Hubble Anisotropy: Directional Hubble Expansion

• Hubble anisotropy is the directional dependence of the cosmic expansion rate, manifesting as variations in the Hubble constant and deceleration parameter across the sky.
• Observational strategies include hemispherical supernova fits, Bayesian maximum-likelihood modeling, and analyses using Tully–Fisher and Cosmicflows data to discern local anisotropic signals.
• Future tests aim to differentiate between local bulk flows, calibration biases, and genuine anisotropic cosmic expansion through enhanced sky-sampling and improved statistical methods.

Hubble anisotropy denotes directional dependence in the inferred cosmic expansion rate, usually formulated as a sky-dependent Hubble parameter, Hubble constant, or low-redshift Hubble diagram. Operationally, it appears when luminosity distance or distance modulus depends on both redshift and line of sight, $D_L(z,\hat n)$ or $\mu(z,\hat n)$, rather than on redshift alone. In the literature, the term covers two distinct but related domains: empirical anisotropy in local determinations of $H_0$ from supernovae, Tully–Fisher distances, and similar tracers; and globally anisotropic cosmologies, especially Bianchi I models, in which different principal directions expand at different rates (Anton et al., 2024, Schucker, 2016).

1. Definition, parameterizations, and observables

At low redshift, anisotropy is usually expressed as a directional Hubble law,

$$D_L(z,\hat n)\approx \frac{z}{H_0(\hat n)}\left[1-\frac{1}{2}\big(1-q_0(\hat n)\big)z+\cdots\right],$$

so that both the intercept $H_0(\hat n)$ and the next cosmographic coefficient $q_0(\hat n)$ may vary across the sky (Anton et al., 2024). A compact phenomenological parameterization is the dipole model

$$H_0(\hat n)=H_{0,\rm iso}\,\left(1+A\,\hat n\cdot\hat p\right),$$

where $A$ is the fractional amplitude and $\hat p$ is the preferred direction (Boubel et al., 2024). In distance-indicator work, the same effect is often written as a dipole in the zero-point or distance modulus, since a directional shift in the absolute calibration translates directly into a directional shift in $H_0$.

In homogeneous but anisotropic cosmology, the standard reference geometry is Bianchi I,

$\mu(z,\hat n)$0

or, in the axially symmetric case, $\mu(z,\hat n)$1 (Schucker, 2016). The corresponding directional Hubble rates are $\mu(z,\hat n)$2, $\mu(z,\hat n)$3, and $\mu(z,\hat n)$4. One convenient anisotropy parameter is the “Hubble stretch” $\mu(z,\hat n)$5, defined by

$\mu(z,\hat n)$6

with present-day value $\mu(z,\hat n)$7 (Schucker, 2016). A different but related background parameterization uses a shear-density contribution,

$\mu(z,\hat n)$8

where $\mu(z,\hat n)$9 measures the present-day fractional shear contribution to $H_0$0 (Yadav, 2023).

These parameterizations separate three conceptually different questions. One is whether the local Hubble diagram contains a dipole or higher multipoles. A second is whether such a signal is caused by local bulk flows, peculiar velocities, calibration anisotropy, or sky-sampling effects. A third is whether the background spacetime itself is anisotropic in the Bianchi sense.

2. Observational strategies and statistical constructions

The classical observational strategy is hemispherical or patch-by-patch fitting of low-redshift supernova Hubble diagrams. In one formulation, the sky is scanned over all possible hemispheres, the Hubble slope is fitted independently on each side, and the asymmetry is quantified by

$H_0$1

This approach was applied to the Constitution sample at $H_0$2 using the truncated low-$H_0$3 luminosity-distance expansion with fixed $H_0$4 (Bahr-Kalus et al., 2012). A related strategy divides the sky into several solid angles with roughly equal supernova counts and compares the $H_0$5 fits between regions, as in the Union2.1 patch analysis (Migkas et al., 2016).

More recent work uses full maximum-likelihood or Bayesian forward modeling with covariance propagation. In Pantheon+ analyses, the fit is performed directly on standardized SN Ia magnitudes with the full or reconstructed covariance matrix, allowing either a dipole in $H_0$6 or a redshift-dependent dipole in $H_0$7 (Sah et al., 2024). Tully–Fisher studies map anisotropy through a direction-dependent zero-point,

$H_0$8

and then interpret $H_0$9 as a spatial variation in $$D_L(z,\hat n)\approx \frac{z}{H_0(\hat n)}\left[1-\frac{1}{2}\big(1-q_0(\hat n)\big)z+\cdots\right],$$0 (Boubel et al., 2024).

A separate methodological line uses exact or numerical spacetime modeling. Ray tracing in inhomogeneous anisotropic solutions computes redshift and luminosity distance directly from null geodesics and Sachs optics, then compares the resulting Hubble diagrams with those inferred from an averaged anisotropic model (Anton et al., 2024). Numerical-relativity simulations instead generate synthetic Pantheon-like catalogs in a general-relativistic large-scale structure spacetime, allowing one to isolate observer-position variance and anisotropic sky-sampling effects in the inferred $$D_L(z,\hat n)\approx \frac{z}{H_0(\hat n)}\left[1-\frac{1}{2}\big(1-q_0(\hat n)\big)z+\cdots\right],$$1 (Macpherson, 2024).

The statistical meaning of a reported anisotropy depends strongly on the chosen frame and correction scheme. Several studies distinguish heliocentric, CMB, Local Group, and peculiar-velocity-corrected redshifts, and the inferred dipole amplitude and direction can change appreciably among them (Sah et al., 2024, Sanejouand, 2023).

3. Supernova constraints and the local Hubble flow

Early low-redshift hemisphere tests already found that the Hubble expansion is not perfectly uniform on the sky, but also that the amplitude is compatible with expected local-structure variance. Using the Constitution set for four different light-curve fitters, the maximal asymmetry direction was found near $$D_L(z,\hat n)\approx \frac{z}{H_0(\hat n)}\left[1-\frac{1}{2}\big(1-q_0(\hat n)\big)z+\cdots\right],$$2, with measured $$D_L(z,\hat n)\approx \frac{z}{H_0(\hat n)}\left[1-\frac{1}{2}\big(1-q_0(\hat n)\big)z+\cdots\right],$$3 and a 95% confidence upper limit $$D_L(z,\hat n)\approx \frac{z}{H_0(\hat n)}\left[1-\frac{1}{2}\big(1-q_0(\hat n)\big)z+\cdots\right],$$4; the amplitude was not in contradiction to expectations from the $$D_L(z,\hat n)\approx \frac{z}{H_0(\hat n)}\left[1-\frac{1}{2}\big(1-q_0(\hat n)\big)z+\cdots\right],$$5CDM model (Bahr-Kalus et al., 2012).

A more detailed Union2.1 analysis divided the sky into nine solid angles and also into Galactic hemispheres. One region, “Group X,” defined by $$D_L(z,\hat n)\approx \frac{z}{H_0(\hat n)}\left[1-\frac{1}{2}\big(1-q_0(\hat n)\big)z+\cdots\right],$$6 and $$D_L(z,\hat n)\approx \frac{z}{H_0(\hat n)}\left[1-\frac{1}{2}\big(1-q_0(\hat n)\big)z+\cdots\right],$$7, contained 82 SNIa and initially showed a non-overlapping $$D_L(z,\hat n)\approx \frac{z}{H_0(\hat n)}\left[1-\frac{1}{2}\big(1-q_0(\hat n)\big)z+\cdots\right],$$8 contour in the $$D_L(z,\hat n)\approx \frac{z}{H_0(\hat n)}\left[1-\frac{1}{2}\big(1-q_0(\hat n)\big)z+\cdots\right],$$9 plane relative to the remaining 464 SNIa, with $H_0(\hat n)$0. However, the discrepancy was driven by three “erratic” SNIa—03D4cx, g050, and 2005hv. Jointly removing them raised the common-area fraction from 0% to $H_0(\hat n)$1, effectively eliminating the apparent anisotropy (Migkas et al., 2016). The same study also noted deviant low-$H_0(\hat n)$2 supernovae, including 1997k, 1997l, and 1997o, as possible signatures of imperfect Galactic extinction correction.

Pantheon+ enlarged the low-redshift sample and shifted the discussion toward the local distance ladder. A hemispherical decomposition found angular variations up to $H_0(\hat n)$3 in $H_0(\hat n)$4 in the SH0ES redshift range $H_0(\hat n)$5, with a maximum contrast $H_0(\hat n)$6 at $H_0(\hat n)$7 and $H_0(\hat n)$8 in the exact CMB dipole direction. The larger $H_0(\hat n)$9 hemisphere encompassed the CMB dipole direction, and the variation was driven largely by hemispheric differences in Cepheid-calibrated absolute magnitudes, with $q_0(\hat n)$0 up to about $q_0(\hat n)$1 mag, though reinforced by Hubble-flow SNe (McConville et al., 2023). The same analysis emphasized that the anisotropy exceeds the quoted SH0ES uncertainty but is not large enough to resolve the early–late $q_0(\hat n)$2 discrepancy.

A complementary Pantheon+ study argued that Planck-parameterized $q_0(\hat n)$3CDM magnitude predictions are not consistent with the full Pantheon+ sample even when $q_0(\hat n)$4 is adjusted, but become consistent if supernovae below $q_0(\hat n)$5 are excluded. It also found that low-redshift subsets roughly centered on the CMB dipole direction can be combined with the high-redshift sample to yield a “quiet flow” with $q_0(\hat n)$6 when both CMB and peculiar-velocity corrections are applied to the redshifts (Sanejouand, 2023). This suggests that low-$q_0(\hat n)$7 anisotropy and frame choice materially affect how well isotropic $q_0(\hat n)$8CDM fits the nearby Hubble diagram.

The most aggressive Pantheon+ anisotropy claim used maximum-likelihood estimators in the heliocentric, CMB, and Local Group frames. In the range $q_0(\hat n)$9, it reported dipolar modulation of the Hubble expansion rate exceeding $$H_0(\hat n)=H_{0,\rm iso}\,\left(1+A\,\hat n\cdot\hat p\right),$$0 in all frames, with best-fit amplitudes of order $$H_0(\hat n)=H_{0,\rm iso}\,\left(1+A\,\hat n\cdot\hat p\right),$$1–$$H_0(\hat n)=H_{0,\rm iso}\,\left(1+A\,\hat n\cdot\hat p\right),$$2, and a redshift-dependent dipole in the deceleration parameter at $$H_0(\hat n)=H_{0,\rm iso}\,\left(1+A\,\hat n\cdot\hat p\right),$$3 significance (Sah et al., 2024). One interpretation advanced there is that the local signal is not a cosmological-constant effect but a general-relativistic consequence of an anomalous bulk flow in a tilted local universe.

4. Tully–Fisher maps, Cosmicflows, and non-supernova approaches

The all-sky Cosmicflows-4 Tully–Fisher sample provides a different route to Hubble anisotropy because anisotropy appears as a differential shift in the TF zero-point rather than as an absolute calibration shift. Using the WISE W1 band and a cut $$H_0(\hat n)=H_{0,\rm iso}\,\left(1+A\,\hat n\cdot\hat p\right),$$4, one study found a best-fit dipole amplitude $$H_0(\hat n)=H_{0,\rm iso}\,\left(1+A\,\hat n\cdot\hat p\right),$$5 mag toward $$H_0(\hat n)=H_{0,\rm iso}\,\left(1+A\,\hat n\cdot\hat p\right),$$6, corresponding to $$H_0(\hat n)=H_{0,\rm iso}\,\left(1+A\,\hat n\cdot\hat p\right),$$7 for $$H_0(\hat n)=H_{0,\rm iso}\,\left(1+A\,\hat n\cdot\hat p\right),$$8, or about a 3% variation, with nominal significance $$H_0(\hat n)=H_{0,\rm iso}\,\left(1+A\,\hat n\cdot\hat p\right),$$9 (Boubel et al., 2024). Yet the same analysis reported that a pure velocity dipole is strongly favored over a pure $A$0 dipole, and that current data do not robustly separate anisotropic expansion from bulk flow.

A later shell-by-shell Cosmicflows-4 analysis mapped $A$1 over equal-area sky patches in radial shells between distance moduli $A$2. For uncorrected observed velocities, the nominal range $A$3 yielded a coherent dipole toward $A$4 with amplitude $A$5; after peculiar-velocity correction, the amplitude dropped to $A$6 and the direction moved closer to the CMB dipole (Salzano et al., 2 Dec 2025). The same work concluded that the dipole decreases with distance, higher multipoles are subdominant, and the effect on the large-scale Hubble tension is limited because the distribution of SN Ia host galaxies used in calibration and Hubble-flow work does not show a strong correlation with the dipole signal.

A distinct, nonstandard explanatory program modeled Hubble anisotropy through intergalactic electron-density gradients. In that framework, line-of-sight averages of $A$7 determine the inferred $A$8, and rim, auto-gravitating, Voronoi, and 2MRS-based models were reported to reproduce the observed average and variance of all-sky $A$9 maps (Zaninetti, 2014). This mechanism is conceptually separate from FLRW, Bianchi, and bulk-flow analyses, but it illustrates the breadth of proposed interpretations of anisotropic $\hat p$0 maps.

5. Homogeneous anisotropic cosmologies and exact spacetime treatments

The standard homogeneous anisotropic template is Bianchi I. Fitting the JLA sample with an axially symmetric Bianchi I model, one analysis obtained

$\hat p$1

described as an “intriguing, yet non-significant” preferred direction (Schucker, 2016). The inferred amplitude was consistent with isotropy at about $\hat p$2, and the paper stressed that improved sampling from LSST should verify or falsify the signal.

A different class of models adds anisotropy only to the averaged background expansion through a shear term proportional to $\hat p$3. In the anisotropic extension $\hat p$4CDM$\hat p$5 constrained by BAO, BBN, CC, Pantheon+, and SH0ES, the allowed anisotropy is extremely small, with upper bounds $\hat p$6 or $\hat p$7 at 95% confidence, but it is positively correlated with $\hat p$8 and can reduce the $\hat p$9 tension by about $H_0$0 in the CMB-independent combinations considered (Yadav, 2023). Because this formalism modifies only the average $H_0$1, it does not itself model directional observables.

More stringent analyses of Bianchi I dynamics argue that such shear cannot produce an observationally relevant present-day Hubble anisotropy. In one treatment, fractional directional differences in the Hubble rates decay as $H_0$2 in radiation domination and $H_0$3 in matter domination, while CMB and BBN constraints drive the present-day directional Hubble anisotropy to a negligible level (Hertzberg et al., 2024). In another exact dust+$H_0$4 solution, present-day bounds such as $H_0$5 imply

$H_0$6

so the maximal directional variation in $H_0$7 is many orders of magnitude too small to account for the Hubble tension (Grøn, 2024). A late-time angle-averaged Bianchi I reanalysis reached the same qualitative conclusion: typical fits give $H_0$8–$H_0$9, implying $\mu(z,\hat n)$00–$\mu(z,\hat n)$01 today, far too small to matter for the SH0ES–Planck discrepancy (Deliyergiyev et al., 21 Oct 2025).

Exact ray tracing in statistically homogeneous anisotropic universes adds an important geometric qualification. In plane-symmetric inhomogeneous dust models, averaged anisotropic Bianchi I or Bianchi V descriptions reproduce the ensemble-averaged Hubble diagram only when a statistical homogeneity scale exists on the averaging foliation; when no such scale exists, the averaged model can fail even for the mean Hubble diagram (Anton et al., 2024). This suggests that “anisotropic expansion” and “anisotropic observations” are not interchangeable notions.

The strongest geometric claim appears in a torsion-based response to the inflationary objection. There, the argument is that if inflation isotropises spatial geometry and spacetime is strictly metric, then the Hubble field must also be isotropic; conversely, a confirmed Hubble anisotropy would require either anisotropy built into inflation or non-metric spacetime geometry, with torsion presented as the natural option (McInnes, 9 Aug 2025).

6. Systematics, interpretation, and future tests

The recurring difficulty in this subject is that local directional structure is easier to produce than a genuine anisotropic background cosmology. Supernova patch analyses show that a few outliers can mimic anisotropy; low Galactic latitude objects can carry extinction-related residuals; non-uniform sky coverage alters the redshift distribution by patch; and keeping global rather than region-specific nuisance parameters can imprint calibration asymmetries (Migkas et al., 2016). This suggests that apparently significant regional tensions must be tested against outlier removal, covariance treatment, and calibration refits before being interpreted cosmologically.

General-relativistic simulations reinforce the same caution. In numerical-relativity synthetic Pantheon-like catalogs, the variance in $\mu(z,\hat n)$02 between different observer positions is about 1–2% when the sky is sampled isotropically, but the inferred value can vary by 4–6% when the same anisotropic Pantheon sample is simply rotated on the observer’s sky (Macpherson, 2024). This indicates that incomplete angular sampling can bias isotropic $\mu(z,\hat n)$03 fits even when the underlying issue is local inhomogeneity rather than a global anisotropic metric.

Current survey forecasts are therefore framed less as searches for a decisive dipole and more as discrimination problems between bulk flow, calibration anisotropy, and genuine expansion anisotropy. For Tully–Fisher data, combined WALLABY and DESI mocks imply that a 1% $\mu(z,\hat n)$04 dipole should be detectable at $\mu(z,\hat n)$05 and distinguishable from the typical bulk flow predicted by $\mu(z,\hat n)$06CDM (Boubel et al., 2024). For ZTF-like SN Ia samples, realistic simulations recover an injected dipole of $\mu(z,\hat n)$07 with an amplitude error of $\mu(z,\hat n)$08 and direction uncertainties of $\mu(z,\hat n)$09 in right ascension and $\mu(z,\hat n)$10 in declination, while remaining insensitive to the fiducial isotropic $\mu(z,\hat n)$11 and to sky footprint (Barjou-Delayre et al., 16 Jan 2026). In the older Bianchi I supernova forecast, LSST was expected to reduce $\mu(z,\hat n)$12 uncertainties to $\mu(z,\hat n)$13 after 1 year and $\mu(z,\hat n)$14 after 10 years, with axis localization to a few degrees (Schucker, 2016).

Across these studies, the empirical pattern is consistent: low-redshift directional structure in the Hubble flow is repeatedly reported at the few-percent level and often near the CMB-dipole or major bulk-flow directions, but robust evidence for a primordial or globally anisotropic Hubble expansion remains unestablished. The central lesson is methodological rather than rhetorical. Hubble anisotropy is a legitimate observable, but its interpretation depends on whether the signal survives control of outliers, peculiar velocities, sky-sampling anisotropy, calibration inhomogeneity, and frame choice. Only after those effects are isolated can a local $\mu(z,\hat n)$15 dipole be promoted from an environmental or survey effect to a statement about cosmic spacetime.