Speaker
Description
The observed matter–antimatter asymmetry of the Universe motivates searches for new sources of CP violation beyond the Standard Model. The Higgs interaction with the top quark, the heaviest known elementary fermion, provides a particularly well-motivated place to search for such effects. A high-energy linear (e^+e^-) collider offers a clean and well-defined initial state, controllable beam polarisation, and the possibility of reconstructing the complete (t\bar tH) final state. In previous studies of (ZH) production, machine-learned CP observables were found to provide improved sensitivity compared with conventional angular observables. Motivated by this result, we investigate whether a similar strategy can exploit the more complex CP information in (t\bar tH), taking advantage of recent developments in event reconstruction and flavour tagging for the ILC.
We present an analysis of (e^+e^- \to t\bar tH) at (\sqrt{s}=550) GeV, using Physsim event generation including the near-threshold NRQCD enhancement and SGV fast detector simulation. The study focuses on semileptonic (t\bar tH) events with (H\to b\bar b). The implementation of Particle Transformer flavour tagging in the ILC reconstruction provides improved heavy-flavour identification as well as charge- and flavour-sensitive information for the reconstructed jets. This information is used both in jet assignment and in the construction of CP-sensitive observables, while also improving multivariate signal-background discrimination. Samples with fully polarised (e^-_Le^+_R) and (e^-_Re^+_L) initial states are studied separately and subsequently combined according to a realistic longitudinal-polarisation running scenario for a future linear collider.
We investigate complementary strategies for extracting the CP-odd scalar–pseudoscalar interference. For physically motivated angular observables, different choices of spin analysers, charge and flavour ordering, and reference-frame definitions are explored. Their generator- and reconstruction-level sensitivities are compared directly, exposing information losses associated with detector reconstruction and ambiguities in identifying the relevant decay products. In parallel, a machine-learned CP observable is constructed directly from reconstructed event information. We explore different choices of input features and event representations, ranging from the kinematics of a minimal set of reconstructed spin analysers to representations incorporating both W-daughter jets, charge-aware flavour information, assignment quality, kinematic-fit quantities, invariant masses, and the fitted neutrino. The resulting angular and machine-learned observables are compared using their sensitivity to the CP-odd interference.