Narrow Dijet Resonance Searches

• Narrow dijet resonance production is the resonant creation of new, short-lived particles decaying into high-energy jets, providing crucial insights into physics beyond the Standard Model.
• Advanced experimental techniques such as jet reconstruction, data scouting, and b-tagging are employed to accurately extract signals and suppress the dominant QCD background.
• Sophisticated statistical methods and machine learning approaches enhance sensitivity and enable precise discrimination among various BSM scenarios by robustly setting exclusion limits.

Narrow dijet resonance production refers to the resonant creation of new, short-lived particles at high-energy colliders that subsequently decay into a pair of highly energetic jets, reconstructed from the hadronization of quarks and/or gluons. The term “narrow” designates resonances whose intrinsic widths are much smaller than the experimental resolution, leading to localized excesses (“bumps”) over the steeply falling QCD dijet invariant mass spectrum. These searches are at the forefront of LHC phenomenology, providing stringent constraints on a diverse range of physics beyond the Standard Model (BSM), including new gauge bosons, colored states, and extensions motivated by string theory, compositeness, or extra dimensions.

1. Theoretical Motivation and Phenomenology

Narrow dijet resonances are motivated by numerous BSM models that predict new color-singlet or colored states with strong couplings to quarks and/or gluons. Representative scenarios include:

• Leptophobic Z′ or W′ gauge bosons, colorons, or axigluons (new vector bosons);
• Excited quarks (q∗), color-octet scalars, or diquarks (scalar or fermionic color-multiplet states);
• Regge excitations in string theory, manifest as resonances in parton scattering amplitudes;
• Dark matter mediators, coupling SM to DM particles in simplified models;
• Randall–Sundrum gravitons and E₆ diquarks.

The signature is a resonance in the dijet invariant mass, $$m_{jj} = \sqrt{(E_1 + E_2)^2 - (\vec{p}_1 + \vec{p}_2)^2}$$, superimposed on a QCD background. The natural width, $\Gamma$, of the resonance is typically much smaller than the detector resolution, justifying the narrow width approximation (NWA). The production cross section factors as $$\sigma(pp \to X \to jj) \approx \sigma(pp \to X) \times \mathrm{BR}(X \to jj)$$, enabling model-independent interpretations and direct mapping to theory predictions (Collaboration, 2010, Torre, 2011, Collaboration, 2013).

2. Experimental Techniques and Data Acquisition

Dijet resonance searches exploit several advanced strategies in data collection and reconstruction:

• Jet reconstruction: Use of the anti-$k_t$ algorithm (commonly $R=0.4$–$0.7$) (Collaboration, 2010, Collaboration, 2011), with wide-jet algorithms merging additional jets within a radius $\Delta R < 1.1$ to improve mass resolution and account for hard QCD radiation, especially important for gluon-initiated final states (Collaboration, 2012, Collaboration, 2011).
• Triggering and data scouting: Significant reliance on hardware and software triggers with thresholds optimized to balance efficiency and bandwidth. Development of the “data scouting” technique (Collaboration, 2016, Collaboration, 24 Oct 2025), whereby only reduced event information (calorimeter jet four-vectors and minimal event-level information) is recorded at high rates, has enabled the extension of searches to lower dijet masses and smaller couplings.
• b-tagging: Deployment of multivariate (CSV, DeepJet) taggers based on secondary vertex and tracking information, in both offline and high-level trigger streams, enhances sensitivity to final states with $b$ quarks (Collaboration, 2012, Collaboration, 2018, Collaboration, 2022).
• Event selection: Kinematic requirements such as $|η| < 2.5$, $|\Delta η| < 1.3$, stringent $p_T$ thresholds, and vertex quality ensure robust signal acceptance and background suppression (Collaboration, 2010, Collaboration, 2012).

3. Background Modeling and Signal Extraction

The dominant background is the QCD multijet process, yielding a smoothly falling $m_{jj}$ spectrum. This is effectively modeled by empirical parameterizations, e.g.,

$$\frac{dσ}{dm} = P_0 (1 - \frac{m}{\sqrt{s}})^{P_1} (\frac{m}{\sqrt{s}})^{P_2 + P_3 \ln(m/\sqrt{s})}$$

with $P_i$ unconstrained fit parameters (Collaboration, 2010, Collaboration, 2011, Collaboration, 2013, Collaboration, 2015). Further refinements (e.g., lower parameter forms, modified exponentials) have reduced background-signal correlations and improved robustness even with growing luminosity and statistics (Collaboration, 24 Oct 2025).

Signal shapes for narrow resonances are simulated using PYTHIA+GEANT, capturing a Gaussian core (from jet energy resolution) with low-mass QCD-induced tails. Gluon-initiated signals exhibit broader, more asymmetric profiles due to soft QCD emission.

Recent work has introduced a two-dimensional fit in the four-jet invariant mass ($m_{4j}$) and the mean dijet mass $(\langle m_{jj} \rangle)$ for pair-produced dijet resonances, as well as rewritten or complementary “ratio methods” for the background using control regions to minimize systematic uncertainties (Collaboration, 23 Jul 2025, Collaboration, 2019).

4. Statistical Analysis and Limit Setting

Upper limits are routinely set on the product

$σ × \mathrm{BR} × A,$

where $σ$ is the cross section, $\mathrm{BR}$ the resonance branching fraction to dijets, and $A$ the acceptance (typically $A \sim 0.6$ for isotropic decays within $|η| < 2.5$ and $|\Delta η| < 1.3$) (Collaboration, 2010, Collaboration, 2013). Both Bayesian (with flat priors and systematic uncertainties incorporated via convolution) and frequentist (profiled likelihood, $CL_s$) procedures are standard (Collaboration, 2011, Collaboration, 2012, Collaboration, 2019, Collaboration, 2019).

When no significant excess is seen, observed and expected $95\%$ CL upper limits are compared with theoretical cross sections to derive mass exclusions for benchmark models:

• String resonances: up to 7.9 TeV (Collaboration, 2019).
• Scalar diquarks: up to 7.5 TeV (Collaboration, 2019).
• Excited quarks: up to 6.0 TeV (Collaboration, 2018).
• Axigluons/colorons: up to 6.1 TeV (Collaboration, 2018).
• Z′ and W′ bosons: up to 2.7–3.3 TeV, depending on the scenario (Collaboration, 2018, Collaboration, 2022).
• b-tagged searches: exclusion of excited $b$ quarks to 4.0 TeV (Collaboration, 2022).

Dedicated low-mass searches using data scouting have established the most stringent upper limits in the $0.35$–$1.8$ TeV window and placed strong constraints on the coupling $g_q$ of dark matter mediators (down to $g_q \sim 0.04$) (Collaboration, 24 Oct 2025, Collaboration, 2019).

5. Model Discrimination and Advanced Analysis Techniques

To discriminate the nature of a discovered resonance, model-independent observables have been proposed. The “color discriminant variable” $D_{col}$ is constructed from the resonance mass, total width, and cross section,

$D_{col} = \frac{M^3}{\Gamma} \, σ_{jj},$

and reformulated as

$$D_{col} = 16π^2\, {\cal N} \left(\sum_{xy=jj} \mathrm{BR}(R \to xy)\right) \left(\sum_{ik} (1+\delta_{ik}) \mathrm{BR}(R \to ik) \left[\tau \frac{dL^{ik}}{d\tau}\right]_{\tau = m_R^2 / s}\right),$$

where the normalization ${\cal N}$ reflects the color and spin factors, and the sum runs over partonic initial and dijet final states. $D_{col}$ depends only on quantum numbers and parton luminosities for a narrow resonance, providing a powerful tool to distinguish, e.g., color-sextet and color-triplet diquarks, colorons, or Z′ bosons once a resonance is observed (Chivukula et al., 2015, Ittisamai et al., 2015).

Machine learning (ML)-based anomaly detection using weakly supervised neural networks trained directly on data has been implemented for model-agnostic searches, operating in reduced-dimensional subspaces such as the jet mass plane ($m_1, m_2$) (Collaboration, 2020). This approach allows enhanced sensitivity in searches for $A \to BC \to$ four-jet final states, increasing discovery reach where traditional resonance-hunting strategies suffer large trials factors or lose efficiency due to jet substructure at high boosts.

6. Benchmark Results and Impact on BSM Parameter Space

Tables of exclusion limits (published in (Collaboration, 2010, Collaboration, 2011, Collaboration, 2012, Collaboration, 2015, Collaboration, 2019, Collaboration, 24 Oct 2025)), upper limits on $σ \cdot \mathcal{B} \cdot A$ for $qq$, $qg$, $gg$ channels, and bounds on effective couplings (e.g., $c_3, c_G, g_q$) across the accessible mass spectrum (from hundreds of GeV to multi-TeV) constitute the principal outputs. Mass limits have been extended with growing luminosity and progressive algorithmic improvements, significantly narrowing the allowed BSM parameter space.

For example, the combination of data scouting and robust background parameterizations has enabled sensitivity to benchmark couplings well below the 0.1 level for dark matter mediators in the $0.5$–$1.8$ TeV mass range (Collaboration, 24 Oct 2025). b-tagged studies have constrained scenarios with preferential couplings to heavy quarks, improving by orders of magnitude the constraints existing prior to advanced trigger development (Collaboration, 2022).

7. Outlook and Future Directions

Future LHC data taking and analysis will continue to benefit from:

• Enhanced low-mass sensitivity through robust data scouting and improved trigger strategies;
• Refined ML techniques for anomaly detection and model-agnostic bump searches, especially in high-dimensional phase space or when dealing with complex decay chains ($A \to BC$);
• Improved systematic control via control-region-based, data-driven background estimations (e.g., the “ratio method” (Collaboration, 2019));
• Extension of analyses to broad resonance scenarios, where natural widths become a non-negligible fraction of the resonance mass and signal shapes require precise modeling of energy-dependent Breit–Wigner forms (Collaboration, 2018, Collaboration, 23 Jul 2025).

Hybrid analyses involving angular distributions, associated production, or double-resonance signatures ($pp \to S \to XX \to (jj)(jj)$) are now standard for protracted excesses (Collaboration, 23 Jul 2025). Continued development of multivariate and machine learning techniques in resonance searches offers a promising avenue to consistently extend exclusion limits and increase new physics discovery potential.

The aggregate body of results places narrow dijet resonance searches at the core of BSM phenomenology at the LHC: systematically benchmarking new models, slicing away large regions of parameter space, and providing immediate constraints on new strongly interacting sectors, dark sectors, and exotic states. The current methodological infrastructure ensures that both anticipated and unanticipated signals—narrow or otherwise—can be robustly scrutinized and, if extant, discovered.