Tese Mestrado

Exploring jet substructure in public LHC data

Alexandre Maria Neves de Barbosa Mendonça

Quinta-feira, 26 de Novembro 2026 das 09:00 às 11:00
Sala P3 (Piso 1 do Pavilhão de Matemática) do IST

Collinear spin correlations, induced by the linear polarisation of intermediate gluons in the parton shower, produce an azimuthal modulation between successive splittings inside jets. This work studies them using CMS Open Data datasets, created from proton-proton collisions at √ s = 13 TeV recorded in 2016 and the corresponding Pythia 8 simulation. A reproducible pipeline was built in order to observe the modulation of the angle ∆ψ12 between the planes of the two hardest splittings in the Lund declustering of jets with pT > 700 GeV, and the azimuthal modulation of the three-point energy correlator. Given that the g → qq¯ and g → gg channels contribute with opposite signs, a two-stage neural-network selection was developed to enrich the samples in g → qq¯ splittings, which were labelled in simulation using the interleaved flavour neutralisation algorithm.

Before any selection was done, the value ⟨cos 2∆ψ12⟩ = −0.078±0.003 was obtained in data, in agreement with the simulation. The implemented selection strengthens the modulation from −0.12 to −0.27 in simulation as the purity rises from 0.18 to about one half, and the data follow the simulation wherever their precision allows. The decomposition by channel shows, however, that every class carries a large non-spin component and that the selection shifts the modulation of each class, so the selected modulation cannot be interpreted as that of the g → qq¯ channel alone.

The energy correlator shows a small positive modulation, nearly insensitive to the channel. These results are consistent with the recent CMS observation, but establishing the spin correlations requires a prediction without them, processed through the same selection.