Institute for Astronomy

Computational Astrophysics PhD Projects

Research projects on offer in our Simulations group:

A new cosmological residual distribution hydrodynamical solver

Sadegh Khochfar

Video: A new cosmological residual distribution hydrodynamical solver
A new cosmological residual distribution hydrodynamical solver

This PhD project proposal aims at developing a new beyond the-state-of-the-art hydrodynamical simulation code. The student will be expected to develop code to implement numerical schemes based on the residual distribution method for highly parallel compute architectures. Emphasis will be put on scalability to large number of compute nodes to allow for the next generation of cosmological simulations.

In current astrophysical simulations, two prevailing numerical methods are the Lagrangian based Smoothed Particle Hydrodynamics (SPH) and the Eulerian based structured-mesh hydrodynamics, often with adaptive mesh capabilities. SPH codes have exceptional adaptive spatial resolution capabilities, as the particles naturally sample the denser regions of the fluid flow. This comes at the cost, however, of poor shock resolution and the suppression of instabilities at contact discontinuities when compared to mesh-based codes. Cartesian mesh-based codes capture shocks and entropy mixing better, however they also suffer drawbacks in resolution (even if using adaptive mesh capabilities) and in poorly resolved bulk flows due to the lack of a Galilean-invariant formulation. As has been noted, many of the drawbacks of both SPH and cartesian mesh-based codes can be eliminated if the mesh is allowed to move with the fluid in an arbitrary Lagrange-Eulerian approach.

Much research has been done to identify a true multi-dimensional upwind scheme over the past few decades. One promising scheme is the residual distribution (RD) method. This method combines many of the advantages of a finite volume method with a genuinely multi-dimensional solution to the hydrodynamical equations. During the project the student will implement such a hydrodynamical solver and investigate its impact on physical processes in the early Universe.

Lensing and Clustering with the Rubin Observatory

Joe Zuntz

 

Combining weak lensing and galaxy clustering statistics yields one of the most powerful probes in cosmology for understanding the distribution of dark matter and the gravitational and cosmic processes that lead to that distribution. We measure and cross-correlate the two-point statistics of each and compare the results to increasingly rich theoretical models. In this project, you'll be part of the Vera Rubin Observatory, the largest upcoming ground-based photometric survey, which will build the largest dark matter maps and statistics over the course of a decade. It will measure the positions and shapes of 10 billion galaxies over the course of its mission You will work with the analysis pipeline that connects our measurements of these galaxies to our theories of physics and cosmology, becoming part of a large international team to do so. You'll help build the supercomputer analysis that will characterise and model Rubin data from core pixel modelling all the way to theoretical modelling, depending on your interests. This will be a heavily computational project, so it would be good to have a strong interest and some experience in that area.

Machine Learning Galaxy Formation

Sadegh Khochfar

In this project we want to develop new machine learning algorithms to investigate galaxy formation in the Universe. The algorithms will be applied to numerical simulations of galaxy formation in a first step and then used on observations. The idea is to break degeneracies between simulations and models used in the community and to identify the most important astrophysical drivers of galaxy formation. The student would work with data sets available in the group and from international collaborations.

Using Machine Learning to Study Galaxy Tidal Features with Next Generation Sky Surveys

Annette Ferguson and Bob Mann

Next generation sky surveys - notably the European Space Agency's Euclid mission and the "Legacy Survey of Space and Time" (LSST) to be conducted at the Vera C. Rubin Observatory in Chile - are poised to be transformative for studies of the low surface brightness Universe. Both will provide very deep optical and near-IR images across huge swathes of the night sky, enabling the first statistical analyses of the frequency and nature of faint tidal debris across the galaxy population. These features arise due to gravitational interactions between galaxies - for e.g. when a small galaxy gets tidally-shredded upon entering the dark matter halo of a larger system, or when two galaxies merge or have a close encounter - and are key signatures of an important, often dominant, mode of galaxy growth at both low and high redshift. Important constraints on models of galaxy assembly can, therefore, be derived from a better understanding of the prevalence and impact of these events, but the practical problem is how to detect galaxies with tidal features from the vast samples of galaxies to be observed in surveys like Euclid and the LSST. Most previous work has detected faint features around galaxies by eye, but that process will not scale to the Euclid and LSST data volumes, even if it is possible to enlist thousands of volunteers through a Citizen Science project. This PhD project will build on our group's previous work (e.g. Walmsley et al. 2019 MNRAS 483 2968) to apply machine learning to the problem of detecting faint tidal features around galaxies. We have found that Deep Learning algorithms, such as Convolutional Neural Networks (CNNs), shows promise but, like all supervised learning approaches, the utility is limited by the lack of training data. The project will investigate a variety of ways to circumvent this issue, as well as explore a variety of other approaches, such as transfer learning and self-training. The PhD student will join the Low Surface Brightness science working groups of both Euclid and Rubin, and will actively contribute to the development and testing of algorithms that can be applied to the first data in 2024. This is a potential CDT project and is suitable for a student who is independent, very adept at programming and statistics, and ideally already familiar with machine learning approaches to image classification.

Under the Galaxy Formation and Evolution projects see also: