Literature cutoff: 24 October 2026.

The universe is expanding. Georges Lemaître derived and estimated the relation in 1927; Edwin Hubble’s 1929 distance–velocity data made the observational case famous. In its nearby-universe form, what we now call the Hubble–Lemaître law is

$$v = H_0 \, d$$

The proportionality constant $H_0$ — the Hubble constant — is the current rate of expansion, in units of km/s/Mpc (kilometres per second per megaparsec, where 1 Mpc $\approx 3.086 \times 10^{22}$ m). Hubble’s original measurement gave $H_0 \approx 500$ km/s/Mpc. It was wrong by a factor of about seven, which is understandable given that he was measuring distances to galaxies in the 1920s using photographic plates and a lot of courage. Over the following decades, as techniques improved, the value converged towards 70 km/s/Mpc. By the 1990s, many people considered the question largely settled: $H_0$ was somewhere between 60 and 80, and the main arguments were about where exactly in that range.

Those arguments have sharpened considerably. We now have a precise local measurement of $H_0$ [[1]] and a precise value inferred from early-universe data within the six-parameter $\Lambda$CDM model [[2]], and they disagree:

$$H_0^{\text{late}} = 73.04 \pm 1.04 \text{ km/s/Mpc} \qquad \text{(distance ladder)}$$

$$H_0^{\text{early}} = 67.4 \pm 0.5 \text{ km/s/Mpc} \qquad \text{(CMB)}$$

The discrepancy is $73.04 - 67.4 = 5.64$ km/s/Mpc. The combined uncertainty is $\sqrt{1.04^2 + 0.5^2} \approx 1.15$ km/s/Mpc. The significance is approximately $4.9\sigma$, which rounds to what cosmologists have taken to calling — with some grimness — the Hubble tension.

Particle physics conventionally uses $5\sigma$ as a discovery threshold, but that convention does not convert this comparison into a discovery of new physics. The quoted significance is conditional on the datasets, error models, and $\Lambda$CDM inference being compared; unknown systematics are not included by the arithmetic above.

This is the kind of problem that keeps me reading papers at unreasonable hours. The rest of this post explains why both routes are taken seriously, where their model dependence enters, and what proposals exist for resolving the mismatch.

What $H_0$ Actually Measures and Why It Matters

Hubble’s law $v = H_0 d$ is valid in the nearby universe, for galaxies whose recession velocities are much less than the speed of light. In the full relativistic framework, the expansion is described by the scale factor $a(t)$, which encodes how distances between comoving points grow with time. The Hubble parameter is defined as

$$H(t) = \frac{\dot{a}}{a}$$

and $H_0 = H(t_0)$ is its value today. The Friedmann equation — derived from general relativity applied to a homogeneous and isotropic universe — gives

$$H(z)^2 = H_0^2 \left[ \Omega_m (1+z)^3 + \Omega_r (1+z)^4 + \Omega_k (1+z)^2 + \Omega_\Lambda \right] \tag{1}$$

where $z$ is the redshift (related to the scale factor by $1 + z = 1/a$), and the $\Omega$ parameters are the present-day fractional energy densities of matter, radiation, spatial curvature, and the cosmological constant (dark energy), respectively. Our standard cosmological model, $\Lambda$CDM, assumes a spatially flat universe ($\Omega_k = 0$). Radiation is negligible today ($\Omega_r \approx 9 \times 10^{-5}$) but dominated before matter–radiation equality at $z \approx 3400$, which matters for the sound horizon below. The Planck 2018 fit gives [[2]]

$$\Omega_\Lambda \approx 0.685, \qquad \Omega_m \approx 0.315, \qquad \Omega_b \approx 0.049$$

where $\Omega_b$ is the baryon (ordinary matter) density, part of $\Omega_m$; flatness makes $\Omega_\Lambda$ and $\Omega_m$ sum to one, up to the tiny radiation term. Dark matter makes up most of $\Omega_m$.

$H_0$ is not just one number among many. It is the normalisation of the entire cosmological distance scale. It appears in the age of the universe:

$$t_0 = \int_0^\infty \frac{dz}{(1+z)\,H(z)}$$

which, for $\Lambda$CDM with the above parameters, gives $t_0 \approx 13.8$ Gyr [[2]]. A higher $H_0$ means a faster expansion rate, which means — for fixed $\Omega$ values — a younger universe. An error in $H_0$ propagates into every cosmological distance and every age estimate. It is not a detail.

Route 1: The Late-Universe Measurement ($H_0 = 73$)

The distance ladder is the name for the chain of calibrated distance measurements that extends from the Earth to the far reaches of the universe. Each rung calibrates the next.

Rung 1: Geometric parallax. For stars within a few kiloparsecs, we can measure distance directly from the shift in apparent position as the Earth orbits the Sun. The parallax angle $\pi$ (in arcseconds) gives the distance $d = 1/\pi$ parsecs. This is pure geometry — it follows from Euclid and Kepler, not from any physical model of stars. The Gaia space mission has measured parallaxes for nearly 1.5 billion sources, typically to 20–30 $\mu$as for stars brighter than magnitude $G = 15$ [[3]], providing the geometric foundation of the ladder in the Milky Way, the first of its three anchors.

Rung 2: Cepheid variables. These are pulsating giant stars whose oscillation period $P$ is tightly correlated with their intrinsic luminosity $L$ — the Leavitt period-luminosity relation, discovered by Henrietta Swan Leavitt in 1912 (an earlier post reconstructs her twenty-five stars). The relation takes the form

$$M = \alpha \log_{10}(P/\text{days}) + \beta + \gamma \left[ \text{Fe/H} \right]$$

where $M$ is the absolute magnitude, and the metallicity term $\gamma[\text{Fe/H}]$ accounts for the chemical composition of the star. Once calibrated against nearby Cepheids with known parallax distances, this relation allows the distance to any galaxy hosting Cepheids to be inferred from their periods and apparent brightnesses. With the Hubble Space Telescope, SH0ES has measured Cepheids in supernova hosts out to about 80 Mpc, most of them within 40 Mpc [[1]].

Rung 3: Type Ia supernovae. These are thermonuclear explosions of white dwarf stars. Their progenitor channels and explosion masses are not all identical, but their light curves can be standardised empirically: peak luminosity correlates with the rate at which brightness declines after peak — the Phillips relation. After this standardisation, Type Ia SNe serve as “standardisable candles” reaching to redshifts $z \sim 2$, far beyond the reach of Cepheids.

The logic of the ladder: geometric anchors (parallaxes in the Milky Way, eclipsing binaries in the Large Magellanic Cloud, a water megamaser in NGC 4258) calibrate nearby Cepheids; those Cepheids calibrate Cepheids in SN Ia host galaxies; those SN Ia establish the absolute luminosity of the standard candle; that calibrated SN Ia sample gives recession velocities (from spectroscopic redshifts) and distances (from luminosities) simultaneously, yielding $H_0$ [[1]]. Only supernovae at $0.023 < z < 0.15$ are used for this last step.

The SH0ES collaboration (Supernovae and $H_0$ for the Equation of State of dark energy) has driven this measurement since 2005. Their 2022 result, using Hubble Space Telescope observations of Cepheids in 37 host galaxies of 42 Type Ia SNe, found [[1]]

$$H_0 = 73.04 \pm 1.04 \text{ km/s/Mpc}$$

This is a 1.4% measurement. For context, Hubble’s original measurement had an error of several hundred percent.

The JWST cross-check. One candidate systematic error was crowding: in HST images of distant galaxies, Cepheids might be blended with unresolved neighbouring stars, artificially brightening them and biasing the distance estimate. JWST resolves individual Cepheids more cleanly in the same host galaxies. Observations of overlapping hosts found HST and JWST Cepheid distances in agreement and strongly disfavoured an unrecognised, distance-dependent HST crowding bias large enough to explain the tension [[4]]. That is an important test of one systematic, not a proof that every distance-ladder systematic has been excluded.

Not every ladder lands at 73. Calibrating the supernovae with the tip of the red giant branch (TRGB) instead of Cepheids, Freedman obtained $69.8 \pm 0.6\,(\text{stat}) \pm 1.6\,(\text{sys})$ km/s/Mpc, consistent with the CMB [[5]], and the Chicago–Carnegie programme’s update with HST and JWST data $70.4 \pm 1.9$ [[6]]. SH0ES attributes the difference to which supernovae the small JWST samples contain and finds $72.6 \pm 2.0$ from all of them combined [[7]]; a 2026 network analysis of most local distance indicators, SH0ES and Chicago–Carnegie distances among them, gives $73.50 \pm 0.81$ [[8]]. As of October 2026 the ladders disagree less with one another than with the CMB, but the question is open.

An independent late-universe measurement comes from time-delay cosmography. Gravitational lensing of a background quasar by an intervening galaxy produces multiple images; the arrival times of photons along different paths differ by an amount that scales as $1/H_0$ and depends on the lens galaxy’s mass profile, the main source of uncertainty [[9]]. The TDCOSMO collaboration found $H_0 = 74.5^{+5.6}_{-6.1}$ km/s/Mpc from seven lensed quasars, entirely independently of the distance ladder [[9]]. This is a completely different physical observable. At ±8% it is consistent with SH0ES and, at about $1\sigma$, with Planck. The 2025 analysis of eight lenses gives $71.6^{+3.9}_{-3.3}$, or $74.3^{+3.1}_{-3.7}$ when lenses without time delays help constrain the mass profiles [[10]].

Route 2: The Early-Universe Inference ($H_0 = 67$)

The cosmic microwave background (CMB) is the thermal radiation left over from recombination — the epoch at $z_\ast \approx 1090$ (about 370,000 years after the Big Bang) when the universe had cooled enough for protons and electrons to combine into neutral hydrogen, allowing photons to free-stream for the first time. The CMB is extraordinarily uniform in temperature ($T \approx 2.725$ K) but carries tiny fluctuations at the level of $\Delta T / T \sim 10^{-5}$.

Before recombination, the universe was a tightly coupled photon-baryon fluid. Perturbations in this fluid oscillated: gravity pulled baryons into overdense regions, while radiation pressure resisted compression. The competition set up acoustic waves — sound waves in the plasma of the early universe. These waves travelled at the sound speed

$$c_s = \frac{c}{\sqrt{3(1 + R)}}, \qquad R = \frac{3 \rho_b}{4 \rho_\gamma}$$

where $\rho_b$ and $\rho_\gamma$ are the baryon and photon energy densities. The waves propagated until recombination froze them in place. The comoving distance they had travelled by then — the sound horizon — is

$$r_\ast = \int_0^{t_\ast} \frac{c_s \, dt}{a(t)} = \int_{z_\ast}^{\infty} \frac{c_s \, dz}{H(z)} \tag{2}$$

This is a comoving length set by the microphysics of the early universe. It depends on the physical densities $\omega_b = \Omega_b h^2$ and $\omega_m = \Omega_m h^2$, with $h = H_0/(100 \text{ km/s/Mpc})$, and on the radiation density, but not on dark energy or on $H_0$ separately. For the best-fit $\Lambda$CDM parameters, $r_\ast = 144.4$ Mpc [[2]]. The value near 147 Mpc that is often quoted is $r_d$, the same integral taken to the end of the baryon drag epoch at $z_d \approx 1060$; it is the ruler for galaxy surveys below.

The frozen acoustic oscillations imprint a characteristic angular scale on the CMB temperature fluctuations. The angular size of the sound horizon as seen from today is

$$\theta_\ast = \frac{r_\ast}{D_M(z_\ast)}, \qquad D_M(z) = (1+z)\,D_A(z) = c \int_0^{z} \frac{dz'}{H(z')} \tag{3}$$

where $D_M$ is the comoving angular-diameter distance (the integral form holds in a flat universe) and $D_A$ the usual angular-diameter distance. Until this revision I wrote this ratio with $D_A$ in the denominator. That was wrong: a comoving length has to be divided by a comoving distance, and with $D_A(z_\ast) \approx 12.7$ Mpc the “angle” would have been 11 radians. The Planck satellite measured $\theta_\ast = 0.5965°$ to 0.03%, one of the most accurately measured parameters in cosmology [[2]]. My own integration of Eqs. (1)–(3) with Planck’s best-fit densities reproduces $r_\ast = 144.4$ Mpc, $D_M(z_\ast) = 13.87$ Gpc and that angle.

Now, here is the key point. $\theta_\ast$ is measured directly from the CMB. But to extract $H_0$, we must model both $r_\ast$ (which depends on the early-universe physics) and $D_M(z_\ast)$ (which depends on the expansion after recombination, and therefore on $H_0$ and $\Omega_m$). The fit varies the six base parameters together against the entire CMB power spectrum, and $H_0$ follows from them.

The Planck 2018 result [[2]]:

$$H_0 = 67.4 \pm 0.5 \text{ km/s/Mpc}$$

This is a 0.7% result — even tighter than SH0ES. And it assumes $\Lambda$CDM. That assumption is crucial, and we will return to it.

Independent CMB experiments give consistent results without Planck: the Atacama Cosmology Telescope combined with WMAP gives $66.78 \pm 0.68$ km/s/Mpc [[11]], and the South Pole Telescope’s SPT-3G data alone give $66.66 \pm 0.60$ [[12]]. That makes a Planck-specific instrumental systematic an unlikely explanation. What they share is the model, $\Lambda$CDM, and the physics of Eq. (2).

The Tension: Numbers and What They Mean

The discrepancy is:

$$\Delta H_0 = 73.04 - 67.4 = 5.64 \text{ km/s/Mpc}$$

The combined statistical uncertainty is:

$$\sigma_{\text{comb}} = \sqrt{1.04^2 + 0.5^2} \approx 1.15 \text{ km/s/Mpc}$$

The significance:

$$\frac{\Delta H_0}{\sigma_{\text{comb}}} \approx 4.9\sigma$$

Riess et al. round this to $5\sigma$ [[1]]. Later comparisons with smaller error bars reach 6–7$\sigma$ [[12]] [[8]], while the Chicago–Carnegie TRGB value sits about $1.5\sigma$ above Planck (Fig. 1).

Figure 1. Methods calibrated by the sound horizon cluster at 66.7–68.5 km/s/Mpc, the SH0ES results and the H0DN network at 72.6–73.5; the TRGB ladder and the lens results lie in between or are too wide to choose. Published central values with 68% intervals: Planck 2018 [[2]], ACT DR6 with WMAP [[11]], SPT-3G D1 alone [[12]], DESI DR2 BAO with a BBN baryon density [[13]], SH0ES with HST Cepheids [[1]] and with JWST Cepheids, TRGB and JAGB stars [[7]], CCHP’s TRGB value with its three error terms in quadrature [[6]], H0DN [[8]], and the lens-only TDCOSMO analyses of seven lenses [[9]] and of eight lenses with a Pantheon+ prior on $\Omega_m$ [[10]]. Bands: the Planck and SH0ES 2022 intervals compared in the text. The rows differ in probe and priors and are not independent: H0DN combines SH0ES, CCHP and other distances, and every sound-horizon row assumes $\Lambda$CDM.

To put this in perspective: for Gaussian errors, a fluctuation of $5\sigma$ or more in one direction has a probability of $2.9 \times 10^{-7}$, the “one in 3.5 million” of Riess et al. [[1]]; counting either sign doubles it. You might win a lottery. You probably should not bet your cosmological model on it.

The tension became more statistically significant as the quoted uncertainties shrank over the past decade. That persistence motivates the term “tension”; it does not establish that the only remaining explanation is new physics.

What Could Explain It

The community has not been idle. The number of proposed solutions runs into the hundreds; one 2021 review needed more than 1,000 references to survey them [[14]]. They can be organised into a few broad categories.

Systematic Errors

The distance ladder has multiple candidate systematics that have been carefully evaluated:

  • Cepheid metallicity dependence: the period-luminosity relation shifts with metal abundance. It is calibrated empirically, including with Gaia parallaxes of Milky Way Cepheids; the residual uncertainty is $\lesssim 0.5$ km/s/Mpc [[1]].

  • Photometric crowding: addressed by JWST [[4]] [[7]].

  • LMC geometry and distance: the Large Magellanic Cloud is one of the three anchors. Its distance from eclipsing binaries is known to about 1.2% [[1]], and the tension survives the removal of any single anchor or type of calibrator [[14]].

  • SN Ia physics: host galaxy dependence of SN Ia peak luminosity, Malmquist bias in flux-limited surveys, potential evolution with redshift. These have been studied extensively; in the SH0ES error budget the supernova terms come to about 1%, or 0.7 km/s/Mpc [[1]].

On the CMB side, Planck estimates that residual modelling uncertainties affect its parameters only at the $0.5\sigma$ level [[2]], and ACT and SPT reach 66.7–66.8 km/s/Mpc with different instruments [[11]] [[12]]. No identified CMB systematic shifts $H_0$ by 5.6 km/s/Mpc.

No single identified systematic currently accounts for the full SH0ES–Planck comparison. Unrecognised or underestimated systematics nevertheless remain a live explanation, alongside extensions to $\Lambda$CDM.

Early Dark Energy

This is one of the most discussed new-physics solutions [[15]]. The idea is to introduce a new energy component that becomes dynamically important shortly before matter–radiation equality, near $z_c \approx 5000$ in the original fits, and decays before recombination. This “Early Dark Energy” (EDE) temporarily increases the expansion rate $H(z)$ at early times.

Why does this help? Recall that the CMB measures $\theta_\ast = r_\ast / D_M$ directly and precisely, Eq. (3). A faster expansion rate before recombination reduces the integral in Eq. (2), shrinking $r_\ast$. The distance $D_M$ also changes, but less sensitively. A smaller $r_\ast$ means that, to reproduce the same observed angle, the model requires $D_M$ to be correspondingly smaller. Smaller $D_M$ implies a higher $H_0$.

How much higher depends on what else moves (Fig. 2). $D_M$ depends on $\Omega_m$ as well as on $H_0$, so a fixed angle defines a curve, not a point. With $\Omega_m \approx 0.30$, as DESI’s BAO measurements prefer, reaching 73 km/s/Mpc needs $D_M$, and therefore $r_\ast$, about 6% smaller than in $\Lambda$CDM. The supernova samples prefer a higher $\Omega_m$ [[13]], which makes the requirement stronger: about 9% at $\Omega_m = 0.33$. In EDE fits the cold-dark-matter density rises too [[15]], so the solution moves along the shifted curve, not along the line of fixed $\omega_m$.

Figure 2. Holding the acoustic angle fixed ties $H_0$ to $\Omega_m$, and a smaller sound horizon lifts the whole relation. Each curve is $D_M(z_\ast) = r_\ast/\theta_\ast$ from Eq. (3), with $\theta_\ast$ at Planck’s value: solid for $r_\ast = 144.4$ Mpc, dashed for a sound horizon 3.3% smaller, the reduction in Poulin et al.’s EDE fit [[15]]. Dotted: $\omega_m = \Omega_m h^2 = 0.143$, the value the peak heights fix in $\Lambda$CDM. Open circle: that EDE fit’s $H_0$, from a fit that included the 2018 SH0ES value, at the $\Omega_m$ I derive from its densities; band: SH0ES [[1]]. My integration of flat $\Lambda$CDM after recombination. Within $\Lambda$CDM, $r_\ast$ itself changes with $\omega_m$, so the CMB’s own degeneracy is somewhat flatter, close to $\Omega_m h^3 = 0.0963$ [[2]].

In Poulin et al.’s preferred model the EDE contributes 5% of the total energy density at its peak. Because $H^2 \propto \rho$, that raises $H$ there by only about 2.6%; together with a higher cold-dark-matter density it shrinks the sound horizon by 3.3% and gives $H_0 = 70.6 \pm 1.3$ km/s/Mpc in a fit that includes SH0ES [[15]]. The field is axion-like and rolls down a periodic potential

$$V(\phi) = m^2 f^2 \left[1 - \cos\left(\frac{\phi}{f}\right)\right]^n$$

where $m$ is its mass and $f$ its decay constant. Frozen at early times, it acts as a cosmological constant; at $z_c$ it begins to oscillate around the minimum and behaves as a fluid with $w = (n-1)/(n+1)$, so for $n = 3$ its energy density falls as $a^{-4.5}$, faster than radiation, leaving no late-universe signature [[15]].

EDE is not without problems. The potential is contrived: Hill et al. note that the $n = 3$ form requires a fine-tuned hierarchy of instanton terms [[16]], and $z_c$ is a free parameter that the data place shortly before matter–radiation equality. More seriously, EDE raises the amplitude of matter fluctuations and so worsens the $S_8$ tension — the $\sim 2$–$3\sigma$ discrepancy between the CMB and several weak gravitational lensing surveys in the parameter $S_8 = \sigma_8 \sqrt{\Omega_m / 0.3}$ (where $\sigma_8$ is the amplitude of matter fluctuations on 8 Mpc/$h$ scales). Without the SH0ES measurement in the fit, CMB and large-scale-structure data show no evidence for EDE [[16]]. Fixing one tension while worsening another is not the behaviour of a correct theory. The $S_8$ side has since moved: the final Kilo-Degree Survey analysis (2025) finds $S_8 = 0.815^{+0.016}_{-0.021}$, consistent with Planck at $0.73\sigma$ [[17]], so that particular objection is weaker than it was in 2020.

A second early-time route changes recombination itself, for instance through a slightly larger electron mass; combined CMB and DESI BAO data mildly prefer such models in a spatially curved universe [[12]].

Modified Gravity and Interacting Dark Energy

A zoo of alternatives modifies either the late-time or early-time expansion history. These include:

  • Phantom dark energy: $w < -1$, which increases $H_0$ inferred from CMB fits, although supernovae and BAO exclude the high-$H_0$ phantom models that Planck alone allows [[2]]
  • Dynamical dark energy: $w \neq -1$, potentially evolving
  • Interacting dark matter/dark energy: energy or momentum transfer between sectors modifying both the background expansion and perturbation growth
  • Modified gravity theories (Horndeski, bimetric gravity, $f(R)$ theories): these change the relationship between curvature and matter, altering $H(z)$

None of these is clearly preferred by the data in isolation, but several of them become more interesting in light of DESI.

The Local Void

A tempting classical explanation: if we happen to live inside a large underdense region (a “Hubble bubble”), the local expansion rate measured by the distance ladder would be higher than the cosmic mean. In an underdense region, there is less gravitational deceleration, so things expand faster locally.

The problem is depth and scale. In linear perturbation theory a region with density contrast $\delta$ expands faster than the mean by

$$\frac{\delta H}{H} = -\frac{f\,\delta}{3}, \qquad f \approx \Omega_m^{0.55} \approx 0.53,$$

where $f$ is the linear growth rate. Raising $H_0$ by 5.64 km/s/Mpc, or 8.4%, needs $\delta \approx -0.47$, half the mean density, where linear theory is only a guide; a 20% underdensity supplies 2.4 km/s/Mpc. Such a void would also have to extend beyond $z \approx 0.5$ to stay hidden in the supernova and BAO Hubble diagrams. We are not in an anomalously underdense region at that scale: estimates of density fluctuations on these scales are about a factor of 20 too small, and the explanation is considered ruled out [[14]].

DESI: Dark Energy May Not Be Constant

In 2024, the Dark Energy Spectroscopic Instrument (DESI) published its first-year results [[18]]. DESI is measuring baryon acoustic oscillations (BAOs) in the distribution of millions of galaxies — the same acoustic physics as the CMB, but imprinted in the late-universe galaxy distribution rather than in photons at $z_\ast \approx 1090$.

The BAO standard ruler is the sound horizon at the end of the baryon drag epoch, $r_d \approx 147$ Mpc, which appears as a preferred separation between galaxy pairs. Across and along the line of sight the BAO peak measures $D_M(z)/r_d$ and $c/[H(z)\,r_d]$; at multiple redshifts this traces $H(z)$ across cosmic time, but on its own it fixes only the product $H_0 r_d$. Calibrating $r_d$ with the baryon density from Big Bang nucleosynthesis gives $H_0 = 68.51 \pm 0.58$ km/s/Mpc from DESI DR2 in $\Lambda$CDM, without any CMB anisotropy data but still assuming standard physics before recombination [[13]]: $3.8\sigma$ below the SH0ES value used here.

DESI DR1 measured BAOs in over 6 million extragalactic objects spanning $0.1 < z < 4.2$ and found a $2.5$–$3.9\sigma$ preference for dark energy that evolves with time (the significance depending on which supernova dataset is used in the combination) [[18]]. The standard model assumes $w = P/\rho = -1$ exactly (a cosmological constant). DESI’s data is better fit by the $w_0$–$w_a$ parameterisation:

$$w(a) = w_0 + w_a (1 - a)$$

where $a = 1/(1+z)$ is the scale factor. With CMB data and the DES five-year supernovae, DR1 gives $w_0 \approx -0.73$ and $w_a \approx -1.05$ [[18]]: a dark energy whose equation of state was more negative in the past, phantom ($w < -1$) before $z \approx 0.35$, and is above $-1$ today. DESI DR2 (released in March 2025) found a preference over $\Lambda$CDM of $3.1\sigma$ for DESI BAO plus CMB. Adding supernovae changed the preference to between $2.8\sigma$ and $4.2\sigma$, depending on the supernova compilation [[13]]. The upper number is therefore not a standalone DESI detection.

The connection to the Hubble tension is model-dependent. The CMB value $H_0 = 67.4$ km/s/Mpc is inferred assuming $w=-1$ within $\Lambda$CDM, so allowing $w(z)$ to vary changes the inferred expansion history. But the phantom-to-quintessence evolution favoured by the DESI DR2 combinations does not automatically move $H_0$ towards SH0ES; published fits can leave or even worsen that tension. DESI’s own $w_0$–$w_a$ fits give $H_0 = 67.51$, $65.91$ and $66.74$ km/s/Mpc with the Pantheon+, Union3 and DES five-year supernovae, and $63.6^{+1.6}_{-2.1}$ with DESI BAO and the CMB but no supernovae [[13]].

DESI may be showing a departure from $\Lambda$CDM, but it has not shown the resolution of the Hubble tension. The distance ladder measures $H_0$ today from local observations; the CMB value is inferred through a cosmological model. Changing that model changes the inference, but not necessarily in the direction required.

This is not yet settled. DESI’s own conclusion is conditional: “Unless there is an unknown systematic error associated with one or more datasets, it is clear that ΛCDM is being challenged by the combination of DESI BAO with other measurements” [[13]]. The statistical significance is below $5\sigma$ and depends on the dataset combination. That dependence is part of the result, not a footnote to it.

JWST and the Early Galaxy Problem

A brief digression — or perhaps not a digression. JWST was designed partly to study the first galaxies. What it has found is unexpected: there are galaxies at $z > 10$ (less than 500 million years after the Big Bang) that are brighter and more abundant than many galaxy formation models within $\Lambda$CDM predicted [[19]]. Early headlines announced that JWST was “breaking cosmology.” The reality is more nuanced: $\Lambda$CDM is not broken by these observations, and some of the most extreme early candidates have been revised to lower redshifts as spectra were taken; one candidate at $z \approx 16$ turned out to lie at $z = 4.9$ [[19]]. But genuine tension persists for some objects.

These are not four measurements of one discrepancy. The Hubble and $S_8$ tensions involve different observables; early-galaxy abundance also depends on astrophysical modelling; and DESI’s dark-energy preference appears only in particular dataset combinations. They are legitimate pressure tests of $\Lambda$CDM, but their significances cannot be added and they do not all point along one parameter direction.

What It Would Mean if $\Lambda$CDM Is Wrong

Let me be clear about what “wrong” means here. $\Lambda$CDM is an extraordinarily successful model. It predicted the angular positions of the CMB acoustic peaks before they were measured. It correctly describes the large-scale structure of the universe across twelve billion years of cosmic history. It accounts successfully for primordial helium and deuterium through Big Bang nucleosynthesis, while the long-standing lithium-7 discrepancy remains. The 2011 Nobel Prize in Physics was awarded “for the discovery of the accelerating expansion of the Universe through observations of distant supernovae” — the observation behind the $\Lambda$ in $\Lambda$CDM.

Abandoning this model is not a decision to take lightly, and no serious cosmologist is proposing to do so. What is being proposed is that $\Lambda$CDM is incomplete in a specific way.

$\Lambda$CDM is not a fundamental theory. In the common six-parameter CMB fit, the sampled parameters describe the baryon and cold-dark-matter densities, the angular acoustic scale, the primordial amplitude and tilt, and the optical depth; $H_0$ is derived within the model rather than directly measured by the CMB. The model does not explain what dark matter actually is — whether it is a WIMP, an axion, a primordial black hole, or something stranger. It does not explain why the cosmological constant $\Lambda$ has the value it does. The model describes; it does not explain.

The Hubble tension, if it survives further scrutiny and grows in significance, would tell us something specific: the expansion history of the universe, or the physics that sets the sound horizon, is not what $\Lambda$CDM predicts. Either there is new physics before recombination that shrinks the sound horizon (EDE, modified recombination, modified gravity at early times) or the later expansion differs from $\Lambda$CDM, although DESI’s evolving dark energy, as fitted so far, does not supply the higher $H_0$. In most proposals the fix is a modification of the Friedmann equation, Eq. (1) — a correction to how we model the $\Omega$ parameters or their time dependence. This is science working as it should: a model that has been extraordinarily successful is now encountering its limits, and those limits are pointing towards something new.

The Top of the Same Ladder

I write on this blog about transit photometry — measuring the dimming of starlight as a planet crosses its star’s disc (see, for instance, exoplanet hunting with smartphones or the gift of transits). Those observations work because we trust the geometric relationship between angular size, distance, and physical size. The same geometric trust is what underpins Rung 1 of the distance ladder: parallax.

What is striking to me, as someone trained in physics, is that the Hubble tension sits at the top of the same ladder I describe at the bottom. Parallax gives distances to nearby stars. Those calibrate Cepheids. Those calibrate supernovae. Those supernovae, at redshifts below 0.15, give $H_0 = 73$ from their recession velocities — measured by spectroscopy and interpreted through general relativity. Meanwhile, the acoustic oscillations in the CMB, a snapshot of the early universe, give $H_0 = 67$ by a completely independent method. The two answers disagree at about $5\sigma$.

The ladder has survived important cross-checks. JWST agrees with the HST Cepheid distances in overlapping hosts. Independent CMB experiments broadly support the Planck inference. DESI finds a dataset-dependent preference for evolving dark energy. The discrepancy is difficult to dismiss, while leaving more than one class of explanation open.

We do not yet know what.

I find this situation — a clean empirical crisis, well-measured, unexplained — to be among the most exciting things happening in physics. Not because I enjoy confusion, but because clean empirical crises are where physics makes progress. The perihelion of Mercury was an annoying discrepancy with Newtonian gravity until Einstein showed it was a signature of spacetime curvature. The ultraviolet catastrophe in blackbody radiation was an embarrassing failure until Planck (Max, not the satellite) introduced the quantum hypothesis. The Hubble tension may be the next one. Or it may turn out to be a mundane systematic that everyone missed. Either answer would be interesting.

For now, a local measurement and a model-dependent early-universe inference do not agree. The work is to find out whether the mismatch lies in the data, the model, or both.


References

Literature and calculations checked through 2026-10-24.

[1] Riess, A. G., Yuan, W., Macri, L. M., Scolnic, D., et al. (2022). A Comprehensive Measurement of the Local Value of the Hubble Constant with 1 km/s/Mpc Uncertainty from the Hubble Space Telescope and the SH0ES Team. The Astrophysical Journal Letters, 934, L7. DOI: 10.3847/2041-8213/ac5c5b

[2] Planck Collaboration (Aghanim, N., et al.). (2020). Planck 2018 results. VI. Cosmological parameters. Astronomy & Astrophysics, 641, A6. DOI: 10.1051/0004-6361/201833910

[3] Gaia Collaboration (Brown, A. G. A., et al.). (2021). Gaia Early Data Release 3: Summary of the contents and survey properties. Astronomy & Astrophysics, 649, A1. DOI: 10.1051/0004-6361/202039657

[4] Riess, A. G., Anand, G. S., Yuan, W., Casertano, S., et al. (2024). JWST Observations Reject Unrecognized Crowding of Cepheid Photometry as an Explanation for the Hubble Tension at 8σ Confidence. The Astrophysical Journal Letters, 962, L17. DOI: 10.3847/2041-8213/ad1ddd

[5] Freedman, W. L. (2021). Measurements of the Hubble Constant: Tensions in Perspective. The Astrophysical Journal, 919, 16. DOI: 10.3847/1538-4357/ac0e95

[6] Freedman, W. L., Madore, B. F., Hoyt, T. J., Jang, I. S., Lee, A. J., & Owens, K. A. (2025). Status Report on the Chicago-Carnegie Hubble Program (CCHP): Measurement of the Hubble Constant Using the Hubble and James Webb Space Telescopes. The Astrophysical Journal, 985, 203. DOI: 10.3847/1538-4357/adce78. An erratum (ApJ, 993, 252) corrects an appendix figure and does not affect $H_0$.

[7] Riess, A. G., Scolnic, D., Anand, G. S., Breuval, L., et al. (2024). JWST Validates HST Distance Measurements: Selection of Supernova Subsample Explains Differences in JWST Estimates of Local H0. The Astrophysical Journal, 977, 120. DOI: 10.3847/1538-4357/ad8c21

[8] H0DN Collaboration (Casertano, S., et al.). (2026). The Local Distance Network: A community consensus report on the measurement of the Hubble constant at ∼1% precision. Astronomy & Astrophysics, 708, A166. DOI: 10.1051/0004-6361/202557993

[9] Birrer, S., Shajib, A. J., Galan, A., Millon, M., et al. (2020). TDCOSMO IV: Hierarchical time-delay cosmography – joint inference of the Hubble constant and galaxy density profiles. Astronomy & Astrophysics, 643, A165. DOI: 10.1051/0004-6361/202038861

[10] TDCOSMO Collaboration (Birrer, S., et al.). (2025). TDCOSMO 2025: Cosmological constraints from strong lensing time delays. Astronomy & Astrophysics, 704, A63. DOI: 10.1051/0004-6361/202555801

[11] Louis, T., La Posta, A., Atkins, Z., Jense, H. T., et al. (2025). The Atacama Cosmology Telescope: DR6 power spectra, likelihoods and ΛCDM parameters. Journal of Cosmology and Astroparticle Physics, 2025(11), 062. DOI: 10.1088/1475-7516/2025/11/062

[12] Camphuis, E., Quan, W., Balkenhol, L., Khalife, A. R., et al. (2026). SPT-3G D1: CMB temperature and polarization power spectra and cosmology from 2019 and 2020 observations of the SPT-3G main field. Physical Review D, 113, 083504. DOI: 10.1103/7wt3-9v2y

[13] DESI Collaboration (Abdul Karim, M., et al.). (2025). DESI DR2 results. II. Measurements of baryon acoustic oscillations and cosmological constraints. Physical Review D, 112, 083515. DOI: 10.1103/tr6y-kpc6. Preprint March 2025: arXiv:2503.14738

[14] Di Valentino, E., Mena, O., Pan, S., Visinelli, L., et al. (2021). In the realm of the Hubble tension — a review of solutions. Classical and Quantum Gravity, 38(15), 153001. DOI: 10.1088/1361-6382/ac086d

[15] Poulin, V., Smith, T. L., Karwal, T., & Kamionkowski, M. (2019). Early Dark Energy can resolve the Hubble tension. Physical Review Letters, 122, 221301. DOI: 10.1103/PhysRevLett.122.221301

[16] Hill, J. C., McDonough, E., Toomey, M. W., & Alexander, S. (2020). Early dark energy does not restore cosmological concordance. Physical Review D, 102, 043507. DOI: 10.1103/PhysRevD.102.043507

[17] Wright, A. H., Stölzner, B., Asgari, M., Bilicki, M., et al. (2025). KiDS-Legacy: Cosmological constraints from cosmic shear with the complete Kilo-Degree Survey. Astronomy & Astrophysics, 703, A158. DOI: 10.1051/0004-6361/202554908

[18] DESI Collaboration (Adame, A. G., et al.). (2025). DESI 2024 VI: Cosmological constraints from the measurements of baryon acoustic oscillations. Journal of Cosmology and Astroparticle Physics, 2025(02), 021. DOI: 10.1088/1475-7516/2025/02/021. Preprint April 2024: arXiv:2404.03002

[19] Arrabal Haro, P., Dickinson, M., Finkelstein, S. L., Kartaltepe, J. S., et al. (2023). Confirmation and refutation of very luminous galaxies in the early universe. Nature, 622, 707–711. DOI: 10.1038/s41586-023-06521-7


Changelog

  • 2026-07-11: Distinguished the local $H_0$ measurement from the $\Lambda$CDM-dependent CMB inference; added DESI DR2 and JWST source records; made the DESI significance dataset-specific; removed the unsupported claim that evolving dark energy resolves the Hubble tension; and qualified claims that all systematics or multiple independent anomalies point in one direction.

  • 2026-03-22: Updated TDCOSMO quasar count to seven lensed systems and the $H_0$ value to match the TDCOSMO-only analysis. Updated DESI DR1 galaxy count to over 6 million extragalactic objects (the previous figure of 14 million corresponds to DR2). Added qualification that the 3.9$\sigma$ significance for evolving dark energy is dataset-dependent (ranging from 2.5$\sigma$ to 3.9$\sigma$ depending on the supernova sample used).

  • 2026-10-03: Corrected the Early Dark Energy timing (the field becomes important around, not after, matter-radiation equality, and begins oscillating and diluting faster than radiation before recombination, not after it); rounded the Planck acoustic angle to the published 0.5965° (the extra digits matched a different dataset combination); changed the Gaia parallax count from “more than” to “nearly” 1.5 billion; removed the claim that the astro-lab@home post describes the CMB acoustic oscillations and the link attached to it.

  • 2026-10-24: Corrected the acoustic angle from $r_s/D_A$ to $r_\ast/D_M$ with $D_M = (1+z)D_A$, and its sound horizon from 147 to 144.4 Mpc (147 Mpc is the BAO ruler $r_d$). Corrected $\Omega_\Lambda$ and $\Omega_m$ from 0.68 and 0.31 to 0.685 and 0.315; $\Omega_r \approx 0$ to $9 \times 10^{-5}$, dominant before $z \approx 3400$; recombination from 380,000 to 370,000 years; the $5\sigma$ chance probability to a one-sided value; Cepheid distances from “period measurements alone” to periods and apparent brightnesses, and their reach from ~50 to about 80 Mpc; SH0ES’s “fifteen years” to since 2005; the LMC distance from “better than 1%” to about 1.2%, as one of three anchors; the early-dark-energy boost of $H$ from ~10% to 2.6% for a 5% energy fraction; the local void from a 20% to a 47% underdensity reaching beyond $z \approx 0.5$; TDCOSMO from “agrees with SH0ES” to consistent with SH0ES and Planck; the closing’s supernovae from $z \sim 2$ to below 0.15; “consistent with systematic errors in the SN Ia data” to DESI’s own conditional sentence. Added the JWST candidate revised from $z \approx 16$ to 4.9, the TRGB ladder and newer CMB, lensing and local values to October 2026, DESI’s BAO-plus-BBN $H_0$ and the $H_0$ of its evolving-dark-energy fits, and two figures.