Supernova and BAO Data Show 2–4σ Deviations from Standard Cosmic Geometry
Modern cosmology rests on a single geometric assumption: that on the largest scales, the universe looks the same in every direction and expands according to one shared clock. That assumption is packaged in the Friedmann-Lemaître-Robertson-Walker (FLRW) metric, the scaffold beneath the ΛCDM model that ties together the cosmic microwave background (CMB), supernova distances, and baryon acoustic oscillations (BAO). Three papers posted to arXiv in April 2026, plus a 1,675-respondent survey from the American Physical Society’s Physics Magazine, suggest the scaffold is wobbling. The data do not yet force a rewrite of the textbook, but they expose a split: the geometry tests fail at roughly 2–4 standard deviations, while physicists surveyed on related puzzles show no settled majority on what would fix them.
The geometry test that should always read zero
The Clarkson-Bassett-Lu (CBL) test, proposed in 2008, combines the angular diameter distance to faraway objects with the line-of-sight expansion rate. In any FLRW universe, the resulting statistic C must equal zero at every redshift, regardless of dark energy, dark matter, or the theory of gravity, as long as light follows geodesics in a metric theory. A companion integral, O, must stay flat. These are null tests: they do not fit parameters; they check whether the geometry itself is internally consistent.
Søren M. Koksbang (University of Southern Denmark) and Asta Heinesen (Queen Mary University of London and the Niels Bohr Institute) spent two companion papers generalizing those tests so a failure points to a physical cause rather than an abstract anomaly. In arXiv:2604.05836, they rewrite C, O, and a new density statistic M in terms of quantities defined in any spacetime: optical shear, Ricci curvature along the light path, and the generalized expansion rate H that BAO surveys actually measure. In arXiv:2604.05822, they apply the framework to real data using bootstrap symbolic regression: 200 resampled fits from Pantheon+ supernovae (1,701 Type Ia events, anchored to the SH0ES local distance ladder) and BAO measurements from BOSS, eBOSS, and DESI, with seven effective redshift bins from z = 0.38 to z ≈ 2.3.
The overlap region where both probes constrain the same geometry runs from z = 0.38 upward. There, the reconstructed C and O deviate from the FLRW prediction of zero. When DESI data release 1 dominates the BAO side, the mismatch reaches 3–4σ at redshifts below z ≈ 0.5; the integrated test O exceeds 4σ at the lowest redshifts in that combination. DESI data release 2 pulls the violation back toward 2–3σ over most of the interval. The authors call these, to their knowledge, the first significant detections of FLRW self-inconsistency in the literature, and they stress that the significance shifts with data cuts and expression-selection rules inside the symbolic-regression pipeline.
The density test M tells a different story. Built to recover the cosmic density field without invoking the Friedmann equations, it should sit near Ωm,0H02 in a ΛCDM universe: 1431.4 (km/s/Mpc)2 for Planck 2018 parameters, or 1799.4 (km/s/Mpc)2 for Pantheon+ combined with SH0ES. Koksbang and Heinesen’s median reconstruction keeps both values within the 16th–84th percentile band (their proxy for 1σ) across most of the shared redshift range. Matter content looks ΛCDM-compatible even while the geometry relations wobble.
Two classes of fixes, and a new discriminator
If the FLRW violation survives tighter data, most tension-fixing models already on the table may be dead ends. Heinesen and Timothy Clifton (Queen Mary) map that problem in arXiv:2604.07244. Evolving dark energy, modified gravity, and new particle species framed inside FLRW geometry cannot explain a failure of FLRW consistency tests; the failure would have to come from how light propagates or from how large-scale structure back-reacts on expansion.
They analyze two concrete non-FLRW routes. The Dyer-Roeder scenario keeps FLRW expansion but lets light beams traverse underdense lines of sight (α < 1 in their notation), mimicking the masking of dense baryonic regions. Cosmological back-reaction, averaged with Buchert’s formalism, instead lets structure formation alter the large-scale expansion rate HD. Each route predicts a distinct signature in the curvature-consistency statistics. Heinesen and Clifton introduce a new test quantity A that equals −K in FLRW but tracks an effective curvature function in the alternative cases, giving observers a way to tell underdense beams from back-reaction when Stage IV surveys tighten the constraints.
What 1,675 physicists think is wrong
Parallel evidence that the field has not converged comes from the Big Mysteries Survey, fielded by Physics Magazine from 28 July to 9 September 2025 and analyzed by Niayesh Afshordi (University of Waterloo and Perimeter Institute) and colleagues in a May 2026 preprint. Of 1,675 respondents, 24.4% chose “no opinion” on the Hubble tension; 22.1% picked early dark energy; only 16.7% blamed systematic errors in supernova data (down from 35% in an 85-person Copenhagen conference sample a year earlier).
On dark energy, a cosmological constant Λ drew 24.0% support, while a time-varying field drew 25.9%. Neither option reached a majority in either survey wave. Inflation fared better on early-universe puzzles (50.8% ± 1.7%), but that is barely over half, not the near-unanimity often implied in public summaries. Hybrid dark-matter models led individual explanations for galactic rotation curves (53.4% when axion, WIMP, primordial-black-hole, and hybrid options are combined), yet modified gravity and quantum-gravity explanations together reached 21.6%, up from 5% in the smaller Copenhagen sample.
The survey is not a random sample of all physicists; it reflects Physics Magazine readers and APS members who chose to respond. Still, at 1,675 participants it is large enough to show that several positions marketed as consensus are, in practice, pluralities or narrow majorities.
What we do not know
Koksbang and Heinesen flag their own uncertainty machinery as unfinished business. Symbolic regression does not ship with native error bars; their bootstrap pipeline depends on manual or semi-automated rules for rejecting pathological fits, and swapping criteria can shift a 4σ FLRW violation to 2σ without changing the qualitative direction. DESI release 2 already softened the tension relative to release 1, which warns against treating any single σ quote as final.
The geometry failure and the density agreement can coexist if the inconsistency sits in light propagation or back-reaction rather than in the total matter budget, but the current data cannot yet pick between those routes. Heinesen and Clifton’s discriminator awaits the denser distance and expansion-rate maps promised by Stage IV surveys.
On the human side, the survey documents disagreement, not which camp is correct. Early dark energy correlates with time-varying dark energy across questions, and quantum-gravity answers cluster across dark matter, dark energy, and the Hubble tension, but correlation is not mechanism.
Why this is human progress
Cosmology has spent a decade fitting parameters inside ΛCDM while H0 from the CMB and H0 from the local distance ladder disagree at roughly 5σ. The April 2026 papers move the argument up a level: they test whether the geometric stage itself is level before adding new actors to the play. A null test that returns 2–4σ instead of zero is more informative once C is expressed in shear, curvature, and expansion terms you can measure, rather than as a single opaque number.
If the violation hardens, the payoff is constraint, not chaos. Back-reaction and underdense light cones are old ideas with new observational signatures; ruling out entire classes of FLRW-based fixes narrows the search. If it fades with better data and cleaner regression, that strengthens ΛCDM’s geometric foundation. Either outcome is progress: the question is no longer only “which dark-energy equation of state fits DESI,” but whether the universe on the largest scales is actually the homogeneous, isotropic arena the standard model assumes.
The survey adds a cultural datapoint with the same moral. Public-facing science often treats Λ, inflation, and inflationary solutions to the Hubble tension as settled. A sample of 1,675 working physicists shows dark energy split nearly evenly, the Hubble tension unresolved, and inflation supported by 50.8%. Honest disagreement, paired with tests that turn distance catalogs into geometry diagnostics, is how a field stops recycling the same parameter tweaks and starts asking which layer of the model broke first.
Drafts may be assisted by AI. Every published article is reviewed and edited by a named member of our staff before publication.
