Periodic Reporting for period 3 - UnDark (New Approaches to Uncover Dark Matter in the post-WIMP Era)
Reporting period: 2023-10-01 to 2025-03-31
The project "UnDark: New Approaches to Uncover Dark Matter in the Post-WIMP Era" addresses the complex and challenging issue of identifying the particle nature of dark matter (DM), which makes up 85% of the matter in the Universe. Despite its ubiquity, the particles that could constitute dark matter remain undetected, with only its gravitational effects on celestial bodies having been observed so far. The problem has been further compounded by the inability to experimentally confirm the previously favored paradigm of weakly interacting massive particles (WIMPs) as the main component of dark matter.
The overall objectives of UnDark are to uncover the nature of dark matter both by widening searches in astrophysical data beyond WIMP models, and by developing new analytical and machine learning methods to improve physical signal models and perform data analysis. The project focuses on axion dark matter and warm dark matter (e.g. sterile neutrinos), which are some of the theoretically most promising dark matter candidates. UnDark aims at providing accurate signal forecasts through detailed numerical modeling of the underlying physical processes, for instance through the modeling of axion-photon conversion in the magnetosphere of neutron stars, or the simulation of galaxy formation in the Universe. Another objective of the project is to introduce new statistical methods for comparing high-dimensional physical models (for instance of strong lensing images) with data, which are based on advanced in deep learning. The project will exploit radio, optical and other astronomical observations for signal searches, as well as a wide range of astrophysical targets from neutron stars to galaxies. The ultimate goal is either to detect signatures of dark matter particles, or to provide the strongest constraints yet on the most promising models.
The UnDark project presents a contribution to fundamental research. It has societal relevance in multiple ways. Dark matter is a mystery that can help to inspire a new generation of scientists. The project requires collaboration among physicists, astronomers, computer and data scientists. Such cross-disciplinary work can lead to unforeseen technological advances, e.g. in terms of transferable data analysis techniques. Furthermore, dark matter research and its intricate debates provide excellent material for educating the public about the scientific method and to increase scientific literacy.
The overall objectives of UnDark are to uncover the nature of dark matter both by widening searches in astrophysical data beyond WIMP models, and by developing new analytical and machine learning methods to improve physical signal models and perform data analysis. The project focuses on axion dark matter and warm dark matter (e.g. sterile neutrinos), which are some of the theoretically most promising dark matter candidates. UnDark aims at providing accurate signal forecasts through detailed numerical modeling of the underlying physical processes, for instance through the modeling of axion-photon conversion in the magnetosphere of neutron stars, or the simulation of galaxy formation in the Universe. Another objective of the project is to introduce new statistical methods for comparing high-dimensional physical models (for instance of strong lensing images) with data, which are based on advanced in deep learning. The project will exploit radio, optical and other astronomical observations for signal searches, as well as a wide range of astrophysical targets from neutron stars to galaxies. The ultimate goal is either to detect signatures of dark matter particles, or to provide the strongest constraints yet on the most promising models.
The UnDark project presents a contribution to fundamental research. It has societal relevance in multiple ways. Dark matter is a mystery that can help to inspire a new generation of scientists. The project requires collaboration among physicists, astronomers, computer and data scientists. Such cross-disciplinary work can lead to unforeseen technological advances, e.g. in terms of transferable data analysis techniques. Furthermore, dark matter research and its intricate debates provide excellent material for educating the public about the scientific method and to increase scientific literacy.
Over the initial 2.5 years of my project, “UnDark: New approaches to uncover dark matter in the post-WIMP era,” my team and I have made substantial progress in terms of dark matter signal modeling, statistical method development, and signal searches.
Indirect searches for dark matter aim at detecting electromagnetic or multi-messenger signals from dark matter. Our attention has here been primarily on axions, a promising candidate for dark matter. We developed an end-to-end ray-tracing simulation code to predict axion-photon conversion in the highly magnetized and dense neutron star environments. This mechanism leads to characteristic narrow radio spectral lines. Additionally, we set up a comprehensive framework to forecast axion signals generated in the polar gap regions of neutron stars, which leads to broad-band signals. While we didn't find definitive evidence for dark matter signals during our search using archival data from the Green Bank Telescope, we imposed the strictest constraints yet on axion-photon coupling yet.
The analysis of astrophysical data is often complicated by the fact that the data, and the corresponding physical models, are very high-dimensional, and require the proper statistical handing of a large number of uncertainties. We developed new statistical methods, based on deep learning, that can handle such analysis challenges efficiently. Our new technique, Truncated Marginal Neural Ratio Estimation (TMNRE), is a special case of simulation-based inference. We showed this method to be effective in tasks involving cosmological parameter inference and studying point source populations. Leveraging TMNRE, we created a new pipeline to constrain the warm dark matter halo mass function directly from multiple lensing images. What is more, we designed a statistical analysis framework for studying stellar streams and developed a dynamic modeling code for stellar streams, which allows for robust inference of mock stellar stream model. Furthermore, we studied the effects of dark matter self-interaction on dark matter halo shapes with N-body simulations (TangoSIDM).
Indirect searches for dark matter aim at detecting electromagnetic or multi-messenger signals from dark matter. Our attention has here been primarily on axions, a promising candidate for dark matter. We developed an end-to-end ray-tracing simulation code to predict axion-photon conversion in the highly magnetized and dense neutron star environments. This mechanism leads to characteristic narrow radio spectral lines. Additionally, we set up a comprehensive framework to forecast axion signals generated in the polar gap regions of neutron stars, which leads to broad-band signals. While we didn't find definitive evidence for dark matter signals during our search using archival data from the Green Bank Telescope, we imposed the strictest constraints yet on axion-photon coupling yet.
The analysis of astrophysical data is often complicated by the fact that the data, and the corresponding physical models, are very high-dimensional, and require the proper statistical handing of a large number of uncertainties. We developed new statistical methods, based on deep learning, that can handle such analysis challenges efficiently. Our new technique, Truncated Marginal Neural Ratio Estimation (TMNRE), is a special case of simulation-based inference. We showed this method to be effective in tasks involving cosmological parameter inference and studying point source populations. Leveraging TMNRE, we created a new pipeline to constrain the warm dark matter halo mass function directly from multiple lensing images. What is more, we designed a statistical analysis framework for studying stellar streams and developed a dynamic modeling code for stellar streams, which allows for robust inference of mock stellar stream model. Furthermore, we studied the effects of dark matter self-interaction on dark matter halo shapes with N-body simulations (TangoSIDM).
During the second half of the UnDark project, my team and I are planning to bring to fruition the developed modeling as well as data analysis techniques by applying them to various data sets. We expect to either falsify or confirm the cold dark matter paradigm by analyzing stellar streams and strong lensing images. Our continued effort to improve signal calculations for axion-induced radio emission from neutron stars is expected to lead to new search channels for axions (in terms of radio transients), which we plan to exploit in dedicated searches. Furthermore, we plan to leverage the TMNRE algorithm to address various analysis challenges in the wider dark matter, astrophysics and cosmology communities, including for instance challenging analyses of gravitational wave signals, with relevance for studying dark matter around black holes.