Dark matter, the hidden force governing the cosmos, holds the key to unlocking the universe’s greatest secrets. represents a staggering 85% of all matter and roughly 27% of the contents of the universe. Despite its pervasive influence, our inability to directly observe it necessitates the study of its gravitational effects on galaxies and other celestial structures.
After years of investigation, the enigmatic nature of dark matter persists as one of the most profound puzzles in science. According to a prominent theory, dark matter could be a type of particle with extremely limited interaction with other matter, except through gravitational forces.
However, some researchers propose that these particles may sporadically interact with each other, a phenomenon known as self-interaction. The detection of such interactions could provide vital insights into the properties of dark matter.
Distinguishing the subtle signs of dark matter self-interactions from other cosmic effects, such as those caused by active galactic nuclei (AGN), has presented significant challenges. AGN feedback can induce similar effects to those of dark matter, making it difficult to distinguish between the two.
Astronomer David Harvey at EPFL‘s Laboratory of Astrophysics has made a significant breakthrough by developing a deep-learning algorithm capable of unraveling these complex signals. The AI-based method has been specifically designed to differentiate between the impacts of dark matter self-interactions and AGN feedback by analyzing images of galaxy clusters—massive assemblages of galaxies held together by gravity. This innovative approach holds great promise for advancing the precision of dark matter studies.
Harvey developed a cutting-edge Convolutional Neural Network (CNN) designed to analyze images from the BAHAMAS-SIDM project, which simulates galaxy clusters under different dark matter and AGN feedback scenarios. Through exposure to thousands of simulated galaxy cluster images, the CNN successfully learned to differentiate between signals caused by dark matter self-interactions and those caused by AGN feedback.
Out of all the CNN architectures tested, the “Inception” model, being the most complex, demonstrated the highest level of accuracy. The AI was trained on two primary dark matter scenarios with varying levels of self-interaction and validated on additional models, including a more intricate, velocity-dependent dark matter model.
Inception demonstrated an impressive accuracy of 80% under ideal conditions, effectively discerning whether galaxy clusters were impacted by self-interacting dark matter or AGN feedback. It maintained its high performance even when realistic observational noise was introduced to mimic data from future telescopes like Euclid.
This indicates that Inception, and AI methods in general, could be extremely valuable for analyzing the vast amounts of space data we collect. Furthermore, the AI’s capability to handle unseen data suggests it is adaptable and reliable, promising great potential for future dark matter research.
AI-based approaches such as Inception could greatly influence our understanding of dark matter. As new telescopes gather unprecedented data, this method will help scientists swiftly and accurately sift through it, potentially revealing the true nature of dark matter.
Journal reference:
- D. Harvey. A deep-learning algorithm to disentangle self-interacting dark matter and AGN feedback models. Nature Astronomy, 2024; DOI: 10.1038/s41550-024-02322-8
