Dijet resonance searches look for a localized excess in the invariant mass of two jets, a possible signature of a new particle produced in a proton-proton collision. Their reach is challenged by the large, smoothly falling background from quantum chromodynamics (QCD). Quark-gluon tagging offers one way to improve separation: it uses differences between jets initiated by quarks and those initiated by gluons to distinguish signal-like events from background.
A study focused on an axial-vector dark matter mediator, Z′, examined whether tagging could strengthen limits in the dijet mass spectrum. Its findings point to a practical but relatively modest benefit for the tested selection methods, alongside greater potential in lower-mass regions.
Why quark-gluon tagging matters in a dijet search
In proton-proton collisions, partons can scatter and produce energetic quarks or gluons. These partons undergo showering and hadronization, forming collimated particle sprays reconstructed as jets. QCD processes produce dijet events in abundance, making them the dominant background for many searches involving hadronic final states.
A new particle that decays to quarks could also produce two jets. If the particle creates a resonance, its contribution may appear as a local excess over the smoothly decreasing background in the dijet invariant mass, mjj. The challenge is to identify a small signal among many background events.
Quark- and gluon-initiated jets differ in their particle-level structure. In this study, the main discriminator was the number of charged-particle tracks associated with a jet: track multiplicity. The analysis explored whether using this information to divide events into samples could make the signal more distinct from the QCD background.
How the tagging strategy was evaluated
The analysis considered a Z′ mediator that couples to quarks and can decay back to Standard Model particles. The specific signal used to optimize the tagger had a quark coupling of gq = 0.2. Because the signal was expected to produce quark-initiated jets, while the lower-mass QCD background was dominated by gluon-gluon dijet events, quark/gluon discrimination offered a potential route to improving the signal-to-background ratio.
Rather than relying only on a fixed track-count threshold, the study tested separation functions that accounted for jet transverse momentum (pT) and dijet invariant mass. Both linear and logarithmic forms were examined, with different starting parameters. This matters because jet properties and the mixture of quark- and gluon-initiated events can vary across the kinematic range. A tagging criterion that works in one region may not provide the same separation elsewhere.
The samples were generated with Pythia 8 using leading-order NNPDF 2.3 parton distribution functions and the A14 tune. Detector response was modeled with Geant 4. The resulting tagged subsamples were compared with an untagged dijet sample designed to represent the 2015–2016 ATLAS dataset, corresponding to 37 fb−1. The comparison focused on the limits obtained for the signal, rather than treating tagging performance as a standalone classification metric.
Ideal separation versus practical performance
The study distinguished between an idealized benchmark and separation based on measurable jet properties. In the ideal case, jets were separated using the initiating parton’s truth identity. This provides a useful reference for the maximum potential of quark/gluon discrimination, but it is not an operational tagging method in observed collision data, where the initiating parton is not directly known.
With this truth-based separation, the improvement in significance reached up to 50%. By contrast, the best tested functions based on jet pT and mjj produced improvements in the range of 2–5%. The difference underscores an important point for analysis design: strong separation in an idealized benchmark does not automatically translate into comparable gains when using practical observables.
For engineers and analysts evaluating similar strategies, the results also show why performance should be assessed in the context of the full search. The relevant question is not only whether a tagger distinguishes jet types, but whether applying it improves the final sensitivity after event selection and background estimation.
Why dijet mass dependence deserves attention
The separation functions incorporated both jet transverse momentum and dijet mass, and the study found that the mass dependence could be especially useful below the mass range analyzed. This is a promising direction, not a demonstrated improvement in that lower range: the finding indicates potential that would require dedicated investigation.
The underlying reason for considering mass is that the composition of the QCD background and the signal can differ across the spectrum. A tagging approach that accounts for this changing context may be better suited to the search than one using a single, mass-independent criterion. However, the reported results do not establish a universal function or guarantee that the same gains will apply to other signal models or datasets.
Practical implications for resonance searches
Quark-gluon tagging is best understood as a potential refinement to a dijet analysis, not a replacement for its core elements. The resonance search still depends on reliable jet reconstruction, event selection, modeling of the smoothly falling background, and evaluation of the resulting limits. Tagging adds another dimension to that process by organizing events according to information about jet structure.
- Use realistic benchmarks: Truth-level separation can show the upper potential of a method, but measurable observables are needed to estimate practical gains.
- Test kinematic dependence: Jet pT and mjj may affect the useful separation, so a selection should be evaluated across the relevant mass range.
- Measure search-level impact: Improvements in a tagging metric matter only if they translate into stronger sensitivity or limits for the target signal.
- Keep conclusions scoped: Results for one Z′ model and simulated dataset should not be assumed to apply unchanged to other signals or analyses.
Conclusion
In the studied ATLAS dijet search, quark-gluon tagging based on track multiplicity offered a way to exploit differences between quark-rich signal events and the QCD background. Ideal truth-based separation showed substantial potential, but practical functions using jet transverse momentum and dijet mass improved significance by 2–5%. The analysis also identified lower dijet-mass regions as an area for further study. For future resonance searches, the central lesson is to validate tagging against realistic observables and judge its value by its effect on the final search sensitivity.