Access Official Technical PDF

Share Summary:LinkedInEmail

A jet substructure measurement examines how energy and momentum are distributed inside a reconstructed jet. In high-energy particle physics, that internal pattern can help distinguish jets produced by the hadronic decays of massive particles, such as W bosons and top quarks, from more common jets initiated by light quarks or gluons. ATLAS studies using 2016 Large Hadron Collider data show how classification, grooming and careful detector corrections work together to make these measurements useful for testing physics models.

Why measure the structure inside jets?

Jets are reconstructed from the spray of particles created when quarks and gluons produced in a collision interact and fragment. A jet’s overall energy and direction do not capture all the information in that spray. Its internal angular and energy structure—its substructure—can provide clues about how it formed.

For example, a high-energy W boson or top quark can decay into multiple quarks. Their products may be reconstructed together as a large-radius jet, with a pattern that differs from a light-parton jet. Substructure variables quantify features of this pattern. They support jet tagging, in which jets are classified by likely origin, and they also provide measurable distributions that can be compared with theoretical predictions and Monte Carlo (MC) simulations.

Jet substructure measurement: classification and comparison

The ATLAS work had two connected aims. One was to develop substructure-based taggers for top-quark and W-boson jets. The other was to measure a range of substructure variables for top, W and light-parton jets, providing data against which simulations could be assessed.

The tagger study used simulated samples to identify variables that help separate signal jets from light-quark and gluon backgrounds. Candidate variables were evaluated in combination with jet mass, and selection thresholds were optimized at different signal-efficiency working points. The study considered, among other inputs, energy correlation observables and N-subjettiness measures. These describe aspects of how energy is distributed across a jet and how well that pattern aligns with a hypothesized number of subjets.

Rather than treating tagging performance as independent of jet momentum, the optimization examined how performance varied with transverse momentum. This matters when a tagger is used across a range of collision conditions: its selection criteria need to reflect how the signal and background distributions change with jet kinematics. The resulting taggers were developed for use in ATLAS analyses of hadronic decays.

Grooming helps focus the measurement

Jet grooming reduces the influence of softer, wider-angle contributions that can obscure the structure of interest. The measurement studied both trimmed and soft-drop groomed jets. These procedures retain selected constituents or branches while removing others according to the grooming algorithm’s criteria.

Grooming does not remove the need to understand the detector. A measured jet is built from reconstructed objects, and the calorimeter’s response and clustering affect the observed energy pattern. As a result, substructure distributions can be sensitive to how calorimeter deposits are grouped into clusters—not just to the physics of the original collision.

Three event selections probe different jet types

The measurement used three distinct regions of phase space to study light jets, top-quark jets and W-boson jets. For the inclusive multijet region, a single-jet trigger selected events with two central jets and no leptons. This provided a sample for measuring light-parton jet substructure.

The top and W regions were selected from semileptonic top-pair events. In this topology, one top quark decays through a leptonically decaying W boson, while the recoiling hadronic system contains the jet being studied. The analysis used the leptonically decaying side to identify the event and then separated candidate top and W jets primarily through their angular distance from the nearest b-tagged jet, alongside jet-mass requirements.

That separation illustrates an important analysis principle: object classification often depends on context. The jet’s own mass and substructure provide information, while nearby reconstructed objects help establish which decay it may belong to.

Detector corrections and uncertainty control

Raw detector-level distributions are not a direct view of the underlying particle-level quantities. Reconstruction can shift events between bins or cause some events to be missed. ATLAS therefore corrected the measured distributions for detector effects using unfolding, including an iterative Bayesian approach. Response matrices describe how particle-level quantities map to detector-level observations within the simulated samples.

Unfolding depends on modelling and must be tested. The analysis performed checks of iteration stability and closure, and examined how using alternative simulations affected the correction. These checks help identify whether a result is unduly shaped by the assumptions used to connect detector observations to particle-level distributions.

Systematic uncertainties also included a novel, bottom-up treatment of calorimeter cluster effects. By varying aspects of cluster reconstruction and evaluating their impact on substructure distributions, the analysis assessed which observables were particularly sensitive to detector clustering. This is significant because a variable can be theoretically useful yet experimentally challenging if small changes in reconstructed inputs produce substantial shifts.

What the results provide

The corrected distributions make it possible to compare measured substructure with predictions from different MC generators. Such comparisons help evaluate how well simulations describe the internal structure of light, W-boson and top-quark jets. Better understanding of this modelling is important in analyses that rely on jet tagging or use substructure to study particle decays.

The practical lesson is that reliable classification and precision measurement are complementary, not interchangeable. Taggers use substructure to select likely signal jets; measurements test whether simulations reproduce the observed distributions. Both depend on well-defined event selections, grooming choices and a transparent accounting of detector and modelling uncertainties.

Conclusion

The ATLAS study demonstrates a complete approach to jet substructure: develop classifiers using simulated data, measure distributions in carefully selected event samples, correct for detector effects, and compare the results with simulation. For technical teams working with complex measurement pipelines, the broader methodological takeaway is to validate each stage—from input reconstruction through classification and correction—rather than treating the final output as independent of the system that produced it.

Access Official Technical PDF

About The Author