Understanding how galaxies form stars over cosmic time remains a fundamental challenge in astronomy, and now, Sinan Deger, Hiranya V. Peiris, and Stephen Thorp, all from the Institute of Astronomy and Kavli Institute for Cosmology at the University of Cambridge, alongside Daniel J. Mortlock from Imperial College London and colleagues, present a new approach to charting this history. The team developed ‘pop-cosmos’, a generative model that learns the patterns of star formation by analysing data from a vast catalogue of 420,000 galaxies, allowing them to reconstruct the star formation rate density over the last 12 billion years. This innovative method reveals detailed patterns in how galaxies form stars and subsequently cease doing so, demonstrating that the quenching of star formation in massive galaxies occurs over a timescale of a few billion years and is closely linked to the activity of supermassive black holes. The findings provide compelling evidence for a critical role of active galactic nuclei feedback in regulating galaxy evolution during this crucial quenching transition, offering new insights into the assembly of galaxies we observe today.
Deep Learning Reconstructs CMB Lensing Potential
Scientists have developed a new method for mapping the cosmic microwave background (CMB), the afterglow of the Big Bang, to reveal the distribution of matter in the universe. This technique uses deep learning, a type of artificial intelligence, to reconstruct the CMB lensing potential, representing how light from the early universe is bent by intervening matter. This new approach overcomes limitations of traditional methods by training a computer network to directly estimate the lensing potential from CMB maps, simplifying the process and reducing errors. The team demonstrates that this deep learning method reduces noise by 25 per cent and improves the accuracy of measurements by 15 per cent. These advancements enable more precise measurements of dark energy, dark matter, and neutrino masses, offering a powerful tool for exploring the fundamental properties of the universe and testing cosmological models.
Lower-Mass Galaxy Star Formation Histories in pop-cosmos
Researchers have analysed the star formation histories of lower-mass galaxies using a detailed simulation called pop-cosmos. This simulation models galaxy formation and evolution, allowing scientists to investigate how star formation rates change over time. Focusing on galaxies with masses between one and ten billion times that of our sun, at a time when the universe was three to four billion years old, the team reconstructed the star formation history of each galaxy to understand how these smaller galaxies evolved differently from their more massive counterparts. The analysis reveals that lower-mass galaxies exhibit more consistent star formation over time compared to more massive galaxies, though this correlation weakens with time. Quenched galaxies show a gradual decline in star formation lasting up to a billion years, with a significant fraction having formed a substantial amount of their stars relatively recently, suggesting the quenching process is not instantaneous. These findings highlight the importance of mass in shaping galaxy evolution.
Reconstructing Galaxy Evolution Over Cosmic Time
Researchers have developed a detailed model, named pop-cosmos, to reconstruct the history of star formation over the last 12 billion years. Trained on data from over 420,000 galaxies, this model accurately simulates galaxy populations and their evolution, allowing researchers to investigate how star formation rates have changed over cosmic time. The team calculated the star formation rate density, finding a peak value at a specific point in the universe’s history, and successfully classified galaxies as either actively star-forming or quiescent. Analysis of the model’s output reveals distinct patterns in the star formation histories of different galaxy types.
Star-forming galaxies exhibit gradually decreasing star formation rates over time, suggesting increasingly random star formation in the early universe. Quiescent galaxies, however, demonstrate a clear transition from active star formation to quiescence, with massive galaxies ceasing star formation over several billion years. Notably, activity from active galactic nuclei peaks as galaxies transition from star-forming to quiescent states, supporting the idea that these black holes play a crucial role in halting star formation.
👉 More information
🗞 pop-cosmos: Star formation over 12 Gyr from generative modelling of a deep infrared-selected galaxy catalogue
🧠 ArXiv: https://arxiv.org/abs/2509.20430