In 1998, two teams of astronomers observed that distant Type Ia supernovae appeared about 30% fainter than expected, interpreting this as evidence that the universe’s expansion is accelerating — a discovery that won the 2011 Nobel Prize and was attributed to “dark energy” or a cosmological constant (Λ). This interpretation rests entirely on the assumption that the universe is perfectly homogeneous and isotropic (the same in every direction and at every location), encoded in the Friedmann-Lemaître-Robertson-Walker (FLRW) metric, a framework adopted in the 1920s when almost no cosmological data existed.
Professor Subir Sarkar of Oxford has spent years challenging this picture. His group’s analyses show that the evidence for cosmic acceleration is far weaker than claimed, that the acceleration signal is not uniform across the sky but dipolar (present in one direction, absent or reversed in the opposite direction), and that the foundational assumption of isotropy — the cosmological principle — is empirically falsified at greater than five standard deviations. If correct, this means dark energy may be a local illusion caused by our motion through an inhomogeneous universe, not a fundamental component of nature.
The stakes are enormous: dark energy is inferred to constitute roughly 70% of the universe’s energy density. Removing it would require a theoretical revolution comparable to the shift from a static to an expanding universe, or from continental drift’s rejection to the acceptance of plate tectonics.
The Cosmological Constant Problem: Why Dark Energy Was Suspicious from the Start
In 1933, Wolfgang Pauli calculated that quantum vacuum fluctuations — the zero-point energy of all quantum fields — should produce a cosmological constant so enormous that the universe could never have expanded beyond a few millimeters. He wrote that “it is more consistent to exclude zero-point energy because, evidently from experience, it does not interact with the gravitational field.”
This is the cosmological constant problem: general relativity says all forms of energy must gravitate, yet vacuum energy manifestly does not — or we would not exist. No proposed solution from any framework (loop quantum gravity, string theory, supersymmetry) has resolved it. String theory, for instance, naturally produces anti-de Sitter space (negative cosmological constant) and must be artificially manipulated to approximate the small positive value astronomers claim to observe.
Sarkar’s point: this unsolved problem should have made physicists deeply skeptical when astronomers announced a small, positive cosmological constant in 1998. Instead, the result was embraced, and the theoretical difficulty was set aside. The cosmological constant problem remains what Steven Weinberg called “the bone in our throat.”
How Supernova Cosmology Works — and Where It Gets Complicated
Type Ia supernovae are used as “standardizable candles” to measure cosmic distances. They are not intrinsically uniform in brightness, but in the 1990s astronomers discovered empirical correlations: brighter supernovae have wider light curves (the Phillips relation), and color corrections further reduce scatter. Using a fitting procedure called SALT2 (Spectral Adaptive Lightcurve Template), the factor-of-10 variation in intrinsic luminosity is reduced to less than a factor of 2, making them useful for cosmology.
The key measurement: distant supernovae at a given redshift are about 0.3 magnitudes (~30%) fainter than they would be in a universe expanding at a constant rate. This dimming is interpreted as the supernovae being farther away than expected, meaning space expanded more than predicted — i.e., the expansion accelerated.
The problem: the corrections applied (stretch, color, host galaxy effects) are themselves of comparable magnitude to the signal being measured. If the corrections are at the 0.15 magnitude level, claiming a 0.3 magnitude effect as new physics is precarious. Sarkar argues that as more data has accumulated, the evidence for acceleration has gotten weaker, not stronger — the opposite of what happens with genuine discoveries like the Higgs boson.
The 2016 Reanalysis: Five Sigma Becomes Three
The original 1998 discovery was based on roughly 50 supernovae per team (about 100 total, with overlap), and the two teams were not fully independent. In 2014, the Joint Light Curve Analysis (JLA) catalog combined ~740 supernovae from multiple surveys into a single, uniformly analyzed dataset.
Sarkar’s group applied principled statistical methods to the JLA data and found that the evidence for acceleration was only ~3 sigma (roughly 99.7% confidence), not the 5 sigma (99.99994%) threshold required for a discovery in physics. This was a significant downgrade from the claimed significance.
Rubin and Hinton (2016) responded by arguing that Sarkar’s team should have used Bayesian marginalization over many nuisance parameters (light curve shape, color, host galaxy mass) rather than profile likelihood. They introduced 12 additional redshift-dependent parameters and recovered a >5 sigma result.
Sarkar’s rebuttal: their “Bayesian” method uses the same maximum likelihood engine dressed in Bayesian language. More importantly, allowing stretch and color corrections to evolve with redshift opens the possibility that the absolute magnitude of supernovae also evolves with redshift — which would completely undermine their use for cosmology. This is not a technicality: if supernovae were intrinsically fainter in the past (because their progenitor stars were younger), this would mimic the dimming attributed to acceleration.
Supernova Progenitor Age: The “Crow Comes Home to Roost”
A group of Korean astronomers has since found that the absolute magnitude of Type Ia supernovae correlates with the age of the progenitor star: younger progenitors produce intrinsically fainter supernovae. Since higher-redshift supernovae come from younger stellar populations, this creates a systematic dimming with redshift that is indistinguishable from cosmic acceleration.
When Sarkar’s group implements this progenitor-age correction, the isotropic (monopole) component of acceleration disappears entirely. What remains is a dipolar signal aligned with our local bulk flow — a local effect, not a cosmological one. The universe, on the largest scales, appears to be decelerating, not accelerating.
The Dipole: Acceleration Is Directional, Not Isotropic
When Sarkar’s team allowed the acceleration to have angular dependence (monopole + dipole) rather than assuming it is the same in every direction, the data overwhelmingly preferred a dipole pattern: acceleration in one direction of the sky, deceleration in the opposite direction. This was first shown with the 740-supernova JLA catalog in 2019 and confirmed with the newer Pantheon+ catalog (~1700 supernovae) in a 2025 paper.
The dipole is aligned with the direction of our local motion (the CMB dipole direction). This is consistent with the idea that we are embedded in a deep bulk flow — a large-scale motion caused by the gravitational pull of inhomogeneous matter distributions — and that this flow creates the illusion of acceleration when we interpret the data using the FLRW metric.
A cosmological constant must be isotropic — the same in all directions — by Lorentz invariance. A directional (dipolar) acceleration cannot be due to dark energy. It is a local, kinematic effect.
The Ellis-Baldwin Test: A Clean, Independent Check
In 1984, George Ellis and John Baldwin proposed a simple, elegant test of whether the universe obeys the FLRW metric. If the CMB dipole (the fact that one half of the sky is slightly warmer than the other) is caused by our local motion at ~369 km/s, then the same kinematic aberration and Doppler effects should produce a matching dipole in the distribution of any cosmologically distant objects (radio sources, quasars) on the sky.
The predicted amplitude of this matter dipole is about 0.5% (5 × 10⁻³), a tiny effect requiring at least a million objects across the whole sky to detect. For decades, the data was insufficient.
Using the NVSS radio survey and, later, a catalog of ~1.5 million quasars from the WISE infrared satellite (analyzed with Nathan Secrest), Sarkar’s team measured the matter dipole. It is in the correct direction but has twice the amplitude it should be if the CMB dipole and matter dipole share the same kinematic origin.
This discrepancy has been confirmed at >5 sigma by multiple independent datasets (radio sources from ground-based telescopes, quasars from a satellite, analyzed by three independent groups). No systematic effect has been identified that can explain it.
What it means: the standard procedure of “boosting” to the CMB rest frame to analyze cosmological data — the procedure used by every collaboration including DESI — assumes that the matter dipole and CMB dipole are the same. They are not. The cosmological principle (statistical isotropy and homogeneity) is empirically falsified.
DESI and the Sum Rule Fallacy
The DESI collaboration has produced spectacular data (12+ million spectroscopic redshifts) and claims evidence for evolving dark energy — dark energy whose density changes with time. However, their analysis begins by assuming the FLRW metric, the very thing Sarkar’s work calls into question.
Sarkar has asked DESI researchers whether the BAO (baryon acoustic oscillation) signal is the same in every direction on the sky. He has received no answer. If the signal varies directionally, the isotropic analysis is invalid.
The “concordance” value of dark energy (ΩΛ ≈ 0.7) is derived from a sum rule: Ωmatter + Ωcurvature + ΩΛ = 1. This sum rule is a direct consequence of the Friedmann equation, which assumes exact homogeneity and isotropy. If the FLRW metric is wrong, the sum rule does not hold, and the inference of ΩΛ = 0.7 collapses.
The CMB does not directly measure Λ. At the time of decoupling (z ~ 1000), matter was a billion times more dominant than Λ. The CMB primarily constrains spatial curvature (Ωk ≈ 0).
Baryon acoustic oscillations constrain Ωmatter.
Supernovae constrain a particular combination of Ωmatter and ΩΛ.
Only by invoking the sum rule (from FLRW) do these combine to give ΩΛ ≈ 0.7. Without FLRW, there is no sum rule, and no necessity for Λ.
The LTB Alternative and the Fine-Tuning Double Standard
If one relaxes the assumption of homogeneity but keeps isotropy, the Lemaître-Tolman-Bondi (LTB) metric allows for radial inhomogeneity — for example, the possibility that we reside in a large underdense region (a “void”). Even this modest departure from FLRW is sufficient to explain the supernova dimming without any acceleration.
Critics object that LTB void models require us to be near the center of the void (within ~1%), which they consider fine-tuned. Sarkar’s response: the cosmological constant requires fine-tuning to one part in 10⁶⁰ (or 10¹²³ if interpreted as vacuum energy). The double standard is glaring — cosmologists accept the far more extreme fine-tuning of Λ while rejecting the modest fine-tuning of an LTB void.
More generally, the Szekeres metrics allow for both inhomogeneity and anisotropy, but the proliferation of free parameters makes them difficult to constrain with current data. Upcoming surveys (Rubin Observatory’s LSST, SPHEREx, Euclid) may provide enough data to fit more general metrics, possibly using machine learning.
Inflation’s Unexamined Assumptions
Inflation theory posits that the early universe underwent a period of exponential expansion driven by the vacuum energy of a slowly rolling scalar field (the inflaton). It predicts the nearly scale-invariant, Gaussian fluctuations observed in the CMB.
Sarkar’s critique: inflation invokes vacuum energy to drive expansion, yet we do not understand whether vacuum energy couples to gravity at all (the cosmological constant problem). If it did couple with its natural magnitude, inflation would have been far too violent, or the universe would have recollapsed instantly. Inflation “uses” vacuum energy and then “loses it” — the inflaton field somehow decays to zero vacuum energy, requiring fine-tuning of 1 part in 10¹⁶⁰ from the top of the potential to the bottom.
The horizon problem (the motivation for inflation) assumes the FLRW metric holds all the way back to t = 0. But the FLRW metric certainly breaks down near the Planck scale, where quantum gravity effects dominate. The integral that defines the particle horizon may not converge if spacetime becomes fractal or otherwise non-smooth at early times. The horizon problem may be an artifact of extrapolating classical general relativity into a regime where it does not apply.
The BICEP2 claim (since retracted) of detecting primordial gravitational waves from inflation would have required this 1-in-10¹⁶⁰ fine-tuning, which is why Sarkar did not believe it.
The Bigger Picture: What Needs to Happen
Sarkar is not merely arguing that one dataset is flawed. He is arguing that the entire inferential chain leading to dark energy — supernovae, CMB, BAO, all analyzed under FLRW — rests on a metric that is empirically falsified. The Ellis-Baldwin test, the supernova dipole, and the matter-CMB dipole mismatch all point to the same conclusion: the universe is not well described by the FLRW metric on the scales assumed.
He draws an analogy to continental drift: the evidence (fossil distributions, continental shapes) was known in the 1920s–1930s but was not accepted for 50 years because there was no theoretical mechanism (plate tectonics). Similarly, the evidence against FLRW may not be accepted until an alternative theoretical framework is developed.
George Ellis proposed 40 years ago a program called the cosmological fitting problem: rather than assuming a metric and fitting data to it, use the data to infer the metric. This was computationally intractable at the time but may now be feasible with modern data volumes and machine learning.
Sarkar’s advice to young researchers: the cosmological constant problem is the biggest unsolved problem in fundamental physics. The fact that it has resisted all attempts at solution — by the brightest minds, using the most sophisticated frameworks — means something very large is missing from our understanding. This is not discouraging; it is an invitation.
On Sigma, Systematics, and Scientific Culture
Sarkar emphasizes the distinction between nominal sigma (the statistical significance quoted in papers, conditional on systematics being correctly modeled) and effective sigma (the actual reliability of a result). Many high-sigma results in particle physics have disappeared due to unrecognized systematic effects (e.g., the 1991 claim of a 17 keV neutrino at Oxford, confirmed at Berkeley, later retracted due to three conspiring systematic effects).
In cosmology, the situation is worse because there is only one universe — you cannot repeat the experiment. Bayesian and frequentist statistics should give the same answer for well-posed questions, but in practice, the choice of priors and the treatment of systematics can dramatically alter conclusions.
There is a publication bias toward results that confirm the standard model. Sarkar notes that of ~30 measurements of Λ following the WMAP paper, only 3 lay outside the 1-sigma range of the “true” value — whereas ~10 should have. The missing outliers suggest selection bias: people publish results consistent with the consensus and shelish those that are not.
He also notes a culture gap between relativists, astronomers, and particle physicists. Astronomers treat Λ as just another parameter in a fit, without appreciating that interpreting it as vacuum energy requires fine-tuning to 60 decimal places. Particle physicists, unable to solve the cosmological constant problem, have largely ignored it because it does not affect collider experiments. Relativists appreciate the theoretical difficulty but often defer to the observational consensus.
