Master Thesis Projects
Note that besides the projects below there are also opportunities for Master Thesis projects at the Max-Planck Institutes. For example in the High Energy Group of MPE here or at the Max-Planck Institute for Astrophysics.
- The Hubble Tension and different cosmic distance measures
Recent direct measurements of the cosmic expansion from Cepheid-calibrated supernovae Ia are in increasing tension with values indirectly inferred from the cosmic microwave background and standard rulers based on baryonic acoustic oscillations created in the early Universe. Currently it is unclear if the tension is due to problems with one (or several) of the measurements, or if it signals a breakdown of our current cosmological standard model. This project tries to test underlying assumptions of the standard model with public cosmological data sets. - Likelihood-free inference with galaxy clusters
Usually in cosmology, we learn about our model parameters (such as the expansion rate or the matter density) by writing down a likelihood for a measurement and evaluating it with the data. This forces us to combine data into summary statistics with (approximately) known likelihood, such as power spectra or total abundance of galaxy clusters. Another approach is to simulate the Universe until it matches the observation, using state-of-the-art machine learning and Bayesian inference tools. This project aims to implement a toy model of this approach before applying it to data. - Diffusion Models for Cosmological Inference
In traditional Bayesian inference, it is necessary to write down an analytic form of the likelihood in order to draw samples from the posterior distribution. However, likelihoods in cosmology are not always known or tractable. In these cases one can make use of modern so-called likelihood-free / simulation-based inference methods, which can deliver posterior estimates by turning the inference into a problem of density estimation. There is an array of density estimators which can be employed for this task, many of which make use of the expressiveness of neural network and deep learning. Other fields of Physics have recently seen a surge in the usage of a special type of deep learning density estimators called diffusion models, which promise an increase in performance compared to previous methods. In this project, we will explore the possible use cases of such diffusion networks in Cosmology, e.g. by applying them to n-body or hydrodynamical simulations. Prior experience with machine learning methods is highly recommended.
- Fast Radio Bursts
Fast Radio Bursts are quite mysterious: they are very short and very bright signals, but their source is still unknown. However, they are definitely extragalactic and visible up to cosmological distances. Since the radio signal undergoes dispersion as it travels through the ionised intergalactic medium, the pulses allow us to probe the large-scale structure of the Universe in a new and exciting way. There are several projects available focusing on theoretical or numerical work depending on your interests. Feel free to get in touch! - Maximum Entropy Reconstruction of the Reionization History
In this project we develop a new method to constrain the reionization history of the Universe with Cosmic Microwave Background data. - Topological Classification of the Cosmic Web
Usually cosmological information is extracted from overdense (clusters) or underdense (voids) regions of the cosmic web. However the structure of the cosmic web is much richer, not only consisting of clusters and voids, but also of filaments and sheets. Detecting and quantifying all these structures is an exercise in classifying the cosmic web topologically. In this project we want to explore the cosmic web with Betti numbers on different scales. - Real-Time Cosmology
We will explore the ability of frequency comb spectrographs, such as the one on the Wendelstein telescope, to measure the time dependence of the redshift factor. This would allow, for example, to constrain intrinsically inhomogenous cosmological models. - Cosmic Voids
The emptiest regions in the Universe may reveal key insights to our understanding of dark energy, dark matter, and other fundamental aspects of cosmology, but their composition and evolution has only begun to be investigated in detail. In this project we will identify voids in simulated and / or observational data, statistically analyze some of their properties, and develop physical models to establish connections to theory. - Cosmic Voids and eRosita
eROSITA will perform the first imaging all-sky survey in the soft energy X-Ray band. Its unprecedented spectral and angular resolution provide us with the unique opportunity of creating a complete mapping of cosmic voids in our cosmos. Voids are typically studied in the distribution of matter. However, to test our theoretical findings with observations, we need to rely on luminous tracers. Recent studies have focussed on how bridge the gap between theory and observations, suggesting that this can easily modelled via the tracer bias. Given this context, eRosita cluster and AGN catalogues represent the exciting next generation of data that will allow us to trace a realistically complete sample of voids, and guide us towards a better understanding of their statistical properties. - Mark correlation of galaxies and halos
Models of galaxy formation have to explain the large scale structure and the properties of the galaxies (luminosity, type, etc.). You will use mark correlation functions to investigate the interplay of spatial distribution and inner properties of galaxies. Theoretical as well as numerical approaches are possible. - Deep Learning the Weak Lensing Mass of Galaxy Clusters
In this project we will explore state of the art learning algorithm to estimate the mass of a galaxy cluster from imaging data of background galaxies. Initially the project will be developed on synthetic catalogs of galaxies but there might be scope to apply this to real observational data.