Speaker
Description
The formation channels of the first supermassive black holes (SMBHs) at z > 6 remain elusive, as current observations are dominated by the most luminous quasars, representing only the tip of the population. To uncover fainter quasars, gravitational lensing provides a uniquely effective natural telescope, extending the accessible luminosity range and offering essential probes of black hole seeds, host galaxy growth, and cosmic reionization.
The ESA Euclid mission will survey 14,000 deg² in optical and near-infrared bands and is expected to reveal ∼2000 strongly lensed quasars, including a rare yet valuable sample at z > 6. To meet this challenge, we present self-supervised learning techniques for discovering lensed quasars in Euclid data. Our framework leverages pretraining on simulated lenses and real sources (galaxies, stars, CCD artifacts) to learn robust image representations without extensive labels, improving completeness and reliability. Applied to the first 63 deg² of Euclid data, this method reduced an initial catalog of ∼1 million sources to ∼10,000 candidates. Preliminary inspection of the top 1,000 ranked systems revealed 40 potential dual point sources and one possible quadruple configuration.
With the forthcoming full data release, our approach will scale to the entire Euclid survey and enable the systematic discovery of lensed quasars at early epochs. This will open a unique window onto the faint end of the quasar luminosity function and the seeding and early growth of the first SMBHs.