Learning and interpreting the galaxy-halo connection in cosmic simulations

  • Harry George Chittenden

Student thesis: Doctoral Thesis (PhD)


In modern galactic astronomy, cosmological simulations and observational galaxy surveys work hand in hand, offering valuable insights into the historical evolution of galaxies on both cosmological scales and an individual basis. As dark matter halos constitute a significant portion of the mass in galaxies, clusters, and cosmic structures, they profoundly impact the properties of galaxies. This relationship is known as the galaxy-halo connection.

Galaxies possess a complex nature necessitating computationally intensive modelling. Accurately and consistently modelling galaxy-halo coevolution across all scales thus presents a challenge, and compromises are usually made between simulation size and resolution. However, it is possible to conduct pure dark matter simulations on larger scales, requiring a fraction of the power of complete simulations. As observational surveys expand in size and detail, however, simulations of this magnitude become crucial in supporting their findings, surpassing the limitations of galaxy simulations.

In this thesis, I present a machine learning model which encodes the galaxy-halo connection within a cosmohydrodynamical simulation. This model predicts the star formation and metallicity of galaxies over time, from properties of their halos and cosmic environment. These predictions are used to emulate observational data using spectral synthesis models, and subsequently the model is applied to a large dark matter simulation.

Through these predictions, the model replicates the correlations responsible for galaxy evolution, as well as observable quantities reflecting this galaxy-halo connection, with similar results in dark matter simulations. The model computes accurate galaxy-halo statistics and reveals important physical relationships; specifically, variables associated with halo accretion influence a galaxy's mass and star formation, while environmental variables are linked to its metallicity. While the predictions from dark matter simulations are reasonably accurate, they are affected by the absence of baryonic processes, the resolution of the simulation, and the calculation of halo properties.
Date of Award29 Nov 2023
Original languageEnglish
Awarding Institution
  • University of St Andrews
SupervisorRita Tojeiro (Supervisor)


  • Galaxies
  • Astrophysics
  • Cosmology
  • Dark matter
  • Dark matter halos
  • Machine learning
  • Galaxy evolution
  • Cosmological simulations
  • Artificial intelligence

Access Status

  • Full text open

Cite this