Using Supervised Learning on maps of total column density to investigate the fractal structure of both the Milky Way and the Andromeda Galaxy.

Bates, Matthew 2023. Using Supervised Learning on maps of total column density to investigate the fractal structure of both the Milky Way and the Andromeda Galaxy. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (16MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (3MB)

Abstract

In this thesis, we design, train, and test multiple Convolutional Neural Networks (CNNs) that can take images of surface-density as inputs and estimate the Hurst parameter, $\HHH$ and, the scaling exponent, $\SSS$ of each image. $\HHH$ is a measure of the power spectrum while $\SSS$ is a measure of the range of the surface-density. The CNNs are trained using exponentiated fractional Brownian motion (xfBm) fields that measure $\npix\times\npix$ pixels in size. We produce three CNNs, one for each $\npix=128,64,\text{and }32$. We find that all three models estimate $\HHH$ more accurately than Delta-Variance. We use these CNNs to analyse Hi-GAL tiles, using a sliding patch, in order to produce maps of $\HHH$ and $\SSS$. Each Hi-GAL tile has been processed using \ppmap to produce maps of total column-density. We find that the size and resolution of the patches, (i.e. the range of scales that the patch captures) affects the resultant statistics. We find evidence of bimodality between sightlines with higher surface density ($\Sigma\gtrsim32\;\mpc$) which correlates with a high $\HHH$ ($\gtrsim0.8$) and $\SSS$ ($\gtrsim1$), and sightlines with lower surface-density ($\Sigma\lesssim32\;\mpc$), which correlate with lower $\HHH$ ($\lesssim0.8$) and $\SSS$ ($\lesssim1$). We find that this bimodality is not present when analysing maps of temperature-differential column density. We find that when $\Sigma\gtrsim32\;\mpc$ the surface densities follow a power law distribution $dP/d\Sigma\propto\Sigma^{-3}$. We also apply these CNNs to \textit{Herschel} maps of the Andromeda Galaxy (M31) that have been processed via \ppmap. We find that the distribution of $\HHH$ for Andromeda has a peak at $\HHH=0.67$, which agrees with the literature.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Physics and Astronomy
Subjects:	Q Science > QC Physics
Uncontrolled Keywords:	Thesis, Dissertation, Degree, Star Formation, ISM, Galaxies, Machine Learning.
Funders:	STFC
Date of First Compliant Deposit:	20 January 2025
Last Modified:	20 Jan 2025 16:01
URI:	https://orca.cardiff.ac.uk/id/eprint/175408

Actions (repository staff only)

Edit Item