Speaker
Description
Compton-thick AGN (CT AGN) are characterised by extreme dust obscuration, and limited visibility in X-ray wavelengths due to the X-ray radiation from the corona being reprocessed into near infrared wavelengths by dust and gas in the torus. New JWST surveys permit investigation of the properties of AGN and host galaxy populations in rest-frame optical between $0.5 < z < 4$; this spans the entire cosmic noon period where both SMBH growth and galaxy star formation rates peak. Prior Hubble Space Telescope observations of a sample of CT AGN at $z ∼ 2$ lack optical clues indicating the AGN. However, in new JWST imagery, we see plenty of these clues which allow us to identify a sample of CT AGN candidates at previously unexplored redshifts. We use a unique visual method independent of typical methods for identifying CT AGN. We use the deep-learning classifier ‘Zoobot’ to identify morphological AGN signatures in the CEERS and COSMOS-Web surveys. This is conducted via a transfer learning method that inherits prior training of Zoobot on ground-based imagery. Predictions of the JWST morphologies are then applied to the large multi-wavelength imaging set of COSMOS-Web, after being fine-tuned using Galaxy Zoo volunteer classifications of CEERS images. We compare our morphological method to other methods of detecting CT AGN, and test recent predictions for the quantity of CT AGN expected to exist at redshift $2 < z < 4$. We also discuss the viability of deep-learning classifiers to find more hidden AGN populations.
| Stream | Science |
|---|