Credit: NASA, ESA, CSA, and STScI

Seeing the Early Universe Through a Simulation

Computational models are empowering researchers to understand observations from the James Webb Space Telescope, painting a virtual picture of the early universe.

The James Webb Space Telescope (JWST), which launched in December 2021, is one of the most sensitive scientific instruments ever built. It’s an eye in the sky, but also a time machine. Light travels at finite speed, meaning that the deeper into space we look, the farther back we can see. In July, researchers released the first JWST images, revealing galaxies formed in the universe’s infancy — more than 13 billion years ago, and less than 300 million years after the Big Bang. The images and the inferences we draw from them are the result of decades of planning, not only to construct the telescope but also to computationally model what it might find.

Of particular interest are the galaxies. “What I’m really looking at is their evolutionary course, which tells us a lot about the universe. How does it work? How did we get here? Are we alone?” says Aaron Yung, an astrophysicist at NASA’s Goddard Space Flight Center who spent four years as a visiting researcher at the Flatiron Institute’s Center for Computational Astrophysics (CCA) while a doctoral student at Rutgers University. Rachel Somerville, an astrophysicist and group leader at the CCA, says that galaxies in themselves are beautiful, each one unique. But she is especially interested in the features they have in common. For instance, spiral galaxies have younger stars than spheroid galaxies and tend to be bluer as a result. “Understanding where those patterns are coming from, I think, is really fascinating,” Somerville says. The simulations she, Yung and others have developed in recent years, outlined in a series of papers, have helped scientists plan the use of the JWST and are now helping them interpret its extragalactic observations.


A model universe

Beginning in 2018, Yung, Somerville and colleagues have published a series of six papers describing a computational model of the universe they developed and the predictions they’ve made about what JWST will reveal. They published the model with several goals in mind. First, they needed to persuade the telescope’s time-allocation committee to allow them precious time to use it. Yung and Somerville are both part of a team called the Cosmic Evolution Early Release Science (CEERS) Survey, aimed at exploring the galaxies in the early universe. Yung summarized their pitch: “‘We’re proposing to spend 72 hours imaging the universe with three scientific instruments on JWST. Based on our simulations, we know for certain that something exciting will show up. So please give us the time.’ And they did.”

Synthetic images of an ultra-deep galaxy survey, with a side-by-side comparison at depths expected to be reached by CEERS (left) and NGDEEP (right). Courtesy of Micaela Bagley,

Second, they wanted to make sure they made the best use of that time. Their survey strategy included choosing the right fields of view, exposure times and light filters. That all depended on what they expected to see. While working out these details, they also created mock images in order to test their data processing pipeline; this included combining images taken through different filters to form color images, as well as extracting galactic properties from those images. In their model, they could create galaxies of known size and distance (“ground truth” properties), then simulate JWST images of them. They compared the extracted properties from those synthetic images with the ground truth until they could reliably extract properties accurately.

Performing these tasks before collecting actual JWST data does more than just expedite data interpretation: It also enhances confidence in the theory undergirding their predictions. “Theorists are often accused of only making postdictions and being guilty of tweaking our models to just fit the observations after we know what they are,” Somerville says. There could be many explanations for a given observation, and you can’t know that you’ve picked the best one, one that will match future observations. If instead you pick a theory before receiving real data, and they match, it’s more likely you’ve nailed the physics.

Somerville began developing the model in 1996, when plans for JWST were still in their infancy. One commonly used approach to modeling galaxy formation is what’s called a numerical simulation. This type of simulation represents matter as particles or filled cells in a grid, and then applies the laws of gravity, hydrodynamics and thermodynamics at each time step. Each simulation can take months to run. Somerville’s model instead is a semi-analytic model (SAM). It’s essentially a bookkeeping exercise, with account values indicating the amount of hot gas or the number of stars or other quantities. A set of theory-based formulas defines how each account (or “reservoir”) influences other accounts. The model ignores where each bit of mass exists in space, making it highly efficient.

View of the simulated universe from the front, just like the way we see the universe. The simulated field has perimeters similar to the observed Hubble Ultra-Deep Field (left). We show a comparison of the simulated field at depths reachable by Hubble (middle) and Webb (right). Credit: Yung et al.

“So instead of tracking millions of particles,” Yung says, “we now track millions or billions of galaxies, which is really neat. We are able to simulate the number of galaxies that actually show up in a survey.” They can also change their physics assumptions, such as how efficiently stars form, and rerun the model. Instead of running a numerical simulation once, they can run a SAM thousands of times. They’ve been using supercomputers at NASA and the Simons Foundation.

Somerville began developing the SAM while a doctoral student at the University of California, Santa Cruz (it’s known as the Santa Cruz SAM), and much of the original framework remains, though she’s made important revisions over the years. Yung notes that his role has been to use the model to make predictions tailored to upcoming JWST surveys and provide support to observing teams and research groups who want to use it, a task he lightly refers to as “customer service.”


Explaining space’s transparency

The farther away galaxies are from us, the farther back in time we’re looking — and the faster they’re moving away from us as the universe expands. As objects recede, their light waves stretch, becoming redder. Astronomers call this phenomenon a redshift. We observe objects with redshifts of 6 to 10 as they were roughly 13 billion years ago. When Yung started his doctoral work, the team anticipated that JWST would be launching in a few years, and that this very early time in the universe would be an interesting topic to study.

In their six papers, Yung, Somerville and collaborators present a broad variety of predictions about these early galaxies. They first make predictions about the observable physical properties of galaxy populations at these epochs. They then show that these predictions agree remarkably well with existing observations.

Side-view of the simulated universe as presented in the “Semi-analytic forecasts for JWST” project (Yung et al. 2022). Each data point represents a galaxy. Larger and darker data points represent galaxies with more mass, and vice versa. The arrows indicate the directions of which the light travels towards us and the "look back effect" from our perspective which enables our study of the universe's past history. Credit: Yung et al.

Early in the universe, neutral hydrogen gas absorbed much of the universe’s radiation, rendering space nearly opaque. But something knocked the electrons off the hydrogen, reionizing it and letting light through. “A major theme of this work, and one of the main reasons that JWST was built, was to try to identify the sources of reionization,” Somerville says. Yung and Somerville showed that their physically grounded model of galaxy formation could reproduce direct measurements of luminous galaxies. Further, the point in cosmic time when the simulation predicted reionization would be complete agreed with the time derived from observations. They also quantified which kinds of galaxies were responsible for producing ionizing photons at different times in cosmic history.


Earth to Maisie

The first five papers consider many parameters to explain the underlying physics of the universe. The sixth picks the most probable parameters and makes the model easy to use by observers. Importantly, it also presents a simulated JWST-like “light cone.” Many simulations output galaxies in snapshots, with their properties as they were at distinct cosmic epochs during the simulation’s evolution in time. However, light cones present the universe along the paths that light takes as it travels to us from distant objects. “The light cone just bridged the gap between boxes to the real universe we see,” Yung says.

In July, Yung, Somerville and other researchers processed some of the first images taken by JWST as part of the CEERS project and announced the discovery of one of the oldest galaxies, from 290 million years after the Big Bang. They called it Maisie’s galaxy, after the project lead’s daughter. The team hadn’t expected many galaxies to exist at such early times. “The universe is full of surprises,” Yung says.

The modeling legwork is paying off. As the researchers look at the data JWST has provided so far, “the most surprising thing is that there are a lot more luminous galaxies at these very early times than we expected,” Somerville says. Maybe dark matter, the invisible stuff that composes 85% of the universe’s mass and contributes to the formation of galaxies, doesn’t work the way we thought. Or maybe the way gas turns into stars is different than hypothesized. People are now using Somerville’s model not to predict what we’ll see but to comprehend what we’ve just seen.


Author’s note: The research team is moving beyond the JWST and expanding their support to other flagship telescopes, including NASA’s Roman Space Telescope, scheduled for launch in 2027. Going forward, the project will be known as ‘Semi-analytic forecasts for the universe‘.


For deeper reading:

Yung LYA, Somerville RS, Finkelstein SL, et al. Semi-analytic forecasts for JWST – I. UV luminosity functions at z = 4–10. Monthly Notices of the Royal Astronomical Society. 2019;483(3):2983–3006.

Yung LYA, Somerville RS, Popping G, et al. Semi-analytic forecasts for JWST – II. Physical properties and scaling relations for galaxies at z = 4–10. Monthly Notices of the Royal Astronomical Society. 2019;490(2):2855–2879.

Yung LYA, Somerville RS, Popping G, et al. Semi-analytic forecasts for JWST – III. Intrinsic production efficiency of Lyman-continuum radiation. Monthly Notices of the Royal Astronomical Society. 2020;494(1):1002–1017.

Yung LYA, Somerville RS, Finkelstein SL, et al. Semi-analytic forecasts for JWST – IV. Implications for cosmic reionization and LyC escape fraction. Monthly Notices of the Royal Astronomical Society. 2020;496(4):4574–4592.

Yung LYA, Somerville RS, Finkelstein SL, et al. Semi-analytic forecasts for JWST – V. AGN luminosity functions and helium reionization at z = 2–7. Monthly Notices of the Royal Astronomical Society. 2021;508(2):2706–2729.

Yung LYA, Somerville RS, Ferguson HC, et al. Semi-analytic forecasts for JWST – VI. Simulated light-cones and galaxy clustering predictions. Monthly Notices of the Royal Astronomical Society. 2022;515(4): 5416–5436.


Recent Articles