Predicting the Kinematics of the Cold Circumgalactic Medium from its Morphology using Convolutional Neural Networks
Connor Jennings, Earl P. Bellinger, Imad Pasha, Pieter van Dokkum, Pratik J. Gandhi
astro-ph.GA
Submitted: 2026-08-04
Comments: 18 pages, 8 figures
Code: https://github.com/qubvel/segmentation
License: http://creativecommons.org/licenses/by/4.0/
The gist: We present a novel approach to predicting plane-of-sky velocities of cold gas clouds in the circumgalactic medium (CGM) of galaxies.
Terminology
Abstract
We present a novel approach to predicting plane-of-sky velocities of cold gas clouds in the circumgalactic medium (CGM) of galaxies. The method uses a convolutional neural network (CNN) trained on simulated emission maps derived from the TNG50 cosmological simulation, with forward modeled noise properties consistent with upcoming observational facilities. Using 182 Milky Way/Andromeda analog galaxies, we generate emission maps in H alpha using Cloudy models, as well as line-of-sight averaged 2D velocity maps. Using a UNet architecture, we train the CNN to take emission maps as input and return plane-of-sky velocity maps as output, which cannot be observationally constrained using traditional methods. Qualitatively, the model is generally able to infer the true overall flow direction. We quantify the effects of Gaussian noise on the network's training and predictive power. At depths expected to be probed by forthcoming telescopes such as MOTHRA, the network has a typical RMS error for the plane-of-sky velocity direction of 0.3-0.5 v vir. This implies that 2D emission maps of sufficient depths will be able to estimate two additional phase space dimensions of cold CGM gas, enabling targeted followup and a better understanding of overall CGM flows.
Sources
- Deeply Uncertain: Comparing Methods of Uncertainty Quantification in Deep Learning Algorithms
- Tracing the Cosmic Evolution of the Cool Circumgalactic Medium of Luminous Red Galaxies with DESI Year 1 Data
- Saas-Fee Lecture Notes: Physics of Lyman Alpha Radiative Transfer
- The 2017 Release of Cloudy
- Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning
- Evaluating Scalable Bayesian Deep Learning Methods for Robust Computer Vision
- Deep Galaxy: Classification of Galaxies based on Deep Convolutional Neural Networks
- Adam: A Method for Stochastic Optimization
- PyTorch: An Imperative Style, High-Performance Deep Learning Library
- Zooming in on the circumgalactic medium with GIBLE: Cloud-scale simulations with cosmological initial conditions
- U-Net: Convolutional Networks for Biomedical Image Segmentation
- Notes on the Behavior of MC Dropout
Related papers
- Apparent Stability in Self-Gravitating Turbulence and the Evolution of Molecular Clouds
- Two sets of potential-density basis pairs for the study of radial perturbations in collisionless spherical stellar systems
- Constraining reionization-era Ly alpha escape with JELS-MUSE: a highly complete H alpha-selected sample at z about6.1
- Deriving volume density profiles of filaments from observed surface densities
- Little Red Dots and Supermassive Black Hole Seed Formation in Ultralight Dark Matter Halos
- MEGATRON: how the first stars can create an iron metallicity plateau in the smallest dwarf galaxies