Understanding the transverse-momentum dependence of constituent-based top-tagging performance

Explain why constituent-based top-tagging classifiers reach peak background rejection at intermediate jet transverse momentum and degrade at very high transverse momentum.

Background

The thesis reports that constituent-based top taggers perform best for jets with transverse momenta of approximately 1–2 TeV. The degradation at very high transverse momentum is expected because the decay products become increasingly collimated, but the enhanced performance at intermediate transverse momentum is not explained. The text suggests that the training-sample distribution may contribute, but does not establish the cause.

References

The degradation of the performance at very high $\pt$ is expected, but the better performance at intermediate $\pt$ is not fully understood.

— A High- and Variable-Dimensional Measurement of the $Z$+jets Differential Cross Section with the ATLAS Experiment and Artificial Intelligence  (2608.28449 - Greif, 28 Aug 2026) in Section 4.4, “Results”