SMILE: Discriminating milli-lens systems in a VLBI pilot project

Dark Matter (DM) remains poorly probed on critical, sub-galactic scales, where predictions from different models diverge in terms of abundance and density profiles of halos. Gravitational lens systems on milli-arcsecond scales (milli-lenses) are expected for a population of dense DM halos (free-floating or sub-halos) and free-floating supermassive black holes in the mass range of $10^6$ to $10^9\,M_\odot$. In this paper, we aim to look for milli-lens systems via a systematic search in a large sample of radio-loud AGN observed with very-long-baseline interferometry (VLBI). We present the observational strategy to discriminate milli-lenses from contaminant objects mimicking a milli-lens morphology. In a pilot project, we have investigated VLBI images from 13,828 sources from the Astrogeo VLBI image database and reduced the number of candidates to 40 in a first step. We present here the images and analysis of new sensitive follow-up observations with the EVN at 5 and 22 GHz and streamline our analysis to reject milli-lens candidates. By using constraints such as the surface brightness ratio, conservation of spectral shape, stability of flux ratios over time, and changes in morphology, we can confidently discriminate between milli-lenses and contaminant objects that mimick them. Using the above constraints, we rule out 31 out of our initial 40 candidates of milli-lens systems, demonstrating the power of our approach. Also, we found many new candidate compact symmetric objects, which are thought to be primarily short-lived jetted radio sources. This serves as a pathfinder for the final sample used for the Search for MIlli-LEnses (SMILE) project, which will allow us to constrain DM models by comparing the results to theoretical predictions. This SMILE sample will consist of $\sim$5,000 sources based on the VLA CLASS survey, including many observations obtained for this project specifically.