Dark Matter and Machine Learning in Multi-Higgs Doublet Models

This thesis investigates extensions of the scalar sector of the Standard Model (SM), in particular models with N Higgs Doublets (NHDMs), applied to Dark Matter (DM) and new sources of CP violation. It combines the study of Multi-Higgs models with modern Machine Learning (ML) techniques for the exploration of high-dimensional parameter spaces. This approach is broadly applicable to any BSM scenario and offers a versatile path to uncover new physics with concrete experimental implications.

First, we study the Z2xZ2 Three Higgs Doublet Model (3HDM) with two stable inert scalars, leading to a multi-component DM framework. A rigorous analysis allows us to derive new conditions for the global minimum of the potential and identify regions where both DM candidates contribute comparably to the observed relic abundance. We impose all theoretical and experimental constraints and explore their interplay and complementarity across the full parameter space. This model displays a rich phenomenology inaccessible to minimal approaches, such as the Inert Doublet Model (IDM), thereby highlighting the predictive power of scalar sector extensions.

Next, we analyse the CP-violating Z2xZ2 C3HDM, which features large pseudoscalar Yukawa couplings that are tightly constrained and challenging to probe using conventional methods. To overcome this, we implement an ML-based scanning algorithm combining an Evolutionary Strategy with a Novelty Reward mechanism. This procedure accelerates the identification of new viable regions and uncovers clear predictions, particularly in the Higgs couplings to the top and bottom quarks.