $qq$, $qg$, and $gg$ final states, and to set limits on Dark Matter (DM) mediators and various BSM models (e.g., $Z'$, $W'$, axigluons, excited quarks).

2. Methodology

A. Data Scouting Technique

The core innovation of this analysis is the use of "Data Scouting" (also known as trigger-level analysis).

• Mechanism: Instead of storing full event data, the High-Level Trigger (HLT) reconstructs jets using only calorimeter information (Calo-jets) and saves a compact representation of the event.
• Advantage: This drastically reduces the event size and reconstruction time, allowing for a significantly lower trigger threshold.
• Trigger Configuration: The analysis uses a trigger requiring the scalar sum of transverse momenta of all jets ($H_T$) to exceed 250 GeV (compared to >1050 GeV for standard offline triggers). This enables access to the 0.6–1.8 TeV mass range.

B. Dataset and Detector

• Collisions: Proton-proton collisions at $\sqrt{s} = 13$ TeV.
• Period: Data collected during 2016, 2017, and 2018.
• Integrated Luminosity: 117 fb$^{-1}$.
• Jet Reconstruction: Jets are reconstructed using the anti-$k_T$ algorithm ($R=0.4$) with the FASTJET package. They are corrected for pileup using jet area methods.
• Event Selection:
  • Two leading jets with $p_T > 30$ GeV and $|\eta| < 2.5$.
  • Wide Jets: Spatially close jets ($\Delta R < 1.1$) are combined into "wide jets" to capture hard gluon radiation and improve mass resolution.
  • Angular Cut: A requirement of $|\Delta\eta| < 1.3$ is applied to suppress the t-channel QCD background (which peaks at large $|\Delta\eta|$) and enhance sensitivity to isotropic resonance decays.

C. Signal and Background Modeling

• Signal: Simulated using PYTHIA 8.205 with the CUETP8M1 tune and GEANT4 detector simulation. Signal shapes are modeled for three parton combinations: $qq$, $qg$, and $gg$.
  • Note: $gg$ resonances are broader with more pronounced low-mass tails due to higher QCD radiation compared to $qq$.
• Background: The QCD background is estimated directly from data using a smooth, monotonically falling parametric function.
  • Function: A modified exponential function with four parameters: $p_0 \exp(p_1 x^{p_2} + p_1(1-x)^{p_3})$, where $x = m_{jj}/\sqrt{s}$.
  • Optimization: To avoid biases and ensure stability, the analysis simultaneously fits 14 distinct data-taking periods (eras) from 2016–2018 rather than fitting the full dataset at once. This prevents correlations between signal shapes and background parameters from creating discontinuous likelihood curves.

D. Statistical Analysis

• Framework: A multibin counting experiment likelihood (product of Poisson distributions) is used.
• Systematics: Major uncertainties include Jet Energy Scale (JES, $\pm 2\%$), Jet Energy Resolution (JER, $\pm 10\%$), integrated luminosity, and background parameterization.
• Limits: Limits are set using the CL$_s$ criterion with the COMBINE tool. The background parameters are treated as nuisance parameters that float freely for each data-taking period.

3. Key Contributions

1. Extended Mass Reach via Data Scouting: Successfully probes the 0.6–1.8 TeV mass range with a trigger threshold ($H_T > 250$ GeV) that is roughly 4 times lower than standard offline triggers, accessing previously unexplored coupling strengths.
2. Robust Statistical Procedure: Introduced a simultaneous fit across multiple data-taking eras to handle the high statistical precision of the scouting data. This approach minimizes correlations between signal and background parameters, resulting in stable $\Delta NLL$ curves and improved sensitivity beyond simple statistical scaling.
3. Model-Independent Limits: Provided limits on the product of cross-section, branching fraction, and acceptance ($\sigma \times B \times A$) for $qq$, $qg$, and $gg$ final states, applicable to any narrow resonance model.
4. Dark Matter Mediator Constraints: Presented the first CMS limits on the coupling of DM mediators to quarks ($g_q$) as a function of mediator mass, reaching coupling values as low as 0.04.

4. Results

• Observation: The dijet mass spectra for all years and the combined dataset are well-described by the smooth background parameterization.
  • No significant excess was observed.
  • The largest local significance was 2.2 standard deviations ($\sigma$) at a resonance mass of 0.8 TeV for the $qq$ channel, which is consistent with a background fluctuation.
• Exclusion Limits (95% CL):
  • BSM Models: The analysis excludes the following benchmark models over the entire mass range (0.6–1.8 TeV) for the considered couplings:
    • New gauge bosons ($W'$, $Z'$)
    • Axigluons and Colorons
    • Excited quarks
    • Scalar diquarks
    • Color-octet scalars
  • Dark Matter: Limits on the universal quark coupling $g'_q$ for a leptophobic $Z'$ mediator were set. The observed limits significantly improve upon previous CMS results at 8 TeV and 13 TeV.
• Sensitivity Gain: The sensitivity achieved is better than expected from a simple scaling of integrated luminosity ($\sqrt{L}$). This improvement is attributed to the reduced number of parameters in the background function and the robust statistical treatment of the data eras.

5. Significance

This paper demonstrates the critical value of Data Scouting in high-energy physics. By bypassing the storage bottlenecks of full event reconstruction, CMS can access lower mass resonances with high statistics.

• Methodological Advance: The paper establishes a new standard for statistical analysis in high-statistics searches, showing that fitting multiple data eras simultaneously yields more stable and sensitive results than global fits.
• Physics Impact: The results provide the most stringent constraints to date on a wide variety of narrow dijet resonance models in the 0.6–1.8 TeV range.
• Dark Matter: The constraints on DM mediators are competitive with the most recent ATLAS results (132 fb$^{-1}$) and represent a significant step forward in constraining simplified models of dark matter interactions at the LHC.

In summary, this analysis successfully utilized a specialized trigger strategy and advanced statistical techniques to push the boundaries of dijet resonance searches, ruling out several popular BSM theories and setting new benchmarks for Dark Matter mediator couplings.