Last year, the Dark Energy Spectroscopic Instrument (DESI) unveiled findings that could challenge our understanding of the universe. Initial results from its first-year observations revealed hints that dark energy—a mysterious force believed to be driving the accelerated expansion of the cosmos—may not be as constant as previously thought.

While scientists continue to validate the findings, the data suggests that dark energy could be evolving over time. If confirmed, this would change cosmology as we know it. Researchers from the DESI collaboration plan to share updated results at the American Physical Society’s Global Physics Summit this week.

“If the DESI result holds up, it means that a cosmological constant is not the origin of cosmic acceleration. It’s much more exciting,” said Andrew Hearin, a physicist at the U.S. Department of Energy’s (DOE) Argonne National Laboratory, which is a DESI member institution. “It would mean that space is pervaded by a dynamically evolving fluid with negative gravity, which has never been observed in any tabletop experiment on Earth.”

To help investigate the observations, Hearin and colleagues turned to the Aurora exascale supercomputer at the Argonne Leadership Computing Facility (ALCF), a DOE Office of Science user facility. By running large-scale simulations of the universe, they are providing a testing ground for researchers to analyze the DESI data.

“When this result came out last year, we got really, really excited,” said Katrin Heitmann, a cosmologist and deputy director of Argonne’s High Energy Physics division. “Our team got together to discuss what we could do to help the community look into this from the simulation side. Simulations play a crucial role in disentangling fundamental physics from systematics in the observations or in the data analysis.”

DESI: Shedding light on dark energy

DESI, located at Kitt Peak National Observatory in Arizona, is an international collaboration of over 900 scientists led by DOE’s Lawrence Berkeley National Laboratory. Mounted on the Mayall Telescope, DESI was designed for spectroscopic surveys. Instead of capturing images, it collects light spectra from distant galaxies to determine their structure, distance, and movement. In its first year of operation, DESI has enabled researchers to create the most detailed 3D map of the universe ever made and track its expansion over the past 11 billion years.

DESI’s initial data largely align with the standard model of cosmology (our current understanding of the universe), but small discrepancies suggest dark energy might be changing over time. This would challenge the current theory for cosmic acceleration: the cosmological constant.

First introduced by Albert Einstein in his equations for general relativity in 1917, the cosmological constant was meant to counteract gravitational collapse and support the idea of a static universe. However, the discovery about a decade later that the universe is expanding led scientists to abandon the idea.

The concept was revived in the 1990s when observations showed that the universe was not only expanding but doing so at an increasingly faster rate. While the simplest explanation was to bring back the cosmological constant, the nature of dark energy remains one of the greatest mysteries in physics.

“The cosmological constant is essentially just an extra term in the equation that everyone has been using for years,” Hearin said. “We don't know why it takes on the particular value that it does, but it’s a very mundane explanation for this unexpected cosmic acceleration.”

Simulations provide test bed for cosmic exploration

A key challenge in cosmology is determining whether observed patterns in the universe are real or distortions caused by how we collect and analyze data. Simulations help researchers address this issue, offering a controlled environment to test different scenarios.

“Since we can’t create a mini universe to conduct experiments, we can test theories by using really big computers like Aurora to simulate the growth of structure in the universe over time,” said Gillian Beltz-Mohrmann, postdoctoral research fellow at Argonne.

To examine the DESI results, the Argonne team used Aurora to run two large-scale simulations—one assuming a constant dark energy and another where it changes over time. Both started with identical initial conditions, enabling researchers to track even the smallest differences as they evolved.

“The idea is that you create a model universe under one set of assumptions, and then you compare your model universe to the real universe. If the agreement is very good, it gives you some confidence that your assumptions are correct,” Hearin said. “But if you have some gross discrepancy, then it tells you that your assumptions don’t align with the real universe and don’t represent the truth.”

While simulations can’t directly confirm DESI’s findings, they provide a critical feedback loop between theory and observation. By testing different measurements within the simulations, researchers can refine their analysis techniques and assess whether the observed patterns could emerge from systematic effects rather than new physics.

“If looking at these two simulations gives us an idea of the type of measurement we should make to help us narrow in on a cosmological model, then we can go back to the real DESI data and make that same measurement and see what it tells us,” Beltz-Mohrmann said. 

Bringing the universe into focus with Aurora

The scale and complexity of running massive cosmological simulations at high resolution over vast volumes can lead to long runtimes, but Aurora allowed the team to complete them quickly.

“Using Aurora’s immense processing power to rapidly run large-scale simulations at sufficiently high resolution, we can respond much faster to new insights from cosmological observations,” said Argonne computational scientist Adrian Pope. “These simulations would have taken weeks of compute time on our earlier supercomputers, but each simulation took just two days on Aurora.”

Achieving high resolution was particularly critical for this study, as Heitmann explained: “Aurora is extremely important because to detect these fine differences, we need incredibly high resolution in our simulations. Without that level of resolution, details can get washed out. Think of it like taking off your glasses and trying to make out a blurry figure.”

To further accelerate their work, the team employed a method called on-the-fly analysis, which allowed them to process simulation data as it was being generated. This approach eliminated bottlenecks in storing and post-processing simulation data, leveraging Aurora’s power to extract insights faster and refine simulations in real time.

“This pair of simulations really illustrates our ability to take a result that’s hot off the presses from a collaboration like DESI, immediately run a simulation based on those results, and then see what it looks like,” Beltz-Mohrmann said.

While DESI’s findings continue to be scrutinized, the team’s simulations provide a valuable tool for refining analysis techniques and exploring new ways to interpret the data. To enable additional studies, the team has made the simulation data publicly available.

“This is a new caliber of simulations in the field,” Hearin said. “Essentially, it provides the community with a test bed to try out ideas for distinguishing between these two universes. The simulations can offer insight into how we might use DESI data to determine which of these competing models of dark energy is the truth.”

The team detailed their work in the study, “Illuminating the Physics of Dark Energy with the Discovery Simulations.”

DESI is supported by the DOE Office of Science. The Argonne team, including Heitmann, Hearin, Beltz-Mohrmann, Pope, Alex Alarcon, Michael Buehlmann, Nicholas Frontiere, Sara Ortega-Martinez, Alan Pearl, Esteban Rangel, Thomas Uram, and Enia Xhakaj, performed the simulations with support from the ALCF’s Aurora Early Science Program and DOE's Exascale Computing Project. Their work was also partially supported by DOE’s Scientific Discovery through Advanced Computing (SciDAC-5) program and NASA’s OpenUniverse project.