CMB Quadrupole-Octopole Alignment

• Quadrupole–Octopole alignment is the observed near-coincidence in the preferred axes of the CMB’s ℓ=2 and ℓ=3 multipoles, challenging the assumption of random isotropy.
• Different analysis methods and preprocessing techniques like masking, inpainting, and foreground subtraction significantly affect the measured alignment and its statistical significance.
• Upcoming polarization studies and cross-checks using alternative data sets aim to determine whether this anomaly is a statistical fluke or indicative of new cosmological physics.

Searching arXiv for recent and foundational work on quadrupole–octopole alignment to ground the article in the literature. Quadrupole–octopole alignment denotes the apparent non-random mutual orientation of the preferred axes associated with the CMB temperature quadrupole ($\ell=2$) and octopole ($\ell=3$). In statistically isotropic Gaussian cosmologies these low-$\ell$ modes are expected to be statistically independent, so an unusually small separation angle between their preferred axes is treated as a large-angle anomaly. The effect has been reported in multiple WMAP and Planck analyses, but its quantitative significance is strongly treatment-dependent: in Planck PR4 Commander the angle is $\theta_{2,3}\simeq 14.8^\circ$, with $p=0.08$ using 100 Planck-like FFP simulations and $p=0.035$ using 5000 pure CMB simulations, i.e. a $\sim2\sigma$ level of marginal significance (Aluri et al., 28 Jun 2025).

1. Definition and null hypothesis

The basic object is a pair of axes, $\hat n_2$ and $\hat n_3$, extracted from the $\ell=2$ and $\ell=3$0 harmonic patterns. Their mutual orientation is usually expressed either as a dot product or as an angular separation,

$\ell=3$1

Equivalently, some analyses define $\ell=3$2 (0909.2495).

Under the null hypothesis of statistically isotropic Gaussian fluctuations, the quadrupole and octopole axes are random and independently distributed on the sphere. In that case $\ell=3$3 is uniformly distributed on $\ell=3$4, with

$\ell=3$5

so small separation angles are rare by construction (Aluri et al., 2010). A frequently quoted benchmark is $\ell=3$6, for which $\ell=3$7 and the corresponding $\ell=3$8CDM probability is $\ell=3$9 when one ignores a posteriori selection effects (Melia, 2012).

The anomaly is therefore not the existence of preferred axes per se, but the claim that the two lowest nontrivial multipoles define axes that are more nearly coincident than expected from an isotropic ensemble. A recurrent source of confusion is that the effect concerns phase geometry, not merely low quadrupole power. Several analyses explicitly separate low power, planarity, and alignment, and recent Planck PR4 work finds that unusually planar modes are a subset of intrinsically anisotropic modes, so they may not be intrinsically planar (Aluri et al., 28 Jun 2025).

2. Operational definitions and statistics

Several non-identical but widely used constructions assign a preferred direction to each multipole.

The angular-momentum-dispersion method searches for the axis $\ell$0 maximizing

$\ell$1

or, equivalently,

$\ell$2

This is the basis of many WMAP- and Planck-era alignment analyses (0909.2495).

The power-tensor formalism instead constructs a $\ell$3 tensor from the harmonic coefficients. In the Planck PR4 analysis,

$\ell$4

with $\ell$5 for an isotropic field. Diagonalizing $\ell$6 yields eigenvalues $\ell$7 and orthonormal eigenvectors; the principal eigenvector is the eigenvector corresponding to $\ell$8 (Aluri et al., 28 Jun 2025). Radio-galaxy analyses at $\ell$9 use the same formalism to define multipole directions outside the CMB context (Tiwari et al., 2018).

A third family of statistics uses multipole vectors. The quadrupole is encoded by two unit vectors and the octopole by three; cross products define one quadrupole area vector and three octopole area vectors. From these, rotation-invariant estimators such as

$\theta_{2,3}\simeq 14.8^\circ$0

probe how unusually parallel the low-$\theta_{2,3}\simeq 14.8^\circ$1 structures are (Polastri et al., 2015).

These statistics are not numerically identical, and that distinction matters. Analyses based on $\theta_{2,3}\simeq 14.8^\circ$2, on $\theta_{2,3}\simeq 14.8^\circ$3, or on multipole-vector $\theta_{2,3}\simeq 14.8^\circ$4 and $\theta_{2,3}\simeq 14.8^\circ$5 can return different tail probabilities for the same sky. This suggests that the anomaly is method-sensitive at the level of inference even when the underlying geometric impression remains similar.

3. Observational history from WMAP to Planck PR4

Early WMAP studies reported very small quadrupole–octopole separation angles. In a local-ISW analysis using WMAP ILC, the pre-subtraction values were $\theta_{2,3}\simeq 14.8^\circ$6 with $\theta_{2,3}\simeq 14.8^\circ$7 for ILC, and $\theta_{2,3}\simeq 14.8^\circ$8 with $\theta_{2,3}\simeq 14.8^\circ$9 for the OT cleaned map (0909.2495). In foreground-cleaning studies, WMAP ILC 5 yr gave $p=0.08$0 with $p=0.08$1, and WMAP ILC 7 yr gave $p=0.08$2 with $p=0.08$3 (Aluri et al., 2010).

Planck-era results remained nontrivial but were less uniform. Copi et al. found that after Doppler-quadrupole correction the p-values for $p=0.08$4 at least as large as observed were $p=0.08$5 for Planck NILC, $p=0.08$6 for Planck SEVEM, and $p=0.08$7 for Planck SMICA; they summarized the mutual alignment as $p=0.08$8 in all three Planck maps studied (Copi et al., 2013). By contrast, analyses using full-sky LGMCA maps found much weaker original alignment in some renditions, including $p=0.08$9, $p=0.035$0, and $p=0.035$1 for PR1-LGMCA before secondary-signal subtraction (Rassat et al., 2014).

A representative set of reported values illustrates the spread:

Data/treatment	Separation angle	Tail probability
WMAP ILC pre-ISW subtraction	$p=0.035$2	$p=0.035$3
WMAP9 ILC pre-inpainting	$p=0.035$4	$p=0.035$5
WMAP9 ILC after inpainting + ISW subtraction	$p=0.035$6	$p=0.035$7
Planck PR4 Commander	$p=0.035$8	$p=0.035$9–$\sim2\sigma$0

The recent Planck PR4 Commander analysis places the effect in a milder regime. Using the power tensor on the Commander map at $\sim2\sigma$1, the quadrupole and octopole principal eigenvectors satisfy $\sim2\sigma$2; the authors conclude that the quadrupole–octopole alignment persists in PR4 Commander at roughly the $\sim2\sigma$3 level, but that its significance has diminished compared to earlier, foreground-corrected analyses (Aluri et al., 28 Jun 2025). The same study also reports that, besides $\sim2\sigma$4, higher multipoles aligned with the quadrupole are insignificant.

4. Sensitivity to masking, foregrounds, and secondary anisotropies

The principal methodological issue is that the lowest multipoles are exceptionally sensitive to sky cuts and low-level additive components. Sparse-inpainting studies found that the raw masked WMAP quadrupole and octopole are extremely well aligned, but that filling the Galactic mask weakens the alignment by shifting the quadrupole axis more than the octopole axis; for WMAP9 ILC, $\sim2\sigma$5, $\sim2\sigma$6 before inpainting becomes $\sim2\sigma$7, $\sim2\sigma$8 after sparse inpainting (Rassat et al., 2013).

Foreground residuals are debated but not negligible. Analytic and numerical IPSE studies found that residual foregrounds and low-$\sim2\sigma$9 negative bias shift the principal eigenvectors only at the few-degree level and in random directions, so they do not systematically push $\hat n_2$0 and $\hat n_2$1 into alignment; if anything, they introduce random misalignments (Aluri et al., 2010). The PR4 analysis reaches a compatible conclusion from a different angle: because no explicit foreground-bias correction map was subtracted before computing $\hat n_2$2, any residual foregrounds will tend to de-align intrinsically correlated multipole axes, making the present $\hat n_2$3 result a conservative underestimate of the true alignment (Aluri et al., 28 Jun 2025).

Other work has argued for a stronger Galactic-residual connection. A variance-based analysis of WMAP 5 yr reported that removing the quadrupole and octopole from both data and simulations makes the variance anomaly disappear, and concluded that the variance anomaly and the previously reported quadrupole and octopole alignment seem to be related and could have a common origin, with Galactic foreground residuals described as the most likely one (Cruz et al., 2010). This does not establish causation, but it shows that the anomaly is entangled with the morphology and cleaning of the lowest modes.

Secondary anisotropies, especially the local ISW signal and the kinetic Doppler quadrupole, materially alter the inferred significance. Subtracting an estimated $\hat n_2$4 ISW map from WMAP data shifts the alignment from $\hat n_2$5 to $\hat n_2$6 in ILC and from $\hat n_2$7 to $\hat n_2$8 in OT, raising p-values to $\hat n_2$9 and $\hat n_3$0, respectively (0909.2495). A broader analysis using 2MASS and NVSS reached a similar trend, finding that after inpainting and ISW subtraction typical separations move into the $\hat n_3$1–$\hat n_3$2 range with $\hat n_3$3–$\hat n_3$4 (Rassat et al., 2013). By contrast, proper frequency-dependent removal of the kinetic Doppler quadrupole can strengthen the anomaly: in SMICA 2013, the alignment significance for $\hat n_3$5 rises from $\hat n_3$6 without DQ removal to $\hat n_3$7 with the correct frequency-dependent factor $\hat n_3$8 (Notari et al., 2015).

A plausible implication is that no single nuisance mechanism has been shown to dominate all analyses. Masking, inpainting, DQ treatment, ISW subtraction, and map-making pipeline all move the estimate by amounts comparable to the nominal anomaly itself.

5. Competing interpretations and model-building

The most conservative reading is that the alignment is either a low-probability realization of an isotropic sky or an unstable statistic dominated by quadrupole-specific systematics. Several papers explicitly caution that posterior selection biases complicate interpretation; one estimate gives the low-$\hat n_3$9 multipole alignment an unlikely level of $\ell=2$0 in $\ell=2$1CDM before accounting for such bias (Melia, 2012).

Beyond that baseline, multiple cosmological mechanisms have been tested. A phenomenological dipolar modulation model of the form

$\ell=2$2

does not alleviate the anomaly: Monte Carlo tests of eight estimators show that the quadrupole/octopole and dipole/quadrupole/octopole alignments remain anomalous at roughly the $\ell=2$3–$\ell=2$4 C.L. in both $\ell=2$5CDM and the dipolar model, with virtually indistinguishable PTEs (Polastri et al., 2015).

A scale-dependent dipolar modulation performs differently. With

$\ell=2$6

Bayesian inference gives $\ell=2$7, $\ell=2$8, and a direction $\ell=2$9 with a $\ell=3$00 $\ell=3$01 contour; under this model the three alignment estimators increase their p-values by about $\ell=3$02, from $\ell=3$03 to $\ell=3$04 for $\ell=3$05, from $\ell=3$06 to $\ell=3$07 for $\ell=3$08, and from $\ell=3$09 to $\ell=3$10 for $\ell=3$11 (Marcos-Caballero et al., 2019). The anomaly is therefore reduced but not fully eliminated.

More speculative anisotropic scenarios have also been proposed. In a Randers–Finsler spacetime with a primordial spectrum

$\ell=3$12

the quadrupole–octopole correlation matrix for $\ell=3$13 picks out the $\ell=3$14–$\ell=3$15 plane perpendicular to the privileged axis, and the probability of two directions lying within $\ell=3$16 rises from $\ell=3$17 on $\ell=3$18 to $\ell=3$19 on a privileged $\ell=3$20 (Chang et al., 2013). Relic-vector-field models likewise generate a scale-dependent quadrupolar modulation $\ell=3$21, with typical alignment angles of order $\ell=3$22–$\ell=3$23 for $\ell=3$24 (Chen et al., 2013). In the $\ell=3$25 cosmology, the predicted probability for $\ell=3$26 is $\ell=3$27–$\ell=3$28, compared with $\ell=3$29 in $\ell=3$30CDM (Melia, 2012).

These models do not have a uniform status. Some only increase the prior probability of near-alignment; others fit different anomalies simultaneously; at least one widely discussed dipolar prescription fails to improve over $\ell=3$31CDM (Polastri et al., 2015). This suggests that model-building around the effect remains exploratory rather than convergent.

6. External cross-checks and prospective tests

A critical question is whether an analogous large-scale alignment appears in tracers other than the CMB temperature. In a full-sky radio-galaxy map built from NVSS and SUMSS, Tiwari and Aluri found no dipole–quadrupole–octopole alignment: the quadrupole direction is roughly $\ell=3$32 away from the dipole, the octopole direction is approximately $\ell=3$33 from the dipole, and the angle between quadrupole and octopole is around $\ell=3$34 degree. Despite large directional errors due to shot noise, the analysis states that dipole–quadrupole and quadrupole–octopole alignment can be ruled out in this data set (Tiwari et al., 2018). This non-detection tightly constrains late-time or large-scale anisotropic models invoked to explain the CMB anomaly.

The most important forthcoming cross-check is large-scale $\ell=3$35-mode polarization. A forecast for AliCPT-1 defines the $\ell=3$36-mode quadrupole–octopole statistic

$\ell=3$37

with isotropic expectation $\ell=3$38 and $\ell=3$39. In 1000 simulations, AliCPT alone gives $\ell=3$40 or $\ell=3$41, biased high by $\ell=3$42 because of partial sky, whereas the combined AliCPT+SO LAT configuration yields $\ell=3$43, statistically indistinguishable from the ideal full-sky noiseless value $\ell=3$44 (Dou et al., 22 Apr 2026). Under the best-case combined configuration, $\ell=3$45 corresponds to $\ell=3$46 CL and $\ell=3$47 to $\ell=3$48 CL.

If the true $\ell=3$49-mode alignment matched the temperature anomaly amplitude $\ell=3$50, the forecasted AliCPT+SO configuration would see $\ell=3$51, i.e. a highly significant detection in polarization (Dou et al., 22 Apr 2026). This suggests that polarization may provide the most decisive near-term discrimination between a temperature-only statistical fluke, a masking-driven artifact, and a genuinely correlated primordial large-angle structure.

The present state of the subject is therefore bifurcated. The alignment remains present enough in cleaned full-sky CMB temperature maps to motivate continued scrutiny, but its nominal significance is not invariant under low-$\ell=3$52 treatment. Independent matter tracers do not reproduce it, while polarization forecasts indicate that a near-cosmic-variance test is feasible with joint northern- and southern-hemisphere data.