Abstract :
Determining the exact nature of dark matter remains a fundamental challenge in modern cosmology. While the standard Cold Dark Matter paradigm successfully describes the large-scale structure of the universe, it faces persistent tensions at sub-galactic scales. Alternative dark matter models (e.g., warm, self-interacting, or fuzzy dark matter) predict substantially different abundances of low-mass halos (below 108 M⊙). Because these low-mass halos lack star formation and are dark matter dominated, they can only be detected via their gravitational influence. Strong gravitational lensing at milliarcsecond scales, known as milli-lensing, provides a direct method to detect these invisible masses. Very Long Baseline Interferometry (VLBI) delivers the extreme angular resolution required to find these systems by combining signals from a global network of radio telescopes to create a single, Earth-sized interferometer. The Search for MIlli-LEnses (SMILE) project aims to utilize this technique for observations of thousands of radio sources to discriminate between competing dark matter models. However, carrying out such a massive radio survey is hindered by a severe technical bottleneck: VLBI data calibration. Traditional calibration is a highly manual, time-intensive process requiring expert supervision, making it impossible to scale to thousands of heterogeneous archival observations.
This doctoral thesis is divided into two parts:
The first part of this work introduces VIPCALs, a novel, fully automated VLBI calibration pipeline. By eliminating the need for human intervention and prior source knowledge, VIPCALs enables the rapid, autonomous processing of raw archival data. This software not only makes the full-scale SMILE survey feasible but also provides a highly accessible, scalable data reduction tool for the broader radio astronomy community.
The second part makes use of VIPCALs to address the primary observational challenge of the SMILE pilot program: distinguishing genuine milli-lenses from potential contaminants, especially compact symmetric objects (CSOs). CSOs are young or frustrated active galactic nuclei whose double-lobed morphologies can perfectly mimic lensed images. By calibrating and analyzing an extensive multi-epoch, multi-frequency dataset of nine promising pilot candidates, I was able to characterize their physical properties and their nature.
Ultimately, this thesis determines the true physical nature of the pilot candidates while establishing a rigorous, multi-wavelength classification framework. By delivering both the automated software infrastructure and the observational methodology required to separate authentic milli-lenses from intrinsic astrophysical contaminants, this work lays the groundwork for the upcoming main SMILE survey.
