A Systematic Census of AGN Variability

California Institute Of Technology, Pasadena CA

Investigators

Abstract

Astronomical surveys are monitoring the night sky and looking for changes in the brightness of stars, galaxies, and asteroids. These surveys produce vast amounts of complex data, so machine learning techniques are being developed to analyze modern data sets. This team will use machine learning and time series techniques to identify active galactic nuclei and catalog the different ways in which their brightness can vary. Future sky surveys can draw on these results to sort out the different models that are developed to explain the variability. This work forms the basis for real-world student projects in data science summer schools. More broadly, the machine learning techniques could be applied to similar collections of time series in other fields, such as neuroscience, seismology, and climate science. Active galactic nuclei (AGN) form one of the largest populations of variable astronomical sources and play a key role in our understanding of accretion physics, relativistic physics, galaxy evolution, and large-scale structure. However, their variability is not simple and remains poorly studied in comparison to other more regular sources, partly due to a lack of appropriate statistical tools and methodologies. A new generation of sky surveys is enabling systematic studies of astrophysical variability, and more sophisticated analyses are now possible through machine learning that can reveal details of the nonlinear nature of the variability. The team will conduct a statistical study of AGN variability employing the Catalina Real-time Transient Survey and Zwicky Transient Facility data sets. These surveys have produced time series data for millions of known and potential AGN spanning almost twenty years with hundreds of observations per source. The team will construct a highly accurate and complete catalog of AGN based on continuous time autoencoders applied to multiband multiepoch photometry, thus combining both color and variability information. Using this, they will systematically identify AGN flares at optical and mid-IR wavelengths and define flare families based on characterizable properties. They will also use deep learning models to provide novel non-parametric discriminating features in the data set. This will establish the knowledge base and methodologies for AGN studies with LIGO and Rubin Observatory. Data products from the project will also form the basis for student projects in various data science and astronomy summer schools. This award addresses/advances the goals of the Windows on the Universe Big Idea. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.