Philosophiae Doctor - PhD / Using two hydrodynamic galaxy formation simulations from the Mufasa project that I helped
develop, we aim to better understand the relationship between galaxy evolution and its cold gas
content commonly known as the neutral hydrogen or Hi. We first look at the environmental
properties of the simulated galaxies and compare to those that are available observationally. As
a proxy, we specifically quantify the so-called galactic conf ormity, which is the concordance
between the properties of galaxies neighbouring the primaries, in chapter 2. We show that the
Hi, the specific star formation rate (sSFR) and the colour of galaxies show galactic conformity
in qualitative agreement with previous observed data, i.e. the Hi-rich primary galaxies are surrounded
by Hi-richer galaxies than the Hi-poor primary galaxies, and similarly for the sSFR
and the colour. We find that environment, quantified by the number of neigbouring galaxies
within a fixed aperture, stellar age and molecular hydrogen (H2) also show conformity. Galactic
conformity also depends on the dark matter halo mass of the primary galaxy. The galactic
conformity signal from the primaries of smaller haloes is weak but extends out to several virial
radii of those structures, whereas the signal is very strong for high mass haloes but lowers
quickly with distances from the primaries. We also find the galactic conformity only emerges
in the later half of cosmic evolution. We next quantify the gas content and star formation
depletion timescales in chapter 3. We use two carefully chosen groups of simulated galaxies
and find that timescales are affected by both the mass of the virialised structure of the first
infall and the galaxy stellar mass at infall: the higher the halo mass or the stellar mass the
shorter the timescale. The gas or Hi depletion timescale is concordant to that of the star formation
quenching, indicative of direct decrease of SFR due to depletion of the extended cold
gas reservoir. The neutral atomic or molecular hydrogen consumption timescale depends on
the Hubble time. Galaxies tend to form stars more efficiently at lower redshift. While the halo
mass of infall affects the consumption timescale of the Hi, it does not correlate with the H2. We
lastly develop machine learning tools to use galaxy photometric data to predict a galaxy’s Hi
mass in chapter 4, to allow predictions for Hi from much larger optical photometric surveys.
The training and testing of the algorithms are done first with the simulated data from Mufasa.
We show that our model performs better than previously done with ad hoc data fitting
approaches. Random Forest (RF) followed by the Deep Neural Networks (DNN) perform best
among the explored machine learning techniques. Extending the trained models to observed
data, namely the Arecibo Legacy Fast ALFA (ALFALFA) and REsolved Spectroscopy Of a Local
VolumE (RESOLVE) survey data, we show the overall performance is slightly reduced relative
to the simulated testing set owing to the small inconsistency between definition of galaxy properties
between simulation and observational data, and DNN perfoms the best in this case. The
application of our methods is useful for galaxy-by-galaxy predictions and anticipated to correct
for incompletness in the upcoming Hi deep surveys done with MeerKAT and eventually
the Square Kilometre Array (SKA).
| Identifer | oai:union.ndltd.org:netd.ac.za/oai:union.ndltd.org:uwc/oai:etd.uwc.ac.za:11394/6581 |
| Date | January 2018 |
| Creators | Rafieferantsoa, Mika Harisetry |
| Contributors | Dave, Romeel |
| Publisher | University of the Western Cape |
| Source Sets | South African National ETD Portal |
| Language | English |
| Detected Language | English |
| Rights | University of the Western Cape |
Page generated in 0.002 seconds