³Ô¹ÏºÚÁÏ

21cm Cosmology

Cosmologists have a detailed understanding of the early Universe up to the formation of the Cosmic Microwave Background thanks to instruments like Planck and of the present Epoch of the Galaxies thanks to observations from JWST and other telescopes. However, the period between 300,000 and 1 billion years after the Big Bang remains largely unobserved. During this period the cosmic web came together, the first stars and galaxies formed and the Universe transitioned from largely neutral to largely ionized.

A promising probe of this unobserved epoch is the 21-cm line in neutral hydrogen. The line arises from the spin flip transition in neutral hydrogen and the relative number of atoms with aligned and anti-aligned electron and proton spins can be characterised as a statistical temperature. The signal, emitted with a wavelength of 21-cm, is redshifted by the expansion of the Universe, into the low frequency radio band on Earth. At different times different process drive the number of atoms in each state and by tracing the evolution of this spin temperature researchers can answer several fundamental questions:

• When the first stars and galaxies formed?
• How big and how bright were these first stars and galaxies?
• What is the nature of dark energy and dark matter?
• When did the first exotic objects like X-ray binaries form?

Credit: CosmoCube team

However, the 21-cm signal remains allusive. While our telescopes are sensitive to the 21-cm signal they are also sensitive to emission from our own Galaxy which is around five orders of magnitude brighter. Observations are further complicated by human-made interference in the radio band, changes in the environment around the telescope with time, extreme calibration requirements and the atmosphere. As a result researchers have to build complex forward models of their data to confidently extract the 21-cm signal. However, these forward models are typically very expensive and need to be evaluated millions of times requiring novel modelling and inference solutions.

At ³Ô¹ÏºÚÁÏ, we are working on addressing these challenges by developing novel machine learning tools to;

• rapidly evaluate computationally expensive signal models,
• explore theoretic models,
• emulate instrument models,
• build enhanced inference algorithms,
• and explore data via simulation based inference techniques.

We play a leading role in the REACH collaboration which operates a radio telescope in South Africa targeting the sky-averaged 21-cm signal and plans to deploy additional instruments in the coming years. We are also part of the CosmoCube team which aims to deploy a mini satellite to lunar orbit in the next 5 years to observe the sky-averaged 21-cm signal at higher redshifts than REACH. CosmoCube will reveal the dark ages before the first stars and galaxies formed helping us understand the early evolution of the Universe and nature of dark energy. In the coming five years the Square Kilometre Array will come online, and our researchers are involved in developing inference pipelines for analysis of the Epoch of Reionization data.

Credit: REACH collaboration

Projects we are involved in:

REACH:  

CosmoCube:  

SKA EoR:  

Recent Research

de lera Acedo, E., Bacon, D., Grainger, W. et al. The CosmoCube lunar mission for probing the dark ages and cosmic dawn via 21-cm cosmology. Nat Astron 10, 1097–1106 (2026).

Tutt, Jacob L., et al. "Towards end-to-end Bayesian forward models in global 21-cm cosmology: surrogate modelling and marginalisation of beam uncertainty." arXiv preprint arXiv:2608.18962 (2026).

Bevins, H. T. J., T. Gessey-Jones, and W. J. Handley. "On the accuracy of posterior recovery with neural network emulators." Monthly Notices of the Royal Astronomical Society 544.1 (2025): 375-390.

de Lera Acedo, E., et al. "The REACH radiometer for detecting the 21-cm hydrogen signal from redshift z≈ 7.5–28." Nature Astronomy 6.8 (2022): 984-998.