Λ_b Baryon Polarisation via the Semileptonic Decays: A Machine Learning Approach

The polarisation of the Λ_b baryon offers a clean window onto b-quark spin, since its spin-singlet light diquark lets the baryon largely retain the polarisation of the parent b quark. Measuring it probes the non-perturbative polarization retention fractions of QCD hadronisation and might provide sensitivity to physics beyond the Standard Model. The polarisation can be extracted from the angular distributions of the semileptonic decays Λ_b → X_c μ⁻ ν̄_μ, where the unobserved neutrino must be reconstructed — a task complicated by a two-fold kinematic ambiguity, untracked neutral hadrons, and uncertainty in vertex positions. To mitigate these limitations, we train machine-learning models on pp collisions at √s = 14 TeV simulated with Pythia 8, comparing a ResNet baseline built on global event features with a Transformer that additionally reads jet-constituent information. The models reach good sensitivity to the b-quark polarisation while needing only a few times more events than the ideal case of perfect event information. The Transformer performs best, and we find the longitudinal component to be robust to vertex-position uncertainty while the transverse components are strongly limited by it.