| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 14.17 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Understanding how galaxies and their constituents, like active galactic nuclei (AGN), evolve and interact across cosmic timescales remains a key challenge in astrophysics, especially during the epoch of reionisation (EoR), when the early Universe transitioned from a neutral to an ionised state. Even though star formation (SF) is considered to be its main contributor, the impact of AGN remains elusive. Recently, huge efforts have been put into the determination of the AGN bolometric radiation output via their search using a multi-wavelength approach.
To address such challenge, this thesis presents a novel machine learning (ML) tool –a pipeline of models– for the efficient selection and redshift characterisation of radio-detectable AGN by using multi-wavelength photometry of sources detected in the infrared (IR). By analysing sources in a wide range of redshift values, this tool enables the exploration of theAGN-galaxy co-evolution across cosmic times. Applied to millions of sources in the Hobby-Eberly Telescope Dark Energy Experiment (HETDEX) Spring and the Stripe 82 (S82) fields, our pipeline has identified almost 100 thousand radio-AGN candidates, with predicted redshift values up to 4.4.
Beyond classification, we can extract the key parameters that most significantly impact candidate selection in our pipeline. This investigation has led to the design of a new AGN colour-colour selection criterion, offering a useful, ML-based, tool for the community.
Furthermore, by extracting radio-AGN candidates from the Evolutionary Map of the Universe Pilot Survey (EMU-PS), we can generate radio luminosity functions (RLFs) with highly constrained uncertainties. Our results are compatible with current knowledge and hint at the existence of a distinct population of bright sources.
Finally, this thesis explores the potential application of our tool to future surveys like the Square Kilometre Array (SKA), that is expected to generate immense radio datasets. Our rapid and reliable method will be instrumental in separating AGN from star-forming galaxies (SFGs) while helping to unveil their interplay across the history of the Universe.
Description
Keywords
Active Galactic Nuclei Radio Galaxies Galaxy classification Redshift Determination Machine Learning Núcleo Galáctico Ativo Radio Galáxias Classificação de Galáxias Determinação do Desvio para o Vermelho Aprendizagem Automática