# When Spectral Modeling Meets Convolutional Networks: A Method for Discovering Reionization-era Lensed Quasars in Multi-band Imaging Data

**Authors:** Irham Taufik Andika, Knud Jahnke, Arjen van der Wel, Eduardo Bañados, Sarah E. I. Bosman, Frederick B. Davies, Anna-Christina Eilers, Anton Timur Jaelani, Chiara Mazzucchelli, Masafusa Onoue, Jan-Torge Schindler  
**arXiv:** [2211.14543](https://arxiv.org/abs/2211.14543)  
**Date:** 2022-11-26  
**Summary page:** https://www.summarizepaper.com/en/paper/2211.14543v1/

## Key points

- Strong-gravitational lenses are difficult to identify, especially reionization-era lensed quasars.
- A new approach was explored by enlarging the permitted spectral parameter space and introducing a new spatial geometry veto criterion implemented via image-based deep learning.
- The first application of this approach involved a systematic search for reionization-era lensed quasars using data from the Dark Energy Survey, the Visible and Infrared Survey Telescope for Astronomy Hemisphere Survey, and the Wide-field Infrared Survey Explorer.
- The search method consisted of two main parts: pre-selection of candidates based on their spectral energy distributions (SEDs) using catalog-level photometry and relative probabilities calculation of being a lens or some contaminant utilizing a convolutional neural network (CNN) classification.
- Deflected point-source lights were painted over actual galaxy images to generate realistic galaxy-quasar lens models optimized to find systems with small image separations, i.e., Einstein radii of $\theta_\mathrm{E} \leq 1$ arcsec.
- Visual inspection was then performed for sources with CNN scores of $P_\mathrm{lens} > 0.1$, which led to the identification of 36 newly selected lens candidates awaiting spectroscopic confirmation.
- Automated SED modeling and deep learning pipelines supported by modest human input are promising routes for detecting strong lenses from large catalogs that can overcome the limitations of primarily dropout-based SED selection approaches.
- By expanding the permitted spectral parameter space and introducing new spatial geometry veto criteria through image-based deep learning techniques, researchers may be able to identify more reionization-era lensed quasars in multi-band imaging data.

## Summary

The search for strong-gravitational lenses has been a challenge in the field of astrophysics, particularly in identifying reionization-era lensed quasars. While around three hundred quasars have been discovered at $z\gtrsim6$ over the last two decades, only one has been identified as being strongly lensed. To address this issue, a new approach was explored by enlarging the permitted spectral parameter space and introducing a new spatial geometry veto criterion implemented via image-based deep learning. The first application of this approach involved a systematic search for reionization-era lensed quasars using data from the Dark Energy Survey, the Visible and Infrared Survey Telescope for Astronomy Hemisphere Survey, and the Wide-field Infrared Survey Explorer. The search method consisted of two main parts: pre-selection of candidates based on their spectral energy distributions (SEDs) using catalog-level photometry and relative probabilities calculation of being a lens or some contaminant utilizing a convolutional neural network (CNN) classification. To construct training datasets, deflected point-source lights were painted over actual galaxy images to generate realistic galaxy-quasar lens models optimized to find systems with small image separations, i.e., Einstein radii of $\theta_\mathrm{E} \leq 1$ arcsec. Visual inspection was then performed for sources with CNN scores of $P_\mathrm{lens} > 0.1$, which led to the identification of 36 newly selected lens candidates awaiting spectroscopic confirmation. These findings demonstrate that automated SED modeling and deep learning pipelines supported by modest human input are promising routes for detecting strong lenses from large catalogs that can overcome the limitations of primarily dropout-based SED selection approaches. By expanding the permitted spectral parameter space and introducing new spatial geometry veto criteria through image-based deep learning techniques, researchers may be able to identify more reionization-era lensed quasars in multi-band imaging data.