Cosmologists have spent decades trying to pin down exactly how fast the universe is flying apart and whether that rate is changing. A new AI-driven method called GAME, short for Genetic Algorithms with Marginalised Ensembles, now claims it can reconstruct the expansion history of the cosmos with dramatically higher precision than existing tools, reporting roughly 95% better accuracy on the derivative quantities that reveal whether dark energy is shifting over time. The technique, detailed in a paper posted to the arXiv preprint server in early 2026, lands just as next-generation sky surveys are flooding the field with fresh data. A note on terminology: the headline refers to “universe-evolution simulations,” but GAME is more precisely a reconstruction tool. Rather than running forward N-body simulations of cosmic structure, it works backward from observational data to reconstruct the universe’s expansion history. The word “simulations” here is a simplification of that reconstruction process.
Why derivatives matter more than you might think
The universe’s expansion rate at different epochs is captured by a value called the Hubble parameter, written as H(z), where “z” represents redshift, a proxy for how far back in time astronomers are looking. Measuring H(z) itself is useful, but the real prize is its derivative: how quickly the expansion rate changes from one epoch to the next. That derivative is what separates a universe powered by a simple cosmological constant, steady and unchanging, from one driven by a dynamic form of dark energy that strengthens or weakens over billions of years.
Traditional reconstruction methods often assume a specific cosmological model before fitting data, which can bias results toward the very framework researchers are trying to test. GAME sidesteps that problem. It runs an ensemble of genetic-algorithm configurations, each one evolving candidate solutions through selection, crossover, and mutation, then combines their outputs through weighted averaging. According to the paper, this ensemble-and-marginalize strategy delivers about 20% improvement in reconstructing H(z) itself and approximately 95% improvement for its derivatives, the quantities most sensitive to dark energy’s behavior.
Genetic algorithms borrow their logic from biological evolution. A population of candidate solutions “competes” across generations; the fittest survive and recombine. In GAME’s case, each configuration explores a slightly different path through the data, and the final answer reflects the collective wisdom of the entire population rather than a single best guess. That built-in diversity is what helps the method avoid locking onto one cosmological model prematurely.
The data pipeline feeding the method
GAME draws on a technique called cosmic chronometry, which estimates H(z) by measuring the ages of the oldest galaxies at different redshifts. If you know how much older one galaxy population is than another at a slightly different distance, you can infer how fast the universe was expanding in between.
Recent cosmic chronometer work has sharpened those age estimates. A study led by Michele Moresco and colleagues at the University of Bologna provided an H(z) measurement at a redshift of roughly 0.7, corresponding to a universe about 6.5 billion years old, using careful sample selection and age-redshift modeling. “The cosmic chronometer approach provides a direct measurement of the expansion rate that does not depend on any assumed cosmological model,” the authors wrote, underscoring the method’s value for testing competing dark energy theories. A separate analysis of galaxies in the LEGA-C survey reported an H(z) value at redshift 0.8 using full-spectrum fitting, a newer technique that analyzes the entire light profile of a galaxy rather than relying on a handful of spectral features. The two approaches agree on broad trends, which strengthens confidence that the underlying expansion signal is real even as individual values carry documented uncertainties.
Combining measurements from different pipelines introduces a subtle hazard: shared assumptions in stellar population models can create correlated errors that look independent if you are not careful. A dedicated study by Moresco and collaborators quantifying the impact of stellar population synthesis model choices addressed this by constructing a full covariance matrix for H(z) data, essentially a map of which errors are entangled. GAME’s ensemble design is built to absorb that complexity, marginalizing over multiple configurations rather than trusting any single modeling pipeline.
On the observational side, the Dark Energy Spectroscopic Instrument (DESI), a Stage IV survey backed by the U.S. Department of Energy, is generating the most detailed three-dimensional map of the universe ever attempted. DESI’s official data release documentation confirms that supporting products, including cosmology analysis chains, are available for outside researchers. That open-access policy means independent teams could, in principle, test GAME against real DESI outputs.
What has not been proven yet
The 95% figure comes from the GAME paper’s own controlled benchmarks, not from application to live survey data. No independent research group has publicly replicated the result, and the paper has not yet passed formal peer review. Those caveats do not invalidate the work, but they place it in the same category as a promising laboratory prototype: technically sound on paper, unproven in the field.
Computational cost is another open question. The paper does not report how GAME’s runtime scales when confronted with the full data volume that Stage IV surveys like DESI produce. A method that delivers superior accuracy but takes prohibitively long to run on large catalogs would have limited practical value. No public statements from DESI collaboration leadership indicate whether GAME integration is being explored.
There is also a ceiling that no statistical technique can break through on its own. Stellar population models, the astrophysical recipes used to estimate galaxy ages, carry their own irreducible uncertainties. Better reconstruction of H(z) derivatives could reveal subtle shifts in dark energy’s equation of state over cosmic time, but if the underlying age estimates are noisy enough, the signal may remain buried. The covariance framework from earlier chronometer research helps contain that problem; it does not eliminate it.
What independent replication on DESI data would prove
For researchers tracking the dark energy question, the practical milestone to watch for is straightforward: a follow-up study that applies GAME to actual DESI data releases or to outputs from another Stage IV survey such as the European Space Agency’s Euclid mission. If the 95% derivative improvement holds up on real observational data rather than simulated benchmarks, it would give cosmologists a meaningfully sharper lens for distinguishing between a static cosmological constant and a dark energy component that evolves.
That distinction matters beyond academic curiosity. Whether dark energy is constant or dynamic determines the long-term fate of the universe: eternal acceleration, a gradual slowdown, or something stranger. GAME offers a new way to interrogate the data, but the data still has to cooperate. As of spring 2026, the method is a well-constructed tool waiting for its first real-world stress test.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
