Simulations of Galaxy Clusters

UoM administered thesis: Phd

  • Authors:
  • Thomas Armitage


Galaxy clusters are the most massive virialised objects in the Universe and powerful cosmological probes. With impending surveys such as Euclid expecting to identify of order 10^4 galaxy clusters it is important that the systematics of cluster mass estimation are understood in order to make full use of the data. In this work we use the C-EAGLE and MACSIS suites of galaxy cluster simulations to explore and test the systematic limitations of dynamical mass estimates. In the first two chapters we provide an overview of modern cosmology, galaxy clusters and the techniques used in cosmological simulations. We also describe the specifics of the C-EAGLE and MACSIS simulations. In Chapter 3 we examine the velocity dispersion - mass scaling relation. It is well known that the DM particle velocity dispersion scales tightly with cluster mass, though there is disagreement in the literature as to whether the galaxies can be used as fair tracers of the DM velocity dispersion in the cluster. We show that selecting galaxies by their total mass, rather than stellar mass causes the measured velocity dispersion to be biased high by ~10 per cent relative to the dark matter dispersion. Selecting galaxies by their stellar mass results in no significant bias. In Chapter 4 we then compare three different dynamical mass estimators; the caustic, Jeans and virial methods. We show that all three suffer from significant scatter ~0.09 - 0.15 dex, but are on average unbiased. We also find a weak correlation between the amount of substructure in a cluster and its mass bias. Finally in Chapter 5 we present a novel way to estimate the mass of clusters via the application of machine learning techniques. We show that training a ridge regression model on a photometric dataset can reduce the scatter by up to a factor of three compared to the methods used in Chapter 4. We find that photometric galaxy data performs just as well as spectroscopic data in many cases. In conclusion we find that machine learning techniques are a very promising way to obtain observationally cheap but accurate mass estimates of clusters, and a viable alternative to current scaling relations.


Original languageEnglish
Awarding Institution
  • Richard Battye (Supervisor)
  • Scott Kay (Supervisor)
Award date1 Aug 2019