Grant: Discovery Grants Program - Individual

While near-future astronomical telescopes will be able to detect fainter light sources (eg. stars, asteroids, galaxies), the images these telescopes collect will be significantly harder to analyze due to blending. When more sources are detected, the chance increases that the light from neighbouring sources overlaps, making the sources blended. Software must be used to deblend the sources, that is, identify neighbouring sources as separate, to accurately measure these sources' positions, brightnesses, and colours. Currently prevalent techniques struggle with deblending, and blending will be a limiting factor for astronomical data analyses with a variety of scientific goals. This research program develops probabilistic cataloguing, a technique that better deblends sources by treating deblending in a Bayesian statistical framework. My previous work has demonstrated that probabilistic cataloguing can deblend extremely crowded images with 1 source every 10 pixels, but its computational cost is a significant barrier to using it on large datasets. This proposal will apply probabilistic cataloguing to a search for binary Kuiper belt objects (KBOs) in images from ground-based telescopes. This search will use a modest amount of data, making it feasible to develop sufficient speed ups to the probabilistic cataloguing code to conduct the search. Probabilistic cataloguing can make full use of ground-based telescope images, complementing searches for binary KBOs using space-based telescopes, for which telescope time is more limited. Searching for binary KBOs will improve constraints on the Solar System's early dynamical history. This program will start by developing and comparing two sampling techniques, Nested Sampling and Hamiltonian Monte Carlo, to determine which is the most computationally efficient (objective 1). Then, the improved probabilistic cataloguing code will be applied to simulated images of binary KBOs and real images of known binary KBOs (objective 2). These tests will allow us to characterize the code's sensitivity to binary KBOs based on the binaries' separation, period, and inclination. After validating the code on known binaries, we will then apply the code to images of KBOs with unknown binarity to search for candidate binaries (objective 3). This research proposal will train undergraduate students to work at the intersection of statistics, computation, and astronomy. The binary KBO search will contribute to constraints on the Solar System's dynamical history, and the associated development will be a stepping stone towards enabling probabilistic cataloguing to be used in larger datasets in astronomy. Blending is just one example of a data analysis challenge that astronomy faces and developing methods like probabilistic cataloguing will allow astronomers to make the most of the data we collect.