Harvard–Smithsonian team develops record-scale early-universe simulation

Lumina, an early-universe simulation from the Center for Astrophysics | Harvard & Smithsonian in Cambridge, MA, helps interpret data from the James Webb Space Telescope and the Nancy Roman Space Telescope.

An arXiv preprint appeared on May 14, 2026. Monthly Notices of the Royal Astronomical Society published the work in 2026 under DOI 10.1093/mnras/stag1469, titled Introducing the Lumina project: large-volume radiation-hydrodynamic simulations of the epochs of hydrogen and helium reionization.

The team calls it the largest cosmological radiation-hydrodynamics simulation ever created, with a volume exceeding 100 times that of previous comparable simulations. It covers a cubic volume roughly 1.5 billion light-years wide on each side, a scale comparable to what today’s most powerful telescopes can observe.

Leren Leren reports that the simulated cosmic web spans approximately 650 million light-years across. Lumina defines a new class of simulation with a 500 cMpc volume and nearly half a trillion resolution elements, where cMpc refers to comoving megaparsecs, a distance measure used in cosmology. The Oak Ridge National Laboratory machine known as Frontier, among the world’s fastest supercomputers in Tennessee, needed more than 15,000 graphics processing units to produce the simulation.

Researchers at the Center for Astrophysics and partner institutions spent years developing and scaling the software architecture, relying on more than 15,000 graphics processing units for parallel work before the dataset and publication in late 2026. Frontier’s computing power let researchers model a large volume while retaining the detail needed to study galaxies, stars and black holes. The model runs on the moving-mesh code AREPO, which joins the IllustrisTNG galaxy-formation model to a GPU-accelerated M1 radiation-transport solver operating across six frequency bins.

It connects IllustrisTNG to radiative transfer that happens on the fly, non-equilibrium chemistry, and ionizing sources such as stars, AGN, X-ray binaries, and shock-heated gas. For its starting conditions, Lumina applies separate transfer functions to baryons and dark matter and includes their relative streaming velocity.

The construction also drew on advanced hydrodynamical codes that solve magnetohydrodynamics equations together with Monte Carlo radiation transport methods. The model follows cosmic history from the Dark Ages to the close of He II reionization at redshift z = 3. It shows how ordinary matter, dark matter and radiation interacted during the first two billion years of cosmic history, giving researchers a way to check theories about cosmic evolution after the big bang by comparing telescope data with the simulation’s predictions.

The simulation indicates that hydrogen reionization happened late and was mainly driven by stars, with the median sub-volume fully ionized by z≈5.2 while leftover neutral HI patches persisted until z≈4.75. In the model, AGN power He II reionization in a self-consistent way, bringing the process almost to completion by z = 3.

No earlier simulation captured helium reionization as it unfolded while letting its effects instantly shape the evolving universe inside the model. By matching JWST, Roman and other telescope observations against Lumina’s predictions, astronomers can check whether their understanding of the laws governing the universe still holds or needs revision. Researchers built it to interpret signals including the Lyman-α forest, 21-cm cosmology, line-intensity mapping, galaxy surveys, quasar absorption spectra, and other tools for studying the evolving ionized structure of the intergalactic medium.

The galaxy mix in Lumina matches the first IllustrisTNG effort, and the bigger volume improves statistics for rare objects, large-scale settings, and cosmic variance. It follows large-scale objects including galaxy clusters and cosmic filaments while also resolving smaller bright objects that drive changes in intergalactic gas.

The output amounts to petabytes of data that researchers can analyze and visualize from broad cosmic webs to single galactic settings, covering dark matter distribution, gas movement, and elemental ionization states as hydrogen and helium ionize under early stellar radiation. Leren Leren describes a dynamic, frame-by-frame visualization where hydrogen shows up in bright gold regions and helium in bold cyan across large cosmic volumes.

Oliver Zier, the lead author of the Lumina paper and an ITC fellow at the Center for Astrophysics | Harvard & Smithsonian, said, “With JWST, the early universe has become an exciting new frontier in astrophysics”. Aaron Smith, an author on the paper and an assistant professor of physics at the University of Texas at Dallas, said researchers can for the first time follow hydrogen and helium changing into their modern form across an enormous volume. Smith added, “We feel incredibly fortunate to be building simulations like Lumina”.

JWST began operations in 2022, and its observations moved the timeline of early galaxy formation earlier than older models had forecast, increasing the need for predictive theoretical models. The foundational framework for large cosmological simulations goes back decades, yet simulating the epoch of reionization stayed difficult because it must follow radiative transfer, gas movement, and gravitational collapse across different scales at the same time. This need lay behind the team’s years of work to build Lumina, set against the growth of supercomputing infrastructure.

Later work will use Lumina to study how gas travels between galaxies, how radiation from the big bang influenced cosmic evolution, and how areas of strong star formation developed over time. It works as a diagnostic sandbox for unusual galaxies that look far larger or more developed than standard models predict, and gaps between simulation and telescope data may reveal limits in standard models, possibly indicating new physics, unusual dark matter traits, or feedback processes not yet accounted for.

A study by Sadain et al. in The Open Journal of Astrophysics compared recombinations in Lumina and THESAN, demonstrating how gas temperature and ionization structure shape the photon budget; the Lumina website lists it as related work. Leren Leren says releasing Lumina datasets and visual media creates an educational tool that turns complex equations and supercomputer calculations into clear pictures of cosmic origins, and that the data will help plan observations for future missions as part of the project’s wider effect on observational and theoretical astrophysics.