DISCnet PhD Studentships
Our 4 year PhD in Data Intensive Science programme offers a unique opportunity to develop advanced knowledge of data intensive methods. The programme combines specialist training in data intensive science, transferable skills development and a 6 month placement.
Optimising Searches for Clusters of Galaxies using Photometric Surveys
Searches for clusters of galaxies in astronomical survey data make use both of modern data mining techniques (such as tree codes) and of astrophysical principles (such as the fact that the constituent galaxies were formed at roughly the same time and the fact that the temperature of the gas that permeates the space between the galaxies is related to the quantity of dark matter).
Simulations of galaxy formation at high redshift
Hydrodynamical simulations of galaxy formation are now beginning to resolve the formation and evolution of galaxies within the correct cosmological context. The results are crucial for interpreting observations from the next generation of cosmological surveys. The student will undertake and analyse such simulations as part of big international collaborations such as Bluetides and Virgo with a particular focus on high redshift galaxies.
Automating galaxy deblending with machine learning
Reliable deblending of merged galaxy images, without fragmenting large spiral galaxies into several pieces, is a challenging and unsolved problem in astronomy. A reliable automated solution is essential for upcoming large imaging surveys such as those to be performed by the Large Synoptic Survey Telescope (LSST) and Euclid. We propose to develop a new software package that will make reliable flux measurements from resolved galaxy images, including those with any combination of irregular structure, overlapping neighbours, and stellar contaminants. We will simultaneously model all overlapping images and sky background, employing Bayesian priors to minimize parameter degeneracies and to avoid unphysical models. Machine learning techniques will be used to probabalisticly allocate the flux from each
separate component to host galaxy or interloper.
Modelling and signal interpretation for the SKA reionization experiment
Within this project the students will be running and analysing large, massively-parallel numerical simulations of the reionization process which occurred during the first billion years of the evolution of the universe. The results from these will be used to create large libraries of models for interpretation of observations from the Square Kilometre Array (SKA) through model fitting using methods like MCMC.
Lensing reconstruction and delensing with forthcoming ground-based data.
Ground-based CMB observations are entering their third and fourth stage, and can map the CMB and its polarization at high sensitivity and resolution. We have recently joined the Simons Observatory which will be the world-leading experiment within the next few years. We will prepare for and analyse this new data for signals of CMB lensing, running estimators on the data sets and calibrating various biases from simulations. We will also work on delensing estimators, to extract the best constraints on primordial gravitational waves from the B-mode polarization. You will use multigrid conjugate gradient techniques for optimal filtering of complex data, and investigate maximum likelihood and sampling based approaches to extracting the signal.
Weak lensing maps for Euclid
We are entering a golden age for cosmological research of our Universe, with state-of-the-art surveys not only providing detailed 3D maps of the galaxy distribution, but also maps of the total matter distribution, through the method of gravitational weak lensing. Sussex is a member of the next generation ESA Euclid mission, and this survey will rely heavily on simulated mock data to make constraints on the underlying physics of our Universe. At Sussex and with our partners in Europe, we are leading work on generation of such synthetic data sets. One of the required tasks for this mission is the generation of full-sky weak lensing maps of the Euclid survey. This would provide critical model input to a wide variety of projects, and not only for Euclid. This would be the frame of a cross disciplinary data intensive PhD. The student would learn to make use of sophisticated parallel programming techniques and data mining tools, to facilitate Euclid science. These tools are also widely used in many areas of the technology industry and so the advancement of them may help lead to breakthroughs in other areas.
Probabilistic models of galaxy formation
Multi-wavelength surveys of the sky to probe galaxy evolution are providing vast data sets with enormous richness with multi-modal data including photometric, morphological and spectral information. These large and complex data sets require correspondingly sophisticated new data analysis techniques. Inspired by the challenges of interpreting Herschel data (in which we have played a leading role) at Sussex we are developing probabilistic models of galaxy formation. This programme could redefine how we model galaxy evolution and formation. These methods take the state-of-the-art Bayesian methodologies and constrain complex, easily extendable, models using all available information, properly quantifying uncertainties rigorously acknowledging the required prior assumptions. These methods will be applied to new STFC related data sets e.g. LOFAR and other SKA pathfinders, LSST, JWST, ALMA, etc. The methodologies are easily exploitable in a wide range of non-STFC applications e.g. epidemiology and we are already working with health informatics partnerships. Students could work on extending the models, applying to specific data sets and projects are easily devised as self-contained parts of the whole programme.