Is dark energy running out of steam?
The largest three-dimensional map of the universe hints that the force driving cosmic acceleration may be weakening. A 2026 reanalysis has made that evidence weaker, and more interesting.
In 1998, two teams studying distant supernovae found that the expansion of the universe is speeding up. The cause was given a placeholder name, dark energy, and the simplest explanation was Einstein’s cosmological constant: an energy of empty space that never changes. For a quarter of a century, that explanation fit almost every observation.
The Dark Energy Spectroscopic Instrument, known as DESI, has begun to suggest otherwise. Its analysis of more than 14 million galaxies and quasars, released in March 2025, prefers a dark energy that was stronger in the past and has been weakening. Combined with other data, the preference ranged from about 2.8 to 4.2 standard deviations, depending on which set of supernova measurements was included.
That dependence turned out to matter. The highest figure came from pairing DESI with a supernova catalog from the Dark Energy Survey, and in 2026 that catalog was reanalyzed.
A ruler written in the sky
DESI sits on the four-meter Mayall Telescope at Kitt Peak in Arizona. It uses 5,000 robotically positioned optical fibers to collect the spectra of thousands of galaxies at once, measuring how far their light has been stretched by cosmic expansion and so how far away they are.
The key measurement is a pattern called baryon acoustic oscillations. Sound waves in the early universe, the same waves imprinted on the cosmic microwave background, left galaxies very slightly more likely to be separated by about 480 million light-years. At a redshift of 1, when the universe was a little under half its present age, that separation spans about two and a half degrees on the sky, five times the width of the full Moon. By measuring how large it appears at different distances, DESI can trace how fast the universe expanded at different times.
One number, two ways of describing it
Cosmologists summarize dark energy with a number called w, the ratio of its pressure to its density. A cosmological constant has w equal to exactly −1, forever. The evolving models allow two parameters: w0, its value today, and wa, how it changes with time. Combining the reanalyzed supernovae with DESI and the microwave background gives w0 = −0.803 ± 0.054 and wa = −0.72 ± 0.21.
The reanalysis, using an improved calibration, cut the significance of the departure from a cosmological constant from 4.2 to 3.2 standard deviations. Here it is useful to hold two statements of the same result side by side. Three point two sigma corresponds to a chance of roughly 1 in 700 that noise alone would produce such a departure, which sounds decisive. But the same paper’s Bayesian comparison, which weighs the fact that the evolving model has more knobs to turn, gives odds of about 5 to 1 in favor of evolution. Those are the odds of a modest bet, and the authors themselves describe the preference as weak.
Neither figure is wrong. They answer different questions. The first asks how surprising the data are if nothing is going on. The second asks how much you should shift your belief between two specific stories. Headlines tend to report the first and readers tend to hear the second.
Headlines tend to report the first and readers tend to hear the second.
The strange shape of the hint
The physics implied by the fit is odd, and that deserves attention. The simplest models of a changing dark energy, slowly rolling fields, keep w above −1. The data prefer a history in which w was below −1 in the past and crossed to above it more recently, a behavior that is hard to obtain from simple theories. Some researchers read this as a clue that the analysis is off, and others as a sign that the physics is richer than anyone assumed.
An independent 2026 study found that with the revised supernova data, the statistical preference for many evolving models falls below the usual thresholds. The evidence has not disappeared. It has simply stopped looking like a discovery in waiting, and become a question that will be settled by better data rather than by the data we have.
What is at stake
If the cosmological constant really is constant, the universe’s future is simple: expansion accelerates forever, and galaxies beyond our local group eventually drift out of contact. If dark energy can change, then the long-term future is open again, and acceleration could eventually stop or even reverse. Very little in cosmology has a larger consequence per unit of decimal place.
What comes next
DESI will keep observing, and its final data set will be substantially larger. The European Space Agency’s Euclid telescope, launched in 2023, is mapping galaxy shapes and positions across a third of the sky. The Vera C. Rubin Observatory in Chile, which released its first images in 2025, will discover supernovae in enormous numbers over its ten-year survey, with a single instrument and calibration, addressing precisely the weakness the reanalysis exposed. Within a few years the question should have a clearer answer, and cosmologists will find out whether the most successful model in their field needs its first major revision since dark energy was discovered.
Sources
Keep reading
More from Cosmos
How old is the universe? 13.8 billion years, and three ways we know
The number comes from running the expansion of space backwards. We did that calculation ourselves, then checked it against the oldest stars, a clock that no expansion measurement can touch.
One light year is about 9.46 lakh crore kilometres. Here is how far that really is
A light year is a distance, not a time. We worked out how long it would take to travel one by train, by plane and by spacecraft, and how far India’s own missions are in light-time.
The universe is expanding at two different speeds
Measure the expansion rate from the early universe and you get one number. Measure it from nearby galaxies and you get another. A decade of better data has made the gap harder to explain away.
Comments
No comments yet. Start the conversation.