GRAPPA MSc Thesis Projects


The following projects are Master Thesis Projects that have been defined by GRAPPA faculty members. Other thesis projects may also be available, please contact the relevant faculty members. See also GRAPPA members’ research interests. Please note that this is an evolving list, so come back in the future.

See the list of completed thesis projects.

Projects with a September 2019 start:


Various projects on accretion around compact objects and particle acceleration in jets

My group works on a variety of topics spanning astrophysics to astro-particle physics, mostly focused on accretion and subsequent jet production around compact objects, how and where particles are accelerated within them, and their effect on the environment.   On the modeling side, we are working on new models that include hadronic processes and their secondaries, so can start to compare with purely ‘leptonic’ models using precision multiwavelength datasets, this would involve learning some software and applying the models, or for more ambitious students who want to pursue theory, even developing new modules/models.   For more experimental types I would be interested in collaborating with KM3NeT folk on techniques to optimize transient searches (likely together with Shin’ichiro Ando and Aart Heijboer), or using CTA “data challenge” data to make predictions for CTA for Galactic X-ray binaries and AGN.   Advanced programmers could have the option of working with our GRMHD simulations.   Please get in touch to discuss options if you are interested!

Contact: Sera Markoff

Searching for dark matter with gravitational waves

The discovery of gravitational waves has opened new exciting opportunities for fundamental physics. One of the most intriguing aspects of this new “window in the universe” is the possibility to study in unprecedented detail the environment around black holes, and a team of GRAPPA researchers has recently shown that these observations can set extraordinarily stringent constraints on the mysterious dark matter that appears to permeate the universe at all scales. In this project we will explore the interplay between gravitational waves, black holes and dark matter. We will focus in particular on the possibility to probe the fundamental nature of dark matter by looking at how it clusters around black holes, and on its subtle impact on the gravitational waveform produced in the merger of black hole binaries. The project will involve both analytical and numerical work, and will be conducted under supervision of G. Bertone, and in collaboration with other GRAPPA staff and postdocs. 

Contact: Gianfranco Bertone

Deep Learning for BSM physics at the LHC

In recent years deep learning has proven to be a useful framework for many problems that arise in physics. One of its key features is the possibility to encode complex and high-dimensional functions in the form of a neural network that allows extremely fast inference. On the other hand, searches for new physics are often limited by their immense computational demand, e.g. for the calculation of pMSSM-19 cross sections. The traditional way of calculating them is via Monte Carlo methods that take several minutes per model point. The goal of this thesis is to substitute the Monte Carlo methods for Deep Learning and to build a tool for the physics community that offers this highly accelerated way of calculating cross sections. In the past, we have already investigated the feasibility of this endeavor and published a paper on arxiv: The student is expected to familiarize her-/himself with the underlying physics and the applied deep learning techniques to then expand the existing tool. Once all the pMSSM-19 particle pairs are done, which is very well possible within a year – especially considering that this is a team effort with international members from the Netherlands, Poland, Spain and South Africa – we will publish both, the tool and another paper, to which the student would of course be invited. 

Contact: S. Otten and Gianfranco Bertone

The artificial physicist

Recent breakthroughs in machine learning, in particular deep learning, have a profound impact on how data analysis and modeling in many scientific areas are done. One exciting development is that sparse modeling of ordinary differential equations and auto differentiation can be used to optimize and learn physics laws (like the Euler equations for fluid dynamics) from raw data. In this project, we will explore how the laws governing various dynamical systems can be learned directly from data. The student will write physics simulators, generate mock data, and write the training pipeline. Several applications are possible:

  • Gravitational wave signals. We would start learning these based on available template waveforms.
  • Gravitational potential. Given the movement of millions of stars (e.g. GAIA data), we will reconstruct the underlying gravitational potential of the Milky Way.
  • Laws of gravity and dark matter structure formation. Using either mock data based on modified gravity, or results from dark matter only N-body simulations, we will infer the laws of (modified) gravity, or effective laws describing the structure grows of dark matter halos in the Universe.

Experience with machine learning and programming is recommended, though not strictly necessary. Other project ideas are welcome as well.

Contact: Christoph Weniger

Falsifying cold dark matter with variational autoencoders

The gravitational effects of small dark matter halos can leave specific signatures in the surface brightness profile of strongly lensed high-redshift galaxies. Future observations with for instance the planned ELT will provide extremely sensitive probes towards these signatures, and a non-discovery of small halos would contradict predictions from cold dark matter. One of the challenges of strong lensing analyses is the correct modeling of the complex mass distribution of galaxies. In this project, we will use machine learning techniques to model the galaxy mass distribution, learned directly from N-body simulations. To this end, we will train variational autoencoders, which are state-of-the-art approaches to obtain high-fidelity parametric generative models directly learned from data. The work will be an important extension of ongoing activities and analysis pipelines, which already include generative models for high-redshift source galaxies (directly trained on observed galaxies).

Contact: Christoph Weniger

Radio searches for dark matter axion-photon conversion with the Murchison Widefield array

QCD axions are one of the theoretically best motivated candidates for cold dark matter (they are produced as by-product of dynamical solutions to the strong CP problem). Dark matter axions would convert into photons in the strong magnetic field of neutron stars, leading to sharp radio line signals that can be observable with existing radio telescopes. In this project, the student will contribute to the analysis of 4 hours of Galactic center observations obtained from the Murchison Widefield array (MWA). The student can either contribute on the side of data reduction (bringing TBs of data down to radio images with high spectral resolution), or on the side of the modeling of the expected signal from neutron stars in the Galactic bulge. The available data is expected to be strong enough for discoveries, and otherwise would provide the most stringent constraints.

Contact: Christoph Weniger

Searches for X-ray lines from sterile neutrino dark matter

Sterile neutrinos are excellent candidates for particle dark matter in the keV mass range. The decay of these particles could show up as monoenergetic line in X-ray observations. Potential line candidates around 3.5 keV are discussed in the literature since a few years. The student will in this project use machine learning tools to analyse available X-ray data is a much more fine-grained way than previously possible. We will use the full spatial and spectral information available in different data sets, and account for the large number of uncertainties that are associated with the different flux components. Machine learning libraries like pytorch or tensorflow allow here to marginalize over millions of parameters, if necessary. We will use probabilistic programming techniques and variational inference to obtain a proper assessment of the uncertainties involved in the analysis.

Contact: Christoph Weniger

Modeling Lyman alpha spectra and stellar streams with deep probabilistic programming

The temperature of dark matter can be probed with observations of Lyman alpha absorption lines in distant quasars (caused by neutral Hydrogen clouds on the line-of-sight), or by the structure of stellar streams in our Milky Way halo. Usually, the power spectrum is used to analyse variations and structure in these observations. In this project, the student will use the latest developments on the front of deep probabilistic programming (a mixture between deep neural networks and variational Bayesian inference) to learn surrogate models for these observations directly from large computer simulations. These surrogate models will encompass all the relevant correlation structure of the original observations, and we will explore if and how they provide a more sensitive probe of the dark matter temperature than conventionally used approaches.

Contact: Christoph Weniger

Searching for dark matter subhalos using Pulsar Timing Array

Pulsars are the most precise clock. Deviation from regular pulses will indicate gravitational perturbation on their way to telescopes. It can be caused by passage of small dark matter structures, called subhalos. Pulsar Timing Array will have great sensitivity to these subhalos, especially with very small masses. In this project, the student will look into sensitivities of the Pulsar Timing Array in detecting the dark matter subhalos, and question if one can test different dark matter candidates.  There will be other methods to probe dark matter subhalos, which might also be a subject of other thesis projects.

Contact: Shin’ichiro Ando

Diffuse supernova neutrino background

Core-collapse supernova explosions are one of the most spectacular events in astrophysics. They will produce copious amount of thermal (and possibly non-thermal) neutrinos, strong gravitational waves as well as electromagnetic waves. Although the occurrence rate is unfortunately small in the Milky Way, it is possible to detect the supernova neutrinos as diffuse background radiation. The purpose of this project is to provide the most realistic theoretical predictions of the flux of the diffuse supernova neutrino background and discuss the detectability at various detectors around the world. It would reveal the most violent aspect of the Universe.

Contact: Shin’ichiro Ando

Searching for Dark Matter with antiprotons

The first non-gravitational signal from Dark Matter is yet to be unveiled. One of the most promising searching methods is that of indirect detection. This consists in identifying excesses in cosmic ray fluxes (electron/positrons, gamma-rays, neutrinos or antiprotons) which could possibly be produced by Dark Matter emissions. Observation of many different astrophysical environments with high concentration of dark matter has not yet revealed a smoking gun signature for dark matter. One notable exception consists of the recent claim of a tentative detection at the level of 4 \sigma statistical significance of an antiproton excess in AMS-02 data consistent with DM particles of mass ~70-90 GeV self-annihilating into the b\bar{b} channel. This claim, however, needs further scrutiny and in this project we will address this using the most robust data analysis methods and modelling techniques.

Contact: Shin’ichiro Ando

XENON1T Data Analysis

The XENON collaboration has used the XENON1T detector to achieve the world’s most sensitive direct detection dark matter results and is currently building the XENONnT successor experiment. The detectors operate at the Gran Sasso underground laboratory and consist of so-called dual-phase xenon time-projection chambers filled with ultra-pure xenon. Our group has an opening for a motivated MSc student to do analysis with the data from the XENON1T detector. The work will consist of understanding the detector signals and applying machine learning tools such as deep neutral networks to improve the reconstruction performance in our Python-based analysis tool, following the approach described in arXiv:1804.09641. The final goal is to improve the energy and position reconstruction uncertainties for the dark matter search. There will also be opportunity to do data-taking shifts at the Gran Sasso underground laboratory in Italy.

Contact: Patrick Decowski

DARWIN Sensitivity Studies

DARWIN is the “ultimate” direct detection dark matter experiment, with the goal to reach the so-called “neutrino floor”, when neutrinos become a hard-to-reduce background. The large and exquisitely clean xenon mass will allow DARWIN to also be sensitive to other physics signals such as solar neutrinos, double-beta decay from Xe-136, axions and axion-like particles etc. While the experiment will only start in 2025, we are in the midst of optimizing the experiment, which is driven by simulations. We have an opening for a student to work on the GEANT4 Monte Carlo simulations for DARWIN, as part of a simulation team together with the University of Freiburg and Zurich. We are also working on a “fast simulation” that could be included in this framework. It is your opportunity to steer the optimization of a large and unique experiment. This project requires good programming skills (Python and C++) and data analysis/physics interpretation skills.

Contact: Patrick Decowski



Projects from last year: 

Various projects on accretion and jet production around compact objects

My group works on a variety of topics spanning astrophysics to astro-particle physics, mostly focused on accretion and subsequent jet production around compact objects, and their effect on the environment.   “Grappa-esque” projects could involve taking existing semi-analytical models and applying them to new multi-wavelength data sets, or even developing new models (or modules).   For more experimental types I would be interested in collaborating with KM3NeT folk on techniques to optimize transient searches (likely together with Shin’ichiro Ando and Aart Heijboer), or using CTA “data challenge” data to make predictions for CTA using our models.   Advanced programmers could have the option of working with our GRMHD simulations.   Please get in touch to discuss options if you are interested!

Contact: Sera Markoff

What impact do turbulent magnetic fields have on particle acceleration models?

For modeling spectral energy diagrams (SEDs) of supernova remnants one often uses simple models in which gamma-ray and radio and X-ray data are combined, using a single zone and a uniform magnetic field. Energetic electrons produce both gamma-ray emission through inverse Compton scattering and radio to X-ray emission through synchrotron radiation.  Multi-zones are already more advantages, but one aspect is always overlooked  for modeling X-ray synchrotron emission: the fact that it requires turbulent magnetic fields. This means that the constant magnetic field assumption is per definition wrong and that it may have led to biases in maximum energy estimates. For this project the idea is to investigate the impact the assumption of a turbulent magnetic field has on modeling basic particle accelerations models, namely what is the maximum electron energy inside the sources.

Contact: Jacco Vink

Modeling the ionization and heating in the interior of Cassiopeia A

My interests are not only confined to particle acceleration related to supernova remnants, and this project is only remotely related to particle acceleration: We recently measured the internal radio absorption of the remnant Cassiopeia A with LOFAR (Arias et al. 2018, ArXiv:1801.04887). This internal absorption informs us of the unshocked (cold) gas inside the shell, which has not been shocked and inform us about the evolutionary state of the remnant. However, to translate absorption into unshocked mass we need to know the temperature. Based on infrared data the temperature is assumed to be 200 Kelvin, suggesting a large mass of 2 Msun. However, it is not clear how the inside can be this hot, as radiation from the bright shell makes 30 Kelvin more likely. So for this project we would like to model the ionisation and heating in the interior of Cas A. It is likely that the state of the gas is not in equilibrium. So we would like to make time dependent modeling, using simple hydrodynamics and using heating from UV/X-ray emission from the hot shell. This is a challenging, but doable project that will require combining different modeling packages. My PhD student Maria Arias will co-supervise.

Contact: Jacco Vink

Particle acceleration efficiency of solar system shocks

With a past master student, I investigated the particle acceleration efficiency of solar system shocks. The idea was to test whether there was a critical Mach number (v/v_sound= M ~2.2) below which acceleration is suppressed. This project showed that for M > 2.2 the efficiency is not Mach number dependent and is about 5-10%. This project is a follow-up: look for shocks that are below 2.2 and find out whether they can accelerate and what the circumstances are under which they accelerate (since we found a few that indeed do not accelerate). In addition, we want to look at the magnetic field turbulence in the shock regions and connect it to the diffusion processes that are essential for the process of diffusive shock acceleration.

Contact: Jacco Vink

Other Projects

Here is a list of MSc research projects that are available at Nikhef.