Understanding Black Hole Growth Across Cosmic Time
Understanding Black Hole Growth Across Cosmic Time
How do supermassive black holes grow?
My research investigates this question through active galaxies caught in unusual stages of accretion. I combine astronomical observations, gravitational lensing, and computational methods to identify and characterize systems that conventional approaches can miss.
My work spans observations of quasars in the early Universe, strong lens discovery and modeling, and machine learning for modern astronomical surveys. Together, these approaches allow me to study black hole activity across a wide range of luminosities, environments, and cosmic epochs.
Rare AGNs and Accretion
Some of the most informative active galaxies are also among the hardest to find. Faint AGNs probe populations below traditional luminosity thresholds, while rapidly changing systems provide direct evidence that accretion can transition between states on observable timescales.
This direction began with high-redshift quasars, including PSO J083+11, a weak-line quasar at z = 6.34 whose small proximity zone points to a short active lifetime. I have since expanded it toward fainter AGN populations and the time domain, to measure the frequency, duration, and contribution of different accretion states.
Artist’s impression of a quasar in the early Universe. Illustration credit: Robin Dienel, courtesy of the Carnegie Institution for Science.
Selected publications: Andika et al. (2020) · Andika et al. (2022) · Andika et al. (2024)
Gravitational Lensing
Strong gravitational lensing magnifies distant sources that would otherwise be difficult to observe. These systems are rare, making their discovery in large astronomical surveys a natural application for machine learning.
I develop methods for both lens discovery and physical inference. These efforts have used convolutional neural networks and vision transformers to search for strongly lensed quasars and to reconstruct lens systems and infer their physical parameters using a physics-informed generative model.
Examples of simulated strong lenses and non-lens sources used to train the lens-discovery models.
Selected publications: Andika et al. (2023) · Andika et al. (2025)
Multimodal Learning for Astronomical Surveys
At LMU Munich, I contribute to the Euclid mission, developing scientific software and scalable analysis methods for its large and heterogeneous datasets.
A major direction of my current research is multimodal representation learning, combining imaging, photometry, catalogs, and spectroscopy within a common framework. These representations can support AGN classification, strong lens discovery, redshift inference, and the retrieval of unusual sources without requiring an entirely separate model for each task.
Publications
A complete list of my refereed publications is available through the Astrophysics Data System (ADS).