Master Thesis defense by Gregers Nordgaard Roland

Title: Learning instead of cutting: Autoencoder-assisted approach to sterile neutrino searches at FCC-ee

Abstract:

This thesis explores the search sensitivity to feebly interacting heavy neutral leptons (HNLs) - nearly sterile gauge singlets that interact only through their small mixing with the standard model neutrinos. The HNLs are introduced to naturally give mass to the light neutrinos, through the introduction of the Seesaw model and the addition of three right-handed neutrinos to the standard model. This is done within the framework of the neutrino minimal standard model 𝜈MSM, which predicts HNLs mass to lie within the range from 150 MeV to 100 GeV. Most of this range can be explored at the high luminosity Z-pole run proposed at the FCC-ee since copious amounts of Z-bosons will be produced, which can decay into HNLs with masses 𝑚𝑁 < 𝑚𝑍. Three studies, using various techniques, are used to find the search sensitivity to the signature of these particles. An analysis of the sensitivity to long-lived HNLs, which decay with a measurable displacement from the interaction point, and two analyses of promptly decaying short-lived HNLs. For low mass points the promptly decaying HNLs give rise to a monojet signature, which can be distinguished from backgrounds using a cut-based analysis.

For higher mass points, the signature is more complicated, and a variety of machine learning models are used to distinguish these signatures. Most notably an autoencoder trained on a simulated standard model background which is used to search for anomalous signatures. This allows the differential selection of the rare HNL events from the background sample. The three searches combine to cover a large region of the phase space within the predicted electro-weak mass range. The search for displaced decays is projected to reach a sensitivity of |𝑈𝜏|2 = 5.14 × 10−12 for the planned L = 150 ab−1 run at Z-pole, and the two prompt searches combine to cover a large part of phase-space where the mass and mixing angle of the HNL is too large for a displaced search.