Banik: Pushing the frontiers of gravitational encounters and collisionless dynamics / Ghosh: Estimating Galaxy Morphological Parameters for ~8 Million Galaxies in the Hyper Suprime-Cam Wide Survey using Bayesian Machine Learning

Oct 31, 2022 - 11:00 am to 12:00 pm

Campus, PAB 102/103 *please note change in location*

Uddipan Banik / Aritra Ghosh (Yale) via zoom

Zoom info:

Banik: The long range nature of gravity complicates the dynamics of self-gravitating many-body systems such as galaxies and dark matter (DM) halos. Relaxation/equilibration of perturbed galaxies and cold dark matter halos is typically a collective, collisionless process. Depending on the perturbation timescale, this process can be impulsive/fast, adiabatic/slow or resonant. First, I shall present a linear perturbative formalism to compute the response (in all three regimes) of disk galaxies to external perturbations such as satellite impacts. I shall elucidate how phase-mixing of the disk response gives rise to phase-space spirals akin to those observed by Gaia in the Milky Way disk, and how these features can be used to constrain the Milky Way’s potential as well as its dynamical history. Next, I shall discuss the secular evolution of a massive perturber due to the back reaction of the near-resonant response of the host galaxy/halo. In this context I shall present two novel techniques to model the secular torque (dynamical friction) experienced by the perturber: 1. a self-consistent, time-dependent, perturbative treatment and 2. a non-perturbative orbit-based framework. These two approaches explain the origin of certain secular phenomena observed in N-body simulations of cored galaxies but unexplained in the standard Chandrasekhar and LBK theories of dynamical friction, namely core-stalling and dynamical buoyancy. I shall briefly discuss some astrophysical implications of these phenomena: potential choking of supermassive black hole mergers in cored galaxies, and the possibility of constraining the inner density profile (core vs cusp) of DM dominated dwarf galaxies and therefore the DM particle nature.

Ghosh: In this talk, I will introduce GaMPEN -- The Galaxy Morphology Posterior Estimation Network. GaMPEN is the only ML framework that can predict full Bayesian posteriors for morphological parameters – it produces 40% more accurate uncertainties at 1/100th of the run-time compared to light-profile fitting algorithms. GaMPEN also contains a Spatial Transformer Network (STN) that automatically crops input galaxy frames to an optimal size before determining their morphology -- this will be crucial in applying GaMPEN to new survey data with no pre-determined radius measurements. Using GaMPEN, we have created one of the largest morphological catalogs currently available, containing ∼ 8 million Hyper Suprime-Cam (HSC) galaxies with estimates of bulge-to-total light ratio, effective radius, flux, and their associated uncertainties. Using a novel technique of first training on simulations and then transfer-learning on real data, we have been able to train GaMPEN with < 1% of our dataset. This catalog has an order of magnitude more galaxies and is three magnitudes deeper compared to current state-of-the-art bulge/disk decomposition catalogs. Currently, we are using this catalog to probe the relationship of morphology with environment and the buildup of mass, shutting down of star-formation in these galaxies.