Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.33757

Dissecting flavor tagging with graph neural networks

Ankita Budhraja·Sunny Kumar Singh·Balbeer Singh
View original source ↗Read source document ↗Export metadata ↓

Abstract

Graph neural networks such as ParticleNet are widely used for quark/gluon jet tagging. We study how the discriminating power of such a network is distributed between constituent multiplicity and the remaining structure of the jet, differentially in jet transverse momentum and multiplicity. To this end, we train ParticleNet on quark and gluon jets from Z+jet production in Pythia, with the two classes matched in transverse momentum and rapidity over the wide range $80\leq p_{T}^{\rm jet}\leq 400$ GeV. We find that the network learns how the fraction of gluon jets changes with the multiplicity in every $p_{T}^{\rm jet}$ bin, including the shift of the multiplicity at which quark and gluon jets are equally likely, which implies that it uses the jet scale (reconstructed from its constituents) jointly with its substructure. At fixed multiplicity, the gluon probabilities it assigns agree on average with the observed gluon fractions, but are widely spread about the mean. This allows us to interpret its output as the sum of the multiplicity log-likelihood ratio and a residual log-likelihood ratio at fixed multiplicity, which allows the network to correctly identify 20-34% of the jets whose multiplicity favors the opposite class. By rescaling the radial spread of the constituents at fixed multiplicity, we find that the network considers broader jets more gluon-like, and that this response is largest at low and intermediate multiplicity, where the multiplicity alone does not decide the class of a jet.

Explore related subjects

hep-phhep-ex

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Ankita Budhraja, Sunny Kumar Singh, Balbeer Singh. 2026-10-24. Dissecting flavor tagging with graph neural networks. https://arxiv.org/abs/2610.33757

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

A Coupled Source Description of Pseudorapidity Distributions from RHIC to LHC: Emergent $1/μ_B$ Scaling and Limiting Fragmentation

One of the most remarkable observations in heavy-ion collisions is the systematic regularity exhibited by pseudorapidity distributions of charged particles across collision energies. While single-source models fail at higher energies and independent multi-source approaches do not reproduce the central dip observed at LHC energies, a unified description across the full RHIC-LHC energy range remains elusive. These distributions from Au+Au collisions at RHIC ($\sqrt{s_{NN}}$ = 19.6--200 GeV) and Pb+Pb collisions at LHC ($\sqrt{s_{NN}}$ = 2.76--5.36 TeV) are analyzed using a novel parametrization based on coupled Gaussian sources where the interaction strength is quantified by parameter $λ$. This coupled two-source model captures the interaction between forward and backward sources through the medium formed in the collision. Remarkably, $λ$ exhibits empirical scaling behavior resembling $1/μ_B$, suggesting sensitivity to baryon stopping and the strongly-interacting medium. All fitting parameters follow systematic energy trends, with the peak-to-peak distance and chemical freeze-out temperature exhibiting identical exponential saturation patterns, indicating that geometric expansion and thermal evolution share a common underlying dynamics governed by QCD phase structure. Furthermore, the approach naturally preserves limiting fragmentation behavior across all energies, in contrast to independent source models that suggest its violation at LHC energies. Although the theoretical basis requires further investigation, these empirical correlations successfully unify charged particle production across nearly two orders of magnitude in collision energy, revealing fundamental connections to underlying collision dynamics.

hep-ph↗
arXiv2026

Flavomons in Matter Gradients: Ray Tracing and Amplitude Evolution

Flavor instabilities develop in neutrino plasmas through the emission of flavomons, collective excitations of the flavor field. Matter gradients can sweep flavomons across the unstable wave-number range, halting their amplification. This effect is parametrically small for fast modes, where it was first identified, but we show that it strongly alters neutrino-mass-induced instabilities below the supernova shock, while leaving those outside largely unaffected. Dispersion relations alone, which are inevitably local, cannot determine the cumulative amplification of slow instabilities. Using the WKB approximation, we derive the covariant evolution equation for the flavomon amplitude along geometrical-optics rays, enabling a global treatment of instability growth and seeding.

hep-ph↗
arXiv2026

Entanglement of momenta at colliders

We investigate the entanglement in the momenta of particle pairs produced at colliders, discarding the information on their spin state. We show how to quantify this entanglement by reducing the continuous momentum variables to a two-qubit system through hemispherical projections. We discuss in detail the production of $τ$-lepton pairs at an electron collider. In the Standard Model, the discretized momenta are in a genuinely maximally entangled pure state for suitable choices of the hemispheres.

hep-ph↗