Topics for graduate theses

Theses are possible at any time in the areas of hardware, simulation and analysis. Some current topics are listed below as examples. For more information and further topics for theses please contact Prof. Alexander Kappes (e-mail), IKP room 224. For topics related to the Einstein Teleskiop you can also contact Dr. Waleed Esmail (E-Mail), IKP Raum 115.

Current topics for bachelor theses:

• Fast Gravitational-Wave Parameter Estimation for the Einstein Telescope using Neural Likelihoods

  Future third-generation gravitational-wave observatories such as the Einstein Telescope will detect orders of magnitude more compact-binary mergers than current detectors, potentially 104–105events per year. This dramatic increase in detection rate makes traditional Bayesian parameter estimation methods (e.g., nested sampling or MCMC used in pipelines like Bilby) computationally challenging, because the cost of likelihood evaluations scales with both waveform duration and the number of posterior samples required. For ET, signals may last hours in the detector band, further increasing the computational burden. A promising direction is therefore to learn compressed representations of gravitational-wave signals using autoencoders and perform parameter inference using neural likelihood estimation, enabling much faster approximate Bayesian inference while retaining astrophysical accuracy. This thesis would investigate whether such ML-based inference techniques can reproduce posterior distributions for compact-binary parameters while significantly reducing computational cost compared to traditional Bayesian pipelines.

• Parameter-Conditioned Autoregressive Waveform Modelling for the Einstein Telescope Era

  This project focus on GW waveform modelling (surrogate generation) for the Einstein Telescope era by taking an existing sequence model like your token-based GPT/Transformer and fine-tuning it to act as a fast waveform surrogate that remains accurate across a broad CBC parameter space. The key challenge for ET is that long, information-rich signals and very high SNR place tight demands on waveform accuracy, the student would structure training around maximizing the noise-weighted overlap (match) between generated and reference waveforms (e.g., minimizing mismatch 1-O) while systematically expanding coverage over m1,m2,χ1z,χ2z,… . Concretely, (i) generate a training set of reference waveforms, (ii) fine-tune the model to predict waveform windows/tokens conditioned on source parameters (your code already supports conditioning via θand tokenized inputs), and (iii) evaluate faithfulness/effectualness using overlap metrics over a held-out grid, reporting where the model fails and how.

Current topics for master theses:

• Improving supernova detection in IceCube Upgrade using machine learning: Neutrinos from core-collapse supernovae typically possess energies in the ballpark of tens of MeVs, rendering them impossible to detect using IceCube's standard reconstruction methods. Currently, IceCube identifies potential supernova neutrino bursts by monitoring for a collective increase in the rates across all optical modules. The forthcoming integration of mDOMs in the IceCube Upgrade heralds a promising shift for detecting these events. Given the configuration of 24 photomultipliers per module, the low-energy neutrino events can produce hits across multiple photomultipliers in a single module in a very short time. This master's study will employ machine learning to efficiently discern supernova neutrinos from background noise, focusing on the analysis of hit patterns and correlations across the array of photomultipliers and optical modules.
• Multi-Messenger Observations of Core-Collapse Supernovae with the Einstein Telescope and IceCube-Gen2: Core-collapse supernovae (CCSNe) are among the most energetic astrophysical phenomena, producing gravitational waves (GWs), neutrinos, and electromagnetic signals. These events provide a unique opportunity for multi-messenger astrophysics, offering insights into the physics of stellar collapse and the formation of neutron stars or black holes. This master project aims to pursue a feasibility study that integrates simulated data from ET and IceCube-Gen2 for multi-messenger analyses of CCSNe. Detection of CCSNe with current observatories like Advanced LIGO (aLIGO), and IceCube is challenging, primarily due to the limitations in their sensitivity and detection methodologies. ET's unprecedented sensitivity, especially in the low-frequency range, will extend the GW detection horizon for CCSNe by an order of magnitude. In addition, the integration of multi-PMT optical modules (mDOMs) in IceCube-Gen2 will enhance sensitivity to MeV-scale neutrinos, allowing individual events to be detected and reconstructed. The projects aims to develop methods to correlate gravitational wave signals from the ET with low-energy neutrino bursts from IceCube-Gen2, focusing on enhancing detection and correlation accuracy.
  
  [1] Timo Peter Butz, Study On The Detection Of Gravitational Waves From Core-Collapse Supernovae With The Einstein Telescope, Master thesis RWTH Aachen, 2024
  [2] Jade Powell, et al., Determining the core-collapse supernova explosion mechanism with current and future gravitational-wave observatories, Phys. Rev. D 109, 063019, 2024