Dark energy flips its sign, but the Hubble tension refuses to budge

For nearly a century, astronomers have known that the universe is expanding. In the late 1990s, two independent teams, the Supernova Cosmology Project, led by Saul Perlmutter, and the High-Z Supernova Search Team, led by Brian Schmidt and Adam Riess, discovered something strange: The expansion is speeding up. The finding earned them the 2011 Nobel Prize in Physics. The leading explanation for this acceleration is “dark energy,” a mysterious force usually modeled as a constant called Lambda, pushing space apart. Combined with cold dark matter, this gives us the LCDM model, the standard picture of the cosmos for the past 25 years.

LCDM is remarkably successful. It fits observations of the cosmic microwave background (CMB), i.e., the leftover glow from the Big Bang, as well as maps of galaxy clustering and the brightness of exploding stars called Type Ia supernovae. But it has one nagging problem: the Hubble tension.

Cosmologists have proposed dark energy that switches sign over cosmic history. A rigorous new analysis published in Physical Review D checks whether it actually closes the gap.

The Hubble tension

The Hubble constant, H₀, describes how fast the universe is expanding today. There are two main ways to measure it. One uses the CMB, essentially “predicting” today’s expansion rate from the physics of the early universe. The other measures it directly, using nearby supernovae calibrated against pulsating stars called Cepheids. These two methods disagree persistently by five to seven standard deviations, far too much to be a coincidence or mere measurement error. Something in our picture of the universe may be missing.

This mismatch has fueled a decade of proposed fixes, from new particles to modified gravity. One popular family of ideas suggests dark energy itself isn’t constant; it changes over cosmic history.

Enter LsCDM: Dark energy that flips sign

One such proposal, called LsCDM, keeps most of LCDM’s ingredients but adds one twist. Instead of always pushing space apart, the cosmological constant is imagined to have once been negative, actually pulling matter together, almost like ordinary gravity, before flipping to positive at some point in the universe’s history, roughly when the universe was less than one-third of its current age. After that flip, it behaves like normal dark energy, pushing space apart as usual. Earlier studies have reported that this single change can ease both the Hubble tension and a related mismatch called the S₈ tension, without hurting the model’s success at explaining the early universe.

But does “easing tension” really mean the models agree?

This is where our new study steps in. Our concern isn’t with the physics of LsCDM itself, but with how cosmologists typically measure “tension” between datasets in the first place.

The usual approach treats the spread of measurements as a bell curve (a Gaussian distribution) and asks how many standard deviations separate two results, much like calculating a batting average’s margin of error. But real cosmological data don’t always behave like tidy bell curves. When one dataset is very precise and another is comparatively loose and lopsided, this simplified approach can badly overstate or understate how serious a disagreement really is.

To test this, we combined the latest CMB data (from the Planck satellite, the Atacama Cosmology Telescope and the South Pole Telescope), the newest galaxy-clustering measurements from the Dark Energy Spectroscopic Instrument (DESI), and the Pantheon Plus supernova catalog calibrated with the SH0ES project’s local Cepheid measurements.

We ran both the standard LCDM model and the LsCDM extension through several statistical tests, not just the standard “rule of thumb,” but also more rigorous techniques that don’t assume Gaussian statistics, plus a check called posterior predictive testing, which asks: If this model were true, how likely is it that we would have measured what we actually measured?

What we found

Our findings tell two very different stories. On the one hand, the picture is unambiguously reassuring: Once we move beyond crude statistical shortcuts and use exact, non-Gaussian methods, the CMB and galaxy-clustering data turn out to agree with each other remarkably well in both models. The foundation of early-universe cosmology is solid.

But throw in measurements from nearby supernovae and the story changes. Both LCDM and LsCDM have real, unsolved tensions. Yes, the sign-flipping model does shift predictions in the right direction, toward the locally measured expansion rate, but the observed H₀ still lies in a surprisingly unlikely region of the model’s predictions. Progress, not standing still.

This is important beyond the specific models we tested. It is a good reminder that claimed breakthroughs in resolving cosmological tensions can sometimes be just an artifact of oversimplified statistics, not real physical insight. Rigorous, non-Gaussian diagnostics cut through that ambiguity: LCDM earns a real but partial victory, but the Hubble tension remains stubbornly unresolved. It will require better data, sharper theory or both to crack it completely.

This story is part of Science X Dialog, where researchers can report findings from their published research articles. Visit this page for information about Science X Dialog and how to participate.

Who’s behind this story?

Lisa Lock

BA art history, MA material culture. Former museum editor, paramedic, and transplant coordinator. Editing for Science X since 2021.

Full profile →

Andrew Zinin

Master’s in physics with research experience. Long-time science news enthusiast. Plays key role in Science X’s editorial success.

Full profile →

Sehjal Khandelwal is a doctoral student at Plaksha University. She is currently working on constraining cosmological models using early- and late-time datasets, in both linear and non-linear regimes. A part of her work also contributes to the statistical techniques used in the process.

Dr. Abraão Jessé Capistrano de Souza is a Brazilian physicist and professor at the Federal University of Latin American Integration (UNILA). His academic work centers heavily on mathematical physics, general relativity, and cosmology, with a specific focus on the dark sector and the geometric models of space-time.

Dr. Suresh Kumar is a Professor of Mathematics at Plaksha University. He works extensively on Bayesian data analysis of cosmological models using observational datasets, employing Monte Carlo Markov Chain (MCMC) methods within CLASS and MontePython, supported by high-performance Linux-based computing clusters.