Deep divefrom Episode 8Cosmology23 October 202620 min

Dark Energy, or Gravity Slightly Weaker Out There?

A deep dive into one paper: what was known before, what the researchers did, what the numbers say and what they don't.

The paper: Is DESI Seeing Dynamical Dark Energy, or a Cosmic Glitch in Gravity? (arXiv preprint)

Chapters

Transcript

The script the two AI voices read from. The times are worked out from the script, not heard, so a line can start a few seconds early or late.

Read the whole episode

Cold open

Host 1You heard the trailer — this is the whole story. Welcome to the Hastrup Review of Science — where artificial intelligence reads real science, so you don't have to pretend you understood the paper. Today, one paper that asks whether the hint of evolving dark energy from the DESI survey could instead be gravity behaving very slightly differently on the scale of the whole universe.

Host 2The paper is by Robin Wen, Lukas Hergt, Niayesh Afshordi and Douglas Scott, posted to the arXiv on the first of October twenty twenty-six, so it is a preprint. Their claim is that one extra number, describing gravity as about eight tenths of a percent weaker for the universe as a whole than in the lab, fits current data at least as well as the two-number dark energy model everyone has been using.

Why this field exists

Host 1Before the glitch, the background. Why does anyone measure dark energy in the first place?

Host 2As established background: in nineteen ninety-eight, two supernova teams found that distant exploding stars were fainter than expected, which meant the expansion of the universe is speeding up. That discovery earned the twenty eleven Nobel Prize in physics. Whatever drives the acceleration got the placeholder name dark energy.

Host 1And the simplest version of dark energy is Einstein's cosmological constant.

Host 2Yes. A constant energy of empty space, the same everywhere and at all times. Combined with cold dark matter and ordinary matter, that gives the standard model of cosmology, called Lambda cold dark matter, and it has fitted almost everything for a quarter of a century.

Host 1Where do the measurements come from?

Host 2Two main sources in this paper. The first is the cosmic microwave background, light released about three hundred and eighty thousand years after the big bang, mapped by the Planck satellite and by two ground telescopes, the Atacama Cosmology Telescope in Chile and the South Pole Telescope. The second is baryon acoustic oscillations, a fixed ruler imprinted on how galaxies cluster.

Host 1Explain the ruler.

Host 2In the early universe, sound waves rippled through the hot plasma and froze in place when it cooled. They left a preferred separation between clumps of matter, roughly five hundred million light-years today, called the sound horizon. Measure that separation in galaxy maps at different distances, and you get the expansion history of the universe, the same way a known ruler at different distances tells you how far away it is.

Host 1And DESI is the survey doing that now.

Host 2DESI is the Dark Energy Spectroscopic Instrument on a telescope in Arizona, recording spectra of millions of galaxies and quasars. This paper uses its second data release, which covers the first three years of observations, measured with seven separate tracer samples spanning from nearby galaxies out to the glow of distant quasars.

What was known before

Host 1So what did DESI see that made people talk about evolving dark energy?

Host 2Again as background: when DESI's distances are combined with the microwave background, and often with supernovae, the fit mildly prefers dark energy that changes over time rather than a constant. The paper calls this a mild but persistent preference for departures from the standard model. The significance quoted in the literature depends heavily on which supernova catalogue is included.

Host 1How is that departure usually written?

Host 2With a two-parameter recipe for the dark energy equation of state, named after Chevallier, Polarski and Linder. One number sets its present value, and a second number sets how fast it drifts as the universe expands. The authors stress that this is an effective description, not a theory: it adds two dials and says nothing about what physics is turning them.

The open question

Host 1So the open question is whether those two dials are actually needed.

Host 2That is how the authors frame it. Are the data asking for two-parameter evolving dark energy, or could they be responding to one physical mismatch between gravity on local and cosmological scales?

Host 1What does a mismatch in gravity even mean?

Host 2There are two places Newton's constant shows up. One is local gravity, what you measure with a torsion balance or by tracking planets. The other is the Friedmann equation, which sets how fast the universe as a whole expands given what is in it; in general relativity the two are the same number.

Host 1And the cosmic glitch lets them differ.

Host 2Exactly that and nothing more. The model, which these same authors developed in earlier papers building on a twenty oh eight proposal co-authored by Afshordi, adds one parameter, called omega g, that measures the ratio between the cosmological and local constants. Omega g of zero is the standard model; a negative value means the smooth universe gravitates slightly more weakly than the solar system does.

Host 1Is that just a convenient knob, or does it come from a theory?

Host 2It is not entirely ad hoc. A difference between the two couplings is a generic low-energy signature of gravity theories with a preferred slicing of spacetime into moments of time, including Horava–Lifshitz gravity, Einstein-aether theory and a model called cuscuton gravity. In the Horava language, a negative glitch corresponds to a parameter called lambda being slightly above its general-relativity value of one.

Host 1Doesn't local physics already rule out gravity theories like that?

Host 2Many versions, yes. Solar System tests constrain preferred-frame effects to about one part in ten thousand and one part in ten million, and the twenty seventeen neutron star merger showed gravitational waves travel at the speed of light to about one part in a million billion. The authors argue that in the limit where the theory's extra field responds instantly, those tests constrain its propagation speed rather than the glitch, so only cosmology can see the glitch.

Host 1Any prior limit from cosmology itself?

Host 2Big bang nucleosynthesis, the formation of light elements in the first minutes, probes the coupling in the Friedmann equation directly. It allows a mismatch of up to about six percent at two sigma. Earlier cosmological analyses quoted tighter percent-level limits, which is the regime this paper is in.

What this paper did

Host 1So what did they actually do?

Host 2They fitted three models to the same data: the standard model, the standard model plus the glitch, and the standard model plus two-parameter evolving dark energy. For each they sampled the six usual cosmological parameters plus whatever the extension adds, with one massive neutrino of sixty milli-electronvolts.

Host 1Which data, specifically?

Host 2For the microwave background, a combined likelihood called Hillik: Planck's fourth data release at large angular scales, the Atacama Cosmology Telescope's sixth release at smaller scales, plus the South Pole Telescope, all with consistent modelling of foreground dust and galaxies. On top of that, DESI's second release of acoustic distances, and in an extended run, gravitational lensing of the microwave background and the Dark Energy Survey's Dovekie supernova compilation.

Host 1How do you stop Planck and the Atacama telescope from counting the same sky twice?

Host 2They split by angular scale. Planck is used below a certain multipole, roughly the level of detail, and the Atacama telescope above it, with the cut chosen at the crossover where the ground telescope's errors drop below Planck's. The South Pole Telescope covers a small patch, and its overlap with the other two is treated as negligible, which is an approximation.

Host 1And the machinery?

Host 2Predictions come from a modified version of a standard cosmology code, with the sampler Cobaya running Markov chains until a convergence statistic falls below one percent. Bayesian evidence, which penalises extra parameters, is computed with nested sampling. The Hillik likelihood carries about seventy nuisance parameters, so for the expensive model comparison they switch to a cheaper, pre-marginalised version.

Host 1There is a subtle problem with treating a gravity change as dark energy, isn't there?

Host 2Yes. In the code, the glitch is represented as an effective dark energy component, and for negative omega g its density is negative at early times and crosses zero before settling into a cosmological constant. At the crossing, its equation of state formally divides by zero.

Host 1Is that a real instability?

Host 2The authors call it a bookkeeping singularity of the fluid description, not a physical instability. They handle perturbations with a standard parameterised scheme and a sound speed equal to light speed as the default, then test a much higher sound speed as a stand-in for the instant-response cuscuton limit, and a run where the sound speed is left free.

Host 1One more design choice I want to flag: the prediction test.

Host 2That is the most interesting part of the method. Besides the joint fit, they fitted each model to the microwave background alone, with no galaxy data, and then asked what each model predicts DESI should measure. A joint fit flatters models with more dials, because two free parameters can absorb almost any smooth distance curve; a prediction cannot.

Results

Host 1Results. What is the headline number?

Host 2With the microwave background plus DESI, omega g comes out at minus zero point zero zero eight one, plus or minus zero point zero zero two five. That is three point three sigma away from zero. Translated, the cosmological gravitational constant is ninety-nine point two percent of Newton's, a weakening of about eight tenths of a percent.

Host 1Does it survive more data?

Host 2Adding lensing and supernovae gives minus zero point zero zero seven three, plus or minus zero point zero zero two two, still three point three sigma. In Horava terms, that is lambda about half a percent above one.

Host 1And the microwave background on its own?

Host 2It already prefers the same value, minus zero point zero zero seven three, at two point one sigma. DESI tightens the constraint without moving its centre, which is what you would expect if both data sets are seeing the same thing.

Host 1Is this a fluke of one new data set?

Host 2The authors trace it over time. Planck twenty eighteen with lensing gave one point nine sigma, Planck's fourth release with lensing gave one point three, adding DESI's first release brought it to two point three, the second release to two point zero with Planck alone, and adding the ground telescopes brings it to three point three. The trend toward negative values is there throughout, though it has not grown smoothly at every step.

Host 1What did the other cosmological parameters do?

Host 2The Hubble constant comes out at sixty-eight point three one, within about a tenth of the standard-model value, so this does not address the Hubble tension. The matter fraction is about thirty percent. By contrast, the evolving dark energy model fitted to the microwave background and DESI alone pulls the Hubble constant down to around sixty-four, with wide uncertainty, until supernovae pull it back to about sixty-seven.

Host 1Now the prediction test. What happened?

Host 2The standard model fitted to the microwave background predicts distances that sit systematically above DESI's measurements below a redshift of about one. The glitch model, fitted only to the microwave background, predicts a one to two percent downward shift of exactly that form, with an uncertainty band about as narrow as the standard model's, and it passes through the DESI points.

Host 1And evolving dark energy?

Host 2Fitted to the microwave background alone, it makes essentially no prediction: its band spans several percent in either direction, because its two dials are unconstrained without nearby data. The authors' summary is that the glitch is not merely a fit to DESI but a prediction from the early universe that DESI confirms.

Host 1They also looked at the known disagreement between the microwave background and DESI.

Host 2In the standard model, the two disagree at two point one sigma on the combination of matter density and the sound-horizon ruler, with DESI preferring less matter. With the glitch, the microwave background constraint stays tight, the DESI constraint loosens along one direction, and the two overlap, leaving a tension of zero point two sigma. With evolving dark energy, both constraints become individually loose and actually drift apart in direction, and the tension only drops to one point three sigma.

Host 2The line from the paper is that a single parameter that makes two data sets agree is more informative than two parameters that make each of them uninformative.

Host 1How does the head-to-head scoring come out?

Host 2On goodness of fit, the glitch improves the chi-squared by eight point six over the standard model, against ten point nine for evolving dark energy, which has twice as many extra parameters. On Bayesian evidence for the microwave background plus DESI, both extensions score plus three point two in log evidence over the standard model, a moderate preference, so the second parameter buys nothing.

A word from our sponsor

Host 2Does your nose hair extend beyond the event horizon of your nostrils? Do people at dinner parties maintain eye contact a little too deliberately? Then you are in luck.

Host 2The Hastrup Review of Science is brought to you by nose-hair trimmers. Compact, rotary, and quietly decisive.

Host 1I use mine every morning. It is the only part of my routine that responds to a single parameter.

Host 2Nose-hair trimmers are very reasonably priced and in stock right now. Whatever is growing out there, they bring it back into tension-free agreement.

Host 1Mine also does ears. I was not expecting it to go there, but I am glad it did.

Host 2Nose-hair trimmers. Because what grows in the dark deserves a closer look. Our referral link is in the show notes.

Why it matters

Host 1So what is the principle here, beyond the numbers?

Host 2It is about how you read an anomaly. A two-parameter fit can tell you the standard model is under strain without telling you why, while a one-parameter model with a physical motivation makes sharp predictions that can fail. The authors' conclusion is that the DESI-era anomaly need not be read first as dynamical dark energy, and that the glitch deserves to be tested alongside the two-parameter model, not after it.

Host 1Who does that change things for?

Host 2Mainly for survey teams and analysts choosing which alternative models to run as standard. Today, results are routinely reported as constraints on the two dark energy dials; this paper argues a gravity-mismatch parameter should sit in the same table. For gravity theorists, it gives a concrete target: a preferred-frame theory with a nearly instantaneous extra field and gravitational waves at light speed.

Host 1And what does it not mean?

Host 2It is not a detection of Horava gravity, which the authors say explicitly; the same mismatch arises in several theories. It does not show that dark energy is a cosmological constant either, and it does not rule out evolving dark energy. And it does not mean Newton's constant is wrong in any measurement you or an engineer would ever make; the claim is about the expansion of the universe as a whole.

Caveats

Host 1Now the caveats, in full.

Host 2First, three point three sigma is a preference, not a discovery; particle physics would want five, and cosmology has seen three-sigma hints fade before. Second, this is a preprint by the group that proposed the model, so an independent reanalysis matters. Third, the significance moves with analysis choices.

Host 1How much does it move?

Host 2Using the cheaper microwave likelihood, the microwave background plus DESI result ranges from two point nine to three point six sigma depending on where Planck hands over to the Atacama telescope. Planck's fourth release on its own, without the ground telescopes, gives only two point zero sigma with DESI. So the headline depends meaningfully on the ground-based small-scale data.

Host 1And the perturbation treatment?

Host 2With the high sound speed, which is the physically allowed regime, the result is minus zero point zero zero five nine and two point nine sigma. Leaving the sound speed free gives two point eight sigma, with a skewed tail toward larger glitches. Very low sound speeds give a much larger glitch, but those are excluded for this field by Solar System bounds and the absence of gravitational Cherenkov radiation.

Host 1The evidence numbers also have fine print.

Host 2Several items. The model comparison uses the cheaper likelihood, not the headline one; evidence depends on the prior width, so halving the allowed range for omega g adds about zero point seven in log evidence, which the authors handle transparently. And they note an unconstrained parameter costs almost nothing in evidence, so the tie with evolving dark energy undersells the glitch's predictive edge, which is their argument, not a neutral finding.

Host 1What about supernovae?

Host 2With lensing and supernovae added, the glitch's evidence advantage over the standard model drops to plus one point one, while evolving dark energy keeps plus two point two. Supernovae prefer more matter than both DESI and the microwave background, and the two-parameter model can accommodate that while the glitch cannot. So on the fullest data set, evolving dark energy is mildly ahead on evidence.

Host 1Anything left undone theoretically?

Host 2The fluid treatment matches the theory exactly only in the limit of infinite sound speed. Mapping a finite sound speed onto an actual Horava action is left for future work, as is a first-principles treatment of the extra field's perturbations.

What to watch next

Host 1So what would confirm or kill this?

Host 2The glitch leaves a fingerprint on how structure grows. Expansion follows the cosmological constant but clumping follows Newton's, so during matter domination density ripples grow slightly faster, and the cumulative growth since the microwave background is enhanced by about three percent in the full calculation.

Host 1What does a survey actually measure?

Host 2The model predicts the clustering amplitude today is about two point eight percent higher than the standard model's extrapolation from the early universe. For galaxy surveys, the growth measure called f sigma eight should be about three percent higher, nearly constant between redshift zero point three and one point five, with the growth rate itself barely changed.

Host 1Who can test that?

Host 2DESI itself, through redshift-space distortions and full-shape analyses, and the Euclid satellite and the Rubin Observatory, which will measure clustering amplitudes at the percent level. The model also predicts no difference between how light and matter respond to gravity, and an early integrated Sachs–Wolfe signal, which the Atacama, South Pole and Simons Observatory telescopes can probe through lensing and cross-correlations with galaxies.

Host 1And simulations?

Host 2Because small-scale dynamics stay Newtonian and only the expansion history changes, the authors say simulations of the glitch should be straightforward. Their chains and plotting code are public on Zenodo and the modified code is on GitHub, so others can rerun it.

Bottom line

Host 1Bottom line?

Host 2The data that hint at evolving dark energy can be fitted equally well, on current evidence, by one physically motivated number: gravity about eight tenths of a percent weaker for the universe than for the solar system. It is a three-sigma preference from the model's own authors, sensitive to analysis choices and weakened by supernovae, but it makes specific growth predictions that the next few years of surveys can check.

Sign-off

Host 1That is the deep dive: Is DESI Seeing Dynamical Dark Energy, or a Cosmic Glitch in Gravity?, by Robin Wen, Lukas Hergt, Niayesh Afshordi and Douglas Scott, on the arXiv. The link is in the show notes.

Host 1Every Friday this same feed carries the roundup episode: five new papers as trailers, and you pick which become deep dives. This has been the Hastrup Review of Science.